catherine only

31.pdf

Home >Chemistry homework help >Pharmacology homework help >catherine only

Nature | www.nature.com | 1

Article

GDF15 mediates the effects of metformin on body weight and energy balance

Anthony P. Coll1,10*, Michael Chen2, Pranali Taskar2, Debra Rimmington1, Satish Patel1, John A. Tadross1, Irene Cimino1, Ming Yang1, Paul Welsh3, Samuel Virtue1, Deborah A. Goldspink1, Emily L. Miedzybrodzka1, Adam R. Konopka4, Raul Ruiz Esponda4, Jeffrey T.-J. Huang5, Y. C. Loraine Tung1, Sergio Rodriguez-Cuenca1, Rute A. Tomaz6, Heather P. Harding7, Audrey Melvin1, Giles S. H. Yeo1, David Preiss8, Antonio Vidal-Puig1, Ludovic Vallier6, K. Sreekumaran Nair4, Nicholas J. Wareham9, David Ron7, Fiona M. Gribble1, Frank Reimann1, Naveed Sattar3,10, David B. Savage1,10, Bernard B. Allan2,10 & Stephen O’Rahilly1,10*

Metformin, the world’s most prescribed anti-diabetic drug, is also effective in preventing type 2 diabetes in people at high risk1,2. More than 60% of this effect is attributable to the ability of metformin to lower body weight in a sustained manner3. The molecular mechanisms by which metformin lowers body weight are unknown. Here we show—in two independent randomized controlled clinical trials—that metformin increases circulating levels of the peptide hormone growth/differentiation factor 15 (GDF15), which has been shown to reduce food intake and lower body weight through a brain-stem-restricted receptor. In wild-type mice, oral metformin increased circulating GDF15, with GDF15 expression increasing predominantly in the distal intestine and the kidney. Metformin prevented weight gain in response to a high-fat diet in wild-type mice but not in mice lacking GDF15 or its receptor GDNF family receptor α-like (GFRAL). In obese mice on a high-fat diet, the effects of metformin to reduce body weight were reversed by a GFRAL-antagonist antibody. Metformin had effects on both energy intake and energy expenditure that were dependent on GDF15, but retained its ability to lower circulating glucose levels in the absence of GDF15 activity. In summary, metformin elevates circulating levels of GDF15, which is necessary to obtain its beneficial effects on energy balance and body weight, major contributors to its action as a chemopreventive agent.

Metformin has been used as a treatment for type 2 diabetes since the 1950s. Recent studies have shown that it can also prevent or delay the onset of type 2 diabetes in people at high risk1,2. At-risk individuals treated with metformin exhibit a reduction in body weight, glucose and insulin levels and enhanced insulin sensitivity3. Although many mechanisms for the insulin-sensitizing actions of metformin have been proposed4, they do not explain the weight loss. The robustness and persistence of metformin-induced weight loss in participants in the Diabetes Prevention Program has drawn attention to its impor- tance to the chemopreventive effects of the drug5. A recent observa- tional epidemiological study6 noted a strong association of metformin use with circulating levels of GDF15, a peptide hormone produced by cells responding to stressors7. GDF15 acts through a receptor complex that is expressed solely in the hindbrain, through which it suppress food intake8–11. We hypothesized that the effects of metformin in

lowering body weight may involve the elevation of circulating levels of GDF15.

Human studies We first measured circulating GDF15 in a short-term human study12 and found that, after two weeks of metformin treatment, there was an increase of about 2.5-fold in mean circulating GDF15 (Fig. 1a). To determine whether this increase was sustained, we measured circu- lating GDF15 levels at 6, 12 and 18 months in all available participants in carotid atherosclerosis: metformin for insulin resistance (CAMERA)13, an-18 month randomized placebo-control trial of metformin in people without diabetes but with a history of cardiovascular disease. In this study, metformin-treated participants lost about 3.5% of body weight with no significant change in weight in the placebo arm13. Metformin

https://doi.org/10.1038/s41586-019-1911-y

Received: 1 July 2019

Accepted: 16 December 2019

Published online: xx xx xxxx

1MRC Metabolic Diseases Unit, Wellcome Trust-Medical Research Council Institute of Metabolic Science, University of Cambridge, Cambridge, UK. 2NGM Biopharmaceuticals, South San Francisco, CA, USA. 3Institute of Cardiovascular and Medical Sciences, University of Glasgow, Glasgow, UK. 4Division of Endocrinology, Mayo Clinic, Rochester, MN, USA. 5Division of Systems Medicine, School of Medicine, University of Dundee, Dundee, UK. 6Wellcome -Medical Research Council Cambridge Stem Cell Institute, Department of Surgery, University of Cambridge, Cambridge, UK. 7Cambridge Institute for Medical Research, University of Cambridge, Cambridge, UK. 8MRC Population Health Research Unit, Clinical Trial Service Unit and Epidemiological Studies Unit, Nuffield Department of Population Health, University of Oxford, Oxford, UK. 9MRC Epidemiology Unit, Wellcome Trust-Medical Research Council Institute of Metabolic Science, University of Cambridge, Cambridge, UK. 10These authors contributed equally: Anthony P. Coll, Naveed Sattar, David B. Savage, Bernard B. Allan, Stephen O’Rahilly. *e-mail: apc36@cam.ac.uk; so104@medschl.cam.ac.uk

https://doi.org/10.1038/s41586-019-1911-y

mailto:apc36@cam.ac.uk

mailto:so104@medschl.cam.ac.uk

2 | Nature | www.nature.com

Article

treatment was associated with significantly (P < 0.0001) increased lev- els of circulating GDF15 at all three time points (Fig. 1b, Extended Data Fig. 1b–e). Furthermore, the change in serum GDF15 from baseline in metformin recipients was significantly correlated (R = −0.26, P = 0.024) with weight loss (Extended Data Fig. 1a).

The correlation of GDF15 increment with changes in body weight, while statistically significant, was modest in size. Although we believe that it does contribute to weight loss in some individuals taking met- formin, we acknowledge that it is not necessary, and there are individu- als with increases in GDF15 that do not exhibit weight loss. However, in the context of a long-term human study with imperfect drug compli- ance and intermittent sampling of GDF15 levels, it is noteworthy that such an association was seen at all. Further, there was no association of weight change with change in GDF15 in the placebo group (R = −0.04, P = 0.740, n = 81).

Mouse studies Following these findings in humans, we performed a series of animal experiments to determine the potential causal link between the changes in GDF15 and weight changes induced by metformin. We administered metformin by oral gavage to mice fed a high-fat diet and measured serum GDF15. A single dose of 300 mg kg−1 of metformin increased GDF15 levels for at least 8 h (Fig. 1c). A higher dose of metformin, 600 mg kg−1, resulted in a sixfold increase in serum GDF15 levels at 4 h and 8 h after the dose, which were sustained above those of vehicle- treated mice for 24 h. The effects of metformin in chow-fed mice were

less pronounced (Extended Data Fig. 2), suggesting an interaction between metformin and the high-fat diet.

To determine the extent to which metformin-induced increase in GDF15 affects body weight, Gdf15+/+ and Gdf15−/− mice were switched from chow to a high-fat diet and dosed with metformin for 11 days. The high-fat diet induced similar weight gain in both genotypes (Fig. 2a). Metformin completely prevented weight gain in Gdf15+/+ mice, but Gdf15−/− mice were insensitive to the weight-reducing effects of met- formin (Fig. 2a, Extended Data Fig. 3a). Metformin significantly reduced cumulative food intake in wild-type mice but this effect was abolished in Gdf15−/− mice (Fig. 2b).

The identical protocol was applied to mice lacking GFRAL, the ligand-binding component of the hindbrain-expressed GDF15 receptor complex. Consistent with the results in mice lacking GDF15, metformin was unable to prevent weight gain in Gfral−/− mice (Fig. 2c, Extended Data Fig. 3b), despite similar levels of serum GDF15 to wild-type mice

2 4 8 24

0 4 12 24

500

1,000

1,500

2,000

2,500

Vehicle Metformin (300 mg kg–1) Metformin (600 mg kg–1)

P = 0.0004 P < 0.0001

P = 0.0003

P < 0.0001

P = 0.029

Time (h)

Time (months) Pl

ac eb

M et

fo rm

G D

F 1 5

(p g

m l–

1 )

500

1,000

1,500

2,000

2,500

1,000

1,200

1,400

1,600

1,800

2,000

2,200 G

D F

1 5

(p g

m l–

1 )

G D

F 1 5

(p g

m l–

1 )

P < 0.0001

P = 0.0001

P < 0.0001

P < 0.0001 a b

Placebo Metformin

Fig. 1 | Effect of Metformin on circulating GDF15 levels in humans and mice. a, Paired serum GDF15 concentration in nine human subjects after two weeks of either placebo or metformin treatment, P value (95% CI) by two-tailed t-test. b, Plasma GDF15 concentration in overweight or obese non-diabetic participants with known cardiovascular disease, randomized to metformin or placebo in CAMER A, using a mixed linear model. Data are mean ± s.e.m. Subject numbers: placebo and metformin, respectively, at time points: baseline, n = 85 and 86; 6 months, n = 81 and 71; 12 months, n = 77 and 68; 18 months, n = 83 and 74. Comparing metformin vs placebo groups, two-sided P = 0.311 at baseline, and P < 0.0001 at 6, 12 and 18 months individually. c, Serum GDF15 levels (mean ± s.e.m.) in obese mice measured 2, 4, 8 or 24 h after a single oral dose of 300 mg kg−1 or 600 mg kg−1 metformin, n = 7 per group, P values by 2-way ANOVA with Tukey’s correction for multiple comparisons.

C h a n g

e in

b o

d y

w e ig

h t

(% )

0 1 2 3 4 5 6 7 8 9

10 11

0 1 2 3 4 5 6 7 8 9 10 11

–8 –6 –4 –2 0 2 4 6 8

Gdf15+/+ vehicle Gdf15+/+ metformin

Gdf15–/– vehicle Gdf15–/– metformin

Gfral+/+ vehicle Gfral+/+ metformin

Gfral–/– vehicle Gfral–/– metformin

P =

0. 00

P =

0. 00

06 P < 0.0001

Time (d)

0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 910 11 0

Time (d)

C u m

u la

ti ve

f o

o d

in ta

ke ( g

)

Gdf15+/+ vehicle Gdf15+/+ metformin

P =

0 .0

0 7 8

P < 0.0001

P =

0 .0

0 1

10 11 0

Time (d)

C u m

u la

ti ve

f o

o d

in ta

ke ( g

)

Gdf15–/– vehicle Gdf15–/– metformin

–8 –6 –4 –2 0 2 4 6 8

C h a n g

e in

b o

d y

w e ig

h t

(% )

Time (d)

P =

0. 03

P =

0. 03

P =

0. 04

P =

0. 01

P =

0. 01

–20

–15

–10

–5

Vehicle + IgG Vehicle + anti-GFRAL Metformin + IgG Metformin + anti-GFRAL

Calo

P =

0. 03

P =

0. 02

P =

0. 00

Antibody

Time (weeks)

C h a n g

e in

b o

d y

w e ig

h t

(% )

GTT

a b

c d

25 35 45 55 65 Body weight (g)

0.4 0.45 0.5

0.55 0.6

0.65 0.7

0.75

E n e rg

y e xp

e n d

it u re

( kc

a l h

–1 )

R2 = 0.90

R2 = 0.86

Effect of metformin P = NS

Vehicle + ctrl IgG Vehicle + anti-GFRAL Metformin + ctrl IgG Metformin + anti-GFRAL

R2 = 0.95

R2 = 0.82

Effect of metformin P = 0.003

25 35 45 55 65 Body weight (g)

0.4 0.45 0.5

0.55 0.6

0.65 0.7

0.75

E n e rg

y e xp

e n d

it u re

( kc

a l h

–1 )

Fig. 2 | GDF15–GFR AL signalling is required for the weight-loss effects of metformin on a high-fat diet. a, Percentage change in body weight of Gdf15+/+ and Gdf15−/− mice on a high-fat diet treated with metformin (300 mg kg−1 day−1) for 11 days. Data are mean ± s.e.m., n = 6 per group except Gdf15+/+ vehicle, n = 7; P values by two-way ANOVA with Tukey’s correction for multiple comparisons. b, Cumulative food intake of mice as in a, P values by two-way ANOVA with Tukey’s correction for multiple comparisons. c, Percentage change in body weight of Gfral+/+ and Gfral−/− mice on a high-fat diet treated with metformin (300 mg kg−1 day−1) for 11 days. Data are mean ± s.e.m., n = 6 per group; P values by two-way ANOVA with Tukey’s correction for multiple comparisons. d, Percentage change in body weight of metformin-treated obese mice dosed with an anti-GFR AL antagonist antibody weekly for five weeks (yellow), starting four weeks after initial metformin exposure (grey). Data are mean ± s.e.m., vehicle + control IgG and metformin + anti-GFR AL, n = 7; other groups, n = 8; P values by two-way ANOVA with Tukey’s correction for multiple comparisons. Calo, period in which energy expenditure measured (see e); arrow, start of oral GTT (Fig. 3e–h). e, ANCOVA of energy expenditure against body weight of mice treated as in d. n = 6 mice per group. Data points show individual mice; P values for metformin calculated using ANCOVA with body weight as a covariate and treatment as a fixed factor.

