Unraveling the Tapestry: Lessons from Multiomics and Spatial Biology in Hepatocellular Cancer

Abstract

Hepatocellular carcinoma (HCC) is a biologically and clinically heterogeneous cancer in which the spatial organization of the tumor microenvironment dictates progression, immune escape, and therapeutic response. Traditional genomic and transcriptomic profiling have revealed key molecular pathways, yet they do not capture the spatial relationships that govern tumor–immune–stromal interactions. Spatial omics technologies now bridge this gap, allowing molecular profiling within intact tissue architecture and revealing how cellular neighborhoods, immune barriers, and stromal niches shape tumor behavior. In this review, we summarize recent advances in technologies like spatial transcriptomics, proteomics, genomics, epigenomics, and metabolomics as applied to HCC. We discuss how these technologies have illuminated the spatial dynamics of the invasive margin, macrophage and fibroblast niches, and immune evasion, identifying both prognostic and predictive biomarkers. We further highlight how spatially resolved analyses have uncovered previously unrecognized axes of communication between cancer cells and key stromal and parenchymal populations that were not apparent from bulk or single-cell molecular analyses alone. These spatially mapped ligand–receptor networks reveal coordinated signaling circuits that define targetable mechanisms of resistance and progression within the microenvironment. Finally, we explore the translational implications where spatial signatures may guide risk stratification, treatment selection, and clinical trial design. By integrating spatial architecture with molecular function, spatial omics is transforming our understanding of liver cancer biology and paving the way for spatially informed precision oncology.

1. Introduction

Hepatocellular carcinoma (HCC) is the most common primary liver malignancy and a leading cause of cancer-related death worldwide (1,2). It most often arises in the setting of chronic liver injury caused by hepatitis B or C virus infection, alcohol-associated liver disease, or metabolic dysfunction–associated steatotic liver disease (MASLD), though aflatoxin exposure, genetic disorders, and other causes of chronic inflammation also contribute. These conditions lead to progressive fibrosis and cirrhosis, establishing a chronically inflamed microenvironment that fosters oncogenic transformation. Despite recent therapeutic advances, HCC remains a cancer with limited responsiveness to locoregional and systemic therapies (3). Immune checkpoint inhibitors (ICIs) such as anti-PD1 or anti-CTLA4 yield objective response rates of only ~15–30% in advanced disease, and reliable biomarkers to guide treatment selection are still lacking (4). Molecular profiling has revealed key pathways and immune signatures, yet these approaches cannot readily explain why tumors with similar clinical stages or even similar transcriptomic profiles exhibit widely variable responses to therapy.

HCC tumor microenvironment forms a complex and spatially organized ecosystem of malignant hepatocytes, immune infiltrates, and stromal components. These cellular populations interact through cytokines, chemokines, and direct cell–cell contacts to generate an immunosuppressive milieu that promotes tumor initiation, progression, and resistance to therapy. Genomic and sequencing studies have greatly advanced our understanding of the HCC microenvironment, however, they usually have relied on bulk profiling, which averages signals across cell populations and obscures spatial and cellular heterogeneity (4,5). Increasing evidence suggests that spatial organization, rather than cellular abundance alone, governs tumor phenotype. This spatial heterogeneity likely underlies the clinical variability in outcomes among patients with comparable stage or treatment exposure. Understanding this dimension of tumor biology is therefore critical for developing more precise, spatially informed therapeutic strategies. In recent years, spatial omics technologies have revolutionized our ability to map the molecular and cellular landscape of the tumor microenvironment in situ (6-8). This review focuses on spatial omics applications in HCC, highlighting how spatial biology technologies have illuminated intratumoral architecture, immune interactions, and tumor heterogeneity. We will first outline major spatial omics technologies, then highlight how they have illuminated key biological processes in HCC, and finally discuss their translational and clinical implications.

2. A primer on the Spatial Omics Technologies Mapping Tumors

Spatial omics refers to a collection of methods that measure molecular profiles of transcripts, proteins, or metabolites in tissue sections while preserving their spatial coordinates. Unlike single-cell sequencing which dissociates cells and loses all positional context, spatial omics can create maps of gene or protein expression across intact tissue. Several platforms have been developed, each with unique strengths. We will review some of the more commonly used ones below (Figure 1, Table 1).

Figure 1. Overview of major spatial omics technologies and their key features.

(A) Spatial transcriptomics methods are broadly categorized into sequencing-based and imaging-based approaches. Sequencing-based platforms (e.g., Visium, Slide-seq, Curio Seeker) capture and barcode RNA transcripts from tissue sections, providing whole-transcriptome coverage at multicellular to near single-cell resolution. Imaging-based methods (e.g., MERFISH, seqFISH, CosMx) use iterative probe hybridization to visualize thousands of targeted RNAs at subcellular resolution, though with limited gene panels.

(B) Spatial proteomics includes multiplex immunofluorescence and imaging mass cytometry, which quantify tens of proteins with subcellular precision. Digital spatial profiling combines barcoded antibodies or RNA probes with optical readouts, enabling high-plex protein and RNA detection from defined regions of interest.

(C) Spatial genomics and epigenomics techniques map DNA and chromatin architecture in situ, spanning from targeted genomic loci to genome-wide resolution. These methods, such as spatial ATAC-seq and spatial-CUT&Tag, allow integration of chromatin accessibility and spatial organization but are currently limited by sample type compatibility.

(D) Spatial metabolomics applies mass spectrometry imaging (MSI) to map metabolite distributions within tissues at subcellular to tissue-level resolution. While label-free and capable of three-dimensional reconstruction, most MSI-based workflows require fresh or frozen samples due to limited FFPE compatibility.

Table 1. Major spatial omics technologies and key features.

Summary of commonly used spatial omics platforms, outlining their principles, resolution, sample compatibility, and main applications in HCC research.

