Metformin sensitizes leukemic cells to cytotoxic lymphocytes by increasing expression of intercellular adhesion molecule-1 (ICAM-1)

Solid tumor cells have an altered metabolism that can protect them from cytotoxic lymphocytes. The anti-diabetic drug metformin modifies tumor cell metabolism and several clinical trials are testing its effectiveness for the treatment of solid cancers. The use of metformin in hematologic cancers has received much less attention, although allogeneic cytotoxic lymphocytes are very effective against these tumors. We show here that metformin induces expression of Natural Killer G2-D (NKG2D) ligands (NKG2DL) and intercellular adhesion molecule-1 (ICAM-1), a ligand of the lymphocyte function-associated antigen 1 (LFA-1). This leads to enhance sensitivity to cytotoxic lymphocytes. Overexpression of anti-apoptotic Bcl-2 family members decrease both metformin effects. The sensitization to activated cytotoxic lymphocytes is mainly mediated by the increase on ICAM-1 levels, which favors cytotoxic lymphocytes binding to tumor cells. Finally, metformin decreases the growth of human hematological tumor cells in xenograft models, mainly in presence of monoclonal antibodies that recognize tumor antigens. Our results suggest that metformin could improve cytotoxic lymphocyte-mediated therapy.


Metformin regulates expression of stress ligands and ICAM-1 in leukemic cells. We have previ-
ously shown in several studies that the presence of wtp53 impacts the effect of several co-treatments involving the so-called metabolic drugs, e.g. metformin 20,27,28 . Hence, we investigated if p53 status can affect the effect of metformin on ligands that are recognized by cytotoxic lymphocytes. We treated with 2 mM metformin for 3 days 3 acute myeloid leukemia (AML) cell lines with different p53 status (OCI-AML3 cells express wt p53, HL60 are p53 null and NB4 express mutant (mut) p53 20,29 ) and analyzed MICA/B and ULBP1 expression on plasma membrane. In addition, we evaluated levels of other ligands such as intercellular adhesion molecule-1 (ICAM-1), a ligand of the lymphocyte function-associated antigen 1 (LFA-1). ICAM-1/LFA-1 interaction is essential for target cell recognition by CL 30,31 . We studied also MHC-I, called HLA in humans, which is the ligand of the killercell immunoglobulin-like receptors (KIRs), the main NK cell inhibitory receptors 22 and the myeloid marker CD33. Metformin did not statistically change expression of the last two molecules in any cell line ( Fig. 1A and Supplemental Fig. 1). In contrast, it increased expression of ULBP-1 and of the integrin ligand ICAM-1 in all of them and of MICA/B in OCI-AML3 cells ( Fig. 1A and Supplemental Fig. 1). Of note, metformin was cytostatic, but not cytotoxic, on these leukemic cells at the selected dose (Fig. 1B). We next tested lower doses of metformin in the cell line that better respond to treatment, i.e. OCI-AML3 (Fig. 1C). Whereas 1 and 2 mM of metformin gave similar effects, concentrations under 1 mM lacked effect. Therefore, for future experiments we used 2 mM.
We challenged these results in another set of hematological cancer cells: the multiple myeloma (MM) cell lines MM1.S (wtp53) and U266 (mtp53) and the pro-monocytic myeloid leukemia cell U937 (nullp53) 20,29,32 . We treated them with metformin and analyzed the previously described antigens, but using the MM marker CD138 instead of CD33 for the MM cell lines. Metformin increased all markers in MM1.S cells, but only MICA/B in U266 and CD33 in U937 cells (Fig. 1D). Hence, in both cell line cohorts, we observed a higher response to metformin in cells expressing wtp53.
Overexpression of antiapoptotic Bcl-2 family members affects metformin effects on leukemic cells. We next investigated the effect of metformin on cells that display chemoresistant mechanisms. Bcl-X L overexpression is associated with bad prognosis and chemoresistance in several types of tumors 33,34 . Mcl-1 overexpression is linked to chemoresistance in multiple myeloma 35,36 . Both proteins are usually overexpressed in tumor cells from chemoresistant patients. Figure 2 and supplemental Fig. 2 showed that metformin induced expression of ICAM-1, ULBP1 and MICA/B on the B-cell chronic lymphocytic leukemia (B-CLL) cell line MEC1 (mutp53). But, it did not increase expression of HLA, the B cell marker CD20 or the death receptors Fas, DR4 or DR5. In cells overexpressing Bcl-X L , metformin only increased expression of the above-mentioned death receptors. Metformin failed to change expression of any of these antigens in cells overexpressing Mcl-1.