Nature | www.nature.com | 3

(Extended Data Fig. 4a, b). In this experiment, the reduction in cumula- tive food intake did not reach statistical significance (Extended Data Fig. 4c).

To investigate the contribution of GDF1 5–GFRAL signal - ling to sustained, metformin-dependent weight regulation, we performed a 9-week study in which mice received approximately 250–300 mg kg−1 day−1 of metformin incorporated into their high-fat diet. The mice lost around 9% of their body weight after 1 month on this diet (Fig. 2d Extended data Fig. 3c). At this time, an anti-GFRAL antago- nist antibody or IgG control was administered. Metformin-consuming mice treated with anti-GFRAL regained about 12% of body weight after 5 weeks, whereas the weight loss seen in IgG control treated mice was maintained, reaching approximately 7% below the starting weight (Fig. 2d). The significant reduction in fat mass seen with metformin treatment and control antibody was not seen in the anti-GFRAL group. (Extended Data Fig. 4d). The delivery of metformin in chow resulted in an initial reduction in food intake in all metformin-treated groups, presumably because of a taste effect. This reduction in food intake will have affected metformin levels and probably affected GDF15 levels, with potential to bias the results. However, it is reassuring to note that any persistence of this would have worked against the detection of a specific effect of GFRAL antagonism, which was clearly demonstrable.

We undertook indirect calorimetry in metformin- and placebo- treated mice treated with anti-GFRAL antibody to establish whether there are additional effects on energy expenditure. Data were analysed by analysis of covariance (ANCOVA) with body weight as the covariate. Metformin treatment resulted in a significant increase in metabolic rate, which was blocked by antagonism of GFRAL (Fig. 2e). Thus under conditions in which GDF15 levels are increased by metformin, body

weight reduction is contributed to by both reduced food intake and an inappropriately high energy expenditure.

GDF15 and glucose homeostasis To examine the extent to which the insulin-sensitizing effects of metformin are dependent on GDF15, we repeated the experiment described in Fig. 2a (see Extended Data Fig. 5), measuring insulin tol- erance in metformin- and vehicle-treated GDF15-null mice and their wild-type littermates (Fig. 3a). Circulating metformin levels in both genotypes were identical (Extended Data Fig. 5d) and consistent with the high end of the human therapeutic range14. Metformin significantly increased insulin sensitivity, as assessed by the area under the plasma glucose curve, with no significant effect of genotype (Fig. 3b). Simi- larly, metformin reduced fasting blood glucose and fasting insulin in a GDF15-independent manner (Fig. 3 c, d).

We also performed oral glucose-tolerance tests (GTTs) on metformin- treated mice given either control IgG or anti-GFRAL antibody for five weeks (Figs. 2d, 3e, f, Extended Data Fig. 6a). Although the effect of metformin glucose disposal on the oral GTT as assessed by the area under the plasma glucose curve did not reach statistical sig- nificance (two-way ANOVA, P = 0.072), there was a significant effect of metformin on insulin (both fasting level and area under the curve (AUC)) after glucose bolus, that was independent of anti GFRAL anti- body (Fig. 3 g–j).

As these mice had different body weights at the time of assessment (Fig. 2d, Extended Data Fig. 3c), we performed intraperitoneal GTTs in a cohort of weight-matched Gdf15+/+ and Gdf15−/− mice that had been fed a high-fat diet for two weeks before receiving a single dose of metformin (300 mg kg−1) (Fig. 3k, l, Extended Data Fig. 6b–d). In these mice, there

0 30 60 90 120

0 30 60 90 120 0 30 60 90 120 0 30 60 90 120

0 30 60 90 120 0 30 60 90 120

0 30 60 90 120

Gdf15+/+ vehicle Gdf15+/+ metformin

Time (min)

G lu

c o

se ( m

m o

l l –1

)

Gdf15–/– metformin Gdf15–/– vehicle

500

1,000

1,500

A U

C ( m

m o

l l –1

m in

)

Gdf15–/–

Effect of metformin P = 0.001 Effect of genotype P = 0.06

Gdf15+/+ Gdf15–/–Gdf15+/+ Gdf15–/–Gdf15+/+

G lu

c o

se ( m

m o

l l –1

)

Effect of metformin P = 0.006

In su

lin ( n g

m l–

1 )

Effect of metformin P = 0.000363

Time (min)

G lu

c o

se ( m

m o

l l –1

)

Vehicle + IgG Metformin + IgG

Time (min)

G lu

c o

se ( m

m o

l l –1

)

Metformin + anti-GFRAL Vehicle + anti-GFRAL

0 1 2 3 4 5 6 7

Time (min)

In su

lin ( n g

m l–

1 ) Vehicle + IgG

Metformin + IgG

0 1 2 3 4 5 6 7

Time (min)

In su

lin ( n g

m l–

1 )

Vehicle + anti-GFRAL Metformin + anti-GFRAL

In su

lin ( n g

m l–

1 )

Vehicle Metformin Vehicle Metformin Vehicle Metformin Vehicle Metformin

Vehicle Metformin Vehicle Metformin Vehicle Metformin Vehicle Metformin Vehicle Metformin Vehicle Metformin

IgG Anti-GFRAL

Effect of metformin P = 0.007

200

400

600 A

U C

( n g

m l–

1 m

in )

IgG Anti-GFRAL

Effect of metformin P = 0.006

Time (min)

G lu

c o

se ( m

m o

l l –1

) Gdf15–/– vehicle Gdf15–/– metformin

Gdf15+/+ vehicle Gdf15+/+ metformin

Time (min)

G lu

c o

se ( m

m o

l l –1

)

a b c d

e f g h

i j k l

Fig. 3 | Effects of metformin on glucose homeostasis. a, ITT (0.5 U kg−1 insulin) after 11 days of metformin treatment (300 mg kg−1) in Gdf15+/+ and Gdf15−/− mice on a high-fat diet. Data are mean ± s.e.m., n = 6 per group, except Gdf15−/− vehicle, n = 7 and Gdf15+/+ vehicle, n = 5. b, AUC analysis of glucose over time in mice from a. Data are mean ± s.e.m.; P values by two-way ANOVA; interaction of genotype and metformin, P = 0.037. c, Fasting glucose (time 0) of ITT from a. Data are mean ± s.e.m.; P values by two-way ANOVA, effect of genotype, P = 0.144; interaction of genotype and metformin, P = 0.988. d, Fasting insulin (time 0) in ITT from a. Data are mean ± s.e.m.; P values by two-way ANOVA, effect of genotype; P = 0.131; interaction of genotype and metformin, P = 0.056. e, f, Glucose over time after oral GTT in metformin-treated obese mice given either IgG (e) or anti-GFR AL (f ) once weekly for five weeks (as in Fig. 2d).

AUC analysis by two-way ANOVA; effect of antibody, P = 0.031; effect of metformin, P = 0.072; interaction of antibody and metformin, P = 0.91. g, h, Insulin over time after oral GTT in mice treated as in e, f. Data are mean ± s.e.m. i, Fasting insulin (time 0) after GTT in mice treated as in e, f. Data are mean ± s.e.m., P values by two-way ANOVA; effect of antibody, P = 0.544; interaction of genotype and metformin, P = 0.691. j, AUC analysis of insulin over time in g, h. Data are mean ± s.e.m.; P values by two-way ANOVA; effect of antibody, P = 0.197; interaction of genotype and metformin, P = 0.607. k, l, Glucose over time after intraperitoneal GTT in mice on a high-fat diet given a single dose of oral metformin (300 mg kg−1) 6 h before the GTT. Data are mean ± s.e.m., n = 8 per group.

4 | Nature | www.nature.com

Article was a significant effect of metformin on glucose levels (plasma glucose AUC) that was independent of GDF15 (Extended Data Fig. 6e).

The effect of metformin in decreasing fasting glucose and insulin and improving glucose tolerance do not require GDF15. Given the a priori expected effect of weight loss on insulin sensitivity it is noteworthy that the effect of GDF15 status on insulin sensitivity as measured by insulin-tolerance test (ITT) (Fig. 3b) fell just short of statistical signifi- cance. In the follow up of the Diabetes Prevention Program study in non-diabetic individuals, weight loss after 5 years of metformin therapy was approximately 6.5% of baseline weight5. We therefore estimated the effect of a 6.5% weight loss on improvements in fasting insulin over 5 years in the Ely Study, a prospective observational population-based cohort study of men (n = 465) and women (n = 634) in the UK (mean age 52 years, mean body mass index 26 at baseline)15, showing that this mag- nitude of weight loss was associated with a reduction in fasting plasma insulin of −5.74 (−9.03, −2.45) pmol l−1 (mean ±95% confidence interval (CI)) in women and −8.78 (−16.24, −1.33) pmol l−1 in men. We conclude that although there are GDF15-independent effects of metformin on circulating levels of glucose and insulin, GDF15-dependent weight loss probably contributes to enhancing insulin sensitivity.

Source of GDF15 production We examined Gdf15 gene expression in a tissue panel obtained from mice fed a high-fat diet (for four weeks) and euthanized 6 h after a single- gavage dose of metformin (600 mg kg−1). Circulating concentrations of GDF15 increased about 5.5-fold compared with vehicle-treated mice (Extended Data Fig. 6f ) and Gdf15 mRNA was significantly increased by metformin in small intestine, colon and kidney (Fig. 4a). In situ hybridization studies demonstrated strong Gdf15 expression in crypt enterocytes in the colon and small intestine and in periglomerular renal tubular cells (Fig. 4b, Extended Data Fig. 7a, b). We confirmed these sites of tissue expression in mice fed a high-fat diet (those used in Fig. 2a) and treated with metformin for 11 days (Extended Data Fig. 8). Further, in organoids derived from human (Fig. 4c) and mouse (Fig. 4d) intestine, grown in two-dimensional (2D) transwells and treated with metformin, we observed a significant induction of Gdf15 mRNA expression and GDF15 protein secretion.

Given the proposed importance of the liver for the metabolic action of metformin, it was notable that the dominant GDF15 expression signal was not from the liver (Fig. 4a, Extended Data Figs. 7a, 8). To determine whether hepatocytes are capable of responding to biguanide drugs with an increase in GDF15, we incubated freshly isolated mouse hepatocytes (Extended Data Fig. 9a) and stem-cell derived human hepatocytes (Extended Data Fig. 9b) with metformin and found a clear induction of GDF15 expression. Additionally, acute administration of the more cell-penetrant biguanide drug phenformin to mice increased circu- lating GDF15 levels (Extended Data Fig. 9c) and markedly increased Gdf15 mRNA expression in hepatocytes (Extended Data Fig. 9d, e). We conclude that biguanides can induce GDF15 expression in many cell types but, at least when given orally to mice, Gdf15 mRNA is mostly induced in the distal small intestine, colon and kidney.

