Longitudinal transcriptional analysis of peripheral blood leukocytes in COVID-19 convalescent donors

Between April-December 2020 (i.e., before the COVID-19 vaccination), 162 CCD and 40 healthy donor controls were enrolled prospectively in an IRB-approved protocol (Clinical Trials Number: NCT04360278) and provided written informed consent to participate in the study. Of the 162 CCD subjects, 93 subjects donated blood once, while 46 donated twice, 12 thrice, 6 four times, and 5 donated five times.

Eligibility criteria for CCD included (1) routine blood donor criteria, (2) molecular or serologic laboratory evidence of past COVID-19 infection, and (3) complete recovery from COVID-19, with no symptoms other than residual loss of taste or smell for ≥ 28 days, or ≥ 14 days with a negative molecular test after recovery and was considered as the first visit post convalescence. We collected donor demographic and biometric data, including age, race, sex, ABO blood type, body mass index, and complete blood counts at the first visit for each subject in the early convalescent period. For each CCD, clinical severity of past COVID-19 infection was categorized as asymptomatic, mild (self-limiting course, symptomatic management at home), moderate (emergency room management or hospitalization), or severe (ICU admission). In all cases, anti-SARS-CoV-2 testing was performed. The minimum interval between plasma donations was 28 days; shorter intervals were acceptable between sample draw visits. Routine plasma donor testing was performed, including standard infectious disease testing, blood group assessment, and human leukocyte antigen antibody testing in female donors. Healthy donor control samples were obtained from research donors (protocol 99-CC-0168) who previously provided consent for the collection of research blood samples and had self-reported to be negative for SARS-CoV-2 exposure.

Anti-SARS-CoV-2 testing was performed using the Ortho-Clinical Diagnostics VITROS® Total (IgA/G/M) and IgG COVID-19 Antibody tests, as well as the SARS-CoV-2 neutralizing assay (NIH/National Institute of Allergy and Infectious Diseases (NIAID) Integrated Research Facility at Fort Detrick, Maryland, USA) as previously described [20].

Five to ten milliliters of human whole blood samples were collected in EDTA-anticoagulated tubes (BD) and centrifuged at 2500 RPM for 15 min. The supernatant plasma was separated for the antibody and multiplex immunoassays. ACK lysis buffer (Quality Biological) was added to the leftover pellet in a 1: 9 concentration, mixed several times, and incubated at room temperature for 15 min. Subsequently, the tubes were centrifuged at 1500 RPM for 10 min, and the supernatant was discarded. The pellet was washed twice with 1XPBS (KD Medical). 700 µL QIAzol lysis reagent (Qiagen) was added to the pellet with mixing and stored at −80 °C. Using the RNeasy Mini Kit (Qiagen), RNA was eluted in 40 µL of Milli-Q water. Following quality (Agilent 2100 Bioanalyzer) and quantity (Nanodrop One, Thermo Scientific) checks, the RNA was stored at −80 °C for further transcriptomic profiling.

Nanostring transcriptomic profiling was performed using the nCounter^® Human Host Response (Additional file 1: Table S1) and the nCounter^® Human TCR diversity panels (Additional file 2: Table S2). Whole blood total RNA (100 ng) was hybridized to reporter and capture probes at 65 °C for 16 h using a thermal cycler (Veriti Applied Biosystems). These hybridized samples were loaded onto the nCounter cartridge, and the post hybridization step and scanning were performed on the nCounter Prep Station and Digital Analyzer.

In a subset of CCD samples with highly perturbed gene expression and in healthy donor controls, we performed cytokine analysis. According to the manufacturer's instructions, a multiplex biometric immunoassay was performed to assess 48 cytokine and chemokine cell signaling molecules (Bio-Plex Human Cytokine Assay; Bio-Rad Inc., Hercules, CA, USA) [21]. The quantified cytokines included interleukins (IL-1α, IL-1β, IL-1Ra, IL-2, IL-2Rα, IL-3, IL-4, IL-5, IL-6, IL-7, IL-8, IL-9, IL-10, IL-12p40, IL-12p70, IL-13, IL-15, IL-16, IL-17A, & IL-18), interferons (IFN-α2 & IFN-γ), tumor necrosis factors (TNF-α, TNF-β, & TRAIL), growth factors (SCF, FGF, β-NGF, HGF, LIF, PDGF-BB, VEGF, SCGF- β, G-CSF, M-CSF, & GM-CSF), and chemokines (CCL247CTACK, eotaxin, GRO-α, CXCL10/IP-10, CCL2/MCP-1, CCL7/MCP-3, MIF, MIGCCL3/MIP-1α, CCL4/MIP-1β, CCL5/RANTES, & SDF-1α). A multiplex array reader from Luminex™ Instrumentation System (Bio-Plex 200 system) was used to determine the cytokine levels. The Bio-Plex Manager Software was used to calculate the cytokine concentrations.

All statistical analyses were performed using R (Version 4.1.1). Raw counts were normalized by scaling each sample by its geometric mean of the panel’s 12 housekeeping genes. Of the 270 samples from CCDs, we removed 2 samples with low signal strength, defined as low outlier values of the housekeeper geometric mean. The normalized data were then log2-transformed. Healthy donor samples were compared to CCD across 4 time windows: 26–89 days, 90–119 days, 120–149 days, and 150–241 days post-symptoms-onset. Within each window, a linear mixed model was fit to each gene’s normalized log2-transformed expression. The model treated CCD/healthy donor status, age, sex, and race as fixed effects and patient ID as a random effect. No patients had multiple samples within the 120–149 day window, so in this window, a linear model was fit with no random effects. The R library lmerTest was used to fit mixed models, and the R function lm was used to fit linear models. For each window, all genes’ p-values were converted to False Discovery Rates using the Benjamini–Hochberg procedure, using the R function p.adjust.

To calculate perturbation scores, we began by standardizing the data to give each gene mean 0 and standard deviation 1 within the healthy donor samples. We then defined a perturbation score as each sample’s Euclidean distance from the mean healthy donor sample in this standardized expression data. To define “highly perturbed” samples, we used the R library Mclust to cluster perturbation scores into two clusters, one high and one low. The highly perturbed samples were clustered into groups P1 and P2 by applying the R functions hclust and cutree to their log2-transformed normalized expression data.

TCR diversity scores were calculated using the Rosalind nCounter TCR Diversity Report, a software tool designed specifically for the nCounter TCR diversity panel. The software calculates the Shannon diversity index for each sample’s TCR gene counts. Gene expression values are first normalized to a “panel standard” reference sample to remove variability due to batch effects. TCR diversity scores were analyzed using the same statistical models applied to gene expression values.

Whole blood samples were collected from 270 donations by CCD and 40 contemporaneously recruited healthy donors from April to December 2020. The 270 CCD donations were consecutive, and they were from 162 donors. Among the 162 CCD, 93 donated once, and 69 donated more than once. All 40 healthy donors only donated once.

Age, sex, ethnicity, and Body Mass Index (BMI) distributions were similar among the 162 CCD and 40 healthy donors (Table 1). With respect to baseline complete blood counts, hemoglobin, platelets, and absolute basophil and eosinophil counts were similar among CCD and healthy donors. However, despite sample collections occurring over several months after convalescence, mean counts for absolute neutrophil counts (ANC), absolute monocyte counts (AMC), and absolute lymphocyte counts (ALC) were significantly higher among CCD compared to healthy donors (Table 1). Cell counts were collected earlier post convalescence for CCD only with their first donation. Among the CCDs, most had mild disease, and anti-SARS-CoV2 levels were highly variable. Some CCDs donated up to 5 times (Table 2).

Age, ethnicity, sex, BMI, and baseline cell counts, as well as ABO blood groups, disease severity and mean total and IgG antibodies, and median neutralizing antibody titers are summarized for CCD donating multiple times (Table 3). Serial antibody titers of individuals who donated 1, 2, 3, 4, or 5 times are shown in Additional file 5: Fig. S1, A-C. The total mean antibody levels increased at the last donation compared to the first donation [Signal/Cutoff (S/Co) 495.0 ± 315.3 versus 376.6 ± 276.4, p < 0.0001]. Otherwise, no differences were observed in this cohort of individuals with respect to changes in antibody titers over collections or when evaluated by other parameters.

CCD samples were stratified into 4 groups based on time since symptoms onset: 26–89 days, 90–119 days, 120–149 days, and 150–241 days. For each gene within each of these groups, a linear mixed model was fit comparing log2-transformed expression levels in CCD vs. healthy donors, adjusting for age, sex, and race, and treating donor ID as a random effect. Results from these models are in Additional file 3: Table S3.

