Microbial signals in primary and metastatic brain tumors

Abstract

Gliomas and brain metastases are associated with poor prognosis, necessitating a deeper understanding of brain tumor biology and the development of effective therapeutic strategies. Although our group and others have demonstrated microbial presence in various tumors, recent controversies regarding cancer-type-specific intratumoral microbiota emphasize the importance of rigorous, orthogonal validation. This prospective, multi-institutional study included a total of 243 samples from 221 patients, comprising 168 glioma and brain metastases samples and 75 non-cancerous or tumor-adjacent tissues. Using stringent fluorescence in situ hybridization, immunohistochemistry and high-resolution spatial imaging, we detected intracellular bacterial 16S rRNA and lipopolysaccharides in both glioma and brain metastases samples, localized to tumor, immune and stromal cells. Custom 16S and metagenomic sequencing workflows identified taxa associated with intratumoral bacterial signals in the tumor microenvironment; however, standard culture methods did not yield readily cultivable microbiota. Spatial analyses revealed significant correlations between bacterial 16S signals and antimicrobial and immunometabolic signatures at regional, neighborhood and cellular levels. Furthermore, intratumoral 16S bacterial signals showed sequence overlap with matched oral and gut microbiota, suggesting a possible connection with distant communities. Together, these findings introduce microbial elements as a component of the brain tumor microenvironment and lay the foundation for future mechanistic and translational studies.

Fig. 1. Study workflow.

a, Schematic of the study workflow. To assess the presence and correlation of bacteria and bacterial elements in brain tumors, we combined integrative methods, including RNAScope FISH and SMI for detection and validation of bacterial signals, culturomics and bioinformatic analyses for taxonomic characterization of intratumoral bacterial signals and spatial technologies to evaluate biological relevance within the TME. Stringent workflows were developed to enhance each methodology. Schematics were created using BioRender.com. b,c, Samples from 221 patients were included in this study, with distribution shown by analytical method (b) and institution (c). d, When sufficient tissue was available, samples were analyzed using at least two methods in the following categories: visualization (FISH, SMI and DSP); sequencing (16S rRNA amplicon and metagenomic shotgun sequencing); and culturomics. The Venn diagram depicts the number of samples analyzed across each methodological category. Seq, sequencing; WGS, whole-genome sequencing.

Fig. 2. Bacterial signals are detectable in primary and metastatic brain tumors.

a, RNAScope 16S rRNA FISH of glioma (left) and BrM (right). Representative images of 15 glioma and 15 BrM samples (female 40%, age range 20–71 years; one full section slide per patient). Nuclei, DAPI, blue; bacterial 16S rRNA, red; GFAP, pan-cytokeratin (pan-CK), green in glioma and BrM, respectively. b, Representative images of H&E staining, 16S rRNA FISH and LPS IHC staining of a brain tumor sample (patient cohort same as a; one sequential full section slide per patient). Nuclei, SYTO13, blue; bacterial 16S rRNA, red; negative control, green. White and black arrows mark 16S rRNA and LPS signals with proximity to nuclei, respectively. c, Schematic demonstrating the experimental workflow for SMI (CosMx platform). Patients: glioma n = 19, BrM n = 9, female 37.1%, age range 28–80 years; FOVs n = 39 per TMA. d, Three-dimensional reconstruction of cells containing high-confidence intracellular 16S signal (light and dark blue) and low-confidence low-level 16S signal (purple). Bacterial 16S signals, red; human transcripts identified in each cell, purple and blue. e, UMAP plot of single-cell clustering of glioma and BrM samples. Each cell cluster is represented by a different color and labeled. f, UMAP plot of high-confidence 16S signal distribution in each cluster and corresponding normalized counts. g,h, Representative images and bar plots demonstrating the percentage of high-confidence 16S-positive cells within glioma (g) and BrM (h) TME. 16S signal, white. Images were obtained using napari software. DC, dendritic cell; ECs, endothelial cells; Mac/Mic, macrophage/microglia; MSC, mesenchymal stem cell; NK, natural killer cell; Oligo, oligodendrocytes; OPC, oligodendrocyte progenitor cell; T_reg, regulatory T cell.

