CD8α marks a Mycobacterium tuberculosis-reactive human NK cell population with high activation potential

Abstract

Natural Killer (NK) cells can recognize and kill Mycobacterium tuberculosis (Mtb)-infected cells in vitro, however their role after natural human exposure has not been well-studied. To identify Mtb-responsive NK cell populations, we analyzed the peripheral blood of healthy household contacts of active Tuberculosis (TB) cases and source community donors in an endemic region of Port-au-Prince, Haiti by flow cytometry. We observed higher CD8α expression on NK cells in putative resistors (Interferon γ release assay negative; IGRA− contacts) with a loss of CD8α surface expression during household-associated exposure and active TB disease. In vitro assays and CITE-seq analysis of CD8α⁺ NK cells demonstrated enhanced maturity, cytotoxic gene expression, and response to cytokine stimulation relative to CD8α⁻ NK cells. CD8α⁺ NK cells also displayed dynamic surface expression dependent on MHC class I in contrast to conventional CD8⁺ T cells. Together, these results support a specialized role for CD8α⁺ NK cell populations during Mtb infection correlating with disease resistance.

Supplementary Information

The online version contains supplementary material available at 10.1038/s41598-025-98367-4.

Introduction

Tuberculosis (TB) remains the leading global cause of death from a single bacterial infection and is a significant public health concern^1,2. Exposure to Mycobacterium tuberculosis (Mtb) bacilli can result in a broad spectrum of clinical outcomes ranging from clearance of primary infection, asymptomatic latent TB infection (LTBI), or symptomatic active pulmonary TB disease^2,3. It is estimated that 25% of the global population has evidence of immune-reactivity to Mtb antigens—measured by Mtb peptide-specific Interferon γ (IFNγ) release assay (IGRA)—and 10% will progress to active pulmonary disease⁴. While IGRA testing is used to identify patients with prior exposure to Mtb, it cannot differentiate viable latent infection from past exposure and bacterial clearance⁵. There is also evidence that Mtb can be cleared by innate cells without adaptive immune priming⁶. The host immunologic determinants of these clinical outcomes remain poorly defined and are a major obstacle to designing effective TB vaccines and host-directed therapies to prevent disease. Most TB immunology studies focus on conventional peptide-specific T cells^7–11, but the presence of these cells have not been shown to correlate with protection^12–17. In contrast, few studies have examined Mtb-reactive innate lymphocytes that contribute to Mtb clearance^18–22.

Here, we focus on the most abundant innate lymphocytes in mammals: Natural Killer (NK) cells²³. NK cells are prevalent in the peripheral blood (0.1–18% of leukocytes) and populate tissue sites such as liver, lung, and mucosal barriers²⁴. They express a broad range of germ-line encoded pattern recognition receptors enabling rapid responses to infection without the need for antigen-specific expansion²⁵. Prior work provides mechanistic evidence for NK cell recognition of intracellular and extracellular Mtb in vitro through natural cytotoxicity receptor engagement of Mtb cell wall components or stress-induced self-ligands^26–38. However, few published studies directly examine NK cell biology in humans naturally exposed to Mtb^31,32,39.

In this study, we interrogate peripheral blood NK cells in a cohort of household TB contacts in an endemic region in Port-au-Prince, Haiti. We analyzed samples from active TB cases (n = 50) and their contacts from the same households (n = 50) in addition to age-, sex-, and IGRA-matched unrelated donors from the source community without household exposure (n = 50). We used flow cytometry to test our hypothesis that specialized NK cell subpopulations expand after household-associated Mtb exposure by comparing IGRA+ donors, who may be latently infected, and IGRA− contacts, who represent possible Mtb resistors. We identify preserved frequency of CD8α⁺ NK cells in the blood of household contacts who remain non-reactive to QuantiFERON-TB Gold testing (IGRA−) with a loss of CD8α surface expression on NK cells in immunoreactive household contacts (IGRA+) and active TB cases. In vitro functional assays of CD8α⁺ NK cells from healthy donors demonstrated enhanced responsiveness to cytokine stimulation compared to CD8α⁻ NK cells and dynamic surface expression in culture. Unexpectedly, most NK cells expressed CD8α protein intracellularly and cellular indexing of transcriptomes and epitopes (CITEseq) revealed largely overlapping transcriptional programs between CD8α^+/− subsets with select cytotoxicity-associated genes differentially expressed in the CD8α⁺ subset. Taken together, our results identify a novel subpopulation of Mtb-reactive CD8α surface-expressing NK cells associated with resistance to primary tuberculosis infection.

Results

Clinical cohort description

To study early events during the immune response to Mtb, we used biobanked samples from a cross-sectional study performed at GHESKIO Centers, Port-au-Prince, Haiti to assemble a clinical cohort of household contacts of active pulmonary TB cases and source community controls between 2015 and 2018 (NIH U19 AI111143, completed). A total of 78 active TB cases, 92 household contacts, and 605 community controls were recruited. We randomly sampled an age-, sex-, and IGRA-matched subset of 50 TB household contacts, 50 active TB cases from the same households, and 50 source community controls without household exposure, based on peripheral blood mononuclear cell (PBMC) availability and quality control (total n = 150). The overall cell viability was similar across all three patient cohorts, active TB (M = 69.4%, SD = 25.0%), household contacts (M = 75.8%, SD = 22.6%), and community controls (M = 81.2%, SD = 14.6%). All available specimens from IGRA− contacts (n = 14) were analyzed.

Age, sex, IGRA status, and relevant sociodemographic variables are summarized in Table 1. No significant variations were observed in the number of household members, household income, smoking status, or alcohol consumption between groups, consistent with previous work¹⁸. Among all recruited participants, 75.8% of household contacts were IGRA+ while 51.6% of community controls were IGRA+, consistent with increased transmission within the household. Fifty-percent of contacts slept in the same room as the active pulmonary TB index case. Among the 14 IGRA− household contacts selected in our study, only two IGRA conversions were detected during 6-month follow-up IGRA testing, supporting the premise that IGRA− contacts may resist Mtb infection without developing Mtb peptide-specific immunoreactivity.