All CCD time windows saw multiple genes with highly statistically significant changes from healthy donors (Fig. 1A). From the panel’s 775 non-housekeeper genes, a fold-change of > 20% and a False Discovery Rate < 0.05 were found for 85 genes in the 26–89-day window (n = 77), 87 genes in the 90–119-day window (n = 56), 178 genes in the 120–149-day window (n = 60), and 30 genes in the 150–241-day window (n = 58).

28 genes departed from the healthy donor mean by > 50% and with FDR < 0.05 in at least one time window (Fig. 1B). These genes participate in diverse biological pathways. While some genes gradually but monotonically return to healthy donor levels, others see their greatest departures from healthy donor expression 120–149 days after onset of symptoms (Fig. 1C). The genes with the greatest average departures from healthy donors include CTLA4, CXCR4, OSM, CXCL2, CCL3/CCL3L1/CCL3L3, IFNA6, and HERC5. CTLA4 and CXCR4 begin upregulated above healthy donor medians and gradually return to normal (Fig. 1C).

In efforts to further study these persistent or recurring and prolonged gene expression aberrations, we sought to delineate samples with immune states perturbed far beyond the average trend. We defined a perturbation score based on the Euclidean distance of each sample’s expression profile from the mean healthy donor sample (Methods). Average CCD perturbation scores were elevated at early time points and returned to the mean healthy donor levels by 200 days (Fig. 2A). CCD perturbation scores were highly right tailed, with a subset of samples from ~ 150 days post-symptoms onset falling far above the healthy donor range. Model-based clustering partitioned the perturbation scores into a large group of “typical” samples and a group of 21 “highly perturbed” CCD samples. This study cannot definitively attribute these highly perturbed immune states to earlier COVID-19; however, detailed donor history and the need for complete absence of symptoms during repeat donations preclude the possibility of re-infections as cause for these changes. Additionally, the study period (carried out early in the pandemic) ruled out the possibility of vaccine induced changes. With this caveat, the below results may offer clues to COVID-convalescent immune dysregulation.

The 21 highly perturbed samples fell into 2 distinct clusters based on their expression profiles. Cluster “P1” (8 samples) was characterized by high expression of PLAU, IL1B, NFKB1, PLEK, LCP2, and other genes (Fig. 2B, C). In a study of time-order transcriptomics to characterize molecular mechanisms which underpin multiple organ dysfunction in COVID-19, PLAU (plasminogen activator, urokinase) was among the genes to induce olfactory and neurological dysfunction [22]. NFKB1, an NF-κB signaling pathway gene, has been involved in the upregulation of inflammatory responses in patients with COVID-19 infection, with TLR4 acting as an intermediary. Additionally, despite a diminished IFN-I response, robust cytokine production and viral replication in SARS-CoV-2 infection are thought to be due to virus-mediated activation of NF-κB in the absence of other canonical IFN-I-related transcription factors [23‐25]. PLEK and LCP2 genes which code proteins, pleckstrin and lymphocyte cytosolic protein 2, respectively, may play a role in COVID-19 pathogenesis [26, 27].

Cluster “P2” (13 samples) was characterized by high expression of IRF3, MTOR, IL18BP, RACK1, TGFB1 and others. TLR3 and TLR4 activate IRF3 (Interferon Regulatory Factor 3) during the viral attack, triggering the type I interferons (IFN-1) transcription and NF-κB activation through the TRIF-dependent pathway during SARS-CoV-2 [28]. This, in turn, changes the expression of many genes that trigger inflammatory and antiviral responses. In severe COVID-19, SARS-CoV-2 triggers a chronic immune reaction that is instructed by TGF-β [29]. Excessively elevated TGF-β activity is also a key feature of COVID-19 cytokine storm [30]. During SARS-CoV-2 infection and replication, mTOR, a serine-threonine kinase involved in cell proliferation and cellular metabolism, was found to be active [31]. mTOR is involved in the interaction of adapter proteins MyD88, TLR9, and IRF-7 in plasmacytoid dendritic cells (pDCs), which leads to the transcriptional activation of type-I interferon (IFN) genes. The interleukin 18 binding protein (IL18BP) gene encodes a soluble inhibitor and carrier that keeps proinflammatory cytokine IL-18, a natural killer (NK) cell amplifier in check [32, 33]. Both T-cells and NK cells both produce IFN-γ, and IL-18 plays a critical role in this process. IL-18, along with other cytokines (IFN-γ, IL-1, IL-6, TNF), are elevated in cytokine storm and are thought to have central immunopathologic roles in COVID-19 [34].

