Multi-omic atlas of NKG2A+ and NKG2C+ single immune cells and patients from infection, autoimmune, and cancer cohorts
We integrated single cell and bulk multi-omic datasets from seven highly phenotyped human clinical cohorts for infection (SARS-CoV-2 n = 296 patients, and chikungunya n = 231 patients), autoimmunity (SLE n = 162 patients), and pan-cancer (n = 11,180 patients) (Fig. 1a, Table S1-7, see Methods)24–29. As NKG2A/C expression is dominantly restricted to CD8+ T cells and NK cells, we focused on these two cell types for single cell analyses and presumed bulk NKG2A/C measurements to represent expression from these two cell types alone (Fig. 1b)30. Since NKG2A/C receptors are known to form heterodimers with CD94 for function, we account for CD94 expression when classifying single cells and bulk samples as NKG2A+ or NKG2C+ biased (Fig. 1c, see Methods)31,32. Given the divergent biological signals sent by these two receptors, inhibitory for NKG2A and stimulatory for NKG2C32, and their restriction to T and NK cells30, we hypothesized that NKG2A or NKG2C are likely involved in shared immunopathology and protection factors across disease contexts. Thus, we took on a pan-disease investigation of how NKG2A+ and NKG2C+ T cell and patient biases associate with biological and clinical markers of disease presence, severity, and patient mortality across three divergent disease states (Fig. 1d).
Patients with infection and NKG2A+ bias have decreased disease severity, mortality and prevalence of post-acute chronic disease.
We start by examining patients with infectious diseases, with a focus on our previously reported longitudinal COVID-19 cohort due to its depth of matched clinical and biological profiling24,25,33–35. Consistent with previous works, our assigned NKG2A+ and NKG2C+ cells presented with divergent phenotypes with NKG2C+ cells appearing more activated, in both the CD8+ T and NK cell compartments, likely due to the stimulatory nature of the NKG2C receptor (Fig. S1a)31. Of note, NKG2C+ CD8+ T cells represented only one percent of total CD8+ T cells, suggesting that they are not CMV-driven populations. NKG2C+ CD8+ T cells are known to expand to a significantly larger fraction of patient CD8+ T cells upon CMV-driven stimulation36. Further, in line with the presence of NK-related NKG2A/C receptors on these unique CD8+ T cells, both NKG2A+ and NKG2C+ cells possessed increased NCAM1 (CD56) gene expression and surface protein levels compared to other CD8+ T cells; CD56 is known to denote NK-like T cells (Fig. S1b)37–40. Further, KIR proteins were markedly upregulated on NKG2A+ and especially NKG2C+ CD8+ T cells; KIRs have been reported on NK-like T cells and to be involved in immunoregulatory behaviors (Fig. S1c)9. We see additional confirmation of our NKG2A+ and NKG2C+ assignment in the NK cell compartment where NKG2C+ NK cells indeed present with an adaptive-like phenotype which NKG2C+ NK cells are known to acquire (Fig. S1d)35,41. Thus, we confirm the validity of our NKG2A+ and NKG2C+ cell assignment in all cellular compartments and demonstrate their phenotypic divergence from each other.
To investigate the biological and clinical impacts of an NKG2A/C immune bias during infection, we assigned each patient as NKG2A+ or NKG2C+ biased based on their ratio of NKG2A+ to NKG2C+ cells; this was done separately for each cell type (see Methods). Interestingly, patients with an NKG2A+ bias had significantly greater odds of surviving than those with an NKG2C+ bias even when accounting for the effects of sex, age, and disease severity (Fig. 2a). Similar trends, albeit less significant, associating an NKG2A+ bias with increased survival were observed in validation cohorts even when accounting for demographic factors (Fig. S2a). Given that expansion of NKG2C+ NK cells are associated with prior CMV infection, a known risk factor in infection contexts, we repeated these analyses based on whether a patient had prior CMV infection as a co-variate (see Methods). Interestingly, even when we accounted for prior CMV infection, NKG2A+ biases were still significantly associated with greater odds of survival suggesting that the benefit derived from an NKG2A+ bias is not explained by CMV infection history alone (Fig. S2b).
To probe for the long-term impact of NKG2A+ biases, we compared long COVID profiles of patients who did and did not have NKG2A+ biased immune systems during acute disease. Intriguingly, we observed significant protection against long COVID in patients with an initial NKG2A+ bias, even when accounting for sex, age, and disease severity (Fig. 2b). We found NKG2A+ biased NK cells to associate with greater protection than CD8+ T cells for all symptom groups except for anosmia/dysgeusia, colloquially called loss of smell/taste. This may be explained by previous observations that anosmia/dysgeusia are driven by fundamentally different pathology than unresolved inflammation42–46 which associates with other post-acute sequelae. As with the survival analyses, we confirmed that NKG2A+ biases, especially those in the NK cell compartment, significantly associate with long COVID protection even when accounting for prior CMV infection (Fig. S2c).
The critical role of inflammation in exacerbating severity in acute and post-acute disease suggests that the benefit derived from an NKG2A+ bias may in some way be related to quelling inflammation. Consistent with this suggestion, patients with an NKG2A+ bias had significantly decreased prevalence of pre-existing chronic conditions even when accounting for demographic differences (Fig. 2c). Many of these co-morbidities are known to have inflammation-driven origins or have pathology intimately tied with inflammation, such as chronic obstructive pulmonary disorder (COPD) and coronary artery disease (CAD)47,48. Thus, these findings strongly suggest, across multiple infection contexts and cohorts, that an NKG2A+ bias confers protection during acute and post-acute disease given findings of reduced mortality, severity, and long-term symptoms.
NKG2A+ biases correlate with reduced pathogenic inflammation during infection
To investigate the suggestion from clinical data that NKG2A+ biases may confer protection through reduced inflammation, we interrogated the biological profiles of NKG2A+ and patients with an NKG2C+ bias for differences in inflammatory proteins and cell types. Consistent with the clinical data, patients with an NKG2A+ bias had significantly downregulated levels of inflammatory proteins (Fig. 2d, Table S8). By contrast, patients with an NKG2C+ bias had plasma proteomes enriched for inflammatory pathways, such as IFNγ response and granulocyte chemotaxis, and upregulated well known inflammatory proteins, such as IFNγ and CXCL1, even into post-acute disease (Table S9)49–51. This elevation of inflammatory proteins in patients with an NKG2C+ bias was confirmed in our validation cohorts, suggesting that this association holds across different infection contexts and cohorts (Fig. 2e, Table S10-12).
Probing deeper into this connection between inflammation and NKG2A/C biases, we compared the immune cell profiles of patients with NKG2A+ biases to those with NKG2C+ biases. Even during post-acute disease, patients with an NKG2C+ bias presented with signs of continued inflammation as observed through significantly higher levels of cytotoxic CD4+ T cells, CD4+ T cell clonal expansion, and perforin secretion capabilities by those cells (Fig. 2f). This continued reactivity in the CD4+ T cell compartment was accompanied by a simultaneous increase in plasmablast percentages (Fig. 2f right). Given the important collaborative roles of CD4+ T and B cells in humoral immunity, we interrogated patients for differences in antibody levels. Notably, while plasmablast levels were upregulated in patients with an NKG2C+ bias, it was patients with an NKG2A+ bias who had greater odds of developing anti-SARS-CoV-2 antibodies (Fig. S3a). In contrast, an NKG2C+ bias associated with increased levels of autoantibodies, particularly anti-IFNα2, and the presence of atypical memory (AtM) B cells during acute disease. AtM B cells are often associated with autoantibodies and autoimmunity (Fig. S3b-c)52,53. This phenomenon of humoral immunity divergence was also observed in the validation cohorts, where we found genes upregulated by patients with an NKG2A+ bias in the chikungunya cohort to display significant overlap with antibody associated genes and near significant overlap for the other COVID-19 cohort (Fig. 2g). Thus, we demonstrate that, across multiple infection contexts and cohorts, NKG2A+ biases not only associate with clinical metrics of protection but also with biological metrics of protection as observed through reduced short- and long- term inflammation, reduced autoreactivity, and increased protective humoral immunity.
NKG2A+ CD8+ T cells associate with protection and reduced inflammation in lupus
Signs that NKG2A+ biases may potentially associate with protection against autoimmunity pushed us to investigate the role of NKG2A+ biases in autoimmune disease. The most well characterized of these is systemic lupus erythematosus (SLE, lupus)26,54. Given recent reports of the importance of CD8+ T cells in lupus settings and their ability to carry NKG2A/C receptors, we focused our autoimmune analyses on the clinical and biological impacts of an NKG2A+ bias in the CD8+ T cell compartment55.
