Single-cell landscape of human liver sinusoidal γδ T cells in the context of latent human cytomegalovirus infection

Abstract

Human γδ T cells are typically divided into Vγ9Vδ2 and non-Vγ9Vδ2 cells, but their detailed heterogeneity remains to be fully elucidated, especially in the liver where they are enriched. Here we analyze liver sinusoidal γδ T cells from healthy donors, with or without latent human cytomegalovirus (HCMV) infection, by performing single-cell RNA sequencing with antibody-derived tags. Vγ9Vδ2 cells are characterized by PLZF expression, and classified into type 1 and type 3 immunity-associated clusters. In contrast, non-Vγ9Vδ2 cells are characterized by Helios expression, and their clusters display naïve-to-effector differentiation processes, which are accelerated by HCMV infection. Notably, among non-Vγ9Vδ2 cells, we identify a novel liver-resident CD56^hiCD161⁻ cell population with a distinct TCR repertoire, which is markedly enriched in HCMV-seropositive donors. These cells exhibit IL-15-induced NK-like cytotoxicity, but limited responsiveness to TCR stimulation. Our findings reveal liver sinusoidal γδ T cells’ heterogeneity in the context of latent HCMV infection, and their functions.

Current understanding of human γδ T cells in the liver remains limited, particularly with regard to their heterogeneity. The authors here provide a comprehensive profile of these cells and identify liver-resident CD56^hiCD161⁻ γδ T cells that exhibit IL-15-induced, NK-like cytotoxicity against cytomegalovirus-infected cells and liver tumor cells.

Introduction

Characterized by unique T-cell receptors (TCRs) composed of γ and δ chains, γδ T cells constitute a distinct subset of unconventional T cells¹. While conventional αβ T cells recognize peptide antigens presented by major histocompatibility complex (MHC) molecules, γδ T cells possess broader antigen recognition capabilities, including the ability to detect various molecular patterns, e.g., stress-induced antigens, phosphoantigens, and non-peptide molecules. γδ T cells serve as important effector cells in various diseases, including infectious disease, cancers, and autoimmune disease².

In mice, γδ T cells are primarily classified based on their TCR γ-chain usage, with distinct subpopulations showing precise tissue-specific localization and specialized functions^3–5. On the other hand, human γδ T cells exhibit distinct classification patterns primarily based on TCR δ-chain expression into Vδ2 and non-Vδ2 subsets, where Vδ2⁺ cells constitute the major population in peripheral blood, and non-Vδ2⁺ cells are enriched in tissues⁶. While γδ T-cell development and functions in mice have been well-investigated, human γδ T cells remain incompletely understood, with most research focused on healthy peripheral blood, thymus, or cancer due to limited sample availabilty^7–10.

The liver serves as an immune sentinel, playing a central role in immune surveillance by defending against gut-derived microbes^11,12. Notably, the liver harbors high numbers of innate lymphocytes—including natural killer (NK) cells and unconventional T cells, such as invariant natural killer T (iNKT), mucosal-associated invariant T (MAIT), and γδ T cells^13–15. γδ T cells play roles in maintaining physiological homeostasis in the liver and modulating pathological conditions, including viral hepatitis, liver cirrhosis, and hepatocellular carcinoma (HCC)^16,17. Their therapeutic potential is exemplified by the successful application of ex vivo-expanded allogeneic γδ T cells in HCC¹⁸.

Human cytomegalovirus (HCMV)—also known as human herpesvirus 5—is ubiquitous and highly prevalent¹⁹. Although immunocompetent HCMV-infected individuals usually remain asymptomatic, HCMV plays a significant role in shaping the immune system²⁰, reportedly driving the differentiation of NK and T cells, including γδ T cells^21,22. Among γδ T cells, HCMV influences the differentiation of non-Vδ2⁺ cells, promoting their transition from a naïve to an effector state with clonal expansion^23,24. However, the detailed process of HCMV-induced γδ T-cell differentiation remains poorly understood.

Recent advances in single-cell RNA-seq (scRNA-seq) have revolutionized our understanding of immune cell heterogeneity. The incorporation of DNA-barcoded antibodies (Ab) against CD markers with scRNA-seq has enabled cluster identification within sequencing data by using specific surface markers^25,26. This multimodal strategy enables more precise identification of cell populations and more direct validation of transcriptomic profiles by flow cytometry.

In the present study, we develop a comprehensive single-cell atlas of human liver sinusoidal γδ T cells from 12 healthy donors, with or without HCMV seropositivity. We classify γδ T cells into two broad lineages—Vγ9Vδ2 and non-Vγ9Vδ2—and then further delineate distinct clusters within each lineage, including clusters enriched by HCMV infection. Notably, we identify previously unrecognized liver-resident CD56^hiCD161⁻ γδ T cells that exert NKG2D-mediated, TCR-independent cytotoxicity against target cells. Moreover, by leveraging public single-cell datasets, we validate that several γδ T-cell subsets identified in the liver are also present in other organs. Overall, our study provides critical insights into the heterogeneity and functional roles of human liver γδ T cells, and establishes a valuable reference resource that can inform future immunological studies and therapeutic strategies.

Results

Single-cell RNA-seq analysis identifies clusters within two distinct populations, Vγ9Vδ2 and non-Vγ9Vδ2, among human liver sinusoidal γδ T cells

Liver perfusates were obtained from healthy donors during living-donor liver transplantation. From these perfusates, we isolated liver sinusoidal mononuclear cells (LSMCs), and further sorted γδ T cells. The sorted γδ T cells from six HCMV-seropositive and six HCMV-seronegative individuals were analyzed by scRNA-seq with DNA-barcoded antibodies against CD markers, also known as antibody-derived tags (ADTs) (Fig. 1a, Supplementary Fig. 1). Unsupervised clustering analysis of scRNA-seq data from 27,722 γδ T cells, using uniform manifold approximation and projection (UMAP), revealed 10 distinct clusters (Fig. 1b, Supplementary Figs. 2 and 3). Cluster distribution strongly correlated with TCR δ-chain transcripts (Fig. 1c). Specifically, C1–C3 expressed TRDV2 and TRGV9, while C4–C9 predominantly expressed TRDV1 (Fig. 1d). C10 uniquely expressed TRDV1 or TRDV3, highlighting clear segregation of TCR δ-chain usage.

Fig. 1. Single-cell profiling of liver sinusoidal γδ T cells reveals distinct non-Vγ9Vδ2 and Vγ9Vδ2 γδ T-cell populations.

a Schematic overview of experimental design. b Uniform manifold approximation and projection (UMAP) projections of liver sinusoidal γδ T cells (n = 27,722 cells) colored according to clusters, labeled C1 to C10. Expressions of TRDV1, TRDV2, TRDV3, and TRGV9 shown on UMAPs (c) and violin plots (d). e Volcano plot of the differentially expressed genes (DEGs) in non-Vγ9Vδ2 and Vγ9Vδ2 γδ T-cell clusters. Left, 80 genes showing a log₂FC < 1 with p < 0.05. Right, 56 genes showing a log₂FC > 1 with p < 0.05. f UMAPs colored according to the transcription factors IKZF2 and ZBTB16. g Representative flow cytometry histograms and cumulative data presenting the expression of Helios and PLZF in non-Vδ2 and Vδ2 γδ T cells (n = 9 biologically independent donors). Data are presented as mean ± s.d. Statistics: Two-sided Mann–Whitney test with Bonferroni correction (e); two-sided Wilcoxon signed-rank test (g). Source data are provided as a Source data file.

We first compared transcriptomic features between Vγ9Vδ2 and non-Vγ9Vδ2 cells. Vγ9Vδ2 cells were characterized by upregulation of multiple NK receptor genes (particularly KLRB1 and KLRC1) and receptor genes for IL-12 and IL-18 (including IL12RB2 and IL18RAP) (Fig. 1e). Non-Vγ9Vδ2 cells exhibited upregulation of different NK receptor genes (including FCGR3A, KLRC2, and KLRC3) along with genes for NK receptor signaling proteins (e.g., TYROBP). Among transcription factors, ZBTB16 was characteristically upregulated in Vγ9Vδ2 cells, and IKZF2 in non-Vγ9Vδ2 cells (Fig. 1e, f). These findings were confirmed at the protein level through flow cytometry analysis for PLZF and Helios, which are encoded by ZBTB16 and IKZF2, respectively (Fig. 1g).

Next, we analyzed characteristics of C1–C3 within the Vγ9Vδ2 population, and C4–C10 within the non-Vγ9Vδ2 population (Fig. 2a and Supplementary Fig. 4). C1 exhibited characteristics of type 3 immunity, including upregulation of RORC, IL23R, CXCR6, CD26 ADT, and CD161 ADT. In contrast, C2 and C3 displayed type 1 immunity profiles, showing upregulation of TBX21, IFNG, and CXCR3. Moreover, C2 exhibited an early-differentiated phenotype, including upregulation of NR4A2, NR4A3, and CD27 ADT, while C3 showed late-differentiated phenotypes, with elevated expressions of GZMB, GNLY, ZEB2, and CD57 ADT.

Fig. 2. γδ T cells exhibit distinct characteristics and differentiation.

