Microbiota triggers STING-type I IFN-dependent monocyte reprogramming of the tumor microenvironment

Summary

The tumor microenvironment (TME) influences cancer progression and therapy response. Therefore, understanding what regulates the TME immune compartment is vital. Here, we show that microbiota signals program mononuclear phagocytes in the TME towards immunostimulatory monocytes and dendritic cells (DCs). Single-cell RNA sequencing revealed that absence of microbiota skews the TME towards pro-tumorigenic macrophages. Mechanistically, we show that microbiota-derived STING agonists induce type I IFN (IFN-I) production by intratumoral monocytes to regulate macrophage polarization and NK–DC crosstalk. Microbiota modulation with high-fiber diet triggered the intratumoral IFN-I–NK–DC axis and improved efficacy of immune checkpoint blockade (ICB). We validated our findings in melanoma patients treated with ICB and showed that the predicted intratumoral IFN-I and immune compositional differences between responder and non-responder patients can be transferred by fecal microbiota transplantation. Our study uncovers a mechanistic link between microbiota and the innate TME that can be harnessed to improve cancer therapies.

Graphical Abstract

eTOC blurb:

Gut microbiota tunes the pro/anti-tumorigenic balance of the tumor microenvironment via STING-type I IFN-dependent mechanism. Feeding high-fiber diet or transferring fecal microbiota from immune checkpoint blockade (ICB) responder melanoma patients triggers this pathway and programs intratumoral mononuclear phagocytes to promote anticancer immunity and ICB efficacy.

Introduction

Tumors develop alongside a highly complex and dynamic microenvironment, and the interplay among all its components influences disease progression and response to therapy (Binnewies et al., 2018). Immune cells are key players of the TME, and studies have shown that classifying tumors based on their immune composition may have predictive value for clinical outcome (Bruni et al., 2020). In this regard, much of the field’s focus has been on the adaptive immune compartment and approaches aiming at awakening T cells, such as immune checkpoint blockade (ICB), have yielded unprecedented results. Unfortunately, however, only a small subset of patients fully benefit of this approach and the determinants of resistance are still unclear (Chen and Mellman, 2017; Ribas and Wolchok, 2018; Sharma et al., 2017). Among the contributing factors to therapy resistance, the immunosuppressive nature of the TME is one of the most challenging. This underscores the need to dissect the signals that regulate the TME.

Mononuclear phagocytes (MPs) [i.e. monocytes (Mo), macrophages (Mac), and dendritic cells (DCs)] are innate immune cells that play instrumental roles in host defense, tissue homeostasis, and repair (Haniffa et al., 2015). MPs are also key components of the TME where they play opposing dual roles: they can induce antitumor responses, but they can also promote tumor immunosuppression and disease progression (Cotechini et al., 2015; Nakamura and Smyth, 2020). Consequently, MPs can be powerful targets to skew the outcome of anticancer immunity. However, the precise nature of the MP network, what cues regulate its composition and functional state, and how this affects antitumor immunity is not well understood, making their therapeutic targeting difficult. With the advent of single cell technologies, we have begun to appreciate the heterogenous nature of MPs in the TME and have seen that they are in a state of flux with high plasticity (Gubin et al., 2018; Katzenelenbogen et al., 2020; Maier et al., 2020; Molgora et al., 2020; Zhang et al., 2019; Zilionis et al., 2019).

Mo can be divided in two main subsets: classical and non-classical (Geissmann et al., 2003; Ziegler-Heitbrock et al., 2010), and they have both been assigned positive and negative roles in cancer (Griffith et al., 1999; Hanna et al., 2015; Qian et al., 2011). They may also impact tumor progression by differentiating into Mo-derived cells (Mac or moDCs) in the TME. However, the signals that influence the fate of Mo in the TME have yet to be deciphered. Tumor-associated Mac are generally assumed to be tumor-promoting (Mantovani et al., 2017). DCs on the other hand, owing to their exceptional antigen presentation capabilities, are thought to be beneficial for the antitumor response. (Binnewies et al., 2019; Broz et al., 2014; Hildner et al., 2008; Wculek et al., 2020).

It is now well appreciated that gut microbiota affect anticancer immunity, controlling spontaneous tumor growth (Arthur et al., 2012; Dejea et al., 2018; Tomkovich et al., 2019) as well as response to chemo- and immunotherapy (Iida et al., 2013; Mager et al., 2020; Paulos et al., 2007; Sivan et al., 2015; Tanoue et al., 2019; Vetizou et al., 2015; Viaud et al., 2013). Recent studies have demonstrated associations between microbiota composition and response to ICB in cancer patients, and evidence suggests that use of antibiotics near the start of treatment can alter therapy efficacy (Chalabi et al., 2020; Chaput et al., 2017; Frankel et al., 2017; Gopalakrishnan et al., 2018; Khan et al., 2021; Matson et al., 2018; Pinato et al., 2019; Routy et al., 2018). However, the cellular and molecular underpinnings of microbiota action remain ill-defined. In fact, most studies have looked at changes in T cell populations and little is known about the impact of microbiota on the intratumoral innate immune compartment. In particular, MPs – which are potent regulators of T cell responses – can be modulated by microbiota at sites distant from the gut (Gorjifard and Goldszmid, 2016). Yet, whether microbiota regulates the functional state of MPs in the TME and the signaling pathways involved are unknown.

In this study, we addressed this fundamental question using a combination of single-cell analysis, microbiota perturbations, and functional cell characterization in several preclinical cancer models. We reveal that microbiota-derived signals are needed to program MPs in the TME towards immunostimulatory Mo and DCs; absence of these signals shifts the MP repertoire towards tumor-promoting Mac. Mechanistically, we demonstrate that microbiota-derived STING agonists such as c-di-AMP induce IFN-I production by intratumoral Mo, which regulates their skewing and NK cell–DC crosstalk. Notably, this mechanism can be triggered by microbiota manipulation with high-fiber diet (FD) to improve antitumor response. We show Akkermansia muciniphila, enriched in FD, produces c-di-AMP and can mediate the FD phenotype. Importantly, we validated the translational relevance of our findings in melanoma patients treated with ICB. We show that transplant of responder, but not non-responder, patient microbiota is sufficient to trigger IFN-I and remodel the innate immune TME. Our study uncovers a mechanism by which microbiota regulates complex interactions among innate immune cells that can be harnessed to overcome the immunosuppressive TME.

Results

Absence of microbiota skews the TME towards protumor Mac at the expense of immunostimulatory Mo and DCs

To interrogate whether microbiota shapes MPs in the TME, we first took a reductionist approach comparing EL4 lymphoma tumors implanted into normobiotic mice (SPF) or animals raised germ-free (GF). We sorted tumor MP populations as Lin^negCD11c^hiMHCII^hi, Lin^negLy6C^hiF4/80^low-neg, and Lin^negLy6C^int-lowF4/80^pos and analyzed their transcriptional profile by NanoString (Figure S1A). We performed gene set enrichment analysis (GSEA) of cells obtained from SPF tumors and corroborated that the predominant cell signatures in these clusters were DCs for the CD11c^hiMHCII^hi population, Mo for Ly6C^hiF4/80^low-neg, and Mo-derived Mac for Ly6C^int-lowF4/80^pos (Figure S1B). When we compared the transcriptional profile of MPs from SPF and GF tumors, principal component analysis (PCA) showed a clear segregation among them with CD11c^hiMHCII^hi cells being the furthest apart along PC1, which explains the most variance in the data (Figures 1A and S1C).

Figure 1. Microbiota shapes the MP landscape in the TME.

Transcriptomic analysis of tumor MPs from SPF or GF mice.

(A-B) NanoString analysis of sorted MPs. PCA (A); Heatmap of DEGs in DCs (FDR < 0.1, |log2(fc)| > 0.6) with ImmGen enriched DC (orange) or Mac (blue) signatures indicated (B).

(C-J) scRNAseq analysis of MPs. UMAP projection of cell clusters or group ID (inset) (C). Dot plot of selected cluster specific genes (D). Trajectory analysis (top) and density plot with the relative contribution of SPF and GF cells (bottom) (E). Pseudotime analysis of E with expression of selected genes correlating with pseudotime (F). Trajectory analysis of Mo and Mac clusters with pie charts showing the proportion of each cluster per state (G). Proportion of cells from each state shown in G in SPF and GF samples (H). Circos plot showing common DEGs between SPF and GF within each cell cluster (FDR < 0.1, |log2(fc)| > 0.5). Outer circle: clusters per group. Inner circle: up-regulated DEGs. Lines connect the same genes to each other (I). IPA Canonical Pathways comparing DC2 populations (J).

Data from 2 experiments combined (A-B). n=2–3/group/exp (A-J). See also Figures S1–S2 and Tables S1–S3.

Further characterization of the CD11c^hiMHCII^hi population by differential gene expression analysis confirmed the enrichment in DC signature genes (e.g. Flt3, Cd209b, Zbtb46, Batf3) in cells from SPF tumors. In marked contrast, the same population obtained from GF display a strong pro-tumorigenic Mac signature (e.g. Mertk, Arg2, C1qa, Apoe, Cd68, Csf1r, Msr1, Trem2) (Figure 1B and Table S1). This skewing was also evidenced by the genes separating SPF and GF in the PCA of CD11c^hiMHCII^hi cells alone (Figure S1D), and the comparison of differentially expressed gene (DEG) clusters between SPF and GF with datasets from the Immunological Genome (ImmGen) Project (Figures 1B and S1E). We also observed skewing towards protumor Mac in the Ly6C^int-lowF4/80^pos population from GF as compared to SPF (Table S1).

Prompted by the striking differences in bulk gene expression analysis between SPF and GF tumor-infiltrating MPs, we next performed single-cell RNA sequencing (scRNAseq) of TME myeloid cells (Figure S1F and Table S2). Unsupervised clustering identified eight distinct MP clusters: three subsets of DCs including conventional DC1 and DC2 as well as a third population of DCs (Ccr7^hi DC) that display high levels of Ccr7, Cd40, Cd86, and Cd274 (PD-L1); two clusters of classical Mo (Mo1, Mo2); and three Mac clusters (Mac1, Mac2, Mac3) (Figures 1C, 1D, and S1F–S1H). Trajectory analysis showed that along component 1, GF had a skewed cell distribution towards Mac while SPF tumors were enriched in Mo and DCs (Figures 1E and 1F). We confirmed the enrichment of the DC marker Flt3 in SPF tumor lysates via RT-qPCR (Figure S1I). In agreement with our NanoString data, further scRNAseq analysis including trajectories of Mo and Mac clusters only, showed an altered cell state in GF tumors that results in the prevalence of Mac expressing canonical protumor genes such as Arg1, Mrc1 (CD206), Trem2, Retnla, C1qa, and Egr2 (Mac2/Mac3). Conversely, Mac in SPF (Mac1) tend to have an antitumor phenotype with higher levels of MHCII genes, Il1b, Cd86, Tlr2, and Nos2 (Figures 1B, 1D, 1G, 1H, and S1H).

Differential gene expression analysis between SPF and GF within each cell cluster revealed that common DEGs are regulated in similar direction across different cell types within each group (Figure 1I). GSEA showed Interferon Alpha Response among the top pathways enriched in Mac and DC populations from SPF compared to GF tumors (Figure S2A); and we confirmed that most MP populations from SPF contained higher number of DEGs associated with IFN-I regulation than those from GF (Figure S2B). Additional pathway analysis showed that Mo and Mac clusters from SPF were enriched in pro-inflammatory pathways, while the same clusters from GF had elevated anti-inflammatory pathways (Table S3). Likewise, when we analyzed DC2, the predominant DC cluster, cells from SPF were enriched in pathways associated with antitumor response such as Dendritic Cell Maturation and Th1 Pathway (Figure 1J). Contrarily, the DC2 cluster from GF tumors was enriched in pathways associated with immunoregulatory and anti-inflammatory functions such as peroxisome proliferator-activated receptors (PPARs) and PD-1/PD-L1 (Figure 1J). Collectively, our transcriptomic analysis showed that absence of microbiota results in impaired intratumoral IFN-I signaling and skewing of MPs towards suppressive Mac at the expense of stimulatory Mo and DCs.

The skewing revealed by our transcriptomic analysis was also observed by flow cytometry analysis. We found a significant reduction in frequency and absolute number of Mo (Lin^negLy6C^hiF4/80^low-neg) and DCs (Lin^negCD11c⁺MHCII⁺CD64^negCD24^+/++FLT3⁺) infiltrating GF compared to SPF tumors (Figures 2A and S2C). Importantly, this is not a GF-specific defect as using broad-spectrum antibiotics (ABX) to deplete microbiota in SPF mice results in a similar phenotype (Figure 2B). Consistent with our scRNAseq data, we found stark differences in Mac subset proportions between SPF and GF mice (Figure 2C). The Mac compartment in tumors from animals devoid of microbiota was skewed towards a pro-tumorigenic phenotype with high CD206 and negative or low expression of MHCII (Figures 2C and 2D). Of note, we did not observe differences in DC proportions in the spleen, tumor-draining or inguinal lymph nodes from SPF and GF mice that were either naive or tumor-bearing (Figures S2D and S2E), agreeing with previous reports (Ganal et al., 2012; Wilson et al., 2008). This indicates a TME-specific rather than a generalized DCpoiesis impairment in mice lacking microbiota.

