Spatial transcriptional profiling of CHB liver biopsies reveals an undetected population of zonally biased HBV-integrated cells

Abstract

Background & Aims

HBV covalently closed circular DNA (cccDNA) and HBV-integrated DNA (iDNA) are features of chronic HBV (CHB). Elimination of both cell populations is likely to be required for functional cure. RNA-based spatial methods, chromogenic in situ hybridization (CISH), and multiplex immunofluorescence (mIF) combined with machine learning methods were used to evaluate archival liver biopsy samples from patients with CHB.

Methods

Modified Visium Spatial Transcriptomics (ST) and Single Nuclei RNA sequencing (snRNA-Seq) were performed on four commercial HBV⁺ biopsies, and six liver biopsies from the clinical trial GS-US-174-0149. mIF for HBV viral proteins and CISH for HBV RNA and DNA were also performed. Machine learning was used to annotate ST data for the presence of cccDNA vs. iDNA. A deep learning (DL) algorithm was used to perform classification based on HBsAg mIF stain patterns and integrated into the ST analytical framework.

Results

The experimental workflow detected HBV RNA and DNA, alongside the human transcriptome. We identified an undescribed population of iDNA cells, which are HBsAg mIF^- but produce comparable levels of viral RNA to iDNA cells that are HBsAg mIF⁺. These cells exhibit a periportal zonal preference in their spatial location within the liver. Analysis of a subset of CHB biopsies by snRNA-Seq confirmed a higher burden (approximately three-to seven-fold) of HBs RNA+ nuclei compared with HBsAg mIF⁺ cells.

Conclusions

Our findings indicate the presence of a previously undescribed population of zonally enriched HBV-integrated hepatocytes that are not detectable by protein-based methods. Intrahepatic viral burden estimates by RNA-based methods were consistently higher than protein-based estimates in all CHB samples profiled.

Impact and implications

This study addresses crucial gaps in understanding HBV infection heterogeneity, zonal distribution, and the interplay between viral RNA, protein expression, and HBV integration events. Protein-based measures of intrahepatic HBV have been recognized as incomplete indicators of HBV burden. Accurate characterization of the viral landscape is essential for developing safe and efficacious therapies aimed at achieving a functional cure. Using RNA-based approaches, we uncovered a hidden HBV burden, which was consistently higher compared with protein-based estimates and was predominantly driven by HBV iDNA. Moreover, we identified two spatially and phenotypically distinct iDNA hepatocyte populations, raising new questions about mutational load and hepatocellular carcinoma risk in regions previously overlooked. Access to RNA-based methods that capture the full extent of HBV burden within the liver is vital for advancing fundamental research, improving disease models, and guiding curative therapies.

Clinical Trials registration

The study is registered under GS-US-174-0149.

Graphical abstract

Highlights

• •Functional cure of HBV requires substantial reduction of HBV⁺ hepatocytes.
• •Accurate quantification of hepatic HBV burden is essential development of novel therapies.
• •RNA-based HBV burden estimates are higher compared with protein-based estimates.
• •A subpopulation of HBV integrated cells appear negative by protein-based assays.
• •Hepatocytes with HBV RNA^- and protein discordance exhibit periportal localization.

Introduction

Approximately 250 million people worldwide are chronically infected with HBV, with a significant increased risk of liver disease.¹ In livers of those with chronic hepatitis B (CHB), hepatocytes containing either HBV covalently closed circular DNA (cccDNA) or HBV integrated into the host genome (iDNA) are present,^2,3 and recent studies have indicated that elimination of both populations is required for HBV cure.⁴ The current standard of care for CHB is treatment with nucleos(t)ide analogs (e.g. tenofovir alafenamide fumarate [TAF] and ‘NUC’), which suppress viral replication, but do not directly impact cccDNA or iDNA. Accordingly, they improve long-term outcomes but rarely lead to cure.⁴ Thus, new therapies to treat CHB are needed, and a better understanding of intrahepatic viral burden and spatial distribution of HBV-positive populations is important for development of safe and effective curative agents.

HBV is a partially double-stranded DNA virus, which is repaired to form cccDNA. Replication occurs through an RNA intermediate (pregenomic RNA, pgRNA) by reverse transcription, resulting in the formation of relaxed circular DNA (rcDNA). HBV double-stranded linear DNA (dslDNA) can also be produced from pgRNA to form iDNA. The structure of dslDNA in iDNA separates the HBV core open reading frame from its promoter, but HBV surface (HBs) and HBV x (HBx) transcripts can still be generated.^3,5

CHB biopsy studies to-date have revealed that staining for the viral proteins HBsAg and HBcAg is largely mutually exclusive at the individual cell level.[5], [6], [7], [8], [9] The presence of HBsAg⁺/HBcAg^- cells is consistent with hepatocytes containing iDNA.[10], [11], [12], [13] Expanded patches of HBsAg⁺/HBcAg^- cells have also been observed, which are likely to be indicative of clonal expansion of HBsAg-expressing iDNA. The lack of HBsAg staining in HBcAg⁺ cells is unexpected because cccDNA⁺ cells express both HBcAg and HbsAg.⁵ This lack of co-staining in presumed cccDNA-containing cells has previously been hypothesized to result from differences in assay detection limits or in the kinetic release of HBsAg in cccDNA vs. iDNA cells.⁶ Interestingly, previous studies have also reported a poor correlation between intrahepatic HBsAg as measured by mIF vs. peripheral HBsAg levels across CHB disease stages,⁶ whereas positive correlations have been observed between liver HBV DNA burden and peripheral HbsAg.^7,14,15 Together, these observations led us to hypothesize that protein-based assays might not fully represent tissue burden.

