Metformin Reduces Desmoplasia in Pancreatic Cancer by Reprogramming Stellate Cells and Tumor-Associated Macrophages

Background Pancreatic ductal adenocarcinoma (PDAC) is a highly desmoplastic tumor with a dismal prognosis for most patients. Fibrosis and inflammation are hallmarks of tumor desmoplasia. We have previously demonstrated that preventing the activation of pancreatic stellate cells (PSCs) and alleviating desmoplasia are beneficial strategies in treating PDAC. Metformin is a widely used glucose-lowering drug. It is also frequently prescribed to diabetic pancreatic cancer patients and has been shown to associate with a better outcome. However, the underlying mechanisms of this benefit remain unclear. Metformin has been found to modulate the activity of stellate cells in other disease settings. In this study, we examine the effect of metformin on PSC activity, fibrosis and inflammation in PDACs. Methods/Results In overweight, diabetic PDAC patients and pre-clinical mouse models, treatment with metformin reduced levels of tumor extracellular matrix (ECM) components, in particular hyaluronan (HA). In vitro, we found that metformin reduced TGF-ß signaling and the production of HA and collagen-I in cultured PSCs. Furthermore, we found that metformin alleviates tumor inflammation by reducing the expression of inflammatory cytokines including IL-1β as well as infiltration and M2 polarization of tumor-associated macrophages (TAMs) in vitro and in vivo. These effects on macrophages in vitro appear to be associated with a modulation of the AMPK/STAT3 pathway by metformin. Finally, we found in our preclinical models that the alleviation of desmoplasia by metformin was associated with a reduction in ECM remodeling, epithelial-to-mesenchymal transition (EMT) and ultimately systemic metastasis. Conclusion Metformin alleviates the fibro-inflammatory microenvironment in obese/diabetic individuals with pancreatic cancer by reprogramming PSCs and TAMs, which correlates with reduced disease progression. Metformin should be tested/explored as part of the treatment strategy in overweight diabetic PDAC patients.


Introduction
The prognosis for patients with pancreatic cancer is dismal, with an overall five-year rate survival of 7% [1]. Obesity and type-2 diabetes mellitus (DM2) have become a pandemic worldwide [2,3]. Recent studies have demonstrated that these metabolic abnormalities are associated with the increased incidence, progression and poor prognosis of PDAC [4][5][6][7][8][9]. At diagnosis, approximately half of PDAC patients are overweight or obese, and up to 80% of patients present with diabetes or glucose intolerance [10][11][12][13][14][15][16][17]. DM2 and obesity may promote PDAC through pro-tumorigenic insulin and insulin-like growth factor-1 (IGF-1) [18][19][20] as well as chronic inflammation [21,22]. Hence, pharmacological interventions that target diabetes/obesity may also produce anti-tumor effects. One such agent, currently under intense investigation, is metformin, the most widely prescribed anti-diabetic generic drug that is also frequently administered to diabetic PDAC patients [23]. Metformin has been shown to improve treatment outcomes in preclinical models of PDAC [24][25][26][27][28][29][30]. In addition, it reduces the incidence of pancreatic cancer in diabetic patients as well as improves survival (reduced risk of death by 32%) in newly diagnosed cases [31][32][33]. However, the mechanisms of action of metformin in pancreatic cancer are not well understood. In vitro studies have addressed the impact of metformin on transcription factors, microRNAs, DNA damage, cancer stem cells and metabolism [34][35][36][37][38]. In addition, metformin has been shown to modulate the function of hepatic stellate cells, reduce oxidative stress in cancer-associated fibroblasts, and decrease tumor inflammation [34,35,39,40]. We and others have shown that reprogramming PSCs reduces the production of extracellular matrix (ECM) components such as collagen-I and hyaluronan (HA), and slows the progression of pancreatic cancer [41][42][43][44][45]. However, the impact of metformin on PSC activation, production of ECM components and tumor desmoplasia has never been described.
The aim of this study is to elucidate the functional mechanisms of metformin within the PDAC fibro-inflammatory tumor microenvironment. In vivo studies of pancreatic cancers to date have been mainly performed in xenograft models in normal weight/non-diabetic mice where metformin has been shown to be less effective [26,[46][47][48]. Hence, we used two immunocompetent syngeneic mouse models that closely mimic obesity / DM2, to better represent pancreatic cancer patients-the majority of pancreatic cancer patients present with either new onset DM2 or impaired glucose tolerance at the time of diagnosis [10][11][12][13][14][15][16][17]-and the target population The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.
Competing Interests: The authors of this manuscript have read the journal's policy and have the following competing interests: RKJ received consultant fees from Ophthotech, SPARC, and SynDevRx. RKJ owns equity in Enlight, Ophthotech, SynDevRx, and XTuit and serves on the Board of Directors of XTuit and the Boards of Trustees of Tekla Healthcare Investors, Tekla Life Sciences Investors, and Tekla Healthcare Opportunities Fund. No reagents or funding from these companies was used in these studies. This does not alter the authors' adherence to PLOS ONE policies on sharing data and materials. of metformin. Furthermore, we complemented our mouse models with in vitro studies as well as with analysis of human samples of pancreatic cancer from normal weight and overweight/obese patients. We found that metformin directly reprograms PSCs and TAMs and alleviates fibrosis and inflammation in obese/diabetic models of PDAC. These effects of metformin correlated with reduced ECM remodeling and epithelial-to-mesenchymal transition (EMT), ultimately affecting metastasis. We confirmed that ECM levels in human PDAC samples in overweight and obese patients were indeed lower in the patient population treated with metformin.

