Differential Integrin Adhesome Expression Defines Human NK Cell Residency and Developmental Stage

This article is featured in Top Reads, p.751

Human NK cells are commonly defined as CD56⁺CD3⁻ cytotoxic innate lymphocytes, and they play a critical role in identifying and killing virally infected or malignant cells. The importance of NK cells in the control of viral infections is underscored by the clinical course of patients with NK cell deficiencies who experience severe and often life-threatening viral infections (1–3). In addition to circulating NK cells found in peripheral blood (PB), tissue-resident NK cell populations are present in organs including liver, lung, spleen, bone marrow, and secondary lymphoid tissue, where they serve unique cytotoxic and regulatory functions (4, 5).

NK cell maturation is marked by the progressive gain of NK cell–associated receptors and functions, and NK cell developmental subsets can be defined as stages 1–6, which are stages of maturation that represent unique cell phenotypes and lineage potentials (6–13). The stage 4 NK cell subset can be further delineated to stages 4A and 4B by expression of NKp80 at stage 4B (10). The predominant NK cell subsets in PB are stages 4B, 5, and 6, often defined as CD56^bright, CD56^dim, and terminally mature CD56^dim, respectively; however, circulating NK cell and innate lymphoid cell precursors are also found at low frequencies (7–9, 14). CD34⁺ NK progenitor cells are thought to enter circulation from bone marrow and subsequently seed peripheral sites to continue NK cell maturation; functionally mature NK cells then re-enter circulation in PB as stage 4B or stage 5 effectors (7). As such, stage 5 cells predominate in circulation, express perforin and granzymes at baseline, and are considered poised for cytolytic function following recruitment to sites of infection or inflammation. Tissue-resident NK cells are generally considered to be stage 4 (CD56^bright) NK cells, yet have distinct phenotypes from PB stage 4 cells and are thought to perform more regulatory functions (4, 8). The differential expression of chemokine receptors and adhesion molecules on NK cells in PB and tissue have been implicated in the mechanisms that mediate tissue localization and homing of distinct NK cell subsets (4, 5).

Integrins act as bidirectional signaling hubs between cellular machinery, namely the actin cytoskeleton and the extracellular microenvironment. As such, integrins play particularly critical roles in lymphocyte activation, immune synapse formation, and B, T, and NK cell development (15–20). Integrin function is finely tuned and can be regulated by changes in expression, localization, and affinity. Their specificity for ligand is dictated in part by the pairing of 18 α subunits and 8 β subunits to form at least 24 unique noncovalently linked obligate heterodimers. NK cells express leukocyte-specific β₂ integrins that mediate cell–cell interactions, as well as those that are more broadly expressed and facilitate cell–matrix interactions, including β1 and β7 integrins (21–27).

Previous studies have shown that VLA-4 (α4β1, CD49d/CD29) and VLA-5 (α5β1, CD49e/CD29) heterodimers are expressed by NK cells found in PB and mediate NK cell adhesion to fibronectin (28). LFA-1, the αLβ2 heterodimer (CD11a/CD18), and Mac-1, the αMβ2 heterodimer (CD11b/CD18), mediate NK cell adhesion to ICAM-1 on target cells, leading to signaling and actin remodeling that stabilize the immunological synapse of NK cells (24–26). Furthermore, a subpopulation of NK cells in human PB are detected with LFA-1 in a partially activated conformation, suggesting that conformational regulation, in addition to expression, is a feature of integrin phenotypes (26). Differences in tissue residency and function between NK cell developmental subsets suggest that distinct integrin repertoires mediate critical functions in lymphocyte development, activation, and migration. However, the expression patterns of integrins and their associated signaling networks from NK cell subsets have not been comprehensively described.

The integrin adhesome is the in silico molecular catalog of signaling protein interactions that occur in response to integrin engagement with ligands, including extracellular matrix (ECM) components and cell adhesion molecules (29–34). Although integrin adhesion complexes are more dynamic in lymphocytes than in larger, slower-moving cells, components of the integrin adhesome are conserved and mediate critical functions in lymphocyte development, trafficking, and activation (29–32). Using RNA sequencing (RNA-Seq) and phenotypic analyses, we sought to define adhesome expression within human NK developmental subsets from tonsil, a known site of NK cell development, PB, and in vitro–derived NK cells. We found that differences in the site of isolation and NK cell developmental intermediates are associated with unique profiles of integrins, actin regulators, and other signaling intermediates. We additionally revealed significant differences in the density of cortical actin between NK cell developmental subsets in PB, thus linking integrin-mediated actin signaling networks with inherent differences in actin networks.

All human tissues used in the RNA-Seq studies were collected under a protocol approved by The Ohio State University Institutional Review Board. Human pediatric tonsils were obtained fresh through the Cooperative Human Tissue Network from Nationwide Children’s Hospital (Columbus, OH), and PB was obtained through the American Red Cross as previously described (10, 35). Single-cell suspensions from blood and tonsil samples were enriched for NK lineage cells using a bivalent Ab RosetteSep (STEMCELL Technologies)–based method (36), and then the resultant enriched NK cell fractions were labeled with Abs (see Table I for a complete list of flow Abs) and finally sorted to purity using a BD FACSAria II cell sorter. Purities were validated postsort, and all samples had purity >99%. Human tonsil NK cell subsets were defined and sorted as follows: stage 3 (Lineage [Lin]^–CD117⁺CD94^–NKp80^–CD16^–), stage 4A (Lin^–CD94⁺NKp80^–CD16^–), stage 4B (Lin^–CD94⁺NKp80⁺CD16^–), and stage 5 (Lin^–NKp80⁺CD16⁺); human blood NK cell subsets were defined and sorted as follows: stage 4B (Lin^–NKp80⁺CD16^–CD57^–), stage 5 (Lin^–NKp80⁺CD16⁺CD57^–), and stage 6 (Lin^–NKp80⁺CD16⁺CD57⁺). For these sorting experiments, Lin = CD3, CD14, CD19, CD20, and CD34.

