Single-cell transcriptomics reveal altered B cell responses and T cell senescence in geriatric non-responders to COVID-19 mRNA vaccination

Summary

The immune response to COVID-19 vaccines is diminished in older individuals. To understand the underlying immunobiology, we analyzed single-cell RNA-seq data from PBMCs of SARS-CoV-2 naive nursing home residents with varying humoral responses following BNT162b2 vaccination and validated via flow cytometry. Responders (R) (>4500 AU/mL anti-spike titers) showed enrichment for naive B cell (IGHD, BACH2, CD22) and naive CD4 T cell and early T follicular helper (Tfh)-related genes (CCR7, TCF7, LEF1, IL6ST, and TGFBR2). Non-responders (NR) (<20 AU/mL) displayed elevated markers of T cell senescence (KLRG1, CCL4, CCL5, and IL32), immune exhaustion (PD-1), and inflammation (TNF-α, IFN-γ). Flow cytometry revealed reduced CD4 T and B cell frequencies but higher CD8 T and NK cells in NR. Despite reduced B cell frequency, NR upregulated plasma B cell genes (PRDM1, XPB1, IRF4), suggesting dysregulated B cell differentiation. Our findings point to impaired lymphocyte responses and increased immunosenescence in NR, emphasizing the need for enhanced vaccine strategies in aging populations.

Graphical abstract

Highlights

• •scRNA-seq identifies immune signatures of COVID-19 mRNA vaccine response in older individuals
• •Vaccine responders show B cell, T cell, and early Tfh signatures of effective humoral immunity
• •Non-responders show altered T cell senescence, exhaustion, and inflammation gene expression
• •Non-responders show plasma cell gene enrichment, suggesting impaired antibody and T cell help

Molecular physiology; Immunology; Transcriptomics

Introduction

Messenger RNA (mRNA) vaccines have profoundly altered the trajectory of the morbidity and mortality of the COVID-19 pandemic, significantly reducing its impact, while paving the way for numerous new vaccines and therapeutics.¹ However, not all individuals mount an adequate immune response to these vaccines, with vulnerable populations, including older persons, patients with cancer, transplant recipients, patients on dialysis or immunosuppressive regimens, as well as those with chronic diseases, exhibiting reduced immune responses.^2,3,4,5 A more complete understanding of the failure to develop a protective immune response to mRNA vaccines in vulnerable populations is thus warranted.

Functional genomic technologies, including single-cell RNA sequencing (scRNA-seq), flow cytometry, and receptor profiling, have provided valuable insights into the adaptive immune response to SARS-CoV-2 vaccination.^6,7,8,9 These studies have identified specific transcriptional changes in key immune cell populations, such as CD4 T cells and B cells, following vaccination. However, these vaccine-induced responses, particularly in B and CD4 T cell compartments, are often attenuated in older individuals, highlighting the need for population-specific analyses to understand immune variability.

Geriatric individuals exhibit diminished responses to mRNA vaccines, including BNT162b2.^3,4,10,11,12 Previously, we reported that SARS-CoV-2-naïve nursing home residents (NHRs) showed weaker responses than younger healthcare workers, with some older participants displaying higher antibody titers than others.³ Consistent with our findings, several studies have shown that older individuals generally mount lower antibody titers than younger individuals.^{4,10,11,12,13,14} An effective vaccine response requires both antibody production and antigen-specific T cell activation. However, in older adults, antigen-specific clonal expansion is reduced, reflecting limited TCR repertoire diversity and contributing to impaired cellular immunity.¹⁴ In addition, older adults exhibit reduced polyfunctionality with lower IFN-γ expression and cytotoxicity compared to younger adults.¹⁵ The antibody response is directly supported by T follicular helper (Tfh) cells, which promote germinal center B cell production of spike-specific immunoglobulins.^{8,16,17,18,19,20} In older individuals, primary vaccination of the BNT162b2 vaccine induces fewer spike-specific Tfh cells and elicits a weaker B cell response than younger individuals, although booster doses enhance B cell responses without restoring Tfh activity.²¹ Early CD4 T cell responses are also attenuated in older adults, including reduced Th1 and CXCR3⁺ circulating Tfh cells following the first mRNA vaccine dose, which correlates with lower antibody titers and diminished CD8 T cell activity.²² In addition, elevated PD-1 expression on Th1 cells was observed in older adults, contributing to limited functional expansion.²² However, these and previous studies largely focused on comparing vaccine responses in older versus younger adults, which identifies broad age-related changes in immunity. They do not reveal which mechanisms within the older adult population enhance or impair the vaccine response, which are therefore most relevant for therapeutic intervention. In contrast, the heterogeneity observed within the geriatric population, where some individuals mount strong antibody responses while others do not, remains poorly understood. Our study directly addresses this gap by investigating the mechanisms that differentiate strong from weak responders (R) exclusively within the elderly, thereby identifying pathways that may be most relevant to improving vaccine protection in this vulnerable group.

A major factor contributing to poor vaccine responses in geriatric populations is immunosenescence, the gradual decline of the immune system due to aging. It is characterized by reduced immune cell function, diminished proliferative capacity, and impaired response to new antigens, partly due to thymic involution, which decreases naive T cell production while accumulating highly differentiated memory T cells.²³ These changes are also accompanied by chronic, low-grade inflammation (“inflammaging”), further compromising immune function.²⁴ Despite these associations, the specific role of immunosenescence in BNT162b2 responses remains unclear. This study provides single-cell transcriptomic and phenotypic evidence of immunosenescence and immune exhaustion in older individuals with poor vaccine responses, contributing to the limited understanding of how these mechanisms affect mRNA vaccine efficacy.

In this study, we aimed to investigate the immunobiology underlying the variable response to the BNT162b2 vaccine in older individuals. Using scRNA-seq and flow cytometry, we analyzed peripheral blood mononuclear cells (PBMCs) from 12 SARS-CoV-2-naive NHRs with high (>4500 AU/mL) or low (<20 AU/mL) antibody titers. R exhibited transcriptional signatures indicative of naive B cells and early Tfh-like CD4 T cells, while non-responders (NR) showed increased expression of genes linked to T cell senescence, exhaustion, and inflammation. Immunophenotyping further revealed skewed lymphocyte composition in NR, with reductions in CD4 T and B cells and a relative increase in CD8 T and NK cells. Despite lower B cell abundance, transcriptional upregulation of plasma cell markers suggested a short-lived, possibly dysregulated B cell response. By identifying the transcriptomic signatures associated with impaired vaccine responses, our study provides insight for developing strategies to enhance vaccine efficacy in geriatric populations. A similar targeted approach could also inform the design of more effective mRNA vaccines or mRNA vaccine therapies for other populations at risk of vaccine failure.

Results

To elucidate factors driving reduced immunogenicity among NHRs vaccinated with the BNT162b2 mRNA vaccine, we conducted a transcriptomic and serologic analysis of peripheral blood from SARS-CoV-2 naive participants at two weeks post-vaccination (Figure 1A). We measured their anti-spike antibody titers and neutralization activity and identified 6 R and 6 NR post initial vaccination series, with anti-spike antibody titers (>4500 AU) and (<20 AU), respectively (Figure 1B). The cohort was balanced for sex (50% male/female, R/NR) and ethnicity, with a comparable age range and median (R: 71.5 [65–99], NR: 86 [76–99]) (Table S1). The age difference was insignificant (Welch’s t test p = 0.10; Mann-Whitney U p = 0.065). We isolated PBMCs and performed scRNA-seq and flow cytometry. Additionally, we correlated antibody levels with B cell and CD4 T cell frequency through flow cytometry. We analyzed the single-cell data using our custom pipeline, allowing us to identify and further subset immune cell clusters for detailed analysis. Finally, we built correlation models to integrate transcriptomic and immunologic data to further validate our findings.

