Impact on population dynamics of periperhal blood immune cells by CD24Fc
We utilized a high dimensional spectral flow cytometry panel with an extensive array of immune population markers (Additional file 1: Table S2) to analyze the systemic effects of SARS-CoV-2 and CD24Fc treatment on PBMCs. Using an unbiased clustering approach based on a multivariate t-mixture model [23], we identified 12 statistically distinct clusters that we visualized in two dimensions using the UMAP algorithm (Fig. 1A). Using clustered heatmap analysis, we correlated expression intensity with clusters to annotate B cells (clusters 1, 6, 8), CD8+ T cells (clusters 7, 11, 12), CD4+ T cells (clusters 2, 3), γδ T cells (cluster 4), natural killer (NK) cells (cluster 10), and myeloid cells (clusters 5, 9) (Fig. 1B). Comparing systemic immune population dynamics (Fig. 1C, D), we found significant increases in plasma B cells (cluster 6; p < 0.01), NK cells (cluster 10; p < 0.001), and terminally differentiated CD8+ T cells (cluster 12; p < 0.05) in baseline (D1) COVID-19 patients vs. healthy donors (HD). Conversely, we found that HD samples were enriched for naïve CD8+ T cells (cluster 11; p < 0.001) and a subset of myeloid cells (cluster 5; p < 0.05). These initial findings are consistent with the established immunopathology of SARS-CoV-2 infection and the critical role of the adaptive immune system in viral pathogen response [35,36,37,38], thus validating our experimental approach.
We next used UMAP contour plots to investigate the effects of CD24Fc treatment on immune population dynamics over time (Fig. 1E, F). From baseline to D8, the CD24Fc group displayed a sharp and steady decline of plasma B cells (cluster 6), which coordinated with a proportional increase in mature B cells (cluster 8). The placebo group showed relatively stable cell proportions for these populations over the same time frame. There were no significant differences between the two groups in mounting an effective anti-Spike protein antibody response (Additional file 1: Fig. S1).
CD24Fc treatment correlates with normalization of CD4+ and CD8+ T cells based on the changes of activation markers
We developed a 25-marker flow cytometry panel to examine the intricacies associated with effector cell (NK and CD4+/CD8+ T cell) activation and differentiation in response to SARS-CoV-2 infection and CD24Fc treatment (Additional file 1: Table S2). Using our unbiased clustering approach, we identified eight distinct clusters within CD8+ T cells from COVID-19 and HD samples (Fig. 2A–C). Clusters 1, 3, and 5 showed naïve and memory like signature with TCF-1 and CD62L expression, and cluster 6 showed increased expression of CD45RO. Cluster 4 showed intermediate T-bet and TOX expression indicating transitory state and cluster 8 expressed multiple activation markers including GZMB, suggestive of hyperactivation in this subset. At baseline, COVID-19 samples showed enriched frequency of clusters 4, 5, 7, and 8, which express markers of activation; HD samples were skewed toward cluster 1, which exhibits a naive phenotype (Fig. 2D–E; p < 0.001 for all clusters). To analyze the impact of CD24Fc on CD8+ T cell activation, we generated UMAP contour plots for each treatment group (Fig. 2F) and analyzed changes to cluster proportions over time (Fig. 2G). CD24Fc treatment correlated with a modest increase in frequency of the phenotypically-naive cluster 1 over time, whereas placebo-treated patients showed marked decline. Conversely, the proportion of cluster 8 cells (a population whose expression pattern is suggestive of highly activated CD8+ T cells) were stagnant in CD24Fc-treated patients, compared to the marked increase seen in the placebo group (Fig. 2G).
