Literature Review

Risk Analysis, Vol. 31, No. 10, 2011 DOI: 10.1111/j.1539-6924.2011.01610.x

An Exposure-Response Threshold for Lung Diseases and Lung Cancer Caused by Crystalline Silica

Louis Anthony (Tony) Cox, Jr.∗

Whether crystalline silica (CS) exposure increases risk of lung cancer in humans without sil- icosis, and, if so, whether the exposure-response relation has a threshold, have been much debated. Epidemiological evidence is ambiguous and conflicting. Experimental data show that high levels of CS cause lung cancer in rats, although not in other species, including mice, guinea pigs, or hamsters; but the relevance of such animal data to humans has been uncer- tain. This article applies recent insights into the toxicology of lung diseases caused by poorly soluble particles (PSPs), and by CS in particular, to model the exposure-response relation between CS and risk of lung pathologies such as chronic inflammation, silicosis, fibrosis, and lung cancer. An inflammatory mode of action is described, having substantial empirical sup- port, in which exposure increases alveolar macrophages and neutrophils in the alveolar ep- ithelium, leading to increased reactive oxygen species (ROS) and nitrogen species (RNS), pro-inflammatory mediators such as TNF-alpha, and eventual damage to lung tissue and ep- ithelial hyperplasia, resulting in fibrosis and increased lung cancer risk among silicotics. This mode of action involves several positive feedback loops. Exposures that increase the gain factors around such loops can create a disease state with elevated levels of ROS, TNF-alpha, TGF-beta, alveolar macrophages, and neutrophils. This mechanism implies a “tipping point” threshold for the exposure-response relation. Applying this new model to epidemiological data, we conclude that current permissible exposure levels, on the order of 0.1 mg/m3, are probably below the threshold for triggering lung diseases in humans.

KEY WORDS: Crystalline silica; dose-response model; exposure-response; lung cancer risk; mathemat- ical model; silicosis

1. INTRODUCTION: IS CRYSTALLINE SILICA HAZARDOUS AT CURRENTLY PERMITTED LEVELS?

Crystalline silica (CS) is one of the most stud- ied, yet most controversial, of substances currently classified as known human carcinogens.(1) Like other poorly soluble particles (PSPs), it has been as- sociated with a variety of possible lung diseases. In addition to silicosis, nonspecific responses such

Cox Associates, 503 Franklin Street, Denver, CO 80218, USA. ∗Address correspondence to Louis Anthony Cox, Cox Associates,

503 Franklin Street, Denver, CO 80218, USA; tel: 303-388-1778; fax: 303-388-0609; [email protected]

as chronic inflammation, fibrosis, lung cancer,(2,3)

and, possibly, chronic obstructive pulmonary disease (COPD)(4) have been suggested as possible conse- quences of high levels of exposure to CS and/or other dusts and respiratory irritants, including cigarette smoke.

Whether CS at currently permitted exposure lev- els (such as OSHA’s PEL-equivalent of 0.1 mg/m3 of respirable CS, or NIOSH’s currently recommended exposure limit of 0.05 mg/m3 for up to a 10-hour workday) creates an excess risk of lung disease has been much debated, but without clear resolu- tion. For decades, scientists, regulators, and occupa- tional health and safety risk managers have wrestled

1543 0272-4332/11/0100-1543$22.00/1 C© 2011 Society for Risk Analysis

1544 Cox

with the following three key questions about human health risks from CS exposures.

(1) Do the causal exposure-response relations be- tween CS exposure and exposure-associated lung diseases have thresholds?

(2) If so, are the exposure levels that cause in- creased risks of such diseases above or below currently permitted exposure levels?

(3) Are risks of some diseases (such as lung can- cer) elevated only at exposures that cause other diseases (e.g., silicosis)?

Expert opinions on all three questions have been sharply divided. Epidemiology, risk assessment, and toxicological research have done much to illuminate the difficulty of answering them decisively,(5,6) but have so far produced few unequivocal answers.

This article examines the causes and exposure- response relations for CS-associated lung diseases, drawing on recent advances in the biology of lung diseases caused by PSPs, which include CS as a special case. For PSPs, chronic inflammation of the lung plays a crucial role in causing lung diseases such as asbestosis, silicosis, fibrosis, COPD, and lung cancer.(2,7−12) We seek to shed new light on the exposure-response relation for CS-associated lung diseases by applying recent insights into this inflam- matory mode of action to model the relation between exposure concentrations and durations and the re- sulting cascade of changes in the lung environment that can hasten the onset and progression of lung diseases.

2. CS EPIDEMIOLOGY IS AMBIGUOUS

A number of epidemiological studies have re- ported that lung cancer risk is elevated among pa- tients with silicosis, especially among those who smoke.(13−15) Others find no such association,(16−20)

and a recent meta-analysis concluded that the associ- ation disappears when confounders (such as smok- ing or occupational coexposures) are correctly ad- justed for.(6) Influential investigators have stated that risks of lung cancer appear to them to be el- evated even at exposure levels below current stan- dards.(21,22) However, we believe that failure to correctly account for exposure measurement errors invalidates this interpretation of the data, as ex- plained below (see Fig. 1). Risk of COPD and re- duced lung function appear to be elevated at esti- mated occupational exposures above 0.1–0.2 mg/m3

of silica dust for at least 30–40 years, independent

of silicosis,(4) but a recent study of Vermont granite workers found no evidence of increased lung cancer risk due to silica exposure in occupational cohorts, even at the high exposure levels where mortali- ties due to silicosis and other nonmalignant res- piratory illnesses were elevated.(23) The apparent paradox of reduced risk of lung cancer in some workplaces with relatively high levels of silica ex- posure has also been noted,(24) further complicat- ing any conjectured causal relation between silica exposure and lung cancer. One possible explana- tion for these differences among studies might be the different (and often highly uncertain) composi- tions of the dusts in different studies.(25) For exam- ple, the toxicity of quartz particles depends on de- tailed properties of the fracture surfaces, with freshly fractured silica typically being more potent than aged silica in elicting various cellular responses, includ- ing production of reactive oxygen species (ROS) by alveolar macrophages.(26) Differences in dust com- position and ages might therefore create heteroge- neous exposure-response relations, perhaps trigger- ing different response mechanisms. In this case, bi- ologically effective doses could be very uncertain, even if respired quantities of dust were measured accurately.

Whether or not silicosis increases lung cancer risk, epidemiological studies have not yet revealed whether silicosis is a necessary precondition for in- creased risk of lung cancer due to CS exposure.(6,27)

Yet, the answer is vital for current practical regu- latory risk management decisions: “If silicosis were the necessary step leading to lung cancer, enforcing the current silica standards would protect workers against lung cancer risk as well. Alternatively, a di- rect silica-lung cancer association that has been sug- gested implies that regulatory standards should be re- vised accordingly.”(24)

Somewhat reassuringly, the increased risk of lung cancer among CS-exposed workers is most ap- parent “when the cumulative exposure to silica is well beyond that resulting from exposure to the recommended limit concentration for a prolonged period of time,”(28) suggesting that enforcing cur- rent standards would protect workers from CS- associated lung cancer risks. However, other re- searchers have cautioned that, “The hypothesis of a silicosis-mediated pathway [for lung cancer], al- though more consistent from an epidemiological per- spective, and reassuring in terms of the effective- ness of current standards in preventing lung cancer risk among silica exposed workers, does not seem to

Lung Diseases and Lung Cancer Caused by Crystalline Silica 1545

explain elevated risks at low silica exposure lev- els.”(29) Thus, the relation between silicosis and lung cancer has remained uncertain, based on vari- ous published interpretations of epidemiological ev- idence. There is no clear evidence that lung can- cer risk is elevated in the absence of silicosis, but the question is unsettled. The following statement(27)

succinctly captures the present state of the art: “A re- cent meta-analysis of 30 studies found a pooled rela- tive risk (RR) of lung cancer of 1.32 (95% CI, 1.23– 1.41) in subjects exposed to CS. In the same investi- gation, the pooled RR was 2.37 (95% CI, 1.98–2.84) in silicotics only (based on 16 studies), whereas no in- crease in risk emerged in non-silicotics (pooled RR = 0.96, 95% CI, 0.81–1.15, based on eight studies). The authors concluded that silica may induce lung can- cer indirectly, probably through silicosis.” Such evi- dence, although not conclusive, favors the hypothesis that lung cancer risk is elevated among silicotics, but not among nonsilicotics.

We believe no credible epidemiological evidence actually shows that CS increases lung cancer risk at exposure levels that do not also cause silico- sis. Rather, the foregoing observation that the “hy- pothesis of a silicosis-mediated pathway . . . does not seem to explain elevated risks at low silica expo- sure levels,” as well as published reports of elevated risk of lung cancer at exposures below those that cause silicosis,(21) misinterpret the available epidemi- ological evidence. They do so by mistakenly inter- preting exposure-response relations estimated from epidemiological studies (all of which have missing and highly uncertain and variable (usually, “recon- structed”) exposure data) as providing valid evidence of “elevated risks (of lung cancer) at low silica ex- posure levels.” But they do not. At most, such stud- ies provide evidence of elevated lung cancer risks at low estimated levels of silica exposure. These are en- tirely different propositions, as explained next. When uncertainties in exposures are accounted for in the risk models, there is no evidence that risks are el- evated at low levels of silica exposure (specifically, at or below those allowed by current standards). Studies that conclude that relatively low exposures to silica (below currently permitted levels, and be- low levels that cause silicosis) increase lung cancer risk are undermined—without exception, as far as we know—by important upward biases in their low- exposure risk estimates. These biases result from imperfect control of potential confounders, ignored model specification errors and uncertainties, and un- modeled errors and uncertainties in exposure esti-

mates. Each of these limitations is briefly discussed next.