Metformin sensitizes leukemic cells to cytotoxic lymphocytes (CL).
We next analyzed metformin effect on CL-mediated tumor cell killing. We generated expanded and activated NK cells (eNK) from umbilical cord blood (UCB) as previously described [24][25][26] and used them against the AML cell lines described in Fig. 1A. As described in Fig. 1, metformin barely affected cell survival but eNK killed between 30 and 60% of target cells, depending on the cell line ( Fig. 3A and supplemental Fig. 3). Target cell pretreatment with metformin doubled eNK-mediated killing in all cell lines. Similarly, in the cell lines described in Fig. 1C, metformin increased killing of MM1.S cells, but not the other cell lines lacking wtp53 (Fig. 3B). We generated expanded cytotoxic T lymphocytes (eCTL) as previously described 37 . Metformin sensitized MM1.S cells to them, but not U266 or U937 (Fig. 3B). This is in agreement with our previous results showing that some of the metformin effects in these leukemic cells depend on p53 status 20 .  were treated with 2 mM metformin for 3 days and the expression of the above described markers plus the MM cell line marker CD138 were analyzed by FACs. Data represent the Mean Fluorescence Intensity (MFI) levels compared to control, non-treated, cells. The bar graphs represent means ± SD of 3 independent experiments; *p < 0.05, **p < 0.01, ***p < 0.005 student t-test compare to control cells.

Overexpression of anti-apoptotic Bcl-2 members blocks metformin-induced sensitization of leukemic cells.
We treated with metformin the cell lines described in Fig. 2 and then tested them to eNK and eCTL cytotoxicity. Metformin clearly sensitized Mec1wt cells to both eNK and eCTL cytotoxicity while it did not have any significant effect on cells overexpressing Bcl-x L or Mcl-1, although they were somewhat sensitive to their cytotoxicity, as previously described 24 (Fig. 4).

Metformin sensitizes primary leukemic cells to NK cells.
We then treated with metformin the primary cells from a B-cell lymphoma patient (BCL-P2) and observed that it significantly increased expression of ULBP1 and ICAM-1 and tended to increase MICA/B and CD20 expression ( Fig. 5A and Supplemental Fig. 4).
In contrast, we did not find a significant effect on HLA surface expression. eNK cells killed BCL-P2 cells and metformin significantly increased killing ( Fig. 5B and Supplemental Fig. 5).
Metformin mainly favors leukemic cell killing through an increase in LFA-1/ICAM-1 interaction. The increase in the expression of stress ligands suggested that metformin-induced sensitization of tumor cells could rely on the interaction of NKG2D with its ligands. Hence, we incubated eNK cells with an anti-NKG2D mAb that blocks the binding of NKG2D with NG2DL before putting them in contact with OCI-AML3 and NB4 cells (Fig. 6A). www.nature.com/scientificreports/ Although our eNK express NKG2D levels similar to non-activated NK cells 25 , we did not observe any effect on cell killing in control or metformin-treated cells. In contrast, when we used a construct (D1D2) that binds to LFA-1 and blocks its binding to ICAM-1 24 , we observed a significant reduction in cell killing in metformintreated cells in three leukemic cell types, including primary BCL-P2 cells (Fig. 6B). We next used another set of antibodies to block the interaction of these receptors with their ligands (Fig. 6C). Blocking ULBP1or MICA/B with specific antibodies did not affect eNK killing. In contrast, when we used an antibody against integrin β2, a component of LFA-1, we observed a significant reduction in cell death in both, control and metformin-treated cells. The strong requirement for LFA1/ICAM1 interaction was also observed in NB4 cells (Fig. 6D). In fact, the decrease on tumor cell recognition by blocking ICAM-1/LFA-1 interaction was strongly dependent on the concentration used.
Metformin delays growth of a fast-growing lymphoma in vivo in the presence of an anti-CD20 mAb. We next investigated the effect of metformin on lymphoma cell growth in vivo. We used BCL-P2 cells that quickly induce tumors when engrafted subcutaneously at 5 million cells per mice in immunosuppressed NSG mice 24 . Metformin alone did not affect tumor development while adoptive transfer of eNK cells barely affected it (Fig. 7A). Consequently, mice survival was not affected by these treatments (Fig. 7B). Tumor growth was delayed by the anti-CD20 mAb rituximab which also increased mice survival (Fig. 7). eNK cell combination with this effective anti-CD20 mAb did not statistically further decrease tumor growth or mice survival (Fig. 7). In contrast, the combinatory treatment of metformin, eNK and anti-CD20 mAb significantly reduced tumor growth with respect to the double combinatory treatments (Fig. 7A), i.e. metformin + eNK and antiCD20 + eNK, and also increased mice survival (Fig. 7B).
Metformin tends to delay growth of a slow-growing lymphoma in vivo. We next investigated metformin effects on the growth of LNH1 cells, which are issued from a diffuse large B-cell lymphoma (DLBCL) patient. These cells grow slower than BCL-P2 when subcutaneously engrafted at 10 million cells per mice 24 . In these experimental conditions eNK cells alone failed to affect tumor growth (Fig. 7C). However, metformin treatment tends to delay tumor development although combination with eNK cells did not have any additional effect (Fig. 7D). The anti-CD20 mAb totally abrogated the growth of these cells (data not shown), making irrelevant the cotreatment with Metformin and/or eNK.