Figure 1.

Single-cell transcriptomics maps immune cell profiles in PBMCs of NR and R post-vaccination, uncovering senescence markers in NR T cells

(A) Schematic of single-cell workflow and study design. Blood samples were collected from 12 SARS-CoV-2 naive nursing home residents 14 days post-initial vaccination series. PBMCs (n = 6, R/NR) were sequenced using single-cell RNA-sequencing and analyzed with a custom pipeline. UMAPs shown are schematic representations and not actual data.

(B) Boxplots showing anti-spike antibody levels (AU/mL) and neutralization ability (pNT50) in NR and R donors (p < 0.001, p = 0.013, respectively; two-tailed Welch’s t test). Boxplots show the median and interquartile range (IQR); whiskers extend to 1.5× IQR. The anti-spike antibody titers and neutralization ability were measured at the same timepoint as the blood sampling for single-cell RNA-seq (14 days after the second dose).

(C) UMAP visualization of PBMC clusters (C1-C15) of all donors post vaccination.

(D) Dot plots of top DEGs in cell type clusters in all donors.

(E) UMAP visualization of PBMCs of NR and R groups.

(F) Volcano plot of the featured top DEGs in PBMCs of R and NR.

(G) Violin plots of T cell clusters show normalized expression levels of upregulated DEGs related to T cell development and function in T cells of R, and those involved in PD-1 signaling and senescence in T cells of NR. Adjusted p-values from the differential expression analysis are reported for each gene.

(H) Heatmap of senescence-associated genes differentially expressed between NR and R. Expression levels of selected upregulated markers from the SenMayo gene set are shown across individual cells, grouped by responder type (R vs. NR). Each column represents a single cell, and each row represents a gene. Color intensity reflects scaled expression levels.

(I) Heatmap of differentially expressed pathways (MSigDB — ImmuneSigDB) in PBMCs of NR vs. R. Average logarithmic fold change and normalized enrichment score (NES) are shown on the right. See also Figure S1.

Single-cell transcriptomics delineates immune cell profiles in PBMCs and reveals signatures of T cell senescence in NR

To investigate factors driving the large variation in response to BNT162b2 vaccination in older adults, we performed scRNA-seq on PBMCs from six R and six NR, selected based on their anti-spike antibody titers (Figure 1B). Cluster analysis identified 15 distinct clusters among all PBMCs (Figure 1C). Cell types were annotated based on the most differentially upregulated canonical markers in each cluster (Figures S1A–S1F). Cluster C2 was identified as containing predominantly CD4 T cells, C3 & C14 as CD8 T cells, C4 as NK cells, C6 & C12 as B cells, C1, C5, C8, C9, and C15 as monocytes, and C11 as dendritic cells (refer to methods). Further, C5 may be classified as non-classical monocytes due to the low expression of CD14 and upregulated expression of FCG3RA (CD16) (Figure 1C). Clusters C7 and C13 were identified as predominantly erythrocytes, and C10 as platelets, due to the high expression of hemoglobin genes and platelet factors, respectively (refer to Methods).

Top differentially expressed genes (DEGs; adj. p < 0.05) were analyzed in clusters of B cells and T cells. Cluster C2 (CD4 T cells) was enriched in genes linked to T cell development (LTB, IL7R, BCL11B, TCF7),^25,26,27,28 activation and differentiation (MAL, RCAN3),^29,30 and homing (CCR7)³¹ (Figure 1D). C3 (CD8 T cells) expressed chemokines (CCL5), granzymes (GZMH, GZMK, GZMA), perforin (PRF1), NKG7, and IL-32, while C14 (CD8 T cells) showed enrichment for IL-32, CCL5, histone genes (HIST1H1B, HIST1H4C), and proliferation markers such as MKI67 (Figure 1D and Table S2). B cell clusters (C6, C12) included markers for B cell signaling (CD79A, TNFRSF13C, BANK1),^32,33 antibody production (IGHA1, IGHM, IGHD, IGKC), and plasma and marginal zone B cells (JCHAIN, MZB1)^34,35 (Figures 1D and S1A).

To compare the transcriptomic profiles of PBMCs between R and NR, we categorized each sample by responder type and visualized the differences by UMAP dimensionality reduction (Figure 1E). Differential gene expression analysis highlighted distinct profiles between R and NR, with the top DEGs (adjusted p-value <0.05) discussed. Mitochondrial genes (MT-ND1, MT-ND3, MT-CYB, MT-CO2, and so forth) and ribosomal genes (RPL13, RPL36, RPL30, RPL12, and so forth) were upregulated in the R, while chemokines (CCL4, CCL5, CCL2) were upregulated in NR (Figure 1F and Table S3). We then isolated the cells identified as predominantly T cells (C2, C3, and C14) and performed differential gene expression analysis between R and NR. Responder T cells had an upregulation of genes involved in helper T cell development and function (SELL, TCF7, LEF1, IL6ST)^28,36,37,38 (Figure 1G and Table S4), while NR T cells were enriched with genes associated with senescence (KLRG1)³⁹ and PD-1 signaling (SH2D1A, MAPK1, FYN, PIK3R1, HLA-DRB1, HLA-DPA1, HLA-DRA)^40,41,42,43 (Figure 1G). Further, gene markers of senescence from the SenMayo gene set (CCL4, CCL5, IL32, IQGAP2, JUN, LCP1, SELPLG, TNFRSF1B) were upregulated in NR T cells (Figure 1H).⁴⁴ Overall, our results indicate increased inflammation in NR, as evidenced by the upregulation of chemokines and pro-inflammatory cytokine genes in PBMCs. Additionally, the enrichment of PD-1 signaling genes and pathways in T cells suggests immune exhaustion in NR.

Next, we applied Gene Set Enrichment Analysis (GSEA) using the list of DEGs in PBMCs of NR vs. R as input, and ImmuneSigDB database as Godec et al.⁴⁵ Only enriched pathways with FDR-adjusted p-value <0.001 are discussed. In NR, we observed significant enrichment of the pathways GSE11057 NAIVE VS MEMORY CD4 T cell DN (NES = 2.09), GSE11057 NAIVE VS CENT MEMORY CD4 T cell DN (NES = 2.09), and KAECH NAIVE VS MEMORY CD8 T cell DN (NES = 2.18), all of which consist of genes more highly expressed in memory T cells. These results indicate that NR samples are enriched for memory-like CD4 and CD8 T cell gene expression, whereas R samples exhibit relatively higher expression of naive T cell-associated genes (Figure 1I).⁴⁶ Additionally, NR were enriched for GSE26495 NAIVE VS PD1HIGH/LOW CD8 T cell DN (NES = 2.49/2.65), both of which consist of genes upregulated in PD-1-low or PD-1-high CD8 T cells relative to naive CD8 T cells. These results indicate that NR cells express gene signatures associated with more activated or exhausted CD8 T cell states (Figure 1I).^47,48 Lastly, we observed significant downregulation of the GSE24574_BCL6_HIGH_TFH_VS_LOW_TFH_CD4_T cell_UP gene set in NR compared to R, consistent with the reduced expression of BCL6-high Tfh cell-associated genes in NR (NES = 2.81) (Figure 1I).⁴⁹ BCL-6 upregulation in both B cells and Tfh cells is essential for B cell germinal center formation and sustained Tfh-B cell interactions.^50,51,52 The enrichment in BCL6-high Tfh-associated pathways in R, may reflect an enhanced Tfh differentiation and Tfh-B cell interactions in R relative to NR.⁴⁹

