Dual Effect of Metformin on Breast Cancer Proliferation in a Randomized Presurgical Trial

There is increasing evidence that hyperinsulinemia and insulin resistance worsen breast cancer prognosis,¹ increase breast cancer risk,² and partly explain the obesity–breast cancer risk association.³ Insulin may promote tumorigenesis via a direct effect on epithelial tissues or indirectly by affecting other modulators, such as insulin-like growth factors, sex hormones, and adipokines.^4,5

Metformin is an oral biguanide that inhibits hepatic gluconeogenesis and sensitizes insulin action at peripheral level. It is the first-line treatment of type 2 diabetes⁶ and can substantially reduce diabetes incidence in obese women with glucose intolerance,⁷ with a good tolerability profile and low cost.⁸ Epidemiologic studies have shown a significant risk reduction in cancer incidence and mortality in patients with diabetes receiving metformin relative to other antidiabetic drugs, including positive results specifically in breast cancer.^9,10 Preclinical studies suggest that metformin may have different mechanisms of tumor inhibition via insulin-dependent and -independent pathways,¹¹ including activation of adenosine monophosphate kinase (AMPK) through liver kinase B1^12,13 and ataxia teleangiectasia-mutated gene kinase,¹⁴ with inhibition of cancer proliferation and apoptosis induction in breast cancer cell lines.^15,16 These findings have prompted great interest in metformin as an anticancer agent,¹⁷ including an ongoing adjuvant trial in nondiabetic women with breast cancer.¹⁸

We assessed the antiproliferative activity of metformin in a window-of-opportunity trial in nondiabetic women with breast cancer who were candidates to surgery. The primary outcome measure was the response of Ki-67, which is used to select drugs in breast cancer because of its prognostic value and its postulated role in predicting drug efficacy.^19–21 Given the different effect of metformin on diabetes incidence depending on body mass index (BMI) or glucose intolerance,⁷ and the putative association between BMI and triple-negative disease,^22,23 we also assessed whether metformin had a greater effect in women with insulin resistance, as assessed by the homeostasis model assessment (HOMA) index (fasting blood glucose [mmol/L] × insulin [mU/L]/22.5),²⁴ and in estrogen receptor (ER)–negative tumors.

From December 9, 2008, to May 3, 2011, 311 women were assessed for eligibility; 111 did not meet inclusion criteria or refused the study, and 200 were randomly assigned to either metformin or placebo for 4 weeks. The participant flow diagram is shown in Figure 1. Three patients receiving metformin and one receiving placebo were not assessed for the primary end point. Baseline patient and tumor characteristics, shown in Table 1, were evenly distributed between arms. Half of women were premenopausal, nearly 40% were overweight or obese, and 27% were insulin resistant. Half of tumors were greater than 25 mm, and more than 85% of women had ER-positive disease. Mean (± SD) treatment duration was 27.1 ± 1.9 versus 26.7 ± 4.4 days and mean (± SD) compliance by pill count was 96% ± 17% versus 96% ± 19% in the metformin and placebo arms, respectively. The median intervals elapsed from last drug intake to blood drawing for circulating biomarkers and to surgery were 36 hours (interquartile range [IQR], 14 to 43 hours) versus 36 hours (IQR, 19 to 43 hours) and 67 hours (IQR, 21 to 74 hours) versus 67 hours (IQR, 21 to 75 hours) in the metformin and placebo arms, respectively.

