Early immune events during SARS-CoV-2 infection impact memory T and B cell responses

Abstract

Immune responses elicited by natural infection of the coronavirus SARS-CoV-2 (COVID-19) show significant heterogeneity in the magnitude and quality of memory T and B cell responses. However, little is known about the contributing factors. In this study, we investigated the early immune factors that contribute to this variability using RNA-seq, targeted proteomics, and flow cytometry analyses. Specifically, we sought to investigate associations between early immune responses and SARS-CoV-2 memory immunity in a longitudinal cohort of 46 individuals hospitalized for COVID-19 from May 2020 to March 2021. These participants returned for follow-up visits up to one-year post-hospitalization where we characterized antibody titers, antibody neutralization, antibody durability, and cellular memory T and B cell responses with multiple assays. Additionally, using integration analysis of Omic measurements, we identified common genes, proteins, and cellular pathways associated with differential memory response outcomes. Our data suggests that high levels of inflammatory proteins, and co-stimulatory molecules during the early stages of COVID-19 lead to enhanced memory T and B cell responses and improved durability. Alternatively, molecules that have a negative effect on dendritic cell maturation including TNFSF11 and FLT3LG correlated with suboptimal memory immune responses. Importantly, we were able to identify early markers that are positively and negatively associated with durable antibody responses in infected participants. This study provides a unique and thorough examination of both innate and memory immunity in the same patients over time, offering valuable insights into the long-term durability of SARS-CoV-2 immunity.

Longitudinal multi-omic analysis of pre-vaccination, hospitalized COVID-19 patients identifies early immune markers predicting durable or suboptimal memory T and B cell responses, revealing potential pathways that shape long-term SARS-CoV-2 immunity.

Introduction

Coronavirus disease 2019 (COVID-19), caused by the human pathogen severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2), has led to significant global morbidity and mortality^1,2. SARS-CoV-2 spread rapidly through the population, escalating into a major public health crisis as the pandemic progressed. COVID-19 presents with a wide range of clinical manifestations, from asymptomatic cases to mild, moderate, and severe disease. While most cases are mild and commonly feature symptoms such as fever, cough, and gastrointestinal distress, severe cases are marked by systemic inflammation, cytokine storms, acute respiratory distress syndrome, and multi-organ damage³. Coronaviruses are RNA viruses, some of which can infect human lung epithelium by binding to the angiotensin-converting enzyme 2 (ACE2) receptor. The SARS-CoV-2 spike protein mediates viral attachment, fusion, and entry⁴. The Spike (S) protein consists of two subunits: S1 and S2. The S1 protein contains the Receptor Binding Domain and binds to host ACE2^4,5. Meanwhile, S2 fuses to the host cell membrane^4,5. The intricate role of Spike receptor binding and attachment to host cells makes this protein complex a major target for neutralizing antibodies.

Upon entry, viral RNA activates endosomal and cytoplasmic sensors such as TLR3/7 and MAVS, respectively. As the infection progresses, infected and neighboring cells recognize pathogen-associated molecular patterns and damage-associated molecular patterns through cytosolic pattern recognition receptors^3,6,7. Concurrently, resident alveolar macrophages release cytokines that recruit innate immune cells, including monocytes, macrophages, neutrophils, natural killer cells, and dendritic cells. Antigen-presenting cells activate CD4 + T helper cells and CD8+ cytotoxic T cells. CD8+ cells mediate the killing of virally infected cells, while CD4+ cells support the immune response by releasing cytokines such as IFN-γ and interacting with B cells to promote antibody production^7,8. Interferons, proinflammatory cytokines, and chemokines are significantly upregulated during COVID-19, including IL-1β, IL-6, TNF, IL-10, IFN-β, IFN-γ, IL-17, IL-18, CXCL10, CCL2, and CCL3, CASP-8 among others^9–11. While cytokine release aids in viral clearance and immune cell recruitment to the lungs, severe COVID-19 can lead to dysregulated immune responses. Excessive cytokine secretion may result in inflammatory cell death (PANoptosis), further releasing cytokines and potentially triggering a cytokine storm. In severe cases, systemic cytokine release creates a positive feedback loop, resulting in life-threatening damage to host tissues and organs^9–11. Thus, a delicate balance in the induction of cytokines and chemokines following SARS-CoV-2 infection can lead to strong immunity and/or severe disease.

Understanding the quality and nature of immune memory responses to SARS-CoV-2 infection became crucial, particularly before vaccine availability. At the beginning of the pandemic, the ImmunoPhenotyping Assessment in a COVID-19 Cohort (IMPACC) study was designed as a prospective longitudinal study¹². This nationwide study collected clinical and biological data to identify immunologic, virologic, proteomic, metabolomic, and genomic features of COVID-19 susceptibility, severity, and disease progress¹². This cohort reflects a naïve population with severe disease leading to hospitalization, recovery, and vaccination, giving insight into hybrid immunity. This study in addition to other longitudinal cohorts following COVID-19 gives us insight into memory immune function following viral infection. Circulating IgG antibodies, memory B cells, and memory T cells persist up to one year post infection, though IgG antibody titers decline the fastest. The estimated half-lives of Spike-binding IgG antibodies and neutralizing antibodies range from 140 to 364 days¹³. However, even after two doses of COVID-19 mRNA vaccines, neutralizing antibodies wane within 7 months of vaccination¹⁴. Circulating memory B cells and long-lived plasma cells exhibit stable frequencies over time, detectable up to eight months following infection¹³. After vaccination, germinal centers remain active, enabling B cells interaction with T follicular helper (Tfh) cells^14,15. This interaction facilitates multiple rounds of somatic hypermutation, resulting in stronger and more specific antibody responses^14,15. Memory CD4 + T cells display slight decay over time post infection, with an estimated half-life of 200 days¹³. These studies highlight the importance of vaccination and booster doses in sustaining long-term protection. However, little is known about the early immune events following SARS-CoV-2 infection associated with memory T and B cell responses.

In this study, we investigated the immune responses in 46 individuals during early infection, and up to one year after infection using a longitudinal cohort. As expected, we observed large variability in memory T and B cell responses. We have shown that cytokines such as CXCL10, CCL23, and IL-6 positively correlated with strong memory B cells and higher neutralizing antibody titers, while type I interferons, and proteins such as FLT3LG and TNFSF11 were negatively correlated with antibody neutralization and decay. Additionally, we found that the frequency of the monocyte subset CD14dimCD16+ that secrete type I interferon is negatively correlated with antibody decay, which suggests a negative impact of type I interferons on antibody durability. This study provides a unique and comprehensive look into how multiple omics and cellular data from patients hospitalized with COVID-19 influence memory immune responses.

Results

A total of 106 individuals hospitalized with COVID-19 in the Drexel site as part of the IMPACC study were recruited for this study, including 49 females and 57 males. Participants were collected as part of the IMPACC cohort, and all participants collected for Drexel University presented with COVID-19 symptoms, with cases ranging from mild to severe. Participant samples were collected at multiple Tower Health locations, including Reading, Pottstown, and Chestnut Hill hospitals, between May 2020 and March 2021. During the study, peripheral blood, nasal swabs, and sputum samples were collected. Blood samples were processed into serum, plasma, and peripheral blood mononuclear cells (PBMCs). Additionally, 46 participants returned for longitudinal blood sampling over several months (3, 6, 9, and 12 months post discharge). These 46 participants were chosen for this study due to multiple longitudinal visits up to 12 months post hospitalization. The median age of participants was 61.5 (20.25 IQR), with 19 participants above the age of 65 at enrollment (range 66–86) (Table 1). Participants reported a median of 3 comorbidities (2 IQR), smoking (18, 39%), hypertension (29, 63%), and diabetes (16, 35%) were the most common. Additionally, 32 (69%) participants had BMIs above or equal to 30 kg/m² (Table 1, Supplementary Data 1). The study coincided with the rollout of SARS-CoV-2 mRNA vaccines, which became publicly available toward the end of the enrollment period. By the 9-month time point, many participants had received vaccinations, and by 12 months, 31 participants (69%) were fully vaccinated with two doses of an mRNA vaccine. These individuals were classified as having hybrid immunity, being both previously infected and vaccinated.

Table 1.