We compared the TCR repertoires of 270 CCDs and 40 healthy donors using Nanostring's TCR diversity panel. The average TCR diversity score in CCD samples did not depart from the average healthy donor sample at any time point (Fig. 3A, B). However, the T-cell receptor (TCR) diversity score in perturbed subset P1 was significantly elevated compared to healthy donors (p = 1.18X10⁻⁷) (Fig. 3C), with unique T cell clonal expansion. Further analysis of the VJ gene combination revealed a significantly increased expression of 7 VJ pairs (TRAV9.1_TCRVA_014.1, TRBV6.8_TCRVB_016.1, TRAV7_TCRVA_008.1, TRGV9_ENST00000444775.1, TRAV18_TCRVA_026.1, TRGV4_ENST00000390345.1, TRAV11_TCRVA_017.1) while 54 pairs declined with FDR < 0.05 (Additional file 4: Table S4). TCR is crucial in T cell-mediated viral clearance and TCR bias is notable in various diseases [35]. Clonotypic T cell receptors (TCRs), which identify a peptide (8–15 amino acids) presented by major histocompatibility complex (MHC), direct the signaling that T cells use to orchestrate adaptive immunity [36, 37]. During the acute stages of infection, peptide-MHC complex (pMHC) recognition by TCR causes naive T cells to become activated and differentiate into diverse functional subsets, which eradicates invasive pathogens [38]. The variable (V), junctional (J), and constant (C) regions make up each of the two TCR chains (α and β) [39]. The diversity (D) region establishes an essential chain by joining the V and J areas. Thus, a functional and highly varied TCR repertoire is created by the TCR recombination process, which also develops highly diverse complementarity-determining regions (CDRs) localized in the TCR α and β chains. Luo et al. studied the blood T cells from recovered COVID-19 patients PBMCs from 1 to 6 weeks, and their TCR repertoires and immune metabolic processes were analyzed using single-cell TCR-seq and RNA-seq [36]. They observed that the TCR repertoire's diversity increased in patients who were discharged but it quickly went back to its baseline levels after 1 week after the SARS-CoV2 virus was eliminated. A significant shift in gene signatures from antiviral response to metabolic adaptation correlated with the dynamics of T cell repertoire in the study. By using ImmunoSEQ technology, Wang et al. studied the TCR repertoires of COVID-19 patients' PBMCs obtained before (baseline), during (acute), and after rehabilitation (convalescent) and they discovered that these patients TCR repertoires differed noticeably from healthy controls in terms of decreased TCR diversity, abnormal complementary-determining region 3 (CDR3) length, different TRBV/J gene usage, and higher TCR sequence overlap [40].

Multiplexed cytokine analysis was performed on 18 healthy donors, 6 P1 CCD, and 10 P2 CCD. Of the 48 cytokines analyzed, the P2 cluster had none that differed significantly from healthy donors, and the P1 cluster had 3 cytokines with significant differences from healthy donor: stem cell factor (SCF), Monocyte Chemoattractant Protein-1 (MCP-1), and Stem Cell Growth Factor-beta (SCGF-b) (Fig. 4). SCF overexpression is notable in inflammatory conditions. Binding of SCF to c-Kit leads to activation of multiple pathways, including phosphatidyl-inositol-3 (PI3)-kinase, phospholipase C (PLC)-gamma, Src kinase, Janus kinase (JAK)/Signal Transducers and Activators of Transcription (STAT) and mitogen-activated protein (MAP) kinase pathways. SCF is an important growth factor for mast cells, promoting their generation from CD34 + progenitor cells. In vitro, SCF induces mast cells survival, adhesion to extracellular matrix, and degranulation, leading to the expression and release of histamine, proinflammatory cytokines, and chemokines. SCF also induces eosinophil adhesion and activation. SCF is upregulated in inflammatory conditions both in vitro and in vivo in humans and mice [41]. MCP-1 elevation is suggestive of increased viral clearance from the CNS [42, 43]. SCGF, which was also marginally elevated in the healthy donor compared to the 2 CCD groups, may be a marker for hematopoietic recovery [44].

Demographic and laboratory data did not identify significant differences between the 2 perturbed clusters compared to healthy donor or “other” CCD samples (Table 4). We were unable to perform a look-back review of donors concerning the occurrence/persistence of signs or symptoms consistent with long-COVID syndrome in our cohort.