Background
Female breast cancer is currently the most diagnosed cancer in many countries worldwide [
1]. Survivors of breast cancer are the largest population of cancer survivors, numbering more than 3.8 million women as of January 2019 in the United States alone [
2] with > 90% survival after 5 years [
3]. Cancer recurrence is a major health concern in this population, particularly in those with overweight or obesity [
4]. Interventions that reduce recurrence risk are needed. In addition, analyses of underlying molecular mechanisms associated with recurrence-reducing interventions can provide insight into biochemical pathways that mediate cancer risk.
Metformin is the most widely prescribed medication to improve glycemic control in individuals with type 2 diabetes. In addition to its glucose lowering effects, metformin use is associated with clinically significant weight loss and improved insulin sensitivity [
5]. Epidemiological studies show that metformin use diminishes cancer occurrence, suggesting that metformin intervention may reduce risk of recurrence in survivors of obesity-related cancers, e.g., breast cancer [
6].
Molecular mechanisms that mediate the metabolic benefits of metformin include inhibition of gluconeogenesis (hepatic and renal) [
7], activation of AMP-activated protein kinase (AMPK) [
8], and inhibition of mitochondrial respiration and glycerophosphate dehydrogenase [
9]. Recent studies demonstrate that additional metabolically beneficial effects of metformin are mediated by the gut, including alterations in enterocytes and microbiota [
10]. However, the mechanisms by which metformin improves metabolic and cancer outcomes are not yet fully understood.
In this study, we employed targeted and untargeted metabolomics approaches to explore metabolites and metabolic pathways associated with metformin treatment in breast cancer survivors. To enhance statistical power, we conducted metabolomic profiling of metformin treatment in plasma samples at baseline and follow-up from two randomized controlled trials, testing the impact of metformin on body weight and the metabolic profile among 373 breast cancer survivors, the Reach for Health Study (US-based) and the MetBreCS study (Italy-based).
Discussion
Our metabolomic analyses, using both targeted and untargeted approaches, revealed previously unreported metabolic pathway alterations as well as several previously reported in preclinical models and humans [
20,
21]. To the best of our knowledge, this study is the first to assess metabolic responses to metformin treatment via metabolomics in breast cancer survivors. Several classes of metabolites were altered following treatment with metformin, including amino acids such as branched-chain amino acids (BCAAs) and their alpha-ketoacid breakdown products, indoles, xanthines, phosphatidylcholines (PCs) and fatty acids.
Participants treated with metformin had significantly higher levels of the BCAAs leucine and isoleucine and their alpha-keto acid derivatives (BCKAs; 3- and 4-methyl-2-oxovalerate, PubChem CID 47 and 70, respectively), which paradoxically are associated with type 2 diabetes risk and insulin resistance in most studies of metformin naïve subjects [
22] but not all [
23]. Possible mechanisms underlying our observation include known effects of metformin to decrease the overall activity of mitochondrial BCAA catabolic and oxidative phosphorylation pathways [
24], enhance gene expression of sirtuin-1 (SIRT 1), and precipitate AMP-activated protein kinase (AMPK) signaling [
23]. Metformin is known to decrease the expression of branched-chain amino acid transaminase (BCAT) 2 [
25]. Metformin inhibition of the oxidative phosphorylation pathway complex I results in the accumulation of mitochondrial NADH, which together with elevated NADH generated from fatty acid oxidation in individuals with overweight obesity, negatively feedback on BCAA catabolism at the irreversible, rate-limiting step of BCKA decarboxylation in the mitochondria catalyzed by branched-chain alpha-keto acid dehydrogenase complex (BCKDH) complex. BCKDH has been previously implicated in pro-inflammatory signaling via MAPK [
26] as well as in the tumorigenesis of colorectal cancer [
27]. Overall effects of metformin to decrease BCAT expression and BCKDH activity may explain the significantly elevated levels of leucine, isoleucine, and their BCKA derivatives. These results merit further investigation of the effects of increased BCAA on tumorigenesis and cancer recurrence.
Effects on metabolism of other amino acid (proline, tyrosine, alanine) were observed in the current study. We observe that proline was significantly increased with metformin treatment, which may be protective against cancer [
20,
28]. Interestingly, increased proline dehydrogenase activity has been shown to fuel proline catabolism and consequently lead to increased growth of BC cells in 3D culture and in vivo metastasis formation [
29]. We observe that tyrosine was significantly decreased with metformin treatment. Elevated levels of the aromatic amino acid tyrosine are strongly associated with the risk of type 2 diabetes and mitochondrial disfunction [
25], while reduced levels of tyrosine are a well-characterized effect of metformin treatment [
22]. Higher levels of tyrosine are associated with poor prognosis and therapeutic response [
30], and the success of tyrosine kinase inhibitors in cancer treatment and management underscores the clinical relevance of high tyrosine levels [
31]. Indeed, a recent randomized phase 2 clinical trial demonstrated that patients with advanced lung adenocarcinoma allocated to a combination tyrosine kinase inhibitor plus metformin treatment demonstrated significantly longer progression free survival as compared to the randomized group receiving only the tyrosine kinase inhibitor [
32]. Our finding that tyrosine was significantly lower in participants receiving metformin suggests reduction of tyrosine levels may be one mechanism by which metformin synergistically contributes to tyrosine kinase cancer treatment. We observe that alanine was significantly increased with metformin treatment. Similarly, a recent comparative metabolomics study of circulating prognostic metabolites found a significant inverse association between serum levels of alanine and the risk factor of high mammographic breast cancer density [
33], suggesting that elevation of alanine levels during metformin treatment may be a key contributor to its association with cancer risk. Cumulatively, our results commensurate with previously published reports and provide a strong basis for further investigation of metformin’s effects on tyrosine, alanine, and proline and associations with cancer risk reduction.
