The microbiome-derived metabolite TMAO drives immune activation and boosts responses to immune checkpoint blockade in pancreatic cancer

Abstract

The composition of the gut microbiome can control innate and adaptive immunity and has emerged as a key regulator of tumor growth, especially in the context of immune checkpoint blockade (ICB) therapy. However, the underlying mechanisms for how the microbiome impacts tumor growth remain unclear. Pancreatic ductal adenocarcinoma (PDAC) tends to be refractory to therapy, including ICB. Using a non-targeted, LC-MS/MS based metabolomic screen, we identified a gut microbe-derived metabolite trimethylamine N-oxide (TMAO) that enhanced anti-tumor immunity to PDAC. Delivery of TMAO intraperitoneally or via a dietary choline supplement to orthotopic PDAC bearing mice reduced tumor growth and associated with an immunostimulatory tumor-associated macrophage (TAM) phenotype and activated effector T cell response in the tumor microenvironment. Mechanistically, TMAO potentiated type-I interferon (IFN) pathway and conferred anti-tumor effects in a type-I IFN dependent manner. Notably, delivering TMAO-primed macrophages intravenously produced similar anti-tumor effects. Combining TMAO with ICB (anti-PD1 and/or anti-Tim3) in a mouse model of PDAC significantly reduced tumor burden and improved survival beyond TMAO or ICB alone. Finally, the levels of bacteria containing CutC (an enzyme that generates trimethylamine, the TMAO precursor) correlated with long-term survival in PDAC patients and improved response to anti-PD1 in melanoma patients. Together, our study identifies the gut microbial metabolite TMAO as a driver of anti-tumor immunity and lays the groundwork for potential therapeutic strategies targeting TMAO.

One sentence summary

The microbial metabolite trimethylamine N-oxide (TMAO) relieves immunosuppression in the tumor microenvironment of pancreatic cancer.

Introduction

Pancreatic ductal adenocarcinoma (PDAC) is a highly lethal cancer with poor prognosis. Although pancreatic tumors exhibit prominent leukocyte infiltrates, immunotherapy has so far failed to improve clinical outcomes in patients with PDAC (1, 2). This is largely due to a highly immunosuppressive tumor microenvironment (TME), characterized by a dense fibrotic stroma, infiltrates of immunosuppressive cells including, most prominently, tumor associated macrophages (TAM) and myeloid derived suppressor cells (MDSC), and low numbers of cancer killing CD8⁺ T cells (3–6). These features of the TME suggest that immunotherapy responses can be improved through strategies that shift the TME from an immunosuppressive state to a more immune-activated state. However, such approaches have been lacking. Cancer progression and response to therapy can be shaped by the tumor and gut microbiomes, via their effects on innate and adaptive immunity (3, 7–13). Long-term PDAC survivors display a distinct tumor microbiome profile, and fecal transplant studies from human into mice show that bacterial communities from long-term PDAC survivors have anti-tumor effects (7). Other studies suggest that the gut abundance of Clostridiales or Bifidobacterium in patients with melanoma associates with improved response to anti-PD1, and fecal transplant from anti-PD1-responders has a clinical benefit in PD1-refractory patients (11, 14). Taken together, these results suggest that improving therapy response requires a better understanding of how the gut microbiome shapes immune responses in the TME.

The gut microbiome produces a variety of metabolites in the process of degrading dietary factors (15); some of these can trigger inflammatory responses and thus contribute to regulating immune function and disease processes (16, 17). One such metabolite, implicated in several diseases including cancer, is Trimethylamine N-oxide (TMAO) (17–19). The generation of TMAO is a two-step process. First, dietary choline, the dominant source of TMAO, is converted to trimethylamine (TMA) by the gut bacterial enzyme choline TMA lyase, which consists of the enzyme pair CutC (catalytic protein) and CutD (activating partner) (20, 21). TMA then enters the portal circulation and undergoes oxidation to TMAO in the liver (22). Importantly, generation of TMAO is completely dependent on TMA production by intestinal bacteria (23). TMAO was initially described as having a role in maintaining the structural and functional integrity of proteins (24). However, recent work strongly supports the idea that TMAO can induce inflammation and immune activation. For example, in models of vascular inflammation or alloreactive T cell responses, TMAO directly induces pro-inflammatory mediators including TNFα, NLRP3 inflammasome, mitochondrial ROS, and NF-κB, and it decreases anti-inflammatory regulators such as IL-10 (25, 26). Clinically, dysregulated TMAO levels are associated with some immune disorders (18, 19). Thus, various interventions have been proposed for either up- or down-regulating the generation of TMAO by the gut microbiome, e.g., supplementing dietary choline to elevate TMAO, or using TMA lyase inhibitors to decrease TMAO (27, 28).

In this study, we discovered that TMAO directly drove an immunostimulatory phenotype in macrophages, which supported effector T cell responses and reduced PDAC burden. Mechanistically, we demonstrated that the anti-tumor effects of TMAO required the type-I interferon (IFN) pathway. Interestingly, TMAO rendered PDAC responsive to immune checkpoint blockade (ICB). Finally, analysis of human data revealed that levels of CutC-containing bacteria correlated with improved survival and response to anti-PD1 in cancer patients. Collectively, our findings demonstrate that the gut microbe-derived metabolite TMAO has immunomodulatory effects and thus may be a therapeutic entry point for boosting anti-tumor immune responses, thereby rendering PDAC responsive to checkpoint immunotherapy.

Results

Gut microbiome derived TMAO drove immune activation in the PDAC TME and reduced tumor growth

To investigate how the gut microbiome influences PDAC growth and immune responses, we determined how alteration of the gut microbiome using the antibiotic metronidazole affects tumor growth in mice. We employed an orthotopic mouse model of PDAC using congenic C57BL/6 PDAC cells isolated from genetically engineered KPC mice (5). Metronidazole was provided in drinking water to tumor bearing mice on d0 after orthotopic PDAC cell injections and tumor growth was assessed on d21. Treatment with metronidazole significantly increased tumor burden (Fig. 1A).

Figure 1. Administration of TMAO or TMA drives immune activation in the PDAC TME.

(A) Measurement of tumor weight in PDAC bearing C57BL/6 (B6) mice on d21 after orthotopic implant of PDAC cells and treatment with metronidazole antibiotic (1g/l) provided in drinking water. n=8 mice in control and n=5 mice in metronidazole treated group. Data are representative of three independent experiments.

(B) Volcano plot comparing metabolite profiles of serum from metronidazole-treated and control mice in (A). Metabolomics analysis was performed by non-targeted LC-MS/MS. n=8 mice in control and n=5 mice in metronidazole treated group.

(C) Schematic representation of the experiment (shown on left) with treatments of TMAO or TMA in PDAC bearing mice. Assessment of tumor weight in PDAC bearing mice on d21 after orthotopic implant of PDAC cells and treatment with TMAO (80mg/kg i.p., 4x per week) or TMA (40mg/kg i.p., 4x per week) starting d7 after tumor cell implants (shown on right). n=5 mice per group. Data are representative of 3–4 independent experiments.

(D) Assessment of tumor weight in PDAC bearing mice d21 after orthotopic implant of PDAC cells and treatment with metronidazole antibiotic and TMAO or TMA. n=5 mice per group.

(E, F, G) Flow cytometry analyses on PDAC tissues from (C) for (E) expression of MHCI, MHCII, and CD86 by TAMs shown as MFI (mean fluorescent intensity), (F)expression of Arg1 by TAMs shown as MFI, and (G) percent IFNγ⁺ TNFα⁺ of CD8⁺ and CD4⁺ T cells. Histogram shows MFI for Arg1 on TAMs and scatter plot shows percent IFNγ⁺ TNFα⁺ on CD8⁺ T cells. n=5 mice per group.

(H) Kaplan–Meier Analysis showing survival of PDAC bearing mice treated with TMAO or TMA. n=10 mice per group.

In (A, C-G), data are presented as mean +/- SD. In (A), p-values were determined by two-tailed Student’s t-tests. In (B), significant change defined as |FC| > 2; q-value < 0.1 (Benjamini-Hochberg FDR-adjusted p-value). In (C-G), p-values were determined by one-way ANOVA with post-hoc multiple comparisons. In (H), p-value was determined using a log-rank (Mantel–Cox) test.

To gain a mechanistic understanding of why tumor burden increased upon treatment with metronidazole, we performed a non-targeted metabolomic screen on serum from the PDAC bearing mice that received metronidazole. The screen identified TMAO as the serum metabolite most reduced by metronidazole, an average of >73-fold (Fig. 1B, Fig. S1A). We then asked whether exogenous TMAO would affect PDAC growth and immune responses in the TME. To address this question, we administered physiologically relevant doses of TMAO or TMA intraperitoneally (i/p) starting d7 after orthotopic PDAC cell injections (28). TMAO levels in tumor tissues increased >8-fold within 2hr of i/p injection, suggesting that TMAO was well disseminated at the tumor site within a relatively short time (Fig. S1B). Two weeks later, we measured tumor weight and assessed phenotypic changes in the immune cell infiltrates by flow cytometry. Administration of either TMAO and TMA (TMAO precursor) significantly decreased tumor size and weight (Fig. 1C). Similar results were observed using a different PDAC cell clone 6419c5 (Fig. S1C). Notably, the increase in tumor burden driven by metronidazole could be reversed by administering TMAO or TMA (Fig. 1D). These data suggested that the increase in tumor burden following metronidazole-mediated depletion of gut bacteria may be due to a loss of TMAO.

