Effect of Exercise or Metformin on Biomarkers of Inflammation in Breast and Colorectal Cancer: A Randomized Trial

Observational studies report that physical activity and metformin after the diagnosis of early stage cancer are associated with a 30% to 40% reduction in the risk of cancer recurrence and mortality (1, 2). The biological processes through which physical activity and metformin may favorably affect clinical outcome remain poorly understood. Inflammation is hypothesized as a key biological mediator of these associations (3, 4).

Inflammation is a hallmark of cancer and is associated with poor clinical outcome in patients with various types of solid tumors (5–7). Inflammation activates the JAK-STAT and NF-κB signaling pathways to promote cell survival, proliferation, migration, and invasion (8–10). Preclinical studies demonstrate that reducing inflammation and targeting inflammatory signaling pathways slow cell growth and delay tumor progression (11, 12). Furthermore, obesity causes chronic inflammation that may promote malignant cell growth (13, 14). An anti-inflammatory benefit of physical activity and metformin may occur, in part, because of reductions in adiposity (15, 16).

These observations provided the scientific rationale to test the effect of exercise and metformin on prespecified inflammation outcome measures in patients with breast and colorectal cancer. We previously reported that exercise and metformin reduced the primary endpoint of fasting plasma insulin, and secondary supportive endpoints of insulin resistance, and adiposity (17). This trial used a 2 × 2 factorial design, which allowed the simultaneous examination of exercise and metformin. This trial was part of the NCI Transdisciplinary Research on Energetics and Cancer (TREC) consortium (18).

The study was a 12-week, multicenter, randomized, 2 × 2 factorial, phase II trial. The study was conducted at three centers in the United States (Dana-Farber Cancer Institute, Duke University, and Yale University). The study was conducted in accordance with Good Clinical Practice and the ethical principles originating in the Declaration of Helsinki. The protocol and informed consent document were approved by the Institutional Review Board for each site. All participants provided informed consent and approval from their physician prior to completing any study activities. The study was registered on Clinicaltrials.gov as NCT01340300.

Eligible participants had stage I–III breast or colorectal cancer; completed surgery, chemotherapy, and radiotherapy ≥1 month(s) prior to enrollment (concurrent endocrine and/or trastuzumab were allowed for participants with breast cancer); were engaging in <120 min·wk⁻¹ of exercise; had an Eastern Cooperative Oncology Group Performance Status of 0–1; had a random glucose <160 mg·dL⁻¹ or fasting glucose <126 mg·dL⁻¹; had adequate renal and kidney function; were age ≥18 years; were English speaking; and were willing to be randomized.

Participants were randomly assigned in an equal ratio to one of four treatment groups for 12 weeks: exercise alone, metformin alone, exercise and metformin, or control (Fig. 1). Participants were stratified by body mass index (<30 kg·m⁻² vs. ≥30 kg·m⁻²), sex (men vs. women), and cancer site (breast vs. colorectal) and then randomized using a permuted block design with fixed block sizes. Participants were not blinded to treatment assignment.

Exercise was performed through a combination of in-person and home-based activity. In-person exercise was supervised by an exercise physiologist. Aerobic exercise was the primary exercise type, with treadmill and outdoor walking as the most common exercise modalities. Exercise intensity was prescribed at 65% to 80% of the age-predicted heart rate (19). During the twice-weekly in-person exercise sessions, participants wore a heart rate monitor to learn the amount of physical exertion consistent with moderate- to vigorous-intensity exercise. Home-based exercise was monitored by self-report using exercise logs that were provided to participants. Participants progressed to the goal of 220 min·wk⁻¹ of exercise. This exercise dose was selected on the basis of observational studies suggesting that higher volumes of activity are associated with a lower risk of recurrence and premature mortality (20, 21). Participants were encouraged to individualize their frequency (days per week), fractionation (sessions per day), and duration (minutes per session) of exercise according to a schedule that promoted optimal adherence to the prescribed exercise volume. The exercise physiologist provided behavioral support and monitored exercise adherence during the study.