For flow cytometry, whole blood was obtained by venipuncture from healthy donors or as discarded apheresis product from patients undergoing routine RBC exchange at Columbia University Medical Center. Alternatively, we acquired RBC-low, leukocyte-enriched buffy coats from the New York Blood Center as an alternative source of primary NK cells for flow cytometric analysis. Primary NK cell subsets from blood were enriched with NK cell RosetteSep (STEMCELL Technologies). Blood was layered onto Ficoll-Paque density gradient, followed by centrifugation at 2000 rpm for 20 min (no brake). NK cells were collected from the density gradient interface and washed with PBS by centrifugation at 1200 rpm for 7 min. NK cells were resuspended in PBS 10% FCS and counted, then either resuspended in PBS for flow cytometry or cryopreserved in FCS with 10% DMSO at a concentration of 1 × 10⁶ to 2.5 × 10⁶ cells/ml.

Tonsil samples for dissociation and flow cytometry analysis were acquired from routine tonsillectomies performed on pediatric patients at Columbia University Irving Medical Center. Tissue samples were placed in a sterile dish with PBS and manually dissociated by mincing into a cell suspension. The cell suspension was then passed through a 40-μm filter to obtain a single-cell suspension and washed with PBS by centrifugation at 1200 rpm for 7 min. Cells were either resuspended in PBS for flow cytometry or FCS 10% DMSO freezing media at a concentration of 5 × 10⁶ to 10 × 10⁶ cells/ml and cryopreserved prior to use.

CD34⁺ precursors for in vitro experiments were isolated from whole blood obtained by venipuncture from healthy donors. Mononuclear blood cells were isolated from donors were incubated with anti-CD34 Ab (clone 581; BioLegend) prior to cell sorting. CD34⁺ cells were isolated by FACS sorting on a BD Aria II cytometer with an 85-μm nozzle at 45 pounds per square inch. Sorted cells were confirmed to be >90% CD34⁺ and were cultured directly after isolation on previously irradiated EL08.1D2 or OP9 cells as described below.

For structured illumination microscopy, primary NK cells were enriched from PB of healthy human donors using RosetteSep (STEMCELL Technologies). Primary cells, apheresis, and tonsils were obtained in accordance with the Declaration of Helsinki with the written and informed consent of all participants under the guidance of the institutional review boards of Ohio State University and Columbia University.

EL08.1D2 cells were a gift from Dr. J. Miller (University of Minnesota) and were cultured as previously described (37) in culture flasks pretreated with 0.1% gelatin. Cells were maintained at 32°C in 40.5% α-MEM (Life Technologies), 50% MyeloCult M5300 (STEMCELL Technologies), 7.5% heat-inactivated FCS (Atlanta Biologicals) with β-mercaptoethanol (1 × 10⁻⁵ M), GlutaMAX (2 mM; Life Technologies), penicillin/streptomycin (100 U/ml; Life Technologies), and hydrocortisone (1 × 10⁻⁶ M; Sigma-Aldrich), supplemented with 20% conditioned media. OP9 cells (American Type Culture Collection) were cultured in nongelatinized culture flasks at 37°C in α-MEM with 20% heat-inactivated FCS (Atlanta Biologicals) and 1% penicillin/streptomycin (100 U/ml; Life Technologies). Prior to in vitro NK cell differentiation, 1 × 10⁴ EL08.1D2 cells were seeded into 96-well, flat-bottom plates precoated with 0.1% gelatin, whereas OP9 cells were similarly seeded into nongelatinized, 96-well, flat-bottom plates. Cells were grown to confluence then subjected to mitotic inactivation by irradiation at 300 Gy.

Following FACS sorting, CD34⁺ cells were cultured in NK cell differentiation media containing Ham F12 media plus DMEM (1:2) with 20% heat-inactivated human AB serum, ethanolamine (50 μM), ascorbic acid (20 mg/ml), sodium selenite (5 μg/ml), β-mercaptoethanol (24 μM), and penicillin/streptomycin (100 U/ml) in the presence of IL-15 (5 ng/ml), IL-3 (5 ng/ml), IL-7 (20 ng/ml), stem cell factor (20 ng/ml), and Flt3L (10 ng/ml) (all cytokines from PeproTech). CD34⁺ cells were seeded onto irradiated EL08.1D2 or OP9 at a density of 2 × 10³ cells per well and incubated at 37°C with weekly half media exchanges.

Freshly sorted blood and tonsil NK cells were pelleted, and total RNA was isolated using the Qiagen RNeasy Mini Kit (Active Motif; QIAGEN). Directional poly-A RNA-Seq libraries were prepared and sequenced as 42-bp paired-end reads on an Illumina NextSeq 500 instrument (Illumina) to a depth of 33.2 × 10⁶ to 48.0 × 10⁶ read pairs (Active Motif). Alignment to human genome (hg19 build) was done using TopHat. Transcriptome assembly and analysis was performed using Cufflinks, and expression was reported as fragments per kilobase of transcript per million mapped reads (FPKM).

FPKM gene expression data were obtained from prefiltered and normalized bulk RNA-Seq raw data and imported to iDEP 0.9 (38). FPKM bulk RNA-Seq data were processed using the source code available on iDEP 0.9 (38) following the recommended parameters. Pathway analysis was performed on iDEP 0.9 (38) transformed RNA-Seq data using Parametric Gene Set Enrichment Analysis ranking and Kyoto Encyclopedia of Genes and Genomes pathways (38, 39), filtering on pathways with 15–2000 genes and false discovery rate cutoff of 0.2. Prism 8.0 (GraphPad Software) was used to visualize data. The data discussed in this publication have been deposited in the National Center for Biotechnology Information’s Gene Expression Omnibus and are accessible through GEO Series accession number GSE169646.