GDF15 expression has been reported to be a downstream target of the cellular integrated stress response (ISR) pathway16–18.Gdf15 mRNA levels were increased in kidney and colon 24 h after a single oral dose of metformin and these changes correlated positively with the fold eleva- tion of Chop (also known as Ddit3) mRNA (Extended Data Fig. 10a, b). As phenformin has broader cell permeability than metformin19, we used it to explore the effects of biguanides on the ISR and its relationship to GDF15 expression in cells. In mouse embryonic fibroblasts (MEFs), which do not express the organic cation transporters needed for the uptake of metformin, phenformin (but not metformin) increased EIF2α phosphorylation, ATF4 and CHOP expression (Extended Data Fig. 10c) and Gdf15 mRNA (Extended Data Fig. 10d), though the changes in EIF2a phosphorylation and ATF4 and CHOP expression were modest

compared with those induced by tunicamycin despite similar levels of Gdf15 mRNA induction. Both genetic deletion of Atf4 and small interfering (si)RNA-mediated knockdown of Chop significantly reduced phenformin-mediated induction of Gdf15 mRNA expression (Extended Data Fig. 10e, f ). In addition, phenformin induction of GDF15

0.0

0.1

0.2

0.3

0.4

G d

f1 5

re la

ti ve

e xp

re ss

io n

(A c tb

n o

rm a liz

e d

)

G d

f1 5

re la

ti ve

e xp

re ss

io n

(A c tb

n o

rm a liz

e d

)

Stomach

0.0

0.2

0.4

0.6

0.8 P = 0.0031

Proximal small intestine

Middle small intestine

0.0

0.1

0.2

0.3

0.4

0.5 P = 0.0002

Distal small intestine

Colon

P = 0.0042

Vehicle Metformin Vehicle Metformin Vehicle Metformin Vehicle Metformin

Vehicle Metformin

Kidney

P = 0.0026

Liver

0.0

0.2

0.4

0.6

0.8

White adipose tissue

0.0

0.1

0.2

0.3

0.4

0.5

Skeletal muscle

Control Metformin

Control Metformin Control Metformin

Control Metformin 0

G D

F 1

5 m

R N

A e

xp re

ss io

n (f

o ld

c h

a n

g e )

P < 0.0001

Rectal

0.8

1.0

1.2

1.4

1.6

G D

F 1 5

se c re

ti o

n (f

o ld

c h

a n

g e )

P < 0.0001 Rectal

G D

F 1

5 se

c re

ti o

n (f

o ld

c h

a n

g e )

Duodenal

P < 0.0001

G D

F 1 5

se c re

ti o

n (f o

ld c h

a n

g e )

Ileal

P < 0.0001

Distal small intestine Colon Kidney

V e h

ic le

M e tf

o rm

50 μm 100 μm 100 μm

Fig. 4 | Metformin increases GDF15 expression in the enterocytes of distal intestine and in renal tubular epithelial cells. a, Gdf15 mRNA expression (normalized to expression levels of Actb) 6 h after a single dose of oral metformin (600 mg kg−1) in tissues from wild-type mice on a high-fat diet. Data are mean ± s.e.m., n = 7 per group; P value (95% CI) by two-tailed t-test. b, In situ hybridization for Gdf15 mRNA (red spots). n = 7 per group. Representative images from the mouse with circulating GDF15 level closest to group median, treated with vehicle or metformin. Mice are from groups described in a. c, GDF15 mRNA expression (left) and GDF15 protein in supernatant (right) of human-derived 2D monolayer rectal organoids treated with metformin. Each colour represents an independent experiment. Data are mean ± s.d., n = 4; P values (95% CI) by two-tailed t-test. d, GDF15 protein in supernatants of mouse-derived 2D monolayer duodenal (left) and ileal (right) organoids treated with metformin. Each colour represents an independent experiment. Data are mean ± s.d., duodenal, n = 5; ileal, n = 3; P values (95% CI) by two-tailed t-test.

Nature | www.nature.com | 5

was markedly reduced by co-treatment with the EIF2α inhibitor ISRIB but, notably, not by the PERK inhibitor GSK2606414 (Extended Data Fig. 10g). Further, GDF15 secretion in response to metformin in mouse duodenal organoids was also significantly reduced by co-treatment with ISRIB (Extended Data Fig. 10h). However, gut organoids derived from CHOP-null mice are still able to increase GDF15 secretion in response to metformin (Extended Data Fig. 10i) indicating the exist- ence of CHOP-independent pathways under some circumstances. These data suggest that the effects of biguanides on GDF15 expression are at least partly dependent on the ISR pathway but are independent of PERK. However, the relative importance of components of the ISR pathway may vary depending on specific cell type, dose and agent used.

Our observations represent an advance in the understanding of the action of metformin, one of the world’s most frequently prescribed drugs. Metformin increases circulating GLP1 levels20–22, but its meta- bolic effects in mice are unimpaired in mice lacking the GLP1 receptor23. Metformin alters the intestinal microbiome24,25 but it is challenging to firmly establish a causal relationship between this effect and the beneficial effects of the drug26.

In this study, we present a body of data from humans, cells, organoids and mice that securely establish a major role for GDF15 in the mediation of the beneficial effects of metformin on energy balance. Whereas these effects probably contribute to the role of metformin as an insulin sensitizer, it has other effects in decreasing glucose and insulin in the absence of GDF15.

Whereas many mechanisms have been suggested for the glucoregula- tory mechanisms of metformin27, there has been less attention paid to its effects on weight. Our discoveries relating to metformin’s effects via GDF15 provide a compelling explanation for this important aspect of its action.

It is notable that the lower small intestine and colon are a major site of metformin-induced GDF15 expression. An emerging body of work strongly implicates the intestine as a major site of metformin action. Metformin increased glucose uptake into colonic epithelium from the circulation28 and a gut-restricted formulation of metformin had greater glucose-lowering efficacy than systemically absorbed formulations29. Our finding that the intestine is a major site of metformin-induced GDF15 expression provides a further mechanism through which met- formin’s action on the intestinal epithelium may mediate some of its benefits.

Online content Any methods, additional references, Nature Research reporting sum- maries, source data, extended data, supplementary information, acknowledgements, peer review information; details of author con- tributions and competing interests; and statements of data and code availability are available at https://doi.org/10.1038/s41586-019-1911-y.

1. Knowler, W. C. et al. Reduction in the incidence of type 2 diabetes with lifestyle intervention or metformin. N. Engl. J. Med. 346, 393–403 (2002).

2. Ramachandran, A. et al. The Indian Diabetes Prevention Programme shows that lifestyle modification and metformin prevent type 2 diabetes in Asian Indian subjects with impaired glucose tolerance (IDPP-1). Diabetologia 49, 289–297 (2006).

3. Lachin, J. M. et al. Factors associated with diabetes onset during metformin versus placebo therapy in the diabetes prevention program. Diabetes 56, 1153–1159 (2007).

4. Rena, G., Hardie, D. G. & Pearson, E. R. The mechanisms of action of metformin. Diabetologia 60, 1577–1585 (2017).

5. Apolzan, J. W. et al. Long-term weight loss with metformin or lifestyle intervention in the Diabetes Prevention Program Outcomes Study. Ann. Intern. Med. 170, 682–690 (2019).

6. Gerstein, H. C. et al. Growth differentiation factor 15 as a novel biomarker for metformin. Diabetes Care 40, 280–283 (2017).

7. Tsai, V. W. W., Husaini, Y., Sainsbury, A., Brown, D. A. & Breit, S. N. The MIC-1/GDF15–GFRAL pathway in energy homeostasis: implications for obesity, cachexia, and other associated diseases. Cell Metab. 28, 353–368 (2018).

8. Mullican, S. E. et al. GFRAL is the receptor for GDF15 and the ligand promotes weight loss in mice and nonhuman primates. Nat. Med. 23, 1150–1157 (2017).

9. Emmerson, P. J. et al. The metabolic effects of GDF15 are mediated by the orphan receptor GFRAL. Nat. Med. 23, 1215–1219 (2017).

10. Yang, L. et al. GFRAL is the receptor for GDF15 and is required for the anti-obesity effects of the ligand. Nat. Med. 23, 1158–1166 (2017).

11. Hsu, J. Y. et al. Non-homeostatic body weight regulation through a brainstem-restricted receptor for GDF15. Nature 550, 255–259 (2017).

12. Konopka, A. R. et al. Hyperglucagonemia mitigates the effect of metformin on glucose production in prediabetes. Cell Rep. 15, 1394–1400 (2016).

13. Preiss, D. et al. Metformin for non-diabetic patients with coronary heart disease (the CAMERA study): a randomised controlled trial. Lancet Diabetes Endocrinol. 2, 116–124 (2014).

14. McCreight, L. J. et al. Pharmacokinetics of metformin in patients with gastrointestinal intolerance. Diabetes Obes. Metab. 20, 1593–1601 (2018).

15. Forouhi, N. G., Luan, J., Hennings, S. & Wareham, N. J. Incidence of Type 2 diabetes in England and its association with baseline impaired fasting glucose: the Ely study 1990-2000. Diabet. Med. 24, 200–207 (2007).

16. Chung, H. K. et al. Growth differentiation factor 15 is a myomitokine governing systemic energy homeostasis. J. Cell Biol. 216, 149–165 (2017).

17. Li, D., Zhang, H. & Zhong, Y. Hepatic GDF15 is regulated by CHOP of the unfolded protein response and alleviates NAFLD progression in obese mice. Biochem. Biophys. Res. Commun. 498, 388–394 (2018).

18. Patel, S. et al. GDF15 provides an endocrine signal of nutritional stress in mice and humans. Cell Metab. 29, 707–718 (2019).

19. Shu, Y. et al. Effect of genetic variation in the organic cation transporter 1 (OCT1) on metformin action. J. Clin. Invest. 117, 1422–1431 (2007).

20. DeFronzo, R. A. et al. Once-daily delayed-release metformin lowers plasma glucose and enhances fasting and postprandial GLP-1 and PYY: results from two randomised trials. Diabetologia 59, 1645–1654 (2016).

21. Preiss, D. et al. Sustained influence of metformin therapy on circulating glucagon-like peptide-1 levels in individuals with and without type 2 diabetes. Diabetes Obes. Metab. 19, 356–363 (2017).

22. Bahne, E. et al. Metformin-induced glucagon-like peptide-1 secretion contributes to the actions of metformin in type 2 diabetes. JCI Insight 3, 93936 (2018).

23. Maida, A., Lamont, B. J., Cao, X. & Drucker, D. J. Metformin regulates the incretin receptor axis via a pathway dependent on peroxisome proliferator-activated receptor-α in mice. Diabetologia 54, 339–349 (2011).

24. de la Cuesta-Zuluaga, J. et al. Metformin is associated with higher relative abundance of mucin-degrading Akkermansia muciniphila and several short-chain fatty acid-producing microbiota in the gut. Diabetes Care 40, 54–62 (2017).

25. Shin, N. R. et al. An increase in the Akkermansia spp. population induced by metformin treatment improves glucose homeostasis in diet-induced obese mice. Gut 63, 727–735 (2014).

26. Forslund, K. et al. Disentangling type 2 diabetes and metformin treatment signatures in the human gut microbiota. Nature 528, 262–266 (2015).

27. Foretz, M., Guigas, B. & Viollet, B. Understanding the glucoregulatory mechanisms of metformin in type 2 diabetes mellitus. Nat. Rev. Endocrinol. 15, 569–589 (2019).

28. Massollo, M. et al. Metformin temporal and localized effects on gut glucose metabolism assessed using 18F-FDG PET in mice. J. Nucl. Med. 54, 259–266 (2013).

29. Buse, J. B. et al. The primary glucose-lowering effect of metformin resides in the gut, not the circulation: results from short-term pharmacokinetic and 12-week dose-ranging studies. Diabetes Care 39, 198–205 (2016).

Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

https://doi.org/10.1038/s41586-019-1911-y

Article Methods

Human studies We analysed samples from nine participants from a study with a placebo-controlled, double-blind crossover design (previously described in ref. 12). In brief, placebo or metformin (week 1, 500 mg twice daily; week 2, 1,000 mg twice daily) was administered following a six-week period of washout. Samples were collected in the morning after overnight fasting. The study was approved by the Mayo Clinic Insti- tutional Review Board and all participants provided written, informed consent (NCT01956929).