Animal experiments
The Institutional Animal Care and Use Committee of the Massachusetts General Hospital approved all experimental use of animals in this study. All animal procedures followed Public Health Service Policy on Humane Care of Laboratory Animals guidelines. Wild-type (WT) C57BL/6 and FVB male mice were originally obtained from The Jackson Laboratory (The Jackson Laboratory, Bar Harbor, Maine) and bred and maintained in our defined-flora animal facility. Mice were maintained on a 12-h light-dark cycle in a temperature-controlled barrier facility, with ad libitum access to food and acidified water. To generate obese/diabetic mouse models, mice (6-weeks old) were given a 60% fat diet (D12492, Research Diets, New Brunswick, NJ) for 10 weeks as previously described [49][50][51] (time of tumor implantation) and continued for the duration of tumor studies. For tumor experiments, the PAN02 and AK4.4 syngeneic PDAC models were used in C57BL/6 and FVB immunocompetent mice, respectively. PAN02 cells (SMAD4-m174) [52] were obtained from ATCC. AK4.4 cells (KrasG12D and p53 +/-) were kindly provided by Dr. Nabeel Bardeesy at MGH. These cells were isolated from mice generating spontaneous pancreatic tumors (Ptf1-Cre/LSL-Kras G12D /p53 Lox/+ ) [53]. Orthotopic pancreatic tumors were generated by implanting a small piece (1 mm 3  the average daily water intake for that cage during a period of 2 weeks prior to treatment initiation, and adjusted every three days based on water consumption and body weight of the animals. The approximate plasma concentration of metformin in patients with type 2 diabetes taking this drug is 0.05 mM, although it may accumulate in tissues and reach higher concentrations locally [35]. For in vitro experiments, a range of concentration from 0.05 to 25 mM depending on the cell line used (please see below for more details).