HBV infection has a heterogenous, patchy nature and therefore, we leveraged spatial transcriptomics (ST), a tissue-based RNA technology, to quantify HBV and the host transcriptome of cells in their native spatial contexts. Visium (10X Genomics, California, USA) generates spot level (55 μm) whole-transcriptome data and is compatible with archival, formalin-fixed paraffin embedded (FFPE) tissue; however, because of the highly degraded nature of FFPE RNA, transcriptional profiling is typically probe based. Single nuclei RNA-sequencing (snRNA-Seq) from 10X Genomics is a similar, probe-based platform adapted for FFPE tissue.

Here, we describe modifications to the Visium and snRNA-Seq workflows by designing HBV-specific probes to enable co-detection of viral RNA and DNA alongside the human transcriptome. To compare infection burden between RNA and protein-based assays, we performed multiplex immunofluorescent (mIF) staining for viral antigens on serial liver biopsy sections. We further describe an analytical pipeline developed to co-register the HBsAg mIF image with the H&E ST image, enabling integration of viral antigen status into the ST analytical framework. This experimental workflow was validated on FFPE blocks from HBV-infected primary human hepatocytes (PHH), the HBV iDNA hepatoma cell line (PLC/PRF/5), and HBV-infected human liver chimeric mice treated with either vehicle or TAF. Following validation, this workflow was implemented on commercially sourced human CHB FFPE liver surgical biopsies, and FFPE liver core needle biopsies collected from patients with CHB enrolled in the GS-US-174-0149 clinical trial.

Data from these studies successfully demonstrated co-detection of HBV RNA and DNA alongside human transcripts. Comparison of cccDNA- and iDNA-containing spot transcriptomes confirmed previously described enrichment in inflammatory pathways. Our RNA-based analysis revealed that tissue burden of iDNA is consistently higher than previously reported by protein-based assays. Thus, we describe a previously undescribed population of HBV-integrated cells, which stain negative for HBsAg by mIF despite producing comparable amounts of HBs RNA to that produced by HBsAg mIF⁺ spots. These HBs RNA⁺/HBsAg mIF^- cells are largely spatially distinct from HBsAg⁺ cells and exhibit zonally biased localization within the liver.

Materials and methods

Spatial transcriptomics and snRNA-Seq probe design

Consensus sequences for HBV GTB and GTC¹⁶ were generated to create a list of potential 50-bp probes tiled across regions of interest in the genome. Probes that aligned to human genome hg38 or did not meet the design guidelines outlined in the 10X Genomics Custom Probe Design for Visium Spatial Gene Expression document CG000621 were removed. The list was further filtered to exclude probes that did not align to multiple subtypes, had reverse complementarity to other probes, or overlapped with probes targeting sequences with better conservation across subgenotypes. Custom probes were purchased from Millipore Sigma with modifications as described in the 10X Genomics Custom Probe Design for Visium Spatial Gene Expression document CG000621.

Visium: Spatial transcriptomics

ST libraries were generated using the protocol outlined in 10X Genomics Visium CytAssist Spatial Gene Expression User Guide (CG000495) with the Visium CytAssist Spatial Gene Expression for FFPE, Human Transcriptome kit (10X Genomics, Cat#PN-1000522). A probe cocktail comprising 24 nM of each custom pair (17 LHS and 17 RHS probes) were combined with the Visium Human Transcriptome Probe Set v2.0 as described in the modified probe hybridization mix of 10X Genomics (CG000621). Additional details are available in the supplementary methods.

Single nuclei RNA-Seq (FFPE): Library preparation

To prepare the snRNA-Seq library, 2 × 25-μm scrolls were sectioned from FFPE donor blocks using a Leica RM2255 microtome and placed in gentleMACS™ C Tubes (Miltenyi Biotec, Cat #130-093-237). The scrolls were processed through the 10x Genomics Sample preparation protocol for GEM-X Flex gene expression workflow for FFPE samples (CG000784- Rev A). Additional details are available in the supplementary methods.

Transcriptome quantification

Spike-in probe sequences were appended to human reference genome refdata-gex-GRCh38-2020-A and annotation file Visium human transcriptome probe set v2.0_GRCh38-2020-A downloaded from 10X Genomics. Illumina sequencing reads were aligned using Spaceranger-3.0.1. Additional details are available in the supplementary methods.