CAMERA was a randomized, double-blinded, placebo-controlled trial designed to investigate the effect of metformin on surrogate markers of cardiovascular disease in patients without diabetes, aged 35 to 75, with established coronary heart disease and a large waist circumference (≥94 cm in men, ≥80 cm in women) (NCT00723307). This single-centre trial enrolled 173 adults who were followed up for 18 months each. A detailed description of the trial and its results has been published previously13. In brief, participants were randomized 1:1 to 850 mg metformin or matched placebo twice daily with meals. Participants attended six monthly visits after overnight fasts and before taking their morning dose of metformin. Blood samples collected during the trial were centrifuged at 4 °C soon after sampling, separated and stored at −80 °C.

All participants provided written informed consent. The study was approved by the Medicines and Healthcare Products Regulatory Agency and West Glasgow Research Ethics Committee, and done in accordance with the principles of the Declaration of Helsinki and good clinical practice guidelines.

Serum GDF15 assays were measured in a fashion blinded to treatment allocation or timing of samples by the Cambridge Biochemical Assay Laboratory, University of Cambridge. Measurements were performed with antibodies and standards from R&D Systems (R&D Systems) using a microtitre-plate-based two-site electrochemiluminescence immuno- assay using the MesoScale Discovery assay platform (MSD).

Mouse studies Studies were carried out at two sites: NGM Biopharmaceuticals, California, and the University of Cambridge.

At NGM, all experiments were conducted with NGM IACUC approved protocols and all relevant ethical regulations were complied with throughout the course of the studies, including efforts to reduce the number of animals used. Experimental animals were kept under con- trolled light (12 h:12 h light:dark cycle, dark 18:30–06:30), temperature (22 ± 3 °C) and humidity (50 ± 20%) conditions. They were fed ad libitum on 2018 Teklad Global 18% Protein Rodent Diet containing 24 kcal% fat, 18 kcal% protein and 58 kcal% carbohydrate, or on high-fat rodent diet containing 60 kcal% fat, 20 kcal% protein and 20 kcal% carbohydrates from Research Diets D12492i, hereafter referred to as 60% HFD.

In Cambridge, all mouse studies were performed in accordance with UK Home Office Legislation regulated under the Animals (Scientific Procedures) Act 1986 Amendment, Regulations 2012, following ethi- cal review by the University of Cambridge Animal Welfare and Ethical Review Body (AWERB). They were maintained in a 12 h:12 h light:dark cycle (lights on 07:00–19:00), temperature-controlled (22 °C) facil- ity, with ad libitum access to food (RM3(E) Expanded Chow (Special Diets Services)) and water. Any mice bought from an outside supplier were acclimatised in a holding room for at least one week before study. During study periods they were fed ad libitum high-fat diet, either D12451i (45 kcal% fat, 20 kcal% protein and 35 kcal% carbohydrates, herein referred to as 45% HFD) or D12492i (Research Diets) as high- lighted in the individual study.

Sample sizes were determined on the basis of homogeneity and consistency of characteristics in the selected models and were suf- ficient to detect statistically significant differences in body weight,

food intake and serum parameters between groups. Experiments were performed with animals of a single gender in each study. Animals were randomized into the treatment groups on the basis of body weight such that the mean body weights of each group were as close to each other as possible, but without using an excess number of animals. No samples or animals were excluded from analyses. Researchers were not blinded to group allocations.

Mouse study 1, acute two-dose metformin and high-fat diet Male C57Bl6/J mice fed 60% HFD for 17 weeks were studied aged 23 weeks (body weight, mean ± s.e.m., 45.6 ± 0.8g). Metformin (Sigma- Aldrich no. 1396309) was reconstituted in water at 30 mg ml−1 for oral gavage and given in the early part of the light cycle. Terminal blood was collected by cardiac puncture into EDTA-coated tubes. GDF15 levels were measured using Mouse/Rat GDF15 Quantikine ELISA Kit (no. MGD-150, R&D Systems) according to the manufacturers’ instruc- tions. RNA was isolated from tissues using the Qiagen RNeasy Kit. RNA was quantified and 500 ng was used for cDNA synthesis (SuperScript VILO; 11754050, ThermoFisher) followed by quantitative (q)PCR. All Taqman probes were purchased from Applied Biosystems. All genes are expressed relative to 18S control probe and were run in triplicate.

Mouse study 2, acute metformin and normal diet Ad libitum group. Male C57BL6/J mice (Charles River) were studied at 11 weeks old. Five-hundred milligrams of metformin was dissolved in 20 ml water to make a working stock of 25 mg ml−1. One hour after onset of light cycle, mice received a single dose by oral gavage of either metformin at 300 mg kg−1 (Sigma, PHR1084-500MG) or a matched vol- ume of vehicle (water). Weight (mean ± s.e.m.) of control and treatment groups were 27.2 ± 0.3 g and 26.7 ± 0.2 g, respectively, on the day of study. After gavage, mice were returned to an individual cage and were euthanized at the relevant time point by terminal anaesthesia (Euthatal by intraperitoneal injection). Blood was collected into a Sarstedt Serum Gel 1.1 ml Micro Tube, left for 30 min at room temperature, then spun for 5 min at 10,000g at 40 °C before being frozen and stored at −80 °C until assayed. Mouse GDF15 levels were measured using a Mouse GDF15 DuoSet ELISA (R&D Systems) which had been modified to run as an electrochemiluminescence assay on the Meso Scale Discovery assay platform.

Fasted group. Mice, conditions and methods as in the ad libitum group, except that male mice were studied at 9 weeks old and 12 h before administration of metformin; mice and bedding were transferred to new cages with no food in the hopper. Weight (mean ± s.e.m.) after fasting and on day of gavage were 22.3 ± 0.5 g and 23.2 ± 0.7 g for control and treatment groups, respectively.

Mouse study 3, metformin and high-fat diet, Gdf15−/− and wild- type mice C57BL/6N-Gdf15tm1a(KOMP)Wtsi/H mice (referred to as Gdf15−/− mice) were obtained from the MRC Harwell Institute, which distributes these mice on behalf of the European Mouse Mutant Archive (https://www. infrafrontier.eu/). The MRC Harwell Institute is also a member of the International Mouse Phenotyping Consortium (IMPC) and has received funding from the MRC for generating and/or phenotyping the C57BL/6N-Gdf15tm1a(KOMP)Wtsi/H mice. The research reported in this publication is solely the responsibility of the authors and does not necessarily represent the official views of the Medical Research Council. Associated primary phenotypic information may be found at https://www.mousephenotype.org/. Details of the alleles have been published30–32.

Experimental cohorts of male Gdf15−/− and wild-type mice were gen- erated by het × het breeding pairs. Mice were aged between 4.5 and 6.5 months. One week before study start, mice were single-housed and three days before the first dose of metformin treatment, mice were

https://www.infrafrontier.eu/

https://www.mousephenotype.org/

transferred from standard chow to 60% high-fat diet. On the day of first gavage, body weight of study groups (mean ± s.e.m.) were 38.2 ± 1.0 g vs 38.8 ± 0.6 g for wild-type vehicle and metformin treatments, respec- tively, and 37.9 ± 0.8 g vs 37.0 ± 1.4 g for Gdf15−/− vehicle and metformin treatments, respectively. Each mouse received a daily gavage of either vehicle or metformin for 11 days, and their body weight and food intake was measured daily in the early part of the light cycle. One data point of 25 food intake points collected on day 11 of the study was lost owing to technical error (Gdf15+/+ mouse, metformin). On day 11, mice were euthanized by terminal anaesthesia 4 h after gavage, and blood was obtained as in mouse study 2. Tissues were fresh frozen on dry ice and kept at −80 °C until the day of RNA extraction.

Mouse study 4, metformin and high-fat diet, Gfral−/− mice Gfral−/− mice on a mixed 129/SvEv-C57BL/6 background were purchased from Taconic (TF3754) and backcrossed for 10 generations to >99% C57BL/6 background at the NGM animal facility. Experimental cohorts were generated by het × het breeding pairs. Study design was as in study 3, except that terminal blood was collected into EDTA-coated tubes.

Mouse study 5, anti GFRAL antibody, metformin and high-fat diet Anti-GFRAL antibody generation. Anti-GFRAL monoclonal antibodies were generated by immunizing C57Bl/6 mice with recombinant puri- fied GFRAL ECD-hFc fusion protein, which was purified via sequential protein-A affinity and size exclusion chromatography techniques using MabSelect SuRe and Superdex 200 purification media, respectively (GE Healthcare), as described in patent US10174119B2 (https://patents. google.com/patent/US10174119B2/en). An in-house pTT5 hIgK hIgG1 expression vector was engineered to include the DEVDG (caspase-3) proteolytic site N-terminal to the Fc domain. The heavy chains of anti- GFRAL monoclonal antibodies were subcloned via EcoR1 and HindIII sites of an in-house engineered pTT5 hIgK hIgG1 caspase-cleavable vector. Light chains of anti-GFRAL monoclonal antibodies were also subcloned in the EcoR1 and HindIII sites in the pTT5 hIgK hKappa vec- tor. The antibodies were transiently expressed in Expi293 cells (Thermo Fisher Scientific) transfected with the pTT5 expression vector, and purified from conditioned media by sequential protein-A affinity and size-exclusion chromatography using MabSelect SuRe and Superdex 200 purification media, respectively (GE Healthcare). All purified an- tibody material was verified endotoxin-free and formulated in PBS for in vitro and in vivo studies. Characterization of anti-GFRAL functional blocking antibodies was carried out using a cell-based RET/GFRAL lucif- erase gene reporter assays, in vitro binding studies (ELISA and Biacore) and in vivo studies, as described in patent number US10174119B2 (https://patents.google.com/patent/US10174119B2/en).

In all studies with anti-GFRAL, purified recombinant non-target- ing IgG on the same antibody framework was used as control. Met- formin was mixed with food paste made from the 60 kcal% fat diet (Research Diet no. D12492) using a food blender at a concentration to achieve an approximate consumption of 300 mg kg−1 metformin per day per mouse. Male mice were single-housed throughout and at the start of study period, body weight (mean ± s.e.m.) was 43.7 ± 1.4 g (vehicle + control IgG), 42.3 ± 1.4 g (vehicle + anti-GFRAL), 41.9 ± 1.1 g (metformin + control IgG) and 43.3 ± 1.3 g (metformin + anti-GFRAL). Recombinant antibodies were administered by subcutaneous injec- tion in the early part of the light cycle. Body composition (lean and fat mass) was analysed by ECHO MRI M113 mouse system (Echo Medi- cal Systems). The metabolic parameters oxygen consumption (VO2) and carbon dioxide production (VCO2) were measured by an indirect calorimetry system (LabMaster TSE System) in open-circuit sealed chambers. Measurements were performed for the dark (from 18:00 to 06:00) or light (from 06:00 to 18:00) period under ad libitum feeding conditions. Mice were placed in individual metabolic cages and allowed to acclimate for a period of 24 h before data collection every 30 min.

Finally, mice underwent a GTT. Mice were fasted for 6 h (07:00–13:00) in a clean cage. Blood samples (~30 μl) were collected as baseline before oral GTT. Mice were orally gavaged with 1 g kg−1 of 20% glucose solution with a dosing volume of 5 ml kg−1. Blood samples were then collected through a tail nick into K2-EDTA-coated tubes (SARSTEDT Microvette; no. 20.1278.100) at 15, 30, 60 and 120 min after glucose challenge. Blood samples were centrifuged at 4 °C and the separated plasma was stored at −20 °C until used for plasma glucose and insulin assays. Glucose assay reagents were obtained from Wako (no. 439-90901) and the insulin ELISA kit was obtained from ALPCO (no. 80-INSMSU-E01).