The flow analysis on tumors revealed no change in percentages of immune infiltrates in mice treated with TMAO or TMA (Fig. S1D, S1E). However, the immune profiles became more activated. Analysis on TAMs showed increases in the expression of co-stimulatory markers such as MHCI, MHCII, and CD86; these were accompanied by a dramatic decrease in the expression of the anti-inflammatory marker Arg1, together suggesting a shift towards an immunostimulatory TAM phenotype (Fig. 1E, 1F). Assessing other immune cell populations, we found a similar shift towards an immunostimulatory phenotype in myeloid-derived suppressor cells (MDSC), CD103⁺ dendritic cells (DC), and plasmacytoid DCs (pDC); in contrast, the B cell immune phenotype remain unchanged in the PDAC TME of mice treated with TMAO or TMA (Fig. S2A-D). Strikingly, effector T cell activation increased substantially, as evidenced by an increase in the percentages of IFNγ⁺TNFα⁺ CD8⁺ and CD4 T cells and an increase in the activation maker CD44 on both CD8⁺ and CD4⁺ T cells (Fig. 1G, S2E-F). Moreover, levels of TMAO in serum correlated positively with percentages of IFNγ⁺TNFα⁺ CD8⁺ T cells (r = 0.7257, p-value = 0.026) and negatively with tumor weight (r = -0.7692, p-value = 0.015) (Fig. S2G). Finally, TMAO or TMA administration significantly improved the survival of PDAC bearing mice (Fig. 1H). These data suggested that TMAO or TMA may suppress tumor growth by reconfiguring the tumor milieu toward an immune activated state.

It was possible that the reductions in tumor weight by TMAO were due to a direct effect on tumor cells. To address this possibility, we monitored PDAC cell proliferation in the presence of increasing TMAO concentrations in vitro by alamarBlue assay. Proliferation of PDAC cells treated with TMAO was indistinguishable from untreated controls (Fig. S2H). We also tested the effects of TMAO on PDAC growth in NSG mice. Treating tumor bearing NSG mice with TMAO did not decrease tumor burden compared to untreated controls (Fig. S2I). These data suggested that the tumor-restraining effect of TMAO is independent of a direct effect on tumor cells.

Tumor growth can also be influenced by systemic changes in immune cells. Therefore, we assessed phenotypic changes in immune cells from peripheral organs and lymphoid tissues (bone marrow, spleen, mesenteric lymph nodes, and colon) from PDAC bearing mice treated with TMAO (Fig. S3A-E). Flow analysis found a few differences in frequencies or phenotype of immune cells. Specifically, we found increased expression of CD86 and MHCII on F4/80⁺CD11b⁺ macrophages in mesenteric lymph nodes, suggesting their activation (Fig. S3B). We found increased percentages of CD11c⁺CD103⁺ DCs in spleen associated with increased expression of activation markers including MHCI and PDL1 (Fig. S3C). We also found increased percentages of CD8⁺ T cells in the intra-epithelial fraction of colonic mucosa associated with increases in activation molecules CD44 and CD69 (Fig. S3E). CD4⁺ T cells from intra-epithelial fraction of colonic mucosa similarly showed increased levels of CD44 and CD69 (Fig. S3E). Taken together, these results suggested that induction of immune activation in peripheral tissues may contribute to the tumor-restraining effect of TMAO.

Dietary choline or gut microbial TMA lyase phenocopied the effects of administration of TMAO

Choline is a major source of circulating TMAO. We thus investigated whether increasing dietary choline would phenocopy the anti-tumor effects of TMAO. We placed PDAC bearing mice on dietary supplementation of 1% wt/wt choline in normal chow (i.e. >10-fold increase in choline from baseline) or control normal diet starting on d0 after orthotopic tumor cell implants. Three weeks later, sera were assessed for changes in metabolites using a non-targeted metabolomic screen. The only significant changes were increases in levels of TMAO and TMA in the choline-supplemented mice compared to controls; notably, these changes associated with a significant decrease in PDAC burden (Fig. 2A, 2B). Flow cytometry analysis of tumor infiltrating immune cells revealed that choline supplementation led to similar immunostimulatory effects as administration of TMAO. Specifically, there were significant increases in expression of co-stimulatory markers such as MHCI and MHCII on TAMs (Fig. 2C). We also found increases in activation markers on other innate immune cells including CD11b⁺Ly6C^highLy6G⁻ monocytic MDSCs (M-MDSC), CD11b⁺Ly6C^lowLy6G⁺ polymorphonuclear MDSCs (PMN-MDSC), and antigen presenting CD103⁺ DCs (Fig. S4A-C). These analyses also revealed marked increases in activated CD8⁺ and CD4⁺ T cells, as evidenced by increases in percentages of IFNγ⁺TNFα⁺ CD8⁺ and CD4⁺ T cells associated with increases in the activation maker CD69 (Fig. 2D, S4D-E). We next treated choline supplemented mice with metronidazole and evaluated tumor response and TMAO levels. Metronidazole dramatically reduced the levels of choline-dependent TMAO and TMA in serum and diminished the tumor reducing effects of choline supplementation (Fig. 2E, S4F). These results indicate that dietary choline can phenocopy the anti-tumor effects of TMAO by generating TMAO in a gut microbiome-dependent manner.

Figure 2. Dietary and gut microbiome interventions phenocopy the anti-tumor effects of TMAO.

(A) Schematic representation of the experiment (shown on left) with supplementation of 1% choline or control diet in PDAC bearing mice. Volcano plot showing differences in serum metabolites in PDAC bearing mice on d21 after orthotopic implant of PDAC cells and supplement of 1% wt/wt choline diet or control diet (shown on right). Metabolomics was performed by non-targeted LC-MS/MS. n=5 mice in control diet and n=5 mice in 1% choline diet group.

(B) Assessment of tumor weight in PDAC bearing mice on d21 after orthotopic implant of PDAC cells and supplement of 1% wt/wt choline diet or control diet. n=7–8 mice per group. Data are representative of 2–3 independent experiments.

(C, D) Flow cytometry analyses on PDAC tissues from (B) for (C) expression of MHCI and MHCII by TAM shown as MFI and (D) percent IFNγ⁺ TNFα⁺ of CD8⁺ and CD4⁺ T cells. n=7–8 mice per group.

(E) Assessment of tumor weight in PDAC bearing mice on d21 after orthotopic implant of PDAC cells and provided with normal chow, 1% wt/wt choline diet, or 1% wt/wt choline diet plus metronidazole. n=5 mice per group.

(F) Quantification of TMAO levels by LC-MS/MS in sera from PDAC bearing mice on d21 after orthotopic implant of PDAC cells and receiving fluoromethylcholine (FMC) or metronidazole in drinking water. n=5 mice per group.

(G) Measurement of tumor weights in orthotopic PDAC bearing mice on d21 after orthotopic implant of PDAC cells and treatment with fluoromethylcholine (FMC). n=8 mice in control and n=5 mice in FMC treated group. Data are representative of two independent experiments. (H, I) Flow cytometry analysis on PDAC tissues from (G) for (H) expression of Arg1 and MHCII by TAM shown as MFI and (I) percent IFNγ⁺ TNFα⁺ of CD8⁺ and CD4⁺ T cells. n=8 mice in control and n=5 mice in FMC treated group.

In (A), significant change defined as |FC| > 2; q-value < 0.1 (Benjamini-Hochberg FDR-adjusted p-value). In (B-I), data are presented as mean +/- SD. In (B-D, G-I), p-values were determined by two-tailed Student’s t-tests. In (E, F), p-values were determined by one-way ANOVA with post-hoc multiple comparisons.

TMAO generation depends on the activity of the gut microbial enzyme cluster CutC/D which degrades choline to TMA. To examine whether the CutC/D enzyme was required for the TMAO driven anti-tumor responses, we asked whether inhibiting CutC/D activity using the previously characterized CutC/D inhibitor fluoromethylcholine (FMC) would reduce circulating TMAO and increase PDAC burden. We provided FMC in the drinking water to PDAC bearing mice. As expected, treatment with FMC dramatically reduced circulating TMAO levels, similar to metronidazole (Fig. 2F). FMC treatment also associated with a significant increase in PDAC burden compared to control mice (Fig. 2G). Flow analysis of tumors revealed that FMC significantly increased the expression of Arg1 while reducing MHCII on TAMs, indicating an overall reduction in TAM immunostimulatory phenotype (Fig. 2H). The immunostimulatory phenotype of CD103+ DCs also decreased, as evidenced by a reduction in MHCI and IL-12p40 expression (Fig. S4G). Last, there was a significant decrease in the percentages of IFNγ⁺TNFα⁺ CD8⁺ and CD4⁺ T cells, suggesting a decreased effector T cell response (Fig. 2I). These data suggested that the TMAO induced immunostimulatory phenotype of TAMs and other immune cells was due at least in part to the activity of the gut bacterial enzyme CutC/D.