Flow cytometry to quantify integrin expression was performed using Abs listed in Table I. Cryopreserved primary NK cells were thawed and resuspended in RPMI 1640–10% FCS, then immunostained. For intracellular phalloidin staining cells were first incubated with Abs for surface receptors, fixed, and permeabilized using CytoFix/CytoPerm (BD Biosciences), then incubated with directly conjugated phalloidin. Data were acquired on a Bio-Rad ZE5 Cell Analyzer, then exported to FlowJo 10 (BD Biosciences) for analysis. Integrin subunit mean fluorescence intensity (MFI) of primary tonsil and PB NK cell subsets was used to directly compare integrin expression of populations of cells collected on the same day. Primary NK developmental subsets were identified and analyzed using the following gating strategy: stage 3 (CD45⁺Lin^–CD34^–CD117⁺CD94^–), stage 4A (Lin^–CD94⁺CD56^brightNKp80^–), stage 4B (Lin^–CD94⁺CD56^brightNKp80⁺CD16^–), stage 5 (Lin^–CD56^dimCD16⁺CD57^–), and stage 6 (Lin^–CD56^dimCD16⁺CD57⁺); Lin = CD3, CD14, and CD19. For flow cytometry of in vitro–derived NK cells, cells were isolated at weekly time points and immunostained. In vitro NK developmental subsets were identified and analyzed using the following gating strategy: stage 3 (CD45⁺Lin^–CD34^–CD117⁺CD94^–), stage 4 (Lin^–CD94⁺CD56⁺CD16^–), and stage 5 (Lin^–CD56⁺CD16⁺); Lin = CD3, CD14, and CD19. The MFI of cells positive for activated integrin β1 and β2 was measured to calculate the density of activated integrins. A similar approach was implemented for cases in which cells were not uniformly positive for a particular integrin or if bimodal expression was observed. Fluorescence minus one controls were used as negative controls to calculate the positive and negative gates for flow cytometric analysis (40). Data were plotted and statistical analysis was performed using Prism 8.0 (GraphPad Software).

Freshly isolated cells were preincubated with anti-CD56 Alexa Fluor 647 (clone HCD56, 1:100; BioLegend) for 20 min in a conical tube then incubated on no. 1.5 coverslips that had been precoated with poly-l-lysine for an additional 20 min at 37°C 5% CO₂. Following incubation, cells were fixed on the coverslip with BD CytoFix/CytoPerm for 10 min. Cells were gently washed with PBS 2% FCS containing 0.1% saponin, then incubated for 30 min with phalloidin Alexa Fluor 568 (1:100; Thermo Fisher Scientific). Coverslips were gently washed again, then mounted with Prolong Gold (Thermo Fisher Scientific).

Imaging was performed on a GE DeltaVision OMX SE in three-dimensional simulation mode through a 60× 1.42 numerical aperture apochromat objective. Three-dimensional images were captured at 125-nm steps, and the pixel size was 79 nm. Images were reconstructed using three orientations and five phase shifts with a Wiener filter constant of 0.001. Acquisition and reconstruction were performed with GE SoftWoRX software, and images were exported to Fiji (41) for further analysis. Calculation of integrated density was performed by applying a uniform default Auto Threshold in Fiji, then measuring area, intensity, and integrated density (area × MFI) for CD56 and actin. Data were graphed, and statistics were calculated in Prism 8.0 (GraphPad Software).

To test the normal distribution of flow cytometric data, we used Shapiro–Wilk test, based on our sample size, with a p value cutoff of 0.05. Our MFI and cell frequency data fell under a normal distribution. Thus, we used an ordinary one-way ANOVA with multiple comparisons to compare the MFIs or cell frequency of NK cell developmental subsets (Figs. 3–9). For nonparametric data, as determined by a Shapiro–Wilk test p value <0.05, we employed a Kruskal–Wallis test with multiple comparisons to compare the mean frequencies of CD29 activation of NK cell developmental subsets from tonsil. We also used a Kruskal–Wallis test with multiple comparisons test to compare the normalized means of all developmental stages and specifically stage 5 and 6 phalloidin intensity to that of stage 4B. Phalloidin intensity of PB and tonsil NK cells was normalized to the MFI of stage 3 cells within the respective sample. The p value cutoff for significance of all statistical tests comparing the means of samples was ≤0.05. Mann–Whitney tests were used to compare unpaired data with a nonnormal distribution identified by Shapiro–Wilk test. All statistical testing was performed in Prism 8.0 (GraphPad Software).

Although differences in integrin expression in NK cell developmental subsets have been previously reported (5, 26, 28), integrin adhesome expression in human NK cells has not been extensively cataloged. Using flow cytometry to sort phenotypically equivalent NK cell developmental subsets, we isolated stage 4B, 5, and 6 NK cells from PB and stage 3, 4A, 4B, and 5 NK cells from tonsils obtained from routine tonsillectomies performed on healthy children (8–10, 35). As CD57⁺CD56^dim stage 6 NK cells are found at very low frequencies in tonsils from healthy individuals, they were not included in our experimental design (Supplemental Fig. 1A). Similarly, PB stage 3 and 4A NK cells were excluded because of their rarity in circulation (10, 14). We pooled RNA from 12 donors into three technical replicates and performed bulk RNA-Seq. Principal component analysis (PCA) using the 18,475 genes detected by RNA-Seq revealed that both tissue residency and developmental stage are determinants of unique gene expression profiles (Fig. 1A). This analysis revealed that the overall gene expression is more significantly affected by tissue specificity (principal component [PC]1: 57% of variance) than the developmental stages (PC2: 12% of variance). As such, when phenotypically equivalent stage 4B and 5 cells from PB or tissue were compared, the two subsets from the same tissue clustered more closely than the two subsets of the same developmental stage (Fig. 1A, Supplemental Fig. 1B, 1C).

To further understand the genes contributing to the separation observed between PB and tonsil NK cells, we plotted the 18,475 genes used for PCA by their PC1 loading weights (Fig. 1A). Genes associated with the generally immature phenotype of tonsil NK cells, such as GZMK, KLRC1, and IL7R, were primarily represented by negative weights in PC1, consistent with the clustering of tonsil subsets with negative PC1 values (Fig. 1B). In contrast, PB NK cells were marked by positive PC1 weights of genes related to NK cytotoxicity, activation, and KIR receptors (B3GAT1, KLRB1, KIR3DL2, KIR2DL1, and IFNGR1) (Fig. 1B). This distribution of gene sets across PC1 suggested that the differences between the transcriptome of tonsil and PB NK cell subsets also reflected differences in the predominant stages of NK cell maturation found at these sites.