Discussion
Metformin has an undoubtedly anticancer value 4,5 , but its mechanism of action is complex and not fully understood. Its immunomodulatory effect should be taken into account to optimize its clinical use 4 . This could be more relevant on hematological cancers in which effector immune cells are in close contact to malignant cells because they niche in the same tissues. Several clinical approaches such as CAR T 38 or CAR NK 39 cells have shown a relevant success to treat blood-borne cancers. However, roughly 50% of patients undergo relapse and new clinical protocols are required to improve the clinical activity of these genetically-modified cytotoxic lymphocytes. Expression of the PD-1 immune checkpoint ligand PD-L1 in target cells efficiently blocks CTL killing. Metformin can promote PD-L1 phosphorylation and degradation 14 . This could increase cytotoxic lymphocyte-mediated tumor cell death. The PD-1/PD-L1 immune checkpoint is less relevant in NK-mediated killing and, additionally, we have observed that our eNK cells express low levels of PD-1 26 . Moreover, in MM samples pembrolizumab does not increase tumor killing by our eNK 26 . Hence, we do not believe that this immune checkpoint plays a main role in our observations. Initially, we believed that metformin sensitizes leukemic cells to NK by induction of stress molecules in target cells. We certainly reproduce previous findings 15 showing that metformin increases expression of stress ligands in multiple cell lines. Our eNK express similar NKG2D levels to naïve cells 25 , but blocking NKG2D or MICA/B and ULBP1 does not decrease metformin effect. However, we have used activated NK cells. Hence, it is possible that naïve NK cells can better sense NKG2DL upregulation on tumor targets. If naïve or activated cells are more representative of the situation in the tumor microenvironment is unclear. It is believed that once in the tumor microenvironment NK cells should be activated by the targets and/or the pro-inflammatory cytokines. In contrast, for tumor cells outside of this environment, e.g. metastatic cells, it could be the opposite. Moreover, in patients the effect of metformin can be different than in our mouse models. In summary, we cannot exclude a major role of metformin in patients.
The clinical results regarding the use of metformin in cancer treatment are inconclusive 4 . In fact, the timing, dose and duration of treatment and the heterogeneity of the patients enrolled make difficult to compare the   4 . In vivo metformin inhibits gluconeogenesis from redox-dependent substrates 5 . In vitro and at higher doses, it increases expression of MICA 15 . But, most importantly, metformin inhibits mitochondrial complex I, fructose 1,6-bisphosphatase and CREB and activates AMPK; 5 . All these effects can target multiple downstream signaling pathways and the most studied is AMPK downstream signaling 5 . AMPK promotes PD-L1 protein degradation 14 , activates p53 17-20 , inhibits lipid synthesis 40 and regulates mTORC1 axis 41 . The role of each of these pathways in ICAM-1 and/or NKG2DL expression is unknown. It is also largely possible that several pathways play different roles, finally leading to expression of those CL ligands and tumor sensitization to CLs. It would be extremely interesting to fully unveil this pathway to better design new immunotherapies to improve the clinical activity of allogeneic CLs 11,12 . Interestingly, the cell lines U266 and U937, which are not sensitized to cytotoxic lymphocytes by metformin, do not upregulate ICAM-1 after metformin treatment. Moreover, Mec-1 cells overexpressing the Bcl-2 family members Bcl-X L or Mcl-1 also failed to upregulate ICAM-1 and are resistant to metformin-induced sensitization. Finally, blocking ICAM-1/LFA-1 interaction strongly decreases metformin effect. Hence, our results support that metformin mediates upregulation of integrin ligands, i.e. ICAM-1, which allows the effective binding and activity of cytotoxic lymphocytes on tumor cells 42 . Cytotoxic lymphocytes travel and roll over multiple substrates and sample different cellular environments. They need to be able to stop, process the stimulating signals of transient cellular contacts and move on 42 . However, when these stimulatory signals are strong enough they should stop, form an adherent junction, stable and strong with the target cell and eliminate it. Metformin-induced ICAM-1 expression on target cells should favor that NK cells establish close contact with them. If the stimulating signals are strong enough, NK cells will proceed to kill the target 30,31 . This has been described in in vitro activated NK cells using a similar expansion protocol as the one we have used here 31 . Interestingly, metformin decreases ICAM-1 expression in non-transformed cells [43][44][45] , including polycystic ovary syndrome (PCOS) subjects 46 and patients with T2D 47 . Hence, it should not favor cytotoxic lymphocyte binding to healthy tissue that could induce autoimmunity.
We did not find any effect in our system. However, we used activated NK cells which are those that should infiltrate the tumor. Hence, it is possible that naïve NK cells can better sense NKG2DL upregulation. Moreover, in patients the effect can be different than in mouse models. We have included these comments in the discussion section. www.nature.com/scientificreports/ Metformin only slightly delays tumor growth and does not improve mice survival either alone or together with eNK. In contrast, it improves survival and decreases tumor growth in the presence of an anti-CD20 mAb and eNK. In our mouse model, eNK are in contact with metformin during their antitumor function, which is not the case of the in vitro experiments. It is well-known that cytotoxic lymphocytes require a glycolytic metabolism for their maximal activity and educated NK cells display a high glycolytic rate 48 , which is essential for their antitumor function 49 . Hence, metformin could affect NK cell metabolism in vivo impairing their function. The efficient anti-CD20 mAb could overpass this effect by inducing maximal NK cell activation through the legation of CD16, which recognizes the Fc moiety of Abs. In this context, metformin improves outcome of DLBCL patients at least partially by sensitizing cells to the anti-CD20 rituximab 50 . However, we do not know if our results apply to naïve NK cells. eNK have a relatively high basal cytotoxicity compared to naïve NK cells and show a mature phenotype [24][25][26] . The more mature NK subsets, which possess higher cytotoxic potential, show the highest activation by LFA-1 30 . Therefore, it is possible that metformin-induced NK sensitization is specific of already activated cytotoxic lymphocytes, something that perhaps is uncommon in cancer patients that carry impaired NK activity 22 .
The concentration of metformin in plasma in T2D patients is around 30 µm 51 . A daily dose of metformin could reach 2,000 mg a day 4 , which is basically 15 mmol and could perhaps locally give higher metformin concentrations. For example, it can reach 140 µm in liver 4 . The intratumor concentration of metformin is difficult to www.nature.com/scientificreports/ evaluate, but it accumulates in ovarian cancer patient biopsies 52 . Moreover, metformin concentration reaches on average 0.41 mmol/kg in the colon of colorectal cancer patients daily treated with an oral low dose of 250 mg/d of metformin, with some patients reaching 1.87 mmol/kg 53,54 . In addition, and remarkably, metformin effects are higher at low glucose concentration (1 mM), which are found in tumor microenvironment 52 . This is probably the case of AML and MM patient bone marrow, which is highly infiltrated by glucose-consuming tumor cells. Hence, metformin concentration in the tumor can be significantly higher than in plasma and/or it can produce specific effects at lower doses in the tumor microenvironment. The antitumor function of metformin is very heterogenous. As previously described, the large variety on age and type of disease of the patients engaged in the clinical trials could explain this fact. We show here that p53 status and/or overexpression of Bcl-x L or Mcl-1 could make tumor cells resistant to the metformin-induced sensitization to cytotoxic lymphocytes. Moreover, the activity of NK cells is largely impaired in leukemia patients 22 and our results suggest that metformin could give better clinical results in patients with sufficient NK activity. Hence, we believe that metformin together with allogeneic activated NK cells could be a future relevant treatment. The ongoing clinical studies of metformin in nondiabetic, cancer, patients will soon show the effect of metformin in several clinical contexts and perhaps support the use of such co-treatment.