Flow cytometric-based phenotypic analysis identifies differences in subset frequencies and higher PD-1 expression in NR

To validate our gene expression results, we measured the frequency of CD4 T cells, CD8 T cells, B cells, and NK cells in R and NR by flow cytometry (Figure 2A). Frequencies of CD4 T cells and B cells in total PBMCs were significantly higher in R than NR (Figure 2A) (p < 0.001, p = 0.03, respectively, one-tailed Student’s t test with Welch’s correction). In contrast, the frequency of CD8 T cells in total PBMCs was significantly higher in NR (p < 0.04, one-tailed Student’s t test with Welch’s correction) (Figure 2A). The frequency of NK cells in total PBMCs was also significantly higher in NR (p < 0.04, Wilcoxon one-tailed rank-sum test) (Figure 2A).

Figure 2.

Flow cytometry reveals altered immune cell frequencies, associations with antibody responses, and elevated PD-1 expression in NR

(A) Distribution of frequencies of CD4 T cells (p < 0.001, one-tailed Welch’s t test), B cells (p = 0.03, one-tailed Welch’s t test), CD8 T cells (p < 0.04, one-tailed Welch’s t test), NK cells (p < 0.04, one-tailed Wilcoxon rank-sum test) in PBMCs in R and NR identified by flow cytometry. Data presented as mean ± SD.

(B) Distribution of PD-1 mean fluorescence intensity (MFI) in CD3 T cells identified by flow cytometry in R and NR (p = 0.04, two-tailed Welch’s t test). Data presented as mean ± SD.

(C) Linear regression models of PD-1 in CD3 T cells identified by flow cytometry and PDCD1 and CD274 gene expression in T cells by scRNA-seq. See also Figure S4. ∗p ≤ 0.05 by Welch’s t test. ∗∗p ≤ 0.05 by Wilcoxon rank-sum test.

We then measured PD-1 levels in T cells of R and NR using flow cytometry and correlated these measurements with scRNA-seq gene expression. PD-1 levels were significantly increased in T cells of NR (p = 0.04, two-tailed t test) (Figure 2B) by flow cytometry. The mean fluorescence intensity (MFI) values of PD-1 protein expression showed a strong correlation with PDCD1 (PD-1, Pearson’s r = 0.69, p-val = 0.01) and CD274 (PD-L1, Pearson’s r = 0.65, p-val = 0.02) gene expression in T cells (Figure 2C). To exclude the possibility that cytomegalovirus (CMV) infection contributed to the observed R versus NR differences, we assessed CMV serostatus across the cohort (Table S5). CMV seropositivity was distributed without a clear pattern between groups, indicating it is unlikely to account for the divergent immune responses observed. As CD4 T cells and B cells are central to the T-dependent vaccine response and underlie our responder classification, we focus our subsequent analyses on these cell types.

R exhibits transcriptomic signatures of naive B cells, while NR displays reduced B cells with features of an altered acute plasma cell response

To further investigate the transcriptomic profiles of B cells from the vaccinees, we combined clusters C6 & C12 for separate analysis using our custom pipeline. Louvain clustering, visualized through UMAP dimensionality reduction at a resolution of 0.2, revealed 7 subclusters with distinct activation and differentiation transcriptomic profiles (Figures 3A, and S2A–S2C). IGHM, IGHD, and CCR7, markers of naive B cells,^53,54 were upregulated in SC1-2; CD27, a memory B cell marker,⁵⁵ was upregulated in SC6; and CXCR3, implicated in the terminal differentiation of memory B cells into plasma cells,⁵⁶ was upregulated in SC5 (Figure 3B). Plasma cell markers—CD38 (SC6), JCHAIN (SC5, SC6), PRDM1(BLIMP1) (SC6), XBP1 (SC5-SC7), and IRF4 (SC5, SC6)—indicated plasma B cell enrichment in SC5-SC7 (Figure 3B).^{34,57,58,59,60} Germinal center markers CXCR4 and CD83 (SC1, SC2) and CD22 (SC1) suggest the presence of GC B cells (Figure 3B).^61,62,63 MZB1 (SC5, SC6) suggests marginal zone B cell enrichment (Figure 3B).³⁵ GSEA revealed upregulation for Hallmark pathways TNF-α via NFKB (NES = 1.32) in SC4, and apoptosis (NES = 1.2) and p53 in SC6 (NES = 0.53) (Figure S2B).

Figure 3.

Single-cell analysis of B cell subset reveals differences between R and NR, with features of a dysregulated acute plasma cell response in NR

(A) UMAP visualization of B cell subclusters (SC1-SC7) of all donors’ post vaccination. Clusters SC5-7 show the enrichment of plasma B cell markers CD38, JCHAIN, PRDM1(BLIMP1), XBP1, and IRF4 (see Results).

(B) Violin plots of canonical markers gene expression per B cells subclusters of all donors.

(C) UMAP visualization of B cell subclusters of NR and R.

(D) Volcano plot of top B cells DEGs in R and NR.

(E) Violin plot shows normalized expression levels of upregulated DEGs involved in immunoglobulin machinery in B cells of NR. Adjusted p-values from the differential expression analysis are reported for each gene.

(F) Network analysis of top DEGs in B cells of R and NR (PPI enrichment p-val <0.001). Oxidative phosphorylation pathway is enriched in R (GO:0006119, FDR-adjusted p-val <0.001), while protein processing in the endoplasmic reticulum (ER), antigen processing and presentation, and protein export pathways are enriched in NR (KEGG hsa04141, hsa04612, hsa03060, respectively, all with FDR-adjusted p-val <0.001). See also Figure S2.

To compare the transcriptomic differences of B cells in R and NR, we assigned each sample to a responder type set, visualized the differences by UMAP dimensionality reduction, and performed differential gene expression analysis (Figure 3C). SC6 had the highest proportion of NR cells relative to R cells, while R cells were mostly in SC1-2, SC4-5, and SC7 (Figures 3C and S2C–S2E). Top DEGs with adjusted p-val <0.05 are discussed. R showed higher levels of expression of IGHD, a marker of naive B cells,⁵⁴ as well as genes associated with naive B cell state (BACH2, CD22),^64,65 and key regulators of B cell development, differentiation, and function (PAX5, IL4R, BCL7A, BTG1, ETS1)^{66,67,68,69,70} (Figure 3D). On the other hand, NR exhibited higher gene expression of plasma cell differentiation markers (PRDM1 (BLIMP1), XBP1, IRF4),^34,57,60 HLA genes (HLA-DQB1, HLA-B, HLA-A), and marginal-zone B cell marker, MZB1,³⁵ and negative regulator, LY9⁷¹ (Figure 3D). Additionally, NR showed increased expression of genes encoding antibodies (JCHAIN, IGHA1, IGHA2, IGKC), particularly enriched in SC6 (Figure 3E). To identify high-confidence pathway activity, we performed STRING-based protein-protein interaction (PPI) network analysis on the full set of upregulated DEGs (positive fold change) in R and NR (Table S7). Networks with significant PPI enrichment are discussed (p < 0.001) (Figure 3F). Functional enrichment of network-associated DEGs revealed distinct pathways, including the oxidative phosphorylation pathway (GO:0006119)⁷² in R, and protein export (KEGG hsa03060)⁷³ and protein processing in the ER (KEGG hsa04141,⁷⁴ and antigen processing and presentation (KEGG hsa04612)⁷⁵ pathways in NR (all with FDR-adjusted p-val <0.001) (Figure 3F). Overall, our findings suggest that NR exhibit a reduced B cell pool that is otherwise enriched for transcriptomic markers of plasma cell differentiation, whereas R retain B cells enriched for markers of naive B cells and B cell development.