In addition, we found cytoskeleton-associated proteins were highly contributing to PC1, suggesting that these are among the genes significantly driving the differences between PB and tonsil NK cell subsets. Specifically, we noted integrins including ITGAD, ITGAE, and ITGA1 within the genes with the lowest PC1 weights (<−0.01), whereas other integrins and actin regulator proteins, such as ITGB2, PXN and ARF1, were within the genes with the highest PC1 weights (>0.01; (Fig 1B). To better understand how integrins were differentially expressed between tissue sites and developmental subsets, we identified the distribution of 229 consensus adhesome genes that have been described previously (29–32) (Supplemental Table I) across PC1 weights (Fig. 1B). Adhesome genes were significantly enriched in the negative end of PC1 (22 adhesome genes in the lowest 5%; (Fig. 1B). Further, Gene Ontology pathway analysis identified pathways associated with cell migration, including regulation of T cell migration, dendritic cell migration, and positive regulation of cell migration, that were associated with genes with low PC loading weights. These observations suggest that genes that are related to integrin-mediated adhesion and cytoskeletal remodeling are in part driving transcriptional differences between tonsil and PB NK cell subsets.

The integrin adhesome network has been described primarily in nonlymphocyte cells (29–32), yet our observations of PC1 weights indicated that distinct patterns of adhesome gene expression are also important in NK cell subsets. To further understand how adhesome genes specifically drive NK cell heterogeneity, we performed PCA only with the 229 consensus adhesome genes (29–32). Similar to the whole transcriptome PCA (Fig. 1A), PC1 separated tonsil NK cells from PB NK cells, whereas PC2 revealed a tonsil specific progression of adhesome gene expression, suggesting that tonsil NK cells undergo additional changes through development (Fig. 1C). Together, these data, including differential gene expression analysis and pathway analysis (Supplemental Fig. 1D, 1E), demonstrate that genes associated with the integrin adhesome are differentially regulated between both developmentally and spatially distinct NK cell subsets. Further, these findings suggest that adhesome gene expression changes are integral parts of NK cell maturation and their adaptation to new environments.

To further define developmental and tissue residency signatures of tonsil and PB NK cell adhesome genes, we identified five clusters (clusters A–E) of adhesome genes with K-means clustering (Fig. 2A, 2B). Cluster A contains genes that are highly expressed in PB NK cells compared with tonsil NK cells. Cluster A includes β1 integrin partners ITGA4 (CD49d), ITGA5 (CD49e), the actin regulator RAC1, and PALLD, a component of actin microfilaments that functions as an actin stabilizer (Fig. 2B, 2C). Similar to cluster A, cluster B includes genes that are highly expressed in PB NK cells, but an important distinction is that cluster B genes are upregulated with maturation. Leukocyte-specific integrins including ITGAL, ITGAM, ITGB2, and calpain 2 (CAPN2), which functions in focal adhesion disassembly (42), are in this cluster. In addition, cluster B reveals that mature PB stage 5 and 6 NK cells upregulate actin and cytoskeletal regulatory proteins like ABI3, ARF1, ARPC2, PXN, and PI3KCA (Fig. 2B, 2C). Cluster C is comprised of genes that are transiently expressed between tonsil stages 4B and 5 and PB stage 5. This includes genes related to the regulation of the cytoskeleton such as PTK2, GAB1, CORO1A, and ITGAX. Together, genes in clusters A–C show that PB NK cells are characterized by upregulation of actin regulatory proteins and leukocyte-associated integrins relative to their tonsil counterparts.

Tonsil NK cell subsets were predominantly defined by their preferential expression of adhesome genes belonging to clusters D and E (Fig. 2A, 2B). Cluster D distinguishes tonsil stage 3 and 4A NK cells from the more mature tonsil stage 4B and 5 as well as PB stages 4B–6 (Fig. 2A, 2B). The genes that constitute cluster D include ITGB7, signaling proteins PRKCA, PAK1, and ADAM12, a disintegrin and metallopeptidase involved in cell migration, proliferation, and invasion, and LRP1, which is predicted to regulate cell migration through Rho GTPases (43, 44) (Fig. 2B, 2C). Cluster E defines tonsil NK cells from PB NK cells (Fig. 2A) and is made up of adhesome genes associated with tissue residency. These genes include integrins ITGA1 and ITGAE, which encode for CD49a and CD103, respectively, and are specifically associated with NK cell tissue residency (21–23, 27) (Fig. 2B, 2C). Cluster E also reveals that tonsil NK cells preferentially express migration-associated signaling proteins RASA1, SRC, TIAM1, HSPB1, and VIM relative to PB NK cells (Fig. 2B, 2C). Finally, layilin (LAYN), which binds to talin and localizes to membrane ruffles, and neuropilin (NRP1), a transmembrane glycoprotein that has not been previously described to play a role in NK cell migration (45, 46), are also found in cluster E and are upregulated by tonsil NK cells relative to PB NK cells (Fig. 2B). Taken together, K-means clustering of adhesome genes reveals the unique gene expression signatures in tonsil and PB NK cell subsets that are associated with both tissue residency and developmental stage.

Our transcriptomic data suggested that there were significant differences in the expression of integrins between NK developmental subsets and sites of isolation. To define the cell surface expression of CD11a/CD18 (LFA-1), CD11b/CD18 (Mac-1), CD11c/CD18, CD103/integrin β7, CD49a/CD29 (VLA-1), CD49d/CD29 (VLA-4), and CD49e/CD29 (VLA-5) integrins that were differentially expressed between these populations, we performed flow cytometric analyses of single-cell suspensions from tonsil and PB donors (see Materials and Methods for gating strategy and Table I for Abs).

Our analysis of leukocyte-specific integrins revealed that PB stage 3–6 and tonsil stage 3–5 NK cells from all donors were found to have uniform cell surface expression of CD11a, CD11b, and CD11c β2 integrins (Fig. 3, histograms), with varying degrees of relative expression detectable by measuring MFI. Expression of CD11a (integrin αL) was found to be significantly higher in NK cell stages 4–5, relative to stage 3 in tonsil, and PB stages 4–6 had increased CD11a MFI relative to stage 3 NK cells (Fig. 3A, Supplemental Table II). Interestingly, our transcriptomic data suggest that ITGAL (integrin αL) is more highly expressed in PB stage 5, but not stage 4, relative to tonsil (Fig. 2C), yet flow cytometric analysis revealed that CD11a was more highly expressed on the surface of both PB stage 4 and 5 NK cells relative to the equivalent subsets in tonsil (Fig. 3A, Supplemental Table II).

