Spatial and functional specialization of human splenic innate lymphoid cells

Abstract

Secondary lymphoid organs, including the spleen, integrate innate and adaptive immune responses. Studies in murine models and humans highlight Natural Killer (NK) cells and other innate lymphoid cells (ILC) as key organizers of splenic immune activity, yet their in situ organization and interactions within immune networks are incompletely understood. Here, we report a spatially resolved single-cell atlas of human splenic NK cells and ILCs. We identify CD39⁺TIGIT⁺CD56^bright NK cells that localize with alternatively activated macrophages outside of B cell follicles. Splenic ILC2s, enriched in vascular-associated regions, express CD40L and can promote B cell survival and IgA secretion. Finally, we define three spatially and functionally distinct ILC3 subsets, namely KLF2⁺ progenitor-like ILC3s, follicle-associated CD38⁺ ILC3s, and IFNGR1⁺ ILC3s, that can promote B cell proliferation, IL-10 production, and IgG class switching. This spatially resolved atlas provides a comprehensive reference of human splenic ILC organization, specialization, and interaction networks.

To date, the spatial and functional characterization of human splenic innate lymphoid cells (ILC) has remained limited. Here, the authors integrate CITE-seq and spatial transcriptomic data and present an atlas of human splenic NK and ILC cells and define high-resolution subset-specific spatial organization and interaction networks, providing insights into the presence of local immune niches in secondary lymphoid organs.

Introduction

Innate lymphoid cells (ILC), including Natural Killer (NK) cells, ILC1s, ILC2s, and ILC3s, acquire tissue-specific and memory-like properties^1,2, and modulate both innate and adaptive immune cell responses, including B and T cell activity^3–7. Murine studies have identified key roles in immune regulation; for example, NK cells in mice traffic from the red pulp (RP) to the white pulp (WP) to eliminate activated CD4⁺ T cells following LCMV infection, in turn reducing antigen-specific B cell responses^5,8. Although transcriptional diversity of human splenic NK cells has been observed⁹, their anatomical localization and contributions to lymphoid tissue immune networks remain poorly understood.

The functions of other ILC subsets in secondary lymphoid organs are also still emerging. Foundational murine studies identified lymphoid tissue inducer (LTi)-like ILC3s as key regulators of splenic architecture, contributing to B and T cell compartmentalization via LT-α, LT-β, and TNF-α³. More recently, additional ILC3 subsets have been implicated in shaping local immunity and influencing tissue repair or organization^3,10–14; for example, marginal zone (MZ) ILC3s were shown to produce BAFF, CD40L, and the Notch ligand Delta-like 1 (DLL1) to support B cell isotype switching¹⁵, and lymphotoxin (LT)α1β2-expressing ILC3s were shown to coordinate with B cells to maintain conventional dendritic cell (cDC) populations¹⁶. Whether these immunoregulatory functions are performed by a single ILC3 population or by functionally and spatially specialized populations has not been discerned. Moreover, the extent to which splenic ILC3s diverge phenotypically and functionally from their circulating or tissue-resident counterparts in other organs has yet to be defined. While recent efforts have mapped ILC distribution across lymphoid tissues¹⁷, a complementary framework integrating spatial localization, protein expression, and functional capacity is essential to fully define how ILC subsets contribute to immune architecture and regulation.

Here, we integrate Cellular Indexing of Transcriptomes and Epitopes by Sequencing (CITE-seq) and spatial transcriptomics to resolve the phenotypic heterogeneity, tissue-adapted programs, and in situ spatial organization of human ILCs within splenic cellular networks. We define unique features of functionally distinct splenic ILC populations and demonstrate their specialized roles in regulating B cell proliferation, differentiation, and antibody production. Predicted cell-cell interactions were validated in co-culture assays, confirming the immunomodulatory capacity of splenic ILC2 and ILC3 populations. Together, these findings establish a high-resolution, spatially resolved atlas of human splenic ILCs and highlight potential mechanisms through which ILC populations may influence immune responses within this critical immune organ.

Results

CITE-seq defines NK cell heterogeneity and diversity across spleen and peripheral blood mononuclear cells

To examine human ILC phenotypes in peripheral blood and spleen, we performed CITE-seq on samples from seven normal spleens obtained from deceased donors and three healthy Peripheral Blood Mononuclear Cell (PBMC) donors (Supplementary Fig. 1a). To enrich for rare ILC and NK cell populations, four splenic donors underwent sorting for NK cells and ILCs prior to sequencing, while three were analyzed as unsorted total splenocytes (Supplementary Fig. 1b). For comparison of tissue-specific versus circulating ILC phenotypes NK cells, other ILCs, and T cells from blood were sequenced and merged for downstream analysis (Supplementary Fig. 1a, c). Cell populations were manually annotated based on protein markers and transcriptomic signatures (Supplementary Fig. 1d–g). ILCs were subclustered to resolve population heterogeneity, with CD56^bright and CD56^dim NK cells distinguished by CD56 and CD16 expression, while ILCs were delineated by high IL7R/CD127 (Supplementary Fig. 1h–k). ILC2s were distinguished by PTGDR2/CRTH2 and ICOS/ICOS, and ILC3s/ILC precursors (ILCp) by KIT/CD117 (Supplementary Fig. 1h–k). ILC2s and CD56^dim NK cells were enriched in PBMCs, while ILCp and ILC3s were more abundant in the spleen (Supplementary Fig. 1l–m).

Given their abundance in both blood and spleen and their known functional diversity and unique tissue-specific phenotypes^18–20, we first focused on defining the phenotypic heterogeneity of splenic NK cells and then performed flow cytometry as validation (Supplementary Fig. 2). To analyze the unique features of NK cells within human spleen, splenic and PBMC-derived NK cells were subclustered based on transcriptomics and surface protein signatures (Supplementary Fig. 3a–e). We identified five CD56^bright, two CD56^dimCD16^lo and, four CD56^dim NK cell clusters in spleen (Fig. 1a). A subset of splenic CD56^dimCD16^lo NK cells was enriched for stress-induced heat shock genes (HSPA1A and HSPA1B), likely reflecting cryopreservation artifacts (Supplementary Fig. 3f). Four CD56^bright and nine CD56^dim NK cell clusters were characterized in PBMC (Fig. 1b). To identify subset-specific NK cell differences and broadly characterize populations, we identified differentially expressed genes (DEGs) and proteins and created a supervised list (Fig. 1c). Analysis of NK cell receptor expression revealed splenic CD56^bright NK cells co-expressed NKG2D, CD94, and NKp46, while CD56^dim NK cells exhibited higher expression of NKp44 (Fig. 1d–f).

Fig. 1. Human splenic and PBMC-derived NK cells display phenotypic differences.

Human NK cells from spleen and PBMC CITE-seq datasets were subclustered and analyzed for differential gene and protein expression. a Clustering of splenic NK cells. b Clustering of PBMC NK cells. c Heatmap of select top differentially expressed genes and proteins of spleen and PBMC NK cells. d NKG2D, CD94, NKp44 and NKp46 protein expression on spleen and PBMC CD56^bright NK cells. e Representative flow cytometry gating of PBMC and spleen expression of CD94 and NKG2D in CD56^bright and CD56^dim NK cells. f Percent positive of NKG2D⁺CD94⁺, NKp44 and NKp46 across spleen (n = 10) and PBMC (n = 5) CD56^bright and CD56^dim NK cells. g Similarity scores of spleen and PBMC CD56^bright and CD56^dim NK cells. h CD56^bright NK cells from spleen and PBMC were subclustered and analyzed for unique subclusters and differentially expressed markers. i Dotplot of differentially expressed markers. j Percent positive of CD39 and CD69 determined by flow cytometry across spleen (n = 10) and PBMC (n = 5) CD56^bright and CD56^dim NK cells. Statistical significance was determined by the Kruskal-Wallis test. Error bars represent standard deviation (SD).

Splenic CD56^dim NK cells highly expressed CD16, NKG7, CX3CR1, PRF1, and GZMB, in line with circulating CD56^dim NK cells (Fig.1c, Supplementary Fig. 3c–e). In contrast, CD56^bright NK cells displayed a transcriptionally distinct profile with high expression of Fas, CD69, CMC1 and CD39, among others (Fig. 1c, Supplementary Fig. 3c–e), and a lower similarity score to PBMC CD56^bright NK cells (Fig. 1g). Gene Set Enrichment Analysis (GSEA) revealed that splenic CD56^bright NK cells are enriched for pathways related to translation and amino acid metabolism (Supplementary Fig. 3g), consistent with a more activated transcriptional state^21,22. Subclustering of CD56^bright NK cells to achieve greater phenotypic resolution indicated that splenic CD56^bright NK cells expressed markers of tissue residency (CD69 and CXCR6), immune regulation (CD39 and TIGIT), and chemokines, including CCL3 and CCL4 (Fig. 1h–j, Supplementary Fig. 3h), further differentiating them from their circulating counterparts. Although largely phenotypically similar to circulating populations, splenic CD56^dim NK cells showed higher chemokine and prostaglandin D2 synthase (PTGDS) expression (Supplementary Fig. 3i). Together, these findings uncovered unique features of splenic human NK cell subsets, in particular highlighting tissue-adapted features of splenic CD56^bright NK cells and potential immunomodulatory functions.