Figure 2. Absence of microbiota skews TME towards pro-tumorigenic Mac at the expense of Mo and DCs.

Tumor leukocytes from mice with (SPF; H2O) or without (GF; ABX) microbiota.

(A-D) MPs from EL4 tumors. Frequency of Mo and DCs in tumors from SPF vs GF (A); or H2O vs ABX (B); pie chart of Mac subsets with center inlet indicating pro- or antitumorigenic phenotype (C); pro/antitumor Mac ratios in tumors from SPF vs GF (left) or H2O vs ABX (right) (D).

(E-G) MPs from MC38, BP, or TUBO tumors. Frequency of Mo of total CD45⁺ cells (E); Frequency of DCs of total CD45⁺ cells (F); pro/antitumor Mac ratio (G).

Data are representative of 2 (B, E-G TUBO) or 3 (A, C) experiments. n=5/group/exp (A, C, E-G MC38), n=3–5/group/exp (B, E-G TUBO), n=9–10/group (E-G BP). Data shown as mean +/− SEM, *p<0.05, **p<0.01. See also Figure S2.

Because the compositional complexity of the TME varies among tumor types, we extended our findings to three additional tumor models with clearly distinct TME immune composition: MC38 colon carcinoma and BP (BRAF^V600E/PTEN^−/−) melanoma in C57BL/6, and the orthotopic TUBO mammary carcinoma in BALB/c mice (Figure S2F). Despite starting with different TME immune repertoire, each of these tumors had significantly lower frequencies of Mo and DCs in absence of microbiota (Figures 2E and 2F). Additionally, we found reduced expression of CD86 and MHCII on DCs from MC38 and TUBO tumors from GF or ABX mice suggesting a less activated phenotype (Figures S2G and S2H). In line with the skewing observed in EL4, we found increased pro/anti-tumorigenic Mac ratio in MC38 and BP, and a similar trend in TUBO tumors from GF or ABX mice indicative of a tumor-permissive environment (Figure 2G). These results help explain the inability of microbiota devoid mice to respond to therapy (Iida et al., 2013; Routy et al., 2018; Vetizou et al., 2015; Viaud et al., 2013). Given the consistency of our findings in several preclinical models, we conclude that microbiota shapes the MP compartment in the TME.

We did not observe obvious differences in the tumor burden of SPF, GF, or ABX mice at the time of immunological analysis (Figure S2I and S2J). This agrees with previous findings from us and others showing that the growth of untreated tumors in mice with or without microbiota is comparable (Iida et al., 2013; Routy et al., 2018; Vetizou et al., 2015; Viaud et al., 2013). Therefore, while physiological levels of microbiota-derived signals may not suffice to control spontaneous tumor growth, they are critical to program MPs for optimal response to cancer therapy.

Microbiota regulates the intratumoral IFN-I–NK cell axis

To understand how microbiota regulates MPs in the TME, we measured cytokines and chemokines that might affect their function. EL4 tumor lysates from GF mice showed a marked reduction in proteins related to recruitment, maintenance, and function of Mo and DCs such as CCL2, GM-CSF, CXCL10, CCL5, IL-15/IL-15R, IL-18, and IL-27 (Figure 3A). In line with our transcriptomic analysis (Figures S2A and S2B), type I IFNs were also decreased in tumors from GF mice both at the protein (IFNα) and mRNA (Ifnb) levels (Figures 3A and S3A). IFN-Is are important for antitumor response (Diamond et al., 2011; Dunn et al., 2005; Fuertes et al., 2011; Zitvogel et al., 2015), and previous studies suggested that MPs required microbiota signals to produce IFN-I following virus infection (Abt et al., 2012; Ganal et al., 2012). Therefore, we surmise that microbiota can regulate IFN-I production in the TME generating a proinflammatory environment that favors antitumor immunity.

Figure 3. Microbiota regulates intratumoral IFN-I signaling and NK–DC axis.

(A-B) Cytokine/chemokine profile of EL4 tumor lysates. Normalized measurements from SPF and GF mice (crossed square indicate below detection limit) (A); difference of fold change means of SPF-GF and WT-Ifnar1^−/− (Kappa=0.313; SE: standard error of the difference) (B).

(C) Frequency of indicated MPs in EL4 tumor of WT or Ifnar1^−/− mice.

(D-E) Survival plot of WT and Ifnar1^−/− (D) or Tmem173^−/− (E) EL4 tumor-bearing mice treated or not with oxa.

(F-G) Frequency of NKs in EL4 tumors of WT or Ifnar1^−/− mice (F) or in indicated syngeneic tumors with or without microbiota (G).

(H) RT-qPCR of Xcl1 expression in EL4 tumors from GF or SPF (normalized to Actb SPF mean).

(I) MFI of Xcl1 in EL4 tumor-infiltrating NKs from GF or SPF.

(J) DEGs between SPF and GF within the NK cell cluster of scRNAseq (FDR < 0.1, |log2(fc)| > 0.5).

(K-M) EL4 tumor-infiltrating cells stimulated ex vivo with cdAMP. Frequency of each cell population of total Ifnb1⁺ CD45⁺ (K); frequency of Xcl1⁺ or Ccl5⁺ NKs of total NKs (L); fold change Ifnb1⁺ Mo cdAMP vs control (M).

(N-O) Intraperitoneal (i.p.) administration of cdAMP. Experimental design (N); frequency of Mo and DCs in the TME (O).

Data from 2 experiments combined (A, B SPF-GF, C-F, I, K) or one representative of 2 (G TUBO, H, L, O) or 3 (G EL4) experiments. n=3/group/exp (A, B SPF-GF), n=5/group/exp (B WT-Ifnar1^−/−, H), n=4–8/group/exp (C, F, G, K, L, M), n=5–10/group/exp (I, O). Data shown as mean +/− SEM, *p<0.05, **p<0.01, ***p<0.001, ****p<0.0001. See also Figures S3, S4, and Table S4.

If IFN-I signaling is a critical step in the microbiota’s effect on the TME immune status, we would predict that absence of IFN-I should result in a similar remodeling of the TME than absence of microbiota. We therefore characterized the immune protein and cellular infiltration profiles of EL4 tumors from wild-type (WT) or IFN-I receptor deficient animals (Ifnar1^−/−) and compared them to GF. We found that the intratumoral cytokine/chemokine profile of Ifnar1^−/−mice resembles that of GF (Figures 3B and S3B). Moreover, Ifnar1^−/− mice displayed an altered TME MP repertoire with reduced Mo and DCs coupled with a skewing towards protumor Mac as observed in GF (Figures 3C and S3C). Characterization of DC subsets in the TME showed that cDC1, which have an essential role in antitumor immunity (Barry et al., 2018; Bottcher et al., 2018; Hildner et al., 2008; Wculek et al., 2020), were decreased in Ifnar1^−/− as compared to WT tumors (Figure 3C). Similarly, mice without microbiota harboring EL4, MC38, or TUBO tumors had reduced cDC1 compared to SPF (Figures S3D and S3E). The reduced levels of IFNs and IFN-regulated genes in GF tumors, together with the remarkable similarities in the cytokine/chemokine profile and the MP repertoire between Ifnar1^−/− and microbiota-deficient mice, support our hypothesis that IFN-I is an important host factor mediating the microbiota’s effect on the TME.

An important consequence of the remodeled TME in absence of microbiota is the loss of therapy efficacy. To study if the same TME remodeling that resulted from absence of IFN-I signaling also affects response to therapy, we used oxaliplatin (oxa) – a chemotherapeutic commonly used in the clinic that has been shown to require microbiota and myeloid cells for its efficacy (He et al., 2021; Iida et al., 2013). We found that Ifnar1^−/− mice display an impaired response to oxa (Figure 3D) akin to that of ABX (Figure S3F) or GF (Iida et al., 2013). In addition, WT mice given anti-IFNAR1 neutralizing antibody also showed a blunted response to oxa (Figure S3G), confirming that – similarly to microbial signals – IFN-I signaling is required for optimal response. Stimulator of interferon genes (STING), a major regulator of IFN-I production, has been shown to play an important role in the response to immuno- and radiotherapy (Deng et al., 2014; Flood et al., 2019; Woo et al., 2014). We therefore asked whether STING, which can be activated by microbial ligands, is also required for chemotherapy efficacy. Indeed, STING deficiency (Tmem173^−/−) resulted in a comparably impaired response to oxa as the absence of Ifnar1 or microbiota signaling (Figure 3E). These results suggest that microbiota impact on therapy efficacy is at least in part via the induction of IFN-I.

We next sought to determine the cellular circuits downstream of the microbiota-induced IFN-I pathway involved in the remodeling of the TME. Our protein data analysis showed that GF and Ifnar1^−/− tumors had significantly lower amounts of IFN-inducible proteins critical for NK cell recruitment and activation such as CXCL10, CCL5, IL-15/IL-15R, and IL-18 (Figures 3A and S3B). Thus, we hypothesized that without microbiota, deficient IFN-I signaling in the TME results in reduced content and impaired function of NK cells (NKs). Accordingly, we found a marked reduction in NK frequency in the TME of Ifnar1^−/− as well as microbiota devoid mice (Figures 3F and 3G). NKs have been identified as regulators of cDC1 abundance in the TME (Barry et al., 2018; Bottcher et al., 2018), in part via the production of DC chemoattractants. Hence, NK impairment would explain the hampered DC recruitment in absence of microbial signals. To confirm this, we performed additional scRNAseq of tumor-infiltrating leukocytes (Figure S3H and Table S4) and found that NKs are the main producers of DC recruiting chemokines Xcl1 and Ccl5 as well as the DC differentiation and survival factor Csf2 (GM-CSF) in EL4 tumors (Figure S3I). Flow cytometry confirmed that NKs are the primary source of Ccl5 in these tumors (Figure S3J). We also observed a significant positive correlation between NKs and DCs in EL4, MC38, and TUBO tumors (Figure S3K). Supporting our hypothesis, we found that Xcl1 transcript as well as CCL5 and GM-CSF proteins were diminished in GF tumor lysates (Figures 3A and 3H), and that NKs from GF tumors produce less Xcl1 compared to SPF (Figure 3I).

We compared the transcriptional profile of NKs from SPF and GF tumors and revealed that NKs from SPF have an activated phenotype with higher expression of effector proteins (Gzmb, Serpinb9b, Ifng, Cxcl10, and Xcl1) and transcription factors known to regulate NK function and homeostasis (Ikzf2 and Klf2). In contrast, NKs from GF expressed higher levels of the gene encoding for transforming growth factor beta-induced protein (Tgfbi), compatible with a pro-tumorigenic environment (Cantelli et al., 2017) (Figure 3J). IFNγ was also significantly reduced in tumor lysates from GF mice (Figure 3A). Validating the scRNAseq and protein data, we observed that NKs from GF tumors were impaired in their ability to produce IFNγ (Figure S3L). From these observations, we conclude that microbiota-induced IFN-I signaling in the TME results in effective recruitment and activation of NKs, which via production of CCL5 and XCL1 in turn recruit DCs. In response to IFN-I, these DCs produce Il15/Il15ra (Figures 3A and S3I) that support and activate NKs (Burkett et al., 2004; Lucas et al., 2007; Mortier et al., 2008), triggering a positive feedback loop to promote antitumor immunity.

Microbial STING agonist c-di-AMP stimulates intratumoral IFN-I–NK cell axis and reprograms MPs

Our data shows that microbiota regulates the IFN-I–NK–DC axis in the TME; and that like microbial signals, IFN-I and STING signaling are required for optimal response to therapy (Figures 3D, 3E, S3F, and S3G). Additionally, previous studies have shown that STING-mediated IFN-I signaling can potentiate antitumoral effects of NKs (Marcus et al., 2018; Nicolai et al., 2020). Consequently, we reasoned that microbiota-derived STING-agonists such as cyclic dinucleotides (CDNs) would be able to regulate the IFN-I–NK–DC axis in the TME. To test this, we stimulated total tumor-infiltrating cells ex vivo with the bacterial CDN c-di-AMP (cdAMP) and measured Ifnb1 transcript by flow cytometry. We identified classical Mo as the main source of IFN-I in the TME in response to cdAMP (Figure 3K). Moreover, cdAMP-induced IFN-I production resulted in increased Xcl1 and Ccl5 expression by intratumoral NKs (Figure 3L). Notably, Mo were also amongst the major producers of the IFN-inducible chemokines Ccl2 and Cxcl10 (important for their own recruitment and that of other leukocytes) as well as of the NK-activating cytokine Il18 (Figure S3I). Interestingly, NanoString analysis of MPs sorted from SPF or GF tumors showed that cells from GF express higher levels of Atf3 (Table S1), a transcriptional repressor of IFNβ production and known negative regulator of inflammatory responses (Labzin et al., 2015). We confirmed the impaired ability of Mo from GF tumors to produce Ifnb1 (Figure 3M), which explains the reduced IFN-I protein, dysfunctional NKs, decreased DCs, and blunted antitumor response observed in these animals.