Metformin was titrated over the first 2 weeks of the study. In week one and two, participants were instructed to consume one metformin capsule at dinner (850 mg). If no gastrointestinal distress or other adverse events were experienced after 2 weeks at 850 mg, participants were instructed to consume one metformin capsule at breakfast and one metformin capsule at dinner, totaling 1,700 mg per day, until the end of the study. Participants who experienced adverse events at 1,700 mg were allowed to continue at 850 mg for the rest of the study. Dosing of metformin for the treatment of prediabetes and diabetes ranges from 500 to 2,500 mg daily, with many individuals requiring ≥1,500 mg daily to achieve adequate glycemic control (22, 23).

Study participants underwent a fasting (≥10 hours) blood draw at baseline and week 12. EDTA-preserved plasma was stored at −80°C. Inflammation measures included high-sensitivity C-reactive protein (hs-CRP), soluble tumor necrosis factor alpha receptor 2 (sTNFαR2), and IL6. hs-CRP, sTNFαR2, and IL6 were selected because of their reported associations with cancer recurrence and mortality in observational studies of patients with breast and colorectal cancer (5–7). hs-CRP was measured as a marker of generalized systemic inflammation (24). sTNFαR2 was measured as an activator of the NF-kB pathway (25); sTNFαR2 is a surrogate marker for TNFα that is more stable in plasma and less sensitive to diurnal variation (26). IL6 was measured as an activator of the JAK-STAT pathway (27). hs-CRP was quantified using an immunoturbidimetric assay (Roche Diagnostics). sTNFα-R2 and IL6 and were quantified using ultrasensitive sandwich enzyme immunoassays (R&D Systems). Baseline and follow-up plasma samples were assayed simultaneously and in duplicate at the end of the study. Blinded quality-control samples were interspersed among cases. Coefficients of variation for all samples were ≤8%. All assays were conducted by staff who were blinded to treatment assignment.

Demographic characteristics including age, sex, and race were self-reported. Clinical information including type of cancer, time since cancer diagnosis, and cancer stage was abstracted from physician records. Body mass and circumferences of the waist and hip were measured in duplicate using standardized techniques.

The sample size was selected to provide sufficient statistical power to detect change in the primary endpoint of fasting plasma insulin (17). Measures of inflammation were prespecified as secondary outcomes. Based on estimates from the Diabetes Prevention Program (DPP) and Action for Health in Diabetes (Look AHEAD) trials (28, 29), this study had sufficient statistical power to detect a standardized mean difference effect size of ≥0.48 for inflammation outcome measures.

All analyses adhered to the intention-to-treat principle. At the time this study was designed, the extent to which exercise and metformin acted independently (e.g., exercise is equally effective whether or not the participant is receiving metformin, and vice versa) was uncertain (30). Therefore, the primary inferential analysis estimated the comparative effect of each of the three intervention groups (e.g., exercise alone, metformin alone, and exercise plus metformin) with the control group (31); conceptually this contrast is a comparison of the cells within a 2 × 2 table (32). The primary modeling strategy evaluated the trial product estimand that was quantified using a generalized linear mixed model with observed data (i.e., no imputation; ref. 33). This model accounts for the correlation between measures and assumes data are missing at random. The secondary modeling strategy evaluated the treatment policy estimand that was quantified using a generalized linear mixed model with predictive mean matching multiple imputation to account for missing data (33, 34). Biomarker concentrations were log transformed in the inferential analysis to improve distributional normality. The baseline value of the dependent variable, randomization stratification factors, and study center were included as covariates in regression models (35). Group-by-time interaction terms were included as fixed-effects in regression models with subject-specific intercepts. A linear contrast of the four individual group means was estimated to determine if the effects of exercise and metformin were more than additive (e.g., multiplicative; ref. 36). In the absence of evidence to suggest a multiplicative interaction, we proceeded to estimate the comparative effects of exercise versus no exercise and metformin versus no metformin, as these main effects represent the most efficient analysis of a 2 × 2 factorial design (37). In a 2 × 2 factorial design, the main effect of one independent variable (e.g., exercise vs. no exercise) represents the overall effect averaged across both values of the other independent variable (e.g., metformin vs. no metformin); conceptually, this contrast is a comparison of the margins of a 2 × 2 table (32).