Spatial distribution and cellular interactions of splenic NK cells

The spleen is essential for immune surveillance and initiating adaptive immune responses through its organized microarchitecture^23–26 (Fig. 2a). In the white pulp (WP), B and T cell zones are distinct, with B cells clustering into follicles, activating in response to cognate antigens, and relying on accessory cells for survival^23–25. The red pulp (RP) houses macrophages and effector cells that clear aged or opsonized cells, and survey for foreign pathogens^23–25.

Fig. 2. Splenic CD56^bright NK cells preferentially interact with alternatively activated myeloid cells, while CD56^dim NK cells cluster with other cytotoxic NK cells.

Spatial transcriptomic analysis was performed on the Xenium platform to assess the spatial context of NK cells within the splenic architecture. a Schematic image of the red pulp (RP) and white pulp (WP) zones of the spleen. Created in BioRender. Crome, S. (2026) BioRender.com/zq40x95. b Xenium image of spleen showing cell types across all zones. c Proportions of cell types in the WP and RP zones. d Proportion of CD56^bright and CD56^dim NK cells in RP and WP. e CD56^dim closest neighbors in RP. f CD56^dim expected versus observed interactions with other CD56^dim NK cells in RP and WP. g Source-target predicted interactions between CD56^dim NK cells and CD56^dim NK cells. h Expression of SPON2 and SELPLG in PBMC and spleen CD56^dim NK cells. i PMA and Ionomycin stimulated spleen (n = 10) and PBMC (n = 5) CD56^dim NK cells were assessed for intracellular cytokine production of granzyme B, perforin, granzyme K, and IFN-γ by flow cytometry. The boxes represent the interquartile range (25th–75th percentiles), the center line indicates the median, and whiskers extend to the minimum and maximum values. j Xenium image of CD56^dim NK cell in clusters in WP and RP with gene expression of GZMK, GZMB, IFNG and PRF1. Scale bars represent 50 μm. k CD56^bright closest neighbors in the WP. l CD56^bright expected versus observed interactions with myeloid and B cells in WP. m Source-target predicted interactions between CD56^bright NK cells and myeloid cells. n Expression of TIGIT and TNFSF12 in PBMC and spleen CD56^bright NK cells. o Expression of CCL4, CCL5 and CCL3 in PBMC and spleen CD56^bright NK cells. p Xenium image of CD56^bright NK cells and myeloid cells in WP with gene expression of TIGIT, CD163, IFNG and IL10. Scale bars represent 200 μm (top) and 50 μm (bottom). Statistical significance was determined by two-tailed Mann-Whitney tests.

Using the spatial transcriptomics platform Xenium, we assessed NK cells and ILC subset (discussed later) localization and interactions in human spleen tissue (Fig. 2b, c). After manually annotating and quantifying cell types (Supplementary Fig. 4a, b), the WP and RP areas could be clearly visualized (Fig. 2b), as the WP was marked by a high density of B cells and CD4⁺ T cells, while the RP contained a mixture of myeloid cells, NK cells, DCs, and CD8⁺ T cells (Fig. 2b, c). NK cell subsets identified by CITE-seq were mapped spatially using markers such as EOMES, CD69, and GZMK for CD56^bright NK cells and FCGR3A and PRF1 for CD56^dim NK cells (Supplementary Fig. 4c, d). Overall, NK cells were more abundant in the RP than WP, and both regions had a greater proportion of CD56^dim than CD56^bright NK cells (Fig. 2c, d).

NK cells survey and target virally infected cells in the RP^5,18,27–30, but also regulate immunity by restricting viral infection-induced germinal center reactions via perforin-dependent targeting of CD4⁺ T cells⁸, impacting antiviral CD8⁺ T cell activity³⁰. To assess NK cell interactions, we employed a centroid-based nearest-neighbor analysis³¹ and a spatial interaction model comparing observed to expected values³². CD56^dim NK cells were most frequently adjacent to B cells in the WP and myeloid cells in the RP (Fig. 2e, Supplementary Fig. 4e, f). Notably, CD56^dim NK cells in both regions displayed higher than predicted self-aggregation (Fig. 2f). Predicted source-target interactions³³ performed on the CITE-seq data showed that heightened self-interactions were mediated via GZMA/GZMB and lymphocyte-recruiting signals via SPON2³⁴ and SELPLG³⁵ (Fig. 2g, h). Additional interactions with B and T cells, mediated by IFNG, IL-16 and CD99 (Supplementary Fig. 4f, g), support broad interactions with other immune lineages.

To examine functional differences, we compared granzyme and cytokine profiles of splenic versus circulating NK cells. Splenic CD56^dim NK cells showed reduced perforin and significantly lower granzyme B but elevated granzyme K, and a trend towards increased IFN-γ (Fig. 2i), with granzyme K having known roles in splenic viral clearance³⁶. Spatial mapping revealed self-aggregating CD56^dim NK cells with GZMK, GZMB, IFNG, and PRF1 transcripts, suggesting coordination of effector cell clustering and recruitment^37,38 (Fig. 2j). These data are consistent with a role for splenic CD56^dim NK cells in local immune responses, potentially involving recruitment or priming of effector NK cells and using distinct cytotoxic pathways. Further studies are needed to directly test this hypothesis.

CD56^bright NK cells comprised 2.56% of WP cells versus 5.0% in RP (Fig. 2c), indicating potential compartment-specific differences in distribution. Spatial analysis showed that CD56^bright NK cells most frequently neighbored B cells (26.89% of cellular neighbors) (Fig. 2k), which likely reflects the high abundance of B cells in these areas (Fig. 2l). However, they did display a higher-than-expected rate of interaction with myeloid cells (Fig. 2l). Predicted ligand–receptor analyses identified chemokines (CCL4, CCL5 and CCL3), and checkpoint molecules (TIGIT and the TNF-related weak inducer of apoptosis (TWEAK) transcript, TNFSF12) as potential signals from CD56^bright NK cells to myeloid receptors (CCR1, CD163 and NECTIN2) (Fig. 2m–o). Prior studies have reported that TIGIT-NECTIN2 engagement inhibits NK cell activation^39–41, and TIGIT deletion on NK cells enhances cytotoxicity⁴², suggesting an immunomodulatory loop. Other predicted interactions included VEGFA with stromal cells and TNFSF10 (TRAIL) with B cells (Supplementary Fig. 4h, i).

Ex vivo stimulation assays showed CD56^bright NK cells can produce granzyme B, perforin, granzyme K, and IFN-γ at levels comparable to PBMC-derived CD56^bright NK cells (Supplementary Fig. 4j); however, gene module scores related to cytotoxicity⁴³, tissue residency⁴⁴ and myeloid-related suppression^45–47 showed splenic CD56^bright NK cells had the lowest cytotoxicity score, coupled with the highest scores for tissue residency and myeloid-related suppression (Supplementary Fig. 4k).

Consistent with this observation, spatial mapping placed CD56^bright NK cells surrounding but not within B cell follicles, in proximity to CD163⁺IL-10⁺ myeloid cells lacking detectable IFNG transcripts (Fig. 2p). These spatial and transcriptional data are compatible with the possibility that alternatively activated myeloid cells may influence CD56^bright NK cell activation states in the spleen. However, future functional studies will be required to directly test this possibility, particularly in inflammatory contexts.

Distinct splenic ILC2 phenotype and spatial localization

We next examined other ILCs present within the human spleen. Although ILC1s were undetected in our dataset, a small proportion of splenic ILCs were ILC2s, which, to our knowledge, have not been previously detected, in addition to ILC3s (Supplementary Fig. 1m, Fig. 3a–c). To investigate ILC2s' tissue-specific features, we compared splenic ILC2s with circulating ILC2s from blood. ILC2s from PBMCs segregated into CD49a⁺ or CD49a^- subsets (Fig. 3d–f). Transcriptomic and proteomic profiling revealed that splenic ILC2s differed by 299 transcripts and proteins from circulating ILC2s, while sharing 131 features in common. Notably splenic ILC2s showed increased expression of CRTh2, IL32, CD25, RORA, IL10RA, and CD49d (Fig. 3g), suggesting a more activated, immunomodulatory phenotype^48,49. Further, splenic ILC2s displayed higher expression of tissue residency and activation markers (CD69/CD69, AREG and CXCR4)^50,51 and expressed checkpoint molecules TIM3⁵² and ICOS⁵³ (Fig. 3h). GSEA indicated an enrichment for pathways related to prostaglandin synthesis and metabolism, as well as epithelial structure maintenance in comparison to other splenic ILCs (Fig. 3i).