Table 1.

Demographics of study participants in Haitian cohort (n = 150).

Variable	Control (n = 50)	Contact (n = 50)	Active TB (n = 50)
Age (mean ± SD)	30.1 ± 9.3	33.9 ± 17.8	20.2 ± 4.26
≤ 18	2 (4)	12 (24)	18 (36)
>18 and < 65	48 (96)	35 (70)	32 (64)
≥ 65	0 (0)	3 (6)	0 (0)
Female sex—no. (%)	26 (52)	30 (60)	26 (52)
IGRA positive—no. (%)	25 (50)	34 (68)	38 (76)

IGRA+ community controls have elevated frequency of mature CD56⁻ NK cells

We first compared the frequency of NK cell subpopulations between clinical cohorts using flow cytometry. Human NK cells were defined by the absence of lineage markers for T cells (CD3), monocytes (CD14), and B cells (CD19), and gated on surface expression of neural cell adhesion molecule (NCAM), CD56, and Fc receptor γRIIIA, CD16 (Fig. 1A). We stratified NK cells into three subsets based on expression of CD56: CD56⁻, CD56^dim, and CD56^bright40,41.

Fig. 1.

Frequencies of NK cell subpopulations are altered after Mtb infection. (A) Gating strategy employed for NK cells from a representative sample. NK cells were defined as CD19⁻, CD3⁻, CD14⁻ lymphocytes that expressed any level of either CD16 or CD56. NK cells were further subdivided into three subsets based on CD56 expression. (B) Proportion of total NK cells among live cells in clinical groups. (C) Proportion of each NK cell subset in clinical groups. Frequency calculated as percent of each subset among total NK cells. Brackets represent Wilcoxon statistical tests, with unadjusted p values. *p ≤ 0.05, **p ≤ 0.01.

We subdivided the community controls and household contacts based on IGRA status. The active TB cohort was stratified by completion of TB treatment. The statistical analysis plan included a priori hypothesis tests as pairwise comparisons between two clinical groups: IGRA− contacts vs. controls, IGRA+ contacts vs. controls, IGRA− contacts vs. IGRA+ contacts, IGRA− controls vs. IGRA+ controls, IGRA+ controls vs. Active TB, IGRA+ contacts vs. Active TB, and Active TB prior to vs. after therapy completion. This enabled testing the overarching hypothesis that specialized NK cell populations responded during asymptomatic household exposure and infection. There was no difference in the frequency of total NK cells or CD56^dim NK cells between clinical groups (Fig. 1B, C center). IGRA+ community controls had fewer CD56^bright NK cells (3.11% vs. 5.45%, p = 0.006) and more CD56⁻ NK cells (19.1% vs. 12.6%, p = 0.002) compared to IGRA− controls or IGRA+ contacts, indicating an accumulation of mature NK cells in immunoreactive donors within the source community (Fig. 1C, right and left).

IGRA+ contacts demonstrate a loss of circulating CD16⁺ NK cells

NK cell activity against Mtb depends upon complex combinatory signaling transduced by activating and inhibiting receptors^42–44. We next investigated markers of NK cell activation ex vivo. We analyzed representative receptors, including CD16a (FcγRIIIA), which binds the F_c region of antibodies and can induce antibody-dependent cellular cytotoxicity (ADCC). Within the major CD56^dim subset, IGRA+ contacts had decreased circulating CD16⁺ NK cells compared to IGRA+ controls (60% vs. 80%, p < 0.0001) with a reciprocal increase in CD16⁻ population, representing downregulation or shedding of CD16 surface expression in IGRA+ household contacts (Fig. 2A, B)⁴⁵. We found no remarkable differences in canonical markers of NK cell activation ex vivo including CD69 and CD25 (Fig. 2C–E). We observed decreased staining for degranulation marker CD107a in CD56^bright NK cells of IGRA− contacts, potentially representing recent activation events during household exposure (Fig. 2C). Expression of the HLA-E restricted receptor NKG2C significantly increased in active TB cases that completed treatment relative to donors actively receiving anti-TB therapy (Fig. 2D).

Fig. 2.

NK cells demonstrate loss of CD16 expression in IGRA+ contacts. (A) Representative flow plots of CD16 gating in IGRA+ contact and IGRA+ control. (B) Frequency of CD16⁺ in CD56^dim NK cells. Brackets represent Wilcoxon tests with unadjusted p-values. Percent positive staining for four activation markers in (C) CD56^bright, (D) CD56^dim, and (E) CD56⁻ NK cells. Brackets represent Wilcoxon tests with unadjusted p values. *p ≤ 0.05, ****p ≤ 0.0001.

NK cells of IGRA− contacts have a depressed functional phenotype

To assess functional activity of NK cells in vitro, we cultured PBMCs in media alone or with K562 cells—a leukemia cell line with low levels of HLA I expression known to activate NK cells—co-incubated at 1:20 K562:PBMC for 15 hours. Cells were then stained, and percent difference was calculated for CD69, CD107a, CD16, and NKG2C (Fig. S1A). Our positive control analyses in Fig. 3A display results from the major CD56^dim subset and demonstrate CD107a, CD69, and NKG2C were significantly increased with K562, while CD16 was significantly decreased, across all clinical groups.

Fig. 3.

IGRA− contacts demonstrate depressed NK cell function in vitro. PBMCs stimulated overnight by co-culture with K562 cell line compared to plain media alone, gated on CD56^dim NK cells. (A) Overall results of activation assay for CD56^dim NK cells for four activation markers. Y-axis represents mean percent positive for each activation marker. P values calculated as a global paired Wilcoxon test including all clinical groups. Each clinical group is displayed separately indicated by line color. (B) Percent difference of four activation markers (stimulated minus resting) compared between clinical groups. Brackets represent Wilcoxon tests with unadjusted p values. *p ≤ 0.05, **p ≤ 0.01, ***p ≤ 0.001, ****p ≤ 0.0001.

