Cancer Immunogenomics Approaches to Characterizing and Treating Malignant Brain Tumors

CHAPTER ONE

Introduction to cancer immunology

1.1 Earliest approaches to immunotherapy

The notion that the immune system can recognize and potentially eliminate cancer has only

gained broad acceptance in the medical community over the last few decades, but initial reports

highlighting the potential of immunotherapy date back to the 19th century. In 1866, Wilhelm Busch

described complete remissions in some sarcoma patients who developed erysipelas, a superficial

skin infection, of their postoperative wound1,2. In 1883, Friedrich Fehleisen identified

Streptococcus Pyogenes as the causative agent for erysipelas and reported surprising responses in

some patients with inoperable metastatic disease inoculated with pure cultures of the bacteria3.

However, it was not until the surgeon William Coley, often recognized as the “Father of Cancer

Immunotherapy”, that these anecdotes were formalized into a crude treatment now referred to as

“Coley’s Toxins”4.

Inspired by the numerous case reports in the literature, Coley performed direct injections

of Streptococcus preparations into patient tumors. Many patients experienced serious side effects

such as high fevers, nausea, malaise, and tachycardia. However, a subset of patients did have

significant regression of their otherwise inoperable tumors which Coley published as a case series

in 18935,6. The serious side effects, in addition to the underlying treatment effect, were likely the

result of substantial innate immune activation in response to the bolus of transferred bacteria, but

these mechanisms were very poorly understood by Coley and the medical community at the time.

Despite some successes, his findings were highly controversial and difficult to consistently

replicate. In 1894, the Journal of the American Medical Association criticized the use of his toxins,

while the American Medical Association referenced them as an “entire failure” and an “alleged

remedy”4,7. Perhaps in part due to the considerable doubt of his medical contemporaries, Coley’s

ideas did not gain widespread acceptance.

At the same time, scientists in the nascent field of immunology were considering not just

the potential for the immune system to be augmented therapeutically but also its role in the natural

surveillance and elimination of tumors. In 1909, noted Germany physician and scientist Paul

Ehrlich was the first to propose that the immune system could potentially eliminate an

“overwhelming frequency” of developing carcinomas8, an early precursor to the theory of cancer

immunosurveillance. However, it would take close to a century of development in the fields of

cancer biology and immunology to develop a framework to fully understand Coley’s early studies

and explore the immune system’s role in the prevention and treatment of cancer.

1.2 Establishment of tumor immunity

Over the first half of the 20th century, many significant developments in tumor immunology

were connected to the study of tissue transplantation. Owing to a lack of established inbred mouse

strains, tumor studies were typically performed by the transplantation of cancer cells between

genetically different organisms or sometimes completely different species. While tissue

transplants between genetically identical hosts were possible, most tissues, whether a malignant

tumor or a benign skin graft, would be normally rejected between genetically dissimilar

individuals. Several studies from James B. Murphy in 1914-15 proposed that the rejection of tumor

grafts was mediated by lymphoid cells and could be enhanced with nonspecific stimulation of

lymphocytes, but his studies gained little traction as the prevailing notion was that lymphocytes

were fixed cells lacking motility9–12. Elegant studies from Ernest E. Tyzzer and Clarence C. Little

in 1916 suggested that the survival of certain tumor transplants in genetically related mouse strains

could be explained by the existence of several genes that, when matched, would allow for tumor

growth13. Several decades later, George Snell set about to isolate the genetic loci responsible for

the acceptance and rejection of grafts, a phenomenon he referred to as ‘histocompatibility’.

Working at the Jackson Laboratory that Clarence Little founded, Snell developed an extensive

panel of congenic mouse strains differing only in their alleles at specific histocompatibility (H)

genes inferred from prior studies14. Using these mice, Snell was able to isolate a that a specific

genetic locus he called H-2 was the strongest barrier to rejection and was extremely polymorphic

between mice15,16. However, a mechanistic understanding of the role of the H-2 locus and the genes

encoded within it was beyond the techniques of the time.

At around the same time, extensive work from Peter Medawar and colleagues defined the

vital role of the cellular components of immunity in the rejection of tissue allografts17–19.

Additionally, they emphasized the faster rate with which a second identical tissue allograft would

be rejected in humans or animal models, highlighting the similar kinetics to other immune-

mediated phenomena. However, these findings posed a problem for the developing field of cancer

immunology. As most studies with common transplanted tumors had been performed between

genetically dissimilar organisms, it was not possible to determine whether the demonstrated

immunity was to antigens common to the tumor itself or simply to any transplanted foreign tissue.

The existence of true tumor-specific immunity was definitively shown in a series of studies from

Ludwik Gross in 1943, Edward Foley in 1953 and Richmond Prehn in 195720–22. All of these

studies displayed that mice could be immunized against syngeneic transplants of tumors induced

by chemical carcinogens, establishing the presence of tumor-specific immunity. Shortly thereafter,

building upon these studies and the earlier ideas of Paul Ehrlich, Sir Macfarlane Burnet and Lewis

Thomas proposed a formal hypothesis of “cancer immunosurveillance” on the role of the immune

system in the natural prevention of malignancy23,24. Burnet and Lewis felt that the immune system

must play a protective role in the host against tumor development going so far as to call it an

“evolutionary necessity that there should be some mechanism for eliminating or inactivating such

potentially dangerous mutant cells”24,25. In summary, Burnet stated “it is by no means

inconceivable that small accumulations of tumor cells may develop and because of new antigenic

potentialities provoke an effective immunological reaction with regression of the tumor and no

clinical hint of its existence.”23 With the combination of these definitive studies from Gross, Foley,

and Prehn together with the conceptual framework of immunosurveillance proposed by Burnet

and Thomas, the role of the immune system in preventing and eliminating cancer appeared to be

coming into focus.

1.3 From immunosurveillance to immunoediting

Following the presentation of the cancer immunosurveillance hypothesis, there was

immense interest in seeking to test its natural conclusion that immunodeficient organisms should

have a higher rate of spontaneous tumor formation. The identification and characterization of the

athymic nude mouse allowed for the examination of the immunosurveillance hypothesis in a

genetically immunodeficient host26,27. In a series of extensive studies, Osias Stutman dealt a major

blow to the acceptance of the immunosurveillance hypothesis. Stutman found that

immunodeficient CBA/H nude mice did not form more tumors than their wild-type counterparts

following carcinogen exposure28. Furthermore, nude mice did not have higher rates of spontaneous

or virally induced tumor formation29,30. Stutman’s results appeared definitive and had failed to

uphold the central tenets of the immunosurveillance hypothesis, leading to widespread skepticism

of its validity. Despite its conceptual elegance, even Lewis Thomas began to doubt his own theory

stating that “the greatest trouble with the idea of immunosurveillance is that it cannot be shown to

exist in experimental animals”31.

However, Stutman’s experiments had multiple caveats he could not have foreseen at the

time. First, later studies would show that nude mice still carried detectable levels of functional ab

T cells and were not fully immunodeficient32,33. Additionally, the CBA/H strain of mice Stutman

used were later shown to have extremely high levels of the aryl hydrocarbon hydroxylase enzyme

necessary for the transformation of the carcinogen used (methylcholanthrene--MCA) into its active

carcinogenic form34,35. Thus, the rate of cellular transformation in this system may have

overwhelmed any endogenous antitumor response. Finally, additional follow-up studies

corroborating these results likely did not follow the mice for sufficient periods of time to assess

the rates of spontaneous tumor formation36.

While Stutman’s work dampened most enthusiasm for the concept of cancer

immunosurveillance, several key studies in the 1990’s renewed interest in the idea. First, interferon

g (IFN-g) was shown to have a protective role against the growth of transplanted tumors and the

development of primary chemically induced or spontaneous tumors37,38. Second, mice lacking

perforin, a component of the cytolytic granules within cytotoxic T cells and NK cells necessary

for target cell lysis, were shown to have higher rates of tumor formation following MCA

challenge39,40. However, the definitive studies proving the existence of a cancer

immunosurveillance process made use of mice lacking recombination activating gene 1 (RAG-1)

or RAG-2. Without these enzymes involved in V(D)J recombination, mice completely lack B and

T cells41. Using 129/SvEv RAG-2-/- mice, Robert Schreiber and his group clearly showed that these

immunodeficient mice developed tumors more rapidly and with a greater frequency than their

wild-type counterparts in response to MCA challenge42. Furthermore, when these mice were aged

and tracked over time, the immunodeficient mice developed far more spontaneous malignancies42.

A set of elegant follow-up transplant experiments provided further insight into the role of the anti-

tumor immune response. When tumors isolated from either RAG-2-/- mice or wild-type mice were

transferred to RAG-2-/- mice, they all grew with similar kinetics suggesting no tumor-cell intrinsic

growth defects in tumors formed in the absence of an intact immune system. However, when these

same tumors were transplanted into wild-type mice, a sizeable fraction of tumors originally derived

from a RAG-2-/- host were unable to form tumors in an immunocompetent host42. This suggested

that tumors formed in the absence of an intact immune system are inherently more immunogenic

than those arising in immunocompetent hosts. Together, these studies definitively proved the role

of the immune system in preventing tumor formation and revived the cancer immunosurveillance

hypothesis of Thomas and Burnet. However, they also demonstrated the ability of the immune

system to sculpt the immunogenicity of tumors that ultimately escape immune recognition.

These dual roles were combined into a “cancer immunoediting” hypothesis presented by

Gavin Dunn and Robert Schreiber containing three stages of tumor-immune interactions:

elimination, equilibrium, escape43–45. The elimination phase represents the initial concept of

immunosurveillance in which highly immunogenic cancer cells are eliminated through an

integrated response of the innate and adaptive immune systems. If the developing tumor is fully

eradicated at this phase, there is no progression to the later stages. However, if the tumor is not

fully eliminated it will progress to the equilibrium phase in which the host immune system exerts

a significant selective pressure that is enough to temporarily contain tumor outgrowth but not

enough to fully eradicate the tumor. Over time, the genetically unstable tumor will accumulate

mutations and tumor subclones will form that are capable of growing under the intense

immunological pressure. It is the growth of these cells that will then result in the formation of a

clinically apparent tumor. This theory of cancer immunoediting provided a unified model to

understand tumor-immune interactions at various stages of carcinogenesis.

1.4 Advancements in fundamental T cell biology

While the cancer immunoediting hypothesis provided a framework to explain the role of

the immune system in both the prevention and formation of tumors, it was a deeper understanding

of the fundamental aspects of T cell biology that ultimately led to the development of novel

immunotherapies in the last few decades. In 1974, Rolf Zinkernagel and Peter Doherty reported

that the interactions between cytotoxic T cells and target cells they recognized required matching

at the H-2 gene complex (also known as major histocompatibility complex) previously defined by

George Snell, a phenomenon now known as major histocompatibility complex (MHC)

restriction46,47. However, the mechanism through which gene products from the MHC regulated

immune responses was not clear. In 1985, a landmark study from Bruce Babbitt, Paul Allen, and

Emil Unanue showed the association of an immunogenic peptide recognized by T cells with MHC

class II molecules48. Together with their previous studies, this work suggested that T cells

recognized short peptide fragments produced by intracellular processing of in-tact proteins that are

then presented by MHC molecules on the surface of antigen-presenting cells49,50. Several years

later, Pamela Bjorkman and Don Craig Wiley elucidated the structure of a human MHC class I

molecule, human leukocyte antigen (HLA)-A2, in complex with a foreign antigen, providing the

most detailed glimpse yet at the cell surface structure recognized by T cells51.

Meanwhile, advancements in T cell biology would enable a more complete understanding

of the other side of this molecular interaction. In the late 1970’s, a series of reports described a

protocol for the selective growth of T cells from normal unfractionated human bone marrow and

peripheral blood52–54. All of the methods involved stimulation with conditioned medium from

phytohemagglutinin (PHA)-stimulated human blood lymphocytes that was theorized to contain a

still unknown molecule referred to as T-cell growth factor. In follow-up work, several groups

reported on the biochemical purification and sequence analysis of this protein that immunologists

had named interleukin-2 (IL-2)55–57. By the mid-1980’s, the gene for human IL-2 had been

identified, recombinant IL-2 had been produced and shown to be biologically active, and IL-2’s

role as an autocrine factor potently stimulating T cell proliferation was becoming clear58–60. The

discovery and characterization of IL-2 would open new doors for studying the fundamentals of T

cell biology.

Throughout the 1970’s and into the early 1980’s, the identification of the T cell receptor

(TCR) mediating antigen-specific reactions in T cells was viewed as the “holy grail” of cellular

immunology. In a series of pivotal studies, Jim Allison and John Kappler characterized the

biochemical features of a cell surface protein that would soon be defined as the TCR61,62. They

described a disulfide-linked dimer with each component containing both constant and variable

regions within both mouse and human T cells. These discoveries set off a race to identify the genes

encoding the TCR. In 1984, a trio of studies from Stephen Hedrick, Mark Davis, and Tak Mak

identified the gene encoding the TCR in both mouse and human T cells63–65. This work defined a

cell-surface protein with constant, variable, and joining regions very similar to the previously

defined structure of immunoglobulins in B cells. In parallel with the studies describing antigen

presentation by MHC molecules, the elusive mechanism through which T cells recognize their

target was coming into focus.

1.5 Molecular basis for anti-tumor immunity

The existence and nature of tumor antigens has been a fundamental question since the

earliest days of cancer immunology, inspiring both enthusiasm and immense uncertainty. In a

widely read 1929 review, William Woglom summarized his skepticism stating that “it would be

as difficult to reject the right ear and leave the left ear intact as it is to immunize against cancer.”66

While the previously described studies of Gross, Foley, and Prehn demonstrated the potential for

mice to be immunized against syngeneic tumors, the underlying mechanisms and the nature of

antigen driving this response were not clear. For many decades, the techniques to probe these

questions were simply not present. However, fundamental discoveries in the B and T cell biology,

including techniques for the production of monoclonal antibodies and the propagation of T cell

clones, allowed for detailed and painstaking studies on the identification of tumor antigens52,54,67.

In a series of detailed molecular studies, Lloyd Old used panels of monoclonal antibodies to

identify potential cell surface antigens in mouse and human cancer lines68,69. However, the antigens

he identified were found to be generally expressed across various cell types or tissue-specific but

not tumor-specific. In his famed 1981 address Cancer Immunology: The Search for Specificity,

Lloyd Old stated that “if there is a basic tenet in the field, it is that cancer cells are antigenically

distinguishable from their normal progenitors,” but he cautioned that70:

“An underlying assumption in much of the work going on in cancer immunology is that antigenic

changes recognizable by the host’s immune system are an invariable accompaniment of malignant

transformation…In my opinion, it is simply too soon to say whether tumor-specific antigens, i.e.,

antigens with an absolute restriction to cancer, exist.”

In the early 1980’s, Thierry Boon’s group was performing a collection of studies on the

mouse mastocytoma line P815. He observed that mutagenesis of P815 resulted in the formation of

tumor variants rejected by syngeneic mice and found to stimulate a variant-specific cytolytic T

cell (CTL) response, suggestive of the generation of new antigens within these tumor variants

following mutagenesis71,72. In a collection of arduous studies utilizing DNA transfection into

antigen-negative clones, Boon’s group isolated the source of T cell reactivity to one of their tumor

variants – a point mutation in a gene they named P91A that resulted in a single amino acid change

in the resulting protein73. Follow-up studies identified additional tumor-specific antigens within

P815 clones, suggesting that T cell recognition of these tumor-specific mutations was a generalized

mechanism of anti-tumor immunity74. Boon’s studies were pivotal in changing our understanding

of cancer immunology, as they demonstrated that tumor mutations producing novel protein

antigens, now termed neoantigens, could function as tumor-specific targets for T cell recognition.

At the same time Boon was performing these studies in mice, Lloyd Old’s group reported

tumor-specific reactivity among autologous T cell clones isolated from a patient with metastatic

melanoma75. Utilizing Boon’s methodologies of anti-tumor CTL clones and antigen-loss tumor

variants, a collection of groups demonstrated the existence of distinct antigens recognized by T

cells on human melanoma lines76–79. Ultimately, this led to the identification of the first tumor

antigens in human patients, members of the multi-protein MAGE family80–82. These genes,

together with related families of BAGE and GAGE genes, were shown to have exclusive expression

to cancer cells and germ cells of the testis, earning them the name Cancer/Testis (CT) antigens83.

Later work from Thierry Boon and Steve Rosenberg would identify additional tumor antigens in

human melanoma that were shown to be lineage-related molecules involved in melanocyte

differentiation such as tyrosinase and MART-184–87. Shortly thereafter in 1995, dual studies

reported on the first human neoantigens identified in melanoma cell lines88,89.

Altogether, these studies proved the existence of mouse and human tumor antigens that

could be broadly grouped into two types. The first, the result of tumor-specific mutations that

produce novel protein epitopes, are termed neoantigens and represent ideal targets owing to their

exquisite tumor specificity. The second represent an enormously diverse class of nonmutated self-

proteins to which T cell tolerance is incomplete owing to restricted patterns of expression such as

CT antigens or tissue differentiation antigens. The specific roles played by each of these classes of

tumor antigens in both spontaneous anti-tumor immunity and the response to immunotherapy

remains a significant question to this day.

1.6 Immune checkpoint blockade

Following his crucial studies involved in the identification of the TCR, Jim Allison

continued to explore the fundamental mechanisms of T cell activation and regulation. While TCR

recognition of peptide:MHC was known to be essential for T cell activation, it was also clear that

a second co-stimulatory signal was required for optimal T cell function. In a pivotal 1992 study,

Fiona Harding in Jim Allison’s lab identified CD28 as the cell-surface molecule providing this co-

stimulatory signal90. Through his interest in T cell co-stimulation, Allison and his lab began

studying a structurally similar protein previously identified as being expressed on activated T cells

known as cytotoxic T-lymphocyte-associated protein 4 (CTLA-4)91. In 1995, Matthew Krummel

in Jim Allison’s lab established the opposing role of CTLA-4 and CD28 in T cell function92. While

CD28 recognition of its ligands B7-1 and B7-2 on antigen presenting cells leads to T cell co-

stimulation, CTLA-4 binds with higher affinity to both B7-1 and B7-2 and produces a potent

inhibitory signal that decreases T cell proliferation and cytokine production92. That same year, Tak

Mak’s group characterized the CTLA4-/- mouse which had a lethal phenotype characterized by

uncontrolled lymphocyte proliferation, highlighting the crucial role of CTLA-4 in T cell

homeostasis93. The following year, Dana Leach and Matthew Krummel in Allison’s group

published a landmark study that would come to change the field of tumor immunology. In it,

treatment of tumor-bearing mice with a blocking antibody against CTLA-4 resulted in the rejection

of pre-established tumors and generated immunological memory against tumor rechallenge94. This

provided the preclinical rationale for the targeting of T cell inhibitory receptors such as CTLA-4

in the treatment of cancer, an approach now known as immune checkpoint blockade (ICB).

Meanwhile in the early 1990’s, Tasuku Honjo’s group was characterizing gene expression

changes in T cells undergoing apoptosis. They isolated a novel member of the immunoglobulin

gene superfamily with normal expression restricted to the thymus that they called programmed

cell death protein 1 (PD-1)95. In contrast to the lethal phenotype observed in the CTLA4-/- mouse,

the PD-1 knockout mouse generated by Honjo’s group displayed a more moderate effect with aged

mice developing symptoms consistent with lupus and PD-1 knockout T cells showing enhanced

functionality96. Collectively, his studies suggested that PD-1 played a crucial role in the

maintenance of peripheral self-tolerance by serving as a negative regulator of T cell function. In

2000, a joint collaboration between Gordon Freeman, Clive Wood and Honjo’s group identified a

member of the B7 gene family as the ligand for PD-1 which they referred to as PD-L197. This work

further characterized the role of PD-1 signaling, describing the expression of PD-L1 and B7-1/B7-

2 on antigen presenting cells and the capacity for PD-1 signaling to inhibit CD28 costimulation in

activated T cells. Recognizing the potential application to tumor immunology, Honjo’s group

published studies in 2002 and 2005 on the role of PD-L1 expression on tumor cells in suppressing

anti-tumor T cell responses and the capacity to reverse this effect with antibody blockade of either

PD-1 or PD-L198,99. In 2006, an elegant study from Rafi Ahmed’s group observed PD-1

upregulation in dysfunctional virus-specific CD8 T cells in the chronic infection model with

lymphocytic choriomeningitis virus (LCMV)100. Crucially, this study described how blockade of

the PD-1/PD-L1 inhibitory pathway could lead to restoration of function in these exhausted CD8

T cells, leading to increased proliferation, cytokine production, and target cell lysis. While the

study was entirely performed using a model of chronic viral infection, it provided a framework for

conceptualizing the role of PD-1 in regulating immune responses more broadly and, together with

Honjo’s previous work, provided a clear rationale for targeting PD-1 in cancer.

These pivotal studies from Allison, Honjo, and numerous others led to the initiation of

clinical trials evaluating the efficacy of blocking antibodies against CTLA-4 (aCTLA-4) or the

PD-1/PD-L1 axis (aPD-1 or aPD-L1) in the treatment of solid tumors. In a landmark clinical trial

published in 2010, aCTLA-4 therapy was shown to provide a survival benefit in patients with

metastatic melanoma101. The next year, the FDA approved the monoclonal antibody ipilimumab

for the treatment of metastatic melanoma, the first drug of any kind ever shown to improve survival

in patients with metastatic melanoma. 2010 also saw the first promising reports on the safety and

tolerability of aPD-1 therapy in patients with refractory solid tumors102. Two years later in 2012,

a major study showed durable responses with aPD-1 therapy in a fraction of patients with

advanced non-small cell lung cancer (NSCLC), melanoma, or renal cell cancer that had failed prior

therapies103. The next year, Science magazine referred to cancer immunotherapy as the

“breakthrough of the year,” and the first aPD-1 therapy was FDA approved in 2014. Since then,

the approved uses for ICB therapy have exploded, with combination therapy consisting of both

aCTLA-4 and aPD-1 indicated for metastatic melanoma, NSCLC, renal cell carcinoma, Hodgkin

lymphoma, and squamous cell carcinoma of the head and neck among numerous others.

The enormous success of ICB therapy in treating a multitude of malignancies has generated

immense interest in understanding the mechanisms of action underlying the effect and

characterizing the correlates of response. Early clinical studies on aPD-1 therapy demonstrated

that the most important factor predicting a successful response was the existence of pre-treatment

CD8 T cell infiltrate together with intratumoral PD-L1 expression, suggesting the presence of an

anti-tumor immune response suppressed by the PD-1/PD-L1 axis104,105. Additionally, numerous

studies revealed that tumor mutational burden correlated with response to either aPD-1 or

aCTLA-4 therapy and was found to correlate with local immune cytolytic activity106–109. The

importance of tumor mutational burden is potentially tied to the relationship between tumor

mutations and neoantigens. Pre-clinical studies proved the importance of neoantigen-specific

immune responses in mediating tumor regression following ICB therapy in mice but a rigorous

approach to this question is more challenging in patients110.

1.7 Cellular immunotherapy

In 1983, Steve Rosenberg’s lab described the generation of “lymphokine activated killers”

(LAK) cells following the activation of peripheral blood from cancer patients with IL-2111. These

LAK cells showed the capacity to lyse NK-cell resistant tumor targets, and a protocol for the

treatment of patients with LAK cells and recombinant IL-2 was quickly initiated112. While close

to half of the original patients experienced objective responses, side effects such as severe fluid

retention were also common. Rosenberg’s group did note particularly strong responses in patients

with metastatic melanoma, a phenomenon they also observed in patients treated with high-dose

recombinant IL-2 alone113. Motivated by numerous studies suggesting anti-tumor immune

responses in melanoma, Rosenberg developed a protocol for the use of autologous expanded tumor

infiltrating lymphocytes (TIL) in conjunction with IL-2114. This combination achieved higher

response rates than previously achieved with LAK cells or IL-2 alone in patients with metastatic

melanoma. While the mechanisms underlying the treatment approaches were not entirely clear,

and LAK cells were likely just activated T cells, these studies paved the way for adoptive cell

therapy (ACT) as a viable treatment approach.

In recent decades, Rosenberg and his group have continuously updated their approaches to

ACT. Following the identification of tumor associated antigens in melanoma, some patients were

treated with activated antigen-specific CD8 T cell clones, achieving responses in a majority of

cases115,116. In 2006, he reported on the first patients with metastatic melanoma treated with

autologous genetically engineered lymphocytes given TCRs specific for tumor-associated antigens

such as MART-1, gp-100, and NY-ESO-1117. Several years later, his group showed that this

approach could induce cancer regression outside of metastatic melanoma, treating patients with

synovial sarcoma with engineered T cells expressing a TCR against the CT antigen NY-ESO-1118.

However, it was more dramatic T cell engineering pursued by Rosenberg and other groups that

would bring forth the next breakthrough in ACT – the use of chimeric antigen receptor (CAR) T

cells.

In 1993, Zelig Eshhar published a report on the generation of cytotoxic lymphocytes that

he had engineered with a chimeric receptor to give these T cells antibody-like specificity119.

Introduction of the chimeric gene consisting of a single-chain Fv domain (scFv) of an antibody

linked to T cell intracellular signaling domains resulted in T cell activation and cytotoxicity upon

recognition of target antigen in an MHC-independent manner. Later, crucial work from Michel

Sadelain’s group would show the essential role of CD28 co-stimulatory domains in the optimal

function of these CAR T cells and display the potential of CD19-directed CARs to eradicate human

B cell tumors in immunodeficient mice120–122. By 2010-11, the first impressive case reports were

published by Carl June and Steve Rosenberg, describing cancer regression in patients with

follicular lymphoma or chronic lymphoid leukemia treated with autologous CAR T cells directed

against CD19123–125. These promising studies were promptly followed by larger trials establishing

the efficacy of this therapy in treating a multitude of B cell malignancies126,127. By 2017, the FDA

approved CAR T therapy for patients with B cell acute lymphoblastic leukemia (B-ALL), a

landmark moment in the field of adoptive cell therapy and cancer immunology. Since then, the

indications for CAR T cell therapy have broadened to include other B cell malignancies such as

diffuse large B-cell lymphoma or follicular lymphoma.

1.8 Cancer immunogenomics

The completion of the human genome project in 2003 laid the foundation for a deeper

understanding of the role of genetic variation in human health and disease128. However, it was the

development of next generation sequencing (NGS) technology enabling far more cost-effective

sequencing that would bring genomics into the mainstream of the clinical and research world. By

2008, the first complete cancer genome had been sequenced, ushering in a new era in the

characterization of cancer129. Since that time, the extensive profiling of human cancer genomes

has produced an in-depth analysis of the genomic alterations at the heart of all cancers and shaped

our understanding of cancer biology. In some cancer types, the identification of specific mutations

has resulted in the development of targeted therapies, providing patients with personalized

treatment approaches130. However, the widespread genomic analysis of cancer has also paved the

way for new approaches and techniques at the intersection with basic immunology in the field of

cancer immunogenomics.

Broadly, cancer immunogenomics represents a complementary approach utilizing the

precision of tumor genomic analysis together with the fundamentals of cancer immunology. The

identification and targeting of candidate tumor-specific neoantigens produced as a result of tumor

genomic alterations lies at the core of cancer immunogenomics. Neoantigens are considered ideal

therapeutic targets owing to the lack of pre-existing T cell tolerance against them and their tumor

specificity. Their restricted expression should allow for tumor-specific targeting without the

accompanying side effects observed with more general immunostimulatory agents such as ICB

therapy or adoptive cell therapy against self-proteins. Furthermore, studies in preclinical systems

have shown that neoantigen-specific responses form the basis for both cancer immunoediting and

the response to ICB therapy110,131. The role of endogenous neoantigen-specific responses in

patients and their involvement in mediating immunotherapy responses is much less clear.

A central component of cancer immunogenomics is the identification of tumor antigens.

