Hepatitis B virus-related hepatocellular carcinoma exhibits distinct intra-tumoral microbiota and immune microenvironment signatures

Abstract

Emerging evidence supports a high prevalence of cancer type-specific microbiota residing within tumor tissues. The intra-tumoral microbiome in hepatocellular carcinoma (HCC), especially in viral (HBV/HCV) HCC, has not been well characterized for their existence, composition, distribution, and biological functions. We report herein a finding of specific microbial signature in viral HCC as compared to non-HBV/non-HCV (NBNC) HCC. However, the significantly diverse tumor microbiome was only observed in HBV-related HCC, and Cutibacterium was identified as the representative taxa biomarker. Biological function of the unique tumor microbiota in modulating tumor microenvironment (TME) was characterized by using formalin-fixed paraffin-embedded (FFPE) tissue-based multiplex immunofluorescence histochemistry (mIFH) allowing simultaneous in situ detection of the liver cancer cells surrounded with high/low density of microbiota, and the infiltrating immune cells. In HBV_HCC, the intra-tumoral microbiota are positively associated with increased tumor infiltrating CD8+ T lymphocytes, but not the CD56+ NK cells. Two subtypes of myeloid-derived suppressor cells (MDSCs): monocytic MDSCs and polymorphonuclear MDSCs, were also found to be positively correlated with the intra-tumoral microbiota in HBV_HCC, indicating an inhibitory role of these microbial species in antitumor immunity and the contribution to the liver TME in combination of chronic viral hepatitis during HCC development.

INTRODUCTION

Hepatocellular carcinoma (HCC) is one of the leading causes of cancer mortality worldwide¹. HCC mostly develops in patients against a background of cirrhotic liver, non-alcoholic fatty liver disease, chronic liver damage, or fibrosis. Cirrhosis is closely related to various etiologies, including chronic hepatitis B virus (HBV) and hepatitis C virus (HCV) infection, chronic alcohol abuse, or metabolic syndrome².

The gut microbiota has been suggested to play an integral role in tumor biology, such as tumor transformation, tumor progression, and the response to anticancer therapies^3,4. Emerging evidence supports an important yet underappreciated role of intra-tumoral microbiota in multiple cancers including prostate cancer⁵, lung cancer^6,7, and melanoma⁸. Exposure of the liver to the gut microbiome was also observed with positive detection of bacterial DNA in the tumor tissue from both rodent model and patients with intestine barrier dysfunction⁹. However, the existence and distribution of bacteria in the liver and the role of intra-tumoral microbiome in patients with HCC remain undefined. Furthermore, chronic HBV and HCV infections represent the leading etiology of HCC (~60–70%)¹ and the chronic viral hepatitis is always associated with gut dysbiosis¹⁰. Thus, it is of interest to investigate the existence of liver tumor-associated microbiome and to identify the unique microbial species that are associated with viral etiology which possess biological functions involved in HCC development.

Liver represents an immunosuppressive environment containing a number of suppressive cell including myeloid-derived suppressor cells (MDSCs)¹¹, Tregs and liver sinusoidal endothelial cells^12–14. MDSCs, a diverse population of immature myeloid cells that have potent immune-suppressive activity, have been reported to promote cholangiocarcinoma and HCC through inhibiting natural killer (NK) cells^9,15 or T-cell function¹⁶. Chronic infections, including viral hepatitis B and C, are commonly associated with monocytic MDSC expansion^17,18. During the process of viral pathogenesis and cancer development, liver colonized microbiota may play a role in synergistically regulating TME, especially in recruitment of suppressive cells, including MDSCs. In this study, we utilized the FFPE tissue-based multiplex immunofluorescence histochemistry (mIFH) for simultaneous in situ detection of the microbiota, the liver cancer cells, and the infiltrating immune cells, allowing us to study the TME mediated by the tissue colonized microbiota and their role in carcinogenesis.

MATERIALS AND METHODS

Patients Information and Tumor Slides.

All de-identified patient tissues utilized in this study (Table 1) were supplied as 3–4 μm FFPE sections, prepared by the Biospecimen Repository and Processing Core of the Pittsburgh Liver Research Center (PLRC) under a blanket Tissue Serum IRB Protocol PRO08010372.

Table 1.

Liver specimens and patient’s clinical characteristics

Etiology	HCC/Background	Sample Category	Sample Counts	Serum Viral Load (IU/ml)**	Gender	Age	Patients’ Tumor Grade
Viral	HCC	HBV_HCC	5	15−2.85E+04	F:0; M:5*	61~66	G1: 2; G1–G2:1; G2:2
		HCV_HCC	5	2.10E+05−1.36E+06	F:0; M:5*	55~69	G1: 1; G2:1; G2–G3:3
	Background	Viral_Background	8		F:1; M:7*	54~81	Background:3; G1–G2:1; G2:4
Non-Viral(NBNC)	HCC	NBNC_HCC	9		F:5; M:4*	62~82	G1:1; G1–G2:1; G2:6; G3:1
	Background	NBNC_Background	7		F:3; M:4*	49~82	G1:1; G2:5; G3:1

^*F: Female; M: Male

^**The most closed data point before surgery

Tumor DNA Extraction.

