Systemic short chain fatty acids limit antitumor effect of CTLA-4 blockade in hosts with cancer

Gut microbiota composition influences the clinical benefit of immune checkpoints in patients with advanced cancer but mechanisms underlying this relationship remain unclear. Molecular mechanism whereby gut microbiota influences immune responses is mainly assigned to gut microbial metabolites. Short-chain fatty acids (SCFA) are produced in large amounts in the colon through bacterial fermentation of dietary fiber. We evaluate in mice and in patients treated with anti-CTLA-4 blocking mAbs whether SCFA levels is related to clinical outcome. High blood butyrate and propionate levels are associated with resistance to CTLA-4 blockade and higher proportion of Treg cells. In mice, butyrate restrains anti-CTLA-4-induced up-regulation of CD80/CD86 on dendritic cells and ICOS on T cells, accumulation of tumor-specific T cells and memory T cells. In patients, high blood butyrate levels moderate ipilimumab-induced accumulation of memory and ICOS + CD4 + T cells and IL-2 impregnation. Altogether, these results suggest that SCFA limits anti-CTLA-4 activity. The gut microbiota has been reported to regulate the efficacy of cancer therapy. Here, the authors show that short-chain fatty acids, which are generated through bacterial fermentation, increases immune tolerance leading to resistance to anti-CTLA-4 immunotherapy in mice and patients with metastatic melanoma.

I n the past ten years therapeutic strategies in oncology have been revolutionized by the use of immune checkpoint blocking antibodies. In clinical practice, CTLA-4 blockade using the therapeutic monoclonal antibody (mAb) ipilimumab (anti-CTLA-4) was approved as standalone or in combo with nivolumab (anti-PD-1) in patients with metastatic melanoma (MM) [1][2][3][4] . Approximately 20 and 60% of patients with MM have an objective clinical response with ipilimumab and with ipilimumab/ nivolumab combination respectively. Recently, studies suggested that gut microbiota composition was associated with antitumor efficacy in patients with MM treated with anti-CTLA-4 5,6 and anti-PD-1 7 mAb. The composition of gut microbiota also seems to be associated with the risk to develop anti-CTLA-4-induced colitis in patient with MM 6,8 . These recent studies highlight specific bacteria associated with clinical response and/or toxicities. Especially, baseline gut microbiota enriched with Faecalibacterium and other Firmicutes was associated with beneficial clinical response to ipilimumab, anti-PD-1, and ipilimumab/anti-PD-1 therapy in melanoma patients 6,7,9 . Based on these independent works, it appears that Faecalibacterium might represent a major feature associated with clinical response in MM patients treated with immune checkpoints. However, scarce findings explain how the gut microbiota composition could have an impact on a distant tumor lesion. In mice, anti-CTLA-4 blocking mAb was shown to induce a dysbiosis favoring the translocation of commensal bacteria that might allow IL-12-secretion by dendritic cells (DCs) as well as the priming of commensal-specific Th1 cells that could migrate to the tumor and recognize tumor cells due to antigen mimicry 5 . Another mechanism was described in mice treated with anti-PD-L1, where specific bacteria (i.e., Bifidobacterium) was associated with an increase DC maturation leading to enhance CD8 + T cell priming and accumulation in the tumor beds under PD-L1 blockade conditions 10 . In patients with MM treated with anti-PD-1, enriched Faecalibacterium was linked to higher CD8 + T cell tumor infiltrate 7 . In addition to the direct effect of commensal bacteria on immune system, it is well known that some bacterial groups produce metabolites that have also immune properties 11 . These de novo synthesized metabolites include short-chain fatty acid (SCFA), mainly acetate (C2), propionate (C3) and butyrate (C4). SCFA mediate several functions especially providing energy to intestinal epithelial cells (IEC) 12,13 . SCFA also play a pivotal role on immune modulation 11 . Butyrate is well known to exert systemic anti-inflammatory activities by affecting immune cell migration, adhesion, cytokine expression as well as affecting cellular processes such as proliferation, activation, and apoptosis 14 . Considering previous results on the association between gut microbiota composition and clinical response and the effect of SCFA on the immune system, even at distant site, we hypothesized that anti-cancer response due to anti-CTLA-4 blockade may be influenced by systemic microbial SCFA. In this study, we demonstrate that microbial systemic SCFA (butyrate and propionate) influence anti-CTLA-4 anti-tumor effect in mice models and in patients with MM and treated with ipilimumab.

Results
Microbiota composition and clinical outcomes in patients. As previously analyzed and replicated with sequencing technology, baseline microbiota enriched in Faecalibacterium prausnitzii (F. prausnitzii) and other Firmicutes was associated with better outcome in a French cohort of 26 MM patients treated with ipilimumab 6 . In the present study, among the fifty MM patients included, 16S rDNA analyses were performed on 38 fecal samples at baseline (V 1 ) (Supplementary information and Supplementary  Tables 1, 2 and 3). We analysed the main genera composition (Fig. 1a). Genera linked to long-term clinical benefit (LTB; progression free survival > 6 months) were Faecalibacterium and Gemminger (Fig. 1b). High proportions of Bacteroides could be found in patients with poor clinical benefit but no statistical significance was reached compared to patients with LTB (Fig. 1b). Other genera were not associated with clinical outcome (Fig. 1b). Altogether, Faecalibacterium might represent a good surrogate marker of LTB. Considering ipilimumab-induced colitis, a tendency for higher proportions of Faecalibacterium, Lachnospiracae and Gemminger, and lower proportion of Prevotella were observed in patients that develop ipilimumab-related colitis even though not significant ( Supplementary Fig. 1). High relative abundance of Faecalibacterium at baseline was linked to overall survival (OS) over than 18 months (Fig. 1c). Kaplan-Meier analyses of patients classified into two groups according to median value of the abundance of Faecalibacterium, showed that high baseline Faecalibacterium was associated with longer progression free survival (PFS) (Fig. 1d). Note that Faecalibacterium, Gemminger 6 , Roseburia as well as Bifidobacterium genera described in another study as associated with clinical efficacy after anti-PD-L1 treatment 10 were positively correlated whereas Bacteroides genera was inversely correlated to Faecalibacterium ( Supplementary Fig. 2).