The earlier studies led by Boon, Rosenberg, and Old describing some of the first tumors antigens

were the culmination of arduous work utilizing classical techniques in molecular and cellular

biology. However, the painstaking nature of these approaches meant they were unlikely to ever

scale for clinical applicability. The remarkable advancements in our capacity to identify tumor

mutations through the widespread utilization of NGS technologies has radically altered approaches

to tumor antigen detection. As early as 2008, Jim Allison’s group reported on the use of in silico

prediction methods for the identification of putative tumor neoantigens from whole-exome

sequencing data132. Numerous computational methodologies have been developed for the

prediction of candidate neoantigens with initial approaches such as NetMHC relying on neural

networks trained on defined sets of peptide:MHC pairs133,134. While all of these approaches

generate putative neoepitopes, none has clearly separated itself as the best option for the

identification of recognized neoantigens, likely owing to the complex multifactorial nature of this

process. However, these approaches will likely improve with time as more patient-derived data

continues to update the underlying models. For now, they provide an initial step towards

personalized cancer immunotherapy by providing patient-specific epitopes and allow for the in-

depth profiling of treated patients.

Numerous studies utilizing these approaches in patients treated with ICB or ACT suggest

a potentially vital role for tumor neoantigens in mediating treatment effect. In addition to the well-

established correlation between mutational load and response, treatment of patients with ICB

therapy has been demonstrated to enhance neoantigen-specific responses135. Furthermore,

comprehensive genomic profiling of tumors treated with ICB therapy suggests a dynamic

evolutionary interplay with the outgrowth of tumor clones that have lost mutation-associated

neoantigens136. Nonetheless, the most striking examples of the importance of neoantigens in

immunotherapy response come from Steve Rosenberg’s work with patients treated with expanded

TIL cultures. While these ACT approaches began long before the technologies existed to easily

probe for antigen-specific responses, numerous reports from his group have identified significant

neoantigen reactivity among TIL cultures given to patients with favorable responses to

treatment137,138. Of note, ACT with TIL cultures containing significant neoantigen-reactivity have

also been shown to mediate tumor regression in cancers beyond metastatic melanoma. One patient

with metastatic cholangiocarcinoma experienced tumor stabilization and ultimate regression

following transfer of a TIL culture with significant CD4 T cell recognition of a tumor neoantigen,

while another patient with metastatic colorectal cancer saw an impressive response from treatment

with a TIL culture containing a polyclonal CD8 T cell response against KRASG12D139,140.

The progress in the genomic profiling of cancer and the development of in silico algorithms

for the prediction of patient-specific neoantigens has also led to renewed enthusiasm for

vaccination against cancer. In one of the earliest genomically guided approaches, Gerald Linette

and his group demonstrated that a dendritic cell vaccine led to an enhancement and broadening of

neoantigen-specific responses in patients with metastatic melanoma141. More recently, several

groups have reported the induction of neoantigen-specific responses in patients with metastatic

melanoma through personalized peptide or nucleic-acid formulations142,143. These same techniques

were also applied and shown to produce responses in the treatment of glioblastoma, a tumor with

a much lower mutational burden and neoantigen target pool144,145. While the optimal vaccination

route for the induction of neoantigen-specific T cells is still an open question, personalized cancer

vaccines represent the culmination of decades of work in cancer immunogenomics. The

enhancement of neoantigen-specific T cell responses, whether through personalized vaccines or

alternative approaches, will likely continue to be a dominant thread in cancer immunotherapy

going forward.

1.9 Cancer immunology today

As the field of immunology has matured over the last several decades, the process through

which anti-tumor immune responses are generated and maintained, specifically the necessity of

cross-talk between the innate and adaptive arms of immunity, has come into clearer focus. Ralph

Steinman’s initial discovery of a novel cell type within lymphoid organs that he termed the

“dendritic cell” did not generate significant enthusiasm at the time146,147. However, the following

decades would elucidate the role of dendritic cells at the intersection of innate and adaptive

immunity. Work from Hy Levitsky and Drew Pardoll would demonstrate the pivotal role of bone-

marrow derived cells in initiating immune responses to MHC class I-restricted tumor antigens148.

Several years later, Nina Bhardwaj’s group showed that dendritic cells function as potent antigen

presenting cells (APCs) capable of acquiring antigens from apoptotic cells and inducing the

generation of CTLs149. A collection of detailed studies from Ira Mellman, Steinman, and Antonio

Lanzavecchia outlined the process through which dendritic cells are activated by microbial

products and pro-inflammatory mediators to become potent APCs and activate CTLs150,151.

Together, these and other studies defined the process through which dendritic cells are

activated in peripheral tissues and migrate to draining lymph nodes where they initiate an adaptive

immune response152. In the context of tumor immunology, this process has been broadly referred

to as the cancer immunity cycle153,154. In summary, this process is initiated by the release of tumor

antigens following cancer cell death and uptake by dendritic cells in the local microenvironment.

Upon activation by pro-inflammatory mediators, these dendritic cells will migrate to local draining

lymph nodes where they prime and activate naïve T cells. Upon proliferation and differentiation

into activated T cells, these effectors must migrate and infiltrate into the tumor microenvironment

where they recognize and kill target cells. The antigen released following tumor cell lysis then re-

initiates this cycle. This cycle provides a framework for conceptualizing the initiation and

maintenance of anti-tumor immune responses and the impact of various interventions such as ICB

therapy or neoantigen vaccines. However, fundamental questions remain at virtually every step of

this cycle. A deeper understanding of the processes regulating myeloid cell activation and

migration as well as T cell recruitment and anti-tumor effector function will continue to inform

future cancer immunotherapy treatment approaches.

Despite immense progress in our understanding of anti-tumor immunity and the

development of the first FDA approved immunotherapies over the last decade, there remain

significant open questions and room for progress. Even among immunotherapy responsive tumors

such as melanoma and NSCLC, a sizeable fraction of patients do not respond. What factors

determine a patient’s response and how can non-responders be prospectively identified and

converted into responders? Furthermore, a number of difficult to treat malignancies such as

pancreatic ductal adenocarcinoma and glioblastoma do not respond to conventional

immunotherapies and have so far proven resistant to more novel approaches. Malignant brain

tumors such as glioblastoma represent a significant challenge in that patients face a particularly

grim prognosis and a paucity of treatment options. This work is specifically focused upon

deepening our understanding of the anti-tumor immune response in malignant brain tumors and

leveraging this for the development of novel immunotherapies.

1.10 Malignant brain tumors with emphasis on glioblastoma

Malignant brain tumors consist of both primary tumors arising from within the central

nervous system (CNS) and secondary metastases originating from extracranial sites. The most

common malignant primary tumor of the CNS is glioblastoma (GBM) with approximately 12,000

new cases per year, whereas secondary tumors typically develop from carcinomas of the lung,

breast, kidney or melanoma155,156. Significant advances in our understanding of the molecular and

genetic features of glioblastoma157,158 have not led to effective new therapies with conventional

standard of care treatment for primary GBM still consisting of surgery followed by concurrent

chemoradiation159. Despite this multimodality treatment, patients have a poor prognosis with a

median progression-free survival of 6.9 months and median overall survival of just 14.6 months159.

The inevitable recurrence is likely due to the highly aggressive and infiltrative nature of GBM,

best exemplified by the Walter Dandy’s 1928 report on the failure of hemispherectomies to prevent

recurrence160.

In contrast, the use of ICB therapy has led to improved outcomes and robust intratumoral

T cell infiltration in a considerable subset of patients with brain metastases (BrMETs)161,162.

However, these treatments and other immunotherapeutic approaches have not been effective in

GBM. Indeed, aPD-1 monotherapy did not improve survival in patients with newly diagnosed or

recurrent disease, and targeted vaccines and cell therapy approaches also have shown limited

efficacy144,145,163–165. Thus far, immunotherapy responses within GBM have been restricted to a

selective group of patients with a hypermutated phenotype caused by germline deficiencies in

DNA replication or repair166,167.

1.11 Molecular and cellular heterogeneity in glioblastoma

The lack of effective treatment options for patients with GBM is not due to an incomplete

understanding of the genomic landscape of GBM. In fact, it was the first study published as part

of The Cancer Genome Atlas and has been the subject of extensive characterization to date157,158.

These studies have defined crucial genetic features of GBM such as frequent copy number

amplifications of EGFR, deletions of CDKN2A/B, and mutations in tumor suppressors such as

TP53, PTEN, NF1, and RB1. However, in-depth genomic profiling has also unveiled remarkable

cellular and molecular heterogeneity within GBM168–171. Gene expression profiling of GBM has

led to the generation of a collection of transcriptionally defined tumor subtypes172. However, an

analysis of spatially distinct regions within GBM demonstrated that approximately half of all

tumors contained regions classified as multiple distinct subtypes within the same mass173. In

essence, each individual GBM is likely composed of numerous unique diseases within the tumor

and the identification of critical targetable dependencies is quite difficult.

In addition to the impact on the development of targeted drug therapies, this heterogeneity

likely poses additional challenges for potential immunotherapy. Studies in NSCLC and melanoma

have defined a strong association between the response to ICB therapy and the frequency of clonal

mutations that likely produce clonal neoantigen targets174,175. In contrast, even oncogenes such as

the epidermal growth factor receptor (EGFR) vIII frequently observed within GBM have been

shown to be expressed by only a fraction of the cancer cells within EGFRvIII-positive tumors176.

This suggests that even targeted immunotherapy against ideal candidates would likely fail due to

immune escape from a resistant subclone.

1.12 The immune system and glioblastoma

A collection of molecular, cellular, and anatomical factors makes GBM a uniquely difficult

tumor to target via immunotherapy. The early work of Peter Medawar, combined with the lack of

conventional lymphoid structures within the CNS, led to the concept of “CNS

immunoprivilege”177. The core principle of this notion was the belief that the CNS was a uniquely

immunosuppressive compartment and that immune reactions to intracranial antigens were

substantially diminished. However, this dogma was frequently challenged over the ensuing

decades by the presence of CNS autoimmunity in the context of demyelinating diseases as well as

clear immunosurveillance against viral, bacterial, and fungal pathogens. More recently, the

definitive identification of lymphatic vessels within the meninges and a collection of studies

defining the cellular components necessary for the initiation of intracranial immune responses has

diminished most enthusiasm for the idea of CNS immunoprivilege178–181. However, there clearly

exist nuances to CNS immunity that are possibly detrimental to generating immune responses but

also potential therapeutic targets. Recent work from Akiko Iwasaki’s lab describing increased anti-

tumor immune responses by the enhancement of lymphatic drainage through vascular endothelial

growth factor C (VEGF-C) treatment represents one intriguing avenue for research182.

GBM also appears capable of generating a significantly immunosuppressive

microenvironment through a combination of systemic and local factors. First, the presence of

systemic T cell dysfunction and lymphopenia has been a noted but unexplained feature for several

decades183,184. While this may be partially driven by treatment with depleting chemotherapy such

as Temozolomide, recent work also suggests the potential for T cell sequestration within the bone

marrow in response to intracranial malignancies185. Furthermore, histological analyses of GBM

reveal high levels of PD-L1 together with relatively sparse T cell infiltration186. The low number

of T cells that do infiltrate into GBM face an extremely immunosuppressive microenvironment

that contributes to significant T cell exhaustion and dysfunction187,188. The features generating this

immunosuppressive microenvironment are likely multifactorial but include high levels of IDO1

expression, production of immunosuppressive cytokines such as TGF-b and IL-10, and infiltration

with myeloid derived suppressor cells (MDSCs)189–192.

1.13 Immunotherapy in glioblastoma

Owing to the scarcity of effective treatment options and the success of immunotherapy in

other difficult to treat tumor types, there has been immense interest in the development of

immunotherapy for GBM. However, most of the conventional approaches have generated mostly

mixed or negative results. Large-scale studies of ICB therapy have demonstrated no survival

benefit in the recurrent setting for adjuvant (post-surgery) ICB but do suggest potential for

neoadjuvant (pre-surgery) ICB therapy to increase survival and CD8 T cell infiltration163,193.

Furthermore, up to twenty percent of GBMs will recur with a treatment-induced hypermutated

phenotype, leading to immense interest in the treatment of these tumors with ICB194. Despite some

case reports hinting at the potential benefit of ICB therapy in these patients, the magnitude of ICB

response is still unclear166,167,194.

Additional trials have focused upon the targeting of tumor-specific or tumor-associated

antigens within GBM. EGFRvIII is a splice variant of EGFR that produces a putative neoepitope

frequently expressed by GBM158,195. Previous studies have demonstrated the ability of this epitope

to induce antibody, CD4, and CD8 T cell responses, making it an ideal antigen target for

immunotherapy in GBM196–198. However, despite encouraging phase II results for a vaccine

targeting this neoantigen, no benefit was observed in the phase III trial164,199. Similarly, a trial of

CAR T cell therapy targeting the EGFRvIII variant displayed clear intratumoral T cell infiltration

and activation but also evidence of immunosuppression and antigen loss preceding immune

escape165. Additional work has targeted the growth factor receptor IL13Ra2 found to be

upregulated on a number of cancer types including GBM200. However, CAR T cell therapy

targeting IL13Ra2 showed similar results as the targeting of EGFRvIII with T cell infiltration in

conjunction with antigen downregulation201.

1.14 Conclusions

Decades of investigation on the fundamentals of cancer biology and immunology have led

to enormous successes with immunotherapy over the last decade, leading many to consider it a

pillar of modern oncology. However, these achievements have also clashed with the reality that

numerous questions still remain on the core aspects of the anti-tumor immune response and most

patients with solid tumors still do not benefit from immunotherapy. One such tumor with limited

response to immunotherapy and a relative scarcity of effective treatment options is GBM. Despite

extensive characterization of its genomic profile and immense interest in leveraging

immunotherapy for its treatment, there has been no change to the standard of care for GBM in over

fifteen years. Cancer immunogenomics, the utilization of complementary approaches in genomics

and cancer immunology, represents a framework for the identification and targeting of tumor-

specific neoantigens within tumors that could open new therapeutic avenues for the treatment of

tumors such as GBM. In this work, we provide a comprehensive immunogenomic profile of a

cohort of malignant brain tumors, develop new systems for the identification and characterization

of tumor-specific responses in mouse and human systems, and establish a preclinical system for

the targeting of brain tumor-specific neoantigens through adoptive cellular therapy.

CHAPTER TWO

Immunogenomic landscape of malignant brain tumors

2.1 Introduction

One cardinal feature of GBM that may be a particularly important contributor to therapy

resistance is its extensive intratumoral molecular and cellular heterogeneity168. Indeed,

intratumoral genetic heterogeneity is found in many cancer types to varying degrees202–205. The

presence of a complex tumor subclonal genomic architecture likely plays a pivotal role in limiting

the efficacy of both targeted therapies as well as immunotherapies. Specifically, studies in non-

small cell lung cancer (NSCLC) and melanoma showed a strong association between checkpoint

blockade immunotherapy response and the frequency of clonal nonsynonymous mutations, which

likely serve as sources of spatially distributed neoantigen targets174,175. In GBM, extensive work

has demonstrated that intratumoral heterogeneity of a range of tumor somatic changes including

mutations, copy number alterations, and transcriptional signatures across spatially distinct tumor

regions is also a hallmark of this disease173,206,207. In contrast, we have a more limited

understanding of the extent of intratumoral heterogeneity in other intracranial malignancies such

as brain metastases (BrMETs), as few corresponding analyses have been performed in these

cancers208. Moreover, further work is needed to understand the relationship between tumor genetic

heterogeneity and other important features of the tumor ecosystem, including the immune

microenvironment. Although numerous recent studies in other solid tumors including NSCLC and

ovarian cancer have examined the relationship between tumor cell intrinsic properties and

immunologic parameters such as the T cell receptor (TCR) repertoire209–212, this analysis has not

been extended either to primary or metastatic malignant brain tumors where the extent of

heterogeneity in the immune microenvironment and its interplay with genomic and transcriptional

diversity is unknown.

To address these questions, in collaboration with Malachi Griffith and Megan Richters, we

performed systematic and comprehensive multi-sector immunogenomic analyses on 93 samples

from a cohort of 30 patients with primary or recurrent gliomas or metastatic brain tumors,

representing the largest cohort of these brain cancers studied spatially to date. For each patient, we

characterized multiple spatially distinct regions using whole exome sequencing (WES), custom

capture validation, RNA, and TCR-sequencing. Our findings underscore the significant

differences in clonal architecture between gliomas and metastatic brain tumors which translates

into distinct neoantigen landscapes and, in turn, tumor infiltrating T cell clonotypic diversity.

These data therefore provide high resolution insights into the immunogenomic landscapes within

malignant brain tumors which may inform tumor-specific therapeutic approaches.

2.2 Results

2.2.1 Genomic features of glioma and BrMET cohorts

We obtained surgically resected tumor tissue and matched peripheral blood from a group

of 30 patients with pathologically confirmed intracranial tumors (Fig. 2-1a-b). Within this cohort,

14 tumors were primary GBM, 4 were recurrent GBM, 1 was an anaplastic oligodendroglioma

and 11 were BrMETs from either lung, breast, or cutaneous malignancies. In total, 21 of these

patients (15 primary gliomas and 6 BrMETs) were newly diagnosed and had received no prior

therapy. The 4 patients with recurrent GBM had undergone standard of care chemoradiation

therapy, while 5 of the BrMET patients had received varying treatment regimens for their primary

tumor (Fig. 2-1b). Importantly, no patients had prior immunotherapy

Figure 2-1: Clinical details of study cohort. Clinical and study-specific details for glioma (a) or BrMET (b)

patients included in the work. The number of spatially distinct regions from each tumor that underwent a

given analysis is indicated in the last four columns. NSCLC - Non-small cell lung cancer.

SCLC - Small cell lung cancer. MGMT - O-6-methylguanine-DNA methyltransferase. WES - Whole-exome

sequencing. TMZ - Temozolomide. RT - radiotherapy. PCV - procarbazine, lomustine (CCNU), vincristine.

Patient ID

Sex

Age

Pathology

Prior Trea tment

WES

Regions

Custom-

Capture

Regions

RNA-Seq

Regions

TCR Seq

Regions

BrMET00 8

NSCLC

Carboplatin,

Pemetrexed

BrMET00 9

NSCLC

BrMET01 0

NSCLC

BrMET01 8

Brea st

Taxol

BrMET01 9

SCLC

BrMET02 3

Brea st

Carboplatin,

Taxotere, Abraxane,

Herceptin/Perjeta

BrMET02 4

Brea st

TDM1, Docetaxel,

Capecitabine,

Tamoxifen,

Exemestane,

Herceptin

BrMET02 5

NSCLC

BrMET02 7

Brea st

Carboplatin,

Taxotere

BrMET02 8

Melanoma

BrMET05 8

NSCLC

treatment, however all patients across the cohort were given pre-operative steroids. Immediately

following surgical resection via craniotomy, each sample was dissociated into multiple (2-4)

spatially distinct tumor regions that each underwent comprehensive genomic and immunological

profiling including DNA WES, RNA sequencing, neoantigen prediction, and TCR sequencing

(Fig. 2-2a). Using WES at an average coverage depth of 156x, we identified a total of 11,923

somatic variants (both single-nucleotide variants and indels) across the tumor cohort. Because of

the subclonal nature of many variant calls derived from tumors with significant intratumoral

heterogeneity, we developed a customized, targeted validation sequencing assay with a set of

probes targeting all initially identified variants in addition to a select group of noncoding sites

(e.g., TERT promoter mutation sites) to re-sequence all initially detected variants to confirm their

presence. Using this approach, 87.6% of the cohort was characterized at a depth of at least 250x

at >90% of positions captured by the custom reagent. This resulted in confirmation of 92%

(10,254/11,181) of the original variants through validation sequencing after removal of those that

could not be targeted (488) or lacked sufficient minimum coverage (244). These data allowed us

to obtain high precision estimates of the VAF of each variant and provided greater confidence that

variants were not missed owing to potential regional variability in neoplastic content or sequencing

depth.

As expected, the median number of aggregate somatic variants per tumor was higher in

BrMETs (504) than in either primary (93) or recurrent glioma (141) (Fig. 2-2b). Within the glioma

cohort, GBM065.Re represented an outlier with 5,750 somatic mutations identified across 4

distinct tumor regions. Most of the variants within this sample displayed the characteristic

mutational signature associated with prior temozolomide treatment (Fig. 2-3a-b), suggesting a

treatment-induced hypermutated phenotype213,214. Within the BrMET specimens,

10 0

10 00

Mutation Burden

ZNFX1

ADGB

BNC 2

CD H9

CSMD 2

EGFR

FCGBP

FOXC2

HC N4

LR P2

NF1

NL RP5

PCM1

VWA2

ZNF831

HC N1

IDH1

MUC 16

PTEN

SLIT3

FRAS1

TP53

0102030

% Mut an t

GBM030

GBM032

GBM051

GBM052

GBM055

GBM056

GBM059

GBM062

GBM063

GBM064

GBM069

GBM070

GBM074

GBM079

AO0 83

GBM018.Re

GBM031.Re

GBM047.Re

GBM065.Re

Mutation Type

Misse nse

Stop Gained

Frame shift

Inframe D ele ti on

Spl ice Re gio n

Ca ncer type

TERT promoter

Sample n =19

Clinical Data

Yes

Prima ry Glioma

Re c. Glioma

10 0

10 00

Mutation Burden

SCN 2A

SLITRK1

SPEN

SPTA1

SYNE1

SYNE2

TENM3

THSD7A

TME M132 C

TSHZ3

VWA5B1

ZFR2

ZNF800

CH D7

COL5 A1

DMD

FAM47C

FLG

LR P1B

ZFHX4

MUC 16

RYR2

TTN

TP53

0204060

% Mut an t

BrMET008

BrMET009

BrMET010

BrMET018

BrMET019

BrMET023

BrMET024

BrMET025

BrMET027

BrMET028

BrMET058

Sample n =11

Brea st Ca ncer

Mel ano ma

NSC LC

SCL C

−1

G−score

Glioma

G−score

BrMET NSCLC

−2

−1

G−score

BrMET BRCA

Variant Count

Breast Cancer

Melanoma

NSCLC

SCL C

Primary Glioma

Rec. Glioma

Ca ncer type

TERT promoter

CDC42BPB

OR5D13

MGAT5B

15 Pri mary Glioma s

4 Recurrent Gliomas

11 Bra in Metastase s

Mul ti -sector tumor sequ encing

92 t umo r re gio ns

Whole exome seque ncing

an d neo antig en pre dictio n

RN A se quencin g and

immu ne pro filin g

TCR sequencing

EGFR

CD KN2A

CD K4

KRAS

ERBB2

10 0

50 0

10 00

20 00

40 00

60 00

Glioma

BrMET

Figure 2-2: Genomic landscape of brain tumor cohort. a, Overview of sample collection, sequencing, and

data analysis. b, Variant counts per tumor pooled across samples. c, Summary of top 25 recurrently mutated

genes (3 or more tumors) in 11 brain metastases. d, Summary of recurrently mutated genes (3 or more tumors)

in 19 primary and recurrent gliomas. e, Cohort-level copy number variation in gliomas determined by the GISTIC

algorithm. Dashed lines indicate significantly recurrent amplifications (red) and deletions (blue) at an FDR < 0.1.

f, NSCLC brain metastasis cohort GISTIC output. g, Breast cancer brain metastasis cohort GISTIC output.

BrMET023

GBM079

GBM032

GBM064

BrMET027

BrMET028

GBM055

GBM065.Pri

GBM052

GBM030

GBM031.Re

GBM074

AO083

GBM069

GBM018.Re

GBM070

GBM065.Re

GBM047.Re

GBM059

GBM056

GBM062

GBM063

BrMET018

BrMET024

BrMET010

BrMET008

BrMET025

BrMET009

BrMET019

BrMET058

Sig nature.1

Un known

Sig nature.16

Sig nature.8

Sig nature.4

Sig nature.3

Sig nature.5

Sig nature.13

Sig nature.9

Sig nature.2

Sig nature.11

Sig nature.7

Sig nature.23

Sig nature.29

Sig nature.28

Sig nature.25

Sig nature.21

Sig nature.19

Sig nature.24

Sig nature.26

Sig nature.12

Sig nature.18

Sig nature.15

Sig nature.27

Sig nature.10

Sig nature.30

Sig nature.6

Sig nature.14

Sig nature.17

Sig nature.20

Sig nature.22

0.00 0

0.00 5

0.01 0

0.01 5

0.02 5

Proportion

A>G

C>T

A>C

C>G

C>A

A>T

10 0

12341

BrMET009 2 31

BrMET010 2 31

BrMET018 2 31

BrMET019 2 31

BrMET023 2 31

BrMET024 2 31

BrMET025 2 31

BrMET027 2 31

BrMET028 2 324121234

GBM052 1 3

GBM055 2 1

GBM056 2 31

GBM059 2 31

GBM062 2 31

GBM063 2 1

GBM069 2 31

GBM070 2 31

GBM074 2 31

GBM079 2 31

AO83 2 1312341

GBM047.Re 2 31234

% Tot al M utation s

BrMET008

BrMET058

GBM030

GBM018.Re

GBM065.Re

GBM031.Re

GBM032

Figure 2-3: Mutation al spectrum and sign atures across cohort. a, Spectrum of single-nucleotide

variants across cohort. b, Distribution of mutational signatures across cohort generated by deconstructSigs.

Signature descriptions at http://cancer.sanger.ac.uk/cosmic/signatures.

recurrent mutations were identified in TP53 across all histologies (7/11 overall; 3/5 NSCLC, 2/4

breast carcinoma) (Fig. 2-2c). KRAS alterations were identified within 2/5 NSCLC tumors and

PIK3CA mutations were present in 2/4 breast carcinomas consistent with their high incidence in

studies of primary samples215,216. Additionally, TERT promoter mutations were observed in both

a melanoma and NSCLC BrMET tumor. Across the GBM samples, we identified recurrent

mutations in canonical GBM-associated genes such as the TERT promoter (14/18), TP53 (7/18),

PTEN (4/18), NF1 (3/18), and EGFR (3/18) (Fig. 2-2d). This high frequency of TERT promoter

mutations mirrors earlier work estimating frequencies upwards of 70-80% among GBM217,218.

Additionally, the frequency of TP53 alterations is consistent with prior studies observing

mutations in 30-40% of GBM samples, while the frequency of EGFR mutations is slightly

lower207,218. We next assessed the prevalence of copy number alterations (CNAs) within our cohort

and identified classical GBM-associated changes such as frequent chromosome 7 amplification

encompassing EGFR (17/18), chromosome 9 deletions of the CDKN2A locus (15/18), and

chromosome 10 deletions of the PTEN locus (17/18) (Fig. 2-2e). These amplifications and

deletions have been detected at comparably high frequencies across several other studies207,218.

The metastatic tumors displayed alterations similar to those previously defined for their

corresponding primary tumor such as amplifications of KRAS (3/5) in NSCLC and ERBB2 (2/4)

in breast carcinomas (Fig. 2-2f,g).

2.2.2 Intratumoral genomic heterogeneity of gliomas and BrMETs

Having characterized all variants in our glioma and BrMET cohorts, we next sought to

characterize the extent of intratumoral genomic heterogeneity within all tumors. First, we

categorized all variants in each sample as either “clonal” (present in all sequenced regions of a

tumor), “subclonal shared” (present in more than one region of a tumor but not all), or “subclonal

private” (only present in one region of a tumor). We observed a striking difference between tumor

types with a significantly (p<0.001; unpaired t-test) greater clonal variant fraction within the

BrMETs (median=0.88) than within gliomas (median=0.41) (Fig. 2-4a,b). Importantly, this

observation was independent of the histology of the primary malignancy, as lung, breast and

melanoma metastases all harbored a high fraction of clonal variants. In contrast, gliomas (primary

and recurrent samples pooled) contained a higher fraction of both subclonal shared (p<0.05;

unpaired t-test) and subclonal private (p<0.05; unpaired t-test) variants than did the BrMETs

(median subclonal private variant fraction 0.28 vs. 0.09) (Fig. 2-4b). Overall, approximately 43%

of mutations within the glioma cohort were categorized as clonal. While this is slightly lower than

two previous studies have reported, almost all of those tumors had only two spatially separate

regions analyzed, in contrast to the three or four sectors sequenced in our study207,219. Focusing

specifically upon cancer driver genes, mutations in TP53, PIK3CA, and KRAS were clonal in all

cases among BrMET samples. In contrast to the generalized heterogeneity of gliomas, a significant

majority of TERT promoter (13/14 clonal), TP53 (6/7), and EGFR (3/3) mutations were clonally

distributed. However, most PTEN (1/4 clonal) and NF1 (0/3) alterations were not clonal within

gliomas. These findings are largely consistent with a prior multisector sequencing study that noted

significant clonality of TERT promoter and TP53 mutations207. However, this same group did note

a majority of identified EGFR mutations were subclonal private in contrast to our cohort207.