Fig. 3. Splenic ILC2s are enriched for markers associated with prostaglandin synthesis in comparison to PBMC ILC2s.

ILC2s from spleen and PBMC were subclustered and analyzed for differential gene and protein expression. a Cell types of splenic ILCs. b Clusters of splenic ILCs. c ILC2-defining surface protein expression in spleen. d Cell types in PBMC ILCs. e Clusters of PBMC ILCs. f ILC2-defining surface protein expression in PBMC. g Shared gene and protein expression of splenic, PBMC CD49a⁻ and PBMC CD49a⁺ ILC2s. h Volcano plot of differentially expressed genes and proteins in splenic and PBMC ILC2s. i Gene set enrichment analysis of splenic ILC2s relative to splenic ILC3s. j ILC2 expression of CRTh2 (protein) and PTGDR2 (gene). Statistical significance was determined by a two-tailed Wilcoxon rank sum test (h).

Indeed, splenic ILC2s had higher expression of the prostaglandin receptor, CRTh2/PTGDR2, compared to PBMC(Fig. 3j). While enrichment in epithelial maintenance pathways was observed, we interpret this as likely reflecting ILC2 involvement in extracellular matrix and adhesion biology, rather than direct epithelial repair. Together, these findings indicate that splenic ILC2s have elevated tissue residency markers, with enhanced activation, immunomodulatory and prostaglandin-related pathways.

Building on these phenotypic insights, we next examined the spatial localization of ILC2s within the human spleen. Spatial transcriptomic analysis revealed that ILC2s were preferentially localized near blood vessels (Fig. 4a), where transitional B cells enter the spleen⁵⁴. Comprising just 0.03% of total cells within vasculature-associated areas (Fig. 4b), ILC2s were found most proximal to B and CD4⁺ T cells (Fig. 4c, d). Source-target analysis indicated ILC2s express LTB, CD40LG and IL-2, while B cells express the CD40 and TRAF3 receptors, and T cells express IL-2 receptors, IL2RA, IL2RB and IL2RG (Fig. 4e–h, Supplementary Fig. 5a). Spatial co-localization further confirmed CD40LG and LTB in ILC2s near B cells expressing CD40 (Fig. 4i).

Fig. 4. ILC2s are located in close proximity to vascular regions in the spleen and interact with B cells through CD40L to induce class switching to IgG3, IgA and IgD subtypes.

Spatial transcriptomic analysis was performed on the Xenium platform to assess the spatial context of ILC2s within the splenic architecture. a ILC2s in close proximity to splenic blood vessels as determined by spatial transcriptomics. Scale bars represent 100 μm. b Proportions of cell types represented in vascular regions. c ILC2 closest neighbors. d ILC2 expected versus observed interactions with B cells, CD4⁺ T cells and myeloid cells. e Source-target predicted interactions between ILC2s and B cells or CD4⁺ T cells. f CD40LG expression. g TRAF3 expression. h CD40 expression. i Spatial transcriptomic images of ILC2s in close proximity with B cells and CD4⁺ T cells in vascular regions showing gene expression of CD40LG, LTB, and CD40. Scale bars represent 100 μm. j–n Naïve B cells and ILC2s were sorted from human spleen and co-cultured for 7 days. Cultures were assessed for B-cell expansion and secreted antibodies and proteins. j Overview of experimental set-up of ILC2-B cell co-cultures. ILC2 cytokines (IL-2, IL-7, IL-33) were added to maintain ILC2 viability. Created in BioRender. Crome, S. (2026) BioRender.com/2588n5y. k Fold change of secreted CD40L and IL-10 in co-cultures (n = 3). l Fold expansion of B cells after co-culture with ILC2s (n = 4). m Concentrations of IgG3 (n = 5), IgA (n = 4) and IgD (n = 5) measured in supernatants. n Fold change of IgG3 (n = 5), IgA (n = 4) and IgD (n = 5) measured in supernatants. Statistical significance was determined by a one-tailed Wilcoxon matched-pairs signed-rank test (k) and a Friedman test with Dunn’s multiple comparisons (l–n). The number of replicates (n) represents biological replicates (donors). Error bars represent standard deviation (SD).

To assess whether these interactions would functionally impact B cell activity, we co-cultured splenic ILC2s with autologous splenic naïve B cells (Fig. 4j, Supplementary Fig. 6) and detected elevated CD40L and IL-10 secretion in the presence of ILC2s (Fig. 4k), with IL-10 previously shown to promote IgA responses^55,56. We also detected secreted IL-5, IL-13, IL-9, and IL-4 from splenic ILC2s (Supplementary Fig. 5b). Notably, splenic ILC2s significantly increased B cell proliferation to levels comparable to addition of exogenous CD40L (Fig. 4l). While B cell cytokine profile and surface marker expression remained unchanged (Supplementary Fig. 5c, d), ILC2s significantly enhanced secretion of IgG3, IgA and IgD (Fig. 4m, n), without affecting other isotypes (Supplementary Fig. 5e, f). Collectively, these findings indicate that splenic ILC2s are capable of supporting B cell proliferation and influencing antibody class switching, thereby contributing to the tuning of humoral immunity within the spleen.

Defining splenic and PBMC ILC3 and ILC precursors populations

We identified three distinct clusters of splenic ILC3s, distinguished by expression of IFNGR1, KLF2 and CD38 (Fig. 5a, b), which were consistently observed across donors (Supplementary Fig. 7a). IFNGR1 ILC3s expressed IFNGR1, NKp44, TNFSF13B and CD69; KLF2 ILC3s were enriched for KLF2, CD11b, NKp46, CD45RA and Fas; and CD38 ILC3s displayed high protein expression of CD38 and HLA-DR, as well as RASD1, ITM2C, TOB1 and SOX4 (Fig. 5b). CD38 ILC3s showed the most transcriptional complexity, with 208 uniquely upregulated genes and proteins (Fig. 5c).

Fig. 5. Splenic ILC3s can be classified into 3 clusters, marked by IFNGR1, KLF2 and CD38, respectively.

ILC3s from spleen and PBMC CITE-seq datasets were subclustered and analyzed for differential gene and protein expression. a Splenic ILC3s form three subclusters marked by IFNGR1, KLF2, and CD38. b Heatmap of selected top differentially expressed genes and proteins across splenic ILC populations. c A Venn diagram of shared and unique gene and protein expression of IFNGR1, KLF2 and CD38 ILC3 populations. d PBMC ILCs group into two ILC2, one ILC3, and three ILC precursor (ILCp) subclusters. e–h Expression of protein markers associated with ILC precursors in PBMC ILC subtypes. e Expression of CD45RA. f Expression of CD7. g Expression of CD73. h Expression of CD38. i Representative flow cytometry gating for splenic ILC3 populations. j Representative flow cytometry gating for PBMC ILC3 populations. k Proportion of splenic ILC3 subtypes in PBMC (n = 8) compared to spleen (n = 10). l Violin plot of ILC lineage-defining transcription factors across spleen and PBMC ILC subclusters. m Human PBMCs (n = 5) and splenocytes (n = 10) were stained ex vivo for intracellular expression of the transcription factor RORγ, and percent positive expression was graphed. n, o Splenocytes and PBMCs were stained for surface markers and assessed by flow cytometry. n Percent positive and representative gating of CD45RA (Splenocytes n = 10, PBMCs n = 5). o Mean fluorescence intensity (MFI) of CD127 (Splenocytes n = 10, PBMCs n = 5). p Similarity score between splenic and PBMC ILC subtypes. Statistical significance was determined by two-way ANOVA. Error bars represent standard deviation (SD).

To assess whether any splenic ILC3s resembled lymphoid tissue inducer (LTi)-like cells, we created a gene module of LTi-associated markers (LTA, LTB, CD25, IL2RA, CXCR6 and CXCR5)^57–59 (Supplementary Fig. 7b). While none of the splenic ILC3 subsets strongly expressed all LTi markers, KLF2 ILC3s had the highest module score (Supplementary Fig. 7b, c), suggesting that this population may share some transcriptional features, though unlikely to represent bona fide LTi cells. We also profiled PBMC ILC3/ILCps (Fig. 5d), and identified four ILCp clusters by using known ILCp markers, CD45RA and CD7⁶⁰, including one cluster marked by high CD73 and one by CD38, respectively (Fig. 5e–h). Additionally, we identified one mature ILC3 cluster, characterized by high CD38 and lower CD45RA expression (Fig. 5d–h).