Then, we applied quantitative PCR (Q-PCR) analyses on baseline feces in the cohort for F. prausnitzii, Bacteroides fragilis (B. fragilis) and Escherichia coli (E. coli) (Supplementary Table 1 and 3). Quantitative-PCR and 16S rDNA analysis data were highly correlated ( Fig. 1e and Supplementary Fig. 3). All patients with an OS longer than 18 months showed higher proportions of F. prausnitzii contrary to B. fragilis (Fig. 2a). Interestingly, abundance of F. prausnitzii and bacterial load were correlated together ( Supplementary Fig. 4) and were linked to improve PFS in patients (Fig. 2b). No such correlation and difference could be observed for B. fragilis 5 or E. coli assessed as controls (Fig. 2a, Supplementary Fig. 5). Altogether, these data indicated that microbiota composition and particularly enrichment with F. prausnitzii using Q-PCR analysis could be linked to clinical outcome in patients treated with ipilimumab as previously suggested using 16S rRNA gene sequencing 6 .
Serum SCFA concentrations and clinical outcomes in patients. The gut microbiota is able to influence immune responses mainly through gut microbiota metabolites such as SCFA (acetate, propionate and butyrate), indoles derivatives, polyamines and others, produced in very large amounts in the colon through bacterial fermentation of dietary fiber 15,16 . Our hypothesis was driven by previous results concerning the association between the response to ipilimumab and the gut microbiota enriched in Faecalibacterium and other Firmicutes 6 , known to be as butyrate producers. These data led us to assume that SCFA might play a mandatory role in ipilimumab efficacy. We determined the concentration as well as proportion, in stools and serum, of acetate, propionate and butyrate in the French cohort of patients with MM treated with ipilimumab ( Supplementary Fig. 6). In stools, quantity and ratio between SCFA (3:1:1 for acetate:propionate: butyrate) were as previously described 17 . Correlations between serum and stool SCFA concentrations were observed for butyrate (r = 0.50; p = 0.0009) and to lower extent for propionate (r = 0.35; p = 0.035) ( Supplementary Fig. 7). Surprisingly, Firmicutes phylum as well as F. prausnitzii were inversely linked with serum concentration of butyrate (Fig. 2c, d and e). No such association could be observed for B. fragilis or E. coli ( Supplementary Fig. 8). We determined SCFA serum levels in an independent Italian cohort of patients with MM and treated with ipilimumab (Supplementary Table 1 and Supplementary Fig. 9). Serum butyrate and propionate were correlated together but not with acetate in both cohorts ( Supplementary Fig. 10). Kaplan Meier analyses of patients classified into two groups according to median value of serum SCFA concentrations, showed that low baseline butyrate was associated with longer PFS in the French cohort (Fig. 2f, p = 0.0054) and low baseline propionate was associated with longer PFS in the Italian cohort (Fig. 2g, p = 0.0010). Serum butyrate concentration tended to be associated with PFS in the Italian cohort (Fig. 2g, p = 0.0688). No such association was observed for serum acetate in both cohorts in Kaplan-Meier analyses (Supplementary Fig. 10). Moreover, serum butyrate and propionate were inversely correlated to PFS and OS ( Fig. 2h and i) while no association could be found for acetate. All data from both cohorts were pooled (n = 85), both butyrate and propionate were negatively correlated to clinical outcomes ( Supplementary Fig. 10d). Moreover, Kaplan Meier analyses of patients classified into two groups according to median value of serum SCFA concentrations, showed that both low baseline butyrate and low baseline propionate was associated with longer PFS in the pooled cohort ( Supplementary Fig. 10e, p = 0.0015 and p = 0.0029 respectively). These results suggest that serum butyrate and propionate concentrations could represent an indirect systemic marker of the microbiota composition associated with clinical outcomes in patients treated with ipilimumab.
Butyrate reduces efficacy of CTLA-4 blockade in mice. To better understand if SCFA may have a direct role on anti-CTLA-4 clinical efficacy, sodium butyrate was given to mice in drinking water at the concentrations of 100 mM and was compared to mice drinking pH-matched water, two weeks before tumor inoculation and all along experiments. Mice also received i.p. injections of anti-CTLA-4 or its isotype control (IgG2b) (Fig. 3a and Supplementary information). We measured SCFA concentrations in mice serum to confirm higher levels of butyrate in supplemented mice. Despite some heterogeneity due to the variability of water intake between mice, concentrations of    . Each patient's microbiota is presented in the graph (n = 38). 0-6 M: overall survival ranging from 0 to 6 months (n = 4), >6-18 M: overall survival ranging from 6 to 18 months (n = 23), >18 M: overall survival >18 months (n = 11). Two-Way ANOVA followed by Tukey's tests for multiple comparisons was performed. p-values are indicated on each graph; ns, not significant. b Kaplan-Meier survival curves of PFS of patients classified into two groups according to median of F. prausnitzii/ng DNA (Low vs high F. prau; left panel) and 16S BACTERIA/ ng DNA (low vs high Bacteria; right panel). Date were performed using Q-PCR analyses on feces at baseline (n = 38), Log-rank (mantel-cox) test was used c Spearman correlation between the relative abundance of Firmicutes assessed by 16S DNA analysis (% of total reads NGS) and serum butyrate at baseline (n = 33). d Q-PCR analyses on feces at baseline of F. prausnitzii (16S F.prau/ng DNA) according to the median concentrations of serum acetate, propionate and butyrate (n = 33). e As in (d) but proportion (n = 33). Each dot represents a mean quantity (assessed in duplicate for each sample) for one patient. Two-tailed Mann-Whitney tests were used (d, e). p-values are indicated on each graph; ns, not significant. Kaplan-Meier survival curves of PFS according to the median value of serum propionate (left panels) and butyrate (right panels) concentrations in French (f, n = 40) and Italian (g, n = 45) cohort, Log-rank (mantel-cox) test was used. h Pearson correlations between serum concentrations of acetate, butyrate and propionate and PFS (n = 40) in French cohort; i As in (h) but for OS (n = 40) in French cohort. Each dot represents one patient. p and r are indicated on each graph. butyrate, as well as the proportion of butyrate among total SCFA, were higher in supplemented mice ( Supplementary Fig. 11). Conversely, the concentrations of acetate and propionate were lower in butyrate-supplemented mice ( Supplementary Fig. 11). Three tumors models were used (CT26, MC38 and MCA101 OVA ). In the CT26 tumor model, stabilization as complete regression of the tumor growth were observed in mice treated with anti-CTLA-4 without sodium butyrate (Fig. 3b) as well as longer survival (Fig. 3c). In butyrate groups, although CTLA-4 blockade could exert some anti-tumor effect (Fig. 3b, c), no significant reduction of the tumor growth was observed at day 20 (D20) compared to control group (Fig. 3d). Moreover a significant difference was observed between anti-CTLA-4+ water and anti-CTLA-4+ butyrate (Fig. 3d). In line with these results, Altogether these data provided evidence that butyrate could diminish anti-CTLA-4 anti-tumor efficacy in mice.