2.1. Imperfectly Controlled Confounding

Perhaps the most familiar threat to valid in- ference from epidemiological studies of CS is con- founding, especially by cigarette smoking and by occupational co-exposures. For example, a recent study(30) reported that: “In a crude analysis adjusted for smoking only, a significant trend of increasing risk of lung cancer with exposure to silica was found for tin, iron/copper miners, and pottery workers. But af- ter adjustment for relevant occupational confounders (arsenic and polycyclic aromatic hydrocarbons), no relationship between silica and lung cancer can be observed.”

The possibility of such confounding has been well recognized and much discussed in the epi- demiological literature on CS, but inability to rigor- ously and fully control for plausible confounders in most past studies continues to limit the validity of the exposure-response relations inferred from these studies.(6) Attempts to adjust for possible confound- ing by smoking, based on subjective estimates of smoking habits and their effects (and an assumed bias model), have modestly reduced the estimated relation (standardized mortality ratio) for silica ex- posure and lung cancer (from 1.6 to 1.43).(21) Other assumptions and models might lead to further re- ductions. Currently proposed methods to account for most of the bias due to confounding by smoking, using differences between COPD and lung cancer rates to estimate bias effects,(31) have not yet been applied to CS, leaving open the question of how much of the apparent relation between CS exposure and lung cancer risk would be eliminated by fully controlling for smoking effects. Similarly, it remains unknown whether fully controlling for occupational co-exposures would fully eliminate the apparent as- sociations between silica exposure and lung cancer risk (in other data sets as well as the one for Chi- nese miners and pottery workers), since most other studies have not provided the needed co-exposure data.(30)

2.2. Unmodeled Errors and Uncertainties in Exposure Estimates Can Inflate Low-Exposure Risk Estimates and Hide True Thresholds

Perhaps the single most important limitation in CS epidemiology is that true individual exposures

1546 Cox

to CS of various types and toxicities are unknown. Therefore, guesses about exposures are used instead, typically based on reconstructions of exposure histo- ries from estimated job exposure matrices, together with simplifying (and inaccurate) assumptions, such as that all silica dust has the same average toxicity or carcinogenic potency value. Exposure-response re- lations are then fit to the guessed-at exposures and observed responses. Although there is a sophisti- cated statistical literature on how to use such un- certain predictors in regression models,(32) these ap- propriate “errors-in-variables,” measurement error, and missing data methods have typically not been used in the CS epidemiology literature. Instead, re- constructed exposure estimates are often treated as if they were true (error-free) data, for purposes of fitting statistical models. Then, unwarranted conclu- sions are drawn that fail to explicitly model and cor- rect for the effects of errors in exposure estimates.(33)

This can create large, unpredictable biases in multi- variate regression coefficients and other measures of exposure-response association.(34)

If the true exposure-response relation is a thresh- old function, then failing to explicitly model errors and uncertainties in exposure estimates can smear out the threshold in the estimated exposure-response models, giving a misleading appearance of a smooth, s-shaped exposure-response function, complete with an apparent (but not real) smooth biological gradi- ent (i.e., higher probabilities of response at higher estimated exposure levels) and elevated risks at esti- mated exposure levels well below the true threshold. Such incorrect modeling will over-estimate excess risks at exposures below the threshold, and underes- timate risks at exposures greater than the threshold.

To illustrate how a smoothly increasing esti- mated exposure-response relation arises from a true threshold relation when there are unmodeled errors in the exposure estimates, consider the following sim- ple hypothetical example. Suppose that true individ- ual exposure rates are uniformly distributed between 0 and 20 mg/m3-years (for 40-year exposure dura- tions), and that the true exposure-response relation has a threshold at 15 mg/m3-years, so that the true risk of lung cancer is 0 for exposures of 15 mg/m3- years or less, and 1 for exposures above 15 mg/m3- years. Suppose that estimates of individual exposures are unbiased, but with some variance around their means, representing estimation errors. For simplic- ity, assume that the ratio of the estimated exposure to the true exposure, for each individual, is uniformly distributed between 0 and 2, with a mean value of

1 (i.e., Estimated exposure = k × True exposure, where k is a random variable, k ∼ U[0, 2], with E(k) = 1). Table I shows true and estimated expo- sures for 10 individuals, based on this simple model of errors in exposure estimates. Fig. 1 shows the es- timated exposure-response relation based on 10,000 individuals.

(For plotting purposes, each estimated exposure is rounded to the nearest integer, from 0 to 40.) The estimated exposure-response relation suggests that risk increases with exposure over the entire range of exposure values, and that it is slightly but signif- icantly elevated even at relatively low exposure lev- els (e.g., 3 mg/m3-years), even though we know that, in this example, the true exposure-response relation has no increase in risk at exposure rates below 15 mg/m3-years. This same conceptual point holds for real data, provided that estimated exposures con- tain errors. However, for real data, we do not know what the correct exposure-response relation is. The use of estimated individual exposures tends to smear out the true but unknown exposure-response rela- tion (e.g., turning a sharp threshold into a gradu- ally increasing curve, as in Fig. 1, or turning a nar- row distribution of individual thresholds into a wider one). Recovering the correct exposure-response re- lation requires additional analysis to correct for this smearing effect by explicitly modeling the relation between true and estimated exposures.(32,35,36) Es- timated exposure-response relations for CS in the epidemiological literature have not made such cor- rections, and therefore they do not provide useful information about possible true exposure-response thresholds or trustworthy evidence that risks at low exposures are truly elevated.

2.3. Model Specification Errors and Uncertainties Can Obscure Threshold Relationships

Many CS epidemiology studies fit parametric sta- tistical models to estimated exposure-response data, and then interpret the estimated model parameters (e.g., odds ratios or regression coefficients) as pro- viding evidence of a positive effect at all exposure levels. This procedure is not justified if different mod- els hold at different exposure levels, as could be the case if there is an exposure threshold, with no in- crease in risk below the threshold and some increase above it.

The assumptions built into a statistical model can drive its conclusions, even if these disagree with the data used to fit the model. As an extreme,

Lung Diseases and Lung Cancer Caused by Crystalline Silica 1547

Table I. Hypothetical Data for True and Estimated Exposures and Resulting Responses

True Exposure Random Multiplier Estimated Exposure Response ∼ U[0, 20] k ∼ U[0, 2], E(k) = 1 = k × ∗True Exposure Threshold Response

1 0.14 1.4 0.19 15 0 2 6.07 0.7 4.30 15 0 3 18.54 0.0 0.75 15 1 4 7.54 1.6 11.99 15 0 5 19.85 0.6 11.31 15 1 6 17.89 0.4 7.52 15 1 7 9.20 1.6 14.74 15 0 8 7.72 1.0 7.77 15 0 9 5.41 1.2 6.75 15 0 10 15.13 0.1 1.81 15 1

Estimated response probabilities and 95% confidence intervals

Response

0 3 6 9 12 15 18 21 24 27 30 33 36 39

Estimated Exposure

0.0

0.2

0.4

0.6

0.8

1.0

R is

k =

P r(

re sp

o n

se )

= E

(r e

sp o

n se

)

Fig. 1. Estimated exposure-response relation for the simulated data in Table I (using 10,000 individuals instead of 10). The correct relation has a threshold at 15: risk = 0 for exposure ≤ 15; risk = 1 for exposure > 15.

hypothetical, example, fitting the regression model Risk = β × Exposure to data that are correctly de- scribed by Risk = 1/Exposure would produce a pos- itive estimate for β, which might be misinterpreted as a positive unit risk factor or potency for the ef- fect of exposure on risk, even though the true relation Risk = 1/Exposure shows that risk actually decreases with increasing exposure. This illustrates how a mis- specified statistical model can override data, and pro- duce a conclusion that risk is increased at low expo- sure levels, even if the data imply nothing of the sort.

To avoid such model specification errors and biases, it is useful to fit nonparametric models to exposure-response data. Fig. 2 presents an example: a spline curve fit to estimated exposure-response data in the influential IARC pooled analysis study.(21) The

authors interpreted this model as “support[ing] the decision by the IARC to classify inhaled silica in oc- cupational settings as a carcinogen, and suggest[ing] that the current exposure limits in many countries may be inadequate.” The y-axis shows estimated RR of lung cancer, with 1 corresponding to no effect.) The authors interpreted Fig. 2 as follows: “Analy- ses using a spline curve also showed a monotonic in- crease in risk with increasing exposure.” However, a more accurate description is that Fig. 2 shows clear evidence of a threshold, with no increase (and, if any- thing, a slight decrease) in risk at low exposure levels.

This finding of an apparent threshold can be converted to a reported finding of a “monotonic increase in risk,” by fitting a parametric statisti- cal model (such as Risk = β × Exposure, having

1548 Cox

Source: Figure from Reference 21.

Fig. 2. A spline curve fit to pooled analysis data suggests a threshold.

parameter β, in the above example), which guar- antees a positive estimate of β (as long as Risk and Exposure values are positive), and hence a monotonic increase in estimated risk even at low exposures, no matter what the data say. (The slope parameter β is necessarily positive when both Risk and Exposure are positive, since the line Risk = β × Exposure necessarily goes through the origin at its lower left, and slopes upward through the posi- tive scatter plot.) The IARC team interpreted the data behind Fig. 2 this way. They fit a similar para- metric model (log relative risk = β × Exposure) to data with positive values of Exposure and log relative risk, and therefore (necessarily) concluded that risks were increased at low exposure levels—a finding that they interpreted as supporting classification of CS as a known human carcinogen that might need tighter regulation. Fig. 2 suggests that a less assumption- laden process could have produced a very different conclusion, that is, that the data do not indicate any increase in risk at low exposures.

In summary, epidemiological evidence on CS and lung cancer have often been interpreted as sug- gesting a causal relation between CS exposure and increased risk of lung cancer,(22) even at relatively low exposure levels that do not cause silicosis. Our

review of CS epidemiology indicates that this inter- pretation is unjustified. CS epidemiological studies and meta-analyses have not corrected for errors in in- dividual exposure estimates, have not applied appro- priate methods to estimate and fully control for con- founding, and have not accepted and interpreted at face value the results of nonparametric analyses that provide clear, model-free, evidence of an exposure- response threshold. As a result, past epidemiological studies do not provide trustworthy information about the presence or absence of thresholds in exposure- response relations, or about the shape of individual or population exposure-response functions. To ob- tain more insight, it is necessary to turn to biological information about how and under what conditions CS increases risks of lung diseases.

3. CS MODE OF ACTION

Over the past decade, molecular biologists and toxicologists have dramatically improved understanding of how PSPs in general, and CS in particular, cause lung diseases. The following steps, reviewed in more detail in Cox for COPD,(10)

are important in the development of many PSP exposure-related lung diseases.

Lung Diseases and Lung Cancer Caused by Crystalline Silica 1549

(1) Sufficient exposure activates alveolar macrophages (AMs) and changes their phenotypes. Intense and prolonged exposure to many PSPs permanently shifts AM popula- tions toward more cytotoxic phenotypes with reduced phagocytic capacity and reduced ability to clear apoptotic cells via efferocy- tosis.(11) For CS, AMs are activated via the MARCO receptor, which plays a crucial role in CS particle recognition and uptake.(12,37)

A shift in AM phenotypes and reduced AM phagocytic capacity has been documented for silica-exposed monkeys,(38) as well as for rodents.(37)

(2) The altered AMs produce increased levels of ROS, reactive nitrogen species (RNS), and pro-inflammatory cytokines, including TNF- α. Exposure to PSPs increases AM production of ROS. Although increases in ROS produc- tion may initially be counterbalanced by com- pensating increases in antioxidants (AOX) (see Ref. 39 for silica and Ref. 40 for a more general overview) sufficient exposure overwhelms and down-regulates AOX in rats, shifting the oxidant-antioxidant balance in the lung toward abnormally high ROS levels and generating oxidative stress.(2) Mechanisms of antioxidant reduction in human bronchiolar epithelial cells (BECs) have started to be elu- cidated in vitro,(41) although more remains to be done (e.g., to clarify the role of the Nrf-2 “master switch” for many antioxidants, and its pathways, such as the Nrf-2-ERK- MAP kinase–heme oxygenase (an antioxi- dant) pathway).(42,43)

(3) A high-ROS environment, in turn, induces AMs (and, to a lesser extent, other lung cell populations, such as BECs) to secrete more pro-inflammatory mediators—most no- tably, tumor necrosis factor alpha (TNF-α), as well as IL-1β, TGF-β1, and other pro- inflammatory cytokines.(44) For CS specif- ically, exposure increases AM production of both ROS and RNS in rats(45) and ac- tivates signaling pathways (including NF- kappaB and AP-1) that promote expression of pro-inflammatory mediators, oncogenes, and growth factors important in lung fibro- sis and cancer.(76,77) Increased ROS stimulates increased secretion of TNF-α by AMs, as ob- served in vivo in silica-exposed rats(78) and in vitro in silica-exposed lung cell lines, in which

ROS activates a specific transcription factor (nuclear factor of activated T cells [NFAT]) that increases TNF-α.(79)

In humans, ROS markers such as 8- isoprostane remain elevated, or increase, in patients with silicosis(80) or COPD(10) even long after exposure stops, suggesting that ex- posure “switches on” a self-sustaining pro- cess (e.g., a positive feedback loop) that keeps ROS permanently elevated. The increase in ROS levels and oxidative stress in the lung environment is considered crucial in caus- ing subsequent exposure-associated lung in- jury and in increasing risk of lung diseases, including fibrosis,(45) silicosis, and lung can- cer.(2,12,46−48)

(4) Increased TNF-α and ROS stimulate an in- flux of neutrophils to the lung. Some specific causal pathways by which TNF-α and ROS at- tract neutrophils into the lung have been par- tially elucidated, as follows.

• TNFα up-regulates interleukin 8 (IL-8) ex- pression.(49) IL-8 (also called CXCL8 lig- and) is a potent chemoattractant for neu- trophils. It recruits additional neutrophils to the lung, via chemotaxis, and activates them (by binding with high affinity to the two chemokine receptors, CXCR1 and R2, on the neutrophil cell surface, stim- ulating their degranulation).(50) The lungs contain a large reservoir of marginated neutrophils, sequestered within the tiny capillaries of the pulmonary microcircu- lation and adhering to the capillary lin- ing (endothelium). In response to IL-8, they squeeze across the alveolar–capillary membrane and into the interstitial air spaces. (How quickly this happens depends on the deformability of the neutrophils, which depends on oxidant–antioxidant bal- ance.(51) IL-8 also increases the cellular ad- hesion of neutrophils (specifically, to fib- rinogen and ICAM-1) via the β2-integrin cell surface adhesion molecule, Mac-1, i.e., CD11b/CD18.(52)) Thus, IL-8 increases the local concentration of activated lung neu- trophils, both by attracting and by retain- ing them. This may be diagrammed as: IL-8 → N (where the arrow indicates that an increase in the quantity on its left (tail) increases the quantity on its right (head).)

1550 Cox

• ROS increases the release of IL-8 from cul- tured macrophages. Specifically, the lipid peroxidation product 8-isoprostane (which is elevated in COPD patients, as well as in the plasma and urine of atherosclerosis pa- tients) increases IL-8 expression in human macrophages in vitro (via a pathway that in- volves both ERK 1/2 and p38 MAPK, but not NF-kappaB).(53)

• ROS also increases IL-8 via the following ROS-EGFR pathway:(10) ROS → TGF-α → EGFR phosphorylation → IL-8, VEGF, MUC5AC, MUC5B (where, again, each ar- row indicates that an increase in the quan- tity on the left (tail) increases the quantity on the right (head) of the arrow). This path- way also increases mucus production in air- ways, via increased expression of the mucin genes MUC5AC and MUC5B. IL-8 is pro- duced by BECs, dendritic cells, and other lung cell populations, following EGFR ac- tivation.

• TNF-α and ROS may also stimulate release of the ligand CXCL2 (i.e., C–X–C motif ligand 2, also called macrophage inflam- matory protein 2-alpha [MIP2-α]), as well as of growth-regulated protein beta (Gro- beta) and Gro oncogene-2 by dendritic cells (DCs), monocytes, and macrophages. CXCL2 is chemotactic for neutrophils, en- hancing their influx into the airways(54)

for murine cells in vitro; see Thatcher et al.(55) for CXCR2 effects on emphysema in smoke-exposed mice in vivo.

In rats exposed to CS, the initial influx of AMs and neutrophils leads to elevated levels of both that persist many months after exposure ceases.(56)

(5) The increased neutrophils and AMs in the lung generate increased ROS levels and ox- idative stress, due in part to their respiratory bursts; in part to the release of neutrophil elastase (NE) from neutrophils; and in part to greatly increased numbers of apoptotic cells (primarily neutrophils, but also AMs and ep- ithelial cells). This completes a positive feed- back loop: ROS → TNF-α from AMs → IL-8 → neutrophils → ROS. NE also fur- ther activates the EGFR pathway (by cleav- ing pro-TGF-α, which stimulates release of mature TGF-α that binds to and phosphory- lates EGFR), and potently stimulates goblet

cell degranulation, contributing to mucus hy- persecretion into the airways.(57) This creates the following positive feedback loop: TGF- α → EGFR phosphorylation → IL-8→ neu- trophils → NE → TGF-α. Activated neu- trophils further amplify the EGFR pathway and inflammation by releasing TNF-α, which increases expression of EGFR on airway ep- ithelial cells.(57) Increases in NE can shift an entire protease–antiprotease network toward a new, high-protease state in which the excess proteases digest lung tissue and cause emphy- sema and COPD, as well as increasing apop- tosis of endothelial and epithelial cells.(10)

(6) High ROS and oxidative stress increase apop- tosis of AMs, neutrophils, and alveolar ep- ithelial cells, leading to lung tissue damage and destruction. Apoptosis of alveolar epithe- lial cells, together with damage to the ex- tracellular matrix (ECM) and alveolar wall from increased proteases, can eventually lead to tissue destruction and remodeling of the ECM, including deposition of collagen lead- ing to scarring and fibrosis in human silico- sis(58) and COPD.(10) Experiments with silica- exposed knockout mice have confirmed that both IL-1β and inducible nitrogen oxide syn- thase (iNOS) are involved in apoptosis and inflammation during murine silicosis.(59) In- creased ROS leading to increased apoptosis of alveolar cells and neutrophils has been ob- served in CS-exposed rats.(60,61) Damaged and dying alveolar epithelial cells (especially Type II alveolar cells) cause the lung parenchyma to secrete, activate, and release transform- ing growth factor beta-1 (TGF-β1), as well as more TNF-α (thus completing still fur- ther positive feedback loops: ROS → TNF- α → IL-8 → neutrophils → ROS → apop- totic cells→ TNF-α). Apoptotic cells (and, even more, necrotic cells, which form if apop- totic cells are not promptly and safely re- moved) also release high levels of ROS into the lung environment. TGF-β1 activates fi- brogenic cells and powerfully attracts AMs (which release more TGF-β1) and other in- flammatory cells (neutrophils and lympho- cytes) into parenchymal tissues.(62) ROS and TGF-β1 stimulate production of new ECM by myofibroblasts, the fibrotic lung’s major collagen-producing cell population.(62) High oxidative stress also decreases the ability of

Lung Diseases and Lung Cancer Caused by Crystalline Silica 1551

AMs to identify and remove apoptotic cells, further increasing their concentration, and hence the concentration of ROS and TGF-β1 in the lung environment.

(7) In rats, damage to lung tissue and altered apoptosis result in epithelial hyperplasia, clonal expansion of preneoplastic cells that would ordinarily be removed via apopto- sis, and increased risk of lung cancer. Ox- idative stress from a high-ROS lung envi- ronment can both reduce apoptosis among some cells (thereby increasing lung cancer risk, if preneoplastic cells are less likely to be detected and removed via apoptosis) and stimulate proliferation and transformation of cells that contribute to increased lung cancer risk.(2) For CS specifically, exposure causes hyperplasia of epithelial cells and fibroblasts in rats, but CS does not induce similar hy- perplasia (or lung cancer) in mice and pri- mates.(7) CS induces hyperplasia of both neu- roendocrine lung cells(63) and Type II alveo- lar cells in rats, although not in mice or ham- sters.(64,81) In rats (but, again, not in mice or hamsters, which do not show elevated lung cancer risk in response to CS exposure), TGF- β1 precursor is localized in hyperplastic alve- olar type II cells and ECM next to granulo- mas (and adenomas, if any).(64,65) This sug- gests a close link between locations of alve- olar cell death and attempted repair of ECM (both of which are associated with TGF-β1) and areas of increased hyperplasia/adenomas. Such usefully detailed biomolecular infor- mation links the process of silicosis (e.g., TGF-β1–mediated collagen production, ECM remodeling, epithelial–mesenchymal transi- tion,(66) and fibrosis) directly to epithelial cell proliferation and increased lung cancer risk (due to increased hyperplasia/adenoma of damaged lung tissue)—the crucial link that epidemiological data alone could not yet provide.

Studies of silica-induced lung cancer in rats—the only species in which CS exposure is known to cause lung cancer—indicate that CS does not act through classical mutational (e.g., KRAS or EGFR mutation) pathways for lung cancer, but rather promotes lung carcinogenesis through indirect epigenetic processes associated with increased proliferative stress and hy-

permethylation of the promoter region of tumor sup- pressor genes (TSGs), specifically including p16.(9)

In humans, aberrant promoter methylation of TSGs is more frequent in serum DNA from silicosis pa- tients with lung cancer than in silicosis patients with- out lung cancer,(67) suggesting that epigenetic gene silencing of TSGs by this mechanism may be rele- vant in silicosis-associated lung cancers in humans, as well as in rats. The p16 gene normally participates in checking and regulating cell division (as part of the p16INK4a-Cyclin D1-CDK4-RB cell cycle con- trol axis).(68) Disruption of p16 gene expression al- lows damaged cells that would normally be removed via apoptosis to undergo mitotic replication instead, increasing the prevalence of damaged (potentially preneoplastic) cells in lung bronchiolar epithelial tis- sue. Epigenetic silencing of p16 by CS-induced hy- permethylation of its promoter region thus presum- ably increases survival and entry of altered (initiated) cells into a clonal expansion phase, thereby promot- ing expansion of preneoplastic cell populations and increasing the risk of lung tumors.(69)

In summary, CS exposure stimulates produc- tion of ROS/RNS, down-regulates counterbalancing antioxidants, and activates immune cells, including AMs (as well as mast cells, and B-lymphocytes).(12)

Activated immune cells release more ROS, cre- ating a positive feedback loop.(2),(7) The resulting high-ROS, chronically inflamed lung environment disrupts normal apoptosis and repair of epithelial and endothelial cells, increases epithelial cell pro- liferation and lung cancer risk, inhibits normal re- pair of damaged epithelial tissue, and promotes ex- cess secretion of collagen and other proteins in the ECM. In rats, and probably in silicosis patients, these changes promote expansion of preneoplastic clonal patches and increase risk of lung cancer, probably in part by epigenetic silencing of TSGs, such as p16. These general features of lung disease processes hold for many PSPs and mineral dusts and fibers, and for CS in particular, as documented in the cited references, although important biochemical details (such as the specific antioxidants generated in re- sponse to initial ROS increases) differ for different compounds.(39)

4. EXPOSURE-RESPONSE MODELING

Although the inflammatory mode of action is complex, one of its main features is obvious: the key quantities and the regulatory relations among

1552 Cox

CS exposure

AM influx & activation EGFR TGF- NE

ROS TNF- from AMs IL-8 neutrophil influx ROS AM influx TNF- from damaged cells apoptotic cells TGF- 1

Fig. 3. Examples of positive feedback loops in a silica disease causal network.

them form a network with multiple positive feed- back loops. Fig. 3 shows examples. In each loop (i.e., each directed cycle among a set of variables, with arrows entering and leaving each variable in it), an increase in one element stimulates an increase in its successor, so that eventually all variables around the loop increase. (Fig. 3 is not intended to be com- plete, e.g., it does not show the direct contribution of CS fragments to ROS, the shift in AM pheno- types toward less effective phagocytosis, the pro- duction of collagen by fibroblasts, or many other biological effects previously discussed. It simply illus- trates some major positive feedback loops involved in CS-associated (and other PSP-associated) lung pathologies.)

If specific quantitative formulas linking the rates of changes of different variables were known, then the dynamic response of such a network to changes in its exogenous inputs (such as CS exposure, in Fig. 3) could be simulated. Even without such detailed quantitative information, however, the method of comparative statics analysis(82) can be used to study how equilibrium levels of variables change in re- sponse to exposure. The basic idea is to compute how equilibrium points change, even though the details of the adjustment process may be (and, for CS, still are) largely unknown. To do this, we focus on some vari- able, such as ROS, that appears in one or more loops. Let’s call the selected variable X. Now, consider the following artificial adjustment process, which is con- structed so that it will lead to the same equilibrium levels of X as the real but unknown adjustment pro- cess. (Throughout, we assume, realistically, that all modeled variables are bounded, and that they adjust to their new equilibrium levels (or quasi-equilibrium levels, for slowly changing variables), in response to any change in inputs, relatively quickly—well within the lifetime of the exposed individual. These assump- tions hold for the variables in more detailed models of COPD.(10)) The artificial adjustment process is it- erative. Each iteration consists of the following two steps.

(i) Hold X fixed at a specified level, denoted by Xt at iteration t. Let all other variables adjust until they are in equilibrium with Xt .

(ii) Next, hold all other variables fixed at their new levels, and let X adjust until it is in equi- librium with them. Denote by X t+1 this new value of X.

If the system were understood in enough detail to allow a full, explicit, dynamic simulation model to be constructed, then the mapping from each value of Xt to the corresponding value of X t+1 could be evaluated numerically. Even without such complete knowledge, we can denote this mapping by some (un- known) function, f , and consider its qualitative prop- erties. By construction, equilibrium values of X (de- fined as values such that X t+1 = Xt ) in the dynamic system are also fixed points of the artificial adjust- ment process represented by f . The model

Xt +1 = f ( Xt ) corresponds to a curve, which we call a model curve, in a graph that plots X t+1 against Xt , as shown in Fig. 4.

Fig. 4 actually shows three different model curves, 1–3, corresponding to successively greater ex- posure levels and/or sensitivities of exposed individu- als. For model curves 1 and 2, there is a unique, glob- ally stable equilibrium value of X, denoted by X ∗, where the model curve intersects the equilibrium line (defined by the 45◦ line X t+1 = Xt ) from above and to the left. This equilibrium is stable because Xt +1 > Xt to its left and Xt +1 < Xt to its right. In other words, if Xt differs from X ∗, then the levels of other vari- ables that are affected by Xt will not adjust to lev- els that sustain Xt , but instead will reach levels that, in turn, cause Xt to move closer to X ∗. Such a glob- ally stable equilibrium represents the normal, home- ostatic equilibrium for the system when no disease is present. Model curve 2 differs from Model curve 1 by showing saturation of X at its right end, that is, a maximum possible level of X. Even a high level of

Lung Diseases and Lung Cancer Caused by Crystalline Silica 1553

Xt X*

Model curve 1: Exposure = 0

Xt+1 equilibrium line: Xt+1 = Xt

Saturated level of X

Model curve 3: Exposure >> 0

Tipping point threshold

X** X ′

Model curve 2: Exposure > 0

Fig. 4. Exposures high enough to destabilize a feedback-control loop create an alternative equilibrium (potential disease) state (X

∗ ∗ ) and a

threshold (X′).

exposure will not lead to an infinite level of X, but will, at most, saturate the response of the feedback loop(s) containing X, sending the affected variables to their maximum levels.

Model curve 3 shows a qualitatively different possibility for an exposed individual for whom the saturated level of X is high enough to intersect the equilibrium line from above and to the left. For such an individual, there are two alternative equilibria: the normal homeostatic equilibrium at X ∗, and an al- ternative, locally stable equilibrium X ∗∗, with X at its saturated level. In between them, for any con- tinuous model curve, there must be a threshold or “tipping point,” denoted by X’ in Fig. 4, such that X will adjust toward X ∗ from any starting point to the left of X′, but will adjust toward X ∗∗ from any starting level to the right of X′. That is, X ′ is an unstable equilibrium separating the two basins of at- traction for the “healthy equilibrium” X ∗ and the potential “disease equilibrium” X ∗∗. (Topologically, such a threshold must exist whenever two alterna- tive stable equilibria exist, for any continuous model curve; it is unique if the model curve is s-shaped.) As explained in detail for a specific parametric model of COPD (consisting of a system of ordinary differ- ential equations and algebraic equations with esti- mated parameter values),(10) exposure that increases a model curve enough to produce a saturated equilib- rium (such as X ∗∗ in Fig. 4) does so by destabilizing the positive feedback loop(s) containing X, causing its variables to escalate until saturation is reached.