Participant demographics

Participant demographics		Overall (n = 46)
Enrollment (years), median (IQR)	(n = 46)	61.5 (20.25)
Age ≥ 65 years at enrollment	≥ 65 years	19 (41%)
Sex at Birth	Female	24 (52%)
	Male	22 (48%)
Race	Asian	2 (4%)
	Black/African American	9 (20%)
	Other	2 (4%)
	Unknown/unavailable	1 (2%)
	White	32 (70%)
Ethnicity	Hispanic or Latino	5 (11%)
	Not Hispanic or Latino	41 (89%)
Comorbidities and BMI
Number of comorbidities, median (IQR)		3 (2)
	Respiratory (no asthma)	6 (13%)
	Any substance	1 (2%)
	Smoker/vape	18 (39%)
	HIV	1 (2%)
	Hypertension	29 (63%)
	Asthma	8 (17%)
	Cardiac	5 (11%)
	Chronic kidney disease	2 (4%)
	Diabetes	16 (35)
	Liver	2 (4%)
	Neoplasm	3 (7%)
Body Mass index (kg.m2), median (IQR)		33.5 (9.55)
BMI category	Class 1–2 obesity (30–39.9)	24 (52%)
	Class 3 obesity (40+)	8 (17%)
	Normal weight	4 (9%)
	Overweight	10 (22%)
Level of admission	Floor	37 (80%)
	ICU stepdown	9 (20%)
Level of respiratory support at admission	Not requiring supplemental O2	17 (37%)
	Non-invasive ventilation or high flow	2 (4%)
	Supplemental oxygen (not high flow)	27 (59%)
SOFA score, median (IQR)		0 (0.75)
Symptom onset to hospitalization	<3 days	7 (15%)
	4–7 days	14 (30%)
	8–14 days	15 (33%)
	>15 days	1 (2%)
Severity & Utilization during acute hospitalization
ICU at any time during hospitalization		10 (22%)
Acute trajectory group	1	23 (50%)
	2	15 (33%)
	3	6 (13%)
	4	2 (4%)
Vaccination status
	Two doses of mRNA vaccine	31 (69%)
	No vaccine	14 (31%)
	Participants vaccinated by 3 months	3 (7%)
	Participants vaccinated by 6 months	11 (24%)
	Participants vaccinated by 9 months	28 (62%)
	Participants vaccinated by 12 months	31 (69%)
Steroid use during hospitalization	No	9 (20%)
	Yes	37 (80%)

Disease severity did not affect the quality or magnitude of memory T cell and B cell responses

Disease severity for the 46 patients was previously characterized by longitudinal modeling of their respiratory ordinal score over the first 28 days post hospital admissions and was performed on the basis of the entire IMPACC cohort¹². Each participant was assigned to a trajectory group (TG) reflecting the severity of their acute disease course¹². Trajectory Group 1 (TG1, n = 23) was characterized by mild respiratory disease and a brief hospital stay with no limitations at hospital discharge. TG2 (n = 15) required more respiratory support and had a longer hospital stay and no limitations at discharge. TG3 (n = 6) was characterized by roughly similar respiratory support and a longer length of stay compared to TG2, but had limitations at discharge. Two additional groups had overall high respiratory support during their hospital stay. TG4 (n = 2) and TG5 received more aggressive respiratory support and had prolonged length of stay, and TG5 specifically experienced fatal illness by day 28 (see Supplementary Data 2). We assessed the association between the trajectory groups and markers of cellular and humoral response to infection, including frequencies of memory B cells specific to the SARS-CoV-2 Receptor Binding Domain (RBD) and SARS-CoV-2 memory T cells measured by the activation induced marker (AIM) assay 1 year following the initial hospital admission as well as RBD-specific IgG titers and neutralizing antibodies measured 3–4 months after hospitalization and before mRNA vaccination (Fig. 1a). Additionally, the Benjamin–Hochberg procedure was used for controlling the false discovery rate in multiple hypothesis testing (adj-p value, Fig. 1a). We did not identify any statistically significant association suggesting that disease severity during the acute phase of COVID-19 was not a major contributor to the variations in cellular and humoral memory responses in our cohort.

Fig. 1. Association between the trajectory groups (established during the acute phase of the COVID-19 disease) with humoral and cellular response measured 6 months to 1 year following hospitalization for SARS-CoV-2 infection.

a Ordinal regression models were fit to assess the association between the trajectory groups and cellular/humoral responses. Regression coefficients (estimate) and corresponding p-values (“Pr(>|z|)”), and adjusted p-values are provided in the table. b Boxplot showing IgG titers as a function of the trajectory groups. Box represents the 1st and 3rd quartiles, while whiskers indicate the 5th to 95th percentile. Dots indicate observed titers for each participant. No significant association was observed between the two variables (z-test: p = 0.108) TG1 n = 23, TG2 n = 15, TG3 n = 6, TG4 n = 2.

Frequency of circulating memory B cells is variable one year post infection and is associated with early immune markers

We characterized the frequency of memory B cells one year post hospitalization using flow cytometry. Using participant PBMCs, we assessed the frequencies of RBD-specific memory B cells. We gated on CD3-CD19 + CD27+IgD− and utilized two streptavidin probes to identify RBD+ cells. Supplementary Fig. 1 depicts the gating strategy with proper controls, RBD+ memory B cells were identified in 36 out of 40 participants (see Supplementary Data 3). The frequencies of RBD+ memory B cells were robustly detectable and were not affected by the vaccination status (Fig. 2a). Specifically, mRNA vaccination was not significantly associated with an increase in the frequencies of circulating memory B cells when assessed 1 year post hospitalization (Fig. 2a). Next, we examined the association between serum protein expression (at admission) and the frequency of RBD-specific memory B cells one year post hospitalization (Fig. 2b). Normalized protein expression (NPX) was assessed in participant serum using Olink panel that included 92 proteins. Additionally, we assessed the frequency of memory B cell association with 65 proteins measured by a multiplex Luminex assay. From these analyses, five proteins were found to correlate with frequencies of antigen-specific memory B cells one year post hospitalization. Memory B cells significantly correlated with CCL23 (p = 0.0002), FLT3LG (p = 0.0048), CXCL10 (p = 0.0459), EIF4EBP1 (p = 0.0087), and CASP8 (p = 0.0178) (Fig. 2c). Among these, only FMS-like tyrosine kinase 3 ligand (FLT3LG) showed a negative correlation with memory B cell responses, indicating that higher levels of FLT3LG protein during the acute phase of the COVID-19 disease are linked to reduced memory B cell frequency 1 year following hospitalization. FLT3LG gene encodes FLT3LG, which binds to FLT3 on cells of myeloid origin^16,17. Mouse studies suggest that FLT3LG is downstream of TLR activation and FLT3LG/FLT3 signaling impacts dendritic cell function and their ability to stimulate memory responses. CCL23 and CXCL10, which play important roles in immunoregulation and inflammatory processes, were positively correlated with the heightened frequency of memory B cells.

Fig. 2. Frequency of antigen-specific circulating Memory B cells is variable one year post hospitalization.

a Frequency of circulating RBD-specific memory B cells (%CD19 + CD27+IgD−) one year post hospitalization. b Study timeline highlighting time points compared in these analysis. This figure compares Olink protein from Day 0 with memory B cells one year post hospitalization. c CCL23, FLT3LG, CXCL10, EIF4EBP1, and CASP8 normalized protein expression (NPX) levels in serum at Visit 1 correlate with RBD-specific memory B cell frequencies. R and p values are reported for each correlation, calculated using Spearman correlation. p < 0.05*, p < 0.01**, p < 0.001***, p < 0.0001****.

Frequency of circulating memory T cells is variable one year post infection and is associated with early immune markers

Next, we assessed the frequencies of memory T cells. SARS-CoV-2 antigen-specific memory CD4 + T cells were identified in 39 participants. PBMCs were stimulated with a SARS-CoV-2 Spike peptide pool, and CD4 + T cells were identified using the AIM assay. AIM + T cells are characterized by markers such as CD69, surface CD40L [sCD40L+], 4-1BB and OX40 (Supplementary Fig. 2). One year post hospitalization, we did not see a clear differentiation between participants with hybrid and natural immunity (Fig. 3a). We examined correlations between AIM+ memory T cells and serum protein levels from participants’ admission (Fig. 3b). Of these, six proteins were of particular interest due to their immunological functions: CXCL1 (p = 0.0113), TNFSF14 (p = 0.0294), IL-33 (p = 0.0279), CCL2 (p = 0.0322), FLT3LG (p = 0.0103), and CXCL13 (p = 0.0353) (Fig. 3c). A total of 20 unique proteins were significantly correlated with memory T cell frequencies one year after hospitalization (Fig. 3d). Interestingly, we observed common proteins that are positively correlated with both frequency of memory B and AIM + T cells. Importantly, we observed the same FLT3LG to be significantly and negatively correlated with the frequency of AIM + T cells, similar to what we observed for memory B cells. These findings suggest that higher FLT3LG expression may influence a regulatory mechanism or immune dysregulation during the early inflammatory response to COVID-19 infection and might have an effect on the frequency of memory B and T cells. Additionally, a significant positive correlation was observed between CXCL13 and memory AIM + CD4 + T cells, suggesting a key role for CXCL13 in sustaining long-term immune memory. In fact, the expression of CXCL13 has been shown to correlate with active germinal centers and is a strong marker for follicular helper T cells in vaccine and infection settings^15,18. Additionally, IL-33, a potent activator of memory T cells, is also positively correlated with the frequency of AIM + T cells and could predict a strong T cell response¹⁹. TNFSF14 (LIGHT) is a protein heavily involved in inducing strong memory T cells and memory B cells^18,20. These findings showed that a higher frequency of memory B cells after infection is associated with specific inflammatory cytokines and costimulatory receptors. These results highlight the connections between the strength of memory immune responses and the levels of inflammatory cytokines and chemokines during early infection.

Fig. 3. Frequency of circulating memory T cells to SARS-CoV-2 is variable one year post hospitalization.

a Frequencies of SARS-CoV-2-specific AIM + T cells (CD69 + sCD40L+, CD69 + 4-1BB+, OX40 + sCD40L+) one year post hospitalization. b Study timeline highlighting time points compared in these analyses. This figure compares Olink protein from Day 0 with AIM+ memory CD4 T cells one year post hospitalization. c CXCL1, TNFSF14, IL-2, IL33, TGFA, CCL2, FLT3LG, CXCL8 normalized protein expression (NPX) levels in serum correlated with AIM+ Antigen-specific memory T cells. d Table of memory T cell subsets correlated with early innate Olink proteins (NPX). R and p values are reported for each correlation, calculated by Spearman correlation. p < 0.05*, p < 0.01**, p < 0.001***, p < 0.0001****.