In agreement with our transcriptomic data, which show increasing levels of integrin αM through PB NK cell maturation, analysis of surface expression of CD11b revealed that expression significantly increased between stage 3 and stages 4B, 5, and 6 of PB NK cells (Fig. 3B, Supplemental Table II). We similarly found that tonsil NK cell subsets also had a significant increase in the expression of CD11b between stages 3 and 5 (Fig. 3B, Supplemental Table II). Expression of CD11c (integrin αX) did not differ between tonsil NK cell stages (Fig. 3C), yet we observed a significant increase in CD11c between PB stages 3–4B (Fig. (3C), followed by a decrease in MFI in stage 5 cells, as suggested by transcriptomic data (Supplemental Table II). We consistently identified a subset of stage 3–5 NK cells in tonsil that had measurable density of CD103 (integrin αE) relative to the majority of their respective populations (Fig. 3D, 3E; Supplemental Table II). Although we did not detect CD103⁺ stage 4 and 5 NK cells in PB, we did detect CD103 on a small subset (<5%) of stage 3 PB NK cells; however, there was no expression of CD103 on PB NK cells from stages 4–6 (Fig. 3D, 3E, Supplemental Table II). Finally, we analyzed the expression of CD18 (integrin β2) and found that in accordance with transcriptomic data, CD18 was significantly upregulated in more mature NK cell subsets, namely stages 4B and 5 in tonsil and stages 5 and 6 in PB relative to stage 3 (Fig. 3F, Supplemental Table II).

Next, we sought to measure the cell surface expression of CD29 (integrin β1) and integrin β7 associated integrin subunits on phenotypically equivalent tonsil and PB NK cell subsets. As suggested by our transcriptomic data (Fig. 2C), PB NK cells did not significantly upregulate or express CD49a (integrin α1) (Fig. 4A, 4B; Supplemental Table II). However, unlike in PB, a subset of tonsil stage 3, 4A, 4B, and 5 NK cells expressed CD49a, which is associated with tissue residency (21, 23) (Fig. 4A). We also found that CD49a⁺ tonsil stage 4 and 5 NK cells had significantly increased CD49a density relative to the CD49a⁺ tonsil stage 3 NK cell population (Fig. 4B, Supplemental Table II) As expected, CD49a⁻ NK cells from both tonsil and PB had undetectable CD49a MFI (Fig. 4B). Both PB and tonsil NK cells were uniformly CD49d⁺ (integrin α4) through maturation, and we found consistently higher expression of CD49d on PB NK cells relative to NK cells in tonsil (Fig. 4C, Supplemental Table II). CD49e (integrin α5), a β1-associated integrin receptor for fibronectin domains, is preferentially expressed by less mature stage 3, 4A, and 4B NK cells and was significantly downregulated through NK cell development in tonsil and PB (Fig. 4D, Supplemental Table II). Further, as suggested by RNA-Seq data (Fig. 2, cluster A), flow cytometric analysis demonstrated that PB stage 4 and 5 NK cells expressed higher levels of CD49e than equivalent subsets from tonsil (Fig. 4D, Supplemental Table II). Moreover, both PB and tonsil NK cells downregulated CD29 as they matured from stage 3, with greater downregulation of CD29 (integrin β1) in PB than in tonsil NK cell subsets (Fig. 4E, Supplemental Table II). Integrin β7, which pairs with CD103 (Fig. 3D) or with CD49d (Fig. 4B), was found to have a high density of expression on a subset of stage 3 PB and tonsil NK cells (10–50%) (Fig. 4F, 4G). Although stage 4 NK cells had a relatively small population with detectable integrin β7 expression (<20%), stage 5 NK cells from both sites and stage 6 from PB had a higher percent (30–90%) of cells with uniformly low expression of integrin β7 (Fig. 4F, 4G). The density of integrin β7 on NK cells significantly decreased after stage 3 and then increased slightly after stage 4 in both PB and tonsil (Fig. 4G, Supplemental Table II). These data, in parallel to our transcriptomic data, demonstrate that PB and tonsil NK cells differentially regulate cell surface expression of integrins both in a developmental and tissue residency–dependent manner.

Based on the differential expression of integrins between NK cell subsets from PB and tonsil, we wanted to understand how NK cells from these respective tissues regulate integrin activation, which has implicit effects on cell behavior, shape, and state (30–32). We included mAb24 and HUTS-4 Abs for detection of open/extended conformation, or activated, CD11a-c/CD18 (LFA-1, MAC-1, and integrin αX/β2), and CD49a-f/CD29 (VLA-1–6), respectively (47–49) (Fig. 5). Tonsil and PB NK cells were both found to have a subset of cells with activated CD29 or CD18 heterodimers (Fig. 5A, 5B). Unlike CD18, the frequency of cells with activated CD29 heterodimers was higher in tonsil than in PB (Fig. 5A, 5B). Although the frequency of NK cells with detectable activated CD29 or CD18 heterodimers did not significantly vary between developmental stage (Fig. 5A, 5B, Supplemental Table II), the density of activated CD29 and CD18 heterodimers on HUTS4⁺ or mAb24⁺ cells varied between developmental subsets isolated from PB (Fig. 5C, 5D, Supplemental Table II). Specifically, when compared with stage 3 and 4A NK cells, more mature stage 5 PB NK cells had a significantly higher density of activated CD18 (Fig. 5C, Supplemental Table II). Similarly, tonsil stage 4A NK cells with detectable open-conformation CD18 heterodimers had a significantly higher density of activated CD18 relative to stage 3 NK cells (Fig. 5C, Supplemental Table II). In addition, tonsil NK cells with activated CD29 had a greater density (MFI) of activated CD29 heterodimers compared with analogous PB subsets, yet in PB, we observed a significant increase in the density of activated CD29 by MFI in stage 5 relative to PB stage 3 and 4A NK cells (Fig. 5D, Supplemental Table II). Taken together, our data suggest that NK cell subsets have differential expression and conformation of integrins on their cell surface, with unique integrin profiles that also reflect their tissue residency.