To rescue the IFN-I signaling deficiency in the TME of GF mice and to validate our findings with cdAMP in vivo, we used a gain-of-function approach mimicking the potential systemic impact of this microbiota-derived product (Figure 3N). Administration of cdAMP into GF animals resulted in increased intratumoral levels of IFN-I genes and a trend towards higher Xcl1 (Figure S4A). More importantly, it restored both Mo and DC infiltration (Figure 3O). Additionally, PCA of Mac populations showed that Mac in the TME of cdAMP-treated GF resemble those from SPF mice (Figure S4B). Conversely, cdAMP injection into SPF mice had no impact on Mo or DC infiltration, indicating that the effect observed in GF is likely due to restoration of a microbiota-mediated mechanism (Figure S4C). Collectively, our results demonstrate that systemic microbiota-derived products, such as the STING agonist cdAMP, regulate innate immune cell interactions in the TME and antitumor responses via IFN-I.

Microbiota modulation with high-fiber diet induces IFN-I, increases DCs, and improves antitumor response

Different gut microbiota compositions have been associated with response to cancer therapy in patients (Sepich-Poore et al., 2021). Thus, we next moved to a more translational and actionable approach to manipulate, but not ablate, gut microbes using dietary intervention or single antibiotics. SPF mice were fed different diets or received single antibiotics in drinking water starting before tumor implantation (Figure S4D). We found that EL4 tumors from animals fed high-fiber diet (FD), which has been associated with lower cancer risk (Trock et al., 1990), had increased frequency and absolute numbers of total DCs and cDC1s compared to mice fed control diet (Figures 4A, S4E, and S4F). Furthermore, FD resulted in better spontaneous tumor growth control (Figures 4B and 4C) in line with a recent study (Li et al., 2020a). Western diet (WD), high in fats and sugars has been associated with increased cancer risk (Fabiani et al., 2016; Xiao et al., 2019). Contrary to FD, mice fed WD had reduced tumor-infiltrating DC (tiDC) numbers and increased tumor size compared to control diet (Figures S4G and S4H). Additionally, we found that perturbing microbiota with a narrow-ranged antibiotic (vancomycin) can result in reduced tiDCs and cDC1s (Figures S4I).

Figure 4. Fiber diet increases tumor DCs and antitumor response.

Mice receiving control or fiber diet as indicated.

(A-C) Frequency of DCs (A); tumor weight (B); average (left) and individual (right) tumor growth curves (C) of EL4 tumors.

(D-H) Frequency of DCs (D); tumor weight (E); average (left) and individual (right) tumor growth curve (F); frequency of indicated MPs (G-H) in MC38 tumors.

(I-J) Tumor growth curve and survival plot of MC38 tumor-bearing mice treated with anti-PD-1 (I) or anti-PD-L1 (J).

Data from 2 (D, E) or 3 (A-C) experiments combined. One representative of 2 (G, H) or 3 (F, I) experiments. (A-H) n=4–8/group/exp. Data shown as mean +/− SEM, *p<0.05, **p<0.01, ***p<0.001, ****p<0.0001. See also Figure S4.

We confirmed the ability of FD to remodel intratumoral MPs using the MC38 model. MC38 tumors from FD mice had increased DCs and Mo as well as reduced Mac, which also resulted in improved spontaneous antitumor response (Figures 4D–4H). More importantly, FD significantly enhanced the therapeutic efficacy of anti-PD-1 and anti-PD-L1 treatments (Figures 4I and 4J). Collectively, our results show that microbiota manipulation can alter the innate landscape of the TME, particularly DC content, and in turn modulate antitumor immunity.

We then focused on FD for its potential clinical applications. We performed FMT from control or FD fed mice into GF animals and confirmed that the beneficial anticancer effect of FD was indeed mediated by microbiota (Figure S5A). We next interrogated the involvement of the IFN-I–NK axis in the FD-induced remodeling of the TME. We found higher levels of IFN-I genes, higher proportion of Ifnb1 producing Mo, and increased Xcl1 expression by NKs in tumors from FD compared to control diet (Figures 5A and S5B). Additionally, Mo from FD tumors produced more Ifnb1 in response to cdAMP (Figure S5C). These results provide an explanation for the augmented DC content and enhanced antitumor response in FD fed mice, further supporting our hypothesis that microbiota regulates the TME innate immune profile via IFN-I.

Figure 5. Fiber diet and A. muciniphila (Akk) trigger IFN-I–NK–DC axis in the TME.

(A) EL4 tumor-infiltrating cells from SPF mice fed control or fiber diet stimulated ex vivo with cdAMP. Fold change of Ifnb1⁺ Mo (left) and Xcl1 MFI on NKs (right).

(B-C) Microbiome analysis of feces from mice fed control or fiber diet. PCoA using weighted Unifrac (B); relative abundance of phyla after diet and before tumor implantation (C).

(D) Transkingdom network analysis of host phenotype and differentially abundant microbes (DAMs) from mice fed control or fiber diet. Nodes = host phenotypes (square) or DAMs (circle); edges = Spearman correlation.

(E) Bipartite betweenness centrality (BiBC) score calculated between DAMs and phenotypes for each DAM in D.

(F-G) GF mice monocolonized or not with Akk and implanted with EL4 tumors. MP characterization (F); tumor growth curve (G).

(H-I) LCMS quantification of indicated cyclic dinucleotides (CDNs) in Akk cell pellets (H) or in cecum content of tumor-bearing SPF or mice from F (I).

(J) Fold change of IRF3 activation in STING reporter cells stimulated with cdAMP (3 ug/ml), heat-killed Akk (MOI 100), or Akk spent medium (1:10) in presence or absence of STING inhibitor H-151.

(K) EL4 tumor-infiltrating cells from mice from F stimulated ex vivo with cdAMP. Fold change of Ifnb1⁺ Mo (left) and Xcl1 MFI on NKs (right).

Data from 2 (A) or 3 (B, C, J) experiments combined or one representative of 2 (F, G) or 3 (H) experiments. Dots in H and J depict technical replicates. n=5–8/group/exp (F, G, K), n=4–10/group/exp (A, I), n=8–15 mice/group/exp (B, C). Data shown as mean +/− SEM, *p<0.05, **p<0.01, ****p<0.0001. See also Figure S5 and Table S5.

Akkermansia muciniphila triggers the IFN-I–NK–DC axis in the TME

To evaluate changes in microbiota composition induced by FD, we performed fecal 16S rRNA sequencing using mice obtained from the same vendor but bred in two different facilities (MD or NY). Feeding FD was sufficient to overcome the initial facility differences and resulted in similar microbiota compositions (Figure 5B). Mice fed FD had reduced alpha diversity, increased relative abundance of phyla Verrucomicrobia and Proteobacteria, and reduced Firmicutes compared to control diet (Figures 5C, S5D, and S5E). Additionally, FD animals were enriched in order-level taxa such as Enterobacteriales, Verrucomicrobiales, and Betaproteobacteriales, and reduced in Clostridiales and Lactobacillales (Figure S5F). Amongst 333 amplicon sequence variants (ASVs), we found 17 differentially abundant microbes (DAMs) higher in FD and 64 DAMs higher in control diet consistent across three independent experiments (Figures S5G, S5H, and Table S5). Next, we employed transkingdom network analysis (Rodrigues et al., 2018) to construct a correlation network using DAMs altered by FD prior to tumor implantation to predict DC infiltration and tumor burden. We created a transkingdom network whose largest connected component consisted of 13 microbial nodes and 4 phenotypic nodes (Figure 5D). Notably, ASVs from the genus Akkermansia, Escherichia-Shigella, and Enterococcus positively correlated with cDC1 and cDC2 infiltration while Lactobacillus was negatively correlated with the same cell populations. Moreover, Akkermansia negatively correlated with tumor burden (Figure 5D), which is in agreement with studies associating the presence of Akkermansia with better tumor control (Li et al., 2020a; Routy et al., 2018).

We then asked if specific microbes were sufficient to induce the immune changes in the TME. To identify causal microbes of the FD phenotype, we used bipartite betweenness centrality (BiBC) analysis (Dong et al., 2015). This analysis is a way of measuring the influence of an individual node (e.g microbe) in the network, allowing us to identify microbial regulators of tumor immune phenotypes independently of their relative abundance. We found that Akkermansia had the highest BiBC score (Figure 5E), which places it as the top microbial candidate to mediate the DC and tumor growth phenotypes regulated by FD. To test the predicted modulatory effects of Akkermansia in this system, we monocolonized GF mice with Akkermansia muciniphila (Akk) before tumor implantation (GF+Akk). We found a significant increase in tiDCs, a clear trend towards higher cDC1s, and a significant reduction in the pro/antitumor Mac ratio in GF+Akk compared to GF mice (Figure 5F). In addition, Akk improved tumor growth control as observed in SPF animals fed FD (Figures 5G). Compared to Akk, monocolonization of GF animals with Lactobacillus reuteri led to lower intratumoral cDC1 content, higher pro/antitumor Mac ratio, and worsening of tumor growth control (Figure S5I), supporting our predictions from the transkingdom network analysis.

The data above shows that Akk favorably remodels the TME. Our model would predict that this occurs via STING-IFN-I pathway. To test this, we first evaluated the ability of Akk to produce CDNs. To do so, we used liquid chromatography-mass spectrometry (LCMS) to measure bacterial-derived CDNs (c-di-AMP, c-di-GMP, and c-GAMP). We found that Akk produced all three CDNs in culture, however, cdAMP was the dominant molecular entity (Figure 5H). We next analyzed mouse cecum content from SPF, GF, or GF+Akk. As expected, the three CDNs were present in samples from SPF mice while none could be detected in GF (Figure 5I). Notably, monocolonization of GF mice with Akk rescued cdAMP levels close to those found in SPF with little to no change in c-di-GMP and c-GAMP (Figure 5I), confirming that Akk predominantly produces cdAMP in vivo as well.

To test the biological activity and potential to induce IFN-I of Akk-derived cdAMP we co-cultured Akk or its spent medium with STING-IRF3 reporter cells in the presence or absence of a selective STING inhibitor. Both heat-killed Akk and the spent supernatant induced STING-mediated IRF3 activation (Figure 5J). We then investigated if, as in the case of FD, Akk also modulates the IFN-I–NK axis in the TME in vivo. We found that GF+Akk had higher frequency of Ifnb1 producing Mo and increased Xcl1 expression by NKs in the TME. Moreover, Akk monocolonization improved the impaired response of GF mice to oxa – which as we showed above depends on IFN-I and STING signaling (Figures 5K and S5J). Altogether, these results provide a direct link between the presence of Akk and the activation of the IFN-I–NK–DC axis in the TME to promote antitumor responses.

The predicted intratumoral Mo–IFN-I–NK–DC interactions correlate with response to ICB in melanoma patients

Recent studies have indicated that microbiota affects response to ICB in cancer patients (Sepich-Poore et al., 2021), but how this happens has not been shown. Therefore, to determine the translational relevance of our findings, we analyzed the innate immune signatures predicted from our preclinical studies in whole tumor RNAseq data from a discovery cohort of melanoma patients treated with ICB (Helmink et al., 2020). Consistent with our data, we found a highly significant positive correlation between IFN-I gene signature and the signatures for classical Mo, NKs, and DCs (Figure 6A). In addition, we observed a significant positive correlation between the signatures for NKs and DCs (Figure 6A). Also, the gene signature for NKs and the DC-recruiting chemokines (XCL1, XCL2, and CCL5) positively correlated with each other and with the signature for cDC1 (Figure S6A), which agrees with previous reports (Barry et al., 2018; Bottcher et al., 2018).

Figure 6. Monocytes and the IFN-I–NK–DC axis correlate with response to ICB in melanoma patients.

Analysis of tumor RNAseq data from melanoma patients treated with ICB. Indicated signatures described in Methods under Patient tumor RNAseq analysis.

(A-B) Discovery cohort tumor samples from multiple timepoints representative of 10 responder (R) and 12 non-responder (NR) patients. Pearson correlations (A); signatures’ Z-score means in R (n=27) and NR (n=35) tumor samples (B).

(C-D) Validation cohort tumor samples post-treatment (n=84 patients) with 45 R and 38 NR. Signatures’ Z-score means in R and NR tumors (C); overall survival of patients after ICB treatment stratified by median expression of the indicated signatures (log-rank p-value shown). Truncated violin plots show median with quartiles, *p<0.05, **p<0.01, ***p<0.001. See also Figure S6.

We next interrogated how these correlations related to response to therapy. Remarkably, the gene signatures for the cell types and cytokines/chemokines that we found significantly decreased in mice lacking microbiota, were similarly reduced in non-responder (NR) as compared to responder (R) patient tumors: classical Mo, type I IFN, NKs, chemokines (XCL1, XCL2, and CCL5), DCs, cDC1, IL-15/IL-15RA, IL-18, and IFNG (Figures 6B, and S6B). In addition, we found that NR patients had a reduced CD8 T cell signature compared to R patients and that intratumoral cDC1 and CD8 T cell signatures positively correlated with each other (Figures S6A and S6C). CIBERSORT analysis to infer immune cell proportions showed an enrichment in Mac in the TME from NR patients (Figure S6D).

