LILRB4 expression is associated with macrophage enrichment and differential survival in hepatocellular carcinoma

Abstract

Background and aims

The clinical and immune-cell context of leukocyte immunoglobulin-like receptor B4 (LILRB4) in hepatocellular carcinoma (HCC) remains incompletely characterized. This study aimed to characterize LILRB4 expression across cell types and examine its associations with survival outcomes and macrophage-related immune features in HCC.

Methods

Public liver hepatocellular carcinoma (LIHC) resources were analyzed using survival, transcriptomic, functional enrichment, immune deconvolution, and single-cell analyses. Survival analyses included 373 patients for overall survival and 365 for disease-specific survival. Kaplan–Meier analyses compared two expression groups (high and low), defined using outcome-selected cutpoints obtained from maximally selected rank statistics. Continuous-expression Cox models provided complementary analyses.

Results

Higher LILRB4 expression was associated with shorter overall survival (OS) in an outcome-selected analysis (n = 373; low-expression group, 250; high-expression group, 123; Cox hazard ratio [HR], 1.524; 95% confidence interval [CI], 1.074–2.162; log-rank p = 0.017). Continuous LILRB4 expression was associated with OS in univariable and adjusted Cox models (adjusted HR, 1.246; 95% CI, 1.010–1.537), whereas continuous expression was not significantly associated with disease-specific survival (HR, 1.137; 95% CI, 0.916–1.412). LILRB4 expression correlated with macrophage enrichment (Spearman r_s = 0.755) and M2-like macrophage estimates from quanTIseq (r_s = 0.544) and CIBERSORT-ABS (r_s = 0.753). Single-cell resources localized LILRB4 mainly to monocyte/macrophage populations.

Conclusion

LILRB4 expression is associated with macrophage-enriched immune features and adverse survival patterns in LIHC. These findings identify LILRB4 as a candidate marker of the macrophage-rich HCC microenvironment that warrants further investigation of its prognostic and immunoregulatory relevance.

Graphical abstract

LILRB4 expression is associated with macrophage-rich transcriptomic and cellular features in LIHC. TCGA-based survival analyses associate higher expression with adverse survival patterns, while single-cell resources and tissue immunofluorescence support a macrophage-related context for LILRB4. Together, these findings support further evaluation of LILRB4 as a marker of immune context and prognosis in HCC. LIHC, liver hepatocellular carcinoma; HCC, hepatocellular carcinoma; TCGA, The Cancer Genome Atlas; TISCH2, Tumor Immune Single-cell Hub 2.The schematic summarizes the data sources, principal associations, and study implications.

Highlights

• •LILRB4 expression is associated with macrophage-related immune features in LIHC.
• •Single-cell datasets localize LILRB4 to monocyte and macrophage populations.
• •Higher LILRB4 expression is associated with poorer survival outcomes.
• •Cox analyses show an exploratory association between LILRB4 and overall survival.
• •Tissue immunofluorescence shows LILRB4/CD163 signal overlap.

1. Introduction

Hepatocellular carcinoma (HCC) is the predominant subtype of primary liver cancer and remains a major global health burden.1, 2 Despite advances in surveillance, locoregional therapy, molecularly targeted agents, and immune checkpoint blockade, advanced HCC continues to exhibit heterogeneous and frequently limited responses.³ A central biological barrier is the liver's distinct immune ecology, in which chronic inflammation, immune tolerance, stromal remodeling, myeloid-cell infiltration, and T-cell dysfunction collectively shape tumor evolution and therapeutic resistance.4, 5

Immune checkpoint blockade has reshaped the systemic treatment landscape of HCC; however, durable clinical benefit remains limited to a subset of patients.3, 4 Tumors with insufficient effector T-cell infiltration or dominant myeloid-mediated immunosuppression frequently respond poorly to therapies targeting programmed cell death protein 1/programmed death-ligand 1 (PD-1/PD-L1) or cytotoxic T-lymphocyte-associated protein 4 (CTLA-4).4, 5 This pattern has focused attention on myeloid-cell immune receptors as potential biomarkers and therapeutic targets.6, 7, 8 The leukocyte immunoglobulin-like receptor family includes activating and inhibitory receptors that modulate innate and adaptive immunity through interactions with major histocompatibility complex molecules and other ligands.6, 7, 8

Leukocyte immunoglobulin-like receptor B4 (LILRB4), also known as immunoglobulin-like transcript 3 (ILT3) or CD85k, is a myeloid inhibitory receptor with intracellular immunoreceptor tyrosine-based inhibitory motifs.6, 7, 8 Expressed mainly in monocytes, macrophages, dendritic cells, and other immune-cell populations, LILRB4 can recruit phosphatases and modulate downstream immune signaling.6, 7, 8 LILRB4 has been linked functionally to immune tolerance, T-cell suppression, leukemia infiltration, and immune escape by solid tumors. 8, 9, 10 In solid tumors, inhibitory signaling through LILRB4 can restrain antitumor immunity; the fibronectin–ILT3 interaction has also been described as a stromal checkpoint that suppresses myeloid-cell activity.10, 11 ILT3/ILT4 expression can also be induced under IL-10-rich immunoregulatory conditions, and ILT3-related ligand interactions, including CD166, have been implicated in immune and tumor-associated signaling.12, 13