While tracking cluster proportions over time provides an unbiased global view of the data, these statistically distinct cell clusters may not always correspond perfectly to biologically distinct cell types. Therefore, we augmented the unbiased clustering analysis with a semi-supervised approach to define a CD8+ T cell activation score. Known markers of CD8+ T cell activation (T-bet, Ki-67, CD69, TOX, and GZMB) were significantly increased in baseline COVID-19 patients compared to HD (Fig. 2H), supporting our hypothesis that SARS-CoV-2 infection increases peripheral T cell activation. To create a unified cell-level activation score, we used PCA to implement dimension reduction of the cell-by-activation marker expression data for all baseline COVID-19 and HD cells. The first principal component (PC1) loadings of each activation marker were used as coefficients in a linear model for defining the activation score (Additional file 1: Table S5). Thus, while we manually selected key T cell activation markers, we determined the relative contribution of each activation marker to the final activation score in a data-adaptive manner, yielding a semi-supervised approach. We observed positive PC1 loadings and positive average log-fold changes for each activation marker, confirming that higher activation scores reflect higher T cell activation (Additional file 1: Table S5). Distributions of activation scores across cell clusters also confirmed that more highly activated cell subsets feature higher activation scores (Fig. 2I). To characterize the effect of CD24Fc treatment on global CD8+ T cell activation, we adopted a GLMM of activation scores over time. While CD8+ T cell activation scores at baseline were not statistically different between groups, the predicted mean activation scores indicate significantly different trajectories between placebo and CD24Fc groups over time (Fig. 2J; p < 0.001). Thus, we conclude that CD24Fc treatment significantly reduced hyperactivation of CD8+ T cells compared to placebo.
CD4+ T cell activation also plays an important role in the immune response to SARS-CoV-2 infection, so we applied the analysis strategy presented above to this population [35]. To comprehensively understand the role of CD4+ T cells and Foxp3+ regulatory T cells (Treg), we analyzed total CD4+ T cells, including Foxp3+ subset (Fig. 3), and then the Foxp3+ Tregs exclusively (Fig. 4). We added Foxp3 to the existing 24-marker flow cytometry panel for the identification of Foxp3+ Tregs. Using our unbiased clustering approach, we identified 10 clusters of statistically distinct CD4+ T cell sub-populations that we projected onto UMAP space to observe global clustering patterns (Fig. 3A, D). To characterize cell clusters in terms of differential marker expression, we computed median expression levels of the 18 markers in the CD4+ flow cytometry panel and plotted cell-level marker expression for each marker on the UMAP space (Fig. 3B, C). Clusters 1 and 2 showed lowest CD44 expression indicating naïve-like phenotype, and cluster 3 showed GZMB expression. Cluster 4 showed highest expression level of CD25 and FOXP3 suggestive of Treg. Clusters 5, 6, 7, and 10 expressed intermediate to high levels of TCF1, and Cluster 9 expressed multiple activation markers. CD4+ T cells revealed dramatic changes in the relative representation of each cluster upon SARS-CoV-2 infection. Similar to the CD8+ T cell activation pattern we observed, CD4+ T cells from COVID-19 patients showed a significant reduction in clusters with lower activation marker expression levels, including clusters 1 (p < 0.001), 2 (p < 0.001), and 8 (p = 0.002), and a significant increase in clusters with higher activation marker expression levels, including clusters 4, 5, 6, 9, and 10 (all p < 0.001). These results suggest that clusters 1 and 2 are largely composed of less activated CD4+ T cells, while other clusters are composed of relatively more activated phenotypes (Fig. 3D, E).
Next, we assessed the short-term longitudinal effect of CD24Fc treatment on CD4+ T cells in COVID-19 patients. Using UMAP contour plots to visualize temporal and treatment-level changes in CD4+ T cell dynamics (Fig. 3F), we quantified fold-changes in populations over time (Fig. 3G). In contrast to our CD8+ T cell results, wherein the phenotypically-naive cluster 1 sustained its level during CD24Fc treatment (Fig. 2), clusters 1 and 2 from the CD4+ T cell population decreased upon CD24Fc treatment, which we believe reflects reduced activation. Clusters 4, 5, and 10 were increased by CD24Fc treatment. Cluster 4 showed high expression level of CD25 and FoxP3 (likely Tregs), while clusters 5 and 10 showed intermediate-to-high levels of CD62L and TCF-1 expression. Cluster 9, which expressed multiple activation markers and was presumably composed of hyper-activated cells, was decreased by CD24Fc treatment, similar to CD8+ T cell results.