We observed similar results in a larger validation cohort of advanced melanoma patients treated with ICB (Liu et al., 2019) (Figures 6C, S6E, and S6F). Moreover, we found that high expression of the following gene signatures was significantly associated with improved patient overall survival after treatment: classical Mo, IFN-I, chemokines, DCs, cDC1, IL15RA (Figures 6D and S6G). Although higher expression of NK signature was not significantly associated with better survival, we observed association with the chemokines produced by them, emphasizing the importance of NK function (Figures 6D and S6G). Notably, the qualitative differences that we observed in the TME of R and NR patients are fully recapitulated in our preclinical models, supporting the notion that microbiota influences patient response to ICB via reprogramming the innate immune TME.

Patient microbiota regulates IFN-I and shapes the MP landscape in the TME to promote response to ICB

To establish a causal relationship between the TME immune compositional differences of R and NR patients and the microbiota, we performed FMT giving feces from melanoma patients treated with ICB to GF mice. Samples from three R and three NR donors were individually transferred into different GF cohorts before implantation with BP melanoma (Figure 7A). Mo in the TME of animals receiving R FMT acquired a stimulatory phenotype and differentiated into inflammatory Mac characterized by high levels of MHCII, intermediate F4/80 and CD68 expression (Mo2 and Mac1) resembling those observed in SPF. Conversely, mice receiving NR FMT showed skewing of MPs towards Mac expressing low MHCII and high levels of F4/80, CD68, and PDL1 (Mo3 and Mac2) (Figures 7B, 7C, and S7A), similar to GF (Figures 1C–1H and 2C–2G). NR FMT mice also had significantly reduced intratumoral DCs and Mo, increased pro/antitumor Mac ratio, and a trend towards decreased NKs (Figures 7D, 7E and S7B) as we observed in GF and Ifnar1^−/− animals. Furthermore, NR FMT tumors had reduced Ifnb1 expression (Figure 7E). This TME remodeling ultimately resulted in poor tumor growth control in NR FMT as compared to R FMT mice (Figure 7F). These results support a cause-effect relationship between patient microbiota, intratumoral IFN-I, and response to ICB.

Figure 7. Patient microbiota regulates IFN-I and shapes the MP landscape in the TME.

(A-F) FMT from 3 R or 3 NR patients individually transferred into GF mice implanted with BP tumors. Experimental design (A); UMAP projection of tumor-infiltrating cells analyzed by FACS (LiveCD45⁺) (B); pie chart of Mo and Mac clusters from B (C); fold change of MP proportions (D) or Ifnb1 expression (RT-qPCR normalized to Actb; E) referred to respective R of each experimental cohort (individual donors indicated by color and experimental cohort by symbol); tumor growth curve (F).

(G-I) Analysis of tumor RNAseq from FMT clinical trial including 3 trial-R and 6 trial-NR (Baruch et al., 2021). GSEA showing Interferon Alpha Response pathway enriched in trial-R vs trial-NR patients post-FMT (G); signatures’ Z-score means in trial-R and trial-NR tumors post-FMT (H); signature change after FMT (I).

Data from 3 experiments combined (D, E) or one representative of 3 (B, C, F) experiments. n=5–8/donor/exp (B-F). Data shown as mean +/− SEM, *p<0.05, **p<0.01, ****p<0.0001. See also Figure S7.

To further corroborate this causal relationship, we analyzed RNAseq data from a phase I clinical trial in which patients with anti-PD-1-refractory metastatic melanoma received FMT from patients who had achieved complete response followed by re-induction of anti-PD-1 (Baruch et al., 2021). We compared tumors from patients that responded to this treatment (trial-R) to those that didn’t respond (trial-NR) and examined the differences induced by microbiota transplantation (pre- and post-FMT). In line with our preclinical data, GSEA showed Interferon Alpha Response as one of the top pathways upregulated post-FMT in trial-R patients (Figures 7G and S7C). Moreover, the gene signatures for Mo, IFN-I, NKs, chemokines, DCs, cDC1, IL-15/IL-15RA, and CD8 T cells were higher in trial-R after but not before FMT (Figures 7H, and S7D). Relative to baseline levels (pre-FMT), these signatures increased post-FMT in trial-R tumors while they remained unchanged in trial-NR (Figure 7I). In contrast to tumor, we found no differences before and after FMT or between trial-R and trial-NR patients when we looked at the same signatures in gut samples from these patients (Figures S7E and S7F), confirming the tumor-specific microbiota effect. Interestingly, Akk, which mediated the IFN-I and immune phenotype in mice, showed a trend toward increased abundance in post-FMT fecal samples from trial-R compared to trial-NR patients (Figure S7G). This concurs with another clinical trial with a similar FMT-ICB design in which Akk was also associated with response to ICB post-FMT (Davar et al., 2021). Together, our findings provide a strong causal link between microbiota, intratumoral IFN-I–NK–DC axis, and response to ICB in cancer patients.

Discussion

There is abundant evidence that microbiota impacts the immune response to cancer (Sepich-Poore et al., 2021). However, our mechanistic understanding is incomplete, particularly as studies often explore other tissues such as lymph nodes, spleen, or gut, which may not necessarily reflect intratumoral occurrences. This lack of knowledge has hampered the development of microbiota-targeted therapies. Here, we uncovered a mechanism by which microbiota shapes the TME innate immune landscape to regulate antitumor immunity. We demonstrate that: 1) Microbiota-derived STING agonists (e.g. cdAMP) induce IFN-I production by intratumoral Mo, which triggers antitumor skewing of the TME. 2) These Mo regulate the recruitment and activation of NKs and subsequent NK–DC crosstalk. 3) If microbiota is unfavorably disrupted, the Mo–IFN-I–NK–DC cascade is halted and Mo differentiate into protumor Mac. 4) Microbiota modulation with FD, monocolonization with cdAMP-producing A. muciniphila, or systemic administration of cdAMP are sufficient to trigger this IFN-I pathway and improve antitumor response. 5) Transplant of microbiota from ICB-responder patients induces intratumoral IFN-I, remodels the TME, and favors response to ICB.

We found that Mo, via IFN-I production, are key regulators of innate immune cell dynamics in the TME, underscoring their potential as therapeutic targets to improve anticancer immunity. Interestingly, in non-cancer studies, IFN-I has also emerged as an important host factor mediating microbiota-immune interactions. However, unlike our observations in tumors, some reports in other physiological contexts identified plasmacytoid DCs as the microbiota-sensing and IFN-I producing cells via different signaling pathways (Antunes et al., 2019; Di Domizio et al., 2020; Schaupp et al., 2020; Steed et al., 2017; Stefan et al., 2020; Swimm et al., 2018; Winkler et al., 2020). Although it remains unexplored, it is possible that other commensal-derived products may also play a role in regulating IFN-I in cancer. Combined, our results and these studies highlight the plasticity of MPs that can respond to different microbiota-derived components activating a variety of signaling pathways.

Our findings from preclinical models as well as RNAseq analysis of the FMT clinical trial, show that microbiota-induced changes observed in the TME are not reflected in other tissues. Also, we show that systemic administration of a microbiota-derived STING agonist can specifically elicit changes in the TME. Nevertheless, the location at which microbial sensing occurs remains unknown. Several, non-mutually exclusive, scenarios are possible: sensing in the bone marrow at the precursor level; in the periphery (gut, circulation); or in the tumor itself. It has become increasingly apparent that tumors distant from the gut harbor their own microbiota (Nejman et al., 2020). In line with this, it has recently been shown that Bifidobacterium colonizes subcutaneous mouse tumors and promotes IFN-I production in response to anti-CD47 treatment (Shi et al., 2020b). In a different report, intratumoral administration of CDNs resulted in IFN-I-induced NK activation (Nicolai et al., 2020). These two studies suggest that intratumoral sensing of microbiota may occur. Nonetheless, while each of these scenarios could be acting independently, they may also work in conjunction whereby systemic sensing initially programs target cells and local signaling amplifies the effects.

Specific antitumor effects of microbiota have been associated with changes in T cell activation and/or recruitment into the tumor or tumor-draining lymph nodes (Matson et al., 2018; Routy et al., 2018; Shi et al., 2020a; Sivan et al., 2015; Tanoue et al., 2019; Vetizou et al., 2015). Some studies have shown direct effects of microbiota-derived metabolites on T cells (He et al., 2021; Mager et al., 2020). Nevertheless, these effects were still reliant on MPs (e.g. costimulation, antigen presentation, cytokines) for improved therapy efficacy. This observation underscores the importance of MPs even if microbial products can directly enhance antitumor T cell function. However, what regulates MP function in the TME remained unclear. Thus, our findings on how microbiota regulates intratumoral MPs fill a crucial knowledge gap not only in the understanding of the role of microbiota in antitumor immunity, but also the cues that control MPs.

Microbial taxa associated with response to ICB in cancer patients differed among different reports (Gopalakrishnan et al., 2018; Matson et al., 2018; Routy et al., 2018). It is likely that the microbiota impact is not dependent on single species and that diverse microbiota compositions as well as several microbial-derived products may result in the same downstream immunomodulatory effect. We demonstrate here that dietary manipulation of microbiota offers a clinically actionable approach to reprogram the TME. Feeding mice a diet enriched in the fiber pectin increased IFN-I production and altered intratumoral MPs to improve tumor growth control and ICB efficacy. In agreement with our observations, two recent abstracts suggested that fiber rich diets were associated with better post-therapy clinical outcomes in cancer patients (Spencer et al., Abstract 2838 AACR Annual Meeting 2019; Richard et al., Abstract 679 SITC Annual Meeting 2020). Notably, several bacterial taxa that have been linked to better ICB response in patients (e.g. Faecalibacterium, Enterococcus hirae, Bacteroides fragilis) are known to hydrolyze pectin (Tan and Nie, 2020). Pectin has been shown to stimulate the secretion of mucin in the gut (Popov et al., 2006) and feeding mucin to mice can reduce tumor burden (Li et al., 2020a). In line with those findings, we found that mucin-degrading Akkermansia – predicted by our network analysis and previously associated with ICB response in patients – produces cdAMP and can recapitulate the FD phenotype. While studies have described both local (i.e. gut) and systemic effects of dietary fiber (Tan and Nie, 2020), these mechanisms remain to be explored in-depth in cancer. Nevertheless, evidence suggests synergism between fiber and predicted bacterial regulators of response to treatment, which can be leveraged as pre- or probiotics to enhance therapy efficacy.

We found that microbiota tailored the TME MP network in several tumor models with distinct immune contextures, pointing to a general mechanism of microbiota immune regulation in cancer. More importantly, we extended our findings to melanoma patients and showed not only that the same innate immune signatures are associated with ICB response, but also that FMT from patient donors into mice is sufficient to induce IFN-I and recapitulate the immune compositional differences between R and NR patients. Analysis of clinical samples from the FMT-ICB trial confirmed that IFN-I induction and the compositional differences transferred by microbiota influence therapy efficacy in patients. Combined, these observations provide strong evidence supporting a causal link between microbiota, IFN-I, and response to ICB.

The consistency of our findings across several models as well as patients’ tumors, suggests that IFN-I and monocyte skewing lie at the root of the immune dysregulation that occurs in tumors and reveals the key role of microbiota in orchestrating this process. One of the most daunting obstacles to effecting response to cancer therapy is overcoming the immunosuppressive TME. Our study provides important insights into the mechanism by which microbiota programs antitumor TME and shall improve our ability to leverage microbiota in the design of cancer therapies.

Limitations of the study

An important question that remains unanswered from our study is where MPs receive microbiota-derived signals. We were unable to detect measurable levels of CDNs in tumor. This could be due to levels being below the detection limit of our method, rapid uptake by host cells, or absence in the tumor. Although we have focused on MPs, future studies should evaluate the relative contribution of myeloid vs lymphoid cells in the microbiota-mediated antitumor effects. While we demonstrate here that microbiota regulates IFN-I–NK–DC axis in the TME, the limited availability of paired gut microbiome and tumor RNAseq datasets made it difficult to establish strong associations between specific microbes and IFN-I phenotype in patients. It will be important to extend our findings to larger patient cohorts across tumor types. Despite these limitations, our study significantly expands our knowledge of microbiota regulation of antitumor immunity, and ties together elements associated with anticancer response into a holistic microbiota-orchestrated mechanism that can be harnessed to improve therapy efficacy.

STAR Methods

RESOURCE AVAILABILITY

Lead contact

Further information and requests for resources and reagents should be directed to and will be fulfilled by the lead contact, Romina S. Goldszmid (rgoldszmid@mail.nih.gov).

Materials availability

This study did not generate new unique reagents.

Data and code availability.

• Single-cell RNA-seq data have been deposited at GEO and 16S rRNA sequencing data have been deposited at SRA. Both datasets are publicly available as of the date of publication. This paper also analyzes existing, publicly available data. All accession numbers are listed in the key resources table.

• This study does not report original code.

• Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request.

KEY RESOURCES TABLE.

