Research ArticleATOPY

Blood natural killer cell deficiency reveals an immunotherapy strategy for atopic dermatitis

See allHide authors and affiliations

Science Translational Medicine  26 Feb 2020:
Vol. 12, Issue 532, eaay1005
DOI: 10.1126/scitranslmed.aay1005

Nurturing NK cells to treat atopic dermatitis

The skin condition atopic dermatitis (AD) is driven by a type 2 immune response. Mack et al. performed high-dimensional immune profiling of patients with AD and revealed deficiencies in certain subsets of natural killer (NK) cells. NK cells showed signs of activation-induced cell death and were restored in patients that responded to immunotherapy. Circulating NK cells were also decreased in a mouse AD model; boosting NK cells with an IL-15 superagonist ameliorated symptoms in the mice. These results suggest that strategies to restore NK cells could help rebalance immunity in AD.


Atopic dermatitis (AD) is a widespread, chronic skin disease associated with aberrant allergic inflammation. Current treatments involve either broad or targeted immunosuppression strategies. However, enhancing the immune system to control disease remains untested. We demonstrate that patients with AD harbor a blood natural killer (NK) cell deficiency that both has diagnostic value and improves with therapy. Multidimensional protein and RNA profiling revealed subset-level changes associated with enhanced NK cell death. Murine NK cell deficiency was associated with enhanced type 2 inflammation in the skin, suggesting that NK cells play a critical immunoregulatory role in this context. On the basis of these findings, we used an NK cell–boosting interleukin-15 (IL-15) superagonist and observed marked improvement in AD-like disease in mice. These findings reveal a previously unrecognized application of IL-15 superagonism, currently in development for cancer immunotherapy, as an immunotherapeutic strategy for AD.


Atopic dermatitis (AD) or eczema is the most common inflammatory skin disorder, has a substantial negative impact on patients’ quality of life, and costs $5.3 billion annually in the United States (13). AD is characterized by elevated production of the type 2 cytokines interleukin-4 (IL-4), IL-5, and IL-13, which promote AD pathogenesis (4). Although classically associated with adaptive T helper type 2 cell responses, more recent work has identified that innate immune cell populations such as basophils and group 2 innate lymphoid cells (ILC2s) are major sources of these cytokines in AD (57). As a result, current treatment strategies in AD have focused exclusively on either broad or selective immunosuppression to combat pathologic type 2 inflammation. However, although it is well understood that type 2 immune cells promote AD pathogenesis, the endogenous mechanisms that maintain immune homeostasis and restrain inflammation in AD remain poorly defined. Beyond skin lesions, moderate-to-severe AD is associated with a systemic immune response involving increased blood eosinophils, elevated immunoglobulin E (IgE), and the development of other atopic disorders such as asthma and food allergy (810). The identification of immune pathways that suppress type 2 inflammation in AD may reveal previously unrecognized cellular pathways that could be therapeutically targeted to ameliorate AD and potentially other allergic diseases.

In addition to robust type 2 inflammation, patients with AD are known to exhibit diminished antiviral immunity and heightened risk for the development of severe disseminated herpesvirus (eczema herpeticum) and vaccinia virus (eczema vaccinatum) infections (11, 12). Natural killer (NK) cells are innate lymphocytes that comprise 5 to 10% of the circulating peripheral blood mononuclear cells (PBMCs) and critically promote antiviral immunity, in part through the production of interferon-γ (IFN-γ). Alterations in NK cell numbers, killing capacity, and IFN-γ production have been described in patients with AD since the early 1980s (1316); however, the specificity and clinical applications of this feature have been largely ignored. IFN-γ has been shown to suppress type 2 inflammation (1719), suggesting that NK cell dysregulation is a functionally relevant feature of AD. Recent studies have demonstrated that NK cells and IFN-γ can restrain ILC2 responses in vitro and during allergic lung inflammation (17, 18, 20), implicating NK cells broadly in regulating allergic inflammation. On the basis of these observations, we hypothesized that NK cells are dysfunctional in patients with AD and contribute to the disease process.


NK cell deficiency is a diagnostic feature of moderate-to-severe AD

We first performed a comprehensive analysis of blood lymphocyte subpopulations in 25 adult patients with moderate-to-severe AD (43 ± 3.4 years; 52% female) and compared them to a control cohort of 363 subjects without AD (50 ± 1 years; 64% female) seen during the study period (Fig. 1, A to D, and fig. S1). Our analysis revealed that patients with AD exhibited a marked and selective reduction in the number of blood NK cells (average 116.7 ± 24.6 NK cells/mm3) compared to controls (178.1 ± 0.7 NK cells/mm3; P = 0.00016) (Fig. 1D and fig. S1E), and 72% of our AD cohort had NK cell numbers below the 2.5th percentile of normal (<105 NK cells/mm3) (Fig. 1E). An analysis of diagnostic value using a receiver operating characteristic (ROC) curve indicated that low blood NK cell numbers can effectively differentiate patients with AD from an all-comers cohort of patients without AD seen across multiple hospital departments (Fig. 1F). Low blood NK cells were even more effective at distinguishing AD from those with moderate-to-severe chronic pruritus of unknown origin (CPUO), another pruritic disorder that currently lacks distinguishing biomarkers (Fig. 1G) (21).

Fig. 1 NK cell deficiency is a diagnostic feature of moderate-to-severe AD compared to non-AD and patients with CPUO.

(A to D) Retrospective clinical laboratory values (number of cells/mm3 peripheral blood) of (A) CD4 T cells, (B) CD8 T cells, (C) B cells, and (D) NK cells in 25 moderate-to-severe AD and 363 non-AD control patient subjects. (E) Histogram of NK cells/mm3 from AD cohort, with the dashed line indicating the bottom 2.5th percentile of normal values for this laboratory test. (F) Receiver operating characteristic (ROC) curve of NK cell number as a diagnostic test for AD versus a non-AD control patient population of all-comers tested during the same time period (except oncology). An optimal sensitivity of 72% and a specificity of 75.9% were found at a cutoff of ≤95 NK cells/mm3. (G) ROC curve of NK cell number for AD versus CPUO cohorts (N = 69 patients with CPUO), with a cutoff of ≤130 NK cells/mm3 providing optimal sensitivity (80%) and specificity (75%). AUC, area under the curve. ***P < 0.001; NS, P > 0.05 by Mann-Whitney U test.

Patients with AD exhibit alterations in specific subpopulations of NK cells

Two functionally distinct NK cell subpopulations, CD56bright and CD56dim NK cells, specialize in cytokine production and target cell killing, respectively (22). Beyond this initial binary classification, recent studies using high-dimensional mass cytometry (CyTOF) technology have revealed a large degree of phenotypic heterogeneity in human NK cells (23). This variation is, in part, due to individual genetic variation, such as human leukocyte antigen (HLA) and killer cell Ig-like receptor (KIR) haplotypes, and environmental exposures (e.g., cytomegalovirus) that shape the NK cell surface receptor repertoire (24). Therefore, we used CyTOF on peripheral blood from patients with AD and controls (Fig. 2A and table S1) in conjunction with viSNE for dimensionality reduction to characterize NK cell subpopulations in AD. NK cells were enriched from total PBMCs by preselecting for lineage (Lin) CD56+ cells before viSNE analysis (Fig. 2, A and B).

Fig. 2 Patients with AD exhibit alterations in circulating NK cell subpopulations.