Fig. 3. Distinct bacterial signatures were identified in brain tumors compared to non-cancerous brain tissue.

a, Tumor samples (glioma n = 77, BrM n = 41, fresh-frozen n = 109, formalin-fixed paraffin-embedded (FFPE), n = 5, OCT (Optimal cutting temperature), n = 4; female 39.8%, age range 20–83 years) and non-cancerous brain tissue (fresh-frozen, n = 13; female 23.1%, age range 20–73 years) were assessed using 16S (V3-V4) rRNA amplicon sequencing. Bubble plot demonstrates the proportion of bacterial taxa identified among samples with detectable bacterial signal in glioma, BrM and non-cancerous prospective cohorts. Taxa identified as potential environmental contaminants after filtration are presented in gray. Tables present the number of samples, ASVs and genera identified pre-filtering and post-filtering. b, Glioma (n = 30) and BrM (n = 15) samples were assessed using metagenomic shotgun sequencing (fresh-frozen n = 45; female 44.4%, age range 26–77 years). Bubble plot demonstrates the proportion of bacterial taxa identified among samples with detectable bacterial signal, presented at genus level. Taxa identified as potential environmental contaminants after filtration are presented in gray. Tables present the number of samples, genera and species identified pre-filtering and post-filtering.

Fig. 4. 16S-high tumor regions are enriched in antimicrobial signatures.

a, Schematic demonstrating the experimental workflow for DSP (GeoMx platform). Patients: glioma n = 31, BrM n = 19; female 34%, age range 20–81 years. Representative image of an ROI containing CD3⁺ cells, taken using the GeoMx platform. b, PCA demonstrating the clustering of human transcripts identified in glioma (R² = 0.48, P = 0.001, two-sided PERMANOVA) and BrM (R² = 0.19, P = 0.001, two-sided PERMANOVA) TMAs based on categorical 16S status analyzed by DSP, background subtracted and Q3 normalized. Low-16S ROIs, blue; high-16S ROIs, red. c, Differentially expressed proteins in glioma 16S-high tumor regions (two-sided LMM, Benjamini–Hochberg test for multiple comparison adjustment, adjusted P < 0.05, log₂FC > 0.58). d, Differentially expressed proteins in BrM 16S-high tumor regions (two-sided LMM, Benjamini–Hochberg test for multiple comparison adjustment, adjusted P < 0.05, log₂FC > 0.58). e, Multiplex fluorescence IHC of BrM samples. Representative images of 16S-low (top panels) and 16S-high (bottom panels) are demonstrated. CD16, green; CD56, red, GZMB, white; yellow arrows, CD16⁺ cells, ×40 magnification. f, Bar plot demonstrating the differential abundance of CD16⁺CD56⁻GZMB⁻ populations in 16S-low (patients n = 6, ROIs n = 16) and 16S-high (patients n = 5, ROIs n = 18) regions in BrM TMA (two-sided LMM of log₂-transformed counts, Benjamini–Hochberg test for multiple comparison adjustment, adjusted P = 0.012, untransformed counts presented). Box plots show median (center line), interquartile range (IQR) (box bounds, 25th and 75th percentiles) and smallest and largest values within 1.5× the IQR (whiskers). Outliers are plotted beyond the whiskers. FDR, false discovery rate; P_adj, adjusted P value; PC, principal component.

Fig. 5. 16S-positive tumor cells and neighborhoods exhibit distinct transcriptional profiles.

a, Representative image of 16S-positive tumor cells assessed for differential transcriptional profile compared to 16S-negative cells; patients with more than 20 16S-positive tumor cells were included (patients with glioma n = 3 (P11, P15 and P26), 16S-positive tumor cells n = 68, 16S-negative tumor cells n = 6,559; patients with BrM n = 4 (P35, P41, P43 and P46), 16S-positive tumor cells n = 155, 16S-negative tumor cells n = 35,221). Tumor cells, teal; 16S signal, red. Image was obtained using napari software. b, Bubble plots demonstrating enriched transcripts in 16S-positive tumor cells across patients with glioma and BrM (two-sided LMM, Benjamini–Hochberg test for multiple comparison adjustment, P < 0.05, log₂FC > 0.58). c, Violin plots demonstrating pathway scores in 16S-positive and 16S-negative tumor cells in patients with BrM (two-sided LMM, Benjamini–Hochberg test for multiple comparison adjustment, P < 0.05, log₂FC > 0.58). Violin plots show smoothed AUCell score distributions, overlaid with box plots indicating the median (center line), interquartile range (IQR) (box bounds, 25th and 75th percentiles) and smallest and largest values within 1.5× IQR (whiskers). d, Representative image of neighborhoods (NHs) within 30-µm radius surrounding 16S-positive tumor cells and those surrounding 16S-negative tumor cells (patients with glioma n = 3 (P11, P15 and P26), cells within 16S-positive neighborhoods n = 731, cells within 16S-negative neighborhoods n = 492; patients with BrM n = 3 (P41, P43 and P46), cells within 16S-positive neighborhoods n = 2,093, cells within 16S-negative neighborhoods n = 8,134). Teal, 16S-positive tumor cell; magenta, 16S-positive neighborhoods; green, 16S-negative neighborhoods. Cells more than 30 μm away from a tumor cell and those between 30 μm and 100 μm of the closest 16S-positive cell were excluded (gray). e, Bubble plots demonstrating transcripts enriched in combined 16S-positive neighborhoods from patients with glioma and BrM (two-sided LMM, Benjamini–Hochberg test for multiple comparison adjustment, P < 0.05, log₂FC > 0.58). f, Violin plots demonstrating pathway scores in 16S-positive compared to 16S-negative neighborhoods in patients with BrM (two-sided LMM, Benjamini–Hochberg test for multiple comparison adjustment, P < 0.05, log₂FC > 0.58). Violin plots show smoothed AUCell score distributions, overlaid with box plots indicating the median (center line), IQR (box bounds) and values within 1.5× IQR (whiskers). g, An example of spatial distribution of different pathways in 16S-positive and 16S-negative neighborhoods, demonstrated for BrM P43.