Mouse study 6, ITT after metformin and high-fat diet, Gdf15−/− and wild-type mice Mice generation and protocol are the same as in study 3, except mice were aged 4 to 6 months. On the day of the first gavage, body weights (mean ± s.e.m.) of study groups were 35.1 ± 1.2 g (wild-type, vehicle), 35.05 ± 1.2 g (wild-type, metformin), 35.08 ± 1.02 g (Gdf15−/−, vehicle) and 35.02 ± 1.47 g (Gdf15−/−, metformin). On day 11, after the final dose of metformin, mice were fasted for 4 h. Baseline venous blood sample was collected into heparinised capillary tube for insulin measurement and blood glucose was measured using approximately 2 μl blood drops using a glucometer (AlphaTrak2; Abbot Laboratories) and glucose strips (AlphaTrak2 test 2 strips, Abbot Laboratories, Zoetis). Mice were given intraperitoneal injection of insulin (0.5 U kg−1, Actrapid, NovoNordisk) and serial mouse glucose levels were measured at time points indicated. Mice were killed by terminal anaesthesia as in study 2. Mouse insulin was measured using a Meso Scale Discovery two-plex mouse metabolic immunoassay kit according to the manufacturer’s instructions and using calibrators provided by Meso Scale Diagnostics. Serum metformin levels were quantified using a stable isotope dilution liquid chromatography–mass spectrometry (LC–MS/MS) method described previously33.

Mouse study 7, GTT after metformin and high-fat-diet, Gdf15−/− and wild-type mice Mice generation was as in study 3, except using female mice aged 3.5 to 5.5 months. Two groups of mice (Gdf15+/+ and Gdf15−/− littermates, body weight (mean ± s.e.m.) 24.1 ± 1.4 g and 24.3 ± 1.3 g, respectively) were fed 60% HFD for two weeks. Each genotype was then further split into vehicle or metformin (300 mg kg−1) treatment groups, given a single gavage dose at 08:00 and fasted for 6 h. At time of GTT, body weights (mean ± s.e.m.) of study groups were 26.4.1 ± 1.5 g (wild-type, vehicle), 26.5 ± 1.0 g (wild-type, metformin, 25.6 ± 1.2 g (Gdf15−/−, vehicle) and 27.1 ± 1.3 g (Gdf15−/−, metformin); one-way ANOVA, P = 0.8722. Baseline testing was as in mouse study 6. Mice then received a single dose of 20% glucose intraperitoneally (2 mg g−1) with serial measurement of glucose levels at time points indicated. Euthanasia and insulin analysis were performed as in mouse study 6.

Mouse study 8, acute single high-dose metformin and high-fat diet Male C57BL6/J mice (Charles River) aged 14 weeks were switched from standard chow to 45% HFD (D12451i) for 1 week then 60% HFD (D12492i,) for 3 weeks. At the time of the study (18 weeks old) body weights (mean ± s.e.m.) were 40.4 ± 1.2 g and 41.1 ± 1.3 g for the vehi- cle and metformin groups, respectively. Five-hundred milligrams of metformin (Sigma, PHR1084-500MG) were dissolved in 8.35 ml water to make a working stock of 60 mg ml−1. Mice received a single dose by oral gavage of either 600 mg kg−1 metformin or a matched volume of vehicle (water). They were returned to ad libitum 60% HFD and 6 h later blood was collected as in study 2. Tissue samples for RNA analysis were collected into Lysing Matrix D homogenization tubes (MP Biomedicals) on dry ice and stored at −80 °C until they were processed. Intestine between pylorus of stomach and caecum was laid out into three equal parts, with tissue taken from the midpoint of each third labelled as

https://patents.google.com/patent/US10174119B2/en

Article ‘proximal’, ‘middle’ and ‘distal’ (adapted from ref. 34). The colon section was from the midpoint between caecum and anus. Tissues for in situ hybridization were dissected and placed into 10% formalin/PBS for 24 h at room temperature, transferred to 70% ethanol and processed into paraffin. Five-micrometre sections were cut and mounted onto Superfrost Plus (Thermo Fisher Scientific). Detection of mouse Gdf15 was performed on formalin-fixed paraffin-embedded sections using Advanced Cell Diagnostics (ACD) RNAscope 2.5 LS Reagent Kit-RED (no. 322150) and RNAscope LS 2.5 Probe Mm-Gdf15-O1 (no. 442948) (ACD). In brief, sections were baked for 1 h at 60 °C before loading onto a Bond RX instrument (Leica Biosystems). Slides were deparaffinized and rehy- drated on board before pre-treatments using Epitope Retrieval Solution 2 (no. AR9640, Leica Biosystems) at 95 °C for 15 min, and ACD Enzyme from the LS Reagent kit at 40 °C for 15 min. Probe hybridization and signal amplification was performed according to the manufacturer’s instructions. Fast red detection of mouse Gdf15 was performed on the Bond RX using the Bond Polymer Refine Red Detection Kit (Leica Biosystems, no. DS9390) according to the ACD protocol. Slides were then counterstained with haematoxylin, removed from the Bond RX and were heated at 60 °C for 1 h, dipped in xylene and mounted using EcoMount Mounting Medium (Biocare Medical, no. EM897L).

Slides imaged on an automated slide-scanning microscope (Axioscan Z1 and Hamamatsu orca flash 4.0 V3 camera) using a 20× objective with a numerical aperture of 0.8. Hybridization specificity was confirmed by the absence of staining in Gdf15−/− mice.

RNA extraction was carried out with approximately 100 mg of tis- sue in 1 ml Qiazol Lysis Reagent (Qiagen 79306l) using Lysing Matrix D homogenization tube and Fastprep 24 Homogenizer (MP Biomedi- cals) and Qiagen RNeasy Mini Kit (no. 74106) with DNase1 treatment following manufacturers’ protocols. Five-hundred nanograms of RNA was used to generate cDNA using Promega M-MLV reverse tran- scriptase followed by TaqMan qPCR in triplicate for GDF15. Samples were normalized to Actb. TaqMan Probes: Mm00442228 m1 GDF15, Mm02619580_g1 Act B, TaqMan; 2× universal PCR Master mix (Applied Biosystems Thermo Fisher, 4318157); QuantStudio 7 Flex Real time PCR system (Applied Biosystems Life Technologies).

Mouse study 9, acute phenformin and normal diet, wild-type mice Male C57BL6/J mice aged 14 weeks with supplier, protocol and methods as in study 2, except that phenformin (Sigma PHR1573, 500 mg) was used instead of metformin.

Organoid studies Duodenal and ileal mouse organoid line generation, maintenance and 2D culture was performed as previously described35. CHOP-null mice were a gift from J. Goodall, with a line from Jackson Laboratory (B6.129S(Cg)-Ddit3tm2.1Dron/J, stock no. 005530). Human rectal orga- noids (experiments approved by the Research Ethics Committee under licence number 09/H0308/24) were generated from fresh surgical specimens (Tissue Bank, Addenbrooke’s Hospital (Cambridge, UK)) following a modified protocol35,36. In brief, rectal tissue was chopped into 5-mm fragments and incubated in 30 mM EDTA for 3 × 10 min, with tissue shaken in PBS after each EDTA treatment to release intestinal crypts. The isolated crypts were then further digested using TrypLE (Life Technologies) for 5 min at 37 °C to generate small cell clusters. These were then seeded into basement membrane extract (BME, R&D Technology), with 20-μl domes polymerized in multiwell (48) dishes for 30–60 min at 37 °C. Organoid medium36 was then overlaid and changed three times per week. Human organoids were passaged every 14–21 days using TrypLE digestion for 15 min at 37 °C, followed by mechanical shearing with rigorous pipetting to break organoids up into small clusters, which were then seeded as before in BME. For Transwell experiments, TrypLE-digested organoids were seeded onto Matrigel (Corning)-coated (2% for 60 min at 37 °C) polyethylene terephthalate

cell culture inserts, pore size 0.3 μm (Falcon) in organoid medium supplemented with Y-27632 (R&D Technology). Organoids were observed through the transparent cell inserts to ensure 2D culture formation (allowing apical cell access for drug treatments). Medium was changed after 2 days and then switched on day 3 to a differentia- tion medium with Wnt3A-conditioned medium reduced to 10% and SB202190 or nicotinamide omitted from culture for 5 days.

For GDF15 secretion experiments, 2D cultured organoid cells were treated for 24 h with indicated drugs, with medium then collected and GDF15 measured at the Core Biochemical Assay Laboratory (Cam- bridge) using the human or mouse GDF15 assay kit as outlined in the CAMERA human study and mouse study 2 above.

RNA was extracted using TRI reagent (Sigma), with any contami- nated DNA eliminated using DNA-free removal kit (Invitrogen). Purified RNA was then reverse-transcribed using superscript II (Invitrogen) as per the manufacturer’s protocol. Quantitative PCR with reverse transcription was performed on a QuantStudio 7 (Applied Biosystems) using Fast Taqman mastermix and the following probes (Applied Biosys- tems): human GDF15 (Hs00171132_m1) and human ACTB (Hs01060665_ g1). Gene expression was measured relative to β-actin in the same sample using the ΔCt method, with fold (relative to control) shown for each experiment.

Hepatocyte studies Primary mouse hepatocyte isolation and culture. Hepatocytes from 8 to 12-week-old C57B6J male mice were isolated by retrograde, non-recirculating in situ collagenase liver perfusion. In brief, livers were perfused with modified Hanks medium without calcium (8.0 g l−1 NaCl, 0.4 g l−1 KCl, 0.2 g l−1 MgSO4.7H2O, 0.12 g l

−1 Na2HPO4.2H2O, 0.12 g l−1 KH2PO4, 3 g l

−1 Hepes, 0.342 g l−1 EGTA and 0.05 g l−1) followed by digestion with perfusion medium supplemented with calcium (0.585 g l−1 CaCl2.2H2O) and 0.5 mg ml

−1 collagenase IV (Sigma, C5138). The digested liver was removed and washed using chilled DMEM:F12 (Sigma) medium containing 2 mM l-glutamine, 10% FBS, 1% penicillin/strepto- mycin (Invitrogen). Viable cells were collected by Percoll (Sigma) gra- dient. The final pellet was resuspended in the same DMEM:F12 media. Cell viability was greater than 90%. Hepatocytes were plated onto primaria plates (Corning). Hepatocytes were allowed to recover and attach for 4–6 h before replacement of the medium overnight before stress treatments the following day for the times and concentrations indicated.

Generation and culture of iPSC-derived human hepatocytes. The human induced pluripotent (iPS) cell line A1ATDR/R used in this work was derived as previously described37,38 under approval by the regional research ethics committee (reference number 08/H0311/201). iPS cells were maintained in Essential 8 chemically defined media39 supplement- ed with 2 ng ml−1 Tgf-β (R&D) and 25 ng ml−1 FGF2 (R&D), and cultured on plates coated with 10 μg ml−1 Vitronectin XFTM (STEMCELL Technolo- gies). Colonies were regularly passaged by short-term incubation with 0.5 mM EDTA in PBS. For hepatocyte differentiation, colonies were dis- sociated into single cells following incubation with StemPro Accutase Cell Dissociation Reagent (Gibco) for 5 min at 37 °C. Single cell suspen- sions were seeded on plates coated with 10 μgml−1 Vitronectin XFTM (STEMCELL Technologies) in maintenance media supplemented with 10 μM ROCK Inhibitor Y-27632 (Selleckchem) and grown for up to 72 h before differentiation. Hepatocytes were differentiated as previously reported40, with minor modifications as listed. In brief, following endo- derm differentiation, anterior foregut specification was achieved after 5 days of culture with RPMI-B27 differentiation media supplemented with 50 ng ml−1 Activin A (R&D)40. Foregut cells were further differenti- ated into hepatocytes with HepatoZYME-SFM (Gibco) supplemented with 2 mM l-glutamine (Gibco), 1% penicillin-streptomycin (Gibco), 2% non-essential amino acids (Gibco), 2% chemically defined lipids (Gibco), 14 μg ml−1 of insulin (Roche), 30 μg ml−1 of transferrin (Roche),

50 ng ml−1 hepatocyte growth factor (R&D), and 20 ng ml−1 oncostatin M (R&D), for up to 27 days.

Cellular studies on ISR Chemicals and reagents. Tunicamycin and ISRIB were purchased from Sigma-Aldrich. Metformin and Phenformin was purchased from Cayman Chemicals and GSK2606414 from Calbiochem. The antibody for GDF15 and CHOP (sc-7351) were obtained from Santa Cruz. Phos- pho S51 EIF2α (ab32157) and Calnexin (ab75801) were purchased from Abcam. D. Ron provided the antibody for ATF4.