Multi-modal image analysis workflow

An image analysis pipeline was developed to co-register the ST and mIF datasets at spot-level resolution. The workflow involved a series of transformations to co-register the mIF image with the ST image from the serial section. Following co-registration, a DL algorithm trained for three-class classification was applied to the HBsAg mIF stained image. Following the mIF-based HBsAg classification, the overlap between cells (mIF image) and ST spots (H&E image) were determined and a group of mIF stained cells were assigned to the corresponding ST spot. Additional details are available in the supplementary methods.

Results

ST with custom spike successfully detects HBV RNAs and DNA in HBV-infected liver biopsies

Liver FFPE blocks were obtained from HBV stably infected urokinase-type plasminogen activator/severe combined immunodeficiency (uPA-SCID) mice (n = 1 Vehicle and n = 1 ‘TAF’), healthy human liver surgical biopsy samples (n = 2), CHB liver surgical biopsy samples (n = 4 ‘POC’), and CHB liver core needle biopsies from the GS-US-174-0149 clinical trial (n = 4 ‘baseline’ and n = 2 ‘NUC’). FFPE cell pellet arrays were prepared from uninfected PHH, HBV-infected PHH (genotype C, GTC), and the iDNA PLC/PRF/5 cell line. Serial 5-μm sections were analyzed using a modified ST protocol with custom spiked HBV probes, mIF, and chromogenic in situ hybridization (CISH) (Fig. 1A). The snRNA-Seq (FFPE) workflow requires 50-μm (2 × 25-μm) FFPE curls and was only performed on the commercially sourced CHB liver (n = 3) and healthy human (n = 1) surgical biopsy samples. Of the 18 GS-US-174-0149 clinical trial samples that passed RNA quality control (DV200 ≥30%; Fig. S1A), only six samples were included in downstream analyses (Fig. S1B). Detailed quality control (QC) metrics and clinical metadata for the POC and GS-US-174-0149 cohorts are provided in Table S1.

Fig. 1.

ST with custom spikes successfully detects HBV RNAs and DNA in HBV-infected liver biopsies.

(A) Schematic of experimental workflow performed on FFPE blocks and (B) locations of HBV-targeting probes. (C) Spatial maps of log-normalized gene expression values for HBs-P, HBx-P, HBc-P, and asDNA ST probes across n = 1 Healthy-, CHB-POC, baseline-, and NUC-treated cohorts. (D) Representative example from the POC cohort of spatial maps for HBs-P and HBc-P by ST and mIF images (scale bars: 800 μm, low magnification, and 50 μm, high magnification) for HBsAg/DAPI (top row), HBcAg/DAPI (bottom row), and CISH images (scale bars: 100 μm) for HBs-P¹ (top row) and HBc-P¹ (bottom row). (E) Representative example from POC cohort of spatial maps of log-normalized gene expression values for HBs-P (top row), HBc-P (middle row), and asDNA (bottom row) and CISH images (scale bar: 60 μm) for HBs-P¹² and PreS-S-X³⁵ probes. Scale bars: 100 μm for all ST spatial maps. CHB, chronic HBV; CISH, chromogenic in situ hybridization; FFPE, formalin-fixed paraffin-embedded; mIF, multiplex immunofluorescence; ST, spatial transcriptomics.

Seventeen HBV-specific probes targeting genotypes B, C, and D were designed (Fig. 1B). Fourteen probes detected both sense DNA and mRNA (‘ST probes’), while three targeted the anti-sense DNA strand (‘asDNA’). ST probes were directed against HBs (HBs-P), HBx (HBx-P), HBV core (HBc-P), and the Direct Repeat 1 (DR1-P) sequence. Probe sequences and their corresponding HBV genomic coordinates are listed in Table S2 and were used for both ST and snRNA-Seq workflows. CISH probes of varying lengths (1ZZ, 12ZZ, and 35ZZ) were designed to detect either RNA and DNA strands or DNA only (Table S2).

To validate and characterize the probes, we performed ST and CISH on two HBV in vitro models: a cccDNA-representative GTC HBV-infected PHH cell line and a replication-incompetent PLC/PRF/5 iDNA cell line. Probe sequences used for the ST and CISH probes are listed in Table S2. Consistent with expectations, HBV-infected PHH cells were positive for HBs-P, HBc-P, and asDNA by ST. These cells were also positive by CISH, using the HBs-P¹² RNAScope probe, which detects both DNA and RNA. By contrast, PLC/PRF/5 cells showed HBs-P positivity but were negative for HBc-P and asDNA by ST (Fig. S2A). To further confirm the presence of replication competent cccDNA in HBV-infected PHH and its absence in PLC/PRF/5 cells, we used an RNAScope CISH probe (PreS-S-X³⁵) targeting the anti-sense DNA strand. High copy number, nuclear and cytoplasmic PreS-S-X³⁵ signals were noted in HBV-infected PHH, consistent with it being replication competent and containing cccDNA. A low copy number, nuclear signal was observed in PLC/PRF/5 cells, consistent with their containing HBV integrations. The absence of an asDNA ST signal in PLC/PRF/5 cells led us to hypothesize that the asDNA ST signal originates from replicating rcDNA and that ST lacks sensitivity for detecting integrated DNA. To test this, we performed ST on uPA-SCID mice stably infected with HBV GTC for 8 weeks, followed by 28 days of treatment with either vehicle or TAF to inhibit viral replication (Fig. S2B)¹⁷. TAF-treated animals showed a near-complete loss of asDNA signal, supporting the hypothesis that most of the asDNA probe signal was derived from rcDNA (Fig. S2C). A detailed overview of ST and CISH probes, mIF stains, and their respective targets (viral RNA, DNA, or protein) is provided in Table S3.

