Relative expansion of CD8 T cells in CVID comprises mainly increased percentages of EM2 and EM3 subpopulations
30 adult patients with CVIDc (17 female, 13 male), 10 adult patients with CVIDio (3 female, 7 male) and 17 healthy adults with no history of immune disease were included in this study. Detailed description of the cohort, including genetic findings, Freiburg and EUROclass classifications, can be found in Table 1. The most common clinical complications were recurrent non-infectious diarrhea (n = 21), interstitial lung disease (ILD, n = 17) and lymphadenopathy (LAD, n = 16), followed by autoimmune cytopenia (AIC, n = 14). Splenomegaly was present in 31 patients, but when found in isolation with no other clinical complications did not qualify the patient as CVIDc. A graph showing co-incidence of various clinical features in the studied cohort can be found in Figure 1A, and the proportion of patients in the different EUROclass phenotypes can be found in Figure 1B.
We next performed a detailed analysis of the differentiation and exhaustion of CD8 T cells in CVID patients. Patients with CVID had higher proportion of CD8 T cells than healthy controls (HD) in the peripheral blood (Figure 1C) (Welch’s ANOVA p = 0.008, unpaired t-tests with Welch’s correction) (CD4:8 ratio in Figure S2), with a shift from naïve (CD45RA+CCR7+) towards effector memory (EM, CD45RA-CCR7-) and terminal effector memory cells (TEMRA, CD45RA+CCR7-) stages (Figure 1D). In particular, EM2 (CD27+28-) and EM3 (CD27-28-) cells were significantly expanded in both CVIDio and CVIDc compared to HD (Figure 1E) (CVIDio EM2 unpaired t-test with Welch’s correction p < 0.001, EM3 p = 0.007, CVIDc EM2 p = 0.002, EM3 p = 0.01).
For detailed exhaustion profiling, where phenotyping of T cells based on differentiation markers is insufficient to precisely identify exhausted T-cell populations, we combined a highly multiplexed phenotyping of exhaustion-biased immune markers with a functional analysis of the expression patterns of (n = 10) cytokines important for exhausted T-cell biology. This advanced approach allows for the identification of relevant exhausted T-cell phenotypes and evaluation of an exhaustion-typical dysfunction pattern (i.e. reduced IL-2 and TNF-α production, but possible production of IFN-γ and XCL-1 chemokine) that can be assessed on a single-cell level using a functional exhaustion score, as previously published [28]. For this reason, we stimulated PBMCs from 4 healthy donors, 4 CVIDio (patient ID 31, 33, 38, 40) and 4 CVIDc (patient ID 5, 10, 17, 23) patients with PMA + ionomycin and analyzed them by using a mass cytometry panel of 44 extracellular markers, transcription factors, cytokines and chemokines (for a complete list of markers see Table S2).
To assess if the analysis of the pre-defined canonical T cell subsets correlated with data-driven approaches, we then performed trajectory-inference analysis using Wanderlust [30] based on 29 phenotypic markers (for a list see Methods). The resulting pseudotime variable corresponded well with these canonical differentiation stages from naïve into central, effector memory and ultimately TEMRA populations (Figure 1F, G), documenting expected changes with gradual decrease and final re-expression of CD45RA, loss of CCR7, preserved in CM, and loss of CD27 and CD28.
When stimulated with PMA and ionomycin, the EM2 and EM3 CD8 T cells had a unique phenotype which on average had decreased production of IL-2, but high production of cytotoxic molecules perforin, granulysin, and the lymphotactin XCL-1 (Figure 1H, dotplots in Figure S3). Especially EM3 cells also displayed a high expression of the senescence marker CD57, increased in chronic immune activation [34], the exhaustion marker Tim3, the ectonucleotidase CD39 and the TNF-α receptor CD120b. Together, this data revealed significant heterogeneity of exhaustion markers and functional T cell profiles in canonical T cell differentiation subsets.
