Spatial and temporal mapping of human innate lymphoid cells reveals elements of tissue specificity

Summary

Innate lymphoid cells (ILC) play critical roles in regulating immunity, inflammation and tissue homeostasis in mice. However, limited access to non-diseased human tissues has hindered efforts to profile anatomically-distinct ILCs in humans. Through flow cytometric and transcriptional analyses of lymphoid, mucosal and metabolic tissues from previously healthy human organ donors, here we have provided a map of human ILC heterogeneity across multiple anatomical sites. In contrast to mice, human ILCs are less strictly compartmentalized and tissue localization selectively impacts ILC distribution in a subset-dependent manner. Tissue-specific distinctions are particularly apparent for ILC1 populations, whose distribution was markedly altered in obesity or aging. Furthermore, the degree of ILC1 population heterogeneity differed substantially in lymphoid versus mucosal sites. Together, these analyses comprise a comprehensive characterization of the spatial and temporal dynamics regulating the anatomical distribution, subset heterogeneity, and functional potential of ILCs in non-diseased human tissues.

eTOC

Innate lymphoid cells (ILC) critically regulate tissue immunity and homeostasis in mice, but limited access to healthy human tissues has hindered efforts to profile anatomically-distinct ILCs in humans. Yudanin and colleagues provide a comprehensive map of the spatial and temporal dynamics regulating the anatomical distribution, subset heterogeneity, and functional potential of ILCs in non-diseased human tissues.

Introduction

Innate lymphoid cells (ILCs) are a recently recognized family of innate immune cells that have been implicated in playing critical roles in regulating immunity, inflammation and tissue homeostasis in the context of infection, chronic inflammation, metabolic disease and cancer (Brestoff et al., 2015; Chang et al., 2011; Eberl et al., 2015; Klose and Artis, 2016; Monticelli et al., 2011; Moro et al., 2010; Neill et al., 2010; Price et al., 2010; Scandella et al., 2008; Sonnenberg et al., 2012, 2013).

Murine ILCs are defined by lack of expression of cell surface markers associated with granulocytes, dendritic cells (DC), macrophages and conventional B and T lymphocytes (termed Lineage negative, Lin-) but can be positively identified by expression of CD127 (IL-7Rα), CD25 (IL-2Rα) and c-Kit (Buonocore et al., 2010; Colonna, 2009; Satoh-Takayama et al., 2010; Takatori et al., 2009; Vonarbourg et al., 2010). Murine ILCs can be delineated into three categories: ILC1 populations express T-bet and include classical NK (cNKs) cells and ILCs that express IFNγ (Diefenbach, 2015; Eberl et al., 2015; Klose and Artis, 2016; Mortha and Diefenbach, 2011); ILC2 populations are comprised of IL-33R⁺ cells that express the transcription factor GATA-3 and secrete the T helper-2 (Th2) cell-associated cytokines IL-5, IL-13 and growth factor amphiregulin (Almeida and Belz, 2016; Klose and Artis, 2016; Liang et al., 2011; Monticelli et al., 2011; Moro et al., 2010; Neill et al., 2010; Price et al., 2010; Sonnenberg et al., 2013) ; and ILC3 populations that are IL-33R-, express RORγT and produce the Th17 cell-associated cytokines IL-17A and IL-22 (Buonocore et al., 2010; Colonna, 2009; Satoh-Takayama et al., 2010; Takatori et al., 2009; Vonarbourg et al., 2010). The majority of these effector cytokines have been shown to directly regulate epithelial and stromal cell responses at barrier surfaces, and many are predominantly secreted by ILC subsets enriched within these sites (Crosby and Waters, 2010; Doherty and Broide, 2007; Klose and Artis, 2016; Lee et al., 2018; Lloyd, 2010; O’Sullivan and Sun, 2017; Reynolds et al., 2010; Saetang and Sangkhathat, 2018; Sonnenberg et al., 2010).

In mice, anatomically-distinct ILC subsets critically provide robust tissue-specific protective responses against local pathogens, and are functionally divergent in lymphoid versus mucosal sites (Klose and Artis, 2016; Mackay and Kallies, 2017, Ricardo-Gonzalez et al., 2018). Tissue-resident murine ILC populations have also been implicated in promoting and sustaining chronic inflammatory diseases, including inflammatory bowel disease (IBD), allergic lung and skin inflammation, and obesity-induced immune dysregulation (Forkel and Mjösberg, 2016; Lund et al., 2017; Saetang and Sangkhathat, 2018; Salimi et al., 2013). Recent studies have identified human ILCs in fetal and adult inflamed or diseased tissues that phenotypically resemble murine ILC2 and ILC3 populations, suggesting they may also exhibit functional tissue specificity (Cella et al., 2008, 2009, 2010; Crellin et al., 2010; Cupedo et al., 2008; Li et al., 2018; Mjösberg et al., 2011). However, spatial distribution of ILCs is dissimilar between humans and mice, and may reflect differences in subset composition and/or function among anatomically diverse populations (Almeida and Belz, 2016; Simoni et al., 2016). Importantly, limited access to non-diseased tissues has hindered efforts to understand how the tissue environment contributes to human ILC distribution, and whether anatomically-distinct subsets are preferentially compartmentalized within these sites remains unclear.

Here, we present a quantitative analysis of ILC distribution and heterogeneity in lymphoid, mucosal and metabolic tissues obtained from a diverse cohort of 44 previously non-diseased organ donors over a wide range of ages and body mass indexes (BMIs). We combine phenotypic and transcriptional profiling of the major ILC subsets from 9 tissues with distribution variance and correlation analyses to identify distinctions in the temporal and spatial dynamics of human ILC populations. Our analyses reveal phenotypic and transcriptional diversity within and among anatomically-distinct human ILCs, which vary in the degree to which site-specific environmental cues impact their resultant distribution, phenotypic heterogeneity, transcriptional profile, and functional potential. Together, our results provide a comprehensive map of the spatial and temporal dynamics of human ILC populations in non-diseased tissues.

Results

Human ILC subsets differentially populate anatomically diverse tissues

ILC populations have been identified in numerous lymphoid and barrier sites where they provide tissue site-specific responses critical for both immune protection and tissue homeostasis. However, how the tissue environment contributes to ILC distribution, and whether specific ILC populations are preferentially compartmentalized within these sites remain poorly defined. While ILCs have been thoroughly characterized in mice, limited access to non-diseased human tissues has hindered efforts to interrogate the phenotype and function of anatomically-distinct ILC populations in non-diseased humans. In an ongoing collaboration with LiveOnNY, we obtained access to multiple lymphoid, mucosal and adipose tissues from previously non-diseased research-consented organ donors (Figure 1A). Specimens acquired from donors comprised nine tissues sampling major mucosal, lymphoid, and metabolic compartments (Figure 1A). All donors were HIV-, HBV-, HCV-, cancer-free and clear of systemic infection prior to brain death caused by head trauma (20.45%), anoxia (27.27%), stroke (50.00%) or other non-infectious traumas (2.27%) (Figure 1B). Comprising both obese and non-obese individuals over a broad spectrum of BMIs (mean BMI=29) (Figure 1B), the 44 donors analyzed here ranged in age from 2 to 70 years (mean age=46.2) (Figure 1C) and were predominantly male (Figure 1D). Additional descriptive statistics, CMV and EBV seropositivity, and individual attributes of each donor used in this study are compiled in Table S1.