TMAO drove anti-tumor transcriptional changes in the PDAC immune infiltrates

We next sought to understand the effect of TMAO on TAMs. We assessed the transcriptome by performing RNAseq on TAMs isolated from tumors of TMAO treated PDAC bearing mice. Our analysis revealed significant changes in 1736 gene transcripts in the TMAO treated group, consistent with a shift toward an immunostimulatory state, compared to controls. Upregulated genes included many involved in inflammation, such as Nr4a3, H-Q5, H2-Q6, Irf3, Irf5, Acod1, Def 6, and Ccrl2 (Fig. 3A). Downregulated genes included Trim29, Tgfb3, Cyp1b1, Pparg, and Mmp12 (Fig. 3A). Ingenuity Pathway Analysis (IPA) of regulators showed enrichment, in TAMs from TMAO treated mice, of several markers that associate with an immune activated phenotype, increased co-stimulation, and loss of immunosuppressive function. Activated regulators included TNF, NFkB, CD40LG, IL15, IFN Beta, and STING1, and inhibited regulators included TGFBR2, IL10RA, and ARG1 (Fig. 3B). These data provided a deeper understanding of how TMAO shifted TAMs to an immunostimulatory phenotype.

Figure 3. TMAO drives immune activation in the PDAC TME.

(A) Heat map showing differential gene expression of indicated RNA transcripts of genes of interest (p-value < 0.05) in FACS-sorted TAMs from tumors of d21 orthotopic PDAC bearing mice treated with TMAO (80mg/kg i.p., 4x per week starting d7 after tumor cell implants). n=4 mice per group.

(B) Ingenuity Pathway Analysis (IPA) of activated (red bars) and inhibited (blue bars) regulators in TAMs from (A).

(C) Uniform Manifold Approximation and Projection (UMAP) plot of scRNA-seq on tumor infiltrating CD45⁺ cells sorted from pooled tumors of d21 orthotopic PDAC bearing mice treated with TMAO. Mice received TMAO treatments as in (A). Tumors from n=7 mice were pooled together per group.

(D) Violin plot of scRNA-seq data in (C) showing number of cells in each immune cell cluster from control and TMAO treated groups.

(E) IPA for regulators of scRNA-seq data in (C) showing activated (red bars) and inhibited (blue bars) regulators in indicated immune cell clusters. Differentially expressed genes passing FDR<0.05 were analyzed.

(F) IPA for functions of scRNA-seq data in (C) showing activated (red circles) and inhibited (blue circles) functions in T cell cluster. Differentially expressed genes passing FDR<0.05 were analyzed.

(G) Measurement of tumor weights in PDAC bearing mice on d21 after orthotopic tumor cell implants and receiving treatments with TMAO or anti-CSF1 plus clodronate or combination. n=5 mice per group.

(H) Assessment of tumor size in PDAC bearing mice on d21 after s/q tumor cell implants and receiving treatments with TMAO or CD8⁺ T cell depleting antibody or combination. n=6–8 mice per group.

In (G, H), data are presented as mean +/- SD. p-values were determined by two-way ANOVA with post-hoc multiple comparisons. ns= not significant.

To gain insights into transcriptional changes induced by TMAO in tumor- infiltrating immune cells, we performed 10x Genomics Chromium droplet single-cell RNA sequencing (scRNA-seq) on FACS-sorted CD45⁺ immune cells. We assessed 10,000 CD45⁺ cells from control and TMAO treated PDAC bearing mice. After quality control and filtering, scRNA-seq analysis identified 7793 cells in the control and 8706 cells in the TMAO treated group. Cell clustering, visualized using a Uniform Manifold Approximation and Projection (UMAP) plot, showed multiple immune cell subgroups segregated into 14 cell types with distinct transcriptional profiles (Fig. 3C).

The frequency of tumor infiltrating immune cells from each cluster did not differ between control and TMAO treated groups (Fig. 3D). However, there were marked differences in the transcriptome profiles. IPA for upstream regulators on the macrophage cluster revealed increases in immunostimulatory markers including type-I IFN responsive transcription regulators (e.g., NFkB, IRF7, and STAT1), inflammatory cytokines (e.g., IL-15, IL12, IFN Beta, and IL-1β), and co-stimulatory molecule (e.g., CD40) and decreases in immunosuppressive markers (e.g., AHR, VEGF, and IL10) (Fig. 3E). IPA on neutrophil and dendritic cell clusters similarly showed activation of some regulators (e.g., TNF, MyD88, and IRF7 in neutrophils and TNF, CD40, and STING1 in dendritic cells) and inhibition of others (e.g., HOXA3, ABCA1, and PPARG in neutrophils and BCL6, IL10RA, and CTLA4 in dendritic cells) (Fig. 3E). These results suggested that TMAO reconfigured the immunosuppressive myeloid cells in the PDAC TME, shifting them to a more activated state.

TMAO also impacted transcriptional profiles in T cells. IPA for upstream regulators on the regulatory T cell cluster showed that these too shifted away from an immunosuppressive phenotype in the TMAO treated group compared to control. We found increases in RICTOR, CPT1B, and TCF7, and decreases in AHR, PTEN, and FOXP3 (Fig. 3E). Interestingly, IPA on the T cell cluster for enrichment of functional categories revealed changes in functions that associate with increased activation and reduced tumor growth. Functions that were activated included proliferation of CD8⁺ T lymphocyte, immune response of cells, antigen presentation, and cell death of tumor cell lines, and those that were inhibited included quantity of regulatory T lymphocytes, proliferation of tumor cells, fibrous tissue tumor, and proliferation of myeloid cells (Fig. 3F). These data indicated that TMAO limited the tumor promoting regulatory T cell response while enriching the cancer killing effector T cell phenotype.

These data prompted us to ask if TAMs and/or CD8⁺ T cells were needed for the TMAO mediated reduction in PDAC burden. We depleted macrophages using anti-CSF1 plus clodronate in PDAC bearing mice and found that after depletion of TAMs, TMAO no longer reduced tumor burden compared to the control group, suggesting TAMs may be involved in the anti-tumor effect of TMAO (Fig. 3G). Similarly, after selective depletion of CD8⁺ T cells, TMAO no longer reduced tumor burden , suggesting that CD8⁺ T cells also are needed for the TMAO mediated anti-tumor effect (Fig. 3H). These combined data suggested that the microbial metabolite TMAO suppressed tumor growth by reconfiguring immune cells in the tumor microenvironment toward a more activated state.

TMAO induced macrophages to acquire an immunostimulatory phenotype

We next determined whether TMAO had a direct effect on macrophage functional phenotype. To investigate this, we examined the in vitro effect of TMAO on pro-inflammatory (M1) or anti-inflammatory (M2) macrophage polarization. To test for M1 polarization, we stimulated mouse bone marrow derived macrophages (BMDM) with LPS or LPS plus IFNγ in the presence of TMAO (300µM) for 8hr. TMAO strikingly elevated the secretion of pro-inflammatory cytokines (IL-6, IL-12p40) and decreased the secretion of anti-inflammatory cytokine IL-10 (Fig. S5A). Testing for M2 polarization, we stimulated macrophages with IL-4 for 8hr. TMAO significantly decreased the expression of immunosuppressive markers Arg1, Fizz1, and Mgl1 (Fig. S5B). Based on these findings, we next performed a transcriptome analysis by RNA-sequencing (RNAseq) to interrogate direct effects of TMAO on gene expression. Exposure of macrophages to TMAO resulted in significant changes in the expression of >1500 genes with fold change ≥1.5-fold and FDR<0.05 (Fig. S5C). IPA revealed increases in multiple pathways contributing to pro-inflammatory response. Pathways that were activated included mTOR signaling, interferon signaling, and production of nitric oxide and reactive oxygen species (ROS) and those that were inhibited included PPAR signaling, sirtuin signaling, and PTEN signaling (Fig. S5D). IPA for functions showed enrichment for immune response of macrophages, quantity and function of lymphocytes, and found decrease in functions for morbidity or mortality, formation of solid tumors, and adenocarcinoma (Fig. 4A). These data suggested that TMAO induced an immunostimulatory macrophage phenotype directly.

Figure 4. TMAO signals by promoting activation of the type-I IFN response.

(A) IPA of RNAseq for activated (red bars) and inhibited (blue bars) regulators in BMDM treated with 300µM TMAO for 8hr. Differentially expressed genes passing FDR<0.05, and |FC|>=1.5 were analyzed. n=3.