Transcriptomic profiling of CD4 T cells reveals naïve-like and early Tfh differentiation signatures in R

To characterize the T cell vaccine response, we analyzed CD4 T cells (originally C2) separately. Louvain clustering (resolution 0.4) identified 7 subclusters with distinct lineage and cytokine profiles (Figure 4A, and S3A–S3E). We identified top DEGs in CD4 T cell subclusters with adjusted p-val <0.05. Regulatory T cell (Treg) markers were enriched in SC12 (IL2RA, CTLA4, IKZF2)^76,77,78 and SC13 (RORA, NEAT1),^79,80 while naive CD4 T cell markers (LEF1, CCR7, TCF7)^37,81,82,83 were enriched in SC9-11 (Figures S3A–S3E). Early activation markers (CD69, CCL5, CCL4) were upregulated in SC14 (Figure S3A). Th1 (ID2, IL2RA, TXK),^84,85,86 Th2/Tfh (GATA3, ICOS),^87,88 and Th17 (CCR6, KLRK1)^89,90 lineage markers and naive and memory CD4 T cell markers (IL7R, PTPRC, SELL)^91,92,93 were enriched across several clusters (Figures S3C–S3E). GSEA revealed IL-2/STAT5 (NES = 1.83) and IFN-γ signaling (NES = 1.60) enrichment in SC12, while SC14 showed enrichment for TNF-α, hypoxia, mTORC1, and apoptosis pathways (NES = 1.51, 1.57, 1.38, 1.34, respectively) (Figure S3B).⁹⁴

Figure 4.

CD4 T cells of R show transcriptomic signatures of naïve-like and early Tfh differentiation, while NR display increased inflammation

(A) UMAP visualization of CD4 T cells subclusters (SC8-SC14) of all donors’ post vaccination. Naive CD4 T cell markers were enriched in SC9-11, while regulatory T cell (Treg) markers were enriched in SC12 and SC13, and early T cell activation markers were upregulated in SC14 (see Results).

(B) UMAP visualization of CD4 T cell subclusters of NR and R.

(C) Dot plots of DEGs per subcluster in CD4 T cells, showing genes associated with naive CD4 T and early Tfh cells in R, and pro-inflammatory response in NR.

(D) Volcano plot of top DEGs of CD4 T cells in R vs. NR.

(E) Dot plot of top 20 leading edge genes of pathway GSE11057_NAIVE_VS_EFF_MEMORY_CD4_T cell_DN (ImmunSigDB, NES = 2.94) ranked by fold change (all DEGs have adjusted p-val<0.05) in CD4 T cells of R vs. NR.

(F) Heatmap of differentially expressed pathways (MSigDB – Hallmark) in CD4 T cells of NR vs. R.

(G) Distribution of PD-1 protein expression (MFI) identified by flow cytometry in CD4 T cells of R and NR, showing significantly higher levels in NR (p = 0.04, by two-tailed Welch’s t test). PD-1 MFI was not statistically significantly different between R and NR CD8 T cells. Flow cytometry analysis also measured CD4⁺ T cell subsets, revealing increased frequencies of CD45RA⁺ cells in R, and increased frequencies of CD45RA⁻ cells in NR (p = 0.02, p = 0.01, respectively, two-tailed Welch’s t test). Data presented as mean ± SD.

(H) Linear regression models of PD-1 (MFI) in CD4 T cells identified by flow cytometry and PDCD1, CD274 gene expression in CD4 T cells by scRNA-seq. See also Figure S3. ∗p ≤ 0.05 by Welch’s t test.

To compare gene expression in CD4 T cells between R and NR, we assigned each sample a responder type and performed differential gene expression analysis. UMAP visualization revealed that SC9 and SC14 had the highest proportion of NR cells relative to R, while SC10-11 had the highest proportion of R cells (Figures 4B, and S3F–S3H). DEGs (adj. p < 0.05) highlighted key differences: R exhibited upregulation of naive CD4 T cell-associated genes (CCR7, LEF1, SATB1, MAL, IL6ST),^{29,37,82,95,96} particularly in SC10 and SC11 (Figures 4C and 4D), and oxidative phosphorylation-related genes (MT-ND3, MT-ND4L, ATP5ME, ATP5F1E, COX7C, TOMM7, PET100),^97,98,99 suggesting the enrichment of naive CD4 T cells in R (Figures 4C and 4D). The upregulation of LEF1, IL6ST, TGFBR2^37,38,100 genes in CD4 T cells and TCF7³⁷ in T cells, and downregulation of Th1-related genes (STAT4, IL2RG, ID2),^84,85,101 is suggestive of early Tfh differentiation in R (Figures 4C and 4D). In contrast, NR CD4 T cells displayed higher expression of genes associated with dysregulation and inflammation, with the upregulation of pro-inflammatory cytokine genes (TNFAIP3, IL-32, IL10RA) and cytotoxicity marker KLRK1 (SC14) (Figures 4C and 4D). Immunomodulators (LGALS1, LGALS3, LTB in SC9) and chemokine CXCR4 were also elevated, along with CASP8, which suggests increased cell death¹⁰² (Figures 4C and 4D). Despite reduced CD4 T cell numbers, NR exhibited enrichment for memory markers (IL7R in SC9, CD44, PTPRC),^91,103 which may be consistent with contraction during long-term memory (Figures 4C, 4D, and Table S9).¹⁰⁴

Next, we performed GSEA with the DEGs of NR vs. R. Only pathways with FDR-adjusted p-val <0.001 are included. Genes within the GSE11057_NAIVE_VS_EFF_MEMORY_CD4_T cell_DN pathway, which are upregulated in effector memory CD4 T cells relative to naive cells, were significantly upregulated in NR compared to R (Figure 4E).^45,46 This supports the enrichment of effector memory-associated transcriptional signatures in NR, as also identified by GSEA (NES = 2.94) (not shown). By contrast, the reduced expression of these genes in R suggests a relatively more naive CD4 T cell-like transcriptional profile. Additionally, Hallmark pathways TNF-α (NES = 2.53), and IFN-γ (NES = 2.20), IL-2/STAT5 (NES = 2.20), as well as mTORC1 (NES = 2.25), hypoxia (NES = 2.52), and p53 (NES = 2.13) pathways were differentially upregulated in NR CD4 T cells (Figure 4F).⁹⁴ Overall, R CD4 T cells are characterized by an enrichment of naive CD4 T cells and early Tfh signatures, whereas NR exhibited pro-inflammatory, stress, and apoptotic signatures.