Prior to de-paraffinization, the FFPE tumor tissue slides were heated in 55°C for 10 min to melt the paraffin. Positioning one shorter edge of the glass slide against a clean 2 mL collection tube, 500 μL xylene were used to wet and dissolve the paraffin from the tissue and allow the excess xylene to flow into the tube. Then the entire melted tissue on the slide was scraped down into the tube. This step was repeated by adding another 500 μL fresh xylene and the remaining tissue was collected into the same tube. After vortexing vigorously for 10 s, tissues were pelleted via centrifugation at 15,000g for 10 min at room temperature (RT, 15–25°C). After removing the supernatant, 1 ml ethanol (96–100%) was added to the pellet-containing tube, which was then vortexed to extract residual xylene. Tissues were pelleted again via centrifugation at 15,000g for 10 min at RT. After carefully removing the residual ethanol, the tissue pellets were air-dried in the lid-opened tubes in a 37°C heat block.

Tissue DNA extraction was started from this step using the QIAamp DNA FFPE Tissue Kit (Qiagen, Cat# 56404) with modifications to maximize the recovery of both Gram-negative and Gram-positive bacteria from tissue¹⁹. Briefly, the tissue pellet was suspended in 180 μL Buffer ATL plus 20 μL proteinase K (provided in the kit). After brief vortexing, the tissues were then incubated at 56°C for 1 h (or until the sample has been completely lysed), followed by incubation at 80°C for 10 min to heat inactivate proteinase K. After the samples were cooled down to RT, 10 μL of 5 mg/mL MetaPolyzyme (Sigma, Cat# MAC4L-5MG in PBS, pH7.5) were added to each sample and incubated at 37°C for 1 h with shaking at 600 rpm. The digested samples in 2 mL collection tubes were briefly centrifuged and then placed on the shaker in the Qiacube for automated DNA extraction. Each extracted DNA sample was eluted in 30 μL of Buffer ATE (provided in the kit) and measured using Nanodrop2000 for concentration and purity. The DNA samples were stored at −20°C.

DNA extraction workflow utilized in this study was also validated to rule out possibility of contamination by including non-tumor containing paraffin slides as the No-Template-Control (NTC) along with the tumor FFPE slides. Using the exact same reagents and instruments, the DNA yield from NTC was “0” by Nanodrop measurement, while positive DNA yields were received from the testing tumor slides. This evidence ensures that no DNA was generated from reagents or paraffin.

16S rRNA gene amplicon library preparation and sequencing.

16S rRNA gene amplicon library preparation was performed using the QIAseq 16S/ITS screening Panel (Qiagen, Cat# 333815) following the manufacturer’s instructions with minor modifications. Briefly, The QIAseq 16S/ITS Panels utilize a 2-stage PCR workflow for targeted enrichment of 16S genes. In the first stage PCR, using the QIAseq 16S/ITS screening Panel, six amplicons (V1V2, V2V3, V3V4, V4V5, V5V7, and V7V9) that unbiasedly interrogate combinations of nine bacterial 16S variable regions covering the entire ribosomal 16S gene were amplified in 3 multiplex PCR reactions incorporated with three phased primer pools (Pool1, Pool2 and Pool3). Each 10 μL PCR reaction contained 2.5 μL of UCP Multiplex Master Mix, 1 μL of Panel Pool 1, 2, or 3, and 6.5 μL of template DNA (5 ng/μL). The conditions for PCR were as follows: 95 °C for 2 minutes to denature the DNA, with 30 cycles at 95 °C for 30 s, 50 °C for 30 s, and 72 °C for 2 min; with a final extension of 7 min at 72 °C to ensure complete amplification.

PCR products for each patient sample amplified using Pool1, Pool2 and Pool3 primers were combined for library preparation. Following a cleanup step with QIAseq Beads, pair-ended eight-base sample indices (QIAseq 16S/ITS Index Kit (Qiagen, Cat#333825)) were introduced in the second stage PCR to support multiplex up to 96 samples on a MiSeq run using the V3 sequencing chemistry. After the final cleanup, the libraries were quality controlled with Agilent 2100 Bioanalyzer and quantified using a fluorometer (Qubit, Thermo Fisher). Each library was normalized and pooled together with a concentration of 2 nM. The pooled libraries were denatured, and further diluted to a final concentration of 10 pM for sequencing on the Illumina MiSeq. Because the libraries were generated using the phased primers to increase the libraries diversity, PhiX spike-in was omitted in this study. Sequencing was performed on an Illumina MiSeq NGS system using a V3 kit with 276 × 2 paired-end run.

Droplet digital PCR (ddPCR).