Effect of metformin on PSCs and macrophages in vitro
Standard MTT assays were performed on PSCs and macrophages treated with metformin in a range of 0.05-25mM, to examine the potential effects on cell viability. RAW 264.7 (mouse leukemic monocyte-macrophages) were obtained from ATCC and used to assess the effect of metformin on macrophages in vitro. Cells were seeded in 10 cm 2 petri dishes in serum/serum-free media and treated with metformin for 48h at concentrations ranging from 0.05 to 0.4 mM (concentrations that do not substantially affect cell viability). Following treatment, cells were collected for RNA and protein extraction in order to perform subsequent analysis of gene expression of cytokines and polarization markers, and for analysis of signaling and metabolic pathways. Human PSCs were seeded in 10 cm 2 petri dishes in media with 2.5% serum and treated with metformin for 48h at concentrations ranging from 0.1 to 10 mM. Cells were collected for protein extraction for analysis of the activation of fibrosis-related pathways. Additionally, PSCs were seeded in an 8-well chamber slide (20,000 cells/well), treated with metformin (1 mM, a concentration that does not affect cell viability) for 48h, and immunofluorescent staining was performed following standard protocols. The cells were fixed with 4% paraformaldehyde and blocked with 5% normal donkey serum for 1 h. They were incubated with the designated primary antibodies overnight at 4°C then for 2h with the appropriate secondary antibodies at RT. PBS was used for all washes, and the stained samples were mounted with Prolong Gold with DAPI. Images were acquired using a confocal microscope. The antibodies used and image acquisition settings are described below. Both cell types were grown in Dulbecco's modified Eagle's medium (Sigma, St. Louis, MO) supplemented with 5% (v/v) heatinactivated fetal bovine serum (FBS; Sigma), 100 units/ml penicillin and streptomycin. Cells were cultured at 37°C in a humidified atmosphere including 5% CO 2 .

Human samples
Human samples of pancreatic cancer were obtained from the MGH tissue repository (http:// www.massgeneral.org/research/resourcelab.aspx?id=31) under an active IRB protocol (Partners Healthcare IRB approval number: 2013P001969). Written informed consent from the donor or the next of kin was obtained for the use of these samples for research purposes. Tumors selected received no prior chemotherapy or radiation therapy before the surgical specimen was collected at the time of tumor resection. A total of 28 samples were randomly selected from this subset of samples. Body mass index was obtained for the respective sample. 7 of the samples (25%) correspond to patients that were medicated with metformin at the time of resection. Paraffin sections were stained for collagen-I and HA as described below. Images were acquired using confocal microscopy and quantified using Matlab. Data were analyzed anonymously.
Multiplex array. Each tumor sample was homogenized directly in lysis buffer for protein extraction. 2ug/ul of sample was used. A multiplex inflammatory multiple cytokines protein array was used (V-PLEX Proinflammatory Panel1 mouse kit, Cat. Number: K15048D) for ELISA analysis.

Immunofluorescence/Immunohistochemistry
For analysis of frozen sections of mouse tumor tissues, the tumors were excised and frozen in optimal cutting temperature compound (Tissue-Tek). Transverse tumor sections, 10 μm thick, were immunostained with specific antibodies. To obtain mosaic images of tumor sections, an Olympus FV1000 confocal laser-scanning microscope was used. A 10x air objective acquired 1260-μm square tiles, and an automated stage scanned throughout the entire cross-section of tumor tissue. The imaged tiles were stitched into a final mosaic image using Olympus software. Antigen expression was quantified by measuring the area occupied by the stain of interest normalized by the area of DAPI-stained nuclei (i.e., unitless), and analyzed using a custom algorithm in MATLAB (The MathWorks). Identical settings and thresholds for analysis were used for all tumors. Antibodies used for immunofluorescence were the following: Collagen-I [LF-68 antibody, 1:50 dilution, provided by Dr Larry Fisher (NIDCR)]; Hyaluronan (biotinylated hyaluronan proteoglycan fragment, 385911, Calbiochem, 1:200 dilution); αSMA (C6198 antibody, Sigma, 1:500 dilution); and F4/80 (MCA497A488 antibody, ABDserotec, 1:200 dilution). Cy3-, Cy5-or FITC-conjugated secondary antibodies were used for the detection of signals by confocal microscopy. Slides were counterstained with DAPI for nuclear staining.

MMP activity assay
Each tumor sample was homogenized directly in lysis buffer for protein extraction. A standard commercial assay (Abcam ab112146) was used to detect the general activity of MMPs in the tumor samples. Tumor lysates were incubated with a fluorescent substrate [fluorescence resonance energy transfer (FRET) peptide] for 1, 2, and 16 hours, and fluorescence was measured using a fluorescent microplate reader.