Butyrate limits anti-CTLA-4-induced DC maturation. Butyrate has been described to modulate immunity and inflammation 18 . Note that within SCFA, butyrate and propionate are often associated with induction of regulatory T cells (Treg) contrary to acetate which has been mainly incriminated in the modulation of B cells immunity [19][20][21][22] . We hypothesized that butyrate could interfere with anti-CTLA-4-induced immune modulation. In the MCA101 OVA model where antigen-specific T cell responses could be monitored, no modification of CTLA-4 expression levels within T cells was observed in tumor-draining lymph nodes (tDLN) in any groups (Supplementary Figs. 12 and 19). Butyrate was shown to prevent DC maturation in vitro 23,24 . In our mouse model, CTLA-4 blockade did not modify the proportion of DC in tDLN (Fig. 4a). However, it increased the proportion of DC expressing CD80 and the density of CD86 on the DC surface.
Modulations of CD80 and CD86 expression were not observed in mice treated with butyrate + anti-CTLA-4 ( Butyrate restrains anti-CTLA-4-induced OVA-specific T cells. We next evaluated OVA-specific CD8 + T cell induction in the tDLN of MCA101 OVA bearing mice. In vitro restimulation of tDLN with MHC class I (MHC-I; SIINFEKL) evidenced in mice that drank butyrate-free water and treated with anti-CTLA-4, a significant induction of IFNγ compared to mice treated with isotype control (Fig. 4g). No such difference could be observed in mice that drank sodium butyrate-supplemented water following anti-CTLA-4 therapy (Fig. 4g). In butyrate mice and not treated with anti-CTLA-4 (isotype group), a slight augmentation of IFNγ was observed following restimulation with 0.1 and 1 µg/mL of SIINFEKL compared to control without SIINFEKL even though not significant (Fig. 4g). This might be explained by the ability of butyrate to up-regulate MHC/antigen complexes through its histone deacetylase (HDAC) inhibitory activity 25,26 . In our hands, a clear up-regulation of MHC molecules was observed in vitro after butyrate treatment at the tumor cell surface in a dosedependent manner ( Supplementary Fig. 13). This capacity could favor an increase of the tumor immunogenicity in butyratesupplemented mice leading to a slight augmentation of IFNγ after ex vivo restimulation. Despite this, the anti-tumor efficacy of anti-CTLA-4 was reduced in mice supplemented with sodium butyrate. These results suggest that butyrate was able to restrain OVA-specific T cell responses following CTLA-4 blockade. Furthermore, previous studies suggested that increased peripheral memory CD4 + T cells 27 as well as ICOS expression on T cells 28 after anti-CTLA-4 treatment was associated with clinical benefit in MM patients. We thus analyzed the proportion of memory CD4 + T cells (CD44 hi CD62L − among CD4 + CD3 + CD45 + ) and ICOS expression in a longitudinal immune monitoring in the whole blood of mice. As previously described in patients, anti-CTLA-4 could induced accumulation of memory T cells as well as ICOS + T cells in the blood of treated mice (Supplementary Figs. 14 and 21). Interestingly, accumulation of memory T cells as well as ICOS induction after anti-CTLA-4 treatment was reduced in butyrate supplemented mice ( Supplementary Fig. 14). We also analyzed the proportion of memory CD4 + T cells in the tDLN and we observed an augmentation of memory CD4 + T cells following anti-CTLA-4 treatment compared to isotype group (Supplementary Figs. 15 and 21) that was not observed in butyrate + anti-CTLA-4 treated mice compared to butyrate + isotype group.
Patient's baseline immune markers and serum butyrate. Butyrate and propionate have been previously described as immunoregulatory metabolites capable of favouring regulatory T cells 19,20 . However no or scarce data exist in humans in vivo, on peripheral Treg cells and the link with systemic SCFA concentrations. We observed that serum butyrate and propionate before introduction of ipilimumab, were correlated with Treg and sCD25 ( Fig. 5a, b, c and d and Supplementary Fig. 22). Note that in butyrate supplemented mice accumulation of Treg cells in the blood and in the tDLN was also observed (Supplementary Figs. 16, 17 and 19). To test if micromolar (µM) concentrations of butyrate could increase proportion of Treg cells, we stimulated PBMC with concentrations close to those found in the serum of patients plus or minus anti-CD3ε stimulation. As shown in Fig. 5e a significant increase of Treg was observed in culture after CD3 stimulation at 50 and 100 µM but not at lower concentration i.e 10 µM. This data seemed consistent with observations made in patients where an expansion of Treg cells was observed only at higher concentrations. SCFA were proposed as anti-inflammatory metabolites through an inhibition of NF-kappaB 29,30 . In our cohort of patients we found that higher concentrations of butyrate and propionate were associated to higher concentrations of inflammatory proteins such as TNFα and MCP-1 ( Fig. 5f and g). These results seemed in discordance with previous studies. However, in most of published studies demonstrating the antiinflammatory properties of SCFA, millimolar concentrations are used in in vitro assays 29,30 . In these different studies, it clearly appeared that NF-KappaB inhibition was dose and time dependent without any effect observed at low concentrations such as concentrations observed in the serum of our patients 30 . Using concentrations close to those found in the serum of patients no inhibition of the production of inflammatory proteins was observed (Fig. 5h) confirming that at these µM concentrations no inhibition of NF-kappaB might be induced. However it doesn't mean that these µM concentrations could induce inflammation. Indeed, we believe that the concentration of butyrate (or propionate) found in the stool and therefore at the periphery is greater when these metabolites are less locally used in the gut thus their action on the intestinal barrier might be limited. To test this hypothesis we measured zonulin a marker of intestinal permeability 31,32 , in the serum of patients. Zonulin was correlated with stools' concentrations of propionate and butyrate (Fig. 5i) suggesting that higher concentrations of butyrate and propionate remaining in the stool of patients were related to an increased intestinal permeability. Since increased intestinal permeability is also related to inflammation, we assessed correlation between zonulin and serum inflammatory proteins; we found that inflammatory proteins such as TNFα and MCP-1 were correlated with zonulin (Fig. 5j). In conclusion, we believe that the link between measurable SCFA and peripheral inflammation was not a direct cause-and-effect relationship, but was rather a refection of intestinal permeability.
Immune blood markers during the course of ipilimumab. We assessed longitudinal modification of immune parameters in patients treated with ipilimumab according to serum butyrate concentration and clinical response. Butyrate levels were not modified during ipilimumab treatment as well as other SCFA ( Supplementary Fig. 6). We and others have previously documented that sCD25 and accumulation of CD4 + memory T cells were rapidly increased by anti-CTLA-4 treatment in patients that will finally have a long-term clinical benefit 6,27,33 . T-distributed stochastic neighbor embedding (t-SNE) analysis of longitudinal fresh whole blood immune monitoring of CD3 + T cells (V 1 versus V 2-3 , Fig. 6a, b) was realized in patients according to clinical response. Modification of t-SNE map between V 1 and V 2-3 was different in patients with long-term clinical benefit compared to patient with poor clinical benefit (Fig. 6c). Accumulation of memory CD4 + and CD8 + T cell populations was observed ( Fig. 6c and Supplementary Fig. 23) only in patients that benefit from treatment. Accordingly in patients with low butyrate concentrations at baseline, higher increase of memory CD4 + T cells was observed following ipilimumab compared to patients with high serum butyrate as demonstrated by post-ipilimumab to baseline ratio (Fig. 6d). In line with this observation, no or weak augmentation of sCD25, a marker of IL-2 impregnation, was observed during the course of ipilimumab in patients with high butyrate compared to those with low butyrate levels ( Fig. 6e) as well as ICOS induction on CD4 + T cells (Fig. 6f). Since butyrate inhibited anti-CTLA-4-induced immunity, we assume that butyrate might counter CD28 signaling. To address this specific issue, we performed in vitro experiments with escalating concentration of butyrate (0, 10, 50, and 100 µM). PBMC were stimulated in complete medium alone or with anti-CD3ε to mimic TCR engagement or anti-CD3ε + anti-CD28. This latest condition recapitulates what should be observed after ipilimumab where the blockade of CTLA-4 favors the engagement of CD28. As shown in Fig. 6, butyrate decreased sCD25 secretion (Fig. 6g) and ICOS induction (Fig. 6h) in a dose dependent manner only in the condition where both CD3 and CD28 were engaged. These data demonstrates that butyrate counters CD28 engagement/ signaling and are in lines with observations made in patients where high levels of butyrate was associated with low induction of ICOS and sCD25 after ipilimumab treatment ( Fig. 6e and f). In conclusion, a weak restoration of the CD28/B7 axis after CTLA-4 blockade is accompanied by a reduced IL-2 impregnation during the course of ipilimumab leading to low accumulation of memory and ICOS + T cell in MM patients with high serum butyrate levels.
Altogether, this study provides insights on how the composition of the gut-microbiota might influence clinical responses to ipilimumab at distant sites through gut-microbial metabolites recovered in the periphery.

Discussion
This work showed that systemic microbial SCFA modulate anti-CTLA-4-induced immune responses and its antitumor efficacy. SCFA, as butyrate and propionate have important metabolic functions and are crucial for intestinal health. In our mice models, serum butyrate was artificially increased without any modification of gut microbiota as demonstrated previously 19 . Modulating only this parameter in mice limited anti-CTLA-4 blockade efficacy. Butyrate has been described to modulate immunity and inflammation 18 . Note that within SCFA, butyrate is always associated with immune modulatory capacities (as for propionate) mainly through induction of Treg cells 21 . In mice, CTLA-4 blockade induces an up-regulation of CD80 and CD86 co-stimulatory molecules, which was abolished by oral administration of sodium butyrate. CTLA-4 acts through cell-intrinsic 34 and cell-extrinsic pathways 35 . Cell-intrinsic pathway involves the cytoplasmic domain of CTLA-4 resulting in negative signal transduction machinery in T cells. The cell extrinsic pathway evidenced in in vivo models 36  quantifications in the stool and zonulin in the serum from patients (n = 9, where both data were available) before ipilimumab treatment (V1). j Spearman correlation between zonulin and serum concentrations of TNFa (light gray squares) and MCP-1 (black squares) (n = 12, where both data were available). Each dot represents one patient; n (number of patients) are indicated on each graph otherwise stated in the legend. P-values are indicated on each graph; Two-tailed unpaired t test (a, b, f, g) and two-way RM ANOVA followed by Tukey's tests for multiple comparisons was used (e and h). ns, not significant.