Dot plots of transcripts and antibody-derived tag (ADT) expression for each Vγ9Vδ2 (a) and non-Vγ9Vδ2 (b) γδ T-cell subcluster. Circle size indicates the percentage of cells expressing the gene or ADT. Color gradient corresponds to increasing mean expression value. c Uniform manifold approximation and projection (UMAP) projections of differentiation status colored according to Cytotrace 2 scores. d Bar graphs showing T-cell receptor (TCR) diversity calculated using the inverse Simpson index for each γδ T-cell cluster. e Heatmap showing Morisita’s overlap index, representing TCR repertoire similarity among γδ T-cell clusters. Source data are provided as a Source data file.

Among non-Vγ9Vδ2 γδ T-cell clusters, C4 exhibited a naïve phenotype, marked by upregulation of TCF7, IL7R, SELL, and CD27 ADT (Fig. 2b and Supplementary Fig. 4), while C5 showed tissue-repair signatures, including AREG^27,28. C6–C9 were characterized by cytolytic and effector functions, with common expression of EOMES, KLRF1 (encoding NKp80), and CD45RA ADT. More specifically, C8 was distinguished by marked upregulation of cytotoxic molecule genes and CD57 ADT (a marker of replicative senescence); C9 exhibited upregulation of CD8 and CD16 ADT; and C10 typically exhibited NK-like features, including upregulation of CD56 ADT, NKG2C ADT, and NKG2D ADT.

To assess the differentiation states of γδ T-cell clusters, we used CytoTRACE2-based differentiation status scores (Fig. 2c)²⁹. Among Vγ9Vδ2 γδ cell clusters, C1 and C2 were in less-differentiated states, while C3 was in a more-differentiated state. Among non-Vγ9Vδ2 cell clusters, C4 and C5 exhibited less-differentiated naïve characteristics, while C6–C10 displayed more-differentiated effector profiles. TCR diversity was evaluated using the inverse Simpson index, revealing that the Vγ9Vδ2 clusters (C1–C3) each displayed similar degrees of TCR diversity (Fig. 2d and Supplementary Fig. 5). In contrast, among non-Vγ9Vδ2 γδ clusters, TCR diversity was high in C4 and C5, while C6–C10 exhibited reduced TCR diversity, indicating that they experienced clonal expansion. We further assessed the clonal relationships between clusters using Morisita’s overlap index analysis (Fig. 2e). Vγ9Vδ2 clusters (C1–C3) clonally overlapped with each other. Among the non-Vγ9Vδ2 γδ clusters, the differentiated effector-like clusters (C6–C9) clonally overlapped with each other. As expected based on its unique TCR δ-chain usage, C10 had a unique TCR profile within the non-Vγ9Vδ2 cell population.

In summary, scRNA-seq analysis with DNA-barcoded antibodies against CD markers revealed two distinct populations (Vγ9Vδ2 and non-Vγ9Vδ2), along with clusters showing different differentiation states among human liver sinusoidal γδ T cells. Notably, the C10 non-Vγ9Vδ2 cluster was a clonally segregated population with NK-like features.

The CD26⁺CD161⁺ cluster is the Vγ9Vδ2 subpopulation showing the highest PLZF expression and cytokine production

We next analyzed functional characteristics of Vγ9Vδ2 cells. First, we performed gene set enrichment analysis using gene sets from various types of liver sinusoidal T cells²⁵. Vγ9Vδ2 cells exhibited significant enrichment for a gene set of MAIT cells (Fig. 3a), which also typically express PLZF³. Moreover, Vγ9Vδ2 cells produced IFN-γ in response to stimulation with IL-12 and IL-18, while both Vγ9Vδ2 and non-Vγ9Vδ2 cells produced IFN-γ when they were stimulated by PMA and ionomycin as positive controls (Fig. 3b). We further aimed to examine the functional heterogeneity of Vγ9Vδ2 cells. To this end, we identified markers that defined C1 and C2/C3, based on distinct surface protein profiles of each cluster obtained from scRNA-seq with ADTs (Figs. 2a and 3c, and Supplementary Figs. 6–8). We determined that C1 could be defined as CD26⁺CD161⁺, C2/C3 as non-CD26⁺CD161⁺ (Fig. 3d).

Fig. 3. Vγ9Vδ2 γδ T cells show functional heterogeneity, with the CD26⁺CD161⁺ cluster exhibiting the highest PLZF expression and cytokine production.

a Gene set enrichment analysis of Vγ9Vδ2 γδ T cells compared with non-Vγ9Vδ2 γδ T cells using published MAIT cell signatures (GSE200173). NES, normalized enrichment score. b Proportion of IFN-γ⁺ cells among non-Vδ2 and Vδ2 γδ T cells after IL-12 and IL-18 stimulation (50 ng/mL each, 18 h), with PMA/ionomycin as a positive control (n = 12 biologically independent donors). c Violin plots showing ADT expression levels of CD26 and CD161 across Vγ9Vδ2 γδ T cell clusters (C1 and C2/C3). Horizontal lines indicate the median expression level. d Representative flow cytometry plots of Vδ2 γδ T-cell clusters, categorized into CD26⁺CD161⁺ (C1) and non-CD26⁺CD161⁺ (C2/C3) cells. e Representative flow cytometry histograms and cumulative data presenting the PLZF expression in the Vδ2 γδ T-cell population (n = 23 biologically independent donors). Violin plots showing expressions of MAIT gene set module score (f) and KLRB1 (g) in Vγ9Vδ2 clusters (C1–C3). Horizontal lines indicate the median expression level. h Proportions of IFN-γ⁺ and TNF⁺ cells among CD26⁺CD161⁺ (C1) and non-CD26⁺CD161⁺ (C2/C3) cells after IL-12 and IL-18 stimulation (50 ng/mL each, 18 h), with PMA/ionomycin as a positive control (n = 12 biologically independent donors). i Expression of IL-18Rα (CD218a) in the Vδ2 γδ T-cell population (n = 12 biologically independent donors). j Representative flow cytometry plots and proportion of IL-17A⁺ cells among CD26⁺CD161⁺ (C1), non-CD26⁺CD161⁺ (C2/C3), and MAIT cells as a positive control from peripheral blood (squares) and liver sinusoidal (circles) mononuclear cells after PMA/ionomycin stimulation (10 ng/mL; 500 ng/mL, 6 h) (n = 8 biologically independent donors). Data are presented as mean ± s.d. Statistics: Gene set enrichment analysis with false discovery rate correction (a); two-sided Wilcoxon signed-rank tests (b, e, h, i, j); Source data are provided as a Source data file.

Among Vγ9Vδ2 cells, flow cytometry analysis revealed that PLZF expression was higher in CD26⁺CD161⁺ cells compared to non-CD26⁺CD161⁺ cells (Fig. 3e). Transcriptomic analysis from the scRNA-seq data further revealed that the MAIT gene set module score was highest in C1 and lowest in C3, among Vγ9Vδ2 cells (Fig. 3f). We additionally confirmed that KLRB1 and ZBTB16 expression was highest in C1 (Fig. 3g and Supplementary Fig. 9).

We also used flow cytometry to investigate functions of Vγ9Vδ2 cells obtained from LSMCs. In accordance with the PLZF expression, we found that the frequency of IFN-γ⁺ or TNF⁺ cells upon stimulation with IL-12 and IL-18 was higher in CD26⁺CD161⁺ cells, compared to non-CD26⁺CD161⁺ cells (Fig. 3h). Moreover, CD26⁺CD161⁺ cells exhibited the higher expression of IL-18Rα than non-CD26⁺CD161⁺ cells (Fig. 3i). We further examined IL-17A production, and found that the frequency of IL-17A⁺ cells was only significantly increased in CD26⁺CD161⁺ cells from LSMCs following 6 h PMA/ionomycin stimulation (Fig. 3j). However, CD26⁺CD161⁺ cells from peripheral blood mononuclear cells (PBMCs) did not show increased IL-17A production. MAIT cells were included in IL-17A ICS assays as a representative IL-17A-producing cell type. Overall, among Vγ9Vδ2 cells from LSMCs, we found that CD26⁺CD161⁺ cells exhibited high PLZF expression and MAIT cell features, and were superior in terms of cytokine production.

HCMV infection leads to clonal expansion and differentiation of non-Vγ9Vδ2 cells

Next, we aimed to examine the impact of latent HCMV infection on γδ T cells. The frequency of γδ T cells among total liver sinusoidal CD3⁺ T cells tended to be increased in HCMV-seropositive individuals compared to HCMV-seronegative individuals, although this difference was not statistically significant (Supplementary Fig. 10a). When cluster distributions were compared between HCMV-seropositive versus HCMV-seronegative individuals, no significant difference was observed for Vγ9Vδ2 cells (Fig. 4a, b and Supplementary Fig. 10b). However, among non-Vγ9Vδ2 cells, three clusters (C8–C10) were expanded in HCMV-seropositive individuals. Notably, C8–C10 exhibited more-differentiated effector profiles, with reduced TCR diversity compared to C4 and C5 (Fig. 2c, d).