Previous studies have demonstrated an immediate and sustained decrease in citrulline levels following administration of metformin in humans [
34], similar to findings of the current study (FDR-corrected
P < 0.001). Citrulline is primarily consumed in the kidney as a substrate for arginine synthesis [
35], interestingly, arginine levels were also significantly reduced in participants assigned to the metformin arm of the current study (FDR-corrected
P < 0.001). These changes upon metformin treatment may be explained by diminished citrulline synthesis in the gut [
36], lowered hepatic production of citrulline [
37], or increased renal uptake of citrulline [
38]. Additionally, reduced citrulline levels are associated with increased intestinal permeability [
39], which could potentially lead to the increased permeation of gut metabolites, e.g., indoxyl sulfate identified in our study, into the bloodstream.
Significant differences in PC species were also observed in response to metformin treatment. A recent study noted significantly lower levels of the long-chain unsaturated PC ae C36:4 following 4–6 weeks of metformin treatment in individuals with type 2 diabetes (T2D) [
34], while an earlier study reported the same finding for PC ae C36:4 in T2D patients under metformin [
40]. These findings were paralleled in participants receiving metformin in this study (FDR-corrected
P < 0.001), thus showing similar metabolic effects of metformin in our population, although less strong. Reports suggest that reductions in PCs and aromatic amino acids (e.g., tyrosine) may not be due to metformin directly, but to the confounding effects of weight loss and subsequently improved metabolic status of study participants [
41,
42]. However, this effect would have been minimized in our analyses as BMI was controlled for as a confounding factor. A study investigating the mechanisms of metformin for cancer treatment found that metformin-treated cells exhibited decreased formation of PCs along with a decrease in PC-synthesizing enzymes, and diminished 14C incorporation into fatty acids for membrane synthesis [
43]. In our study, levels of four acyl chain PCs were significantly reduced after metformin treatment.
Our fatty acids results suggest metformin treatment effects on several desaturase enzymes, including those encoded by the stearoyl-CoA 9-desaturase genes SCD1 and 5 and the fatty acid desaturase genes FADS5 and 6. Metformin treatment profiles indicate an increase in SCD activity (18:1n9/18:0 desaturation index), however, we are unable to differentiate the enzymatic isoform origin of increased activity (i.e., SCD1 vs. SCD5). SCD activity and association with cancer risk factors in humans varies by cancer type and SCD isoform. Metformin suppresses SCD1 expression via AMPK modulation [
44], so the apparent change in stearoyl-CoA 9-desaturase activity that we observe may in fact be due to increased SCD5 activity. A recent study in > 4900 breast cancer patients reports that relatively higher expression of SCD5 improves relapse-free survival [
45]. Also, via AMPK activation, metformin treatment reduces the activity index and gene expression of FADS2 [
46‐
48], as observed in our data via greater proportion of metformin-treated participants with reduced percentage of long chain omega-6 fatty acids and stabilized percentage of 20:2n6 compared to that in the placebo group. A more generalized downregulation of desaturase enzymes is observed in metformin-treated cells, with significant reductions both in activity and expression of FADS1-3 [
49]. Metformin specifically reduces FADS1 activity [
46] and individuals with FADS1 genetic variants with reduced activity have decreased breast cancer risk [
50]. Less FADS1 and 2 activity leads to a reduction in long chain omega-6 fatty acids and a less pro-inflammatory tissue milieu that may aid in reducing cancer risk and growth promotion [
51]. Future studies should devote more investigational efforts to understanding this intricate relationship between metformin effects on various lipid species, desaturase activities and the consequences for cancer biology.