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
Anti-CD16/32 (2.4G2)	BioXcell	Cat# BE0307
Anti-PD-1 (RMPI-14)	BioXcell	Cat# BE0146
Rat IgG2a isotype (2A3)	BioXcell	Cat# BE0089
Anti-PD-L1 (6E11.mlgG1.WT)	Genentech	6E11.mlgG1.WT
Mouse IgG1 isotype (anti-HIV gp120)	BioXcell	Cat# BE0154
Anti-IFNAR1 (MAR1-5A3)	BioXcell	Cat# BP0241
Mouse IgG1 isotype (MOPC-21)	BioXcell	Cat# BP0083
Anti-CD4 (GK1.5)	BD Biosciences	Cat# 584298
Anti-CD11b (M1/70)	BD Biosciences	Cat# 741934
Anti-CD24 (M1/69)	BD Biosciences	Cat# 563545
Anti-CD45 (30-F11)	BD Biosciences	Cat# 612975
Anti-CD45.1 (A20)	BD Biosciences	Cat# 741517
Anti-CD45.2 (104)	BD Biosciences	Cat# 612778
Anti-CD49b (DX5)	BD Biosciences	Cat# 553858
Anti-CD64 (X54-5/7.1)	BD Biosciences	Cat# 741024
Anti-CD135 (A2F10.1)	BD Biosciences	Cat# 562537
Anti-CD274/PDL1 (MIH5)	BD Biosciences	Cat# 563369
Anti-Ly6C (AL-21)	BD Biosciences	Cat# 561237
Anti-Ly6G (1A8)	BD Biosciences	Cat# 563978
Anti-MHC II (M5/114.15.2)	BD Biosciences	Cat# 566086
Anti-NK1.1 (PK-136)	BD Biosciences	Cat# 551114
Anti-Siglec F (E50-2440)	BD Biosciences	Cat# 747316
Anti-Siglec H (440C)	BD Biosciences	Cat# 566581
Anti-TCR-β (H57-597)	BD Biosciences	Cat# 562839
Anti-Ter119 (TER-119)	BD Biosciences	Cat# 563850
Anti-CCR7 (4B12)	BioLegend	Cat# 120124
Anti-CD8a (53-6.7)	BioLegend	Cat# 100704
Anti-CD11b (M1/70)	BioLegend	Cat# 101261
Anti-CD19 (6D5)	BioLegend	Cat# 115509
Anti-CD86 (GL1)	BioLegend	Cat# 105043
Anti-CD103 (2E7)	BioLegend	Cat# 121409
Anti-CD206 (C068C2)	BioLegend	Cat# 141720
Anti-F4/80 (BM8)	BioLegend	Cat# 123147
Anti-Ly6C (HK1.4)	BioLegend	Cat# 128036
Anti-Ly6G (1A8)	BioLegend	Cat# 127645
Anti-NKp46 (29A1.4)	BioLegend	Cat# 137606
Anti-NK1.1 (PK136)	BioLegend	Cat# 108706
Anti-Siglec H (551)	BioLegend	Cat# 129603
Anti-XCR1 (ZET)	BioLegend	Cat# 148218
Anti-CD68 (FA-11)	BioLegend	Cat# 137010
Anti-CCR7 (4B12)	Thermo Fisher	Cat# 12-1971-83
Anti-CD8a (53-6.7)	Thermo Fisher	Cat# 13-0081-85
Anti-CD11b (M1/70)	Thermo Fisher	Cat# 47-0112-82
Anti-CD11c (N418)	Thermo Fisher	Cat# 35-0114-80
Anti-CD19 (eBio1D3)	Thermo Fisher	Cat# 35-0193-82
Anti-CD49b (DX5)	Thermo Fisher	Cat# 15-5971-82
Anti-CD103 (2E7)	Thermo Fisher	Cat# 17-1031-82
Anti-CD209a (MMD3)	Thermo Fisher	Cat# 50-2094-82
Anti-CD135 (A2F10)	Thermo Fisher	Cat# 46-1351-82
Anti-MHC-II (M5/114.15.2)	Thermo Fisher	Cat# 13-5321-82
Anti-NK1.1 (PK136)	Thermo Fisher	Cat# 56-5941-82
Anti-TCRγδ (eBioGL3)	Thermo Fisher	Cat# 15-5711-82
Anti-TCRβ (H57-597)	Thermo Fisher	Cat# 15-5961-83
Anti-Ter119 (TER-119)	Thermo Fisher	Cat# 35-5921-82
Anti-CD3 (145-2C11)	Thermo Fisher	Cat# 15-0031-83
Bacterial and virus strains
Akkermansia muciniphila	ATCC	BAA-83
Lactobacillus reuteri	Dr. Giorgio Trinchieri	N/A
Biological samples
Human stool specimens	University of Texas, MD Anderson Cancer Center	IRB PA15-0232
Chemicals, peptides, and recombinant proteins
Live Dead Fixable Blue Dead Cell Stain kit	Invitrogen	L34962
FOXP3/Transcription Factor Staining Buffer kit	Invitrogen	00-5521-00
Primaxin	Merck & Co	NDC 0006-3516-59
Vancomycin	Hospira	NDC 0409-4332-01
Neomycin	VetOne	NDC 13985-578-16
Oxaliplatin	Teva	NDC 00703-3986-01
HPLC grade Water	Fisher Scientific	Cat #: W6 4
HPLC grade Methanol	Fisher Scientific	Cat #: A456 4
HPLC grade Isopropanol	Fisher Scientific	Cat #: A461 4
HPLC grade Acetic Acid	Fisher Scientific	Cat #: A11350
HPLC grade Tributylamine	Acros Organics	Cat #: AC139321000
c-di-GMP	Invivogen	tlrl-nacdg
c-di-AMP	Invivogen	tlrl-nacda
3’3-cGAMP	Invivogen	tlrl-nacga
2’3-cGAMP	Invivogen	tlrl-nacga23-02
H-151	Invivogen	inh-h151
Critical commercial assays
Prime Flow RNA Assay	Thermo Fisher Scientific	88-18005-210
type 1 probe for Xcl1 AF647	Thermo Fisher Scientific	VB1-19578-PF
type 6 probe for Ccl5 AF750	Thermo Fisher Scientific	VB6-14424-PF
type 10 probe for Ifnb1 AF568	Thermo Fisher Scientific	VB10-3282108-PF
Cytokine & Chemokine Convenience 36-Plex Mouse ProcartaPlex™ Panel 1A	Invitrogen	EPXR360-26092-901
THP1-Dual KI-mSTING Cells	Invivogen	thpd-mstg
Deposited data
Mouse tumor scRNAseq	This paper	GSE181745
Mouse fecal 16s rRNA sequencing	This paper	PRJNA754385
Patient tumor RNAseq	Helmink et al., 2020	EGAD00001004352
Patient tumor RNAseq	Liu et al., 2019	https://static-content.springer.com/esm/art%3A10.1038%2Fs41591-019-0654-5/MediaObjects/41591_2019_654_MOESM3_ESM.txt
Patient tumor & gut RNAseq	Baruch et al. 2020	GSE162436
Experimental models: cell lines
EL4 lymphoma	ATCC	TIB- 39
MC38 colon carcinoma	Dr. Giorgio Trinchieri	N/A
TUBO breast carcinoma	Drs. Elda Tagliabue and Mario Colombo	N/A
BP (BRAFV600E/PTEN−/−) melanoma	Dr. Jennifer Wargo	N/A
Experimental models: organisms/strains
C57BL/6NTac germ-free (GF) mice	Gnotobiotic Facility of the Laboratory Animal Sciences Program of the Frederick National Laboratory.	In house
C57BL/6NTac	Taconic Farms Inc	B6-F and B6-M
BALB/cAnNCrl mice	Charles River Laboratories	Strain code 028
NCI B6-Ly5.1/Cr	Charles River Laboratories	Strain code 564
B6-Ifnar1 (Ifnar1 KO)	Cancer and Inflammation Program Core at the NCI Frederick	In house
C57BL/6J-Tmem173/J (STING KO)	Cancer and Inflammation Program Core at the NCI Frederick	In house
Oligonucleotides
Actb (B-actin) F and R	DOI: 10.2337/db17-1150	See table S7
Rn18S (18S rRNA) F and R	DOI: 10.1038/srep34345	See table S7
Flt3 F and R	PrimerBank: 26337657a1	See table S7
Xcl1 F and R	PrimerBank: 6678712a1	See table S7
Ifnb1 F and R	DOI: 10.1128/JVI.00013-07	See table S7
Ifna4 F and R	PrimerBank: 6754294a1	See table S7
Ifna5 F and R	Designed with PrimerBlast	See table S7
Ifna6 F and R	Designed with PrimerBlast	See table S7
Recombinant DNA
Software and algorithms
BD FACSDiva	BD Biosciences	https://www.bdbiosciences.com/en-us/instruments/research-instruments/research-software/flow-cytometry-acquisition/facsdiva-software
FlowJo (v.10.6.0)	BD Biosciences	www.flowjo.com
R package Cytofkit2 2.0.1	Chen et al., 2016	https://github.com/JinmiaoChenLab/cytofkit2.git
nSolver Data Analysis	NanoString Technologies	www.nanostring.com
Partek Genomic Suite	Partek	www.partek.com
GSEA_4.0.3	UC San Diego	https://www.gsea-msigdb.org/gsea/index.jsp
R package scatterplot3d_0.3-41	Uwe Ligges and Martin Mächler	https://cran.r-project.org/web/packages/scatterplot3d/index.html
R package ComplexHeatmaps 1.20.0	Gu et al., 2016	https://github.com/jokergoo/ComplexHeatmap
R package Seurat_3.1.4	Butler et al., 2018	https://satijalab.org/seurat/
R package monocle_2.14.0	Qiu et al., 2017	http://cole-trapnell-lab.github.io/monocle-release/docs/
R package DESeq2_1.30.1	Love et al., 2014	https://github.com/mikelove/DESeq2
R package fgsea_1.16.0	Korotkevich et al., 2021	https://github.com/ctlab/fgsea
R package msigdbr_7.2.1		https://github.com/igordot/msigdbr
DADA2	Callahan et al., 2016	https://github.com/benjjneb/dada2
QIIME 2 2019.4	Bolyen et al., 2019	https://qiime2.org
iTOL	Letunic and Bork., 2019	https://itol.embl.de
GraphPad Prism 9.1.0	GraphPad Software	https://www.graphpad.com
Sciex Analyst v1.7	Sciex	https://sciex.com/products/software/analyst-software
Sciex MultiQuant v3.0.2	Sciex	https://sciex.com/products/software/multiquant-software
CIBERSORT	Newman et al., 2015	https://cibersort.stanford.edu/
Other
30% pectin fiber diet	Envigo	TD.170555
45%kcal fat Western diet	Envigo	TD.08811
Waters Atlantis T3 Column	Waters Corporation	Cat #: 186003722

EXPERIMENTAL MODEL AND SUBJECT DETAILS

Mice

C57BL/6NTac germ-free (GF) mice were bred and maintained in the Gnotobiotics Facility of the Laboratory Animal Sciences Program of the Frederick National Laboratory. C57BL/6NTac specific-pathogen-free (SPF) controls were obtained from Taconic Farms Inc. BALB/cAnNCrl and B6-Ly5.1/Cr mice were obtained from Charles River Laboratories (CRL). B6-Ifnar1 (Ifnar1 KO) and C57BL/6J-Tmem173/J (STING KO) mice and their respective wild-type controls were bred and maintained in a specific pathogen free environment in the Cancer and Inflammation Program Core at the NCI Frederick. All animal studies were approved by the Institutional Animal Care and Use Committee of National Cancer Institute and were conducted in accordance with the IACUC guidelines and the National Institutes of Health Guide for the Care and Use of Laboratory Animals. 8 to 10-weeks-old mice from both genders were randomly distributed into the experimental groups for this study. Chow and drinking water were provided to mice ab libitum. When not specified and for control of modified diets, mice were fed NIH-31 Open Formula Autoclavable diet. For the patient fecal matter transplant (FMT) experiment performed at The University of Texas MD Anderson Cancer Center, the study was approved by the Institutional Animal Care and Use Committee (IACUC) at The University of Texas MD Anderson Cancer Center, in compliance with the Guide for the Care and Use of Laboratory Animals. B6 germ-free mice were purchased from and transported by the gnotobiotic facility of Baylor College of Medicine (Houston, TX). Mice were housed at the MD Anderson Cancer Center gnotobiotic mouse facility.

Human subjects

For patient fecal sample collection, six stage IV M1a-d melanoma patients seen at the University of Texas MD Anderson gave informed consent during their clinic visit. IRB protocol PA15–0232 was reviewed by the Institutional Review Board IRB1 Office of Protocol Research at the University of Texas MD Anderson in accordance with the Institutional Review Board (IRB) Policy for Continuing Review of Research and the federal regulations governing human subjects research [45 CFR 46.109(e) and 21 CFR 56.109(f)]. All patients received immune checkpoint blockade (ICB) treatment. Response was defined as achieving complete response or partial response by RECIST 1.1 criteria. Three responder (R) and three non-responder (NR) patients were used as donors for fecal matter transplant in this study. Mean age 60.6 years (49–76). Gender: female/male 3/3 (50/50%). Mean BMI 28.16 kg/m² (22–41). Stage: IV M1a 1 (16.66%), IV M1b 3 (50%), IV M1d 2 (33.33%).