Fig. 4. Non-Vγ9Vδ2 γδ T cells undergo clonal expansion and differentiation in latent human cytomegalovirus (HCMV) infection.

a Uniform manifold approximation and projections (UMAPs) split by HCMV-seronegative and HCMV-seropositive donors, colored according to clusters. b Alluvial plots showing the comparison of average proportions of γδ T-cell clusters from HCMV-seronegative and HCMV-seropositive donors. c T-cell receptor (TCR) diversity, as assessed by inverse-Simpson index, across γδ T-cell clusters in HCMV-seronegative (green) and HCMV-seropositive (purple) cohorts. d UMAP projections colored according to TCR clonality, split by HCMV-seronegative and HCMV-seropositive donors. TCR clonality is categorized as single (n = 1), small (1 <n ≤ 5), medium (5 <n ≤ 20), large (20 <n ≤ 100), or hyperexpanded (100 < n ≤ 500). e Violin plots showing ADT expression levels of CD27, CD57, CD8, CD16, CD56 and CD161 across non-Vγ9Vδ2 γδ T cell clusters (C4, C8, C9 and C10). Horizontal lines indicate the median expression level. f Representative flow cytometry plots of non-Vγ9Vδ2 γδ T-cell clusters, categorized into CD56^hiCD161⁻ (C10), CD27⁺CD57⁻ (C4), and CD27⁻CD57⁺ (C8) populations. g Frequencies of CD27⁺CD57⁻ (C4), CD27⁻CD57⁺ (C8), CD16⁺CD8⁺ (C9), and CD56^hiCD161⁻ (C10) cells among γδ T cells from HCMV-seronegative and HCMV-seropositive donors (n = 8 biologically independent donors per group). h Frequencies of CD27⁻CD45RA⁻ (effector memory) and CD27⁻CD45RA⁺ (CD45RA⁺ terminally differentiated) cells among CD56^hiCD161⁻ and CD27⁻CD57⁺ non-Vδ2 γδ T cells (n = 16 biologically independent donors). Data are presented as mean ± s.d. Statistics: Two-sided Mann–Whitney U test (g); two-sided Wilcoxon signed-rank tests (h). Source data are provided as a Source data file.

We directly compared the inverse-Simpson index and found that TCR diversity in C1–C3 was comparable between the groups with and without latent HCMV infection (Fig. 4c). On the other hand, among non-Vγ9Vδ2 cells, C6–C10 exhibited reduced TCR diversity in the HCMV-seropositive group compared to the HCMV-seronegative group, compatible with their more-differentiated effector profiles. Notably, C9 was absent among HCMV-seronegative individuals. This analysis of TCR clonality also revealed that hyperexpanded and large clones within C6–C10 were predominant among HCMV-seropositive individuals, whereas small or single clones were more frequently observed in seronegative individuals (Fig. 4d). In summary, our results indicated that HCMV infection impacts non-Vγ9Vδ2 cells by promoting clonal expansion and more advanced differentiation.

Flow cytometry was performed to further examine functional characteristics of clusters within the non-Vγ9Vδ2 population. We focused on C4, a cluster with naïve phenotypes, and C8–C10, which were expanded in HCMV-seropositive individuals. We analyzed distinct surface protein profiles obtained from scRNA-seq with ADTs (Figs. 2b and 4e, and Supplementary Figs. 7 and 8), and identified markers defining each cluster: C4 could be defined as CD27⁺CD57⁻, C8 as CD27⁻CD57⁺, C9 as CD16⁺CD8⁺, and C10 as CD56^hiCD161⁻ (Fig. 4f). Flow cytometry analysis confirmed the increased frequencies of CD27⁻CD57⁺ cells (C8) and CD56^hiCD161⁻ cells (C10) in HCMV-seropositive individuals compared to HCMV-seronegative individuals. CD16⁺CD8⁺ cells (C9) did not show a significantly increased frequency in HCMV-seropositive individuals (Fig. 4g), and thus were not further examined. We analyzed differentiation markers and found that compared to CD56^hiCD161⁻ cells, CD27⁻CD57⁺ cells exhibited a significantly lower frequency of effector memory T cells, and a significantly higher frequency of CD45RA⁺ terminally differentiated T cells (Fig. 4h).

The CD56^hiCD161⁻ cluster is a non-Vγ9Vδ2 subpopulation that exerts IL-15-driven NK-like cytotoxicity

We examined the functionality of CD27⁺CD57⁻ cells, CD27⁻CD57⁺ cells, and CD56^hiCD161⁻ cells. In response to anti-CD3/anti-CD28 stimulation, we observed a significantly increased frequency of IFN-γ⁺TNF⁺ cells among CD27⁺CD57⁻ cells, but not among CD27⁻CD57⁺ cells or CD56^hiCD161⁻ cells (Fig. 5a). This finding indicated that CD27⁻CD57⁺ and CD56^hiCD161⁻ cells were hypo-responsive to TCR-mediated stimulation. The same stimulation yielded a significant increase in the frequency of IFN-γ⁺TNF⁺ cells among total Vδ2⁺ cells.

Fig. 5. CD56^hiCD161⁻ cluster exhibits IL-15-driven natural killer (NK)-like cytotoxicity among non-Vγ9Vδ2 γδ T cells.

a Proportion of IFN-γ⁺TNF⁺ cells following CD3/CD28 stimulation (n = 11 biologically independent donors). Liver sinusoidal mononuclear cells (LSMCs) were stimulated for 6 h, using Dynabeads^TM Human T-Activator CD3/CD28 at a bead:cell ratio of 1:25. b Representative flow cytometry histograms and cumulative data showing cytotoxicity of sorted CD8⁺ αβ T cells, γδ T cells, and NK cells following IL-15 stimulation (10 ng/mL, 48 h) and co-culture with PKH-26-labeled K562 cells at a 20:1 effector:target ratio for 12 h (n = 8 biologically independent donors). Dead cells were stained with TO-PRO-3, and then cytotoxicity was analyzed by flow cytometry. c Frequency of CD107a⁺ cells among CD27⁺CD57⁻, CD27⁻CD57⁺, and CD56^hiCD161⁻ non-Vδ2 cells, and among total non-Vδ2 and Vδ2 γδ T cells, following 48 h stimulation with or without IL-15 (10 ng/mL), and incubation with or without K562 cells (n = 14 biologically independent donors). d NKG2D level following IL-15 stimulation (10 ng/mL) for 48 h in the non-Vδ2 γδ T-cell population (n = 14 biologically independent donors). e Frequencies of IFN-γ⁺TNF⁺ cells following IL-15 stimulation (10 ng/mL) for 36 h, in the non-Vδ2 γδ T-cell population (n = 10 biologically independent donors). f IL-2/IL-15Rβ (CD122) expression in the non-Vδ2 γδ T-cell population (n = 10 biologically independent donors). g–i Human neonatal primary dermal fibroblast (HDFn) cells (either mock- or HCMV-infected) were co-cultured with LSMCs carrying fully or partially matched HLA-I alleles (n = 10 biologically independent donors) at a 4:1 E:T ratio for 16 h with or without IL-15 (10 ng/mL). Frequency of CD107a⁺ cells among CD56^hiCD161^– (C10) cells (g). Frequency of CD107a⁺ cells among CD56^hiCD161^– (C10) cells with or without HLA-I blocking Ab (10 μg/mL) (h), and with or without MIC-A/B blocking Ab (10 μg/mL) (i). j–l Co-culture was performed under the same conditions as in (g–i), except for using LSMCs carrying mismatched HLA-I alleles (n = 6 biologically independent donors). Frequency of CD107a⁺ cells among CD56^hiCD161^– (C10) cells (j). Frequency of CD107a⁺ cells among CD56^hiCD161^– (C10) cells with or without HLA-I blocking Ab (10 μg/mL) (k), and with or without MIC-A/B blocking Ab (10 μg/mL) (l). Data are presented as mean ± s.d. Statistics: Two-sided Wilcoxon signed-rank tests (a–f); Friedman tests with Dunn’s multiple-comparisons tests (c); two-way repeated-measures analyses of variance with Šídák’s multiple-comparisons tests (g–l). Source data are provided as a Source data file.

We also investigated the NK-like functions of γδ T cells. Specifically, we examined TCR-independent IL-15-induced NK-like cytotoxic activity against K562 cells lacking HLA class I (HLA-I) expression, which has been previously studied in various types of T cells^30,31. We found that total γδ T cells exhibited IL-15-induced NK-like cytotoxicity, which was higher than that of CD8⁺ αβ T cells and lower than that of NK cells (Fig. 5b). We further investigated which γδ subpopulations exerted this IL-15-induced NK-like cytotoxicity. In all examined γδ subpopulations, IL-15 stimulation followed by K562 co-culture led to a significantly increased frequency of degranulating cells marked by CD107a staining (Fig. 5c). Among non-Vδ2⁺ subpopulations, the frequency of CD107a⁺ cells was significantly higher in CD56^hiCD161⁻ cells, compared to CD27⁺CD57⁻ or CD27⁻CD57⁺ cells. The frequency of CD107a⁺ cells was also significantly higher among CD56^hiCD161⁻ cells, compared to total Vδ2⁺ or non-Vδ2⁺ cells. These results indicate that CD56^hiCD161⁻ cells are the γδ subpopulation specialized for TCR-independent IL-15-induced NK-like cytotoxicity.