Eukaryotic cell lines and treatments. Mouse embryonic fibroblast (MEF) cell lines were obtained from D. Ron (CIMR/IMS, Cambridge) and maintained as previously described18. MEFs were transfected with 30 nM control siRNA or a smartpool on-target plus siRNA for mouse Chop (Dharmacon L-062068-00-0005) using Lipofectamine RNAi MAX (Inv- itrogen) according to the manufacturer’s instruction. 48 h post siRNA transfection, cells were processed for RNA and protein expression analysis. All cells were maintained at 37 °C in a humidified atmosphere of 5% CO2 and seeded onto 6- or 12-well plates before stress treat- ments for the times and concentrations indicated. Vehicle treatments (for example, DMSO) were used for control cells when appropriate.

RNA isolation, cDNA synthesis and qPCR. Following treatments, cells were lysed with Buffer RLT (Qiagen) containing 1% 2-mercaptoethanol and processed through a Qiashredder with total RNA extracted us- ing the RNeasy isolation kit according to manufacturer’s instructions (Qiagen). RNA concentration and quality was determined by Nanodrop. 400 ng–500 ng of total RNA was treated with DNase1 (Thermofisher Sci- entific) and then converted to cDNA using MMLV Reverse Transcriptase with random primers (Promega). Quantitative RT–PCR was carried out with either TaqMan Universal PCR Master Mix or SYBR Green PCR mas- ter mix on the QuantStudio 7 Flex Real time PCR system (Applied Biosys- tems). All reactions were carried out in either duplicate or triplicate and Ct values were obtained. Relative differences in gene expression were normalized to the expression levels of the housekeeping genes HPRT or GAPDH for cell analysis, using the standard curve method. Primers used for this study: mouse Gdf15 (Mm00442228_m1, ThermoFisher Scientific), human GDF15 (Hs00171132_m1, ThermoFisher Scientific), human GAPDH (Hs02758991_g1, ThermoFisher Scientific), mouse Hprt (forward AGCCTAAGATGAGCGCAAGT, reverse GGCCACAGGACTAG AACACC).

Immunoblotting. Following treatments, cells were washed twice with ice-cold D-PBS and proteins collected using RIPA buffer supplemented with cOmplete protease and PhosStop inhibitors (Sigma). The lysates were cleared by centrifugation at 13,000 rpm for 15 min at 4 °C, and protein concentration determined by a Bio-Rad DC protein assay. Typi- cally, 20–30 μg of protein lysates were denatured in NuPAGE 4 × LDS sample buffer and resolved on NuPage 4–12% Bis-Tris gels (Invitrogen) and the proteins were transferred by iBlot (Invitrogen) onto nitrocel- lulose membranes. The membranes were blocked with 5% non-fat dry milk or 5% BSA (Sigma) for 1 h at room temperature and incubated with the antibodies described in the reagents section. Following a 16 h incubation at 4 °C, all membranes were washed five times in Tris- buffered saline/0.1% Tween-20 before incubation with horseradish peroxidase (HRP)-conjugated anti-rabbit immunoglobulin G (IgG) or HRP-conjugated anti-mouse IgG (Cell Signalling Technologies). The bands were visualized using Immobilon Western Chemiluminescent HRP Substrate (Millipore). All images were acquired on the ImageQuant LAS 4000 (GE Healthcare).

Statistical analyses CAMERA data were analysed using a mixed linear model with restricted maximum likelihood to investigate the metformin effect on GDF15. This is analogous to conducting a repeated measures ANOVA, but is

a more flexible analysis and allows for missing observations within subjects. The 0–18 months difference in weight and GDF15 correlation was tested using Spearman’s coefficient. CAMERA data were analysed using STATA v.15.1.

Other statistical analyses were performed using Prism 7 and Prism 8, using unpaired two-tailed t-tests, or two-way ANOVA, with multiple comparison adjustment by Tukey’s or Sidak’s test. Metabolic rate was determined using ANCOVA with energy expenditure as the dependent variable, body weight as a covariate and treatment as a fixed factor. ANCOVA and analyses of glucose and ITT in mice were performed using SPSS 25 (IBM).

Reporting summary Further information on research design is available in the Nature Research Reporting Summary linked to this paper.

Data availability Source Data for Figs. 1–4 and Extended Data Figs. 1–6, 8–10 are provided with the paper. Other data that support the findings of this study are available from the corresponding authors upon request. The CAMERA trial dataset is held at the University of Glasgow and is available on request from the investigators subject to a signed agreement operating within the confines of the original ethics application. 30. Skarnes, W. C. et al. A conditional knockout resource for the genome-wide study of

mouse gene function. Nature 474, 337–342 (2011). 31. Bradley, A. et al. The mammalian gene function resource: the International Knockout

Mouse Consortium. Mamm. Genome 23, 580–586 (2012). 32. Pettitt, S. J. et al. Agouti C57BL/6N embryonic stem cells for mouse genetic resources.

Nat. Methods 6, 493–495 (2009). 33. McNeilly, A. D., Williamson, R., Balfour, D. J., Stewart, C. A. & Sutherland, C. A high-fat-

diet-induced cognitive deficit in rats that is not prevented by improving insulin sensitivity with metformin. Diabetologia 55, 3061–3070 (2012).

34. Ortega-Cava, C. F. et al. Strategic compartmentalization of Toll-like receptor 4 in the mouse gut. J. Immunol. 170, 3977–3985 (2003).

35. Goldspink, D. A. et al. Mechanistic insights into the detection of free fatty and bile acids by ileal glucagon-like peptide-1 secreting cells. Mol. Metab. 7, 90–101 (2018).

36. Sato, T. et al. Long-term expansion of epithelial organoids from human colon, adenoma, adenocarcinoma, and Barrett’s epithelium. Gastroenterology 141, 1762–1772 (2011).

37. Rashid, S. T. et al. Modeling inherited metabolic disorders of the liver using human induced pluripotent stem cells. J. Clin. Invest. 120, 3127–3136 (2010).

38. Yusa, K. et al. Targeted gene correction of α1-antitrypsin deficiency in induced pluripotent stem cells. Nature 478, 391–394 (2011).

39. Chen, G. et al. Chemically defined conditions for human iPSC derivation and culture. Nat. Methods 8, 424–429 (2011).

40. Hannan, N. R., Segeritz, C. P., Touboul, T. & Vallier, L. Production of hepatocyte-like cells from human pluripotent stem cells. Nat. Protoc. 8, 430–437 (2013).

Acknowledgements The CAMERA trial is funded by a project grant from the Chief Scientist Office, Scotland (CZB/4/613). D.P. is supported by a University of Oxford British Heart Foundation Centre of Research Excellence Senior Transition Fellowship (RE/13/1/30181). N.S. and P.W. acknowledge support from a BHF Centre of Excellence award (COE/ RE/18/6/34217). We thank P. Barker, K. Burling and other members of the Cambridge Biochemical Assay Laboratory (CBAL) .This project is supported by the National Institute for Health Research (NIHR) Cambridge Biomedical Research Centre. The views expressed are those of the authors and not necessarily those of the NIHR or the Department of Health and Social Care. A.P.C., D. Rimmington, J.A.T., I.C., Y.C.L.T. and G.S.H.Y. are supported by the Medical Research Council (MRC Metabolic Diseases Unit (MC_UU_00014/1)). Mouse studies in Cambridge are supported by S. Grocott and the Disease Model Core, with pathology support from J. Warner and the Histopathology Core (MRC Metabolic Diseases Unit (MC_UU_00014/5) and Wellcome Trust Strategic Award (100574/Z/12/Z). D.B.S. and S.O. are supported by the Wellcome Trust (WT 107064 and WT 095515/Z/11/Z), the MRC Metabolic Disease Unit (MC_ UU_00014/1) and The National Institute for Health Research (NIHR) Cambridge Biomedical Research Centre and NIHR Rare Disease Translational Research Collaboration. We thank J. Jones and other members of the Histopathology and ISH Core Facility, Cancer Research UK Cambridge Institute, University of Cambridge, Li Ka Shing Centre. D. Ron is supported by a Wellcome Trust Principal Research Fellowship (Wellcome 200848/Z/16/Z) and a Wellcome Trust Strategic Award to the Cambridge Institute for Medical Research (Wellcome 100140). A.V.-P., S.R.-C. and S.V. are supported by the BHF (RG/18/7/33636) and MRC (MC_UU_00014/2). A.M. is supported by a studentship from the Experimental Medicine Training Initiative/ AstraZeneca. R.A.T. and L.V. are supported by an ERC advanced grant NewChol and core support from the Wellcome Trust and Medical Research Council to the Wellcome–Medical Research Council Cambridge Stem Cell Institute. M.Y., D.A.G., E.L.M., F.M.G. and F.R. are supported by the MRC (MC_UU_00014/3) and Wellcome Trust (106262/Z/14/Z and 106263/Z/14/Z). M.Y. is supported by a BBSRC-DTP studentship. A.R.K., R.R.E. and K.S.N. are supported by NIH Grants R21 AG60139, UL1 TR000135 and T32DK007352 and acknowledge K. Klaus for technical assistance. N.J.W. is supported by the MRC (MC_UU_12015/1) and is an

Article NIHR Senior Investigator. We acknowledge J. Luan for statistical assistance. CHOP-null mice were a gift from J. Goodall.

Author contributions The overall conceptualization of studies included in this work was done by A.P.C., N.S., D.B.S., B.B.A. and S.O.: these authors contributed equally to this work. A.P.C., M.C., P.T., D. Rimmington, I.C. and Y.C.L.T. designed, managed, performed and analysed data from mouse experiments. S.V. designed experiments and analysed data. A.M. and G.S.H.Y. contributed to conceptualization of experiments and data analysis. J.A.T. performed in situ hybridization experiments. S.P. designed, managed and performed cell based assays along with E.L.M., S.R.-C., R.A.T., H.P.H., A.V-P., L.V. and D. Ron. J.T.-J.H. performed measurement of serum metformin levels. M.Y., D.A.G., F.M.G. and F.R. designed, performed and analysed organoid experiments. A.R.K., R.R.E. and K.S.N. designed and performed short-term metformin studies in humans. N.J.W. analysed the Ely Study Cohort. P.W., D.P. and N.S. designed, analysed and interpreted data arising from the CAMERA study. A.P.C., D.B.S., B.B.A. and S.O. wrote the paper, which was reviewed and edited by all the authors.

Competing interests P.W. has received grant support from Roche Diagnostics, AstraZeneca and Boehringer Ingelheim. N.S. has consulted for AstraZeneca, Boehringer Ingelheim, Eli Lilly, Napp, Novo Nordisk and Sanofi, and received grant support from Boehringer Ingelheim. M.C., P.T. and B.B.A. are or were employees of NGM Biopharmaceuticals and may hold NGM stock or stock options. F.R. and F.M.G. have received support from AstraZeneca and Eli Lilly. F.M.G. has provided remunerated consultancy services to Kallyope. S.O. has provided remunerated consultancy services to Pfizer, AstraZeneca, Novo-Nordisk and ERX Pharmaceuticals. All other authors declare no competing interests.

Additional information Supplementary information is available for this paper at https://doi.org/10.1038/s41586- 019-1911-y. Correspondence and requests for materials should be addressed to A.P.C. or S.O. Peer review information Nature thanks Samuel Breit, Daniel Drucker, Jerrold M. Olefsky, and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. Reprints and permissions information is available at http://www.nature.com/reprints.

https://doi.org/10.1038/s41586-019-1911-y

http://www.nature.com/reprints

Extended Data Fig. 1 | Expanded CAMER A dataset. a, Linear association between change in body weight and change in plasma GDF15 between 0 and 18 months among metformin-treated participants (n = 74, Spearman correlation r = –0.26, two-sided P = 0.024). The red line is the linear regression slope, and grey area is 95% CI for the slope. b, Absolute and relative differences in plasma GDF15 concentration between metformin and placebo groups at each time point (total 625 observations in 173 participants). c, d, Individual measures of

plasma GDF15 levels in placebo group (c) and metformin group (d) over time. e, Plasma GDF15 concentration (95% CI) in overweight or obese non-diabetic participants with known cardiovascular disease randomized to metformin or placebo in CAMER A; modelled using a mixed linear model as per Fig. 1 and grouped as ‘all participants’ and ‘all participants not reporting diarrhoea and vomiting’. Model includes all participants.