To explore the translational implications of this heterogeneity, we calculated the fraction

of total tumor variants that would have been identified from sequencing a single glioma or BrMET

tumor site, which simulates the information that would be obtained from a single-site biopsy at

0.2

0.4

0.6

0.8

1.0

BrMET

Primary

Glioma

Fraction Variants Identified

Breast Cancer

Melanoma

NSCLC

SCL C

Primary Glioma

Rec. Glioma

Single Site Sequencing

N.S.

25 0

50 0

75 0

1 Site

2 Sites

3 Sites

Tot al Variants Identified

BrMET

10 0

15 0

20 0

1 Site

2 Sites

3 Sites

Glioma

GBM030

GBM032

GBM051

GBM052

GBM055

GBM056

GBM059

GBM062

GBM063

GBM064

GBM069

GBM070

GBM074

GBM079

AO083

GBM018.Re

GBM031.Re

GBM047.Re

GBM065.Re

CNV Count

Glioma

BrMET008

BrMET009

BrMET010

BrMET019

BrMET025

BrMET058

BrMET NSCLC

BrMET018

BrMET023

BrMET024

BrMET027

BrMET BRCA

*** **

0.00

0.25

0.50

0.75

1.00

BrMET

Proportion

Clonal Subclonal Shared Subclonal Privat e

Recurrent

Glioma

Subclonal Privat e

Subclonal Shared

Clonal

c d

Glioma

Figure 2-4: Intratumoral genomic heterogeneity of variants and copy number alterations. a, Variant

clonality per tumor. b, Proportion of clonal, subclonal shared, and subclonal private variants in brain metastases

and gliomas. Significance determined by unpaired t-test. *p< 0.05, ***p< 0.001. c, Proportion of total identified

variants that would have been captured through the sequencing of a random single site from within each tumor.

Significance determined by unpaired t-test. *p< 0.05, **p< 0.01. d, Total variants identified per tumor if one,

two, or three samples were pooled for analysis. Significance determined by unpaired t-test. *p< 0.05. e, Copy

number variation (CNV) clonality per tumor in the gliomas (left), NSCLC brain metastases (middle), and breast

cancer brain metastases (right) cohorts.

BrMET008

BrMET009

BrMET010

BrMET018

BrMET019

BrMET023

BrMET024

BrMET025

BrMET027

BrMET028

BrMET058

GBM030

GBM032

GBM051

GBM052

GBM055

GBM056

GBM059

GBM062

GBM063

GBM064

GBM069

GBM070

GBM074

GBM079

AO083

GBM018.Re

GBM031.Re

GBM047.Re

GBM065.Re

40 0

80 0

12 00

60 00

Variant Co unt

surgery. We observed that a higher fraction (p<0.01; unpaired t-test) of the total variants were

identified within BrMETs (median=0.92) compared to primary gliomas (median=0.73) when

sampling a single site (Fig. 2-4c). Given the limitations of single-site sampling to capture tumor-

wide variant information in gliomas, we determined the extent to which multi-region sequencing

could lead to the identification of additional tumor variants within our cohort. Among the BrMETs,

sampling additional tumor regions did not identify significantly more variants. However, among

glioma samples, sequencing three regions instead of one raised the median number of identified

variants from 61 to 98 (Fig. 2-4d). Thus, multi-region sequencing in gliomas captures a more

complete picture of the genomic landscape but provides only limited improvement in the

characterization of BrMETs due to their increased comparative spatial genomic homogeneity.

Having characterized glioma and BrMET genomic architecture at the variant level, we next

sought to characterize the intratumoral heterogeneity of CNAs to determine if there is evidence for

spatial architecture that is similar to that of the variants. Strikingly, the landscape of CNAs within

gliomas was markedly more spatially heterogeneous than the pattern observed within BrMETs

(Fig. 2-5). When we classified each CNA event as clonal, subclonal shared, or subclonal private,

we identified a significant fraction of clonal CNAs within BrMETs in contrast to the

predominantly subclonal CNAs within gliomas (Fig. 2-4e). Thus, at both the variant and CNA

level, gliomas are significantly more spatially heterogeneous than BrMETs.

Finally, to characterize these samples at the transcriptional level, we made use of a

previously defined gene expression-based molecular classification of GBM into proneural, neural,

classical, and mesenchymal subtypes172. We identified a distribution of transcriptional subtypes

with 12 neural, 12 classical, 16 mesenchymal, and 6 proneural subtypes across the 46 primary or

recurrent GBM tumor sectors. Intriguingly, in a majority (9/16) of the tumors with multiple regions

22q12.2

21q22.3

20q11.1

19q13.43

19q13.42

19q13.31

19q13.12

19q13.2

19p13.12

19p13.11

19p13.2

18p11.1

17p13.2

17p13.1

17p11.2

16q23.1

16q21

16q11.1

16p13.11

16p13.3

16p12.3

16p12.1

16p11.2

15q14

14q24.1

14q11.2

13q21.33

13q13.1

12q24.31

12q24.23

12q15

12q14.1

12q13.12

12q11

12p13.33

11q12.3

11p11.11

10q23.31

10q11.1

9q11

9p21.3

9p13.3

8p11.1

7q36.3

7q34

7q31.32

7q22.1

7q11.23

7p22.1

7p11.2

7p11.1

6q21

6p22.2

6p21.33

6p12.2

5q32

5q31.3

5p13.3

4q34.3

4p16.3

3q11.1

3p14.3

2q31.2

2q11.2

2q11.1

2p11.1

1q32.1

1q21.2

1p36.33

1p36.23

GBM051 2 3

GBM052 2 3

GBM055 2 3

GBM056 2 3

GBM059 2 3

GBM062 2 3

GBM063 2 3

GBM064 2 3

GBM069 2 3

GBM070 2 3

GBM074 2 3

GBM079 2 3

AO8 3 2 3

GBM018.Re 2 3

GBM031.Re 2 3

GBM047.Re 2 3

Glioma

20q11.22

19p13.2

19p12

18p11.1

17q25.3

17q24.1

14q11.2

13q33.2

12p12.1

11q13.2

11p11.11

9p11.1

8p21.1

8p12

7p22.1

7p11.1

6q26

6q24.3

6p25.1

6p23

6p21.33

5p15.2

5p13.2

3q27.2

3q25.1

3q11.1

2q24.2

2q11.1

1q42.2

1q21.3

1q21.2

1p36.33

1p36.12

Br ME T00 8 2 3

Br ME T00 9 2 3

Br ME T01 0 2 3

Br ME T01 9 2 3

Br ME T02 5 2 3

Br ME T05 8 2

BrMET NSCLC

21q22.3

19p13.3

17q25.1

17q24.1

17q23.2

17q21.33

17q21.31

17q12

17p12

17p11.2

16p13.11

16p11.2

15q26.3

15q26.1

15q22.2

14q32.31

13q33.3

13q14.2

13q13.1

13q12.3

12q24.23

12q24.12

12q24.11

12q11

11q13.3

11q13.1

11p11.11

10q23.31

9p24.1

8q24.13

8q21.13

8q13.3

8p21.3

8p11.1

7q31.1

7q22.1

7q11.23

7p11.1

6q27

6q25.3

6q25.1

6q24.1

6q23.3

6q22.33

6q22.31

6q21

6q15

6q14.1

6q12

6p21.1

4q35.2

4q21.1

3q26.2

3q11.1

1q44

1q32.2

1q21.2

Br ME T01 8 2 3

Br ME T02 3 2 3

Br ME T02 4 2 3

Br ME T02 7 2 3

BrMET BRCA

−2

−1

GBM032

GBM030

GBM065.Re

Figure 2-5: Intratumoral landscape of recurrent copy number variation events. Copy number variation

events in gliomas (left), non-small cell lung cancer brain metastases (middle) or breast cancer brain metastases

(right). Colors depict relative intensity of copy number alterations per sample. Positive scores in red indicate

amplifications and negative scores in blue indicate deletions.

Figure 2-6: Molecular subtype classification of GBM. Circos plot displaying the classification of each tumor

region among the four molecular subtypes of GBM. Classification was determined by gene set enrichment

analysis on previously defined gene signatures (see Methods).

Neural

Mesenchymal

Classical

Proneural

GBM018.Re

GBM030

GBM031.Re

GBM032

GBM047.Re

}

GBM051

GBM052

}

GBM055

GBM056

}

GBM059

GBM062

GBM063

GBM064

GBM065.Re

}

GBM069

GBM070

GBM074

GBM079

analyzed, we observed intratumoral heterogeneity of these transcriptional subgroups (Fig. 2-6), in

line with prior reports173.

2.2.3 Intratumoral heterogeneity of tumor antigens in gliomas and BrMETs

Numerous clinical trials developing personalized neoantigen vaccines for GBM and other

brain cancers are ongoing144,145. To determine the consequences of tumor genetic heterogeneity on

immunologic features of each tumor, we applied the pVacSeq neoantigen prediction pipeline220,221

to define the neoantigen landscape across all glioma and BrMET samples. When we aggregated

the total number of predicted neoantigens for each tumor from all sampled regions, BrMETs

harbored a higher number of HLA class I neoantigens per tumor (median=186) than either primary

(median=39) or recurrent gliomas (median=51). Moreover, BrMET class I neoantigens were

significantly (p<0.001; unpaired t-test) more clonal than those identified in gliomas (Fig. 2-7a,b).

Of note, all EGFR mutations did yield predicted clonal class I neoantigens. However, despite all

carrying the IDH1 R132H mutation, only one of the four IDH-mutant patients within the cohort

had a predicted class I neoantigen from this variant, displaying the HLA haplotype dependence of

these results. BrMETs also exhibited a greater number of HLA class II neoantigens (300) than

primary (60) or recurrent gliomas (81), and these neoantigens were significantly (p<0.001;

unpaired t-test) more clonal in their tumor distribution (Fig. 2-8a,b). For most patients, the spatial

distribution of both class I and class II neoantigens closely mirrored the underlying variant

distribution (Fig. 2-9a-c). Finally, whereas BrMET HLA class I neoantigens are mostly captured

by single site tissue sampling, additional glioma neoantigens continue to be identified as additional

tissue sites are sampled (Fig. 2-7c). Taken together, these data demonstrate that neoantigens are

distributed heterogeneously in gliomas compared to BrMETs.

Survivin

NY−ESO −1

EPHA2

HER 2/Neu

IL13RA2

MAGEA3

MAGEA4

MAGEA6

MAGEA10

MART1

gp 100

SART1

TERT

WT1

Survivin

NY−ESO −1

EPHA2

HER 2/Neu

IL13RA2

MAGEA3

MAGEA4

MAGEA6

MAGEA10

MART1

gp 100

SART1

TERT

WT1

0.0

0.5

1.0

1.5

2.0

2.5

CT Antigen Score

Intrat umoral Variance

BrMET Glioma

Clonal Subclonal Shared Subclonal Privat e

N.S.

10 0

20 0

30 0

1 Site

2 Sites

3 Sites

Total Class I Neoant igens Identified

BrMET

1 Site

2 Sites

3 Sites

Glioma

Breast Cancer

Melanoma

NSCLC

SCL C

Primary Glioma

Recurrent Glioma

a b

c e

WT1

TE RT

Survivin

SART1

NY−ESO−1

MART1

MAGEA 6

MAGEA 4

MAGEA 3

MAGEA 10

IL13RA2

Her2/Neu

gp100

EPHA2

GBM032 2

GBM051 2

GBM055 1

GBM056 2

GBM059 2

GBM062 2

GBM063 2

GBM064 2

GBM069 2

GBM074 2

GBM079 2

AO083 23

GBM018.Re 2

Glioma

BrMET009 23

BrMET010 23

BrMET018 23

BrMET019 23

BrMET023 23

BrMET024 23

BrMET027 23

BrMET028 23

BrMET058 2

BrME T

−10

−5

GBM030

GBM052

GBM070

GBM031.Re

GBM047.Re

GBM065.Re

BrMET008

BrMET025

Figure 2-7: Intratumoral neoantigen and cancer/testis antigen heterogeneity. a, Class I neoantigen

clonality per tumor. b, Proportion of clonal, subclonal shared, and subclonal class I neoantigens in brain

metastases and gliomas. Significance determined by unpaired t-test. **p< 0.01, ***p< 0.001. c, Impact of

multi-region sequencing on total class I neoantigen load. Significance determined by unpaired t-test.

**p< 0.01. d, Heat map of cancer/testis (CT) antigen scores for each sample calculated by normalizing tumor

expression to normal “Brain-Cortex” expression (see Methods). e, Plot of the average intratumoral variation

in CT scores between regions of the same tumor by the average cancer/testis antigen score for each gene

among all brain metastases or gliomas.

BrMET008

BrMET009

BrMET010

BrMET018

BrMET019

BrMET023

BrMET024

BrMET025

BrMET027

BrMET028

BrMET058

GBM030

GBM032

GBM051

GBM052

GBM055

GBM056

GBM059

GBM062

GBM063

GBM064

GBM069

GBM070

GBM074

GBM079

AO083

GBM018.Re

GBM031.Re

GBM047.Re

GBM065.Re

20 0

40 0

60 0

16 00

Clas s I Neoant igen Count

0.00

0.25

0.50

0.75

1.00

Prop ortion

BrMET

Gl i om a

***

N.S.

***

N.S. **

0.00

0.25

0.50

0.75

1.00

BrMET Glioma

Proportion

Figure 2-8: Class II neoantigen clonality. a, Class II neoantigen clonality per tumor. b, Proportion of clonal,

subclonal shared, and subclonal private class II neoantigens in brain metastases and gliomas. Significance

determined by unpaired t-test. **p< 0.01, ***p< 0.001.

Clo nal Subclonal Shared Subclonal P rivate

a b

BrMET008

BrMET009

BrMET010

BrMET018

BrMET019

BrMET023

BrMET024

BrMET025

BrMET027

BrMET028

BrMET058

GBM030

GBM032

GBM051

GBM052

GBM055

GBM056

GBM059

GBM062

GBM063

GBM064

GBM069

GBM070

GBM074

GBM079

AO083

GBM018.Re

GBM031.Re

GBM047.Re

GBM065.Re

200

400

600

800

3000

Class II Neoantigen Count

Cl ona l Subclonal Shared Subclonal Private

2333

Class I Neoantigen Intersections

BrMET009−3

BrMET009−2

BrMET009−1

0204060

Class I Neoantigens

16 1

55864

10 0

15 0

Class II Neoantigen Intersections

BrMET009−3

BrMET009−2

BrMET009−1

05010 0

15 020 0

Class II Neoantigens

20 20

Variant Intersections

GBM062−3

GBM062−2

GBM062−1

025507510 0

Varia nt s

1 1

Class I Neoantigen Intersections

GBM062−3

GBM062−2

GBM062−1

0102030

Class I Neoantigens

34 6

72 3 3 3 1

10 0

20 0

30 0

Varia nt Intersecti ons

BrMET009−3

BrMET009−2

BrMET009−1

010 020 030 0

Varia nt s

12 12

Class II Neoantigen Intersections

GBM062−3

GBM062−2

GBM062−1

010203040

Class II Neoantigens

FIgure 2-9: Spatial distribution of variants and neoantigens. Spatial distribution of variants (a), class I

neoantigens (b), and class II neoantigens (c) for two represenative examples. Variants/neoantigens are

grouped based upon the set of tumor regions in which they are shared.

In addition to neoantigens, cancer/testis (CT) antigens represent another group of tumor-

specific antigens that can be recognized by the immune system. These antigens have highly

restricted expression in normal tissue, can be expressed in reproductive cells, and are often

upregulated in malignancies. Although CT antigens are targeted in a range of clinical trial

efforts201,222–224, their expression and distribution in brain cancers has not been described

previously. We therefore sought to characterize CT antigen expression and spatial distribution

within our tumor cohort. Since CT antigens are wild-type proteins, normal tissue expression

impacts both the degree of anticipated off-target effects from directed immunotherapeutic efforts

as well as the extent of immunological tolerance during development. Thus, we scored each

candidate antigen from a curated list of CT antigens based upon its log-transformed expression

relative to normal brain tissue. Among gliomas, some of the highest scoring CT antigens were

BIRC5 (Survivin), PMEL (gp100), and IL13RA2 (Fig. 2-7d). While BIRC5 was also the highest

scoring antigen among the BrMET samples, they displayed relatively higher expression of many

prominent CT antigens such as the MAGE family proteins, gp100, MART1, and HER2/neu as

compared to gliomas. One notable exception of this was IL13RA2 which was the lowest scoring

antigen among BrMETs but ranked third among gliomas. These findings were consistent

regardless of whether brain or matched primary site tissue was used to generate BrMET CT antigen

scores (Fig. 2-10a). In contrast to the marked heterogeneity of tumor neoantigens, we observed no

significant difference in the spatial distribution of CT antigen scores overall between gliomas and

BrMETs as assessed by a cosine similarity index (Fig. 2-10b). However, these characteristics do

vary between CT antigens with TERT, CTAG1B (NY-ESO-1), and the MAGE family trending to

higher intratumoral variance (greater heterogeneity of expression) than the more clonal BIRC5 and

gp100 (Fig. 2-7e).

WT1

TE RT

Survivin

SART1

NY −ESO−1

MA RT1

MAGEA6

MAGEA4

MAGEA3

MAGEA10

IL13RA2

He r2/N eu

gp100

EPHA2

BrME T009 2

BrME T010 2

BrME T018 2

BrME T019 2

BrME T023 2

BrME T024 2

BrME T027 2

BrME T028 2

BrME T058 2

−10

−5

BrMET - normalized to primary site

N.S.

0. 2

0. 4

0. 6

0. 8

1. 0

CT Antigen Similarity

Breast Cancer

Melanoma

NSCLC

SCL C

Primary Glioma

Rec. Glioma

Figure 2-10: Intratumoral cancer/testis antigen similarity. a, Heat map of cancer/testis (CT) antigen

scores for each BrMET tumor sample calculated by normalizing tumor expression to matched primary tissue

site expression for each gene (in contrast to “Brain-Cortex” expression as done in Fig. 3). b, Comparison of

intratumoral CT antigen similarity as assessed by a pairwise cosine similarity metric on vectors composed

of the scores (normalized to “Brain-Cortex” expression) for each CT antigen.

BrMET Glioma

BrME T008

BrME T025

2.2.4 Spatial resolution of immune landscapes in gliomas and BrMETs

We next sought to define the immune cell infiltration within each tumor and to describe

the extent of intratumoral heterogeneity within the local immune microenvironments. We adopted

previously published immune deconvolution methods that resolve immune cell populations from

transcriptional data such as Danaher immune scores225 and CIBERSORT226 to characterize the 80

tumor regions for which there was sufficient RNA to generate RNA sequencing data. We

calculated Danaher immune scores for all regions and observed substantial intertumoral variation

specifically among “CD8 T cell” and “Cytotoxic cell” scores (Fig. 2-11a). Surprisingly, we

detected no significant difference in the aggregate immune scores between regions from glioma

and BrMET samples.

To probe potential differences in more detail, we next performed differential gene

expression analysis on tumor regions from gliomas and BrMETs. We detected significantly higher

levels of CD274 (PD-L1) and significantly lower levels of CXCL9 in gliomas compared to

BrMETs (q<0.01; t-test with multiple comparison correction), consistent with the severe

immunosuppression appreciated in the former188 (Fig. 2-11b). Additionally, because there was a

robust infiltration of macrophages in both gliomas and BrMETs as determined by CIBERSORT

(Fig. 2-12a), we explored two previously published gene sets227,228 to further characterize this

population. While we observed a slight polarization towards the immunosuppressive M2

phenotype characterized by higher expression of STAT3 and MRC1 (CD206) within gliomas, we

identified a major distinction within macrophage ontogeny. Specifically, the BrMETs had a

significant skewing towards higher expression of genes associated with monocyte-derived

macrophages relative to gliomas, which were enriched for microglial-specific genes (Fig. 2-11c).

This is consistent with recent work in the field utilizing single-cell analyses229,230.

NK CD56dim cell s

NK cells

Ne utroph ils

Mast cells

Macro pha ges

T-reg

Th1 cells

T ce lls

Exha usted CD8

Cytotoxic cells

CD 8 T cells

CD 45

B cells

Br ME T009 2

Br ME T010 2

Br ME T018 2

Br ME T019 2

Br ME T023 2

Br ME T024 2

Br ME T027 2

Br ME T028 2

Br ME T058 2

GBM032 2

GBM051 2

323

GBM055 1

GBM056 2

GBM059 2

GBM062 2

GBM063 2

GBM064 2

GBM069 2

3121

GBM074 2

GBM079 2

AO08 3 23

GBM018.Re 2

12342312

Br ME T 02 5

Br ME T 00 8

GBM030

GBM052

GBM070

GBM031.Re

GBM047.Re

GBM065.Re

−2

−20 2 4

PC1 (5 6. 2% explained var.)

PC2 (9 .4 % explained var.)

GBM018.Re

GBM031.Re

GBM032

GBM051

GBM056

GBM062

GBM063

GBM064

GBM069

GBM074

GBM079

AO083

−2

−20 2 4

PC1 (5 6. 2% explained var.)

PC2 (9 .4 % explained var.)

BrMET008

BrMET009

BrMET010

BrMET018

BrMET019

BrMET023

BrMET024

BrMET027

BrMET028

PD−L1 CXCL9

M2−M1 Skew

BrMET Glioma

0.0

2.5

5.0

MDM−Microglia S kew

GliomaBrMET

Brea st Cancer

Mel ano ma

NSC LC

SCL C

Primary Gl ioma

Re curren t Glio ma

Lo g( TPM )

BrMET Glioma

Lo g( TPM )

BrMET Glioma

Brea st Cancer

Mel ano ma

NSC LC

SCL C

Primary Gl ioma

Re curren t Glio ma

Figure 2-11: Spatial diversity of the immune microenvironment. a, Heat map of the immune cell scores for

all samples as estimated by the method from Danaher et al (see Methods). Each column represents a tumor

region and each row represents an immune population. Scores represent the average of the log-transformed

expression of a collection of subset-specific genes. b, Difference in log-transformed expression of PD-L1 (left)

and CXCL9 (right) between tumor types for all samples within cohort. Significance determined by t-test with

Benjamini-Hochberg multiple test correction. **q < 0.01. c, Difference in macrophage polarization (left) or

ontogeny (right) based upon previously published gene sets (see Methods) between tumor types for all

samples within cohort. Significance determined by two-sided t-test. *p<0.05. **p<0.01. d, Danaher scores

for each sector from a tumor with three or more samples plotted in PC1-PC2 space following principal

components analysis.

Figure 2-12: Immune microenvironment profiling and intratumoral heterogeneity.

a, Simplified CIBERSORT (see Methods) output estimating abundance for each immune cell type for each

tumor sample in cohort. b, Quantification of immune intratumoral heterogeneity or similarity through either

the area encompassed by regions from each tumor in PC1-PC2 space (left) or an intratumoral pairwise

cosine similarity metric on vectors composed of the Danaher immune scores for each region (right).

Breast Cancer

Melanoma

NSCLC

SCLC

Primary Glioma

Rec. G lioma

0.00

0.25

0.50

0.75

1.00

Br ME T009 2

Br ME T010 2

Br ME T018 2

Br ME T019 2

Br ME T023 2

Br ME T024 2

Br ME T027 2

Br ME T028 2

Br ME T058 2 1

GBM032 2

GBM051 2

GBM055 1

GBM056 2

GBM059 2

GBM062 2

GBM063 2

GBM064 2

GBM069 2

GBM074 2

GBM079 2

AO0 8 3 2 3

GBM018.Re 2

GBM047.Re 3

Prop ortion

Cell Type

B cells

CD 4+ T cells

CD 8+ T cells

DC s

Macro pha ges

Mon ocytes

NK cells

Other

Br M E T 00 8

Br M E T 02 5

GBM030

GBM052

GBM070

GBM031.Re

GBM065.Re

N.S.

0. 0

0. 1

0. 2

0. 3

0. 4

Area

Danaher ITH

BrMET Glioma

0. 95

0. 96

0. 97

0. 98

0. 99

Cosine Similarity

Intratumoral Danaher Similarity

BrMET Glioma

Finally, we assessed the degree of immunologic intratumoral heterogeneity as estimated

by the Danaher scores. We performed principal components analysis on the Danaher immune

scores for all tumors with RNA from at least three sectors. Plotting each region in PCA space, we

observed that most regions from the same tumor clustered together (Fig. 2-11d). To quantify the

extent of heterogeneity, we calculated the area in PC1-PC2 space (termed Danaher ITH) of the

triangle with vertices corresponding to each region of a tumor. In contrast to the substantial

differences in variant and neoantigen heterogeneity, we detected no difference in the degree of

intratumoral immune cell heterogeneity between the tumor types (Fig. 2-12b). A similar result was

obtained by calculating pairwise cosine similarity for Danaher immune scores between regions

from the same tumor (Fig. 2-12b). Overall, these data suggest that the intratumoral spatial variation

in the immune microenvironment is similar between gliomas and BrMETs and generally is less

than the intertumoral variation.

2.2.5 T cell receptor clonotypic diversity and heterogeneity

To evaluate the diversity, heterogeneity, and degree of clonal expansion of TCR clonotypes

within the infiltrating T cell populations of GBM and BrMET tumor samples, we performed TCR

sequencing on 65 regions from 22 tumors in the cohort. The TCR b-chain complementarity-

determining region 3 (CDR3) is highly diverse and plays a significant role in antigen recognition.

Therefore, the TCR b chain CDR3 sequences can function as unique barcodes of individual T cell

clones as they are activated and undergo clonal expansion within the tumor. We classified

clonotypes within each tumor region as either the dominant clone (Clone 1) or in predetermined

clonotype groups based on frequency (i.e., Clones 2-5, 6-20, 21-100, 101-1000, or more than 1000

clones) (Fig. 2-13a & Fig. 2-14a). Unexpectedly, we observed substantial clonal expansion within

0.00

0.04

0.08

0.12

T Cell Fraction

0.0

0.1

0.2

0.3

Simpson Clonality

Breast Cancer

Melano ma

NSC LC

SCLC

Primary GBM

Recurren t GBM

25 %

50 %

75 %

10 0%

BrMET009 2

BrMET010 2

BrMET018 2

BrMET019 2

BrMET025 2

BrMET027 2

BrMET028 2

GBM052 2

GBM055 2

GBM056 2

GBM059 2

GBM070 2

GBM074 2

GBM079 2

Percentage of Repe rtoire

GBM032

GBM063

GBM064

Clone 1

Clones 2−5

Clones 6−20

Clones 21−100

Clones 101−1000

Clones 1000+

GBM047.Re 2

GBM065.Re

GBM079 2

GBM074 2

GBM070 2

GBM059 2

GBM056 2

GBM055 2

GBM052 2

GBM047.Re 2

GBM052 2

GBM055 2

GBM056 2

GBM059 2

GBM070 2

GBM074 2

GBM079 2

GBM

BrMET028 2

BrMET027 2

BrMET025 2

BrMET019 2

BrMET018 2

BrMET010 2

BrMET009 2

BrMET010 2

BrMET018 2

BrMET019 2

BrMET025 2

BrMET027 2

BrMET028 2

BrMET

0. 00

0. 25

0. 50

0. 75

Morisita Overlap

GBM064

GBM063

GBM065.Re

GBM032

GBM063

GBM064

GBM065.Re

Figure 2-13: Intratumo ral T cell repertoire clonality and diversity. a, The proportion of each sample’s T cell

repertoire consisting of clones from a given rank position when all TCR sequences are ordered by

descending frequency in the sample. b, Comparison of T cell fraction (left) and Simpson clonality (right)

between regions from GBMs or BrMETs. T cell fraction and Simpson clonality were calculated as described

in Methods. P values calculated by two-sided t-test: *P<0.05. c, Heat maps showing the frequencies of

the ten most expanded intratumoral beta-chain sequences per tumor in different regions. Each row

represents one unique sequence and each column a tumor region. d, Quantification of TCR intratumoral

heterogeneity by calculation of the Morisita overlap (see Methods) between pairs of distinct tumor regions

in either GBM (left) or BrMETs (right). Values range from 0 (indicating no similarity) to 1 (identical

TCR repertoires).