Fig. 6. Intratumoral bacterial 16S signal correlates with the oral and gut microbiota of patients with brain tumor.

a, Differentially abundant taxa in the oral and gut microbiome of patients with glioma and BrM (age range 18–40 years, female 46.7%) compared to healthy individuals (HMP, age range 18–40 years, female 46.4%), assessed by ANCOM-BC, ANCOM-BC2 and MaAsLin2 (two-sided, Holm and Benjamini–Hochberg test for multiple comparison adjustment, respectively). Taxa with differential abundance by at least two methods were included. Heatmap values demonstrate the log₂FC of taxa with significant differential abundance in patients with brain tumor compared to HMP. If a taxon was not significantly different by two of the three methods, the corresponding cell was colored in white (adjusted P values are presented in Supplementary Table 7). Schematic was created using BioRender.com. b, The proportion of the identified intratumoral bacterial sequence (seq.) length that overlapped with bacterial sequences in matched oral or gut microbiome (metagenomic shotgun sequencing). Schematic was created using BioRender.com. c, Stacked bar plot demonstrating the number of identified taxa in tumor with or without sequence overlap with salivary or stool bacterial taxa.

Extended Data Fig. 1. Distribution of 16S signal within cells and across patient samples.

a,b, RNAScope 16S rRNA FISH of (a) glioma, BrM, and (b) healthy brain tissue; bacterial 16S rRNA, red; negative control probe, green; DAPI, blue, 20X objective; 3D reconstructions, 63X objective (patients: glioma n = 15, BrM n = 15; female 40%, age range 20-71 years; 1 full section slide per patient; Normal n = 40; female 32.9%, age range 15-54 years; 1 TMA, 2 cores per patient). c, RNAScope 16S rRNA FISH and Lipopolysaccharide (LPS) staining of normal tissue adjacent to tumor (NAT) in glioma and BrM samples (patients: glioma n = 13, BrM =3, female 57%, age range 28-71 years; 1 full section slide per patient). Arrows demonstrate staining with intracellular pattern. Nuclei, SYTO13, blue; 16S rRNA, red; negative control, green. d, Density plots demonstrating the distribution of the bacterial 16S median (red line) in relationship to all human transcripts within a cell at X, Y, Z axes. The two black lines indicate where 50% of human transcripts are located. high-confidence cell, upper panels; low-confidence cell, lower panels. e, Tables demonstrating the total cell number per patient, and total number and percentage of 16S-positive cells per patient. f, Table demonstrating the total number of cells per cell type, percentage of each cell type among the total cells, and the percentage of 16S-pos cells for each cell type.

Extended Data Fig. 2. Expression of transcripts across cell clusters.

a,b, Bubble plots demonstrating the marker transcripts in each cell cluster identified for (a) glioma and (b) BrM. MSC, mesenchymal stem cell; Mac/Mic, macrophage/microglia; DC, dendritic cell; NK, natural killer cell; Oligo, oligodendrocyte; OPC, oligodendrocyte progenitor cell; ECs, endothelial cells; Treg, regulatory T cells; T, tumor.