Figure 1. Donor descriptive statistics and analysis overview.

A. Analysis overview depicts 9 tissues (left), from which 4 ILC populations (right) were isolated, sampled metabolic (Abd., abdominal; Mes., mesenteric), lymphoid (mLN, mesenteric lymph node; ldLN, lung-draining lymph node), and mucosal (LP, lamina propria; IEL, intra-epithelial lymphocytes) sites. Classical NK (cNK) (gray), ILC1 (red), ILC3 (green) and ILC2 (blue), subsets were further subdivided by expression of NKp44 and CD56 (ILC1 and ILC3), or CRTH2 (ILC2), respectively. Colors and abbreviations are used consistently throughout. B. Donor cause of death (COD; anoxia (white), head trauma (gray), stroke (black), other traumatic injury (red)) and body mass index (BMI; underweight (<18, white), normal (18–25, gray), overweight (25–30, black), and obese (>30, red)) distributions shown include all 44 donors analyzed. C. Age distribution and D. gender (black, male; white, female) of 44 donors. Additional descriptive statistics and individual donor information is compiled in Table S1. E. Schematic of gating strategy used to identify human ILCs; lineage-negative (Lin^–; CD3, CD5, FcεRI, CD11c, CD11b, CD14, CD19) CD56⁺CD127^– cNK and CD56^–CD127⁺ ILC subsets were delineated into CRTH2⁺ ILC2, CRTH2^–CD117^–ILC1 and CRTH2^–CD117⁺ ILC3, and further subdivided by their expression of NKp44 into NKp44⁺ (ILC1⁺, ILC3⁺) or NKp44^– (ILC1^–, ILC3^–) populations. F. Representative flow cytometry plots show frequencies (percent) of CD45⁺Lin-CD127⁺CD56^+/− ILC populations identified within each tissue from an individual donor. For donor information and all individual values shown see Table S1 and Tables S2–S3. See also Table S6 for a summary of donors and tissues used per figure panel.

To assess systemic distribution patterns, we performed flow cytometric analysis of ILCs isolated from human adipose (abdominal (Abd. Fat), mesenteric (Mes. Fat)), spleen, lymph nodes (lung-draining (ldLN), mesenteric (mLN)), lung, small intestine lamina propria (SI LP) (jejunum, ileum), colon and intraepithelial lymphocytes (IEL) (Figure 1A, Table S2, Table S6). We designed a single-panel, surface marker-based flow cytometric identification strategy consistent with the standard in the field (Bernink et al., 2013; Björklund et al., 2016; Lim et al., 2017; Simoni et al., 2016), which allowed us to reliably and reproducibly distinguish these cell subsets across tissues while minimizing contamination from myeloid, T cell, and other cell lineages. Specifically, in all tissues we identified populations of CD56⁺ CD127^– cNKs, CD56^+/− CD127⁺ CRTH2^– CD117^– NKp44^+/− ILC1s, CD56^– CD127⁺ CRTH2⁺ ILC2s, CD56^+/− CD127⁺ CRTH2^– CD117⁺ NKp44^+/− ILC3s that lacked expression of lineage markers associated with T cells, B cells, DCs, macrophages and multiple granulocyte populations (CD3, CD5, CD11b, CD11c, CD14, FcεR1 and CD19) as previously reported in the field (Bernink et al., 2013; Björklund et al., 2016; Lim et al., 2017; Simoni et al., 2016) (Figure 1E, 1F).

To ensure gating stringency of ILCs and cNKs, we took several approaches. First, inclusion of CD5 was necessary to eliminate T cell contamination (Lim et al., 2017; Simoni et al., 2016). Second, to ensure appropriate separation between ILCs and cNKs, we examined the tissue-specific pattern of cNK cell markers CD94 and NKp80. While we confirmed that these markers are a useful way to identify cNK cells in blood, the expression pattern of these receptors on cNK cells was much more dim and very heterogeneous across tissues, with some sites showing almost no expression (particularly the colon and jejunum) (Figure S1A, S1B). Further, we did not detect appreciable expression of CD94 or NKp80 on CD127+ ILCs from any tissue, except on a minor proportion in the lung and draining lymph node (Figure S1A, S1B), although we cannot rule out that expression of these receptors may be affected by enzymatic cleavage during digestion. Collectively, these data illustrate that the ILC subset gating strategy we employed here is not likely to include contamination of cNK cells. Lastly, to avoid myeloid cell contamination, we examined expression of myeloid lineage markers CD16, CD11b, CD11c on ILCs and cNKs. Use of CD16 can aid in reducing contamination without compromising tissue ILC identification as CD127⁺ ILCs did not express CD16 (Figure S1C), although a small proportion of lung-specific cNKs can express CD16 (Figure S1C), and therefore we did not include this marker in our lineage gate. However, we found that inclusion of CD11b and CD11c in the lineage gate was essential to avoid substantial myeloid cell contamination, particularly in mucosal tissues. Critically, we found that in non-diseased human lymphoid and mucosal tissues, few of the CD127⁺ ILC subsets expressed CD11b or CD11c once contaminating myeloid cells were excluded. (Figure S1D–S1F). Therefore, our stringent gating strategy was unlikely to be inadvertently excluding rare ILC populations from analysis. Collectively these data confirm the validity of our flow cytometric gating strategy as a robust, single-panel design for identifying human cNK and ILC subsets.

Spatial and temporal analyses reveal phenotypic heterogeneity within and among human ILC populations

To discern whether ILCs in humans are compartmentalized within different tissues, we first analyzed the relative frequency of ILC subsets among nine mucosal, lymphoid and metabolic tissues from 44 diverse donors. Composition and phenotypic heterogeneity was assessed within each tissue by flow cytometry, using markers established in mice (Figure S2A, S2B) and those used previously in limited human studies to delineate human ILC subsets and survey their expression of transcription factors (RORγT GATA3, EOMES, TBET), surface receptors (IL-7Rα (CD127), CRTH2, NKp44) and maturation markers (c-Kit) (Figure 1E, Figure 2). We found extensive heterogeneity in subset composition within human ILCs isolated from distinct tissue sites, with markedly different distribution patterns than murine ILC subsets (Figure 2A, Figure S2).

Figure 2. Spatial and temporal distribution of human ILC subsets.

Frequency of lymphocytes in 9 tissue sites isolated from 44 organ donors was quantified by flow cytometry. A. Mean frequency (+SEM) of cNK (gray), ILC1 (red), ILC2 (blue) and ILC3 (green) subsets expressed as a percent of total CD45⁺Lin^– is shown for each tissue. Representative flow cytometry histograms depict transcription factor expression by B. lymphoid and mucosal ILC1, ILC3, and cNK or C. lung-derived ILC1, ILC2 and ILC3 subsets from an individual donor. D. Distribution variance between subset frequencies was assessed for each tissue by two-way ANOVA with post-hoc Holm-Sidak correction. Significant frequency variances between tissue pairs are indicated by p values, with red- and pink-shaded boxes (p < 0.001, red; p < 0.005, light red; p < 0.05, pink) and white boxes (ns). E. Non-parametric spearman correlation analysis of changes in ILC subset tissue distribution with increasing age and F. BMI is shown by tissue, ordered from adipose to lymphoid to mucosal sites. Correlation strength, directionality (r = −1, bottom to r = +1, top) and significance for each site is denoted by shading (p <0.05, red; NS, not significant). For donor information and all individual values shown see Table S1 and Tables S2–S3. See also Figure S2 for murine ILC distribution and Table S6 for a summary of donors and tissues used per figure panel.