Using the univariate cell-level activation workflow described above, we determined CD4+ T cell activation scores. Known markers of CD4+ T cell activation (T-bet, Ki-67, CD69, TOX, and PD1) were significantly increased in baseline COVID-19 patients compared to HD (Fig. 3H). Distributions of activation scores across cell clusters also confirmed that more highly activated cell subsets feature higher activation scores (Fig. 3I). Predicted mean activation scores indicate significantly different trajectories between the placebo and CD24Fc groups over time; CD24Fc-treated samples had significantly lower CD4+ cell activation levels relative to placebo (overall p < 0.001; Fig. 3J). Baseline values for CD4+ T cell activation were not statistically different between groups. In contrast, mean activation scores were significantly different between placebo and CD24Fc-treated at all other time points (D2, p = 0.001; D4, p < 0.001; D8, p < 0.001), with the most marked difference on day 8. Thus, we conclude that the attenuation of lymphocyte hyperactivation extends to the CD4+ T cell compartment.
We performed the same analyses on Foxp3+ Tregs exclusively (Fig. 4) and found that COVID-19 was associated with hyperactivation in this population as well. Cluster 1 showed lowest expression of CD44 suggesting less activated phenotype. Clusters 3 and 4 showed highest expression levels of PD1 and CD69, respectively, and cluster 2 exhibited highest TCF1 expression level. Clusters 6, 7, and 8 expressed multiple activation markers including CTLA4. Upon SARS-CoV-2 infection, clusters 1 and 3, which represent less activated phenotype, were downregulated and clusters 6, 7, and 8 reflecting more activated phenotype increased when COVID-19 patient samples were compared with HD samples. CD24Fc treatment was associated with a substantial reduction by day 8 (Fig. 4G) in the proportion of cells belonging to the most hyperactivated cell cluster (Treg cluster 8; Fig. 4I). Using the GLMM activation score model, we found a significant reduction in Foxp3+ Treg activation associated with CD24Fc treatment by day 8 (p < 0.001), while we failed to detect a significant difference in predicted activation scores between treatment groups at earlier time points (Fig. 4J).
CD24Fc reduces NK cell dysregulation
The increased number of NK cells in samples from patients with COVID-19 (Fig. 1C, D, cluster 10) implies they play an important role in SARS-CoV-2 infection. We investigated the activation and functional status of NK cells using our unbiased clustering and visualization approach and identified 12 statistically distinct NK cell clusters, which we visualized on heatmaps and UMAPs (Fig. 5A–C). Clusters 7 and 10 showed CD3 expression indicating NKT cell property, and cluster 11 expressed multiple activation markers suggesting hyperactivated phenotype. Clusters 1 and 5 showed minimal expression of activation markers indicative of resting NK cells, and clusters 3 and 12 showed CD11b expression. Cluster 5, the most highly represented cluster in HD samples, displayed an expression pattern suggestive of a less activated population; it was significantly downregulated in COVID-19 patients (p < 0.001; Fig. 5D–E). Samples from COVID-19 patients also revealed a significant reduction in cluster 2 (p = 0.003) and expansion of clusters 1, 4, 6, 8, 9, 11, and 12 (Fig. 5D–E; p = 0.04 for cluster 1; p = 0.002 for cluster 9; p = 0.03 for cluster 12; p < 0.001 for clusters 4, 6, 8, 11).
To understand the role of CD24Fc treatment on NK cell population dynamics, we generated UMAP contour plots to visualize temporal and treatment-based changes (Fig. 5F), and quantified these differences (Fig. 5G). Clusters 1 and 2, which showed a more naive phenotype, were increased by CD24Fc, whereas cluster 11, which expresses multiple activation markers, was decreased. To visualize activation, known NK cell activation markers (TOX, GZMB, KLRG1, Ki-67, and LAG3) were assessed (Fig. 5H) and plotted per cluster (Fig. 5I). Using a GLMM of activation scores over time, we found that while baseline values for NK cell activation were not statistically different, the mean activation scores were significantly different between placebo and CD24Fc groups throughout the study duration (p < 0.001; Fig. 5J). Thus, CD24Fc treatment rapidly normalized NK cell activation status, and the impact was sustained throughout the study period.
CD24Fc attenuates systemic cytokine response
The profound changes in lymphocyte dynamics after CD24Fc treatment indicate that CD24Fc exerts its effect by regulating the systemic cytokine levels. To test this hypothesis, we compared plasma cytokine concentrations from HD and COVID-19 patients treated with CD24Fc or placebo. We used multiplex ELISA and Luminex analysis platforms to test 37 cytokines in total. Fifteen out of 37 tested cytokines were significantly elevated (p < 0.05) during SARS-CoV-2 infection (p < 0.05, Fig. 6A, Additional file 1: Fig. S2A-B). These included cytokines associated with type 1 (IL-12p40, CXCL9, IL-15) and type 3 (IL-1α, IL-1β, RANTES) immunity, and chemokine MCP-1 (CCL2) that recruits monocytes and T cells to the sites of inflammation. Only three of 37 cytokines, including TNRFII, Eotaxin-2 and IL-8 were significantly downregulated in COVID-19 patients (p < 0.05; Fig. 6A).