Absolute bacterial quantification was performed using the QX200 Droplet Digital PCR system (Biorad). Reactions were assembled in a final volume of 22 μL containing 11 μL of 2× QX200 ddPCR Probe Supermix, 0.9 μM of each primer and 0.25 μM probe. Primers and probes used in ddPCR reactions are listed in Table S1. After generating the droplets in the automated droplet generator, PCR reactions were performed with the cycling conditions as one step of 95 °C for 5 min followed with 40 cycles of 30 s at 95 °C and 2.5 min at 55 °C. After cooling the reaction at 4 °C for 5 min, the reactions were heated up at 90°C for 5 min for enzyme deactivation, and finally holden at 10°C. Set the ramp rate for each step to 2°C/sec. After read-out on the QX200 Droplet Reader, the raw data was analyzed using the QuantaSoft Analysis Pro Software to calculate absolute copy numbers. Cutibacterium abundance was calculated as the ratio of copy numbers for Cutibacterium 16S rRNA gene to copy numbers for pan 16S rRNA genes²⁰ in per μL template DNA.

Multiplex immunofluorescence histochemistry (mIFH).

3 μm FFPE tissue slides were baked at 60 °C for one hour in a clean oven, followed by dewaxing with xylene (3 × 10 min) and rehydration through a graded series of ethanol solutions: (100% 1 × 5 min; 95% 1 × 5 min; and 70% 1 × 2 min), and then slides were washed in deionized water for 2 min. After slide fixation by 10% NBF (Neutral buffered formalin) for 20 min and washing in deionized water for 2 min, slides were treated for antigen retrieval and manually stained using the Opal Polaris 7-Color Manual IHC Detection Kits (AKOYA BIOSCIENCE, Cat# NEL861001KT, 6 reactive fluorophores and Spectral DAPI) according to the manufacture’s instruction. Cancer cells, immune cells and microbes were stained by specific primary antibodies (Table S2) in these multiple rounds of staining, resulting in that each specific cell surface marker was paired with designated Opal fluorophore signal. After the last round of antibody staining, slides were counterstained with Spectra DAPI, then mounted with ProLong^™ Diamond Antifade Mountant (Thermo Fisher, Cat#P36961) and fluorescence-grade microscopy coverslip.

Whole-slide multispectral scanning was performed on the PhenoImager^™ HT (Akoya Biosciences). Autofluorescence control slides without primary antibody but with the same staining and imaging conditions were included to subtract autofluorescence from the overall signal. Images were visualized and inspected using Phenochart (Akoya Bioscience) and analyzed using QuPath(0.4.1). Regions of interest (ROIs) were defined by positive detection of CK19+ cancer cells and colonized with either High Microbiota ROI (HM ROI, central LPS signal >100μm/pixel) or Low Microbiota ROI (LM ROI, central LPS signal <40μm/pixel) in size of area of 200 mm². Using QuPath user-trained algorithms²¹, tumor/stroma and marker+/− cells immune cells quantity were evaluated statistically from each ROI.

Bioinformatic Analyses.

Bioinformatics was run through Qiagen CLC Genomics Workbench 23, plugged in with Microbial Genomics Module and connected to licensed server. High-dimensional taxonomic biomarker discovery was performed using the LDA Effect Size (LEfSe) algorithm provided from Hunttenhower Lab Galaxy Server 2.0²². Briefly, 16S rRNA data was preprocessed to remove the background reads and trim reads on quality and performed Operational Taxonomic Unit (OTU) Clustering through SILVA taxonomy classification against SILVA 16S v132 99%−1 database²³. OTU matrix was cleaned by filtering out the unclassified, low abundant (≤4), and possible contaminated OTUs generating from 16S rRNA amplification, such as Chloroplast, Cucumis sativus (cucumber) and Solanum chacoense (Chaco potato). Then the cleaned OTUs were visualized for taxonomic profiling with relative abundances in bar chart at the taxonomic level from Phylum to Genus. Heatmap was created for the top 50 features in completed linkage measured with Euclidean distance. Using this cleaned OTU matrix, an ANOVA like linear model differential abundance test was performed on groups of samples to identify the top abundant species toward Viral_HCC (fold increased≥2 and p≤0.05) as compared to either Non-Viral_HCC or Viral_Background. The α-diversity was calculated for Shannon index using the Kruskal-Wallis test, to characterize the richness and evenness of the microbiome in each group of samples. β-diversity was evaluated by PCoA for Bray-Curtis dissimilarity or Jaccard distance.

Statistical analyses.

Intra-tumoral microbiota α-diversity analysis (e.g., Shannon entropy) in sample groups (such as Viral_HCC and NBNC_HCC) was determined by Kruskal-Wallis test and p-value <0.05 was considered as statistically significant. β-diversity analysis (e.g., Bray-Curtis Similarity and Jaccard distance) between sample groups were determined by PERMANOVA (99,999 permutations) with p<0.05. Differential abundance analysis (ANOVA-like comparison) across groups (Viral_Background vs. Viral_HCC and NBNC_HCC vs. Viral_HCC) were identified with p-value<0.05, FDR p-value≤0.05 and folder change≥4. Taxonomy with significant abundance differences between groups was identified using LEfSe at FDR 0.05, LDA ≥2.0 or ≤−2.0. All tests were two-sided. Multiple Mann-Whitney test was used to analyze the significance of marker+ cells in the defined ROI using the GraphPad Prism 10 software, p<0.05 was considered as significance.