Macrophage isolation
After cutting the tumor tissue into pieces 0.5 mm in diameter, tumors were dissociated with collagenase and hyaluronidase for one hour. Dissociated tumors were washed with sorting buffer (0.5% BSA in PBS) and passed through cell strainers before they were blocked using Fcr and subsequently incubated with a F4/80 biotinylated primary antibody (eBiosciences) for 15 min. Anti-biotin conjugated magnetic beads (Miltenyl Biotech) were added to the cell suspension and incubated for 15 min before exposing the cells to magnetic columns (MACS LS-columns from Miltenyl Biotech) for cell separation.

Statistical analysis
Statistical differences between groups were assessed by a two-tailed Student t-test. Two-way analysis of variance was used when comparing body weight over time between the control and metformin-treated groups. The incidence of metastasis and wall invasion was analyzed using chi-square. All statistical analyzes were performed using Prism Graphpad software. All results were considered statistically significant when the P value was less than 0.05 when calculated with the appropriate statistical test. Results are presented as the mean ± standard error.

Metformin reduces desmoplasia in mouse and human PDACs
We have shown that the reduction of desmoplasia-lowering ECM components such as HA and collagen-I-in PDACs potentiate anti-tumor treatments [41][42][43]. Hence, we examined the relationship between metformin treatment and desmoplasia in PDACs in chemo/radiotherapy naïve PDAC surgical samples. We found that in overweight or obese patients under treatment with metformin, levels of HA were 30% lower than in patients not taking metformin (Fig 1i  and 1ii). Interestingly, no difference in HA levels was observed in metformin treated patients of normal weight. On the other hand, treatment with metformin appeared to have no impact on collagen-I expression in either body weight group (S1 Fig). To evaluate if we could recapitulate these findings in pre-clinical models, we used FVB and C57BL/6 mice fed with a high-fat diet. We, along with others have shown that a high-fat diet induces obesity and metabolic abnormalities typical of DM2, elevated glucose, insulin and IGF-1 [50,[57][58][59][60] in these strains. After 10 weeks on the high-fat diet, AK4.4 and PAN02 tumors were orthotopically implanted in obese FVB and C57BL/6 mice, respectively. The animals were randomly assigned to metformin in drinking water (300 mg/Kg) or no treatment at day 7 until day 21 when plasma and tumors were collected. Treatment with metformin correlated with reduced expression of HA by 64% (Fig 2Ai and 2Aii) and of collagen-I by 35% (Fig  2Ai and 2Aiii) in AK4.4. tumors. Furthermore, the percentage of activated PSCs (as determined by the expression of alpha-smooth muscle antigen, αSMA) that co-express HA and collagen-I decreased by 58 and 38%, respectively (Fig 2Bi-2Biii). In a second, less desmoplastic model (PAN02), metformin decreased the expression of HA by 40% and of collagen-I by 22%, although it did not reach statistical significance (S2i- S2iii Fig). Nonetheless, metformin significantly reduced the density of collagen-I positive activated PSCs in tumors by 54% in this model (S2v Fig). The density of HA positive activated PSCs in PAN02 tumors was also reduced by 57% (S2iv Fig) but did not reach significance. We have shown previously that Angiotensin-II receptor 1 (AT1) is critical for HA and collagen-I production in PDACs [43]. Indeed, we observed that metformin was able to reduce the expression of AT1 (Fig 2C). Taken together, these data indicate that metformin reduces desmoplasia in PDACs in overweight/obese hosts.