Since IL-15-induced NK-like cytotoxicity is mainly mediated by NKG2D³², we next examined IL-15-induced NKG2D upregulation. We found that CD56^hiCD161⁻ cells exhibited significantly higher levels of NKG2D expression following IL-15 stimulation, compared to CD27⁺CD57⁻ or CD27⁻CD57⁺ cells (Fig. 5d). Additionally, the IL-15-induced production of IFN-γ and TNF was significantly higher among CD56^hiCD161⁻ cells, compared to CD27⁺CD57⁻ or CD27⁻CD57⁺ cells (Fig. 5e). The expression of CD122 (IL-2/IL-15Rβ) was also significantly higher on CD56^hiCD161⁻ and CD27⁺CD57⁻ cells, compared to CD27⁻CD57⁺ cells with the highest level on CD56^hiCD161⁻ cells (Fig. 5f). scRNA-seq analysis also confirmed that C4 and C10 exhibit higher IL2RB expression than C8 (Supplementary Fig. 11). These results may explain why CD56^hiCD161⁻ cells were more responsive to IL-15 stimulation compared to the other γδ subpopulations. We also examined IL-12/IL-18-induced IFN-γ and TNF production and found that all non-Vγ9Vδ2 subsets were inferior to Vγ9Vδ2 cells (Supplementary Fig. 12).

Next, we examined whether CD56^hiCD161⁻ cells directly exert cytotoxic activity against HCMV-infected cells. We co-cultured HCMV-infected or non-infected fibroblasts with IL-15-treated or untreated LSMCs that were selected based on HLA-I allele matching with the fibroblasts. The frequency of CD107a⁺ cells among CD56^hiCD161⁻ cells was significantly increased against HCMV-infected cells compared to non-infected cells, and it was further increased by IL-15 treatment (Fig. 5g). When HLA-I blocking Ab was added to the co-culture, the frequency of CD107a⁺ cells was significantly increased in the presence of IL-15 (Fig. 5h), suggesting that the cytotoxic activity might be inhibited by the interaction between HLA-I and KIRs. Notably, MIC-A/B blocking Ab significantly reduced the frequency of CD107a⁺ cells among IL-15-treated CD56^hiCD161⁻ cells (Fig. 5i), showing that NKG2D contributes to the cytotoxicity under this condition. Given that NKG2D mediates this cytotoxicity, we also performed the cytotoxicity assays using CD56^hiCD161⁻ cells from HLA-I allele-mismatched donors. As a result, the frequency of CD107a⁺ cells among CD56^hiCD161⁻ cells was significantly increased against HCMV-infected cells compared to non-infected cells (Fig. 5j), indicating that the cytotoxic activity is not restricted by HLA-I. Cytotoxic activity was increased in the presence of HLA-I blocking Ab (Fig. 5k) and reduced in the presence of MIC-A/B blocking Ab (Fig. 5l), again indicating NKG2D-mediated cytotoxicity that is inhibited by the HLA-I–KIR interaction. Taken together, these data demonstrate that CD56^hiCD161⁻ cells exert NKG2D-mediated cytotoxicity against HCMV-infected cells, which can be regulated by KIRs.

The CD56^hiCD161⁻ cluster is enriched in the liver compared to peripheral blood

To examine liver-enrichment of each γδ T-cell cluster, we performed additional scRNA-seq analysis using matched PBMC samples (HCMV seropositive, n = 4; HCMV seronegative, n = 2) obtained from the same donors whose LSMCs were originally sequenced in this study. Integration of data from liver sinusoidal (27,722 cells) and peripheral blood (19,565 cells) γδ T cells revealed 10 clusters (Fig. 6a, b and Supplementary Figs. 13 and 14), each of which showed transcriptional similarity to one of the original clusters (Fig. 6c).

Fig. 6. CD56^hiCD161⁻ γδ T cells are enriched in the liver sinusoids compared to peripheral blood.

Uniform manifold approximation and projection (UMAP) projections of integrated liver sinusoidal and peripheral blood γδ T cells (total 47,287 cells), colored by clusters (a), and split by peripheral blood mononuclear cells (PBMCs) and liver sinusoidal mononuclear cells (LSMCs) (b). c Heatmap showing transcriptomic similarity using Pearson correlation calculated from Harmony embedding, between integrated clusters and original LSMC clusters (C1–C10). d Frequencies of each γδ T cell clusters in scRNA-seq between PBMCs and LSMCs from matched donors (n = 6 biologically independent donors). Blue circles indicate HCMV-seronegative donors and red circles indicate HCMV-seropositive donors. e Flow cytometry-based comparison of γδ T cell subsets (CD26⁺CD161⁺ (C1), non-CD26⁺CD161⁺ (C2/3), CD27⁺CD57^– (C4), CD27⁻CD57⁺ (C8), CD16⁺CD8⁺ (C9), and CD56^hiCD161^– (C10) cells between PBMC and LSMC (n = 16 biologically independent donors). f Violin plots showing ADT expression of CD69, CD103, and CD49a, and transcript expression of ZNF683 across γδ T cell clusters. Horizontal lines indicate the median expression level. Data are presented as mean ± s.d. Statistics: Two-sided Wilcoxon signed-rank test (d, e). Source data are provided as a Source data file.

We compared the frequency of each cluster between PBMCs and LSMCs on the basis of scRNA-seq data (Fig. 6d). Among the Vγ9Vδ2 T-cell clusters, C1 showed a trend toward enrichment in liver sinusoids compared to in peripheral blood, whereas C2/3 tended to be less abundant in liver sinusoids. When gene expression of C1 was compared between PBMCs and LSMCs, C1 exhibited significantly higher expression of type-3 immunity-associated genes including RORC and CCR6, alongside higher ZBTB16 and KLRB1 expression in LSMCs compared to PBMCs (Supplementary Fig. 15). Among the non-Vγ9Vδ2 T-cell clusters, C5, C6, and C10 were significantly enriched in liver sinusoids compared to in peripheral blood, while C4 was significantly decreased in liver sinusoids relative to peripheral blood.

We confirmed these results by flow cytometry analysis with additional paired PBMCs and LSMCs (Fig. 6e). Among Vγ9Vδ2 T cell subsets, CD26⁺CD161⁺ cells exhibited significant enrichment in LSMCs compared to PBMCs, while non-CD26⁺CD161⁺ cells showed an opposite result. Among non-Vγ9Vδ2 T cell subsets, CD56^hiCD161⁻ cells were exclusively found in LSMCs and nearly absent among PBMCs, while CD27⁺CD57⁻ cells were more enriched in PBMCs than in LSMCs.

To examine the tissue-resident features of γδ T cells, we analyzed the expression of markers associated with tissue residency, including CD69 ADT, CD103 ADT, CD49a ADT, and ZNF683 (which encodes Hobit, a well-known transcription factor associated with tissue residency³³) (Fig. 6f). CD69 was upregulated in several clusters, while CD103 expression was negligible across all clusters, as reported among liver sinusoidal CD8⁺ αβ T cells³⁴. CD49a expression was specifically elevated in C10, aligning with its identification as a marker of tissue-resident memory γδ T cells¹⁷. Furthermore, high ZNF683 expression was observed in C4 and C10. Taken together, these results strongly suggest that C10 is a liver-resident γδ T-cell population.

The CD56^hiCD161⁻ cluster is enriched in the liver compared to other tissues

We further validated the γδ T-cell heterogeneity by analyzing public datasets from various human organs—including liver, jejunum, ileum, lung, and lymphoid organs^10,28,35,36. We integrated our scRNA-seq data from LSMCs and PBMCs with the publicly available data (Fig. 7a, b, Supplementary Fig. 16). Subsequent unsupervised clustering analysis revealed 11 distinct γδ T-cell clusters (PC1 to PC11), which we mapped to our original LSMC clusters based on marker expression profiles (Fig. 7c–e). PC1 corresponded to C1 (type-3 immunity: RORC, IL23R, and DPP4); PC2 to C2 (type-1 immunity: TBX21, CXCR3, and IFNG); PC3/4 to C3 (late-differentiated phenotypes); PC5 to C4 (naïve phenotypes: CD27 and TCF7); PC6 to C5 (tissue-repair signatures; AREG); PC7/8/9 to C6–C9 (cytolytic and effector functions; EOMES and KLRF1); and PC10 to C10 (NK-like features; NCAM1 and KLRC2). PC11 was a gut-resident subset (ITGAE and CCR9), and did not correspond to any of the original LSMC clusters. Tissue distribution analysis was performed to identify the tissue type-specific distribution of each cluster (Fig. 7f, g). The NK-like PC10 was enriched in the liver, PC9 was predominant in the lung, and gut-resident PC11 was enriched in jejunal and ileal tissues. These findings demonstrated the tissue type-specific distribution of γδ T-cell subpopulations.

Fig. 7. CD56^hiCD161⁻ γδ T cells are enriched in the liver compared to other tissues.