Methods/Commercial Platform	Technology	Features	Limitations	References
Spatial transcriptomics
Sequencing-based
10x Genomics- Visium/Visium HD	Spatially barcoded 3′ mRNA capture sequences array spots that encode a spatial barcode, which can be resolved by NGS to reconstruct the spatial location.	2-55μm resolution Printed barcodes Suitable for FFPE/fresh-frozen Whole transcriptome (A-tailed RNAs)	Not single cell resolution Lack of subcellular granularity	9
Curio Seeker- Slide Seq and HDST	Beads-based spatially barcoded 3′ mRNA capture sequences that can be spatially resolved.	2-10μm resolution Bead-based barcodes Suitable for fresh-frozen Whole transcriptome (A-tailed RNAs)	Not single cell resolution No FFPE samples	10,11,12
Imaging-based
Vizgen- MERFISH and seqFISH+	RNA detection based on multiple reporter probe hybridization to target RNAs over multiple rounds	Single cell, subcellular resolution Targeted RNAs (upto 10K) Suitable for FFPE/fresh-frozen High gene detection efficiency Single molecule resolution	Multiple rounds of probe hybridization Cell segmentation challenges	13,14
10x Genomics- Xenium and NanoString- CosMx	Xenium utilizes padlock probe amplification, followed by fluorescence in situ sequencing CosMX uses branched probes against target RNAs detected by reporters	Single cell, subcellular resolution Targeted RNAs (Xenium, 5K, CosMx, 6K) Suitable for FFPE/fresh-frozen Can co-detect RNA and proteins Large scale spatial transcriptomics	Limited gene expression Cell segmentation challenges	15,16
Bio-Techne- RNAscope	RNA detection by multiple probes for specific RNAs, with rounds of pre-amplification, amplification, and fluorophore binding	Single cell, subcellular resolution High sensitivity Low cost, and easy to use FFPE/fresh-frozen Suitable for validations or clinical use Can co-detect RNAs and proteins	Low number of targets upto 12 genes Limited multiplexing	17
Spatial proteomics
Multiplexed Immunofluorescence
Akoya Biosciences- PhenoCycler (CODEX)	Cyclic immunofluorescence based on in situ hybridization of fluorescent probes that hybridize to index barcodes tagged to antibodies.	Single cell, subcellular resolution Targeted proteins (20-100) Suitable for FFPE/fresh-frozen	Low number of targets in a test Limited availability of antibodies (upto 350)	19
Imaging mass cytometry	Protein labeling by heavy metal-conjugated antibodies detected with mass spectrometry	Single cell resolution Targeted proteins (30-50) Suitable for FFPE/fresh-frozen	Low number of targets in a test Tissues gets ablated	20
Spatially resolved proteogenomic profiling
Bruker- nanoString GeoMx	Detection of RNAs or proteins with specific photocleavable probes	Single cell resolution Whole transcriptome and proteins (570+) Suitable for FFPE/fresh-frozen High sensitivity	Pre-selected region of interest in a tissue Limited subcellular granularity	21
Spatial Genomics
DNA seqFISH+, Slide DNA-seq	seqFISH+ uses sequential rounds of fluorescently labeled probes to detect DNA directly in situ Slide DNA-seq capturesDNA by spatially barcoded beads	Single cell resolution 3660 chromosomal loci to whole genome Suitable for cells/fresh-frozen	Not suitable for FFPE samples Low spatial granularity	23
Spatial Epigenomics
Spatial CUT&Tag, Spatial ATAC	CUT&Tag utilizes Tn5 tethered to specific histone marks ATAC uses Tn5-inserted adapters into open chromatin	Chromatin mapping at 20-55μm resolution Suitable for cells/fresh-frozen	No single cell resolution Not suitable for FFPE samples	24,25
Spatial Metabolomics
Mass Spectrometry Imaging (IMS)	Uses a focused beam (SIMS)/laser (MALDI) or charged microdroplets (DESI) to ionize molecules	SIMS (subcellular resolution) MALDI (single cell) DESI (35-200μm of resolution) Suitable for fresh-frozen Label-free method 3D possibility with SIMS	Limited compatibility with FFPE samples Variability in ionization efficiency	26,27,28

2.a. Spatial Transcriptomics (ST)

This approach captures RNA transcripts along with their location within the tissue. Broadly, current ST platforms can be divided into two categories: capture-based high-throughput sequencing (HTS) approaches and imaging-based in situ hybridization or sequencing methods. Each employs distinct principles and offers complementary strengths and limitations. Capture-based HTS approaches rely on barcoded oligonucleotide arrays or beads to capture polyadenylated RNA from tissue sections. After reverse transcription and sequencing, the barcodes enable reconstruction of spatially resolved transcriptomes. The widely used Visium Spatial Gene Expression platform (10x Genomics) places tissue sections on glass slides containing printed capture spots (~55 μm diameter), each representing the pooled transcriptome of 1–10 cells (9). This design allows whole-transcriptome profiling across large tissue areas but with multicellular resolution. Next-generation array-based systems, such as Slide-seqV2 and High-Definition Spatial Transcriptomics (HDST), improve spatial precision by using densely packed barcoded beads, as small as 10 μm, that capture transcripts at near single-cell resolution (10-12). Similarly, new-generation barcoded substrates like Curio Seeker combine fine spatial resolution with higher RNA capture efficiency through hydrogel-based surface chemistry. These HTS-based methods provide broad, unbiased transcriptome coverage ideal for discovery, but their main limitations include loss of subcellular information, and the need for complex computational alignment between sequencing data and histological images.