NKG2A+ and NKG2C+ CD8+ T cells, when projected onto a transcriptome defined UMAP, present with canonical CD8+ T cell phenotypes along with a prominent short-lived effector cell (SLEC)-like population (Fig. 3a). SLEC-like cells displayed upregulated levels of CD57, confirming their terminal phenotype, and increased IFNγ mRNA, which suggests that SLEC-like cells may play a role in lupus pathology given previous demonstrations of IFNγ-dependent inflammation and activation of autoreactive cells in patients with lupus (Fig. 3b)56–58. When we interrogated SLEC-like cells for NKG2A+ or NKG2C+ biases, we found a strong NKG2C+ bias that was accompanied by a near complete clinical bias towards patients with lupus (Fig. 3c upper). In contrast, non-SLEC-like cells were nearly all NKG2A+ and almost solely derived from healthy donors (Fig. 3c lower). We statistically confirmed both the increased prevalence of NKG2C+ cells in patients with lupus and the lupus-specific CD8+ T cell bias towards a SLEC-like phenotype (Fig. 3d). Further, we confirmed the SLEC-like phenotype of NKG2C+ cells through flow cytometry where we found NKG2C+ cells to present significantly more frequently than NKG2A+ cells as CD57+IL7R−, which is the literature definition of SLEC cells (Fig. S4a-c, Table S13)59,60. Thus, we both demonstrate the increased prevalence of NKG2C+ CD8+ T cells in patients with lupus, but we also experimentally validated their inflammatory SLEC-like phenotype that may, possibly through IFNγ secretion, allow them to play pathogenic roles.
To explore the larger immunological impacts of an NKG2A+ bias, we assigned each patient with lupus and each healthy donor as NKG2A+ or NKG2C+ biased based on their ratio of NKG2A+ and NKG2C+ CD8+ cells (see Methods). Consistent with our CD8+ T cell analyses, patients with an NKG2C+ bias were more likely to have lupus even when accounting for demographic co-variates (Fig. 3e). When we compared the percentages of other immune subtypes against patients with either an NKG2A+ or an NKG2C+ bias, we observed that an NKG2C+ bias associated with a clear increase in inflammatory classical monocytes and cytotoxic T cells. In contrast, an NKG2A+ bias associated with increased percentages of naïve CD4+ and naïve CD8+ T cells (Fig. 3f). These immune cell associations match what was observed for patients in infection contexts. For example, inflammatory cell types are overrepresented in patients with an NKG2C+ bias. Thus, we demonstrate that, across disease classes, NKG2A+ biases are consistently associated with both clinical protection as well as biological markers of protection, such as reduced inflammation and hyper-activation.
Patients with cancer and NKG2A+ bias have increased survival across cancer types
While we observed NKG2A+ biases as associated with protection for infection and autoimmune contexts, these are diseases where inflammation is closely tied to, if not the source of, disease pathology. Thus, an NKG2A+ bias is understandably protective, as it provides cells an additional method to receive inflammation quelling signals. However, in cancer contexts, inflammation has been claimed as both beneficial and harmful; beneficial for the promotion of immune activation and infiltration and harmful due to inflammation- and activation- induced apoptosis of anti-tumor immune cells11,61–63. Thus, the impact of NKG2A+ biases in cancer contexts is unclear. To address this, we compiled single cell and bulk profiles of 397,810 tumor-infiltrating CD8+ T cells and 11,180 patients with cancer across 33 different cancer types (Fig. 4a)28,29. In addition, we gathered spatial transcriptomics data from five different patients for breast, prostate, and ovarian cancer to understand how cell-cell interactions differ between NKG2A+ and NKG2C+ CD8+ T cells across cancer types64.
To assess the prevalence of NKG2A+ and NKG2C+ biases across different cancer types, we measured the percentages of NKG2A+ and NKG2C+ CD8+ T cells in cancer types for which we had large numbers of patients with single cell data. Interestingly, not only did we observe widespread prevalence of these immune cell subsets amongst cancer types, but cancer types differed in their tendency for an NKG2A+ and NKG2C+ biased response (Fig. 4b). Interestingly, melanoma (MELA), which is well known for fostering an immunogenic tumor microenvironment (TME)65 appeared significantly NKG2A+ biased. In contrast, thyroid cancer (THCA), which presents as a cold tumor with few immunogenic antigens66 was significantly NKG2C+ biased. In line with these observations, we found NKG2A+ biased cancer types to mildly associate with greater tumor immunogenicity (Fig. S5a). This suggestion of NKG2A+ association with pro-immune response TMEs prompted us to compare the long-term survival of patients with either an NKG2A+ and or an NKG2C+ biased tumor. Interestingly, NKG2A+ biased tumors associated with significantly increased pan-cancer patient survival rates, with especially prominent survival benefits for specific cancers, such as certain renal cancer subtypes (Fig. 4c, S5b). For no cancers did an NKG2C+ bias confer a survival advantage. Thus, across a broad range of cancer types, we find that NKG2A+ biases are positively associated with cancer survival. In contrast to previous suggestions that NKG2A may function as a druggable immune checkpoint23, our analyses suggest that instead, treatments designed to promote NKG2A+ biased immune responses may benefit patients with cancer.
