Research Article

Cross-reactivity between tumor MHC class I–restricted antigens and an enterococcal bacteriophage

See allHide authors and affiliations

Science  21 Aug 2020:
Vol. 369, Issue 6506, pp. 936-942
DOI: 10.1126/science.aax0701

Phages and cancer immunity

Gut bacteria are involved in the education of T cell immune responses, and the intestinal ecosystem influences anticancer immunity. Fluckiger et al. report microbial antigens that might cross-react with antigens associated with tumor cells. They found that a type of intestinal bacteria called enterococci harbor a bacteriophage that modulates immune responses. In mouse models, administration of enterococci containing the bacteriophage boosted T cell responses after treatment with chemotherapy or programmed cell death protein 1 (PD-1) blockade. In humans, the presence of the bacteriophage was associated with improved survival after PD-1 immunotherapy. A fraction of human T cells specific for naturally processed melanoma epitopes appeared to be able to recognize microbial peptides. This “molecular mimicry” may represent cross-reactivity between tumors and microbial antigens.

Science, this issue p. 936


Intestinal microbiota have been proposed to induce commensal-specific memory T cells that cross-react with tumor-associated antigens. We identified major histocompatibility complex (MHC) class I–binding epitopes in the tail length tape measure protein (TMP) of a prophage found in the genome of the bacteriophage Enterococcus hirae. Mice bearing E. hirae harboring this prophage mounted a TMP-specific H-2Kb–restricted CD8+ T lymphocyte response upon immunotherapy with cyclophosphamide or anti–PD-1 antibodies. Administration of bacterial strains engineered to express the TMP epitope improved immunotherapy in mice. In renal and lung cancer patients, the presence of the enterococcal prophage in stools and expression of a TMP–cross-reactive antigen by tumors correlated with long-term benefit of PD-1 blockade therapy. In melanoma patients, T cell clones recognizing naturally processed cancer antigens that are cross-reactive with microbial peptides were detected.

Unleashing immune responses against tumor-associated antigens through chemotherapy, radiotherapy, targeted therapies, or immune checkpoint inhibitors has formed the basis of successful cancer treatments (1, 2). The recent discovery that the gut microbiota affects the cancer-immune set point, thus influencing the clinical outcome of cancer therapies, has rekindled the concept that microbes or their products modulate not only intestinal but also systemic immunity (3, 4). Indeed, memory responses by interferon-γ (IFNγ)–secreting CD4+ and CD8+ T cells specific for Enterococcus hirae, Bacteroides fragilis, and Akkermansia muciniphila are associated with favorable clinical outcome in cancer patients (58), suggesting that microbe-specific T lymphocytes may contribute to antitumor immune responses. The mechanisms through which microbes trigger chronic intestinal inflammation and systemic autoimmune disease have not been resolved (9). The theory of “molecular mimicry” (1015) posits that T cells elicited by bacteria or viruses accidentally recognize autoantigens as they “escape” from self-tolerance–inducing mechanisms (such as clonal deletion or inactivation). Although major histocompatibility complex (MHC) class I– and class II–binding epitopes encoded by bacterial genomes may be immunogenic (1014), very few reports have demonstrated that microbe-specific CD4+ or CD8+ T lymphocytes attack normal or neoplastic tissues (1618).

Cyclophosphamide (CTX) is a chemotherapeutic drug that can induce the translocation of E. hirae from the gut lumen to the mesenteric and splenic immune tissues. This results in CD4+ and CD8+ T lymphocytes producing interleukin-17 (IL-17) and IFNγ and correlates with improved anticancer immune responses in mice (6, 19). Broad-spectrum antibiotics abolished the therapeutic efficacy of CTX unless E. hirae was supplied by oral gavage (6). When comparing a panel of distinct E. hirae strains (table S1 and fig. S1A) for their capacity to restore the antibiotic-perturbed anticancer effects of CTX, we found that only a few E. hirae isolates (such as 13144 and IGR11) were efficient to reduce MCA205 tumor size (Fig. 1, A and B) (6). Given that the therapeutic efficacy of the combination of CTX and E. hirae 13144 is abrogated by the depletion of CD8+ T cells or the neutralization of IFNγ (6), we screened the differential capacity of E. hirae strains to elicit memory T cell responses after priming of the host, measured as the ex vivo recall response (IFNγ secretion) of splenic CD8+ T cells against various E. hirae strains loaded onto dendritic cells (DCs) (fig. S2A). Whereas E. hirae 13144 triggered specific CD8+ T cell responses (that were not cross-reactive against irrelevant enterococci), E. hirae 708 and 13344 (two prototypic inefficient strains) (6) failed to do so (fig. S2A).