Treatment effects were calculated as the treatment effect ratio, which quantifies the percent change in geometric means from baseline to 12 weeks (e.g., a treatment effect ratio of 0.75 indicates a 25% reduction), with 95% confidence intervals (CI). Model fit was assessed using a combination of numeric and graphical techniques. Interaction terms of group, time, and randomization stratification factors were included in regression models to quantify heterogeneity of treatment effect. Exploratory analyses quantified the extent to which change in body mass and circumferences of the waist and hip mediated the observed treatment effect (38).

Between September 2011 and December 2015, 139 participants were recruited and randomized with primary data collection ending in May 2016. Baseline characteristics of study participants were balanced (Table 1).

At baseline, the geometric mean (SD) hs-CRP was 0.55 (1.04) mg·L⁻¹, sTNFαR2 was 7.80 (0.33) pg·mL⁻¹, and IL6 was 1.04 (0.82) pg·mL⁻¹, indicating low to moderate inflammation. Among participants randomized to exercise, 77% and 17% completed ≥50% and ≥90% of their initially prescribed exercise volume, respectively. Among participants randomized to metformin, 67% and 31% consumed ≥50% and ≥90% of their initially prescribed metformin dose, respectively. At 12 weeks, 91 (65%) participants completed their assigned intervention; reasons for premature discontinuation have been described (17). Participants who did not complete the study were more likely to be of white race [multivariable-adjusted OR: 3.59 (95% CI, 1.14–11.36)]; no other measured factors, including randomized group assignment and baseline concentrations of inflammation, were associated with study completion.

By pairwise effects analysis (e.g., contrasting the cells within the 2 × 2 table), compared with control, exercise alone statistically significantly reduced hs-CRP: −30.2% (95% CI, −50.3, −1.0) and IL6: −30.9% (95% CI, −47.3, −9.5); but did not statistically significantly change sTNFαR2: 1.0% (95% CI, −10.4, 13.9; Table 2). Compared with control, metformin alone did not statistically significantly change hs-CRP: −13.9% (95% CI, −40.0, 23.4), sTNFαR2: −10.4% (95% CI, −21.3, 2.0), or IL6: −22.9% (95% CI, −42.3, 2.0). Compared with control, exercise and metformin statistically significantly reduced sTNFαR2: −13.1% (95% CI, −22.9, −1.0) and IL6: −38.7% (95% CI, −52.3, −18.9); but did not statistically significantly change hs-CRP: −20.5% (95% CI, −44.0, 12.7). The combination of exercise and metformin was not synergistic for hs-CRP (P = 0.35), sTNFαR2 (P = 0.66), or IL6 (P = 0.69). Intervention adherence was not associated with magnitude of treatment effect; participants who adhered even minimally to either intervention achieved an inflammation lowering benefit. The correlations with exercise adherence with change in inflammation were: hs-CRP (R = −0.03; 95% CI, −0.26, 0.21); sTNFαR2 (R = −0.12; 95% CI, −0.34, 0.12); and IL6 (R = 0.01; 95% CI, −0.24, 0.22). The correlations with metformin adherence with change in inflammation were: hs-CRP (R = 0.06; 95% CI, −0.18, 0.29); sTNFαR2 (R = 0.04; 95% CI, −0.20, 0.27), and IL6 (R = 0.04; 95% CI, −0.20, 0.27). Heterogeneity of the treatment effect did not substantively differ between any randomization stratification subgroups. Results were similar using predictive mean matching multiple imputation (Supplementary Table S1).

By main effects analysis (e.g., contrasting the margins of the 2 × 2 table), compared with no exercise, exercise statistically significantly reduced IL6: −23.7% (95% CI, −36.9, −8.6; Table 3); but did not statistically significantly change hs-CRP: −19.0% (95% CI, −36.6, 2.0) and sTNFαR2: 0.0% (95% CI, −7.7, 9.4). Compared with no metformin, metformin statistically significantly reduced sTNFαR2: −12.2% (95% CI, −18.9, −3.9); but did not statistically significantly change hs-CRP: 3.0% (95% CI, −18.1, 30.9) and IL6: −13.9% (95% CI, −28.1, 4.0). Intervention adherence was not associated with magnitude of treatment effect. Heterogeneity of the treatment effect did not substantively differ between any randomization stratification subgroups. Results were similar using predictive mean matching multiple imputation (Supplementary Table S2).