In mice, ILC2 and ILC3 were the major non-cNK subsets in tissues, with strict compartmentalization of ILC2s to epididymal (eWAT) and inguinal (iWAT) white adipose tissues and lungs, and ILC3 to intestinal sites (Figure S2C, S2D). In contrast, in humans, ILC1 and ILC3 subsets comprised the predominant non-cNK ILC populations in all tissues, and were proportionally more frequent mucosal and lymphoid sites, particularly in the colon and ileum IEL, respectively (Figure 2A). Analysis of transcription factor expression revealed that RORγT⁺ ILC3s preferentially populated the spleen and LP jejunum (Figure 2B), while GATA3 expression was predominant in lung ILC2s (Figure 2C). Moreover, in a majority of individuals, ILC2 comprised a major non-cNK population within the lungs and metabolic tissues (Figure S3C, S3D). Although we identified both ILC2 and ILC3 in the human intestine (Figure 2A), the frequency of intestinal ILC2 varied widely among individuals (Figure S3E, S3F) and the ILC3 population potentially may also include progenitor cells, as described in a recent publication (Lim et al., Cell 2017). ILC1 and cNK subsets, which populated lymphoid and mucosal tissues were heterogeneous in their expression of EOMES and TBET (Figure 2A, 2B, Figure S3G, S3H).

The proportion of ILCs within each tissue was consistent among diverse donors (Figure S3), allowing us to statistically test whether fluctuations in subset frequency vary significantly (p <0.05) as a function of tissue localization. To analyze differential ILC subset compartmentalization among lymphoid, mucosal and adipose tissue sites, we quantified the cell numbers and mean subset frequency variance in each tissue (Figure 2D, Table S2). Although mucosal sites were preferentially populated by, and had significantly (p <0.05) greater frequencies of ILC3s, the proportion of ILC1 and ILC3 cells in lymphoid tissues was similar (Figure 2D, Table S2), suggesting that ILC subset distribution is less strictly compartmentalized in humans than in mice (Figure S2E).

Analyzing donors by age and BMI also enabled us to determine whether ILC subset composition varied with age or in metabolic disease. To assess quantitatively age and disease-dependent effects, we calculated the correlation between ILC subset frequencies within each tissue and donor age or BMI (Table S3). The impact of donor age or metabolic health on ILC subset composition differed among anatomically distinct populations. Intestinal ILC3 frequencies decreased with increasing age in a tissue-dependent manner, with a compensatory increase of intestinal cNK, while subsets in lymphoid tissue sites remained unchanged (Figure 2E, Table S3). Conversely, increased donor BMI was correlated with a reduction in ILC1 and ILC2 frequencies within mucosal and abdominal adipose tissues (Figure 2F, Table S3). Together, these results indicate that tissue localization differentially impacts ILC distribution within human tissues in a subset-dependent manner.

The human ILC subset-dependent effects of tissue localization, age, and disease on subset distribution prompted us to probe whether ILC subsets restricted within individual tissues are phenotypically and/or functionally distinct. Previous studies have reported that ILC subsets can vary in phenotype depending on activation state and/or tissue localization. We assessed NKp44 and CD56 expression on human ILCs isolated from 9 tissues and found phenotypic heterogeneity within and among distinctly-localized subsets. In humans, expression of the cNK cell receptor, NKp44, was exhibited by ILC3s and intestine-derived ILC1s (Figure 3A, Crellin et al., 2010; Glatzer et al., 2013). Additionally, these populations heterogeneously expressed CD56, a cNK cell receptor also marking activated T cells. Overall, spleen- and intestine-derived ILCs were more phenotypically heterogeneous than in lung and adipose, exhibiting greater variance in proportion of NKp44⁺ and CD56⁺ cells (Figure 3A, 3B, Figure S3). Expression of NKp44 was highest in mucosal ILC1 and ILC3 subsets and lowest in populations within lymphoid and metabolic tissue sites (Figure 3A, 3C). This phenotypic heterogeneity exhibited by ILC1s and ILC3s was consistent, with little variation among donors (Figure S3), and distinctly compartmentalized within each tissue (Figure 3D). Moreover, the impact of donor age and BMI differentially affected phenotypically distinct populations in a subset-dependent manner. Specifically, increasing age was associated with significant reductions (p <0.05) in both NKp44⁺ and NKp44^– intestinal ILC3s, but had no impact on either population of ILC1s (Figure 3E, Table S3). Conversely, obesity correlated with reduced proportions of NKp44⁺ but not NKp44^– ILCs in lymphoid and mucosal sites, although trends differed between IEL versus LP compartments of the SI and colon (Figure 3F, Table S3). Moreover, we could also identify differences in ILC subset composition within related anatomic groups, as illustrated by altered frequencies of NKp44⁺ ILCs observed between two types of draining lymphoid tissue (ldLN versus mLN) in the context of age (ILC3s) or obesity (ILC1s) (Figure 3E, 3F). These results reveal distinct phenotypic heterogeneity within and among human ILC populations, with subset-specific effects in the context of age or obesity.

Figure 3. Spatial and temporal dependence of human ILC subset heterogeneity.

Mean frequency (± SD) of A. NKp44⁺ and B. CD56⁺ cells within ILC1 and ILC3 subsets depicts varied subset heterogeneity among mucosal, lymphoid, and adipose tissues. C. Mean frequency (+SEM) of total CD45⁺Lin^–CD127⁺NKp44⁺ (solid) and NKp44^– (striped) ILC subsets is shown within each tissue, and compiled from 44 donors. D. Distribution variance between ILC1 and ILC3 subset frequencies was assessed for each tissue by two-way ANOVA and adjusted for multiple comparisons using Holm-Sidak correction. Significant frequency variances between tissue pairs are indicated by p values, with red- and pink-shaded boxes (p < 0.001, red; p < 0.005, light red; p < 0.05, pink) and white boxes (ns). E. Non-parametric spearman correlation analysis of changes in ILC1 and ILC3 subset tissue distribution with increasing age and F. BMI is shown by tissue, ordered from adipose to lymphoid to mucosal sites. Correlation strength, directionality (r = −1, bottom to r = +1, top) and significance for each site is denoted by shading (p <0.05, red; NS, not significant). For donor information and all individual values shown see Table S1 and Tables S2–S3. See also Figure S3 for distribution by donor, Table S6 for a summary of donors and tissues used per figure panel.