Metabolite signatures in our analyses indicate an effect of metformin on the gut microbiota. Prior research has shown that metformin influences the activity of gut bacteria and suggests that the metabolic benefit of metformin may in part be mediated by these effects. A recent randomized controlled trial found that germ-free mice which were inoculated with fecal microbiota from humans with type 2 diabetes and receiving metformin treatment, showed significantly improved glucose intolerance [
52]. In humans, this association may be explained by the modulatory workings of the estrogen-gut microbiome axis, a bidirectional relationship mediated via the actions of estrogen and β—glucuronidase [
53]. Additionally, a recent review noted that specific classes of microbiota-derived metabolites, most notably BCAAs and indole derivatives, have been previously implicated as potential biomarkers in metabolic disorders, such as cancer [
54], a finding which was echoed in the current study as well (see significant changes in leucine, isoleucine, tyrosine and indoxyl sulphate in Tables
2,
3). Notably, our results highlight probable perturbations in indole metabolism and aryl hydrocarbon signaling [
54], as indicated by significant changes observed in indoxyl sulphate. Future studies would do well to incorporate gut microbiome sequencing to further test this possible association.
An interesting signature of increased caffeine metabolism emerged in the metformin treatment profiles. Caffeine metabolites paraxanthine and theophylline are primarily formed by the enzyme CYP1A2 in the liver. Paraxanthine is the primary metabolite of caffeine (~ 80%) [
55] with slower clearance than caffeine. The higher paraxanthine/caffeine and theophylline/caffeine ratios following metformin treatment suggest a higher metabolic activity of CYP1A2. To our knowledge, metformin has not previously been shown to impact the activity of CYP1A2. Metformin is not known to be a substrate of CYP1A2. Although we do not know coffee consumption details for our study participants, potential changes in the habitual consumption of coffee due to the metformin treatment may contribute to the observed changes. We could not find any literature evidence of coffee consumption changes with metformin use. A currently ongoing clinical trial aims to study a six-drug cocktail of probes for CYP enzymes and metformin interactions, including caffeine [
56]. The randomized controlled design of our studies minimizes impact of other factors on our observations such as variation in coffee consumption frequency or timing or CYP1A2-modifying drug use across participants.
Nevertheless, the significant association between metformin treatment and decreased levels of caffeine and its downstream metabolites warrants further investigation of CYP1A2 expression and activity given existing evidence supporting a role for CYP1A2 involvement in BC pathogenesis. Notably, CYP1A2 is known to be a key enzyme in BC etiology, contributing variably to carcinogen activation as well as estrogen synthesis and anti-inflammatory pathways [
57]. Furthermore, specific isoforms of CYP1A2 have shown reduced activity and led to increased BC risk, whereas the − 3860A variant has consistently demonstrated increased metabolic clearance of caffeine and concomitant reduction in BC risk [
57,
58]. Moreover, CYP1A2 activity has also been linked to type 2 diabetes mellitus [
59,
60] and has demonstrated interaction effects with coffee consumption and BRCA1 mutation [
61,
62]. In addition, CYP1A2 has monoxygenase and epoxygenase activities which result in the generation of anti-inflammatory metabolites of omega-6 and -3 polyunsaturated fatty acids [
63‐
65]. Epoxide metabolites generated from docosapentaenoic acid (DPA) and eicosatetraenoic acid (ETA) have a variety of anti-cancer-related activities in vivo including suppressing inflammation, angiogenesis, and growth and metastasis of human breast and prostate cancer cell lines. Thus, one mechanism potentially underlying the protective effects of metformin may be modulation of CYP1A2.
Strengths of this study include the broad metabolomics approaches utilized, including both targeted and untargeted analyses as well as fatty acid analyses. The study included plasma samples from a large number of participants (n = 373) enrolled in two different randomized controlled trials with sample collections at baseline and study end for 352 participants (paired samples). Limitations of this study include the fact that the two pooled populations come from two different study designs and study centers, which resulted in several baseline differences between the two populations of women, including age, baseline ΒΜΙ and menopausal status. Moreover, the US study also included a weight loss intervention for half the sample, which was not present in the Italian study. These differences lead to the necessity of controlling for several factors in multivariate analysis, reducing the parsimony of the models. However, a series of sensitivity analyses performed in the study and the control for confounders validate the findings and allowed exploration of the effects the BMI, the weight-loss intervention and center adjustments had in the selected models. Furthermore, the pooled analysis including the two different cohorts allowed to generalize the results beyond the single trial, since the results were consistent when they were analyzed by study. Indeed, we observed equidirectional changes in all 11 significant metabolites of the targeted analysis between Italian and USA cohorts, while five of these 11 metabolites retained significance in the Italian cohort. Among the 20 significant features identified by the pooled untargeted analysis, 19 features exhibited changes in the same direction between study cohorts and eight out of 20 features were significant in the smaller Italian cohort. Cumulatively, results between the two study cohorts are well-aligned. Moreover, as in any untargeted metabolomics approach, the need for controlling multiple testing might lead to over-penalization of
P-values and loss of relevant metabolites [
66]. However, the approach of performing targeted analysis in the same set of samples allowed the discovery of additional metabolites, which were not picked up by the untargeted approach. Finally, our methods did not enable identification of the exact composition of the glycerophospholipids detected restricting our ability to interpret the altered phosphatidylcholines’ role in our study.