(B) IPA of RNAseq for activated (red bars) and inhibited (blue bars) regulators in BMDM pre-treated with TMAO (300µM) for 1hr or left untreated and then stimulated with 20% PDAC tumor conditioned media (TCM) for 8hr. Differentially expressed genes passing FDR<0.05, and |FC|>=1.5 were analyzed. n=3.

(C) RT-PCR showing relative expression of indicated genes in BMDM pre-treated with TMAO (300µM) for 1hr or left untreated and then stimulated with ISD (interferon stimulatory DNA) or poly(dG:dC). The mRNA relative expression is shown compared to β-actin. n=4.

(D) Flow cytometry analyses of indicated activation markers on BMDM pre-treated with TMAO (300µM) for 1hr or left untreated and then stimulated with ISD (interferon stimulatory DNA). Histograms show MFI for CD86 on BMDM. n=3–4.

(E) Schematic representation of the experiment (shown on left) with treatments with anti-IFNAR1 and TMAO in PDAC bearing mice. Assessment of tumor size in PDAC bearing mice on d21 after s/q PDAC cell implants and receiving treatments with TMAO or IFNAR1 blocking antibody or combination (shown on right). n=7 mice per group.

(F) Kaplan–Meier Analysis showing survival of PDAC bearing mice from (E).

In (C-E), data are presented as mean +/- SD. p-values were determined by two-way ANOVA with post-hoc multiple comparisons. In (F), p-value was determined using a log-rank (Mantel–Cox) test. ns= not significant.

To gain an understanding of TMAO effects on macrophages in the context of the tumor milieu, we exposed BMDM to PDAC tumor conditioned media (TCM) in the absence or presence of TMAO for 8hr and performed RNAseq. Addition of TCM alone resulted in significant changes (FDR<0.05) in expression of >9300 genes (Fig. S6A). As examples, TCM increased expression of immunosuppressive genes including Arg1, Cd274, and Tgfβ1, while decreasing expression of genes contributing to pro-inflammatory response including Clec12a, IL12rb2, and Ciita (Fig. S6A). IPA for regulators revealed increases in NFE2L2, MYC, HIF1A, and ATF4, and decreases in IRF3, IRF7, and STAT1 (Fig. S6B). This suggested that in vitro conditioning of macrophages with TCM drove changes in the transcriptome that are characteristics of immunosuppressive TAMs.

Addition of TMAO to the TCM exposed macrophages resulted in ~3900 significantly altered genes (FDR<0.05); 2742 of these were also significantly altered by TCM alone (Fig. S6C). TMAO drove a striking decrease in expression of genes related to immunosuppression (including Cxcl5, Arg1, Ccl22, and S100A9), matrix metalloproteinases (including Mmp8, Mmp9, and Mmp13), and extracellular matrix remodeling (including Col7a1 and Vcan) (Fig. S6D-E). In contrast, TMAO increased expression of immunostimulatory genes (including Ccl5, Clec12a, and H2-DMa) (Fig. S6D). IPA revealed decreases in some regulators (e.g., NFE2L2, IL4, TGFB1, and HIF1A) with remarkable increases in regulators of the type-I IFN pathway (e.g., IRF7, IRF3, STING1, and STAT1) (Fig. 4B). It is particularly notable that in vivo TMAO also significantly increased regulators of the type-I IFN pathway (e.g., IRF7, IFN Beta, STING1, and STAT1) in TAMs (Fig. 3B, 3E). Collectively, these results showed that TMAO induced an immunostimulatory macrophage phenotype and this phenotype associated with an increased type-I IFN response; this increased type-I IFN response may be a key mechanism underlying the anti-tumor effects of TMAO.

TMAO potentiated the type-I IFN response and conferred anti-tumor effects in a type-I IFN dependent manner

To validate the observation that TMAO induced type-I IFN activation, we stimulated BMDM with known activators of the type-I IFN pathway including ISD (interferon stimulatory DNA) or poly(dG:dC) for 8hr in the absence or presence of TMAO and assessed changes in transcripts by RT PCR. BMDM responded to ISD and poly(dG:dC) with increases in expression of type-I IFN response genes (Fig. 4C). TMAO strikingly induced the expression of Irf7, Irf5, and Ifnα1 to both ISD and poly(dG:dC) (Fig. 4C), but only induced expression of Ifnβ1 to poly(dG:dC) (Fig. 4C). TMAO also increased surface expression of activation markers including CD86, MHCII, and PDL1, and the percent of MHCII⁺PDL1⁺ macrophages upon stimulation with ISD (Fig. 4D). These results suggested that TMAO potentiated the type-I IFN response and may promote the acquisition of an immunostimulatory macrophage phenotype by stimulating the type-I IFN pathway.

These data prompted us to examine whether the type-I IFN pathway was required for the TMAO driven anti-tumor responses. We treated PDAC bearing mice with anti-IFNAR1 blocking antibodies starting on d0 and with TMAO starting d7 after tumor cell implants. In control mice (no block of type I IFN pathway), TMAO significantly decreased tumor growth and improved survival as expected; however, in mice that received anti-IFNAR1, the anti-tumor effects of TMAO were completely abrogated (Fig. 4E, 4F). These results suggested that the tumor suppressive action of TMAO required the type-I IFN pathway.

TMAO directly altered macrophages to a phenotype able to support T cell responses and reduce PDAC burden

The observation that treatment of macrophages with TMAO significantly changed gene transcription (e.g., of transcripts associated with immune response of antigen presenting cells, quantity and function of lymphocytes, and CD8 T cell proliferation, Fig. 4A) suggested that TMAO may enhance the ability of macrophages to drive T cell proliferation. We co-cultured OT-I T cells (SIINFEKL-specific CD8⁺ T cells) with cognate peptide-pulsed dendritic cells for 3-days in the presence of BMDM that were primed with TCM +/- TMAO for 24 hours. TCM priming was used to mimic the immunosuppressive effects of the tumor milieu. As expected, co-culture with TCM-stimulated BMDM, which contributes to a highly immunosuppressive phenotype, drove significant reduction in T cell proliferation compared to activated controls, with most T cells achieving only 4 rounds of cell division (Fig. 5A). In contrast, co-culture with BMDM that were primed with TCM in the presence of TMAO reversed the suppressive capacity of BMDM and rescued T cell proliferation, with most T cells achieving 6 rounds of cell division (Fig. 5A). Similar results were observed when T cells received proliferative stimuli αCD3 and αCD28 (Fig. 5B).

Figure 5. TMAO shapes macrophage responses to promote effector T cell activity and reduce PDAC growth.

(A) Flow cytometry analysis of Cell Trace Violet (CTV) labelled OT-I T cell proliferation after co-culture with TCM stimulated BMDM (+/- TMAO) in the presence of DCs and OVA_257–264 peptide for 3 days. n=3.

(B) Flow cytometry analysis of Cell Trace Violet (CTV) labelled CD8⁺ T cell proliferation after co-culture with TCM stimulated BMDM (+/- TMAO) in the presence of anti-CD3 and anti-CD28 for 3 days. n=3.

(C) Measurement of tumor weights in PDAC bearing mice on d21 after orthotopic tumor cell implants and receiving i.v. injections of BMDM (either primed with TMAO for 24hr or control, 3x per week, 10⁶ per mouse). n=5 mice per group. Data are representative of two independent experiments.

(D, E) Flow cytometry analysis on PDAC tissues from (C) for expression of IFNγ, Ki67, CD103, and CD44 on CD8⁺ T cells shown as MFI. Histogram shows MFI for indicated markers on CD8⁺ T cells.

(F) RT-PCR showing relative expression of indicated genes in human monocyte derived macrophages (HMDM) pre-treated with TMAO for 1hr and then stimulated with 20% PDAC conditioned media (TCM) for 8hr. TCM was generated using PANC-1 human PDAC cell line. The mRNA relative expression is shown compared to β-actin. n=4.

Data are presented as mean +/- SD. In (A, B), p-values were determined for comparing division 5 to 6 between the groups by one-way ANOVA with post-hoc multiple comparisons. In (C, E), p-values were determined by one-way ANOVA with post-hoc multiple comparisons. In (F), p-values were determined by two-tailed Student’s t-tests. ns= not significant.

This prompted us to test whether TMAO-primed macrophages could also restrain PDAC growth. We treated tumor bearing mice with 10⁶ TMAO-primed macrophages given i.v. 3x per week for 2 weeks, starting d7 after orthotopic tumor implantation. The TMAO-primed macrophages were generated by treating BMDM with TMAO for 24hr. Macrophages injected i.v. reached the spleen and tumor quickly (2hr after injection) (Fig. S7A). Analysis of tumor burden on d21 after orthotopic implantation of PDAC cells showed that mice that received control macrophages had a slightly increased tumor burden compared to untreated controls (Fig. 5C). Intriguingly, mice that received TMAO-primed macrophages had a more than 2.4-fold average decrease in tumor burden compared to mice receiving control macrophages (Fig. 5C). To evaluate if this shrinkage of tumors was due to effector T cell responses, we assessed tumor infiltrating T cells for phenotype by flow cytometry. Mice that received control macrophages did not exhibit changes in expression of activation markers (e.g., IFNγ, Ki67, CD44), but only showed decreased CD103 on both CD8⁺ and CD4⁺ T cells (Fig. 5D-E, S7B), compared to mice not receiving any macrophages. However, mice that received TMAO-primed macrophages showed a robust activation profile, with significant upregulation of IFNγ, Ki67, CD103, and CD44 on both CD8⁺ and CD4⁺ T cells, compared to mice receiving control macrophages (Fig. 5D-E, S7B). Together, these results suggested that TMAO shaped macrophages to a phenotype that enhanced effector T cell response and decreased PDAC growth.