BrMET

GBM

BrMET

GBM

Clono type

Reg. 1

Reg. 2

Reg. 3

BrMET009

Reg. 1

Reg. 2

Reg. 3

BrMET018

Reg. 1

Reg. 2

Reg. 3

GBM074

Reg. 1

Reg. 2

Reg. 3

GBM079

0.00

0.04

0.08

0.12

0.16

20 %

40 %

60 %

80 %

Cl one 1

Cl one s 2-5

Cl one s 6-20

Cl one s 21-10 0

Cl one s 101-1 ,000

Cl one s 1,0 01 - 20,000

Proportion of repertoire

BrMET

GBM

0.00

0.25

0.50

0.75

Co sine Similarity

GBM079 2

GBM074 2

GBM070 2

GBM059 2

GBM056 2

GBM055 2

GBM052 2

GBM047.Re 2

GBM052 2

GBM055 2

GBM056 2

GBM059 2

GBM070 2

GBM074 2

GBM079 2

GBM

GBM032

GBM063

GBM064

GBM065.Re

GBM032

GBM063

GBM065.Re

GBM064

BrMET028 2

BrMET027 2

BrMET025 2

BrMET019 2

BrMET018 2

BrMET010 2

BrMET009 2

BrMET010 2

BrMET018 2

BrMET019 2

BrMET025 2

BrMET027 2

BrMET028 2

BrMET

Br e ast Ca nc er

Melanom a

NSCLC

SC L C

Primary GBM

Recurrent GBM

0.0

0.2

0.4

0.6

0.8

1.0

Intratumoral C osine Simil arit y

1.0

0.0

0.2

0.4

0.6

0.8

Intratumo ral Mo risi ta Overla p

Figure 2-14: Intratumoral T cell repertoire heterogeneity. a, The proportion of the T cell repertoire for

each sample comprised of clones within a given clonotype group when clones are ranked in descending

order of frequency. b, Quantification of TCR intratumoral heterogeneity by calculation of the cosine similari-

ty (see Methods) between pairs of distinct tumor regions in either GBM (left) or BrMETs (right). Values

range from 0 (indicating no similarity) to 1 (identical TCR repertoires). c, Comparison of intratumoral T cell

repertoire similarity between GBM and BrMETs via either Morisita overlap (left) or cosine similarity (right).

Each data point represents the pairwise comparison between regions from the same tumor. Significance

determined by unpaired t-test. *p<0.05.

BrMET GB M BrMET GBM

GBM. For example, each region of GBM055 harbored a dominant clone comprising >17% of the

T cell repertoire. In addition, regions from GBM047.Re, GBM056, GBM059, GBM065.Re, and

GBM079 all contained dominant clones present at frequencies >12%. In contrast, of the sequenced

BrMETs, only 1 region from BrMET025 (NSCLC) harbored a dominant T cell clone present at a

frequency >10%. Overall, the T cell fraction among all cells (estimated through TCR sequencing;

see Methods) was significantly (p<0.05; unpaired t-test) higher within the BrMETs, while the TCR

repertoires within GBM were determined to have a higher degree of clonality (p<0.05; unpaired

t-test) (Fig. 2-13b).

To investigate the degree of intratumoral heterogeneity of the T cell repertoires, we next

explored the distribution of the top 10 clones within each tumor. The distribution of the expanded

TCRs within each tumor differed greatly between patients, with some exhibiting marked T cell

homogeneity among all examined regions while others showed profound intratumoral diversity

(Fig. 2-13c). Several of the GBMs were particularly heterogeneous with locally expanded T cell

clones present at frequencies greater than 10% within one region of the tumor but significantly

less than 1% in all other regions. We then quantified this diversity at a repertoire-wide level

through pairwise comparisons using the Morisita overlap index (MOI), a measure of similarity

between populations based on the number of shared sequences and their relative frequencies. As

expected, the MOI values approached 0 (no similarity) for repertoires from different patients but

were variable within patients (Fig. 2-13d). Comparing the tumor types, we detected a higher level

of intratumoral repertoire similarity among BrMETs than among GBM whether the MOI or a

similar cosine similarity index was used (Fig. 2-14b,c). Therefore, the increased spatial

heterogeneity of variants and neoantigens within GBM is recapitulated at the TCR repertoire level.

Ultimately, to further validate that the expanded clones were specifically enriched within

tumor tissue, we performed TCR sequencing on the peripheral blood from 8 of these patients (5

GBM/3 BrMET). As expected, the T cell repertoire from peripheral blood (PBMC) trended

towards lower clonality than the matched TIL (Fig. 2-15a). Intriguingly, the degree of repertoire

similarity between PBMC and TIL appeared greater (p=0.07) within BrMETs than within GBM,

perhaps suggestive of a stronger systemic immune response in patients with metastatic disease

(Fig. 2-15b). Tracking the most highly expanded intratumoral clones across each patient confirmed

that the majority were detectable in peripheral blood but at substantially reduced frequencies (Fig.

2-15c). In addition to this clonotypic diversity, we also observe differential V and J gene usage

between peripheral blood and matched TIL, suggestive of a combination of both VJ-dependent

and VJ-independent divergence as previously described (Fig. 2-16)231.

Figure 2-15: Clonality and similarity of peripheral blood T cell repertoire. a, Comparison of T cell

repertoire Simpson clonality between peripheral blood and matched TIL. b, Quantification of TCR similarity

between peripheral blood and matched TIL through calculation of the Morisita overlap (see Methods).

Values range from 0 (indicating no similarity) to 1 (identical TCR repertoires). c, Alluvial plots tracking the

frequencies of the top 5 clonotypes from within each region of the tumor. Each clonotype is defined below

the associated plot by the amino acid sequence of its CDR3 and the V gene used.

p= 0.13

0.0

0.1

0.2

0.3

Simpson Clonality

Melanom a

NSCLC

Primary GBM

Recurrent GBM

PB MC

TIL

Clonotype

CASRHVQET QYF TRBV6−5

CASSQQGAGDNSPLHF TRBV7−9

CASSQVIWDRDRSSYNSPL HF TRB V14−1

CASSSAWRTY E Q Y F T RB V 5−1

CASSY QGTTEAFF TRBV6−2/ 06−3

CAST IKSGRDTG ELFF TRBV6−1

CSARDGTSGSPSYE QYF TRBV20−1

BrMET009

Clon ot ype

CASRTG L AH E Q Y F T R BV 11 −2

CASSLKAGLYGYTF TRBV7−9

CASSLRGAGNQPQH F TRBV 13−1

CASSLTW Q S S YE QY F T R BV 28 −1

CASSPAGGSGYN EQF F T RBV 7−9

CASSS GLAGEST DTQY F TRB V11 −1

CASSSLYVAGGPFF TRB V 28−1

CASSS TSVRAYEQYF TRBV7−8

CASSTKQGDYGYTF TRBV5−1

CASSY MEYEQYF TRBV6−2 /06−3

CAST SPSVYY EQYF TRBV6 −5

CATS VAGGYE QYF TRBV24−1

RASSLTG NQ PQH F TR B V7 −3

GBM047.Re

0.0

0.1

0.2

Prop ortion

PBMC Tumor

Re g. 1

Tumor

Re g. 2

Tumor

Re g. 3

Clon ot ype

CARSLAQGS P LHF T RBV5−3

CASSAGGGS SYE QY F T RBV 9−1

CASSHVPGQGESEQ YF TRBV4 −3

CASSIGWARNNEQFF TRBV19−1

CASSLGGGET QYF TRBV12−3 /12−4

CASSLGGRANSYE QYF TRBV11−1

CASSLTE Q GE Q F F T RB V 7−2

CASSP SYSYEQYF T RBV 9−1

CASSQDFGTGVAYEQYF TRBV1 4−1

CSARMG SEPAEAFF TRBV20−1

CSATP GR A N NSP L H F TR B V20 −1

GBM079

0.0

0.1

0.2

Prop ortion

PBMC Tumor

Re g. 1

Tumor

Re g. 2

Tumor

Re g. 3

0.0

0.1

0.2

0.3

Prop ortion

PBMC Tumor

Re g. 1

Tumor

Re g. 2

Tumor

Re g. 3

PBMC Tumor

Re g. 1

Tumor

Re g. 2

Tumor

Re g. 3

Clonotype

CASRLAAEQAFF TRBV12−2

CASRPQGRPDEQYV TRBV6 −5

CASSAGT GGNQPQ HF TRB V 2−1

CASSFQG SFNTEAF F TRBV28−1

CASSGGTGGNSPLHF TRBV2 −1

CASSHGRSNQPQHF TRBV1 1−2

CASSLDAGGSY EQY F TRBV5−5

CASSRDTLPGETQYF TRBV12−1

0.00 0

0.02 5

0.05 0

0.07 5

0.10 0

0.12 5

Prop ortion

PBMC Tumor

Re g. 1

Tumor

Re g. 2

Tumor

Re g. 3

BrMET010

Clon ot ype

CASSFGNT IYF TRBV5−4

CASSFSGQNEAF F TRBV27 −1

CASSLLEGNYEQYF TRBV7 −9

CASSMTGGGTEAFF TRBV5 −6

CASSP GNEQFF TRBV11−2

CASSQETGSYEQYF TRBV4−1

CASSQGTGDTGELFF TRBV4−3

CASSQSQNTEAF F TRBV4−1

CASSRRVGGY TF TRB V 11−2

CATRAGE GY GYT F T RBV 5−1

CSARGYTF TRBV29−1

BrMET028

0.00

0.03

0.06

0.09

0.12

Prop ortion

PBMC Tumor

Re g. 1

Tumor

Re g. 2

Tumor

Re g. 3

0.00

0.05

0.10

0.15

0.20

Prop ortion

GBM074

PBMC Tumor

Re g. 1

Tumor

Re g. 2

Clon ot ype

CAAKGLTE P TKLVANEKLF F TRBV10−1

CASLSPSRSYEQYF TRBV2 8−1

CASRAGVD EQF F T RBV2−1

CASSE TPNSPLHF TRBV10−1

CASSFRGL QETQYF TRBV5−4

CASSLTG MTE A FF TRB V5 −4

CASSP GFNYGYTF TRBV11 −1

CASSP GLAGT YY SYNEQFF TRBV18−1

CASSQDGTAISNQPQHF TRBV3−1/03 −2

CASSQRTG EP S YE QY F T R BV 7−9

CASSWGTGVDQGYTF TRBV7 −9

CSASLGPFDEQFF TRBV20−1

CSASRPG HG YTF TRBV20−1

CSVEDRDRGDYGYTF TRBV29 −1

CSVGEGGWEL FF TRB V2 9−1

Tumor

Re g. 3

p= 0.07

0.0

0.2

0.4

0.6

0.8

PB MC/TIL Morisita Overlap

BrMET

GBM

Melanom a

NSCLC

Primary GBM

Recurrent GBM

Figure 2-16: Distribution of T cell repertoire V-J usage. Modified circos plots displaying the prominent

V-J gene combinations within select tumors and matched peripheral blood. The frequency of a given V or J

gene segment within the repertoire is represented by its arc length along the circle, while the width of the

connection between a given V-J pair corresponds to the frequency of that combination. Within each patient,

the top 5 V-J pairs within any one region are highlighted across all matched samples. All plots correspond

to V-J usage among TCR β chain sequences.

BrMET009

GBM047.Re

GBM074

PBMC Tumor Reg. 1 Tumor Reg. 2 Tumor Reg. 3

2.3 Discussion, Future Directions and Conclusions

Despite numerous advances in both targeted and immune-based therapies, the prognosis

for patients with malignant brain tumors such as GBM or BrMETs remains poor. Within GBM,

one potential rationale for the high rate of treatment failure is the extensive intratumoral cellular

and molecular heterogeneity168 owing to complex tumor clonal dynamics. However, the impact of

this tumor cell diversity upon the tumor-immune microenvironment remains unclear. Furthermore,

the spatial heterogeneity of both tumor and immune landscapes within BrMETs requires further

study.

To address these knowledge gaps, we performed comprehensive immunogenomic profiling

on multiple spatially distinct regions from a cohort of 30 patients with primary or secondary

malignant brain tumors. We observed a striking distinction in the distribution of somatic variants,

with a majority of mutations usually shared by all analyzed regions within BrMETs while gliomas

were markedly heterogeneous and subclonal. This dichotomy extended to the distribution of

candidate neoantigens within these tumors, whereas the intratumoral distribution of targetable CT

antigens was more homogeneous. Furthermore, the intratumoral TCR repertoire was significantly

more similar between spatially distinct regions of BrMETs than gliomas, which often harbored

locally expanded T cell clones.

Previous studies have reported on the significant genomic and transcriptional variability

between spatially distinct regions of gliomas173,206,207 and hypothesized as to the consequences of

heterogeneity on treatment resistance. However, our results suggest that this spatial heterogeneity

of genomic alterations is minimal within metastatic brain tumors, which instead display a

markedly more clonal distribution. We envisage that this difference may be due to the separate

evolutionary trajectories of these tumor types. Recent studies suggest that GBMs arise from the

slow accumulation of somatic mutations in neural stem cells during which time multiple subclones

can develop before presenting as a clinically apparent tumor232. However, secondary BrMETs

likely develop quickly from the rapid growth of an already transformed subclone upon arrival into

the CNS. This malignant clone will quickly develop into an apparent tumor allowing less time for

the development of genetically disparate subclones. Previous studies have shown that these

metastatic clones can accumulate additional genomic alterations relative to the matched primary,

thereby providing potentially unique therapeutic targets that are likely clonal within the metastatic

tumor208,233. Future work should explore whether this tumor spatial homogeneity represents a

specific feature of BrMETs or is a more general characteristic of secondary metastatic tumors.

Importantly, the comparative distinctions we observed extend to the distribution of

candidate class I and class II neoantigens, as patients with metastatic tumors harbor a significantly

higher proportion of clonal neoantigens. This dichotomy between BrMETs and gliomas could have

profound implications on anti-tumor immunity as studies have reported that T cell

immunoreactivity against clonal neoantigens drives sensitivity to checkpoint blockade

treatment174,175. However, additional work is needed to characterize the degree of anti-tumor

immunity within malignant brain tumors and determine the relative contributions of clonal and

subclonal neoantigens in stimulating immune responses. In particular, detailed analysis of the

antigenic targets of the T cell clones within tumors will be critical to understanding how antigen

clonality shapes immunogenicity.

Recent studies have broadened our understanding of the immune microenvironment within

gliomas and BrMETs through methods such as mass cytometry, RNA sequencing, and

immunofluorescence229,230. Our work builds on this by exploring the spatial heterogeneity of the

immune response and performing a deeper analysis on the tumor infiltrating lymphocyte

populations. Through immune profiling from RNA-seq data, we did not detect significant

intratumoral differences in either the immune infiltrate or activation state for either tumor type in

contrast to the significant heterogeneity of variants and neoantigens within gliomas. However,

many of the immune profiling analyses performed are to some extent limited in their ability to

resolve specific immune cell populations and activation states owing to their reliance on bulk

RNA-seq data. Further spatial analysis with more sensitive metrics such as flow cytometry,

immunofluorescence, single-cell RNA-seq, and/or spatial transcriptomics would be needed to

clarify whether this is true immune homogeneity or a limitation of bulk RNA-seq analysis.

In contrast, TCR repertoire sequencing did uncover substantial spatial heterogeneity. While

both GBM and BrMETs demonstrated evidence of expanded intratumoral T cells, many of the

dominant clones within GBM samples were highly spatially restricted. These data suggest that

tumors harbor complex immune microenvironments in which a given tumor may contain pockets

of clonally expanded T cells adjacent to regions with minimally expanded T cells. Whether this T

cell heterogeneity is due to the recognition of spatially diverse subclonal antigens or the result of

extrinsic regional features such as inflammatory cytokines that promote clonal expansion requires

further study.

Taken together, these results carry immediate significance for the design and

implementation of clinical studies in malignant brain tumors. The extensive intratumoral

heterogeneity within gliomas indicates that single-site genomic analysis will not capture the

totality of targetable mutations and neoantigens. This is of particular importance in the design of

targeted therapy studies and/or neoantigen vaccines, as the analysis of multiple regions identifies

a significantly higher number of targetable neoantigens than one site alone. We have already begun

to implement this approach in a trial of a personalized neoantigen vaccine in patients with newly

diagnosed GBM (NCT03422094). An additional benefit of this approach is increased confidence

in the clonality of targeted antigens, as clonal neoantigens would presumably represent ideal

targets. However, it remains to be seen whether sampling additional regions beyond the number

in this study and the aforementioned clinical trial would shed further light on the molecular

landscape. Furthermore, the complexity of the intratumoral T cell repertoire argues that multiple

sites should be analyzed for the potential expansion of neoantigen-reactive T cells or isolation of

tumor-specific TCRs for therapy. On the other hand, the relative homogeneity of BrMETs suggests

that a single region is sufficient for genomic and immunological phenotyping.

Overall, this chapter provides an immunogenomic profile of malignant brain tumors

through a multi-sector approach showcasing substantial intratumoral heterogeneity within gliomas

while highlighting surprising homogeneity among BrMETs. These distinctions hold for both

cancer cell-intrinsic (genomic alterations) and cancer cell-extrinsic (TCR repertoire) features. A

growing understanding of the tumor-immune microenvironment and an appreciation of its spatial

complexity may improve the efficacy of immunotherapy in patients with malignant brain tumors.

CHAPTER THREE

Immunological characterization of malignant brain tumors

3.1 Introduction

In contrast to the low efficacy of ICB therapy in the treatment of primary brain

malignancies such as GBM, metastatic brain tumors have shown a surprising degree of

responsiveness161,162. This suggests that despite decades of skepticism rooted in the notion of CNS

immunoprivilege, the immune system can be leveraged to eradicate intracranial malignancies.

However, it also points to fundamental distinctions in the immunological state of primary tumors

such as GBM and secondary BrMETs. Despite extensive genomic profiling of these tumors,

comparatively less is known about their immunological microenvironments.

Furthermore, while primary GBM classically presents with fewer than 100 exome-wide

mutations, a unique subset of approximately 20-30% of recurrent gliomas exhibit a treatment-

induced hypermutated phenotype157,158,206,219. Conventional standard-of-care treatment for GBM

includes the chemotherapeutic Temozolomide (TMZ) which functions by alkylating DNA at

guanine residues. Mutations in components of the DNA mismatch repair pathway including MSH6

have been implicated in mediating TMZ resistance and are suspected to be involved in the

development of the hypermutated recurrent state234–236. This is likely due to an inability to detect

TMZ-induced DNA damage resulting in continued cancer cell proliferation and incorporation of

genomic defects into resulting daughter cells. Meanwhile, several studies across different cancer

types have reported enhanced aPD-1 therapy responsiveness in tumors with a high mutational

burden due to mismatch repair deficiency237,238. Thus, there has been immense interest in applying

this principle to the treatment of hypermutated recurrent GBM with ICB therapy. Early reports

have suggested some degree of responsiveness in patients with a hypermutated phenotype owing

to germline defects in DNA repair pathways, but the generalizability of these results to treatment-

induced hypermutation is not clear166,167,194. Furthermore, there has been no comprehensive

immunological profiling of these hypermutated recurrent tumors to assess their similarities and

differences to primary GBM.

Recent advances in single-cell phenotyping such as mass cytometry (CyTOF) and single-

cell RNA sequencing (scRNA-seq) have allowed for a more detailed picture of the immunological

state of GBM and BrMETs to emerge. A predominant focus of much immune profiling in

malignant brain tumors has been the tumor infiltrating myeloid cells, specifically those of the

macrophage or microglial lineage. This spotlight is well earned as tumor-associated macrophages

(TAMs) and microglia have been shown to comprise a significant fraction of the bulk tumor mass

in GBM and play a critical role in promoting tumor invasion, survival, and proliferation239–242.

Recent studies have utilized modern technologies to more deeply probe the immunological

microenvironment of both GBM and BrMETs with particular interest in the tumor-infiltrating

myeloid compartment. One study using CyTOF on a cohort of human tumors identified

significantly greater overall leukocyte infiltration into BrMETs with a higher fraction of

monocyte-derived tumor-associated myeloid cells230. Another study utilizing several

complementary methods produced a similar result with greater monocyte-derived macrophages

within BrMETs that appeared to express more pro-inflammatory mediators than the more

immunosuppressive microglia present in GBM229. These findings are consistent with our own data

outlined earlier that utilized bulk transcriptomics.

Perhaps owing to the relative scarcity of lymphocytes within gliomas, comparatively less

is known about the expression programs and functional states of tumor infiltrating T cells within

GBM and BrMETs. Flow cytometry-based profiling has revealed a state of severe exhaustion

among T cells within GBM characterized by the upregulation of numerous inhibitory receptors

such as PD-1, CTLA-4, and LAG-3187,188. A recent study utilizing scRNA-seq technology

identified NK cell-like expression signatures among cytotoxic glioma-infiltrating T cells with

expression of inhibitory receptors such as CD161243. However, these works did not provide a

comparative analysis between GBM and BrMETs of the expression programs and functional states

of tumor-infiltrating T cells. Furthermore, the TCR b chain sequencing described in the previous

chapter unveiled a substantial degree of intratumoral clonal expansion among both tumor types,

with GBM often harboring significantly expanded but spatially restricted T cell clones. Thus, we

sought to more deeply characterize the immune microenvironment within a subset of the 30 tumors

that underwent the spatial characterization described in chapter two. To do so, we performed

scRNA-seq on sorted CD45+ immune cells from within 5 primary GBM, 1 hypermutated recurrent

GBM, and 5 BrMET samples.

3.2 Results

3.2.1 Single-cell immunological characterization of GBM & BrMET cohorts

From the previously discussed cohort of 30 patients (15 primary glioma, 4 recurrent

glioma, and 11 brain metastases), a group of samples had direct ex vivo TIL frozen down at the

time of resection. In total, we sorted live CD45+ cells from 11 of these for scRNA-seq with

additional 5’ VDJ enrichment to provide T cell clonotype information. These 11 samples were

pooled for subsequent analysis performed with the aid of Anthony Wang, a graduate student in

our lab. Focusing specifically on CD3+ T cells, we identified 18 clusters that were grouped into 9

distinct cell types based upon canonical marker expression (Fig. 3-1). These cell types included

UMAP1

UMAP2

910

CD4+ Effector

T cells 2

Treg

Exhausted CD8+ T cells

Cytotoxic CD8+ T cells

Terminally Differentiated CD8+ T cells

Naive/CM T cells

CD4+ Memory T cells

CD8+ EM T cells

CD4+

Effector

T cells 1

Figure 3-1: T cell profile of GBM and BrMET cohorts. UMAP visualization of T cells collected from 5 BrMET and 6 GBM

samples identified by cell types (C0, C7, C12: Naive/CM T cells; C1, C10, C13: CD4+ Memory T cells; C15: CD4+ Effector T

cells 1; C3, C9: CD8+ EM T cells; C2, C6, C11: Cytotoxic CD8+ T cells; C4: Terminally Differentiated CD8+ T cells; C5,

C16, C17: Exhausted CD8+ T cells; C14: CD4+ Effector T cells 2; C8: Tregs).

four groups of CD8 T cells (Exhausted CD8+ T cells, CD8+ Effector Memory T cells, Cytotoxic

CD8+ T cells, Terminally Differentiated CD8+ T cells), four groups of CD4+ T cells (CD4+

Effector T cells 1, CD4+ Effector T cells 2, CD4+ Memory T cells, Tregs) and a population of

naïve T cells. Strikingly, we observed significant segregation by tumor type with certain clusters

and cell types belonging almost exclusively to GBM or BrMET samples, respectively (Fig. 3-2).

This distinction was present among both CD4 and CD8 T cell subsets, pointing to substantially

discordant T cell states between the two tumor types.

In contrast to prior reports on dramatic T cell exhaustion in GBM, we observed

comparatively few GBM CD8 T cells within the canonically exhausted CD8 T cell group (clusters

5, 16, 17) characterized by higher expression of LAG3, HAVCR2, TIGIT, and PDCD1 (Fig. 3-3).

Meanwhile, T cells from GBM displayed enrichment for a CD8 T+ cell subset (cluster 4)

characterized by expression of KLRG1. The expression of KLRG1 has been associated with a

terminally differentiated T cell subset characterized by proliferative dysfunction and decreased

effector functions in various models of chronic viral infection244–246. Notably, while these cells

expressed similar levels of cytotoxic molecules such as GNLY and GZMB relative to the exhausted

CD8 T cell groups, they expressed lower levels of exhaustion-associated markers such as LAG3,

PDCD1, and TOX (Fig. 3-3).

Furthermore, GBM and BrMET tumors appear to contain dramatically different CD4 T

cell compartments. Cluster 14 (CD4+ effector T cells 2) is comprised of effector CD4+ T cells

almost exclusively derived from BrMET samples, while cluster 15 (CD4+ effector T cells 1) is

predominant CD4+ T cells from GBM. Comparing the two groups, we observe substantially higher

levels of the co-stimulatory molecule TNFRSF4 (OX40) and pro-inflammatory cytokine IFNG

among the cluster 14 CD4+ T cells within BrMETs (Fig. 3-3). However, these CD4 T cells also

BrMET

GBM

UMAP 1

UMAP 2

0.0

0.5

1.0

BrMET GB M

Proportion of cells

Naive/CM T cell

CD4+ Memory T cell

CD8+ Memory T cell

CD4+ Effector T cell 1

CD8+ Cytotoxic T cell

CD8+ Ter minally Differentiated T cell

CD8+ Exhausted T cell

CD4+ Effector T cell 2

Tr egs

Figure 3-2: Distribution of T cell states by tumor typ e. a, UMAP projection of T cells collected from GBM & BrMET tumors

labeled by sample tumor type. b, Bar plot displaying relative frequency of cell types among each tumor type.

a b

CD8A

CD8B

CD4

SELL

CCR7

TCF 7

LEF1

IL7R

CD27

CD28

NKG7

PRF1

GZ MB

KLRG1

GZ MA

GZ MH

GN LY

GZ MK

IFNG

EOMES

SLAMF7

PDCD1

LAG3

TOX

ENTPD1

HAVCR2

TNF RSF 9

PRDM1

TIG IT

TNF RSF 14

BTLA

CD40LG

B3GAT1

CD84

CXCR4

CXCR5

ICOS

IL6R

IL21R

TNF RSF 4

SLAMF1

BCL6

MAF

STAT3

CTLA4

FOXP3

CD14

SPI1

12 7 0 10 13 1 9 3 15 11 6 2 4 17 516 14 8

Percent Expressed

Average Expression

Naive markers

Memory markers

Cytotoxic mar kers

Exhaustion markers

TF H-li k e markers

Treg mark ers

Naive/CM T cell

CD4+ Memory

T cell

CD8+ Memory

T cell

CD4+ Effector

T cell I

CD8+ Cytotoxic

T cell

CD8+ Terminally

Differentiated T cell

CD8+ Exhausted

T cell

CD4+ Effector

T cell 2 Tregs

Myeloid markers

Figure 3-3: Gene expression of T cell clusters. Dotplot of gene expression values across clusters for select cell type-

associated markers.

express substantially higher levels of markers typically associated with CD8 T-cell exhaustion

such as TIGIT, HAVCR2, and TOX. These two clusters also express unique cytotoxicity markers

with enrichment of GZMB and PRF1 among cluster 14 in BrMETs but higher levels of GZMK and

GZMH among cluster 15 in GBM samples. Additionally, in line with previous work utilizing

CyTOF, BrMET tumors display a substantially greater fraction of immunosuppressive Tregs230.

Finally, we investigated the degree of clonal expansion and gene expression profile of

clonally expanded T cells within both tumor types. Despite the regional clonal expansion described

for GBM in the previous chapter, at a bulk level most gliomas do not contain significantly

expanded clones (Fig. 3-4a). The notable exception to this is GBM065.Re which will be discussed

in detail in the next section. BrMET009 and BrMET010 (both NSCLC) have the most significant

clonal expansion among the BrMETs, and these expanded clones primarily lie within cluster 5

which was classified as an exhausted CD8 T cell cluster (Fig. 3-4b). In contrast, the expanded

clones from GBM065.Re lie within cluster 6 and did not express very high levels of canonical

exhaustion-associated genes despite significant clonal enrichment. Interestingly, the moderately

expanded clones observed in a majority of GBM samples did fall within the terminally

differentiated cluster 4 characterized by KLRG1 expression. While most of the expanded clones

belonged to CD8 T cells, the previously described cluster 14 consisting predominantly of BrMET

CD4 T cells with expression of TNFRSF4 (OX40) and IFNG did contain some of the only clonally

expanded CD4 T cells across all samples (Fig 3-4b).