To compare ILC3s across tissues, we used CITE-seq-derived surface markers to isolate IFNGR1, KLF2, and CD38 ILC3 subsets by flow cytometry (Supplementary Fig. 8). ILC3 populations were isolated by flow cytometry sorting using CD38⁺CD11B^- to isolate CD38 ILC3s, CD38^+/-CD11B⁺ to isolate KLF2 ILC3s, and the absence of these markers (CD38^-CD11B^-) for IFNGR1 ILC3s (Fig. 5i, j, Supplementary Fig. 8). Across all splenic donors, IFNGR1 ILC3s represented the most abundant splenic population (61.4% +/−17.6), followed by KLF2 ILC3s (22.4% +/−16.0) and CD38 ILC3s (15.9% + /‒ 7.0) (Fig. 5k). In blood, however, 79.5% + /−23.7 of ILC3/ILCps expressed KLF2 ILC3-associated markers, while 17.2% + /−21.0 matched IFNGR1 ILC3s and 2.9% + /−2.7 matched CD38 ILC3s (Fig. 5k).

We next compared expression of lineage-defining transcription factors across PBMC and splenic ILCs. Splenic CD38 and IFNGR1 ILC3s expressed the highest levels of RORC, while ILC2s and PBMC ILCps expressed more GATA3, in line with an ILCp phenotype^61,62(Fig. 5l). Flow cytometry analysis confirmed greater RORγT expression in all splenic ILC3 populations compared to PBMC ILC3/ILCps, with CD38 ILC3s expressing the highest levels (Fig. 5m). While PBMC IFNGR1 and KLF2 ILC3s had higher CD45RA (associated with progenitor/naïve cells), splenic ILC3s had significantly higher expression of CD127 (Fig. 5n, o, Supplementary Fig. 7d, e), consistent with a more differentiated state. Similarity analysis revealed the high similarity between PBMC ILCps and spleen KLF2 ILC3s, whereas PBMC CD38 ILCs had the highest similarity to their splenic counterparts. Notably, IFNGR1 ILC3s lacked a clear PBMC counterpart, underscoring their tissue specificity (Fig. 5p).

We next identified DEGs and proteins and validated the defining markers for IFNGR1 ILC3s by flow cytometry, confirming expression of IFNGR1, CD69, and NKp44 (Fig. 6a–f). Compared to circulating ILCps/ILC3s, splenic IFNGR1 ILC3s upregulated AREG, IL7R, IFNGR1, and CXCR4, while PBMCs showed higher CD45RA, CD11b, GATA3, and SELL, reinforcing a more activated, tissue-resident phenotype (Fig. 6g, h).

Fig. 6. Splenic ILC3s have distinct transcriptional and functional profiles.

CITE-seq data from spleen and PBMC ILCs were used to assess the similarity of populations from each source. a UMAP of IFNGR1 expression on splenic ILC3s. b Expression of IFNGR1 ILC3-associated genes and proteins on splenic ILC3s. c Representative flow cytometry gating for IFNGR1 and CD69 on splenic ILC3 populations. d–f Flow cytometry quantification of ILC3 populations. d Mean fluorescence intensity (MFI) of IFNGR1 (n = 10). e Percent positive CD69 (n = 10). f Percent positive NKp44 (n = 10). g Volcano plot of differentially expressed genes and proteins in splenic IFNGR1 ILC3s and PBMC ILCps/ILC3s. h Expression of IFNGR1 ILC3-associated genes and proteins on PBMC ILCps and splenic IFNGR1 ILC3s. i UMAP of KLF2 expression on splenic ILC3s. j Expression of KLF2 ILC3-associated genes and proteins on splenic ILC3s. k Representative flow cytometry gating for CD11b and NKp46 on splenic ILC3 populations. l Percent positive of NKp46 (n = 10). m Expression of ILCp-associated genes and proteins on PBMC ILCps and splenic KLF2 ILC3s. n PBMC, ILC expression of KLF2. o PBMC ILC expression of CD11b. p Trajectory analysis of ILC populations. q UMAP of CD38 expression on splenic ILC3s. r Expression of CD38 ILC3-associated genes and proteins on splenic ILC3s. s Representative flow cytometry gating for CD38 and HLA-DR on splenic ILC3 populations. t Percent positive of HLA-DR (n = 10). u Expression of CD38-associated genes and proteins in PBMC ILC3s/ILCps and splenic CD38 ILC3s. v Expression of antigen presentation and processing-associated transcripts on splenic ILC3s. Statistical significance was determined by a two-tailed Wilcoxon rank sum test (g) and a Friedman test with Dunn’s multiple comparisons (d, e, f, l, t). The number of replicates (n) represents the matched biological replicates (donors). Error bars represent standard deviation (SD).

Using publicly available scRNA-seq datasets from human colon, lung, blood, and tonsil⁶³ we confirmed that IFNGR1 ILC3s are also present in the colon, lung and tonsil, but absent from circulation (Supplementary Fig. 7f–j). GSEA revealed IFNGR1 ILC3s were enriched for pathways related to translation, innate mucosal immunity, and protein synthesis (Supplementary Fig. 9a), suggesting a poised effector state.

We next assessed DEGs and proteins on KLF2 ILC3s, and validated NKp46 surface expression ex vivo by flow cytometry (Fig. 6i–l). Because similarity scores link KLF2 ILC3s most closely to PBMC ILCps (Fig. 5p), we assessed common ILCp markers^60,64–67. Although splenic KLF2 ILC3s had lower CD45RA, TCF7, GATA3, CD7 and KLRB1 than ILCps, these markers were still highly detected on KLF2 ILC3s (Fig. 6m). In contrast, SELL was absent on KLF2 ILC3s but high on PBMC ILCps, while KLF2 ILC3s expressed ZBTB16, which is expressed by ILCps⁶⁸ (Fig. 6m). Further, PBMC ILCps also expressed KLF2 and CD11b, mirroring KLF2 ILC3s (Fig. 6n, o). We noted KLF2 ILC3s could be detected within the tonsil (Supplementary Fig. 7i) but were rare in other organs. Interestingly, pseudotime analysis predicted that KLF2 ILC3s are an upstream precursor population, giving rise to IFNGR1 and CD38 ILC3s (Fig. 6p). GSEA further supported a quiescent phenotype, with downregulation of glycolysis, ATP biosynthesis, and innate immune response in mucosa pathways²¹ (Supplementary Fig. 9b). These findings indicate that KLF2 ILC3s represent an ILC3-precursor-like population, sharing several ILCp-associated features, and highlight that KLF2 may serve as a marker for ILCp populations in both PBMC and spleen.

Splenic CD38 ILC3s were defined by expression of RASD1, HLA-DR, ITGB7 and SOX4, and HLA-DR expression was validated by flow cytometry (Fig. 6q–t). Exploring the PBMC ILC3 and ILCp populations also characterized by high CD38 surface expression revealed shared expression of HLADR, ITGB7 and SOX4 (Fig. 6u). However, unique to splenic CD38 ILC3s was RASD1, which plays a role in inhibiting migration and mTOR activity⁶⁹, and surface CD69, a marker of both activation and tissue residency⁷⁰ (Fig. 6u), implying restricted migration and enhanced tissue residency of splenic CD38 ILC3s.

CD38 ILC3s were found only to be present in PBMCs and spleen (Supplementary Fig. 7j), while GSEA revealed enrichment of cytokine signaling pathways, such as TGF-β, IL-2 and NOTCH1, as well as antigen presentation pathways (Supplementary Fig. 9c). Indeed, CD38 ILC3s showed higher expression of antigen presentation-related genes (CD74, PDIA3, CALR, PSMB1, TAP1 and TAPBP) in comparison to other splenic ILC3s (Fig. 6v), suggesting a potential role in antigen presentation and immune modulation.

We next assessed the cytokine profile of each ILC3 population by intracellular flow cytometry staining. Ex vivo, KLF2 ILC3s exhibited increased GM-CSF and IFN-γ expression, CD38 ILC3s produced IL-17F and IL-10, and both IFNGR1 and KLF2 ILC3s expressed TNF (Supplementary Fig. 10a). Low IL-22 expression was also observed across all subsets (Supplementary Fig. 10a).

However, following stimulation with either IL-2, IL-7 & IL-23 or IL-2, IL-7, IL-23 & IL-1β, all ILC3 subsets expressed GM-CSF and amphiregulin (Supplementary Fig. 10b, c). Notably, IFNGR1 ILC3s produced high IL-22, TNF, and IL-10 but little IFN-γ when IL-1β was absent, whereas adding IL-1β decreased IL-22, IL-10, TNF and increased IFN-γ (Supplementary Fig. 10c, d). This IL-1β–driven shift appeared unique to IFNGR1 ILC3s (Supplementary Fig. 10c).

Because KLF2 ILC3s displayed a precursor-like signature, we investigated their plasticity by culturing in NK cell, ILC2, or ILC3-activating media (Supplementary Fig. 11a). Even after 21 days, no significant changes in KLF2 ILC3 cytokine profiles were observed (Supplementary Fig. 11a), indicating commitment to the ILC3 lineage.