In contrast to array-based capture technologies, imaging-based spatial transcriptomics visualizes RNA molecules in situ through iterative rounds of hybridization or sequencing within intact tissue sections. These methods achieve true single-cell or even subcellular resolution while preserving tissue architecture. Techniques such as MERFISH, seqFISH+, and CosMx Spatial Molecular Imager employ large panels of fluorescently labeled probes to detect hundreds to thousands of predefined transcripts across multiple imaging cycles (13,14). This enables highly multiplexed detection with exceptional spatial precision, resolving transcript localization within individual cells or even specific cellular compartments. More recently, Xenium (10x Genomics) has extended this in situ imaging paradigm by using barcoded oligonucleotide probes to directly quantify hundreds to thousands of RNA species within tissue sections at subcellular resolution (15). While comparable to CosMx in spatial precision, Xenium is optimized for seamless integration with Visium datasets, enabling multiscale analyses that link whole-transcriptome discovery with targeted high-resolution validation. By providing transcript-level spatial coordinates with cell segmentation and flexible panel design, Xenium supports quantitative neighborhood, proximity, and niche analyses that are particularly well suited for studying complex tumor microenvironments. The CosMx SMI platform provides additional flexibility by allowing investigators to predefine regions of interest (ROIs) for high-plex spatial analysis (16). This targeted acquisition strategy allows researchers to focus on specific niches such as tumor margins, immune aggregates, or fibrotic septa, optimizing both data richness and cost-efficiency. RNAscope represents a complementary in situ hybridization technology that combines high sensitivity with single-transcript resolution (17). Although traditionally used for target validation, newer multiplexed RNAscope assays can simultaneously visualize multiple RNA targets, typically up to 12 per tissue section, while preserving the same subcellular precision. This makes RNAscope a powerful orthogonal tool for validating and extending spatial transcriptomic findings. While imaging-based methods offer unmatched spatial resolution and morphological fidelity, their trade-offs include limited gene coverage, extended imaging cycles, and complex data processing relative to sequencing-based platforms. Nonetheless, they remain indispensable for delineating the high-resolution architecture of gene expression within complex tissues such as the liver.

2. b. Spatial Proteomics and Multiplexed Imaging

Spatial proteomics refers to a diverse array of technologies that enable multiplexed protein detection while preserving tissue architecture. These methods have been designated as the “Method of the Year 2024” by Nature Methods, reflecting their transformative impact on understanding biological complexity (18). Proteins are the primary executors of biological function, but they do not always correlate directly with transcript abundance due to post-transcriptional regulation, stability differences, and post-translational modifications. Moreover, many clinically relevant biomarkers and therapeutic targets in HCC are best characterized at the protein level. To profile proteins in situ, researchers have developed three major techniques- multiplexed immunofluorescence, mass-spectrometry-based Imaging, and spatially resolved proteogenomic profiling. CODEX (Co-Detection by Indexing) is a notable example of iterative multiplex immunofluorescence staining and imaging, where antibodies are conjugated to unique DNA barcodes; after binding and washing steps, the barcodes are sequenced or fluorescently tagged across cycles, enabling visualization of 40–100 proteins on the same tissue (19). In contrast, imaging mass cytometry (IMC) uses antibodies labeled with heavy metal isotopes and an ion beam to detect their abundance at each spot. These yield single-cell resolution images with panels of ~30–50 proteins, with no spectral overlap issues since detection is by mass spectrometry (20). One significant advantage is that both CODEX and IMC can work on formalin-fixed, paraffin-embedded (FFPE) tissue, which is invaluable for analyzing clinical samples and archived cohorts. Another technique, the Nanostring GeoMX Digital Spatial Profiler (DSP), bridges genomics and proteomics, it uses photocleavable oligo-tagged antibodies or RNA probes to measure 100s of targets in user-defined regions of interest (ROIs) (21). For instance, researchers can select tumor vs. stromal areas on a slide and get protein expression profiles for each region (21,22). DSP is useful for validating spatial differences in larger cohorts or in FFPE material when high resolution single-cell mapping is not needed. Collectively, spatial proteomic methods complement transcriptomic technologies by revealing the functional state of cells within their native architecture. Together, they provide a bridge between high-resolution discovery and clinically translatable biomarker development in HCC.

2.c. Other Spatial technologies

Beyond transcriptomic and proteomic profiling, emerging spatial genomics, epigenetics, and metabolomics approaches are expanding the molecular depth of spatial biology. Spatial genomics enables the localization of DNA alterations such as copy number changes, mutations, or chromosomal rearrangements within intact tissue architecture, using methods like in situ sequencing or spatial DNA FISH (23). Spatial epigenetic profiling maps chromatin accessibility, histone modifications, or DNA methylation patterns at cellular resolution, illuminating how local chromatin states shape gene regulation within distinct microenvironments (24,25). Meanwhile, spatial metabolomics employs advanced imaging mass spectrometry (e.g., MALDI-IMS, DESI-MS) to visualize the distribution of small molecules and metabolites directly in tissue sections, linking metabolic zonation or reprogramming to histologic features (26-28). Together, these modalities complement spatial transcriptomics and proteomics by providing a multidimensional view of tissue organization, gene regulation, and function.

2.d. Choosing the right spatial technology

Several factors determine the right spatial biology technology employed for a particular question-

Tissue Type and Preservation-

Formalin-fixed, paraffin-embedded (FFPE) specimens dominate clinical archives, making FFPE compatibility a practical priority for translational studies. Methods such as NanoString GeoMX DSP, CosMx SMI, CODEX, and imaging mass cytometry are all validated for FFPE tissues and can be applied retrospectively to large patient cohorts. However, RNA integrity in FFPE material is often compromised, limiting the performance of capture-based spatial transcriptomics that require intact RNA (e.g., 10x Visium). Fresh-frozen tissue, while less abundant, preserves RNA quality and is preferable for discovery-oriented transcriptomic studies using Visium, Slide-seq, or Curio Seeker. Similar tissue preservation considerations apply to spatial epigenetic and metabolomic workflows, which often require fresh or flash-frozen material to maintain native chromatin and metabolite integrity.