RESULTS

Detection of intra-tumoral microbiota in the HCC samples.

Distinctive intra-tumoral microbiota of several cancer types including colorectal cancer²⁴, stomach cancer²⁵, breast cancer²⁶, and pancreatic ductal adenocarcinoma⁵, are associated with TME and cancer onset and exist in both cancer and immune cells⁸. In terms of liver cancer, few studies have been done for these low biomasses⁹, and the spatial relationships among bacterial species, tumor cells, and immune cells have not been reported. We thus first investigated the spatial location of bacteria in the malignant loci of HBV-related HCC (HBV_HCC) patient liver tumor samples through mIFH detection of both malignancy biomarker Cytokeratin 19 (CK19) and the intra-tumoral microbiota surface antigens, including lipopolysaccharide (LPS) and lipoteichoic acid (LTA) for Gram-negative and Gram-positive bacteria, respectively. Representative Hematoxylin and Eosin (H&E)-stained slides indicate the whole FFPE slide (Fig. 1A) or loci with macrotrabecular massive HCC²⁷ (Fig. 1D). Gram-negative (LPS+) or Gram-positive (LTA+) tumor microbiota were found by the whole slide scanning (Fig 1B) in a tumor sample with the positive HCC malignant cells (CK19+, Fig. 1C). In another HBV HCC tumor with HCC morphology as macrotrabecular liver with vascular invasion (Fig. 1D), tumor microbiota was detected in the same loci with the presence of tumor cells (Fig. 1E and 1F), indicating a co-existence of microbiota and HCC malignancy with geographical overlap. Furthermore, microbial bioburden in NBNC_HCC and Viral_HCC samples were also quantified via droplet digital PCR (ddPCR) using the pan-16S rRNA specific primers/probe (Suppl. Fig. 1). Since it is practically impossible to culture the microbes from FFPE tissues, we therefore conducted both the immunoassay (mIFH) and the molecular assay (ddPCR) to demonstrate the existences of the microbial components in the liver tumor. Based on the evidence that intra-tumoral bacteria have been detectable through electron microscopy in various types of solid tumors⁸, it is conceivable that microbial colonization may also occur within the liver tumor microenvironment.

Figure 1. Co-existence of intra-tumoral microbiota and malignant cancer cells in liver cancer tissues.

(A) Morphology of a representative HBV_HCC FFPE whole slide stained with Hematoxylin and Eosin (H&E). (B) Whole slide scan and detection of Gram-negative and Gram-positive intra-tumoral bacteria in mIHC with antibodies against lipopolysaccharide (LPS) (cyan) and lipoteichoic acid (LTA) (yellow), respectively. Scale bars are 1mm. White arrows indicate the representative area with positive microbiota signals. (C) Detection of CK19+ HCC malignant cells (magenta) (white arrow). Another representative HBV_HCC FFPE slide were visualized in a zoomed region with the macrotrabecular HCC morphology (D) by H&E staining (E) by mIHC for LPS+ (cyan) and LTA+ (yellow) tumor microbiota (F) by mIHC for CK19+ tumor cells (magenta). Scale bars are 100μm.

Identification of unique probiotics in the Viral_HCC and NBNC_HCC.

Microbiota profiling was conducted by 16S rRNA sequencing to illustrate the microbial composition at the Phylum level for each sample (Fig. 2A) or at Genus level (Fig. 2B) grouped as Viral_HCC, Viral_ Background, NBNC_HCC and NBNC_Background, from the cohort of liver cancer patients with either viral infection or NBNC etiology (Table 1). The abundance features were clustered in complete linkage with Euclidean distance measurement for the top 50 species in a heat map shown in Fig. 2C. In this heatmap, microbial clusters that are highly associated with Viral_HCC as compared to the other three groups of samples, including Bacteroidales, Parabacteroides, Peptoniphilus, Ruminococcus 2, Lachnoclostridium, Cutibacterium, [Eubacterium] coprostanoligenes group and Burkholderiaceae. To specify the microbial candidates that are statistically more abundant toward Viral_HCC as compared to either NBNC_HCC or Viral_Background, a linear abundance model (FDR p≤0.05 and minimal fold=1.5) analysis was performed, and 50 overlapped candidates were listed (Fig. 2D). The top selections from the candidates were determined using the cutoff as p<0.05 and fold increase (Viral_HCC) ≥ 2, which are listed in Fig. 2D.

Figure 2. Microbial signatures of tumor (HCC) or background liver tissues with viral infection or NBNC etiology.