A pro-inflammatory effect of TMAO was observed in human macrophages

We next tested whether TMAO would induce a pro-inflammatory phenotype in human macrophages. We exposed human monocyte derived macrophages (HMDM) to TCM generated from human pancreatic cancer cell lines PANC-1 or Capan-2 in the absence or presence of 300µM TMAO. Addition of TMAO to TCM-stimulated macrophage cultures significantly increased the expression of pro-inflammatory markers (including IFNR1, IL-6, CCL-5, TNFA, and IL-1B for PANC-1 derived TCM, and IFNB, IL-6, CCL-5, IL-1B, NFKB for Capan-2 derived TCM) and significantly decreased the expression of anti-inflammatory marker IL-10 (Fig. 5F, S7C-D). These results suggested that TMAO also drove a pro-inflammatory response in human macrophages and may function in humans in a fashion similar to that observed in mice.

TMAO sensitized PDAC to checkpoint immunotherapy and improved survival in tumor bearing mice

Immune checkpoint inhibitors such as anti-PD1 do not reduce PDAC burden in patients (1). Our data suggested that TMAO enhanced the immune activated state in the PDAC TME, with heightened activation of myeloid and T cells (Fig.1, 3, and S2). Concurrently, TMAO treated tumors displayed increased expression of immune checkpoint molecules on immune cells. Specifically, administration of TMAO or TMA, or dietary supplementation with choline, in PDAC bearing mice significantly increased tumor infiltrates of PD1⁺Tim3⁺ CD8⁺ T cells (Fig. 6A-B). Likewise, levels of PDL1 on TAMs, MDSCs, and B cells were increased (Fig. S8A). This suggested to us that combining TMAO with anti-PD1 or anti-Tim3 may increase efficacy over either intervention alone. First, we treated PDAC bearing mice with either TMAO, anti-PD1 antibody, or combination starting d7 after orthotopic tumor implantation for two weeks. As anticipated, anti-PD1 alone did not reduce tumor burden, however a combination of anti-PD1 plus TMAO did significantly reduce tumor weight compared to TMAO alone or control (Fig. 6C). Flow analysis of tumors revealed a striking increase in the cell surface marker MHCI on myeloid cells, including TAMs, MDSCs, and CD103+ DCs in the combination treated group compared to single treatments or untreated controls, suggesting that combination treatment increased the immunostimulatory phenotype of these cell populations (Fig. 6D). This was associated with a remarkable effector T cell response, comprising significant increases in the percentages of CD44⁺Ki67⁺ and IFNγ⁺TNFα⁺ CD8⁺ and CD4⁺ T cells, suggesting considerable proliferation and activation state of T cells (Fig. 6E, S8B). Similar results on tumor burden were observed with anti-Tim3 antibody (Fig. 6F). These results suggest that TMAO renders PDAC responsive to anti-PD1 or anti-Tim3 checkpoint therapy.

Figure 6. Administration of TMAO or TMA renders PDAC responsive to ICB.

(A, B) Flow cytometry analysis of percent PD1⁺Tim3⁺ CD8⁺ T cells assessed from tumor tissues of PDAC bearing mice on d21 after orthotopic implant of PDAC cells and receiving (A) treatments with TMAO or TMA and (B) supplementation of 1% choline diet or control diet. Scatter plot shows percent PD1⁺Tim3⁺ on CD8 T⁺ cells. (A) n=5 mice per group and (B) n=7–8 mice per group.

(C) Measurement of tumor weights in PDAC bearing mice on d21 after orthotopic PDAC cell implants and receiving treatments with TMAO or anti-PD1 or combination. n=5 mice per group. Data are representative of 2–3 independent experiments.

(D, E) Flow cytometry analysis on PDAC tissues from (C) for (D) the expression of MHCI by TAMs, CD103⁺ DCs, and MDSCs shown as MFI, and (E) percent CD44⁺ Ki67⁺ or IFNγ⁺ TNFα⁺ of CD8⁺ T cells.

(F) Measurement of tumor weights in PDAC bearing mice on d21 after orthotopic PDAC cell implants and receiving treatments with TMAO or anti-Tim3 or combination. n=5 mice per group.

(G) Assessment of tumor size in PDAC bearing mice on d21 after s/q PDAC cell implants and receiving treatments with TMAO, TMA, anti-PD1, anti-Tim3, or indicated combinations. n=7–8 mice per group.

(H) Kaplan–Meier Analysis showing survival of PDAC bearing mice from (G).

In (A-G), data are presented as mean +/- SD. In (A, C-G), p-values were determined by one-way ANOVA with post-hoc multiple comparisons. In (B), p-values were determined by two-tailed Student’s t-tests. In (H), p-value was determined using a log-rank (Mantel–Cox) test. ns= not significant.

To better explore the therapeutic potential of TMAO, we employed a subcutaneous tumor model and started treatment of mice when tumors reached 3–5 mm in size. Mice receiving TMAO, or TMA, or a combination of anti-PD1 plus anti-Tim3 had a significantly lowered tumor burden than controls (Fig. 6G). Interestingly, the tumor restraining effect was significantly augmented when we combined TMAO with anti-PD1 plus anti-Tim3, or TMA with anti-PD1 plus anti-Tim3; combinations also increased survival (Fig. 6G, 6H). Collectively, these results suggested that the choline metabolites TMAO or TMA, when combined with ICB (anti-PD1 plus anti-Tim3), improved PDAC responsiveness to ICB and survival of tumor bearing mice.

Levels of TMA-producing bacteria and of CutC gene expression correlated with improved survival and response to anti-PD1 in cancer patients

Microbes in the gut and tumors impact host physiology, cancer growth, and treatment response (7, 9, 11, 29). Riquelme et al. suggested that enrichment of several bacterial species in tumors from long-term survivors of PDAC are predictive of improved survival (7). Fecal microbiota transplantation (FMT) from these long-term PDAC survivors (LTS) into PDAC bearing mice conferred anti-tumor immune effects compared with FMT from short-term PDAC survivors (STS) or healthy controls. It is suggested that bacteria found in tumors have translocated from the gut to colonize the pancreas, where they influence the composition of the tumor microbiome and anti-tumor immunity (7, 8). Because our studies implied that the anti-tumor effects of TMAO required the activity of the bacterial enzyme CutC, we hypothesized that the tumor microbiome of LTS patients had an overabundance of CutC containing bacteria. Various bacterial taxa belonging to Clostridia, Bacilli, Desulfovibrionia, and Gammaproteobacteria contain the CutC gene and are significant producers of TMA, which contributes to TMAO levels in the circulation (30–32). To gain insights into whether the tumor microbiome of these LTS patients contained TMA-producing bacteria, we analyzed the tumor microbiome 16S ribosomal RNA gene sequencing data of LTS and STS patients from Riquelme et al. TMA-producing bacteria considered in this analysis were identified by Rath et al. in human fecal samples by a gene targeted assay for choline TMA lyase (CutC) (30). Examining the genus-level bacterial counts, we found that the relative abundance of Bacillus and Paenibacillus was significantly greater in LTS compared to STS PDAC patients (Fig. 7A). These results suggested that the presence of TMA-producing bacteria correlated with improved survival in PDAC patients.

Figure 7. Abundance of CutC containing bacteria and elevated CutC gene expression correlate with improved survival and response to anti-PD1 in cancer patients.

(A, B) Box-and-whisker plot showing relative abundance of CutC containing bacterial counts of (A) Genus Bacillus (top) and Genus Paenibacillus (bottom), within the tumor microbiome of short-term survivors (STS) and long-term survivors (LTS) of PDAC and (B) Genus Bacillus, within the gut microbiome of melanoma patients who were responders (R) or non-responders (NR) to anti-PD1 checkpoint immunotherapy after FMT.

(C) Heat map showing normalized CutC level in CutC expressing bacterial strains between responders (R) vs non-responders (NR) to anti-PD1.

(D) Box-and-whisker plot indicating relative abundance of CutC containing bacterial families Clostridiaceae and Enterococcaceae in responders (R) vs non-responders (NR) to anti-PD1.

(E) Box-and-whisker plot of CutC gene expression shown as ppm across responders (R) vs non-responders (NR) to anti-PD1.

(F) Overall survival of patients stratified by expression of CutC.