(A) Sample collection and analysis schematic for CyTOF-based profiling of peripheral blood NK cells in six AD and five control subjects. (B) Representative viSNE plot colored by subpopulation identity. See Materials and Methods and figs. S2 to S4 for additional gating information. tSNE, t-distributed stochastic neighbor embedding. (C) Marker signal intensity for select markers indicated above the plots on a representative AD sample. Blue lines indicate subpopulation gates. (D) Heatmap of median arcsine-normalized signal intensities for each viSNE-based subpopulation. (E) Frequency of cells in each subpopulation as a percentage of total NK cells. *P < 0.05 by two-way ANOVA.

Manual gating was performed on the basis of previously described subsets of CD56bright, immature CD56dim, mature CD56dim, and adaptive NK cells (Fig. 2C and figs. S2 and S3) (25). Briefly, mature CD56dim cells are characterized by enhanced lytic granule protein contents, including granzyme B and perforin, expression of CD57, and expression of KIRs (Fig. 2D). To identify comparable populations between genetically diverse donors, we considered cells expressing any of the KIRs in our CyTOF panel to be mature. In contrast, immature CD56dim cells have enhanced NKG2A/CD94 expression in the absence of CD57 and KIRs (fig. S4A) (2628). Last, adaptive NK cells, which are expanded in response to cytomegalovirus exposure, are classically defined by NKG2C expression in the absence of FcεRγ (2932).

Our viSNE analysis and gating approach identified a total of eight distinct cellular subsets (Fig. 2D). Two of these were determined to be non–NK cell populations, because they were uniformly negative for all canonical NK cell markers other than CD56 (Fig. 2D). These non-NK populations are likely composed of CD56 populations captured through a generous pregate, although they could also contain CD56-expressing innate lymphocytes present at very low frequencies in the blood (3335). Of the six NK cell populations, patients with AD exhibited a selective reduction in mature CD56dim NK cells, but not CD56bright, immature CD56dim, or adaptive populations (Fig. 2E). Patients with AD also displayed elevated frequencies of a nonclassical natural cytotoxicity receptor–negative population, which we called NCR, that lacked expression of natural cytotoxicity receptors (NCRs) (NKp30, NKp80, and NKp46), NKG2D, and CD16 but did express NK cell–associated transcription factors (T-bet and Eomes) and effector molecules (granzyme B and perforin) (Fig. 2E and fig. S4, B to D). Together, these findings demonstrate that patients with AD have not only a global reduction in their blood NK cells but also a shift in subpopulations of NK cells that reflects a loss of mature CD56dim effector cells.

Type 2 cytokine blockade reverses NK cell defects in patients with AD

Dupilumab is an anti–IL-4 receptor α (IL-4Rα) monoclonal antibody (mAb) that is highly effective for the treatment of moderate-to-severe AD (3638). We therefore sought to investigate whether the NK cell alterations observed in patients’ blood are reversed by dupilumab treatment. We examined patients with AD before and after receiving dupilumab (Fig. 3A and table S2) for disease severity, as measured by the Investigator Global Assessment (IGA) score, clinical flow cytometry, and CyTOF. After treatment, all patients exhibited improvement in their IGA score (Fig. 3B) as well as a reduction in the inflammatory, AD-associated serum biomarkers TARC (CCL17), IL-4, and IL-13 (Fig. 3, C to E, and table S3). In association with the clinical response, the majority (67%) of patients demonstrated recovery of mature CD56dim NK cells, whereas all patients (100%) had a significant reduction in the frequency of the NCR population by CyTOF analysis (P = 0.0279; Fig. 3F). The two patients whose mature NK cells did not recover after dupilumab had the highest numbers of total NK cells and highest frequency of the mature subset before treatment. Although the recovery of mature CD56dim NK cells was not statistically significant for the whole cohort (P = 0.4498), there was a significant treatment effect on mature CD56dim NK cells in patients with NK cells below the ROC diagnostic cutoff of 95 cells/mm3 (P = 0.0391), suggesting that the degree of NK cell deficiency may affect this population’s response to treatment. Notwithstanding this, there was a significant recovery of the total NK cell population as determined by clinical flow cytometry after dupilumab treatment (P = 0.0156) (Fig. 3G). This demonstrates that NK cell deficiency associated with AD is reversible by type 2 cytokine blockade.

Fig. 3 Type 2 cytokine blockade rescues NK cell deficiency in human AD.

(A) Schematic of anti–IL-4Rα mAb (monoclonal antibody; dupilumab) treatment monitoring. Dupilumab was administered every other week (q2w) subcutaneously (s.c.) for at least 3 months. (B) Investigator Global Assessment (IGA) score for each patient before and after treatment. Scoring is conducted on a six-point scale, in which 0 = no disease and 5 = very severe disease. N = 6 patients with AD. **P < 0.01 by Mann-Whitney U test. (C to E) Luminex ELISA measurements of plasma (C) thymus and activation-regulated cytokine (TARC; also known as CCL17), (D) IL-4, and (E) IL-13 in healthy controls (N = 11) and patients with AD (N = 10) before and after dupilumab treatment. *P < 0.05 by unpaired, two-tailed t test. P < 0.05 and ††P < 0.01 by paired, two-tailed t test between pre- and posttreatment values for each individual. (F) Subpopulation frequencies as determined by CyTOF multidimensional analysis per patient before and after treatment (N = 6 patients). *P < 0.05 by paired two-tailed t test. P < 0.05 when excluding patients with NK cells ≥95 cells/mm3 (open circles) in a two-tailed, paired t test. (G) Number of NK cells (cells/mm3 blood) from clinical laboratory testing for total Lin (CD3, CD19) CD56+/CD16+ cells before and after treatment. *P < 0.05 by paired, two-tailed t test.

AD NK cells exhibit cellular features of activation-induced cell death

To identify cellular programs that may underlie the loss of mature CD56dim NK cells in AD, we performed RNA sequencing (RNA-seq) of sort-purified CD56dim NK cells from both patients with AD and control individuals (Fig. 4A, fig. S5A, and table S4). Principal components analysis (PCA) revealed distinct transcriptional profiles between AD and control NK cells (Fig. 4B) with a total of 567 differentially expressed genes (P < 0.05) (Fig. 4C and table S5). Genes differentially expressed between AD-associated and control NK cells were enriched for gene ontology (GO) terms associated with endocytosis, cytokine signaling, cell migration, and cell division, suggestive of an enhanced activation state (Fig. 4D, fig. S5B, and table S6). Together with the selective reduction in mature NK cells observed by CyTOF, we hypothesized that NK cells from patients with AD may be undergoing activation-induced cell death (AICD).

Fig. 4 NK cells from patients with AD exhibit evidence of AICD.

(A) Schematic of CD56dim NK cell isolation for RNA-seq (N = 4 AD and 5 control individuals). (B) PCA of CD56dim NK cells from AD and control individuals. (C) Volcano plot with significantly differentially expressed genes, with >2-fold change highlighted in red (up-regulated in AD) and blue (down-regulated in AD). (D) Broad categories of differentially expressed GO terms plotted by their enrichment scores (see Materials and Methods). (E) Enrichment scores of significantly differentially enriched (P < 0.05) apoptosis-associated gene sets by GSEA. (F) Select apoptosis gene set–associated genes differentially expressed in AD versus control CD56dim NK cells, represented as z-scored RPKM (reads per kilobase per million reads). (G) Frequency of CD56dim NK cells that are cleaved caspase-3+ by flow cytometry from either control or AD PBMCs. N = 5 healthy control and 5 patients with AD. (H) Schematic of AICD in vitro assay. (I) Percent change in the frequency of live CD56dim NK cells of Lin (CD3, CD19, CD14, CD34) cells after 3 hours of in vitro CD16 ligation. *P < 0.05 by unpaired, two-tailed t test. N = 9 non-AD patient controls and 8 patients with AD. (J) GSEA enrichment plot of AD versus control gene expression compared to a previously published IL-2, IL-12, and IL-18 cytokine-induced gene set (40). FDR, false discovery rate. NES, normalized enrichment score; FWER, familywise-error rate. (K and L) Plasma (K) IL-12 and (L) IL-18 measured by Luminex multiplex ELISA in samples taken from 10 patients with AD before and at least 3 months after initiation of dupilumab treatment as well as from 11 healthy control donors. *P < 0.05 with an unpaired t test, P < 0.05 with a paired t test between pre- and posttreatment values for each individual. (M) Frequency of cleaved caspase-3+ CD56dim NK cells by flow cytometry from control PBMCs after overnight culture. ***P < 0.001 by two-tailed t test. (N) Percent change from baseline of live CD56dim NK cells 3 hours after CD16 ligation in control PBMCs preincubated with either medium alone or medium with IL-12 and IL-18 overnight. *P < 0.05 by two-tailed t test.