Microbiota profiling conducted by 16S rRNA sequencing to illustrate the microbial composition at the (A) Phylum level or (B) Genus level. (C) The abundance features were clustered and visualized by a heat map. (D) Linear abundance model (FDR p<=0.05 and minimal fold= 1.5) shows the microbial candidates that are statistically more abundant toward Viral_HCC as compared to either NBNC_HCC or Viral_Background. In the 50 overlapped candidates, selected microbial candidates are listed below with the cutoff criteria as p<0.05 and fold increase (Viral_HCC) >= 2.

However, the intra-tumoral microbiome from Viral_HCC was not found with significant diversity in microbial richness and evenness by α-diversity analysis measured with Shannon entropy (Kruskal-Wallis p>0.05) as compared to other sample types, as shown in Fig. 3A. Moreover, no significant (PERMANOVA p=0.1) inter-group microbiome β-diversity was found from the comparison between Viral_HCC and NBNC_HCC in Principal Coordinates Analysis (PCoA) for Bray-Curtis distance measurement (Fig. 3B). We can conclude that although significantly abundant microbial species were found in Viral_HCC, the microbiome toward Viral HCC do not have significant differences as compared to NBNC_HCC.

Figure 3. Microbiota diversity analysis between HCCs with viral or non-viral etiology.

(A) Microbiome richness and evenness was analyzed in Shannon entropy and determined by Kruskal-Wallis in the box-and-whisker plot. Each boxplot represents median, Interquartile Range (IQR), the lowest and the highest value of the first and the third quartiles with IQR of 1.5, and the outliners. (B) PCoA of Bray-Curtis dissimilarity between Viral_HCC (clustered in red) and NBNC_HCC (clustered in black).

To further specify the viral etiology, we dissected the Viral_HCC samples into two categories as HBV_HCC and HCV_HCC. These two types of samples, along with NBNC_HCC, were reanalyzed as shown in Fig. 4A. Surprisingly, HBV_HCC shows significantly low α-diversity compared to HCV_HCC (p=0.009) (Fig. 4B). Moreover, significant microbiome β-diversity was found in HBV_HCC (clustered in black) as compared to HCV_HCC (clustered in red) (p=0.02) or to NBNC_HCC (clustered in purple) (p=0.01), indicated by PCoA with Jaccard distance, implying that HBV infection results in drastic patient dysbiosis and unique colonized tumor microbiota signature (Fig. 4C).

Figure 4. Specific microbial signatures of HBV_HCC as compared to HCV_HCC and NBNC_HCC.

(A) Microbiota profiling conducted by 16S rRNA sequencing for HBV_HCC, HCV_HCC and NBNC_HCC at the Family and the Genus levels. (B) α-diversity analysis in Shannon entropy was determined by Kruskal-Wallis test. (C) β-diversity was evaluated by PCoA of Jaccard distance between HBV_HCC (clustered in black), HCV_HCC (clustered in red) and NBNC_HCC (clustered in purple).

We performed a linear discriminant analysis of effect size (LEfSe) to characterize the contributing species in HBV_HCC compared to NBNC_HCC and the species rejected at FDR 0.05 with LDA ≥2.0 or ≤−2.0 are considered significant. By doing this analysis, taxa favored in HBV_HCC were identified as shown in red, while that in NBNC_HCC were identified as shown in green (Fig. 5A). Selected taxa candidates identified in Fig. 5A were demonstrated by their abundances, shown in Fig. 5B. With the cutoff of 200 OTUs, Dolosigranulum, Prevotella 9, Cutibacterium and Nocardioides were selected as the favored taxa biomarkers in HBV_HCC while Chryseobacterium was selected in NBNC_HCC. Having the highest OTU abundance among the selections, Cutibacterium is considered as the representative taxa biomarker in HBV_HCC. This taxa biomarker was further confirmed by ddPCR to calculate the absolute abundance of Cutibacterium shown as the ratio of Cutibacterium 16S rRNA to pan-16S rRNA. Cutibacterium was found significantly (p<0.05) more in HBV_HCC as compared to NBNC_HCC (Fig. 5C).

Figure 5. High-dimensional taxonomic biomarker identification toward HBV_HCC.

(A) In LEfSe analysis at FDR 0.05 with LDA score ≥2.0 or ≤−2.0, selected taxa favored in HBV_HCC are shown in red, while that in NBNC_HCC are shown in green. (B) The abundances of each selected taxa in either HBV_HCC samples (black) or NBNC_HCC samples(red) are visualized in bar chart with standard deviation. (C) Absolute abundance of Cutibacterium was calculated by ddPCR showing the gene copies ratio of the Cutibacterium 16S rRNA to the pan 16S rRNA. Median and Interquartile Range have been labeled in each group of results (dot: NBNC_HCC; square: HBV_HCC). (*p<0.05; two-sided Mann-Whitney test).

Intra-tumoral microbiota are positively associated with increased tumor infiltrating CD8+ T lymphocytes, but not the CD56+ NK cells, in HBV_HCC.