Early immune markers are associated with neutralizing antibody magnitude and decay rate

We measured neutralizing antibody response in our cohort at all time points from hospital admission to 1 year post hospitalization. This allowed us to identify early immune markers affecting the magnitude and decay of the response. In Fig. 4a, we show peak neutralizing response at day 4–7 post infection, and this is maintained with some decline for 1 year post hospitalization for both the Wuhan and Alpha strain. These strains were selected based on their predominant circulation during the study period (May 2020–March 2021). At three months post infection, all participants exhibited high IC50 values for neutralizing antibodies against the Wuhan and Alpha variants with high variability (Fig. 4a). To explore potential drivers of these responses, we assessed correlations between Visit 1 Olink protein levels and neutralizing antibody IC50 values at three months for both variants. Neutralizing antibody responses were correlated with CXCL10 (p = 0.004), CSF1 (p = 0.0002), IL-18 (p = 0.004), PD-L1 (p = 0.0001) and CD40 (0.0011) (Fig. 4b and Supplementary Fig. 3). Specifically, neutralizing IgG IC50 values for the Wuhan strain showed negative correlations with CXCL5 (p = 0.0241) and TNFSF11 (p = 0.0246), while displaying strong positive correlations with key inflammatory proteins, such as IFNγ (p = 0.0231), IL-6 (p = 0.0120), IL-18 (p = 0.0047), and CSF1 (p = 0.0002). Additional proteins significantly correlated include IL-10RB (p = 0.0001), CCL7 (p = 0.0010), CXCL10 (p = 0.0044), TNF (p = 0.0054), CCL2 (p = 0.0083), and CXCL8 (p = 0.0290) (Supplementary Fig. 3). Of note, the expression of CD40 on antigen-presenting cells leads to strong memory T cell responses. Overall, these findings supplement our cellular data, with proinflammatory cytokines and chemokines positively correlating with stronger neutralizing antibody responses, underscoring the integral role of early inflammatory signals in shaping memory immunity.

Fig. 4. Circulating IgG neutralizing antibodies to SARS-CoV-2 variants.

a Neutralizing antibodies to three different SARS-CoV-2 strains, Wuhan, Alpha, and Beta from hospitalization to one year follow up. b CXCL5, TNFSF11, IFNγ, IL-6, IL-18, CD40, PD-L1, and CSF1 normalized protein expression (NPX) levels correlated with Neutralizing IgG Antibodies to SARS-CoV-2 Wuhan. R and p values are reported for each variant and correlation, calculated using one-way ANOVA and Spearman Correlation. p < 0.05*, p < 0.01**, p < 0.001***, p < 0.0001****.

We then determined the decay of neutralizing antibody response over 1 year post hospitalization. The decay rate was determined for all infected participants using a negative exponential decay model at all time points for infected participants and time points before vaccination for participants with hybrid immunity. Fig. 5a shows the decay rate for 24 participants, where some have durable responses (minimum decay after 1 year) while others have shorter durable responses (decay occurs after 3 months). We then measured the associations of early immune markers (at hospital admission) with neutralizing antibody decay (durability) using Olink, Luminex, and flow cytometry to identify markers and innate immune subsets. In Fig. 5b, we show immune markers positively and negatively associated with a durable antibody response. We found that specific proinflammatory cytokines and chemokines are positively associated with a durable neutralizing antibody response. These include CCL8 (p = 0.0113), MMP10 (p = 0.0297), TNFSF10 (p = 0.0204), MIG (p = 0.0221). On the other hand, some proteins were negatively associated with neutralizing antibody durability. These include CCL23 (p = 0.025), HGF (p = 0.0135). Importantly, we found that the frequency of CD14dim and plasmacytoid dendritic cells (pDCs), both known producers of type I interferons, to be negatively associated (p = 0.011 and p = 0.04, respectively) with a durable neutralizing antibody response.

Fig. 5. Decay of neutralizing antibody response in infected participants.

a Decay rate for 24 participants using negative exponential decay model. These graphs show infected participants at all timepoints before vaccination and hybrid immunity. b Associations of early immune markers positively and negatively associated with neutralizing antibody durability. Antibody half-life correlated with multiple Olink proteins, Luminex proteins, and flow cytometry frequencies R and p values are reported for each comparison, as measured Spearman Correlation. p < 0.05*, p < 0.01**, p < 0.001***, p < 0.0001****.

Identification of early immune markers associated with affinity of anti-RBD and anti-S1 antibodies

We measured the antibody affinity using a rapid surface plasmon resonance (SPR) screening assay for initial rates of binding. Using SPR kinetics analysis of IgG binding over time allowed us to measure quantitative aspects of antibody binding, such as equilibrium association constants (affinity, KA). KA values measure the amount of antibody-antigen complex formed at equilibrium, independent of concentration (Fig. 6a). We wanted to assess the relationships between Olink proteins and antibody quality three months after hospitalization in infected participants (Fig. 6b). We found several immune markers that are either positively or negatively associated with RBD and S1 IgG KA values (Supplementary Fig. 4). We reported seventeen significant correlations with various Olink proteins. Once again, we saw correlations with inflammatory cytokines, including CXCL8 (p = 0.0296), IL-7 (p = 0.018), IL-17A (p = 0.0155), PD-L1 (p = 0.0416), CXCL10 (p = 0.0366), and CD40 (p = 0.0049; Fig. 6c). There was one negatively correlated protein, TNFSF11 (p = 0.0482). TNFSF11 is a protein commonly detected in the lungs, thymus, and lymph nodes. TNFSF11 is shown to be a dendritic cell survival factor and is involved in T-cell immunology²¹. It is also thought to be involved in dendritic cell maturation. These correlations informed us that higher amounts of inflammatory markers during early infection can influence the quality of the antibody response post infection. Overall, this shows us that cytokines and chemokines involved in inflammation and regulation have a role in influencing antibody quality and quantity following infection.

Fig. 6. Normalized protein expression (NPX) of early proteins correlates with longitudinal IgG antibody quality.

a RBD and S1 IgG Antibody Binding Association three months after hospitalization. b Study timeline highlighting which time points are compared in these analyses. This figure compares Olink protein from Day 0 with IgG antibody KA three months post hospitalization. c Normalized protein expression (NPX) levels in serum correlated with IgG and S1 IgG KA three months post hospitalization. Longitudinal plasma antibody affinity (KA) was determined using surface plasmon resonance at multiple timepoints for RBD IgG and Spike IgG. R and p values are reported for each comparison, as measured Spearman Correlation. p < 0.05*, p < 0.01**, p < 0.001***, p < 0.0001****.

Innate immune subsets associations of acute monocyte frequencies and longitudinal SARS-CoV-2 antibody quantity and quality

Next, we investigated whether early innate immune cells are associated with humoral memory responses. Using flow cytometry, we analyzed CD3⁻CD19⁻HLA-DR⁺ innate immune cells, focusing on three dendritic cell subsets: pDCs (CD11c-CD1c-CD303 + CD141−), conventional dendritic cells (cDCs, CD11c-CD1c-CD303-CD141+), and myeloid dendritic cells (mDCs, CD11c + CD1c + CD303-CD141−). Full gating scheme of cellular subsets is depicted in Supplementary Fig. 5. Both mDCs (p = 0.0213) and pDCs (p = 0.0473) were positively correlated with RBD-specific IgG titers (Fig. 7a, b). pDCs are critical in viral infections, capable of rapidly producing large quantities of type I and type II interferons, which are essential for antiviral defense. mDCs, the largest population of dendritic cells in the blood, are key to initiating adaptive immunity by activating and priming T cells²². Although dendritic cells do not directly regulate antibody responses, their role as antigen-presenting cells facilitates interactions with T helper cells, particularly Tfh cells²². This interaction is essential for enhancing antibody quality, quantity, and durability. Overall, our findings suggest that early innate immune cell activity, particularly through dendritic cell interactions, can influence the affinity of RBD-specific IgG, underscoring their role in shaping long-term humoral immunity.

Fig. 7. Acute dendritic cell subsets correlated with RBD IgG Abundance.

a Study timeline highlighting which time points were compared in this analysis. This figure compares acute dendritic cells and antibodies 3 months post hospitalization. b Acute myeloid dendritic cells correlated with RBD IgG Abundance. Acute plasmacytoid dendritic cells correlated with RBD IgG Abundance. R and p values are reported for each comparison, as measured Spearman Correlation. p < 0.05*, p < 0.01**, p < 0.001***, p < 0.0001****.

Associations of nasal viral load and SARS-CoV-2 memory immunity

We next assessed the association of airway SARS-CoV-2 viral load and our memory immune responses one year post hospitalization (Supplementary Fig. 6). Nasal viral loads were taken around hospital admission and determined using qPCR. For this assay, a lower Ct value indicates higher viremia. All cellular and humoral responses were not impacted by early viral loads. We assessed whether nasal viral loads were associated with the frequency of innate immune subsets at admission. Similarly, we did not see any significant correlations between these two groups (Supplementary Fig. 6). Overall, viremia in early infection did not significantly influence memory B cells, antibody responses, or associated with early innate immune subsets.

Gene expression

We also performed gene expression analysis on PBMC at the admission visit and determined the association with multiple memory response outcomes at later visits. We have identified multiple immune pathways, including several inflammatory-related pathways such as complement (with genes CR2 coding for the receptor of the complement component C3 and FCN1 encoding a protein part of the lectin pathway of the complement system), inflammatory response (with genes CCL2 coding for the monocyte/DCs chemoattractant MCP1 and the gene NFKBIA coding for the inhibitor of NFkB) and TNF-signaling via NFkB (with several TNFa signaling target genes such as ATF3 and TNFAIP3, as well as upstream activators of NFkB such as TLR2 and PLAUR) to be significantly induced in early immune events and to be positively correlated with RBD-specific memory B cells, AIM + T cells and t_1/2 of neutralizing antibodies (Fig. 8). This corroborates our findings that heightened inflammatory markers during the acute phase of COVID-19 disease are associated with improved cellular and humoral response 1 year after hospitalization

Fig. 8. Heatmap showing gene expression at Visit 1 positively correlated with a humoral response measure at Visit 10.