Resolution and Scale-

The required spatial resolution depends on the biological scale of interest. For microenvironmental mapping at the level of cellular neighborhoods or immune niches, single-cell imaging platforms such as CosMx, MERFISH, or CODEX are ideal. For broader tissue-level gradients, such as zonation patterns or fibrosis-associated gene expression, capture-based methods (Visium, Slide-seqV2) or region-based profiling (GeoMX DSP) provide sufficient resolution while covering larger areas. In general, we balance resolution against throughput- high-resolution imaging for mechanistic insight, and lower-resolution capture for cohort-level validation.

Discovery vs. Hypothesis-Driven Studies-

Unbiased, whole-transcriptome sequencing-based approaches (Visium, Slide-seq) are well suited for discovery since they can identify new molecular programs or unexpected cellular states. In contrast, targeted imaging assays (CosMx, MERFISH, RNAscope, CODEX) are optimal for hypothesis-driven studies, where specific pathways, signaling molecules are being interrogated in situ. Spatial genomics and epigenomics methods, though still emerging, increasingly enable targeted exploration of copy-number changes, DNA accessibility, or methylation landscapes within defined niches thus bridging molecular genetics with spatial biology.

Molecular Layer of Interest-

Transcriptomics provides insight into gene regulation and cellular identity, while proteomics directly reflects functional states, receptor activation, and post-translational modifications. Spatial epigenomics adds another layer by mapping chromatin accessibility and histone marks that regulate transcriptional activity, while spatial genomics localizes DNA alterations such as copy-number variation or clonal mutations. Spatial metabolomics complements these by visualizing metabolic gradients and zonation patterns that are especially relevant in liver cancer. In immune-oncology contexts such as HCC, where checkpoint expression and cytokine signaling define therapeutic response, integrating transcriptomic discovery with proteomic, epigenetic, or metabolomic validation can yield the most actionable information. Platforms like GeoMX DSP and CosMx SMI uniquely enable multi-platform RNA–protein analysis from the same tissue section, bridging these layers.

Pathological Feature of Interest-

In addition to tissue type, resolution, and molecular layer, the dominant pathological feature being studied should guide the choice of spatial technology. Different features, such as vascular architecture, tumor invasion patterns, immune compartmentalization, fibrosis, or zonation, place distinct demands on spatial resolution and preservation of tissue context. For example, in highly vascular organs such as the liver and in vascularized tumors such as HCC, questions related to vessel density, microvascular invasion, or immune cell localization relative to blood vessels are best addressed using imaging-based spatial platforms that preserve histologic architecture. High-plex imaging approaches, including imaging mass cytometry and cyclic multiplex immunofluorescence, enable simultaneous visualization of endothelial cells, pericytes, tumor cells, and immune populations. These platforms support quantitative analyses of perivascular niches and intravascular versus extravascular immune localization. In contrast, when the pathological feature of interest involves broader transcriptional programs, such as angiogenic signaling or hypoxia-associated states, capture-based spatial transcriptomic methods are well suited for discovery. Similar considerations apply to other pathological features, including fibrotic remodeling, tumor–stroma interfaces, or clonal expansion patterns. Together, these examples highlight the importance of aligning spatial modality selection with the specific structures under investigation.

Practical Considerations-

Instrument access, assay cost, and computational infrastructure also shape platform choice. Imaging assays require specialized hardware and extensive image analysis pipelines, whereas sequencing-based methods leverage standard NGS workflows. Pilot experiments comparing FFPE and frozen sections or cross-validating RNA, protein, and metabolites help define an optimal workflow for a given tissue.

Ultimately, the “right” spatial technology is not defined by any one feature alone but by its fit to the biological question, tissue constraints, and intended application, whether for discovery, biomarker validation, or clinical translation in liver cancer research.

3. Spatial Omics Reveals HCC Biological Insights into HCC Pathogenesis and Therapy

The spatial organization of the liver cancer microenvironment represents a fundamental determinant of tumor behavior, immune evasion, and therapeutic response. Recent spatial omics studies have demonstrated that the HCC microenvironment is not merely a collection of diverse cell types, but a highly organized ecosystem in which spatial positioning shapes cellular function, intercellular communication, and clinical outcomes. In this section, we summarize the current understanding of HCC biology that has emerged from spatial profiling studies, organized across five major themes: spatial dynamics at the invasive tumor margin; spatially defined macrophage niches and immune evasion; fibroblast-driven stromal remodeling and architecture; spatial biomarkers associated with outcomes and therapeutic response; and spatially resolved therapeutic targets. Macrophages and fibroblasts are discussed in dedicated sections because recent studies in HCC have most consistently implicated these populations as spatial regulators of HCC biology, while other immune populations are discussed where spatially resolved and mechanistically informative data are available. Together, these themes highlight how spatial omics studies have informed HCC biology, disease progression, and treatment response (Figure 2).

Figure 2. Spatial architecture shaping the hepatocellular carcinoma (HCC) microenvironment.

The top panel conceptually illustrates how spatial omics technologies can uncover key biological insights in HCC, including patterns of cancer cell heterogeneity, immune cell infiltration, intercellular communication, and the emergence of distinct cellular neighborhoods within the tumor ecosystem.

The bottom panel highlights several key findings from recent spatial studies that have mapped the molecular and cellular circuits driving HCC progression. Spatial profiling has identified critical molecular programs such as SPP1–CD44 axis in ECM remodeling, and macrophage-derived TGFB1-mediated cancer stemness, CAF-mediated PD1 signaling driving T-cell exhaustion, and macrophage-derived VEGFA-induced angiogenesis, all of which contribute to tumor growth and immune evasion. At the tumor invasive edge, a tumor–immune barrier forms where SPP1⁺ macrophages and cancer-associated fibroblasts (CAFs) establish a fibrotic and immunosuppressive niche that excludes CD8⁺ T cells, while CXCL6-secreting tumor cells and SAA⁺ hepatocytes reinforce local inflammation and invasion.