Next, we directly compared NK cell response to K562 between clinical groups. IGRA+ contacts demonstrated significantly more degranulation (CD107a) and NKG2C expression after stimulation relative to IGRA− contacts (Fig. 3B, top left and bottom left). IGRA+ contacts had an attenuated CD69 response relative to IGRA+ controls (Fig. 3B, top right). No significant differences in CD16 expression were detected. The same analyses applied to CD56⁻ and CD56^bright NK cells demonstrated similar responses to stimulation, but no significant differences between clinical groups, save an attenuated CD69 response in CD56^bright NK cells of active TB cases (Fig. S1B–E). Taken together with our ex vivo data, these in vitro restimulation results indicate that CD56^dim NK cells of putative Mtb resistors (IGRA− contacts) have depressed function, potentially representing sequelae of household-associated Mtb exposure.

Loss of circulating CD8α⁺ NK cells during household-associated Mtb infection

Despite their canonical classification as short-lived innate lymphocytes, NK cells are increasingly recognized for their marked diversity and ability to differentiate into longer-lived effectors^{19,21,31,46–49}, including after mycobacterial stimuli⁵⁰. The most extensively studied is an NK cell memory population co-expressing NKG2C and CD57 that expands in blood during human cytomegalovirus (CMV) infection^31,51. Multiple putative markers of NK cell memory during Mtb infection have been suggested, including CD57, CD8α, CD161, and CD27^32,39,50. We compared expression of these markers across clinical groups and found a loss of CD8α expression in IGRA+ contacts and active TB cases among CD56^dim NK cells (Fig. 4A). A similar trend was observed with the marker of terminal differentiation, CD57 and there were significantly fewer CD57⁺CD56^dim NK cells in IGRA+ contacts compared to IGRA− contacts (Fig. 4B). The frequency of CD8α⁺ T cells was also reduced in IGRA+ contacts relative to controls (Fig. S2A). CD8α⁺ NK cell frequency increased with age but was not significantly associated with sex (Fig. S2B). CD8α expression among CD56^bright NK cells was also significantly decreased in IGRA+ vs. IGRA− controls (Fig. S3A). CD57⁺CD56^bright NK cells expanded in IGRA+ contacts relative to IGRA+ controls (S3A). CD57 expression among CD56⁻ NK cells significantly decreased in TB cases who completed treatment vs. those on treatment (Fig. S3C). We did not observe any significant differences in the expression of CD27, CD127, or CD161 between clinical groups (Fig. S3A-C). Together, these results indicate that donors who develop household-associated immunoreactivity to Mtb antigens or progress to active TB disease demonstrate decreased CD8α surface expression on CD56^dim NK cells. In turn, CD8α⁺ NK cell frequency significantly correlates with resistance to primary household-associated infection as measured by IGRA immunoreactivity.

Fig. 4.

CD8α⁺ NK cells are depleted in the blood of cases with household-associated Mtb infection. (A) Representative gating and frequency of CD8α⁺CD56^dim NK cells in clinical groups. (B) Representative gating and frequency of CD57⁺CD56^dim NK cells in clinical groups. (C) Expression of select surface markers in CD8α⁺ vs. CD8α⁻ CD56^dim NK cells of healthy donors, with active TB excluded. (D) Activation assay with overnight K562 co-culture versus no K562 comparing CD8α⁺ vs. CD8α⁻ CD56^dim NK cells. Y-axis displays the percent difference for CD69 and CD107a. Asterisks represent paired Wilcoxon tests. (E) As in panel (D), instead displaying CD56⁻ NK cells. P values represent unadjusted Wilcoxon testing. *p ≤ 0.05, **p ≤ 0.01, ***p ≤ 0.001, ****p ≤ 0.0001.

CD8α⁺ NK cells represent a mature population with high functional potential

Despite the identification of cytotoxic CD8α⁺ NK cells over 30 years ago^52–54, the function of CD8α on NK cells during health and disease is not well-characterized and is of emerging interest as a cellular target against immune-mediated diseases^52,54–56. We next used flow cytometry to define co-expressed receptors of CD8α⁺ NK cells in our cohorts. In the Haitian cohorts, a mean of 27% of NK cells expressed CD8α on their surface. Among NK cell subsets, CD56^dim NK cells had the largest proportion of CD8 positivity (median: 36%, range: 1–74%) (Fig. S4A). We then defined co-expressed receptors on CD8α⁺ NK cells among all healthy Haitian donors, excluding active TB cases with chronic symptomatic disease. CD8α⁺ NK cells expressed significantly higher levels of CD57 and CD16, and lower levels of CD25 and CD127 (Fig. 4C). This indicates that CD8α⁺ NK cells represent more mature effectors compared to their CD8α⁻ counterparts. When compared to T cells, median fluorescence intensity (MFI) of CD8α surface expression was significantly lower in NK cells (Fig. S4B).

To compare CD8α^+/− NK cell function, PBMCs from the Haitian cohorts were co-incubated with K562 cells for 15 hours and compared to resting conditions as described above. CD8α⁺ NK cells were more responsive to K562 induction measured by CD69 upregulation compared to CD8α⁻ NK cells but demonstrated lower levels of degranulation as measured by CD107a staining (Fig. 4D). The propensity for CD8α⁺ NK cells to upregulate CD69 upon K562 stimulation was more pronounced among the CD56⁻ compartment, where the CD8α⁻ fraction had little to no response (Fig. 4E). Overall, our data suggest that CD8α expression among NK cells marks a population with higher activation potential, specifically among terminally differentiated CD56⁻ NK cells, that are reported to have depressed effector functions⁵⁷.