Metformin affects desmoplasia by directly reducing TGF-β signaling and production of collagen-I/HA by PSCs
Collagen-I and HA are essentially produced by activated PSCs in PDACs [43,61]. To determine if metformin treatment can directly reduce collagen-I/HA expression in PSCs, we incubated PSCs in vitro with metformin. We found that, at a dose (1 mM) that does not substantially affect the viability of PSCs (S3 Fig), metformin decreased the expression of HA and collagen-I (Fig 3Ai-3Aiii). This indicates metformin acts on PSCs to reduce the  production of these critical ECM components. Furthermore, we found that metformin (0.1-1mM) reduced expression of AT1, TGF-β and downstream signaling via SMAD-2, as well as PDGF-β, all key players in ECM production by PSCs (Fig 3B). In addition, metformin also affected the activation of canonical signaling pathways that promote PSC activation and fibrosis [62], in particular ERK, p38, and STAT3, although at relatively higher doses (1-10 mM) ( Fig 3B). Importantly, in whole tumors, metformin decreased activation of STAT3 in both models, with a trend for reduced p38 activation in PAN02 (S4i-S4iii Fig), suggesting that metformin can accumulate at concentrations high enough to affect these signaling pathways in vivo. Taken together, these data indicate that metformin affects desmoplasia by directly reducing AT1/TGF-β/STAT3 signaling and production of collagen-I/HA by PSCs.

Metformin also improves desmoplasia by preventing recruitment and M2 polarization of macrophages in PDACs
The inflammation that occurs in PDACs is a major component of desmoplasia [63]. In particular, tumor-associated macrophages (TAMs) are a major source of cytokines that aggravate desmoplasia in PDACs [64] and negatively affect disease outcome [65]. Therefore, we determined here the effect of metformin on TAM infiltration in tumors. We found that TAM levels were 60% lower with metformin treatment in the AK4.4 model (Fig 4A). They also tended to be lower (~30%, not significant) in the PAN02 model ( S5A Fig). To determine a direct effect of metformin on macrophages, we incubated macrophages with metformin in vitro for 48h at increasing concentrations. We found that metformin affected the viability of macrophages at doses of 0.4 mM or higher (S6 Fig). Metformin at a concentration of 0.05 mM (similar to the concentration measured in plasma of patients taking metformin) reduced M2 markers such as Arg-1 and IL-10, while metformin at doses higher than 0.2 mM reduced both M1 and M2 markers (Fig 4B). In flow-sorted TAMs from PAN02 tumors in vivo, we indeed confirmed an effect of metformin on the expression of M2 markers Arg-1 (~1/2) and IL-10 (~2/3) without significantly affecting the expression of M1 markers (Fig 4C). We then sought to understand how metformin affects these cells. Several canonical and non-canonical signaling pathways can be activated in PDACs during inflammation and promote expression of M2 markers on TAMs [66][67][68][69]. Consistent with the effects on TAMs, metformin reduced activation of STAT3, JNK, AKT and p38 in macrophages in vitro at concentrations lower than 0.2 mM (Fig 4D). As mentioned above, in whole tumors metformin decreased activation of STAT3 in both models, with a trend for reduced activation of p38 in PAN02, in line with our in vitro results (S4i-S4iii Fig).
It has been shown that STAT3 activity is decreased by metformin via activation of metabolic energy sensor AMP-activated protein kinase (AMPK) in multiple cell types [70]. Indeed, we found that the inhibitory effects of metformin on STAT3 signaling in macrophages in vitro associated with activation of AMPKα and downstream enzyme Acetyl-CoA Carboxylase (ACC) (the latter evident only in serum added media) in these cells (Fig 4D and S7 Fig). Taken together, these data indicate that metformin reduces TAM infiltration as well as expression of M2 markers, which may be mediated at least in part via AMPK/STAT3 signaling inhibition in macrophages. In addition, inflammation in tumors is characterized by an excess of inflammatory cytokines that promote desmoplasia, and metformin has been shown to affect multiple inflammatory mediators [34,35]. Here, we found that metformin reduced the expression of IL-1β and CXCL1 in AK4.4 tumors (Fig 4E). A reduction of IL-1β after metformin treatment also occurred in the PAN02 tumor model (S5Bi and S5Bii Fig). In addition, a broader panel of   Fig). In conclusion, we found that metformin reduces the production of desmoplastic cytokines (e.g., IL-1β) as well as infiltration and M2 polarization of TAMs, which may be mediated at least in part via AMPK/STAT3 signaling inhibition in macrophages.

