Identification of bacteria-derived HLA-bound peptides in melanoma

Abstract

A variety of species of bacteria are known to colonize human tumours^1-11, proliferate within them and modulate immune function, which ultimately affects the survival of patients with cancer and their responses to treatment^12-14. However, it is not known whether antigens derived from intracellular bacteria are presented by the human leukocyte antigen class I and II (HLA-I and HLA-II, respectively) molecules of tumour cells, or whether such antigens elicit a tumour-infiltrating T cell immune response. Here we used 16S rRNA gene sequencing and HLA peptidomics to identify a peptide repertoire derived from intracellular bacteria that was presented on HLA-I and HLA-II molecules in melanoma tumours. Our analysis of 17 melanoma metastases (derived from 9 patients) revealed 248 and 35 unique HLA-I and HLA-II peptides, respectively, that were derived from 41 species of bacteria. We identified recurrent bacterial peptides in tumours from different patients, as well as in different tumours from the same patient. Our study reveals that peptides derived from intracellular bacteria can be presented by tumour cells and elicit immune reactivity, and thus provides insight into a mechanism by which bacteria influence activation of the immune system and responses to therapy.

Extended Data Fig. 1 |. Similarity of bacterial composition between metastases.

A Jaccard index was calculated to determine the similarity between the bacterial composition of the different metastases on the species level. Colour code indicates the Jaccard index. The highest similarity was observed between metastases from the same patient, but metastases of different patients also showed similarity. Black boxes indicate tumour samples taken from the same patient.

Extended Data Fig. 2 |. Visualization of bacterial 16S rRNA in tissue sections from melanoma tumours.

a, 16S rRNA fluorescence in situ hybridization (FISH) staining of tissue sections from melanoma tumours using pan-bacteria EUB338 probe (red) and DAPI (blue). b, 16S rRNA FISH staining of tissue microarray sections of melanoma tumours (red) and DAPI (blue). Slice name indicates the position in the tissue microarray. Images are presented at 20× magnification. Scale bars, 100 μm. c, Representative control using 16S FISH nonspecific control probe. Asterisks mark the region that was selected for higher magnification. Figures are representative of at least three independent experiments.

Extended Data Fig. 3 |. Taxonomic analysis of 108 whole-genome-sequenced melanoma samples identify a bacterial composition similar to that found in our tumour cohort.

a, Alpha diversity, measured as the number of observed species in a tumour (green) or blood (purple) sample. P values from paired two-tailed Wilcoxon test between tumour and blood taxonomic diversity. b, Microbiome similarity within and between groups. Bray–Curtis dissimilarity measured between each pair of samples, then stratified into four groups. P values from two-tailed Wilcoxon test. c, Comparison of the relative abundance between tumour samples and associated blood samples. P values from paired two-tailed Wilcoxon test between tumour and blood taxonomic abundance. ***P < 0.001, **P < 0.01, *P < 0.05. d, List of groups of bacteria that are more abundant in the tumour samples plotted in c. P values from paired two-tailed Wilcoxon test between tumour and blood taxonomic abundance. P values with asterisks survived multiple hypothesis correction (false-discovery rate of 5%). In the box plots, the centre lines represent the medians, the boxes represent the range between the 25th and 75th percentile, and the whiskers represent the range between the smallest and largest data point.

Extended Data Fig. 4 |. Pipeline for the identification of bacteria -derived HLA peptides.

Tumour samples were 16S-sequenced to identify their bacterial composition and analysed using HLA peptidomics. Searching the data according to the bacteria resulted in the identification of bacteria-derived peptides. Peptides were filtered according to their identification quality and ability to bind to the HLA alleles of the patient. Selected bacterial peptides were then tested for reactivity. Identification of bacterial peptide presentation was validated by a HLA peptidomics analysis of cell lines cocultured with bacteria.

Extended Data Fig. 5 |. Length distribution of bacteria-derived peptides.

The length distribution of bacteria-derived peptides is similar to the expected length of HLA-I and HLA-II peptides.

Extended Data Fig. 6 |. Bacterial proteomes contain a higher amount of hydrophobic amino acids compared to the human proteome.

a, For each metastasis, the percentages of bacterial and human peptides that match each HLA-A, HLA-B and HLA-C allele of the patient is indicated. The allele with the best per cent rank binding prediction by NetMHCpan was assigned to each peptide. b, Kyte-Doolittle hydrophobicity index was calculated for bacterial and human peptides. The hydrophobicity of HLA-I bacterial peptides is higher than that of human-derived peptides (indicated P value is from an unpaired two-sample Wilcoxon test). c, The percentage of hydrophobic and nonhydrophobic amino acids was calculated for bacterial proteomes and the human proteome. Two groupings were used for selecting hydrophobic amino acids: L, I, V, F and M, or L, I, V, F, M, W, Y and A. The percentage of hydrophobic and nonhydrophobic amino acids from bacterial proteomes is plotted in the box plot. In the box plots, the centre lines represent the medians, the boxes represent the range between the 25th and 75th percentile, and the whiskers represent the range between the smallest and largest data point. The percentages representing the human proteome are marked by a red dashed line. d, Two-sided Student’s t-test comparing the percentage of hydrophobic and non-hydrophobic amino acids between bacterial proteomes and the human proteome. The P values and false-discovery rates are indicated in the table.