To further validate the transcriptomic differences of CD4 T cells between R and NR, we measured PD-1 protein levels in CD4 T cells by flow cytometry and correlated them with PDCD1 gene expression. PD-1 was expressed at significantly higher levels in NR than R CD4 T cells (p = 0.04, by two-tailed Welch's t test) (Figure 4G), and its MFI intensity was moderately positively correlated with PDCD1 (PD-1) (Spearman r = 0.51, p-val = 0.09), and with CD274 (PD-L1) (Spearman r = 0.49, p-val = 0.10) gene expression in scRNA-seq data (Figure 4H). In contrast, PD-1 expression within CD8 T cells did not differ between NR and R by flow cytometry (Figure 4G), indicating the increased PD-1 signal is not explained by the higher CD8 frequency in NR (Figure 2A). We also measured the frequency of CD45RA⁺ and CD45RA⁻ CD4 T cells in R and NR (Figure 4G). The frequency of CD45RA⁺ CD4⁺ T cells was significantly higher in R, whereas CD45RA⁻ CD4 T cells were significantly increased in NR (p = 0.02, p = 0.01, respectively, by two-tailed Welch's t test) (Figure 4G). Together, these findings suggest that NR exhibit a loss of naive CD4 T cells, accompanied by an expansion of memory CD4 T cells.

Discussion

While diminished antibody and antigen-specific T cell responses are known to contribute to vaccine inefficacy in older adults, the variation in COVID-19 vaccine responses observed in geriatric populations remains poorly understood. In this study, we examined the immune response of 12 NHRs who received two doses of the BNT162b2 mRNA vaccine, using scRNA-seq and flow cytometry phenotyping of their PBMCs. Geriatric R exhibited higher CD4 T and B cell frequencies and displayed enrichment of genes and pathways associated with naive B cell response, naive CD4 T cell, and early Tfh cell differentiation. In contrast, NR exhibited lower CD4 T and B cell frequencies, higher CD8 T cell and NK cell frequencies, and gene expression signatures associated with a short-lived plasma B cell response and immunological T cell senescence. Our findings suggest that differences in naive T cell and early Tfh enrichment, as well as T cell immune senescence and exhaustion, may underlie the heterogeneity in B cell and antibody responses among older individuals vaccinated with BNT162b2, addressing an important gap in understanding response heterogeneity within this population.

Responder B cells were enriched for gene markers associated with naive B cells (IGHD, BACH2, CD22), oxidative phosphorylation, and B cell development, differentiation, and function (PAX5, IL4R, BCL7A, BTG1). This enrichment in naive B cell-associated genes in R is consistent with a recent study demonstrating that SARS-CoV-2 vaccination can activate spike-reactive naive B cells in unexposed individuals, which subsequently undergo affinity maturation and clonal expansion post-vaccination to contribute to the evolving memory B cell pool and antibody response.¹⁰⁵ Importantly, ETS1 was also expressed at significantly higher levels in B cells from R compared to NR. One of the critical processes in the optimization of an antibody response is class switch recombination, where somatic mutation occurs within the variable regions along with isotype changes in the constant region of the antibodies. ETS1 and PAX5 transcription factors have both been shown to recruit AID (activation-induced cytidine deaminase), the major enzyme responsible for this process, directly to IgH locus sequences.¹⁰⁶ In addition, ETS1 has been shown to inhibit plasmacytic differentiation, in part by blocking BLIMP1’s (PRDM1, upregulated in NR, see later in discussion) ability to bind to DNA, which allows for higher expression of its target genes such as PAX5.¹⁰⁷ We hypothesize that ETS1 may be a key driver in restricting the R' B cells from plasmablast differentiation and propose that higher levels of expression of both ETS1 and PAX5 may result in increased antibody diversity in R as well as a reduced propensity for the terminal differentiation and conservation of the naive B cell compartment.

In contrast, despite a moderated B cell pool, NR exhibited an enrichment of plasma cell-associated (PRDM1(BLIMP1), XPB1, IRF4, MZB1) and immunoglobulin genes (JCHAIN, IGHA1, IGHA2, IGKC), but not CD38. This may reflect an abortive or short-lived plasma cell response lacking full maturation, possibly due to inadequate T cell help or survival signaling.^108,109 Moreover, NR exhibited reduced abundance of B cell clusters enriched for germinal center (CXCR4, CD22) and memory B cell (CD27, CXCR3) markers, which may suggest the impaired formation of immunological memory following vaccination, potentially contributing to their diminished antibody responses. Similar patterns have been observed in low R to the HBV vaccine, where deficient memory B cell responses were associated with weaker immunity, and higher memory B cell frequencies correlated with stronger responses in older individuals.^110,111

Tfh cells play a crucial role in B cell activation and antibody production, and R exhibited transcriptomic signatures consistent with early Tfh differentiation, including the upregulation of TCF7, LEF1, IL6ST, and TGFBR2, downregulation of Th1- and activation-related genes, and enrichment of a BCL6-high Tfh gene signature by GSEA.^37,38,100 However, the downregulation of MAF and upregulation of SATB1, key regulators of mature Tfh cells, along with the absence of differential expression of canonical markers BCL6, CXCR5, and ICOS, suggest a Tfh-primed but not fully committed state. ^112,113 Studies in mouse models have shown LEF1or TCF7 deletion in CD4 T cells impairs Tfh differentiation, leading to reduced B cell responses and germinal center formation, whereas forced LEF1 expression enhances Tfh differentiation by regulating mechanisms upstream of BCL6.³⁷ This regulatory axis, which preferentially guides CD4 T cells toward Tfh differentiation, appears to be downregulated in NR, which may underlie their diminished B cell response. Diminished Tfh responses have been associated with suboptimal vaccine outcomes in various contexts, including kidney transplant recipients, who exhibit blunted germinal center B cell responses and lower frequencies of spike-specific cTfh, CD4, and CD8 T cells post-BNT162b2 vaccination.^17,18 Similarly, low R to the HBV vaccine demonstrate reduced circulating Tfh frequencies compared to high R.¹¹⁴ Furthermore, R exhibited elevated expression of naive CD4 T cell- and oxidative phosphorylation-associated genes as well as a higher frequency of CD45RA⁺ CD4 T cells by flow cytometry, suggesting an enrichment for naive CD4 T cells in R. Recent work has shown that antigen-experienced CD4 T cells can retain a naïve-like phenotype while contributing to long-term immune memory, indicating that such cells may represent a durable and functionally relevant component of vaccine-induced immunity.¹¹⁵

The immune response of NR also exhibited features of T cell senescence and exhaustion. Our scRNA-seq analysis revealed an upregulation of KLRG1 gene expression, and TNF-α, IFN-γ, and mTORC1 signaling pathways in NR T cells, indicating increased immunosenescence. This was further supported by the upregulation of SenMayo gene set markers, which identify senescent cells (Figure 1H).⁴⁴ Additionally, NR showed an enrichment of memory T cell pathways relative to naive T cell pathways, consistent with flow cytometry results demonstrating higher frequencies of CD45RA⁻ CD4 T cells and lower frequencies of CD45RA⁺ naive CD4 T cells. Together, these findings suggest an accumulation of memory T cells at the expense of naive T cells.^23,46 Increased PD-1 signaling and PD-1 protein expression in NR T cells further suggest immune exhaustion.²³ Notably, IL2RG gene expression was upregulated, possibly as a compensatory mechanism to maintain T cell survival and function by enhancing responsiveness to cytokines such as IL-7 and IL-15.¹¹⁶ This may help sustain T cell homeostasis and partial functionality despite inhibitory conditions, such as PD-1-mediated T cell suppression or functional decline due to senescence.¹¹⁷ Further, reduced CD4 T cell counts and downregulation of SELL (CD62L) in T cells, which mediates lymph node homing, suggest impaired T cell trafficking in NR, further contributing to T-B cell interactions and diminished antibody response.¹¹⁸ Together, these findings align with observations of senescence features in geriatric NR to the HBV vaccine.¹¹⁹ In addition, the observed transcriptional downregulation of naive B cell and CD4 T cell markers in NR may reflect a loss of antigen-inexperienced lymphocyte repertoire, diminishing the capacity to mount de novo immune responses. Such a phenotype is consistent with age-associated immunosenescence and may contribute to the poor vaccine responsiveness observed in NR.²³ Reduced thymic function, memory T cell accumulation, and impaired T cell responses have been previously observed after BNT162b2 vaccination, while frailty correlates with increased CD8⁺CD28⁻ TEMRA cells, a hallmark of immunosenescence, in mRNA-1273 vaccine older recipients.^13,15 We presume that immunosenescence could play a role in the impaired response of geriatric NR to the BNT162b2 mRNA vaccine.

Recent evidence, including our findings, highlights the significant role of immunosenescence in vaccine responses in older populations. This underscores the importance of developing more personalized vaccination strategies, such as dose adjustments or specific adjuvants, to enhance response in this vulnerable population. For instance, while recombinant zoster vaccines are less affected by immunosenescence, influenza vaccines require higher doses for comparable immune responses in older adults.^120,121

Understanding the immune mechanisms of BNT162b2 vaccine NR may also provide insights into other mRNA vaccines, including those for influenza, RSV, and CMV, as well as personalized cancer vaccines such as mRNA-4157/V940 for melanoma. Currently, there are 14 mRNA vaccine candidates in Phase I trials, 6 in Phase II, and 1 in Phase III/IV across these indications.^1,122,123 Thus, a deeper understanding of the mRNA vaccine-induced immune response can facilitate the development of more effective vaccines and targeted therapeutics to overcome impaired immune response in NR.

Limitations of the study

This study has several limitations. We cannot necessarily determine whether the observed effects on the immune response to the BNT162b2 vaccine in older persons are due to the mRNA platform itself or the spike-specific antigen response, a distinction in optimizing vaccine strategies. Indeed, lipid nanoparticle (LNP) formulations have been shown to have intrinsic adjuvant activity that can induce Tfh, germinal center B cell, long-lived plasma cell, and memory B cell responses, and outperform other adjuvants such as Addavax.¹²⁴ Our findings suggest that geriatric NRs exhibit impairments in these immune compartments and thus warrant further investigation into LNP’s role in boosting vaccine efficacy. Also, while we were able to match R and NR vaccinees by age and biological sex, our study is limited in sample size owing to enrollment and sample collection limitations within SARS-CoV-2 naive NHRs. Although key transcriptomic findings were supported by orthogonal flow cytometry analyses, limited sample availability precluded additional independent or functional validation experiments. Nonetheless, the cohort was balanced for sex and ethnicity, with a comparable age range, and we believe that the insights provided, along with the accompanying genomic dataset, will serve as a valuable public resource for advancing the biological understanding of COVID-19 mRNA vaccine non-response at the single-cell level.

Resource availability

Lead contact

Further information and requests for resources and reagents should be directed to and will be fulfilled by the lead contact, Cheryl Cameron (cheryl.cameron@case.edu).

Materials availability

This study did not generate new unique reagents.

Data and code availability

• •The raw and processed single-cell RNA-seq data have been deposited at the Gene Expression Omnibus and are publicly available under accession number GSE300265.
• •All original code and any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request.

Acknowledgments

This work was supported by the National Institutes of Health AI129709-03S1 (SG, DHC, and MJC), U01 CA260539 (CLK), P30AI036219 Sub-Project ID 9213 (MJC), and Centers for Disease Control and Prevention 200-2016-91773. The views and opinions expressed are those of the authors and do not represent the policy of the US Dept of Veterans Affairs.

Author contributions

S.G., D.H.C., C.M.C., and M.J.C. conceived and designed the study and obtained funding. B.T., O.A.O., D.K., M.P., D.W., L.C., H.A., K.S., M.S., D.H.C., and C.M.C. coordinated and collected clinical samples. B.T. and C.M.C. performed the RNA-Seq experiments. B.T., J.P., H.A., and C.M.C. performed the flow cytometry experiments. O.A.O., L.C., A.B.B., and D.H.C. performed the serology and microneutralization experiments. J.I., B.T., M.R., J.P., B.M.W., D.H.C., C.M.C., and M.J.C. performed the analysis. B.R., L.Z., S.G., and C.L.K. provided additional scientific guidance. J.I., B.T., M.R., D.H.C., C.M.C., and M.J.C. wrote the manuscript.

Declaration of interests

SG and DHC are recipients of investigator-initiated grants to their universities from Pfizer to study pneumococcal vaccines, and Sanofi Pasteur and Seqirus to study influenza vaccines, and SG from Genentech on influenza antivirals. SG and DHC are recipients of a collaborative grant to their universities from Moderna to study respiratory infections, and SG from GSK to study the shingles vaccine. SG also receives consulting fees from AstraZeneca, GlaxoSmithKline, Icosavax, Janssen, Merck, Moderna, Novavax, Pfizer, Sanofi, Seqirus, Shionogi, and Vaxart, and has received fees for speaking for AstraZeneca, GlaxoSmithKline, Janssen, Moderna, Pfizer, Sanofi, and Seqirus. All other authors declare no competing interests.