In vitro differentiation of NK cells from CD34⁺ precursors has been well described as a method of studying human NK cell development (50, 51). Although innate lymphocytes undergoing maturation in such systems are thought to progress through stages of differentiation similarly to those in situ (9, 13, 52, 53), the use of xenogenic stromal cells and exogenous cytokines in such systems make it difficult to define the role of the microenvironment in this process. Given the differences between analogous subsets of NK cells isolated from different tissues, we sought to define the expression of integrins on NK cells generated from in vitro differentiation. Primary human CD34⁺ cells from PB were isolated and cultured with EL08.1D2 or OP9 stromal cell lines in the presence of cytokines (FLT3L, SCF, IL-3, IL-7, and IL-15) (50, 51, 54, 55). Following 4 wk of in vitro NK differentiation, stage 4 and 5 NK cells represented a significant proportion of the CD45⁺ population (Supplemental Fig. 2A, 2B).

We performed flow cytometry with our integrin panel described above after 4 wk of differentiation on EL08.1D2 or OP9 stromal cells. At this time point, the frequency of stage 1 and 2 NK cells significantly decreases relative to mature stages 3–5, which represent over 50% of the culture; thus, only stage 3, 4, and 5 NK cells were analyzed (Supplemental Fig. 2A, 2B). We did not detect NKp80 or CD57 expression on in vitro–derived cells, which is consistent with previous findings that OP9 stromal cells do not support their expression (56); therefore, we did not distinguish between stage 4A and 4B or include stage 6 cells in our analysis.

In NK cells from tonsil and PB, the density of CD11a expression on CD11a⁺ cells significantly increased with maturation at stages 4 and 5 (Fig. 3A). Similarly, frequency and density of CD11a increased with maturation in both EL08.1D2 and OP9 in vitro conditions (Fig. 6A, 6B, Supplemental Fig. 2C, 2D). Interestingly, we observed a CD11a^low population, suggesting the presence of in vitro–specific CD11a^high and CD11a^low phenotypes (Fig. 6A, 6B). Further, stage 4 and 5 in vitro NK cells from OP9 conditions had slightly, but consistently, increased density (MFI) of CD11a expression relative to their equivalent subsets from EL08.1D2 condition (Fig. 6A, 6B, Supplemental Fig. 2C, 2D). In vitro–derived NK cells from EL08.1D2 and OP9 conditions were uniformly positive for CD11b in all observed stages (Fig. 6C, 6D; Supplemental Fig. 2C, 2D). A significant subset of in vitro–derived NK cells from both conditions expressed CD11c (Fig. 6E, 6F). As in vitro NK cells matured into later stages, the percentage of CD11c⁺ cells decreased, whereas the cell surface density of CD11c on individual cells did not change (Fig. 6E, 6F; Supplemental Fig. 2C, 2D). Although we failed to detect cells with high CD103 density as observed in tonsil, a significant subset of in vitro NK cells from EL08.1D2 and OP9 conditions had low, but measurable, CD103 expression (Fig. 6G, 6H). Unlike in PB and tonsil, CD103 was significantly upregulated on stage 5 in vitro–derived NK cells (Fig. 6G, 6H; Supplemental Fig. 2C, 2D). In parallel with CD11a expression, the frequency of CD18⁺ cells and density of CD18 (integrin β2) was significantly increased in stage 4 and 5 in vitro NK cells relative to stage 3 regardless of the in vitro stromal cell condition, and we noted higher density of CD18 on stage 4 and 5 cells from the OP9 condition (Fig. 6I, 6J; Supplemental Fig. 2C, 2D).

Next, we analyzed the expression of β1- and β7-associated integrins on in vitro–derived NK cells from EL08.1D2 and OP9 conditions. In vitro stage 3 NK cells from EL08.1D2 and OP9 conditions had heterogenous expression of CD49a, but by stages 4 and 5, almost all cells became CD49a⁺ (Fig. 7A, 7B, Supplemental Fig. 2C, 2D). CD49d was detected on in vitro NK cells from both conditions, yet the frequency of CD49d⁺ NK cells, particularly from OP9 cultures, decreased after stage 3 (Fig. 7C, 7D, Supplemental Fig. 2C, 2D). Similar to CD49d, the frequency of CD49e⁺ cells significantly decreased after stage 3 in the NK cells on OP9, but not on EL08.1D2 (Fig. 7E, 7F, Supplemental Fig. 2C, 2D). Unlike in PB and tonsil, we did not detect changes in CD29 density among in vitro NK cell developmental subsets; all cells were uniformly CD29⁺ (Fig. 7G, 7H, Supplemental Fig. 2C, 2D). A subpopulation of stage 3 cells expressed integrin β7, whereas in vitro stage 4 and 5 NK cells from EL08.1D2 and OP9 cultures were predominantly positive for integrin β7 (Fig. 7I, 7J, Supplemental Fig. 2C, 2D). With all β1- and β7-associated integrins, the differences were observed in frequencies, rather than densities (MFI), at the single-cell level (Fig. 7, Supplemental Fig. 2C, 2D).

To understand the conformational profile of integrins on in vitro–derived NK cells, we next quantified the relative density of activated conformation CD18 and CD29 heterodimers. The frequency of in vitro NK cells with activated CD18 heterodimers increased through maturation in both EL08.1D2 and OP9 conditions (Fig. 8A, 8B). We did not detect a significant difference in the percentage of in vitro NK cells with activated CD29 heterodimers between developmental subsets (Fig. 8C, 8D, Supplemental Fig. 2C, 2D). The MFIs of activated conformations followed a similar trend, showing an increase in open-conformation CD18 heterodimers and no significant change in the density of open-conformation CD29 heterodimers through development (Fig. 8A–E, Supplemental Fig. 2C, 2D). Notably, we did not observe a significant difference in integrin expression or conformational activation between in vitro–derived NK cells that were generated by culture with OP9 or EL08.1D2 feeder cells.