No HBV-specific probe signal was detected in healthy liver samples (Fig. 1C). In CHB tissues, ST revealed patchy positivity for HBs-P, HBx-P, HBc-P, and asDNA, likely reflecting clonal expansion of iDNA and cccDNA-containing cells. ST-positive regions for HBs-P and HBc-P corresponded with HBsAg⁺ and HBcAg⁺ regions by mIF (Fig. 1D). BaseScope CISH probes (HBs-P¹ and HBc-P¹; Table S2) were designed to target the same sequences (<50 bp) as the ST probes. Consistent with the ST results, regions of the tissue that were positive for HBs-P or HBc-P by ST, were positive for HBs-P or HBc-P by CISH, respectively (Fig. 1D). Additional representative images of HBsAg and HBcAg mIF staining in CHB and healthy liver, cell line controls and uPA-SCID humanized mice, alongside corresponding spatial maps for HBs-P and HBc-P are provided in Fig. S3.

In CHB biopsies, ST-positive spots for HBs-P, HBc-P, and asDNA, along with PreS-S-X³⁵ CISH positivity, showed increased cytoplasmic and nuclear foci, indicative of cccDNA and replicating rcDNA (Fig. 1E, solid black arrowheads). By contrast, presumed iDNA regions (HBs-P⁺, HBc-P^-, asDNA^-) exhibited fewer nuclear-only HBV DNA foci (Fig. 1E, open black arrowheads), consistent with integrated HBV DNA. Comparative ST, CISH, and mIF targeting profiles for cccDNA^- and iDNA regions in CHB samples are summarized in Table S4.

Automated ST guided annotation confirms previously described enrichment of inflammatory pathways in HBV-infected liver tissue

To classify HBV-infected regions within liver tissue, ST signals from HBs-P, HBx-P, HBc-P, and asDNA was used to train a supervised machine learning model (Random Forest Classifier, RFC). The model annotated ST spots into three categories: (1) uninfected; (2) iDNA-positive; and (3) cccDNA positive (Fig. 2A; Fig. S4A). Training was based on manual annotations informed by the number and localization of PreS-S-X³⁵ CISH probe foci. A published scRNA-Seq dataset¹⁸ was used to exclude spots that were estimated to contain <25% hepatocyte transcriptional content from further analysis. ST spots classified as cccDNA were HBs-P⁺/HBx-P⁺/HBc^-P⁺/asDNA^±. Spots classified as iDNA were HBs-P⁺/HBx-P⁺/HBc-P^-/asDNA^-. Uninfected spots were negative for viral RNA and DNA by ST.

Fig. 2.

Automated ST guided annotation confirms previously described enrichment of inflammatory pathways.

(A) Schematic of spot-labeling matrix based on Random Forest Classifier-derived outputs. (B) Representative example of the predicted spot categories on a sample from the baseline cohort. Dashed circles are used to illustrate the low-frequency ‘uninfected’ and ‘low hepatocyte’ spots. (C) Log-normalized gene expression values of HBs, HBc and asDNA across viral categories in POC, baseline and NUC cohorts. (D) Dot plots comparing expression of IFIT2, IFI27, CXCL9, and CXCL10 across all three cohorts. (E) Spatial maps of log-normalized gene expression of IFI27 and CXCL9 across viral categories in a representative POC sample. Bottom row ST scale bar: 100 μm. IF, interferon-stimulated gene; ST, spatial transcriptomics.

Given that each ST spot could encompass between two and eight hepatocytes, classifications represent dominant signal profiles but might include mixed cell populations. For example, cccDNA-classified spots might contain iDNA or uninfected cells, and iDNA-classified spots might include uninfected hepatocytes. Uninfected spots were derived from CHB samples with no detectable viral signal, whereas ‘healthy’ spots originated from healthy liver samples with sufficient hepatocyte content and no viral signal. Representative images for regions classified as uninfected, iDNA-positive, and cccDNA-positive from a clinical biopsy are shown in Fig. 2B. As expected, HBs-P ST signal was present in both iDNA- and cccDNA-containing spots, whereas HBc-P and asDNA ST signals were restricted to cccDNA positive regions (Fig. 2C, top and middle panels). Notably, NUC-treated samples exhibited a marked reduction in asDNA signal within cccDNA classified spots, consistent with suppression of viral replication (Fig. 2C, bottom panel).