Gene set enrichment analysis was performed to identify hallmark pathways (left annotation) enriched among genes associated with at least one humoral response measurement. Leading edge genes of those hallmark genesets are presented in the heatmap (rows). Gene expression was scaled to a mean of 0 and standard deviation of 1 (z-score) across Visit 1 samples (columns). Humoral response measurements are presented as the top annotation on the heatmap.

We generated a riverplot to illustrate all early immune markers associated with multiple memory responses, including antibody quality and frequency of memory T and memory B cell responses and accentuate those that are shared with multiple cellular and humoral responses. We identified several early markers of humoral response to SARS-CoV-2 infection, including the proinflammatory cytokine IP-10 (CXCL10) expression at Visit 1 positively correlated with memory B cell and Tfh frequencies, IgG against RBD and neutralizing antibody levels 1 year after infection (Fig. 9). TNF and TNFSF14 protein levels, as well as TNF signaling identified by RNA-seq were respectively associated with neutralizing antibody titers at Visit 7 and memory B cells and AIM + T cells at Visit 10. Both IP-10 and TNF signaling suggest that proinflammatory signaling at Visit 1 is associated with heightened cellular/humoral responses at Visit 7/10, suggesting that type I interferons may be detrimental to cellular/humoral responses at Visit 7/10 (Figs. 9 and 10). This contrasts with type II interferons, including interferon-gamma, positively correlated with Visit 7 IgG against RBD and neutralizing antibodies (Figs. 9 and 10). Genes downstream of interferon-gamma, including HLA-A and TAP1, also implicated in antigen processing, were positively correlated with Visit 7 neutralizing antibodies (Fig. 8), suggesting a direct regulation of antigen processing by interferon-gamma following infection with SARS-CoV-2 leading to higher magnitude of neutralizing antibodies detected even 1 year following infection.

Fig. 9. Riverplot showing the correlation between Visit 1 analytes (measured within 4 days of hospitalization) and Visit 10 markers of humoral response to SARS-CoV-2 infection (measured 1 year after hospitalization).

Each box corresponds to an analyte (left) or humoral response measure (right); the ribbon connecting an analyte and a measurement is colored based on their Spearman correlation (rho); the heights of the boxes are proportional to the number of significant correlations. All correlations presented in the figure were greater than 0 and passed a t-test nominal p-value cutoff of 0.05.

Fig. 10. Riverplot showing the correlation between Visit 1 analytes (measured within 4 days of hospitalization) and Visit 10 markers of humoral response to SARS-CoV-2 infection (measured 1 year after hospitalization).

Each box corresponds to an analyte (left) or humoral response measure (right); the ribbon connecting an analyte and a measurement is colored based on their Spearman correlation (rho); the heights of the boxes are proportional to the number of significant correlations. All correlations presented in the figure were less than 0 and passed a t-test nominal p-value cutoff of 0.05.

Discussion

SARS-CoV-2 continues to be a global burden, and individuals are still being infected five years later. With the emergence of multiple variants of concern, several people in the United States continue to be hospitalized with COVID-19 infection. Understanding the longevity of cellular and humoral to SARS-CoV-2 remains a major priority, especially with the emergence of hybrid infection. Our lab sought to understand early immune events that influence memory immune responses following COVID-19 infection. Overall, our data shows that individuals with specific inflammatory cytokine expression develop higher frequencies of antigen-specific B and T cells, and increased neutralizing antibodies compared to individuals without a robust cytokine profile.

In this study of 46 adults hospitalized with COVID-19 during the first year of the pandemic, we identified multiple specific markers and early immune cell frequencies associated with cellular and humoral memory responses 1 year after hospitalization. Notably, inflammatory markers with positive correlations suggest a significant role in enhancing memory cell responses. These inflammatory proteins are linked to improved viral control and a more robust and sustained memory immune response. While we did not see a significant difference between participants with hybrid immunity and infection alone for memory B cell frequencies, we did see a trend between time from vaccination and memory cell frequencies. Looking at correlations with memory B cells, all significantly positively correlated proteins play crucial roles in immune responses or viral infection mechanisms. Interestingly, FLT3 ligand negatively correlates with antigen-specific memory B cell responses, suggesting higher FLT3LG levels are associated with reduced memory B cell frequencies. FLT3LG is essential for the development of pDCs and cDCs, with deficiencies in FLT3LG leading to reduced dendritic cell levels²³. Additionally, FLT3LG is essential for increasing DC numbers in response to a loss of regulatory T cells^24,25. In turn, after dendritic cells increase in response to FLT3LG, T regulatory cells increase²⁶. This feedback loop between FLT3LG and Treg cells could be responsible for decreased frequencies of memory B cells and memory T cells in the long term and could be a possible future investigation.

CXCL13 is another early immune protein significantly correlated with memory T cells. CXCL13, in particular, has been implicated in Tfh cell activity during SARS-CoV-2 infection^15,27. Tfh cells enhance B cell responses and facilitate somatic hypermutation to generate high affinity antibodies, and also are important in establishing immunological memory²⁷. CXCL13 levels in plasma have been shown to correlate with GC activity in the draining lymph nodes and correlate with the magnitude of antibody responses^28,29. This protein is an important mediator in establishing memory Tfh cell responses.

IL-33 is another protein positively correlated with memory T cell responses. Interestingly, IL-33 has been shown to increase memory responses in both TH1 and Th2 responses^19,28,29. IL-33 contributes to the differentiation of tissue resident memory T cells in asthma, and in response to vaccine-induced memory^28,29. IL-33 used as an adjuvant also boosts vaccine-induced memory T cell responses and has been shown to protect against subsequent heterologous viral challenges²⁹. IL-33 secretion from endothelial cells in the lungs directly influences Th1 CD4 + T cells to secrete IFN-γ, which is positively correlated with the magnitude and quality of SARS-CoV-2 IgG antibodies in our work and is an important protein which seems to be influencing the memory immunity in our participants following SARS-CoV-2 infection.

Our data also shows that soluble TRAIL is negatively correlated with Neutralizing IgG Antibodies 3 months after hospitalization, suggesting that higher levels of this protein during acute COVID-19 infection weaken the neutralizing capabilities of antibodies (Fig. 10). Soluble TRAIL rapidly induced apoptosis through caspase and MAPK activation^30,31. Dendritic and NK cells have been shown to mediate apoptosis in multiple cell types^32,33. TRAIL secretion and surface protein are upregulated on monocytes and macrophages after stimulation with type I interferons, supporting our observations that Type I interferons negatively regulate memory immune responses following SARS-CoV-2 infection³⁴. Additionally, TRAIL expression has a role in CD4 + T cell regulation and generation of CD8+ memory T cells, contributing to the establishment of memory immune responses³⁵.

When investigating correlations between antibody responses and early immune proteins, we observed positive associations of PD-L1, IL-6, IL-18, CASP8, and CD40 with multiple antibody measures, including neutralizing antibody responses and overall antibody quality and quantity. Neutralizing antibodies targeting Wuhan and Alpha SARS-CoV-2 were correlated with CSF1, TNFRSF11B, PD-L1, and CD40. Dysregulation of the PD-1/PD-L1 axis has been associated with COVID-19 severity and epithelial cells have been shown to upregulate PD-L1 in response to SARS-CoV-2 infection³⁶. PD-L1 binding to PD-1 on T cells and drives a regulatory phenotype, and the PD-1 pathway is needed for establishing T cell memory^37,38. PD-L1 signaling on memory T cells has been implicated in inflammatory responses and induces conversion into suppressive T cells³⁸.

Additionally, the proinflammatory cytokines IL-6, IL-18, and CASP8, which are well-documented components of the COVID-19 cytokine storm, were positively correlated with stronger antibody responses^9,10. These cytokines are associated with severe disease and adverse outcomes. Together, these findings suggest that early COVID-19 infection elicits a robust innate immune response characterized by monocyte and macrophage activation and the release of inflammatory cytokines and chemokines. While excessive upregulation may lead to cytokine storm and poor outcomes, participants who expressed specific cytokines and chemokines have benefited from this inflammatory milieu, as it likely contributed to the development of stronger memory immune responses. To our knowledge, this study is among the first to directly link cytokine storm-driven inflammation to enhanced memory immunity in the same participants one year after hospitalization.

The role of interferons in shaping the durability of adaptive immune responses provides an important framework for interpreting our findings. Type I interferon (IFN-I) responses are crucial for early antiviral defense, but their effects can vary depending on the timing and magnitude^39,40. During acute infection, an exaggerated IFN-I response is associated with cytotoxic inflammation and impaired immune memory formation. In contrast, mRNA vaccination elicits strong adaptive responses in the absence of this inflammatory IFN signature^39–41. Recent work also highlights that IFN-I signaling can restrict mRNA vaccine efficacy by suppressing antigen uptake in cDCs. These findings highlight the importance of regulation in the production of IFN-I and associated pathway proteins⁴². In contrast, Type II interferon (IFN-γ) is a correlate of durable immunity. IFN-γ production following mRNA vaccination correlates with the development of long-term T-cell memory and the durability of neutralizing antibodies^43,44. In line with our results, a balanced interferon response during early SARS-CoV-2 infection appears to support the development of durable immune memory, likely by enhancing antigen presentation and minimizing the inflammation linked to excessive IFN-I signaling. However, more research is needed to elucidate these mechanisms.