In support of this, gene set enrichment analyses (GSEAs) revealed that caspase-associated and apoptotic gene sets were enriched in AD CD56dim NK cells (Fig. 4E). Both initiator and effector caspases increased in expression along with stress-induced genes such as TP53, BCL2L1, and PARP1 (Fig. 4F). To confirm these results, we performed flow cytometry on blood from AD and healthy control donors and found increased activity of the proapoptotic effector caspase-3 in AD CD56dim NK cells (Fig. 4G; fig. S6, A and B; and table S7). However, despite the reduced frequency of total CD56dim cells in their PBMCs (fig. S6C), we did not observe an increase in dead CD56dim NK cells by 7–aminoactinomycin D (7-AAD) staining at steady state from AD donors (fig. S6D). Therefore, we hypothesized that NK cell death in AD may require additional stimuli. NK cells have been reported to undergo AICD in response to ligation of the low-affinity Fc receptor CD16, a key mediator of antibody-dependent cellular cytotoxicity (39). We observed a selective reduction in live CD56dim NK cells from AD blood compared to healthy control blood in response to CD16 ligation (Fig. 4, H and I; fig. S6E; and table S8). These studies indicate that AD-associated NK cells exhibit a baseline proapoptotic phenotype and are more susceptible to AICD.

Previous studies have demonstrated that AICD of NK cells requires priming by activating cytokines such as IL-2 and IL-12 (39). We compared our RNA-seq data from control and AD CD56dim NK cells to a previously published dataset derived from human NK cells that were stimulated in vitro with IL-2, IL-12, and IL-18 (40). Using GSEA, we found significant enrichment of this cytokine-stimulated gene set in our AD NK cell transcriptomes (P = 0.026) (Fig. 4J and fig. S7A). We also found elevated IL-12 (Fig. 4K) and IL-18 (Fig. 4L) in the sera of patients with AD compared to sera from healthy controls (table S3). Furthermore, IL-18 in patients with AD decreased after type 2 cytokine blockade (Fig. 4L), suggesting that IL-18 may be associated with disease activity. To test whether elevated IL-12 and IL-18 may contribute to patients’ NK cell pathology, we preactivated control NK cells with IL-12 and IL-18 and observed increased cleaved caspase-3 staining (Fig. 4M) and loss of NK cells after CD16 ligation (Fig. 4N) in both bulk PBMCs and purified NK cell cultures (fig. S7, B to D). In addition, although they had a similar amount of FCGR3A mRNA (fig. S7E), CD56dim NK cells from patients with AD had, on average, less CD16 surface protein (fig. S7, F and G). Because CD16 is known to be posttranslationally regulated by proteolytic cleavage in response to activation (41), this is consistent with an activated state of CD56dim NK cells in vivo. Together, these findings suggest that, in the context of AD, blood CD56dim NK cells are exposed to elevated IL-12 and IL-18 and primed for AICD.

NK cells are enriched in lesional AD skin and limit type 2 inflammation

The strong association between AD and systemic NK cell dysregulation prompted us to investigate whether NK cells are altered in AD skin lesions. To examine this, we performed RNA-seq of paired lesional and nonlesional skin biopsies from six patients with AD to look for transcriptional evidence of NK cell activity (Fig. 5A and table S9). PCA readily distinguished the transcriptome of lesional skin from nonlesional skin (fig. S8A). To determine which pathways are altered in lesional skin, we looked at the differentially enriched GO terms. Among the most highly enriched GO terms, we found a number involving NK cell activation, cytotoxicity, and migration (Fig. 5B and table S10). Analysis of the differentially expressed genes driving this GO term enrichment revealed genes known to be involved in NK cell function, such as CD226 (DNAM-1), CRTAM, PRDM1, and GZMB (fig. S8B and table S11). To test whether these gene expression changes may be indicative of enhanced NK cell abundance in the lesional skin, we used the CIBERSORT platform to generate a predicted cellular composition of the complex tissue (42). Consistent with the GO enrichment analysis, CIBERSORT imputed an enrichment of activated NK cells in lesional AD skin compared to nonlesional skin (Fig. 5C).

Fig. 5 NK cells are enriched in lesional AD skin and limit type 2 inflammation.

(A) Sample acquisition schematic for RNA-seq of lesional (L) and nonlesional (NL) skin biopsies of six patients with moderate-to-severe AD. (B) Select differentially expressed gene ontology (GO) terms (P < 0.05) between lesional and nonlesional skin plotted by the enrichment score (see Materials and Methods). (C) Abundance (in arbitrary units) of activated NK cells imputed with CIBERSORT in lesional versus nonlesional patient skin biopsies (N = 6 pairs of biopsies). *P < 0.05 by paired, two-tailed t test. (D) Schematic of AD-like disease induction using MC903 or ethanol (EtOH, vehicle) application (n = 4 to 5 mice per group). (E) Select differentially expressed GO terms (P < 0.05) in murine AD-like skin compared to control skin plotted by their enrichment score (see Materials and Methods). (F) Flow cytometric analysis of murine AD-like (MC903) and control (EtOH) skin measuring the frequency of Lin (CD3, CD5, CD19, FcεRIα) NK1.1+ NK cells. (G) Diagram of the murine AD model for the comparison of WT and Il15−/− mice. (H) Frequency of Lin (CD3, CD11c, CD19) NK1.1+ cells in the lesional skin of WT and Il15−/− mice measured on day 12 by flow cytometry. (I) Gene expression of Ifng by quantitative reverse transcription PCR in lesional skin on day 12. (J and K) Frequency on day 12 by flow cytometry of (J) Lin (CD3, CD5, CD11b, CD11c, CD19, NK1.1) KLRG1+ ST2+ ILC2s and (K) Siglec-F+ eosinophils in lesional skin. *P < 0.05 by unpaired two-tailed t test.

We validated this approach by repeating it on a well-established murine model of AD-like disease. AD was induced using a standardized protocol of topical MC903 (calcipotriol) application daily for 12 days (Fig. 5D and fig. S8C) (43, 44). This model recapitulates the central features of AD including erythema (redness), scaling, blood eosinophilia, serum IgE elevation, itch behavior (pruritus), and histopathologic features of AD including acanthosis (epidermal thickening), hyperkeratosis (stratum corneum thickening), spongiosis (epidermal edema), and mixed dermal lymphocyte and eosinophil infiltration (5, 6, 43, 45). Similar to human AD skin, GO terms related to NK cell function were enriched in murine AD-like skin such as “NK cell activation” and “NK cell activation in immune response” (Fig. 5E). Last, as predicted by the RNA-seq signature, AD-like skin exhibited a higher frequency (Fig. 5F) and number (fig. S8D) of NK1.1+ NK cells compared to skin of control mice by traditional flow cytometry. Together, these findings suggest that NK cells are enriched and activated in lesional AD and AD-like skin of both humans and mice, respectively.