To further characterize the transcriptional landscape, differentially expressed genes (DEGs) were identified between cccDNA- and iDNA-containing spots across the patient cohorts (log₂ fold-change >1.0, adjusted p <0.05; Fig. S4B). In both the POC and baseline cohorts, cccDNA- and iDNA containing spots showed elevated expression of interferon-stimulated genes (ISGs) (e.g., IFIT2 and IFI27) and genes encoding chemokines (e.g., CXCL9 and CXCL10) compared with uninfected spots, with the highest expression observed in cccDNA-positive regions (Fig. S2D). Spatial maps from a representative POC sample demonstrated localized enrichment of IFI27 and CXCL9 (Fig. 2E). Pathway analysis further supported these findings, revealing significant enrichment of inflammatory pathways in cccDNA-positive spots from both POC and baseline cohorts. By contrast, cccDNA-positive spots from NUC-treated samples showed reduced inflammatory pathway activation, potentially reflecting diminished immune responses resulting from suppressed viral replication or prevention of new infection events (Fig. S4C).

Paired mIF and ST reveal a previously undescribed population of HBV integrated cells

Consistent with previous studies, mIF identified two distinct HBV-positive hepatocyte populations in CHB liver samples: (1) HBcAg⁺/HBsAg^± cells, likely representing cccDNA-containing hepatocytes; and (2) HBcAg^-/HBsAg⁺ cells, presumed to harbor integrated HBV DNA (iDNA). These populations were corroborated by ST: HBcAg⁺/HBsAg^± mIF cells were HBs-P⁺/HBc-P⁺ by ST, consistent with cccDNA presence (Fig. 3Aa), whereas HBcAg^-/HBsAg⁺ mIF cells were HBs-P⁺/HBc-P^- by ST, consistent with iDNA (Fig. 3Ab).

Fig. 3.

Paired mIF and ST reveal previously undescribed population of HBV integrated cells.

(A) CHB mIF: (a) (top row) representative cccDNA: HBs-P⁺ by ST and CISH, HBc-P⁺ and asDNA⁺ by ST and HBcAg+/HBsAg- by mIF; (b) (middle row) representative of iDNA: HBs-P⁺ by ST and CISH, HBc-P^- and asDNA^-, and HBcAg^-/HBsAg⁺ by mIF. Panel c (bottom row): HBs-P⁺ by ST and CISH, HBc-P^- and asDNA-, and HBsAg⁺/HBcAg⁺ by mIF. CISH scale bars: 60 μm, mIF scale bars: 20 μm. (B) Spatial maps for HBs-P and mIF staining (HBsAg/DAPI) for uninfected (top row), HBV-infected PHH (middle row), and PLC/PRF/5 (bottom row). (C) Representative region of CHB mIF: (a) (top row) ‘Weak HBsAg’ and (b) (bottom row) ‘Strong HBsAg’; mIF scale bars: 50 μm. (D) Schematic of mIF and ST H&E multi-modal integration workflow. (E) Decisional matrix for spot labels based on HBsAg mIF patterns. (F) Spot labels for mIF staining transferred to the ST framework. Scale bars for all ST spatial maps: 100 μm. CHB, chronic HBV; iDNA, integrated DNA; mIF, multiplex immunofluorescence; PHH, primary human hepatocytes; ST, spatial transcriptomics.

Unexpectedly, ST revealed a third hepatocyte population characterized by HBs-P⁺ via ST and CISH, but lacking HBcAg and HBsAg expression by mIF (Fig. 3Ac). These HBcAg^-/HBsAg^-/HBs-P⁺ cells were also negative for HBc-P and asDNA by ST, suggesting that they represent a previously undescribed subset of iDNA-containing hepatocytes. This transcript–protein discordance mirrors findings in the PLC/PRF/5 iDNA cell line, which expresses and secretes HBsAg despite being HBsAg negative by mIF[19], [20]^, (Fig. 3B; Fig. S5A).

Further analysis of the HBsAg⁺ mIF regions revealed two subpopulations based on signal intensity: ‘Weak HBsAg’ and ‘Strong HBsAg’ (Fig. 3C). To systematically classify these populations, a multi-modal image analysis workflow was developed to automate HBsAg classification and spatially co-register mIF and ST data, integrating protein expression into the transcriptional landscape (Fig. 3D, Fig. S5B and C). Spots were labeled as ‘Weak’ or ‘Strong’ if ≥80% of constituent cells matched the respective classification. Spots were labeled as ‘Mixed’ if no dominant cell type was present, and ‘Negative’ if all cells were HBsAg^- by mIF (Fig. 3E). The spatial distribution of these classifications for the region shown in Fig. 3C is presented in Fig. 3F. A similar integrative pipeline was developed for HBcAg to assess whether infection burden estimated by RNA-based ST differed from protein-based mIF staining, enabling a more nuanced understanding of HBV transcriptional and translational heterogeneity in CHB liver tissue.