Uniform manifold approximation and projection (UMAP) projections of integrated γδ T cells from LSMCs, PBMCs, and public datasets (total 67,544 cells), colored by clusters (PC1–PC11) (a), and by dataset (b). c Heatmap showing transcriptomic similarity using Pearson correlation calculated from Harmony embedding, between integrated clusters (PC1–PC11) and original LSMC clusters (C1–C10). d Dot plot showing average expression of marker genes across for each γδ T cell cluster (PC1–PC11). Dot size represents the percentage of cells expressing each feature, and dot color indicates the scaled average expression level. UMAP projections colored by Vδ chain usage (e) and by tissue of origin from public datasets (f). g Proportional distribution of integrated clusters (PC1–PC11) across tissues. LIV liver, LNG lung, SP spleen, BMA bone marrow aspirate, PB peripheral blood, JEJLP jejunum lamina propria, JEJEPI jejunum epithelium, JEJ jejunum, ILE ileum. Source data are provided as a Source data file.

Next, we performed analyses using publicly available human liver spatial transcriptomics datasets with single-cell resolution (CosMx SMI Human Liver RNA FFPE Dataset). Upon integrating our LSMC scRNA-seq data with the spatial transcriptomics data (total 31,003 cells), unsupervised clustering identified six distinct γδ T-cell clusters in the spatial dataset (Supplementary Fig. 17a). We mapped these clusters to our original LSMC clusters based on marker expression profiles, and found that Cluster A corresponded to C1 (expressing ZBTB16, KLRB1, and DPP4); Cluster B to C7 (expressing TOX, EOMES, and HLA-DRA); Cluster C to C3/C8 (showing high expression of cytotoxic markers TBX21, GZMB, and GNLY); Cluster D to C10 (exhibiting high expression of NK-related genes, including NCAM1, TYROBP, and KLRK1); and Cluster E to C6 (Supplementary Fig. 17b, c). These findings confirm that our perfusion-derived populations are, in fact, present within liver tissues. We further examined whether each cluster exhibited preferential localization to specific hepatic zones—from zone 1 (periportal zone) to zone 3 (pericentral zone). However, none of the γδ T-cell clusters showed any specific zonal enrichment (Supplementary Fig. 17d). Analysis of spatial proximity to various cell types also revealed that all γδ T-cell clusters were distributed with no particular proximity to any cell type (Supplementary Fig. 17e).

The CD56^hiCD161⁻ cluster is a γδ T-cell subpopulation enriched in hepatocellular carcinoma

To examine potential roles of γδ T-cell clusters in HCC, we performed scRNA-seq analysis by integrating our LSMC scRNA-seq data with the publicly available scRNA-seq data obtained from HCC tissues^37–39. Unsupervised clustering analysis identified seven distinct γδ T-cell clusters (A to G), which we have mapped to our originally identified LSMC clusters based on gene expression profiles (Fig. 8a, b and Supplementary Fig. 18). A corresponded to C1 and C2, exhibiting Vγ9Vδ2 features (ZBTB16, and KLRB1); B to C3/C8/C9, exhibiting differentiated phenotypes (GZMB, and GNLY); C to C4, with naïve phenotypes (CD27 and TCF7); D to C5, with tissue-repair signatures (AREG); E to C7 and F to C6 with cytolytic and effector functions (EOMES and KLRF1). Importantly, we found that cluster G corresponded to C10 from LSMC, exhibiting NK-like features (NCAM1 and KLRC2). Notably, clusters C, E, and G were enriched in tumor compared to adjacent normal tissue (Fig. 8c).

Fig. 8. CD56^hiCD161⁻ γδ T cells are enriched in hepatocellular carcinoma and associated with favorable outcomes.

a Uniform manifold approximation and projection (UMAP) projections of integrated γδ T cells from liver sinusoidal mononuclear cells (LSMCs) and public hepatocellular carcinoma (HCC) data (total 32,187 cells), colored by clusters (A–G) and split by normal and tumor tissue. b Heatmap showing transcriptomic similarity using Pearson correlation calculated from Harmony embedding, between integrated clusters (A–G) and original LSMC clusters (C1–C10). c Proportional distribution of integrated clusters (A–G) across normal and tumor tissue. d Frequency of CD107a⁺ cells among CD56^hiCD161^– (C10) cells following co-culture with Huh-7 cells with or without IL-15 stimulation (10 ng/mL) (n = 6 biologically independent donors). e Kaplan–Meier survival curves of 319 HCC patients from The Cancer Genome Atlas, stratified according to C10 signature expression using the xCell algorithm. Data are presented as mean ± s.d. Statistics: Friedman test with Dunn’s multiple-comparisons test (d); two-sided log-rank (Mantel–Cox) test (e). Source data are provided as a Source data file.

We also experimentally validated the functionality of C10 cells against HCC tumor cells. To this end, we performed co-culture experiments using the HCC cell line Huh-7 (Fig. 8d). Co-culture with Huh-7 cells significantly increased the CD107a-degranulating cytotoxic activity of CD56^hiCD161⁻ (C10) cells in the presence of IL-15, although this activity was not observed in the absence of IL-15. These data indicate that IL-15-stimulated C10 γδ T cells exert cytotoxicity against HCC cells without HLA-I restriction, because HLA-mismatched HCC cells could be targeted.

We assessed the clinical relevance of C10, which exhibited liver-resident features and high IL-15-induced NK-like cytotoxicity. Using The Cancer Genome Atlas (TCGA) dataset of hepatocellular carcinoma (HCC), we analyzed the association between transcriptomic features of C10 (Supplementary Table 1) and patient outcomes, noting that the C10 transcriptomic signature does not exclusively represent C10 γδ T cells due to overlapping features with NK cells (Fig. 8e). A total of 319 HCC patients were stratified into C10 signature-high and C10 signature-low groups, based on the expression levels of C10 signature genes. We found that disease-specific survival was significantly better in the C10 signature-high group, compared to the C10 signature-low group (p = 0.0092). These results suggest that the liver-resident C10 population may exert a protective function against HCC progression.

To assess whether the C10 signature showed prognostic relevance beyond HCC, we applied the same signature to multiple non-liver TCGA tumor cohorts—including colon adenocarcinoma, lung adenocarcinoma, kidney renal clear cell carcinoma, and glioblastoma multiforme (Supplementary Fig. 19). Although the non-liver tumor types exhibited significantly higher C10 signature score than HCC (Supplementary Fig. 20), the significant prognostic association observed in HCC was not reproduced in any of these non-liver tumor types, supporting that the prognostic relevance of the C10 signature is specific to liver cancer.

Discussion

Recent advances in single-cell technologies have revolutionized our understanding of immune cell heterogeneity, revealing previously unappreciated functional diversity across various lymphocyte populations, including NK and MAIT cells^40,41. While emerging studies have applied single-cell sequencing to human γδ T cells^9,28, comprehensive analyses of the deeper heterogeneity of γδ T cells remain limited, particularly in the hepatic environment where they are enriched. In the present study, we performed single-cell sequencing with DNA-barcoded antibodies to enable the in-depth characterization of liver sinusoidal γδ T cells. We identified multiple γδ T-cell clusters with diverse differentiation states, some of which were expanded by latent HCMV infection. Moreover, we defined cluster-specific markers and assessed their functional properties.

Vγ9Vδ2 cells are characterized by their semi-invariant TCR and share common features with other unconventional T cells, including MAIT and iNKT cells^6,42. These cells can be activated by IL-12 and IL-18 in a TCR-independent manner, and exhibit strong upregulation of CD161 and the transcription factor PLZF, which is a master regulator of unconventional T cells^3,43,44. Previous studies have explored Vγ9Vδ2 cells’ heterogeneity using markers, such as CD26, CD161, and CCR6^45–47. In the present study, we confirmed these features and further resolved the Vγ9Vδ2 population into three subsets—two type 1 subsets (early-differentiated and late-differentiated) and one type 3 subset—based on differential expression of CD26, CD161, CD27, and CD57. Notably, the type 3 subset was defined by high CD26 and CD161 expressions, exhibited the strongest PLZF expression, and closely resembled MAIT cells, which is in accordance with recent findings⁹.

Several studies have shown that HCMV infection induces the expansion of non-Vγ9Vδ2 γδ T cells, which contribute to protective antiviral responses^23,48–50. In the present study, we observed clonal expansion of effector non-Vγ9Vδ2 γδ T cells in HCMV-seropositive individuals. Moreover, these expanded cells segregated into two distinct clusters: a replicative senescent CD57⁺CD27⁻ population (C8) and a unique liver-resident CD56^hiCD161⁻ population (C10). While both clusters exhibited clonal expansion, C10 exhibited a uniquely restricted TCR repertoire that showed minimal overlap with those of other clusters, highlighting its distinct clonal origin. This divergence suggests that different mechanisms underlie clonal expansion among the non-Vγ9Vδ2 cells. Although TCR ligands for non-Vγ9Vδ2 γδ T cells have been identified⁵¹, it has been suggested that the HCMV-associated expansion of non-Vγ9Vδ2 cells is not driven by HCMV-encoded antigens but rather by infection-induced self-antigens or TCR-independent stimuli^22,52–54. Indeed, we found that CD56^hiCD161⁻ cells among non-Vγ9Vδ2 cells exert cytotoxicity against HCMV-infected cells, which is mediated by the interaction between NKG2D and MIC-A/B ligands. Our results highlight the potential role of non-Vγ9Vδ2 γδ T cells in HCMV immunity, supporting further investigation of their therapeutic applications. There also remains a need to clearly address whether the CD16⁺CD8⁺ (C9) cluster is expanded by HCMV infection, and responds to HCMV-infected cells, as we observed that the C9 cluster tended to be increased in HCMV-seropositive donors.