Extended Data Fig. 3. Filtration of 16S rRNA and metagenomic shotgun sequencing datasets.

a, Proportion of ASVs and prevalence of genera identified in controls pooled from all sequencing batches (controls n = 24). Inflection point was calculated via ChangePoint analysis. The inner proportion plot depicts a higher magnification of the inflection point. Red arrow, inflection point. b, Alluvial plot demonstrating the taxa identified across all glioma, BrM, and non-cancerous brain samples pre-filtration and after each filtration step. Table demonstrates the number of ASVs and genera present at each filtration step. c, Histograms demonstrate bacterial richness (alpha diversity) in brain tumors post-filtration, based on 16S amplicon sequencing or metagenomic shotgun sequencing. Both methods were applied to 16 samples, with 9 samples yielding reads by both methods and presented separately. d, 3D reconstruction of cells with intra-cellular genus-specific bacterial signal. Porphyromonas, green; Fusobacterium, blue; Prevotella, magenta; human transcripts, gray.

Extended Data Fig. 4. Distribution of 16S signal and the associated enriched proteins and transcripts.

a, Bacterial score in glioma and BrM tumor regions compared to the normal tissue adjacent to tumor (glioma NAT n = 7, p value = 0.037, BrM NAT n = 1). Two-sided Linear Mixed Model (LMM), Benjamini-Hochberg test for multiple comparison adjustment, adjusted P value < 0.05. Boxplots show median (center line), interquartile range (box bounds, 25^th and 75^th percentiles), and smallest and largest values within 1.5× the IQR (whiskers). Outliers are plotted beyond the whiskers. b, Bacterial score in glioma and BrM 16S-high (1^st quartile; glioma patients n = 11, ROI n = 21; BrM patients n = 11, ROI n = 19) vs. 16S-low tumor regions (4^th quartile; glioma patients n = 12, ROI n = 21; BrM patients n = 6, ROI n = 19). Boxplots show median (center line), interquartile range (box bounds, 25^th and 75^th percentiles), and smallest and largest values within 1.5× the IQR (whiskers). Outliers are plotted beyond the whiskers. c,d, Volcano plot demonstrating the proteins enriched in (c) glioma and (d) BrM 16S-high tumor regions. Two-sided LMM, Benjamini-Hochberg test for multiple comparison adjustment, adjusted P value < 0.05, log2FC > 0.58. e,f, Differentially expressed transcripts in (e) glioma and (f) BrM 16S-high tumor regions. Two-sided LMM, Benjamini-Hochberg test for multiple comparison adjustment, adjusted P value < 0.05, log2FC > 1.5. g, Correlation of the identified anti-microbial, metabolic signatures, and reference tumor inflammation signatures with the 16S bacterial signal.

Extended Data Fig. 5. 16S-pos tumor cells and neighborhoods exhibit distinct transcriptional scores, per patient.

a, Violin plots demonstrating biological pathway scores in 16S-positive and -negative tumor cells in individual patients (BrM 16S-pos/16S-neg tumor cells: P35 40/2247, P41 35/5776, P43 58/12609, P46 22/14589). Two-sided Wilcoxon signed-rank test. Violin plots show smoothed AUCell score distributions, overlaid with boxplots indicating the median (center line), interquartile range (box bounds, 25th and 75th percentiles), and smallest and largest values within 1.5× IQR (whiskers). b, heatmap demonstrating the z-score of 16S-associated pathways in glioma 16S-positive tumor cell and neighborhoods and 16S-negative neighborhoods. c, violin plots demonstrating biological pathway scores in 16S-positive and -negative neighborhoods in individual patients (BrM cell number in 16S-pos/16S-neg neighborhoods: P41 557/1061, P43 1024/1800, P46 512/5273). Two-sided Wilcoxon signed-rank test. Violin plots show smoothed AUCell score distributions, overlaid with boxplots indicating the median (center line), interquartile range (box bounds, 25th and 75th percentiles), and smallest and largest values within 1.5× IQR (whiskers). d, heatmap demonstrating the z-score of 16S-associated and general inflammatory control pathways in 16S-positive tumor cell and neighborhoods and 16S-negative neighborhoods (containing 16S-negative tumor cells).

Extended Data Fig. 6. Oral and gut bacterial signatures are associated with brain tumor progression.

a,b, Differentially abundant taxa in the oral and gut microbiome of (a) glioma patients and (b) BrM patients with post-resection intracranial progression compared to patients with no progression, assessed by ANCOM-BC, ANCOM-BC2, and MaAsLin2 (two-sided; Holm and Benjamini-Hochberg test for multiple comparison adjustment, respectively). Taxa with adjusted p value < 0.05, log2FC > 2 and with differential abundance by at least two methods are presented. Heatmap demonstrates the log2FC of each taxon with significant differential abundance. If a taxon was not significantly different by 2 of the 3 methods, the corresponding cell was colored in white (adjusted p values are presented in Supplementary Table 9). Boxplots show median (center line), interquartile range (box bounds, 25^th and 75^th percentiles), and smallest and largest values within 1.5× the IQR (whiskers).