3. a. Spatial dynamics at the invasive margin

The interface between tumor and surrounding liver tissue has emerged as one of the most biologically important spatial domains in HCC. The invasive margin is now recognized as the most active region of the HCC microenvironment and is defined by its pronounced transcriptional, cellular, and architectural complexity, mediating both tumor progression and therapeutic resistance. Spatially resolved analysis often describes this critical boundary as a 500 μm-wide zone centered on the tumor border as being associated with strong immunosuppression, metabolic reprogramming, and damaged hepatocytes (29,30). Using Stero-seq with high resolution (220 nm/spot) to study the invasive edge of primary human HCCs, Wu et al found that hepatocytes immediately at the tumor–normal interface were “damaged” and expressed high levels of serum Amyloid A1/2 (SAA1, SAA2) (29). Using spatial analysis, investigators mapped a paracrine interaction at the invasive edge, showing that CXCL6-secreting tumor cells activate JAK–STAT3 signaling in adjacent hepatocytes, leading to SAA overexpression, a relationship that would be difficult to detect without spatial context (29). Another study analyzing this zone using spatial mapping identified a distinctive niche known as the tumor–immune barrier (TIB) which was noted to contribute to non-responsiveness to ICIs (30). This TIB was found to be composed primarily of interacting hypoxia-induced SPP1⁺ macrophages and ECM-producing cancer-associated fibroblasts (CAFs) whose coordinated activity establishes a physical and chemical shield against cytotoxic T lymphocytes (30). Together, these studies highlight the invasive margin as a spatially dynamic zone where tumor–stromal and tumor–hepatocyte interactions shape immune exclusion and local inflammation.

Beyond these cellular and molecular interactions, the structural architecture of the invasive front itself also contributes to intratumoral heterogeneity. Using high-resolution spatial mapping, Wu et al demonstrated that the presence or completeness of the fibrous capsule strongly influences the spatial continuity and transcriptomic diversity of surrounding stromal and immune populations (31). Tumors with an intact fibrous capsule showed greater spatial continuity and lower transcriptomic diversity than those with discontinuous or absent capsules, highlighting how physical boundaries regulate the organization of stromal and immune elements (31). Subsequent studies have highlighted the prognostic importance of immune cell positioning within the tumor edge. For instance, a higher proportion of CD3-CD57+ NK cells at the invasive front correlated with lower recurrence of HCC, and a prognostic model (the TIMES score) built using both biomarker expression (SPON2, ZFP36L2, ZFP36, VIM, HLA-DRB1) and their spatial localization within tumor regions greatly improved predictive accuracy (32,33). In another study, Yang et al identified a novel population of PDL1+CD103+ dendritic cells which interact with regulatory and exhausted T cells through multiple ligand–receptor pathways, reinforcing local immunosuppression associated with early postoperative recurrence (34). Extending this line of evidence, multiple integrative spatial proteomic and transcriptomic analyses have identified that immunosuppressive macrophage subsets enriched in peri-tumoral regions orchestrate angiogenic, fibrotic and immune-evasive remodeling (32,33,35). Together, these spatially resolved findings underscore that microanatomical context dictates immune evasion and therapeutic response in HCC.

While the studies above focus on spatial programs within established tumors and their microenvironment, spatial omics analyses are also beginning to illuminate how normal liver architecture shapes the earliest steps of tumor initiation. Beyond invasive and stromal programs, recent work implicates intrinsic liver zonation as a determinant of HCC origin. Using targeted in situ spatial transcriptomics with the 10x Genomics Xenium platform, Guo et al. demonstrated that premalignant clones and early tumor lesions preferentially arise within specific metabolic zones of the liver, indicating that HCC initiation is influenced by zonation-defined hepatocyte states rather than occurring uniformly across the liver (36). Together, these findings extend the impact of spatial organization in HCC from tumor progression to tumor origin, highlighting zonation as a spatial axis that may shape early evolutionary trajectories and future therapeutic vulnerabilities.

3. b. Macrophage niches and immune evasion

Macrophages are among the most dynamic and functionally diverse immune populations in HCC, and spatial omics analyses now delineate how their positioning within perivascular, fibrotic, and invasive niches governs tumor behavior. In our lab, multimodal spatial mapping of treatment-resistant residual HCC revealed that M2-like PD-L1⁺ macrophages promote tumor stemness and CD8⁺ T-cell exhaustion within spatially constrained neighborhoods through the secretion of TGF-β, thereby sustaining immune suppression and driving recurrence (37). Keeping in line with this, Fan et al have identified HCC regions where SPP1⁺ macrophages cluster near cancer stem-like cells, creating hypoxic niches that promote extracellular matrix remodeling, angiogenesis, and immune exclusion through SPP1–CD44/ITG signaling (38). Similarly, in MASH-HCC, imaging mass cytometry revealed that PDL1+/ICOS+ macrophages interact with exhausted PD1+CD8T cells and myeloid-derived suppressor cells, forming spatially organized immunosuppressive networks distinct from viral HCC (39). Extending these observations to the clinical scale, an integrative analysis of single-cell and spatial transcriptomics across human HCC samples demonstrated that presence of dense macrophage clusters correlated with lipid metabolic reprogramming, cytotoxic T lymphocyte (CTL) exhaustion, and poor outcomes, yet paradoxically predicted favorable response to anti–PD-1 therapy (40). Further highlighting their structural role in shaping the tumor microenvironment, another spatial proteomic analysis identified macrophage-coated tumor clusters (MCTCs) that physically encapsulated tumor cells, enhanced invasiveness and local immune evasion (41). Lastly, connecting macrophage localization to vascular remodeling, high-plex imaging combined with spatial transcriptomics revealed that VEGF-expressing perivascular macrophages formed immune cell neighborhoods around blood vessels, and also interacted with endothelial cells to promote the hypervascularization characteristic of HCC (42). Collectively, these studies position macrophages as spatially anchoring cells that sculpt specific niches, regulate hypoxia-driven signaling, and determine immune evasion within the HCC microenvironment.