A recent study investigated hepatic γδ T cells and demonstrated that liver-infiltrating Vδ2^– γδ T cells exhibit enrichment of clonally expanded effector populations, including both circulating (CD27^lo/–CD45RA^hi) and tissue-resident subsets (CD27^lo/–CD45RA^lo)¹⁴. Our study extends this framework by providing higher-resolution dissection of the hepatic γδ T-cell compartment through scRNA-seq with ADTs, and identifying ten transcriptionally distinct clusters, including Vδ2 cells. The present analysis confirmed the predominance of clonally expanded effector cells in liver sinusoids, and further revealed functionally distinct subsets that were not previously distinguished by surface markers alone. Notably, C10 includes a higher frequency of CD27^–CD45RA^– cells, consistent with the previously described tissue-resident subset. The tissue-resident features of C10 were further supported by the expression of markers, including CD49a and ZNF683, in our study.

To further characterize the novel CD56^hiCD161⁻ (C10) γδ T-cell population, we assessed its functional properties. These cells exhibited an NK-like phenotype and robustly responded to IL-15 stimulation, even in the absence of TCR stimulation. Notably, their characteristics closely resemble those of liver sinusoidal CD56^hiCD161⁻ CD8⁺ αβ T cells described in our previous work²⁵, suggesting that the hepatic environment may contribute to the development of these NK-like T cells. Given the abundant IL-15 production in the liver—which is a critical cytokine for the maintenance of innate-like lymphocytes⁵⁵—it is plausible that IL-15 may play key roles in the expansion and effector function of both CD56^hiCD161⁻ non-Vγ9Vδ2 γδ T cells and CD8⁺ αβ T cells. Notably, recent studies have highlighted IL-15 as a central mediator of tissue-resident cytotoxic lymphocytes, including innate-like killer T cells and type 1 innate lymphoid cells, which exert potent anti-tumor responses^56,57. In particular, type 1 innate lymphoid cells are expanded and activated by IL-15 in a TCR-independent manner, and exhibit sustained cytotoxicity without exhaustion⁵⁶. In line with these findings, our present data showed that CD56^hiCD161⁻ γδ T cells exert cytotoxicity against HCC cells, likely through TCR-independent, IL-15-mediated activation and NK-like cytotoxicity. Together, these findings suggest that IL-15 may contribute to coordinating hepatic immunity and to the tumor surveillance of innate-like cytotoxic lymphocytes, including non-Vγ9Vδ2 γδ T cells.

While our single-cell profiling offers substantial depth and resolution, this study also has several limitations. First, we primarily focused on healthy liver, and thus did not fully explore γδ T cells’ heterogeneity in disease contexts, such as liver cirrhosis and cancer. Second, comparisons based on HCMV serostatus are inherently correlative; therefore, additional studies are required to establish causality. Third, our functional validation was limited to selected stimuli, such that other functional properties of γδ T cells may remain uncharacterized.

A recent comprehensive study explored the characteristics of human γδ T cells across diverse tissues and throughout life stages, offering a broad overview of their heterogeneity²⁸. Although our present study was limited to liver sinusoidal γδ T cells, it has unique strengths. Specifically, we analyzed γδ T cells on a larger scale, enabling clear delineation between Vγ9Vδ2 and non-Vγ9Vδ2 cells, and revealing distinct transcriptional heterogeneity within both populations. Additionally, by incorporating DNA-barcoded antibodies targeting CD markers, we extended our analysis beyond transcriptomics, enabling the definition of clusters and investigation of the functional properties of γδ T cells by flow cytometry. Moreover, we examined the TCR repertoire of γδ T-cell clusters in the context of latent HCMV infection.

In conclusion, our present study provides comprehensive insights into the immunological and functional heterogeneity of liver sinusoidal γδ T cells. The identification of liver-resident CD56^hi non-Vγ9Vδ2 γδ T cells that exert IL-15-mediated NK-like functions offers a novel perspective regarding tissue-specific immune surveillance. These findings deepen our understanding of human γδ T-cell biology and serve as a valuable resource for exploring the roles of these cells in health and disease, including cancer and chronic viral infections.

Methods

Human blood and liver sinusoidal samples

Paired peripheral blood and liver perfusate samples were collected from healthy living donors during liver transplantation. Graft livers were perfused with Custodiol HTK solution (Essential Pharmaceuticals) during the bench procedure. Of the total 1000 mL of perfusate, the first 500 mL was discarded, and the second 500 mL was collected and filtered³⁴. From this filtered perfusate, LSMCs were isolated by density gradient centrifugation using lymphocyte separation medium (cat #15-5960-18; Corning). After separation, PBMCs and LSMCs were cryopreserved in fetal bovine serum (FBS) (cat #35-015-CV; Corning) with 10% dimethyl sulfoxide (DMSO) (cat #D8418; Sigma-Aldrich), and stored in a liquid nitrogen storage tank.

For scRNA-seq analysis, samples from 12 donors were used, comprising 6 HCMV-seropositive and 6 HCMV-seronegative individuals. For flow cytometry validation experiments, additional subjects from the cohort were included to validate key scRNA-seq findings with greater statistical power. All donors were healthy living liver transplant donors who underwent rigorous pre-transplant screening without significant comorbidities. Detailed information including sex, age, and HCMV serostatus is provided in Supplementary Table 2. Informed consent was obtained from all participants included in the scRNA-seq and flow cytometry analyses. This study was reviewed and approved by the institutional review board of Severance Hospital (Seoul, Republic of Korea; approval numbers 4-2014-0261 and 4-2018-0777) and was conducted in accordance with the Declaration of Helsinki.

scRNA-seq library preparation and sequencing

From isolated LSMCs, γδ T cells were enriched by depleting CD14⁺ and CD19⁺ cells using microbeads (cat #130-097-052 and cat #130-050-301; Miltenyi Biotec). From the remaining cells, γδ T cells were sorted using a FACSAria Fusion instrument (BD Biosciences), and then subjected to library preparation using the Chromium Next GEM Single Cell 5′ Kit v2 and the 5′ Feature Barcode Kit (10X Genomics). Libraries for transcriptome and ADT profiling were generated following the manufacturer’s protocols. Sequencing was performed on an Illumina HiSeq X platform, targeting ~35,000 reads per cell for gene expression and ~5000 reads per cell for ADTs. A similar pipeline was applied to PBMC sequencing, using the Chromium Next GEM Single Cell 5′ Kit v3, 5′ Feature Barcode Kit, and Human TCR Amplification Kit. Primers for γδ TCRs were designed following a previously reported protocol⁵⁸.