0.00

0.25

0.50

0.75

1.00

Oc cupi ed Repe rtoir e Space

Clonal Indices

[1:1]

[2:5]

[6:20]

[21:100]

[101:1000]

Br ME T 00 9 Br ME T 01 0 Br ME T 01 8 Br ME T 02 7 Br ME T 02 8 GBM056 GBM062 GBM063 GBM064 GBM065Re GBM074

Hyperexpanded (50 < X <= 500)

Large (20 < X <= 50)

Medium (5 < X <= 20)

Small (1 < X <= 5)

Single (0 < X <= 1)

Figure 3-4: T cell clonal expansion among GBM & BrMETs. a, Representation of the proportion of total T cell repertoire space

occupied by T cell clones with a given clonal index in each sample. b, UMAP representation of T cells with clonal frequently overlaid.

UMAP2

UMAP1

3.2.2 Genomic and immunological profiling of a hypermutated recurrent GBM

Owing to a unique immunogenomic profile and growing literature studying the genomics

of hypermutated GBM, we devote the following section to an in-depth analysis of a specific tumor

with this phenotype. GBM065.Re presented with tumor progression of an IDH1 mutant anaplastic

astrocytoma 5 years after initial resection. In the interim, the patient was treated with four cycles

of vincristine, CCNU, and procarbazine, proton radiation therapy, and six cycles of high-dose

temozolomide (TMZ) (Fig. 3-5a). Using DNA WES of 4 tumor regions, we observed a substantial

increase in the mutational burden across all regions of the recurrent tumor (mean 1587

variants/sector) relative to the prior primary tumor (219 variants) (Fig. 3-5b). Variant analysis

revealed that two regions from the recurrent tumor contained the MSH6 T1219I variant previously

identified in Lynch syndrome and known to act in a dominant-negative manner247,248 (Fig. 3-6a).

The other two regions both contained unique MSH6 mutations not previously reported (G1148S

& G1116D) but computationally predicted as “likely to impair molecular function” by a previously

published tool for the prediction of MSH6 variant significance249. Additional mutations in DNA

mismatch repair genes such as MLH3 and POLD3 were detected in some but not all regions of the

recurrent tumor. This heterogeneity of potentially pathogenic mismatch repair defects in

hypermutated recurrent GBM was observed in one prior patient and suggests either the emergence

of multiple unique routes to hypermutation occurring within the same tumor or a common

alternative mechanism unrelated to these genes219. An analysis of the mutational signatures within

the recurrent tumor revealed a significant enrichment for signature 11, known to be associated

with prior treatment with TMZ213,214 and previously reported to be enriched in hypermutated

recurrent gliomas218.

FLAIRT1- po st

GB M065 P rim ar y

Anaplastic Ast ro cytoma

GB M065.Re

4X PCV

6X TMZ

2013 2018

TRBV14

TRGV10

TRBV12−4

TRAV5

KLRC3

ANXA1

HL A−DPA1

IFITM2 HL A−DRB1

CD7 4

GZ M K

TRAV1 4 DV4

GPR 18 3 KLRC4

TRBC1

CPNE2

SLAMF1

−2−1012

lo g (Fold Cha nge)

−log

10(P−Val ue)

50 0

10 00

15 00

GBM065.Prima ry

GBM065.Re 1

GBM065.Re 2

GBM065.Re 3

GBM065.Re 4

Variant Count

a b

15 30 25131

441

1208

1273

1593

1178

198

500

1000

1500

Variant Intersections

GBM065.P

rimar y

GBM065.Re−4

GBM065.Re−3

GBM065.Re−2

GBM065.Re−1

500

1000

1500

Varia nts p er Sampl e

Fig. 3-5: Immunogenomic profile of hypermutated recurrent GBM. a, Description of clinical course.

RT: radiation therapy. PCV: procarbazine, CCNU, and vincristine. TMZ: temozolomide. b, Total variants

identified from whole-exome sequencing for primary tumor and each analyzed region of recurrence.

c, UpSet plot of the distribution of all identified variants grouped by the set of tumor samples in which a variant

is shared. d, Alluvial plot displaying the frequency of the five most expanded intratumoral TCR β-chain

sequences from either region of recurrent tumor across peripheral blood and tumor samples. e, Uniform

manifold approximation and projection (UMAP) dimensionality reduction of the scRNA-seq data of T cells

from GBM065.Re. The dashed outline delineates the general population each cell belongs to and the color

indicates the degree of expansion for that T cell clone (defined by alpha/beta pair). f, Volcano plot of

differentially expressed genes within hyperexpanded clones relative to rest of T cells.

e f

−2.5

0.0

2.5

5.0

−5.0 −2.5 0.0 2. 5 5. 0

UMAP 1

Hyper

expanded (50 < X )

Large (20 < X <= 50)

Medium (5 < X <= 20)

Sm all ( 1 < X <= 5)

Sin gle ( 0 < X <= 1)

Cytotoxic C D8+ T cells

CD4+ re gul atory-li ke T cel ls

Effe ctor Me mory

T ce lls

Naive/Central

Memory T cells

UMAP 2

Clonal

Subclonal Shared

Subclonal Private

Clono type

CASS AYS GTGAFF TR BV 6−6

CASSDGTVKSYEQYV TRBV2−1

CASSLGL AGHEQFF TRBV 12−1

CASSQRSGELFF TRBV12−3/12−4

CASSYQSGTQHF TRBV12−3/12−4

CATR RTSGGFNEQFF TRB V24−1

0.0

0.2

0.4

0.6

Cl ona l Propo rtion

PBMC Re g. 2

Re g. 1

cluster

1 2 3 4 5 6

GBM065.Primary

1 2 3 4 5 6

GBM065.Re 1

1 2 3 4 5 6

GBM065.Re 2

1 2 3 4 5 6

GBM065.Re 3

1 2 3 4 5 6

Cluster

GBM065.Re 4

GBM065−Pri GBM065.Re−1 GBM065.Re−2 GBM065.Re−3 GBM065.Re−4

00.25 0.50 0. 75 1. 00 0. 25 0. 50 0. 75 1.00 0.25 0. 50 0. 75 1. 00 0.25 0. 50 0. 75 1. 00 00.25 0.50 0.75

0.25

0.50

0.75

0.25

0.50

0.75

1.00

0.25

0.50

0.75

1.00

0.25

0.50

0.75

1.00

0.25

0.50

0.75

1.00

Variant

Other

TP53 R280G

TP53 T125T

IDH1 R132H

MSH6 T12 19I

00 0

Fig. 3-6: Clonal landscape of GBM065. a, Pairwise VAF comparisons for GBM065 primary and recurrent

samples. b, Clonal architecture of GBM065 primary and recurrent tumor samples generated by ClonEvol.

Overall clonal composition of each sample is represented by fraction of shaded spheres

We observed that the vast majority of variants identified within each region of the recurrent

tumor were subclonal private and not spatially distributed (Fig. 3-5c & Fig. 3-6b). A small fraction

(30 total) of variants were shared between all regions of the recurrent tumor, and an even smaller

number (15 total) were shared by the primary and all sectors of the recurrence. Importantly, these

included likely drivers of the tumor such as TP53 and IDH1 alterations. This remarkable

heterogeneity is consistent with a prior report in which two regions of a hypermutated recurrent

tumor were sequenced and shared less than 2% of all identified mutations219. In contrast to the

variant and neoantigen heterogeneity within this tumor, the TCR Vb repertoires within two

analyzed regions were more similar (MOI = 0.85) and had the highest clonality across all samples

in the cohort. Remarkably, the top three T cell clonotypes within GBM065.Re made up more than

37% of the intratumoral repertoire, suggesting a significant degree of clonal expansion.

Additionally, these dominant clones were all present at substantially reduced frequencies (<1%)

within the patient’s peripheral blood, confirming the specific intratumoral expansion of these cells

(Fig. 3-5d).

Finally, to more deeply characterize the immunological landscape of GBM065.Re, we

performed single-cell RNA sequencing (scRNA-seq) with TCR enrichment on sorted CD45+

immune cells (n=1,728) isolated from this tumor immediately following surgical resection. We

found that a majority of the cells were of the lymphoid lineage and were predominantly cytotoxic

CD8 T cells (59% of sequenced cells) with smaller populations of naïve and CD4 regulatory-like

T cells (Fig. 3-5e). The highly expanded T cell clones identified through bulk TCR Vb sequencing

also were represented in the scRNA-seq data but were found dispersed throughout the CD8 T cell

clusters. We performed differential expression analysis on the hyperexpanded clonotypes and

found them enriched for markers of activation such as KLRC3, KLRC4, and GZMK together with

MHC Class II genes classically upregulated by activated T cells (Fig. 3-5f). Thus, despite

substantial heterogeneity at the variant and neoantigen level, the hypermutated GBM065.Re

contains clonally expanded T cells with an activated phenotype distributed throughout the tumor.

3.3 Discussion, Future Directions, and Conclusions

Owing to significantly different responses to ICB therapy and shared location within a

unique anatomical site, several recent comparative studies have provided insight into the

immunological microenvironment of GBM and BrMETs229,230. However, these initial studies have

been primarily focused upon dissecting the tumor myeloid population that makes up a majority of

intratumoral immune cells. Herein, we provide a scRNA-seq-based characterization of tumor

infiltrating T cells in a cohort of 11 malignant brain tumors. To our knowledge, this is the largest

comparative study of the lymphoid compartment between GBM & BrMETs and uncovered tumor

type specific differences.

Our results suggest that BrMETs and GBM harbor significantly different intratumoral T

cell infiltrates with BrMETs containing clonally expanded CD8 T cells characterized by

expression of canonical exhaustion markers. In contrast, most GBM tumors do not contain

significantly clonally expanded CD8 T cells in bulk but rather harbor regions of local expansion

as discussed in chapter 2. Furthermore, the moderately expanded clones in GBM appear to adopt

a terminally differentiated phenotype characterized by expression of KLRG1 but do not express

the same levels of inhibitory receptors observed in clonally expanded CD8 T cells in BrMETs.

The expression of KLRG1, canonically considered a marker of T cell senescence, has also been

demonstrated to restrict antitumor T cell function245,246,250. It is intriguing to consider whether this

distinction is a factor in the differential responses to ICB therapy, as the CD8 T cells in GBM in

our data do not develop into classically exhausted CD8 T cells with high expression of multiple

inhibitory checkpoints. This would suggest that the PD-1/PD-L1 axis is not always a prominent

aspect of immune escape in GBM. The factors responsible for this dichotomy between canonically

exhausted CD8’s in BrMETs and KLRG1+ terminally differentiated CD8’s in GBM necessitate

further study.

Another significant feature of the immune microenvironment within BrMET tumors

relative to GBM is the existence of moderately clonally expanded CD4 T cells. These cells appear

to have an activated phenotype characterized by expression of OX40 and relatively high levels of

IFN-g in addition to numerous inhibitory checkpoints such as PD-1, TIGIT, and TIM-3. The exact

role of CD4 T cells in mediating anti-tumor immune responses is an area of active research, but

numerous studies have demonstrated their essential roles in the development of anti-tumor CD8 T

cells and maintenance of a pro-inflammatory tumor microenvironment251–254. The relative absence

of activated CD4 T cells within GBM coupled to the generalized CD4 lymphopenia observed in

these patients suggests that CD4 T cell dysfunction is a prominent mode of immune escape in

GBM183–185.

The hypermutated recurrent tumor GBM065.Re stood in stark contrast to the relative lack

of clonal expansion and CD8 T cell activation in most GBM tumors. Tumor mutational burden

has been shown to correlate with checkpoint blockade response in some studies107,175, and

hypermutated tumors resulting from germline or acquired DNA repair deficiencies have been

shown to be uniquely responsive to immunotherapy across other tumor types237. Within

hypermutated GBM, which occurs in up to 20% of recurrent disease, the efficacy of checkpoint

blockade and other immunotherapies remains an open question. Several reports have observed

clinical responses from checkpoint blockade in patients with hypermutant GBM harboring

germline DNA repair defects, but a recent retrospective analysis found no benefit in mismatch

repair deficient GBM patients treated with PD-1 blockade166,167,194. Ongoing clinical trials are

designed to address this question, such as Alliance study A071702/NCT04145115 [A Study

Testing the Effect of Immunotherapy (Ipilimumab and Nivolumab) in Patients With Recurrent

Glioblastoma With Elevated Mutational Burden]. Although characterization of more patients is

needed to generalize our findings, the immunogenomic profiling of GBM065.Re revealed

potentially unique features of these tumors. First, an overwhelming majority of mutations and

associated neoantigens within this hypermutant tumor were subclonal private. These data, together

with our other findings and published work207, indicate that single-site profiling analysis would be

inadequate and dramatically underestimate both tumor complexity and neoantigen burden.

Second, despite this dramatic heterogeneity of variants, we observed remarkably similar TCR

repertoires with highly expanded CD8+ T cell clones within different regions. scRNA-seq analysis

on these expanded clones shows evidence of T cell activation suggesting potential tumor

reactivity. Whether these clones react to the small subset of clonal neoantigens, the thousands of

regional neoantigens, overexpressed CT antigens, or are responding in a non-specific way to

inflammatory stimuli will require further analysis. However, our data indicates that hypermutant

tumors can contain highly activated and clonally expanded T cells while also representing an

extreme of mutational and neoantigen heterogeneity. Ongoing work is directed at understanding

how the immune system may direct responses to antigen targets in the context of this

heterogeneity.

CHAPTER FOUR

Approaches to mouse and human neoantigen identification

4.1 Introduction

The foundational work of Thierry Boon, Lloyd Old, Steve Rosenberg and many others led

to the identification of the first mouse and human tumor antigens, launching a flurry of activity

focused on determining the antigenic basis for anti-tumor immunity73,80. These initial studies were

slow, arduous, and difficult to replicate across a variety of tumor types and under various treatment

conditions. Since that time, the advent of high-throughput next generation sequencing has

drastically changed the field, allowing relatively rapid identification of a complete set of putative

neoantigens and expressed CT antigens from patient tumors. This has led to a collection of studies

that have defined tumor neoantigens recognized by infiltrating T cells in the settings of ACT and

ICB therapy135,137. However, an equally critical avenue of research has been the characterization

of endogenous neoantigen reactivity in a wide array of human tumors, including those resistant to

conventional immunotherapy.

The earliest reports that comprehensively profiled neoantigen reactivity arose shortly after

the approval of ICB therapy in the early 2010’s. The underlying methodology behind most of these

studies was the screening of expanded TIL against libraries of candidate peptides or tandem

minigenes containing tumor mutations that were introduced to either patient peripheral blood or

artificially engineered APCs. The ultimate functional read-out was generally cytokine production

from the patient TIL. Initial studies were predominantly focused upon immunotherapy-responsive

tumors such as metastatic melanoma and NSCLC and demonstrated the presence of endogenous

CD4 and CD8 neoantigen-specific responses106,137,138,255,256. However, more recent work has

expanded these findings to lower-mutational burden tumors and identified neoantigen-specific

CD8 responses across a host of tumor types including ovarian, bladder, and pancreatic cancer257–

260. Intriguingly, a recent study even reported high rates of neoantigen-specific responses in

pediatric acute lymphoblastic leukemia, suggestive of the potential for reactivity in very low

mutational burden tumors261. However, despite immense interest in the development of

immunotherapy for the treatment of GBM, the antigen specificity of the endogenous anti-tumor T

cell response has not been characterized. A deeper understanding of the nature and breadth of the

antigens recognized by the TIL in GBM could aid in the development of novel therapies.

In the past decade, there has been immense interest in the development of unbiased high-

throughput screening systems for the identification of tumor antigens. While recent technological

advancements have allowed for broad high-throughput screens of the antigens recognized by the

humoral immune response in various settings, the correspondingly higher affinity of the

antibody:antigen interaction relative to the TCR:pMHC interaction does not allow these

techniques to be easily generalized262. Meanwhile, the widespread adoption of scRNA-seq has

provided previously unparalleled insight into the tumor-infiltrating T cell compartment across a

variety of malignancies, including the capacity to obtain both expression signatures and paired

TCR a/b chains from individual T cells. Several groups have utilized scRNA-seq for tumor

antigen detection by introducing intratumoral TCRs into donor T cells and subsequent co-culture

with matched peripheral blood loaded with peptides or tandem minigenes263,264. At the same time,

numerous groups have reported novel approaches to identify TCR:pMHC pairs without extensive

reliance on patient-derived reagents. For example, Christopher Garcia’s lab developed a yeast-

display system utilizing a diverse peptide library expressed on yeast subjected to an affinity-based

selection using bead-multimerized TCRs265. Furthermore, several recent studies from David

Baltimore’s group have developed unique systems based upon trogocytosis or the generation of

APCs containing fusion proteins of peptide:HLA’s linked to signaling domains266,267. However,

all these approaches are not easily generalizable to additional HLA alleles beyond HLA-A*02:01

or to collections of TCRs isolated from tumors with no known MHC restriction. Thus, there exists

a need for approaches to screen TCRs isolated from intratumoral T cells against candidate antigens

in a way that minimizes reliance on patient-derived reagents but is properly scalable beyond model

systems.

Finally, the identification of tumor antigens in preclinical murine systems allows for more

deeply mechanistic studies on the role of neoantigen-specific responses, either endogenously or in

response to targeted therapies such as vaccination or cellular therapy. Previous work in our lab

identified the Imp3D81N (mImp3) and Odc1Q129L (mOdc1) point mutations as endogenously

recognized class I neoantigens within the murine glioma models GL261 and SMA-560,

respectively268. Additional work from our lab demonstrated that the Epb4H471L, Pomgnt1R497L, and

Plin2G332R mutations are endogenously recognized class I neoantigens in the more aggressive and

infiltrative CT2A glioma model269. Furthermore, we showed that neoantigen vaccine against these

epitopes synergized with ICB therapy and led to tumor regression in some mice, displaying the

potential utility of neoantigen identification in murine and ultimately human systems. However,

we have yet to identify tumor antigens recognized by CD4 T cells, despite clear evidence of their

necessity in mediating anti-tumor immune responses (unpublished data). Furthermore, we are

interested in the identification of additional class I restricted neoantigens to more deeply

characterize these workhorse mouse models.

Thus, we sought to develop flexible systems to probe for antigen reactivity among the TIL

in both human patients and mouse preclinical systems through the extraction of TCR a/b chains

from scRNA-seq data and incorporation into reporter systems. These approaches enable screening

for tumor antigens in both human and mouse tumors in a flexible manner that minimizes the need

for patient-derived reagents.

4.2 Results

4.2.1 Development of system for human tumor antigen identification

With the ultimate goal of identifying tumor antigens, we developed a protocol for the

expansion of tumor-infiltrating lymphocytes from malignant brain tumors. In this process, brain

tumors were minced into several millimeter sections and cultured in media containing high-dose

IL-2. Following a week of expansion in this media, myelin and acellular debris was removed prior

to lymphocyte expansion in media containing high-dose IL-2 and aCD3/28 beads for a period of

3-5 weeks. The period of expansion necessary was determined by serial assessments of each TIL

culture via flow cytometry. For a majority of tumors, we were able to generate expanded TIL

cultures consisting of almost exclusively T cells with variable ratios of CD8/CD4 T cells (Fig. 4-

1a). This TIL expansion protocol was performed for all 30 tumors in the previously discussed

cohort.

In order to characterize the expanded TIL, we performed TCR sequencing on DNA isolated

from a select group of TIL cultures and matched patient tumor. As mentioned in a prior chapter,

this data allows us to use the Vb chain CDR3 as a barcode for specific T cell clones to assess their

maintenance over the expansion process. Strikingly, we observed a significant difference with

most of the enriched intratumoral T cell clones either not present in the expanded TIL or present

at substantially decreased frequencies (Fig. 4-1b). This effect was seen in TIL cultures derived

from either GBM or BrMET tumors and was similarly observed in TIL grown with just high-dose

Figure 4-1: Characterization of expanded brain tumor TIL cultures. a, Representative flow cytometry plots

of 3-5 week expanded TIL cultues derived from a GBM (top) or BrMET patient (bottom). b, Frequency

comparison of the top 5 most expanded clones in matched tumor and expanded TIL cultures in BrMET (left)

and GBM (right). Each color indicates a specific patient. Significance calculated by paired t-test.

CD 4

CD 8

CD 4

CD 8

Tumor

Expanded TIL

0.00001

0.0001

0.001

0.01

0.1

Frequ ency

Brain Metastase s

** *

Tumor

Expanded TIL

0.00001

0.0001

0.001

0.01

0.1

Frequ ency

GBM

GBM047.Re

TIL

BrMET009

TIL

IL-2 (data not shown). Meanwhile, the most expanded clones within the TIL cultures were often

present at significantly lower frequencies in the matched tumor. This significant clonal drift upon

expansion suggested that the use of the TIL cultures for antigen identification would be flawed

and potentially lead to false negatives and false positives. Thus, we sought to develop an

alternative approach to tumor antigen identification in this cohort of patients.

As described in the previous chapter, we accumulated extensive scRNA-seq data on a

subset of these GBM and BrMET tumors. In addition to providing insight on the expression profile

of TIL in these two tumor types, this data also allowed for the isolation of specific TCR a/b pairs.

Across both tumor types, we observed significantly expanded TCRs in some tumors that were

predominantly belonging to CD8 T cells (Fig. 4-2a). Presumably, these clones would be the ideal

candidates to screen for the identification of tumor antigens owing to their intratumoral

enrichment. Therefore, we aimed to generate a system through which these expanded T cell clones

could be screened for antigen reactivity in a TCR-specific manner.

A predominant focus in developing this system was to minimize the need for patient-

derived reagents to allow for the flexible application of this approach across patient samples.

Accordingly, we chose to engineer several immortalized cell lines including K562’s and a TCR-

deficient clone of Jurkat cells for this approach (Fig. 4-2b). Owing to their lack of endogenous

MHC class I expression and rapid proliferative capacity, K562’s have been previously engineered

to express HLA-A*02:01 to function as an artificial APC (aAPC) for the stimulation of CD8 T

cells270,271. We significantly broadened this approach by generating a panel of K562 aAPCs

expressing 30 single HLA-A/B/C molecules (Fig. 4-3a). Furthermore, the K562’s were made to

express CD80 to facilitate their ability to stimulate target cells. As an effector cell, we obtained

the TCR-deficient clone of Jurkat cells from Paul Thomas’s lab that had been engineered to

Figure 4-2: Outline of human neoantigen screening system. a, Tables describing the top 5 most clonally

expanded CD8 TCRs within a representative BrMET (top) and GBM (bottom) tumor. b, Outline of neoantigen

screening system generated for TCR antigen identification.

BrMET009

Clone

Fr equency

Alpha Chain

Alpha CD R3

Beta Chain

Beta CD R3

23.6%

TRAV20/TRAJ21/ TRA C

CAVLLYNFNKFYF

TRBV6-2/TRBJ2-

5/ TRB C2

CASRHVQETQYF

8.2%

TRAV19/TRAJ45/ TRA C

CALSEWPGGGADGLTF

TRBV30/TRBD1/TRBJ2-

5/ TRB C2

CAWSRGQEGETQYF

4.4%

TRAV20/TRAJ17/ TRA C

CAVQAVAAG NK LTF

TRBV30/TRBJ1-

2/ TRB C1

CAWSVYYGYTF

4.2%

TRAV29/TRAJ53/ TRA C

CAVSGGSNYKLTF

TRBV9/TRBD2/TRBJ1-

5/ TRB C1

CASSAWMGNQPQHF

2.5%

TRAV29/TRAJ56/ TRA C

CAASGRIGANSKLTF

TRBV5-1/ T RB D2 /TRBJ2-

7/ TRB C2

CASSSAWRTYEQ YF

GBM074

Clone

Fr equency

Alpha Chain

Alpha CD R3

Beta Chain

Beta CD R3

2.8%

TRAV17/TRAJ17/ TRA C

CATDAGIKAAG NK LTF

TRBV6-5/ T RB D1 /TRBJ2-

4/ TRB C2

CASSYSRTGGWDIQYF

2.2%

TRAV3/TRAJ9/TRAC

CAVRDEDTGGFKTIF

TRBV29-1/ TRB D2 /TRBJ2-

2/ TRB C2

CSVGEGGWELFF

1.9%

TRAV24/TRAJ4/TRAC

CAFREGGYNKLI F

TRBV11-1/ TRB D1 /TRBJ1-

2/ TRB C1

CASSPGFNYGYTF

1.7%

TRAV12-1/ TRAJ35/TRAC

CVVNGGFGNVLHC

TRBV6-1/ T RB D2 /TRBJ2-

1/ TRB C2

CASSVRTSGLNE QFF

1.5%

TRAV21/TRAJ27/ TRA C

CAVLNAGKSTF

TRBV6-5/ T RB D1 /TRBJ1-

2/ TRB C1

CASRGAGTNYGYTF

Figure 4-3: Generation and validation of human neoantigen screening system. a, Representative flow

cytometry plots of the generation of the K562:CD80:HLA-A*02:01 cell line. b, Representative flow cytometry plots

of the generation of the Jurkat:Nur77-GFP:MART1-TCR reporter cell line. c, Jurkat:Nur77-GFP:MART1-TCR

reporter cells were co-cultured with K562:CD80:HLA-A*02:01 cells loaded with either irrelevant peptide or the

MART1 epitope for 24 hours prior to flow analysis for GFP & CD69 upregulation. d, Dose-response titration of

GFP and CD69 upregulation in response to MART1 epitope.

0.001 0.01 0.1 110 100

Co ncent ration Peptide (uM )

% GFP+

GFP Response Titration

Irrelevant peptideMART1

0.001 0.01 0.1 110 100

Concentration Peptide (uM)

% CD69+

CD69 Response Titration

K562

K562:CD80:

HLA-A*02:01

Jurk at :Nur7 7-GFP

Jurk at :Nur7 7-GFP:

MART1-TCR

MART1 peptide

Irrelevant peptide

express GFP under control of the Nur77 promoter, one of the initial genes downstream of TCR

stimulation. To facilitate our antigen screens, we engineered these Jurkat cells to express either

CD8 or CD4 and patient-specific TCRs.

To validate the functionality of our system, we transduced the Jurkat:Nur77-GFP cells with

a TCR specific for the MART1 epitope previously targeted in several adoptive transfer

studies117,223 (Fig. 4-3b). Following co-culture of these transduced cells with K562:CD80:A*02:01

cells loaded with the MART1 peptide, we observed robust antigen-specific stimulation with

concordant upregulation of GFP and the T cell activation marker CD69 (Fig. 4-3c). Furthermore,

this response displayed a clear dose-dependency with a loss of detectable signal as peptide

concentrations reached below 100 nM (Fig. 4-3d). This response was not observed if

Jurkat:Nur77-GFP cells were lacking the MART1-specific TCR or if K562 aAPCs were not

engineered to express HLA-A*02:01 (data not shown). Therefore, this system appears to function

as intended in the detection of antigen-specific responses in a TCR-specific manner. We are now

actively applying this approach for the extensive screening of expanded intratumoral TCRs against

candidate neoantigens, CT antigens, and viral antigens across this cohort of patients.

4.2.2 Development of system for mouse tumor antigen identification

While prior work from our lab identified endogenous class I neoantigens in a collection of

murine glioma models, we are interested in more comprehensively profiling the class I-restricted

neoantigen-specific responses and identifying class II-restricted neoantigens in these systems. Our

previous studies utilized IFN-g ELISPOT screens on intracranial TIL for the detection of these

responses. Since that time, we have more extensively profiled these tumors including scRNA-seq

with 5’ VDJ enrichment to provide T cell clonotype information in both untreated and ICB treated

conditions. Similarly to the human tumors, we have utilized this approach for the identification of

the most expanded intratumoral TCRs (Fig. 4-4a). Noting the high levels of clonal expansion from

CD8 T cells in both of these models, particularly after ICB therapy, we sought to develop a system

for the screening of CD8 and CD4 T cells for neoantigen reactivity in a TCR-specific manner (Fig.