For a biological interpretation, suppose that X represents ROS, and that the mechanism by which

long-term exposure increases the model curve is to shift cell populations (such as AMs) toward phe- notypes that produce higher levels of ROS (and/or higher levels of the causal drivers of increased ROS in Fig. 3). Then X ∗∗ represents a high-ROS equilib- rium, in which ROS and all the other variables in Fig. 3 (which participate in positive feedback loops with ROS) have increased levels. If long-term ex- posures produce a model curve with two alternative equilibria (such as model curve 3), and if short-term exposure transients can then temporarily increase the level of X, then any exposure history that in- creases X past its tipping-point threshold will trigger a self-sustaining escalation in levels of X (and of all other variables that participate in a positive feedback loop with X, including all variables shown in Fig. 3) until the high-ROS (saturated-equilibrium) state is reached. If defensive and repair resources are insuf- ficient to counter the damage done in this high-ROS state, then tissue destruction and other clinical man- ifestations of lung disease may result. The threshold model in Fig. 4 predicts that progression to the high- ROS potential disease state will occur, even in the absence of further exposure, once the tipping point has been passed.

The preceding threshold model is motivated by current understanding of the biology of lung re- sponses to PSP exposures in general, and to CS exposures in particular, but it does not require de- tailed knowledge of the biological mechanisms in- volved, many of which remain uncertain. For exam- ple, with sufficient knowledge and data, each of the links between variables in Fig. 3 could be further

1554 Cox

elucidated, perhaps expanding into an entire sub- network showing molecular-level details of how an increase in the variable at the tail of an arrow prop- agates through signaling pathways and other mecha- nisms to cause an increase in the variable at the ar- row’s head. But such a detailed description would not change the basic topology of the network, nor its properties derived from the fact that multiple posi- tive feedback loops dominate its qualitative behav- ior. The exposure-response threshold in Fig. 4 does not depend on such details, and hence is robust to uncertainties about them. Although further biologi- cal information may eventually allow more detailed simulation and prediction of the time courses of lung disease initiation and progression, it should leave in- tact the insights that comparative statics analysis, of the type performed in this section, provides today.

4.1. Confirmatory Data: How Well Does the Theory Match Observations?

The analysis of alternative equilibria in Fig. 4 im- plies the existence of an exposure threshold, below which lung damage is largely reversible (although the homeostatic equilibrium X ∗ can be shifted rightward if exposure shifts the whole model curve up), and above which escalation of ROS, and of the other vari- ables in Fig. 3, to permanently elevated levels will progress, even without further exposure. It is use- ful to compare this theoretical prediction to avail- able data, which come largely from a series of stud- ies in rats, undertaken by NIOSH. Porter et al.(70)

found experimentally that “the time course of rat pulmonary responses to silica inhalation as bipha- sic, [with] the initial phase characterized by increased but controlled pulmonary inflammation and dam- age. However, after a threshold lung burden was exceeded, rapid progression of silica-induced pul- monary disease occurred.” They reported: “During the first 41 days of silica exposure, we observed ele- vated but relatively constant levels of inflammation and damage, with no fibrosis. Subsequently, from 41 to 116 days of exposure, rapidly increasing pul- monary inflammation and damage with concomitant development of fibrosis occurred. This suggested that pulmonary defense mechanisms were initially able to compensate and control silica-induced pulmonary in- flammation and damage, but after a certain threshold lung burden was exceeded, these control mechanisms no longer were adequate to prevent the progres- sion of silica-induced pulmonary disease.” In terms of Fig. 4, these data could be interpreted as indicat-

ing that exposure initially moves the model curve up- ward, thus moving the homeostatic equilibrium right- ward (yielding the reported controlled, reversible increases in levels of loop variables). Continued exposure moves the model curve further upward (e.g., because it selects for macrophages that pro- duce higher levels of ROS for the same exposure), eventually creating a tipping point threshold and an irreversible disease state (saturated equilibrium), yielding the reported rapid progression of pulmonary disease.

Such a coincidence between qualitative predic- tions and experimental observations in rats, while perhaps encouraging, does not prove that our con- ceptual model is correct. To test the specific biolog- ical interpretation (suggested by Fig. 3) that a high- ROS equilibrium accounts for silica-induced lung diseases, it would be necessary to assess the levels of ROS in conjunction with the initiation and pro- gression of silica-induced lung diseases. Fortunately, such experiments have been done. Porter et al.(71) ex- amined the mechanism by which injury progresses in rat lungs even after exposure ceases, and found that it is indeed mediated by a continuing increase in the production of ROS (and also RNS). They reported that “even after silica exposure has ended, and de- spite declining silica lung burden, silica-induced pul- monary nitrogen oxide (NO) and ROS production increases, thus producing a more severe oxidative stress. . . . iNOS and NO-mediated damage are asso- ciated anatomically with silica-induced pathological lesions.” This is fully consistent with the prediction (from Fig. 4) that, once the tipping point threshold has been passed, the system will be in the basin of at- traction for a high-ROS equilibrium, to which it will move (thus increasing the levels of all the loop vari- ables positively linked to ROS) even after silica ex- posure has ended. A similar tipping-point threshold between two basins of attraction has been reported in an explicit dynamic simulation model of COPD.(10)

Thus, this key feature of our theoretical analysis ap- pears to be consistent with some limited available data.

Of course, rats are not people, and the relevance of experimental findings in rats to disease processes in people can be questioned. However, Porter et al.(70) note that in human occupational populations, too, “[h]uman epidemiologic studies have found that silicosis may develop or progress even after occupa- tional exposure has ended, suggesting that there is a threshold lung burden above which silica-induced pulmonary disease progresses without further

Lung Diseases and Lung Cancer Caused by Crystalline Silica 1555

exposure.” Thus, we believe there is empirical support for the inference that CS, like other PSPs that cause lung diseases following chronic inflam- mation,(2) induces a high-ROS state as a possible alternative equilibrium to the usual, lower-ROS, homeostatic equilibrium—at least in susceptible individuals (defined as those in whom exposure shifts the model curve up enough to create the alternative stable equilibrium state, X ∗∗). Exposures that push the dynamic system of interacting variables in the lung (see Fig. 3) into the basin of attraction of this high-ROS state then trigger progression to the high-ROS state, even if no further exposure occurs. Depending on an individual’s capacity to repair the multiple types of damage caused by the high-ROS state (see Fig. 3), a variety of lung diseases, from silicosis to lung cancer, can result. We propose this as a unifying conceptual model for understanding the induction and progression of inflammation-mediated lung diseases caused by inhalation of PSPs.

5. DISCUSSION: USING THE MODEL TO ADDRESS POLICY-RELEVANT QUESTIONS

Epidemiological investigations that do not in- clude careful, well-validated modeling of exposure estimation errors may not yet be capable of deliver- ing convincing answers to the policy-relevant ques- tions raised in the introduction: whether exposure- related diseases occur together; whether CS has an exposure-response threshold for causing lung dis- eases; and, if so, whether currently permissible expo- sure limits lie above or below the threshold. How- ever, combining available, imperfect epidemiological evidence with recent advances in understanding of lung responses to poorly soluble particulates (PSPs) in general, and CS in particular, as outlined in the previous two sections, allows us to shed new light on each of these practical questions.

5.1. Existence of an Exposure-Response Threshold

There are strong empirical, as well as theoretical, grounds for expecting a threshold in the exposure- response relation. In theory, knowledge that CS acts through positive feedback loops (Fig. 3) suggests the presence of an exposure-response tipping point threshold (such as X ′ in Fig. 4). Empirically, rel- atively low exposures have been observed to in- duce largely self-limiting and reversible effects in rats (consistent with a homeostatic equilibrium, X ∗),

while high exposures have been observed to trig- ger a self-sustaining escalation to a permanent high- ROS state (consistent with an alternative equilib- rium X ∗∗).(70,71) Our review of CS epidemiology in Section 2 suggests that existing epidemiology is fully consistent with the biologically-based understand- ing of PSP mode of action and the two alternative- equilibria theory in Figs 3 and 4, and with their implied exposure-response threshold for exposure- related increases in lung disease risks (as observed for many PSPs in rats),(8) once a clear distinction is drawn between exposure-response curves for es- timated exposures and exposure-response curves for true but unknown exposures. The former may lack a threshold, even if the latter have one (Fig. 1).

5.2. Quantitative Estimation of the Exposure-Response Threshold: ≥ 0.4 mg/m3 A potentially useful quantitative contribution

from CS epidemiology is the observation by Rush- ton(4) that lung function appears to be diminished in some studies at estimated occupational exposure concentrations in excess of 0.1–0.2 mg/m3 of res- pirable silica dust for durations of at least 30–40 years, in the presence of other occupational dust exposures. If this finding is confirmed, and if con- founding by cigarette smoking and occupational co- exposures is eventually ruled out as an explanation (perhaps by building on recent innovative statistical methods(31)), then 0.1–0.2 mg/m3 of silica dust for 30– 40 years might be accepted as a useful point of depar- ture for estimating the exposure threshold that must be exceeded to create a disease state.