RNA-based assays reveal a higher burden of HBV-positive hepatocytes compared with protein-based detection

ST identified HBs-P⁺ spots that lacked detectable HBsAg protein by mIF, suggesting that RNA-based methods may capture a greater number of HBV-positive hepatocytes than protein-based assays alone. Unlike the single-cell resolution provided by mIF (Fig. 3A), ST spots encompassed multiple hepatocytes, which could include both HBcAg⁺/HBsAg^± (cccDNA) and HBcAg^-/HBsAg⁺ (iDNA) phenotypes. To assess detection sensitivity, we combined cccDNA and iDNA classifications to compare the frequency of HBV-positive spots identified by ST vs. mIF. ST-based detection (defined by HBs-P+ signal) and consistently identified a higher proportion of HBV-positive spots across all three cohorts compared to mIF-based detection (defined by HBsAg ‘Weak’, ‘Mixed’, or ‘Strong’ and/or HBcAg positivity) (Fig. 4A). The variability in ST-detected cccDNA and iDNA spot frequencies across samples likely reflects the inherent heterogeneity of HBV infection and known sampling variability (Fig. 4B).⁶ The discrepancy between ST and mIF detection appears to be driven, at least in part, by differences in HBs ST vs. HBsAg mIF positivity. The frequency of HBsAg⁺ mIF spots among cccDNA and iDNA categories is detailed in Fig. S6A, and per-sample distributions are shown in Fig. S6B.

Fig. 4.

HBV tissue positivity by RNA-based assays is higher than estimates based on protein-based assays.

(A) Tissue spot positivity across all cohorts comparing the frequency of HBsAg^+/- and/or HBcAg⁺ spot positivity by mIF with iDNA or cccDNA ST positivity. (B) Tissue spot positivity across all cohorts representing relative distributions of iDNA and cccDNA by ST. (C) Comparison of spot-level HBs-P by ST and spot-level HBsAg/HBcAg by mIF vs. HBs-P at the single nuclei level by snRNA-Seq and HBsAg/HBcAg mIF at the single cell level. (D) Distribution of HBs-P log-normalized expression values across viral categories and HBsAg mIF spot labels in POC (top row), baseline (middle row) and NUC (bottom row). The Wilcoxon rank-sum test was used to compare means between HBsAg mIF-negative and HBsAg mIF-positive classes. (E) Distribution of HBsAg mIF staining patterns across all three cohorts. cccDNA, covalently closed circular DNA; iDNA, integrated DNA; mIF, multiplex immunofluorescence; snRNA-Seq, single nuclei RNA sequencing; ST, spatial transcriptomics.

Given that ST spots could contain between two and eight hepatocytes and were scored as HBV positive if even a single cell was infected, we further evaluated infection burden at the single cell level using snRNA-Seq (FFPE) on three samples from the POC cohort with matched ST and mIF data. A healthy liver FFPE block processed in parallel served as a negative control and showed no HBV-positive cells. Consistent with ST findings, snRNA-Seq revealed a three-to seven-fold higher frequency of HBV-positive hepatocytes compared with mIF-based quantification (Fig. 4C, Fig. S6C and D).

To determine whether the absence of HBsAg protein signal was due to reduced HBs-P levels, we compared HBs-P signal across spots classified by mIF as ‘Weak’, ‘Strong’, ‘Mixed’, or ‘Negative’ for HBsAg (Fig. 4D). Across all three cohorts, mean HBs-P levels varied by less than one-fold between groups, with no statistically significant differences observed. The distribution of the HBsAg mIF classifications across uninfected, cccDNA, and iDNA spots is shown in Fig. 4E.

HBs-P⁺/HBsAg^- hepatocytes are enriched in the periportal zone of the liver

Spatial analysis revealed distinct zonal localization patterns of HBV-infected hepatocytes based on HBsAg expression. Strong HBsAg⁺ iDNA cells were predominantly located near the central vein (zone 3, pericentral, denoted as CV), as confirmed by colocalization with the canonical zone 3 marker, CYP2E1 (Fig. 5A). By contrast, HBsAg^- mIF^- iDNA cells were enriched near the portal vein (zone 1, periportal, denoted as PV), validated by expression of the canonical zone 1 marker, HAL and corresponding H&E staining. Importantly, ST confirmed that the absence of HBsAg protein in periportal regions was not due to reduced HBs RNA levels, because the HBs-P signal remained detectable (Fig. 5A).

Fig. 5.

HBs-P ST-positive, HBsAg mIF-negative spots display a periportal preference.