Fig. 1 Phage tail length TMP is a specific antigenic sequence in E. hirae 13144.

(A and B) C57BL/6 mice bearing MCA205 sarcomas were conditioned with broad-spectrum antibiotics (ATB) (streptomycin, colistin, ampicillin, vancomycin) for 3 days before performing oral gavages with E. hirae strain 13144 and intraperitoneal injections of CTX. (B) Tumor size was recorded for each mouse at sacrifice on day 25. Mean tumor sizes at sacrifice are depicted. (C to E). Naive C57BL/6 mice were conditioned with antibiotics, gavaged with indicated E. hirae strains and treated with CTX (C). (D) Day 11 purified CD8+ T splenocytes [from (C)] were restimulated ex vivo in a recall assay with bone marrow–derived DCs loaded with the indicated peptides (group 7 in table S2) to quantify IFNγ-secreting CD8+T cells. The number of IFNγ-producing cells per 2 × 105 CTL is depicted for three independent experiments. (E) H-2Kb/TMP1 (TSLARFANI) or H-2Kb/SIINFEKL tetramer-binding CD8+ splenocytes [from (C)] were detected by cytofluorometry at day 11. The percentages of tetramer-binding CTL in the CD8+ T cell gate are depicted for three independent experiments. Also, fig. S2D presents tetramer stainings in tumor-draining LNs. Each graph assembles results from two to three independent experiments containing groups of five to six mice. ANOVA statistical analyses (Kruskal-Wallis test): *P < 0.05, **P < 0.01, ***P < 0.001. The statistical report is in the supplementary materials.

To identify relevant T cell epitopes, we aligned the sequences of bacterial genes encoding putative cell wall and secreted proteins for immunogenic (13144) versus nonimmunogenic (708 and 13344) E. hirae strains, followed by the in silico identification of 13144-specific nonapeptides with strong affinity (<50 nM) for the MHC class I H-2Kb protein (table S2). Subsequently, we recovered splenic CD8+ T cells from mice that had been exposed to E. hirae 13144 and CTX (Fig. 1C), restimulated them in vitro with pools of potentially immunogenic nonapeptides from E. hirae 13144 to measure IFNγ production (table S2 and fig. S2B), and then split the most efficient pool (no. 7) into individual peptides (Fig. 1D). This approach led to the identification of one dominant epitope (one-letter amino acid code: TSLARFANI, abbreviation TMP1) in position 187 to 197 of the amino acid sequence of the bacteriophage tail length tape measure protein (TMP, 1506 amino acids) from a 39.2-kb prophage of E. hirae 13144 (Fig. 1D, fig. S3, and tables S2 and S3). Temperate and lytic bacteriophages are bacterial viruses that transfer virulence, antimicrobial resistance genes, and immunogenic sequences to new bacterial hosts with a strict specificity (20). The TMP protein, which contains a variable number of tandem repeats with highly conserved tryptophan and phenylalanine residues at fixed positions, is encoded by the genome of Siphoviridae phages (21, 22).

The 39.2-kb prophage (i.e., bacteriophage genome) encodes 65 genes, including one shared between all 18 E. hirae genomes and 38 specific to E. hirae 13144 (fig. S1B), encoding capsid, portal, and tail structures characteristic of Siphoviridae bacteriophages. The TMP1 epitope of the 39.2-kb prophage from E. hirae 13144 and the prophage fragment contained in E. hirae IGR11 showed 100% sequence identity (figs. S3 and S4A). Accordingly, E. hirae IGR11 was as efficient as E. hirae 13144 in reducing the growth of MCA205 sarcomas treated with CTX (Fig. 1, A and B). By contrast, the absence of a bona fide TMP1 epitope (observed in E. hirae 708 and 13344) (fig. S1B) and a mutation in position 3 of the TSLARFANI peptide (L→F observed in E. hirae ATCC9790) (fig. S4A) correlated with the lack of anticancer effects of these E. hirae strains (Fig. 1B) (6). Enzyme-linked immunosorbent spot (ELISpot) assays designed to detect peptide-specific IFNγ-producing T cells revealed that mice gavaged with E. hirae 13144 or IGR11 mounted a CD8+ T cell response against TMP1 (but not against the control peptides TMP2 and TMP3), but mice receiving E. hirae strains lacking TMP1 (strains 708 and 13344) or a strain possessing a mutated TMP1 (strain ATCC9790) were unable to stimulate a response (Fig. 1D). We used a fluorescent H-2Kb/TSLARFANI tetrameric complex [and its negative control H-2Kb/SIINFEKL binding to ovalbumin (OVA) specific CD8+ T cells] to detect the frequency and distribution of TMP1-specific cytotoxic T lymphocytes (CTLs) in naive and MCA205 sarcoma-bearing C57BL/6 mice. We observed a specific increase in splenic CD8+ T cells that recognized the TMP1 peptide (but not the OVA peptide SIINFEKL) at day 7 after treatment with CTX and gavage with E. hirae 13144 (Fig. 1E), as well as in tumor-draining lymph nodes (LNs) of tumor-bearing mice at day 14 after treatment with CTX and gavage with E. hirae 13144 (fig. S2, C and D). Splenic TMP1 (but not OVA)–specific (H-2Kb/TSLARFANI tetramer-positive) CTLs also increased in their frequency after gavage with E. hirae IGR11 (but not 13344 or ATCC9790) (Fig. 1E). The H-2Kb/TSLARFANI tetramer-positive CTLs were specifically enriched in the CXCR3+CCR9+ fraction of CD8+ T cells from secondary lymphoid organs (fig. S2C). Even in mice colonized with human fecal materials, CTX administration and oral gavage with E. hirae 13144 induced an anticancer effect (fig. S2E) and an expansion of H-2Kb/TSLARFANI tetramer-positive CTL in tumor-draining LNs at day 7 and in tumor beds at day 17 but were not detectable in mesenteric LNs (fig. S2, F and H). Hence, immunogenic E. hirae elicits an H-2Kb–restricted CTL response against the TMP-derived peptide TMP1/TSLARFANI.

To explore the capacity of TMP1-specific H-2Kb–restricted T cells to control the growth of MCA205 cancers, we subcutaneously (sc) immunized naive C57BL/6 mice with DCs loaded with heat-inactivated E. hirae 13144 (positive control), the naturally occurring TMP1/TSLARFANI peptide from 13144 and IGR11, the L→F mutant from E. hirae ATCC9790 (mut3) (Fig. 2A and fig. S4A), or other nonimmunogenic bacterial peptides (group 1 in fig. S2B). In this prophylactic setting, DCs pulsed with TMP1 (but not mut3) were as efficient as the whole E. hirae extract in reducing tumor growth (Fig. 2, B and C). Next, we explored whether the TMP1 peptide would be able to confer immunogenicity to the usually inefficient bacterium Escherichia coli strain DH5α in the therapeutic setting, in which antibiotic treatment is followed by gavage with different bacterial strains and CTX-based chemotherapy (Fig. 1A) (6). E. coli engineered to express TMP1 (fig. S5) was as efficient as E. hirae 13144 in restraining MCA205 tumor growth (fig. S4B and Fig. 2D) and eliciting tetramer-binding CTL in the spleen (Fig. 2E). By contrast, E. coli expressing an irrelevant sequence [encoding mouse enhanced green fluorescent protein (EGFP) protein], mut3, or mutant TMP1 bearing an S→A exchange in the anchor position 2 (mut2) (Fig. 2A) failed to induce such a cancer-protective immune response (Fig. 2, D and E).

Fig. 2 Prophylactic and therapeutic immunization using phage tail length TMP against murine sarcomas.

(A) Sequence of the immunogenic epitope TMP1 (TSLARFANI) with the engineered and naturally occurring mutations in positions 2 and 3, respectively. (B and C) Prophylactic vaccination: TLR3 ligand–exposed DCs were pulsed with peptides or heat-inactivated bacteria and then inoculated (sc) twice into mice. One month later, MCA205 sarcomas were implanted in the opposite flank, and tumor size was monitored [mean ± SEM are in (B), individual results are in (C)]. (D and E) Therapeutic vaccination: MCA205 tumor–bearing mice were treated with CTX and gavaged with E. hirae 13144 or E. coli (as shown in Fig. 1A) that were genetically modified to express the indicated peptides or EGFP as a negative control. (D) Tumor size at sacrifice (each dot representing one tumor per mouse) and (E) the frequency of H-2Kb/TMP1 tetramer-binding splenic CD8+ T cells were monitored (each dot representing one flow cytometric staining). Results are shown for 12 to 18 animals, gathered from two to three independent experiments. ANOVA statistical analyses (Kruskal-Wallis test) were used: *P < 0.05, **P < 0.01. The statistical report is in the supplementary materials.

To explore the mechanism by which TMP1 exerts its anticancer activity against MCA205 tumors in C57BL/6 mice, we investigated whether H-2Kb–restricted mouse tumor antigens with high identity to the TMP1 peptide (TSLARFANI) exist. Using the National Center for Biotechnology Information (NCBI) BLASTP suite, we found that the peptide (GSLARFRNI) belonging to the proteasome subunit beta type-4 (PSMB4) located at amino acid positions 76 to 84 shared a strong homology with TMP1 (seven out of nine amino acids with identical amino acids at the MHC class I anchoring positions 2 and 9) (Fig. 3A). We queried for potential neoepitopes of MCA205 but found no significant homology with TMP1, prompting us to focus on the nonmutated PSMB4 peptide. Some mouse tumors (such as MCA205 sarcomas and TC1 lung cancers) overexpress the PSMB4 antigen compared with their normal tissues of origin, whereas others (such as MC38 colon cancers) failed to do so (Fig. 3B). This correlates with the fact that MCA205 and TC1 tumors respond to the treatment with CTX+E. hirae 13144, whereas MC38 tumors do not (fig. S6, A and B). PSMB4 is an oncogenic driver involved in proliferation and invasion (23) in a variety of malignancies such as glioblastoma (24), melanoma (25), and breast cancers (26), associated with poor prognosis (23, 24, 26). CRISPR-Cas9–mediated genomic knock-in of the PSMB4 sequence in MCA205 cells, replacing GSLARFRNI by GALARFRNI (with an S→A exchange in position 2) or GSFARFRNI (with an L→F exchange in position 3 equivalent to mut3 of TSLARFANI) (fig. S7) significantly affected tumor growth kinetics (fig. S6, C and D), suggesting that this PSMB4 epitope contributes to the oncogenic activity of PSMB4. Whereas these knock-in mutations did not interfere with the efficacy of CTX treatment alone, they substantially blunted the anticancer effects of E. hirae 13144 (Fig. 3, C and D). We extended these findings to a second tumor model in which the anticancer effects of the combination of CTX+E. hirae 13144 were additive even in the absence of antibiotic-induced dysbiosis. Introducing a knock-in mutation in position 3 of PSMB4 into TC1 lung cancer cells again compromised the antitumor effects of CTX (Fig. 3E). Moreover, in the setting of PD-1 blockade, administration of E. hirae 13144 without prior conditioning with antibiotics reduced the growth of parental but not PSMB4-mutated MCA205 cancers (fig. S6E). These results support the idea that the TSLARFANI TMP1 peptide encoded by E. hirae 13144 induces T cell responses against the PSMB4-derived GSLARFRNI peptide across different tumor types and therapy modalities.

Fig. 3 Molecular mimicry between phage tail length TMP and the oncogenic PSMB4 epitope in murine tumors.

(A) Sequence alignment of the enterophage TMP1 peptide and a PSMB4 epitope with its two experimental mutants. (B) Relative (Rel.) expression of PSMB4 mRNA in MCA205 sarcoma, TC1 lung cancer, and MC38 colon carcinomas as compared with their healthy tissue of origin (mean ratio ± SEM, n = 3). (C and D) Therapeutic response of wild type (WT) versus knock-in mutants of MCA205 to CTX alone or in combination with immunogenic E. hirae strain 13144 (as outlined in Fig. 1A). Results are shown as tumor growth kinetics (mean ± SEM) for (C) selected MCA205 clones or as (D) individual results (one data point corresponds to one mouse) on day 25. (E) Therapeutic response of WT versus mutated TC1 lung cancers to CTX alone or in combination with E. hirae 13144 (as outlined in Fig. 1A, but without antibiotic preconditioning), reflected by tumor growth kinetics and individual tumor sizes at sacrifice. Results are shown as mean ± SEM. Mann-Whitney test or ANOVA statistical analyses (Kruskal-Wallis test) were used: *P < 0.05, **P < 0.01. The statistical report is in the supplementary materials.

Reinforcing the notion of molecular mimicry between phage-encoded and cancer antigens, flow cytometric analyses using fluorescent-labeled tetramers H-2Kb/TSLARFANI (from TMP1) and H-2Kb/GSLARFRNI (from PSMB4) identified a subset of double-positive CTLs that infiltrate MCA205 tumors from CTX and E. hirae 13144–treated mice (fig. S6F) and that was as frequent as CTLs recognizing the PSMB4 peptide only (Fig. 4A). We purified the splenic CD8+ T cells using either the TMP1–H-2Kb– or PSMB4–H-2Kb–specific tetramers and stimulated them with irrelevant (OVA-derived SIINFEKL) versus relevant (TMP-derived TSLARFANI or PSMB4-derived GSLARFRNI) peptides (Fig. 4B). CD8+ T cells binding H-2Kb–TMP1 tetramers produced IFNγ not only in response to TMP1 (up to fivefold increase in IFNγ-secreting T cells) but also in response to the PSMB4 epitope (twofold increase, as much as with heat-killed E. hirae 13144 processed by DCs) (Fig. 4C, fig. S6G). Similarly, CD8+ T cells binding H-2Kb–PSMB4 tetramers functionally recognized TMP1, albeit less efficiently than the PSMB4 epitope (fig. S6G). We analyzed the T cell receptor (TCR) repertoire of these two tetramer-reactive CD8+ T cell subsets. In accordance with the functional data, half of the CD8+ T cells labeled with PSMB4-H-2Kb tetramers shared clonotypes with the much wider TCR repertoire of T cells labeled with the TMP1-H-2Kb–specific tetramers (Fig. 4D and tables S4 and S5), but not with the negative fraction (fig. S6H). Therefore, T cells recognizing the TMP1 epitope of immunogenic E. hirae can cross-react with a peptide contained in the oncogenic driver PSMB4 and vice versa.

Fig. 4 Phage tail length TMP cross-reacts with the PSMB4 cancer epitope.

(A) Flow cytometry analysis of CD8+ tumor-infiltrating lymphocytes (from tumors treated as outlined in Fig. 1A) after costaining with two different tetramers (H-2Kb/TMP1 and H-2Kb/PSMB4, sequences in Fig. 3A). Each data point indicates one tumor, and error bars indicate SEM. The graphs represent compiled results of three independent experiments with five mice per group. (B and C) Purified CD3+ T splenocytes from animals treated with CTX and E. hirae 13144 were restimulated ex vivo with bone marrow–derived DCs (BM-DC) loaded with TMP1 or PSMB4 peptide. One week after ex vivo restimulation, peptide-specific CD8+ T cells were purified after staining with the corresponding tetramer to measure IFNγ secretion in response to DC loaded with peptides (TMP1, PSMB4, SIINFEKL as negative control) or heat-inactivated E. hirae 13144. These results were performed in parallel on the tetramer-binding versus nonbinding fraction and were normalized to the PBS controls (Ctrl). Each dot represents one culture, and error bars represent SEM. Mann-Whitney test or ANOVA statistical analyses (Kruskal-Wallis test) were used: *P < 0.05, ***P < 0.001. (D) Venn diagram of TCRα and β chains from tetramer-positive CD8+ T cells specific for PSMB4 (yellow) or TMP1 (green). (E) Lysogenic conversion of E. gallinarum by the E. hirae Siphoviridae phage in vivo. Ileal content was obtained from naive mice or from mice receiving E. hirae together with CTX, followed by cultivation and isolation of bacterial colonies, MALDI-TOF identification, and PCR-based detection of TMP. Results are from five mice per group, and SEMs within the five ilea are indicated for the CTX+E. hirae 13144 group. (F) Transmission electron microscopy of the phage produced by E. hirae 13144. (G) Kaplan-Meier survival plots of 76 patients with NSCLC or renal cell cancer subjected to PD-1–targeting immunotherapy, stratified according to the presence or absence of TMP in at least five E. faecalis or E. hirae colonies per patient. Univariate log-rank (Mantel-Cox) analysis was used. The statistical report is in the supplementary materials. mOS, mean overall survival; Neg, TMP phage negative; Pos, TMP phage positive.

Some bacteriophages have the potential to transfer immunogenic sequences to other strains in the host ecosystem (2022). To investigate the capacity of the E. hirae 13144 prophage to lysogenize other bacterial species in vivo, we performed culturomic analyses of the ileal content from C57BL/6 mice subjected to oral gavage with E. hirae 13144 and systemic CTX therapy, followed by polymerase chain reaction (PCR) analyses seeking TMP sequences (fig. S8, A and B). We tested 7 to 18 bacterial colonies from each animal and a total of 76 colonies. We only found lysogenic conversion of Enterococcus gallinarum by the E. hirae–temperate phage in vivo, as confirmed by sequencing of the phage genome in the second host (Fig. 4E and fig. S8, B and C). By contrast, none of the 90 colonies (mostly of E. gallinarum) isolated from naive mice harbored the TMP sequence (fig. S8A) and table S6. Similarly, in vitro coculture of TMP+ E. hirae 13144 together with TMP- E. gallinarum spp. at a 1:1 ratio uncovered a significant (~15%) rate of lysogenic conversion (fig. S8D). Examination of a preparation admixing E. hirae 13144 and E. gallinarum at a 1:10 ratio by means of transmission electron microscopy revealed numerous phages with the typical Siphoviridae morphology in the medium, whereas control cultures (bacteria separately) were free of such phages (Fig. 4F). These results indicate that the TMP1 peptide–encoding Siphoviridae prophage of E. hirae 13144 disseminates within enterococci.

We next explored the possible pathophysiological relevance of these findings. We first screened a total of 3027 adult and mother-infant metagenomes (27), validated by a second independent metagenomic assembly–based screening of 9428 metagenomes (28), to assess the breadth of coverage (BOC) of the E. hirae genome and its phages (fig. S9A). E. hirae was present with 100% confidence (BOC > 80%) in fewer than 150 fecal samples from disparate geography, age, and datasets. This phage (and its host) can be vertically transmitted from mothers to infants and then colonizes the neonate. There was an increased prevalence of the phage (57%) in fecal microbiomes from children (representing 16% of all metagenomes, Fisher’s test P value < 0.00001) (table S7). The E. hirae 13144 phage was detectable in many samples lacking the presence of the E. hirae core genome, suggesting that other bacteria in addition to E. hirae can host this phage (table S7). All host genomes belonged to the Enterococcus genus (except two assigned to Coprobacillus), in particular Enterococcus faecalis (80 genomes), Enterococcus faecium (23 genomes), and E. hirae (15 genomes), suggesting that phage 13144 (and its homologs from E. hirae 708, and 13344) are genus-specific but not species-specific (table S8).

Contrasting with metagenomics that has a low sensitivity to detect poor abundance species, culturomics followed by MALDI-TOF (matrix-assisted laser desorption ionization time-of-flight mass spectrometry) provides a technology for detecting rare E. hirae colonies in the stool of healthy individuals (29) or cancer patients (8). PCR analyses of each single cultivatable enterococcal colony (up to five per species and individual) from 76 cancer patients led to the detection of the TMP sequence encompassing the TMP1 peptide in 34% of the patients, only in E. faecalis and E. hirae (figs. S9B and S10). Advanced renal and lung cancer patients [cohort described in Ref. (8)] with detectable fecal TMP at diagnosis exhibited prolonged overall survival after therapy with immune checkpoint inhibitors targeting PD-1 (Fig. 4G). Therefore, we screened 16 TMP-derived nonapeptides predicted to bind the human MHC class I human leukocyte antigen (HLA)-A*0201 with high affinity for their ability to prime naive CD8+ T cells from six healthy volunteers in vitro. We found 6 out of 16 epitopes capable of triggering significant peptide-specific IFNγ release that were located in two distinct regions of the TMP protein (504 to 708 and 1397 to 1462) (fig. S11, A and B, and table S9). Using the NCBI BLASTP suite, we searched the human cancer peptidome [of the Cancer Genome Atlas (TCGA) database] for a high degree of homology with these six HLA-A*0201–restricted immunogenic nonapeptides. We found that only the TMP-derived peptide KLAKFASVV (amino acids 631 to 639) shared significant homology (seven out of nine amino acids, with identical residues at the MHC anchoring positions 2 and 9) with a peptide contained in the protein glycerol-3-phosphate dehydrogenase 1-like (GPD1-L) (fig. S11C). GPD1-L reportedly counteracts the oncogenic hypoxia-inducible factor 1α–dependent adaptation to hypoxia, and its expression is associated with favorable prognosis in head and neck squamous cell carcinomas (3032). The TCGA transcriptomics database unveiled that high expression of GPD1-L is associated with improved overall survival in lung adenocarcinoma and kidney cancers (fig. S11D). Moreover, high expression of GPD1-L mRNA by tumors at diagnosis was associated with improved progression-free survival in three independent cohorts of non–small cell lung cancer (NSCLC) patients (n = 157) (table S10) treated with anti-PD1 antibodies (fig. S11, E and F). Expression of GPD-1L failed to correlate with that of PD-L1 in NSCLC (fig. S11G). Mutations in or adjacent to the 631 to 639 amino acid sequence of GPD-1L gene could rarely be identified in several types of neoplasia (fig. S12).

We derived an HLA-A*0201–restricted, phage peptide (KLAKFASVV)–specific T cell line from peripheral blood mononuclear cells (PBMCs) of a human volunteer. Clones from this line also recognized the HLA-A*0201–restricted GPD-1L epitope (KLQKFASTV) (fig. S13, A to C). Moreover, we detected CD8+ T cells binding HLA-A*0201/KLAKFASVV tetramers exhibiting hallmarks of effector functions after in vitro stimulation of PBMCs with the KLAKFASVV phage epitope in three out of six NSCLC patients (fig. S13, D to F). In the reverse attempt, searching for molecular mimicry between well-known and naturally processed nonmutated melanoma differentiation antigens recognized by human T cell clones (such as HLA-A*0201–binding MART-1 or MELOE epitopes) and gut commensal antigens, we found microbial analogs in the public microbiome databases (figs. S14 and S15 and tables S11 and S13). Some of these microbial peptides are recognized by the corresponding TCR (tables S11 and S13) with similar affinities as the parental (tumoral) epitope.

Altogether, our results suggest that microbial genomes code for MHC class I–restricted antigens that induce a memory CD8+ T cell response, which, in turn, cross-reacts with cancer antigens. Several lines of evidence plead in favor of this interpretation, as exemplified for the TMP1 epitope found in a phage that infects enterococci. First, naturally occurring (mut3 in E. hirae strain ATCC9790) or artificial mutations (mut2 or mut3 in E. coli) introduced into the TMP1 epitope suppressed the tumor-prophylactic and therapeutic potential of bacteria expressing TMP1. Second, transfer of the TMP1-encoding gene into E. coli conferred immunogenic capacity to this proteobacterium, which acquired the same antitumor properties as TMP1-expressing E. hirae. Third, when cancer cells were genetically modified to remove the TMP1–cross-reactive peptide in the PSMB4 protein, they formed tumors that could no longer be controlled upon oral gavage with TMP1-expressing E. hirae. Fourth, cancer patients carrying the TMP phage sequence in fecal enterococci or the GPD1-L tumoral antigen homologous to TMP epitopes exhibited a better response to PD-1 blockade, suggesting that this type of microbe-cancer cross-reactivity might be clinically relevant.

Recent reports point to the pathological relevance of autoantigen–cross-reactive, microbiota-derived peptides for autoimmune disorders such as myocarditis, lupus, and rheumatoid arthritis (15, 33, 34). Given the enormous richness of the commensal proteome (35), we expect the existence of other microbial antigens mimicking auto- and tumor antigens. We have extended these findings to naturally processed melanoma-specific antigens that have microbial orthologs recognized by the same TCRs (figs. S14 and 15 and tables S10 to S12). Global phage numbers have been estimated to reach as high as 1031 particles with the potential of 1025 phage infections occurring every second (36, 37). Thus, the perspective opens that in the microbiota, bacteriophages may enrich the therapeutic armamentarium for modulating the intestinal flora and for stimulating systemic anticancer immune responses.