To investigate the biological function of the intra-tumoral microbiota in modulating TME, we employed mIFH to simultaneously detect malignant cancer cells and infiltrating immune cells in situ within a defined region of interest (ROI), with either high or low density of microbial signal. The expression patterns and geographic associations between microbiota, CD8+ T cells, CD56+ NK cells and CK19+ cancer cells are shown in Fig. 6. Fig. 6A shows the expression pattern of either High Microbiota ROI (HM ROI) or Low Microbiota ROI (LM ROI) with defined size of area (mm²), from an HBV-HCC sample. In the HM ROI, surrounding to the CK19+ tumor cells, massive LPS+ Gram-negative and LTA+ Gram-positive bacterial signals have been observed. In the same region, the infiltrating CD8+ T cells and CD56+ NK cells were also detected. Interestingly, in the LM ROI, CD8+ T cells quantity was significantly lower than that in the HM ROI; CD56+ NK cells were mainly localized to the macrotrabecular-massive HCC²⁷ without significant reduction of quantity in LM ROI (Fig. 6B).

Figure 6. Tumor loci with high density of intra-tumoral microbiota are associated with increased CD8+ lymphocytes, but not the CD56+ NK cells, in HBV_HCC.

(A) ROIs in malignant tumor loci were determined by CK19+ tumor cells surrounding with massive colonized LPS+ Gram-negative and LTA+ Gram-positive bacteria in defined 200 mm² area. A representative High Microbiota (HM) ROI or Low Microbiota (LM) ROI from an HBC_HCC sample is shown by mIHC for CK19+ tumor cells (magenta), CK19+LPS+ (magenta, cyan) and CK19+ LTA+ (magenta, yellow). (B) In the same High Microbiota ROI or Low Microbiota ROI or as shown in (A), infiltrating CD8+ T cells (orange) and CD56+ NK cells (red) were also detected. Scale bars are 50 μm. (C) CD8+ lymphocytes and CD56+ NK cells were quantified from 5 of High Microbiota ROIs and 5 of Low Microbiota ROIs from (C) HBV_HCC patient 1 tumor sample and (D) HBV_HCC patient 2 tumor sample. Increased CD8+ infiltrating was found significantly (p<0.05) associated with the higher quantity in both samples. In contrast, CD56+ NK cells were not found to have significantly quantitative difference in either HM or LM ROIs (ns, p>0.05).

Using the QuPath user-trained algorithms²¹, immune cells quantity were evaluated statistically from two HBV_HCC samples. In each sample, 5 of HM ROIs and 5 of LM ROIs were selected for quantification. Increased levels of tumor infiltrating CD8+ T lymphocytes were found significantly associated with the higher quantity of the intra-tumoral microbiota (p<0.05) (Fig. 6C). In marked contrast, CD56+ NK cells, which has been reported to play a regulatory and proangiogenic function linked to liver injury²⁸, were not found to have significantly quantitative difference in either HM or LM ROIs (Fig. 6D). These data indicate that the tumor microbiota alter the TME primarily through recruiting CD8+ T cells. Interestingly, such correlations between the mass of intra-tumoral microbiota and recruitment of CD8+ T cells and CD56+ NK cells were not found in NBNC_HCC samples (Suppl. Fig. 2). This result implies that chronic HBV infection may result in a unique microenvironment that favors the distinct microbiota colonization, which in turn alter the TME by increasing CD8+ T cells infiltration.

Intra-tumoral microbiota are positively associated with increased tumor infiltrating monocytic MDSCs and polymorphonuclear MDSCs in HBV_HCC.

MDSCs have been reported to promote cholangiocarcinoma and viral pathogenesis^17,18, We set out to assess the association between these immune suppressive cells recruitment and tumor microbiota by counting two subtypes of MDSCs: monocytic MDSCs (M-MDSCs, CD14+CD11b+) and polymorphonuclear MDSCs (PMN-MDSCs, CD15+CD11b+)²⁹ in tumor cell (CK19+) containing ROIs colonized with either high (Fig 7A, image A and B) or low density of microbiota (Fig 7A, image C and D). Fig. 7B shows the spatial imaging of CD11b+, CD14+, CD15+, CD11b+CD14+ and CD11b+CD15+ cells in either HM or LM ROIs. Through quantifying the specific immune cells from HM ROIs and LM ROIs to evaluate the immune contexture about myeloid derived immune-suppressor cells, we found that, although the total leukocyte (CD11b+) showed no difference in the two types of ROIs, both CD14+ and CD15+ cells were statistically more abundant in the HM ROIs (Fig. 7C). This result indicates that intra-tumoral microbiota in HBV_HCC promote MDSCs recruitment and potentially accelerate the disease progression via inhibiting host antitumor immunity. However, neither monocytic MDSCs nor polymorphonuclear MDSCs exhibited positive correlation with the amount of microbiota in the NBNC_HCC (Supple. Fig. 3), inferring that the recruitment of MDSCs may be a synergistical effect from both viral pathogenesis and tumor microbiota.