STAR★Methods

Key resources table

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
Human Phycoerythrin-conjugated Donkey F(ab)2 IgG, with Fcγ	Jackson ImmunoResearch	RRID: AB_2340519
Human monoclonal HLA-DR FITC	Biolegend	RRID: AB_314682
Human monoclonal CD19 APC	eBioscience	RRID: AB_10804519
Human monoclonal CD38 Alexa Fluor 700	Biolegend	RRID: AB_2072781
Human monoclonal CD14 APC-Cy7	Biolegend	RRID: AB_830693
Human monoclonal CD56 Pacific Blue	Biolegend	RRID: AB_10612566
Human monoclonal CD25 BV605	Biolegend	RRID: AB_11218989
Human monoclonal CD8 BV786	BD	RRID: AB_2687487
Human monoclonal PD-1 PE-Cy-7	Biolegend	RRID: AB_2159324
Human monoclonal CD3 BUV395	BD	RRID: AB_2744387
Human monoclonal CD45RA PE Texas Red	Invitrogen	RRID: AB_10372222
Human monoclonal CD74 BUV805	Biolegend	RRID: AB_2075505
Human monoclonal CD4 BUV805	BD	RRID: AB_2744423
Bacterial and virus strains
Stabilized full-length S protein (aa 16-1230 with furin site mutated)	Canaday et al., 2021	N/A
SARS-CoV-2 pseudotyped pHAGE-CMV-Luc2-IRES-ZsGreen-W lentivirus	Wilfredo et al., 2021	N/A
SARS-CoV-2 ΔC18 pseudotyped pHAGE-CMV-Luc2-IRES-ZsGreen-W lentivirus	Wilfredo et al., 2021	N/A
SARS-CoV-2 ΔC18 D614G pseudotyped pHAGE-CMV-Luc2-IRES-ZsGreen-W lentivirus	Wilfredo et al., 2021	N/A
Chemicals, peptides, and recombinant proteins
Dulbecco’s Phosphate-Buffered Saline (DPBS)	Invitrogen	Cat # 14190144
Bovine Serum Albumin (BSA)	Invitrogen	Cat # AM2616
RPMI (1640) Medium	Corning	Cat # 10-040-CV
Fetal Bovine Serum	Gibco	Cat # 10438026
DMSO	Sigma	Cat # D2650-100ML
Critical commercial assays
Chromium Next GEM Single Cell 3' Kit v3.1	10X Genomics	Cat # 1000121
Chromium Next GEM Chip G Single Cell	10X Genomics	Cat # 10001210
Single Index Kit N, Set A	10X Genomics	Cat # 1000212
Live Dead Fixable Aqua Viability Kit	Invitrogen	Cat# NC0180395
Deposited data
Processed single cell RNA-seq data	Gene Expression Omnibus	GSE300265
Software and algorithms
Cellranger v4.0.0	10X Genomics	RRID:SCR_0173
R (programming language) v4.1.1	R	RRID:SCR_001905
Seurat R package v4.1.1	Hao et al. 2021	RRID:SCR_016341
Harmony R package v0.1.0	Broad Institute	RRID:SCR_022206
ggplot2 R package	R	RRID:SCR_014601
BioRender	BioRender	RRID:SCR_018361
Gene Set Enrichment algorithm (GSEA)	Broad Institute	RRID:SCR_003199
Cytoscape v3.10.3	Cytoscape Consortium	RRID:SCR_003032
Drug Perturbation Enrichment algorithm (dpGSEA)	Fang et al., 2021	RRID:SCR_025351
Limma package (Bioconducter)	Ritchie et al., 2015	RRID:SCR_010943
Python v3.9.7	Python	RRID:SCR_008394
FlowJo Version 10	Flowjo.com	RRID:SCR_008520
FACS Diva software	BD Biosciences	RRID: SCR_001456
Prism v10	GraphPad Software	RRID: SCR_002798
Molecular Signature Database (mSigDB) v7.5.1	Liberzon et al., 2011, 2015	RRID:SCR_016863

Experimental model and study participant details

Study design

The goal of the study was to examine the gene expression differences in the adaptive immune response to the BNT162b2 mRNA vaccine among the geriatric population.

Human participants

In this study, vaccine R (n=6) and vaccine NR (n=6) were selected from a cohort comprising 12 SARS-CoV-2 naïve nursing home residents. The detailed description of the cohort of the longitudinal observational study was previously outlined.³ This study was approved by the Western-Copernicus Group Institutional Review Board (protocol #STUDY20211074). All participating residents or their legally authorized representatives provided informed consent. The study complied with the 1964 Declaration of Helsinki and its later amendments. The median age of R was 71.5 [range: 65-99], while for NR, it was 86 [range: 76-88]. The difference in age was not statistically significant (Welch’s t-test p = 0.10; Mann-Whitney U p = 0.065). Both groups had a 50% male composition. All R were Caucasian, whereas 83% of NR were Caucasian and 17% were African American. We identified prior SARS-CoV-2 infection through polymerase chain reaction (PCR), antigen testing, and/or the presence of high antibody titers against the SARS-CoV-2 spike protein and receptor-binding domain (RBD). We collected blood samples post initial BNT162b2 mRNA vaccination series, 14 ± 3 days after their second dose.

Method details

Anti-spike assay

The anti-spike assay and AU measurement methods were outlined previously.³ Briefly, to measure anti-spike antibody levels in our cohort, we used stabilized full-length S protein (aa 16-1230, with furin site mutated) conjugated to magnetic microbeads (Luminex). We then measured antigen-specific IgG in patient serum using anti-human Phycoerythrin-conjugated F(ab)2 IgG (Jackson ImmunoResearch). Mean fluorescent index (MFI) was measured using the Magpix assay system (BioRad). To ensure consistency, we employed an internal standard from convalescent plasma and ran a standard curve of half-log dilutions starting at 1:100, assigning a relative antibody unit (AU) value of 4000 for this dilution. AU values were interpolated from this curve. A receiver operating curve, based on pre–SARS-CoV-2 serum controls and SARS-CoV-2 convalescent samples, determined cutoff values for convalescent antibody levels with 99% specificity and 98% sensitivity using S protein antibodies.

SARS-CoV-2 pseudovirus neutralization assay

The neutralization assay method was previously described.³ To compare the neutralization ability of our cohort’s sera against SAR-CoV-2, we used lentiviral particles pseudotyped with vaccine strain spike protein (Wilfredo et al., 2021). We conducted three-fold serial dilutions (1:12 to 1:8748) for each serum sample, followed by incubation with 50–250 infectious units of pseudovirus for 1 hour. The assay and readouts were performed using Fluent Automated Workstation liquid handler (Tecan) and 348-well plates (Grenier). We determined the percentage of neutralization by subtracting background luminescence measured in cell control wells (containing cells only) from sample wells and then dividing by the luminescence measured in virus control wells (containing virus and cells only). Lastly, we calculated pseudovirus neutralization titers (pNT50) by taking the inverse of the 50% inhibitory concentration value for all samples exhibiting a pseudovirus neutralization value of 80% or higher at the highest serum concentration.

PBMC isolation and cryopreservation

To isolate peripheral blood mononuclear cells (PBMCs) from whole blood, we purified the cryopreserved cells from buffy coats by centrifugation over a Ficoll-Hypaque gradient (GE Healthcare) and then cryopreserved in 10% DMSO/ 90% FBS in liquid nitrogen. PBMCs were later thawed using standard protocol, resuspended in Dulbecco’s phosphate-buffered saline (DPBS) DPBS with 0.04% bovine serum albumin (BSA). We determined cell count and viability using trypan blue on a Countess II FL (Thermo Fisher).

Single-cell RNA-sequencing

To sequence PBMCs using single cell RNA-seq technology, ∼10,000 PBMCs per sample were immediately loaded into a Chromium Single Cell Chip (10X Genomics). GEM generation was performed, followed by reverse transcription, cDNA amplification for 11 cycles of PCR, and library generation according to the 10x Genomics 3’ gene expression (v3.1 chemistry) user guide. Sequencing was performed on a Novaseq S2 flow cell on a Novaseq 6000 (paired-end, 100 cycle run design) to obtain approximately 40,000 reads/cell.

Flow cytometry assay

For immune cell phenotyping, cells were first stained with Live/Dead Aqua (Invitrogen) followed by cocktails of monoclonal antibodies recognizing the following cell surface markers: CD3, CD4, CD8, CD56, CD19, CD14, HLA-DR, CD38, CD25, CD74, CD45RA and PD-1. Cells were washed, fixed and resuspended in staining buffer. All events were collected on a BD ARIA-SORP instrument. Data was analyzed using FlowJo software (TreeStar).

Quantification and statistical analysis

Antibody titers statistical analysis

To analyze the differences in antibody levels in our cohort, we performed Welch’s t-tests between two groups, R and NR, after confirming normal distribution but unequal variances (Figure 1B). The scale in the figure is normalized to log10. P-values and standard error bars are reported. No significant differences between females and males were observed in any of the titers (not shown). To confirm that the observed effect is not influenced by the differences in age between R and NR, we ran linear regression analyses involving anti-spike levels, responder type, and age (not shown). A simple linear regression model between age and anti-spike showed that there is no association. After controlling for age as a covariate in a linear regression model, anti-spike levels remained significantly different between R and NR. Additionally, age was not significantly associated with anti-spike levels when controlling for the effect of responder type. As such, age does not confound the relationship between responder type and anti-spike levels in our cohort. All analyses and visualizations were performed using R programming language v4.1.1.

Single cell RNA-seq data processing

To preprocess the raw single cell RNA-seq data, we used cellranger software v4.0.0 (10X Genomics) with the default parameters. First, we demultiplexed raw base call files (BCL) files for each flow cell directory into FASTQ files using cellranger mkfastq. This process separated reads by sample while adding cell barcodes and Unique Molecular Identifiers (UMIs) to read headers. We then aligned reads to the GRCh38 reference genome and quantified gene expression by counting UMIs associated with each gene, generating a gene-barcode matrix using cellranger count. In total, this yielded an estimated number of 140,674 cells. Finally, we employed the Scrublet and Scanpy packages (Python) to detect potential cell doublets within our dataset. This entailed computing doublet scores for each cell, setting a specific threshold (0.25), and labeling cells as doublets if their scores surpassed this threshold. Identified doublets were subsequently removed resulting in an estimated total of 133,328 cells for downstream analysis.

Clustering and batch-correction

To analyze the preprocessed single cell RNA-seq data, we utilized a custom pipeline based on Seurat v4.1.1 (R programming language v4.1.1). Initially, we applied several filtering steps: cells with fewer than 4000 detected features were removed, while total counts were limited to a range of 750-15000. Additionally, we removed reads exhibiting high mitochondrial gene expression (>25%). In total, we analyzed 121,322 cells post quality control processing. We then normalized the gene expression by applying a logarithmic transformation using the NormalizeData() function. We then identified and selected highly variable features and scaled the normalized gene expression by a linear transformation using FindVariableFeatures(), and ScaleData(), respectively. We performed linear dimensionality reduction by Principal Component Analysis (PCA) using RunPCA(). We then identified cell clusters by calculating the neighborhood graph using FindNeighbhors() with 13 dimensions and performed community detection by the Louvain algorithm using FindClusters() with 0.2 resolution. Lastly, we performed dimension reduction analysis using Uniform Manifold Approximation and Projection (UMAP). Following the initial clustering, we used RunHarmony() v0.1.0 (Broad Institute) to correct for batch effect in our dataset by aligning cell embeddings across different experimental batches while preserving the biological signal. Finally, we re-clustered the cells as described earlier to refine the clusters.

Subset identification and analysis

We identified immune cell subsets from peripheral blood mononuclear cells by examining the gene expression of canonical markers within each cluster. Differentially expressed genes (DEGs) were determined using the FindAllMarkers() function in Seurat (v4.1.1) with default parameters, a logFC threshold of 0.25, and a non-parametric test (Wilcoxon rank-sum) for ranking genes, applying a p-value cutoff of < 0.05. Multiple testing correction was performed using the Bonferroni method, and only DEGs with an adjusted p-value < 0.05 were considered significant and discussed in the results. As such, CD4 T cells were identified based on their high expression of canonical DEGs: CD3D, CD3E, CD3G, CD4, CD28, GATA3,¹²⁵ STAT3,¹²⁶ RORA.¹²⁷ CD8 T cells were identified based on: CD3D, CD3E, CD3G, CD28, CD8A, CD8B. B cells were identified based on: MS4A1 (CD20), CD22, CD79A, IGHM (IgM), IGHD (IgD), PAX5.⁶⁶ NK cells were identified based on: NCAM1 (CD56), FCGR3A (CD16), KLRD1 (CD94). Monocytes were identified based on: CD14, FCGR3A (CD16), CD64. Dendritic cells based on: CD1C,¹²⁸ CLEC10A.¹²⁹ Erythrocytes were identified based on the high expression of hemoglobin genes (HBB, HBA1, HBA2, HBD, HBM) and platelets based on high expression of platelet factors (PF4, PPBP, ITGA2B¹³⁰) (Figures 1D, S1A and S1B). Average log2FC, p-values, and adjusted p-values for clusters of PBMCs, and subclusters of B cells, and CD4 T cells subsets can be found in Tables S2, S6, S8 respectively. Subsequently, we isolated the B cells and CD4 T cells subsets for further analysis, following the previously described methodology. Our analysis encompassed a total of 6,263 cells utilizing 7 dimensions with a resolution of 0.2 for the B cell subset and 23,200 cells, utilizing 7 dimensions with a resolution of 0.4 for the CD4 T cell subset.

Differential gene expression analysis

To identify top DEGs in immune R and NR, we categorized each sample according to responder type and used FindAllMarkers() (Seurat v4.1.1) with same parameters as described above. After quality control, we analyzed 62,705 PBMC from R and 58,617 from NR. Only DEGs with an adjusted p-val <0.05 were considered significant and discussed in the results. Average log2FC, p-values, and adjusted p-values for R and NR of PBMCs, T cells, B cells, and CD4 T cells subsets can be found in Tables S3, S4, S7, S9 respectively. We created visualizations using ggplot2 package (R programming language v4.1.1).

Pathway and network analysis

To examine differentially expressed pathways in cell clusters and in R and NR, we ranked the generated gene lists using a composite score incorporating both statistical significance (−log10(p)) and magnitude of change (log2FC), weighted by direction of expression, as input for Gene Set Enrichment Analysis (GSEA) using the fgsea() function with default parameters and 10,000 permutations (R programming language v4.1.1). Only gene set pathways with FDR-adjusted p-val < 0.001 are discussed in the manuscript. We downloaded and used MSigDB gene set libraries including Hallmark, C2 curated (Reactome and KEGG), C5 Gene Ontology (GO), and C7 immunologic signature (ImmuneSigDB, Vaccine Response) gene sets.^45,94 To perform network analysis, we selected the top 100 DEGs in R and NR ranked by highest fold change and used String database and software v12.0. Only networks with PPI Enrichment Score < 0.001 are included in the analysis. We created all visualizations using ggplot2 package (R programming language v4.1.1) and Cytoscape v3.10.3.

Flow cytometry statistical analysis

To quantify and analyze the characteristics of cells in R and NR, we analyzed the flow cytometry data using FlowJo software v10.0. The gating strategy is shown in Figure S4. We then performed statistical analysis and generated figures using GraphPad Prism v10. Normality was assessed using standard tests (p >0.10). For normally distributed data, unpaired Welch’s t-tests (two-tailed) were used. For non-normally distributed data, comparisons were performed using Wilcoxon rank-sum tests (two-tailed). One-tailed Welch’s t-tests were applied when a directional hypothesis was defined by the study design. Statistical significance was defined as p-val ≤ 0.05. ∗p ≤ 0.05 by Welch’s t-test. ∗∗p ≤ 0.05 by Wilcoxon rank-sum test.

Correlation models

To integrate antibody levels and gene expression with flow cytometry data, we built linear correlation models. We first correlated B cell and CD4 T cell frequencies in PBMCs identified by flow cytometry with anti-spike and anti-neutralization titers. Next, we correlated single-cell RNA-seq data with flow cytometry data, comparing cell type frequencies identified by gene expression with those characterized by flow cytometry, and correlating MFI values of specific markers to their gene expression. We assessed the normality of variables using the Shapiro-Wilk test and histograms, applying Pearson correlation for normally distributed variables and Spearman correlation for non-normal variables.

Published: April 16, 2026