CD8α protein is ubiquitously expressed in NK cells

We next interrogated the immune phenotype and function of CD8α⁺ NK cells in a distinct biorepository of healthy human PBMCs (IRB2021-00478; PI: Vorkas) to validate our findings and perform additional functional assays. First, we confirmed the presence of CD8α expression in NK cells by extracellular staining with CD8α and CD8β antibodies. About 60% of NK cells did not express surface CD8, CD8αα was found in ~ 40% of NK cells, while rare CD8αβ NK cells were detected (Fig. 5A). In blood, CD8α was expressed in all three compartments of NK cells, and most frequently among CD56^dim cells, with median expression comparable to Haitian donors (Fig. 5B). We also validated our prior result from the contact cohort (Fig. 4C) demonstrating CD16 and CD8α were significantly co-expressed (Fig. 5C). As our TB contact studies only assessed surface CD8α on NK cells, we next analyzed intracellular CD8α expression by flow cytometry. We observed that nearly all CD56^dim NK cells expressed CD8α protein intracellularly, contrasting with T cells whose CD8 expression did not significantly differ between surface and intracellular staining (Fig. 5D). CD56^bright and CD56⁻ NK cells also demonstrated ubiquitous expression of CD8α protein intracellularly (Fig. S5). To test the stability of surface CD8α on NK cells over time, we incubated PBMCs for 1 or 7 days in pro-survival cytokine IL-2. We observed significantly increased CD8α surface expression following prolonged incubation, indicating that surface presentation of NK cell CD8α may be more dynamic than in T cells where it is a constitutively expressed surface receptor (Fig. 5E).

Fig. 5.

CD8α⁺ NK cells are prevalent in healthy US donors with ubiquitous intracellular CD8α expression. (A) Flow cytometry plot and percentage of CD8αα⁻, CD8αα⁺ and CD8αβ⁺ among total NK cells (n = 3). (B) Percentage of CD8α⁺ NK cells in different subsets based on CD56 expression (n = 10). (C) Frequency of CD16 expression in CD8α⁺CD56^dim NK cells. (D) Frequency and representative flow cytometry plots of CD8α expression comparing CD56^dim NK cells and T cells with extracellular (EC), IC (Intracellular), and EC+IC CD8α staining. (E) Percent CD8α⁺ after EC, IC, and EC+IC staining at day 1 and day 7 in CD56^dim NK cells and T cells. Statistical significance was measured using unpaired t tests, ns—no significance, *p < 0.05, **p < 0.005, ***p < 0.001, ****p < 0.0001.

CD8α⁺ NK cells have enhanced responses to cytokine stimulation

We next hypothesized that surface CD8α was associated with enhanced responsiveness to inflammatory signals, following prior reports of increased cytotoxicity against leukemic cells relative to CD8α⁻ NK cells^53,56. We stimulated PBMCs with either the potent IL15-IL15Rα sushi domain fusion construct (IL15-15RαSu)⁵⁸, IL15-15RαSu/IL12/IL18, IL12/IL18 alone, or virulent Mtb Erdman whole cell lysate and assessed intracellular IFNγ expression by flow cytometry. Cytokine stimulation led to decreased CD8α⁺ surface staining while Mtb stimulation did not alter CD8 surface expression, suggesting that cytokine stimulation induces downregulation of the CD8α receptor or cell death (Fig. 6A). Both IL15/IL12/IL18 and to a lesser degree Mtb lysate conditions induced IFNγ responses in CD56^dim NK cells compared to resting (Fig. 6B). When stratified by CD8α, we observed increased intracellular IFNγ with cytokine stimulation in both CD16^−/+ CD8α⁺ NK cells relative to CD8α⁻ cells, but no differences upon co-incubation with Mtb lysates (Fig. 6C). Responses were more robust among the CD8α⁺CD16⁻ NK cell population, likely representing highly activated cells downregulating the F_c Receptor. Percentage of CD25, CD69, and intracellular Granzyme B responses were also enhanced in cytokine-induced CD8α⁺ NK cells across CD56-stratified subsets (Fig. S6).

Fig. 6.

CD8α⁺ CD56^dim NK cells have enhanced response to cytokine stimulation. (A) Pairwise analysis of percent CD8α⁺ NK cells upon stimulation with IL15/12/18 and MtbLys 1:200 relative to resting condition (n = 9). (B) Pairwise analysis and representative flow plots of intracellular IFNγ in CD56^dim NK cells after stimulation with IL15/12/18 and MtbLys at dilution 1:200 relative to resting condition (n = 9). (C) Percent intracellular IFNγ⁺ in CD8α⁺ (triangles) versus CD8α⁻ (circles) CD56^dim NK cells comparing IL-15-Su, IL15-Su/12/18, IL-12/18, or MtbLys conditions (n = 9). Both the CD16⁺ (right) and CD16⁻ (left) CD56^dim NK cell subpopulations are displayed. (D) Representative flow cytometry plots, with %CD8α⁺ (top) and %CD8α⁻(bottom) of CD56^dim NK cells alongside absolute count per 10⁴ live cells after anti-MHC class I blockade compared to no antibody (rest)  or IgG2a isotype controls. Statistical significance was measured using paired t-test, with threshold of *p < 0.05, **p < 0.005, ***p < 0.001. (E) Same experimental setup as D, gated on CD8⁺ T cells. IFNγ Interferon γ, MtbLys Mtb whole cell lysate.

CD8α surface expression on NK cells is MHC class I-dependent

To determine a mechanism by which CD8α may modulate NK cell function, we hypothesized that CD8α on NK cells engages MHC class I molecules like CD8⁺ T cells^59,60. To test this, we incubated whole PBMCs with MHC class I blocking antibody (clone: W6/32) compared to IgG2 isotype and no antibody controls and measured surface CD8α expression. Strikingly, surface CD8α expression on NK cells was significantly reduced after MHC class I blockade (Fig. 6D). As MHC class I blockade may activate NK cells due to loss of inhibitory Killer-like immunoglobulin Receptor (KIR) engagement and potentially induce apoptosis, we measured frequency and CD107a staining to assess degranulation of CD8α^+/− NK cells during MHC class I blockade. We found that MHC class I blockade led to a significant decrease in the frequency and absolute numbers of total NK cells (Fig. S7A), accompanied by a significant increase in CD107a staining among both CD8α^+/− NK cells (Fig. S7B). These effects were unique to MHC class I blockade conditions and not observed with isotype control antibodies. Blockade was further validated by loss of surface staining of anti-HLA I antibodies with MHC class I blockade, but not isotype control (Fig. S7C). NK cell depletion with anti-MHC class I was explained by absolute loss of CD8α-expressing NK cells (Fig. 6D), without a consistent reciprocal increase in CD8α⁻ absolute numbers (Fig. 6D). These results may represent preferential induction of CD8α⁺ NK cell death or CD8α surface downregulation after MHC class I blockade. Importantly, CD8α surface expression on CD8⁺ T cells was not affected (Fig. 6E). Together, our results indicate that MHC class I engagement of NK cells may stabilize CD8α surface expression.

CD8α⁺ NK cells have overlapping transcriptomes with CD8α⁻ NK cells

The transcriptional programs associated with surface expression of CD8α in NK cells and whether these relate to homologous functions observed in CD8α⁺ T cells during disease is an area of emerging interest^{52–54,56,61}. To interrogate the transcriptional profile of CD8α^+/− NK cells, we sorted innate lymphocyte subsets from PBMCs including NK cells, γδ T cells, Mucosal-associated invariant T (MAIT) cells, and invariant Natural Killer T cells (iNKTs) by fluorescence-activated cell sorting (FACS) from two healthy donors. We performed cellular indexing of transcriptomes and epitopes single cell sequencing (CITE-seq) of these populations directly after sorting or after PBMC coincubation for seven days in IL2 supplemented media +/− Mtb lysates to define surface protein and transcript expression in CD8α^+/− NK cells. We discriminated NK cells from surrounding innate lymphocytes using surface antibody barcodes for CD56 and CD16 together with expression of canonical NK cell gene transcripts CD3E, NCR1, and NCAM1 (CD56) (Fig. S8)^52,62,63. Dimensionality reduction with UMAP identified four clusters of NK cells with distinct transcriptional signatures (Fig. 7A–D). To assess pathogen-specific responses, we first performed differential expression between the ‘rest’ and ‘Mtb lysate’ conditions and display the top eight most differentially expressed genes in Fig. 7E. The cytotoxic serine protease GZMB (Granzyme B) and killer immunoglobulin-like receptor, KIR3DL1, were the most upregulated genes in the Mtb lysate condition.

Fig. 7.

CITE-seq analysis of CD8α⁺ vs. CD8α⁻ NK cells from two healthy donors reveals overlapping transcriptomes. (A) Results of clustering and uniform manifold approximation (UMAP) projection for all NK cells. (B) UMAP reduction colored by Donor. (C) UMAP reduction colored by condition. (D) Violin plots representing top genes that were differentially expressed between the four NK cell clusters. (E) Violin plots of the eight differentially expressed genes between resting and Mtb lysate condition. (F) Distribution of NK cell expression of CD8 using barcoded antibodies against CD8α protein and CD8A transcript. (G) Violin plot of the expression of CD8A and CD8B transcripts, as well as CD8α surface expression in the four NK cell clusters. (H) Correlation of CD8α expression between transcript and protein measurement. (I) Annotated UMAP showing CD8α⁺ and CD8α⁻ NK cells based on CD8A transcript expression. (J) Percent of NK cells expressing CD8A transcript in stimulation conditions. (K) Dot plot of genes differentially expressed in CD8A^+/− NK cells.

We next analyzed the distribution of CD8A transcript and CD8α protein expression in NK cells using both transcript and surface antibody barcodes for CD8α. NK cells did not cluster independently based on the presence of CD8A/CD8α (Fig. 7F). The expression of the CD8A gene and CD8α surface protein was distributed throughout all NK cell clusters and a positive correlation was observed between protein and transcript expression (Fig. 7G–I). Low amplification of surface antibody barcodes for CD8α was observed in NK cells, in contrast with other CD8α-expressing immune subsets sequenced (e.g., MAIT cells; Fig. S9) consistent with CD8α MFI differences detected by flow cytometry (Fig. S4B).

To define transcriptional signatures associated with CD8A transcript expression, we stratified NK cells expressing greater than zero CD8A transcripts and performed differential expression relative to CD8A⁻ cells. Half of all NK cells expressed CD8A transcripts (Fig. 7J). Differential gene expression revealed few statistically significant differences between CD8A^+/− cells. Notable differentially expressed genes in CD8A⁺ NK cells include cytotoxicity-associated proteins natural cytotoxicity receptor 3 (NCR3) and the serine protease granzyme M (GZMM) (Fig. 7K).

Discussion

Identifying the correlates of protection against natural Mtb infection is an area of critical importance that will inform the next generation of TB vaccines and host-directed therapies^{8,42,50,64–67}. In this study, we hypothesized that investigation of household-associated Mtb exposure would reveal specialized NK cell populations that represent early events following high-risk transmission. Our results identify a CD8α-expressing NK cell subpopulation with preserved frequency in the blood of contacts who lack immunoreactivity to Mtb peptide pools ESAT6 and CFP10 (IGRA-; QuantiFERON-TB Gold negative) accompanied by decreased CD8α surface expression among immunoreactive (IGRA+) household contacts and active TB cases. Few studies have interrogated CD8α⁺ NK cells, as this receptor is most commonly understood as a lineage marker for T cells⁶⁸ and is absent in murine NK cells⁶⁹. Recent work reporting an association of CD8α⁺ NK cells with protection against human infectious, autoinflammatory and malignant disease has generated interest in targeting this subpopulation for host-directed therapeutics^44,52,53,56. However, little is known how CD8α receptors may modulate NK cell activity and potentially enhance cytotoxic functions during infectious and non-infectious diseases^53,55,61. Here, we show that the majority of peripheral blood NK cells can express CD8α protein intracellularly, but neither CD8A transcript nor CD8α protein marked a distinct transcriptional state in healthy donors. This indicates that the majority of NK cells have the potential to express CD8α at their surface, but the conditions through which this expression is induced or maintained is not well understood. Our study demonstrates that surface expression of CD8α on NK cells is attenuated by MHC class I blockade and that CD8α marks an NK cell population with high activation potential, including enhanced responsiveness to cytokine stimuli in vitro. Together, this work highlights a novel innate effector associated with human tuberculosis infection and provides important insights into NK cell biology and innate TB immunology.

To our knowledge, there have been only two published household TB contact studies that specifically examined NK cell biology during household-associated Mtb exposure and infection^31,39. Our study expands upon the sample size of this prior work and provides additional analyses of NK cell subpopulations stratified by CD56 not previously addressed^47,70,71. Specifically, we observed decreased frequency of immature CD56^bright NK cells with reciprocal increase in mature CD56⁻ NK cells among community IGRA+, in line with findings during chronic viral infections such hepatitis C and HIV^72–74. We also observed that CD16 expression was increased in community IGRA+ relative to household-associated IGRA+, which may represent a late effect after exposure associated with immunoreactivity within the CD56^dim subset⁶⁷. Alternatively, this observation could be explained by a relative loss of CD16 cell surface expression following household-associated exposure, potentially due to receptor engagement leading to downregulation or shedding⁴⁵. Together, these changes in frequency of CD16 and CD56 expression indicate a gradual expansion of mature NK cells in peripheral blood after asymptomatic infection^{33,45,70,71,73,75}, accompanied by both markers of maturity and dysfunction (loss of CD56), but also enhanced potential for ADCC (gain of CD16).

Two phenotypic markers—CD8α and CD57—demonstrated surface expression that was both independently and inversely associated with infection in households. CD8α has long been recognized as a surface marker on human NK cells^52–55,76 and non-human primate NK cells^69,77. In humans, CD8α expression on NK cells has been linked to enhanced cytotoxicity^53,56,61 and has even been proposed in one study as a potential Mtb memory marker⁵⁰. Recent studies in prospective longitudinal human cohorts report that the loss of CD8α expression—at both the transcript (CD8A) and protein levels—correlates with HIV progression^44,54, relapsing multiple sclerosis⁵², and recurrence of acute myeloid leukemia after stem cell transplantation⁵⁵. In this context, CD8α⁺ NK cells are hypothesized to play cytotoxic and regulatory roles in disease prevention. Sorted surface CD8α⁺ NK cells also demonstrate enhanced anti-leukemia cytotoxicity in vitro compared to sorted CD8α⁻ cells⁵⁵ and most recently IL-15 induced CD8α⁺ NK cells were found to have potent anti-leukemia activity⁷⁸. We make novel contributions to this literature by reporting that human CD8α⁺ NK cells are highly prevalent in both Haitian and healthy US donors with ubiquitous intracellular CD8α protein expression. Our finding indicates that—in contrast to CD8α⁺ T cells that express surface CD8α protein constitutively—NK cell CD8 expression is dynamic and most NK cells have the potential to express CD8α on the cell surface and enhance effector function. Moreover, we report that CD8α⁺ NK cells have high activation potential and are very sensitive to cytokine stimulation, suggesting complex roles during inflammatory immune responses against disease. As CD8α surface staining was strongly dependent on MHC class I expression^55,60,79, our study raises critical questions for future investigation that will define the molecular mechanisms of NK cell CD8α engagement with MHC class I and to determine whether this may be modulated by binding to antigen-specific peptide-MHC class I complexes. Taken together with previously published work^52–54, CD8α⁺ NK cell subpopulations emerge as an attractive target in vaccination and host-directed approaches against infectious and noninfectious diseases⁸⁰.

In addition to these mechanistic insights, we present the first multi-omic analysis of CD8α⁺ NK cells using CITE-seq and show that in contrast to T cells, CD8α does not mark a distinct transcriptional state of NK cells in healthy donors assayed. We observed few differentially expressed cytotoxic genes in CD8A+ NK cells including NCR3 and GZMM. These results will need to be validated in other healthy donors as well as compared to disease states, including TB contact cohorts.

There are certain limitations inherent to our human immunology study design. While household contacts experienced high-risk exposures to pre-treatment smear+ active TB cases, this does not exclude the possibility they may have been exposed or infected in the community. We control for this using source community donors to account for background transmission. Similarly, IGRA− community controls may have also been recently exposed to tuberculosis, though this group represents the most rigorous control for IGRA− household contacts. Our study was conducted in one TB endemic population in Haiti and will need to be validated in other cohorts. Due to the largely cross-sectional study design, we were unable to follow study volunteers beyond 6 months for IGRA− contacts, of whom only two converted to IGRA+ providing insufficient sample size to analyze correlates of IGRA conversion prospectively. Finally, all our biological samples in TB cohorts were derived from human blood and may not represent the abundance or function of NK cells within infected lung tissue. Future studies in Mtb-infected donors with multi-omic validation are needed to understand the role of CD8α⁺ NK cells at primary sites of infection.

In sum, our study adds to the growing literature highlighting a protective role for CD8α⁺ NK cells against both infections and non-infectious immune-mediated disease. Ongoing work will define how surface CD8α expression on NK cells may enhance targeting of Mtb-infected cells and will inform future cellular therapies and vaccination approaches to prevent or treat TB disease.

Materials and methods

Study design and participant recruitment

Haitian donors were enrolled through the Tri-Institutional Tuberculosis Research Unit (TBRU) at the Groupe Haitien d’etude du Sarcome de Kaposi et des Infections Opportunistes (GHESKIO) Centers under Institutional Review Board Approval at Weill Cornell Medicine and GHESKIO Centers in Haiti. All research was performed in accordance with the Declaration of Helsinki. All participants provided written informed consent, including consent for future use of biobanked samples for research. A dedicated clinical field team at the GHESKIO Centers in Port-au-Prince, Haiti recruited these study volunteers as part of a NIH-funded TBRU (AI111143). Immunoreactivity to Mtb was detected using QuantiFERON-TB^® Gold (QIAGEN), and active TB was excluded by clinical screening for signs and symptoms of active TB including fever, cough, night sweats, weight loss, and abnormal findings on Chest X-ray. Most contacts were recruited close to the time of diagnosis of the index case and were living in the same house as the active TB case for at least 1 month in the 6 months prior to diagnosis. The mean time to contact recruitment from TB case diagnosis was 14 days and the IQR was 0–1 days; 80% of contacts were recruited within 24 hours of index case diagnosis. To select for high transmission settings, each household was required to have a history of two active pulmonary TB cases. Healthy donors without reported TB exposure from the same community were recruited as controls. All cases with active pulmonary TB received periodic follow-up appointments while on treatment, and anyone with known contact with an active TB patient who remained IGRA− received 6-month follow-up and was rescreened with IGRA. The median smear grade of the active pulmonary TB cases in the households was 2–3+, conferring high risk for person-to-person transmission⁸¹. All donors provided informed consent prior to peripheral blood donation for PBMC isolation and cryopreservation in 90% FBS/10% DMSO. All donor samples were de-identified on site using a barcode system and stored in −80 °C in Port-au-Prince for no more than 1 month prior to transfer to liquid N₂ tanks in New York, New York and eventual transfer to Stony Brook University for analysis. Healthy donors were also recruited at Stony Brook University (SBU) under IRB2021-00478 (PI: Vorkas) and the New York Blood Center (NYBC), and were used for in vitro functional assays and CITE-seq.

PBMC isolation

PBMCs were isolated from peripheral blood at GHESKIO (Port-au-Prince, Haiti) using the Ficoll-Paque (GE Healthcare) density centrifugation, frozen in 5 × 10⁶ cells/ml aliquots in 90% FBS/10% DMSO (Thermo Fisher Scientific) and stored at  −80 °C for no more than one month prior to being shipped on dry ice to New York and transferred to liquid N₂. PBMCs from healthy donors recruited at SBU and New York Blood Center (NYBC) were isolated using Ficoll-Paque, BD Vacutainer CPT cell preparation tube (Becton, Dickinson and Company) or SepMate tubes (Stem Cell Technologies) using the manufacturer protocol. PBMCs isolated at SBU were cryopreserved in serum free Bambanker medium (GC Lymphotec) and stored in liquid N₂.

Flow cytometry acquisition and analysis

A complete list of fluorescent antibodies is summarized in Supplementary Table 1: Reagents and resources. All antibodies were commercially acquired, and multicolor panels were designed and validated using fluorescent minus one (FMO) controls. MR1 tetramers were obtained from the NIH Tetramer Core Facility (Atlanta, GA). In brief, cryopreserved PBMCs were thawed, plated, Fc receptor blocked (eBioscience), and underwent direct ex vivo extracellular staining for 15 min at room temperature (RT). For intracellular staining, PBMCs were incubated with 1X Brefeldin A (BioLegend) for the final hour of incubation prior to staining with intracellular antibodies. Samples were then washed twice with FACS buffer and acquired using an Aurora Spectral Analyzer (Cytek). Flow cytometry analysis, including manual gating, was performed using FCS Express 7 (De Novo Software) or R (v4.3.1) with the following library packages downloaded from Bioconductor: flowWorkspace, ggcyto, and flowCore.

Mtb whole cell lysate

Mtb whole cell lysate was prepared in Biosafety Level 3 facility at SBU. Fifty mL of Mtb Erdman 7H9 culture was grown to OD₆₀₀ 0.6, centrifuged at 4000×g for 10 min at RT and pellet was resuspended in 2.5 ml of 10 mM Tris-HCl buffer (pH 7.5) and rotated for 2 h at RT. The mixture was centrifuged at 500×g for 15 min at RT and aqueous phase was collected and filtered twice through 0.22 µm spin filter (Corning Costar Spin-X) at 6000×g for 5 min. The aliquots were stored at −80 °C for future use.

NK cell co-culture with K562 cell line

K562 cells (ATCC) were incubated in Iscove’s Modified Dulbecco’s Medium (IMDM) supplemented with 10% FBS. Cryopreserved K562 cells were thawed at 37 °C water bath and cells were cultured at 1 × 10⁵ viable cells/ml in T-75 flask as per manufacturers protocol. Once the cell density reached 1 × 10⁶ cells/ml, K562 was co-incubated with PBMCs at 1:20 K562 to PBMC in a round bottom clear plate (Greiner) for 15 hours, at 37 °C with 5% CO₂. At the beginning of the incubation period, anti-CD107a antibody was added to the culture. After incubation was complete, cells were washed, stained and acquired by flow cytometry.

NK cell stimulation assay

Cryopreserved healthy donor PBMCs from SBU and NYBC were used for NK cell stimulation assays. Thawed PBMCs were cultured 1 × 10⁵ cells per well in round bottom clear plate (Greiner) for 15 h, at 37 °C with 5% CO₂. Samples were stimulated and incubated overnight in complete RPMI 1640 media (1X) with 2 mM L-Glutamine, 10 mM HEPES, 1 mM MEM non-essential amino acids, 1 mM sodium pyruvate, 0.05 mM 2-mercaptoethanol, and 10% (v/v) heat-inactivated fetal bovine serum (Gibco) supplemented with IL-2 (50 U/ml; PeproTech). The cytokine stimulation conditions were performed with commercially available recombinant IL-12 (100 ng/mL; BioLegend), recombinant IL-18 (100 ng/mL; Biolegend) and IL-15-Su construct (1 µg/ml) containing a truncated Sushi domain of IL-15Rα synthesized by the Cheung lab (MSKCC), that is demonstrated to be a more potent and stable compound with longer half-life than recombinant IL-15⁵⁸. For Mtb stimulation conditions, 1:100 and 1:200 dilutions of Mtb whole cell lysate were used.

MHC class I Blockade

Cryopreserved healthy donor PBMCs from SBU and NYBC were used for experiments with MHC class I blockade. PBMCs were thawed at 37 °C and incubated in complete RPMI media with 10% heat-inactivated FBS and supplemented with IL-2 (50 U/ml; PeproTech). 1 × 10⁵ PBMCs were plated per well in round bottom clear plates (Greiner) and incubated with or without 5 µg InvivoMAb anti-human MHC class I (clone W6/32; Bio X cell) or 5 µg Invivo MAb mouse IgG2a isotype control (clone C1.18.4; Bio X cell) for 15 hours, at 37 °C with 5% CO₂ before staining and analysis. CD107a (PE-Cy5, clone H4A3; BD Pharmingen) was added during overnight incubation using the manufacturer’s recommended concentration i.e. 20 µl for 1 × 10⁶ cells. After incubation was complete, cells were washed, stained and acquired by flow cytometry.

scRNA-seq and CITE-seq library generation and sequencing

PBMCs were stained with fluorescent antibodies for cell sorting of MAIT cells (CD3⁺, CD161⁺, MR1-5-OP-RU tetramer⁺), NK cells (CD3⁻, CD19⁻, CD14⁻, CD16^+/− ,CD56^+/−), γδ T cells (CD3⁺, TCRγδ⁺) and iNKT cells (CD3⁺CD161⁺ Vα24-Jα18⁺) using FACS. The sorted cells were blocked with human Trustain, FcX blocking reagent (BioLegend) for 10 min at 4 °C and then stained with TotalSeq-C barcoded tagged and hashtags antibodies for 30 min at 4 °C in a total volume of 100 μL. The residual TotalSeq-C and hashtags antibodies were removed by centrifugation at 400xg for 5 min at 4 °C. A complete hashtag and TotalSeq-C antibody list is provided in Supplementary Table 1: Reagents and Resources.

Gene expression and surface antigen library preparation was conducted at the Stony Brook University Single Cell Genomics Facility (Director: Rosati). Briefly, cell viability and concentrations were determined on a Countess 3FL with trypan blue staining. Approximately 5,000 live cells from each condition were loaded in a single 10× Genomics 5′ NextGEM Single Cell assay reaction (30,000 total cells per reaction), to perform 5′ single cell immune profiling assay using a Chromium iX instrument (10× Genomics). Gene expression (GEX) and surface antigen (SA) libraries were prepared, and quality controlled using Chromium NextGEM 5′ Single Cell with Feature Barcode assay and a TapeStation 4200, according to the manufacturer’s instructions. The libraries were sequenced at a depth of approximately 25,000 (GEX) and 10,000 (SA) paired reads/cell on an Illumina NovaSeqX sequencer (Novogene, Inc).

The gene expression and surface feature barcode raw FASTQ files of single cell RNA-seq were analyzed using Cell Ranger (10× Genomics, v.7.0) (https://www.10xgenomics.com/analysis-guides/demultiplexing-and-analyzing-5%E2%80%99-immune-profiling-libraries-pooled-with-hashtags) multi-pipeline with human reference transcriptome (GRCh38) downloaded from 10× Genomics website and a hashtag configuration file, which contained the barcodes related to each different samples.

Single-cell RNAseq analysis

The generated gene expression matrix files were analyzed with R (4.3.2) using the Seurat (5.0) package and cells were filtered based on following criteria: (1) the number of genes was ≥ 200 and ≤ 6000 and (2) the mitochondrial content was less than 5%. Using the FindVariableFeatures() function we selected the top 3000 features with highest variance. Then, different samples were integrated using the harmony algorithm and scaled. Log-normalization and principal component analysis (PCA) was performed on integrated values. The top 20 principal components were selected for clustering and visualization with UMAP using the default Louvain algorithm. Differential gene expression was performed using FindAllMarkers() with a minimum threshold of 25% of cells expressing the gene of interest.

Statistical analysis

Statistical analysis was performed using R (4.3.2) and GraphPad Prism (10.1.2). Non-parametric testing with Wilcoxon tests were used to compare the means between groups for flow cytometric markers. For flow cytometric analysis of Haitian TB contact cohort PBMCs, only comparisons a priori hypothesized to represent biological features were tested and p values were not adjusted for multiple comparisons. The statistical analysis plan included pairwise comparisons between two clinical groups: IGRA− contacts vs. controls, IGRA+ contacts vs. controls, IGRA− contacts vs. IGRA+ contacts, IGRA− controls vs. IGRA+ controls, IGRA+ controls vs. Active TB, IGRA+ contacts vs. Active TB, and Active TB prior to vs. after therapy completion. This enabled testing the overarching hypothesis that specialized NK cell populations responded during asymptomatic household exposure and infection.

Electronic supplementary material

Below is the link to the electronic supplementary material.

Author contributions

The following authors contributed to the manuscript in various ways. Conceptualization: KH, JBL, CKV. Methodology: NKC, AP, RK, KCH, JBL, CKV. Investigation: NM, AP, RK, TK, BR, NKC, HX, JBL and CKV. Visualization: NM, AP, RK, JB, BR, DEC, JBL, and CKV. Funding acquisition: CKV. Writing—original draft: NM, AP, CKV. Writing—review and editing: all authors.

Funding

This study was supported by the Tri-I TBRU, part of the TBRU Network (U19 AI11143). CKV acknowledges support from the NIAID T32 Pathogenesis of Infectious Diseases Training Program (T32AI007613-18; Roy Gulick, Weill Cornell Medicine), the IDSA Education and Research Foundation Postdoctoral Fellowship Award in Infectious Diseases, a Potts Memorial Foundation Award, NIAID K08AI132739, R21AI171578, and R21AI183259.

Data availability

De-identified clinical information will be made available upon request to the corresponding author charles.vorkas@stonybrookmedicine.edu. Transcriptomic data has been deposited on the NCBI GEO database under the bioproject accession ID: PRJNA1219683 and the SRA IDs SRR32309740 - SRR32309745.

Declarations

Competing interests

The authors declare no competing interests.