The potential relevance of LILRB family members in HCC is emerging. Flow-cytometric analyses of samples from patients with HCC showed greater LILRB4 expression in tumor-associated macrophages than in paired tumor-free liver tissues, suggesting a possible role in the HCC immune microenvironment.¹⁴ Bioinformatic studies have likewise associated LILRB family expression with liver cancer prognosis and immune infiltration.15, 16 Our recent glioblastoma study linked LILRB4-positive tumor-associated macrophages (TAMs) to macrophage polarization and tumor progression through a STAT3/IL10-associated axis.¹⁷ Together, these observations suggest that a related myeloid checkpoint program may operate in HCC. Nevertheless, the cell-type specificity, macrophage-polarization context, and immune-microenvironmental significance of LILRB4 in liver hepatocellular carcinoma (LIHC) are insufficiently characterized.

TAMs are abundant components of the HCC microenvironment and have important functional roles. 18, 19, 20, 21, 22 Rather than fitting a simple binary M1/M2 classification, macrophages occupy highly plastic, context-dependent activation states.23, 24 Within cancer inflammation and immune regulation, STAT3 signaling is widely recognized as a central mediator of tumor-associated immune suppression.²⁵ M2-like TAM programs involve markers and mediators such as CD163, MRC1/CD206, MSR1, IL10, TGFB1, and related molecules and are widely associated with immune suppression, tissue remodeling, angiogenesis, invasion, therapeutic resistance, and poor prognosis.20, 21, 22, 26 Whether LILRB4 specifically marks an M2-like macrophage-enriched suppressive microenvironment in LIHC remains to be systematically evaluated.

We integrated public survival data, transcriptomic correlations, functional annotations, immune deconvolution, single-cell evidence, and descriptive tissue immunofluorescence to define the clinical and immune-cell context of LILRB4 in LIHC. We examined associations with macrophage-related transcriptional features and evaluated how strongly the available survival analyses support prognostic interpretation. The study was designed to generate hypotheses for subsequent validation rather than to establish a mechanism or a clinically validated predictor.

2. Materials and methods

2.1. Study design

This retrospective, exploratory study combined public transcriptomic, clinical, immune-infiltration, and single-cell resources. Analyses used web-based bioinformatics workflows and the statistical methods and software packages described below. We selected LILRB4 for focused analysis because of its observed survival associations among the five LILRB family genes and its established myeloid-receptor biology. The public-data observations were complemented by immunofluorescence analysis of archived LIHC tissue.

2.2. Data acquisition and preprocessing

For the survival analyses, we obtained TCGA-LIHC STAR-processed RNA-sequencing data in TPM format and corresponding clinical annotations from the Genomic Data Commons ((https://portal.gdc.cancer.gov/)) and the TCGA Pan-Cancer Clinical Data Resource.²⁷ Expression values were transformed using log₂(TPM + 1). Normal samples and records lacking clinical information were excluded. Of 374 records included in the workflow, 373 had overall survival (OS) information and 365 had disease-specific survival (DSS) information; 130 OS and 79 DSS events were recorded. The complete-case multivariable OS and DSS models each included 259 observations. However, LILRB4 was not included in the final DSS model; therefore, no adjusted DSS effect estimate for LILRB4 was available. Single-cell resources were queried using Tumor Immune Single-cell Hub 2 (TISCH2; http://tisch.comp-genomics.org/),28, 29 and cell-type information was obtained from the Human Protein Atlas (HPA; https://www.proteinatlas.org/).³⁰ The Supplementary Materials provide detailed survival methods, coefficient tables, model diagnostics, and analysis specifications.

2.3. Survival analysis of LILRB family members

Kaplan–Meier analyses compared two groups, designated high expression and low expression, for each LILRB family gene and survival endpoint. Expression cutpoints were determined using survminer::surv_cutpoint, which implements maximally selected rank statistics,³¹ and survival curves were compared using log-rank tests. For the LILRB4 OS comparison, 373 patients were divided into a low-expression group (n = 250; 76 events) and a high-expression group (n = 123; 54 events). Numerical cutpoints and group sizes for the remaining nine gene–endpoint comparisons were not available in the retained analysis reports; these comparisons are therefore presented descriptively. Continuous-expression Cox analyses estimated hazard ratios (HRs) for each one-unit increase in log₂(TPM + 1). Candidate covariates comprised pathologic T, N, and M stage, tumor status, sex, age, residual tumor, histologic grade, AFP, and LILRB4. A univariable p < 0.10 criterion was used to screen candidate covariates for exploratory multivariable Cox analysis. The final OS model included T stage, M stage, tumor status, and LILRB4, whereas LILRB4 was not included in the final DSS model. Proportional hazards diagnostics and variance inflation factors were examined for the fitted models. Analyses were performed with R version 4.2.1 (R Foundation for Statistical Computing, Vienna, Austria), with the survival (version 3.3.1), rms (version 6.3-0), survminer (version 0.4.9), and ggplot2 (version 3.4.4) packages. Numerical results appear in Table 1, Table 2.