Innate immune cell responses are also critical during early SARS-CoV-2 infection. Monocytes, macrophages, and dendritic cell subsets are key players in maintaining inflammation and initiating adaptive immune responses through antigen presentation. Our data revealed that two subsets of dendritic cells—mDCs and pDCs—correlated with higher levels of RBD-specific IgG antibodies three months after infection. mDCs, often referred to as inflammatory dendritic cells, are instrumental in T cell priming during viral infections^22,45. pDCs, on the other hand, are central to interferon production, which activates interferon-stimulated genes crucial for controlling viral replication in the early stages of infection. cDCs also play a pivotal role in antigen presentation and cross-priming^22,45. Each of these dendritic cell subtypes has been implicated in determining COVID-19 infection severity and patient outcomes by shaping cytokine and chemokine responses during early infection. Our findings indicate that higher frequencies of these dendritic cell subsets support a more robust adaptive immune response, resulting in higher RBD-specific IgG titers.

Strengths of our study include a large cohort of COVID-19 positive participants, both vaccinated and unvaccinated. This study also spans longitudinally over 12 months and contains a mix of both male and female participants. We also include the antibody durability for infection. This study also has limitations, including the median age of participants in their 60 s. In addition, these samples were collected in one geographical location. These participants were hospitalized and did not contain participants with severe COVID-19 infection nor a control group for comparison. In summary, our study demonstrates that elevated systemic inflammatory markers are associated with stronger cellular and humoral memory immune responses. These findings align with published data on the early release of inflammatory proteins during cytokine storms triggered by SARS-CoV-2 infection. This dataset is unique in directly comparing early immune responses and their long-term effects on memory durability within the same cohort. The results suggest that individuals with heightened inflammatory cytokine expression during early infection develop more robust memory immunity, offering better protection against reinfection with SARS-CoV-2 variants of concern and mitigating the severity of secondary infections.

Methods

Study design and setting

NIAID staff conferred with the Department of Health and Human Services Office for Human Research Protections (OHRP) regarding potential applicability of the public health surveillance exception [45CFR46.102(l)(2)] to the IMPACC study protocol. OHRP concurred that the study satisfied criteria for the public health surveillance exclusion and the IMPACC study team sent the study protocol and participant information sheet for review and assessment to institutional review boards (IRBs) at participating institutions. Twelve sites elected to conduct the study as public health surveillance, while 3 sites with prior IRB-approved biobanking protocols elected to integrate and conduct IMPACC under their institutional protocols. Participants enrolled under the public health surveillance exclusion were provided information sheets describing the study, samples to be collected, and plans for data de-identification and use. Those who requested not to participate after reviewing the information sheet were not enrolled. In addition, participants did not receive compensation for study participation while inpatient, and subsequently were offered compensation as outpatients.

Study participants

Several site-dependent methods were used to identify eligible participants, including (but not limited to) acquisition of hospital-based service lists and notification of the research team of any positive SARS-CoV-2 PCR testing. Drexel University was chosen as one of the fifteen sites across the country to participate in a public health surveillance study aimed at responding to the SARS-CoV-2 pandemic early in 2020. The Institutional Review Board of Drexel University approved (IRB Protocol # 2004007753) this study, aimed at identifying longitudinal immune responses in hospitalized patients with COVID-19 (IMPACC). Given its provision as a public health surveillance study, it was approved with no need for continuing review. All human subjects were assessed for capacity using a standardized and approved assessment. Subjects deemed to have capacity voluntarily gave informed consent prior to being enrolled in the study. When subjects were deemed to lack capacity, they were enrolled only if an appropriate surrogate was identified and gave informed consent. Individuals or their surrogates did not receive compensation for their participation in the study, although travel reimbursement was provided for outpatient visits. Study inclusion criteria included subjects with a diagnosis of COVID-19 and positive PCR-based testing for SARS-CoV-2 and ongoing symptoms and/or clinical findings consistent with acute COVID-19, who were over the age of 18 years, regardless of disease severity, race, ethnicity, gender, pregnancy, or nursing status. Study exclusion criteria included clinically significant anemia or short life expectancy. Blood and nasal swabs collected from patients with acute, symptomatic COVID-19 at various stages of illness at time of enrollment (within 24 hours of hospitalization and then again 3, 7, 14, 21, and 28 days as inpatients. They were then brought back for blood and nasal swabs collection as outpatients 3, 6, 9, and 12 months post-hospitalization. If they were discharged prior to 28 days, subjects were brought in as outpatients for the 28 days timepoint. All subjects enrolled were hospitalized within the Tower Health system at one of three local hospitals, including Pottstown Hospital, Chestnut Hill Hospital, and Reading University Hospital. All subjects were assessed for positive SARS-CoV-2 PCR-based testing, clinical disease, and/or chest imaging consistent with COVID-19 prior to enrollment. All of these donors were identified as having COVID-19 based on a positive PCR-based test for SARS-CoV-2 (GenMark ePlex SARS-CoV-2 Test for SARS-CoV-2, Roche cobas SARS-CoV-2, Abbott m2000 COVID-19, GenMark ePlex SARS-CoV-2, or Luminex ARIES SARS-CoV-2 assay) or a rapid point of care molecular test for SARS-CoV-2 (Abbott ID NOW COVID-19) performed at the Reading Hospital\clinical laboratory. Donors were also included if they had positive testing for SARS-CoV-2 at an outside laboratory based on clinical documentation, self-report, or confirmation of positive testing from an outside laboratory. Whole blood from all hospitalized donors was collected in EDTA tubes and stored at room temperature briefly prior to processing for PBMC and plasma isolation. Whole blood was also collected in additional serum separator tubes for serum isolation. The maximum blood volume collected (for any purpose) within any 8-week period was set at 500 mL per the IRB-approved protocols. Samples were de-identified prior to analysis. Other efforts to maintain the confidentiality of participants included referring to specimens and other records via an assigned, coded identification number. All ethical regulations relevant to human research participants were followed.

PBMC and plasma isolation and handling

Whole blood was collected in EDTA tubes (BD Vacutainer tubes, Franklin Lakes, NJ, USA) and stored at room temperature prior to processing for PBMC isolation. Whole blood was processed as previously described (Grifoni et al., 2020a). In brief, PBMC were isolated by density-gradient sedimentation, as described below. Sterile technique was used to transfer the blood to conical tubes. Whole blood was diluted 1:2 in room temperature RPMI (Corning, Manassas, VA, USA), then layered over an appropriate volume of room temperature Lymphoprep (Stemcell technologies), then centrifuged for 25 min at 600 × g at room temperature with no brake to separate the cellular fraction and plasma. The plasma was then carefully removed, aliquoted, and stored at 80 °C. The PBMC buffy coat was then collected and washed with RPMI. An aliquot of cells was counted by Countess II (Thermofisher). Isolated PBMC were then cryopreserved in cell recovery medium containing 10% DMSO (GIBCO) and 90% heat inactivated fetal bovine serum (FBS; Hyclone Laboratories, Logan, UT), placed in a Mr. Frosty freezing container (Thermo Scientific, USA) and frozen overnight at 80 °C, and then transferred to liquid nitrogen for further storage until used in the assays. All blood samples and blood products were handled in a BSL-2 laboratory with the use of appropriate personal protective equipment and safety precautions. Where appropriate, plasma was heat inactivated for 30 min at 54–56 °C prior to use.

Flow cytometry

For all flow cytometry assays of PBMCs, cryopreserved cells were thawed by diluting them in 10 mL pre-warmed complete RPMI containing 5% human AB serum (Gemini Bioproducts) in the presence of benzonase (20 μl/10 mL) and spun at 1200 rpm for 10 min. Supernatants were carefully removed by pipetting and cells were resuspended in warm medium, counted, and apportioned for assays.

Flow cytometry of monocyte and DC subsets

The stimulated PBMCs from adult donors were prepared and incubated with fluorochrome-conjugated antibodies for flow cytometry. Briefly, cells were washed twice with fluorescence-activated cell sorting (FACS) buffer (PBS containing 2% FBS), surface-stained using antibodies for 30 min in 100 μL of FACS buffer, and then fixed using 2% PFA for 15 min at 37 °C. All antibody dilutions were used at 1:1000 unless otherwise stated. The following fluorochrome-conjugated antihuman antibodies were used: CD3 (Clone: HIT3α, Cat: 300324), CD19 (Clone: HIB19, Cat: 302216), CD11c (Clone: 3.9, Cat: 301606), CD16 (Clone: 3G8, Cat: 612786), CD14 (Clone: HCD14, Cat: 325634), HLA-DR (Clone: LN3, Cat: 327014), CD1c (Clone: L161, Cat: 331538), CD303 (Clone: 201A, Cat: 354238), were all from BioLegend. CD141(Clone: AD5-14H12, Cat: 130-113-317) was from Miltenyi Biotech. LIVE/DEAD Fixable Dead Cell Stain (Life Technologies, Cat: L34957) was used to gate on live cells. Samples were acquired on a BD^TM LTR Fortessa (BD Biosciences), and analysis was conducted using FlowJo V10.10 Software.