CD8 T cell landscape is grossly disturbed in CVID, with enriched activated and exhausted cellular populations and cells with immunoregulatory potential, but lacking progenitor capability
To understand the exhaustion programs of CD8 T cells in CVID in more detail, we analyzed the high-dimensional dataset in a data-driven, unbiased manner. Dimensionality reduction on 29 phenotypic markers was performed by UMAP, revealing major differences between HD, CVIDio and CVIDc cohorts (Figure 2A). Clustering of high-dimensional data by a self-Organizing Maps algorithm (FlowSOM) revealed 20 distinct cell clusters (optimal number of clusters selected using the elbow method) (Figure 2B), the proportion of which was significantly different between the three cohorts (Figure 2C, Figure S4). Cluster 8 was significantly expanded in CVIDc compared to HD (unpaired t-test with Welch’s correction p = 0.0002), cluster 16 was significantly expanded in CVIDio compared to HD (p = 0.0002, CVIDc × HD p = 0.1), whereas cluster 18 was significantly diminished in both CVIDio and CVIDc compared to HD (CVIDio × HD p = 0.01, CVIDc × HD p = 0.009). Several other clusters (cluster 3, 6, 10, 12) were insignificantly expanded in both CVID cohorts. The clusters provided higher granularity compared to the canonical differentiation stages shown in Figure 1 and showed different localization across the canonical developmental pseudotime (Figure 2E).
We then compared the expression of all phenotypic and functional markers across all 20 clusters (Figure 2D). Further, we calculated a functional exhaustion score (FES) as described previously to identify clusters with a polyfunctionality pattern typical for exhausted T cells [28], based on TNF-α, IFN-γ, IL-2 and XCL-1 production. FES was mediocre or high in clusters expanded in CVIDc, with a similar trend in CVIDio, and low in clusters more prevalent in healthy donors, and it was significantly different between clusters (Welch’s ANOVA p < 0.0001, Figure S5). These findings link clusters with an exhaustion phenotype and FES to CVID, in particular in CVIDc patients.
Clusters 3 and 16, which were upregulated in CVID, displayed strong expression of the activation markers CD38, HLA-DR, the proliferation-associated nuclear protein Ki67, and produced IL-10, IL-17A, to a lesser extent also the effector cytokine IFN-γ, but little or none TNF-α as well as no perforin and granulysin. Their expression patterns then diverged, with cluster 3 having low expression of most exhaustion-associated markers, whereas cluster 16 displayed an elevated proportion of cells staining positive for CXCR5, higher expression levels of transcription factors TOX, FOXO1, Helios and Eomes, and inhibitory molecules such as PD-1, TIGIT, LAG3, Tim3 and CTLA-4, and had a particularly high FES. The phenotype of cells in cluster 16 is reminiscent of the elusive regulatory CD8 T cell population, based on their production of IL-10, PD-1 expression and higher average expression of the transcription factor FoxP3 in comparison to other CD8 T cells (Figure S6) [35,36].
Cluster 6 bore the hallmarks of senescent cells expressing CD57 and effector capabilities.
Cluster 8, also expanded in CVIDc, displayed an above average expression of exhaustion-associated phenotypic markers Eomes, PD-1, TIGIT, LAG-3, 2B4 and CD160, with low Tbet and TCF1, suggesting terminal differentiation. These cells retained their capacity to produce TNF-α, IFN-γ and XCL-1, but did not produce large amounts of cytotoxic molecules such as granulysin or perforin and unlike cluster 16 were not activated or proliferating. These cells together with cluster 12, which had very high expression of PD-1, fulfil most canonical signs of being exhausted and have a correspondingly high FES.
Cluster 18, severely diminished in CVID, had the hallmarks of naïve CD8 T cells, including expression of CD45RA, CCR7, CD27, CD28, CD73, CD127 and CD7.
Cluster 20, a small population but seemingly lower in CVID compared to HD, clustered together with the naïve CD8 T cells of cluster 18, but was notable for its high expression of TCF1 and FOXO1, transcription factors associated with progenitor capabilities of effector cells, in conjunction with CCR7. Similarly, cluster 19 also expressed TCF1, but also high levels of CXCR5 intermediate levels of PD-1 and no TOX, which bears similarity to progenitor population of exhausted T cells [37,38]. Cluster 19 was comparably prevalent in both CVID patients and HD.
Thus, the CyTOF analysis revealed the reduction of naïve CD8 T cells and CCR7+TCF1+FOXO1+ stem-cell-like progenitors, but not exhaustion-specific CXCR5+PD-1intTCF1+ progenitors. In contrary, two distinct groups of cells were more prevalent in CVID, a group of highly differentiated classical exhausted cells with high FES lacking signs of ongoing activation and production of cytokines, and a group of highly activated and proliferating cells with immunoregulatory features such as the production of IL-10 and expression of CXCR5 and higher average expression of FoxP3.
Bulk CVID CD8 T cells show features of exhaustion, activation and cytotoxicity
Following the markers of interest discovered by unbiased clustering approach shown above in a smaller cohort assessed by mass cytometry, we verified the expression pattern of exhaustion-associated markers PD-1, TIGIT, CD127, 2B4, Tbet, Eomes and TCF1, activation markers CD38 and HLA-DR, nuclear protein Ki67 and the cytotoxicity markers perforin and granzyme B across several flow cytometric panels in the larger cohort of 40 patients and 17 healthy donors (Figure 3A and S7).
On average, bulk CVIDc derived CD8 T cells were activated (Figure 3B), proliferating (Figure 3C) and functionally active by producing cytotoxic molecules (Figure 3D), corresponding to the expanded clusters 3, 8, 12 and 16. These features were generally more pronounced in CVIDc derived CD8 T cells compared to CVIDio, although the difference only reached significance for the percentage of activated HLA-DR+ cells. CVID CD8 T cells on average had higher expression of the exhaustion-associated marker TIGIT and lost the expression of the IL-7 receptor CD127, while the proportion of PD-1+ cells was comparable between HD, CVIDio and CVIDc (Figure 3E). These changes did not merely reflect changes in the distribution of the different memory stages shown in Figure 1, but were also present within the distinct naïve and EM subpopulations (Figure S8).
To assess cells corresponding to the dysregulated clusters seen in CyTOF analysis using the larger cohort of 40 patients and 17 healthy donors stained with flow cytometry, we gated on CD127-PD-1+TIGIT+CD28- cells (phenotype equivalent of clusters 8 and 12, hereafter referred to also as Cluster 8/12-like cells) and CD127-PD-1highTIGIT+CD28+ cells (phenotype equivalent of cluster 16, hereafter referred to also as Cluster 16-like cells). Cluster 8/12-like cells were significantly more prevalent in CVIDc compared to HD, with a similar but insignificant trend for cluster 16-like cells (Figure 3F). The prevalence of cluster 8/12-like and cluster 16-like cells determined by flow cytometry correlated to CyTOF-determined clusters 8+12 and cluster 16 cells in those patients assessed by both methods (n = 12) (Figure 3G) (Pearson correlation p = 0.04 and 0.0024, R2 = 0.354 and 0.618, respectively).
Retrospectively re-assessing the CyTOF data of 4 HD, 4 CVIDio and 4 CVIDc patients, we verified that the expression pattern of CD127-PD-1+TIGIT+CD28- cells mirrored that of clusters 8 and 12, the expression pattern of CD127-PD-1highTIGIT+CD28+ mirrored that of cluster 16 (Figure 3H), and when overlaid on the UMAP visualization these populations overlapped with their respective clusters (Figure 3I). Functionally, CD127-PD-1highTIGIT+CD28+ expressed high levels of Ki67, IL-10 and IL-2, but unlike their CD127-PD-1+TIGIT+CD28- counterparts lacked the cytotoxic effectors perforin and granulysin.