The following TotalSeq antibodies were used for ADT profiling: anti-CD4 (RPA-T4; Biolegend), anti-CD5 (UCHT2; Biolegend), anti-CD8 (SK1; Biolegend), anti-CD10 (HI10a; Biolegend), anti-CD11c (S-HCL-3; Biolegend), anti-CD14 (63D3; Biolegend), anti-CD16 (3G8; Biolegend), anti-CD19 (HIB19; Biolegend), anti-CD20 (2H7; Biolegend), anti-CD25 (BC96; Biolegend), anti-CD26 (BA5b; Biolegend), anti-CD27 (O323; Biolegend), anti-CD28 (CD28.2; Biolegend), anti-CD38 (HIT2; Biolegend), anti-CD44 (BJ18; Biolegend), anti-CD45 (2D1; Biolegend), anti-CD45RA (HI100; Biolegend), anti-CD49a (TS2/7; Biolegend), anti-CD49b (P1E6-C5; Biolegend), anti-CD49d (9F10; Biolegend), anti-CD56 (5.1H11; Biolegend), anti-CD57 (QA17A04; Biolegend), anti-CD62L (DREG-56; Biolegend), anti-CD69 (FN50; Biolegend), anti-CD85j/ILT2 (GHI/75; Biolegend), anti-CD94 (DX22; Biolegend), anti-CD95/Fas (DX2; Biolegend), anti-CD103 (Ber-ACT8; Biolegend), anti-CD107a (H4A3; Biolegend), anti-CD122 (TU27; Biolegend), anti-CD127 (A019D5; Biolegend), anti-CD137/4-1BB (4B4-1; Biolegend), anti-CD152/CTLA-4 (BNI3; Biolegend), anti-CD158/KIR2DL1/S1/S3/S5 (HP-MA4; Biolegend), anti-CD158b/KIR2DL2/L3 (DX27; Biolegend), anti-CD158e1/KIR3DL1 (DX9; Biolegend), anti-CD158f/KIR2DL5 (UP-R1; Biolegend), anti-CD159a/NKG2A (S19004C; Biolegend), anti-CD159c/NKG2C (S19005E; Biolegend), anti-CD161 (HP-3G10; Biolegend), anti-CD178/Fas-L (NOK-1; Biolegend), anti-CD183/CXCR3 (G025H7; Biolegend), anti-CD186/CXCR6 (K041E5; Biolegend), anti-CD195/CCR5 (J418F1; Biolegend), anti-CD197/CCR7 (G043H7; Biolegend), anti-CD218a/IL-18Rα (H44; Biolegend), anti-CD226/DNAM-1 (11A8; Biolegend), anti-CD244/2B4 (C1.7; Biolegend), anti-CD257/BAFF (1D6; Biolegend), anti-CD279/PD-1 (EH12.2H7; Biolegend), anti-CD284/TLR4 (HTA125; Biolegend), anti-CD307c/FcRL3 (H5/FcRL3; Biolegend), anti-CD314/NKG2D (1D11; Biolegend), anti-CD335/NKp46 (9E2; Biolegend), anti-CD336/NKp44 (P44-8; Biolegend), anti-CD337/NKp30 (P30-15; Biolegend), anti-CD360/IL-21R (17A12; Biolegend), anti-CD366/Tim-3 (F38-2E2; Biolegend), anti-TCR α/β (IP26; Biolegend), anti-TCR γ/δ (B1; Biolegend), anti-TCR Vα7.2 (3C10; Biolegend), anti-TCR Vα24-Jα18 (6B11; Biolegend), anti-TCR Vβ13.1 (H131; Biolegend), anti-TCR Vγ9 (B3; Biolegend), anti-TCR Vδ2 (B6; Biolegend), anti-CX3CR1 (K0124E1; Biolegend), anti-HLA-A,B,C (W6/32; Biolegend), anti-HLA-DR (L243; Biolegend), anti-HLA-E (3D12; Biolegend), and anti-TIGIT (A15153G; Biolegend). Catalogue numbers and antibody dilutions are provided in Supplementary Table 3.

Data processing and analysis

Raw sequencing data were processed with Cell Ranger software (v7.0.1). The mkfastq pipeline was used for demultiplexing, and the multi pipeline aligned reads to the GRCh38 human reference genome (refdata-gex-GRCh38-2020-A). Filtered gene expression matrices were analyzed using R (v4.2.2) and the Seurat package (v4.1.1) for each GEM. Donor identity was confirmed using TotalSeq-C anti-human hashtag ADT expression, and vireo & cellsnp-lite for single nucleotide polymorphism (SNP)-based demultiplexing. Cells were excluded if they exhibited mitochondrial gene expression of ≥5% or fewer than 200 detected features. To identify and remove low-quality cells, clustering (PCA dimensions 1–30, resolution: 0.1) was applied through the Seurat standard workflow. Clusters with a low number of detected features, low CD3E expression, or high mitochondrial gene expression were excluded before doublet identification. Doublets were filtered using DoubletFinder (v2.0.3). Dying cells were excluded by removing clusters showing elevated stress or apoptosis-associated genes.

For PBMCs, a similar pipeline was applied with the following tools: Cell Ranger (v9.0.1), GRCh38 human reference genome (refdata-gex-GRCh38-2024-A), R (v4.4.2), Seurat (v5.3.1), and DoubletFinder (v2.0.6). Donor identity was confirmed using vireo & cellsnp-lite with SNPs derived from the LSMC data. Genes not shared between LSMC and PBMC datasets were excluded.

We applied standard Seurat clustering and UMAP-based dimensionality reduction. Differentially expressed genes and cell-type-specific markers were identified using the FindMarkers function, applying the following thresholds: minimum percentage (min.pct) > 0.1 and log2 fold change ≥0.5. Batch effects were corrected using Harmony integration. To assess transcriptional similarity with the original LSMC clusters, Pearson correlation analysis was applied to the Harmony embeddings.

Single-cell TCR analysis

TRUST4 software was used to reconstruct TCR sequences from scRNA-seq BAM files generated during the multi-pipeline alignment. The analysis utilized the provided hg38_bcrtcr.fa and human_IMGT+C.fa reference files, ensuring accurate identification of TCR alpha (TRA) and beta (TRB) chains, as well as gamma (TRG) and delta (TRD) chains. For PBMCs, TCR sequences were aligned from the γδ TCR library through the Cell Ranger multi pipeline. The VDJ reference file (refdata-cellranger-vdj-GRCh38-alts-ensembl-7.1.0) was downloaded from the 10x Genomics website (https://www.10xgenomics.com/support/software/cell-ranger/downloads#reference-downloads). Due to the low number of filtered contigs, all contigs were retained and matched to the gene expression data.

Differentiation state analysis

Differentiation states were inferred using CytoTRACE2, an unsupervised algorithm that ranks cells along a continuum from less-differentiated to more-differentiated phenotypes based on transcriptional diversity and gene expression patterns. Raw log-normalized RNA expression matrices from Seurat were converted to a cell-by-gene format and analyzed using the cytotrace2 function implemented in the CytoTRACE2 R package, with default parameters (species = “human”). CytoTRACE2 generates a normalized relative score (range 0–1), with higher values indicating transcriptionally less-differentiated stem-like states and lower values representing more-differentiated states. The predicted scores were mapped onto UMAP embeddings to visualize the spatial gradients of differentiation.

Survival analysis of C10 signature across TCGA tumor cohorts

C10 signature genes were identified using the FindMarkers function with thresholds set at a minimum expression percentage of 0.2 and a log₂ fold change ≥1. C10 signature scores for the TCGA HCC dataset were calculated using the GSVA method, based on data retrieved from the UCSC Xena platform. Patients were stratified into signature-high and signature-low groups based on the median C10 signature score. Survival analysis was performed using the log-rank test and visualized as Kaplan–Meier curves. To evaluate the tumor specificity of the C10 prognostic signature, the same analysis was applied to additional TCGA tumor cohorts, including colon adenocarcinoma, lung adenocarcinoma, kidney renal clear cell carcinoma, and glioblastoma multiforme.

Analysis of public scRNA-seq datasets

For multi-organ integration, public multi-organ scRNA-seq datasets of γδ T cells were integrated with our LSMC and PBMC data using the Seurat pipeline, incorporating four public datasets^10,28,35,36. Duplicate cells were removed to avoid redundancy. In datasets lacking explicit γδ T-cell annotation, γδ T cells were manually identified via clustering with Harmony batch correction across donors and filtering based on TCR and CD3 gene expression. Only genes shared across all datasets were retained for integration. Batch correction was performed using Harmony (dimensions 1–10, resolution = 0.5), aligning transcriptional profiles while excluding TCR genes such as TRDV1 and TRDV2 to avoid potential bias. Principal components were calculated using non-TCR genes shared across datasets, enabling identification of conserved and tissue-specific features of γδ T cells. Pearson correlation analysis of Harmony embeddings was used to assess transcriptional similarity with the original LSMC clusters.

For spatial transcriptomic integration, a NanoString CosMx spatial single-cell transcriptomic dataset of healthy liver (https://nanostring.com/products/cosmx-spatial-molecular-imager/ffpe-dataset/human-liver-rna-ffpe-dataset/) was integrated with our LSMC dataset using the Seurat pipeline. γδ T cells were pre-annotated by the data providers. Only genes shared between our dataset and the public dataset were retained for integration. TCR genes were absent from the spatial dataset. SCTransform was applied to correct inherent differences between sequencing and nucleic acid hybridization techniques. Batch correction was performed using Harmony. Transcriptional similarity with the original LSMC clusters was assessed using Pearson correlation analysis of Harmony embeddings, as described above.

For HCC integration, public HCC scRNA-seq datasets were integrated with our LSMC dataset using the Seurat pipeline, incorporating three public datasets^37–39. γδ T cells were manually identified through clustering and filtering based on TCR and CD3 gene expression. Only genes shared across all datasets were retained for integration. Batch correction was performed using Harmony, aligning transcriptional profiles while excluding TCR genes such as TRDV1 and TRDV2 to avoid potential bias. Transcriptional similarity with the original LSMC clusters was assessed using Pearson correlation analysis of Harmony embeddings, as described above.

Flow cytometry

Cryopreserved LSMCs or PBMCs were thawed, and then stained with fluorochrome-conjugated antibodies targeting surface markers for 15 min at room temperature. Dead cells were excluded using LIVE/DEAD fixable viability dye (cat #L23102; Invitrogen). After surface staining, cells were fixed and permeabilized using a FoxP3 staining buffer kit (cat #88-8824-00; Invitrogen), and then stained with intracellular markers for 30 min at 4 °C. Multicolor flow cytometry was performed using FACSymphony A3 instrument (BD Biosciences), and data were analyzed with FlowJo software (Treestar).