Activation induced cell marker (AIM) assay

Assays were conducted as previously described (Grifoni et al., 2020a; Morou et al., 2019; Reiss et al., 2017). Cells were cultured for 24 h in the presence of SARS-CoV-2-specific spike peptide pool [2 μg/mL], in 96-well U-bottom plates at 1 × 10⁶ PBMC per well in complete RPMI containing 5% Human AB Serum (Gemini Bioproducts). A stimulation with an equimolar amount of DMSO was performed as a negative control (NC). Staphylococcal enterotoxin B (SEB, 1 μg/mL) was included as a positive control. In each experiment, PBMCs from an unexposed individual were included to ensure consistent sensitivity and specificity of the assay. Live cells were stained using the LIVE/DEAD Fixable Aqua Stain Kit (Thermo Fisher Scientific, catalog no. L34962) and incubated with BD human FC block (BD Biosciences, San Diego, CA) for 15 min at room temperature (RT). Surface Stain was resuspended in PBS and incubated at 4 °C for 10 min 1 × 10⁶ PBMCs were resuspended in 100 μL PBS with 2% FBS (FACS buffer). Antibody mix containing the rest of the surface antibodies was then added directly to cells and incubated for 30 min at 4 °C in the dark. Following surface staining, cells were washed once with FACS buffer and resuspended in 100 μL BD fixation/permeabilization solution (BD Biosciences, San Diego, CA) and incubated at 4 °C for 45 min in the dark. Cells were then washed twice with permeabilization buffer and stained with intracellular and intranuclear antibodies for 25 min at 4 °C in the dark. After staining, cells were washed once with permeabilization buffer and resuspended in FACS buffer. All antibody dilutions were used at 1:1000 unless otherwise stated. The following fluorochrome-conjugated antihuman antibodies were used: CD3 (Clone: HIT3α, Cat: 300324), CD4 (Clone: RPA-T4, Cat: 300518), PD-1 (Clone: EH12.2H7, Cat: 329904), CD8 (Clone: HIT8a, Cat: 300922), CXCR3 (Clone: G025H7, Cat: 353716), 4-1BB (Clone: 4B4-1, Cat: 309828), CD69 (Clone: FN50, Cat: 310932), CD25 (Clone: BC96, Cat: 302608), CD154 (Clone: 24–31, Cat: 310832), CD25 (Clone: BC96, Cat: 302608), CD40L (Clone: 24–31, Cat: 310832), IL-2 (Clone: MQ1-17H12, Cat: 500332), TNFα (Clone: Mab11, Cat: 502940), IFNγ (Clone: 48.B3, Cat: 502509) were all from BioLegend. CXCR5 (Clone: RF8B2, Cat: 558113) and CCR6 (Clone: 11A9, Cat: 562724) were from BD Biosciences. LIVE/DEAD Fixable Dead Cell Stain (Life Technologies, Cat: L34957) was used to gate on live cells. Samples were acquired on a BD^TM LTR Fortessa (BD Biosciences), and analysis was conducted using FlowJo V10.10 Software.

RBD memory B cell assay

For flow cytometry analysis of SARS-CoV-2-specific B cells, biotinylated protein antigens multimerized on fluorescently labeled streptavidin were used as probes to detect antigen-specific B cells. Biotinylated RBD proteins were purchased from Sino Biologic (catalog no. 40592-V08H-B). Biotinylated RBD was incubated with 1 × 10⁶ PBMC per well in complete RPMI for 30 min in U-bottom 96-well plates and then stained with 50 μL of streptavidin APC and PE at 4 °C for 1 h, followed by staining for surface markers in Brilliant Stain Buffer at 4 °C for 30 min. All antibody dilutions were used at 1:1000 unless otherwise stated. The following fluorochrome-conjugated antihuman antibodies were used: CD3 (Clone: HIT3α, Cat: 300324), CD19 (Clone: HIb19, Cat: 302252), CD10 (Clone: H110a, Cat: 312212), CD38 (Clone: HIT2, Cat: 303519), CD27 (Clone: 0323, Cat: 302828), CD20 (Clone: 2H7, Cat: 302336) were from Biolegend. IgD (Cat: 555778), CD21 (Clone: B-ly4, Cat: 561381), and 4-1BBL (Clone: C65-485, Cat: 744397) were all from BD Biosciences. LIVE/DEAD Fixable Dead Cell Stain (Life Technologies, Cat: L34957) was used to gate on live cells. Samples were acquired on a BD^TM LTR Fortessa (BD Biosciences), and analysis was conducted using FlowJo V10.10 Software. In each experiment, PBMCs were stained with just streptavidin (no biotinylated RBD protein) as a decoy control.

Cytokine and chemokine analysis

Supernatants collected from PBMCs during stimulation were analyzed for chemokine/cytokine levels using the human immune monitoring 65-Plex ProcartaPlex™ Panel (Invitrogen™). This kit was used to determine the levels of 65 cytokines, chemokines, growth factors, and soluble receptors produced by MDDCs, 24 h after stimulation, and PBMCs 6 and 24 h after stimulation. The following human chemokine/cytokine premixed panel was used according to the manufacturer’s protocol: G-CSF (CSF-3), GM-CSF, IFN alpha, IFN-g, IL-1a, IL-1b, IL-2, IL-3, IL-4, IL-5, IL-6, IL-7, IL-8 (CXCL8), IL-9, IL-10, IL-12p70, IL-13, IL-15, IL-16, IL-17A (CTLA-8), IL-18, IL-20, IL-21, IL-22, IL-23, IL-27, IL-31, LIF, M-CSF, MIF, TNF-a, TNF-b, TSLP, BLC (CXCL13), ENA-78 (CXCL5), Eotaxin (CCL11), Eotaxin-2 (CCL24), Eotaxin-3 (CCL26), Fractalkine (CX3CL1), Gro-alpha (CXCL1), IP-10 (CXCL10), I-TAC (CXCL11), MCP-1 (CCL2), MCP-2 (CCL8), MCP-3 (CCL7), MDC (CCL22), MIG (CXCL9), MIP-1a (CCL3), MIP-1b (CCL4), MIP-3a (CCL20), SDF-1a (CXCL12), FGF-2, HGF, MMP-1, NGF-b, SCF, VEGF-A, APRIL, BAFF, CD30, CD40L (CD154), IL-2R (CD25), TNF-RII, TRAIL (CD253), TWEAK. Data was acquired on a Luminex™ FLEXMAP 3D™ System using bead regions defined in the protocol and analyzed using Belysa Curve Fitting Software (Sigma Aldrich). Standard curves were generated, and sample concentrations were calculated in pg/mL.

Variant antibody detection by Luminex

Tetracore^® FlexImmArray™ 36-Plex SARS-CoV-2 Human IgG Antibody Test (RUO), Catalog # TC-RD-036-096, was used to assess the presence of SARS-CoV-2 human IgG antibodies to 32 different coronavirus antigens and variants among infected, recovered, asymptomatic, and vaccinated individuals. The following 36-Plex assay panel was used according to the manufacturers protocol: SARS-CoV-2 Receptor Binding Domain, Nucleocapsid Protein-SARS-CoV-2, SARS-CoV-2 Hybrid NP-RBD, Spike Protein SARS-CoV-2, Spike Protein S1-SARS-CoV-2, Spike Protein S1-SARS-CoV-2, SARS-CoV-2 Receptor Binding Domain, SARS-CoV-2 Receptor Binding Domain, Variant SARS-CoV-2 RBD, Variant (BRZ, SA, UK) SARS-CoV-2 RBD, Variant (SA) SARS-CoV-2 RBD, Variant (BRZ, SA) SARS-CoV-2 RBD, Variant (India) SARS-CoV-2 RBD, Variant (BRZ) SARS-CoV-2 RBD, Spike Protein S1-SARS-CoV-2 (D614G), Spike Protein S1-SARS-CoV-2, Variant (SA) SARS-CoV-2 Spike Protein S1, Variant (UK) SARS-CoV-2 Spike Protein S1, Variant (UK) SARS-CoV-2 Spike Protein S1, Spike Protein S1-SARS-CoV, Spike Protein S1-MERS-CoV, Spike Protein S1-HCoV-229E, Spike Protein S1-HCoV-NL63, Spike Protein S1-HCoV-OC43, Spike Protein S1-HCoV-HKU1, Spike Protein S1-HCoV-HKU1, Nucleocapsid Protein-SARS-CoV-2, Nucleocapsid Protein-SARS-CoV, Nucleocapsid Protein-MERS-CoV, Nucleocapsid Protein-HCoV-229E, Nucleocapsid Protein- HCoV-NL63, Nucleocapsid Protein- HCoV-HKU1, Nucleocapsid Protein- HCoV-OC43.

To test human serum or plasma samples, each sample was centrifuged at room temperature for 2 minutes at 10,000 rpm using the tabletop centrifuge. The sera or plasma samples were serially diluted in 1× Sample Dilution Buffer (SDB) in a non-binding 96-well sample dilution plates to get a final dilution of 1 in 400. The centrifuged samples were diluted first at 1 in 20 by adding 10 µL of centrifuged sample to 190 µL of SDB and transferring 10 µL of 1:20 diluted sample to a well containing 190 µL SDB, to give 1:400 dilution.

NC, Low positive control (LPC), and High positive control (HPC) samples were included in the kit for use in the testing. NC, HPC, and LPC are tested at least in duplicate wells in each plate. Three-fold serial dilution of the HPC is made and tested for measuring a range of positive samples in each plate. HPC was tested at 1, 3, 9, 27, 81, and 243-fold dilutions to construct a standard curve. LPC is used as a normalizer and the average of two replicate MFI values of LPC was used for calculating the relative fluorescence intensity (RFI).

25 µL of 36-Plex magnetic microsphere mix was added to each well of the 96-well clear-bottom assay plate from the kit. 25 µL each of various dilutions of HPC or ready to use LPC, or NC or 1:400 dilution of serum or plasma samples was added to each designated well. The mixture of sample and the antigen coated microsphere were incubated at room-temperature for 30 min while shaking at 800 rpm. This allows for antibodies from the controls and samples to bind to the antigens on the microspheres. After the first incubation, the assay plate was washed using automated magnetic plate washer (BioTek 405 TS Washer) to remove unbound antibodies by washing four times with 200 µL assay wash buffer/well during each wash step. After the first wash, 25 µL of Fluorescent Reporter (Anti-human IgG-phycoerythrin) conjugate from the kit was added to each well and incubated for 30 min at room-temperature with shaking at 800 rpm. After the second incubation, the assay plate was washed as described for the first wash. 150 µL of wash buffer was added to each well and the contents of the plate were re-suspended by shaking at 800 rpm for 5 min at room-temperature before analysis. The assay plate was analyzed using Luminex® FLEXMAP 3D^® instrument with xPONENT Software Ver. 4.3.309.1 was used to measure fluorescence on the microspheres and reported as median fluorescent intensity (MFI) for each bead set.

Result for each well was reported as RFI for each coronavirus antigen was calculated by normalizing the MFI value of each sample with respect to the average LPC MFI value of the same antigen as shown below.

$\mathrm{RFI}\mathrm{for}\mathrm{each}\mathrm{antigen}=\frac{SampleMFIforeachcoronavirusantigen}{AverageoftheduplicatevaluesoftheLowpositivecontrolMFIforeachcoronavirusantigen}$

Surface plasmon resonance (SPR) assays

SPR experiments were performed on a Biacore S200 biosensor (Global Cytiva Lifesciences) at 25 °C using PBS-P (10 mM Phosphate, 150 mM NaCl, pH 7.4, 0.05% P-20) as the running buffer by Gabriella Canziani in Dr. Irwin Chaiken’s lab at Drexel University College of Medicine. A CM5 sensor chip (Cytiva, Marlborough, MA) was docked and derivatized by amine coupling with RBD using freshly prepared 100 mM NHS (N-hydroxysuccinamide) and 400 mM EDC (1-ethyl-3-(3-(dimethylamino) propyl) carbodiimide) reagents (Cytiva) mixed 1:1. Flow cell 1 was activated and blocked and remained as the control of binding to RBD (Purified Wuhan recombinant Spike RBD, Sino Biological, BDA, Beijing, RBD-His 40592-VNAH, FC2). RBD directly coupled to the sensor was validated using ACE2 (10108-H08H) and VHH-72-huFc, the latter a generous gift from Integral Molecular (Philadelphia, PA). The anti-RBD titration standard was an RBD-specific neutralizing mouse monoclonal antibody (Mab) (40592-MM57). Serum from post-immunization bleeds were diluted in sample buffer with 20 mg/mL CM-Dextran saline, or NSB reducer (Cytiva) to minimize non-specific binding. The final serum sample dilutions were 1/15 and 1/20 (1.3-fold apart) in naïve serum. Sample matching naïve sera mixed with CM-Dextran in equal proportion, were used as NCs, while Mab standards between 1.2 and 250 nM were also spiked in naïve serum mixed CM-Dextran were prepare. A surface density of 50 RU/KDa RBD was experimentally found to offer partial mass transport limited (MTL) binding of serum-spiked antibody standards injected over control and RBD flow cells at 5 μL/min. Under MTL, the antibody binding rates are dependent on the concentration of RBD-specific antibody binding. The anti-RBD Mab standard was used to calculate antibody concentrations of individual mouse sera. Duplicates of the Mab standards, controls and serum samples were injected over all surfaces for 1 minute (association phase) at flow 5 µL/min and washed with running buffer for 5 min (dissociation phase). The remaining bound antibodies were removed with two 40-second pulses of 15 mM HCl (pH 2.0). Using the anti-RBD Mab standard, serum polyclonal Ab (pAb) initial rates of binding were translated into concentrations using Phenom, a software developed to systematize multiple data analysis. Anti-RBD MAb standards against each domain at four concentrations spanning 1.2 to 250 nM were subsequently injected in a kinetic format at flow 50 µL/min for one minute to measure the kinetic profiles of each coupled domain. To calculate the kinetics of pAb binding to RBD, two individual sera dilutions were injected in duplicate over all surfaces at flow 50 µL/min for one minute and bound pAb were washed with running buffer for 5 min followed by surface regeneration. All binding profiles (sensorgrams) were double referenced to minimize the impact of instrument and solvent noise. Raw data analysis for the calculation of binding rates and kinetics was blinded to sample identity. Two types of calculations were carried out to fully characterize individual sera: pAb epitope specific concentration (dR/dt = kt*C) and binding kinetics (dR/dt =  ka*C*(Rmax-R)- kd*R), where Rmax and R were measured in RU (response units), C was the antibody concentration, kt was the mass transport constant, ka and kd were the kinetic association and dissociation constants. Specific pAb concentrations were calculated using Phenom v0.7.202101-alpha (Meritoki). Standard Mab binding profiles were globally fit to a 1:1 binding model to calculate the kinetic constants ka and kd and maximum binding capacity for each domain surface (or Rmax) using Scrubber 2.c (BioLogic Software, AU). In the same way, the profiles of pAb binding to RBD were analyzed by fixing the Rmax to the previously determined binding capacity for the RBD surface. The average kinetic parameters generated from two independent datasets (1/15 and 1/20 dilutions) were identical and were used to calculate the equilibrium association constants KA (ka/kd) M-1.

SARS-CoV-2 pseudovirus neutralization assay

All pseudoviruses used in this study were produced by co-transfecting HEK293T cells with the corresponding plasmid expressing the appropriate S with the pHIV-1NL43 ∆Env-NanoLuc reporter virus plasmid⁴⁶as previously described⁴⁷. Cell supernatant containing the pseudovirus was harvested 48 h post-transfection and stored at −80 °C until further use. To test the neutralization activity as described previously⁴⁷), HEK293T cells expressing the SARS-CoV-2 receptor ACE2⁴⁶were seeded in poly-L-lysine-coated 96-wells plates. The next day triplicate serial dilutions of heat-inactivated serum samples were prepared, mixed 1:1 with SARS-CoV-2 pseudovirus, incubated for 1 h at 37 °C, and then added in a 1:1 ratio to the cells. After 48 h, the cells were lysed, transferred to half-area 96-wells white microplates (Greiner Bio-One), and Luciferase activity was measured using the Nano-Glo Luciferase Assay System (Promega) with a GloMax system (Turner BioSystems). Relative luminescence units were normalized to the units from cells infected with pseudovirus in the absence of serum. Neutralization titers (IC50) were the serum dilution at which infectivity was inhibited 50%.

Antibody decay

A negative exponential decay model was fitted to the titers of IgG against RBD and neutralizing antibodies against the Wuhan, UK, and South-Africa strains of SARS-CoV-2. Only samples from unvaccinated individuals and samples taken prior to vaccinations were used when fitting the decay model. T1/2 was calculated as the time after hospitalization when half of the maximum titers were reached.

Olink

This data was generated by the IMPACC core. Study samples were assayed in plate batch layouts following a centralized randomized scheme that we described previously. In every plate, three samples (IMPACC_Serum, IMPACC_Plasma, and IMPACC_Plasma_Stim) were used as IMPACC inter-plate references (Reference samples). All samples (participant sera and reference) were subjected to a PEA (Olink) multiplex assay, Inflammatory panel (Olink Bioscience, Uppsala, Sweden), according to the manufacturer’s instructions. This inflammatory panel included 92 proteins associated with human inflammatory conditions. An incubation master mix containing pairs of oligonucleotide-labeled antibodies to each protein was added to the samples and incubated for 16 h at 4 °C. Each protein was targeted with two epitope-specific antibodies, increasing the assay’s specificity. The presence of the target protein in the sample brought the partner probes in close proximity, allowing the formation of a double-strand oligonucleotide polymerase chain reaction (PCR) target. On the following day, the extension master mix in the sample initiated the specific target sequences to be detected and generated amplicons using PCR in 96-well plates. For the detection of the specific protein, Dynamic array integrated fluidic CircuitN 96 × 96 chip was primed, loaded with 92 protein-specific primers, and mixed with sample amplicons, including three inter-plate controls (IPS) and three NC. For the target protein quantification, real-time microfluidic qPCR was performed in Biomark (Fluidigm, San Francisco, CA).

Serum proteomics targeted (Olink) quantification and statistical analysis

This data was generated by the IMPACC core. Data were analyzed using Real-time PCR analysis software via the ΔΔCt method and NPX manager. NPX is calculated in three steps from the Cq values: (i) ΔCqsample = Cqsample–Cqextensioncontrol, (ii) ΔΔCq = ΔCqsample–ΔCqinterplatecontrol, (iii) NPX = Correction factor–ΔΔCqsample. Data were normalized using internal controls in every sample, inter-plate control (IPC), NCs, and correction factors, and they were expressed on a log2 scale proportional to the protein concentration. One NPX difference equals the doubling of the protein concentration.

Batch normalization was performed to account for potential batch effects caused by re-assayed samples that were not able to adhere to the study randomization scheme or assay condition changes, including those due to assay kit lot# changes or differences in study collection phases. Olink Data Analysis Normalization employed identical reference samples in all plates. NPX value for each analyte was adjusted based on the adjustment factor that makes the median of all reference samples the same for all plates. Sequential steps included: (1) the reference sample the-inter-plate-median was calculated; (2) for each assay, the pairwise difference from the interplate median was calculated in first step 1 for each of the reference samples on all plates; 3) plate- and assay-specific differences in step 2 were used as normalization factors; and 4) plate and assay-specific normalization factors were added from step 3 to each value for each assay and plate.

SARS-CoV-2 RT-qPCR

This data was generated by the IMPACC core. To assess SARS-CoV-2 viral load, nasal swabs were collected and placed in 1 ml of Zymo-DNA/RNA shield reagent (Zymo Research). RNA was extracted using the quick DNA-RNA MagBead kit (Zymo Research) following the manufacturer’s instructions. RT-PCR for SARS-CoV-2 was performed on RNA extracts using the SARS-CoV-2 (2019-nCoV) CDC qPCR Probe Assay with target genes 2019-nCoV_N1, 2019-nCoV_N2, and control human RNase P (Integrated DNA Technologies)³⁷⁴. A ceiling cutoff of 40 cycles was applied as the limit of detection. Samples with an undetectable human RNase P were considered collection failures and were excluded.

Peripheral blood mononuclear cell (PBMC) gene expression

This data was generated by the IMPACC core. RNA was extracted from cells (2.5 × 10⁵ PBMCs), homogenized in 200 μL of Buffer RLT (Qiagen) and then extracted using the Quick-RNA MagBead Kit (Zymo) with DNase digestion. RNA quality was quantified using Qubit HS RNA assays and assessed using a Fragment Analyzer (Agilent). Library preparations were performed using the SMART-Seq v4 Ultra Low Input RNA Kit (Takara Bio) to synthesize full-length cDNA from an input of 10 ng of RNA. After a bead-based clean-up to purify the cDNA, the Nextera XT kit was used to create libraries through a process of tagmentation and fragment amplification and appended with dual-indexed bar codes using the NexteraXT DNA Library Preparation kit (Illumina). Libraries were validated by capillary electrophoresis on a Fragment Analyzer (Agilent), pooled at equimolar concentrations, and sequenced on an Illumina NovaSeq6000 (Emory) at 100 bp, paired-end read length targeting ∼25 million reads per sample. Repeated measures from a group of PBMC samples collected from healthy controls and repeated measures of a subset of IMPACC samples were used across library prep and sequencing batches to assess inter-site batch effects throughout the study. Universal Human References controls were included to assess intra-site batch variation.

PBMC transcriptomics data processing and quality control

This data was generated by the IMPACC core. Processing and quality control were performed using an internal Snakemake workflow for RNA-Seq analysis (Github: https://github.com/yerkes-gencore/IMPACC-RNA_Seq). Reads were trimmed for adapter sequence and quality score with cutadapt v1.14112. Reads were aligned with STAR v2.4.2a to a composite reference of human (GRCh38) reference sequence with gene annotations from Ensembl release 91 and SARS-CoV-2 (NCBI strain MN908947.3). Transcript abundance estimates were calculated internally to the STAR aligner using the algorithm of htseq-count94. Sequencing quality metrics were determined using FastQC77 (v0.11.5), alignment quality metrics with Picard tools (v2.22), and STAR logs and gene counts, including average quality per read > Q30, percent and absolute counts of reads uniquely mapped to annotated transcripts.

Gene counts were transformed to log2-counts per million using the voom function of the R package limma. A linear model with the log2 gene counts as a dependent variable and humoral response measurements as an independent variable was fitted to each gene. A moderated t-test was used to assess the significance of the correlation between a gene and humoral response (i.e., regression coefficient different from 0). Geneset enrichment analysis using the fgsea function of the R package fgsea was used to assess the enrichment of the Molecular Signature DataBase hallmark genesets (MSigDB version 7.3, module Hallmark) among genes ranked by their t-statistic. GeneSet Variation Analysis was used to average the expression of genes in the hallmark genesets for each sample.

Statistical methods

Data and statistical analyses were done in FlowJo 10.10 and GraphPad Prism 10.4, unless otherwise stated. The statistical details of the experiments are provided in the respective figure legends. Data plotted in linear scale were expressed as mean + standard deviation (SD). Data plotted in logarithmic scales were expressed as geometric mean + geometric standard deviation (SD). Correlation analyses were performed using Spearman, while Mann–Whitney or Wilcoxon tests were applied for unpaired or paired comparisons, respectively. Details pertaining to significance are also noted in the respective legends. Spearman correlation and Student t-test were used to assess the significance of the correlation between Visit 1 analytes and Visit 7/Visit 10 humoral response. T cell data have been calculated as background-subtracted data or stimulation index. Background-subtracted data were derived by subtracting the percentage of AIM+ cells after SARS-CoV-2 stimulation from the DMSO stimulation.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Author contributions

K.W. contributed conceptualization, data curation, formal analysis, investigation, methodology, writing-review & editing. D.J., J.C., and M.B. contributed to data curation and formal analysis. G.C and I.C. contributed data curation and formal analysis. J.B., M.J.G., J.R., and W.O. contributed to data curation and formal analysis. K.V.V. and N.V contributed to data curation and formal analysis. S.E.B., W.E., M.C.A., H.B., H.S., J.D.A., O.L., H.T.M., S.K.S., A.O. contributed by providing resources, investigation, writing-review & editing. The IMPACC network contributed data curation and formal analysis. R.P.S. contributed experimental design. R.S. contributed experimental design. M.A.K. and C.B.C. provided resources and funding acquisition. S.F. and E.K.H contributed conceptualization, investigation, providing resources, funding acquisition, project administration, writing, review & editing.

Peer review

Peer review information

Communications Biology thanks the anonymous reviewers for their contribution to the peer review of this work. Primary Handling Editors: Emily Lee and Dario Ummarino.

Data availability

Data used in this study is available at ImmPort Shared Data under the accession number SDY1760 and in the NLM’s Database of Genotypes and Phenotypes (dbGaP) under the accession number phs002686.v2.p2. The source data generated in the figures and Supplementary Figs. are provided as a Source data file. Source data are provided with this paper.

Code availability

All code is deposited on Bitbucket (bitbucket.org/kleinstein/impacc-public-code) and on Zenodo (zenodo.org/records/18200211] (https://zenodo.org/records/18200211).

Competing interests

The Icahn School of Medicine at Mount Sinai has filed patent applications relating to SARS-CoV-2 serological assays and NDV-based SARS-CoV-2 vaccines, which list Florian Krammer as co-inventor. Mount Sinai has spun out a company, Kantaro, to market serological tests for SARS-CoV-2. Florian Krammer has consulted for Merck and Pfizer (before 2020), and is currently consulting for Pfizer, Seqirus, 3rd Rock Ventures, Merck, and Avimex. The Krammer laboratory is also collaborating with Pfizer on animal models of SARS-CoV-2. Viviana Simon is a co-inventor on a patent filed relating to SARS-CoV-2 serological assays (the “Serology Assays”). Ofer Levy is a named inventor on patents held by Boston Children’s Hospital relating to vaccine adjuvants and human in vitro platforms that model vaccine action. His laboratory has received research support from GlaxoSmithKline (GSK) and is a co-founder of and advisor to Ovax, Inc. Charles Cairns serves as a consultant to bioMerieux and is funded for a grant from Bill & Melinda Gates Foundation. James A Overton is a consultant at Knocean Inc. Jessica Lasky-Su serves as a scientific advisor of Precion Inc. Scott R. Hutton, Greg Michelloti, and Kari Wong are employees of Metabolon Inc. Vicki Seyfer-Margolis is a current employee of MyOwnMed. Nadine Rouphael reports grants or contracts with Merck, Sanofi, Pfizer, Vaccine Company, Quidel, Lilly, and Immorna, and has participated on data safety monitoring boards for Moderna, Sanofi, Seqirus, Pfizer, EMMES, ICON, BARDA, Imunon, CyanVac, and Micron. Nadine Rouphael has also received support for meetings/travel from Sanofi and Moderna and honoraria from Virology Education. Adeeb Rahman is a current employee of Immunai Inc. Steven Kleinstein is a consultant related to ImmPort data repository for Peraton. Nathan Grabaugh is a consultant for Tempus Labs and the National Basketball Association. Akiko Iwasaki is a consultant for 4BIO, Blue Willow Biologics, Revelar Biotherapeutics, RIGImmune, Xanadu Bio, and Paratus Sciences. Monika Kraft receives research funds paid to her institution from NIH, ALA; Sanofi, Astra-Zeneca for work in asthma, serves as a consultant for Astra-Zeneca, Sanofi, Chiesi, GSK for severe asthma; is a co-founder and CMO for RaeSedo, Inc, a company created to develop peptidomimetics for treatment of inflammatory lung disease. Esther Melamed received research funding from Babson Diagnostics and honorarium from Multiple Sclerosis Association of America and has served on the advisory boards of Genentech, Horizon, Teva, and Viela Bio. Carolyn Calfee receives research funding from NIH, FDA, DOD, Roche-Genentech, and Quantum Leap Healthcare Collaborative as well as consulting services for Janssen, Vasomune, Gen1e Life Sciences, NGMBio, and Cellenkos. Wade Schulz was an investigator for a research agreement, through Yale University, from the Shenzhen Center for Health Information for work to advance intelligent disease prevention and health promotion; collaborates with the National Center for Cardiovascular Diseases in Beijing; is a technical consultant to Hugo Health, a personal health information platform; cofounder of Refactor Health, an AI-augmented data management platform for health care; and has received grants from Merck and Regeneron Pharmaceutical for research related to COVID-19. Grace A McComsey received research grants from Redhill, Cognivue, Pfizer, and Genentech, and served as a research consultant for Gilead, Merck, Viiv/GSK, and Janssen. Linda N. Geng received research funding paid to her institution from Pfizer, Inc. All other authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at https://doi.org/10.1038/s42003-026-09927-x.