We also assessed whether IFN-γ influences IFNGR1 ILC3s (Supplementary Fig. 11b), and surprisingly found no change in cytokine expression profiles after 7 days in culture with either ILC3 cytokines alone or ILC3 cytokines and IFN-γ (Supplementary Fig. 11b), indicating IFNGR1 ILC3s cytokine profile is unaltered with IFN-γ stimulation.

Spatial locations, cellular interactions and functional characterizations of splenic ILC3s

ILC3s comprised ~0.3% of total splenocytes in both the WP and RP, with IFNGR1 ILC3s accounting for 58.07% of total ILC3s, KLF2 ILC3s for 23.04%, and CD38 ILC3s for 18.89% (Fig. 7a, b). IFNGR1 and CD38 ILC3s were primarily localized to the WP, whereas KLF2 ILC3s were enriched in the RP (Fig. 7c, d).

Fig. 7. Splenic KLF2 ILC3s preferentially interact with red pulp CD8⁺ T cells, while CD38 ILC3s interact with germinal center CD38-expressing plasma cells.

Spatial transcriptomic analysis was performed on the Xenium platform to assess the spatial context of ILC3s within the splenic architecture. a Proportion of ILC3 subsets as a percentage of total ILC3s. b Percent of total ILC3s in the RP and WP. c ILC3 subtype proportions in the RP and WP. d Xenium images of ILC3 subtypes in the RP and WP. Scale bars represent 200 μm. e KLF2 ILC3 closest neighbors in RP. f KLF2 ILC3 expected versus observed interactions with CD8⁺ T cells and myeloid cells in RP. g Source-target predicted interactions between KLF2 ILC3s and CD8⁺ T cells. h Xenium images of KLF2 ILC3s and CD8⁺ T cells in RP with gene expression of GZMB and TNF. Scale bars represent 50 μm. i Splenocytes (n = 10) were stimulated ex vivo with PMA/ionomycin and stained with ILC3 subset markers and intracellular expression of cytokines. The percent positive of granzyme B is shown. j CD38 ILC3 closest neighbors in WP. k CD38 ILC3 expected versus observed interactions with B cells and proliferating cells in WP. l Source-target predicted interactions between CD38 ILC3s and plasma cells. m Xenium images of CD38 ILC3s, B cells and proliferating plasma cells in WP with gene expression of TNFSF13B, CD38, and TNFRSF13B. Scale bars represent 100 μm (left) and 20 μm (right). n Splenocytes (n = 10) were stimulated ex vivo with PMA/I and stained with ILC3 subset markers and intracellular expression of cytokines. The percent positive of BAFF is shown. Statistical significance was determined by the Friedman test with Dunn’s multiple comparisons. Error bars represent standard deviation (SD).

Given their predominance in the RP, we examined the spatial interactions of KLF2 ILC3s in this area. Nearest-neighbor analysis revealed that myeloid cells (33.33%) and CD8⁺ T cells (26.67%) were their most frequent neighbors (Fig. 7e). Notably, observed interactions with CD8⁺ T cells exceeded expected frequencies (Fig. 7f). Source-target analyses suggested that CD8⁺ T cells engage KLF2 ILC3s via TNF and other TNF family ligands (TNFSF14, TNFSF10 and TNFSF12) (Fig. 7g). Prior studies have shown that TNFR2 engagement with TNF-α induces the expression of granzyme B⁷¹, which was observed in KLF2 ILC3s by both spatial transcriptomics and intracellular staining (Fig. 7h–i). Additional predicted KLF2 ILC3 interactions include B cell interactions via TNFSF13B and JAG1, among others (Supplementary Fig. 12a). Xenium confirmed co-localization of KLF2 ILC3s with CD8⁺ T cells and the expression of TNF by CD8⁺ T cells (Fig.7h). Thus, despite their precursor-like transcriptional profile, KLF2 ILC3s retain the capacity to express pro-inflammatory mediators.

Transitional B cells mature in follicles via BCR and B cell activating factor (BAFF) signaling, differentiating into either marginal zone B (MZB) cells through elevated Notch2 signaling and their BCR⁷², or short-lived germinal center-derived plasma cells and plasmablasts^73–76. CD38 is a known marker of activated B cells and antibody-secreting cells^77–80; however, a functional role for CD38 ILC3s has not been described. CD38 ILC3s were predominantly localized in the WP (Fig. 7c) and found most proximal to B cells, CD4⁺ T cells and proliferating cells expressing SDC1, a hallmark of plasma cells and plasmablasts⁸¹ (Fig. 7j). Proliferating cells also expressed CD19 and BANK1, in line with a plasmablast identity (Supplementary Fig. 4a). Observed interactions with plasmablasts were ~three-fold higher than expected, while those with B cells were only marginally larger-than-expected (Fig. 7k), suggesting a preferential localization near antibody-secreting cells.

Plasma cell maturation begins with antigen stimulation, where mature B cells are activated in the presence of cytokines, such as BAFF, and then differentiate into antibody-secreting cells involved in immune activation and regulation⁸². ILC3s have been shown to induce B cell activation via BAFF (TNFSF13B) and the Notch ligand Delta-like 1 (DLL1)¹⁵, while LTα1β2 (LTB)-expressing ILC3s were shown to regulate splenic responses¹⁶. Interestingly, source-target interaction analysis identified interactions between CD38 ILC3s and plasmablasts that are likely mediated via CD38, DLL1, LTB and TNFSF13B (Fig. 7l, Supplementary Fig. 12b), suggesting that CD38 ILC3s may be positioned to support antibody-secreting cells, paralleling known mechanisms of BAFF-mediated B cell activation⁷⁴. Spatial analysis confirmed CD38 ILC3s co-localize with CD38⁺ B cells and proliferating plasmablasts within and surrounding the B cell follicle, and notably, the ILC3s expressed TNFSF13B (BAFF) (Fig. 7m, n).

Follicular dendritic cells (FDC) are traditionally responsible for retaining and displaying native antigen to B cells in secondary lymphoid organs⁸³. However, the presence of HLA-DR⁺ CD38⁺ ILC3s expressing BAFF and Notch ligands, coupled with their enriched antigen presentation-related gene expression (Fig. 6v), as well as prior studies which report a role for BAFF, DLL1 and LTB-expressing ILC3s in promoting B cell responses¹⁵, suggests a potential supportive role for CD38⁺ ILC3s in plasmablast maturation and antibody production in the WP.

IFNGR1 ILC3s, the most abundant splenic ILC3 population not found in circulation, were observed in both the WP and RP (Fig. 7c). Interaction analysis indicates they are in close proximity with B cells, myeloid cells, and CD4⁺ T cells (Fig. 8a–d). In the RP, predicted B cell interactions exceeded expected levels, whereas interactions with CD4⁺ T cells were also marginally elevated in both compartments (Fig. 8a–d). IFNGR1 ILC3s were predicted to interact with B cells via LTB, JAG1 and TNSFSF13B, and, with CD4⁺ T cells via ADAM10 binding to IL6R (Fig. 8e). Additional IFNGR1 ILC3-cell interactions include B cell interactions via TNFSF10 (TRAIL), and ICAM1-ITGAL-ITGB2 interactions with CD4⁺ T cells, among others (Supplementary Fig. 12c). Spatial analysis confirmed co-localization of IFNGR1 ILC3s with B cells and CD4⁺ T cells in both the WP and RP, as well as expression of TNFSF13B and IL6 by ILC3s (Fig. 8f, g) and ex vivo cytokine analysis confirmed that IFNGR1 ILC3s express IL-6 and BAFF upon stimulation (Figs. 7n, 8h).

Fig. 8. IFNGR1 ILC3s interact preferentially with B cells and induce an activated phenotype characterized by increased IL-10 and BAFF expression, as well as class switching to IgG1, IgG2, and IgG4 isotypes.

Xenium was used to define interactions of IFNGR1 ILC3s within the spleen, and interactions were validated by in vitro co-cultures. a IFNGR1 ILC3 closest neighbors in WP. b IFNGR1 ILC3 expected versus observed interactions with B cells and CD4⁺ T cells in WP. c IFNGR1 ILC3 closest neighbors in RP. d IFNGR1 ILC3 expected versus observed interactions with B cells and CD4⁺ T cells in RP. e Source-target predicted interactions between IFNGR1 ILC3s and B cells or CD4⁺ T cells. f, g Xenium images of IFNGR1 ILC3s in clusters with B cells and CD4⁺ T cells with gene expression of TNFSF13B, IL6, and TNFRSF13C. Scale bars represent 200 μm (left) and 20 μm (right). f IFNGR1 ILC3 interactions in the WP. g IFNGR1 ILC3 interactions in the RP. h Splenocytes (n = 10) were stimulated ex vivo with PMA/ionomycin and stained with ILC3 subset markers and intracellular expression of cytokines. Percent positive of IL-6. i–m Naïve B cells and IFNGR1 ILC3s were sorted from human spleen and co-cultured together for 7 days. After 7 days, cultures were assayed for B cell expansion, expression of activation markers, intracellular cytokines and secreted antibodies. i Overview of experimental set-up of IFNGR1 ILC3-B cell co-cultures. Created in BioRender. Crome, S. (2026) BioRender.com/b50dkls. j Fold expansion of B cells after co-culture with IFNGR1 ILC3s (n = 5). k Representative gating of CD86 and CD38 and fold change in expression of CD86 (n = 4) and CD38 (n = 3). l Fold change and concentration of supernatant levels of IgG1, IgG2 and IgG4 (n = 5). m Representative gating of intracellular IL-10 and BAFF and fold change in intracellular cytokine expression on B cells (n = 4) and concentrations in supernatant (n = 4). Statistical analysis was determined by Kruskal-Wallis tests (j), one-tailed Wilcoxon matched-pairs signed-rank test (k, l, m) and Friedman test with Dunn’s multiple comparisons (h, l). The number of replicates represents biological replicates (donors). Error bars represent standard deviation (SD).

Given IL-6 has been shown to induce IL-10 in splenic B cells⁸⁴, and spatial transcriptomics supported in vivo co-localization with B cells (Fig. 8f, g), we hypothesized that IFNGR1 ILC3s could impact germinal center responses. To test this possibility, we co-cultured splenic IFNGR1 ILC3s with donor-matched naïve B cells for 7 days and assessed ILC3s’ ability to influence B cell surface markers, Ig isotype switching, and cytokine production (Fig. 8i, Supplementary Fig. 13). B cell numbers significantly increased with the addition of IFNGR1 ILC3s (Fig. 8j). Surface marker analysis showed a trend towards increased CD86 and CD38 on B cells (Fig. 8k), in line with an activated, mature B cell phenotype⁸⁵ and showed no changes in CD24, CD27, CD40, or CD138 (Supplementary Fig. 13c, d). IFNGR1 ILC3s also increased levels of IgG1, IgG2 and IgG4 (Fig. 8l), with limited differences in isotypes such as IgG3, IgE, IgA, IgM or IgD (Supplementary Fig. 13a, b). Intracellular cytokine staining identified that B cells co-cultured with IFNGR1 ILC3s had a nonsignificant increase in expression of BAFF, and interestingly, IL-10 (Fig. 8m, (Supplementary Fig. 13f–g). These results demonstrate that splenic IFNGR1 ILC3s can support B cell survival, activation, and functional maturation, including the potential induction of an IL-10⁺ Breg-like phenotype⁸⁶ in vitro, potentially contributing to splenic immune homeostasis.

Discussion

Understanding how ILCs contribute to immune regulation in human secondary lymphoid tissues remains limited, particularly within the spleen, a central site for blood-borne antigen surveillance and systemic humoral immunity. Here, we present a spatially resolved, multimodal atlas of human splenic ILCs, revealing tissue-specific transcriptional states and functional specializations across NK cells, ILC2s, and ILC3s. By integrating transcriptomics, surface proteomics, spatial localization, and functional assays, our study defines splenic ILC subsets and their interactions with other immune cells, uncovering previously underappreciated associations with B cell survival, class switching, and phenotypic states. These findings position splenic ILCs as spatially organized modulators of adaptive immunity within the human spleen.

Splenic NK cell subsets displayed compartmentalized phenotypes, with CD56^bright NK cells characterized by CD39 and TIGIT expression and spatially associated with alternatively activated macrophages, consistent with a putative regulatory checkpoint that may aid in restricting NK cell-mediated activation within germinal centers. In contrast, CD56^dim NK cells maintained a cytotoxic profile, including granzyme B and perforin, resembling murine splenic NK cells known to regulate T follicular helper and GC B cells via perforin-dependent mechanisms^5,8.

We identified a previously unexplored role for splenic ILC2s, which were enriched for activation markers and preferentially localized near vascular regions, possibly contributing to transitional B cell survival⁵⁴. These ILC2s expressed CD40L, in addition to type 2 cytokines and IL-10, which may contribute to B cell survival, proliferation and IgA secretion based on in vitro co-culture assays. While ILC2-driven IgA class switching has been reported in other tissues⁸⁷, in our co-cultures, ILC2s promoted IgA production from splenic B cells, suggesting that ILC2-derived signals can influence IgA responses even in a non-classical anatomical context. While we do not interpret this to imply splenic B cells are a physiologic source of IgA in vivo, these findings underscore the capacity of ILC2s to influence humoral responses outside of canonical mucosal contexts.

Three transcriptionally and spatially distinct ILC3 populations were identified. KLF2 ILC3s, resembling ILC precursors, were predominantly in the red pulp, co-localizing with activated CD8⁺ T cells, and expressed granzyme B, a feature not typically associated with ILC3s. CD38 ILC3s resided in B cell follicles together with proliferating B cells and plasma cells, expressed HLA-DR, produced BAFF, and showed phenotypic overlap with follicular dendritic cells⁸⁸. These could indicate that CD38 ILC3s are capable of providing auxiliary antigen presentation and promoting B cell differentiation in germinal centers. IFNGR1 ILC3s, the most abundant splenic population and present in both the WP and RP, induced regulatory features in B cells in vitro, including IL-10 expression, while also promoting secretion of IgG1, IgG2, and IgG4 antibodies. These findings complement prior reports of ILC3-mediated B cell regulation in human tonsil⁸⁹, extending this paradigm to the spleen.

Our study prioritized high-resolution, multimodal characterization of rare splenic ILC and NK cell populations, leveraging both sorted and unsorted spleen samples to broaden immune context and allow integration across compartments. Although CITE-seq can capture all immune cells, our approach prioritized depth and resolution of ILC biology over complete cellular breadth. Larger-scale, unbiased whole-transcriptome spatial approaches across more donors will be essential to build a comprehensive, generalizable spleen atlas. Given the species-specific differences in ILC composition⁹⁰ and splenic architecture, our decision to focus on human samples reflects the need for human-centered mapping. Nonetheless, complementary murine studies could provide mechanistic validation of the cell-cell interactions and functional phenotypes identified here.

By defining subset-specific spatial organization and interaction networks within the human spleen, this study establishes a high-resolution reference for splenic ILC biology and provides mechanistic insight into how ILC populations are integrated into local immune architecture in secondary lymphoid organs. Notably, we report the first spatial and functional characterization of human splenic ILC2s and identify IFNGR1⁺ ILC3s as a previously unrecognized, dominant ILC population in human spleen, that is capable of tuning B cell responses. This high-resolution framework provides a foundation for exploring the effects of ILC modulation of splenic immune responses in health and disease.

Methods

PBMCs and spleen preparation

Human spleen and blood specimens were collected in accordance with biosafety and ethical protocols approved by the University Health Networks Research Ethics Board (20-5206 and 17-6229, respectively). Healthy peripheral blood was obtained from donors through the Canadian Blood Services Blood4Research program, with each donor providing written, informed consent and processed as previously described⁷. Spleens from normal deceased donors were received from the University of Alberta IsletCore or the Ajmera Transplant Center Islet Transplant Program. Samples were cut into small pieces using a sterile scalpel, then placed into gentle MACS C-tubes (Miltenyi Biotec, 130-093-237) with 10 mL PBS plus 2% FBS and placed in a gentleMACS dissociator using the m_spleen_01_01 setting. After dissociation, the slurry was mashed through a 70 µm cell strainer, and ACK lysed to remove red blood cells⁹¹. Donor information can be found in Supplementary Table 1. Donors were selected to achieve balanced representation (across all sequencing, 6 female, 7 male); however, sex was not included as a variable in the analysis due to limited statistical power. Donor ages ranged between 26–72.

CITE-seq staining of total and ILC-enriched spleen and PBMCs

Surface staining was performed as described in the BioLegend protocol⁹². As ILCs are present in the spleen and PBMC at very low abundance and display significant overlap in protein and RNA level expression of molecules expressed by T cells, we also performed an ILC enrichment on 4/7 spleen samples and 3/3 PBMC samples. For all 3 PBMC samples, T cells were also enriched for using the EasySep Human T Cell Isolation Kit Human (StemCell Technologies, 17951), and CITE-seq was performed on these samples concurrently. 5 µL of TruStain FcX Fc Blocking reagent (BioLegend, 422302) was first added, and cells were incubated for 10 min at 4 °C. Next, a lineage cocktail consisting of FITC-conjugated antibodies against B cells, T cells, and myeloid cells was used to exclude immune cells from ILCs. Antibodies used for lineage exclusion are listed in Supplementary Table 2. Cells were then washed and resuspended in EasySep Buffer (StemCell Technologies, 20144) and enriched using the EasySep FITC Positive Selection Kit II (StemCell Technologies, 17662) per manufacturer instructions. Enriched cells were then stained with flow cytometry antibodies as well as the TotalSeq-C antibody cocktail concurrently at a 1:1 ratio, and flow sorted with the panel found in Supplementary Table 3 before sequencing. All TotalSeq-C antibodies added are listed in Supplementary Table 4. Cells were sorted as either CD56^dim NK cells, CD56^bright NK cells or other ILCs (Supplementary Fig. 1a). CD56^dim NK cells were sorted as Lineage⁻, CD56^dim, CD16^-, CD56^bright NK cells were sorted as Lineage⁻, CD56^bright, CD16⁻, and ILCs were sorted as Lineage⁻, CD127⁺, CD16⁻. For splenic cells, after sorting, cells were washed 3x, combined into one tube (each tube containing 25% CD56^dim NK cells, 25% CD56^bright NK cells, and 50% helper ILCs), resuspended at 1200 cells/µL in PBS (ThermoFisher, 10010049) plus 0.05% BSA (ThermoFisher, AM2616) and sent for sequencing. For PBMCs, after sorting, cells were washed 3x, and NK cells were combined into one tube (containing 50% CD56^dim NK cells and 50% CD56^bright NK cells) while ILCs were left in a separate tube, and each tube was resuspended at 1200 cells/µL in PBS (ThermoFisher, 10010049) plus 0.05% BSA (ThermoFisher, AM2616) and sent for sequencing. For total spleen samples (non-ILC enriched), splenocytes were stained for CITE-seq as above, washed 3x and resuspended at 1200 cells/µL in PBS (ThermoFisher, 10010049) plus 0.05% BSA (ThermoFisher, AM2616) and sent for sequencing.

CITE-seq data analysis pipeline

Samples were prepared for sequencing using the 10X Genomics Single Cell 5’ v2 platform in accordance with the manufacturer’s instructions for capture of 12,000 cells per sample. Reverse transcription, cDNA amplification and sequencing libraries were generated using 10X Genomics Single Cell 5’ v2 reagents. Across samples, cells were sequenced to a target depth of 40,000 reads per cell. Read alignment to the reference human genome (GRCh38/hg38) and gene expression matrices were generated by the 10X Genomics CellRanger pipeline (version 6.1.2).

Data were loaded into R, and Seurat (Seurat v4.3.0) objects were generated for each individual sample. Quality control for splenic samples included removing high mitochondrial content cells by filtering cells with >10% of Unique Molecular Identifiers (UMI) mapped to mitochondrial genes, between 500 & 5000 unique genes and to remove doublets, between 500 & 20000 total genes. Quality control for PBMC samples included removing high mitochondrial content cells by filtering cells with >5% of Unique Molecular Identifiers (UMIs) mapped to mitochondrial genes, between 500 & 5000 unique genes and to remove doublets, between 500 & 30000 total genes.

Data from Spleen and PBMC were merged and normalized with SCTransform⁹³, respectively. The individual datasets were integrated using Harmony (RunHarmony, v1.2.3)⁹⁴. Principal component analysis was used for dimensionality reduction (RunPCA), and cells were clustered using the Louvain algorithm with 30 principal components (FindNeighbors and FindClusters) in Seurat⁹⁵. Clusters were visualized using the Uniform Manifold Approximation and Projection (UMAP) algorithm⁹⁶.

Clusters suspected of containing doublets or aggregates were inspected manually, re-filtered with more stringent thresholds (e.g., requiring >1000 detected genes, >1000 UMIs, <10,000 UMIs, <5000 genes, and 1–5% mitochondrial reads), re-integrated, and re-clustered until the aggregate signal was eliminated. Cells with low transcript abundance (<200 features) and high antibody expression, indicative of antibody aggregates, were also removed. Integrated data was then normalized following the same process above. Cell types and lineages were annotated by analyzing the top differentially expressed genes and/or proteins (FindMarkers, min.pct = 0.25) and manually labeled.

For differential gene expression (DGE), we used the Wilcoxon rank-sum test implemented in the presto package. Genes were considered significant if they met both of the following thresholds: adjusted p-value < 0.05 (Benjamini–Hochberg correction) and log₂ fold change > 0.25. We then used top DEGs for gene set enrichment analysis (GSEA), using the GSEA Pre-ranked tool. Gene ranking was performed using the Presto Wilcoxon output: specifically, each gene was assigned a score equal to the sign of the log fold change × –log₁₀(p-value), ensuring that both effect size and statistical significance contributed to ranking. Genes were then ordered by this score, and the ranked list was exported for input into GSEA. In the GSEA software, we selected the “No_Collapse” chip platform option and used the most recent gene set collection from the Bader Lab at the time of analysis (Human_GOBP_AllPathways_no_GO_iea_August_08_2023_symbol.gmt).

Cell-cell predicted interactions were inferred using LIANA³³. Cell lineage and pseudotime inference were performed using the package slingshot⁹⁷, and cell similarity scores were determined by the ClusterFoldSimilarity package⁹⁸.

To quantify functional transcriptional programs across NK cell subsets, we computed three gene module scores representing cytotoxicity⁴³, tissue residency⁴⁴ and myeloid-related suppression^45–47 using Seurat’s AddModuleScore function (v5). The cytotoxicity module included canonical effector genes (PRF1, GZMB, GZMH, GZMA, GZMK, GNLY, NKG7, CTSW, and IFNG). The tissue-residency module consisted of established TRM-associated markers (CD69, ITGAE, CXCR6, ITGA1, ZNF683, and PRDM1). The myeloid suppression module captured inhibitory receptor and cytokine-response genes known to mediate myeloid-driven NK dysfunction (LAG3, TIGIT, PDCD1, HAVCR2, CD96, CTLA4, TGFBR2, SMAD2, SMAD3, SOCS1, SOCS3, STAT3, ADORA2A, PTPN6, PTPN11, CBLB, HIF1A, IL10RA, and ZFP36L2). Using Seurat’s AddModuleScore function, the average normalized expression of module genes was calculated. These gene module scores were visualized using the VlnPlot function in Seurat.

To compare splenic ILC populations to those from publicly available datasets, we downloaded data from publicly available single-cell RNA-seq data of human ILC3s from blood, colon, lung, and tonsil, which were obtained from the Gene Expression Omnibus (GEO; accession number GSE150050⁶³). Cells were subset to those annotated as ILC3s in the original dataset. Quality control excluded cells with fewer than 500 detected genes, >40% mitochondrial gene content, or >10% ribosomal gene content, resulting in 5,056 cells retained for analysis. Data were normalized and variance-stabilized using SCTransform (Seurat v4.3.0) with mitochondrial gene percentage regressed out. Dimensionality reduction was performed using PCA, followed by batch correction using Harmony with “Origin” as the integration variable. UMAP was computed from the top 30 Harmony components, and clustering was performed with a resolution of 0.75 using the shared nearest neighbor (SNN) graph. Gene module scores for general ILC3s (genes used: ID2, KIT, IL7R), IFNGR1 ILC3s (genes used: IFNGR1, TNFSF13B, CD69), CD38 ILC3s (genes used: CD38, SOX4, RASD1), and KLF2 ILC3s (genes used: KLF2, JUND, IFITM2) were calculated using Seurat’s AddModuleScore function. Module scores were then visualized to compare enrichment across tissues.

Flow cytometry analysis

Surface marker staining was performed for 30 min after 15 min blocking with human TruStain FcX (BioLegend, 422302). Surface antibodies assessed are listed in Supplementary Table 5. For intracellular staining, cells were fixed and permeabilized using the FOXP3/Transcription Factor Staining set (eBioscience, 00-5523-00) and incubated with intracellular antibodies for 30 min at room temperature (RT). For intracellular cytokine analysis, cells were stimulated with 1x Cell Stimulation Cocktail (eBioscience, 00-4970-93) for 6 h, with GolgiStop and GolgiPlug (BD Biosciences, 554724, 555029) added after 2 h. Intracellular antibodies used are listed in Supplementary Table 6. Samples were acquired on a LSR Fortessa or Symphony A3 flow cytometer (BD Biosciences), and data were analyzed using FlowJo v10 software.

Xenium workflow

Spleen from deceased donors was cut into small sections, fixed overnight in 10% neutral buffered formalin (Sigma-Aldrich, HT501128). Following overnight fixation, the tissue was rinsed in water and stored in 70% ethanol. Paraffin embedding, tissue slicing and Hematoxylin and Eosin (H&E) staining were performed by the University Health Network (UHN) Pathology Research Program (PRP). For spatial transcriptomic studies, the Xenium platform (10X Genomics) was applied to tissue samples by the Princess Margaret Genomics Center (PMGC), according to the manufacturer’s protocol. Briefly, slides were deparaffinized by incubation at 60 °C on a thermal cycler for 2 h and subsequently immersed in xylene and ethanol. Next, slides were decrosslinked by adding decrosslinking buffer to the Xenium cassette. Probe hybridization mix was then added to each well, and the Xenium cassette was placed on the thermocycler at 50 °C overnight (16–24 h). The following day, slides were washed, and post-hybridization wash buffer was added and incubated at 37 °C for 30 min. Slides were then washed and incubated with the ligation mix at 37 °C for 2 h. Next, slides were incubated with autofluorescence solution for 10 min at room temperature in the dark to quench autofluorescence and following autofluorescence quenching, were incubated with Xenium Nuclei staining buffer for 1 min at room temperature. Slides were then loaded into the Xenium instrument and left to run for approximately 2 days. Xenium samples were probed with the standard colon panel, listed in Supplementary Table 7 and a custom 100-gene panel, listed in Supplementary Table 8.

Xenium analysis

The raw gene expression matrix was imported from the data files using the LoadXenium function in the Seurat (v5.3.0) package. Cells with 0 counts were removed, data were normalized and visualized as described above and clusters were manually annotated. Cell ID’s and annotations were loaded into Xenium Explorer 3 for cell type and gene expression visualization.

Nearest neighbors were calculated using k = 5 nearest neighbors in the FNN (version 1.1.4.1) package. Expected versus observed interactions were calculated by: (population 1 cells * population 2 cells) / (total cells²) = probability of interaction probability of interaction * k nearest neighbors * total cells = expected number of interactions

Vascular regions were identified by manual annotation based on tissue morphology visible in the Xenium spatial transcriptomics images. Regions were defined as lumen-like, acellular or low-cellularity areas exhibiting characteristic vessel architecture, including rounded or elongated tubular structures consistent with vascular morphology.

All Xenium image analysis shown are representative from Spleen 1 (Supplementary Table 1). Comparable results were obtained from two additional independent samples (Spleens 2 and 3).

Ex-vivo sorting of splenic ILC subsets

Thawed splenocytes were incubated with human TruStain FcX (BioLegend, 422302) and stained with a cocktail of lineage antibodies conjugated to FITC, listed in Supplementary Table 2. Splenocytes were then washed in FACS buffer (98% PBS (ThermoFisher, 10010049) + 2% FBS (R&D Systems, S11550H)), resuspended in EasySep Buffer (StemCell Technologies, 20144), and ILCs enriched using the EasySep FITC Positive Selection Kit II to remove lineage-positive cells per manufacturer instructions (StemCell Technologies, 17662). Enriched cells were stained with the antibody cocktail listed in Supplementary Table 9 and sorted using a BD Symphony S6 (BD Biosciences). Isolated ILCs were then expanded in complete X-VIVO 15 media (Lonza, BEBP04-744Q) supplemented with 5% human AB serum (MilliporeSigma, H4522-100ML), 100 U/mL Penicillin-Streptomycin (ThermoFisher, 15140122) and 1x GlutaMAX (ThermoFisher, 35050061).

ILC2 and IFNGR1 ILC3:B cell co-cultures

Naïve B cells were isolated from cryopreserved spleen and cultured with autologous sorted IFNGR1 ILC3s or ILC2s. ILCs were sorted as described above. Naïve B cells were isolated using an EasySep Human Naïve B Cell Isolation Kit (Stemcell Technologies, 17254) and plated in complete X-VIVO 15 media (Lonza, BEBP04-744Q) supplemented with 5% human AB serum (MilliporeSigma, H4522-100ML), 100 U/mL Penicillin-Streptomycin (ThermoFisher, 15140122) and 1x GlutaMAX (ThermoFisher, 35050061) with ILC2s or IFNGR1 ILC3s at a ratio of 1:5. In both co-cultures, 5 mg/mL CpG ODN 2006 (Miltenyi, 130-100-105), which engages TLR9, a powerful B cell-activating receptor⁹⁹, was added to all co-cultures as a baseline stimulus to induce naïve B cell activation and proliferation in vitro. In cultures with ILC2s, ILC2 cytokines (100 IU/mL IL-2 (Proleukin) (SteriMax Inc, DIN:021301811), 20 ng/mL IL-7 (BioLegend, 581908) and 20 ng/mL IL-33 (BioLegend, 581806) were added to maintain ILC2 viability. A control well including 0.1 mg/mL CD40L (BioLegend, 591704) was also added to cultures containing ILC2 cytokines. In cultures with IFNGR1 ILC3s, ILC3 cytokines (100 IU/mL IL-2 (Proleukin) (SteriMax Inc, DIN:021301811), 20 ng/mL IL-7 (BioLegend, 581908) and 20 ng/mL IL-23 (BioLegend, 574106) were added to maintain ILC3 viability. On days 3 and 5, supernatants were replenished with fresh CpG and cytokines. After 7 days, supernatants were collected, and B cells were assessed for alterations in cell number, surface marker expression and intracellular cytokine production. If staining for cytokines, cells were first stimulated as described above. Supernatants were assessed for secreted antibodies and cytokines by cytokine bead array using the LEGENDplex Human Immunoglobulin Isotyping Panel (8plex) (BioLegend, 740638) and the LEGENDplex Human B Cell Panel (13-plex) (BioLegend, 740527), respectively.

Statistical analysis

Statistical significance was determined using a Kruskal–Wallis test with Dunn’s multiple comparisons for comparisons involving three or more unpaired groups. For two-group unpaired non-parametric comparisons, a Mann–Whitney test was used. For paired donor analyses used to assess B cell:ILC cocultures, a one-tailed Wilcoxon matched-pairs signed-rank test was applied. For experiments involving repeated measures across more than two matched conditions, a Friedman test with Dunn’s multiple comparisons was used. Comparisons across two factors (PBMC and spleen) were analysed using a two-way ANOVA with Geisser–Greenhouse correction. The number of biological replicates (n) is indicated in each figure legend. Error bars represent standard deviation (SD) unless otherwise stated. Data analysis was performed using GraphPad Prism v9.5.1.

Reporting summary

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

Author contributions

S.J.C., K.T.R., J.M.M., and S.Q.C. designed and implemented CITE-seq studies on spleen and PBMC. S.J.C., A.H., M.E.M., C.S.R., and S.Q.C. designed and implemented Xenium spatial studies. S.J.C., S.M., A.H., H.M., S.J., M.L.M., and J.A.M.processed tissue, performed in vitro validations and co-culture studies. S.J.C., K.T.R., J.M.M., J.A.M., and S.Q.C. developed ILC isolation and expansion methods. S.J.C., S.M., A.H., M.E.M., and H.M. analyzed the data. S.Q.C., H.M., & S.J.C. wrote the REB. S.Q.C. and C.S.R. supervised the work. S.J.C. and S.Q.C. wrote the manuscript, which all authors reviewed and edited.

Peer review

Peer review information

Nature Communications thanks Andrea Cerutti, who co-reviewed with Celia Corral-Vazquez, Zachary Rogers 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 made possible by funding from the Canadian Institutes of Health Research (169084, 198149), the New Frontiers Research Fund, Natural Sciences and Engineering Research Council of Canada (RGPIN-2021-03672), and the Canadian Foundation for Innovation (38308). S.Q.C holds a Tier 2 Canada Research Chair in Tissue-Specific Immune Tolerance. SJC was supported by a Banting and Best Novo Nordisk Scholarship, a Queen Elizabeth II/Dr. Dina Gordon Malkin Graduate Scholarship in Science and Technology, Ontario Graduate Scholarship and a Canadian Society of Transplantation (CST) Research Training Award. SJ was supported by a Banting and Best Novo Nordisk Scholarship, an Ontario Graduate Scholarship and a Canadian Graduate Scholarship – Doctoral award. MLM was supported by a Canada Graduate Scholarship – Master's award and an NSERC Cell and Engineering Approaches to Preserve and Rejuvenate Organs (CEAPRO) Collaborative Research and Training Experience (CREATE) award.

Data availability

The CITE-seq data have been deposited in GEO under the accession number GSE293758. The Xenium data have been deposited in GEO under the accession number GSE293759. Publicly available single-cell RNA-seq data of human ILC3s from blood, colon, lung, and tonsil were obtained from the GEO accession number GSE150050. All other data are included in the Supplementary Information or available from the authors, as are unique reagents used in this article. The raw numbers for charts and graphs are available in the Source Data file whenever possible. Source data are provided with this paper.

Competing interests

The authors declare the following competing interests: KTR, SJC, JAM, JMM and SQC have filed a provisional patent for ILC isolation and expansion methods (Provisional Patent Application#: 63/353,823 and #: 63/469,234).

Supplementary information

The online version contains supplementary material available at https://doi.org/10.1038/s41467-026-76100-7.