Bacteria

Akkermansia muciniphila (ATCC BAA-83) was inoculated into BD Brain Heart Infusion (BHI) broth (Millipore, 53286) supplemented with 0.4% mucin (Worthington Biochemical Corporation, LS002976) and Remel vitamin K/Hemin solution (Thermo Fisher Scientific, cat: R450951) and incubated under anaerobic conditions at 37 °C for 24 h. Lactobacillus reuteri, kindly provided by Dr Giorgio Trinchieri (CCR-NCI-NIH), was inoculated into a BD Difco Lactobacilli MRS broth (BD, Cat: 288130) supplemented with Remel vitamin K/Hemin solution (Thermo Fisher scientific, cat: R450951) and incubated under anaerobic conditions at 37 °C for 24 h. Cell density was estimated using OD600 of 1 = 8 × 10⁸ cells/ml before freezing in PBS +25% glycerol at 10⁹ cells/0.2mL to use directly in experiments. All handling of bacteria was performed in a Whitley A35 Anaerobic Workstation and all reagents were vented in an anaerobic atmosphere for at least 24 hours prior to use to limit contact of anaerobic bacteria to oxygen.

Cell lines

C57BL/6-derived EL4 lymphoma and MC38 colon carcinoma (female) were grown in 10% FBS RPMI medium supplemented with 1mM pyruvate, 2mM L-glutamine and 0.1mM non-essential amino acids at 37 °C 5% CO₂ in air atmosphere. C57BL/6-syngeneic BP (BRAF^V600E/PTEN^−/−) melanoma cells were grown in 10% FBS DMEM medium supplemented with L-glutamine 37 °C 5% CO₂ in air atmosphere. TUBO breast carcinoma cells (female) established from a Balb/c-neuT spontaneous lobular carcinoma, kindly provided by Drs. Elda Tagliabue and Mario Colombo (Fondazione IRCCS “Istituto Nazionale dei Tumori”, Milan, Italy), were grown in 10% FBS RPMI medium supplemented with L-glutamine 37 °C 5% CO₂ in air atmosphere. All cell lines used for in vivo tumor models were tested for specific mouse pathogens (MTBM test) but were not authenticated. THP1-Dual KI-mSTING Cells (cat thpd-mstg, InvivoGen) were cultured following manufacturer instructions and used for in vitro assays only.

METHOD DETAILS

Tumor implantation and monitoring

B6 mice were shaved and injected in the right flank subcutaneously with 5×10⁵ EL4 lymphoma, 2×10⁵ MC38 colon carcinoma, or 8×10⁵ BP melanoma cells. BALB/c mice were injected into the right mammary fat pad with 5×10⁵ TUBO breast carcinoma cells. For tumor growth kinetic and therapy studies, tumor sizes were measured three times/week and animals were euthanized at endpoint. Tumor volume (mm³) was calculated using the following formula: $Volume=\frac{(longestdiameter)\ast {(shortestdiameter)}^2}{2}$

Anti-tumor therapy and other treatments

For anti-PD-1 efficacy studies, MC38 tumors were treated with anti-PD-1 (BioXCell, BE0146) or isotype control (BioXcell, BE0089) when tumors reach approximately 50mm³. Antibodies were administered intraperitoneally (IP) at a dose of 200ug per mouse every three days for a total of 3 doses. For anti-PD-L1 efficacy studies, MC38 tumors were treated with anti-PD-L1 (kindly provided by Genentech, 6E11.mlgG1.WT) or isotype control (BioXcell, BE0154) when tumors reach approximately 100mm³. Antibodies were administered intraperitoneally (IP) at a dose of 200ug per mouse every three days for a total of 3 doses. For response to chemotherapy, EL4 tumor bearing mice were injected intraperitoneally once with 10mg/kg oxaliplatin when tumors reach 100–200mm³ size, around a week after tumor implantation.

For IFNAR-1 blockade experiments, EL4 tumor bearing mice were injected intraperitoneally with 0.5mg of IgG1 isotype control (BioXCell, MOPC-21) or anti-IFNAR-1 antibody (BioXCell, MAR1–5A3) every three days starting at day four post tumor implantation for a total of four injections. For in vivo administration of c-di-AMP (InvivoGen), mice were injected IP with 25ug diluted in PBS 72 hours and 3 hours prior to euthanasia and tissue harvesting.

Microbiota perturbations

For modified diets, 30% pectin fiber diet (Envigo, TD.170555) and 45%kcal fat Western diet (Envigo, TD.08811) were given starting 2–3 weeks prior to tumor implantation and continuing throughout the study. For fiber diet, we used mice obtained from the same vendor and with the same genetic background but bred in two different facilities (MD or NY), which allowed us to study the effect of dietary intervention on animals with different initial microbiota compositions. For depletion of microbiota in SPF mice, a cocktail containing 0.5g/L Primaxin, 0.5g/L Vancomycin, and 0.7g/L Neomycin was provided in autoclaved drinking water (changed every other day) starting 3 weeks before tumor implantation and continuing throughout the study. For single antibiotic experiments, 0.5g/L Vancomycin was provided in drinking water (changed weekly) following the same regimen. When indicated, fecal matter transplants were performed using cecum from mice fed 30% fiber diet or standard chow as control, to mice raised under GF conditions. For this, cecum contents of donor mice were collected in anaerobic conditions, filter through a 100um cell strainer and stored at −80°C in 25% glycerol PBS until use. Mice received 200ul of fecal slurry via oral gavage three times in one week and rested one week before tumor implantation.

Flow cytometric analysis of leukocytes

Tumors were harvested from mice 7–9 days post-implantation or when they reached the 200–400 mm³ depending on the model. To obtain single cell suspensions, tumors were mechanically disrupted prior to enzymatic digestion with 0.5% FCS RPMI containing 1mg/ml DNase (Roche, 10104159001) and 200U collagenase IV (Gibco, 17104–019) for 1 hour at 37C. The digested tissue was then passed through a 40um filter followed by 5min of red blood cell lysis with ACK. Spleens and lymph nodes were injected with 1mL and 0.5mL, respectively, of the same digestion media and incubated for 25 min at 37C before passing through a 40um filter and performing of red blood cell lysis with ACK. Single cell suspensions were resuspended in PBS and stained with LIVE/DEAD Fixable Dead Cell Stain kit (Invitrogen) followed by the staining with the corresponding antibody cocktail prepared in Brilliant Stain Buffer (BD Biosciences, 563794). Fc receptors were blocked with anti-CD16/32 antibody (2.4G2, BioXcell) and the following anti-mouse antibodies were used: CD4 (GK1.5), CD11b (M1/70), CD24 (M1/69), CD45 (30-F11), CD45.1 (A20), CD45.2 (104), CD49b (DX5), CD64 (X54–5/7.1), CD135 (A2F10.1), CD274/PDL1 (MIH5), Ly6C (AL-21), Ly6G (1A8), MHCII (M5/114.15.2), NK1.1 (PK-136), SiglecF (E50–2440), SiglecH (440C), TCR-β (H57–597), Ter119 (TER-119), all purchased from BD Biosciences; CCR7 (4B12), CD8a (53–6.7), CD11b (M1/70), CD19 (6D5), CD86 (GL1), CD103 (2E7), CD206 (C068C2), F4/80 (BM8), Ly6C (HK1.4), Ly6G (1A8), NKp46 (29A1.4), NK1.1 (PK136), SiglecH (551), XCR1 (ZET), CD68 (FA-11), all purchased from BioLegend; CCR7 (4B12), CD8a (53–6.7), CD11b (M1/70), CD11c (N418), CD19 (eBio1D3), CD49b (DX5), CD103 (2E7), CD209a (MMD3), CD135 (A2F10), MHC-II (M5/114.15.2), NK1.1 (PK136), TCRγδ (eBioGL3), TCRβ (H57–597), Ter119 (TER-119), CD3 (145–2C11), all purchased from Thermo Fisher. When biotinylated antibodies were used, cells were subsequently incubated with fluorochrome-conjugated streptavidin. For intracellular markers, cells were fixed and permeabilized using FOXP3/Transcription Factor Staining Buffer kit (eBioscience) according to the manufacturer’s instructions after surface staining and incubated with the corresponding antibodies.

Samples were acquired in a FACS Symphony A5 running under BD FACSDiva software (BD Biosciences) and data were analyzed with FlowJo software (v.10.6.0). Further analysis was performed with the R package Cytofkit2 2.0.1 (J. Chen’s laboratory) (Chen et al., 2016) as described (Araya and Goldszmid, 2020). The preliminary gating strategy for all analysis is: (1) FSC-A vs SSC-A for debris exclusion; (2) FSC-H vs FSC-A for doublet exclusion; (3) SSC-A vs Live/Dead for dead cell exclusion; (4) SSC-A vs CD45 for positive cells (tumor infiltrating leukocytes). t-Distributed Stochastic Neighbor Embedding (t-SNE) or Uniform Manifold Approximation and Projection (UMAP) algorithm was run with a concatenation of all samples per group simultaneously. For cluster analysis, Rphenograph was selected in the same package (k = 30) and the clusters were annotated based on expression levels of the markers of interest. Clusters representing the populations of interest were selected and quantified using FlowJo. Cell populations were defined as indicated in Supplementary Table 6. Frequencies from total leukocytes (Live CD45⁺ cells) were determined and absolute numbers per mg of tumor were calculated.

Detection of mRNA by flow cytometry

Detection of mRNA was done by PrimeFlow RNA Assay (ThermoFisher) using probes type 1 for Xcl1 AF647 (ThermoFisher, VB1–19578-PF), type 6 for Ccl5 AF750 (ThermoFisher, VB6–14424-PF) or type 10 for Ifnb1 AF568 (ThermoFisher, VB10–3282108-PF) according to the manufacturer’s instructions. For ex vivo stimulation with c-di-AMP, single cell suspensions of total tumor infiltrating cells were incubated at 37C, 5% CO₂, for 3 hours with or without 100ug/ml c-di-AMP (InvivoGen) in RPMI supplemented with 10% FBS, 1mM sodium pyruvate, 10mM HEPES, 50uM 2-mercaptoethanol, 0.1mM non-essential amino acids, and 2mM L-glutamine.

NanoString gene expression analysis

Single cell suspensions of EL4 tumors from SPF and GF mice were sorted as LiveCD45⁺CD3^negCD19^negNK1.1^negLy6G^neg with additional markers for each population of interest: dendritic cells (DCs; CD11c^hiMHCII^hi), monocytes (Mo; Ly6C^hiF4/80^low-neg), monocyte-derived macrophages (Mo-derived Mac; Ly6C^int-lowF4/80^pos). Sorted populations were resuspended in RLT buffer and analyzed using nCounter analysis system (NanoString Technologies) with a custom panel of 530 genes related to myeloid cells and inflammation according to manufacturer’s instructions (Goldszmid et al., 2012). For analysis of expression, each sample was normalized to technical controls and housekeeping genes: Asb7, Eif4h, Gsk3a, Nol6, Oaz1, Pex1, Poldip3, Rnf214, Sap130, Sdha. The data was then log2 transformed and batch-effect removal was performed between two independent experiments using Partek Genomics Suite. To verify cell identity, Gene Set Enrichment Analysis (GSEA_4.0.3) was used to associate gene expression patterns with ImmGen gene sets. One-way ANOVA was used to compare groups and p-values were corrected for multiple comparisons using Benjamini-Hochberg procedure (FDR). Principal component analysis (PCA) was performed in R using the singular value decomposition approach and visualized with the package scatterplot3d_0.3–41. Heatmaps were created in R using ComplexHeatmaps 1.20.0 (Gu et al., 2016). Hierarchical clustering was used to extract gene clusters which were subsequently assessed in Mac and DCs ImmGen populations (GSE122108) using mean expression of all genes in that cluster within the population of interest.

Single-cell transcriptomic analysis

Single cell suspension of EL4 tumors from SPF or GF mice were obtained as described above. Cell hashing with barcoded CD45 antibodies were used to pool biological replicates. Samples were then sorted as either LiveCD45⁺CD3^neg for tumor infiltrating leukocytes or LiveCD45⁺CD3^negCD19^negNK1.1^negTCRb^negTCRgd^negTer119^neg for TME myeloid cell compartment followed by library preparation and scRNA sequencing using the 10X Genomics protocol by the NCI CCR Single Cell Analysis Facility.

All analyses were carried out on NIH biowulf2 high performance computing environment. We used the following R packages for the analyses: Seurat_3.1.4, monocle_2.14.0, DESeq2_1.30.1, fgsea_1.16.0, and msigdbr_7.2.1. The Seurat package (Butler et al., 2018) was used for clustering and UMAP analysis. The low-quality cells were removed using the following criteria: 1) percent of mitochondrial genes above 15%; 2) doublet cells or negative cells with multiple hashing antibodies or no hashing antibody, respectively. For integrating SPF and GF data, we used the functions of FindIntegrationAnchors (anchor.features = 5000) and IntegrateData. Clustering analysis was performed with FindClusters (resolution = 0.5, Table S2 and Table S4). Differential gene expression between SPF and GF was performed using DESeq2 (Love et al., 2014). We aggregated cells based on hashing antibodies, which were used as biological replicates for DESeq2 analysis. Trajectory analysis was performed with Monocle 2 (Qiu et al., 2017). For analysis of mononuclear phagocytes (MPs), cluster gene expression from LiveCD45⁺ CD3^neg CD19^neg NK1.1^neg TCRb^negTCRgd^negTer119^neg samples were mapped to previously published mouse immune cell type signatures (Zilionis et al., 2019). Clusters mapping to monocytes (Mo), macrophages (Mac), or dendritic cells (DCs, excluding plasmacytoid dendritic cells) were extracted for downstream analysis. Mean expression of cluster specific genes (FDR < 0.1, |log2(fc)| > 0.5) were mapped to ImmGen populations (GSE122108) to identify cell types further. Dot plots created using node color as average expression across cells within cluster (hierarchical clustering parameter) and node size as percentage of cells within clusters expressing the specific gene. Differentially expressed genes (DEGs) were imported into Ingenuity Pathway Analysis (IPA) for further analysis. Common DEGs were visualized using Circos. The statistics from the DESeq2 analysis as rank list was used to perform GSEA analysis (FDR < 0.01) with FGSEA (Korotkevich et al., 2021) using the Hallmark gene sets from MSigDB (https://www.gsea-msigdb.org/gsea/msigdb/index.jsp). GSEA plots were generated with plotEnrichment function. To assess the possible gene regulation by different Interferons (IFNs), DEGs between SPF and GF MPs were contrasted with the Interferome data base (Rusinova et al., 2013). The number of genes regulated by type I, type II and type I/II IFNs were obtained and plotted for each population.

RT-qPCR analysis

Tumor pieces were mechanically disrupted in Buffer RLT Plus (Qiagen) supplemented with 2-mercaptoethanol followed by homogenization with QIAshredder (Qiagen) and total RNA extraction with the RNeasy Plus Mini Kit (Qiagen). RNA concentration and purity were assessed using DS-11 Series Spectrophotometer (DeNovix). Reverse transcription was performed using iScript Reverse Transcription Supermix for RT-qPCR (Bio-Rad). Subsequent cDNA was used for qPCR with Power SYBR Green PCR Master Mix (Applied Biosystems) in duplicate wells on QuantStudio 6 Flex Real-Time PCR System (Applied Biosystems) with the following cycle conditions: activation 95°C for 10 min; 40 cycles of denaturation at 95°C for 15 sec and annealing/extension at 60°C for 60 sec; melting curve for assessment of amplicons. Primer pairs (Table S7) were used at a final concentration of 500 nM.

Whole tumor protein expression

EL4 tumors were homogenized using 500ul Cell Lysis Buffer (Invitrogen, EPX-99999–000) for every 100mg of tissue with a rotor-stator homogenizer. Supernatant was extracted and total protein concentration was measured using the BCA Protein Assay Kit (Thermo Fisher Scientific, 23227). Samples were diluted to 10mg protein/mL with PBS and analyzed with the Cytokine & Chemokine Convenience 36-Plex Mouse ProcartaPlex™ Panel 1A (Invitrogen, EPXR360–26092-901) according to manufacturer’s instructions. Analyte concentrations were normalized to mean value of control group (fold change) within each independent experiment before combining for downstream analysis.

16S rRNA gene amplicon sequencing

Fresh fecal pellets collected from individual mice were stored at −80°C until analysis. One pellet per sample was loaded individually into wells of a 96- well plate for DNA extraction followed by 16S ribosomal RNA (rRNA) gene amplicon sequencing at the National Cancer Institute Microbiome and Genetics Core Facility using their standard workflow previously described (Bae et al., 2020; Li et al., 2020b). Briefly, extracted DNA underwent a two-step PCR protocol to amplify and barcode the 16S V4 region (515F-806R). Samples were sequenced on the Illumina MiSeq platform. Paired-end fastq reads were subsequently analyzed using DADA2 (Callahan et al., 2016) and QIIME 2 2019.4 (Bolyen et al., 2019).

Microbiome and transkingdom network analysis

Three independent experiments were used consisting of B6-Ly5.1/Cr mice from CRL obtained from one of two different facilities (MD or NY). The bacterial abundance table for each of the three independent experiments were relativized and quantile normalized. Mann-Whitney tests were calculated per experiments for the 333 amplicon sequence variants (ASVs) that were common in the three experiments. ASVs were considered to be differentially abundant microbes (DAMs) if they had the same direction of (Control vs Fiber) fold-change in all three experiments, with Fisher-pvalue and FDR of < 10%. Visualization of ASVs and DAMs was performed using iTOL (Letunic and Bork, 2019). We used transkingdom network analysis (Lam et al., 2018; Rodrigues et al., 2018; Zhang et al., 2020) to associate fecal DAMs prior to tumor implantation as a predictor for tumor phenotypes 1 week post implantation. Two independent experiments were used for which bacterial abundances and mouse tumor phenotype data (e.g. tumor weight, dendritic cells/mg tumor) were available. First, Spearman’s rank correlations were calculated between all pairs of microbes and phenotypes by pooling the fiber samples from both experiments. Correlations were retained if they had p-values < 15% and satisfied principles of causality (i.e. satisfied fold change relationship between the two partners in fiber vs control) (Yambartsev et al., 2016). Thereafter, correlations were calculated per experiment in fiber samples. Finally, edges obtained from pooling were retained if they had the same sign of correlation coefficient as those calculated per experiment and false positive edges due to pooling experiments were removed to make the transkingdom network. Bipartite betweenness centrality analysis (Dong et al., 2015) was used to identify “bottleneck” microbes connecting the microbial network to phenotypes. Microbes with high betweenness centrality and strongly connected to host phenotypes are more likely key regulators of host outcomes (e.g. changes in tumor weight).

Bacterial monocolonization

For GF monocolonization experiments, mice received an oral gavage of 0.2mL (10⁹ cells) of A. muciniphila or L. reuteri twice in one week followed by one week of break before tumor implantation. Prior to each dose, mice received an intraperitoneal injection of 0.3mg Famotidine (Fresenius Kabi USA, NDC 63323–738-09) and 0.02mg Sincalide (GoldBio, W-190–2).

Cyclic dinucleotide (CDN) measurement

For all liquid chromatography-mass spectrometry (LCMS) methods, LCMS grade solvents were used. Tributylamine was purchased from Millipore Sigma. LCMS grade water, methanol, isopropanol, and acetic acid were purchased through Fisher Scientific. All standards were purchased from InvivoGen. Spectral information for each CDN standard was acquired via direct injection in negative mode and collision energy was ramped. The top four most abundant signals for each CDN species were selected and instrument conditions were optimized to maximize signal. Signals were validated to avoid background noise by spiking standard into sample and the top two multiple-reaction monitoring (MRM) pairs were selected for each analyte. Declustering potential (DP), Entrance Potential (EP), Collision Energy (CE), Collision Cell Exit Potential (CXP) were set as follows:

Molecule	MRM pair	DP	EP	CE	CXP
c-di-AMP	657.0/328.0*	−70	−10	−40	−15
c-di-AMP	657.0/134.0	−70	−10	−50	−15
c-di-GMP	689.0/344.1*	−70	−10	−50	−15
c-di-GMP	689.0/150.0	−70	−10	−50	−15
c-GAMP	673/79*	−70	−10	−120	−15
c-GAMP	673/134	−70	−10	−70	−15

All source and collision parameters are in units of volts and * denotes those used for quantification. Samples were extracted by addition of 1:1 methanol and water on ice. Samples were centrifuged at 16k ×g for 20 minutes at 4 °C to remove macromolecules. Volume was reduced as necessary using an SPD130 Savant SpeedVac. All molecular analysis was performed using a series of targeted MRM methods. All samples were separated using a Sciex ExionLC™ AC system and analyzed using a Sciex 5500 QTRAP® mass spectrometer. All samples were separated across a Waters Atlantis T3 column (100Å, 3 μm, 3 mm X 100 mm) with a binary gradient from 5 mM tributylamine, 5 mM acetic acid in 2% isopropanol, 5% methanol, 93% water (v/v) to 100% isopropanol over seven minutes. Concentration was calculated using an 8 point standard curve. Putative unique fragment peaks were selected from 2’3’ and 3’3’ -c-GAMP however these signals were unable to distinguish between these isomers and a single set of MRMs was developed for c-GAMP species. However, c-GAMP was not detected in cecum from GF mice while it was present in that from SPF animals, indicating that the c-GAMP measured corresponds to the bacterial related 3’3’-c-GAMP and not the mammalian host associated 2’3’-c-GAMP.

STING pathway functional assay

STING pathway functional assessment was performed using THP1-Dual™ KI-mSTING Cells (thpd-mstg, InvivoGen) according to the manufacturer’s instructions. Briefly, 90,000 cells were incubated in a 96-well plate with or without the STING-inhibitor H-151 (inh-h151, InvivoGen) at 400 ng/ml for 2 hours at 37C 5% CO₂. Heat-killed A. muciniphila was prepared in PBS heating at 95°C for 30min. Heat-killed A. muciniphila at MOI 100 or 20ul of spent supernatant was added and incubated for 20–24 hours. When indicated, 3ug/ml of c-di-AMP was added to the culture as a positive control. IRF3 activation was quantified using QUANTI-Luc (rep-qlc1, InvivoGen) bioluminescent assay in a plate reader using a 100ms integration time. Measurements were normalized to their respective vehicle controls before analysis.

Patient tumor RNAseq analysis

FPKM normalized tumor RNAseq data was obtained from 10 R and 12 NR patients treated with nivolumab (anti-PD-1) alone or in combination with ipilimumab (anti-CTLA-4) (Helmink et al., 2020). Samples from multiple timepoints (baseline, on-treatment, surgery) were used collectively for downstream analysis. FPKM normalized tumor RNAseq data was additionally obtained from an independent validation cohort of advanced melanoma patients treated with nivolumab or pembrolizumab (anti-PD-1) (Liu et al., 2019). Only samples obtained from skin biopsies were retained to eliminate any potential inter-tissue variation (n=84). R patients were defined using best radiographic response (RECIST 1.1) criteria as complete response (CR), partial response (PR), or stable disease (SD) without progression for at least 180 days. NR patients were defined using RECIST 1.1 criteria as progressive disease (PD) or stable disease with progression within 180 days. Mixed response patients (n=1) were excluded from R/NR comparisons.

TMM normalized tumor and gut RNAseq data was obtained from 3 R and 6 NR refractory metastatic melanoma patients treated with fecal matter transplant (FMT) along with a reinduction of nivolumab (anti-PD-1) (Baruch et al., 2021). Differential gene expression between R and NR tumor samples at each timepoint (pre- or post-FMT) was performed using DESeq2. The statistics from the DESeq2 analysis as rank list was used to perform GSEA analysis (FDR < 0.01) using the Hallmark gene sets from MSigDB (https://www.gsea-msigdb.org/gsea/msigdb/index.jsp). GSEA plots were generated with plotEnrichment function.

Each cohort was analyzed separately but followed the same procedure for the quantification of immune signatures. Briefly, expression values were log2-transformed, and z-scores were calculated across all samples based on the mean expression value of signature genes. Genes missing from datasets were excluded. The following gene signatures were used as previously described (Bottcher et al., 2018): chemokines (XCL1, XCL2, CCL5), NK cells (NCR1, NCR3, KLRB1, CD160, PRF1), cDC1 (CLEC9A, XCR1, CLNK, BATF3), and CD8 T cells (CD8A, CD8B, CD3E). We additionally used the following signatures: classical monocytes (LYZ, ITGAL, VCAN, CSF1R, CD14, FCGR3A, CCR2, S100A9), dendritic cells (CD1C, CD1A, LY75, LTB, CD86, FLT3, CD80, CD40, SIRPA), and type I IFN (CXCL10, IFI16, IFI27, IFI30, IFI6, IFIH1, IFIT1, IFIT2, IFIT3, IFITM1, IFITM2, IFITM3, IFNA1, IFNA2, IFNA4, IFNAR1, IFNAR2, IFNB1, IFNE, IFNW1, IRF1, IRF2, IRF3, IRF5, IRF7, IRF9, ISG15, ISG20, JAK1, JAK2, OAS1, OAS2, SOCS1, STAT1, STAT2, STAT3, TMEM173, TYK2). For assessment of macrophage infiltration, CIBERSORT (Newman et al., 2015) was used to estimate cell fractions across samples.

Patient fecal microbiota transplantation

Fecal sample collection and FMT from R and NR metastatic melanoma patients treated with immune checkpoint blockade therapy at The University of Texas (UT) MD Anderson Cancer Center were performed as previously described (Gopalakrishnan et al., 2018). Fecal samples from three responder (R) or three non-responder (NR) human donors were individually transferred into independent cohorts of 5–8 GF mice per donor. 200μl cleared supernatant from 0.1 g/μl human fecal suspension was obtained using a 100 μm strainer and gavaged into mice for 3 doses over 1 week, followed by a 1-week break to allow for the establishment of the microbiota. Mice were then injected in the right flank subcutaneously with 8×10⁵ BRAF^V600E/PTEN^−/− (BP) cells. Tumor growth and survival were assessed, and tumors were harvested and processed as described above for analysis.