We next studied the impact of CD24Fc on cytokine expression in patients with COVID-19. As shown in Fig. 6B, several cytokines (GM-CSF, IL-5, IL-7, IL-10) and chemokines (MIG, MIP-1α, MIP-1β) were down-modulated over time. Serum levels dynamics of selected individual cytokines are shown in Fig. 6C, D and Fig. S2C. At one week after treatment initiation, many cytokines and chemokines were reduced by tenfold or more. Notably, many of these inflammatory proteins were selectively reduced in the CD24Fc-treated patients or downregulated more rapidly compared to placebo-treated patients. Specifically, CD24Fc significantly down-modulated plasma levels of IL-10 and IL-15 (p = 0.05 and p = 0.002, respectively; Fig. 6C, D). Other tested cytokines implicated in COVID-19 pathogenesis, including IL-6 and GM-CSF [39], exhibit a similar trend toward a selective downregulation by CD24Fc, albeit these did not reach the levels of statistical significance (Fig. 6B, Additional file 1: Fig S2C). To increase the statistical power of the analysis of the influence of CD24Fc on systemic cytokine response, we calculated a cytokine scores for each treatment group by integrating expression of all markers tested by multiplex ELISA platform using weighted sum approach. Analysis of cytokine scores demonstrated a significant decrease in CD24Fc-treated groups compared to placebo (p < 0.001; Fig. 6E). This finding was independently confirmed using Autoencoder [28] and PCA (Additional file 1: Fig. S2D).
To better understand the global modulation of systemic cytokine response by CD24Fc treatment, we studied correlations between individual cytokines across groups. Correlation matrices wherein darker red lines indicate stronger correlation (Fig. 6F) showed that only a few groups of cytokines were co-expressed by HD. However, the numbers of co-regulated cytokines dramatically increased in baseline COVID-19 samples (vs. HD controls) indicating activation of coordinated cytokine response. Remarkably, samples from CD24Fc-treated patients (pooled over time) showed a decline in cytokine correlations compared to baseline or placebo treatment. Similarly, cytokine network plots connecting cytokines with moderate and strong associations (Pearson correlation r > 0.4 [29]) showed lower overall interconnectedness in CD24Fc group as compared to baseline or placebo treatment (right two panels, Fig. 6G). The overall cytokine network correlations and connectivity in CD24Fc-treated patients were significantly different from baseline or placebo treatment (Fig. 6H, I; p < 0.001 for both).
To understand the relevance of decreased correlation and connectivity of the cytokine network in CD24Fc-treated patients to disease severity and therapeutic effect, we analyzed a previously published dataset of cytokine expression in serum from patients with COVID-19 that were either treated in the intensive care unit (ICU patients) or did not require ICU treatment (non-ICU patients) [40]. Notably, we found that inter-cytokine correlation and connectivity were lower in non-ICU patients than ICU patients (Fig. 7). These data suggest that the increased blood cytokine network correlation and connectivity analysis we developed are associated with increased COVID-19 disease severity, while mild disease (without the need for ICU treatment) is characterized by lower correlation and connectivity. Therefore, decreased correlation and connectivity of the cytokine network is likely a novel useful tool to examine the therapeutic efficacy of anti-inflammatory agents.
To identify factors that may play an important role in response to CD24Fc, we calculated centrality scores [32] for individual cytokines based on their connectivity and correlations within the global cytokine network (Additional file 1: Table S6). The variances of the centrality scores of 30 cytokines were lower in baseline and placebo-treated COVID-19 patients compared to HD and CD24Fc-treated COVID-19 patients (Fig. 6J). These data indicate that distinct cytokines are highly heterogeneous in terms of their interconnectedness with other cytokines (centrality) in healthy individuals. Upon SARS-CoV-2 infection, cytokine centralities become more uniform, and subsequent CD24Fc treatment abrogates this effect (Fig. 6J).