As in other epidemiological studies, there is large uncertainty in this review about true exposures, implying that any real exposure-response thresh- old is likely to be significantly greater (perhaps by several-fold) than the level at which the estimated exposure-response threshold shows elevated risks (see Fig. 1). To obtain a clear estimated concentra- tion threshold between 0.1 and 0.2 mg/m3, it is neces- sary to modify the example in Table I. For example, Fig. 5 shows a simulated exposure-response curve when the true exposure is uniformly distributed be- tween 0 and 1 mg/m3 and there is a true response threshold at 0.4 mg/m.3 (With the true probabil- ity of response, i.e., exposure-induced illness, be- ing 0 for concentrations below this threshold and 1 above it. In reality, of course, different individuals might have different thresholds, reflecting their own model curves and X ′ values, but it remains true that

1556 Cox

Plot of Means and Conf. Intervals (95.00%)

for Simulated Response Probabilities: Pr(Response | Estimated Exposure)

Response

0 0.06 0.12 0.18 0.24 0.30 0.36 0.42 0.48 0.54 0.60 0.66 0.72 0.78

Estimated Exposure Concentration (mg/m^3)

-0.2

0.0

0.2

0.4

0.6

0.8

1.0

1.2

E st

im a

te d

r e

sp o

n se

f ra

ct io

n ,

P r(

re sp

o n

se )

Fig. 5. A true threshold at 0.4 mg/m3

produces an estimated threshold between 0.1 and 0.2 mg/m3. (N = 10,000 samples; k ∼ U[0.3, 1.7]; true exposure ∼ U[0, 1] mg/m3.)

unmodeled error, even in unbiased exposure esti- mates, smears out and decreases the apparent thresh- old level of exposure at which excess population risks start to occur.) In the absence of detailed study of real-world exposure estimation errors, such hy- pothetical examples suggest that an estimated ex- posure concentration threshold between 0.1 and 0.2 mg/m3 might correspond to a true threshold value of about 0.4 mg/m3 for the concentration threshold that must be exceeded before adverse health effects occur among susceptible workers.

However, this rough estimate of 0.4 mg/m3 is contingent on as-yet unproved assumptions, includ- ing that the adverse health effects in Rushton(4)

were caused by CS, rather than by other exposures. We have assumed only a rather modest degree of variability in estimated exposures around the corre- sponding true values (namely, a uniform distribution around the mean, k ∼ U[0.3, 1.7], with no outliers or heavy tails). The true threshold could be substan- tially higher than 0.4 mg/m3 if exposure estimates have greater variability than this. (As an extreme ex- ample, the true threshold could be as high as 2 mg/m3

and still give an estimated threshold of 0.1 mg/m3 if (a) each individual with an estimated exposure of 0.1 has a 5% probability of having been exposed to 2 mg/m3 and a 95% probability of having been exposed to 0 mg/m3, for an average exposure of 0.05 × 2 + 0.95 × 0 = 0.1 mg/m3; and (b) the power of the study is such that at least 5% of individuals in an exposure group must respond in order for an excess risk to be

detected.) Thus, to better estimate the true level at which adverse health effects associated with the high- ROS state are induced, it will be essential for future studies to more carefully characterize the error dis- tribution of estimated exposures around true expo- sure levels, perhaps using more detailed simulations of workplace daily exposure distribution means and variances.

Meanwhile, it appears plausible that currently permitted exposure levels of 0.1 mg/m3 of respirable CS could be well below (possibly by a factor of 2 to 10, based on the hypothetical examples just de- scribed) the levels that might increase risks of ad- verse health effects. This conclusion becomes more robust if, instead of there being different thresholds for different CS-induced lung diseases, there is one large dichotomy, as illustrated in Fig. 4, between a low-ROS homeostatic equilibrium and a high-ROS disease state equilibrium (which can then produce different ROS-mediated diseases in susceptible in- dividuals, based on different vulnerabilities in their defensive and repair resources for responding to ox- idative stress injuries). We now consider further the implications of such a dichotomy.

5.3. Is Increased Risk of Silicosis Necessary for Increased Risk of Lung Cancer?

The study of Rushton(4) examines estimated con- centrations for longitudinal effects, so that even long- delayed health effects can eventually be counted.

Lung Diseases and Lung Cancer Caused by Crystalline Silica 1557

This is very useful when the alternative-equilibria theory in Fig. 4 is combined with an assumption that the high-ROS equilibrium is necessary (although perhaps not sufficient, if defensive and repair ca- pabilities are sufficiently strong) to cause increased risk of ROS-mediated lung diseases. Together, these assumptions imply that if increased rates of ROS- mediated lung diseases do eventually occur in an ex- posed occupational population, then exposure must have been sufficient to create the high-ROS state in susceptible individuals—and, therefore, high enough to have increased risks of several different diseases associated with the high-ROS state among individu- als susceptible to each type (e.g., due to limited ca- pacity for alveolar epithelial tissue repair, for em- physema; or ECM repair, for fibrosis; or apoptosis of premalignant cells, for lung cancer; and so forth). Conversely, this understanding of the disease pro- cess implies that protecting against any of the high- ROS diseases, by keeping exposures below the lev- els that induce a high-ROS state in an individual or species, will protect against all of them, from silicosis to inflammation-mediated lung cancer. This makes it plausible that exposures that are too low to cause in- creased risk of silicosis (even among susceptible in- dividuals) will also not cause increased risk of lung cancer, even if silicosis is not a necessary precondi- tion for CS-induced lung cancer: failure to create the high-ROS alternative equilibrium protects against both. According to this logic, increased risk of sil- icosis (and other indicators of the high-ROS state) in susceptible individuals should be expected as a necessary accompaniment to increased risk of other high-ROS diseases (such as inflammation-mediated lung cancer caused by CS(2,9)), whether or not silico- sis causally contributes to CS-induced lung cancer.

6. CONCLUSIONS

Postulating an exposure-response threshold for lung diseases (including lung cancer) associated with exposure to CS and other PSPs is not new. It has long been discussed for CS, with rat data, human data, and mechanistic information being cited in support of thresholds.(8) For example, in 1995, researchers from California’s Department of Toxic Substances Con- trol(72) reviewed the then-available evidence on the carcinogenicity of CS, and concluded: “The weight of evidence for both rats and humans indicates that fibrotic and silicotic lesions in the lung result from in- halation exposure to CS and that lung cancer is sec- ondary to those lesions in the lung. Thus CS should

be considered to have a threshold for causing can- cer. The critical exposure criterion is that exposure level which does not produce a fibrogenic or silicotic response; thus it is necessary to determine the no ob- served adverse effect level (NOAEL) for fibrogene- sis.”

Our analysis supports these earlier conclusions. To do harm, exposures to PSPs such as CS must be large enough and last long enough to trigger the chronic inflammatory responses and progression to a high-ROS state that can eventually lead to diseases. In vitro evidence in cell cultures, as well as in vivo experiments in rats, indicate exposure thresholds for inflammation,(73) oxidative stress, and resulting diseases, including lung cancer.(8) Moreover, normal lung cell populations interact via homeostatic (neg- ative) feedback loops that stabilize and maintain oxidant–antioxidant balance(74,75) and other (e.g., proteinase/anti-proteinase) equilibria.(10) Disease risk is not increased by exposures while homeosta- sis is maintained. Disrupting normal homeostasis requires activating positive feedback loops (Fig. 3) capable of damaging tissue (respiratory epithelium) and overwhelming normal repair processes. Both rat data(8) and mathematical modeling of inflammation- mediated lung diseases (Fig. 4) indicate that these responses to PSPs have exposure-response thresh- olds. Of course, these data and models are limited, and much remains to be learned about the details of the biological inputs and feedback loops that they describe, as well as others that may yet be discovered. Thus, we cannot completely exclude the possibility that a threshold does not exist. But our model-based analysis may add to previous weight-of- evidence conclusions by suggesting how exposure- response thresholds naturally arise between alternative basins of attraction in positive feed- back loop systems.

For CS and many other PSPs, sufficient expo- sure triggers AM activation and phenotype change, release of ROS and RNS, attraction of monocytes, AMs, and neutrophils to inflamed areas, damage and destruction of alveolar epithelial tissue and ECM, disruption of normal apoptosis and epithelial tissue repair and ECM repair, sustained epithelial prolif- eration and hyperplasia, and possible promotion of lung cancer. These disease processes may be mod- eled as networks of damaging positive feedback loops that are either “switched on” (meaning that the loop is attracted to a new, stable equilibrium with increased values of its variables, such as X ∗∗ in Fig. 4) or “switched off” (meaning that the loop

1558 Cox

remains in the basin of attraction of the healthy equi- librium, X ∗ in Fig. 4). Excess risk of inflammatory lung diseases and lung cancer arises only at expo- sure intensities and durations that are large enough to switch on these disease processes. For CS, these trigger levels may be on the order of 0.4 mg/m3 or more of silica dust, depending on the distribution of exposure estimation errors around true values. Such levels significantly exceed currently permissible lev- els (e.g., 0.05–0.1 mg/m3), implying that further re- ductions in permitted exposure levels—if permitted levels are enforced—should not be expected to pro- duce further reductions in human health risks.

ACKNOWLEDGMENTS

This work was supported in part by the Crys- talline Silica Panel of the American Chemistry Coun- cil. I am grateful to members of the Panel for stimu- lating discussions on crystalline silica epidemiology, biology, and risk assessment. All research questions addressed, methods used, and conclusions reached are mine alone.

REFERENCES

1. IARC. IARC Monographs on the Evaluation of Carcino- genic Risks to Humans, 1997; Vol. 68. Silica. Available at: http://monographs.iarc.fr/ENG/Monographs/vol68/volume68. pdf, Accessed March 28, 2011.

2. Azad N, Rojanasakul Y, Vallyathan V. Inflammation and lung cancer: Roles of reactive oxygen/nitrogen species. Journal of Toxicology and Environmental Health. Part B, Critical Reviews, 2008;11(1):1–15. Available at: http:// www.informaworld.com/smpp/section?content=a789269849. fulltext=713240928, Accessed March 28, 2011.