The expansion of TIGIT+ CD8 T cells was strongly associated with the EURO classification based on B cell phenotyping (see Methods and [3]) (one-way ANOVA p = 0.002) (Figure 3J), as was the expression of PD-1 (one-way ANOVA p = 0.035) and crucially so were the exhausted cluster 8/12-like cells (CD127-PD-1intTIGIT+CD28-) (one-way ANOVA p = 0.0429). The association between the EURO classification and immunoregulatory cluster 16-like cells (CD127-PD-1highTIGIT+CD28+) did not reach significance (one-way ANOVA p = 0.58) (Figure 3K).
These results show activation and elevated expression of exhaustion-associated molecules on bulk CD8 T cells in CVID patients. Further, they demonstrate a surrogate gating strategy to assess exhausted cells corresponding to clusters 8 and 12 and immunoregulatory cells corresponding to cluster 16 using low-parametric flow cytometry and show that these exhausted cells and markers are associated with the EURO classification system of CVID.
Clinical complications of CVID such as interstitial lung disease and autoimmune cytopenia strongly associate with more pronounced features of exhaustion, whereas diarrhea associates with T cell activation
As the clinical phenotype of CVIDc is varied and patients can exhibit many different complications, we tested the association between CD8 T cell phenotype and individual clinical features seen in patients with complex disease. Figure 4A shows a table with unpaired student’s t-test with Welch’s correction non-adjusted p-values calculated between all CVID patients with and without each symptom for select easy-to-measure parameters, exhausted cluster 8/12-like and immunoregulatory cluster 16-like cells. Whereas the expression of inhibitory receptors, loss of CD127 and the exhausted cluster 8/12-like cells were significantly associated with AIC, lymphadenopathy and ILD, recurrent non-infectious diarrhea was most strongly associated with CD8 T-cell activation and the activated immunoregulatory cluster 16-like cells.
ILD was strongly associated with the proportion of cluster 8/12-like cells, loss of CD127 and expression of TIGIT (Figure 4B). Cluster 8/12-like (CD127-PD-1+TIGIT+CD28-) cells offered best combination of sensitivity and specificity when searching for ILD in CVID patients (p = 0.0057, area under the receiver operating characteristic (ROC) curve (AUC) = 0.791, >12% cluster 8/12-like cells had 77% sensitivity and 74% specificity for ILD), however TIGIT+ cells also performed very well (p = 0.0123, AUC = 0.734, >66% TIGIT+ cells had 64% sensitivity and 74% specificity for ILD) (Figure 4F).
AIC was also associated with loss of CD127 and expression of TIGIT (Figure 4C). While TIGIT+ cells were significant predictors of AIC (p = 0.014, AUC = 0.736, >68% TIGIT+ cells had 64% sensitivity and 81% specificity for AIC), TIGIT/CD127 ratio was even more helpful (p = 0.009, AUC = 0.751, >1.5 ratio of TIGIT+/CD127+ cells had 64% sensitivity and 81% specificity for AIC).
Diarrhea in CVID was not associated with an expansion of the regulatory cluster 16-like cells as it was seen in patients without diarrhea when compared to healthy donors (Figure 4D). Instead, the expression of activation markers CD38 (p = 0.002, AUC = 0.784, >41% CD38+ cells had 71% sensitivity and 84% specificity for diarrhea) and HLA-DR (p = 0.006, AUC = 0.753, >6% HLA-DR+ cells had 71% sensitivity and 63% specificity for diarrhea) were both good predictors of this complication.
A similar distribution of cluster 16-like cells and additionally bulk CD8 T cells was seen for the manifestation of hepatopathy (Figure 1E). The best predictor for hepatopathy in CVID patients was the low proportion of EM1 cells (p = 0.028, AUC = 0.763, <9.4% EM1 cells had 88% sensitivity and 67% specificity for hepatopathy).
These data show a strong association between exhausted phenotype of CD8 T cells, ILD and AIC, and strong association between CD8 T cell activation and non-infectious diarrhea. While the prevalence of CD127-PD-1+TIGIT+ CD28- exhausted cluster 8/12-like cells was a strong predictor of ILD, the proportion of TIGIT+ cells also performed well in distinguishing such patients.