We next asked whether the systemic loss of NK cells we observed in patients with AD affects the disease process. Recent studies have found that NK cells and IFN-γ suppress ILC2 proliferation and cytokine production in vitro and in allergic lung inflammation (1719). Because ILC2s are critical pathogenic drivers of AD (57), we hypothesized that systemic NK cell deficiency in patients with AD may contribute to their inflammatory skin disease. First, we measured peripheral blood NK cells in our murine model and found that, like humans, mice given MC903 had a decreased frequency of circulating NK cells compared to controls over time (fig. S8E). To determine whether this reduction contributes to disease, we induced AD-like disease in Il15−/− mice, which are developmentally NK cell deficient (Fig. 5G) (46). As expected, these mice had substantial reductions in both NK cells (Fig. 5H and fig. S9A) and Ifng expression (Fig. 5I) in lesional skin. In addition, we observed an increase in ILC2s (Fig. 5J and fig. S9A) and eosinophils (Fig. 5K and fig. S9B) in skin lesions as well as the skin-draining lymph nodes (fig. S10, A to D) of Il15−/− mice compared to controls. Il15−/− mice also had an increased frequency of ILC2s at steady state (fig. S10E), suggesting that NK cell regulation of ILC2s may be a homeostatic function. Furthermore, to confirm the effect of NK cell deficiency on ILC2s during AD-like disease, we depleted NK cells from both wild-type (WT) and T cell–deficient Rag1−/− mice during induction of AD-like disease using anti-NK1.1 mAbs and observed similar effects of NK cell reduction on ILC2 frequencies (fig. S10, F to M). Together, these findings demonstrate that NK cells limit innate type 2 inflammation in the skin in a murine model of AD.

IL-15 superagonism promotes NK cell–dependent resolution of AD-like inflammation

Despite the robust effect of NK cell depletion on skin ILC2s and eosinophils, we did not observe a notable exacerbation of clinical disease in NK cell–deficient mice (fig. S10, N and O). This suggests that physiologic NK cell responses are insufficient to control disease under AD-like conditions. Therefore, we hypothesized that boosting NK cells to supraphysiologic levels may promote their regulatory function and ameliorate disease. IL-15 superagonists (SAs) are an established method to promote NK cell survival and proliferation in vivo (47, 48) and are currently being used in clinical trials either alone or in combination with checkpoint inhibitors for cancer immunotherapy (4951). However, whether an NK cell immunotherapy strategy could be applied to allergic diseases such as AD remains unexplored. Thus, we generated a murine IL-15 SA, in which a soluble IL-15Rα–Fc complex is stably loaded with IL-15 (Fig. 6A and fig. S11A), which we confirmed was able to induce NK cell expansion in vivo in a dose-dependent manner (fig. S11, B and C). After 4 days of AD-like disease induction in WT mice, we administered systemic murine IL-15 SA or an IgG1 isotype control and monitored disease progression (Fig. 6B). Mice that received IL-15 SA had more NK cells in the blood (Fig. 6C) and skin (Fig. 6D). They also had reduced numbers of skin ILC2s and eosinophils (Fig. 6, E and F) compared to mice that received isotype control treatment. These cellular changes were accompanied by a robust reduction in disease severity (Fig. 6G), including clinical scoring (Fig. 6H), skin thickness (Fig. 6I), and histopathology (Fig. 6, J and K). Initiation of IL-15 SA treatment at an even later time point, on experimental day 8, was also able to reduce disease symptoms (fig. S11, D to F), indicating that this method has robust therapeutic potential in the setting of established disease.

Fig. 6 IL-15 superagonism boosts NK cells and alleviates AD-like disease in mice.

(A) Diagram of IL-15 superagonist (SA) administered to mice. (B) Schematic of IL-15SA or IgG1 isotype administration during AD model. (C and D) Number of Lin (CD3, CD11c, CD19) NK1.1+ CD49b+ NK cells on experimental day 8 in the (C) blood and (D) skin. (E and F) Numbers of (E) ILC2s and (F) eosinophils in AD-like skin on experimental day 8. *P < 0.05, **P < 0.01, and ***P < 0.001 by unpaired two-tailed t test. (G) Representative images of lesional skin on experimental day 12 of IgG1 isotype- and IL-15SA–treated mice. (H) Clinical scores and (I) ear thickness (percent change from baseline) of isotype- and IL-15SA–treated mice during the course of treatment. **P < 0.01 by two-way ANOVA. (J) Representative H&E histopathology images of ear skin on experimental day 12. Scale bars, 100 μm. Arrowheads indicate inflammatory vasodilation. (K) Histopathology score of H&E sections on experimental day 12. *P < 0.05 by unpaired, two-tailed t test. (L) Treatment schematic of IL-15SA or isotype treatment and AD-like disease induction in Cd8−/− mice. (M) Ear thickness (percent change from baseline) and (N) clinical scores of isotype control- and IL-15SA–treated Cd8−/− mice. *P < 0.05 by two-way ANOVA. (O) Treatment schematic of IL-15SA or isotype treatment and AD-like disease induction in NK cell–deficient Ncr1-iCre Rosa-DTA (Ncr1DTA) mice. (P) Ear thickness (percent change from baseline) and (Q) clinical scores of isotype- and IL-15SA–treated Ncr1DTA mice. NS, P > 0.05 by two-way ANOVA.

Although IL-15 SA treatments have been shown to primarily target NK cells in both mice and patients (47, 51), IL-15 is also important in generating memory CD8 T cell responses (46). To determine the relative contributions of NK cells and CD8 T cells in IL-15 SA-mediated disease reduction, we gave IL-15 SA to Cd8−/− mice after AD-like disease induction (Fig. 6L). Despite the lack of CD8 T cells, IL-15 SA significantly reduced ear thickness (P = 0.0189; Fig. 6M) and clinical scores (P = 0.0138; Fig. 6N), suggesting that NK cells are sufficient to mediate the therapeutic effect. To test whether NK cells are specifically required, we generated NK cell–deficient mice using Ncr1-iCre mice crossed with Rosa-stop-floxed-DTA mice (Ncr1DTA) (Fig. 6O and fig. S12). Conditional deletion of NK cells abrogated the ability of IL-15 SA to reduce both ear thickness (Fig. 6P) and clinical scores (Fig. 6Q). Collectively, these studies indicate that restoring NK cell deficiency through IL-15 superagonism is an effective and promising therapeutic strategy for AD.


AD is a systemic immune disorder in a family of type 2 inflammatory conditions including asthma and food allergy that are characterized by elevated IgE, eosinophilia, and a predisposition for allergen sensitization across barrier surfaces. Although most of the research on AD pathology thus far has focused on cutaneous mechanisms of barrier dysfunction and inflammatory cell recruitment, how the systemic immune system in AD is negatively affected is poorly defined. In this study, we show that low peripheral blood NK cells in patients with AD have diagnostic value in distinguishing AD from both our cohort of non-AD patients and specifically patients with CPUO. Beyond reduced numbers, CyTOF and RNA-seq analysis of both control and diseased NK cells demonstrated that AD-associated NK cells have a distinct transcriptional program indicative of AICD and a selective loss of a subset of mature CD56dim NK cells. These findings suggest that chronic inflammation associated with AD can promote activation, maturation, and global loss of blood NK cells.