Finally, given the changes in integrin adhesome expression between developmental subsets, we sought to define the density of cellular actin in NK cells generated in vitro. We found that as NK cells matured, they exhibited a significant increase in density of actin content following cell permeabilization as measured by intracellular detection of phalloidin by flow cytometry (Fig. 8F). Together, our results indicate that in vitro–derived NK cells have conserved patterns of integrin expression and conformation compared with ex vivo NK cells while exhibiting distinct actin content in a stage-specific manner.

Given that flow cytometric analysis of in vitro–derived NK cells unexpectedly revealed increased intensity of actin in mature NK cells, we sought to investigate the nature of cortical actin in ex vivo NK cells. Components of the actin cortex include actin, myosin, and regulators of actin polymerization and turnover, including the Arp2/3 complex and formins. Pools of actin monomers provide substrate for the generation and maintenance of the actin cortex, as well as the leading edge and immune synapse. Nonmuscle mammalian cells contain two structurally similar isoforms of actin monomers, γ-actin and β-actin, encoded by the ACTG1 and ACTB genes, respectively (57, 58). Given the ubiquitous nature of actin within all cells, it was unsurprising that we found that although there was a trend toward increased ACTB expression in more terminally mature NK cell subsets, there was no significant difference in expression of ACTB or ACTG1 between the isolated subsets of human NK cells in our bulk RNA-Seq dataset (Fig. 9A).

To characterize the nature of actin in PB and tonsil NK cells, we performed intracellular detection of phalloidin by flow cytometry as we had previously done for in vitro–derived NK cells (Fig. 8F). Strikingly, we found that there was a clear demarcation in actin content between PB stage 4 and 5 NK cells (Fig. 9B). More extensive analysis of PB NK cell subsets showed that stage 4A and 4B NK cells had significantly decreased intensity of phalloidin than stage 3, 5, and 6 NK cells (Fig. 9C). In contrast to PB, tonsil stage 4A, 4B, and 5 NK cells all had decreased cortical actin density when compared with stage 3 cells (Fig. 9C).

Given the differences we found by flow cytometry, we sought to directly visualize and measure the cortical actin network in freshly isolated PB NK cells. We performed superresolution structured illumination microscopy of actin by phalloidin detection in freshly isolated stage 4 and stage 5 cells from PB. Imaging confirmed our flow cytometry results, and we observed a significantly greater density of cortical actin in more mature cells (Fig. 9D). As predicted, CD56 intensity measured by integrated density (MFI × area) was significantly greater in stage 4 CD56^bright NK cells than stage 5 CD56^dim NK cells (Fig. 9E). Similar measurements of actin further validated our observations as we found that integrated density of actin was significantly greater in stage 5 NK cells (Fig. 9F). Although such measurements could be reflective of differences in cell size, we found that CD56^dim NK cells did not have a significantly greater volume than CD56^bright cells (Fig. 9G). Therefore, stage 5 NK cells have greater actin density relative to stage 4 cells, and this is independent of cell size or activation.

Finally, we sought underlying differences in gene expression that could drive the differences in actin density that we observed between NK cell developmental subsets. Further analysis of our RNA-Seq datasets identified key actin nucleation–promoting factors and cytoskeletal regulators that were differentially expressed, including ARPC2 (the ARPC2 subunit of the Arp2/3 complex) and profilin (PFN1) (Fig. 9H). Therefore, although there are a number of adhesome-related genes that function in adhesion and migration that have different patterns of differential expression between tissue and PB, actin nucleating pathways seem to be specifically upregulated in terminally mature NK cells in PB. This is reflected by a greater density of cortical actin that is detectable in the absence of cellular activation and not linked to expression of actin monomers ACTB and ACTG1.

Our data suggest that NK cells receive both intrinsic and extrinsic cues that dictate integrin adhesome expression. Tonsil NK cells are representative of secondary lymphoid tissue-resident NK cells (9), suggesting that the expression of adhesome components involved in migration and ECM interactions, such as collagen-specific integrins, are preferentially expressed in these tissues relative to PB. Conversely, PB NK cells upregulate integrin adhesome components that promote cellular integrity in an environment with constant shear flow and enable cells to bind endothelial cells lining blood vessels during extravasation (48).

Previous studies have found that ITGA1, ITGAE, and ITGAD, which are all upregulated in tonsil NK cell populations relative to PB NK cells, are associated with lymphocyte tissue residency and migration (5, 23, 59). ITGAE (CD103) is upregulated in tissue-resident memory CD8⁺ T cells from lung and spleen, as well as in CD56^brightCD16^– NK cells from tissue sites such as lung, endometrium, nasal mucosa, and intestine, thus providing support for our observations that the adhesome expression of tonsil NK cells represents previously characterized tissue-resident lymphocytes (5, 22, 59, 60). Our flow cytometric data also define a subset of CD103⁺ stage 3 cells in PB, which may include ILC1s, as these would have been included in our FACS gating strategy (35, 61). Previous associations of CD103 expression with NK cell tissue residency, together with the presence of CD103⁺ tonsil NK cells, suggests that CD103⁺ NK cell subsets may represent tissue-resident NK cells; however, this was not confirmed by measuring expression of CD69 or other tissue residency markers (5, 21). Although less well characterized than integrin αE, integrin αD is expressed on human NK cells and is associated with increased adhesion to ECM components, inside-out signaling-dependent cytokine secretion, and homing to inflammatory sites (62). Our ability to distinguish stage 4A and 4B NK cells and define the upregulation of β2 integrins, including ITGAL (CD11a), ITGAM (CD11b), and ITGB2 (CD18), as NK cells mature into stage 4B in tonsil allowed us to further define differences in adhesome profiles that occur at this stage of maturation associated with commitment to NK cell maturation (10). Based on these observations and the higher expression of cytoskeletal signaling and regulator genes including focal adhesion kinase (PTK2), CAPN2, and TIAM1, we show that tonsil NK cells have a unique adhesome profile that is defined by higher expression of both integrins and the cytoskeletal machinery that mediates their interactions with tissue.