4-4b). In contrast to the significant HLA polymorphism and polygenism in humans, in-bred

C57BL/6 mice only express two MHC class I molecules, Db and Kb, and one MHC class II

molecule, I-Ab. Thus, we chose immortalized mouse embryonic fibroblasts (MEFs) isolated from

a C57BL/6 mouse as a source of b haplotype immortalized antigen presenting cells for class I

antigen detection. For the detection of CD4 T cell antigens, we made use of the immortalized

C57BL/6-derived dendritic cell line JAWSII, which has previously been shown to express

significant MHC class II and the ability to stimulate CD4 T cells following LPS stimulation272.

As an effector cell, we obtained the TCR-deficient 58 hybridoma cell line from David

Kranz’s lab273. Following introduction of a TCR and antigen-specific stimulation, this cell line has

been shown to express canonical T cell-associated transcription factors such as NFAT and robustly

produce IL-2274,275. To facilitate these antigen screens, the original 58 hybridoma cell line was

engineered to express either CD4 or CD8 (Fig. 4-5a). Furthermore, to allow for flow cytometry-

based screens, we introduced an NFAT-GFP reporter construct into these cells in which expression

of GFP is controlled by an NFAT-inducible promoter (provided by Ken Murphy with human CD4

as a selection marker).

To validate the functionality of our system, we made use of the canonical model antigen

ovalbumin and its associated class I and class II TCRs. Following introduction of the OTI TCR

into the 58 reporter system, we observed robust CD69 and GFP upregulation after co-culture with

MEFs loaded with the SIINFEKL peptide (Fig. 4-5b). This response is paralleled by significant

Figure 4-4: Outline of murine neoantigen screening system. a, Tables describing the top 5 most clonally

expanded CD8 TCRs within two murine tumor models with and without anti-PD-L1 therapy. b, Outline of

neoantigen screening system generated for TCR identification.

GL261

Clone

Frequency

Alpha Chain

Beta Chain

2.7%

TRAV6 -

6/TRAJ43/T RAC

TRBV26/TRBJ2 -7/ T RBC1

2.0%

TRAV7D -

2/TRAJ7/T RAC

TRBV16/TRBD2 /TRBJ2 -

7/T RBC1

1.3%

TRAV13 -

1/TRAJ39/T RAC

TRBV20/TRBJ1 -1/ T RBC1

1.2%

TRAV12D -

2/TRAJ16/T RAC

TRBV20/TRBJ1 -1/ T RBC1

1.1%

TRAV7 -

4/TRAJ4/T RAC

TRBV5/TRBD1/TR BJ2-

1/T RBC1

GL261+anti-PD-L1

Clone

Frequency

Alpha Chain

Beta Chain

18.4%

TRAV14 -

3/TRAJ45/T RAC

TRBV12 -1/T RBD2 /T RBJ2 -

3/T RBC2

5.8%

TRAV12 -

2/TRAJ57/T RAC

TRBV13 -2/T RBD1 /T RBJ2 -

4/T RBC1

2.8%

TRAV12N -

1/TRAJ48/T RAC

TRBV12 -2/T RBD1 /T RBJ2 -

5/T RBC1

1.9%

TRAV8N -

2/TRAJ5/T RAC

TRBV14/TRBD2 /TRBJ1 -

1/T RBC1

1.6%

TRAV14 -

3/TRAJ37/T RAC

TRBV2/TRBJ1 -1/ T RBC1

CT2A

Clone

Frequency

Alpha Chain

Beta Chain

5.5%

TRAV12D -

2/TRAJ17/T RAC

TRBV13 -1/T RBD2 /T R BJ2 -

4/T RBC1

5.3%

TRAV10N/TRAJ 7/

TRAC

TRBV13 -1/T RBD2 /T R BJ2 -

4/T RBC1

3.5%

TRAV14D -

2/TRAJ40/T RAC

TRBV13 -2/T RBD2 /T R BJ2 -

7/T RBC1

2.5%

TRAV5 -

4/TRAJ49/T RAC

TRBV1/TRBD1/TR BJ1-

5/T RBC1

2.3%

TRAV8N -

2/TRAJ26/T RAC

TRBV4/TRBD2/TR BJ2-

1/T RBC1

CT2A+anti-PD-L1

Clone

Frequency

Alpha Chain

Beta Chain

19 .6%

TRAV12 -

1/TRAJ26/T RAC

TRBV16/TRBD2 /TRBJ2 -

5/T RBC1

4.4%

TRAV7D -2/TRAJ7

/TRAC

TRBV5/TRBD1/TR BJ2-

2/T RBC1

2.6%

TRAV16N/TRAJ 12

/TRAC

TRBV13 -1/T RBD2 /T RBJ2 -

3/T RBC1

2.1%

TRAV10N/TR AJ7

/TRAC

TRBV5/TRBD1/TR BJ1-

1/T RBC1

1.9%

TRAV13 -1/TRAJ7

/TRAC

TRBV20/TRBD2 /TRBJ2 -

7/T RBC1

Figure 4-5: Validation of murine antigen screening system. a, Schematic depicting generation of 58

hybridoma reporter lines expressing mouse CD4/CD8 and NFAT-GFP reporter construct. b, Co-culture of

58:CD8:NFAT-GFP:OTI cells co-cultured with MEFs loaded with either an irrelevant peptide or SIINFEKL. Flow

cytometry analysis performed after 24 hours of stimulation. c, IL-2 ELISPOT of 58:CD8:NFAT-GFP:OTII cells

co-cultured with JAWSII cells loaded with either the minimal OTII epitope or a longer frament following overnight

LPS stimulation of JAWSII cells.

Irrelevant peptide SIINFEKL

58:CD8:NFAT-GFP:OTI 58:CD8:NFAT-GFP:OTII

IL-2 production detected by an IL-2 ELISPOT (data not shown). To assess the ability of the system

to identify class II-restricted antigens, we introduced the OTII TCR into the 58 reporter cell line

and observed significant IL-2 production following co-culture with stimulated JAWSII cells

loaded with the ovalbumin class II epitope (Fig. 4-5c). We are now actively applying this approach

using the 58:CD8 and 58:CD4 reporters for the extensive screening of expanded intratumoral

TCRs against candidate neoantigens in our murine preclinical models.

4.3 Discussion, Future Directions and Conclusions

The identification of the antigens driving anti-tumor immune responses has been a core

facet of cancer immunology since its inception. While the earliest tumor antigens were identified

through laborious and strenuous protocols, significant advances in both immunology and

genomics have made the process more rapid and high-throughput. Despite this, the antigen

specificity of most intratumoral T cell clones in patient tumors remains undefined, and the extent

of tumor antigen-reactivity within the TIL of many cancer types is unknown. With the growing

interest in the targeting of neoantigens as a form of personalized immunotherapy, the

characterization of both endogenous and treatment-induced neoantigen-specific responses will

only grow in importance.

The last decade has seen an explosion of new techniques for T cell antigen discovery, many

of them utilizing advances in next generation sequencing technologies to facilitate high-throughput

screens. Unfortunately, most of these approaches are validated in idealized model systems and

difficult to generalize to the broad scale needed for the diversity of TCR candidates and HLA

alleles in patient samples. Therefore, we sought to generate flexible systems to screen for mouse

and human tumor antigen reactivity in a TCR-specific manner. Ultimately, we hope this approach

utilizing TCRs identified through scRNA-seq will enable us to pair T cell reactivity to expression

profile, defining the populations enriched for tumor-specificity. However, it is likely that

significant evolution of the antigen screening systems will be needed to more completely define

the landscape of T cell reactivity. For now, we have been unsuccessful in generating a tandem

minigene-based approach that would enable us to migrate away from large scale peptide screens.

This is likely due to the lower sensitivity of the Jurkat-based reporter, which suggests that

introduction of TCRs into matched PBMC might yield a more sensitive effector readout. Our

reliance on peptide screens in turn necessitates a reliance on neoantigen prediction pipelines.

While these approaches are continuously improving, their predictions for class II antigens and rare

HLA’s are considerably less accurate276,277. Furthermore, the detection of reactivity to self-

antigens is significantly hindered by a peptide-based approach, as one is restricted to defined

epitopes as a manner of practicality. Thus, while the described system can perform peptide-based

screens with a high degree of flexibility for patient or mouse tumors, the development of more

generalized and less biased screening approaches should be a top priority.

One indirect approach to the targeting of neoantigens that began with Steve Rosenberg’s

group in the 1980’s is the significant ex vivo expansion of patient TIL prior to subsequent re-

infusion. These same expanded TIL cultures have been shown to have robust neoantigen reactivity

and anti-tumor efficacy in a sizeable portion of patients with metastatic melanoma and some

epithelial malignancies114,137,278. We sought to follow a similar protocol for the subsequent

expansion and screening of brain tumor TIL within our cohort of patients. However, we observed

significant clonal drift with a majority of expanded intratumoral TCRs present at much lower

frequencies in the TIL cultures. The inability of the most expanded intratumoral T cell clones to

proliferate following stimulation is somewhat consistent with the phenotype observed on scRNA-

seq in which the dominant clones in GBM or BrMET tumors were either terminally differentiated

or significantly exhausted. However, this points to a significant problem in attempting to pursue

adoptive TIL therapy in patients with GBM. Numerous studies have hinted that clonally expanded

T cells within the tumor are enriched for tumor antigen reactivity279,280. As these clones are

specifically lost following ex vivo expansion, the resulting TIL cultures may have lost any pre-

existing tumor reactivity. Any attempts at pursuing adoptive TIL therapy in malignant brain tumors

must wrestle with this issue and develop protocols to maintain tumor clonal architecture such as

modified stimulation conditions or treatment with genetic or epigenetic modifiers to reverse

tumor-induced senescence. These difficulties in the expansion of potentially tumor-specific T cells

only emphasizes the need for linking tumor antigen specificity to expression profiles within TIL

samples for the potential identification of TCRs to be used in ACT.

Driven by the immense success of cancer immunotherapy in certain tumor types, large-

scale analyses of endogenous neoantigen-specific T cell responses have now been performed in a

number of patients across numerous malignancies135,137,138,255,258,260. These studies clearly describe

T cell immune responses against both class I and class II-restricted neoantigens in a variety of

tumor types. However, they also all demonstrate that only a very small fraction of potential

neoantigen targets produce detectable reactivity among tumor-infiltrating T cells. Why is the

fraction of potential targets eliciting a response so low and what does this mean for neoantigen

targeted therapy in tumors with a lower mutational burden? One possibility is that a majority of

immunogenic candidates have been edited out in the course of tumor formation, leaving only

neoantigens that drive relatively poor immune responses. Therefore, a high mutational burden, as

is frequently seen in metastatic melanoma, may be needed to generate enough neoantigen

candidates to produce a strong anti-tumor response. However, recent data showing frequent

neoantigen-specific responses in pediatric patients with acute lymphoblastic leukemia, a very low

mutational burden tumor, suggests that a high neoantigen load is not necessary to induce anti-

tumor immunity261. Another possibility is the concept of immunodominance. This phenomenon,

well-described in viral immune responses, occurs when the immune system becomes fixated in its

response to a particular antigenic epitopes at the expense of alternative targets281. Previous

preclinical studies have demonstrated its relevance in the response to murine sarcoma antigens282.

Finally, it is possible that the endogenous anti-tumor T cell response is more diverse than

previously appreciated but simply produces responses not detected through conventional

approaches. Regardless, the development of personalized immunotherapy against tumor-specific

neoantigens will necessitate a deeper understanding of the endogenous anti-tumor T cell response

to aid in the development of neoantigen vaccine or adoptive cellular therapy approaches. While

the presence of neoantigen-specific TCRs within GBM has not been definitely proven, studies in

preclinical mouse models and other immunologically suppressive human tumors hints at their

existence. The remainder of this work will be focused upon how the identification of neoantigen-

specific TCRs can be leveraged for both mechanistic and therapeutic studies.

CHAPTER FIVE

Generation of a neoantigen-specific TCR transgenic

5.1 Introduction

One of the initial focuses of our lab was the immunogenomic profiling of a collection of

murine glioma lines (GL261, CT2A, SMA-560) to facilitate deeper analyses of the mechanisms

underlying neoantigen-specific responses and how to leverage these responses therapeutically. By

screening TIL in tumor-bearing mice for IFN-g production in response to candidate neoantigens,

we identified the endogenous neoantigens mImp3, mOdc1, and mEpb4 in GL261, SMA-560, and

CT2A, respectively268,269. Recent studies from our lab demonstrated that neoantigen vaccination

against the mImp3 epitope led to a survival benefit in GL261 and that a polyvalent neoantigen

vaccine combined with ICB therapy produced cures in a majority of CT2A-bearing mice269.

However, vaccine approaches do not allow in vivo tracking of adoptively transferred cells and

necessitate very early treatment (day 3 post-tumor implantation) to achieve an effect. An

alternative approach for neoantigen targeting is the adoptive transfer of neoantigen-specific T cells

derived either from bulk TIL cultures or genetic modification of infused cells. For that purpose,

we sought to isolate a mImp3-specific TCR and generate a preclinical system for studying

neoantigen-specific adoptive cell therapy.

The identification of paired TCR a/b chains has historically necessitated the laborious

process of generating T cell clones. However, within the last decade, numerous methods have been

reported for the cloning of antigen-specific TCRs without the establishment of T cell clones. Some

of these approaches utilize deep sequencing technologies and attempt to identify TCR a/b pairs

through statistical analysis or DNA barcoding283–285. Lately, the widespread use of scRNA-seq

technology has provided the ability to simultaneously isolate TCRs and gene expression profiles

from thousands of individual cells, dramatically altering this field of TCR identification286. An

alternative technique utilizes single-cell RT-PCR on sorted T cells, which allows greater flexibility

in its ability to isolate TCRs from a much smaller starting population287–290. Recent studies have

greatly streamlined this process, allowing the identification of antigen-specific TCRs in less than

one week’s time, offering a significantly faster, more flexible, and more cost-effective route to

TCR identification than NGS technologies.

Numerous approaches exist for the generation of mouse T cells with a defined antigen-

specific TCR including viral gene transduction and the generation of retrogenic or transgenic

mice. While the transduction of primary mouse T cells represents the quickest and most flexible

option, achieving a consistently high efficiency of transduction is difficult and would be

enormously laborious for large-scale projects291,292. Retrogenic mice, in which hematopoietic

stem cells are retrovirally transduced with a given TCR prior to reintroduction into otherwise

lethally irradiated hosts, allow for relatively rapid generation of mice bearing a single TCR on

peripheral T cells293,294. However, as these mice do not integrate the TCR into their germline,

they cannot pass it on to their progeny and would need to be consistently remade for use in larger

studies. Transgenic mice, in which the specific TCR is encoded by transgenes introduced into the

mouse’s genome, represent the slowest but ultimately best approach for long-term studies owing

to the ability to generate large colonies of transgenic mice bearing a single TCR.

Thus, we aimed to isolate a mImp3-specific TCR and generate a transgenic mouse

bearing this TCR. This platform will enable a wide range of studies on the role of neoantigen-

specific T cell responses in malignant brain tumors.

5.2 Results

5.2.1 Identification of neoantigen-specific TCRs

To provide cells for the isolation of a mImp3-specific TCR, three separate populations of

mImp3-specific T cells were generated. Splenocytes were taken from mice that had rejected

GL261 once following aPD-L1 therapy (1X), once following aPD-L1 therapy and an additional

two times from memory rechallenge (3X), or had been vaccinated in a prime-boost manner with

mImp3 synthetic long peptide and Poly(I:C) adjuvant (Vax). Each of these populations were

separately stimulated with low-dose mImp3 peptide and IL-2 with a weekly replacement with

fresh naïve splenocytes as APCs. Following 6-8 weeks of stimulation, single mImp3-specific CD8

T cells were tetramer sorted into PCR plates (Fig. 5-1a).

In order to isolate mImp3-specific TCRs, we adapted a previously published single-cell

PCR protocol290. In brief, a two-step nested PCR reaction was performed in which the first RT-

PCR reaction included a pool of 41 Va and 39 Vb primers specific to the leader sequences of all

possible TCR a or b chains. Subsequent PCR reactions utilized the common 5’ adapter added to

the end of the first PCR product (Fig. 5-1a). The second-step PCR products were ran on a gel and

purified prior to undergoing sanger sequencing. The benefit of this specific approach relative to

other single-cell PCR protocols is the incorporation of primers specific to leader sequences and

addition of a 5’ adapter for the second PCR reaction. This ensures that the resulting second-step

PCR product will contain the entirety of a Va or Vb gene segment from the start codon through

past the CDR3. Alternative protocols often produce TCR fragments lacking the 5’ end of the Va

or Vb gene segment, necessitating the comparison to reference sequences and potentially

ambiguous assignment of specific Va or Vb gene segments. In total, we obtained TCR a/b pairs

from a total of 42 tetramer-sorted single-cells across three distinct populations (Fig. 5-1b). All of

Figure 5-1: Identification of candidate mImp3-specific TCRs. a, Schematic depicting the isolation of

candidate mImp3-specific TCRs. Tetramer-positive T cells were single-cell sorted into PCR plates prior to

a nested multiplexed PCR reaction. b, Distribution of T cell clonotypes isolated from three separate sources

of mImp3-specific cells. c, Table describing the final TCR candidates selected for further characterization.

TCR

4.3 D − TR AV13D/TRBV13−2

4.1 B − TRAV6D−6/TRBV13 −2

TR AV6/TRBV13−2

TCR

3x.2 .5C − TRAV12D/TRBV26

3x.2 .4D − TRAV12D/TRBV26

3x.1 .1C − TRAV3D/TRBV12

3x.1 .4A − T RAV12D/TRBV26

TR AV6/TRBV13

TR AV13D/TRBV14

TR AV12D/TRBV31

TR AV7/TRBV5

TR AV6D/TRBV13

TCR

V1.1D − TRAV6/TRBV13

V1.6C − TRAV6/TRBV29

V1.3E − TRAV12D/TRBV3

TR AV12D/TRBV15

TR AV13D/TRBV13

TR AV3D/TRBV12

TR AV9D/TRBV15

1x Rej ect e d 3x Rej ect e d Vaccinated

TCR

Alpha Chain

Alpha CDR3

Beta Chain

Beta CDR3

4.1B

TRAV6D/TRAJ47

CALGAEDYANKMIF

TRBV13/TRBJ 2/TRBD1

CASGDWVGAETLYF

4.3D

TRAV13D/TRAJ 26

CALEYAQGLTF

TRBV13/TRBJ 2/TRBD2

CASGGLGDQDTQYF

V1.1D

TRAV6/TRAJ22

CVLGDSGSWQLIF

TRBV13/TRBJ 2/TRBD2

CASGDAMGGAETLYF

V1.6C

TRAV6/TRAJ22

CVLAHASSGSWQLIF

TRBV29/TRBJ 2/TRBD2

CASPTGGLAK TLYF

V1.3E

TRAV12D/TRAJ 18

CALSDRGSALG RL HF

TRBV3/TRBJ2/TRBD1

CASSLEQGGG YNYA EQFF

3x1.1C

TRAV3D/TRAJ17

CAVGGSNSAG NK LTF

TRBV12/TRBJ 1/TRBD1

CASSLEDREGS DY TF

3x1.4A

TRAV12D/TRAJ6

CALVPGGNYKPTF

TRBV26/TRBJ 2/TRBD1

CASSPDSYEQ YF

3x2.4D

TRAV12D/TRAJ6

CALIPGGNYKPTF

TRBV26/TRBJ 2/TRBD1

CASSPDSYEQ YF

3x2.5C

TRAV12D/TRAJ6

CALSEGGNYKPTF

TRBV26/TRBJ 2/TRBD1

CASSPDSYEQ YF

the expanded populations displayed the enrichment of specific T cell clones with a collection of

lower frequency clonotypes only observed once. From these isolated TCRs, we selected nine

candidates for further analysis, with specific focus given to the TCRs present at greatest frequency

in the expanded populations (Fig. 5-1c).

5.2.2 Characterization of neoantigen-specific TCRs

To characterize these nine TCR candidates, full-length TCR chains were generated by

combining sequencing results with IMGT reference sequences for a or b chain constant regions.

For each TCR, gene blocks were ordered consisting of b chain – P2A - a chain and cloned into

the pMIG vector backbone. For screening each candidate, we utilized the previously described 58

hybridoma cell line that lacks an endogenous TCR but generates robust antigen-specific responses

following introduction of a specific TCR274,275. Each TCR was retrovirally introduced into these

58 hybridoma cells, generating a library of TCR-expressing immortalized cell lines. Initially, all

nine candidate TCRs were screened for their ability to bind the mImp3 tetramer. Despite being

isolated from tetramer-sorted CD8 T cells, the V1.3E and 3x.1.4A TCRs did not bind to the mImp3

tetramer (Fig. 5-2b). However, the other seven TCRs all displayed variable levels of tetramer

binding, with the 3x.1.1C TCR demonstrating the strongest response. Notably, the 3x.1.1C TCR

was the only candidate to bind to the mImp3 tetramer in a CD8-independent manner (data not

shown).

The seven candidate TCRs with appreciable levels of tetramer binding were then assessed

for their ability to stimulate cytokine production in TCR-expressing 58 hybridoma cells. To do so,

each 58 hybridoma cell line expressing a given TCR was co-cultured with naïve splenocytes

loaded with either the mImp3 epitope or an irrelevant Db-restricted antigen (mOdc1). While all

Figure 5-2: Functional characterization of candidate mImp3-specific TCRs. a, Schematic of system for

screening TCRs by introduction into 58 hybridoma cells. b, Representative tetramer stains on 58 hybridoma

cells transduced with candidate TCRs. c, IL-2 production from TCR-expressing 58 hybridoma cells following

overnight co-culture with splenocytes loaded with 10 micromolar peptide. d, Dose-response titration of 3x.1.1C

TCR-expressing 58 cells following overnight co-culture with splenocytes loaded with indicated peptide.

c d

mImp3 Tetramer

V1.1D V1.3E 3x.1.1C

10 0

15 0

4.1B 4.3D V1.1D V1.6C 3x.1.1C 3x.2.4D 3x.2.5C

IL−2 (pg/mL)

Peptide

mImp3

mOdc1

120

110 100 1000 10000

Co ncen tratio n (nM)

IL−2 (pg/mL)

Peptide

mImp3

wtI mp3

candidates demonstrated the ability to induce IL-2 production from the 58 hybridoma cells, the

3x.1.1C TCR produced significantly higher levels than all other TCRs (Fig. 5-2c). To further

profile this TCR, we assessed its potential cross-reactivity to the wild-type Imp3 epitope. In a

similar co-culture experimental design, no wild-type reactivity was observed across a range of

concentrations (Fig. 5-2d). Meanwhile, the 3x.1.1C TCR displayed a clear dose-dependent

response to the mImp3 antigen (Fig. 5-2d). Thus, the 3x.1.1C TCR was an ideal candidate for

further study owing to clear tumor neoantigen reactivity.

5.2.3 Generation of neoantigen-specific TCR transgenic

Owing to the challenges of viral transduction in primary murine splenocytes and the

desire for a large supply of mImp3-specific T cells, we proceeded to generate a TCR transgenic

expressing the 3x.1.1C TCR. To do so, we made use of the previously validated pCD2 and p428

transgene vectors provided by the lab of Paul Allen295,296. The TCR a/b chains of the 3x.1.1C

TCR were cloned into the pCD2 and p428 vectors, respectively. Following linearization of the

transgene fragments, they were co-injected into the pronucleus of C57BL/6 zygotes in work

performed by the Transgenic, Knockout, and Micro-Injection core. An initial group of 54

C57BL/6 mice were screened by PCR on genomic tail DNA with TCR transgene-specific

primers. Unfortunately, none of these initial mice carried either transgene fragment. Owing to

increased success using mice on a mixed background, we proceeded to perform microinjections

in C57BL/6 x 129 hybrid mice. In the first batch of 19 C57BL/6 x 129 mice, we identified two

potential founders (Fig. 5-3a).

Figure 5-3: Generation of MISTIC Transgenic Mouse. a, Genotyping result confirming integration of

transgenes containing both alpha and beta chains of 3x.1.1C TCR in two mice (7.14 & 7.18). b, Flow

cytometry profiling of the peripheral blood of the transgenic founder mouse and resulstant offspring after

crossing with a CD45.1+ congenic.

!"#$%&'()*

To characterize these potential founder mice, we profiled their peripheral blood via flow

cytometry. In contrast to naïve mice, a significant majority of the T cells in these mice were CD8

T cells, all of which bound strongly to the mImp3 tetramer (Fig. 5-3b). Following crossing of these

founder mice with CD45.1+ C57BL/6 males, we confirmed the germline integration of the

transgene in resulting offspring. These transgenic mice expressing the 3x.1.1C TCR are hereafter

referred to as Mutant Imp3-Specific TransgenIC (MISTIC) mice.

5.3 Discussion, Future Directions and Conclusions

With the tremendous interest in the targeting of neoantigens for mechanistic and

therapeutic studies, the rapid identification of TCRs with defined specificity becomes critical. In

the past, the identification of a TCR against a defined antigen would usually entail the generation

of a T cell clone. However, recent approaches utilizing either deep sequencing technologies or

single-cell nested multiplexed PCR have greatly streamlined this process resulting in candidate

TCR identification in as little as one week’s time. For our studies, we adapted a previously

published nested multiplexed PCR protocol for the isolation of a set of TCR candidates specific to

the mImp3 neoantigen within GL261.

While we were ultimately successful in the identification of a set of neoantigen-specific

TCR candidates, the approach does carry some significant drawbacks. Owing to the exquisite

sensitivity of nested PCR reactions, we encountered significant challenges with contamination

upon initial optimization. Ultimately, the single-cell nested PCR protocol necessitated the use of

specialized equipment and bench space to eliminate these concerns. Furthermore, performing PCR

reactions on 96-well plates and isolating the resulting products via gel purification is both

technically challenging and laborious. Widespread utilization of scRNA-seq now enables the

identification of TCR a/b pairs for an entire population and presents an appealing alternative to

our chosen approach. However, the identification of TCRs via scRNA-seq is slower, more costly,

and requires far greater cell numbers than single-cell PCR protocols. The ideal procedure for

identifying TCRs with a defined specificity is ultimately very project-specific.

It is generally assumed that clonally expanded T cell populations within a given tissue will

be enriched for antigen-specificity due to localized antigen-driven expansion. This phenomenon

has been clearly demonstrated in both human and murine systems279,280. On the other hand, our

data set offers an intriguing counterpoint. While the most clonally expanded T cells within our

analyzed populations were all mImp3-specific, the most highly reactive TCR that was ultimately

chosen for incorporation into the MISTIC mouse was only detected once among thirteen

sequenced cells from that population. It is not clear whether this is the result of clonal drift over

the course of the antigen-specific expansion, or whether this clone was never present at a

significant frequency from the onset. Regardless, the high affinity and reactivity of the 3x.1.1C

TCR points to the complex factors governing T cell expansion both in vivo and in vitro and

cautions against over-reliance on clonotype frequency in inferring T cell functionality.

100

CHAPTER SIX

Characterization of neoantigen-specific adoptive cell therapy

6.1 Introduction

There is immense interest in adoptive T cell therapies for tumors such as GBM lacking

significant baseline intratumoral T cell infiltration. The earliest attempts at adoptive T cell

therapies were Steve Rosenberg’s landmark studies in the 1980’s which demonstrated impressive

results in the treatment of metastatic melanoma with expanded TIL112,114. Since then, his group

has broadened this treatment to a wide array of malignancies such as cholangiocarcinoma and

colon carcinoma and demonstrated that these infusion products often contain significant

neoantigen reactivity137,139,140. However, this approach necessitates the expansion of incredibly

large numbers of T cells from the patient TIL, and these infusion products are likely broad

polyclonal populations of cells both with and without tumor specificity. Thus, since the

identification of the first tumor antigens in the 1990’s, Rosenberg and others have had an interest

in adoptive cell therapies using T cells engineered to express a given TCR with defined tumor

antigen specificity.