Tissue localization drives transcriptional heterogeneity in an ILC subset-dependent manner

Studies of murine ILCs have revealed that transcriptional differences among anatomically distinct subsets reflect tissue-specific heterogeneity in their function and protective capacity (Ricardo-Gonzalez et al., 2018). To investigate whether differentially localized human ILCs are also transcriptionally distinct, we performed genome-wide analysis of sort-purified ILC1, ILC2, ILC3, and cNK subsets from jejunum, spleen, and lungs of 5 individual donors using high-throughput RNA sequencing. Due to technical limitations of sequential, plate-based cell sorting, not all ILC populations could be isolated from every tissue. Despite this limitation, unbiased principal component analysis revealed tight clustering among ILC subsets regardless of localization, suggesting that the tissue environment is not the predominant determinant of transcriptional heterogeneity in human ILCs (Figure 4). Furthermore, clustering was not distinct among samples from different donors (Figure S4A, S4B).

Figure 4. Ontogeny, not localization, predominantly drives transcriptional heterogeneity in human ILCs.

Transcriptional sequencing of mucosal and lymphoid ILCs from lung, jejunum, and spleen of 5 individuals was performed by RNAseq. Principal component (PC) analysis of resultant transcriptomes shows stronger clustering among A. cNK (gray), ILC1 (red), ILC2 (blue) and ILC3 (green) subsets than B. within jejunum (yellow), lung (white), and spleen (black) tissues C. Heatmap depicts expression of previously characterized genes and unsupervised clustering among cNK (gray), ILC1 (red), ILC2 (blue) and ILC3 (green) subsets. Relative gene expression is indicated by row z-score, ranging from −2 (low, blue) to 2 (high, red). See also Figure S4 for extended clustering, Table S1 for donor information, and Table S6 for a summary of donors and tissues used per figure panel.

Of all ILC subsets, ILC3s were most transcriptionally distinct, overlapping least with other populations regardless of localization. Likewise, ILC2 samples clustered together with little overlap, although samples from the same tissue exhibited greater similarity. The least transcriptionally distinct population was comprised of ILC1s, which overlapped extensively with both splenic and intestinal cNK cells and shared some attributes with ILC3s (Figure 4A, 4B). ILC and cNK subsets in all tissues distinctly expressed transcription factors and surface markers previously established in delineating innate immune cell populations, including TBX21 (TBET), EOMES, KIT (CD117), RORC (RORγT), GATA3, ID2, IL7R (CD127), and NCR1 (NKp46) (Figure 4C). In concordance with previous studies (Bernink et al., 2013; Björklund et al., 2016; Krämer et al., 2017; Simoni et al., 2016b), and data from our phenotypic analyses (Figure S3), we found that TBX21 and EOMES expression by cNK was heterogeneous and tissue-dependent. Specifically, TBX21 was expressed by all cNK (Figure 4C), though protein expression was highest among subsets within intestinal tissues (Figure S3G). While all cNK expressed EOMES transcriptionally (Figure 4C), protein expression was markedly higher in splenic and lymphoid cNK than in intestinal-derived cNK (Figure S3H). Among ILC subsets, TBX21 was predominantly enriched in ILC1, which also expressed EOMES heterogeneously (Figure S3, Figure 4C). Conversely, neither ILC2 nor ILC3 expressed EOMES or TBX21 (Figure 4C). RORC expression was highest in ILC3 and not seen in ILC2, while GATA3 was expressed by both subsets (Figure 4C), as previously reported (Li et al., 2018; Mjösberg et al., 2012). Of note, while hierarchical clustering reflected similarly strong grouping by subset, ILC1s further formed small transcriptionally distinct cliques resembling either ILC2, ILC3 or cNK populations (Figure 4C). These findings were in contrast to results from murine studies in which fate mapping studies indicate that ILC1s comprise a lineage distinct from other helper-like ILC subsets (Diefenbach, 2015; Eberl et al., 2015; Klose and Artis, 2016; Mortha and Diefenbach, 2011). Together, while these data validate our gating strategy to identify human ILC2 and ILC3, the heterogeneity in transcription factor expression among cNK and ILC1 (Figure S3) suggest that delineation by canonical transcription factors alone may not accurately depict the complexity of these populations in diverse human tissue microenvironments.

Subset- and tissue-specific transcriptional signatures distinguish mucosal versus lymphoid ILCs

The highly conserved, subset-specific gene signature exhibited by human ILCs suggests that subset identity, rather than tissue localization, may be a stronger determinant of ILC functional potential. Indeed, hierarchical clustering of subsets within spleen or jejunum clearly delineated cNK from ILC2 and ILC3 populations (Figure S4C, Figure 5A, 5B). Overall, human ILC populations were transcriptionally distinct, particularly among ILC2, ILC3 and cNK cells, reflecting subset-specific functional potential that remained remarkably consistent between donors. Both splenic and intestinal cNK cells were enriched in genes related to host defense (Figure 5C, Table S5), expressing a conserved network of cytotoxicity-associated transcripts including GNLY, GZMB, PRF1, GZMA, CD244, FASLG, IFNG, CD244 and SH2D1A (Figure 5E). Similarly, ILC3 subsets in both sites distinctly expressed a network of genes associated with responses to cytokine or lipid signaling, including IL7R, KIT, AQP3, PTGS2, and TNFRSF25 (Figure 5D, 5F, Table S5). Compared to ILC3s or NKs, we found that ILC2s had elevated expression of IL1R2, PTGS2, IL17RB, IL10RA, and CD74, genes that have been demonstrated to have in vivo significance in murine models (Stehle et al., 2016; Yu et al., 2016) (Figure S5A, Table S5). Moreover, subset-specific genes enriched in ILC2s encode markers that function in multiple pathways associated with type 2 immune responses, including IL13, IL1RL1, IL17RB, AREG, and CCR4 (Figure S5B, S5C) (Izuhara, 2017; Schuijs and Halim, 2018; Zhou, 2012). We also identified a number of genes that previously have not been associated with ILCs, such as PLAUR, PLIN2, SGK1, C5AR1, and IRAK3 (Figure S5), suggesting that this dataset can serve as a useful resource for other investigators to pursue targets and pathways of interest in human ILC development and function.

Figure 5. Subset- and tissue-specific transcriptional signatures distinguish mucosal versus lymphoid ILCs.

Shown here are significant (q<0.1, p<0.001) differentially expressed genes by each subset from the A. jejunum or B. spleen. Relative gene expression is indicated by row z-score, ranging from −2 (low, blue) to 2 (high, red). Bar plots of mean normalized counts (+SEM) depict genes differentially expressed by C. cNK and D. ILC3 populations within spleen (black) and LP jejunum (white). Corresponding KEGG pathways are indicated in brackets for select significant differentially expressed genes. E, F. cNK and ILC3s differentially express cohesive interaction networks of integrins, chemokine receptors, and their associated transcription factors. Known corresponding protein interactions (STRING 10.0) include inhibition (red T line), binding (orange line), activation (arrow), or other reaction (black line). Also indicated are additional interactions derived from text-mining databases (dashed line) and tissue-specific expression (green, LP jejunum; black, spleen). See also Figures S4–S5 and Table S5 for tissue- and subset-specific signature genes, Table S1 for donor information, and Table S6 for a summary of donors and tissues used per figure panel.