3. American Thoracic Society. Adverse effects of crystalline sil- ica exposure. American Journal of Respiratory and Critical Care Medicine, 1997; 155(2):761–768.

4. Rushton L. Chronic obstructive pulmonary disease and oc- cupational exposure to silica. Reviews on Environmental Health, 2007; 22(4):255–272.

5. Soutar CA, Robertson A, Miller BG, Searl A, Bignon J. Epi- demiological evidence on the carcinogenicity of silica: Fac- tors in scientific judgement. Annals of Occupational Hygiene, 2000; 44(1):3–14.

6. Erren TC, Glende CB, Morfeld P, Piekarski C. Is exposure to silica associated with lung cancer in the absence of silico- sis? A meta-analytical approach to an important public health question. International Archives of Occupational and Envi- ronmental Health, 2009; 82(8):997–1004.

7. Mossman BT. Mechanisms of action of poorly soluble partic- ulates in overload-related lung pathology. Inhalation Toxicol- ogy, 2000; 12(1–2):141–148.

8. Oberdörster G. Toxicokinetics and effects of fibrous and nonfibrous particles. Inhalation Toxicology, 2002; 14(1): 29–56.

9. Blanco D, Vicent S, Fraga MF, Fernandez-Garcia I, Freire J, Lujambio A, Esteller M, Ortiz-de-Solorzano C, Pio R, Lecanda F, Montuenga LM. Molecular analysis of a multistep lung cancer model induced by chronic inflammation reveals

epigenetic regulation of p16 and activation of the DNA dam- age response pathway. Neoplasia, 2007; 9(10):840–852.

10. Cox LA. A causal model of chronic obstructive pulmonary disease (COPD) risk. Risk Analysis, 2011; 31(1):38–62.

11. Gulumian M, Borm PJ, Vallyathan V, Castranova V, Donald- son K, Nelson G, Murray J. Mechanistically identified suitable biomarkers of exposure, effect, and susceptibility for silico- sis and coal-worker’s pneumoconiosis: A comprehensive re- view. Journal of Toxicology and Environmental Health. Part B, Critical Reviews, 2006; 9(5):357–395.

12. Huaux F. New developments in the understanding of im- munology in silicosis. Current Opinion in Allergy and Clinical Immunology, 2007; 7(2):168–173.

13. Kurihara N, Wada O. Silicosis and smoking strongly increase lung cancer risk in silica-exposed workers. Industrial Health, 2004; 42(3):303–314.

14. Ulm K, Gerein P, Eigenthaler J, Schmidt S, Ehnes H. Silica, silicosis and lung-cancer: Results from a cohort study in the stone and quarry industry. International Archives of Occupa- tional and Environmental Health, 2004; 77(5):313–318.

15. Amabile JC, Leuraud K, Vacquier B, Caër-Lorho S, Acker A, Laurier D. Multifactorial study of the risk of lung cancer among French uranium miners: Radon, smoking and silicosis. Health Physics, 2009; 97(6):613–621.

16. Hessel P, Sluis-Cremer GK, Hnizdo, E. Silica exposure, sili- cosis, and lung cancer: A necropsy study. British Journal of Industrial Medicine, 1990; 47:4–9.

17. Chan CK, Leung CC, Tam CM, Yu TS, Wong TW. Lung can- cer mortality among a cohort of men in a silicotic register. JOEM, 2000; 42:69–75.

18. Chen, W, Chen J. Nested case-control study of lung cancer in four Chinese tin mines. Occupational and Environmental Medicine, 2002; 59:113–118.

19. Carta P, Aru G, Manca P. Mortality from lung cancer among silicotic patients in Sardinia: An update study with 10 more years of follow up. Occupational and Environmen- tal Medicine, 2001; 58:786–793.

20. Yu ITS, Tse LA, Leung CC, Wong TW, Tam, CM, Chan, AC. Lung cancer mortality among silicotic workers in Hong Kong—No evidence for a link. Annals of Oncology, 2007;18:1056–1063.

21. Steenland K, Mannetje A, Boffetta P, Stayner L, Attfield M, Chen J, Dosemeci M, DeKlerk N, Hnizdo E, Koskela R, Checkoway H, International Agency for Research on Cancer. Pooled exposure-response analyses and risk assessment for lung cancer in 10 cohorts of silica-exposed workers: An IARC multicentre study. Cancer Causes Control, 2001; 12(9):773– 784.

22. Stayner L. Silica and lung cancer: When is enough evidence enough? Epidemiology, 2007; 18(1):23–24.

23. Vacek PM, Verma DK, Graham WG, Callas PW, Gibbs GW. Mortality in Vermont granite workers and its associ- ation with silica exposure. Occupational and Environmen- tal Medicine, 2010. Available at http://www.ncbi.nlm.nih.gov/ pubmed/20855299.

24. Brown T. Silica exposure, smoking, silicosis and lung cancer— Complex interactions. Occupational Medicine (London), 2009; 59(2):89–95.

25. Dahmann D, Taeger D, Kappler M, Büchte S, Morfeld P, Brüning T, Pesch B. Assessment of exposure in epidemiolog- ical studies: The example of silica dust. Journal of Exposure Science and Environmental Epidemiology, 2008; 18(5):452– 461.

26. Porter DW, Barger M, Robinson VA, Leonard SS, Landsit- tel D, Castranova V. Comparison of low doses of aged and freshly fractured silica on pulmonary inflammation and dam- age in the rat. Toxicology, 2002; 175(1–3):63–71.

27. Pelucchi C, Pira E, Piolatto G, Coggiola M, Carta P, La Vec- chia C. Occupational silica exposure and lung cancer risk: A

Lung Diseases and Lung Cancer Caused by Crystalline Silica 1559

review of epidemiological studies 1996–2005. Annals of On- cology, 2006; 17(7):1039–1050.

28. Lacasse Y, Martin S, Gagné D, Lakhal L. Dose-response meta-analysis of silica and lung cancer. Cancer Causes Con- trol, 2009; 20(6):925–933.

29. Cocco P, Dosemeci M, Rice C. Lung cancer among silica- exposed workers: The quest for truth between chance and ne- cessity. Medicina Del Lavoro, 2007; 98(1):3–17.

30. Chen W, Bochmann F, Sun Y. Effects of work related con- founders on the association between silica exposure and lung cancer: A nested case-control study among Chinese miners and pottery workers. Int Arch Occup Environ Health, 2007; 80(4):320–326.

31. Richardson DB. Occupational exposures and lung cancer: Adjustment for unmeasured confounding by smoking. Epi- demiology, 2010; 21(2):181–186.

32. Carroll RJ, Chen X, Hu Y. Identification and estimation of nonlinear models using two samples with nonclassical mea- surement errors. Journal of Nonparametric Statistics (Print), 2010; 22(4):379–399.

33. Cassidy A, ‘t Mannetje A, van Tongeren M, Field JK, Zaridze D, Szeszenia-Dabrowska N, Rudnai P, Lissowska J, Fabi- anova E, Mates D, Bencko V, Foretova L, Janout V, Fevotte J, Fletcher T, Brennan P, Boffetta P. Occupational exposure to crystalline silica and risk of lung cancer: A multicenter case- control study in Europe. Epidemiology, 2007; 18(1):36–43.

34. Veierød MB, Laake P. Exposure misclassification: Bias in category specific Poisson regression coefficients. Statistics in Medicine, 2001; 20(5):771–784.

35. Cheng D, Branscum AJ, Stamey JD. Accounting for response misclassification and covariate measurement error improves power and reduces bias in epidemiologic studies. Annals of Epidemiology, 2010; 20(7):562–567.

36. Lu C, Lyles RH. Misclassification adjustment in thresh- old models for the effects of subject-specific exposure means and variances. Proceedings of the Joint Statistical Meeting, 2008. Available at: http://www.amstat.org/sections/ srms/proceedings/y2008/Files/302008.pdf, Accessed March 28, 2011.

37. Thakur SA, Beamer CA, Migliaccio CT, Holian A. Critical role of MARCO in crystalline silica-induced pulmonary in- flammation. Toxicology Science, 2009; 108(2):462–471.

38. Hildemann S, Hammer C, Krombach F. Heterogeneity of alveolar macrophages in experimental silicosis. Environmen- tal Health Perspectives, 1992; 97:53–57.

39. Janssen YM, Marsh JP, Absher MP, Hemenway D, Vacek PM, Leslie KO, Borm PJ, Mossman BT. Expression of an- tioxidant enzymes in rat lungs after inhalation of asbestos or silica. Journal of Biological Chemistry, 1992; 267(15):10625– 10630.

40. Comhair SA, Erzurum SC. Antioxidant responses to oxidant- mediated lung diseases. American Journal of Physiology. Lung Cellular and Molecular Physiology, 2002; 283(2):L246– L255.

41. Antognelli C, Gambelunghe A, Del Buono C, Murgia N, Talesa VN, Muzi G. Crystalline silica Min-U-Sil 5 induces ox- idative stress in human bronchial epithelial cells BEAS-2B by reducing the efficiency of antiglycation and antioxidant enzymatic defenses. Chemico-Biological Interactions, 2009; 182(1):13–21.

42. Eom HJ, Choi J. Oxidative stress of silica nanoparticles in hu- man bronchial epithelial cell, Beas-2B. Toxicology In Vitro, 2009; 23(7):1326–1332.

43. Guo RF, Ward PA. Role of oxidants in lung injury during sepsis. Antioxidants and Redox Signaling, 2007; 9(11):1991– 2002.

44. Rimal B, Greenberg AK, Rom WN. Basic pathogenetic mechanisms in silicosis: Current understanding. Current Opinion in Pulmonary Medicine, 2005; 11(2):169–173.