In (A, B), Mann-Whitney non-parametrical test was used for comparison of relative abundance values between LTS vs STS and R vs NR groups with Benjamini-Hochberg used for correction for multiple testing. Significant results that passed FDR<0.05 were shown. In (C, D), CutC presence was associated with response through Fisher exact test. Bacterial CutC with Fisher exact p < 0.01 and Mann-Whitney p < 0.03 are presented. In (E), p-value was determined by the Wilcoxon rank-sum test, and (F) p-value was determined by the log-rank test.

Our murine studies showed TMAO rendered PDAC responsive to anti-PD1. This suggested that levels of TMAO may dictate response to anti-PD1. We therefore hypothesized that the clinical response to anti-PD1 may correlate with the presence of CutC containing bacteria or the expression of the CutC gene in the fecal microbiome. To test this, using data from Davar, et al. (14), we assessed whether the gut microbiome from patients who responded to anti-PD1 therapy after FMT contained increased levels of TMA-producing bacteria compared to non-responders. Indeed, the relative abundance of CutC containing Bacillus was higher in responders compared to non-responders (Fig. 7B). To correlate anti-PD1 response with CutC containing taxa and CutC gene expression, we analyzed the shotgun metagenomic sequencing and fecal microbiome 16S ribosomal RNA gene sequencing data from Matson, et al. (9). This uncovered significant associations between anti-PD1 response and CutC expression in 18 bacterial strains including Clostridium tetani, Clostridium tepidum, Clostridium culturomicium, Enterococcus asini, Enterococcus sp. DIV1298c, and Enterococcus phoeniculicola (Fig. 7C). These strains belong to two bacterial families, namely Clostridiaceae and Enterococcaceae, which are indeed enriched in responders compared to non-responders (Fig. 7D). Last, analysis of data from McCulloch JA, et al. (29) similarly showed a significant increase in the expression of CutC in responders compared to non-responders (Fig. 7E), and that high expression of CutC significantly associated with improved overall survival after anti-PD1 (Fig. 7F). Taken together, these results suggest that the presence of CutC containing bacteria and CutC gene expression correlate with improved survival and response to anti-PD1 in cancer patients.

Discussion

Increasing evidence suggests that gut flora is linked to immune cell function and to improved response of cancers to ICB (9–12, 14, 16). However, the underlying mechanisms remain unclear but likely involve microbial metabolites. We found an unforeseen link between the gut microbiome-derived metabolite TMAO and anti-tumor immune responses in PDAC. TMAO induced an immunostimulatory phenotype in macrophages, boosted effector T cell function, and rendered PDAC responsive to ICB. Clinically, CutC containing bacteria correlated with improved response to anti-PD1. Collectively, our study identified the microbial metabolite TMAO as a driver of anti-tumor immunity in PDAC.

Similar to other published reports (25, 26), we found that TMAO was a significant inducer of inflammatory effects. Our data suggested that TMAO drove activation of myeloid cells (TAMs, MDSCs, and DCs) and T cells both in the TME and systemically. TMAO increased inflammatory drivers including NFkB, type-I IFN, STING, STAT1, and Notch in myeloid cells. These inflammatory drivers are known to shape immune cell function and thus can regulate tumor growth and therapy response. For example, type-I IFN activation has suggested to restrict generation of TAMs while promote polarization towards immunostimulatory phenotype (13, 33). STING agonists shape TAM/myeloid cells to acquire a dominant immunostimulatory phenotype and render tumors responsive to checkpoint therapy in pre-clinical models of many cancer types including PDAC (34, 35). Our data also suggested that TMAO inhibited immunosuppressive markers including AHR, MYC, EGFR, TGFB1, VEGF, HIF1A, and STAT3, which help establish a highly suppressive and therapy resistant TME. Many studies including ours suggested that approaches inhibiting immunoregulatory molecules (e.g., AHR, IDO1, or ATF4) or growth factors (e.g., EGFR, VEGF, or TGFB1) to improve anti-tumor immunity (3, 6, 36–39). Together, these data suggest that TMAO is a driver of anti-tumor effects and may do so by reconfiguring tumor immune infiltrates to an immune activated phenotype.

Mechanistically, TMAO appears to exert its immunostimulatory effects by potentiating the type-I IFN pathway. The type-I IFN pathway has emerged as a key link between the gut microbiome and host immune responses in multiple pathophysiological situations including cancer (13, 40). The type-I IFN response in the macrophage/myeloid compartment is a major driver of anti-tumor immunity and response to ICB (13, 41–43). In our work, transcriptome analysis of TAMs from bulk and single cell studies demonstrated that TMAO increased the activation of regulators of the type-I IFN pathway (e.g., IRF7, IFN⍺/β, IFN Beta, STING, and STAT1). Interestingly, TMAO specifically boosted type-I IFN activation to the known activator ISD (interferon stimulatory DNA), which associated with a significant increase in expression of activation markers (e.g., CD86, MHCII). Importantly, when mice were given anti-IFNAR1 blocking antibody, the anti-tumor effects of TMAO were abolished. Although the precise mechanism by which TMAO directly affects type-I IFN pathway remains to be determined, our study demonstrated the comprehensive immunomodulatory effects of TMAO that shape macrophage phenotype to support anti-tumor immunity.

Chronic exposure to TMAO increases the risk for cardiovascular disease, renal failure, and diabetes (18, 19, 44), and increased TMAO levels are linked to risk of chronic inflammatory malignancies like colorectal cancer (45, 46). Our studies employed strategies that transiently increased TMAO to transform the immunosuppressive TME into an immunogenic state that can respond to checkpoint immunotherapy. This suggests that strategies that acutely increase TMAO may be reasonable interventions. In this vein, we employed a dietary intervention with a choline supplement (available as nutritional supplement) and found effects similar to those of direct administration of TMAO. Additionally, our studies evaluated an alternative to systemically delivery of TMAO, in which we delivered TMAO-primed macrophages. These mimicked the effects of direct administration of TMAO. In the longer-term, it may be important to determine the links between TMAO, inflammation and cancer during both short-term and extended TMAO exposures.

The diversity and composition of the gut microbiome and/or the tumor microbiome dictate tumor immunity and response to immunotherapy (7, 9, 11, 29). One way in which the microbiome composition influences immune responses in the TME is by production of metabolites that confer biological effects on immune cells. For example, triple-negative breast cancers enriched with immune activated cells contain bacteria belonging to Clostridiales and abundant levels of TMAO (17). Our data analysis suggests that CutC containing TMA-producing bacteria (Genus Bacillus and Paenibacillus) are significantly increased in long-term compared to short-term PDAC survivors. Interestingly, the same genus, Bacillus, is overabundant in melanoma patients who respond to anti-PD1 checkpoint therapy. Shotgun metagenomics analysis of fecal microbiome revealed that levels of CutC containing bacteria within families Clostridiaceae and Enterococcaceae and expression of the CutC gene correlate with improved response to anti-PD1 in melanoma. Moreover, we found that two gut microbiome dependent metabolites, TMA and TMAO, drove robust immune activation in the PDAC TME and rendered tumors responsive to ICB. Our studies shed light on possible mechanisms connecting the microbiome and immune responses in the TME. This may lay the groundwork for developing new microbiome-based or diet-based therapies as novel, adjunct therapeutics for PDAC.

Current immunotherapy approaches lack efficacy in treating PDAC. Given the strong immunosuppressive and tumor promoting role of TAMs and their predominance in the PDAC TME, there is growing interest in targeting TAMs through strategies that impede their recruitment and survival or reprogram them to an immunostimulatory phenotype able to bolster effector CD8 T cells. Our studies demonstrate that the microbial metabolite TMAO enabled TAMs to become immunogenic and promote effector T cell activity, transformed the TME to an immune activated state, and rendered PDAC sensitive to checkpoint immunotherapy, suggesting that strategies that alter levels of TMAO could be a promising clinical intervention to manage PDAC.

Materials and methods

Study Design

The main objectives of this study were to elucidate how the gut microbe-derived metabolite TMAO influences the host immune responses in the PDAC TME, to determine if TMAO would synergize with immune checkpoint therapy to reduce tumor growth, and to assess whether there is a clinical correlation between TMAO production and survival in PDAC patients. The overall design did employ some in vitro approaches but relied on in vivo mouse model systems, including orthotopic tumor implantation, to achieve the first two objectives; to test for a clinical correlation, we assessed datasets of bacterial species generated by three other groups. We employed flow cytometry and multiomic approaches to determine the impact of TMAO on the phenotype of immune infiltrates in the TME. Mice were randomly assigned to the experimental groups and studies were not blinded. The sample sizes of 5 to 10 mice per experimental group for tumor growth studies were determined based on prior experiments with 80% power at a two-sided type I error rate of 0.05 to detect an effect size of 1.3 or larger. The number of mice used in each experiment is indicated in figure legends.