Materials and methods
Ethical statement. Experiments

In vivo experiments.
In vivo experiments were carried out using 6-8 weeks/old male NOD scid gamma (NSG) mice, which were bred and housed in pathogen-free conditions in the animal facility of the European Institute of Oncology-Italian Foundation for Cancer Research (FIRC), Institute of Molecular Oncology (Milan, Italy). For engraftment of human cells, mice were subcutaneously engrafted with 5 × 10 6 BCL-P2 or 10 × 10 6 LNH1 primary tumor cells derived from a BCL (P2) patient or a diffuse large B-cell lymphoma (DLBCL) patient (LNH1). Metformin was given ad libitum through the drinking water at 2 mg/ml. At day 4, we engrafted 10 (BCL-P2) or 10 (LNH1) million e-NK cells and at day 6, mice were treated i.p. with the anti-CD20 rituximab (in saline medium) 3 mg/kg once a week × 3 weeks; or with a combination of both. Tumor growth was monitored at least once a week using a digital calliper, and tumor volume was calculated according to the formula: L × W 2 /2 = mm 3 , where W represents the width and L the length of the tumor mass.
Counting and determination of cell viability. After treatment, hematopoietic cells were stained with Muse Count and Viability Reagent, and then analyzed on the Muse Cell Analyzer (Millipore) to identify cell number and survival 28 . Statistical analysis. The statistical analysis of the difference between means was performed using the 2way ANOVA test or the Student't test using the software Prism9 from GraphPad Software, LLC. The results are given as the confidence interval (*:p < 0.05, **:p < 0.01, ***:p < 0.005).