Antibodies for flow cytometry

Flow cytometry analysis was performed using the following antibodies: anti-CD3 (UCHT1; BD Biosciences), anti-CD8 (SK1; BD Biosciences), anti-CD14 (MφP9; BD Biosciences), anti-CD16 (3G8; BD Biosciences), anti-CD19 (HIB19; BD Biosciences), anti-CD26 (BA5b; Biolegend), anti-CD27 (M-T271; BD Biosciences), anti-CD56 (NCAM16.2; BD Biosciences), anti-CD57 (NK-1; BD Biosciences), anti-CD107a (H4A3; BD Biosciences), anti-CD122 (TU27; Biolegend), anti-CD161 (HP-3G10; Biolegend), anti-CD218a/IL-18Rα (H44; Biolegend), anti-CD314/NKG2D (1D11; Biolegend), anti-TCR γδ (11F2; BD Biosciences), anti-TCR Vδ1 (REA173; Miltenyi Biotec), anti-TCR Vδ2 (B6; BD Biosciences), anti-TCR Vα7.2 (3C10; Biolegend), anti-TCR Vα24 (6B11; Biolegend), anti-IFN-γ (4S.B3; BD Biosciences), anti-TNF (MAb11; BD Biosciences), anti-IL-17A (BL168; Biolegend), anti-Helios (22F6; Biolegend), and anti-PLZF (R17-809; R&D Systems). Catalogue numbers and antibody dilutions are provided in Supplementary Table 3.

In vitro stimulation

To evaluate T-cell activation, thawed LSMCs were stimulated with Dynabeads Human T-Activator CD3/CD28 (bead-to-cell ratio: 1:25; cat #11131D; Gibco) for direct TCR activation, or with PMA (10 ng/mL; cat #P1585; Sigma-Aldrich) and ionomycin (500 ng/mL; cat #I3909; Sigma-Aldrich) to assess maximal activation potential. Cultures were incubated at 37 °C for 6 h. After the first hour of stimulation, brefeldin A (cat #555029; BD Biosciences) and monensin (cat #554724; BD Biosciences) were added, and their presence was maintained for the remaining 5 h.

For intracellular cytokine staining (ICS) following IL-12 and IL-18 stimulation, cells were treated with recombinant human IL-12 (50 ng/mL; cat #200-12; PeproTech) and IL-18 (50 ng/mL; cat #9124-IL; R&D Systems) at 37 °C for 18 h. Brefeldin A and monensin were added during the final 5 h of incubation. For ICS following IL-15 stimulation, cells were cultured with recombinant human IL-15 (10 ng/mL; cat #200-15; PeproTech) at 37 °C for 36 h, with the addition of brefeldin A and monensin after the first 24 h. All ICS assays were performed with a minimum of 200,000 total PBMCs and LSMCs per condition. For NKG2D expression and cytotoxicity-associated molecule analysis, cells were cultured with IL-15 at 37 °C for 48 h.

For IL-17A production analysis, LSMCs and matched PBMCs from the same donors were stimulated with PMA (10 ng/mL; Sigma-Aldrich) and ionomycin (500 ng/mL; Sigma-Aldrich) for 6 h at 37 °C. Brefeldin A and monensin were added after the first hour and maintained for the remaining 5 h. This protocol enables TCR-independent assessment of the intrinsic IL-17A production capacity across γδ T cell subsets. This paired design allowed direct comparison of type-3 functional capacity between peripheral blood and liver sinusoidal compartments. MAIT cells were simultaneously analyzed under the same stimulation conditions as a positive control for IL-17A production.

CD107a degranulation assay against target cells

To evaluate the degranulation potential of liver sinusoidal γδ T-cell subpopulations, LSMCs were co-cultured with K562 cells (cat #CCL-243; ATCC) or Huh-7 cells (cat #60104; KCLB) as target cells at an effector-to-target ratio of 1:1 for 12 h. Anti-CD107a antibody was added at the initiation of co-culture. After 1 h, brefeldin A and monensin were added, and the cells were further incubated for 11 h. Co-cultures with K562 cells were performed in RPMI 1640 medium (cat #10-040-CVRC; Corning) supplemented with 10% FBS (cat #35-015-CV; Corning) and 1% penicillin–streptomycin (cat #LS202-02; WELGENE), whereas co-cultures with Huh-7 cells were performed in DMEM supplemented with 10% FBS and 1% penicillin–streptomycin. Following co-culture, cells were washed and stained with fluorochrome-conjugated antibodies against the indicated surface markers for 15 min at room temperature in the dark.

HCMV infection and co-culture for CD107a degranulation

Human neonatal primary dermal fibroblast cells (HDFn; cat #PCS-201-010, ATCC) were maintained in Dulbecco’s Modified Eagle Medium (DMEM; Corning) supplemented with 10% fetal bovine serum (FBS) (cat #35-015-CV; Corning), and 1% penicillin–streptomycin (cat #LS202-02; WELGENE). HDFn cells carry following HLA-I alleles: HLA-A*01:01, HLA-B*27:02, HLA-B*37:01, HLA-C*02:02, and HLA-C*06:02.

HCMV strain JHC, a previously described clinical isolate from bone marrow transplant patient⁵⁹ was provided by Prof. Jin-Hyun Ahn (Sungkyunkwan University, South Korea). For HCMV infection, HDFn cells were seeded in 96-well flat-bottom plates at a density of 500,000 cells/ml 12 h before infection. Cells were then mock-infected or infected with HCMV at a multiplicity of infection (MOI) of 1 in complete DMEM. After overnight incubation, the virus-containing medium was replaced with fresh complete DMEM, and the cells were maintained for an additional 5 days. The medium was changed every 2 days, and cells were monitored daily by microscopy for cytopathic effects characterized by enlarged cell size, enlarged nuclei, and rounded morphology. Wells in which 50–100% of cells showed cytopathic effects were co-cultured with LSMCs for CD107a degranulation assays as described above. LSMCs carrying fully or partially matched (n = 10) or mismatched (n = 6) HLA-I alleles were selected for the co-culture. For blocking assay, anti-HLA-A/B/C antibody (10 μg/mL; cat #311428; Biolegend) and anti-MIC-A/B antibody (10 μg/mL; cat #320902; Biolegend) were used. Additional information is provided in Supplementary Table 3.

Cytotoxicity assay

γδ T cells, CD8⁺ αβ T cells, and NK cells were sorted from thawed LSMCs using a FACSAria Fusion instrument (BD Biosciences), and were then cultured, with or without IL-15 (10 ng/mL), for 48 h prior to the cytotoxicity assay. K562 target cells were labeled with PKH26 (cat #PKH26GL; Sigma-Aldrich), and co-cultured with γδ T-cell effector cells at a 20:1 effector-to-target ratio, in 96-well plates for 12 h. After co-culture, TO-PRO-3-iodide (0.5 μM; cat #T3605; Invitrogen) was added, and the cells were immediately analyzed by flow cytometry. To determine the background and maximum TO-PRO-3-iodide staining, target cells were incubated with medium and detergent, respectively. Specific cytotoxicity (%) was calculated as follows: (TO-PRO-3-iodide⁺PKH26⁺ cells in co-culture − background)/(maximum − background) × 100.

Statistical analysis

Statistical analyses were conducted using GraphPad Prism (version 8.4; GraphPad Software). Two-sided Wilcoxon signed-rank tests were used for comparisons between two paired groups, and two-sided Mann–Whitney U tests were used for comparisons between two unpaired groups. Friedman tests followed by Dunn’s multiple-comparisons tests were used for comparisons among more than two conditions. Two-way repeated-measures analyses of variance (ANOVA) followed by Šídák’s multiple-comparisons tests were used for repeated comparisons involving two factors. A p value of <0.05 was considered statistically significant.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Author contributions

D.U.K. and E.C.S. designed the study. J.Y.P., D.J.J., H.J.P., J.G.L., D.G.K., E.K.M., and M.S.K. acquired and provided human clinical samples and information. D.U.K, B.H.M., S.K., and Y.J.K. performed experiments and collected data. D.U.K, B.H.M., S.K., Y.J.K, S.H.P., and E.C.S. were involved in data analysis and interpretation. D.U.K, B.H.M., Y.J.K., and S.K. performed statistical analysis. D.U.K and E.C.S. drafted the manuscript, with comments from all authors.

Peer review

Peer review information

Nature Communications thanks Emile Voest, who co-reviewed with Miguel Parra-Martínez and Allard van Renterghem; Li Kai TAN and the other anonymous reviewer(s) for their contribution to the peer review of this work. A peer review file is available.

Funding

This work was supported by grants from the National Research Foundation of Korea (NRF) funded by the Korean government (MSIT) (RS-2024-00439160). This work was also supported by a grant from the Ministry of Food and Drug Safety (RS-2025-02213409) and by a grant from the Korea Health Technology R&D Project through the Korea Health Industry Development Institute (KHIDI), funded by the Ministry of Health & Welfare, Republic of Korea (RS-2025-25460003).

Data availability

All single-cell RNA-sequencing data generated in this study have been deposited in the GEO database under the GEO Series accession number GSE301771. Further information and requests for resources should be directed to and will be fulfilled by the Lead Contact, E.-C.S. (ecshin@kaist.ac.kr). Source data are provided with this paper.

Code availability

All the custom codes used in this study have been deposited in Zenodo under https://doi.org/10.5281/zenodo.21720285⁶⁰.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at https://doi.org/10.1038/s41467-026-77255-z.