Figure 7. Intra-tumoral microbiota are positively associated with increased monocytic MDSCs and polymorphonuclear MDSCs in HBV HCC.

(A) Representative HM ROI and LM ROI with CK19+ tumor cell(magenta), colonized LPS+ Gram-negative(cyan) and LTA+ Gram-positive(yellow) bacteria in defined 200 mm² area. (B) In the same High Microbiota ROI or Low Microbiota ROI as shown in (A), spatial imaging of CD11b+ (green), CD14+ (orange), CD15+ (red), CD11b+CD14+(green, orange) and CD11b+CD15+ (green, red) cells were visualized by mIHC. Scale bars are 50 μm. (C) From 4 of High Microbiota ROIs and 4 of Low Microbiota ROIs in the HBV HCC sample, CD11b+ leukocytes were not seen difference in the two types of ROIs, but both of CD14+ and CD15+ cells were statistically more abundant in the HM ROIs (ns: p>0.05, *p<0.05; two-sided Mann-Whitney test).

DISCUSSION

Hepatocarcinogenesis is well-known for its heterogeneity and generally develops in the setting of chronic hepatitis or cirrhosis^2,30. Various factors contribute to the etiology of HCC, with virus-associated mechanisms driving hepatocarcinogenesis accounting for approximately 80–90% of cases³¹. HBV and HCV are often categorized together as one risk factor in HCC studies. However, HBV and HCV are distinct viruses and cause HCC via different molecular mechanisms. While both viruses encode gene products that may directly or indirectly affect hepatocyte proliferation and differentiation to induce HCC via a plethora of virus-host interactions, HBV is a DNA virus, and the random integration of viral DNA into the host genome is potentially an additional cancer driver causing alteration of oncogenes and/or tumor suppressor genes expression as well as host genome instability^32,33. Contrary to HBV, HCV is a single-stranded RNA virus without integration in viral life cycle³⁴. Nonetheless, it is generally acknowledged that the long-term hepatic inflammation induced by chronic HBV or HCV infection shapes an immune microenvironment to promote liver disease progression towards HCC development^32–34.

Chronic viral hepatitis is associated with liver fibrosis, cirrhosis and always linked with increased gut permeability and gut dysfunction¹⁰. Among viral HCC patients, overrepresented genera Bacteroides and Veillonella were found in the HBV HCC patients^35,36, whereas a reduction of microbial diversity in association with overrepresented Streptococcus and Lactobacillus was observed in the HCV HCC patients³⁷. This taxonomic disparity indicates that different gut microbiome-mediated biological pathways may be involved in HCC across different etiologies.

Liver has unique anatomic features, notably enriched vessels with heightened permeability and a dual vascular supply from both the systemic arterial and portal venous systems rendering it susceptible to the seeding of disseminated tumor cells³⁸ and exposure of gut-derived products¹⁰. To mitigate immune responses against self-antigens, including microbiota and intestinal products, and to avoid liver damage³⁸, the liver creates an immunosuppressive environment housing various suppressive cell types including MDSCs¹¹, Tregs and liver sinusoidal endothelial cells^12–14. These physiological conditions provide an opportunity for microbiota translocation and colonization in the liver, as evidenced by the detection of bacterial DNA in the liver tumor tissue from both rodent model and patients with intestine barrier dysfunction⁹. However, spatial evidence demonstrating the presence of microbiota in the liver is lacking, and the biological function of these microbial species remains unclear.

In this study, we characterized the HCC intra-tumoral microbiome in human liver samples obtained from the PLRC Biospecimen Repository. To identify the specific microbiota toward the etiology of viral infection, microbiome composition and diversity analysis have been performed. Based on the composition and abundance of the microbiota, a set of bacteria were found to colonize more in the Viral_HCC, as compared to NBNC_HCC (Fig. 2D), including Ruminococcus 2, Alcaligenes, Bacteroidales and Flavonifractor. This result is consistent with another cohort study in which Ruminococcus gnavus was identified as taxa biomarker for Viral HCC³⁹. However, in our study, there was no significant diversity difference between Viral_HCC and NBNC_HCC (Fig. 3). The limited sample size could potentially account for the lack of significance in diversity; however, this finding may also suggest heterogeneity within liver cancer and its associated microbiota. Interestingly, after dissecting the viral tumor samples into subgroups of HBV_HCC and HCV_HCC, we found that HBV_HCC-associated tumor microbiota possesses significant low α-diversity as compared to HCV_HCC, as well as significant β-diversity as compared to either NBNC_HCC or HCV_HCC. The reasons for such virus-specific phenomena could be multifaceted but considering that HBV and HCV cause HCC via different mechanisms encompassing both host and viral factors, chronic hepatitis B may create a distinct environment conducive to unique microbiota colonization. It is worth noting that, while the enrolled HBV_HCC patients had received antiviral nucleoside analogue treatment, but none had eliminated the virus, highlighting the role of HBV in HCC development.