Change in body mass, waist circumference, or the waist-to-hip ratio did not mediate the observed treatment effect of exercise on IL6 or the treatment effect of metformin on sTNFαR2 (Table 4). No serious or unexpected adverse events were reported; nonserious adverse events have been described (17).

In this randomized 2 × 2 factorial trial of 139 survivors of breast and colorectal cancer, exercise reduced concentrations of IL6 and metformin reduced concentrations of sTNFαR2 over 12 weeks. The combined effect of exercise and metformin was not multiplicative, although statistical power was limited. The observed treatment effect was consistent across randomization stratification variables including baseline body mass index, sex, and cancer type. Change in body mass, waist circumference, or the waist-to-hip ratio did not mediate the observed treatment effect of exercise and metformin on inflammation outcome measures. In pairwise effects analyses comparing each intervention group with the control group, exercise reduced hs-CRP and IL6, and the combination of exercise and metformin reduced sTNFαR2 and IL6.

One of the mechanisms by which physical activity and metformin are hypothesized to exert anticancer effects is through their impact on the host microenvironment by reducing inflammation (3, 4). Our results provide evidence that inflammation is reduced when 12 weeks of exercise or metformin are administered to patients with breast and colorectal cancer. In pairwise analysis, the combination of exercise and metformin reduced both IL6 and sTNFαR2 compared with control. In main effects analysis, exercise reduced IL6 and metformin reduced sTNFαR2. IL6 activates the JAK/STAT pathway and TNFα in part through its receptor, sTNFαR2, activates the NF-kB pathway (11, 12). Our results suggest that exercise and metformin inhibit distinct inflammatory processes, and the combination of exercise and metformin more comprehensively inhibit the physiology of distinct inflammation-related signaling pathways than each intervention alone.

Obesity is associated with poor clinical outcome after cancer diagnosis (39). One mechanism through which obesity is hypothesized to exert procancer effects is increased inflammation caused by hypertrophic metabolically active adipocytes (13, 14). Exercise reduces adipose tissue and increases lean mass, despite stability of body weight (40). In patients with type 2 diabetes, metformin causes modest weight loss, preferentially through reductions in adipose tissue mass (41). We previously reported that exercise and metformin reduced body mass, waist circumference, and the waist-to-hip ratio (17). In our exploratory analysis, we observed no evidence that the treatment effect of exercise or metformin on inflammation outcome measures was mediated by change in body mass or anthropometric surrogate measures of body composition (e.g., waist circumference or the waist-to-hip ratio).

The results of this trial complement the Reach for Health Trial, also conducted as part of the NCI TREC consortium (42). Reach for Health used a similar 2 × 2 factorial trial design to evaluate the effect of metformin or behavioral weight loss in overweight and obese patients with breast cancer. By main effects analysis, over 24 weeks, no change in hs-CRP was observed with metformin: −14.9% (95% CI, −32.9, 3.1) or behavioral weight loss: −12.4% (95% CI, −30.4, 5.5). Our study found no main effect of exercise or metformin on hs-CRP; in pairwise analysis, exercise reduced hs-CRP by 30% relative to control. This observation is consistent with a meta-analysis of six randomized controlled trials demonstrating that exercise reduces hs-CRP in patients with cancer (43). However, the absence of an effect of metformin on hs-CRP is in contrast to an analysis of 492 patients with breast cancer enrolled in the MA.32 trial, where metformin reduced hs-CRP by 6.7% versus placebo (44).

There are several limitations to this trial. The main limitation is the small sample size, which limited our ability to identify multiplicative interaction effects between exercise and metformin on inflammation outcome measures. The small sample size may have also limited our ability to detect small, but potentially clinically meaningful, main effects for exercise or metformin. The intervention duration was 12 weeks, which limits our ability to understand the benefits of exercise and metformin over longer time horizons. The study sample was not enrolled on the basis of having elevated biomarkers of inflammation at baseline, which limits our understanding of the treatment effect in patients with acute or chronic inflammation. Intervention adherence was modest, however adherence to exercise or metformin was not correlated with the magnitude of treatment effect. Follow-up at 12 weeks was modest, however results and conclusions of our primary analysis were robust to various missing data and statistical modeling assumptions.