Metformin reduces ECM remodeling, EMT, and metastasis
In addition to producing ECM components, PSCs also promote ECM remodeling and EMT to facilitate invasion and metastasis [43,71,72]. Hence, we determined whether the effects on PSCs by metformin also extend to these processes. Indeed, metformin reduced the expression in AK4.4 tumors of multiple genes involved in ECM remodeling (including MMPs) and EMT (Fig 5A). In addition, metformin treatment unregulated genes that prevent ECM remodeling (Fig 5A). Though to a lesser extent, similar findings were observed in PAN02 tumors (S8A Fig). At the protein level, we also observed a reduction of metalloproteinase 9 (MMP-9, Fig 5Bi  and 5Bii) by 70% in the AK4.4 model with metformin treatment. Average MMP-2 levels were also approximately half (not significant) in metformin-treated PAN02 tumors (S8Bi and S8Bii  Fig). Consistently, we confirmed in vitro that metformin decreased protein levels of MMP9 in PSCs (S8C Fig). Furthermore, MMP activity in tumors was also decreased in metformintreated animals compared to control mice (Fig 5C). In addition to ECM remodeling, EMT was also affected. At the protein expression level, the EMT marker vimentin was decreased and Ecadherin was increased in AK4.4 with similar trends in the PAN02 model, confirming reduced EMT (Fig 5Bi and 5Bii, S8Bi and S8Bii Fig). Consistent with these effects on the tumor microenvironment, metformin reduced the incidence of mesenteric peritoneal and retroperitoneal wall metastasis metastasis (percentage of mice affected) (Fig 5D) as well as the average number of metastasis per mouse (Fig 5E). These effects were particularly evident in the more metastatic model PAN02, although similar trends were obtained for the less metastatic AK4.4 model.