(A) Representative iDNA region of HBsAg mIF staining from the POC CHB cohort highlighting the transition from ‘Strong HBsAg’ in hepatocytes surrounding the CV to ‘Negative HBsAg’ in hepatocytes surrounding the PV, alongside spatial maps showing log-normalized gene expression values for HBs-P, HAL and CYP2E1 by ST. (B) Dot plot comparing expression of zone 1 and zone 3 curated gene lists across the HBsAg mIF spot labels. The size of the dot represents the percentage of spots expressing the genes of interest, and the colors represent the amount of expression per spot, with blue representing high expression and gray representing low expression. Scale bars: 50 μm and 100 μm (A). CHB, chronic HBV; CV, central vein; iDNA, integrated DNA; mIF, multiplex immunofluorescence; PV, periportal vein; ST, spatial transcriptomics.

To further investigate zonal gene expression patterns, curated gene sets representing canonical zone 1 and zone 3 markers were analyzed across HBsAg mIF categories in all three cohorts. Dot plots in Fig. 5B show that iDNA_Strong spots were enriched for zone 3 genes, whereas iDNA_Negative spots preferentially expressed zone 1 genes (Fig. 5B). These zonal gene enrichments were statistically significant in both the POC and baseline cohorts (Table S5), although not in the NUC cohort, likely because of a limited sample size. Pathway-level analysis supported these findings (Fig. S7).

Together, these data suggest the existence of two distinct classes of iDNA-containing hepatocytes: (iDNA_S) cells, which secrete HBsAg and localize to the pericentral zone, and (iDNA_A) cells, which accumulate HBsAg and are enriched in the periportal zone (Fig. 6). This zonal bias might influence the efficiency of HBsAg secretion and contribute to the spatial heterogeneity of HBV infection within the liver.

Fig. 6.

HBV iDNA phenotypes. Schematic depicting the RNA/protein landscape of iDNA hepatocytes. Pericentrally enriched iDNA-A are hepatocytes with iDNA accumulate detectable HBsAg by IHC/mIF, whereas periportally enriched iDNA-S are hepatocytes with iDNA that likely secrete HBsAg with high efficiency and, thus, do not stain by HBsAg IHC/mIF. iDNA, integrated DNA; IHC, immunohistochemistry; mIF, multiplex immunofluorescence.

Discussion

This study presents a comprehensive spatial and molecular profiling of HBV infection in human liver biopsies through the integrated application of ST, CISH, and mIF, with key findings validated by snRNA-Seq. By combining these modalities, we were able to interrogate the spatial landscape of HBV infection and compare infection burden estimates derived from RNA-vs. protein-based assays. To minimize manual bias, we developed DL algorithms to automate classification of HBsAg mIF staining patterns (Negative, Weak, or Strong) and implemented a pipeline to integrate protein expression data into the transcriptional landscape. This enabled high-throughput, spatially resolved analysis of HBV-infected tissue across thousands of ST spots. Robust detection of viral RNA and DNA was achieved in cell line models, HBV-infected mouse livers, commercially sourced CHB surgical resections, and needle core biopsies from the GS-US-174-0149 clinical trial. Although the sample numbers analyzed was limited, the ‘mini bulk’ nature of ST provided statistically powered insights into infection dynamics. Notably, the declining availability of CHB biopsies and the stringent RNA quality requirements of ST technologies constrain inclusion for such types of analysis. Surprisingly, despite meeting the RNA integrity thresholds, over half of the needle core biopsies from GS-US-174-0149 failed to yield usable ST data and had to be excluded from analyses. This is likely because of the age of the liver biopsy, differences in storage temperature (ambient vs. cold storage at 4 °C), exposure to ambient light and storage in nonhygroscopic conditions. This is an important consideration for future studies aimed at profiling archival clinical trial biopsies. Consistent with previous work showing increased CD8⁺ T cell infiltration near HBcAg⁺ cells compared to HBsAg⁺ cells, our spatial analyses revealed enrichment of ISGs and inflammatory pathways in proximity to cccDNA-containing hepatocytes relative to iDNA⁺ cells and uninfected hepatocytes.²¹

A key observation was the discordance between RNA and protein detection, with the number of HBs-P⁺ spots identified by ST approximately double that of HBsAg⁺ and/or HBcAg⁺ spots detected by mIF. To address the limitations of spot-level resolution, we performed snRNA-Seq on a subset of CHB samples and found that RNA-based detection revealed a three-to seven-fold higher infection burden compared with mIF. Although this analysis was limited to three samples and tissue sections were ∼100 μm apart, the findings underscore the sensitivity of RNA-based approaches and highlight the need for larger-scale studies spanning the natural history and treatment phases of CHB.

Importantly, HBs RNA⁺, HBsAg mIF^- iDNA-containing spots were consistently observed across all CHB biopsies, suggesting that RNA-based methods detect a broader spectrum of HBV-infected cells. Despite comparable HBs RNA levels between HBsAg⁺ and HBsAg^- spots, only a subset of iDNA cells were detectable by mIF, raising questions about differences in HBsAg synthesis, stability, or secretion. Transcriptionally guided zonal analysis, alongside qualitative histological review, revealed that HBsAg^- iDNA cells were enriched in the periportal zone, whereas HBsAg⁺ cells localized to the pericentral zone. This spatial distribution aligns with previous findings that pericentral hepatocytes support viral replication and HBcAg expression,[22], [23], [24] although zonal differences in iDNA localization have not been previously reported Our findings indicate that transcriptional differences between iDNA_Strong and iDNA_Negative populations were largely attributable to metabolic specialization associated with liver zonation. As previously mentioned, we cannot entirely exclude the possibility that the absence of HBsAg mIF staining in regions where expression is anticipated might be due to limitations to the technical sensitivity of the assay. Nonetheless, given the comparable levels of HBs mRNA and consistent zonal spatial enrichment, it is unlikely that technical factors alone account for the HbsAg-negative population.

Previous laser capture microdissections studies of HBsAg-positive regions in CHB tissues revealed mutations and deletions in the PreS region of integrated HBV, which have been linked to altered HBsAg accumulation and cellular stress responses.[25], [26] These alterations are thought to contribute to the early stages of HCC development. Given that the association between PreS mutations and HCC risk has been primarily based on analysis of HBsAg-positive regions, our current findings prompt the question of whether a similar mutational burden and corresponding HCC risk exists within the iDNA_HBsAg mIF negative population. Notably, a recent study reports the incidental presence of both iDNA and PreS deletions in HbsAg-negative regions of tissue at frequencies comparable to those observed in HBsAg positive- and tumor regions.²⁷

At a bulk tissue level, PreS mutations/deletions were absent or present at very low frequencies in our cohort, suggesting they do not account for the observed HBsAg patterns (data not shown). This low frequency of PreS deletions was too small to account for the expanded patches of brightly positive HBsAg cells and, thus, HBsAg accumulation in our current samples was unlikely to be related to PreS deletions. These PreS mutations have also been linked with upregulation of cellular stress,[25], [26] neither of which were observed in our study. Furthermore, periportal hepatocytes are known to exhibit enhanced protein secretion capacity,²⁸ which could explain the lack of intracellular HBsAg detection by mIF despite active transcription.[19], [20] Given the probe-based nature of our assays, we were unable to assess the impact of integration site variability on the observed iDNA phenotypes. Nonetheless, our findings reveal a previously unrecognized dichotomy among iDNA-containing hepatocytes: those that accumulate HBsAg intracellular (iDNA-A) and those that efficiently secrete it (iDNA-S). This distinction, along with the spatial and transcriptional heterogeneity of HBV infection, underscores the complexity of the intrahepatic viral landscape.

In conclusion, HBV infection within the liver is more heterogeneous than previously appreciated, shaped by the interplay of cccDNA vs. iDNA, clonal integration events, and zonal variation in HBsAg retention. The identification of iDNA-A and iDNA-S populations provides new insights into HBV biology and highlights the need for therapeutic strategies that address the full spectrum of infected hepatocyte phenotypes.

Abbreviations

cccDNA, covalently closed circular DNA; CHB, chronic HBV; CISH, chromogenic in situ hybridization; cv, central vein; DEG, differentially expressed gene; DL, deep learning; dslDNA, double-stranded linear DNA; FFPE, formalin-fixed paraffin-embedded; iDNA, integrated DNA; ISG, interferon-stimulated gene; mIF multiplex immunofluorescence; pgRNA, pregenomic RNA; PHH primary human hepatocytes; pv, portal vein; rcDNA, relaxed circular DNA; RFC, Random Forest Classifier; snRNA-Seq, Single Nuclei RNA sequencing; ST, Spatial Transcriptomics; TAF, tenofovir alafenamide fumarate; uPA-SCID, urokinase-type plasminogen activator/severe combined immunodeficiency.

Authors’ contributions

SM conceived and designed the study, executed ST/CISH assays, generated figures, and wrote the manuscript. RR designed probes and analyzed ST and snRNA-Seq data, and generated analytic outputs. HM developed the multi-modal integration pipeline. CM developed and performed mIF staining. AM performed ST and snRNA-Seq assays. EV performed sectioning. RW performed CISH. SC, HCC, and KHS performed sequencing of ST and snRNA-Seq libraries. AA reviewed mIF data. AR processed ST data. LL oversaw analysis of data. SPF and SB provided critical review of manuscript. LD oversaw conception and execution of the study and supported data interpretation. MMH provided control reagents, oversaw study execution and analysis, interpreted data and wrote the manuscript. All authors read and approved the final manuscript.

Data availability

Sequencing files from PHHs, PLC/PRF/5, PXB livers, and commercially procured liver biopsies have been deposited in the NCBI Gene Expression Omnibus (www.ncbi.nlm.nih.gov/geo/) under GSE306575. Data from clinical trial samples are available upon request.

Financial support

This work was funded by Gilead Sciences Inc.

Conflicts of interest

SM, RR, JM, CM, EV, RW, SC, HC, KS, AA, AR, LL, SPF, SB, LD, and MMH are employees and stockholders of Gilead Sciences, Inc.

Please refer to the accompanying ICMJE disclosure forms for further details.

Supplementary data

The following are the Supplementary data to this article: