We studied IBD (16) and non-IBD HCW subjects (38) (n=521 patients total, Table 1), enrolled in an IRB-approved prospective registry at Cedars-Sinai between January and June 2021 (9, 10). For IBD patients, samples were collected longitudinally at the time of SARS-CoV-2 vaccine dose 1, dose 2 (when available), and 2 and 8 weeks after dose 2 (after dose 1 for vector vaccine participants) when possible. For HCW subjects, samples were collected at dose 1, and 8 weeks after dose 2. HCW subjects for this study were chosen from the available HCW registry by matching for the IBD age distribution. We quantified spike-specific and nucleocapsid SARS-CoV-2 antibody levels using the SARS-CoV-2 IgG-II assay (Abbott Labs, Abbott Park, IL). Self-reported COVID-19 were excluded from analyses except where specifically indicated.

Immunosequencing of the CDR3 regions of human TCRβ chains was performed on blood genomic DNA using the immunoSEQ Assay (Adaptive Biotechnologies, Seattle, WA), which includes bias-controlled multiplex PCR, high-throughput sequencing, identification and quantitation of absolute abundance of unique TCRβ CDR3 regions, and quantitation of the corresponding T cell fractions by template count normalization (33). Attribution of TCR sequences to SARS-CoV-2 spike or other non-spike SARS-CoV-2 protein specificities were assigned as described by Alter et al. and Snyder et al. (32, 39). Briefly, SARS-CoV-2-associated TCRβ sequences were identified using a one-tailed Fisher’s exact test comparing TCRβ presence in SARS-CoV-2 PCR-positive samples (n=1954) with negative controls (n=3903). Subsets of these SARS-CoV-2-associated sequences were assigned to spike and non-spike antigens based on data from multiplexed antigen stimulation assays (40). A total of 917 TCRs were assigned to the SARS-CoV-2 spike protein and 1564 to non-spike viral proteins. The breadth metric was calculated as the number of unique annotated rearrangements among total number of unique productive rearrangements in the individual sample’s dataset. The depth metric was calculated by the summation of two elements multiplied pairwise; (a) the raw frequency of each rearrangement in the total repertoire in the individual sample’s dataset, and an estimate of clonal generations of the lineage represented by each rearrangement. The resultant depth metric estimates the relative number of clonal expansion generations across the TCRs, normalized by the total number of TCRs sequenced in the sample. Hence, the metric can range from negative to positive values. We infer whether a SARS-CoV-2-associated sequence is a CD4+ or CD8+ T cell by statistically associating each sequence to a Class II or Class I HLA. HLA associations are derived from a set of 657 SARS-CoV-2 positive individuals who have genotyped HLAs. We built a binary logistic regression classifier with L1 regularization to determine which HLA best predicts the observed distribution of a given SARS-CoV-2-associated sequence across all HLA-typed cases. The L1 regularization strength was tuned to yield a single non-zero coefficient, giving a single inferred HLA association for each enhanced sequence. The inferred HLA associations are validated against the subset of SARS-CoV-2-associated sequence which overlap with our multiplex antigen stimulation assays.

All data was analyzed using R version 4.1.2. The MHC Class-I/II residual score was measured by calculating the linear regression of Class-I and Class-II breadth with no restrictions on the y-intercept. To calculate the scaled and normalized z-score, the mean of the residuals was subtracted from each residual and divided by the standard deviation. The residual and z-scores describe how far each patient’s response lies from the regression in an absolute manner, rather than a proportional manner (i.e. fold change), which limits Class-I/II variation noise from patients with low breadth and depth metrics. Patients with z-score >1 or <-1 were denoted as extreme responders and either Class-I biased or Class-I deficient, respectively.

Statistics were performed using t-tests, generalized linear models, ANOVA and Tukey’s test, and chi-squared tests as indicated. Generalized linear models were used unless otherwise noted in the text. Comparison of TCR breadth and depth used a Mixed Linear Model across time points and Generalized Linear Model within time points. Where possible, inverse normal transformation was performed, and age and sex were included as covariates. Confidence intervals for binomial probabilities were computed using exact methods. Other analyses are specified in the individual figures. Analyses were restricted to individuals with mRNA vaccines and no prior COVID-19 experience unless stated otherwise. Figure symbols and p-values serve as a visual guide, exact p-values are described in the text.

The demographics of the study cohorts are shown in Table 1. Both IBD and HCW cohorts were disproportionately female (IBD, 56%; HCW, 71%). Participant ages ranged from 19 to 83 years. Participants were grouped by decades of age into 6 groups, with the largest group being 30-39 years (n=158). By study design, the age distributions of the HCW subjects were matched to the IBD subjects. The IBD and HCW groups were similar in Hispanic/Latino ethnicity (~5-6%), but differed by race with a higher proportion of Asian subjects among HCWs (Table 1). All HCWs received BNT162b2 except for 1 subject who received mRNA-1273, while those with IBD received all 3 vaccines (n=154 BNT162b2, n=128 mRNA-1273, n=15 Ad26.COV2.S). IBD patients had either Crohn’s disease (CD, 211) or Ulcerative Colitis/Indeterminant Colitis (UC/IC, 93).

The time course of the T-cell response is shown in Figures 1A, B for IBD and HCW. These trends were observed in IBD patients and the HCW control population (after adjustment for age, sex, and IBD type). Both TCR breadth and depth increased significantly over time after vaccination compared to pre-vaccination (Figures 1A, B). For the IBD cohort, the fold-responses for clonal depth at dose 2, 2 weeks post-dose2, and 8 weeks post-dose 2, compared to dose 1, were 1.8, 2.7, and 1.9 fold (p<1E-10 each). Similarly, clonal breadth increased over time relative to baseline (p<1E-10). In the HCW cohort, the response was also elevated at the one evaluated time post-vaccine (8 weeks post-dose 2, p<1E-10). There was no significant difference by race or between IBD types and HCW cohorts for either depth or breadth metrics (Figures S1A, B).

Age was a significant factor in the T-cell response (Figures 1C, D), but this effect was restricted to clonal depth (p=5.8E-9), not clonal breadth (p=0.63). The impact of age on clonal depth was not significantly different for IBD and HCW cohorts (ANOVA, p=0.07). Sex was not significantly associated with clonal breadth and depth overall (Figures 1E, F). However, we observed variations over time that may suggest that sex affects the dynamics of the T-cell clonal response. A sex difference was observed at Dose 2, with a higher depth in female IBD patients versus male patients, but this was conflated with a disparity in age distribution and hence not significant (p=0.13, IBD only, Figures 1F).

We also assessed the impact of vaccine type on the T-cell response (Figures 1G, H). Among the relatively small subgroup (N=15) receiving vector vaccine (Ad26.COV2.S), we observed a significantly lower clonal breadth (p=0.0034) and depth (p=0.015), at 2 weeks and 8 weeks post vaccination compared to mRNA vaccines. Responses to the two mRNA vaccines were generally comparable, but patients vaccinated with mRNA-1273 generated a higher clonal breadth (p=0.0044) but not depth response (p=0.53) in patients at 8 weeks.

Our data included a small number of subjects with self-reported prior COVID-19 infection (n=27). Despite a small sample size, this group was associated with significantly higher breadth and depth at dose 1, dose 2, and 8 weeks post vaccination (p=0.036E-7, Figures 1I, J). In accordance with the TCR response after COVID-19 infection, we also observed an elevated response non-spike, SARS-CoV2-specific T breadth and depth in subjects with prior infection. These cells are not stimulated by the vaccine and therefore unchanged following vaccination in subjects with no prior COVID-19 infection (p<1E-10, Figure S1C). Among the top 10% of non-spike breadth values, 27% of samples reported prior COVID-19 infection (26/95) compared to 2.1% (19/875) prior infections in the bottom 90% of non-spike breadth values. Similarly, of the top 10% of non-spike depth values, 20% (19/97) came from reported prior infection samples compared to 3% of the bottom 90% of non-spike depth values.

We further analyzed the change in T-cell response by stratifying patients by the MHC Class-I/Class-II MHC-specific response. As stand-alone metrics, the Class-I- and Class-II-specific clonal breadth and depth responses display the same trends as the overall breadth and depth (Figures 2A, C). There was a consistent proportion of Class-I- to Class-II-specific clones for both breadth and depth (about 85% Class-II).

We tested whether there was any disproportionate distribution of Class-I/Class-II variation (from Class-I-biased to Class-I-deficient/relatively Class-II-biased TCR responses) that might be dependent on clinical factors. To measure the Class-I/II variation, we performed a linear regression between Class-I and Class-II breadth and depth and measured the residual (Figures 2B, D) and z-score of Class-I/II bias for each patient’s SARS-CoV-2-specific response (Figures S2A, B).

Lower z-scores indicate less Class-I-specific T-cell activity. We observed a small but significant decrease in z-score for breadth and depth with increasing age group (Figures S2C-F, ANOVA, breadth p=1.7E-8, depth p=1.8E-7). Z-score and thus Class-I-specific T cells increased in female participants (breadth p=0.003, depth p=0.006, Figure 2E). No significant changes were associated with the vaccination time (breadth p=0.44, depth p=0.24) or IBD status (breadth p=0.75, depth p=0.68). We also observed a divergence in breadth and depth z-scores over time by vaccine administered, with those that received mRNA-1273 having significantly lower z-scores (8 weeks breadth p=0.02, depth p=0.0035, Figure 2F).

Some immunomodulatory treatments significantly affected the Class-I/II balance. Patients receiving anti-IL23 (n=86, Figure 2G) had lower breadth z-scores (p=0.039) but not depth z-scores (p=0.11). No significant changes were observed in breadth or depth z-scores for patients receiving anti-integrins (n=40, breadth p=0.33, depth p=0.058) or immunomodulators (n=49, breadth=0.86, depth=0.26). Different types of anti-TNF treatment resulted in significant differences (n=103, infliximab n=63 and adalimumab n=37). After correcting for age, sex, and time after vaccination, patients receiving adalimumab had significantly higher depth Z-scores (p=0.012, Figure 2H) but not breadth Z-scores (p=0.091), while those receiving infliximab (breadth p=0.83, depth p=0.44) did not. Finally, patients reporting prior COVID-19 infections also had significantly higher z-scores and Class-I bias (p<0.05), especially at Dose 1 and at 8 weeks after vaccination (Figure 2I).

For ~85% of all patients measured, a typical balance between measured Class-I and Class-II breadth and depth was maintained with Class-I:Class-II at ~1:6 ratio, indicated by Class-I/II z-scores between -1 and 1. Other extreme responder patients were isolated (Figures 3A, B) and associated with clinical features. The most significant associations were observed at 8 weeks after vaccination, with both age and vaccine type emerging (chi-squared test, breadth age p=0.032 and vaccine type p=0.038, depth age p=0.0095 and vaccine type p=0.018). Patients with breadth or depth z-scores greater than 1 (Class-I-biased) were significantly younger, and vice versa for z-scores less than -1 (Class-I-deficient). Of the three vaccines studied, patients receiving mRNA-1273 resulted in lower z-score extremes, while those receiving Ad26.COV2.S resulted in higher z-score extremes in the limited sample size.

We measured antibody abundance in patients to determine if TCR metrics were associated with antibody production (Figure 3C). As previously reported (16, 38, 41, 42), IgG-spike antibody levels were highly dependent on time after vaccination, prior SARS-CoV-2 infection, and vaccine type, and slightly dependent on IBD status (Figure 3D). No significant association was found based on age or gender (p=0.91, p=0.35, respectively). Antibody levels were highly correlated with standard TCR metrics of breadth and depth. We observed increased antibody levels with increased breadth z-score, independent of the overall clonal breadth (p=0.028).