Article

Extended Data Fig. 2 | Effect of single oral dose of metformin in chow-fed mice. Serum GDF15 levels in male mice measured 2, 4 or 8 h after a single gavage dose of metformin (300 mg kg−1). a, Mice fed ad libitum overnight before gavage. b, Mice fasted for 12 h before gavage. Data are mean ± s.e.m. (a; n = 6 per group, b; n = 4 per group); P values by two-way ANOVA with Tukey’s correction for multiple comparisons.

Extended Data Fig. 3 | Body weight changes with metformin treatment in mice with disrupted GDF15–GFR AL signalling. a, Absolute body weight in Gdf15+/+ and Gdf15−/− mice on a high-fat diet treated with metformin (300 mg kg−1 day−1) for 11 days, mice as in Fig. 2a. Data are mean ± s.e.m., P values by two-way ANOVA with Tukey’s correction for multiple comparisons. b, Absolute body weight in high-fat diet-fed Gfral+/+ and Gfral−/− mice given an

oral dose of metformin (300 mg kg−1) once daily for 11 days, mice as in Fig. 2c. Data are mean ± s.e.m. c, Absolute body weight of metformin-treated, obese mice dosed with an anti-GFR AL antagonist antibody or with control IgG weekly for five weeks, starting four weeks after initial metformin exposure; mice as in Fig. 2d. Data are mean ± s.e.m. P values by two-way ANOVA with Tukey’s correction for multiple comparisons.

Article

Extended Data Fig. 4 | Response of high-fat diet-fed Gdf 15−/− and Gfral−/− mice to metformin. a, Circulating GDF15 levels in high-fat diet-fed Gdf15+/+ and Gdf15−/− mice given oral dose of metformin (300 mg kg−1) once daily for 11 days. Data are mean ± s.e.m., mice as in Fig. 2a. All Gdf15−/− samples were below lower limit of the assay (<2 pg ml−1); P values by two-way ANOVA with Tukey’s correction for multiple comparisons. b, Circulating GDF15 levels in high-fat diet-fed Gfral+/+ and Gfral−/− mice given oral dose of metformin (300 mg kg−1) once daily for 11 days. Data are mean ± s.e.m., mice as in Fig. 2c; P values by two- way ANOVA with Tukey’s correction for multiple comparisons. c, Cumulative food intake in high-fat diet fed Gfral+/+ and Gfral−/− mice on a high-fat diet given an oral dose of metformin (300 mg kg−1) once daily for 11 days. Data are

mean ± s.e.m., mice as in Fig. 2c; no statistically significant difference in vehicle versus metformin by two-way ANOVA. d, Fat mass (left) and lean mass (right) in metformin-treated obese mice dosed with anti-GFR AL antagonist antibody weekly for five weeks, starting four weeks after initial metformin exposure (mice as in Fig. 2d). Body composition was measured using MRI after 4 weeks of metformin exposure, before receiving anti-GFR AL (week 4), after 6 weeks of metformin exposure and 2 weeks after receiving anti-GFR AL (week 6) and after 9 weeks of metformin exposure and 5 weeks after receiving anti-GFR AL (week 9). Data are mean ± s.e.m. (n = 7, vehicle + control IgG and metformin + anti-GFR AL; n = 8, other groups); P values by two-way ANOVA with Tukey’s correction for multiple comparisons.

Extended Data Fig. 5 | Response of second, independent cohort of high-fat diet fed Gdf 15+/+ and Gdf 15−/− mice to metformin. a–c, Percentage change in body weight (a), absolute body weight (b) and cumulative food intake (c) of Gdf15+/+ and Gdf15−/− mice on a high-fat diet treated with metformin (300 mg kg−1 day−1) for 11 days. Data are mean ± s.e.m. (n = 6 per group, except

Gdf15−/− vehicle, n = 7); P values by two-way ANOVA with Tukey’s correction for multiple comparisons. d, Circulating metformin levels in mice 6 h after final dose of metformin on day 11. Data are mean ± s.e.m. (n = 6 per group, except Gdf15 +/+ vehicle, n = 4; Gdf15−/− vehicle, n = 7); P values by two-way ANOVA with Tukey’s correction for multiple comparisons.

Article

Extended Data Fig. 6 | Glucose, insulin and GDF15 response to metformin. a, Fasting glucose from oral GTT as in Fig. 3e, f. ANOVA; effect of antibody, P = 0.028; effect of metformin, P = 0.271; interaction of antibody and metformin, P = 0.707. b, Circulating GDF15 in mice undergoing intraperitoneal GTT after a single dose of metformin as in Fig. 3 k, l. P values by two-way ANOVA with Tukey’s correction for multiple comparisons. c, d, Fasting glucose (c) and fasting insulin (d) at time 0 of intraperitoneal GTT as in Fig. 3k, l; not

statistically significant by two-way ANOVA. e, AUC analysis of glucose levels as in Fig. 3k, l. P values by two-way ANOVA, effect of genotype, P = 0.392; interaction of genotype and metformin, P = 0.883. a–e, Data all mean ± s.e.m. f, Circulating GDF15 levels in high-fat-diet-fed Gdf15+/+ mice after single oral dose of metformin (600 mg kg−1). Samples were collected 6 h after dosing, data are mean ± s.e.m., n = 7 per group; P values (95% CI) by two-tailed t-test.

Extended Data Fig. 7 | In situ hybrididation for Gdf 15 mRNA expression in gut, liver and kidney. a, Representative images from the mouse with circulating GDF15 level closest to the group median shown in Fig. 4b, with images from other regions of the gut and from liver. b, In situ hybridization for

Gdf15 mRNA expression (red spots) in colon. Tissue collected from high-fat- diet-fed wild-type mice, 6 h after a single dose of oral metformin (600 mg kg−1) (right, red box, M1–M7) or vehicle gavage (left, blue box, V1–V7); n = 7 mice per group, mice as in Fig. 4.

Article

Extended Data Fig. 8 | Analysis of Gdf 15 mRNA expression (normalized to expression levels of Actb) in tissue from high-fat diet-fed Gdf 15+/+ mice. Metformin treatment (300 mg kg−1) once daily for 11 days (see Fig. 2a). Data are mean ± s.e.m., n = 6 metformin, n = 7 vehicle; P values (95% CI) by two-tailed t-test.

Extended Data Fig. 9 | Hepatic GDF15 response to biguanides. a, b,Gdf15 mRNA expression in primary mouse hepatocytes (a) or human iPS-cell-derived hepatocytes (b) treated with vehicle control (Con) or metformin for 6 h. mRNA expression is presented as fold expression relative to control treatment (set at 1), normalized to Hprt and GAPDH in mouse and human cells, respectively. Data are expressed as mean ± s.e.m. from four (a) or two (b) independent experiments. P values (95% CI) by one-way ANOVA with Tukey’s correction for multiple comparisons. c, d, Circulating levels of GDF15 (c) and hepatic Gdf15 mRNA expression (d) (normalized to β2-microglobulin) in chow-fed, wild-type mice 4 h after a single oral dose of phenformin (300 mg kg−1). Data are mean ± s.e.m., n = 6 per group; P values (95% CI) by two tailed t-test. e, Representative image of in situ hybridization for Gdf15 mRNA expression (red spots) of fixed liver tissue derived from animals treated as described in c and d.

Article

Extended Data Fig. 10 | Role of the ISR in biguanide-induced Gdf 15 expression. a, b, mRNA levels in kidney (a) and colon (b) isolated from obese mice 24 h after a single oral dose of metformin (600 mg kg−1). Data are mean ± s.e.m. (n = 5 per group, except forn = 4 for colon metformin Slc22a1). P values (95% CI) by two-tailed t-test. Gdf15 mRNA fold induction 24 h after metformin (600 mg kg−1) is positively correlated with Chop mRNA induction in both kidney (a, right) and colon (b, right). Black line shows linear regression analysis. c–g, Immunoblot analysis of ISR components (c) and Gdf15 mRNA expression (d) in wild-type mouse embryonic fibroblasts (MEF) treated with vehicle control (Con), metformin (Met, 2 mM) or phenformin (Phen, 5 mM) or tunicamycin (Tn, 5 μg ml−1, used as a positive control) for 6 h. e–g, Gdf15 mRNA expression in ATF4 knockout (KO) MEFs (e), in control siRNA and CHOP siRNA transfected wild-type MEFs treated with Tn or Phen for 6 h (f ), or in wild-type MEFs pre-treated for 1 h with either the PERK inhibitor GSK2606414 (GSK, 200 nM) or eIF2α inhibitor ISRIB (ISR, 100 nM) then co-treated with

phenformin for a further 6 h (g). mRNA expression is presented as fold expression relative to its respective control treatment (set at 1) or phenformin- treated samples (set as 100) with normalization to Hprt gene expression. Data are mean ± s.e.m. from two (c, d) or at least three (e–g) independent experiments. P values (95% CI) by two tailed t-test relative to phenformin- treated control wild-type and control siRNA-treated samples. h, GDF15 protein in supernatant of mouse derived 2D duodenal organoids treated with metformin in the absence or presence of ISRIB (1 μM). Data are expressed as mean ± s.e.m. from two independent experiments. At least duplicate protein measurements for each sample. P values by two-way ANOVA with Sidak’s correction for multiple comparisons. i, GDF15 protein in supernatants of mouse-derived 2D duodenal organoids from wild-type and Chop-null mice treated with metformin from two independent experiments. At least duplicate protein measurements for each sample. Data are mean ± s.e.m.; P values (95% CI) by two-tailed t-test.

n ature research

| rep ortin

g sum

m ary

O ctober 2018

Corresponding author(s): Anthony P.Coll , Stephen O'Rahilly

Last updated by author(s): Dec 11, 2019

Reporting Summary Nature Research wishes to improve the reproducibility of the work that we publish. This form provides structure for consistency and transparency in reporting. For further information on Nature Research policies, see Authors & Referees and the Editorial Policy Checklist.

Statistics For all statistical analyses, confirm that the following items are present in the figure legend, table legend, main text, or Methods section.

n/a Confirmed

The exact sample size (n) for each experimental group/condition, given as a discrete number and unit of measurement

A statement on whether measurements were taken from distinct samples or whether the same sample was measured repeatedly

The statistical test(s) used AND whether they are one- or two-sided Only common tests should be described solely by name; describe more complex techniques in the Methods section.

A description of all covariates tested

A description of any assumptions or corrections, such as tests of normality and adjustment for multiple comparisons

A full description of the statistical parameters including central tendency (e.g. means) or other basic estimates (e.g. regression coefficient) AND variation (e.g. standard deviation) or associated estimates of uncertainty (e.g. confidence intervals)

For null hypothesis testing, the test statistic (e.g. F, t, r) with confidence intervals, effect sizes, degrees of freedom and P value noted Give P values as exact values whenever suitable.

For Bayesian analysis, information on the choice of priors and Markov chain Monte Carlo settings

For hierarchical and complex designs, identification of the appropriate level for tests and full reporting of outcomes

Estimates of effect sizes (e.g. Cohen's d, Pearson's r), indicating how they were calculated

Our web collection on statistics for biologists contains articles on many of the points above.

Software and code Policy information about availability of computer code

Data collection No software was used.

Data analysis Prism v8 for MacOSX , Prism V 8 for WIndows and Microsoft Excel 13were used for Statistical analysis. CAMERA data analysed using STATA version 15.1. ANCOVA and analysis of data from glucose homeostais experiments was performed using SPSS 25 (IBM).

For manuscripts utilizing custom algorithms or software that are central to the research but not yet described in published literature, software must be made available to editors/reviewers. We strongly encourage code deposition in a community repository (e.g. GitHub). See the Nature Research guidelines for submitting code & software for further information.