There are several strengths to this trial. The randomized design and use of two distinct interventions that are both hypothesized to favorably affect inflammation outcome measures allowed for a time- and cost-efficient comparison of causal effects. Our study included patients with breast and colorectal cancer, which allowed examination of heterogeneity of the treatment effect between cancer sites. The use of three biomarker measures of inflammation allowed for a detailed physiologic investigation of treatment benefit.

In one of the first randomized clinical trials evaluating two different metabolic interventions in patients with cancer, this study demonstrates that exercise and metformin reduced inflammation. The findings from this randomized trial are useful to begin to understand the biological mediators of the relationship between physical activity and metformin with clinical outcome in patients with cancer. Results from ongoing phase III randomized clinical trials with disease endpoints will inform the utilization of exercise and metformin in clinical practice, and the correlative studies embedded into these trials will offer unprecedented insight into mechanisms of treatment benefit (44, 45).

J.C. Brown reports grants from NIH, grants from American Institute for Cancer Research, and grants from Susan G. Komen Foundation during the conduct of the study. L.W. Jones reports other from Pacylex, Inc. (Stock ownership) outside the submitted work. B. Cartmel reports grants from NCI during the conduct of the study. S.M. Tolaney reports grants and personal fees from AstraZeneca (Research funding to institution; honorarium for advisory board/consulting), grants and personal fees from Eli Lilly (Research funding to institution; honorarium for advisory board/consulting), grants and personal fees from Merck (Research funding to institution; honorarium for advisory board/consulting), grants and personal fees from Novartis (Research funding to institution; honorarium for advisory board/consulting), grants and personal fees from Nektar (Research funding to institution; honorarium for advisory board/consulting), grants and personal fees from Pfizer (Research funding to institution; honorarium for advisory board/consulting), grants and personal fees from Genentech/Roche (Research funding to institution; honorarium for advisory board/consulting), grants and personal fees from Immunomedics (Research funding to institution; honorarium for advisory board/consulting), grants and personal fees from Bristol-Myers Squibb (Research funding to institution; honorarium for advisory board/consulting), grants and personal fees from Eisai (Research funding to institution; honorarium for advisory board/consulting), grants and personal fees from Sanofi (Research funding to institution; honorarium for advisory board/consulting), grants and personal fees from Odonate (Research funding to institution; honorarium for advisory board/consulting), grants and personal fees from Seattle Genetics (Research funding to institution; honorarium for advisory board/consulting), grants from Exelixis (Research funding to institution), grants from Cyclacel (Research funding to institution), personal fees from Puma (honorarium for advisory board/consulting), personal fees from Celldex (honorarium for advisory board/consulting), personal fees from Silverback Therapeutics (honorarium for advisory board/consulting), personal fees from G1 Therapeutics (honorarium for advisory board/consulting), personal fees from Abbvie (honorarium for advisory board/consulting), personal fees from Athenex (honorarium for advisory board/consulting), personal fees from OncoPep (honorarium for advisory board/consulting), personal fees from Daiichi-Sankyo (honorarium for advisory board/consulting), personal fees from Kyowa Kirin Pharmaceuticals (honorarium for advisory board/consulting), personal fees from Samsung Bioepsis Inc. (honorarium for advisory board/consulting), and personal fees from CytomX (honorarium for advisory board/consulting) outside the submitted work. E.P. Winer reports personal fees from Carrick Therapeutics (consultant), personal fees from G1 Therapeutics (consultant), grants and personal fees from Genentech/Roche (Research to Institute and Consultant), personal fees from Genomic Health (consultant), personal fees from GSK (consultant), personal fees from Jounce (consultant), personal fees from Leap (consultant), personal fees from Lilly (consultant), personal fees from Novartis (unpaid), personal fees from Seattle Genentics (consultant), and personal fees from Syros (consultant) outside the submitted work. K. Ng reports grants from National Cancer Institute, grants from Department of Defense, and grants from Cancer Research UK during the conduct of the study; grants and non-financial support from Pharmavite, grants from Revolution Medicines, non-financial support from Evergrande Group, grants from Genentech, grants from Gilead Sciences, grants from Tarrex Biopharma, personal fees from Bayer, personal fees from Seattle Genetics, personal fees from Array Biopharma, and personal fees from X-Biotix Therapeutics outside the submitted work. C.S. Fuchs reports personal fees from Agios, personal fees from Bain Capital, personal fees from Amylin Pharma, personal fees from CytomX Therapeutics, personal fees from Daiichi-Sankyo, personal fees from Eli Lilly, personal fees from Entrinsic Health, personal fees from Evolveimmune Therapeutics, personal fees from Genentech, personal fees from Merck, personal fees from Taiho, and personal fees from Unum Therapeutics outside the submitted work; and He also serves as a Director for CytomX Therapeutics and owns unexercised stock options for CytomX and Entrinsic Health. He is a co-Founder of Evolveimmune Therapeutics and has equity in this private company. He had provided expert testimony for Amylin Pharmaceuticals and Eli Lilly. J.A. Meyerhardt reports personal fees from COTA Healthcare, personal fees from Taiho (NCCN grants), and personal fees from Ignyta (Single advisory board) outside the submitted work. No potential conflicts of interest were disclosed by the other authors.