CD28. This mechanism is specific of CTLA-4/CD28 ligands and operates in an antigen-dependent manner in vivo 35 . Butyrate was shown to prevent DC maturation in vitro 23,24 . In these studies, butyrate was shown to act directly on DC maturation through the modulation of expression of several genes indicating a potent anti-inflammatory activity. In our settings, adding butyrate in mice treated with IgG2b did not induce modification of CD86/ CD80 expression (in tumor-bearing mice). These data suggest that blocking the interaction between CTLA-4 and CD86/CD80 is mandatory to observe butyrate-induced regulation. CTLA-4 being blocked we can rule-out the possibility that butyrate could augment trans-endocytosis by T cells; butyrate might rather prevent DC maturation through a direct mechanism as suggested by in vitro studies 23,24 . In the MCA101 OVA tumor model, blocking CTLA-4 could increase antigen-specific T cell responses in the tDLN; addition of butyrate could limit these responses. Several studies observed that sodium butyrate may induce T cell anergy 37 or reduce T cell priming capability 38 . In these studies butyrate did not act directly on T cells but limited the capacity of DC to stimulate T cells. In our work, since butyrate abolished anti-CTLA-4-induced CD80/CD86 up-regulation in vivo, restoration of the B7/CD28 axis might remain insufficient to promote the expansion of memory T cells and antigenspecific T lymphocytes.
In line with mice data where butyrate was shown to regulate the size and function of the colonic Treg pool by up-regulating IL-10 19,39 as well as accelerating extrathymic peripheral differentiation of Treg cells 20,22 , an immune regulatory profile was associated with high level of serum butyrate before introduction of ipilimumab in patients. As in mice, longitudinal immune monitoring in patients with MM, provided evidence that high butyrate was associated with less accumulation of memory and less ICOS induction on T cells in hosts with cancer and treated with anti-CTLA-4. No or weak increase of sCD25 in patients with high butyrate levels during ipilimumab treatment could be evidenced. Finally in vitro stimulation of Human PBMC demonstrated that butyrate could counter CD28 signaling pathway. Altogether, these data suggested a poor restoration of CD28 signaling after CTLA-4 blockade in patients with high levels of blood butyrate. Since butyrate was able to inhibit ipilimumab-induced DC maturation and CD28 signaling pathway, restoration of CD80/CD86 through the use of TLR agonists or other DC maturating agents might not be sufficient to counter the deleterious effect of these metabolites. In line with this, CpG agonists could not overcome the influence of butyrate in mice treated with anti-CTLA-4 ( Supplementary Fig. 18). In consequence the specific modulation of butyrate/propionate signaling pathway might be mandatory to restore anti-CTLA-4 efficacy in  patients with high systemic levels of SCFA. SCFA are known to act also through specific receptors (GPR41, GPR43 and GPR109A) expressed on immune cells, the use of GPR specific inhibitors might be a better option. However this has to be demonstrated and will be part of future research. In this study patients with high butyrate concentration had higher concentration of inflammatory proteins in the serum. This data seemed contradictory with previous published works. However and as demonstrated, at µM concentrations no inhibition of inflammatory proteins production could be observed. This is in lines with previous studies where dose escalation were used; a minimal concentration of 3 mM (3000 µM) butyrate could induce a reproducible and significant diminution of inflammatory proteins such as TNFα and IL-6 30 .
The level of faecal SCFA is influenced by several factors such as the gut microbiota composition 40 , the transit time and the available substrata 41 and the gut integrity. Firmicutes and F. prausnitzii are well known for its butyrate producing capabilities. Faecal concentrations of SCFA measured in the stools (and serum) depend on both the production by microbiota and their consumption by bacteria (cross-feeding) and colonocytes 13 , playing a crucial role in proliferation, differentiation of epithelial cells and gut permeability 17 . Systemic concentrations of SCFA depend on both the production and absorption rates in the gut, acetate being absorbed in the highest amount in the liver and reaches the highest peripheral concentrations (19-160 μmol/l) compared to propionate (1-13 μmol/l) and butyrate (1-12 μmol/l) 17 . Peripheral concentrations measured in some patients with MM were clearly above these previously published values. Unfortunately, very few studies assessed SCFA in serum. Most data accumulated on SCFA production rely on metabolites production by bacteria in culture 42 , or metagenomics analysis and detection of pathways used by bacteria for SCFA formation 43,44 . Considering that these data do not measure the real concentrations, it appears difficult to conclude on the balance between production and consumption. In the current study, zonulin was correlated with stools' concentrations of propionate and butyrate suggesting that higher concentrations of butyrate and propionate remaining in the stool of patients were related to an increased intestinal permeability. This result suggests that high concentrations of SCFA in stools of patients might reflect a poor consumption by colonocytes and/or bacteria increasing gut permeability. Since increased intestinal permeability is also related to inflammation, it was not surprising to observe correlations between zonulin and systemic inflammatory proteins.
It has been shown that gut microbiota impacts on the level of expression of enzymes involved in SCFA metabolism. Particularly butyrate-producing microbes increase the level of acyl CoA dehydrogenase involved in β-oxidation 13 and colonic mitochondrial 3-hydroxy 3-methyl glutaryl CoA (mHMGCoA) synthase responsible for the biosynthesis of ketone bodies from butyrate 45 . On the contrary, absence of butyrate-producing microbes led to low levels of expression of these enzymes and low metabolism of butyrate 13,45 . Recently, a negative correlation between the abundance of fecal Firmicutes and fecal SCFA levels was observed in healthy volunteers 46 . Moreover in mice, a high-amylose maize starch butyrylated (HAMSB) diet seems to be associated with expansion of Bacteroides species as well as higher concentrations of systemic butyrate leading to resolution of type I diabetes in mice models through induction of immunosuppression 22 . In line with these studies, we brought evidence in patients that microbiota enriched with Faecalibacterium is negatively correlated with Bacteroides genera and associated with lower systemic butyrate levels.
Our study has limitations; in patients due to low number of patients we could not realize multivariate analysis including major clinical and biological parameters already known to be associated with PFS and OS. It pushed us to validate our first observation in an independent cohort of patients with MM and treated with ipilimumab. Due to the discovery nature of our work, one's cannot exclude that it won't be validated in larger cohorts of patients. Nowadays, ipilimumab is not used as a standalone treatment in many patients, thus first we will have to validate if SCFA are also associated to clinical outcomes in patients treated with anti-PD-1/PD-L1 since Faecalibacterium was also associated to clinical outcomes in MM patients 7 . In mice, we used butyrate supplementation in water; however the use of HAMSB diet as described elsewhere 22 could have been a more direct manner to incriminate metabolites produced by microbiota and response to anti-CTLA-4 blocking mAbs. Finally, we described immune parameters in mice and patients associated with high systemic levels of butyrate but we did not identified a specific pathway affected by butyrate that could be blocked in order to restore antitumor activity of anti-CTLA-4 blockade in hosts with high systemic butyrate.
In conclusion, systemic butyrate (and propionate) appeared to limit antitumor activity of anti-CTLA-4. This study provides a vision on the relationship between the composition of the gutmicrobiota and the clinical response to ipilimumab via microbiota-derived metabolites. Finally, the identification of a specific pathway responsible for butyrate/propionate-associated immune-modulatory effects should represent an exciting research area that could allow the discovery of targets to improve anti-CTLA-4 response rate in patients with MM.

Methods
Mice and cell lines. C57Bl/6 and BALB/c mice were purchased from Harlan Laboratories (Gannat, France) between 8 and 14 weeks of age. Mouse colon carcinoma MC38 and CT26 and fibrosarcoma MCA101 OVA cell lines were cultured in complete medium (RPMI 1640 medium supplemented with 1 mM sodium/pyruvate, 1 mM of non-essential amino acids (NEAA), 100 IU/mL penicillin/streptomycin, 2 mM L-Glutamine and 10% fetal bovin serum (FBS, Life Technologies)). For the MCA101 OVA cell line RPMI complete medium was supplemented with 1 mg/ml hygromycin B (Thermofisher).
Choice of syngeneic tumor models in mice. CT26, MC38 were demonstrated as partial responder models (CT26 > MC38) to immune checkpoints thus providing good models for combination treatment effect (here Butyrate + anti-CTLA-4) (https://www.crownbio.com/wp/murine-model-guide-immuno-oncology; 47 ). Note that melanoma preclinical models in mice, such as B16 models, are not responding to immunotherapy due to low MHC molecules expression, B16 appeared to be immunologically "silent" with low expression of immune cell type-specific genes thus did not appeared as a good preclinical model to test immune-checkpoints in mice 47 . The fibrosarcoma MCA OVA was used for its expression of OVA antigen thus allowing the "easy" monitoring of H-2Kb:OVAp (peptide sequence SIIN-FEKL)-specific T cell responses. Moreover in our hands MCA OVA was also a partial responder model after anti-CTLA-4 treatment.
Tumor inoculation in mice and antitumor treatments. For all mouse experiments in the butyrate group, mice were treated 2 weeks before tumor inoculation and all along the experiment with sterile sodium butyrate in the drinking water at the concentrations of 100 mM (pH-matched) and changed every week.
Five hundred thousand of CT26 tumor cells were inoculated subcutaneously (s. c) in the right flank of Balb/c mice. Five hundred thousand MCA101 OVA cells were inoculated s.c. in the right flank of C57Bl/6 mice. Five hundred thousand of MC38 cells were inoculated s.c. in the right flank of C57Bl/6 mice. Mice received intraperitoneally (i.p.) injections of 100 μg of anti-CTLA-4 (clone 9D9) or its isotype control control (IgG2b) at D7, D10 and D13. In the experiment with CpG agonists, s.c injections near to the tumor site were done with 30 μg of CpG (Invivogen) or ODN control (Invivogen) at D7, post tumor inoculation. Tumor growth was followed-up three times a week by measuring the length and width with a caliper and was calculated as follows: size (mm 2 ) = length x width. Mice were euthanized when the tumor size was ≥300 mm 2 or boundary points were reached according to the French and European laws and regulations for the use of mice for scientific purposes. To analyze draining lymph nodes immune population, mice were sacrificed 14 days post tumor inoculation and draining lymph nodes were removed. Single cells were obtained after mechanical disruption. Serum samples were collected at the same time and were stored at −80°C until further analysis of SCFA dosages. For longitudinal blood immune monitoring in whole blood of mice, tumor bearing mice (CT26 model) were bleed before the first anti-CTLA-4 injection (D6 post tumor injection) and 2 days after the third injection of anti-CTLA-4 (D15) in both PH-matched water and sodium butyrate drinking groups.
Ex vivo experiments in mice. Ex vivo restimulation of T cells from tDLN: mice were sacrificed 21 days after MCA101 OVA inoculation, tDLN were removed and after mechanic dissociation 2.10 5 mononuclear cells were stimulated with escalating doses of Ovalbumin CD8 epitope 257-264 (SIINFEKL; Genecust) at 0, 0.1, and 1 µg/ml. The supernatants were assayed at 48 h by ELISA for mouse IFN-γ (Becton Dickinson) according to manufacturer's instructions.
Patients. Eighty-five patients were prospectively enrolled in the French study (GOLD cohort) at Gustave Roussy Cancer Campus (Villejuif) between March 2013 and June 2015. Among them, 28 patients did never receive ipilimumab, four patients received only a single infusion of ipilimumab and died soon thereafter due to their disease and three patients were lost for the follow-up. They were therefore excluded from further analyses. The remaining fifty patients with MM (48/50) or metastatic prostate carcinoma (2/50) were followed-up for at least 6 months. Fifty-Italian patients with MM treated with ipilimumab were prospectively enrolled at Instituto Nazionale Tumori Fondazione G. Pascale (Napoli) between July 2014 and March 2016 respectively. Among them, five patients received only a single infusion of ipilimumab and were therefore excluded from further analyses. The remaining forty-five patients with MM were followed-up for at least 6 months. Patients were informed of the study and consented to participate. French patients had a prespecified clinical workup; feces and blood were collected at baseline (V 1 ), prior to each ipilimumab infusion (V 2 , V 3 , V 4 ). Patient's characteristics are described in Supplementary Table 1. In the GOLD cohort, all blood, serum and fecal samples used in this study are listed in Supplementary Table 2. In the Italian cohort only serum were collected at baseline for all patients. Ipilimumab was administered intravenously every 3 weeks, at a dose of 3 or 10 mg/kg and could be continued after V 4 , at a maintenance dose of one infusion every 12 weeks, in patients whose disease was controlled (response or stable disease). Peripheral blood mononuclear cells (PBMC) were isolated by Ficoll density gradient, frozen in Hyclone media (Thermo Scientific) and stored at −196°C for further in vitro experiments.
In vitro experiments in human PBMC. PBMC were isolated by Ficoll density gradient from four healthy donors and four MM patients before ipilimumab treatment and cryopreserved in liquid nitrogen. 1.5 × 10 5 thawed PBMC were stimulated, into 96-well flat bottom Maxisorp (Thermofisher, USA) with escalating concentrations of sodium butyrate (0, 10, 50 and 100 μM) in complete medium alone, with pre-coated anti-CD3ε alone (0.5 µg/ml; BD Pharmingen) or anti-CD3ε + soluble anti-CD28 (0.1 µg/ml; BD Pharmingen) during 4 days at 37°C, 5% CO 2 . At day 4, PBMC were collected to evaluate the percentage of Treg and ICOS expression among CD4 + T cells using flow cytometry analysis. At day 4, supernatants were collected and cryopreserved at −80°C until the day of quantification. sCD25, TNFα, MCP-1 and IL-10 were quantified using Bio-Plex® MAGPIXTM Multiplex assay (Millipore) according to manufacturer's instructions and plates were analyzed on a MAGPIX (MAGPX13337701, Luminex). Determination of cytokines concentrations was assessed with XPONENT software solutions.
Flow cytometry analysis. In mouse experiments, before staining, Fc receptors were blocked for 15 min at 4°C using anti-CD16/32 functional grade purified antibodies (eBioscience) at 0.5 µg/10 6 cells. For staining on PBMC, cells were incubated for 30 min at 4°C with antibodies for cell surface staining. For surface staining on whole blood, after Fc receptor blocking and surface staining, red cells were lysed with a mix of Versalyse lysing solution + Fixative solution (1000:25, Beckman Coulter) during 15 min at room temperature. Cells were then washed two times with PBS 1×. For FoxP3 staining, cells were fixed and permeabilized after cell surface staining according to the FoxP3 kit protocol (eBioscience). Antibody characteristics are described in Supplementary Table 4. Samples were acquired on a Gallios Cytometer (Beckman Coulter) and fluorescence was analyzed using Kaluza software (Beckman Coulter). DCs were defined among CD45 positive cells by lineage negative (CD3/CD19/NKp46) as classe II MHC (IAb + ) and CD11c-positive cells. In humans, immune monitoring on whole blood and on PBMC was performed as described previously 6 .
Short-chain fatty acid dosages. Quantitation of SCFA including acetate, butyrate and propionate from serum and fecal samples was performed using LC-MS/MS method adapted from Lu et al. 48 . Hydrochloric acid was added to serum samples (25 µL) and to faeces samples (20 mg) prior hydrazinquinolin (HQ) derivatization and before protein precipitation using acetonitrile. Organic phase (25 µL) was diluted in the mobile phase (475 µL) prior to injection for mass spectrometry analysis. SCFA free materials were used. The analyses were performed on a Acquity ® UPLC system (Waters, Milford, MA, USA) that included a binary solvent manager, a cooled auto-sampler and a column oven coupled with a Xevo ® TQ-MS (Waters, Milford, MA, USA). A Waters BEH-C18 (2.1 × 100 mm, 1.7 µm) from Waters (Milford, MA, USA) was used. The column was maintained at 40°C. The Waters Xevo ® TQ-MS mass spectrometer was interfaced to the Acquity ® UPLC system via an electrospray ionization (ESI) source used in positive mode. The system control and data acquisition were performed using MassLynx ® software (v4. Quantification of bacteria by quantitative PCR. In the French cohort, for each patient, 200 mg of feces were used. DNA extraction included a first step of mechanical lysis: samples were diluted in 400 µL of ASL buffer in presence of 0.5 mm glass beads. Lysis was performed at 30 Hz during 30 min with the help of a TissueLyser (Qiagen). Next, samples were centrifuged at 10000 g 12000 rpm during 4 min and 200 µL of the supernatant was further extracted using the QIAamp DNA Stool Mini Kit (Qiagen), according to the manufacturer's instructions. Extractions series were realized during five consecutive days (5 series), for each series, stool sample from the same patient (patient #30) was extracted, this sample confirmed the reproducibility of analyzes. All DNA samples were quantified using Qubit™ 3 Fluorometer and Qubit™ dsDNA BR Assay Kit (ThermoFisher) according to manufacturer instructions. For each sample dilutions were realized to test 10 ng of DNA per qPCR. All qPCR were performed using the TaqMan™ Gene Expression Master Mix kit (LifeTechnologies) in 25 μl final volume (4 μl of genomic DNA added per tube, 300 nM each primer and 100 to 200 nM of each probe, 10 3 copies of an internal amplification control (IAC). All samples were analyzed in duplicate. All negative, very weak or discordant results were verified by another series of qPCR (in duplicate). All primers and probes used in this study (all Bacteria, Faecalibacterium prausnitzii, Escherichia coli, Bacteroides fragilis and Internal Amplification Control (IAC)) are presented in Supplementary Table 3 and were described in supplementary bibliography [49][50][51][52] . Standard curves were performed using genomic DNA preparations: F. prausnitzii A2-165 (from 0.4 ng/Q-PCR (~1.2 × 10 5 16S targets) to 0.04 pg/Q-PCR (~120 16S targets); E. coli CIP 7621 (~1.2 × 10 6 16S targets) to 0.4 pg/Q-PCR (~120 16S targets)) and B. fragilis CIP 7716 (from 1.2 ng/Q-PCR (~1.2 × 10 6 16S targets) to 0.12 pg/Q-PCR (~120 16S targets). All primers, probes and IAC'sDNA were purchased from Sigma Life Science (Woolands, TX, USA). Amplifications were carried out using the following ramping profile: 1 cycle at 50°C for 1 min, 1 cycle at 95°C for 10 min, followed by 40 cycles of 95°C for 30 s, 60°C for 1 min. Excepted for the B. fragilis Q-PCR, the annealing was at 58°C for 1 min. Q-PCR data were normalized to the expression levels of total bacteria measured by the all bacterial Q-PCR with E. coli standard curve. Final analyses according to serum butyrate levels were realized on data from 33 patients for whom both serum butyrate concentrations and stool were available at V 1 .
Ion S5 torrent sequencing analysis. In the same French cohort, 5 ng of stool DNA genomic were amplified with Ion 16S metagenomics Kit (ThermoFisher Scientific) (part Number A26216) using 20 amplification cycles according to the manufacter's instructions (Manual MAN0010799). As recommended, E. coli DNA control (2 ng/PCR) was used in parallel. All library preparations were quantified with Universal Quantitation Kit (Part number A26217). Templating and enrichment for Ion Torrent sequencing were performed by using Ion Torrent CHEF (ThermoFisher Scientific) according to the manufacter's instructions (Manual MAN0010846). To optimize the 16S sequences read number, 11differents barcoded samples were used for sequencing in 530 Ship. Sequencing was performed on an Ion S5 sequencer (ThermoFisher Scientific), using 400-bp sequencing kits with 850 run flows. Base calling and run demultiplexing were performed by using Torrent_Suite Software, version 4.0.2, with default parameters for the General sequencing application. All sequencing data were exported in Ion Reporter Software, version 5.6 with Metagenomic workflow which detects the bacterial diversity from a metagenomics sample from Ion semiconductor reads from the Ion 16S Metagenomics Kit. The experimentally generated sequence data were compared to two libraries of representative reference sequences download from (1) Curated MicroSEQ (R) 16S reference Library (V2013.1) and (2) Curated Grengenes data base (V13.5).
Statistical analyses. Statistical analyses were performed using the Prism 7.0 (GraphPad) Software. All tests used are systematically indicated on graphs. The paired T test or Wilcoxon matched-pairs signed rank tests could be used to compare paired groups, while the unpaired T test with Welch's correction or Mann-Withney tests could be performed to compare two unmatched groups. Oneway ANOVA was used when more than two groups were compared. Two-way ANOVA multiple comparisons was used when two or more grouping variables were analyzed. Two-way RM ANOVA multiple comparisons was used when two or more grouping matched variables were analyzed. Correlations were assessed using the Spearman (don't assume that data are sampled from Gaussian distribution) or Pearson (assume that data are sampled from Gaussian distribution) correlation coefficient. Contingency could be determined using the Fisher's exact test. Unsupervised cell population analysis was performed using t-distributed stochastic neighbor embedding (t-SNE) algorithm using the online R software (version 3.5.0, cytofkit package). The "T cell 1" flow cytometry panel (CD95/CCR7/HLA-DR/ CD25/ICOS/ CD45RA/CD3/CD127/CD4/CD8); as previously described 53 was used, after setting the compensation matrix cells were manually gated using Kaluza Analysis (version 2.0) on size/granularity discrimination and then on the expression of CD3 and logicle transformation was applied to extracted data for all samples. t-SNE analyses was achieved on 8500 cells CD3 + for each sample, using T cell 1 panel markers for T cell differentiation status (CCR7/CD45RA/CD4/CD8). Supervised analysis was performed using Kaluza Flow Cytometry Software (Beckman Coulter). OS and progression-free survival (PFS) according to SCFA dosages were estimated using the Kaplan-Meier method and compared using the log-rank test. OS and PFS analyses were carried out using PROC lifetest of SAS software, Version 9.4 (SAS Institute, Inc., Cary, NC). All tests are two-tailed and were considered significant at a p value < 0.05. Exact p values were annotated on graphs otherwise ns or p > 0.05 was annotated for non-significant data.
Human and mouse studies approval. Immune monitoring in the blood and the serum of patients were approved by the Kremlin Bicêtre Hospital Ethics Committee (SC12-018; ID-RCB-2012-A01496-37) for the French cohort and by the internal ethics board of the Instituto Nazionale Tumori IRCCS Fondazione "G. Pascale" of Napoli (number of registry 33/17) for the Italian cohort. The Declaration of Helsinki protocols were followed for both cohorts. Patients provided their written informed consent to participate in these studies prior to inclusion in the study. Blood samples of healthy volunteers (HV) were obtained after inform consent signature and were collected according to a procedure validated by the CCPSL UNT-N°12/EFS/079. All animal experiments were carried out in accordance with French and European laws and regulations and approved by the French Animal Experimentation Ethics Committee n°26 (02004.02).
Reporting summary. Further information on research design is available in the Nature Research Reporting Summary linked to this article.

Data availability
Data such as human demographic data are not publicly available in accordance with the General Data Protection Regulation since it could compromise research participant privacy/consent. All other data that support the findings of this study are available from the corresponding author [N.C.] upon reasonable request. We have submitted all source data of this manuscript to Springer Nature Research Data Support, the submission number is: RDS-SPRN-00421 54 .