Data Policy information about availability of data

All manuscripts must include a data availability statement. This statement should provide the following information, where applicable: - Accession codes, unique identifiers, or web links for publicly available datasets - A list of figures that have associated raw data - A description of any restrictions on data availability

The data that support the findings of this study are available from the corresponding authors upon request.The CAMERA trial dataset is held at the University of Glasgow and is available on request from the investigators subject to a signed agreement operating within the confines of the original ethics application.

n ature research

| rep ortin

g sum

m ary

O ctober 2018

Field-specific reporting Please select the one below that is the best fit for your research. If you are not sure, read the appropriate sections before making your selection.

Life sciences Behavioural & social sciences Ecological, evolutionary & environmental sciences

For a reference copy of the document with all sections, see nature.com/documents/nr-reporting-summary-flat.pdf

Life sciences study design All studies must disclose on these points even when the disclosure is negative.

Sample size Sample sizes were determined on the basis of homogeneity and consistency of characteristics in the selected models and were sufficient to detect statistically significant differences in body weight, food intake and serum parameters between groups, while also ensuring no more animals than necessary were used. These assessments are based on extensive expertise with these models and endpoints. In vitro sample sizes were based on previous extensive experience with reagents and systems used.

Data exclusions One data point of 25 food intake points collected on day 11 of mouse study 3 was lost due to technical error (see methods). One animal in mouse study 6 which was otherwise well and healthy did not progress to ITT after metformin due to technical and acute behavioural issues at time of study. One blood sample from mouse study 7 was lost due to technical error.

Replication Acute dosing of metformin to chow fed animals has been replicated and reproduced across two laboratories. Phenfomin data has been replicated. Mouse study 1 has not been replicated exactly but the acute response to metformin in HFD fed animals has been reproduced across two laboratories. Longer term dosing studies of metformin to HFD Gfral and Gdf15 null mice (mouse study 3 and 4) have been replicated; two independent cohorts of Gdf15 null mice appear in the manuscript. ITT post metformin dosing of Gdf15 null has been done once. Mouse study 5 has not been replicated exactly but similar data were generated using anti-GFRAL antibody given over a different time scale but with similar metformin exposure. Mouse studies7 and 8 have been done once. All in vitro and cell based experiments have been replicated successfully and reliably reproduced with replicate numbers reported in legends.

Randomization Animals were randomised into treatment groups based on body weight such that the mean body weights of each group were as matched as possible but without using excess numbers of animals.

Blinding Serum/plasma GDF15 measurements were blinded by the investigators. Investigator undertaking ISH analysis of tissue was blinded to treatment during tissue processing and labelling.

Reporting for specific materials, systems and methods We require information from authors about some types of materials, experimental systems and methods used in many studies. Here, indicate whether each material, system or method listed is relevant to your study. If you are not sure if a list item applies to your research, read the appropriate section before selecting a response.

Materials & experimental systems n/a Involved in the study

Antibodies

Eukaryotic cell lines

Palaeontology

Animals and other organisms

Human research participants

Clinical data

Methods n/a Involved in the study

ChIP-seq

Flow cytometry

MRI-based neuroimaging

Antibodies Antibodies used ATF4 antibody was obtained from David Ron. CHOP was obtained from Santa Cruz (Cat# sc-7351; RRID: AB_627411). pEIF2a

Ser51 was obtained from Abcam (Cat# ab32157; RRID: AB_732117). Calnexin was obtained from Abcam (Cat# ab75801; RRID: AB_1310022). Anti-GFRAL functional blocking antibody generated by NGM.Secondary antibodies used were horseradish peroxidase (HRP)-conjugated anti-rabbit immunoglobulin G (IgG), HRP-conjugated anti-mouse IgG (Cell Signalling Technologies)

Validation ATF4 and CHOP antibody has been previously validated in KO and siRNA knockdown samples (this study and PMID: 30639358). pEIF2a antibody has been validated previously (PMID: 27297692) and we have independent confirmed this during the course of our studies for PMID: 30639358 (data not shown). Calnexin is a well established commercially available antibody that has been used by numerous investigators and published frequently. Characterization of anti-GFRAL functional blocking antibodies was carried out using a cell-based RET/GFRAL luciferase gene reporter assays, in vitro binding studies (ELISA and Biacore) and in vivo studies as described in patent number; US10174119B2, https://patents.google.com/patent/US10174119B2/en.

n ature research

| rep ortin

g sum

m ary

O ctober 2018

Eukaryotic cell lines Policy information about cell lines

Cell line source(s) Mouse Embryonic Fibroblasts (MEFs) was obtained David Ron (CIMR). The hiPSC line A1ATDR/R was obtained from Ludovic Vallier ( Cambridge Stem Cell Institute).

Authentication MEFs have been previously validated (PMID 12667446 and 12667446) and the hIPS cell line validated (PMID 20739751 and 21993621)

Mycoplasma contamination MEFs and A1ATDR/R cells were tested negative for Mycoplasma contamination

Commonly misidentified lines (See ICLAC register)

no commonly misidentified lines were used.

Animals and other organisms Policy information about studies involving animals; ARRIVE guidelines recommended for reporting animal research

Laboratory animals Mice studied ranged between 2 to 6 months of age. At NGM, animals were kept under controlled light (12hour light and 12hour dark cycle, dark 6:30 pm - 6:30 am), temperature (22 ± 3°C) and humidity (50% ± 20%) conditions. In Cambridge, were maintained in a 12-hour light/12-hour dark cycle (lights on 0700–1900), temperature-controlled (22°C) facility, with ad libitum access to food (RM3(E) Expanded chow, Special Diets Services, UK) and water. Any mice bought from an outside supplier were acclimatised in a holding room for at least one week prior to study. Study diets as outlined in relevant section of "Methods". Age , sex and body weight of mice used as detailed in relevant "Methods" section. Unless otherwise stated, male mice were used a danimals housed singly during studies with environmental enrichment within cages. All drug administration and testing performed during the light cycle. Wild type mice used in Cambridge (C57BL6/J mice ) from Charles River, Margate, UK. C57BL/6N-Gdf15tm1a(KOMP)Wtsi/H mice were obtained from the MRC Harwell Institute, UK. Gfral-/- mice were purchased from Taconic (#TF3754) on a mixed 129/SvEv-C57BL/6 background and backcrossed for 10 generations to >99% C57BL/6 background at NGM’s animal facility.

Wild animals The study did not involve wild animals

Field-collected samples The study did not involve samples collected from the field.

Ethics oversight At NGM, all experiments were conducted with NGM IACUC approved protocols and all relevant ethical regulations were complied with throughout the course of the studies, including efforts to reduce the number of animals used. In Cambridge, all mouse studies were performed in accordance with UK Home Office Legislation regulated under the Animals (Scientific Procedures) Act 1986 Amendment, Regulations 2012, following ethical review by the University of Cambridge Animal Welfare and Ethical Review Body (AWERB).

Note that full information on the approval of the study protocol must also be provided in the manuscript.

Human research participants Policy information about studies involving human research participants

Population characteristics CAMERA was a randomized, double-blinded, placebo-controlled trial designed to investigate the effect of metformin on surrogate markers of cardiovascular disease in patients without diabetes, aged 35 to 75, with established coronary heart disease and a large waist circumference (≥ 94cm in men, ≥80 cm in women) . Mean age (yrs)+/-SD; metformin 63(8), placebo 64(8); male sex , metformin 79(81%), placebo 63( 72%).Baseline characteristics did not differ substantially between treatment groups. Nine participants completed the study by Konopka and colleagues; 7 had a family history of T2DM and 8 were metformin naïve. One participant had previously used metformin but discontinued more than 2 years before the study commenced.

Recruitment 3000 potential participants were identified from electronic searches of Glasgow general practice databases, supplemented by patients from hospital cardiology clinics. Of those invited, 805 replied and 356 were screened. 173 were enrolled and randomly assigned (86 to metformin, 87 to placebo).Key eligibility criteria included the use of statin therapy, history of coronary heart disease, large waist circumference (≥94cm in men, ≥80cm in women), and no history or biochemical evidence of type 2 diabetes. Eligible participants were randomly assigned to metformin or placebo (1:1) with a randomisation sequence generated independently by computer with permuted blocks of four without stratification. In the study by Konopka and colleagues, inclusion criteria were: obesity (body mass index >30 kg/m2), sedentary (<1 hour of structured activity per week), nonsmoking, and not taking any medication to control blood glucose.

Ethics oversight All participants in CAMERA study provided written informed consent and were followed up for 18 months. This study was approved by the Medicines and Healthcare Products Regulatory Agency and West Glasgow Research Ethics Committee, and done in accordance with the principles of the Declaration of Helsinki and good clinical practice guidelines. The study by Konopka and colleagues was approved by the Mayo Clinic Institutional Review Board and all participants provided written, informed consent .

Note that full information on the approval of the study protocol must also be provided in the manuscript.

n ature research

| rep ortin

g sum

m ary

O ctober 2018

Clinical data Policy information about clinical studies All manuscripts should comply with the ICMJE guidelines for publication of clinical research and a completed CONSORT checklist must be included with all submissions.

Clinical trial registration The CAMERA trial is registered with ClinicalTrials.gov, number NCT00723307. The study by Konopka and colleagues is registered, number NCT01956929.

Study protocol Trial protocol outlined in PMID: 24622715 and supplied on submission.

Data collection Participants were randomized 1:1 to 850mg metformin or matched placebo twice daily with meals. Follow up study visits were conducted between 2009 and 2012 at the Glasgow Clinical Research Centre. Participants attended six monthly visits after overnight fasts and before taking their morning dose of metformin. Blood samples collected during the trial were centrifuged at 4 degrees Celsius soon after sampling, separated and stored at -80°C. Bodyweight, body fat (by bio-impedance with a Tanita BIA body fat analyser [Tanita Corporation, Tokyo, Japan]), waist circumference (measured midway between lowest rib and iliac crest), and hip circumference (measured around widest part of the buttocks) were measured at each visit. In the study by Konopka and colleagues, placebo or metformin (week 1, 500mg twice daily; week, 2 1000mg twice daily) were administered following a six week period of washout. Samples were collected in the morning after overnight fasting.

Outcomes The primary endpoint was progression of mean distal carotid intima-media thickness (cIMT) over 18 months in the modified intention-to-treat population. Further descriptions and results for the pre-specified outcomes for the CAMERA trial are provided in a previous publication PMID: 24622715. As outlined in a previous publication (PMID: 27160898) the study by Konopka and colleagues aimed toi nvestigated whether metformin inhibited glucagon-stimulated endogenous glucose production (EGP) in humans. The study measured EGP using stable isotope methodology under basal, glucagon-deficient, and glucagon-stimulated conditions.

GDF15 mediates the effects of metformin on body weight and energy balance

Human studies
Mouse studies
GDF15 and glucose homeostasis
Source of GDF15 production
Online content
Fig. 1 Effect of Metformin on circulating GDF15 levels in humans and mice.
Fig. 2 GDF15–GFRAL signalling is required for the weight-loss effects of metformin on a high-fat diet.
Fig. 3 Effects of metformin on glucose homeostasis.
Fig. 4 Metformin increases GDF15 expression in the enterocytes of distal intestine and in renal tubular epithelial cells.
Extended Data Fig. 1 Expanded CAMERA dataset.
Extended Data Fig. 2 Effect of single oral dose of metformin in chow-fed mice.
Extended Data Fig. 3 Body weight changes with metformin treatment in mice with disrupted GDF15–GFRAL signalling.
Extended Data Fig. 4 Response of high-fat diet-fed Gdf15−/− and Gfral−/− mice to metformin.
Extended Data Fig. 5 Response of second, independent cohort of high-fat diet fed Gdf15+/+ and Gdf15−/− mice to metformin.
Extended Data Fig. 6 Glucose, insulin and GDF15 response to metformin.
Extended Data Fig. 7 In situ hybrididation for Gdf15 mRNA expression in gut, liver and kidney.
Extended Data Fig. 8 Analysis of Gdf15 mRNA expression (normalized to expression levels of Actb) in tissue from high-fat diet-fed Gdf15+/+ mice.
Extended Data Fig. 9 Hepatic GDF15 response to biguanides.
Extended Data Fig. 10 Role of the ISR in biguanide-induced Gdf15 expression.