Table 1.

Exploratory grouped survival summaries for LILRB family genes.

Endpoint	Gene	HR	95% CI	Log-rank p
OS	LILRB1	1.455	0.997–2.122	0.037
OS	LILRB2	0.723	0.470–1.114	0.105
OS	LILRB3	1.266	0.866–1.850	0.200
OS	LILRB4	1.521	1.052–2.199	0.017
OS	LILRB5	0.544	0.351–0.844	0.001
DSS	LILRB1	1.366	0.878–2.124	0.171
DSS	LILRB2	0.789	0.469–1.329	0.342
DSS	LILRB3	1.718	1.064–2.775	0.050
DSS	LILRB4	1.810	1.097–2.988	0.046
DSS	LILRB5	0.574	0.334–0.987	0.019

Table 2.

Continuous-expression LILRB4 survival estimates and interpretive limits.

Endpoint/model	n	HR (95% CI)	p	Interpretation
OS univariable	373	1.187 (1.008–1.397)	0.040	Exploratory association
OS adjusted	259	1.246 (1.010–1.537)	0.040	Global PH violation; not independent validation
DSS univariable	365	1.137 (0.916–1.412)	0.244	No clear association
DSS adjusted	259	Not estimated	Not applicable	LILRB4 was not included in the final model

Supplementary Table S1 presents grouped survival summaries, Table S2 provides continuous-expression estimates, Table S3a–h Cox results and diagnostics, and Table S4 analysis specifications.

2.4. Correlation, enrichment, and network analyses

Spearman rank correlations ranked genes associated with LILRB4 expression; positive and negative correlations were displayed separately. The top 30 positive and negative correlates were displayed in expression heatmaps. We analyzed GO biological process, cellular component, and molecular function categories and KEGG pathways with clusterProfiler.³² STRING (https://string-db.org/) provided the protein-association networks and functional annotations.³³ These analyses characterized the immune and myeloid biological context associated with LILRB4 expression.

2.5. Immune microenvironment and macrophage analyses

The immune context of LILRB4 expression was examined with ESTIMATE-derived stromal and immune scores,³⁴ gene-set-based immune enrichment,³⁵ and CIBERSORT-ABS and quanTIseq deconvolution estimates.36, 37 TIMER (http://timer.comp-genomics.org/) -derived infiltration outputs provided the displayed macrophage correlations.³⁸ Spearman coefficients quantified associations involving LILRB4. The marker-gene analysis included CD163, MSR1, IL10, TGFB1, and MRC1. Using GEPIA2021 (http://gepia2021.cancer-pku.cn/), we compared deconvolution-derived M1-and M2-related expression estimates³⁹; these estimates do not directly measure expression within isolated macrophages.

2.6. Single-cell expression analysis

TISCH2 was used to assess LILRB4 expression across the resource's cell-type annotations.28, 29 The six displayed TISCH2 entries comprised LIHC_GSE125449_aPDL1aCTLA4, LIHC_GSE140228_10X, LIHC_GSE140228_Smartseq2, LIHC_GSE146115, LIHC_GSE146409, and LIHC_GSE166635. Because the two GSE140228 technologies do not constitute independent cohorts, the six entries represent five GEO accession families. We did not perform new integration, re-clustering, or differential-expression analysis of the raw single-cell data. The representative GSE140228 Smart-seq2 display includes only annotated immune-cell populations and therefore cannot establish whether LILRB4 is absent from cell types not represented in the display.¹⁸ The resource views provided qualitative evidence of cell-type context rather than independent validation of survival associations.

2.7. Tissue immunofluorescence

Archived specimens from five patients with pathologically confirmed HCC were randomly selected from the archives of the Department of Pathology, Chinese PLA General Hospital, Beijing, China. No additional inclusion or exclusion criteria were applied. Sections containing HCC and sections of adjacent liver tissue underwent immunofluorescence staining. The specimens were collected between August 2015 and March 2016. The patients comprised four men and one woman, with a median age of 54 years (range, 37–71 years). According to the available pathology records, two tumors were poorly differentiated and two were moderately differentiated; differentiation was not recorded for one patient. Other clinical characteristics were unavailable in the retained records. For each of the five patients, three spaced sections from HCC tissue and three from adjacent liver tissue were examined. Five non-overlapping fields per section were selected by systematic sampling following a random start, yielding 15 fields per tissue type and 30 fields per patient. The tissue findings were presented descriptively, without patient-level inferential testing. Formalin-fixed, paraffin-embedded sections (4 μm) were blocked with normal goat serum (Sigma-Aldrich, St. Louis, MO, USA; cat. no. G9023) containing 0.3% Triton X-100 (Solarbio, Beijing, China; cat. no. T8200) at room temperature. Primary antibodies were incubated overnight at 4 °C, followed by three washes with phosphate-buffered saline (PBS, pH 7.4; Gibco, Grand Island, NY, USA; cat. no. 10010023). CD163 was detected with a directly conjugated antibody (Proteintech Group, Inc., Rosemont, IL, USA; cat. no. CL594-16646; 1:200). LILRB4 was detected using a rabbit polyclonal antibody (Abcam, Cambridge, UK; cat. no. ab229747; 1:200) and FITC-conjugated goat anti-rabbit IgG (H + L) (Proteintech Group, Inc., Rosemont, IL, USA; cat. no. SA00003-2; 1:200), with secondary antibody incubation for 1 h at room temperature. Nuclei were stained with DAPI (Solarbio, Beijing, China; cat. no. C0060). Images were acquired using a FLUOVIEW FV3000 confocal microscope (Olympus, Tokyo, Japan). Individual fluorescence channels and merged images were assessed in ImageJ (version 1.54h; National Institutes of Health, Bethesda, MD, USA) to characterize the distribution and spatial overlap of LILRB4 and CD163. Pixel-level overlap was displayed as grayscale channels and a colocalized-pixel display.

2.8. Statistical analysis

Each patient record was the analysis unit for survival analyses. Kaplan–Meier curves were compared using log-rank tests, with expression groups defined by maximally selected rank statistics as described in Section 2.3. Univariable and multivariable Cox models reported HRs with 95% confidence intervals (CIs). Model diagnostics included term-specific and global proportional-hazards tests and variance inflation factors (VIFs). Source-reported model-level likelihood-ratio, Wald, and score statistics and C-indices are provided in Supplementary Table S3g. Correlation analyses used Spearman rank correlation, denoted by r_s throughout. p values are reported as labeled in the source: nominal for survival and correlation analyses, and adjusted, where indicated, for enrichment. For panels presented descriptively, numerical p values are reported as provided by the source without significance claims; test and adjustment settings that were not reported are not inferred. Records lacking survival time or event information were excluded from the corresponding endpoint analysis, and multivariable models used complete cases. Tissue patient characteristics were summarized as counts and median age with range. Immunofluorescence findings were descriptive, and no patient-level hypothesis test was performed. Supplementary Tables S1–S4 provide the available sample sizes, software versions, model specifications, effect estimates, and diagnostic results.

3. Results

3.1. Exploratory survival associations of LILRB4 in LIHC

In outcome-selected Kaplan–Meier comparisons, higher LILRB4 expression was associated with shorter OS and DSS (Fig. 1; Table 1). The verifiable LILRB4 OS report included 373 records, with 250 in the low-expression group (76 events) and 123 in the high-expression group (54 events). The log-rank p value was 0.0174. The corresponding univariable Cox HR for high versus low expression was 1.524 (95% CI, 1.074–2.162; Cox p = 0.018). A separate platform log-rank row reported HR 1.521 (95% CI, 1.052–2.199), which should not be conflated with the Cox estimate. The supplied grouped DSS summary reported an HR of 1.810 (95% CI, 1.097–2.988) with a log-rank p = 0.046. Because that summary did not identify a separate Cox estimate or its group counts, we report it as a platform summary and interpret it cautiously. Table 1 presents associations for the other LILRB genes; these analyses did not adjust for outcome-based cutoff selection or family-gene screening.

Fig. 1.

Exploratory survival comparisons for LILRB family genes in LIHC. (A) OS; (B) DSS. The curves compare outcome-selected high- and low-expression groups; the p values are nominal log-rank p values, without correction for cutoff selection or testing across five genes and two outcomes. For the LILRB4 OS analysis, n = 373 (low, 250; high, 123), with 130 events (low, 76; high, 54). Table 1 reproduces the effect estimates and confidence intervals from the original family analyses. Because the source summaries omit both numerical cutpoints and group totals for the other panels, the caption makes no cross-panel assumption about a shared sample size or clinically applicable threshold. The remaining nine comparisons are presented descriptively; their source-reported p values do not support significance claims. OS, overall survival; DSS, disease-specific survival; LILRB, leukocyte immunoglobulin-like receptor subfamily B; LIHC, liver hepatocellular carcinoma.

When LILRB4 was modeled continuously, the univariable OS HR was 1.187 (95% CI, 1.008–1.397; n = 373; p = 0.040), whereas the DSS HR was 1.137 (95% CI, 0.916–1.412; n = 365; p = 0.244). The selected multivariable OS model yielded an HR of 1.246 (95% CI, 1.010–1.537; n = 259; p = 0.040) after adjusting for T stage, M stage, and tumor status. However, its global PH test was significant (p = 5.02 × 10⁻⁵), and the timing of tumor status assessment relative to baseline had not been established. Accordingly, the analysis cannot demonstrate prognostic information independent of the other modeled factors. Because the DSS model excluded LILRB4 and yielded an unstable estimate for tumor status, it provides no adjusted DSS effect estimate for LILRB4 (Table 2; Supplementary Table S2 and S3a–h). Additional proportional-hazards diagnostics, model-level test statistics and C-indices, and VIF results are presented in Supplementary Tables S3e, S3g, and S3h, respectively.

The platform-summary values were copied as supplied; the source did not identify them as Cox-model estimates. The supplementary OS report separately provides the LILRB4 Cox HR as 1.524 (95% CI, 1.074–2.162; Cox p = 0.018). The p column in Table 1 contains log-rank tests, which need not correspond numerically to a Wald test of the displayed HR. Outcome data determined the cutpoints, and the analysis did not report adjustment for multiple testing. Only the LILRB4 OS comparison had verifiable group totals (low/high, 250/123). Reference-category direction and group totals remain unspecified for all other comparisons; accordingly, they are presented descriptively without treating the supplied p values as statistically significant findings. HR, hazard ratio; CI, confidence interval; OS, overall survival; DSS, disease-specific survival.

For continuous expression, each HR quantifies the effect of a one-unit log₂(TPM + 1) increment. The OS-adjusted model included pathologic T stage, M stage, tumor status, and LILRB4, as selected by the platform workflow. The final DSS model included pathologic T stage, M stage, and tumor status in 259 observations, but excluded LILRB4. Thus, no adjusted DSS HR or corresponding p value for LILRB4 was estimated. The global PH p value for the OS model was 5.02 × 10⁻⁵. The report did not establish when tumor status was assessed. The full endpoint datasets include 130 OS and 79 DSS events; these totals cannot be transferred to the smaller multivariable samples because their numbers of events were unavailable. Complete clinical-variable results and diagnostics are in Supplementary Table S2 and S3a–h. PH, proportional hazards; TPM, transcripts per million.

3.2. Genes correlated with LILRB4 and functional annotation

LILRB4 expression was positively correlated with numerous genes associated with immune and myeloid biology (Fig. 2A). The reported Spearman coefficients included CD86 (r_s = 0.923), CYBB (r_s = 0.917), HCK (r_s = 0.911), HAVCR2 (r_s = 0.909), ITGB2 (r_s = 0.909), C3AR1 (r_s = 0.905), NCKAP1L (r_s = 0.904), SPI1 (r_s = 0.903), PILRA (r_s = 0.899), CD84 (r_s = 0.895), C1QC (r_s = 0.894), SIGLEC9 (r_s = 0.893), FCER1G (r_s = 0.892), LAPTM5 (r_s = 0.890), TYROBP (r_s = 0.886), and MSR1 (r_s = 0.864). Negative correlates are displayed separately (Fig. 2B). Thus, these associations may reflect differences in immune-cell composition and cannot distinguish cell abundance from regulation within individual cells.

Fig. 2.

Transcriptional correlates of LILRB4 and their functional annotations in LIHC. (A, B) The 30 highest-ranking positive and negative LILRB4 correlates, respectively, according to Spearman coefficients. (C–E) GO biological process, cellular component, and molecular function enrichment. (F) KEGG pathway enrichment. (G) STRING protein-association network. (H) Functional annotations of the displayed network. Enrichment color scales indicate platform-reported adjusted p values, where labeled. This analysis is descriptive; the source information is insufficient to reconstruct the input features, reference universe, number of bulk samples, or multiple-testing procedure. The displayed correlation coefficients and source-reported adjusted-p value/FDR color scales are presented descriptively; no significance claim is made from this figure. The 30 genes in A/B are feature counts rather than patient counts. GO, Gene Ontology; KEGG, Kyoto Encyclopedia of Genes and Genomes; BP, biological process; CC, cellular component; MF, molecular function; PPI, protein-protein interaction; MHC, major histocompatibility complex; FDR, false discovery rate; LIHC, liver hepatocellular carcinoma.