NKG2A+ CD8+ T cells in tumors associate with increased immune infiltration and tumor-immune interactions
To more deeply investigate the biological underpinnings behind the seemingly beneficial NKG2A+ bias, we sought to understand how NKG2A+ CD8+ T cell tumor infiltrates differ from their NKG2C+ counterparts. To achieve this, we examined the transcriptomic profiles of nearly 400,000 CD8+ T cells from 21 cancer types (Fig. 4d) and assigned them as NKG2A+, NKG2C+, or neither, based on mRNA expression levels (see Methods). Interestingly, we observed NKG2A+ and NKG2C+ cells to occupy distinct phenotypes with NKG2A+ cells significantly biased for a TRM (T resident memory) phenotype and NKG2C+ cells biased towards TEMRA and TKLR (killer cell lectin like receptor expressing T cell) phenotypes (Fig. 4e)67,68. The preference of NKG2A+ cells toward a TRM phenotype and NKG2C+ cells towards more strongly activated phenotypes was confirmed through their differential expression of key marker genes (Fig. 4f, Table S14). TRM cells have been well characterized as potent supporters of anti- infection and cancer immune responses and their presence is associated with survival. TRM cells are also known to present with an activated phenotype that allows for effector responses while avoiding activation induced cell death (AICD)15,18. These characterizations not only match the observed benefit NKG2A+ biases provide patients with cancer, but also align with clinical and biological characterizations of NKG2A+ CD8+ T cells and biases we observed in infection and autoimmunity contexts. This suggests that the pan-disease benefit of NKG2A may arise from NKG2A+ cells biasing towards TRM or analogous phenotypes that allow for control of a given target without pathogenic inflammation.
To further characterize the biological nature of an NKG2A+ bias in cancer settings, we compared the TME profiles of patients with NKG2A+ and NKG2C+ biases. In agreement with our previous suggestions of increased immune activity in NKG2A+ biased tumors, patients with NKG2A+ biases presented with increased CD8+ T cell infiltration, TCR engagement, dendritic cell presence, and decreased prevalence of M2 macrophages (Fig. 4g). The first three factors not only confirm that patients with an NKG2A+ bias have increased immune infiltration in their tumors, but also suggests that patients with an NKG2A+ bias are equipped with the proper immune machinery of dendritic cells to prime and present CD8+ T cells with tumor antigens to permit cancer cell recognition and killing. M2 macrophages are well known to be pro-tumorigenic, have been frequently associated with metastasis and are claimed as direct players in fostering cold immunosuppressive TMEs69–73. Thus, their decreased prevalence in patients with NKG2A+ biases demonstrates that NKG2A+ biases not only associate with increased immune infiltration of key anti-tumor players but also positively correlates with immune cell behaviors that facilitate, not hinder, anti-tumor responses.
Inspired by these suggestions of NKG2A+ biases fostering pro-survival anti-cancer TMEs, we gathered spatial transcriptomics samples from five separate patients that comprised three different cancer types: breast, prostate, and ovarian (Fig. 4a)64. In particular, one of the breast cancer samples was already well characterized by a pathologist and thus we focused our analyses on that sample. Reminiscent of their divergent phenotypes, NKG2A+ and NKG2C+ spots occupied distinct spatial regions of the tumor (Fig. 4h). While NKG2C+ spots did differ from NKG2A+ spots by sitting in an interferon expressing zone of the tumor, both subsets were either in or directly adjacent to tumor tissue, perhaps suggesting active tumor engagement and possibly killing. Further, consistent with the observed association between NKG2A+ biases and immune infiltration, differentially upregulated genes in NKG2A+ spots were significantly enriched for antibody related genes (Fig. 4i). This phenomenon matches our observations across viral infection contexts, and thus suggests that this association may hold true across very diverse disease settings. To confirm the active immune cell and tumor engagement suggested to occur in NKG2A+ spots, we performed cell-cell interaction analysis by looking for ligand-receptor pairs differentially enriched in NKG2A+ CD8+ T cell spots compared to NKG2A− CD8+ T cell spots across all five spatial transcriptomics datasets (Fig. 4j). Interestingly, not only did we observe enrichment for HLA-E:NKG2A, which suggests that there may be active engagement of NKG2A in tumors, but we also observed increased chemokine, cytokine, and exhaustion receptor engagement. This suggests active recruitment of immune cells, consistent with increased immune infiltration in bulk RNA-seq samples, and suggests the possibility of tumor-immune cell engagement, for example through the well-known PD-1:PD-L1 axis74–76. Thus, we demonstrate across dozens of cancer types, thousands of patients, and hundreds of thousands of tumor-infiltrating CD8+ T cells, that NKG2A+ biases associate with increased survival and immune infiltration of tumors, as confirmed through in situ measurements, possibly due to the acquirement of a TRM phenotype by NKG2A+ CD8+ T cells that allows for sustained anti-tumor effector T cell responses.