Our analyses revealed little distinction between spleen- and gut-derived ILC subsets, which were only distinct in expression of a few genes (Figure S6A, S6B). However, these genes were similarly and consistently expressed by cNK and ILC3 populations within each site, suggesting an important role in tailoring tissue-specific functions (Figure 5A, 5B). Compared to intestinal cNK, splenic cNK had significantly (FDR <0.1) increased expression of chemokine- and integrin-encoding genes, including CCL3, CCL4, and ITGB2, as well as activation and/or migration-related genes, such as S1PR5 and SELL (CD62L) (Figure 5B, 5C, Figure S6A). Similarly, jejunum-derived subsets upregulated known intestinal migration genes, including CCR9 and ITGAE (Figure 5A, 5C, Figure S6B). Tissue-specific distinctions predominantly comprised immune-associated transcripts downregulated in intestinal versus lymphoid tissues, such as IL1R1, RUNX2, and PCDH9 in ILC3s (Figure 5D) or FCGR3A, PRF1, and KLRC1 in cNK (Figure 5C). The global reduction of these transcripts within gut-derived ILC populations may be indicative of enhanced immune suppression at barrier sites with greater pathogen exposure, ostensibly to prevent unnecessary tissue damage.

Transcriptional profiles of anatomically-distinct human ILC1s reflect heterogeneity in activation and function

Growing interest in the potential plasticity of ILC subsets has resulted in particular focus on deciphering phenotypic and functional distinctions among ILCs, although the contribution of environmental cues to subset-specific characteristics remains unclear, particularly for populations within human tissues. Our results suggest that, although ILCs are less strictly compartmentalized in humans than in mice, localization differentially affects ILC population composition and distribution in a subset-dependent manner. Tissue-specific distinctions were particularly apparent for ILC1 populations, the distribution of which was significantly (p <0.05) more affected by both local and systemic perturbations caused by obesity or aging. Furthermore, phenotypic and transcriptional heterogeneity varied widely among ILC1 subsets in lymphoid and mucosal sites, which resembled cNK and ILC3 populations, respectively (Figure 4A, Figure 5A, 5B, Figure S4).

Recent studies describe nearly indistinguishable ILC1 and cNK populations during inflammation (Bernink et al., 2013; Björklund et al., 2016; Krämer et al., 2017; Simoni et al., 2016b), revealing an increasingly multifaceted relationship between group 1 ILCs that is further highlighted by our findings, which suggest that anatomic localization provides additional layers of complexity. To investigate whether tissue site-specific environmental cues drive group 1 ILC heterogeneity, we analyzed the transcriptional profiles of cNK and ILC1 subsets isolated from mucosal and lymphoid compartments and found marked differences in population heterogeneity between distinctly localized subsets (Figure 6). Specifically, gut-derived ILC1s exhibited increased transcriptional heterogeneity compared to splenic ILC1s, as determined by hierarchical clustering of sample distances (Figure 6A, Figure 6B), comprising multiple groups functionally resembling and/or distinct from cNK cells. Although intestinal ILC1s preferentially expressed transcripts characteristic of helper-like innate lymphocytes, including those encoding surface markers, such as ICOS, IL18, IL7R, IL2RA, IL18R1, and IL4R (Figure 6A), few of these genes distinguished ILC1 subsets across tissues (Figure 6C, 6D). Conversely, we found less transcriptional heterogeneity among splenic ILCs, which exhibited little distinction from co-localized cNKs (Figure 6B, 6C).

Figure 6. Transcriptional heterogeneity among anatomically-distinct ILC1s reflects functional potential.

Shown here are significant (q<0.1, p<0.001) differentially expressed genes by cNK and ILC1 in A. jejunum and B. spleen. Relative gene expression is indicated by row z-score, ranging from −2 (low, blue) to 2 (high, red). C. Bar plots of log normalized counts (mean+SEM) depict select significant genes differentially expressed by cNK (gray) and ILC1 (red) populations within the spleen (striped) and jejunum (solid). Genes previously associated with particular ILC subsets are in bold. D. Violin plots of mean log normalized counts across subsets in all tissues show expression distribution of select subset-specific genes by ILC1s (red) and cNKs (gray). E. Comparison of tissue-independent significant differentially expressed genes enriched in ILC1 versus cNK populations (ILC1, red) to those distinguishing Lin^–CD34⁺CD38⁺CD123^– CD45RA⁺CD7⁺CD10⁺CD127⁺ common lymphoid progenitors (CLP) from CD3^–CD56⁺NKp46⁺ cNK cells (CLP, dark gray) revealed 22 overlapping transcripts. Heatmap shows relative expression of common genes in CLP and ILC1 cells, as indicated by row z-score ranging from low (black) to high (red). These and previously published sequencing data sets are, or will be made, publicly available on GEO (GSE60448). See Figure S6 and Table S5 for tissue- and subset-specific signature genes, Table S1 for donor information, and Table S6 for a summary of donors and tissues used per figure panel.

The lack of heterogeneity among ILC1 subsets in spleen may reflect their diminished functional capacity, their potential contamination by other ILC subsets due to a particular gating strategy, or indicate the presence of progenitor cell populations. To distinguish between these possibilities, we mined a previously published dataset (GSE60448) containing transcriptome-wide sequences of Lin-CD34⁺CD38⁺CD123^–CD45RA⁺CD7⁺CD10⁺CD127⁺ common lymphoid progenitors (CLP) and CD3-CD56⁺NKp46⁺ cNK cells isolated from human umbilical cord blood (UCB), and compared significant transcriptional differences between these populations to those in our current data set. Of the 74 significantly differentially expressed genes (sDEG; FDR <0.1) between cNK and ILC1 subsets in our data, 22 were also sDEG between UCB-derived CLP and ILC1 populations (Figure 6E). Among these, HLA-DOA, BCL11A, KIT and IL7R were highly expressed by both ILC1 and CLP (Figure 6E). While KIT (CD117) and IL7R (CD127) are known markers of both CLP and ILC1 subsets (Vivier et al., 2018), expression of BCL11A could be unique to lymphoid progenitors or progenitor-like cells. Future studies utilizing single-cell sequencing will be necessary to fully characterize heterogeneity within mature lymphoid and progenitor-like cell subsets.

Discussion

While studies in mouse model systems indicate that ILCs serve as important regulators of immunity, inflammation and tissue homeostasis in barrier sites (Brestoff et al., 2015; Chang et al., 2011; Eberl et al., 2015; Klose and Artis, 2016; Monticelli et al., 2011; Moro et al., 2010; Neill et al., 2010; Price et al., 2010; Scandella et al., 2008; Sonnenberg et al., 2012, 2013), analysis of the phenotypic and functional characterization of human ILCs has been more limited. Recent studies have reported the existence of ILCs in some human fetal or diseased adult tissues (Mjösberg et al. 2011; Mjösberg and Spits 2016; Simoni et al. 2016; Brestoff et al. 2015; Monticelli et al. 2011). However, little is known about the composition of anatomically diverse ILC populations within healthy humans, and whether their site-specific heterogeneity reflects distinct responsiveness to host-, environmental- or microbial-derived signals remains unclear.

Here, using a unique resource comprised of previously healthy organ donor tissues (Thome et al., 2014; Granot et al., 2017; Kumar et al., 2017), we have been able to comprehensively characterize ILCs across 9 tissues from 44 donors spanning over 50 years of life. Our data reveal differences in phenotypic heterogeneity both within and among human ILC subsets, with consistent expression patterns exhibited in each tissue. We found that although ILCs were less strictly compartmentalized in humans than in mice, localization differentially impacted population composition and distribution in an ILC subset-dependent manner. Our results support the possibility of distinctive effector functions among anatomically diverse ILC subsets, as human ILCs exhibit site-specific transcriptional signatures and responses to aging and metabolic disease.

While human and murine ILCs shared expression of some surface markers, there were enough distinctions and inconsistencies between them to raise fundamental questions about the biology of human ILCs (Klose and Artis, 2016; Mjösberg and Spits, 2016; Montaldo et al., 2016; Simoni and Newell, 2018). Murine ILC populations comprise three distinct subsets, whose functional capacities are tailored to the environments in which they reside (Ebbo et al., 2017, Kim et al., 2016, Klose and Artis, 2016). Whether anatomically-distinct human ILCs are similarly affected by environmental signals has been difficult to examine due to limited tissue access. Our analysis of ILC subset distribution and phenotype across 9 lymphoid, mucosal and adipose tissues within an individual and across many donors revealed a more complex relationship between localization and heterogeneity than that in mice. ILCs in spleen and intestines were more phenotypically heterogeneous than those in lung and adipose, particularly in their expression of CD56 and NKp44. While NKp44 and CD56 have been noted on ILC3s and ILC1s, respectively, in previous publications using diseased tissues (Crellin et al., 2010; Glatzer et al., 2013), our study examines how expression of these receptors changes with regard to tissue microenvironment in the context of healthy human tissues. Overall, our data illustrate that ILC1s are the most varied in phenotype. These results are consistent with a recent study of human ILC1s, which suggested that they may not be a homogeneous cell subset (Simoni et al., 2016). Additionally, it is possible that current gating strategies for ILC1s, and to a lesser extent ILC3s, could potentially contain GATA3⁺ c-kit^+/− ILC2s that have downregulated CRTH2 upon entry into the tissue environment from the blood (Tait Wojno et al., 2015). Phenotypic differences between tissue-derived ILCs versus blood-derived ILCs are an important aspect of human ILC biology to consider for future studies.

To gain additional insight into the environmental impact on human ILC heterogeneity, we examined changes in subset composition among anatomically distinct sites within the donor population as a function of both age and BMI. Previous studies have relied on limited sample sizes and/or diseased human tissues, and have not been able to model changes over time or with health versus disease (Mjösberg and Spits, 2016; Mjösberg et al., 2011; Simoni et al., 2016). Although directly probing tissue-specific responses of ILCs is not possible within the same individual, the wide range of BMIs and age represented by our donors, as well as the consistency of tissue subset composition among them, allowed an unprecedented quantitative analysis of the spatial and temporal dynamics of human ILCs in health and disease over life. However, given these are human samples from previously non-diseased, deceased donors, we cannot explore causation and can only infer potential correlations based on the data presented. Nevertheless, our approach revealed critical insights into the distinct impact of aging and tissue perturbation on anatomically diverse ILC populations. Results from these studies show that frequencies of intestinal, but not lymphoid, ILC3 decrease with increasing donor age. Moreover, NKp44⁺ ILC1 were reduced in obesity, with a compensatory increase in cNK populations. There could be numerous reasons for this selective increase in cNK cells (but not other ILC subsets) following the reduction of ILC1s. One explanation may be that because NKs are developmentally and functionally most similar to ILC1s compared to the other ILC subsets, NKs are more likely to take advantage of the cytokine milieu, including IL-15, to expand when there are fewer ILC1s to compete with for those signals in the tissue. Together, these data suggest that ILC subsets are differentially impacted by local signals, and in particular, implicate tissue localization in driving ILC1 heterogeneity.

Recent examination of human ILCs have begun probing whether these subsets transcriptionally resemble those in mice, as well as how their tissue-dependent transcriptional heterogeneity may influence resultant function (Björklund et al., 2016; Mjösberg and Spits, 2016; Sonnenberg et al., 2013). Consistent with results from these studies, we found few tissue-specific differences among ILC subsets in spleen versus lymph nodes. Moreover, we observed no phenotypic distinctions between ILC2s and ILC3s among mucosal and lymphoid sites. In contrast, ILC1 populations were most transcriptionally diverse among tissues, with splenic ILC1 transcriptionally mirroring classical NK, while intestinal ILC1 more closely resembled ILC3 subsets. ILC1-like populations have been identified in intestinal epithelia, tonsils, skin, liver, and tumor infiltrates (Bernink et al., 2013; Björklund et al., 2016; Lim et al., 2017; Simoni et al., 2016). However, the most widely accepted definition of ILC1s relies on the absence of markers (CRTH2^– CD117^–), and T-bet expression in this population has been shown to be bi-modal (Mjösberg and Spits, 2016). Therefore the ‘true’ ILC1 identity in non-diseased tissue remains controversial. The reduced heterogeneity and transcriptional similarity to CLP we observed among splenic ILC1 suggests that current definitions of this cell population may not sufficiently exclude progenitor-like cells. Full characterization of ILC1 heterogeneity necessitates the use of single-cell approaches sampling tissues across anatomic sites within an individual to truly distinguish ILC1 and cNK. To that end, findings detailed here are complemented and supported by results from a recent study (Crinier et al., 2018) using single-cell RNA sequencing (RNAseq) to compare human and murine cNK signatures within spleen and blood. Of note, the spleen-specific cNK profiles described therein closely mirror our data from bulk RNAseq of splenic cNK, both exhibiting tissue-specific expression of numerous genes, including KLRC1, PRF1, CCL3, IFNG, and ITGB2. Further, our RNAseq analyses revealed that some ILC1s cluster more closely with ILC3s versus NKs in lymphoid versus intestinal tissues. Combined with the data in the recent study (Crinier et al., 2018) these data provide a step forward in unraveling the complexity of ILC1 heterogeneity and relationship to the broader ILC family.

We demonstrate human ILCs widely vary in the degree to which site-specific environmental cues impact their resultant anatomic distribution, phenotypic heterogeneity, transcriptional profile, and functional capacity. By generating a comprehensive map of human ILC composition and spatial dynamics in non-diseased tissues over life, our results provide insights that cannot be inferred from previous studies limited to few anatomic sites, small sample sizes, or variable disease states of donors. We anticipate that these datasets will aid in revealing critical insights into the role of ILCs in promoting human health and tissue-specific pathologies.

STAR Methods

Mice

C57BL/6 mice were purchased from the Jackson Laboratory (Bar Harbor, ME). Male and female mice analyzed were aged between 7–20 weeks. Within individual experiments, all animals were age- and sex-matched. Mice were maintained in specific pathogen-free facilities at Weill Cornell Medicine following protocols approved by the Weill Cornell Medicine Institutional Animal Care and Use Committee (IACUC).

Acquisition of human tissues

Human tissues were obtained from deceased organ donors at the time of organ acquisition for clinical transplantation through an approved research protocol and MTA with LiveOnNY (Thome et al., 2014; Granot et al., 2017; Kumar et al., 2017). All donors were free of chronic disease and cancer, were Hepatitis B, C, and HIV-negative. Tissues were collected after the donor organs were flushed with cold preservation solution and clinical procurement process was completed. The study does not qualify as “human subjects” research, as confirmed by the Columbia University IRB as tissue samples were obtained from brain-dead (deceased) individuals.

Murine lymphocyte isolation

For intestinal lamina propria lymphocyte preparations, intestines were isolated, attached fat removed, and tissues cut open longitudinally. Luminal contents were removed by shaking in cold PBS. Intraepithelial lymphocytes (IELs) were collected after incubation in 1 mM EDTA, 1 mM DTT, shaking for 30 min at 37°C. IELs were furth er purified using a 40% Percoll. The lamina propria lymphocytes were isolated by digesting the remaining tissue in 1 mg/ml Collagenase/Dispase (Roche, Nutley, NJ) and 0.1 mg/mL DNase I (Sigma-Aldrich) for 1 hr. at 37°C. The lungs were perfused with 10 ml PBS throug h the right ventricle of the heart prior to removal. Lungs were cut into small pieces and digested in Collagenase D (2 mg/ml) in RPMI Media for 45 min at 37°C shaking. The spleen, mLN a nd lung-draining LNs were forcefully passed through a 70 μm filter and spun down. iWAT and eWAT were minced finely and placed in Collagenase II (Sigma Aldrich, St Louis, MO) at a concentration of 1 mg/ml for 1 hr. shaking at 37°C. After isolation, all cells were passed thr ough a 70 μm cell strainer, spun down and any remaining red blood cells were lysed.

Human lymphocyte isolation

Spleen specimen were cleaned of fat and chopped into small pieces. The pieces were placed into 50 ml conical tubes with 10% FCS/RPMI Media, minced manually with scissors and were incubated by mechanical shaking at 37°C for 1 hr. A fter this incubation period it was forcefully passed through a stainless steel tissue sieve, subsequently through a 70 μm filter (Miltenyi, Bergisch-Gladbach) and pelleted through centrifugation. Centrifugation with Ficoll (GE Healthcare, Wauwatosa, Wisconsin) was used to separate lymphocytes. Residual red blood cells (RBC) were lysed using AKC lysis buffer, incubated for 5 min on ice and cells were washed with RPMI media. The surplus of cells recovered from all organs were frozen down in ice cold 90% DMSO+10% FCS.

For intestinal specimens, the fat was removed, the specimen was cut into small pieces and was cleaned with ice cold PBS by vigorous vortexing. The specimen was incubated in 1mM DTT (Sigma Aldrich, St Louis, MO) and 1 mM EDTA (Invitrogen, Grand Island, NY) for 1 hr, shaking at 37°C. The supernatant was collected, filtered th rough a 70 μm filter (Miltenyi, Bergisch-Gladbach), spun down and IELs were recovered by 40% Percoll. Enzymatic digestion media (collagenase D (Roche, Nutley, NJ) [2 mg/ml], trypsin inhibitor (life technologies, Carlsbad, CA) [1 mg/ml], and DNase I [0.1 mg/ml]), was injected submucosally into the intestinal samples and inflated tissues were kept at 37°C for 30 min after which tissues were minced manually with scissors and incubated shaking at 37°C for another 1 hr. After this incubation period the residual tissue was passed through a stainless steel tissue sieve, subsequently through a 70 μm filter (Miltenyi, Bergisch-Gladbach) and pelleted through centrifugation. The surplus of cells recovered from all organs were frozen down in ice cold 90% DMSO+10% FCS.

Lung tissues were minced manually into 50 ml conical tubes with scissors and specimen were incubated shaking at 37°C in enzymatic digestion me dia (as described above). After this incubation period the residual tissue was passed through a stainless steel tissue sieve, subsequently through a 70 μm filter (Miltenyi, Bergisch-Gladbach) and pelleted through centrifugation. Residual red blood cells (RBC) were lysed using AKC lysis buffer, incubated for 5 min on ice and cells were washed with RPMI media. The surplus of cells recovered from all organs were frozen down in ice cold 90% DMSO+10% FCS.

Mesenteric LN and lung-draining LN were minced finely and placed in collagenase D (Roche, Nutley, NJ) [2 mg/ml], trypsin inhibitor (life technologies, Carlsbad, CA) [1 mg/ml], and DNase I [0.1 mg/ml]) for 1 hr shaking at 37°C. After the di gest the cell suspension was filtered through a 70 μm filter (Miltenyi, Bergisch-Gladbach) and pelleted through centrifugation. Residual red blood cells (RBC) were lysed using AKC lysis buffer, incubated for 5 min on ice and cells were washed with RPMI media. The surplus of cells recovered from all organs were frozen down in ice cold 90% DMSO+10% FCS.

Mesenteric and abdominal fat were cleared from as much connective tissue as possible, minced finely and placed in Collagenase II (Sigma Aldrich, St Louis, MO) at a concentration of 1 mg/ml for 1 hr shaking at 37°C. The obtained cell suspens ion was passed through a 70 μm filter (Miltenyi, Bergisch-Gladbach) and spun down. After centrifugation residual fat (floating on the top) was discarded and red blood cells (RBC) were lysed using AKC lysis buffer, incubated for 5 min on ice and cells were washed with RPMI media. The surplus of cells recovered from all organs were frozen down in ice cold 90% DMSO+10% FCS.

Flow cytometric analysis and cell sorting

Murine and human cells were stained with LIVE/DEAD Fixable Aqua Dead Cell Stain Kit (405 nm) (life technologies, Carlsbad, CA) for 20 min in PBS on ice, washed and consecutively stained with fluorophore-conjugated antibodies against surface markers. No pre-enrichment kits were used. For intracellular staining, cells were fixed and permeabilized utilizing a commercially available kit (eBioscience) and stained with fluorophore-conjugated monoclonal antibodies against transcription factors. After staining cells were washed and fixed in 2% paraformaldehyde (Bioworld, Dublin, OH) for flow cytometric analysis. For cell sorting, cells were kept on ice in 1% FCS in PBS with RNaseOUT Recombinant Ribonuclease Inhibitor (Life Technologies, Waltham, MA) and single, viable cells of interest were sorted using a BD Aria II. A complete list of antibodies used, as well as individual clone and source information, can be found in the Key Resources Table.

KEY RESOURCES TABLE

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
Anti-Human CD11B Alexa Fluor 700	eBioscience	M1/70
Anti-Human CD11C Alexa Fluor 700	eBioscience	3.9
Anti-Human CD16 Alexa Fluor 700	BD Pharmingen	3G8
Anti-Human CD14 Alexa Fluor 700	Biolegend	HCD14
Anti-Human CD19 Alexa Fluor 700	eBioscience	HIB19
Anti-Human CRTH2 Alexa Fluor 647	BD Pharmingen	BM16
Anti-Human CD127 APC-eFluor780	eBioscience	eBioRDR5
Anti-Human CD56 Brilliant Violet 421	Biolegend	HCD56
Anti-Human T-BET Brilliant Violet 421	Biolegend	4B10
Anti-Human CD45 Brilliant Violet 605	Biolegend	HI30
Anti-Human EOMES FITC	eBioscience	WD1928
Anti-Human GATA3 PE	BD Pharmingen	L50–823
Anti-Human NKp44 PE	Biolegend	P44–8
Anti-Human RORgT PE	BD Pharmingen	Q21–559
Anti-Human CD4 PE-TR	Invitrogen	S3.5
Anti-Human CD3 PE-Cy7	eBioscience	UCHT1
Anti-Human CD5 PE-Cy7	eBioscience	UCHT2
Anti-Human FCERI PE-Cy7	eBioscience	AER-37
Anti-Human CD117 PerCP-eFluor710	eBioscience	104D2
Anti-Human CD161 FITC	Biolegend	HP-3G10
Anti-Human CD94 PE	Biolegend	DX22
Anti-Human NKp80 APC	Biolegend	5D12
Anti-Human CD14 Brilliant Violet 650	Biolegend	M5E2
Anti-Human CD11B Brilliant Violet 650	Biolegend	ICRF44
Anti-Human CD11C Brilliant Violet 650	Biolegend	3.9
Anti-Human CRTH2 Brilliant Violet 711	Biolegend	BM16
Anti-Human NKp44 BUV395	BD Pharmingen	P44–8
Anti-Mouse Eomes Alexa Fluor 488	eBioscience	TBR2
Anti-Mouse Cd90.2 Alexa Fluor 700	Biolegend	30-H12
Anti-Mouse B220 APC-eFluor780	eBioscience	RA3–6B2
Anti-Mouse Cd11b APC-eFluor780	eBioscience	M1/70
Anti-Mouse Cd11c APC-eFluor780	eBioscience	N418
Anti-Mouse Cd127 Brilliant Violet 421	Biolegend	A7R34
Anti-Mouse Cd45 Brilliant Violet 605	Biolegend	30-F11
Anti-Mouse Cd4 Brilliant Violet 650	Biolegend	RM4–5
Anti-Mouse T-bet eFluor 660	eBioscience	4B10
Anti-Mouse Rorgt PE	eBioscience	B2D
Anti-Mouse Cd3e PE-Cy7	eBioscience	145–2C11
Anti-Mouse Cd5 PE-Cy7	eBioscience	53–7.3
Anti-Mouse FcerI PE-Cy7	Biolegend	42430
Anti-Mouse Nkp46 PE-eFluor 610	eBioscience	29A1.4
Anti-Mouse Gata3 PerCP-eFluor710	eBioscience	TWAJ
Bacterial and Virus Strains
Biological Samples
Chemicals, Peptides, and Recombinant Proteins
Chloroform	Sigma-Aldrich	496189
MEM α (Minimum Essential Medium α)	ThermoFisher	12561072
RNaseOUT	ThermoFisher	10777–019
ZombieUV Fixable Viability Kit	Biolegend	423107
Critical Commercial Assays
Deposited Data
Raw and analyzed data	This paper	GEO:
Experimental Models: Cell Lines
Experimental Models: Organisms/Strains
Oligonucleotides
Recombinant DNA
Software and Algorithms
STARaligner	https://www.github.com/alexdobin/STAR	STAR: ultrafast universal RNA-seq aligner. Dobin A, Davis CA, Schlesinger F, Drenkow J, Zaleski C, Jha S, Batut P, Chaisson M, Gingeras TR. Bioinformatics. 2013 Jan 1;29(1):15–21. doi: 10.1093/bioinformatics/bts635. Epub 2012 Oct 25. PMID: 23104886
rSubread		Liao Y, Smyth GK and Shi W (2013). “The Subread aligner: fast, accurate and scalable read mapping by seed-and-vote.” Nucleic Acids Research, 41, pp. e108.
DEseq2		Love MI, Huber W, Anders S. 2014. Moderated estimation of fold change and dispersion for RNA-seq data with DESeq2. Genome Biology 15:550.
NetworkAnalyst		Xia J, Gill E, and Hancock REW (2015) “NetworkAnalyst for Statistical, Visual and Network-based Approaches for Meta-analysis of Expression Data” Nature Protocols 10, 823–844
Enrichr		Chen EY, Tan CM, Kou Y, Duan Q, Wang Z, Meirelles GV, Clark NR, Ma’ayan A. Enrichr: interactive and collaborative HTML5 gene list enrichment analysis tool. BMC Bioinformatics. 2013;128(14).
Degust	Degust by David R. Powell	http://victorian-bioinformatics-consortium.github.io/degust/
STRINGdb 10.0	STRING Consortium 2017	Szklarczyk D, Franceschini A, Wyder S, Forslund K, Heller D, Huerta-Cepas J, Simonovic M, Roth A, Santos A, Tsafou KP, Kuhn M, Bork P, Jensen LJ, von Mering C. STRING v10: protein-protein interaction networks, integrated over the tree of life.Nucleic Acids Res. 2015 Jan; 43:D447–52.https://string-db.org/
Rstudio version 1.1.442	RStudio, Inc. (2018)	https://www.rstudio.com
R version 3.4.2	R Foundation for Statistical Computing (2017)	https://www.R-project.org
IPA	Qiagen	Causal analysis approaches in Ingenuity Pathway Analysis. Bioinformatics. 2014 Feb 15;30(4):523–30. https://www.qiagenbioinformatics.com/products/ingenuity-pathway-analysis

RNAseq

For each sample, 100 cells/well were sorted into TCL buffer (Qiagen Germantown, MD,USA) in a Twin.tec PCR 96-well plate (Eppendorf; Hauppauge, NY, USA). Illumina SmartSeq2 RNA libraries were prepared by the Broad Institute, and paired-end sequencing was performed on an Illumina HiSeq at depth of 50M reads. Alignment and transcript counting was performed using STAR aligner and FeatureCounts. Linear dimensionality reduction and sample clustering were performed by principal component analysis (PCA). Because bulk RNAseq data is generated from pooled cells of a few individuals, PCA is generally preferred over non-linear clustering techniques, such as tSNE, which generate more distinct “clusters” and are often used for scRNAseq data. Normalized read counts and differential gene expressions were quantified using the R-based package DESeq2. Genes were defined as statistically significant at a nominal p-value < 0.001 and a Benjamini-Hochberg adjusted FDR value of q <0.1. Functional network and pathway analysis was performed for statistically significant differentially expressed genes using DAVID (NIAID, NIH), KEGG (Kanehisa Laboratories), and STRING 10.0 (SIB, CPR, EMBL).

Statistical analysis and visualization

All samples were acquired using BD Fortessa or BD LSRII (both BD Biosciences) and data were analyzed using FlowJo software (Tree Star Inc.). All statistical tests were selected with consideration to sample distribution and variance. Descriptive statistics (percent means, standard deviations, counts) were calculated for each cell subset and tissue in Microsoft Excel. Frequency variance was determined for each subset and tissue by Holm-Sidak post-hoc correction for multiple comparisons following two-way ANOVA to exclude subset-dependent effects in GraphPad PRISM (Graphpad software, Inc., La Jolla, CA). Frequency comparison p-values for each tissue versus each remaining tissue were graphed using Microsoft Excel. Correlation analysis of subset distribution vs. age and BMI was calculated in GraphPad PRISM by nonparametric Spearman correlation for non-Gaussian distributions. Resulting two-tailed p-values and R-values were graphed in Microsoft Excel and using GraphPad Prism.

Highlights.

1. In contrast to mice, human ILCs are less strictly compartmentalized

2. ILC subset composition is differentially impacted by tissue localization

3. Tissue environment drives transcriptional heterogeneity in a subset-dependent way

4. ILC1 exhibit greater transcriptional heterogeneity in mucosal vs. lymphoid sites