Animals

Mouse experiments were performed following National Institutes of Health (NIH) guidelines and were approved by the Institutional Animal Care and Use Committee (IACUC) of The Wistar Institute. C57BL6/J (B6), OT-I (TCR transgenic on B6 background) mice were obtained from Jackson Laboratory and/or were bred at the animal facility of The Wistar Institute. NSG mice were obtained from the animal facility of The Wistar Institute. All mice were maintained in a specific pathogen-free barrier facility at The Wistar Institute in accordance with the Association for Assessment and Accreditation of Laboratory Animal Care (AAALAC). Female or male mice at 8–10 weeks old were used in all experiments.

PDAC cell lines and culture

Mouse PDAC cell clones (2838c3, 6419c5) were isolated from late-stage primary tumors from genotypically proven C57BL/6 background KrasLSL-G12D/+; Trp53LSL-R172H/+; Pdx1-Cre; Rosa26YFP/YFP (KPCY) mice as previously described (5). These clones were authenticated to be free of rodent pathogens using IMPACT Rodent Pathogen Testing by the IDEXX BioAnalytics (Columbia, MO) and were routinely tested for mycoplasma contamination. PDAC cell lines were cultured in DMEM (Corning, MA, 10–013-CV) with 10% FBS (Corning, MA, 35–016-CV) and 100 U penicillin and streptomycin antibiotics (Corning, MA, 30–002-CI).

PDAC orthotopic and subcutaneous murine model

For orthotopic tumor cell implant, mice were anesthetized using isoflurane and provided s/q injection of analgesic Buprenorphine SR (Covetrus North America, O72117) peri-operatively. The surgical area was prepared and a skin incision was made in the left side of the abdomen. Using the forceps spleen was exposed to gain access to the pancreas and 50µl single-cell suspensions of PDAC cells (0.75 to 1.5 X 10^5 per mouse) diluted in the matrigel and 1x PBS at 1:1 was injected into the tail region of the pancreas using insulin syringes (29 Gauge). The successful injection was verified by the appearance of a fluid bubble without signs of intraperitoneal leakage. The pancreas and spleen was put back gently and abdominal wall was sutured with absorbable Vicryl suture (McKesson, VA, 1034496) and the skin was closed with wound clips (Braintree Scientific, 205016). Mice were observed for signs of tumor growth by palpation. For the subcutaneous model, 50µl single-cell suspensions of 2 X 10^5 mouse PDAC cells in 1x PBS were injected s/q. After three weeks, mice were sacrificed for the assessment of serum, tumor growth, and immune profiling. For subcutaneous model, tumor sizes were assessed by caliper measurements of tumor length (L) and width (W) and calculated as tumor volume (L*W²/2). For survival analysis, tumor volumes of 500 mm³ were used as an endpoint.

Preparation of tumor conditioned media (TCM) and macrophage polarization

The tumor conditioned media (TCM) was generated by dissociating orthotopic PDAC tissues in DMEM media (200mg/ml) and incubating the tissue suspension for 2hr at 37⁰C. First, tissue suspension was spun at 400g for 5min and then the supernatant was spun at 21000g for 10min to remove cellular/tissue debris. Then the supernatant was filtered using sterile 0.2µM filters (VWR, 28145–477) and was used at 20% to stimulate mouse macrophages in vitro. Additionally, macrophages were polarized to M1 phenotype by LPS (100ng/ml; Sigma, L2880) or LPS (100ng/ml) plus IFNγ (100ng/ml; BioLegend, 575306), or to M2 phenotype by IL-4 (10ng/ml; BioLegend, 574304) stimulation for 8hr.

Generation of primary mouse and human macrophages

Mouse bone marrow derived macrophages (BMDM) were generated by culturing bone marrow cells with macrophage colony stimulating factor (MCSF) (BioLegend, 576406) as previously described (47). Human monocyte derived macrophages (HMDM) were generated by culturing human monocytes with human MCSF (BioLegend, 574804). More details can be found in the supplemental materials.

RT-PCR

Mouse or human macrophage lysates were processed to obtain purified RNA using RNeasy Plus Mini Kit (Qiagen, 74136). iScript^™ Reverse Transcription Supermix kit (Bio-Rad, 1708841) was used to reverse transcribe 100ng of RNA to cDNA. RT-PCR reactions were carried out on Applied Biosystems Quant Studio 7 Flex Real-Time PCR system using 1X SYBR Green PCR Master Mix (Applied Biosystems). Primer sequences are listed in Supplementary Table 1. The relative mRNA expression is analyzed compared to the expression of β-actin. Data was analyzed using manufacturer’s instructions.

ELISA

Supernatants from BMDM culture polarized with M1 stimuli were assessed for cytokine production by Enzyme-linked immunosorbent assays (ELISAs). Commercially available kits from Invitrogen were used to measure IL-6 (Invitrogen, 88–7064), IL-12p40 (Invitrogen, 88–7120), and IL-10 (Invitrogen, 88–7105) according to the instructions provided by the manufacturer.

Type-I interferon activation in vitro

Interferon stimulating DNA (ISD) and Poly(dG:dC) were obtained from InvivoGen (tlrl-pgcn, tlrl-isdn). To activate type-I IFN response in macrophages, ISD or poly(dG:dC) were mixed with Lipofectamine 3000 (Invitrogen, L3000150) at a ratio of 1:1 (v/w), and then added to cells at a final concentration of 1 μg/ml.

T cell suppression assay

For T cell suppression assay, we primed BMDM with 20% TCM in the presence of 300µM TMAO for 24hr. OT-I T cells (CD8⁺) were isolated from spleens of OT I mice using EasySep^™ Mouse CD8⁺ T Cell Isolation Kit (Stem Cell Technologies, 19853A) and were labelled with CellTrace^™ Violet dye (Invitrogen, C34557A) at a final concentration of 5µM. Dendritic cells (DCs) were isolated from spleens of wild type mice using EasySep^™ Mouse Pan-DC Enrichment Kit II (Stem Cell Technologies, 19863). Co-cultures were set up in triplicates with BMDM primed as above, OT I T cells, and DCs in the presence of OVA_257–264 peptide (Ana Spec Inc., AS-60193–1) at 0.5ng/ml for 3 days. DCs were added at 1:10 ratio to T cells. T cell proliferation was assessed by flow cytometry for dilution of CellTrace^™ Violet stain on CD8⁺ T cells. In experiments using anti-CD3 and anti-CD28 induced T cell proliferation, CD8⁺ T cells were isolated from spleens of wild type mice and were co-cultured with BMDM primed as above in the presence of 0.1 µg of anti-CD3 (Biolegend, 100340) and anti-CD28 (Biolegend, 102116).

In vivo treatments

TMAO (Sigma Aldrich, 317594) at 80mg/kg and TMA (Sigma Aldrich, T72761) at 40 mg/kg were administered via intraperitoneal (i.p.) injection 4x per week given on alternate days starting d7 after tumor cell implantation. Metronidazole (Cayman Chemical Company, 900249) treatment was given in drinking water at 1mg/ml with change of water bottle two times a week. To inhibit CutC, mice received treatment with fluoromethylcholine (FMC) (BLD pharm, BD130578) provided in drinking water from day 0 at 100mg/kg. In experiments with dietary supplementation of choline, mice received 1% choline diet (Envigo Teklad Diets, TD.130328) and ingredients matched normal chow control diet (Envigo Teklad Diets, TD.00588).

For ICB, mice received anti-PD1 (200μg i.p. 3x per week, clone RMP1–14, BioXCell) or anti-Tim3 alone (100μg i.p. 3x per week, clone RMT3–23, BioXCell, BE0115). To deplete CD8⁺ T cells, mice were injected with anti-CD8 depletion antibody (250 μg/mouse, i.p., clone 2.43, BioXcell, BE0061) starting d0 of tumor cell injections and given every three days until the end of the study. Control mice received 1x PBS i.p. To deplete TAMs, mice were treated with CSF1 neutralizing antibody (0.5 mg per mouse, i.p., clone 5A1, BioXCell, BE0204) and clodronate-containing liposome (100 μl per mouse, Liposoma BV, C-005). Both treatments were given starting d3 prior to tumor cell implantation and on every fourth day post tumor implantation until the end of the study. For IFNAR1 blocking experiments, mice were treated with neutralizing IFNAR1 antibody (BioXCell, BE0241, clone MAR1–5A3) at 200 μg/mouse i.p. on day 0 and 100 μg/mouse on every fourth day post tumor implantation until the end of the study.

Flow cytometry

Fluorescent conjugated antibodies used for flow cytometry are listed in the Supplementary Table 2. Single cell suspensions of tumor, spleen, mesenteric lymph node, bone marrow, and colon mucosa were prepared as previously described (3, 47, 48) and stained for surface and intracellular markers. Samples were acquired on BD FACS Symphony flow cytometer. Data were analyzed using FlowJo v10 (Treestar Inc.). More details can be found in the supplemental materials.