Throughout the last two decades, this approach has been utilized for the targeting of CT or

differentiation antigens such as MART-1, NY-ESO-1, and MAGE-A3 in a broad variety of solid

tumors117,118,297. These studies have been simultaneously promising for the degree of anti-tumor

immunity observed but also deeply troubling for the significant toxicities. Despite seeming tumor

specificity, the targeting of these shared self-antigens has resulted in severe on-target off-tumor

effects such as neurological toxicity or severe colitis due to low level expression of these antigens

in normal host tissues297–299. These studies have indirectly led to increased enthusiasm for the

101

targeting of neoantigens, as they represent ideal therapeutic targets owing to the lack of any pre-

existing immune tolerance against them and their tumor-specific expression. However, unlike the

targeting of CT antigens, which utilize TCRs against epitopes presented by common HLA alleles,

cell therapy with engineered T cells targeting tumor-specific neoantigens would also necessitate

the identification of patient-specific TCRs. Despite this, the widespread utilization of both tumor

genomic sequencing and scRNA-seq profiling of TIL has made this approach far more feasible

than it once seemed300,301.

Despite significant interest in adoptive T cell therapy for solid tumors such as GBM and

the clear benefits of targeting tumor-specific neoantigens, very few preclinical systems exist to

accurately model it. This is largely based on practicality, as relatively few endogenous tumor

neoantigens have been identified in murine models and even fewer transgenic systems exist to

easily target them. Most preclinical systems for studying adoptive T cell immunotherapy target

significantly overexpressed and highly immunogenic foreign proteins introduced into the tumors

or shared antigens also expressed by normal tissue302,303. However, neither overexpressed foreign

proteins nor shared antigens accurately recapitulate the tumor-specific expression profile and

degree of immunogenicity of an endogenous tumor neoantigen. Thus, there exists a need for a

platform to investigate neoantigen-directed cellular therapy and the role of neoantigen-specific T

cells in GBM.

To that end, we generated and characterized the Mutant Imp3-Specific TransgenIC

(MISTIC) mouse engineered with a TCR specific for the mImp3 neoantigen within the murine

GBM model GL261. To our knowledge, this is the first TCR transgenic generated against a tumor-

specific neoantigen in a mouse GBM model. This system displays the potential for neoantigen-

102

targeted cellular therapy in the treatment of GBM and serves as a platform for both translational

and basic investigation on the role of neoantigen-specific T cells in GBM.

6.2 Results

6.2.1 Functional validation of MISTIC T cells

We first sought to characterize the in vitro functionality of MISTIC T cells in response to

their cognate antigen. To do so, we harvested splenocytes from MISTIC mice and stimulated this

bulk population for 5 days with mImp3 peptide and low-dose IL-2. Over this short time frame, we

observed substantial proliferation, and CD8 T cells represented an overwhelming majority of the

population at day 5. Virtually all of these CD8 T cells had an activated phenotype characterized

by high levels of CD44, and a majority also expressed CD62L indicative of a central memory

phenotype (Fig. 6-1a). To confirm the active proliferation of these cells, we isolated and CFSE-

labeled CD8 T cells from MISTIC mice and incubated them for 72 hours with naïve splenocytes

loaded with different peptides at 100 nM. Using the CD45.1 congenic marker for identification of

MISTIC T cells, we demonstrated significant proliferation in response to the mImp3 peptide but

no stimulation by either the wild-type Imp3 epitope or an irrelevant peptide (SIINFEKL) (Fig. 6-

1b). To assess the responses of activated MISTIC T cells, we stimulated MISTIC splenocytes for

5 days as previously described, isolated CD8 T cells from the resulting population, and rested them

for 6 hrs without peptide or IL-2. These rested cells were then co-cultured with naïve splenocytes

loaded with different peptides at varying concentration, and the level of IFN-g secreted into the

supernatant was measured. The activated MISTIC T cells produced high levels of IFN-g in

response to the mImp3 antigen but again showed no response to the wild-type epitope or an

irrelevant peptide (mOdc1) (Fig. 6-1c). Finally, we wished to investigate the capacity of these

103

100

150

0.1 110 100 1000

Concent ration (nM)

IFN−γ (pg/mL)

Peptide

mImp3

mOdc1

wtI mp3

Figure 6-1: Functional characterization of MISTIC T cells. a, Flow cytometry profiling of CD8 T cells from

the spleens of MISTIC mice both before (left) and after (right) in vitro stimulation with mImp3 peptide. b, Naive

MISTIC T cells were CFSE-labeled and stimulated for 72 hrs with mImp3 pepitde prior to assessment of cell

division. c, IFN-γ dose-response from D5-stimulated MISTIC T cells following overnight co-culture with naive

splenocytes loaded with peptides at the indicated concentration. d, Relative IFN-γ following co-culture of

D5-stimulated MISTIC T cells with IFN-γ-stimulated tumor targets.

GL261

CT2A

CT2A:mImp3

Fold Change IFN-γ

104

MISTIC T cells to respond to antigen-bearing tumor targets. To do so, we again utilized D5-

stimulated/rested MISTIC T cells that were co-cultured with IFN-g stimulated tumor targets. The

MISTIC T cells produced high levels of IFN-g in response to mImp3-expressing GL261 but

significantly less in response to CT2A, which does not express the mImp3 antigen (Fig. 6-1d).

When CT2A was engineered to overexpress the mImp3 antigen, we again detected significant

levels of IFN-g, suggesting the mImp3 antigen is sufficient to mediate this response.

6.2.2 Adoptive cell therapy with MISTIC T cells

We next aimed to test the efficacy of our MISTIC T cells in a model of adoptive cellular

therapy (ACT) against mice with intracranial GL261. Most established techniques for cellular

therapy in patients necessitate in vitro stimulation for the purpose of either genetic engineering or

TIL expansion. Thus, we developed an ACT protocol in which tumor-bearing mice receive an

intravenous infusion of in vitro expanded CD8 MISTIC T cells on day 7 following tumor

implantation (Fig. 6-2a). These mice all receive a lymphodepleting dose of 5 Gy total body

irradiation the day prior to cell therapy infusion and are given supplemental IL-2 for several days

thereafter. Control mice receive the same total body irradiation and supplemental IL-2 but no T

cell infusion. The MISTIC T cell therapy resulted in a significant treatment benefit to GL261-

bearing mice, with a majority of mice cured of their tumors (Fig. 6-2b). When the tumor was given

a longer duration to establish and the initiation of treatment was delayed until day 14, this treatment

benefit is lost (Fig. 6-2c).

To assess the potential mechanisms underlying this treatment effect, we performed the

same survival analysis in RAG2-/- mice lacking any endogenous B or T cells. In this system, the

105

b c

Figure 6-2: MISTIC T cell therapy against GL261. a, Schematic depicting the treatment protocol for MISTIC

T cell therapy. b, Survival curve of GL261-bearing mice treated with MISTIC cell therapy on day 7 post-tumor

implantation. c, Survival curve of GL261-bearing mice treated with MISTIC cell therapy on day 14 post-tumor

implantation. d, Survival curve of GL261-bearing RAG KO mice treated with MISTIC cell therapy on day 7

post-tumor implantation. e, Survival curve of GL261-bearing Delta32 mice treated with MISTIC cell therapy

on day 7 post-tumor implantation. All data represents two separate biological replicates.

p < 0.000 1

0.00

0.25

0.50

0.75

1.00

020 40

Days

Survival P roba bility

Treatment

Control

MISTIC Treated

+ +

p = 0.89

0.00

0.25

0.50

0.75

1.00

020 40 60 80

Days

Su rvival P roba bility

Treatment

Control

MISTIC Treated

p = 0 .1

0.00

0.25

0.50

0.75

1.00

010 20 30

Days

Survival Probability

Treatment

RAG -- Control

RAG -- MISTIC Treated

p = 3 e−0 4

0.00

0.25

0.50

0.75

1.00

020 40 60

Days

Survival Probability

Treatment

Delta32 -- C ontrol

Delta32 -- MISTIC Treated

106

profound survival benefit is lost, indicating the necessity of endogenous host lymphocytes for

mediating the response to MISTIC T cell therapy (Fig. 6-2d). We then sought to investigate the

role of the conventional dendritic cell 1 (cDC1) subset in MISTIC T cell therapy. Prior studies

from our lab and others has demonstrated the essential role of these cells in cross-presenting tumor

antigens and priming naïve CD8 and CD4 T cells304,305. Recent work from Ken Murphy’s lab has

identified the transcriptional network controlling cDC1 differentiation and generated a knockout

mouse (Delta32) lacking a crucial enhancer necessary for cDC1 development306. When GL261-

bearing Delta32 mice were given MISTIC T cell therapy, the survival benefit was diminished but

not completely abrogated as in the RAG2-/- system (Fig. 6-2e). Therefore, cDC1’s are also required

for optimal therapeutic effect but not to the same extent as endogenous host lymphocytes.

Finally, we sought to confirm the anti-tumor effect by direct in vivo imaging. For this, a

cohort of control and treated mice underwent serial MRI imaging on a weekly basis following

MISTIC T cell therapy. All of the control mice eventually developed aggressive and infiltrative

tumors clearly visible on T1-weighted MRI (Fig. 6-3). However, most of the treated mice either

did not develop a detectable tumor or demonstrated significantly delayed tumor formation.

6.2.3 Distribution and maintenance of MISTIC T cells

Immune cell infiltration into brain tumors is often considered restricted by the blood-brain

barrier. Additionally, several studies have hinted at T cell sequestration in the bone marrow as a

means for preventing intratumoral T cell infiltration185,302. Thus, we sought to characterize the

distribution of MISTIC T cells following their infusion into tumor-bearing mice. To do so, mice

were harvested three days following MISTIC T cell therapy (10 days post-tumor implantation),

and the localization of MISTIC T cells was assessed via flow cytometry. Just three days following

107

Figure 6-3: MRI profiling of treated mice. T1-weighted MRI images of GL261-bearing mice treated on day 7

with MISTIC T cell therapy. Each panel represents an individual mouse on day 21 post-tumor implantation and

is representative of the cohort overall.

D21 Control

D21 MISTIC Treated

108

cell therapy infusion, a significant fraction of CD8 T cells in the tumors were the adoptively

transferred MISTIC T cells (Fig. 6-4a). In contrast to the cells in other tissue sites, the intratumoral

MISTIC T cells had already significantly upregulated expression of PD-1 (Fig. 6-4a). The MISTIC

T cells also comprised a very high fraction of CD8 T cells in the bone marrow, spleens, and

peripheral blood of treated mice (Fig. 6-4b). Furthermore, they were present in both tumor-

draining and non-draining lymph nodes at comparable frequencies (Fig. 6-4b).

This significant intratumoral MISTIC T cell infiltration resulted in profound changes to the

immunological microenvironment of these tumors. In agreement with our prior studies, a majority

of control mice contained CD4-predominant TIL (Fig. 6-4c). However, treatment with the MISTIC

T cell infusion resulted in a significant skewing of the tumor microenvironment towards being

CD8-predominant (Fig. 6-4c). Additionally, this treatment resulted in the formation of long-lasting

MISTIC T cell populations. On day 50 post-tumor implantation, more than 6 weeks after T cell

infusion, virtually all of the treated mice retained significant levels of circulating MISTIC T cells.

In contrast to the predominance of effector cells within the tumor microenvironment, these cells

appeared to be a circulating central memory population (Fig. 6-4d).

6.2.4 Limitations of MISTIC T cell therapy

Despite the considerable treatment benefit from MISTIC T cell therapy in GL261-bearing

mice, it is noteworthy that a select group of mice will still succumb to disease from tumors

escaping immune control. We sought to comprehensively profile these tumors for the purpose of

identifying crucial immune evasion mechanisms. When the mice succumbed to disease, we

analyzed the immune infiltrate into these escape tumors by flow cytometry. Strikingly, the CD8 T

cell infiltrate within these tumors was comprised almost entirely of MISTIC T cells (Fig. 6-5a).

109

c d

Figure 6-4: Distribution and maintenance of MISTIC T cells. a, Representative flow cytometry plots of the

infiltration of MISTIC T cells into the tumor and their prevalence in peripheral blood in treated mice three days

post-transfer. The two plots on the left are gated on CD8 T cells, while the two plots on the right are gated on

MISTIC T cells. b, Distribution of MISTIC T cells across various tissues in treated mice three days post-transfer.

c, Impact of MISTIC T cell therapy on the CD8/CD4 ratio in treated mice. d, Frequency and phenotype of

MISTIC T cells in the peripheral blood of treated mice 50 days post-tumor implantation.

BM cLN iLN PBMC SPL TIL

%CD45.1+ of CD8 T cells

Condition

Control

Treated

Co ntrol Treated

CD8/CD4 Ratio

Condition

Control

Treated

%CD45.1+ of CD8 T cel ls

D50 Treated PBMC

110

All of these cells had appeared to adopt an exhausted phenotype characterized by significant

upregulation of PD-1 (Fig. 6-5a). We have also performed scRNA-sequencing on the immune

infiltrate from a cohort of treated escape tumors with analysis currently in progress.

We then wanted to profile these escape tumors by whole-exome and RNA-sequencing to

identify potential escape mechanisms at play in the face of this significant tumor-specific T cell

infiltration. In comparing the expression profile from treated escape tumors to untreated control

tumors, we identified significantly higher levels of Cd3d, Cd8a, and Trbv12-1, the specific Vb

within the 3x.1.1C TCR (Fig. 6-5b). Furthermore, we detected higher levels of the chemokine Xcl1

involved in the recruitment of cDC1s (Fig. 6-5b). However, we could not pinpoint any differences

in the expression of components of antigen presentation or the Imp3 gene, suggesting antigen or

class I downregulation did not occur. Altogether, the expression data is consistent with our

observations from flow cytometry but does not offer any novel insight into potential mechanisms

of immune evasion at play. We are actively awaiting the results of the whole-exome sequencing

to determine whether the escape tumors retained the mImp3 antigen.

With extensive work from our own and other labs characterizing the extensive intratumoral

heterogeneity of GBM, it is likely that many neoantigen targets would be subclonal. However, a

vast majority of preclinical systems do not incorporate antigen heterogeneity. To address this, the

Genome Engineering and iPSC Center generated a CRISPR clone of GL261, hereafter referred to

as GL261:E8, in which the Imp3D81N mutation was reverted to wild-type. GL261:E8 grew at a

similar rate to GL261 in vitro and resulted in the death of tumor-bearing mice with a nearly

identical time-course in vivo (Fig. 6-6a). Furthermore, the capacity of MISTIC T cells to induce

tumor cell lysis as measured by 7-AAD staining was substantially diminished against GL261:E8

tumor targets (Fig. 6-6b).

111

Figure 6-5: Escape tumor profiling. a, Flow cytometry characterization of the TIL in treated escape tumors.

The left-sided plots are gated on CD8 T cells. b, Expression values (in FPKM) for a subset of chosen genes

between control and treated escape tumors.

D54 Escape

Tumor

D69 Escape

Tumor

0. 1

10 0

10 00

Cd3d Cd8a Trbv12−1 Xcl1 Imp3 H2−D1 H2−K1 B2m

FPKM

Condition

Control

Treated

112

a b

Figure 6-6: MISTIC T cell treatment in heterogeneous system. a, Survival curves for naive mice implanted

with either GL261 or GL261:E8. b, MISTIC T cell cytotoxicity against tumor targets measured by 7-AAD+

following overnight co-culture. c, Survival curves for naive mice implanted with either GL261 or GL261:E8 and

mice previously cured of GL261 with MISTIC T cell therapy implanted with either GL261 or GL261:E8.

d, Survival curves for mice implanted with either GL261, GL261:E8, or a 50/50 mixture of the two cell types

treated on day 7 with MISTIC T cell therapy. The curves in panel (a) represent two separate biological replicates

while the curves in (c) and (d) have been once with additional replicates in progress.

p = 0 .3 7

0.00

0.25

0.50

0.75

1.00

010 20 30

Days

Survival Probability

Treatm ent

GL261

GL261:E8

0.00

0.25

0.50

0.75

1.00

020 40 60

Days

Survival Probability

Treatm ent

50/50−−Cont rol

50/50−−M ISTIC Treated

GL261−−Cont rol

GL261−−MISTIC Treated

GL261:E8−−Cont rol

GL261:E8−−MIS TIC Treated

0.0

2.5

5.0

7.5

10 .0

12 .5

GL261 GL261:E8

% 7−AAD+

0.00

0.25

0.50

0.75

1.00

010 20 30

Days

Survival Probability

Treatment

Naive−−GL261

Naive−−GL261:E8

Rechall en ge−−GL261

Rechall en ge−−GL261:E8

113

To assess the generation of long-lasting immunological memory and the protection against

antigen escape variants, mice previously cured of GL261 with MISTIC T cell therapy were

rechallenged in the contralateral hemisphere with either GL261 or GL261:E8. As expected, these

mice were predominantly protected against GL261 rechallenge. However, a majority of mice

previously cured of GL261 by MISTIC T cell therapy were unable to survive rechallenge with

GL261:E8 (Fig. 6-6c). Finally, we sought to determine whether the MISTIC T cell therapy could

prove effective in a primary challenge against a heterogeneous tumor. To model this, mice were

injected with either GL261, GL261:E8, or a 50/50 mixture of the two cell lines. Strikingly, the

mice implanted with either GL261:E8 or the 50/50 cell mixture received no survival benefit from

the MISTIC T cell therapy (Fig. 6-6d).

6.3 Discussion, Future Directions and Conclusions

Tumors such as GBM that lack significant baseline intratumoral T cell infiltration represent

potentially ideal targets for adoptive cellular therapy. The rapid development of genomic

technologies that facilitate neoantigen identification has led to a focus on the development of

neoantigen-directed cellular therapy, with the ultimate hope that this approach will mediate

significant anti-tumor effects without the toxicity of targeting shared antigens. In this chapter, we

presented a preclinical system for investigations of neoantigen-specific cellular therapy in brain

tumors using the MISTIC mouse we generated together with the murine glioma model GL261.

Adoptive cell therapy with MISTIC T cells exhibited both significant tumor control under certain

conditions but also highlighted the challenges of successful glioma immunotherapy. We believe

that this system presents an invaluable platform to investigate both mechanistic and translational

questions in brain tumor immunology.

114

The adoptive transfer of a huge number of activated tumor-specific CD8 T cells should

theoretically have the capacity to lyse and eliminate tumors without engaging the host immune

system, as the initial steps of the cancer immunity cycle have been completely bypassed. However,

our studies in the RAG2-/- and Delta32 systems suggest that the endogenous host immune system,

specifically the lymphocyte compartment, is necessary for optimal treatment effect. A potential

reason for the lack of treatment benefit in RAG2-/- mice is the complete absence of anti-tumor CD4

T cells, which have been shown to be necessary for the optimal function of anti-tumor CD8 T cells

in numerous studies251–254. The residual treatment benefit seen in the Delta32 mice may then be

mediated by the low but non-zero levels of non-cDC1-dependent CD4 T cell activation observed

in other work304. Regardless, a deeper analysis of the interactions between the adoptively

transferred MISTIC T cells and the endogenous immune system should identify areas for further

investigation and potential therapeutic targets.

While successful MISTIC T cell therapy displays the potential for the targeting of

neoantigens in brain tumors, several studies utilizing CAR T cells have demonstrated some of the

major challenges facing adoptive cell therapy in the treatment of GBM. Thus, we consider the

study of mechanisms of immune evasion and tumor escape with this system to be an equally

important aspect. To that end, the generation and subsequent experiments with the GL261:E8

subclone are crucial. One of the major hopes for successful immunotherapy is the development of

epitope spreading, whereby targeting of a specific antigen leads to the generation of immune

responses against previously unrecognized antigens. Generally, this is thought to be mediated by

significant tumor cell lysis leading to antigen release and subsequent initiation of the cancer

immunity cycle. Our results suggest that the MISTIC T cell therapy does not generate a significant

degree of epitope spreading, as evidenced by the lack of immunity to GL261:E8 in previously

115

cured mice and the lack of efficacy in treating tumors comprised of a 50/50 mixture of GL261 and

GL261:E8. This has major implications for the treatment of GBM and other heterogeneous solid

tumors, suggesting that a potent tumor-specific therapy will not necessarily mediate widespread

anti-tumor immunity against new targets. On one hand, this points to the need for the targeting of

clonal neoantigens within tumors. However, the number of clonal neoantigen targets is incredibly

low in heterogeneous tumors such as GBM. Therefore, the process through which an antigenically

diverse anti-tumor immune response can be generated must be a significant focus going forward.

Otherwise, most targeted T cell therapies within GBM and other solid tumors will inevitably fall

victim to the outgrowth of an antigen-negative subclone.

116

CHAPTER SEVEN

Discussion, Future Directions, and Conclusions

The concept that the immune system can be leveraged to eliminate and prevent cancer

circulated on the fringes of both cancer biology and immunology for more than a century.

However, largely driven by a deeper understanding of the fundamentals of T cell biology, cancer

immunology has experienced a rebirth over the last several decades. In many respects,

immunotherapy is now viewed as a fundamental pillar of cancer therapy on par with more

conventional approaches such as radiation and chemotherapy. While cancer immunotherapy has

been transformative in certain cancers, many tumors such as GBM have seen no benefit.

It is noteworthy that the conventional FDA-approved immunotherapies, CAR T cells and

ICB therapy, do not have any tumor-specific component to them. ICB therapy appears to enhance

endogenous anti-tumor immune responses but relies on the prior generation and expansion of these

responses by the host. CAR T cells, on the other hand, utilize the targeting of a cell surface

molecule (CD19) expressed by both B cell malignancies and host B cells and can only be tolerated

because the loss of endogenous B cells is a clinically manageable toxicity. At the same time, an

explosion of genomic technologies has paralleled the advances in cancer immunotherapy and has

resulted in the development of a field of cancer immunogenomics. Broadly, cancer

immunogenomics seeks to utilize both tumor genomic analyses and cancer immunology to identify

and target tumor-specific neoantigens. Owing to their restricted expression pattern and a lack of

central tolerance against them, neoantigens have long been considered ideal therapeutic targets.

However, challenges in neoantigen identification have long hindered attempts at targeting them

clinically. Widespread adoption of next generation sequencing technologies, computational

117

prediction algorithms, and the development of single-cell RNA sequencing have provided

substantial new insight into tumor-immune interactions in a variety of tumor types and present a

path forward for the targeting of tumor-specific neoantigens. This work seeks to further our

understanding of the immunogenomic state of human malignant brain tumors with an emphasis

on neoantigen identification and establish a preclinical system for targeting these brain tumor

neoantigens through adoptive cellular therapy.

Despite extensive genomic profiling, comparatively less is known about the

immunological microenvironment of GBM. Recent studies have focused primarily upon tumor

infiltrating myeloid cells, which make up a significant fraction of the bulk tumor mass but devoted

less attention to the lymphoid compartment. Furthermore, nothing is known about the antigens

recognized by endogenous tumor infiltrating T cells in malignant brain tumors. To that end, we

recruited a cohort of 30 patients with primary or secondary malignant brain tumors to

comprehensively profile their immunological microenvironment and identify tumor-specific

antigens recognized by infiltrating T cells. Our multi-sector analysis revealed substantial spatial

differences in the genomic characteristics of these tumors, supporting the need for multi-sector

analysis in the design of small molecule or neoantigen-directed trials against gliomas. In contrast,

single-site analysis provided a fairly accurate picture of the immunogenomic state of secondary

metastatic brain tumors.

While we uncovered considerable spatial differences between primary gliomas and

secondary brain metastases, we also observed significant clonotype loss in our expanded TIL

cultures from both tumor types. This forced us to reconfigure our approach to neoantigen

identification and generate a TCR-based screening system for antigen identification utilizing TCRs

isolated from our scRNA-seq data. While our system offers flexibility with regards to patients and

118

TCRs, it is limited by its dependence on peptides and is not particularly high-throughput.

Therefore, while it can be utilized for screens against candidate neoantigen peptides, the

development of a more unbiased and high-throughput system should be of utmost importance.

Whether such an approach utilizes tumor cell lines and cDNA screens or fully artificial systems

such as yeast display will require extensive further study. However, the presence of tumor-specific

neoantigens recognized by the endogenous host immune response remains a fundamental question

in anti-glioma immunity. The ultimate goal should be the comprehensive characterization of the

antigen specificity of tumor infiltrating lymphocytes within GBM including neoantigens, viral

antigens, or overexpressed self antigens.

Ultimately, the direct generation of neoantigen-specific responses in patients will arise

from either neoantigen vaccination or cellular therapy. While neoantigen vaccines are more easily

implementable, they have thus far generated relatively low numbers of cells and primarily

produced CD4 T cell responses despite targeting predicted CD8 epitopes142,143. Cellular therapy

has historically been based upon the expansion of endogenous TIL from patient tumors, but the

sparse T cell infiltrate in GBM combined with our own challenges in maintaining TIL clonal

architecture suggests this approach will be difficult in GBM. The implementation of CAR T cell

therapy has demonstrated the potential of autologous engineered T cells. Together with the rapid

identification of intratumoral TCRs from scRNA-seq data, the generation of personalized TCR-

engineered T cells has become feasible. However, a fundamental question with this approach is

the choice of TCR. Ideally, systems will exist whereby tumor-specific TCRs can be predicted from

single-cell profiling without necessitating the need for extensive antigen screens.

To that end, our generation of the MISTIC mouse and subsequent MISTIC T cell therapy

represents a preclinical model for this neoantigen-directed cellular therapy and a means to probe

119

these exact questions. While the 3x.1.1C TCR was isolated from an in vitro stimulated population

and was found to be the most functional TCR, it was not the most expanded T cell clone in this

population. This suggests that a myriad of yet unknown factors determines the ideal choice of

therapeutic TCR, even when restricted to the response against a specific neoantigen:MHC

complex. Future work can utilize the targeting of mImp3 in GL261 to focus on this specific

question, characterizing numerous neoantigen-specific TCRs both in vitro and in their ability to

induce anti-tumor immune responses in vivo.

While the MISTIC mouse facilitates numerous studies on adoptive cellular therapy, it also

opens the door for more mechanistic studies on the migration and activation of neoantigen-specific

T cells in brain tumors. Owing to the relative lack of preclinical systems with defined endogenous

neoantigens, most adoptive transfer studies utilize genetically modified tumor systems engineered

to express the targets of pre-existing TCR transgenic mice. By generating our own TCR transgenic,

we have the means to target an endogenous tumor-specific neoantigen. Extensive work from our

own and other labs has demonstrated the pivotal importance of the type I conventional dendritic

cell (cDC1) in mediating anti-tumor CD4 and CD8 T cell responses. However, the exact location

of T cell priming within brain tumors is not clear owing to the lack of conventional lymphatic

structures within the CNS. Through adoptive transfer and high-resolution imaging studies, we

hope to clarify these aspects of CNS immunity. Furthermore, we aim to characterize the role of

the meninges, specifically the dural layers, in modulating T cell responses to intracranial antigens.

Prior studies from our own and other labs have hinted at the localization of immune cells to the

meninges in the setting of CNS immune responses, but the functional impact of this is unknown307.

Additionally, the MISTIC T cell system provides a means to investigate the specific roles of CD4

T cell responses in anti-tumor immunity, specifically in the context of aiding and maintaining

120

functional CD8 T cell responses. Future studies should characterize potential indirect interactions

between transferred MISTIC T cells and the endogenous CD4 T cell compartment including sites

of interaction and impact on MISTIC T cell function. Moving forward, the potential identification

of endogenous class II-restricted neoantigens within our preclinical models would greatly facilitate

this work.

Antigen heterogeneity is undoubtedly one of the most significant challenges facing targeted

immunotherapies. This work comprehensively profiled the immunogenomic landscape of a cohort

of malignant brain tumors, outlining substantial spatial genomic and antigenic heterogeneity

among malignant gliomas. Strikingly, this same heterogeneity was observed in the distribution of

the T cell repertoire, with GBM harboring spatially restricted but expanded clonotypes. While it

is possible that the antigenic heterogeneity produces this intratumoral immune heterogeneity,

further work will be needed to clarify whether this phenomenon is antigen-driven. Regardless, this

well-defined antigenic heterogeneity is often ignored in the context of preclinical models. Even

when tumor antigenic heterogeneity is studied, it normally utilizes drastically overexpressed

highly immunogenic foreign antigens present in a proportion of the tumor cells, which may not

accurately recapitulate the responses to endogenous neoantigens.