Enrichment results included terms related to leukocyte-mediated immunity, regulation of immune effector processes, immune-cell activation, and myeloid-cell activation (Fig. 2C–E). The secretory granule membrane appeared among the cellular-component annotations. The KEGG results included complement and coagulation cascades, natural killer cell-mediated cytotoxicity, osteoclast differentiation, Staphylococcus aureus infection, and Fc epsilon RI signaling (Fig. 2F). These annotations are consistent with immune-cell biology, but do not demonstrate the activation of individual pathways.³²

The STRING-derived network comprised HLA-F, HLA-E, HLA-G, ALCAM, APOE, PTPN6, SYK, CD300C, and CD86 (Fig. 2G). Network annotations encompassed major histocompatibility complex (MHC) class I/Ib protein complexes, antigen processing and presentation, inhibitory cell-surface receptor signaling, negative regulation of T-cell cytokine production and T-cell receptor signaling, and a CTLA4 blockade-related signature (Fig. 2H). These annotations provide context based on existing knowledge but do not demonstrate direct physical interactions, pathway activity, or response to CTLA4 blockade in the analyzed LIHC samples.³³

3.3. LILRB4 expression correlates with immune-cell enrichment

We next assessed the association between LILRB4 expression and tumor immune infiltration. LILRB4-high LIHC tumors had higher ESTIMATE, immune, and stromal scores than LILRB4-low tumors, consistent with higher LILRB4 expression being associated with an immune- and stromal-cell-enriched tumor microenvironment (Fig. 3A). Among immune-cell subsets, macrophages had the strongest positive correlation with LILRB4 expression (r_s = 0.755), followed by Th1 cells (r_s = 0.631), immature dendritic cells (r_s = 0.614), T cells (r_s = 0.598), activated dendritic cells (r_s = 0.564), T helper cells (r_s = 0.468), follicular helper T cells (r_s = 0.449), effector memory T cells (r_s = 0.448), Th2 cells (r_s = 0.446), B cells (r_s = 0.432), natural killer (NK) CD56 dim cells (r_s = 0.387), cytotoxic cells (r_s = 0.377), dendritic cells (r_s = 0.311), and neutrophils (r_s = 0.301). By contrast, LILRB4 expression was negatively associated with the Th17-cell signature (r_s = −0.346) (Fig. 3B).34, 35, 36, 37, 38

Fig. 3.

Immune-cell context of LILRB4 and correlations with macrophage-related estimates. (A) ESTIMATE, immune, and stromal scores in the displayed LILRB4-high and -low groups. (B) Spearman correlations with immune-cell enrichment signatures. (C) Comparison of immune signatures between the high- and low-expression groups. (D) Cell-type expression context from HPA. (E) TISCH2 views of six entries derived from the five accession families specified in Section 2.6. (F, G) Immune-cell annotations and LILRB4 expression in GSE140228 Smart-seq2. (H) Source expression-comparison display. (I, J) Spearman correlations with quanTIseq and CIBERSORT-ABS M2-like macrophage estimates, respectively; nominal p = 6.52 × 10⁻²⁸ and 3.42 × 10⁻⁶⁴. (K) Spearman correlations with CD163, MSR1, IL10, TGFB1, and MRC1. (L) GEPIA2021 deconvolution-derived M1/M2-related expression comparison. ESTIMATE, Estimation of stem and immune cells in malignant tumor tissues using Expression data; HPA, Human Protein Atlas; TISCH2, Tumor Immune Single-cell Hub 2; UMAP, uniform manifold approximation and projection; Th, T helper; NK, natural killer; DC, dendritic cell; aDC, activated DC; iDC, immature DC; pDC, plasmacytoid DC; Tcm, central memory T cell; Tem, effector memory T cell; Tfh, follicular helper T cell; Tgd, gamma delta T cell; Treg, regulatory T cell; ILC, innate lymphoid cell; Mono/Macro, monocyte/macrophage; MAIT, mucosal-associated invariant T cell; Tex, exhausted T cell annotation; CIBERSORT-ABS, absolute-mode CIBERSORT; TPM, transcripts per million.

The displayed high- and low-expression groups differed in macrophage, dendritic-cell, cytotoxic-cell, T-cell, and B-cell signatures (Fig. 3C). Although consistent with differences in immune-cell admixture, these comparisons did not directly measure cell numbers or immunosuppressive function.

3.4. Single-cell context and M2-like macrophage associations

HPA data indicated macrophage-associated expression of LILRB4 (Fig. 3D).³⁰ The six TISCH2 resource entries demonstrated LILRB4 expression in monocyte/macrophage populations, with variable expression in other annotated immune populations (Fig. 3E). These entries are complementary resource views rather than six independent clinical validation cohorts. In the representative GSE140228 Smart-seq2 immune-cell display, LILRB4 expression was detected in the monocyte/macrophage compartment (Fig. 3F and G).18, 29 Accordingly, cell types absent from the display's annotations cannot be evaluated. The source expression-comparison display in Fig. 3H is descriptive only; no claim of differential-expression significance is made from its incompletely specified comparison.