3. c. Fibroblast remodeling and stromal architecture

Another key cellular player in HCC is the fibroblast, particularly cancer-associated fibroblasts (CAFs), which act as major architects of both structural and functional heterogeneity within the tumor microenvironment. Recent spatial multi-omics studies have delineated spatially distinct CAF subtypes that organize the stromal landscape of HCC. Cheng et al identified at least two dominant CAF subtypes- CAF-FAP and CAF-C7 that define distinct stromal archetypes within HCC tissue (43). CAF-FAP fibroblasts localize primarily within the tumor core, where they form inflammatory hubs enriched for PD1+ exhausted CD8T cells and secrete pro-tumor cytokines such as CCL2 and CXCL12 (43). In contrast, CAF-C7 fibroblasts accumulate along the tumor edge, activating wound-healing and extracellular matrix (ECM) remodeling programs, and co-localize with SPP1+ macrophage clusters to promote fibrosis and invasion through TGFβ signaling (43). Beyond their structural roles, spatial analyses have uncovered metabolically specialized functions of CAFs that are tightly linked to their spatial localization within the tumor. Integrative spatial metabolomics showed that CAFs function as metabolic hubs, secreting lactic acid and long-chain polyunsaturated fatty acids that drive macrophage polarization toward immunosuppressive M2 states and enhance tumor invasiveness (44). Additional multimodal spatial analyses have also revealed novel CAF phenotypes beyond the canonical subsets. For instance, F5-CAFs were identified as a subtype which form peritumoral nests supporting cancer stem cell maintenance, revealing stromal niches that sustain tumor self-renewal (45). CAFs also exert potent immunomodulatory influence through cross-talk with immune cells. Certain CAF subsets secrete factors that recruit or potentiate other immunosuppressive cells, for instance, a newly identified lipid‐metabolic CAF subset, CD36+ CAFs, was shown to attract CD33+ myeloid-derived suppressor cells via macrophage migration inhibitory factor (MIF) signaling (46). Furthermore, CAFs serve as key structural enablers of immune evasion niches, as demonstrated by their contribution to the tumor-immune barrier formation which actively limits immune infiltration of lymphocytes into the tumor core, a feature particularly observed in patients unresponsive to ICIs (30). Collectively, these high-resolution spatial multiomics studies establish fibroblasts as spatially and functionally diverse regulators that shape stromal topology, metabolic gradients, and immune evasion in HCC.

3. d. Spatial biomarkers of clinical outcomes and response to therapies

Extending mechanistic insights into clinical relevance, spatial omics studies have revealed that prognostic and predictive biomarkers in HCC can be defined by spatial analyses of the tumor microenvironment. Several key spatial transcriptomic and proteomic studies now define molecular signatures that predict recurrence, survival, and therapeutic response with potentially greater precision than bulk profiling. For example, the spatial study of the invasive margin discussed earlier revealed that hepatocytes expressing SAA at the tumor invasive zone marked a high-risk population, whose presence was strongly associated with early recurrence and reduced survival (29). In another study, Jia et al (2025) demonstrated that the density and spatial localization of CD3-CD57+ NK cells at the tumor invasive zone predicted improved recurrence-free survival (32). Building on this, they derived the TIMES (Tumor Immune Microenvironment Spatial) score, integrating five spatially weighted genes, SPON2, ZFP36L2, ZFP36, VIM, and HLA-DRB1, whose combined spatial expression at the invasive front accurately stratified recurrence risk (32). Importantly, this predictive power was lost when spatial information was disregarded, underscoring that spatial distribution, not abundance alone, defines clinical relevance (32). Similarly, in our study on therapy-resistant HCC, we found that spatial interactions between M2-like macrophage and stem-like tumor cells predicted recurrence (37). Extending this principle to the broader immune landscape, Maestri et al. (2024) used spatial single-cell imaging across two large cohorts to define prognostically distinct immune classes (35). Using spatial single-cell imaging (CODEX), researchers identified an immune class (IC2) associated with the best prognosis, characterized by frequent CD8T–tumor and CD4T–CD8T-cell interactions and a high cytotoxicity gene signature (35). Crucially, incorporating spatial interaction data outperformed cell composition alone in survival prediction.

While these studies define how spatial organization predicts recurrence and survival, similar principles are now being leveraged to predict treatment response. For example, the tumor-immune barrier (TIB) structure composed of SPP1+ macrophage and CAFs was found to be associated with non-response to anti-PD1 treatment (30). Small neoadjuvant “window-of-opportunity” studies utilizing high-plex spatial omics have begun to reveal how the tumor–immune landscape distinguishes responders from non-responders to systemic therapy in HCC. In a phase I trial of cabozantinib plus nivolumab, Mi et al. (2022) used imaging mass cytometry (IMC) to show that responding tumors exhibited a distinctive immune spatial architecture characterized by the proximity of B cells and CD8T cells to PDL1+ macrophages (47). Conversely, non-responders displayed immune-excluded architectures dominated by tumor cells and enriched for arginase1+ macrophages in close proximity to quiescent CD8T cells (47). Follow-up spatial transcriptomic (ST) profiling of the same cohort further identified that tumor cells in responders exhibited activation of the PAX5 transcription network, linked to B cell maturation (48). In contrast, non-responders maintained cancer cell-intrinsic proliferative and metabolic pathways without activation of immune-related pathways (48). Together, these early clinical studies highlight how spatial context, including cellular proximity and immune barrier formation, within the TME can stratify therapeutic response and guide the development of spatially informed predictive biomarkers in HCC.

3. e. Spatial analysis reveals therapeutic targets

Apart from biomarkers, spatial omics can also identify new therapeutic targets in HCC. In our own work on post-TACE residual HCC, we identified spatial ligand–receptor TGF-β and PD-L1 signaling as dominant suppressive pathways within the macrophage–tumor neighborhoods, and combined blockade of both in preclinical models disrupted the niche, restored T-cell activity, and prevented recurrence, highlighting this macrophage–tumor axis as a spatially defined therapeutic target in HCC (37). Other spatial studies have converged on additional pathways that sustain immune evasion and stromal remodeling. The SPP1–CD44 signaling axis, mediating communication between macrophages and CAFs, has emerged as a particularly compelling target (49). While not yet in clinical use, this pathway is pharmacologically targetable through developing anti-osteopontin and anti-CD44 agents. Similarly, spatial profiling of tumor–vascular interfaces revealed that galectin-9 (LGALS9) from tumor cells interacts with TIM3 on endothelial cells, identifying an immune–vascular checkpoint with therapeutic potential, as TIM3 and galectin inhibitors are already in oncology trials (50). Spatial metabolomic analyses have further expanded the scope of intervention, revealing immune-suppressive metabolic niches such as those dominated by adenosine signaling, where adenosine receptor antagonists may function as metabolic checkpoint inhibitors to locally reverse immune suppression (51). Therapy-based spatial studies are also uncovering actionable mechanisms of resistance. In neoadjuvant cabozantinib plus nivolumab study, spatial transcriptomics of nonresponders revealed immune–CAF interactions involving FOS and JUN activation, suggesting that targeting the FOS–JUN pathway could remodel the ECM and reprogram immune function (48). Their ligand–receptor mapping further implicated PAX5-driven B-cell signaling and CXCL6–JAK/STAT3 loops as spatially organized axes that may enhance response to immunotherapy (48). In the same clinical setting, imaging mass cytometry identified CD163+/Arg1+ macrophages whose spatial proximity to T cells marked non-response, underscoring their potential as a distinct macrophage population for therapeutic targeting (47). Collectively, these studies demonstrate that spatial omics not only pinpoint molecular targets but also define the microanatomical niches where therapeutic pressure must be applied.

4. Toward Translation: Clinical Implications and Future Directions

Spatial omics in HCC is entering a translational era in which tissue architecture and molecular mapping are increasingly informing clinical decision-making. As these approaches move from discovery toward clinical application, their relevance can be considered along the continuum of patient care. In this section, we discuss how spatial biology may inform prognostic risk assessment at diagnosis and after resection, guide treatment selection and prediction of therapeutic response, and support dynamic monitoring of tumor evolution during therapy. We then address key practical and technical challenges that must be overcome for clinical implementation, including issues related to assay standardization, tissue availability, and computational infrastructure. Finally, we highlight the emerging role of artificial intelligence in integrating spatial data with histopathology and genomics to enable scalable, clinically actionable decision support. Together, these themes outline a framework for potentially translating spatial omics from mechanistic insight into clinical practice in HCC (Figure 3).

Figure 3. Translating spatial analysis of hepatocellular carcinoma (HCC) into clinical applications.

Spatial multi-omic technologies integrate histology, genomics, transcriptomics, and proteomics to generate high-dimensional tissue maps that link molecular states to spatial context. These approaches have enabled four major clinical applications in HCC:

(1) Predict clinical outcomes- Spatially resolved biomarkers integrating tumor, immune, and stromal architecture can stratify patients by recurrence risk and overall survival.

(2) Personalized medicine- Spatial multi-omic profiling differentiates immune, stromal, and molecular tumor states to guide the selection of systemic, targeted, or combination therapies.

(3) Response prediction- Spatial signatures of immune proximity, ligand–receptor pairing, and checkpoint expression can be translated into predictive classifiers to identify likely responders before therapy initiation.

(4) Identify new therapeutic axes- Mapping cellular interactions at single-cell resolution uncovers novel pathways within the tumor microenvironment that can be targeted to modulate disease progression and overcome treatment resistance.

Together, these advances illustrate how spatial biology bridges tissue architecture with actionable clinical insight in liver cancer.

4.a. Prognostic Biomarkers for recurrence in early HCC

As discussed earlier, spatial features such as the proximity of CD8T cells to tumor cells, the presence of tumor–immune barriers, and the abundance of angiogenic CAF clusters carry strong prognostic value. This has fueled growing interest in developing clinically deployable spatial pathology assays. For example, an immunohistochemistry or multiplex immunofluorescence panel encompassing 6–8 key markers could be paired with quantitative image analysis software to measure parameters such as the mean distance of CD8T cells from the nearest tumor cell or the density of CD163+ macrophages or SAA+ hepatocytes at the tumor margin. As these approaches are validated in larger, multi-institutional cohorts, they could enable risk stratification after resection or prior to systemic therapy. For instance, a patient with a high spatial immune exclusion score might be predicted to experience early recurrence and could be prioritized for adjuvant therapy or intensified surveillance.

4. b. Predictive Biomarkers for Response to Systemic Therapy

Similarly, in the context of systemic therapy for advanced HCC, spatial biomarkers could help predict which patients are most likely to respond. For instance, ICIs may only be effective if CD8T cells can physically access tumor cells. Patients whose biopsies reveal pre-existing CD8T-cell infiltration or lack a well-formed immune barrier might therefore be the best candidates for ICI therapy. Conversely, tumors characterized by prominent CAF/macrophage barriers or sparse immune infiltration (“immune-cold” tumors) may require combination regimens such as ICI plus anti-VEGF therapy. Post hoc spatial analyses of clinical trial samples are beginning to test these hypotheses (47,48). Emerging data from other cancer types further support this concept, suggesting that spatial immune architectures can robustly predict response to immunotherapy (52,53). As spatial understanding of the HCC microenvironment deepens, treatment selection may increasingly rely on tumor ecosystem subtype rather than stage alone.

Spatial omics can potentially be used not only to identify optimal therapies but also to monitor how the tumor microenvironment evolves during treatment. For example, in immunotherapy trials, repeat biopsies are being analyzed with spatial tools to assess whether treatment dismantles immune barriers and promotes CD8T-cell infiltration changes that could inform early therapy adaptation in an adaptive-trial framework (54,55). Similarly, in neoadjuvant studies, spatial analysis of resected tumors is being tested as a surrogate endpoint to evaluate whether therapy successfully remodels the immune landscape, for instance, by converting an immune-cold tumor into a hot one (56-58). Ultimately, integrating spatial biomarkers into clinical workflows could shift HCC management from empiric treatment selection toward dynamically guided, microenvironment-informed precision therapy. A combination of spatial transcriptomics, proteomics, and even spatial metabolomics on a patient’s tumor could provide a “tissue GPS map” that guides precision medicine.

4. c. Challenges to Clinical Implementation

We acknowledge that several challenges must be addressed before spatial omics can be routinely implemented in clinical practice. First, high-dimensional spatial assays remain costly and technically demanding, requiring specialized instrumentation and trained personnel for both acquisition and analysis. Translating a 40–50 marker research panel to a hospital pathology lab will require major simplification and standardization. However, initiatives such as the Human Tumor Atlas Network (HTAN) and NIH Bridge2AI are already developing minimal predictive panels (4–8 markers) and cross-platform calibration standards, while commercial platforms like Akoya’s PhenoImager HT and Leica BOND RX are working toward CLIA-compatible workflows that could make multiplex spatial assays clinically viable within the decade. Second, the computational burden and lack of standardized analytics pipelines remain barriers to broader use. Spatial datasets require robust image segmentation, single-cell identification, and spatial neighborhood analysis, all of which currently depend on expert bioinformaticians. This challenge can be addressed through automation and AI integration, as open-source tools, and digital pathology vendors begin embedding spatial analysis within FDA-cleared clinical platforms. Cross-validation through public reference datasets and federated learning approaches may further enhance reproducibility and confidence in spatial biomarker quantification.

Third challenge is tissue availability, which represents a fundamental constraint, particularly in HCC. Liver tumors are rarely biopsied unless clinically indicated, and when obtained, a single core may not capture the biological diversity of multifocal or heterogeneous lesions. This issue could be mitigated by embedding small “spatial biopsy cores” within existing interventional procedures and by developing linked biobanks that integrate resection, explant, and autopsy samples with comprehensive clinical annotation. Spatial profiling of archival FFPE tissue, which is increasingly feasible with newer chemistries, also expands the pool of accessible material. Last, there remains a need for clinical integration and interpretive consensus. Even as predictive spatial biomarkers emerge, their translation into actionable clinical metrics will require agreement on how to score, report, and act on them. Establishing standardized spatial scores, analogous to PDL1 IHC, HER2, or OncotypeDX, validated across multicenter cohorts, will be crucial. Collaborative frameworks and multi-institutional consortia are beginning to harmonize spatial data collection and define the regulatory pathways for spatial diagnostics.

Together, these challenges represent not roadblocks but developmental milestones in the evolution of spatial biology from discovery to deployment. As costs fall, computational tools mature, and regulatory frameworks adapt, spatial omics is poised to become a cornerstone of precision oncology providing pathologists and clinicians with a high-dimensional, spatially resolved view of tumor biology that complements genomics and histopathology. As algorithms learn to integrate spatial, genomic, and histologic data in real time, clinicians may soon have access to AI-augmented pathology reports that not only describe what is present in the tumor, but also predict how it behaves and responds to therapy, bringing the vision of truly spatially informed precision oncology within reach.

4. d. Promise of artificial intelligence (AI)-driven spatial biology integration to the clinic

Advances in AI, machine learning (ML), and spatial omics methods offer an exciting opportunity to infer and impute biological signals by training models that align molecular profiles with spatially matched histopathologic or radiologic features. By integrating these complementary data modalities, AI can bridge molecular and morphologic information. Using AI/ML for integrative modeling of spatial omics can enable the systematic dissection of the tumor microenvironment within their native spatial context. Such models could help predict therapy response, recurrence, and metastasis, while also guiding the development of new therapeutic strategies (59). For example, Zeng et al developed an AI-based pathology model capable of estimating the atezolizumab–bevacizumab response signature (ABRS) directly from HCC histology slides using a multiple-instance learning pipeline (60). The predicted ABRS-P scores correlated strongly with spatial transcriptomic features and progression-free survival, demonstrating how AI can impute complex molecular signatures from morphology to guide therapy selection (60). Applying similar principles, Jia et al applied ML to integrate multiplex imaging, spatial transcriptomics, and proteomics to generate the TIMES score, a spatially informed five-gene signature that outperformed TNM and BCLC staging in predicting post-resection recurrence (32). These studies illustrate how AI can bridge histology and molecular biology by spatially imputing gene-level information from standard tissue images. Extending this approach to therapeutic modeling, Zhang et al combined ML and spatial multi-omic data to construct a spatial Quantitative Systems Pharmacology (spQSP) model of HCC, enabling virtual clinical trials that simulate response to anti-PD1 plus cabozantinib therapy (61). Collectively, these studies highlight the growing role of AI-driven modeling in HCC, marking a new era of AI-driven, spatially omics.

5. Conclusion

In closing, spatial omics has provided an unprecedented window into the liver cancer microenvironment. In HCC, it has underscored that where matters as much as what; that it is not merely the presence of particular cells or molecules, but their spatial relationships that dictate tumor behavior and therapeutic response. By mapping the tumor as a dynamic ecosystem of interacting cellular communities, spatial technologies are revealing actionable insights regarding which interactions to disrupt, which cellular players to target, and how to deliver therapies more effectively. As these technologies become increasingly accessible, they are poised to reshape clinical trial design and therapeutic decision-making in HCC and beyond. The coming years may see the first trials in HCC that stratify patients by spatial phenotypes or incorporate spatial endpoints to measure therapeutic efficacy. Ultimately, the integration of spatial omics with AI-driven analytics and digital pathology holds the promise of truly personalized liver cancer care.