Extended Data Fig. 7 |. Hydrophobicity of bacterial and human peptides per allele.

Kyte–Doolittle hydrophobicity index was calculated for bacterial and human peptides and plotted in a box plot for each HLA allele. In the box plots, the centre lines represent the medians, the boxes represent the range between the 25th and 75th percentile, and the whiskers represent the range between the smallest and largest data point. a, The hydrophobicity of the bacterial peptides that bind to the HLA-C*03:04, HLA-C*03:03 and—to a lesser extent—HLA-A*02:01 was higher compared to the hydrophobicity of other alleles (marked in red). Additional alleles also show this trend, but they were derived from a lower number of tumour samples and therefore are not indicated. b, Bacterial peptides are marked in red and human peptides are marked in grey.

Extended Data Fig. 8 |. Gentamicin assay demonstrating the entry of bacteria into melanoma cells.

a, b, Colony-forming units (CFU) of A. odontolyticus and S. caprae after coculture with 51AL and 55A3 melanoma cells. Less-invasive bacteria (L. animalis and L. plantarum) were used as a control, and show lower CFU. c, d, CFU of S. capitis (c) (isolated from tumour 58) after coculture with 58A melanoma cells, or S. succinus (d) (isolated from tumour Mel261) after coculture with 51AL or 55A3 cells. Cells were cultured with the indicated bacteria for 4 and 8 h. ‘Supernatant’ refers to the CFU of medium taken from samples incubated with gentamicin after coculture with the bacteria (grey). ‘No gentamicin’ refers to samples not treated with gentamicin after the coculture (red). ‘With gentamicin’ refers to samples treated with gentamicin for 1 h after the coculture (blue). Bars represent the average of s.e. between biological replicates (n = 3). P values from Student’s t-test between the supernatant sample and the without gentamicin or with gentamicin samples; P values between S. caprae or A. odontolyticus with gentamicin to L. animalis or L. plantarum with gentamicin control samples are from a one-way analysis of variance followed by Tukey’s test, and are presented in Supplementary Table 4.

Extended Data Fig. 9 |. Immunofluorescence staining of melanoma cells cocultured with aerobically grown bacteria, demonstrating the ability of the bacteria to enter the cells.

a, Melanoma cells expressing GFP (green) were cocultured with the aerobically grown bacteria S. caprae, S. capitis or S. succinus stained with antibacterial antibody lipoteichoic acid (LTA) (red); cell nuclei were stained with DAPI (blue). White arrows indicate the location of bacteria that entered the melanoma cells. b, A representative image of 51AL cells expressing GFP (green) cocultured with S. caprae and stained without a primary LTA antibody (red), to exclude nonspecific staining. Images are presented at 63× magnification. Scale bars, 10 μm. Figures are representative of at least three independent experiments.

Extended Data Fig. 10 |. Immunofluorescence staining of melanoma cells cocultured with anaerobically grown bacteria, demonstrating the ability of the bacteria to enter the cells.

a, Melanoma cells stained an anti-HLA antibody (red) were cocultured with anaerobically grown bacteria F. nucleatum or A. odontolyticus. These bacteria were labelled with click chemistry (green). Cell nuclei were stained with DAPI (blue). White arrows indicate the location of bacteria that entered the melanoma cells. b, A representative image of 51AL cells stained with the anti-HLA antibody (red) cocultured with F. nucleatum that were not grown with D-GalNAz and labelled with Alexa Fluor F488 (green), to exclude nonspecific staining. Images are presented at 63× magnification. Scale bars, 10 μm. Figures are representative of at least three independent experiments.

Extended Data Fig. 11 |. CLEM images showing entry of bacteria into melanoma cells.

Fusobacterium nucleatum was grown with D-GalNAz and then labelled with DIBO-Alexa Fluor 488. Actinomyces odontolyticus and S. caprae were incubated with an anti-LTA antibody, and then with an anti-mouse secondary antibody labelled with Alexa Fluor 488. The 51AL and 55A3 cell lines were coincubated with the bacteria for 8 h. Ultra-thin sections were analysed by fluorescence microscopy to identify the bacteria (green and blue labelling are for bacteria and nucleus, respectively), followed by transmission electron microscopy (TEM) of the same cells for high-resolution morphology. Left and middle panels show CLEM and TEM images, respectively. Scale bars, 5 μm. The right panel shows high-magnification TEM image of the area in the black box in the corresponding middle panel. Scale bars, 1 μm. Bacteria that entered the melanoma cell are indicated with a white arrow. N, nucleus; M, mitochondrion; ER, endoplasmic reticulum. Figures are representative of at least three independent experiments.

Extended Data Fig. 12 |. TIL reactivity towards bacteria-derived antigens.

Flow cytometry analysis of IFNy-secreting TILs after coculture of 51AL and 55A3 TILs with B cells loaded with the indicated bacterial peptide. Each peptide was loaded on a different B cell that exhibited the HLA alleles to which the peptide was predicted to bind. TILs were stained with anti-IFNγ and anti-CD45 bifunctional antibody, which binds secreted IFNγ. The value indicates the ratio of IFNγ-secreting cells with the peptides to those with the DMSO control. Grey dots indicate the results of n = 3 biological replicates, and blue dots represent the average of replicates. Bars represent s.e. between replicates.

Fig. 1 |. Identification of intratumoral bacteria in melanoma.

Schematic phylogenetic tree of the bacterial composition of 17 melanoma metastases that originated from 9 patients. The analysis is based on rRNA 16S gene sequencing. The different colours and shades in the circles indicate the different classifications of bacteria at the genus (inner circle), order (middle circle) and class (outer circle) level. Each patient is colour-coded (as in the index), and different metastases from the same patient are depicted in different shades of the same colour.

Fig. 2 |. Characteristics of bacterial peptides.

a, The number of bacterial peptides presented on HLA-I and HLA-II in each patient sample (patient number indicated at top) is indicated in a blue colour scale (left). White indicates that no peptides were identified in the sample, and grey indicates that the bacterium was not identified in this metastasis (NA, not applicable). The total number of bacterial HLA-I and HLA-II peptides from each bacterium is noted in the bar plot on the right. Species names marked in red are known to be intracellular bacteria (Supplementary Table 6). b, Bacterial peptides that were identified in a few metastases from the same patient or in different patients are indicated. Peptides identified in the sample are marked green, and white denotes peptides that were not identified in the sample (although the metastasis has the required HLA allele for this peptide presentation and the species of bacteria). Grey indicates samples that lack the HLA allele and bacteria to produce the peptide. c, For each metastasis, the percentages of bacterial and human peptides that match each HLA-A (left), HLA-B (middle) or HLA-C (right) allele of the patient is indicated. The allele with the best per cent rank binding prediction (by NetMHCpan) was assigned to each peptide; the full allele list is indicated in Extended Data Fig. 6.

Fig. 3 |. Evidence of bacteria entry and presentation by melanoma cells.

a, Tumour 422 was digested and CD45⁺ and CD45⁻ populations were subjected to HLA peptidomics. The table lists the peptides identified in each sample. b, Immunofluorescence staining detecting bacteria in 51AL cells stained with an anti-HLA antibody (red) that were cocultured with F. nucleatum (green). Cell nuclei were stained with DAPI (blue). Left, representative merged image from the z-stack centre, presented at 63× magnification. Right, z-stack 3D image (bacteria, turquoise; melanoma cells, magenta; nucleus, blue). Scale bars, 10 μm. c, We co-incubated 55A3 cells with F. nucleatum (green). Cell nuclei were stained with DAPI (blue). Ultramicrotome sections were analysed by CLEM. Left, the area of intracellular bacteria is marked by a black box. Scale bar, 5 μm. Right, zoomed-in view of the bacteria. Scale bar, 1 μm. d, Number of bacterial peptides identified by HLA peptidomics in the 51AL and 55A3 cell lines that were cocultured with F. nucleatum and S. caprae, respectively. B2M and CIITA were knocked out (KO) to reduce the levels of HLA-I and HLA-II, respectively. Compared with cells infected with scrambled control single-guide RNA (SC), the cells with B2M or CIITA knocked out show a lower number of identified bacterial peptides. Less-invasive species of bacteria (L. animalis and L. plantarum) were used as a control to show that the identified peptides are HLA ligands that resulted from intracellular bacteria. e, Different peptides were derived from the pyruvate–ferredoxin (flavodoxin) oxidoreductase protein from F. nucleatum. The protein sequence is described until the 900th amino acid rather than the end of the protein (1,188 amino acids in total), as no peptides were identified after this region of the protein.

Fig. 4 |. TIL reactivity towards bacteria-derived antigens.

IFNγ-secreting 51AL and 55A3 TILs were detected after 6 h of coculture with B cells loaded with a bacterial peptide or dimethyl sulfoxide (DMSO) control, using flow cytometry. TILs were also tested for the presence of the CD69 reactivity marker. The image presents 4 representative immunogenic peptides (3 out of 7 immunogenic peptides of patient 51 and 1 out of 1 peptides of patient 55) that showed at least a 2-fold change between the peptide and DMSO control. The percentage of positive IFNγ-secreting or CD69-expressing TILs is an average of three independent experiments (Extended Data Fig. 12, Supplementary Figs. 10, 11). a, Patient 51. b, Patient 55.