Studies over the past two decades have found various abnormalities in NK cell populations in the blood of patients with AD (1316). However, the precise nature of these defects and their relationship to disease status has been unclear. Our data are consistent with previous studies measuring decreased total CD56dim NK cells in patients with AD (15), including one study that specifically observed a reduction in CD57+ NK cells (14). Furthermore, the increased apoptotic gene signature in CD56dim NK cells is supported by a previous study showing that AD NK cells had enhanced apoptosis in vitro after phorbol 12-myristate 13-acetate (PMA)/ionomycin stimulation (16). This preferential apoptotic response was dependent on the presence of donor monocytes (16), although the signals required for this interaction to induce apoptosis are not clear. It is possible that monocytes are a source of proinflammatory cytokines, such as IL-12 and IL-18, in the blood of patients with AD that may contribute to both the activated phenotype and enhanced sensitivity to AICD. However, the activating signals triggering NK cell death in vivo are unknown.

In addition to the total NK cell reduction and loss of mature CD56dim cells, we also observed an outgrowth of a small, nonclassical population of CD56dim cells that lacked canonical NK cell receptors, which we termed NCR cells. These cells are present at very low frequency in control subjects but expanded substantially in the setting of AD. The study design limited our ability to deeply phenotype or functionally characterize these cells, because the multidimensional analysis used to identify them exceeds the number of parameters available for cell sorting. However, nonclassical populations of NK cells with reduced NCR expression and impaired lytic properties have been observed in humans with chronic viral infections such as human immunodeficiency virus and hepatitis C virus (5254). These cells share several features with the NCR population observed in our dataset such as decreased NKp46, NKp80, and perforin expression. Thus, understanding the relationship between chronic inflammation, NK cell deficiency, and antiviral immunity in AD may provide insight into a common mechanism underlying their expansion in chronic diseases.

In conjunction with the loss of peripheral blood NK cells, we identified an elevated NK cell transcriptional signature in lesional skin of patients with AD. The presence and developmental origins of NK cells in healthy human skin remain poorly characterized. Current data support the presence of a population of CD3 CD56+ CD16 NK cells in healthy and inflamed skin that most closely resembles the CD56bright population in the blood (55). Furthermore, a recent study identified a proliferative, skin-homing population of CD56bright NK cells during acute dengue virus infection that represented most of the skin NK cells in these patients (56). However, whether a conventional population of CD56dim CD16+ NK cells exists in human skin or traffics into the skin in response to noninfectious inflammatory conditions remains unclear. In mice, two populations of NK cells have been identified in the skin: tissue-resident and recirculating (57). We identified an increased number and frequency of NK cells in AD-like skin compared to control skin, which further increased after systemic IL-15 SA stimulation. Although the number of NK cells leaving the blood for the skin would likely be insufficient to account for the substantial loss in total NK cells in patients’ blood, we hypothesize that some degree of migration of NK cells from the blood to the skin supports the expanded numbers of NK cells in AD lesions and may represent an endogenous regulatory response to aberrant type 2 skin inflammation.

We propose that the observed changes in peripheral blood NK cells, in addition to having diagnostic value, may have broader implications for both protective immunity and inflammation in the setting of chronic AD. In our preclinical model of AD, we found that a loss of NK cells in mice, through both genetic and pharmacologic approaches, resulted in an exacerbation of pathogenic ILC2 responses, suggesting that NK cells can regulate ILC2s in skin lesions. Although our study was limited to a single model system of AD, these findings are consistent with recent studies in the lung, showing that loss of NK cell–derived IFN-γ was accompanied by an increase in ILC2s and allergic lung inflammation (17). Therefore, our findings in the skin indicate that this NK cell–ILC2 inhibitory axis may be an evolutionarily conserved regulatory mechanism present at multiple barrier surfaces. In light of these findings, the systemic loss of NK cells that we identified in patients with AD may not only impair antiviral immunity in patients but also contribute to the unchecked type 2 inflammation and skin lesions. In addition, the restoration of NK cell numbers in patients after type 2 cytokine blockade with dupilumab indicates that NK cell reduction may occur secondary to allergic inflammation, creating a vicious disease cycle.

Although this study focused on the effects of inflammatory cytokine signaling on NK cell numbers, the ability of NK cells to limit AD-associated inflammation suggests that reduced NK cell numbers and/or function could be a predisposing risk factor for AD. In support of this perspective, a previous study of two separate European cohorts detected a potentially protective effect of KIR2DS1 against the development of AD (58). Another study found a single-nucleotide polymorphism (SNP) in KIR2DS2 associated with AD and asthma (59). Although these observations implicate NK cell homeostasis in AD and atopy, the functional relationship between these SNPs and NK cell function or disease outcomes is undefined. Furthermore, the current study is limited with respect to the severity of its study population. Although we found NK cell deficiency in moderate-to-severe patients, whether NK cells are reduced in the blood of patients with less severe symptoms or correlate with disease severity was not evaluated.

In a departure from the current immunosuppressive treatment approaches for AD, these findings offer a new paradigm in which reversing NK cell deficiency in patients may provide therapeutic benefit. NK cell agonism (4951) and NK cell checkpoint inhibition (60) strategies in cancer immunotherapy have proven highly effective in both expanding host NK cells and boosting their function. Subcutaneous administration of the IL-15 SA complex ALT-803 has been well tolerated in patients with cancer, with the main adverse events being an injection site reaction that resolves without intervention and transient hypertension (61). In another trial in which IL-15 SA complexes were given in combination with the checkpoint inhibitor nivolumab over the course of 6 months (49), IL-15 SA adverse events lessened over time, indicating that this may be a viable approach for chronic treatment. We therefore propose NK cell–based immunotherapy as a treatment approach for AD.


Study design

The rationale for the design of human studies was to undertake either a case-control study approach or perform a basic observational study in a diseased population (cases) in response to a highly effective treatment (dupilumab) over time to determine whether blood NK cell populations are different in frequency, number, phenotype, and/or identity between cases and controls or in response to treatment. Given the observational nature of the translational studies, there was no randomization or formal blinding process for the investigators. Where possible, measurements were acquired in a blinded manner and then unblinded after results were obtained. Although there was no predefined power analysis performed, interim power analyses were performed when trends were observed, and additional cases and/or controls were added to achieve the determined cohort size. Sample acquisition was stopped upon reaching statistical significance (P < 0.05).

The rationale for the design of murine studies was to use only enough mice in each experiment to observe a statistically significant difference between groups. These numbers were not based on predefined power analyses but previous experience with our well-validated model system and numbers of mice used previously (5, 45, 62). Age-, sex-, and strain-matched controls were always used. Whenever possible, measurements were acquired in a blinded manner and then unblinded after results were obtained. Interim power analyses were performed when trends were observed, and experiments were replicated to achieve the necessary cohort size. Data shown in the figures are a representative of at least three independent experiments or pooled across experiments when a larger cohort size was required.

For retrospective analysis of blood lymphocyte populations, we extracted Lymphocyte-13 flow cytometry data from Cerner, a centralized data management software program used by the Barnes-Jewish Hospital (BJH) laboratory, for all patients seen at BJH between January 2015 and December 2018 [Institutional Review Board (IRB) no. 201703135]. Patients who were seen in an oncology clinic or who had a history of any malignancy were excluded. This resulted in a total of 363 patients designated as controls. The diagnosis of AD (cases) was made on the basis of the revised Hanifin and Rajka criteria (63). Patients with AD who visited the Washington University School of Medicine (WUSM) specialty itch clinic between January 2015 and December 2018, were ordered a Lymphocyte-13 laboratory test, and had an IGA score of ≥3 (moderate-to-severe diagnosis) were selected for inclusion in the analysis. This resulted in 25 moderate-to-severe AD cases. A total of 69 patients with CPUO (64) were represented in the control group and, in some analyses, were also compared directly against patients with AD. CPUO was diagnosed on the basis of the presence of chronic pruritus for >6 weeks in the absence of a primary skin rash, endocrine disease, metabolic disorders, uremia, hepatobiliary disease, malignancy, infection, neurologic disease, drug reactions, or psychiatric etiology (65, 66). This cohort was selected from patients who visited WUSM specialty itch clinic within the same study period and received a diagnosis of CPUO. These retrospective analyses of existing clinical data qualified for an informed consent waiver. Primary data are reported in data file S1.