The pre- versus post-treatment changes in Ki-67 by allocated arm and according to specific subgroups are shown in Table 2 and Appendix Table A1 (online only). Overall, metformin did not significantly change Ki-67 relative to placebo, with a mean proportional increase of 4.0% (95% CI, −5.6% to 14.4%) 4 weeks apart. However, the effect of metformin according to HOMA index was heterogeneous (P_interaction = .045, not corrected for multiplicity). In women with HOMA index more than 2.8, Ki-67 at surgery was lower on metformin, with a mean proportional decrease of 10.5% (95% CI, −26.1% to 8.4%), whereas in women with HOMA less than or equal to 2.8, Ki-67 increased by 11.1% (95% CI, −0.6% to 24.2%). The effect of metformin was also significantly modified by HOMA index in the subgroup of 119 women with luminal B tumors (P_interaction = .05). Although the overall Ki-67 proportional change was −1.75 (95% CI, −11.4 to 8.9, P = .7), it was 5.01 (95% CI, −7.2 to 18.8, P = .4) in women with HOMA index less than or equal to 2.8 (n = 86) and −16.0 (95% CI, −30.9 to 2.1, P = .08) in women with HOMA index more than 2.8 (n = 33). Similar effect modifications of the metformin effect on Ki-67 were noted according to BMI (P_interaction = .143), waist/hip girth ratio (P_interaction = .058), and CRP (P_interaction = .08), with a trend to a decrease of Ki-67 in the upper quartile and a trend to an increase of Ki-67 in the lowest quartiles (Table 2 and Appendix Table A1). Likewise, women who consumed alcohol showed a significant decrease of Ki-67 on metformin, whereas nondrinkers exhibited a reversed trend (P_interaction = .005). Figure 2 shows the subpopulation treatment effect pattern plot analyses of the Ki-67 change by HOMA index, BMI, and CRP. The plots show a reversal of metformin effect on Ki-67 for HOMA index values between 2 and 3, without any trend to a greater effect for higher HOMA values (Fig 2A, all assessable tumors; Fig 2B, luminal B tumors). Conversely, for BMI and CRP, a continuous trend was seen, suggesting a greater decrease of Ki-67 induced by metformin for high BMI and CRP values (Figs 2C and 2D). The effect of metformin on Ki-67 response was not significantly affected by the interval from treatment cessation to surgery (P = .4), ER/PgR status, molecular subtype, metabolic syndrome, and all variables listed in Table 1 (not shown).

The changes in insulin, glucose, CRP, and total cholesterol levels by allocated arm and according to BMI are shown in Table 3 and Appendix Table A2 (online only) for log-transformed data. Overall, metformin did not significantly modulate insulin and glucose levels, but there was a trend toward a significant effect modification by BMI (P_interactions = .092 and = .015, respectively), with a reduction of both biomarkers in women with BMI more than 27. Moreover, an overall reduction of CRP and total cholesterol was noted under metformin. There was no significant influence of the interval from treatment cessation to blood drawing on any circulating biomarker, including insulin and glucose levels (P = .4 and P = .6, respectively).

Adverse events are listed in Table 4. Diarrhea (P = .05) and nausea (P < .05) were the most frequent adverse events observed in the metformin arm. There was one grade 3 event (abdominal pain) and one serious adverse event (dizziness, vomiting), both in the placebo arm.

Therapeutic targeting of aberrant tumor metabolism has gained attention because the abnormally high proliferative activity of malignant cells requires high levels of nutrients to meet the increased demands for energy consumption and protein biosynthesis.⁴⁰ Abnormalities of host glucose and insulin metabolism are critical in promoting breast cancer,⁴¹ and preclinical¹⁵ and epidemiologic evidence^9,10 indicates that metformin may have therapeutic activity against breast cancer through indirect (insulin mediated) and direct mechanisms. We assessed in a proof-of-principle trial whether metformin decreased breast cancer proliferation in nondiabetic women using short-term Ki-67 response as surrogate biomarker.^19–21

Overall, metformin did not significantly decrease breast cancer proliferation, thus failing to support our primary hypothesis. However, in a planned subgroup analysis, the effect of metformin on Ki-67 was modified by insulin resistance status (HOMA index), with a nonsignificant 10.5% decrease of breast cancer proliferation in women with insulin resistance and a nonsignificant 11% increase in women without insulin resistance. The nominal statistical significance of the interaction test (P = .045) must be considered with caution because correction for multiple testing would bring it to nonsignificant levels. Yet, the qualitative interaction between metformin and insulin resistance may have important clinical implications and warrants confirmation in independent studies.

In additional unplanned analyses, similar modifications of metformin effects on Ki-67 by HOMA index were noted in women with luminal B tumors and in women who were overweight or obese, had abdominal obesity, had moderate alcohol intake, and had elevated CRP levels. In contrast to our secondary hypothesis, metformin did not differentially affect cancer proliferation according to ER status, although the statistical power of this analysis was less than 60% due to the lower than expected number of ER-negative tumors. Finally, a heterogeneous effect of metformin was noted on circulating insulin and glucose according to BMI, further suggesting a host-dependent therapeutic effect. Importantly, these findings are in line with the preventive effect of metformin on diabetes incidence,⁷ which was restricted to obese or glucose-intolerant women, with a greater benefit for increasing BMI and glucose levels.