Discussion
Metformin improves cancer outcomes in preclinical models of PDAC [25,26] and diabetic patients with pancreatic cancer [32,35] though the underlying mechanisms are not well understood. Hence, there is a need to continue to study and elucidate the mechanisms of action of metformin in PDACs. Fibrosis and inflammation are critical components of the desmoplasia that characterize PDACs [73]. We and others have previously shown that reprogramming the central instigator of the desmoplastic microenvironment-PSCs-can be an effective intervention in the treatment of PDACs [43,61,74]. Furthermore, metastases in PDACs are facilitated by the active desmoplastic fibro-inflammatory microenvironment that promotes ECM remodeling, EMT and tumor invasion [43][44][45]75]. We uncovered in this study a previously unknown role of metformin on the activity of PSCs, TAMs, tumor fibrosis and inflammation, and how it impacts systemic dissemination of the disease. We found that in overweight and obese patients -which appear to have increased levels of ECM components in tumors -, metformin treatment reduced tumor levels of HA. We confirmed that metformin robustly affected HA as well as collagen-I, though to a lesser extent, in preclinical obese/diabetic mouse models of syngeneic PDACs. Furthermore, we found that the alleviation of desmoplasia occurred due to a direct effect on HA and collagen-I production by PSCs, which was associated with a reduction of AT1/PDGF-ß expression and TGF-ß/SMAD-2 signaling. Although the precise mechanisms for the preferential effect of metformin on HA over collagen-I in all preclinical models and human cancer patients studied so far are not clear, it is an intriguing and potentially important finding. Nonetheless, we have recently observed that HA plays an equally important role as collagen-I in reducing therapy delivery and efficacy in PDACs [43,76].
In addition to reducing fibrosis, we found here that metformin reduces the production of pro-metastatic cytokines. In both tumor models, metformin reduced the secretion of IL-1ß, which has been shown to promote metastasis in a PDAC model [30]. IL-1β in tumors is typically produced by PSCs, inflammatory and tumor cells, is involved in macrophage recruitment and PSC activation, and both IL-1β and CXCL1 worsen desmoplasia [77][78][79][80]. Consistently, it has been shown that IL-1β mediates at least in part the effects of metformin on the malignant transformation of mammary epithelial cells [81]. Furthermore, we found that metformin reduced the levels of chemokines involved in TAM recruitment and function (e.g. CSFs, CCL3) [82][83][84], and consistently, metformin reduced the recruitment of TAMs and their expression of M2 markers in vivo and in vitro at clinically relevant doses. Our data corroborate the previous findings of Karnevi [17]. STAT3 promotes polarization of TAMs to an M2 phenotype [85], and It has been shown that STAT3 activity is decreased by metformin [81] in part via activation of AMPKα in multiple cell types [70,86]. We found here that the effects of metformin on macrophage polarization associated with activation of AMPKα/ACC and reduction of STAT3 signaling.
Activated PSCs and M2 TAMs have been shown to promote ECM remodeling and EMT [87,88]. Here we found that, consistent with the effects on these cells, metformin reduces ECM remodeling and EMT. This is also in line with a recent report describing the modulation of EMT by metformin in the PAN02 model [89], and with the finding that metformin impeded TGF-ß-promoted EMT in breast cancer cells [90]. Importantly, ECM remodeling and EMT have been shown to promote tumor invasion and metastasis [43][44][45]75], and as expected we observed a decrease in metastasis in mice treated with metformin.
The modulation of systemic and local metabolism has been the major focus of studies evaluating the effect of metformin in PDAC. Metformin could improve systemic levels of Insulin/ IGF-1 and glucose, and affect Insulin/IGF-1 signaling and AMPK/ACC activation. However, we found that this only occurred (and mildly) in one of the models (PAN02). This suggests that the effects of metformin on desmoplasia do not correlate with global activity of metabolic pathways. In fact, despite earlier reports suggesting that IGF-I may be involved in cancer risk and outcome [91], subsequent clinical studies failed to establish anti-IGF-I agents as cancer therapeutics [92]. In addition, there is no convincing evidence for a carcinogenic role of any insulin derivative currently used in therapy for diabetes [93]. Similarly, the beneficial effects of metformin did not correlate with levels of blood sugar in patients [33], and in a pre-clinical model Franco and colleagues have shown that the effects of metformin may be more dependent on the direct effect on tumors rather than on systemic metabolism [94]. In addition, despite the report that autophagy can be affected by metformin to reduce tumor progression [95,96], we found no evidence of this in our study. Of note, metformin did not affect body weight through the experiment (S10 Fig), suggesting a body weight-independent effect on tumor growth. Importantly, two very recent studies-one retrospective [97] and the first prospective study [98] indicated that metformin might not be uniformly beneficial. The benefit in some but not all studies suggests that a subset of tumors may not respond to metformin and that a careful selection of patients may be required for metformin to be effective. We found that metformin's effect on desmoplasia in patients only occurred when their BMI was higher than 25 (overweight and obese patients). This indicates that metformin may not be beneficial in normal weight patients, and suggests that BMI should be explored as a potential biomarker of response to this drug.

Conclusion
In conclusion, this study indicates that in an overweight/obese condition, metformin reprograms the fibro-inflammatory tumor microenvironment and ultimately reduces metastasis. We found that metformin directly reduces AT1/PDGF-ß and TGF-ß signaling and ECM production by PSCs-preferentially HA. Metformin also reduces inflammation-another key element of desmoplasia-through reduction of cytokine production, and recruitment and M2-polarization of TAMs. This was associated with AMPK activation and STAT3 signaling inhibition in macrophages. Finally, the alleviation of desmoplasia by metformin was associated with reduced ECM remodeling, EMT and systemic metastasis (Fig 6). Importantly, the effects on desmoplasia observed in human samples seem restricted to an overweight/obese population, which appear to have tumors with increased content of ECM components. With nearly 200 trials ongoing to address the effect of metformin on diabetic and non-diabetic cancer patients, understanding the yet elusive mechanisms of action of metformin may provide an opportunity to uncover potential biomarkers of response and define strategies of patient stratification for the judicious use of this highly promising yet generic drug.