To address this, we generated a CRISPR clone of GL261, GL261:E8, to facilitate

preclinical studies on tumor heterogeneity utilizing an endogenous tumor neoantigen. The concept

of epitope spreading, characterized by a broadened immune response to non-targeted antigens

following targeted immunotherapy, is often invoked as a means for the immune system to respond

to polyclonal heterogeneous tumors. We sought to directly address this utilizing our MISTIC T

cell therapy in GL261 and GL261:E8. Strikingly, the survival benefit appears to be lost upon

treatment of a heterogeneous tumor consisting of an equal mixture of GL261 and GL261:E8,

121

presumably due to the eventual outgrowth of antigen-negative GL261:E8. More surprisingly, mice

previously cured of GL261 with MISTIC T cell therapy do not appear to have generated significant

immunity against GL261:E8. This suggests that, despite the thousands of additional predicted

neoantigens present within GL261, tumor regression with MISTIC T cell therapy did not lead to

the generation of protective immunity against additional tumor targets. Given the highly

heterogeneous nature of GBM and other solid tumors, this indicates that strategies to invoke

epitope spreading and the generation of a more diverse immune response must be pursued.

Recently, several major studies have implicated tertiary lymphoid structures, ectopic lymphoid

aggregates that develop in non-lymphoid tissues at sites of chronic inflammation, in the successful

response to immunotherapy308,309. While the mechanisms underlying this are not clear, it is feasible

that these additional lymphoid aggregates facilitate the development of a more robust and diverse

anti-tumor immune response. The combination of MISTIC T cell therapy with GL261/GL261:E8

will allow for further studies on how to best facilitate epitope spreading.

A major byproduct of the success of CAR T cell therapy has been the widespread

acceptance of autologous T cell engineering in the treatment of cancer. One approach to enhance

MISTIC T cell function and potentially induce epitope spreading is further genetic modifications

of these cells prior to infusion. This could be done by the deletion of T cell inhibitory factors or

the incorporation of gene cassettes inducing additional cytokine or chemokine production from the

MISTIC T cells. The feasibility of this approach has already been demonstrated in patients with a

study utilizing autologous T cells with endogenous PD-1 deleted and an NY-ESO-1-specific TCR

introduced via CRISPR224. The possibilities for further T cell engineering utilizing the MISTIC T

cell system are broad.

122

This work has helped characterize the immunogenomic state of malignant brain tumors

utilizing comprehensive sequencing technologies, created systems for the antigen screening of

intratumoral TCRs, and developed a system for targeting tumor neoantigens with adoptive cellular

therapy in a preclinical model of GBM. Ultimately, it has shed insight on the potential for

immunotherapy in GBM but has also highlighted significant challenges both specific to GBM and

in the treatment of solid tumors in general. Our ultimate hope is that further investigation of the

anti-tumor immune response in both primary human samples and preclinical mouse models will

aid the development of novel immunotherapeutic approaches to improve disease outcomes in

patients with GBM.

123

CHAPTER EIGHT

Methods

Patient recruitment. All study participants were neurosurgical patients at Barnes-Jewish Hospital

with pathologically confirmed stage III/IV glioma or metastatic disease. Prior to surgery, we

obtained written informed consent from the patients for a Washington University School of

Medicine Institutional Review Board-approved protocol (#201111001) for the analysis of tumor

tissue and peripheral blood and sharing of genomic data. All procedures and experiments were

performed in accordance with the ethical standards of the 1964 Helsinki declaration.

Clinical sample processing & nucleic acid extraction. Tumor samples were processed

immediately following surgical resection under sterile conditions. Tissue was thoroughly washed

in phosphate-buffered saline to eliminate peripheral blood leukocyte contamination. For tumor

samples that were resected en bloc, spatially distinct sectors were dissected out from the mass via

scalpel. In other cases where multiple discrete regions demonstrated enhancement on MRI, tissue

was separated at the time of surgery and a representative sample from each sector was chosen for

analysis. In all cases, tissue was immediately flash frozen and initially stored at -80oC or in liquid

nitrogen until further use. Peripheral blood was separated through Ficoll-Paque PLUS density

gradient (GE), and the buffy coat was collected and frozen for matched normal genomic DNA.

Total RNA and genomic DNA was extracted from peripheral blood mononuclear cells or

frozen tissue using the Qiagen AllPrep DNA/RNA Kit (catalog #80204) according to the

manufacturer's instructions (Qiagen, Valencia, CA). As melanin is a known inhibitor of enzymatic

reactions and coprecipitates with RNA310, further purification was performed for the melanoma

124

sample as described311, modified with a RNeasy (Qiagen, Valencia, CA) column-based clean-up

to remove the additives used to bind the melanin. For GBM065.Re, regions 3-4 and the sample

from the primary tumor came from FFPE tissue cores. In these cases, total RNA and genomic

DNA was extracted using the Qiagen AllPrep DNA/RNA FFPE Kit (Catalog #80234).

Sequencing and somatic variant detection. All sequencing was performed on the Illumina

NovaSeq (S4) platform. From each patient a normal sample and 2-4 samples from a single tumor

were subjected to whole exome sequencing (WES). 80 of 93 tumor samples had sufficient tissue

to perform RNA sequencing. WES and RNAseq from 20 tumors were performed at the McDonnell

Genome Institute (MGI), and 10 tumors underwent WES and RNAseq at the Institute for Genomic

Medicine at Nationwide Children's Hospital. WES was performed on 2 additional recurrent

GBM065 regions and a GBM065 primary tumor by Novogene. Read alignment, somatic variant

calling, variant filtering, variant effect prediction, additional variant annotation, and RNA

expression estimation was performed using a tumor analysis pipeline defined in the Genome

Modeling System as previously described312. Briefly, all fastq files were aligned to the human

reference genome build GRCh38 with HISAT2(RRID:SCR_015530)313 for RNA and BWA-

MEM314 for DNA. Somatic variant calling was performed using Strelka315, VarScan316, Mutect317,

and Pindel318. To remove any false positive variants and discover TERT promoter mutations,

custom capture validation sequencing was performed to an average depth of 582x for all unique

variant sites using NimbleGen SeqCap EZ Prime Choice Probes (12,388 total probes created from

11,425 variants and 51 genes; 1.49 Mb of sequence targeted for capture). Tumor purity was

estimated using TPES319 for samples with sufficient variant count (> 20), and for those with

insufficient variants, the median variant allele frequency was used (2 x median VAF) . SciClone320

125

and ClonEvol321 were used to assess the clonality of mutations and subclonal evolution of primary

and recurrent GBM065 tumors.

CNV detection. Matched tumor/normal WES data was used to predict CNVs with CNVkit322.

Each tumor region sample was compared pairwise against the single matched normal data for the

patient. CNVkit was run (via “cnvkit batch”) using default parameters. Regions of alignment were

summarized to a resolution of 10kb and these data were subjected to segmentation using circular

binary segmentation. The resulting segmentation files were divided into three cohorts: 1) primary

and recurrent GBMs 2) breast cancer BrMETs 3) NSCLC BrMETs. GISTIC2323 was run on the

segments of each cohort with a q-value cutoff of 0.1. Relative amplitude thresholds were used to

create the heatmaps and clonality plots in Figure 2.

GBM Molecular Subtyping. Single-sample GSEA (ssGSEA) scores324 were calculated for each

sample in the cohort using the previously defined gene sets172 and the GSVA R package325. The

circlize (RRID:SCR_002141) R package was used for the visualization of these molecular

subtypes326.

Neoantigen prediction. Clinical class I and class II HLA typing was performed on each normal

sample by Histogenetics to two field resolution. For each tumor sample, a VCF file containing all

passing somatic variants was annotated with Ensembl VEP327 (RRID: SCR_002344) using the

parameters --everything, --flag_pick, and --plugin (Wildtype and Downstream). The VCF was

further annotated with DNA and RNA read counts using VAtools. Using this VCF as input, a

containerized version of pVACtools221 (DockerHub: griffithlab/pvactools:1.5.0) was used to

126

predict and annotate likely neoantigens in each sample as previously described328. Briefly, using

the sub-module “pvacseq run” we performed peptide:MHC binding affinity predictions with eight

class I and four class II algorithms (NNalign, NetMHC, NetMHCIIpan, NetMHCcons,

NetMHCpan PickPocket, SMM, SMMPMBEC, SMMalign, MHCflurry, MHCnuggetsI,

MHCnuggetsII). For this analysis, to be considered a neoantigen candidate it must have arisen

from a variant with tumor DNA VAF > 5% and median binding affinity (across all algorithms) <

500 nm. No sequences were excluded based on tumor RNA VAF, coverage, or gene expression

values to ensure that no high quality candidates were excluded based solely on RNA data.

Neoantigens that pass this filter are designated ‘candidate neoantigens’ in the manuscript.

Genomic alteration distribution. All genomic alterations (variants and associated neoantigens,

CNVs) were determined independently with each region as an individual sample. They were then

aggregated together for the plots in Figure 1 to provide an overall description of each tumor. We

then determined the distribution of each alteration between regions from the same tumor. If the

same change was observed in all regions of the tumor, it was defined as “clonal”. If it was

identified in more than one but not all regions it was defined as “subclonal shared”. Alterations

only observed in one region of each patient sample were defined as “subclonal private”.

Cancer/testis antigen analysis. A subset of cancer/testis antigens were chosen for analysis based

upon a combination of prior work targeting them in GBM201 (e.g. IL13Ra2) and/or previous

studies in other malignancies222,224 (e.g. NY-ESO-1, MAGE family proteins). Cancer/testis antigen

scores were generated by performing a log2 transform on the expression (TPM) of each gene

normalized to the median tissue expression for “Brain-Cortex” in the Genotype-Tissue Expression

127

(GTEx) Project database. Cancer/testis antigen scores were generated for the metastatic samples

through either normalizing to “Brain-Cortex” expression or the associated matched primary tissue

site (“Breast Mammary Tissue”, “Lung” or “Skin Sun Exposed”). The intratumoral spread of

cancer/testis antigen expression was assessed by calculating the variance of scores within the

regions of a tumor or by calculating the cosine similarity between regions, where each antigen’s

score in a given region represents one component of the vector.

RNA-Seq analysis and immune microenvironment profiling. Gene and transcript

quantification was performed using kallisto329. Differential expression analysis between tumor

types was then performed using sleuth330. To quantify immune cell populations in the tumors,

CIBERSORT226 was run using the LM22 signature gene file and disabled quantile normalization

for 100 permutations. Relative immune cell abundance was obtained using gene expression as

previously described in Danaher et al225. Principal components analysis dimensionality reduction

was implemented on Danaher scores for each tumor region in R version 3.6.2 using the ggbiplot

package.

Intratumoral similarity was performed by representing Danaher scores as vectors and

calculating the normalized dot product between pairs of regions from the same tumor. The immune

intratumoral heterogeneity (ITH) was defined for tumors with three distinct regions as the area in

PC1-PC2 space of the triangle whose vertices were represented by each region’s Danaher score.

The myeloid-specific analyses were performed through the generation of log-transformed scores

making use of gene sets defined in previous publications. A total of 11 and 10 genes were used to

define the M1 and M2 scores, respectively, based upon gene expression signatures of in vitro

polarized M1 or M2 macrophages227. A total of 6 distinct genes for each were used to define the

128

scores for both microglial and monocyte-derived macrophages228. These genes were selected

based upon differential expression in tumor-associated macrophages from distinct lineages in

single-cell RNA-sequencing data of human gliomas. To account for differential abundance of

macrophages between samples as a reason for findings, the difference between these scores

(corresponding to a ratio of gene expression) was taken to generate an “M2-M1 skew” or “MDM-

Microglial skew”.

TCR repertoire sequencing. Sequencing of the CDR3 regions of TCR b-chains was performed

through the ImmunoSEQ Assay (Adaptive Biotechnologies) and used the same DNA extracted

from tumor samples utilized for whole exome sequencing. Initial analysis was performed on the

immunoSEQ ANALYZER 3.0. Additional analysis was performed using the immunarch package

in R. Visualization of the V-J gene usage among T cell repertoires was performed using the circlize

(RRID:SCR_002141) package in R326. The intratumoral T cell fraction was calculated by dividing

the number of productive TCR templates by the number of nucleated cells estimated by the

amplification of reference genes. The Simpson’s clonality for each region was calculated by taking

the square root of the Simpson’s diversity index for all productive rearrangements with possible

values ranging from 0 (very polyclonal) to 1 (predominantly monoclonal or oligoclonal).

The similarity between TCR repertoires was assessed through either the normalized dot

product (cosine similarity) or Morisita overlap between the vectors of TCR clonotype abundances.

Both of these metrics are based upon pairwise comparisons between two repertoires. The T cell

repertoires for two regions are represented by vectors with indices covering the union of TCRs

observed in either of the corresponding regions. Each position of the vector then represents the

129

abundance (as a count) of a given clonotype within that region. The cosine similarity of the T cell

repertoires in regions A and B can then be given by:

𝐶𝑜𝑠𝑖𝑛𝑒'𝑆𝑖𝑚𝑖𝑙𝑎𝑟𝑖𝑡𝑦=' 𝐴

⋅'𝐵

𝐴

∗

𝐵

∑

𝐴!∗'𝐵!

!#$

8∑

𝐴!

!#$

8∑

𝐵!

!#$

where

𝐴!

and

𝐵!

represent the counts of clonotype

𝑖

within region A or B, respectively, and

𝑛

represents the number of unique clonotypes observed in either A or B. Similarly, the Morisita

overlap between the T cell repertoires in regions A and B is defined as:

𝑀𝑜𝑟𝑖𝑠𝑖𝑡𝑎'𝑂𝑣𝑒𝑟𝑙𝑎𝑝=' 2∗(𝐴

⋅'𝐵

)

@∑

𝐴!

!#$

𝑁&

∑

𝐵!

!#$

𝑁'

∗'𝑁&∗𝑁'

where

𝐴!

𝐵!

, and

𝑛

represent the same values as before and

𝑁&

and

𝑁'

represent the total number

of productive rearrangements observed in region A or B, respectively. Both of these metrics

provide a value between 0 and 1 where 0 represents no similarity (orthogonal vectors or completely

disparate populations) and 1 represents complete similarity (parallel vectors or completely

identical populations). Comparisons between tumor types were performed by taking all

intratumoral comparisons between GBMs and BrMETs and grouping them.

Sample preparation for single-cell sequencing. For single-cell samples, the fresh surgical

sample was rinsed with PBS to remove visual blood contaminant and manually dissociated using

130

frosted microscope slides and gentle trituration. The resulting single-cell suspension was passed

through 100 mM and 70 mM filters before undergoing Percoll (GE Healthcare Life Sciences)

density gradient centrifugation to remove myelin contamination. Following this separation, the

resulting cell pellet underwent RBC lysis with ACK Lysis Buffer (Lonza Biosciences) and was

frozen in 90% FBS and 10% DMSO at -80oC and later stored in liquid nitrogen until further use.

The tumor sample was later thawed, and the single-cell suspension was stained with anti-

CD45, anti-CD11b, anti-CD3, and Zombie NIR Viability Dye (BioLegend). Live CD45+ single

cells were purified by fluorescence-activated cell sorting on a BD FACSAria II with an 85 mM 45

psi nozzle into a buffer of PBS with 0.04% BSA. A total of 13,000 were submitted for analysis.

Single-cell library preparation. cDNA was prepared after the Gel Beads in Emulsion (GEM)

generation and barcoding, followed by the GEM-RT reaction and bead cleanup steps. Purified

cDNA was amplified for 10-14 cycles before being cleaned up using SPRIselect beads. Samples

were then run on a Bioanalyzer to determine the cDNA concentration. TCR target enrichment was

done on the full length cDNA. GEX and Enriched TCR libraries were prepared as recommended

by the 10x Genomics Chromium Single Cell V(D)J Reagent Kits (v1 Chemistry) user guide with

appropriate modifications to the PCR cycles based on the calculated cDNA concentration. For

sample preparation on the 10x Genomics platform, the Chromium Single Cell 5’ Library and Gel

Bead Kit (PN-1000006), Chromium Single Cell A Chip Kit (PN-1000152), Chromium Single Cell

V(D)J Enrichment Kit, Human, T Cell (96rxns)(PN-1000005), and Chromium Single Index Kit T

(PN-1000213) were used. The concentration of each library was accurately determined through

qPCR utilizing the KAPA library Quantification Kit according to the manufacturer's protocol

(KAPA Biosystems/Roche) to produce cluster counts appropriate for the Illumina NovaSeq6000

131

instrument. Normalized libraries were sequenced on a NovaSeq6000 S4 Flow Cell using the XP

workflow and a 151x8x151 sequencing recipe according to the manufacturer's protocol. A median

sequencing depth of 50,000 reads/cell was targeted for each Gene Expression Library and 5000

reads/cell for each V(D)J (T cell) library.

Single-cell sequencing analysis. Raw sequencing data was processed with Cell Ranger, version

3.0.1331, from 10X genomics, mapped onto a human genome reference (GRCh38-2020-A).

Downstream analysis was performed using the Seurat (RRID: SCR_007322) R package version

3.2.0332. Low quality cells and potential doublets were accounted for by removing cells that

contained fewer than 500 expressed genes, a nCount value greater than the nCount value of the

93rd percentile of the total sample, or more than 10% mitochondrial transcripts. Genes that were

expressed in fewer than 100 cells were also removed. For each cell, expression of each gene was

normalized to the sequencing depth of the cell, scaled to a constant depth (10,000), and log

transformed. Variable genes were selected with default settings and principal component analysis

was performed on the variable genes. Dimensionality reduction and visualization were performed

with the UMAP algorithm (Seurat implementation) using the first 15 PCA dimensions.

Unsupervised graph-based clustering of cells was performed using the mentioned PCA dimensions

with a resolution of 0.8.

Enriched gene expression levels in each cell cluster were identified by a Wilcoxon Rank

Sum test-based function. These genes, along with common cell type markers, were used to

establish the cell identity of each cluster. Projection of average expression of marker genes into

UMAP or violin plots was used for cell type identification. Gene expression signatures used for

definition of clusters were as follows: CD3E, CD3D, CD3G (T cells), NKG7, PRF1, GZMH, CD3-

132

(Natural killer cells), MS4A1, CD79B, CD3- (B cells), and CD14, S100A8, S100A9, C1QC, CD68,

CTSD, HLA-DR+ (Monocytes/macrophages).

Second-level clustering of T cells was performed by subsetting only T cell clusters and

rerunning scaling, log transformation, and variable gene selection. In addition, PCA was

performed again on the new variable genes. Dimensionality reduction and visualization were

performed with the UMAP algorithm using the first 10 PCA dimensions. Unsupervised graph-

based clustering of cells was performed using the mentioned PCA dimensions with a resolution of

1.0. Gene expression signatures used for definition of clusters were as follows: CCR7, SELL,

TCF7, KLF2 (Naive/central memory T cells), Il7R, KLRB1, SELL- (Effector memory T cells),

TIGIT, CTLA4, CD4 (CD4+ regulatory-like T cells), and NKG7, PRF1, CCL5, GZMH, CD3,

CD8A, CD8B (Cytotoxic CD8+ T cells).

V(D)J libraries were processed with CellRanger V(D)J, version 2.0.0, from 10x Genomics

mapped onto a human VDJ reference (GRCh38-2.0.0). Clonotype analysis was performed with

the scRepertoire (version 0.99.17) R package333. Clonotypes were defined as the combination of

the genes of the TCR A and B chains and nucleotide sequences as previously discussed334.

Statistical Analysis on Chapters 2-3. Data analysis and visualization in R was performed using

the tidyverse package. Statistical significance for variant, neoantigen, CNV clonality estimates,

heterogeneity estimates, T cell fraction, and T cell clonality were performed using an unpaired t-

test. Significance for differential expression was determined by a multiple t-test with Benjamini-

Hochberg adjustment with an FDR = 0.05.

133

Human neoantigen screening system. TCR-deficient Jurkat-Nur77:GFP cells were obtained

from Paul Thomas’s lab at St. Jude Children’s Research Hospital, while K562 cells were obtained

from Todd Fehniger’s lab here at Washington University. For additional modification of the Jurkat

cells, gene blocks encoding human CD4 or CD8 were ordered and cloned into the pLX304

expression system prior to retrovirus production and Jurkat transduction. For additional

modification of the K562’s, a gene block encoding human CD80 was ordered and cloned into the

pLX304 expression system prior to retroviral transduction.

Full-length human TCRs were constructed by utilizing the 5’ VDJ single-cell sequencing

data in conjunction with the IMGT reference. Full-length TCR gene blocks were ordered

consisting of b chain -- P2A -- a chain and were cloned into the pMSCV-IRES-RFP expression

plasmid prior to retroviral transduction.

Patient-specific HLA’s were identified by performing clinical haplotyping on DNA

isolated from peripheral blood of patients through Histogenetics. Full-length human HLA’s were

then constructed by utilizing the European Bioinformatics Institute database. Gene blocks

encoding these HLA’s were ordered and cloned into the pBabe-puromycin backbone prior to

retroviral transduction of K562’s and puromycin selection.

Mouse neoantigen screening system. TCR-deficient 58 hybridoma cells were obtained from

David Kranz’s lab at the University of Illinois Urbana-Champaign. For additional modification of

these cells, gene blocks encoding mouse CD4 or CD8 were ordered and cloned into the pLX304

expression system prior to retroviral production and transduction. Full-length TCR gene blocks

were ordered consisting of b chain -- P2A -- a chain and were cloned into the pMSCV-IRES-RFP

expression plasmid prior to retroviral transduction.

134

Animals. All animal studies were approved by the Washington University Animal Studies

Committee and all mice were housed in accordance with the Institutional Animal Care and Use

Committee standards. Male or female mice between the ages of 6-16 weeks of age were used for

tumor injections. TCR transgenic mice were used until 1 year of age. C57BL/6 and C57BL/6:129

hybrid mice were purchased from Taconic Biosciences. CD45.1 congenic, and RAG2 KO mice

were purchased from Jackson Laboratory. C57BL/6 IRF8+32kb-/- (Delta32) mice were obtained

from Kenneth Murphy.

Cell Lines & Media. GL261 was obtained from the National Cancer Institute Tumor Repository,

and CT2A was obtained from Dr. Peter Fecci (Duke University). All mImp3 overexpression lines

were generated by cloning a 132 bp segment of the mutated IMP3 gene into the pBabe backbone.

Transduction of target cell lines was performed as described335. The GL261:E8 clone was

generated by the Genome Engineering & iPSC Center (GEiC) at Washington University in St.

Louis. Briefly, guide RNAs were designed to target near the mutation site. The parental GL261

cells were nucleofected with CRISPR constructs and single strand DNA donors (ssODNs) carrying

the correction. Single cell GL261 clones were screened for the presence of the N81D correction.

All tumor cell lines were maintained in D10 media consisting of DMEM (Gibco) supplemented

with 10% FBS, 1% L-Glutamine (Corning), 1% MEM Nonessential Amino Acids (Corning), 1%

Sodium Pyruvate (Lonza Bioscience), and 1% Pen/Strep (Gibco).

58 hybridoma cells were obtained from David Kranz (University of Illinois Urbana-

Champaign) and cultured in media consisting of RPMI (Gibco) supplemented with 10% FBS,

0.5% HEPES (Corning), 1% Sodium Bicarbonate (Corning), 1% L-Glutamine (Corning), 1%

Pen/Strep (Gibco), and 50 uM b-Mercaptoethanol (Sigma-Aldrich).

135

Primary splenocytes were cultured in R10-BME media consisting of RPMI (Gibco)

supplemented with 10% FBS, 1% L-Glutamine (Corning), 1% Sodium Pyruvate (Lonza

Bioscience), 1% Pen/Strep, 0.5% Sodium Bicarbonate (Corning), and 50 uM b-Mercaptoethanol

(Sigma-Aldrich). When indicated, this media was supplemented with specific concentrations of

peptide and/or cytokines.

Single-Cell PCR & TCR Isolation. To provide cells for TCR isolation, three separate populations

of mImp3-specific T cells were generated. Splenocytes were taken from mice that had rejected

GL261 once following aPD-L1 therapy (1X), once following aPD-L1 therapy and an additional

two times from memory rechallenge (3X), or had been vaccinated in a prime-boost manner with

50 µg mImp3 peptide and 100 µg Poly(I:C) adjuvant (Vax). Each of these populations were

stimulated with 1 nM mImp3 peptide and 50 IU/mL rhIL-2 with a weekly replacement with fresh

naïve splenocytes as antigen presenting cells. Single mImp3-specific CD8 T cells were tetramer

sorted into PCR plates in a buffer of PBS with 0.1% BSA Fraction V (Sigma-Aldrich) and 200

U/mL RNase inhibitor (New England BioLabs) following 6-8 weeks of stimulation to enrich for

tetramer-positive cells.

To isolate mImp3-specific TCRs, we adopted a previously published single-cell PCR

protocol290. In brief, a two-step nested PCR reaction was performed in which the first RT-PCR

reaction included a pool of 41 Va and 39 Vb primers specific to the leader sequences of all possible

TCR a or b chains. Triton X-100 detergent (Sigma-Aldrich) was added to a concentration of 0.1%

for this first step to facilitate cell lysis. PCR product from this first reaction was diluted 1:100 prior

to separate second-step reactions for a or b chains. The specific reagents used and PCR reaction

conditions can be found in previously published work290.

136

The second-step PCR products were ran on a gel and isolated via QIAquick Gel Extraction

Kit (Qiagen) before undergoing sanger sequencing. Full-length TCR chains were generated by

combining sequencing results with IMGT reference sequences for a or b chain constant regions.

Gene blocks for each TCR were ordered (IDT) consisting of b chain – P2A - a chain and cloned

into the pMSCV-IRES-GFP (pMIG) (AddGene) backbone through Gibson Assembly (New

England BioLabs).

58 Hybridoma Functional Assays. Full-length TCRs for each candidate were constructed

utilizing the sanger sequencing results together with the IMGT reference for the constant regions

of the a and b chain. Gene blocks were ordered consisting of b chain -- P2A -- a chain and were

cloned into the pMSCV-IRES-GFP (pMIG) expression plasmid prior to retroviral transduction.

For the assessment of TCR functionality, TCR-expressing 250K 58 hybridoma cells were co-

cultured with 200K splenocytes loaded with peptide at varying concentration in 500 uL. After

overnight co-culture, 200 uL of supernatant were taken and analyzed for IL-2 production by the

Milliplex Mouse High Sensitivity IL-2 Multiplex Assay.

Generation of MISTIC Transgenic Mouse. Full-length TCR a or b chains for the 3x.1.1C TCR

were cloned into the transgene plasmids pCD2 and p428, respectively. Transgene fragments were

digested and isolated via gel extraction prior to resuspension in a filtered microinjection buffer.

Pronucleus injections were then performed by the Transgenic, Knockout, and Micro-Injection

Core at Washington University in St. Louis.

137

Transgenic T Cell Stimulation. Naïve splenocytes from transgenic mice were isolated by

mechanical dissociation and filtration through a 100-micron cell strainer. Mononuclear cells were

isolated through Ficoll-Paque PLUS density gradient (Cytiva) centrifugation and stimulated in

R10-BME media with 1 µM mImp3 peptide and 30 IU/mL rhIL-2. Cells were split 1:1 after 3 days

of stimulation. After 5 days of stimulation, CD8 T cells were isolated through negative selection

via EasySep Mouse CD8+ T Cell Isolation Kit (StemCell Technologies). The resulting cells were

then used for select in vitro functional assays or in vivo adoptive transfer studies.

Intracranial Injections. For intracranial tumor injections, cells were harvested following at least

one passage and having reached 60-90% confluency. 50,000 tumor cells in 5 uL PBS were injected

into the right hemisphere 2 mm posterior to bregma at a depth of 3.5 mm using a Stoelting

stereotactic headframe. Following tumor injection, mice were tracked daily and euthanized when

moribund.

Adoptive Cell Therapy. One day prior to receiving cell therapy, tumor-bearing mice received 5

Gy of total-body irradiation. Stimulation of naïve splenocytes from transgenic mice was performed

as described. After 5 days of stimulation and cell isolation, CD8 T cells were resuspended in PBS

at 2.5 x 107 cells/mL and 200 µL was given intravenously via the tail vein with a 27-gauge needle.

In addition, mice received daily injections of 180,000 IU of rhIL-2 intraperitoneally on the day of

cell transfer and for two days thereafter.

MRI Imaging. Following treatment on day 7, mice were serially imaged on a weekly basis

beginning on day 14 until day 35. After that point, imaging was performed on a biweekly basis.

138

All imaging was performed by the Small-Animal Magnetic Resonance Facility core at Washington

University in St. Louis.

Cancer Immunogenomics Approaches to Characterizing and Treating Malignant Brain Tumors

Students also viewed