Primary human sample collection

Functional assays and RNA-seq on control NK cells were performed on peripheral blood obtained from Mohs surgery patients (IRB no. 201507042). After obtaining informed consent, blood and skin samples were obtained from patients with moderate-to-severe AD (IGA ≥ 3) seen in the Division of Dermatology at WUSM/BJH from November 2015 to September 2018 (IRB no. 201410014). CyTOF analysis was performed on PBMCs from healthy volunteers, following informed consent (IRB no. 201503172). All samples were obtained from peripheral blood draw, and PBMCs were isolated by Ficoll density gradient purification and frozen at −80°C until assayed.

Cases and controls were age- and sex-matched whenever possible. However, two studies were not sex-matched (Figs. 1 and 2). These studies had opposite sex biases and found similar results, which were confirmed in a separate, sex-matched analyses (fig. S6C), suggesting that our findings are not dependent on sex differences. Detailed characteristics of research cases and controls are in tables S1 to S4 and S7 to S9.

Research animals

WT C56Bl/6J, Rag1−/−, and Rosa-stop-floxed-DTA mice were initially purchased from The Jackson Laboratory and bred in house. Cd8−/− mice were directly purchased from The Jackson Laboratory. Il15−/− mice were originally generated by Kennedy and Peschon (46), obtained from Taconic, and bred in house. Ncr1-iCre mice were generated by E.V. (67) and bred in house. All experiments were conducted with the approval of the Washington University Institutional Animal Care and Use Committee. Animals were housed on a standard 12:12 light:dark cycle with free access to food and water. Experiments were performed on independent cohorts of male and female mice. For induction of AD-like disease, 8- to 12-week-old mice were treated with 1 nmol MC903 (Tocris Bioscience) in 10 μl of 100% ethanol (EtOH) vehicle, or vehicle alone, on the bilateral ear skin daily for 7 or 12 days. Body weight and ear thickness were measured daily with a digital scale and analog caliper by the same investigator. For tissue harvest, animals were euthanized by CO2 inhalation.

Mass cytometry

Mass cytometry was performed as previously described (48). Briefly, metal-tagged antibodies were purchased from Fluidigm or custom-conjugated using the Maxpar X8 Antibody Labeling Kit according to the manufacturer’s instructions (Fluidigm). All antibodies were titrated before use. PBMCs were stained with metal-conjugated antibodies (table S12) with the following protocol: PBMCs were washed and counted, and 3 × 106 cells were stained with primary and then secondary surface antibodies for 30 min each on ice. Cells were then washed and stained with cisplatin for viability, fixed for 30 min on ice, permeabilized using the FoxP3 Transcription Factor Staining Kit (eBioscience) per the manufacturer’s instructions, and left in CyTOF Cell Staining Buffer (Fluidigm) overnight. The next day, cells were repermeabilized, barcoded with Cell-ID 20-Plex Pd Barcoding Kit (Fluidigm), pooled, and stained with intracellular primary and secondary antibodies on ice for 30 min each. Last, Cell-ID Intercalator-Ir (Fluidigm) was added to detect nuclei. Cells were diluted in distilled deionized water containing 10% EQ Calibration Beads (Fluidigm) at 106 cells per ml and acquired on a CyTOFII instrument (Fluidigm) at the Bursky Center for Human Immunology and Immunotherapy Programs Immunomonitoring Lab core facility. The data were randomized with Fluidigm acquisition software and normalized with MATLAB bead normalization (68).

Samples were debarcoded using the MATLAB Nolan laboratory single-cell debarcoder v0.2 (nolanlab/single-cell-debarcoder; GitHub) (69) as live, single cells (Bead Cisplatin DNA1/2+) and then imported into Cytobank. Samples were normalized to machine controls to reduce batch effects of multiple run days. To do this, pregated CD19 CD14 CD3 CD56+ events were exported from Cytobank for each sample. Median signal intensities were extracted for each machine control, and a normalization vector was generated as the ratio of a machine control to a benchmark run using a custom code in R (doi:10.5281/zenodo.3568404). This normalization vector was then applied to each sample in the run, and the normalized files were reimported as a single experiment into Cytobank for further analysis. Dimensionality reduction was performed with Cytobank viSNE using equal sampling, 10,000 iterations, perplexity 50, and θ of 0.5.

Flow cytometry

For in vitro human studies, cells were harvested from cell culture, stained with primary antibodies on ice for 30 min, washed, stained with 7-AAD (BioLegend), and acquired on a BD Fortessa X-20. For animal studies, ear skin was digested in 500 μl of Liberase TL (0.25 mg/ml) (Roche) in Dulbecco’s modified Eagle’s medium (Sigma-Aldrich) at 37°C and 5% CO2 for 90 min. Skin and draining lymph nodes were then manually homogenized through a 70-μm cell strainer to obtain a single-cell suspension. All cells were stained with Zombie UV dye (BioLegend) for viability at room temperature for 20 min, followed by primary antibodies on ice for 30 min (table S13). Secondary streptavidin-conjugated fluorophores were stained on ice for 30 min. Cells were then fixed with BD Cytoperm/Cytofix reagent on ice for 30 min or overnight at 4°C before data acquisition on a BD LSRFortessa X-20 special order research product. Data were analyzed with FlowJo 10 (Tree Star).

In vitro stimulation assays

For the CD16 ligation assay, 0.5 × 106 to 1 × 106 PBMCs were stimulated in 96-well round-bottom plates in a 37°C incubator with 5% CO2. PBMCs were thawed and cultured for 12 to 14 hours with basal medium [RPMI 1640 containing 10% Human AB serum (Sigma-Aldrich), 10 mM Hepes (Corning), 1× nonessential amino acids (Corning), 1 mM sodium pyruvate (Corning), 1× penicillin (100 IU/ml)–streptomycin (100 μg/ml) solution (Gibco), 2 mM l-glutamine (Gibco)] and recombinant human IL-15 (rhIL-15) (1 ng/ml) (Miltenyi). Before stimulation, medium was changed to fresh medium containing indicated concentrations of anti-CD16 (BD Biosciences) antibody–conjugated MACS iBeads (Miltenyi). After 3 hours, cells were harvested for flow cytometric analysis as described above. For cytokine priming assays, replicates of 106 PBMCs harvested from a healthy donor Leukopak (STEMCELL) were incubated overnight in basal medium containing rhIL-15 (1 ng/ml), rhIL-12 (10 ng/ml) (BioLegend), and rhIL-18 (100 ng/ml) (Gibco). For purified NK cell stimulation, NK cells were isolated from Leukopak PBMCs using the Human NK Cell Negative Selection Kit (STEMCELL) in a 96-well round-bottom plate using an EasyPlate magnet (STEMCELL) per the manufacturer’s instructions.