Our data demonstrate that the largest changes in adhesome gene expression in both PB and tonsil occur early in NK cell development, and we found many adhesome genes differentially expressed between PB stages 4B and 5, whereas none were differentially expressed between tonsil stages 4B and 5. This suggests that there are still unknown differences between representative stage 4B and 5 NK subsets in tonsil and PB that go beyond the differences in expression of NK markers commonly used to discriminate developmental subsets between tissues. Still, our observations are in line with previous studies showing that tonsil- and lymph node–resident stage 5 NK cells resemble a more immature-like stage 4B NK cell, with increased expression of CD56 and decreased expression of cytotoxic function-related machinery, whereas stage 4B and stage 5 NK cell subsets in PB are unique both transcriptionally and phenotypically (4, 5). Furthermore, flow cytometric analysis revealed that both PB and tonsil stage 3 NK cells had unique adhesome expression relative to mature NK cells. The differences observed in integrin expression between stage 3 and stage 4 NK cells in tonsil and PB details undefined phenotypic changes that occur during NK cell commitment. It should also be noted that the gating strategy we used to identify stage 3 cells and sort stage 3 cells from tonsils likely contain a heterogeneous population of cells that includes bona fide stage 3 NK cells and ILC precursors (35, 63).

When considering these differences between NK cells isolated from different sites, we sought to better define how in vitro–derived cells align with primary cells. In vitro studies have demonstrated a functional requirement for integrins, specifically VLA-4 (α4β1), in facilitating T cell precursor interactions with OP9-DL1 stroma and priming double-negative T cells to receive Notch signaling (64). When we consider CD11a and integrin β2 expression, we observed that similar to primary human NK cells, in vitro–derived NK cells have changes in integrin expression between stages 3 and 4 of development. Specifically, NK cells transition from having relatively low integrin β2 in early stages of in vitro NK cell development to having high integrin β2 expression during stages 4–5. The transition between low to high expression of integrin β2 suggests that the upregulation of integrin β2 is a key feature of NK maturation in vitro and occurs in concert with the acquisition of CD94 and downregulation of CD117 (c-kit). Although we found similarities in the patterns of expression of CD11a and integrin β2 between in situ– and in vitro–derived NK cells, namely increasing expression through development, there are notable differences in the pattern of expression of β1 and β7 integrins. Our data suggest that in vitro NK cells resemble tissue-resident NK cells based on their expression of CD49a and integrin β7, yet ultimately have a unique adhesome phenotype from ex vivo NK cells (21, 65). We also found that the relative frequencies of cells with activated conformation of LFA-1, Mac-1, and integrin β1 differed between in vitro– and in situ–derived cells. Measuring the amount of activated integrin on the surface of NK cells allowed us to understand differences in the regulation of these subunits between tissue sites and demonstrated that integrin activation is dependent on both the microenvironment and developmental stage of NK cells. Factors that could contribute to the observed differences in integrin expression and activation between in vitro– and in situ–derived cells include differences in ECM components and the exogenous use of cytokines, particularly IL-15 (65). Although the phenotypes of NK cells generated on EL08.1D2 and OP9 cells were very similar, we noted slightly higher density of CD11a/CD18 subunits on stage 4 and 5 NK cells from OP9 conditions and differences in β1 integrin–associated subunits CD49d and CD49e. These differences between OP9 and EL08.1D2-derived cells speak to the unique microenvironments provided by these cell lines that could reflect their differing physiological origins (54, 55, 66). Together, this comprehensive profiling of integrins from in vitro–derived cells provides insight into the nature of cell–cell contacts that likely shapes NK cell development.

Our finding that NK cells at later stages of development, namely stage 5 in PB and stages 4 and 5 in vitro, had increased density of cortical actin was surprising. Although we did see a slight increase in the expression of the actin monomer ACTB, the difference in cortical actin density as measured by both flow cytometry and superresolution microscopy was orders of magnitude greater in stage 5 (CD56^dim) NK cells than stage 4 cells. Further, it should be noted that we did not activate the cells with integrin or activating receptor ligation. As such, the differences we observed were that of the cortical actin meshwork, not activation-induced actin remodeling. In contrast with other cell surface receptors such as L-selectin and CD94 that appear to have intermediate expression as cells progress from stage 4 to stage 5 (12, 67), we found a sharp demarcation in phalloidin intensity between CD56^bright and CD56^dim cells. Shear flow, such as that found in circulation, induces rapid lymphocyte morphological changes prior to tissue extravasation, suggesting that the CD56^dim NK cells found primarily in circulating PB may have increased cortical density in response to shear stresses or actin polymerization induced by tethering or transendothelial migration. However, given that the CD56^bright (stage 4) population isolated from PB had uniformly lower phalloidin staining than stages 3 or 5, it seems unlikely that this is the case unless stage 3 and CD56^dim NK cells have uniquely undergone transendothelial migration. Although the mechanism that underscores the unique actin phenotype in stage 4 PB NK cells is unknown, the increased expression of actin nucleating proteins, including the ARPC2 subunit of the Arp2/3 complex, and profilin in stage 5 PB cells suggests that the higher expression of these correlates with this higher actin density, and ARPC2 is a known regulator of cortical actin thickness (68). Although beyond the scope of this study, more detailed investigations into actin dynamics and architecture in primary NK cells will be necessary to link the phenotypic differences that we observed with functional outcomes.

Although in this study we have been guided by the consensus integrin adhesome, what remains to be defined is the spatial information regarding the formation of integrin adhesion complexes and signaling islands. Studies of the integrin adhesome are generated by proteomic data and further probed by high- and superresolution microscopy that provides critical spatial information about how such complexes are formed and function (29, 30, 32, 34, 69–71). An additional caveat of our approach is that both flow cytometry and gene expression data consider only discrete integrin subunits, yet their function and ligand specificity must be considered in the context of their obligate heterodimeric structure. However, by defining the expression of adhesome components at the gene expression level and validating some of these by protein expression, we lay the foundation for future studies that will better define the role of integrins through multiscale approaches.

We thank Michael Kissner for technical support and acknowledge the use of shared resources of the Columbia Stem Cell Initiative flow cytometry core. This study includes work using the Radiation Research Shared Resource of the Herbert Irving Comprehensive Cancer Center at Columbia University. We thank Evelyn Hernandez and Carlos Aguilar Breton for assistance with procuring and processing tonsil samples.

The authors have no financial conflicts of interest.