45. Fubini B, Hubbard A. Reactive oxygen species (ROS) and re- active nitrogen species (RNS) generation by silica in inflam- mation and fibrosis. Free Radical Biology and Medicine, 2003; 34(12):1507–1516.

46. Ding M, Shi X, Castranova V, Vallyathan V. Predisposing factors in occupational lung cancer: Inorganic minerals and chromium. Journal of Environmental Pathology, Toxicology and Oncology, 2000; 19(1–2):129–138.

47. Shi X, Castranova V, Halliwell B, Vallyathan V. Reactive oxygen species and silica-induced carcinogenesis. Journal of Toxicology and Environmental Health. Part B, Critical Re- views, 1998; 1(3):181–197.

48. Schins RP, Knaapen AM. Genotoxicity of poorly soluble par- ticles. Inhalation Toxicology, 2007; 19(Suppl 1):189–198.

49. Smart SJ, Casale TB. TNF-alpha-induced transendothelial neutrophil migration is IL-8 dependent. American Journal of Physiology, 1994; 266(3 Pt 1):L238–L245.

50. Pease JE, Sabroe I. The role of interleukin-8 and its receptors in inflammatory lung disease: Implications for therapy. Amer- ican Journal of Respiratory Medicine, 2002; 1(1):19–25.

51. MacNee W. Pulmonary and systemic oxidant/antioxidant im- balance in chronic obstructive pulmonary disease. Proceed- ings of the American Thoracic Society, 2005; 2(1):50–60.

52. Takami M, Terry V, Petruzzelli L. Signaling path- ways involved in IL-8-dependent activation of adhesion through Mac-1. Journal of Immunology, 2002; 168(9):4559– 4566.

53. Scholz H, Yndestad A, Damås JK, Waehre T, Tonstad S, Aukrust P, Halvorsen B. 8-isoprostane increases expression of interleukin-8 in human macrophages through activation of mitogen-activated protein kinases. Cardiovascular Research, 2003; 59(4):945–954.

54. Mortaz E, Kraneveld AD, Smit JJ, Kool M, Lambrecht BN, Kunkel SL, Lukacs NW, Nijkamp FP, Folkerts G. Effect of cigarette smoke extract on dendritic cells and their impact on T-cell proliferation. PLoS One, 2009; 4(3):e4946.

55. Thatcher TH, McHugh NA, Egan RW, Chapman RW, Hey JA, Turner CK, Redonnet MR, Seweryniak KE, Sime PJ, Phipps RP. Role of CXCR2 in cigarette smoke-induced lung inflammation. American Journal of Physiology, Lung Cellular and Molecular Physiology, 2005; 289(2):L322–L328.

56. Absher MP, Trombley L, Hemenway DR, Mickey RM, Leslie KO. Biphasic cellular and tissue response of rat lungs after eight-day aerosol exposure to the silicon dioxide cristobalite. American Journal of Pathology, 1989; 134(6):1243–1251.

57. Kim S, Nadel JA. Role of neutrophils in mucus hypersecre- tion in COPD and implications for therapy. Treatments in Respiratory Medicine, 2004; 3(3):147–159.

58. Delgado L, Parra ER, Capelozzi VL. Apoptosis and extracel- lular matrix remodelling in human silicosis. Histopathology, 2006; 49(3):283–289.

59. Srivastava KD, Rom WN, Jagirdar J, Yie TA, Gordon T, Tchou-Wong KM. Crucial role of interleukin-1beta and nitric oxide synthase in silica-induced inflammation and apoptosis in mice. American Journal of Respiratory and Critical Care Medicine, 2002; 165(4):527–533.

60. Leigh J, Wang H, Bonin A, Peters M, Ruan X. Silica-induced apoptosis in alveolar and granulomatous cells in vivo. Environ Health Perspectives, 1997; 105(Suppl 5):1241–1245.

61. Zhang DD, Hartsky MA, Warheit DB. Time course of quartz and TiO(2) particle-induced pulmonary inflammation and neutrophil apoptotic responses in rats. Experimental Lung Research, 2002; 28(8):641–670.

62. Kisseleva T, Brenner DA. Fibrogenesis of parenchymal or- gans. Proceedings of the American Thoracic Society, 2008; 5(3):338–342.

63. Elizegi E, Pino I, Vicent S, Blanco D, Saffiotti U, Montuenga LM. Hyperplasia of alveolar neuroendocrine cells in rat lung carcinogenesis by silica with selective expression of

1560 Cox

proadrenomedullin-derived peptides and amidating enzymes. Laboratory Investigation, 2001; 81(12):1627–1638.

64. Williams AO, Flanders KC, Saffiotti U. Immunohistochem- ical localization of transforming growth factor-beta 1 in rats with experimental silicosis, alveolar type II hyperplasia, and lung cancer. American Journal of Pathology, 1993; 142(6):1831–1840.

65. Williams AO, Saffiotti U. Transforming growth factor beta1, ras and p53 in silica-induced fibrogenesis and carcinogene- sis. Scandinavian Journal of Work Environment and Health, 1995; 21(Suppl 2):30–34.

66. Corvol H, Flamein F, Epaud R, Clement A, Guillot L. Lung alveolar epithelium and interstitial lung disease. Interna- tional Journal of Biochemistry and Cell Biology, 2009; 41(8– 9):1643–1651

67. Umemura S, Fujimoto N, Hiraki A, Gemba K, Takigawa N, Fujiwara K, Fujii M, Umemura H, Satoh M, Tabata M, Ueoka H, Kiura K, Kishimoto T, Tanimoto M. Aberrant promoter hypermethylation in serum DNA from patients with silicosis. Carcinogenesis, 2008; 29(9):1845–1849.

68. Cox LA Jr. Could removing arsenic from tobacco smoke sig- nificantly reduce smoker risks of lung cancer? Risk Analysis, 2009; 29(1):3–17.

69. Kuilman T, Michaloglou C, Vredeveld LC, Douma S, van Doorn R, Desmet CJ, Aarden LA, Mooi WJ, Peeper DS. Oncogene-induced senescence relayed by an interleukin- dependent inflammatory network. Cell, 2008; 133(6):1019– 1031.

70. Porter DW, Hubbs AF, Mercer R, Robinson VA, Ramsey D, McLaurin J, Khan A, Battelli L, Brumbaugh K, Teass A, Castranova V. Progression of lung inflammation and damage in rats after cessation of silica inhalation. Toxicology Science, 2004; 79(2):370–380.

71. Porter DW, Millecchia LL, Willard P, Robinson VA, Ram- sey D, McLaurin J, Khan A, Brumbaugh K, Beighley CM, Teass A, Castranova V. Nitric oxide and reactive oxygen species production causes progressive damage in rats after cessation of silica inhalation. Toxicology Science, 2006; 90(1): 188–197.

72. Klein AK, Christopher JP. Evaluation of crystalline sil- ica as a threshold carcinogen. Scandinavian Journal of Work Environment and Health, 1995; 21:95–98. Available

at: http://www.sjweh.fi/show abstract.php?abstract id=96, Accessed March 28, 2011.

73. Donaldson K, Borm PJ, Oberdorster G, Pinkerton KE, Stone V, Tran CL. Concordance between in vitro and in vivo dosimetry in the proinflammatory effects of low-toxicity, low- solubility particles: The key role of the proximal alveolar re- gion. Inhalation Toxicology, 2008; 20(1):53–62.

74. Liu B, Chen Y, St Clair DK. ROS and p53: A versa- tile partnership. Free Radical Biology and Medicine, 2008; 44(8):1529–1535.

75. D’Autréaux B, Toledano MB. ROS as signalling molecules: Mechanisms that generate specificity in ROS homeostasis. Nature Reviews, Molecular Cell Biology, 2007; 10:813–824.

76. Castranova V. Signaling pathways controlling the production of inflammatory mediators in response to crystalline silica ex- posure: Role of reactive oxygen/nitrogen species. Free Radi- cal Biology and Medicine, 2004; 37(7):916–925.

77. van Berlo D, Knaapen AM, van Schooten FJ, Schins RP, Albrecht C. NF-kappaB dependent and independent mech- anisms of quartz-induced proinflammatory activation of lung epithelial cells. Particle and Fibre Toxicology, 2010; May 21: 7–13.

78. Gossart S, Cambon C, Orfila C, Séguélas MH, Lepert JC, Rami J, Carré P, Pipy B. Reactive oxygen intermediates as regulators of TNF-alpha production in rat lung inflammation induced by silica. Journal of Immunology, 1996;156(4):1540– 1548.

79. Ke Q, Li J, Ding J, Ding M, Wang L, Liu B, Costa M, Huang C. Essential role of ROS-mediated NFAT activation in TNF- alpha induction by crystalline silica exposure. American Jour- nal of Physiology, Lung Cellular and Molecular Physiology, 2006; 291(2):L257–L264.

80. Pelclová D, Fenclová Z, Kacer P, Kuzma M, Navrátil T, Lebedová J. Increased 8-isoprostane, a marker of oxidative stress in exhaled breath condensate in subjects with asbestos exposure. Industrial Health, 2008;46(5):484–489.

81. Saffiotti U. Silicosis and lung cancer: A fifty-year perspective. Acta Biomed, 2005;76(2):30–37.

82. Cox LA Jr. Risk Analysis of Complex and Uncertain Sys- tems. New York: Springer, 2009, Chapter 13. Available at: www.springerlink.com/content/jn57472874131283/, Accessed March 28, 2011.

Copyright of Risk Analysis: An International Journal is the property of Wiley-Blackwell and its content may

not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written

permission. However, users may print, download, or email articles for individual use.