One of the Editors-in-Chief of Cancer Prevention Research is an author on this article. In keeping with AACR editorial policy, a senior member of the Cancer Prevention Research editorial team managed the consideration process for this submission and independently rendered the final decision concerning acceptability.

The content is solely the responsibility of the authors and does not necessarily represent the official views of the NIH. The funding agency had no role in the design and conduct of the study; collection, management, analysis and interpretation of the data; preparation, review or approval of the article; and decision to submit the article for publication.

J.C. Brown: Conceptualization, resources, data curation, formal analysis, supervision, funding acquisition, validation, investigation, visualization, methodology, writing-original draft, project administration, writing-review and editing. S. Zhang: Data curation, software, formal analysis, visualization, methodology, writing-review and editing. J.A. Ligibel: Conceptualization, resources, data curation, supervision, funding acquisition, investigation, methodology, project administration, writing-review and editing. M.L. Irwin: Conceptualization, resources, supervision, funding acquisition, investigation, project administration, writing-review and editing. L.W. Jones: Conceptualization, resources, data curation, supervision, funding acquisition, investigation, methodology, writing-review and editing. N. Campbell: Resources, data curation, investigation, project administration, writing-review and editing. M.N. Pollak: Conceptualization, resources, data curation, formal analysis, supervision, funding acquisition, validation, investigation, methodology, project administration, writing-review and editing. A. Sorrentino: Data curation, investigation, project administration, writing-review and editing. B. Cartmel: Resources, data curation, supervision, validation, investigation, methodology, project administration, writing-review and editing. M. Harrigan: Data curation, supervision, investigation, methodology, project administration, writing-review and editing. S.M. Tolaney: Data curation, supervision, investigation, methodology, project administration, writing-review and editing. E.P. Winer: Data curation, supervision, investigation, methodology, project administration, writing-review and editing. K. Ng: Data curation, supervision, investigation, methodology, project administration, writing-review and editing. T.A. Abrams: Data curation, supervision, investigation, methodology, project administration, writing-review and editing. T. Sanft: Data curation, supervision, investigation, methodology, project administration, writing-review and editing. P.S. Douglas: Conceptualization, data curation, supervision, investigation, methodology, project administration, writing-review and editing. F.B. Hu: Conceptualization, resources, supervision, funding acquisition, investigation, methodology, project administration, writing-review and editing. C.S. Fuchs: Conceptualization, resources, supervision, funding acquisition, validation, investigation, methodology, project administration, writing-review and editing. J.A. Meyerhardt: Conceptualization, resources, data curation, software, formal analysis, supervision, funding acquisition, validation, investigation, visualization, methodology, writing-original draft, project administration, writing-review and editing.

This work was supported by the NCI of the NIH under Award Numbers R00-CA218603, R25-CA203650, and U54-CA155626 and the National Institute of General Medicine Sciences of the NIH under Award Number U54-GM104940. J.A. Meyerhardt is supported by the Douglas Gray Woodruff Chair Fund, the Guo Shu Shi Fund, Anonymous Family Fund for Innovations in Colorectal Cancer, Project P Fund, and the George Stone Family Foundation.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.