IL-15 SA

IL-15 SA was administered by intraperitoneal injection of 1 μg of SA in 100 μl of phosphate-buffered saline (PBS) daily on days 4 to 7 of MC903 treatment. IL-15 SA was prepared as previously described (47). Briefly, 20 μg of recombinant murine IL-15 (rmIL-15; eBioscience or STEMCELL) was combined with 90 μg of a chimeric sIL-15Rα fused to the Fc domain of human IgG1 (R&D Systems) at a concentration of 0.1 mg/ml of IL-15 in PBS. The mixture was then vortexed, incubated at 37°C for 20 min, and diluted to 10 μg/ml of IL-15 in PBS. SA concentration was calculated with reference to IL-15 for in vivo dosing. Stable loading was confirmed by measuring free rmIL-15 in solution after coincubation of rmIL-15 and IL-15Rα–Fc proteins by enzyme-linked immunosorbent assay (ELISA) (R&D Systems, DuoSet). Actual values based on a standard curve were comparable to values predicted based on molar ratios. Isotype control solution was prepared in the same fashion with rhIgG1 (R&D Systems), 0.1% bovine serum albumin, and 1 μM glycine in PBS. Aliquots of SA or isotype solution were frozen at −20°C and thawed just before injection. Clinical scoring was adapted from the eczema area and severity index (EASI) (70), performed by a treatment-blinded investigator, and calculated as the sum of a redness score (0 = none, 5 = severe) and a scaling score (0 = none, 5 = severe).

Histological analysis

For murine AD-like histopathology analysis, ear tissues were harvested on experimental day 12, fixed in 4% paraformaldehyde, and embedded in paraffin before sectioning and staining with hematoxylin and eosin (H&E). Slides were imaged using the NanoZoomer 2.0-HT System (Hamamatsu). Images were scored by a blinded investigator, and histopathology score was calculated as the sum of the following criteria: keratin thickness (average of 3 measurements per 40× image × 3 images), epidermal thickness (average of 3 measurements per 40× image × 3 images), epidermal spongiosis (0 to 3 rating), microabscesses (0 to 3 rating), vascular area (sum per treated side/1000), and inflammatory infiltrate (0 to 10 rating per whole ear).

Plasma cytokine measurements

Plasma was isolated from peripheral blood draw from AD or healthy control subjects by Ficoll gradient separation and frozen at −80°C. Plasma was diluted 1:1 in assay diluent and blocked by preincubation on Protein L–coated plates (Thermo Fisher Scientific) for 90 min at room temperature on an orbital shaker before the detection assay. Cytokines were measured using a 27-plex custom Luminex ELISA kit (R&D Systems), and data were collected on a FLEXMAP three-dimensional system (Thermo Fisher Scientific).

Quantitative reverse transcription polymerase chain reaction of murine skin

MC903-treated, AD-like ear skin and EtOH-treated control skin were harvested on day 12 of treatment, placed in RNAlater (Invitrogen) overnight at 4°C, and stored at −80°C. RNA was isolated after tissue homogenization with a bead homogenizer in buffer RLT (Qiagen) with 142 mM ß-mercaptoethanol using the RNeasy Mini Kit (Qiagen) per the manufacturer’s instructions. Genomic DNA was removed with a TURBO DNA-Free kit (Invitrogen) before complementary DNA (cDNA) synthesis with the High Capacity cDNA Reverse Transcription Kit (Applied Biosystems). For relative quantification of Ifng mRNA, 10 ng of cDNA was used to perform quantitative polymerase chain reaction (qPCR) with a commercial primer-probe assay (assay ID Mm.PT.58.41769240; Integrated DNA Technologies) and the TaqMan Gene Expression Master Mix (Applied Biosystems) on a StepOnePlus machine (Applied Biosystems).

RNA-seq of skin

Murine MC903- and vehicle EtOH-treated skin RNA-seq data were obtained from a previously published study (45). For sequencing of human skin, 4-mm punch biopsies were placed in RNAlater (Life Technologies) overnight at 4°C and stored at −80°C until further processing. Skin was homogenized with a bead homogenizer in RNA lysis buffer, and RNA was isolated with the RNeasy Mini Kit (Qiagen). Library preparation, alignment, and transcript abundance were performed by the Genome Technology Access Center (GTAC) at WUSM as previously described (45). Briefly, after deoxyribonuclease treatment (TURBO DNase, Invitrogen), ribosomal RNA was removed with Ribo-Zero kit (MRZH11124; Illumina) and RNA was reverse-transcribed using SuperScript II RT enzyme (Invitrogen). Human samples were sequenced with an average of 60 million 1 × 50 single reads on an Illumina HiSeq3000. Reads were aligned to Ensembl release 76 human genome assembly using STAR (71), gene counts were determined with Subread:featureCount (72), and sequence performance was assessed with RSeQC (73).

RNA-seq of sort-purified NK cells

Live, CD45+ CD3 CD56dim NK cells (30,000 to 100,000) were sort-purified from cryopreserved PBMCs on an Aria II (BD Biosciences) into 200 μl of lysis buffer RA1 (Macherey-Nagel) containing tris(2-carboxyethyl)phosphine (TCEP) per the manufacturer’s instructions. Cells were then vortexed for 30 s, frozen on dry ice, and stored at −80°C. RNA isolation was performed with the NucleoSpin RNA XS Kit (Macherey-Nagel). Library preparation, alignment, and transcript abundance were performed by the GTAC at WUSM. Ribosomal RNA was removed, and cDNA was generated with the SMARTer Kit (Clontech) with 10 ng of total RNA per sample. Samples were sequenced to an average depth of 34 million 1 × 50 reads on a HiSeq3000 (Illumina). Reads were aligned to Ensembl release 76 human genome assembly using STAR (71), gene counts were determined with Subread:featureCount (72), and sequence performance was assessed with RSeQC (73).

RNA-seq analyses

Genes were filtered for rowSums() > 10 counts and protein coding designation. Differential gene expression analysis was conducted using the Bioconductor package DESeq2 (74) in R v3.5.1 using default parameters. GSEAs were performed with GAGE (generally applicable gene set enrichment) (75) and GSEA (Broad Institute). Differentially enriched GO terms were grouped into larger biological categories based on keywords (table S6). Enrichment score for GO terms was calculated as the Stat.mean score divided by P value from GAGE output in R. Enrichment score for GO categories is the mean of the enrichment scores of GO terms within each category. Genes associated with GO terms were determined to be genes with differential expression between AD and control groups (P < 0.05), with indicated differentially expressed GO assignments extracted using AnnotationDbi package (Bioconductor).

Statistical analysis

Data are presented as means ± SEM unless otherwise specified. Murine results are representative of at least two independent experiments. Graphical results and statistical testing for RNA-seq and retrospective laboratory testing analysis were conducted in R v3.5.1. Graphical results and statistical testing for remaining studies were conducted with GraphPad Prism 8. Data were tested for normality using the Shapiro-Wilk test, and a nonparametric test (Mann-Whitney U or Wilcoxon) was used for data that were deemed nonnormal. Otherwise, a t test or analysis of variance (ANOVA) was performed, where indicated. For tests with multiple comparisons, the Sidak (two-way ANOVA) or Sidak-Holm (t tests) correction for multiple comparisons was used. ROC curve analyses were conducted in SPSS v25.0 for Mac. ROC curve analyses were conducted assuming that a lower NK cell number was a positive test result.


Fig. S1. Clinical lymphocyte laboratory results from patients with AD.

Fig. S2. Signal intensities and gating for phenotypic analysis of human peripheral blood NK cells by CyTOF.

Fig. S3. Density plots of viSNE analysis of CyTOF data for Lin CD56+ cells by subject.

Fig. S4. Detailed signal intensities of CyTOF viSNE populations.

Fig. S5. RNA-seq of human CD56dim NK cells.