RNAseq analysis

Total RNA was extracted from BMDM using RNeasy Plus Mini Kit (Qiagen, 74136) or from sorted F4/80⁺CD11b⁺ TAMs using Single cell purification kit (Norgen, 51800). RNA quality was validated using the TapeStation High Sensitivity RNA ScreenTape (Agilent). Libraries were prepared using Illumina Stranded mRNA Prep kit (500 ng DNase1 treated total RNA was used). Final library QC was done on the bioanalyzer using High Sensitivity DNA kit (Agilent). Next Generation Sequencing was performed with 100 bp paired end, on NovaSeq 6000 (Illumina) using SP v1.5 200 cycle kit. RNA-seq data was aligned using bowtie2 (49) algorithm against hg19 human or mm10 muouse genome version and RSEM v1.2.12 software (50) was used to estimate read counts and RPKM values using gene information from human Ensemble transcriptome version GRCh37.p13 or mouse Ensemble transcriptome version GRCm38.89. Raw counts were used to estimate significance of differential expression difference between any two experimental groups using DESeq2 (51). Overall gene expression changes were considered significant if passed FDR<0.05 thresholds with additional fold change cutoffs when stated. Gene set enrichment analysis was done using QIAGEN’s Ingenuity^® Pathway Analysis software (IPA^®, QIAGEN Redwood City,www.qiagen.com/ingenuity) using “Canonical pathways”, “Diseases &Functions” and “Upstream Regulators” options. Selected results that passed FDR<0.05 threshold were considered unless stated otherwise. Additional cutoffs on predicted activation state Z-score were applied where specified. Pathway clustering analysis was performed using Cytoscape (52). Selected pathways or functions network was imported with edges weight between any two pathways calculated as % of overlapped genes with overlaps of <20% not considered. Clustering was performed using Cytoscape plugin clusterMaker2 (53) with “Community Cluster (GLay)” option.

Single cell RNA-sequencing (scRNA-seq)

The scRNA-seq was performed on fluorescence-activated cell sorting (FACS)-enriched live CD45⁺ cells from tumors of orthotopic PDAC bearing mice. The scRNA-seq data was preprocessed using Cell Ranger Suite (v3.1.0, https://support.10xgenomics.com) with refdata-cellranger-mm10–3.0.0 transcriptome as a reference to map reads on mouse genome (mm10) using STAR (54). Low-quality cells with less than 200 genes with reads and cells with over 10% mitochondrial content were filtered out. Cell clustering, marker identification and visualization were performed using Seurat v4 (55). R package SingleR (56) was used to determine cell types of the clusters using ImmGen data set as a reference for cell-specific gene signatures. Further resolution of cell types was carried out using known gene markers associated with different immune cell sub-types. Genes differentially expressed between TMAO and control samples within clusters of interest were identified using non-parametric Wilcoxon rank sum test and statistically significant genes (FDR<0.05, and fold cut-offs when specified) were subjected to enrichment analysis using QIAGEN’s Ingenuity^® Pathway Analysis software (IPA^®, QIAGEN Redwood City, www.qiagen.com/ingenuity)

Microbiome analysis

We analyzed 16S microbiome data for long- and short-term surviving patients from Riquelme et al (7) (NCBI BioProject Accession Number: PRJNA542615) using QIIME2 software (57). Shotgun sequencing data related to response to PD-1 blockade treatment after fecal microbiota transplant (14) was derived from NCBI BioProject PRJNA672867 and analyzed using Kraken2 (58) and Bracken (59) pipeline. Genus-level bacterial counts were exported and relative abundance values were calculated by normalizing raw counts to the total number of reads in the samples. Mann-Whitney non-parametrical test was used for comparison of relative abundance values between LTS vs STS and Responder vs non-responder groups with Benjamini-Hochberg used for correction for multiple testing. Significant results that passed FDR<0.05 were overlapped with a list of TMA-producing bacteria identified by Rath et al (30) and reported.

We additionally analyzed shotgun metagenomic sequencing and 16S sequencing from Matson, et al. for anti-PD1 responding and non-responding patients (9). To quantify CutC expression, shotgun sequencing FASTQ files were downloaded from through SRP116709. Reads were aligned against a collection of CutC reference protein sequences that was downloaded through NCBI Reference Sequence Database bacterial (RefSeq), using BlastX (60)with E-value<1e-7. For each sample, the number of each RefSeq bacterial CutC reads was normalized to the total number of reads. The normalized reads were compared between responders and non-responders through Mann-Whitney non-parametrical test. In addition, CutC presence was associated with response through fisher exact test. Bacterial CutC with fisher exact p < 0.01 and Mann-Whitney p < 0.03 are presented in Fig. 7C. Family-level 16S of Enterococcaceae and Clostridiaceae (which were identified through CutC expression analysis) were obtained from Matson, et al. and compared between responders and non-responders through Mann-Whitney non-parametrical test.

We additionally utilized the normalized bacterial CutC gene expression levels provided in McCulloch JA, et al. (29) which combines data from five different projects (PRJNA762360, PRJNA399742, PRJNA541981, PRJNA397906 and PRJEB22893). Pan-cohort CutC levels were compared between responders and non-responders through Mann-Whitney non-parametrical test. Pan-cohort CutC levels were additionally used to compare survival between patients with low vs. high CutC expression levels, for patients with available survival information through McCulloch JA, et al. using log-rank test (where patients were grouped by median CutC expression level).

Metabolomics analysis

Standards used for analysis: TMAO (Sigma, 317594), TMA (Sigma Aldrich, T72761), TMAO-d9 (Cambridge Isotope Labs, DLM-4779-PK), and TMA-13C3 (Sigma, 591599). Non-targeted polar metabolite profiling was performed as described previously with some modifications (61). Briefly, polar metabolites were extracted from serum samples with 10 times volume of ice-cold 80% methanol. Samples requiring absolute quantification of TMAO or TMA, were spiked with 2.7 µM TMAO-d9 and TMA-13C3 stable isotope-labeled internal standards prior to extraction. For these samples, a full MS scan range of 50–750 m/z was used in positive ion mode. Non-targeted and targeted data analysis was carried out on the same raw data. Non-targeted analysis was performed using Compound Discoverer 3.1 (ThermoFisher Scientific). Periodic analysis of a sample pool was used to correct metabolite quantifications over the course of a batch analysis. Serum data were normalized per volume serum injected, which was equivalent for all samples. Targeted analysis was performed using TraceFinder 4.1 (ThermoFisher Scientific). Targeted metabolites were quantified by peak areas from [M+H]+1 adducts in full MS scans with 10 ppm mass error as follows: TMAO (76.07569 m/z), TMA (60.08078 m/z), TMAO-d9 (85.13218 m/z), and TMA-13C3 (63.09084 m/z). Peaks were selected for integration based on the retention time of standards. Metabolite concentrations were determined from endogenous to internal standard ratios and calibration curves with linear fit and 1/X weighting.

Statistical analysis

Statistical analysis was performed using GraphPad Software version 9. To compare differences between two groups, two-group t-test or Wilcoxon rank sum test was used, and to assess differences between multiple groups, One-way ANOVA, Two-way ANOVA or Kruskal-Wallis test with post-hoc multiple comparisons was used. Post-hoc multiple tests with Bonferroni’s procedure was applied to control type I error. The difference in survival of mice was calculated by Kaplan-Meier with log-rank analysis. Statistical analysis of non-targeted metabolomics data sets was performed using Perseus (62). Data were log2-transformed, and pairwise comparisons between experimental conditions used two-tailed, unpaired Student’s t-test with Benjamini-Hochberg false discovery rate correction for multiple hypothesis testing (q-value).

Supplementary Material

Supplementary Figures

Supplemental figure 1. TMAO does not influence percentages of immune infiltrates in the PDAC TME.

Supplemental figure 2. TMAO drives immune activation in the PDAC TME.

Supplemental figure 3. TMAO influences immune cell activation systemically.

Supplemental figure 4. Influence of choline diet and FMC on TMAO driven immune responses in the PDAC TME.

Supplemental figure 5. TMAO induces a pro-inflammatory phenotype in macrophages.

Supplemental figure 6. TMAO drives immunostimulatory phenotype in TCM conditioned macrophages.

Supplemental figure 7. Macrophages injected intravenously reach tumor tissue, TMAO-primed macrophages promote CD4⁺ T cell activation, and TMAO induces pro-inflammatory response in human macrophages.

Supplemental figure 8. Efficacy of TMAO in combination with anti-PD1 immune checkpoint blockade.

Supplementary Table 1

Supplemental table 1: List of primers used for RT-PCR

Supplementary Table 2

Supplemental table 2: List of antibodies used for flow cytometry

Supplementary Table 3

Supplemental table 3: Data file

Data and materials availability

The RNA-seq and scRNA-seq data are available in GEO database (GEO accession GSE207947). The metabolomics data are available at the NIH Common Fund’s National Metabolomics Data Repository (NMDR) website, the Metabolomics Workbench, https://www.metabolomicsworkbench.org where it has been assigned Study ID: ST002216 and ST002217. The data can be accessed directly via its Project DOI: http://dx.doi.org/10.21228/M8PM6K. . All data needed to evaluate the conclusions in the paper are present in the paper or the Supplementary Materials