QUANTIFICATION AND STATISTICAL ANALYSIS

Statistical analysis

Statistical analyses were performed using GraphPad Prism software or R. Data were analyzed for normal distribution prior to comparisons. Outliers were detected using Grubbs’ test and excluded from the analysis. Values are presented as the mean ± SEM; individual values are presented as scatter plots with column bar graphs or truncated violin plots. Data in pie charts represent mean of samples in group. Statistical significance between two groups was determined using two-tailed unpaired Student’s t-test (parametric) or Mann-Whitney (non-parametric). For the comparison of multiple groups, statistical significance was determined by one-way ANOVA followed by Tukey’s post-hoc test (parametric) or Kruskal–Wallis test followed by Dunn’s test (non-parametric). For tumor outcome experiments, the Log-rank (Mantel-Cox) test was used for long-term survival analyses and mixed-effects analysis or two-way ANOVA were used for tumor growth measurements. For long-term survival analyses, mice were censored when they exit the study for reasons not related with the study endpoint. For the comparison of protein expression profiles, positive or negative mean differences were categorized and assessed using the Kappa statistic (Landis and Koch, 1977). The Benjamini–Hochberg procedure (FDR) was used to adjust p-values for multiple testing. All measurements shown are distinct biological replicates unless otherwise noted. For animal studies, sample size was defined based on past experience with the models. The minimum number of animals necessary to achieve the scientific objectives was used for ethical reasons. Animals were allocated randomly to each treatment group and different groups were processed identically.

Supplementary Material

1

Table S1. Differentially expressed genes of sorted intratumoral mononuclear phagocytes, Related to Figures 1 and S1

Differentially expressed genes from NanoString analysis of sorted populations from EL4 tumor of SPF or GF mice (FDR < 0.1, |log2(fc)| > 0.6; SPF/GF).

2

Table S2. Cluster specific genes from LiveCD45⁺CD3^negCD19^negNK1.1^negTCRb^neg TCRgd^negTer119^neg scRNAseq clusters, Related to Figures 1 and S1

Differentially expressed genes between each cluster vs all other clusters using only genes detected in at least half of the cells in either of the two populations. (FDR < 0.1, |log (fc)| > 0.5).

3

Table S5. Differentially abundant microbes in feces of mice fed fiber diet, Related to Figures 5 and S5

Differentially abundant microbes (DAMs) between feces of mice fed control or fiber diet (Fisher-p-value and FDR < 0.1). Values summarized across 3 independent experiments.

4

Table S4. Cluster specific genes from LiveCD45⁺CD3^neg scRNAseq clusters, Related to Figures 3 and S3

Differentially expressed genes between each cluster vs all other clusters using only genes detected in at least half of the cells in either of the two populations. (FDR < 0.1, |log (fc)| > 0.5).

5

Figure S1. Transcriptional analysis of EL4 tumor-infiltrating MPs from SPF or GF mice, Related to Figure 1

(A-E) NanoString analysis of sorted MPs. t-SNE projection of high-parametric flow cytometry analysis of leukocytes (LiveCD45⁺) with schematic for MP cell sorting (left) and overlay of selected surface expression markers (right) (A); gene set enrichment analysis (GSEA) of cells sorted from SPF mice compared to ImmGen gene sets (B); quantification of PC1 from Figure 1A (C); PCA of DCs (left) and top 20 genes in the loading of PC1 in positive and negative directions (right) (D); mean expression of genes in clusters from Figure 1B in ImmGen DC and Mac populations (E).

(F-H) scRNAseq analysis of LiveCD45⁺CD3^negCD19^negNK1.1^neg sorted cells. UMAP projection with labelled cell clusters and identification (F); mapping of cluster gene expression to major immune cell type signatures (# indicates clusters mapping to monocytes, macrophages, or DCs, excluding pDCs) (G); mean expression of cluster specific genes of MoMacDCs (FDR < 0.1, |log2(fc)| > 0.5) in selected ImmGen populations (H).

(I) RT-qPCR of Flt3 expression in whole tumor (normalized to Actb SPF mean).

Data from 2 experiments combined (A-E, I). n=2–3/group/exp (A-H), n=4–8/group/exp (I). Data shown as mean +/− SEM, ***p<0.001, ****p<0.0001

6

Figure S2. Differential gene expression and flow cytometry analysis of MPs from mice with or without microbiota, Related to Figures 1 and 2

Analysis of MPs from mice with (SPF; H2O) or without (GF; ABX) intact microbiota.

(A-B) Downstream analysis of DEGs between SPF vs GF EL4 tumors of MoMacDC scRNAseq clusters (FDR < 0.1, |log2(fc)| > 0.5). GSEA (FDR < 0.01) (A); number of type I and II interferon regulated genes enriched in each group per cluster (B).

(C) Absolute number of DCs and monocytes in EL4 tumors.

(D) Frequency of DCs in inguinal lymph node and spleen of non-tumor bearing C57BL/6 mice.

(E) Frequency of DCs in tumor-draining lymph node and spleen of EL4 tumor-bearing mice.

(F) Pie chart of CD45⁺ cells in indicated tumors from SPF mice (mean of one representative experiment).

(G-H) Median fluorescence intensity (MFI) of MHCII and CD86 in tumor-infiltrating DCs from mice harboring MC38 (G) or TUBO (H) tumors.

(I-J) Tumor weight of EL4 (I) or MC38, BP, and TUBO (J) in mice with or without microbiota.

Data from 2 (D spleen) or 4 (E) experiments combined or one representative of 2 (H, I, J MC38 and TUBO) or 4 (C) experiments. n=2–3/group/exp (A-B), n=3–5/group/exp (D, E, H), n=5/group (G), n=5–10/group/exp (F, I, J). Data shown as mean +/− SEM, *p<0.05, **p<0.01

7

Figure S3. Role of IFN-I in response to therapy and regulation of MPs in the TME, Related to Figure 3

(A) RT-qPCR of Ifnb1 expression in whole EL4 tumor (normalized to Actb SPF mean).

(B) Selected protein measurements from EL4 tumor lysates of WT or Ifnar1^−/− mice (normalized to WT mean).

(C) Ratio of pro/antitumor Mac in EL4 tumors of WT and Ifnar1^−/− mice.

(D)Frequency of cDC1 in EL4 tumors from mice treated with or without broad-spectrum antibiotics (ABX).

(E) Frequency of tumor infiltrating cDC1 in indicated- tumors of mice with (SPF; H2O) or without (GF; ABX) microbiota.

(F) Survival plot of EL4 tumor-bearing mice given or not ABX in drinking water and treated with oxaliplatin (oxa) or PBS.

(G) Survival plot of EL4 tumor-bearing mice administered with four injections of anti-IFNAR-1 or isotype control and treated with oxaliplatin (oxa) or PBS.

(H-I) scRNAseq of LiveCD45⁺CD3^neg sorted cells from EL4 tumors. UMAP projection with cell cluster identification (H); dot plot expression of selected proteins reduced in GF tumors shown in Figure 3A (I).

(J) Frequency of each cell population of total Ccl5⁺ CD45⁺ EL4 tumor-infiltrating cells.

(K) Spearman correlation of frequency of NKs and DCs in EL4, MC38, and TUBO tumors including mice with or without microbiota.

(L) Frequency of IFNγ producing NKs in EL4 tumors from SPF or GF mice.

Data from 2 experiments combined (G, J) or one representative of 2 (D, E TUBO, F) or 3 (C, E EL4) experiments. n=2–3/group/exp (H-I), n=3–5/group/exp (C, D, E), n=5/group/exp (A, B, J, L). Data shown as mean +/− SEM, *p<0.05, **p<0.01, ***p<0.001, ****p<0.0001

8

Figure S4. Response to systemic cdAMP administration and microbiota perturbations, Related to Figures 3 and 4

(A) RT-qPCR of type I IFN genes and DC recruiting chemokines (normalized to 18S) in EL4 tumors from GF mice treated with cdAMP or PBS (A).

(B) PCA of Mac from SPF or GF mice injected with PBS or GF + cdAMP using MFI of cell surface markers (left). Marker with highest loading in each direction along PC1 indicated. Quantification of PC1 in each group (right).

(C) Frequency of Mo and DCs in EL4 tumors from SPF mice treated with cdAMP or PBS.

(D) Experimental design of diet and antibiotics experiments.

(E-F) EL4 tumors from mice given control or fiber diet. Absolute number of DCs (E) and frequency and absolute numbers of cDC1 (F).

(G-H) EL4 tumors from mice given control or western diet. Frequency and absolute numbers of DCs (G); tumor weight (H).

(I) Frequency and absolute numbers of DCs and cDC1 in EL4 tumors from mice given vancomycin in drinking water.

Data from 3 experiments combined (E-F) or one representative of 2 experiments (A-B, G-H). n=4–5/group/exp (G-I), n=5/group/exp (C), n=4–8/group/exp (E-F), n=5–10/group/exp (A-B). Data shown as mean +/− SEM, *p<0.05, **p<0.01, ***p<0.001

9

Figure S5. Fiber diet and A. muciniphila trigger antitumor response, Related to Figure 5

(A) MC38 tumor growth of GF mice receiving FMT from control or fiber-fed mice.

(B) RT-qPCR of type I IFN genes in EL4 tumors from control or fiber mice (normalized to Actb control).

(C) EL4 tumor-infiltrating cells stimulated ex vivo with cdAMP from control or fiber mice. Fold change Ifnb1⁺ Mo cdAMP vs control.

(D-H) Analysis of fecal microbiota from control or fiber mice. Paired alpha diversity before and after 2–3 weeks on diet (D); relative abundance of phyla after diet and before tumor implantation (E); heatmap of order-level taxa in fecal samples before tumor implantation (significance using FDR < 0.05; *higher in control, ĥigher in fiber) (F); interactive Tree of Life (iTOL) visualization of amplicon sequence variants with fold change of differentially abundant microbes (DAMs) enriched in either control (brown) or fiber (green) diet (Fisher p-value < 0.01, FDR < 0.01) in circumference (each experiment indicated in stacked bars) (G); top 10 DAMs in each direction shown as last known taxonomy (H).

(I) GF mice monocolonized with L. reuteri or A. muciniphila and implanted with EL4 tumors. MP characterization (left) and tumor growth (right) (I).

(J) Tumor growth of EL4 tumors from GF mice monocolonized with A. muciniphila and treated with oxaliplatin (oxa).

Data from 3 experiments combined (D-H) or one representative of 2 experiments (C). n=4–5/group/exp (B), n=5–9/group/exp (C), n=8–15/group/exp (D-H), n=8/group/exp (A, I). Data shown as mean +/− SEM, *p<0.05, **p<0.01, ****p<0.0001

10

Figure S6. RNA-seq analysis of tumors from melanoma patients treated with ICB, Related to Figure 6

Analysis of tumor RNAseq data from melanoma patients treated with ICB. Indicated signatures described in Methods under Patient tumor RNAseq analysis section.

(A-B) Discovery cohort tumor samples from multiple timepoints representative of 10 responder (R) and 12 non-responder (NR) patients. Pearson correlations (A); gene counts (B) and signatures’ Z-score means for indicated signatures (C) in R (n=27) and NR (n=35) tumor samples; CIBERSORT estimated cell fraction of macrophages in melanoma patient tumors at baseline (D).

(E-G) Validation cohort tumor samples post-treatment (n=84 patients) with 45 R and 38 NR. Pearson correlations (E). signatures’ Z-score means in R and NR tumors (F); overall survival of patients after ICB treatment stratified by median expression of the indicated signatures (log-rank p-value shown) (G).

Truncated violin plots show median with quartiles, *p<0.05, **p<0.01, ***p<0.001, ****p<0.0001.

11

Figure S7. Analysis of RNA-seq data from melanoma patients treated with FMT and ICB, Related to Figure 7

(A) MFI of selected markers for clusters of BP tumor-infiltrating Mo and Mac from Figures 7B and 7C.

(B) Fold change of NK proportions from same tumors as Figure 7D and 7E referred to respective R of each experimental cohort (individual donors indicated by color and experimental cohort by symbol).

(C-F) Analysis of tumor and gut RNAseq from FMT phase I clinical trial including 3 trial-R and 6 trial-NR (Baruch et al., 2021). GSEA showing Hallmark pathways enriched in trial-R vs trial-NR patient tumors pre- or post-FMT (FDR < 0.01) (C); signatures’ Z-score means in trial-R and trial-NR tumors pre-FMT (D); gut signature change after FMT (E), signatures’ Z-score means in trial-R and trial-NR gut post-FMT (F)

(G) Frequency of A. mucinphila in patient feces 65 days after starting FMT treatment.

Data from 3 experiments combined (B) or one representative of 3 (A) experiments. n=5–8/donor/exp (A, B). Data shown as mean +/− SEM.

12

Table S6. Definitions for immune cell populations obtained by high-parametric flow cytometry, Related to STAR Methods

Table S7. Mouse primers for RT-qPCR, Related to STAR Methods

13

Table S3. Canonical Pathways between SPF and GF in intratumoral monocytes and macrophages, Related to Figures 1 and S1

Canonical Pathways from Ingenuity Pathway Analysis (IPA) of DEGs between SPF and GF (FDR < 0.1, |log2(fc)| > 0.5) in monocyte and macrophage clusters. Missing values indicates pathways absent or below significance threshold for that population.

Highlights:

• Microbiota-induced type I IFN programs antitumorigenic mononuclear phagocytes (MPs)

• Monocytes in the TME produce type I IFN in response to microbial STING agonists

• High-fiber diet induces typer I IFN, remodels MPs in the TME, and improves ICB efficacy

• Microbiota from ICB responders induces IFN-I and shapes the MP landscape in the TME