An inherent limitation of our study is the relatively small sample size. Despite the scarcity of available liver cancer tumor samples, we acquired an equal number of HBV_HCC and HCV_HCC tumor samples (5 each) along with 10 nonviral HCC samples, enabling us to conduct this preliminary study and derive statistically significant metagenomic data. Gender disparities in HCC incidence have been observed globally^40,41, and our study similarly indicates a higher prevalence of Viral_HCC patients among males compared to females (Table 1). While there is no reported evidence demonstrating a direct relationship between gender and the tumor microbiome, research has shown that sex hormones such as estrogen and testosterone can influence the gut microbiome⁴². Moving forward, we aim to expand our study by including more viral HCC samples as they become accessible through PLRC and other available biobank resources. In order to address limitations found in previous intra-tumoral microbiome studies dealing with specimens with low microbial content¹⁹, we developed a specific workflow to maximize the recovery of both Gram-negative and Gram-positive bacteria from tissue samples, and the taxa biomarkers identified through metagenomic approaches were validated by the absolute quantification method using ddPCR. By doing so, Cutibacterium was identified as a representative taxa biomarker for HBV_HCC (Fig. 5), which significantly towards to each HBV_HCC sample regardless the viral load of HBV measured before surgery. Interestingly, Cutibacterium (formerly Propionibacterium) has been found to be highly prevalent in prostate tissue among patients diagnosed with prostatitis and prostate cancer^43,44. A higher burden of Cutibacterium acnes was also found to be associated with a distinct sub-group of thyroid cancer with poor prognosis⁴⁵. Long term exposure to this microbial species can incite carcinogenesis through activation of cancer-associated pathways, such as AKT, ERK, EZH2 and ATF2, to trigger cell transformation and inflammation⁴⁶.

Although HCC-associated tumor microbiome has been investigated by others before³⁹, the previous efforts were mainly focused on taxonomic profiling of microbial composition using metagenomic approaches. In this study, we characterized the spatial signal of microbiota-derived components, liver cancer cells, and immune cells to assess the association between microbiota density and specific immune cells populations in the background of HBV infection (Fig. 6–7). It has been reported that the accumulation of MDSCs occurs in tumors with hepatic metastasis in response to various cancer-related growth factors and cytokines⁴⁷. MDSCs exhibit antitumor activity by inducing T cell dysfunction through the production of reactive oxygen species (ROS), arginase-1 (ARG1), and nitric oxide synthase (NOS)⁴⁸. MDSCs have also been reported to promote cholangiocarcinoma and HCC through inhibiting natural killer (NK) cells^9,15 or T-cell function¹⁶. Surprisingly, in HBV_HCC, inflamed tumor region with high density of microbiota was found with increased CD8+ T cells and both PMN-MDSCs and M-MDSCs (Fig. 6–7). Chemokines serve as crucial mediators of inflammation and play an important role in controlling viral and bacterial infections by facilitating chemotactic cell recruitment, including the MDSCs recruitment to tumor sites. In previous studies, higher levels of CXCL2 and IL-8 (also known as CXCL8), the chemotactic factors for neutrophils, have been observed in acute-on-chronic liver failure⁴⁹, and the accumulation of hepatic MDSCs has been found to be associated with gut microbiota through a TLR4/CXCL1/CXCR2129-dependent mechanism⁹. Therefore, we reason that chronic HBV infection and tumor microbiota colonization synergistically accelerate the expressions of peripheral chemokines and chemokine receptors within the high microbiota loci, resulting in the elevated MDSCs infiltration into the tumor. In contrast, recruitment of NK cells, the important innate immune cell subset involved in immune surveillance of hematological malignancies and solid tumors as well as metastatic spreading, were not found to be significantly correlated with intra-tumoral microbiota in HBV_HCC (Fig. 6). In future studies, profiling the cytokines and chemokines within HBV_HCC tumors exhibiting high and low microbiota levels will offer insights into the distinct recruitment patterns of immune cells within specific TMEs.

The pathogenesis and clinical manifestations by chronic HBV infection involve increased intestine permeability¹⁰, MDSC expansion^14,15, liver injury, and disorganized and leaky vasculature, creating conditions for microbiota translocation into the liver and the colonization at the malignant sites. This unique microbial colonization leads to significant infiltration of immune suppressive cells, which hinders host antitumor immunity and potentially impacts the cancer treatment outcomes. In a subset of melanoma patients resistant to PD-1 blockade treatment, their baseline microbiota showed strong association with MDSCs accumulation in tumor biopsies⁵⁰. In these regards, a deeper understanding of tumor microbiome functions across different cancer etiologies may open avenues for developing strategies to manipulate the tumor microbiome as a therapeutic approach in cancer treatment.

DATA AVAILABILITY STATEMENT

Datasets of 16S rRNA sequencing in this study have been deposited in the NCBI BioProject database (https://www.ncbi.nlm.nih.gov/sra/PRJNA1074337). Other data that support the findings of this study are available on request from the corresponding authors upon reasonable request.