The trend to a decrease in breast cancer proliferation in women with high HOMA index suggests that indirect, insulin- and glucose-mediated effects are the main mechanism of anticancer effect of metformin in human breast cancer and attenuate the role of direct effects on specific pathways of the AMPK–phospatidylinositol 3-kinase–mammalian target of rapamycin cascade,¹⁸ possibly because of the much higher concentrations used in vitro^15,32 and in vivo^16,42 relative to humans.⁸ An insulin-mediated mechanism is also in line with the vast majority of epidemiologic studies in patients with diabetes,⁹ in whom metformin is the first-line treatment and is particularly effective in reducing major diabetes complications in overweight and obese patients.^7,35 In its initial phase, which accounts for a large fraction of the Western population, type 2 diabetes is characterized by insulin resistance and is associated with the highest increase in breast cancer risk,^43,44 presumably as a result of the proliferative effects of elevated insulin, glucose, insulin-like growth factors, adipokines, sex hormones, and low sex hormone–binding globulin levels.^5,45 Thus the public health implications of a 10% decrease in Ki-67 on breast cancer risk in insulin-resistant women, coupled with the remarkable reduction of diabetes onset in women with glucose intolerance or obesity,⁷ could be substantial and supportive of preventive studies in women with these characteristics.¹ This is especially important in the United States, where the prevalence of obesity and insulin resistance is at least twice as much as our study population, where only 13% had a BMI more than 30 and approximately a quarter were insulin resistant. Likewise, the heterogeneous effect in luminal B cancers, although derived from a post hoc analysis, has potential therapeutic implications given the high frequency of this partially endocrine resistant tumor type.

The trend to an increased proliferation in women without insulin resistance, albeit nonsignificant, suggests a complex biologic effect of metformin in women without metabolic alterations. Animal studies have shown that metformin may have tumor suppressive effects where a metabolic phenotype of high caloric intake, metabolic syndrome, and diabetes exists, but no effect or even a growth promoting effect under normal energy intake.^16,46 In humans, metformin has shown differential effects according to grade of obesity and glucose intolerance, exercise, diet, alcohol use, and lipid levels, supporting the notion of a drug that regulates metabolic homeostasis depending on energy balance.^{7,13,16,35–37} Further mechanistic studies are necessary to understand these heterogeneous findings. Because regular and moderate alcohol use is a significant risk factor for breast cancer,⁴⁷ presumably through increased estrogen levels, the observation that metformin prevented the increased cancer proliferation in moderate alcohol users is noteworthy. One mechanism could be the decrease of testosterone levels observed in our study (unpublished data), similarly to what is observed in women with polycystic ovary syndrome.⁴⁸ Interestingly, a synergistic interaction between metformin and moderate alcohol use in lowering diabetes incidence was noted in the Diabetes Prevention Program trial,³⁷ suggesting the existence of common biologic pathways on energy regulation.

A presurgical study⁴⁹ recently conducted in Scotland in approximately 50 patients with breast cancer showed a significant decrease of Ki-67 under metformin, with modulation of different target genes. There was no evidence of specific subgroup effects, so a direct comparison with our study is difficult. However, the higher insulin levels compared with our study and the likely higher BMI levels of the Scottish population supports the notion that metformin lowers breast cancer proliferation in insulin resistant or overweight women.

Adverse events were in line with the known toxicity profile of metformin, with a higher incidence of gastrointestinal symptoms, mostly low-grade diarrhea and nausea of transient duration.⁸ Notably, compliance was extremely high, and there were no cases of hypoglycemia in this nondiabetic population.

The strengths of our study include use of an international reference laboratory for breast cancer tissue biomarkers, a large cohort representing a wide range of host and tumor characteristics that enabled subgroup analyses, and a sufficient power for the predicted interaction with HOMA index. A potential limitation, which was dictated by safety issues, is the interval from last drug intake to surgery (median, 67 hours; IQR, 21 to 74 hours), which is longer than the blood half-life of metformin (approximately 18 hours⁸). However, there was no significant association between any biomarker changes and the interval from treatment cessation, and metformin can reach higher tissue:blood concentrations,⁵⁰ which should not decrease its biologic effects up to several days from drug cessation. If anything, however, the wash-out time diluted our findings toward the null hypothesis without affecting overall conclusions.

In conclusion, our results suggest a heterogeneous effect of metformin on breast cancer proliferation depending on insulin resistance and other factors reflecting altered energy balance, with a trend to a decreased proliferation in women with elevated HOMA index and an opposite trend in women with normal insulin sensitivity. Although our observation is consistent with the current epidemiologic evidence derived from diabetic women, further clinical studies are necessary to confirm these findings in the general population.

We thank Serena Mora and Chiara Bresciani for the data management assistance and Brandy Heckman for her thoughtful scientific advice.

The author(s) indicated no potential conflicts of interest.