Supplementary Materials

Materials and Methods

Figs. S1 to S15

Tables S1 to S12

Statistical Report

References (3852)

References and Notes

Acknowledgments: We are thankful to the animal facility team of Gustave Roussy and all the technicians from Centre GF Leclerc. We are indebted to O. Kepp for figure design and to H. G. Rammensee (University of Tübingen, Germany) for his careful guidance in peptide selection and reading of the paper. Funding: L.Z. and G.K. were supported by RHU Torino Lumière (ANR-16-RHUS-0008), ONCOBIOME H2020 network, the Seerave Foundation, the Ligue contre le Cancer (équipe labelisée); Agence Nationale de la Recherche (ANR) – Projets blancs; ANR Francogermanique ANR-19-CE15-0029 under the frame of E-Rare-2, the ERA-Net for Research on Rare Diseases; Association pour la recherche sur le cancer (ARC); Cancéropôle Ile-de-France; Fondation pour la Recherche Médicale (FRM); a donation by Elior; the European Research Council (ERC); Fondation Carrefour; High-end Foreign Expert Program in China (GDW20171100085 and GDW20181100051), Institut National du Cancer (INCa); Inserm (HTE); Institut Universitaire de France; LeDucq Foundation; the LabEx Immuno-Oncology; the SIRIC Stratified Oncology Cell DNA Repair and Tumor Immune Elimination (SOCRATE); CARE network (directed by X. Mariette, Kremlin Bicêtre AP-HP), and the SIRIC Cancer Research and Personalized Medicine (CARPEM). The results shown here are based upon data generated by the TCGA Research Network: I.C. was supported by National Research, Development and Innovation Fund of Hungary Project (no. FIEK_16-1-2016-0005). Z.S. was supported by the Research and Technology Innovation Fund (NAP2-2017-1.2.1-NKP-0002) and Breast Cancer Research Foundation (BCRF-17-156). Z.S. and I.C. were supported by the Novo Nordisk Foundation Interdisciplinary Synergy Programme Grant (NNF15OC0016584). P.N. was supported by the Italian Association for Cancer Research (AIRC IG 19822). Mouse TCR sequencing was performed by the TRiPoD ERC-Advanced EU (322856) grants to D.K. N.S. is supported by the European Research Council (project ERC- STG MetaPG-716575). L.D.S. is funded by a Roux-Cantarini fellowship from the Institut Pasteur (Paris, France). Bristol-Myers-Squibb provided resources to support translational research related to the NIVOREN clinical trial enrolling kidney cancer patients. Competing interests: R.D., D.R., L.Z., and G.K. are cofounders of everImmune, a biotech company devoted to the use of commensal microbes for the treatment of cancers. R.D. is a full-time employee of everImmune. L.Z., A.F., V.C., and R.D. are inventors on a patent application (WO/2019/129753) submitted by Institute Gustave Roussy/INSERM/University Paris Saclay that covers “Immunogenic sequences from a Phage Tail Length Tape Measure Protein, bacteria expressing the same and their use in treating a cancer.” Data and materials availability: E. hirae 13144 isolate and E. hirae isolates IGR4 and IGR11 are available from U1015 INSERM under a material transfer agreement (with Gustave Roussy Technology Transfer) or from the repository of Institut Pasteur listed as Enterococcus hirae deposited at the Collection Nationale de Cultures de Microorganismes (CNCM) of Pasteur Institute, Paris: strains 13144 (EHFS001), deposited on November 7, 2013, under the number I-4815; IGR4 deposited on November 27, 2017, under the number CNCM I-5260; and IGR11 deposited on November 27, 2017, under the number CNCM I-5261. Bacteria genomes sequenced in this study have been deposited in the NCBI GenBank under the accession number PRNJA639126. Microbiome-related mass spectrometry data have been deposited (

Stay Connected to Science

Navigate This Article