LILRB4 expression correlated positively with M2-like macrophage estimates from quanTIseq (r_s = 0.544; nominal p = 6.52 × 10⁻²⁸) and CIBERSORT-ABS (r_s = 0.753; nominal p = 3.42 × 10⁻⁶⁴; Fig. 3I and J). Positive correlations with marker genes included CD163 (r_s = 0.749), MSR1 (r_s = 0.864), IL10 (r_s = 0.748), TGFB1 (r_s = 0.590), and MRC1 (r_s = 0.375; Fig. 3K). These concordant associations are consistent with macrophage-rich transcriptional features. However, these associations do not establish a discrete M2 phenotype because deconvolution estimates share bulk-expression inputs and the marker genes are not exclusive to one cell type or activation state.23, 24, 36, 37 The GEPIA2021 display suggests higher M2-related than M1-related expression estimates (Fig. 3L); however, this model-derived comparison does not directly measure expression within purified macrophages.³⁹

3.5. Descriptive tissue immunofluorescence shows LILRB4/CD163 signal overlap

Images of LIHC sections showed overlapping LILRB4 and CD163 immunofluorescence signals (Fig. 4A–H). The image-level colocalization coefficient was reported as 0.5719 (Fig. 4I). These images cannot establish strong colocalization, macrophage-exclusive expression, or regulation of macrophage polarization by LILRB4.

Fig. 4.

Descriptive immunofluorescence analysis of LILRB4 and CD163 in LIHC tissue. (A) DAPI; (B) LILRB4; (C) CD163; (D) merged fluorescence channels; (E) enlarged merged view. White arrowheads denote representative overlapping signals; red arrowheads denote CD163-positive regions showing little corresponding LILRB4 signal. (F, G) Grayscale channels for LILRB4 and CD163. (H) Colocalized-pixel map. (I) The reported image-level coefficient, Rcoloc = 0.5719, is reported only as an original image-level output. Tissue specimens from five independent patients were evaluated (n = 5). For each patient, three spaced sections per tissue type and five non-overlapping fields per section were examined, yielding 15 fields each for HCC and adjacent liver tissue. The displayed coefficient is an image-level result, not a patient-level mean. No patient-level test or p value is reported. Scale bars in panels D and E represent 10 μm, based on author-confirmed panel-specific information. Spatial calibration is not inferred for the other panels. Rcoloc is the label in the retained image-level output. The staining methods and technical reporting limitations are detailed in Section 2.7. DAPI, 4′,6-diamidino-2-phenylindole; LIHC, liver hepatocellular carcinoma.

Although the limited tissue observation complements the public-data associations, it does not establish the prevalence of LILRB4/CD163 overlap across patients with HCC, M2 functional identity, or a survival relationship in a local cohort. A larger, appropriately characterized series would be required to assess these outcomes. No expanded cohort or tissue microarray was analyzed during this revision.

4. Discussion

This study shows consistent associations among LILRB4 expression, macrophage-related transcriptional features, and adverse survival patterns in LIHC. Bulk-expression and deconvolution findings indicate macrophage enrichment; public single-cell datasets localize LILRB4 predominantly to monocyte/macrophage populations, and tissue immunofluorescence provides complementary spatial context of LILRB4/CD163 overlap. Collectively, these observations support LILRB4 as a macrophage-associated immune marker in LIHC and justify further clinical and mechanistic investigation.

The bulk-expression associations may arise partly from variation in the abundance of infiltrating myeloid cells. This interpretation accords with the correlations between LILRB4 and multiple leukocyte-associated genes and with its expression in monocyte/macrophage populations in single-cell views. The data cannot determine whether the signal reflects greater transcript abundance within each cell or a larger population of LILRB4-expressing cells. The resource displays therefore do not establish tumor-cell specificity or expression restricted to macrophages.

Previous studies have implicated LILRB4 in myeloid immune regulation and reported its expression in HCC-associated immune cells.6, 7, 8, 9, 10, 14, 15 The present analysis integrates LILRB4-related bulk expression, immune deconvolution, and single-cell observations in LIHC. The tissue images add spatial context but do not provide clinical validation, and overlap among the underlying public datasets limits the study's contribution. Multiple analytical views of the same cohort offer convergent perspectives without constituting external replication.

Correlations with CD163, MSR1, IL10, TGFB1, and MRC1 align with macrophage-associated programs described in tumor biology.20, 21, 22, 23, 24, 26 However, a binary M1/M2 framework does not capture the continuum of macrophage states, and none of these genes is a specific functional measure. The deconvolution and marker results are therefore best described as M2-like macrophage-associated transcriptional features. Accordingly, these results do not show that LILRB4 creates an immunosuppressive environment, causes T-cell dysfunction, or predicts response to immune checkpoint blockade.

The protein-association network and functional annotations are consistent with established antigen presentation and myeloid inhibitory signaling biology.6, 7, 8, 9, 10, 11, 12, 13, 32, 33 Nevertheless, database network links do not provide direct experimental evidence of these interactions in HCC. Similarly, enrichment of an annotation does not establish activation or inhibition of the corresponding pathway. Experimental observations in other tumor systems motivate HCC-specific evaluation of LILRB4-mediated immunologic effects but provide no evidence for its functional effect in the present cohort.10, 11

Our glioblastoma study previously reported experimental findings involving LILRB4-positive TAMs and STAT3/IL10-associated signaling.¹⁷ Because those findings derive from another tumor system, they do not functionally validate the pathway in HCC. Although LILRB4 correlated with IL10 expression in the present analysis, the study did not include LILRB4 perturbation, measurement of STAT3 phosphorylation, cytokine-secretion assays, or macrophage rescue experiments. A role for LILRB4 in STAT3/IL10-associated macrophage biology in HCC therefore remains a hypothesis informed by prior work and the current associations.12, 25, 26 The direction, cellular context, and causal relevance of this association remain unresolved.