Fig. S6. Gating strategies for in vitro assays on CD56dim NK cells.

Fig. S7. AD NK cells demonstrate evidence of AICD.

Fig. S8. NK cell enrichment in AD skin of mice and humans.

Fig. S9. Gating strategy for murine skin immune cell populations.

Fig. S10. NK cell depletion results in enhanced ILC2 responses.

Fig. S11. IL-15 SA induces a dose-dependent NK cell expansion and ameliorates AD-like disease at later time points.

Fig. S12. Ncr1-DTA mice lack NK cells but have intact CD8 T cells.

Table S1. Demographics for subjects analyzed in Fig. 2, NK cell CyTOF.

Table S2. Demographics for subjects analyzed in Fig. 3, NK cell CyTOF before/after dupilumab treatment.

Table S3. Demographics for subjects analyzed in Fig. 3, plasma cytokines.

Table S4. Demographics for subjects analyzed in Fig. 4, NK cell RNA-seq.

Table S5. Differential gene expression table output from DESeq2 analysis of CD56dim NK cell RNA-seq.

Table S6. Differentially enriched GO terms from GAGE analysis of CD56dim NK cell RNA-seq.

Table S7. Demographics for subjects analyzed in Fig. 4, cleaved caspase-3 staining.

Table S8. Demographics for subjects analyzed in Fig. 4, CD16 ligation.

Table S9. Demographics for subjects analyzed in Fig. 5, skin RNA-seq.

Table S10. Differentially enriched GO terms from GAGE analysis of lesional versus nonlesional human skin RNA-seq.

Table S11. Differential gene expression table output from DESeq2 analysis of lesional versus nonlesional human skin RNA-seq.

Table S12. Human NK cell CyTOF antibodies.

Table S13. Flow cytometry antibodies.

Data file S1. Primary data.


Acknowledgments: We would like to thank all of the patients who participated in this study for their generous contribution and support. We also thank the clinical faculty and staff in the Division of Dermatology at WUSM. We thank the Immunomonitoring Lab (IML) at the Bursky Center for Human Immunology and Immunotherapy Programs (CHiiPs), especially O. Malkova, for their assistance with CyTOF and flow cytometry. We also thank the GTAC for sequencing support, the Alafi Neuroimaging Laboratory for NanoZoomer access, and the Digestive Disease Research Core Center (DRCC) for histology support. We thank the Alvin J. Siteman Cancer Center at WUSM and BJH in St. Louis, MO and the Institute of Clinical and Translational Sciences (ICTS) at WUSM for the use of the Tissue Procurement Core, which provided biobanking service. We also thank A. Guggisberg for scientific editing services provided on behalf of ICTS. Funding: We thank the Alvin J. Siteman Cancer Center at WUSM; BJH in St. Louis, MO; and the ICTS at WUSM for the use of the Tissue Procurement Core, which provided human biospecimen storage service. We also thank the Siteman Cancer Center and ICTS for the use of the IML, which provided CyTOF services. The Siteman Cancer Center is supported, in part, by National Cancer Institute (NCI) Cancer Center Support Grant P30 CA091842, and the ICTS is funded by the NIH’s National Center for Advancing Translational Sciences (NCATS) Clinical and Translational Science Award (CTSA) program grant UL1TR002345. We thank the GTAC, which provided sequencing service and is supported by NCI grant P30CA91842 and by ICTS Clinical and Translational Science Award UL1TR000448 from the National Center for Research Resources (NCRR). The Alafi Neuroimaging Laboratory is supported by NCRR Shared Instrumentation grant 1S10RR027552. The DDRCC is supported by National Institute of Diabetes and Digestive and Kidney Diseases (NIDDK) grant P30DK052574. This publication was also made possible, in part, by UL1RR024992 from the NIH NCRR and NIH CTSA grant UL1TR002345 from the National Center for Advancing Translational Sciences (NCATS). B.S.K. was supported by grants from the NIH National Institute of Arthritis and Musculoskeletal and Skin Diseases (NIAMS) (R01AR070116 and K08AR065577), the Doris Duke Charitable Foundation, the American Skin Association, and LEO Pharma. M.R.M. and A.M.T. were supported by NIH National Institute of Allergy and Infectious Diseases (NIAID) training grant T32AI716339. A.M.T. was also supported by NIH National Heart, Lung, and Blood Institute (NHLBI) training grant T32HL007317. A.Z.X. was supported by the Howard Hughes Medical Institute (HHMI) Medical Fellows Program. The Vivier laboratory at CIML and Assistance-Publique des Hôpitaux de Marseille is supported by funding from the European Research Council (ERC) under the European Union’s Horizon 2020 research and innovation program (grant agreement no. 694502; TILC), the Agence Nationale de la Recherche including the PIONEER Project (ANR-17-RHUS-0007), Equipe Labellisée “La Ligue” (Ligue Nationale contre le Cancer), MSDAVENIR, Innate Pharma, and institutional grants to CIML (INSERM, CNRS, and Aix-Marseille University) and to Marseille Immunopole. J.R.B. was supported by the Children’s Discovery Institute (MI-F-2019-795), Burroughs Wellcome Fund (CAMS 1019648), and NIH Office of the Director (DP5 OD028125). M.M.B.-E. was supported by NCI F32CA200253 and NIH NHLBI T32HL00708843. T.A.F. was supported by NIH NIAID R01AI102924 and NCI R01CA205239. E.M.O. and M. Colonna were supported by NIH NIAID R01AI134035. D.J.M. was supported by NIH NIAMS R01AR069062 and NIH NIAMS R01AR070873. Author contributions: M.R.M., J.R.B., M.M.B.-E., H.N., E.P., A.Z.X., M.M., A.M.T., F.W., and T.-L.B.Y. conducted experiments. M.R.M., M.M.B.-E., J.R.B., P.L.C., J.A.W., E.P., P.L., M. Colonna, M. Cella, W.M.Y., E.M.O., T.A.F., and B.S.K. designed experiments and interpreted the data. M.R.M., D.J.M., D.S., F.W., N.D.B., H.N., A.Z.X., C.H., F.K., R.C., M.L.C., and B.S.K. saw patients, managed clinical research samples, or obtained consent. E.V. provided key reagents. M.R.M. and B.S.K. wrote the manuscript. Competing interests: B.S.K. has served as a consultant for AbbVie Inc., Concert Pharmaceuticals, Incyte Corporation, Menlo Therapeutics, and Pfizer Inc. B.S.K. is founder and chief scientific officer of Nuogen Pharma Inc. B.S.K. has participated in advisory boards for Cara Therapeutics, Celgene Corporation, Kiniksa Pharmaceuticals, Menlo Therapeutics, Regeneron Pharmaceuticals Inc., Sanofi, and Theravance Biopharma. M.L.C. has served as a consultant for MDOutlook. D.J.M. has consulted for LEO Pharma, Pfizer Inc., Sanofi, and Regeneron Pharmaceuticals Inc. M.R.M., J.R.B., and B.S.K. are authors of a patent application related to this work (018984/US). All other authors declare that they have no competing interests. Data and materials availability: All data associated with this study are present in the paper or Supplementary Materials. Human NK cell CyTOF data files (Figs. 2 and 3) are available on (FR-FCM-ZYV4). Custom code for CyTOF analysis (Figs. 2 and 3) is available at doi:10.5281/zenodo.3568404. Human NK cell RNA-seq raw data (Fig. 4) are available through the Gene Expression Omnibus (GEO) (GSE125916). Human AD lesional and nonlesional skin RNA-seq data (Fig. 5) are also available through GEO (GSE140227).

Stay Connected to Science Translational Medicine

Navigate This Article