The survival results should be interpreted with particular caution. The grouping thresholds were derived from the outcomes, and selection of LILRB4 followed review of multiple family members and endpoints. The nominal p values and HRs may therefore overstate the reproducibility of the observed separation. In the univariable reports, continuous expression was related to OS but not clearly to DSS. The adjusted OS estimate was obtained from a smaller, selected complete-case set and a model with a global PH-assumption violation; the adjustment set did not jointly incorporate age, AFP, cirrhosis, and liver function. Tumor status could incorporate post-baseline information because the report did not define its assessment time. LILRB4 was excluded from the DSS model, which also yielded an unstable estimate for tumor status. Consequently, the analysis supports neither independent prognostic utility nor a clinically applicable cutoff. Establishing such claims would require a prespecified baseline-covariate model, appropriate diagnostics, and independent validation.

Several further limitations apply. Bulk expression cannot separate changes in cell abundance from regulation within individual cells, and deconvolution methods may rely on correlated input signals. Survival and LILRB4–macrophage findings were not tested in an external bulk HCC cohort, and the single-cell resource views cannot substitute for such validation. Tissue immunofluorescence was limited to five patients and provided descriptive spatial evidence. The reported colocalization coefficient was an image-level output rather than a patient-level estimate, and the tissue observations did not establish a relationship with survival in a local cohort. Incomplete parameter-level reporting for some online analyses and for the tissue component further reduces reproducibility. Mechanistic experiments are required to evaluate effects on macrophage polarization and the STAT3/IL10 axis in HCC. In light of these constraints, the findings should be regarded as hypothesis-generating rather than confirmatory.

5. Conclusions

LILRB4 expression is associated with macrophage-related transcriptional features and adverse survival patterns in LIHC. Together, bulk transcriptomic, immune-deconvolution, and single-cell observations identify LILRB4 as a candidate marker of the macrophage-rich HCC microenvironment. These exploratory findings support further clinical validation and functional investigation of its prognostic and immunoregulatory relevance.

CRediT authorship contribution statement

Qi Liu: Writing – original draft, Visualization, Validation, Software, Project administration, Methodology, Investigation, Formal analysis, Conceptualization, Data curation. Yuyang Liu: Writing – original draft, Visualization, Validation, Software, Methodology, Investigation, Formal analysis, Data curation. Junyi Chen: Writing – original draft, Visualization, Validation, Software, Methodology, Investigation, Formal analysis, Data curation. Tao Wan: Investigation. Chao Li: Investigation. Ning Zhang: Writing – review & editing, Visualization, Supervision, Project administration, Methodology, Investigation, Formal analysis, Conceptualization. Zhanyu Yang: Writing – review & editing, Visualization, Supervision, Project administration, Methodology, Investigation, Formal analysis, Conceptualization.

Informed consent

The analyses of public data used publicly available, deidentified transcriptomic, clinical, and single-cell resources. No participants were newly enrolled, and the secondary analyses did not access personally identifiable information. Ethical approval and consent for primary data collection are reported in the cited source studies.

Ethics statement

Immunofluorescence used archived pathological specimens from a parent study approved by the Ethics Committee of Chinese PLA General Hospital on 20 January 2022 (approval no. S2021-608-02). This approval number refers to the parent study and is not presented as a separate approval issued specifically for the present LILRB4 analysis. Participants provided written informed consent permitting use of their medical records and pathological specimens in subsequent research. The retained specimens were analyzed under that permission for secondary use.

Organ donation

Not applicable.

Animal treatment

Not applicable.

Data availability statement

The Genomic Data Commons (https://portal.gdc.cancer.gov/) and TCGA Pan-Cancer Clinical Data Resource provide public access to the TCGA-LIHC expression and clinical datasets.²⁷ Section 2.6 identifies the analyzed TISCH2 entries and accession families, which can be accessed at https://tisch.comp-genomics.org/; HPA provides the cell-type information at https://www.proteinatlas.org/. Supplementary Tables S1–S4 document the survival procedures, grouped summaries, Cox-model coefficients, diagnostic results, missing-data information, and analysis specifications. Supporting immunofluorescence tissue images appear in Fig. 4.

Declaration of generative AI and AI-assisted technologies in the writing process

Not applicable.

Funding

This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors.

Declaration of competing interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Appendix A. Supplementary data

The following is the Supplementary data to this article: