How B-cell recovery affects mRNA vaccine responses in stem cell transplant patients

AlloHCT recipients (n = 77), CAR‐T recipients (n = 3) or healthy adults (n = 29) were vaccinated with two doses of the BNT162b2/Comirnaty® mRNA vaccine 3 weeks apart (Figure 1a).3 The study cohort characteristics at study start are summarised in Table 1 and Supplementary table 1 (S2). Data types across study participants are reported in Supplementary table 1. Eight alloHCT recipients were grouped as COVID‐19‐experienced on the basis of detectable antibodies for SARS‐CoV‐2 at study start (n = 7) and/or PCR‐verified infection before study day 35 (n = 2, one also with pre‐existing antibody). These study participants were grouped as COVID‐19‐experienced and reported separately in Figure 1.

At 14 days after boost vaccination, there were detectable IgG binding titres against both spike (S) and the receptor binding domain (RBD) in all healthy controls assessed (n = 14), as well as 69/77 (90%) of alloHCT recipients for S, and 57/77 (74%) for RBD. No detectable S‐ or RBD‐specific IgG titres were found in recipients of B cell targeted CAR‐T therapy (Figure 1b and c). AlloHCT recipients vaccinated < 12 months (± 30 days) post‐transplant mounted lower vaccine‐specific antibody levels than healthy controls, while antibody responses in patients vaccinated later than 12 months post transplant were comparable to healthy controls (Figure 1b and c). Low‐ or undetectable S‐/RBD‐specific titres were noted in both HCT groups. S‐specific IgG antibody levels significantly correlated with time elapsed between alloHCT and first vaccine dose (Figure 1d). However, the correlation was limited and outliers observed, indicating that factors other than time also influence the magnitude of the vaccine response. Further subdivision of the > 12mo HCT group did not indicate that a different time cut‐off (within a 12–36 months range) would consistently separate responders from non‐responders (Supplementary figure 2A).

Stratification of alloHCT recipients by ongoing pharmacologic immunosuppression (Supplementary figure) and cGvHD at the time of vaccination (Supplementary figure) indicated that both variables negatively impacted the magnitude of Spike‐specific IgG in serum, though significance was only seen in alloHCT recipients < 12mo post alloHCT. 3A 3D

S‐specific memory T cells, as assessed by peptide recall and intracellular cytokine staining, showed detectable CD4⁺ and CD8⁺ T‐cell responses in a majority of the participants assessed, including the CAR‐T recipients (Supplementary figure 4A and B). The S‐specific T‐cell response was Th1 polarised as evidenced by the majority of responding cells producing IFN‐γ or IL‐2, as well as TNF.

Serum neutralising capacity against live SARS‐CoV‐2 (WuHu‐1 strain) after two doses of mRNA vaccine was assessed in a subset of alloHCT recipients (n = 29) and healthy controls (n = 19). Selection was based on both sample availability and demonstrated seroconversion by S‐ and RBD‐binding ELISA. Seroconverted alloHCT recipients demonstrated similar neutralising titres as the healthy controls (Figure 1e), as well as neutralisation potency index (ratio of neutralising titre to binding ELISA titre) (Figure 1f). An assessment of antibody binding strength by chaotropic ELISA showed on average similar binding strength in alloHCT recipients and healthy controls (Figure 1g). We observed a trend towards a markedly higher Spike‐binding IgG avidity in SARS‐CoV‐2 infection‐experienced participants, in line with prior data from healthy vaccinees.25 A proportion of alloHCT recipients had antibody avidity below quantifiable level (no detectable binding remaining after chaotropic treatment, here reported as avidity index = 1%), which was not seen in the controls. Assessment of antibody cross‐reactivity to the S proteins of the gamma and beta strains of SARS‐CoV‐2 showed similar levels in vaccine‐responsive alloHCT recipients and healthy controls (Supplementary figure 3G). The induction of antibodies targeting different regions of the S/RBD protein was investigated by performing ELISA with the competition of well‐characterised monoclonal antibodies (mAbs) representing four previously defined antibody classes26 for a subset of participants. Among alloHCT recipients with measurable antibody levels allowing for accurate quantitation, recipients and healthy controls exhibited similar levels of mAb‐competing binding for all five mAbs analysed (Supplementary figure 3H). In summary, in the alloHCT recipients who developed quantifiable antibody responses to mRNA vaccination, antibodies exhibited similar avidity, cross‐reactivity and neutralising capacity as in a healthy population. However, the underlying reasons for the observed highly variable vaccine responses in the alloHCT patients remained unclear.

To better understand how immune reconstitution and the composition of the B‐cell pool may impact the vaccine response, we performed IgM and IgG repertoire sequencing of the BCR immunoglobulin heavy chain variable region (VH) for bulk B cells from 20 individuals representing both early (< 12) and late (> 12) months post alloHCT recipients as well as healthy controls. Given that alloHCT recipients constitute such a highly heterogeneous group, we selected both individuals with (n = 8) and without (n = 8) GvHD for this analysis. Detailed sequencing information per group is described in Supplementary Table 2.

Clonotype analysis was performed with the IgDiscover software, using the well‐established criteria of identical V‐J pairing, same HCDR3 length and 80% amino acid sequence identity in the HCDR3. Both IgM and IgG repertoires were analysed (Figure 2a and b). The IgG repertoires showed variable clonal composition between alloHCT recipients, with some participants exhibiting proportionally expanded clones comprising > 5% (up to approximately 15%) of the total repertoire. Such proportionally expanded IgG clones were not observed in healthy controls. Despite the significantly expanded BCR clones in some alloHCT recipients, median clonal proportions were similar between alloHCT recipients and healthy controls in both IgG and IgM repertoires (Figure 2c). Distribution of VH and JH gene usage in IgG and IgM repertoires was similar between alloHCT recipients and controls (Figure 2e, Supplementary figure 5C). The overall polyclonal repertoires were further supported by the estimation of IgG and IgM BCR clonotype diversity using species richness index Chao1, which yielded similar results in the alloHCT recipients and controls irrespective of time elapsed after transplantation (Supplementary figure 5D).

We estimated the degree of IGHV somatic hypermutation (SHM) compared with germlines assigned using the OGRDB AIRR‐C database (version 7) as reference. Post‐alloHCT repertoires showed a similar median SHM for IgM but reduced SHM of IgG repertoires in the < 12mo group compared with controls (Figure 2d). We also observed that IgG SHM correlated with time post‐alloHCT (Figure 2f), indicating time‐dependent maturation of the class‐switched B‐cell pool. Subdividing alloHCT recipients by the presence/absence of cGvHD did not show a clear effect on IGHV‐SHM (Supplementary figure 7B) or clonal size (Supplementary figure 7C) for either IgM or IgG repertoires.

Given the differences in the overall BCR repertoire between alloHCT recipients and healthy controls, we investigated whether differences would also be observed in the antigen‐specific B cells induced by vaccination. The frequencies of S‐ and RBD‐specific memory B cells (MBCs) were measured using fluorescently tagged recombinant S and RBD protein probes (Figure 3a). As with the antibody titres, the frequency of both S‐specific class‐switched IgG+ MBCs was significantly lower in the early post‐alloHCT recipients than in healthy controls (Figure 3b). IgM+ S‐specific B cells displayed a similar pattern, with significantly lower frequencies in the early alloHCT recipient group than in healthy controls (Figure 3c). The fraction of RBD‐binding out of total S‐specific IgG+ B cells was similar between alloHCT groups and controls (Figure 3d). Single S/RBD‐specific IgG+ B cells were index‐sorted, and the immunoglobulin heavy chains sequenced to estimate clonality and degree of SHM. SHM of S‐specific BCR sequences was highly variable between individuals (Supplementary figure 6A), and overall similar when comparing pooled sequences from the different study groups, albeit with a slightly higher median level in the < 12mo alloHCT group at the clonotype level (Figure 3e). Clonotype analysis of sorted S+ single cells showed a highly polyclonal response, with occasional expanded clones consisting of > 10 highly similar sequences (Figure 3f). S‐specific sequences frequently used IGHV genes IGHV3‐30, IGHV3‐30‐3, IGHV1‐69 and IGHV3‐15 (Supplementary figure 6B and C), in line with previous data.26, 27, 28, 29, 30 In summary, < 12 months post alloHCT recipients developed lower levels of S‐ and RBD‐specific peripheral B‐cell memory. However, S‐specific BCR sequences from class‐switched B cells indicated that BCR sequences from HCT recipients were polyclonal and accumulated SHM to at least a similar level as in healthy controls. Taken together, these data indicate a quantitative, but not necessarily qualitative, restriction of the memory B‐cell vaccine response in alloHCT recipients vaccinated < 12mo post transplant.

To improve coverage of the circulating S‐specific B cell pools in alloHCT recipients and better estimate clonal composition and diversity, we queried our bulk IgG and IgM libraries for alloHCT recipients (n = 15, excluding one COVID‐19‐experienced individual) and controls (n = 4) for sequences highly similar to antigen‐specific clonotypes identified through the single‐cell sequencing described above (Figure 3g). Criteria for identification as antigen‐specific were identical HV‐HJ gene pairing, identical HCDR3 length and 80% amino acid sequence identity in the HCDR3 to at least one sequence in the single‐cell reference dataset. S‐specific clonotype size (Figure 3h) and IGHV SHM (Figure 3i) were similar between alloHCT recipients (n = 15) and healthy controls (n = 4), irrespective of the 12‐month cut‐off in time elapsed from transplantation and first vaccination. One S‐specific clonotype comprising >5% (approx. 7%) of total repertoire was identified in an alloHCT recipient (Supplementary figure 7A and G). Diversity analysis of antigen‐specific B‐cell clones using the Chao1 index indicated that alloHCT recipients were able to develop BCR repertoires with similar diversity as healthy controls in response to vaccination (Figure 3j).

To identify potential factors underlying the variable vaccine responses in alloHCT recipients, we extended our study by performing detailed phenotyping of B cells in a subset of participants at study start (Day 0). Participants with a range of vaccine responses in the < 12 and > 12 months post alloHCT groups were analysed (n = 35 HCT, 2 CAR‐T, 7 HC).

Relative proportions of total CD19⁺ B cells were not significantly different between the alloHCT groups and healthy controls, but were clearly reduced in the CAR‐T cell recipients (significance not tested due to low n), as expected (Figure 4a). Both subgroups of alloHCT recipients had similar proportions of B cells with a CD27⁻, CD21⁺ resting naïve phenotype compared to the healthy controls (Figure 4c). The healthy controls had higher proportions of resting CD27⁺, CD21⁺ memory B cells compared to healthy controls (Figure 4d). We observed elevated percentages of activated memory (CD27⁺CD21⁺) (Figure 4e) and atypical (CD27⁻CD21⁺CD11c⁺IgD⁻) B cells (Figure 4f) in a few alloHCT recipients, but proportions did not differ significantly between alloHCT and healthy controls at the group level. There was no clear influence of cGvHD on the frequencies of B cell subpopulations at Day 0, except for the observation that cGvHD+ alloHCT recipients <12mo from transplant differed more from healthy controls in proportions of CD21⁺ CD27⁺ resting memory B cells than cGvHD‐ (Figure S8C). A trend was seen towards higher proportion of TrB in cGvHD⁺ < 12mo alloHCT group than that of controls (Supplementary figure 8F), but the difference was not significant. Comparison of alloHCT recipients with and without ongoing immunosuppression showed that alloHCT recipients > 12mo without immunosuppression had slightly higher proportions of naïve B cells (Supplementary figure 9B) than that of controls.

As transitional B cells (TrB cells) have been reported to be elevated in alloHCT recipients early after transplantation,11, 16, 31, 32, 33 we quantified the proportions of TrB cells, identified as CD24hi CD38hi CD19⁺ cells (Figure 4g, Supplementary figure 15). During assay establishment, we confirmed that the CD24hi CD38hi TrB population expressed higher levels of CD10 (MFI) than that of populations identified as resting naïve B cells and resting memory B cells (Figure 4h). We found that proportions of TrB cells were significantly elevated in alloHCT recipients < 12 months after transplantation compared with healthy controls, although there was a high degree of variability within both the < 12 and > 12 months alloHCT groups (Figure 4i). Within the TrB population, group differences were most pronounced in the CD27⁻CD21⁻ (Figure 4j) or CD21lo (Figure 4k) compartments. CD21lo TrB cells have been described as more functionally immature than CD21hi.33

Where cell numbers allowed, we additionally assessed the proportions and phenotypes of T cells, including circulating T‐follicular helper cells (cTFH). AlloHCT recipients showed similar levels of total T cells as healthy controls (Supplementary figure 10A). The CD4/CD8 ratio was decreased in alloHCT recipients (Supplementary figure 10B and C), though the proportion of cTFH cells out of CD4⁺ was similar to that of healthy controls (Figure 4l).

To understand the impact of the peripheral immune cell pool composition on the response to vaccination, we correlated proportions of different B‐cell subpopulations at the study start to the level of S‐ and RBD‐specific IgG attained after two vaccine doses (Day 35). There was a moderate negative correlation between fractions of TrB cells and S‐specific antibodies IgG (Day 35) (Figure 4m). TrB was also negatively correlated to time elapsed post‐alloHCT (Figure 4m). To better understand the relative contributions of TrB and clinical parameters, a multivariable linear regression analysis including time since transplantation, immunosuppression status and TrB frequency was performed. The TrB emerged as the strongest independent predictor of vaccine antibody titres (Supplementary figure 11), indicating that baseline transitional B‐cell frequency is a robust determinant of post‐vaccination antibody responses in this cohort.

We performed functional in vitro cultures of PBMCs from alloHCT recipients to measure whether their B cells had intact capacity to respond to stimulation, differentiate to antibody‐secreting cells and produce soluble antibody. PBMCs from a subset of < 12 months and > 12 months alloHCT recipients (n = 15), as well as from healthy controls (n = 5) were cultured with or without addition of anti‐human IgM Fab2 fragments, IL‐21, CD40L and CpGB for 144 h (6 days) (Figure 5a). Cultured cells were analysed for expression of CD27 and CD38, indicative of differentiation to the antibody‐secreting cell (ASC) state (Figure 5b). Selection of samples for ex vivo stimulation was based on the availability of PBMCs as well as B‐cell levels (participants expected to have near‐total B‐cell aplasia were not selected). Polyclonal activation induced similar levels of ASC differentiation (CD19⁺CD38hiCD27hi) (Figure 5c, Supplementary figure 12C), Ki‐67 expression in CD19⁺ cells (Figure 5d, Supplementary figure 12D) as well as IgG (Figure 5e, Supplementary figure 12E) and IL‐10 (Figure 5f, Supplementary figure 12F) production in the assessed alloHCT recipients compared to healthy controls. The fold change in IgG concentration was slightly lower in some alloHCT recipients than in healthy controls (Supplementary figure 12E); however, substantial antibody production was induced in all tested samples except one. There was a moderate correlation between the proportion of CD21hi TrB cells (as observed in phenotypic analysis in Figure 4i) and the fold change of ASCs after polyclonal activation (Supplementary figure 12G). We also observed a correlation between frequency of resting memory B cells and induction of IgG in stimulated culture supernatants (fold change compared to unstimulated controls) (Supplementary figure 12G). In summary, PBMC from both the < 12 and > 12 months post alloHCT recipients tested in this assay showed similar capacity to differentiate toward ASC phenotype and produce antibodies as cells from healthy controls. These results suggest that B cells in the tested alloHCT recipients were functional when provided with sufficient stimuli.

To assess whether there is a link between B‐cell phenotype, BCR repertoire composition and vaccine responses, we performed correlation analysis using datasets from Day 0 and Day 35. Diversity (Chao1) of the Day 35 S+ IgG pool was positively correlated with the Day 35 antibody avidity, as well as the pre‐vaccination proportions of total and naïve B cells (Figure 5g). Similar to IgG, total IgM diversity was positively correlated with the proportions of total and naïve B cells. Both total and S‐specific IgG IGHV SHM negatively correlated with baseline proportions of transitional B cells, while total IgG IGHV SHM positively correlated with baseline proportions of activated and resting memory B cells (Figure 5g). There was no significant correlation between IgG diversity and TrB cells. Taken together, these data support the conclusion that while mRNA vaccination of alloHCT recipients can in many cases elicit a highly diverse antigen‐specific B‐cell repertoire with IGHV mutational load similar to that of a control population, the presence of immature B‐cell populations may limit SHM accumulation.

The study participants described here were part of a previously published prospective clinical trial, COVAXID.36 This trial was registered at EudraCT (no. 2021‐000175‐37) and clinicaltrials.gov↗ (NCT04780659↗). The Swedish Medical Product Agency (ID 5.1‐2021‐5881) and the Swedish Ethical Review Authority (ID 2021‐00451) approved the study. The original clinical trial protocol included two vaccine doses and immunogenicity measurements until 6 months after the second dose and was subsequently extended with permission from the Swedish Ethical Review Board and the Swedish Medical Products Agency (no. 2021‐06046‐02 and no. 5.1‐2021‐92 151, respectively). The primary endpoint was seroconversion after two vaccine doses at Day +35 from first vaccination. Immune reconstitution was a secondary endpoint. Study participants who did not receive their second immunisation were excluded from analysis. Study participants who had pre‐existing immunity at baseline, a positive PCR at baseline, or PCR‐verified COVID‐19 infection between Day 0 and Day 35 were considered COVID‐19+. These participants are reported separately in Figure 1 and are excluded from further analysis to avoid misconstruing the assessment of vaccine immunogenicity. The exception to this is one COVID‐19+ participant who was included in the immunoglobulin repertoire sequencing analysis and is reported in Figure 2 (which pertains to the overall IGH repertoire), but is not included in the Spike+ B‐cell repertoire analysis in Figure 3.

Both male and female alloHCT recipients were included in the study. We included both male and female participants in all sub‐selections made for specific analyses. Because selection was also based on cGvHD⁺ vs. cGvHD⁻ and sample availability, we were not always able to maintain a 50‐50 ratio of male–female participants, for example in the case of BCR repertoire sequencing experiments. However, our assessment is that we have a relatively well‐balanced study population in terms of biological sex. For this study population, we do not expect the sex of study participants to have significant bearing on alloHCT engraftment or cGvHD. Prior data have demonstrated sex‐dependent differences in both innate and immune functions, as reviewed by Klein et al.51 and Dunn et al.52 In the context of vaccination, female sex has been associated with higher antibody responses to influenza vaccination53 and mouse models have shown greater diversity of germinal centre B cells after influenza immunisation.54 Conversely, male sex has been associated with higher disease severity of COVID‐19.55 However, for mRNA SARS‐CoV‐2 vaccination, early efficacy studies of the BNT162b2 (Pfizer/BioNTech)56 and mRNA‐1273 (Moderna Therapeutics)57 showed similar vaccine efficacy for male and female participants. The previously reported quantitative analysis of Spike‐specific antibody responses in the Phase IV clinical trial by Bergman et al.,3 which includes the study population reported here, also did not demonstrate significant impact of biological sex on antibody titres at Day 35.

Samples from Day 0 and Day 35 were collected according to the study protocol described in Bergman et al.3 Briefly, samples were collected into sodium heparin Vacutainer tubes (BD Biosciences). For plasma separation, a small aliquot of whole blood was centrifuged at 500 g for 10 min, and the plasma fraction was collected and stored at −20°C until use. Whole blood was diluted 1:1 with phosphate‐buffered saline (PBS) layered onto a Lymphoprep (Stemcell) or Ficoll‐Paque (Cytiva) gradient in SepMate tubes (Stemcell) according to manufacturer instructions. A 10‐min centrifugation at 1200 g was used for separation of PBMC from diluted whole blood. The PBMC‐containing fraction was transferred to a clean tube and washed twice with phosphate‐buffered saline (PBS) containing 2% heat‐inactivated fetal bovine serum (FBS). PBMCs were resuspended in PBS and counted using trypan blue and an automated cell counter. PBMC were cryopreserved in FBS containing 10% dimethylsulfoxide (DMSO) and stored frozen at −180°C until use.

Prior to serological assessment, plasma/serum samples were incubated at 56°C for 30 min. ELISA analysis for quantitation of S‐ and RBD‐specific antibodies was carried out at previously described,58, 59 with modifications. Half‐area ELISA plates (Greiner Bio‐One or Corning) were coated at 4°C overnight with 50 ng well⁻¹ of either SARS‐CoV‐2 HexaPro Spike proteins or soluble RBD (kind gifts of Neil King, University of Washington), at 1 μg mL⁻¹ in PBS. ELISA plates were washed 3× using PBS containing 0.05% Tween‐20 (PBS‐T); all subsequent washing steps were performed identically unless otherwise specified. Plates were blocked for 1 h at room temperature (RT) using 5% (w/v) milk powder in PBS. Plasma or serum samples were diluted 5‐fold in PBS with 5% milk. 50 μL serially diluted samples was added to ELISA plates and incubated for 2 h at RT, followed by a wash step. For assessment of antibody avidity by chaotropic ELISA, parallel plates were treated with either PBS or 1.5 M NaSCN, 50 μL well⁻¹, for 10 min followed by washing. Subsequently, plates were incubated 1 h at RT with 50 mL well⁻¹ peroxidase‐conjugated anti‐human‐IgG or anti‐IgA secondary antibody (Jackson Immuno) diluted 1:5000 (IgG) or 1:2000 (IgA) diluted in PBS + 5% milk. Plates were then washed. For detection, 50 mL well⁻¹ 1‐step Ultra TMB substrate (Thermo Fisher) was added and incubated for 5 min at RT. Peroxidase activity was stopped using 50 mL well⁻¹ 1 m H₂SO₄. Plates were read at 450 nm with 570 nm background correction using a VarioSkan Lux Multimode reader (Thermo Fisher). For IgG binding titres and avidity assessments, plates were read at 450 nm with 550 nm background correction using an EnSpire Multilabel Reader (PerkinElmer). Optical density (OD) at 570/550 nm was subtracted from the OD at 450 nm and the resultant OD was used in all subsequent calculations. A four‐point logarithmic (4PL) non‐linear regression curve fit was performed using Graphpad Prism version 9.1 (San Diego, California). Titres giving half‐max OD (ED50) was used as readout for plasma. For assessment of SARS‐CoV‐2 Spike‐specific avidity, the avidity index was calculated as the percentage of binding remaining in 1.5 m NaSCN treated plates as compared with untreated plates: ED50(NaSCN)/ED50(PBS) × 100%. For assessment of binding to SARS‐CoV‐2 variants of concern, the ratio of ED50 (variant spike) to ED50 (ancestral spike) was used as readout.

For mAb competition ELISA, monoclonal SARS‐CoV‐2 Spike‐binding antibodies were biotinylated using the EZ‐Link Sulfo‐NHS‐LC‐biotin kit (Thermo Fisher) according to manufacturer instructions. ELISA was performed as above with the following changes: Plates were coated with recombinant Spike protein at 1 μg mL⁻¹, or 2 μg mL⁻¹ for CR3022 competition. Plasma/serum was serially diluted threefold with a starting dilution 1:5. Samples were incubated for 30 min at RT, after which a fixed concentration of biotinylated monoclonal diluted in 5% milk buffer was added and incubated a further 90 min at RT. Plates were then washed as described for binding ELISA, and incubated with NeutrAvidin‐HRP (Thermo Fisher) at 1:5000 in PBS for 60 min at RT. Plates were washed, developed using TMB, and read as described above. A standard curve of monoclonal antibody in competition with its biotinylated counterpart was run in parallel with each sample batch. A 4‐point logarithmic non‐linear regression curve fit was performed using Graphpad Prism 10 for Mac (Boston, Massachusetts, USA) for each sample, and the ED50 used as read‐out. Curve fits of poor quality (visual assessment) or with competition too low for accurate quantitation were excluded from analysis. ED50 < LLOQ (1:5) were adjusted to LLOQ for statistical comparison.

Live virus neutralisation assay was performed as previously described60 with minor modifications. Briefly, SARS‐CoV‐2 wild‐type strain (SARS‐CoV‐2/01/human/2020/SWE) was grown on Vero E6 cells. For virus neutralisation assay, samples were serially diluted fivefold in serum‐free Dulbecco's Modified Eagles Medium (DMEM) supplemented with 0.2% penicillin/streptomycin. Diluted serum samples were incubated with 1000 PFU well⁻¹ SARS‐CoV‐2 virus for 30 min at 37°C. Following pre‐incubation, serum‐virus mix was added to Vero E6 cells, seeded in Greiner CELLSTAR® 96‐well plates (Greiner Bio‐One, Austria) at a density of 10⁴ well⁻¹, 1 day prior to the experiment. Cells were infected for 8 h, then fixed in 4% formaldehyde for 40 min. Plates were washed with PBS, and cells permeabilised for 10 min at RT using 0.5% Triton‐X and 20 mm glycine in PBS. Plates were blocked for 30 min at RT using 2% bovine serum albumin (BSA) in PBS followed by staining of infected cells using a rabbit‐anti‐SARS‐CoV‐2 nucleocapsid antibody (Sino Biological) for 1 h at RT, followed by a goat‐anti‐rabbit IgG (H + L) AF‐488 secondary antibody for 30 min at RT. Plates were counter‐stained with DAPI (0.1 mg mL⁻¹) for 10 min. Fluorescent signal was measured using a TROPHOS plate RUNNER HD instrument (TROPHOS SA, Marseille, France). The half‐maximal inhibitory dilution reciprocal (ID50) was determined by 4‐parameter nonlinear regression performed in Graphpad Prism 9.0 (Boston, Massachusetts, USA).

Frozen PBMC were thawed in a water bath at 37°C, resuspended in R10 medium, washed twice (1500 RPM × 5 min), rested for 1–3 h in an incubator at 37°C with 5% CO₂ and counted.

1–3 × 10⁶ PBMC were collected for B‐ and T cell phenotyping. PBMC were stained using one of two panels of fluorescently labelled antibodies. For B‐cell phenotyping: CD19⁻ECD (J3‐119, Beckman Coulter), CD10⁻BUV395 (MEM‐78, BD Biosciences), CD11c⁻AF700 (3.9, BioLegend), IgM‐PerCP‐Cy5.5 (G20‐127, BD BioSciences), CD38⁻APC‐Cy7 (HIT2, BioLegend), CD21⁻BV711 (B‐ly4, BD Biosciences), CD123⁻BV510 (6H6, BioLegend), CD3⁻BV510 (SP34‐2, BD Biosciences), CD24⁻BV650 (ML5, BD Biosciences), CD27⁻PE‐Cy7 (M‐T271, BioLegend), IgG‐BV786 (G18‐145, BD Biosciences), CD16⁻BV510 (3G8, BD Biosciences), CD56⁻BV510 (B159, BD Biosciences), CD20⁻BV605 (2H7, BioLegend), CD14⁻BV510 (M5E2, BioLegend), IgD‐BV488 (Polyclonal, Southern Biotech) and 7AAD live/dead viability stain (Invitrogen). Surface staining was performed at 4°C for 20 min. Cells were washed using PBS‐FCS and permeabilised using the BD Transcription Factor Buffer kit (BD Biosciences). Cells were resuspended in 200 μL of kit Fix per Perm buffer and incubated for 20 min at RT. Cells were thereafter washed with the kit perm/wash buffer and stained intracellularly using a panel of intracellular antibodies Ki‐67‐PE (B56, BD Biosciences), IRF4‐Pacific Blue (IRF4.3E4, BioLegend), BLIMP‐1‐AF647 (6D3, BD Biosciences) diluted in perm/wash. Cells were washed using perm/wash buffer, resuspended in PBS containing 1% FA, and acquired using a BD LSRFortessa flow cytometer. For T cell and monocyte phenotyping, surface staining was performed using the following markers: CCR6‐BV786 (11A9, BD Biosciences), CXCR3 BUV395 (1C6, BD Biosciences), CXCR5 E‐Fluor610 (MU5UBEE, Invitrogen), CD38 AF488 (HIT2, BioLegend), CD3 APC‐Cy7 (SP34‐2, BD Biosciences), CD16 AF700 (3G8, BD Biosciences), CCR7 BV421 (G043H7, BioLegend), CD20 BV510 (2H7, BioLegend), CD19 BV510 (HIB19, BD Biosciences), CD56 BV510 (B159 BD Biosciences), HLA‐DR BV510 (L243 BioLegend), CD4 PE‐Cy5 (SK3 BD Biosciences), CD8a BV711 (RPAT‐T8, BioLegend), ICOS PerCP‐Cy5.5 (C398.4A, BioLegend), CD14 AF677 (M5E2, BioLegend), CD45RA BV650 (5H9, BD Biosciences), and intracellular staining using FoxP3 PE (206D, BioLegend). Data were analysed using Flowjo version 10. Gating strategies used for the identification of B‐cell, T‐cell and monocyte populations shown in Supplementary figure 15.

Frozen PBMC were thawed as described for phenotypic analysis above. After aliquoting of PBMC for phenotyping, samples with sufficient remaining PBMC were resuspended in R10 medium at 10 × 10⁶ cells mL⁻¹. Cells were then cultured for 144 h in the presence or absence of polyclonal activators as previously described,61 with modifications. Stimulation was performed in R10 medium containing 0.5 μg mL⁻¹ anti‐IgM Fab2 fragments (Southern BioTech), 1 μg mL⁻¹ soluble CD40L (Stemcell), 50 ng mL⁻¹ IL‐21 (Stemcell) and 2.5 μg mL⁻¹ CpGB class B ODN (Invivogen). Briefly, 100 μL of cell suspension containing 1 × 10⁶ cells were plated onto 48‐well treated tissue culture plates (Corning‐Falcon). One to five replicate wells were plated per sample and condition depending on sample availability. Four hundred microliters of R10 containing polyclonal activators was added to stimulated wells. Four hundred microliters R10 without activators was added to control wells. Cells were cultured for 144 h (6 days), after which replicate wells were pooled in FACS tubes and centrifuged (450 g × 5 min). Supernatants were collected and stored frozen at −20°C for later analysis by ELISA. Cultured PBMC were washed with PBS containing 2% FCS and stained for surface and intracellular markers using the same antibody panel described above for phenotyping. Gating strategy used for the identification of CD27⁺CD38⁺ ASCs and Ki67⁺ B cells is shown in Supplementary figure 16.

For detection of total IgG, an in‐house sandwich ELISA set‐up was used. Briefly, half‐well ELISA plates (Greiner BioOne) were coated with 2 μg mL⁻¹ anti‐human IgG antibody (clone MT145) overnight at 4°C. Plates were washed 3× with PBS‐T and blocked for 1 h at RT with 5% milk powder in PBS. Blocking buffer was removed and supernatant samples diluted in 5% milk‐PBS were added to plates. An 11‐point standard curve of human IgG (Sigma‐Aldrich) was run in each plate alongside samples. Samples were incubated for 2 h at RT. Plates were washed 3× with PBS‐T and 50 μL horseradish peroxidase conjugated goat‐anti‐human IgG (Jackson Immuno) diluted 1:7500 in 5% milk‐PBS was added to wells and incubated 1 h at RT. Plates were washed 3x and developed using 1‐step TMB substrate (Thermo Fisher) for 5 min, after which the reaction was stopped using 1 m H₂SO₄. Plates were read at 450 nm with 570 nm background correction using a Varioskan LUX ELISA reader instrument (Thermo Fisher). Background subtracted OD values were used as readout. For detection of IL‐10, the MabTech human IL‐10 (HRP) ELISA Flex kit was used according to manufacturer instructions and using half‐well ELISA plates (Greiner BioOne). Plates were developed and read as described for anti‐IgG ELISA. For both IgG and IL‐10, 5‐point logarithmic standard curve fits and interpolation of sample ODs were performed in Graphpad Prism 10 to calculate concentrations of the respective analytes. Samples below LLOQ were adjusted to LLOQ for fold change calculations.

Fluorescently conjugated antigen probes were prepared as previously described.58, 59 Briefly, SARS‐CoV‐2 S2‐P Spike and soluble RBD proteins (kind gifts of Neil King, University of Washington, USA) were biotinylated using Sulfo‐NHS‐biotin (Thermo Fisher) according to manufacturer instructions. Excess biotin was removed using Slide‐A‐Lyzer mini dialysis devices (Thermo Fisher). Prior to staining of PBMCs, fluorescent Spike and RBD tetramers were constructed by stepwise incubation of biotinylated proteins with either Streptavidin‐APC, Streptavidin‐PE or Streptavidin‐BV421 (BioLegend) on ice, for a 4:1 final molar ratio of protein to streptavidin conjugate. Frozen PBMCs were thawed in a 37°C water bath and transferred to RPMI complete medium (HyClone) supplemented with 10% FBS (Gibco), 1% penicillin/streptomycin and 1% L‐glutamine (R10). Cells were washed twice with R10, rested for 1–3 h (37°C, 5% CO₂), and counted using Trypan blue and an automated cell counter. One to three million PBMCs was transferred to FACS tubes, washed with cold PBS containing 2% FBS and stained with 100 ng each of Spike‐PE tetramer, Spike‐APC tetramer and RBD‐BV421 tetramer for 20 min at 4°C. Cells were then surface stained with anti‐human IgM‐PerCP‐Cy5.5 (G20‐127; BD Biosciences), CD3⁻ BV510 (SP34‐2; BD Biosciences), CD123⁻BV510 (6H6; Biolegend), CD19⁻ECD (J3‐119; Beckman‐Coulter), CD16⁻BV510 (3G8; BD Biosciences), HLA‐DR‐ BV650 (L243; Biolegend), IgG‐BV786 (G18‐145; BD Bioscience), CD20⁻BV605 (2H7; Biolegend), CD14⁻BV510 (M5E2; BioLegend) and IgD‐FITC (Polyclonal; Southern Biotech) for 20 min at 4°C. Samples were washed with cold PBS with 2% FBS, resuspended in R10 media containing 0.25 μg mL⁻¹ 7AAD viability dye (Invitrogen) and acquired using a FACSAria III Fusion instrument. Single cell index sorting of Spike‐PE and Spike‐APC double positive IgG+ B cells was conducted simultaneously with data collection. Sorting was performed at the Biomedicum Flow Cytometry Core Facility, which receives funding from the Infrastructure Board at Karolinska Institutet. Cells were sorted dry into 96‐well PCR plates (Bio‐Rad), immediately transferred onto dry ice, and stored at −80°C until downstream use. Data was analysed using Flowjo version 10 (FlowJo Inc.). Representative gatings are shown in Supplementary figure 13A (sorting) and Supplementary figure 13B (quantitative analysis).

RNA from single Spike‐specific sorted memory B cells was extracted and reverse‐transcribed into cDNA with random hexamers using the Superscript III kit (Invitrogen) following the manufacturer's instructions. Single‐cell nested PCR was then performed using human IgG heavy‐chain VDJ primers previously reported by Doria‐Rose et al.,62 and using the HotStarTaq Plus Polymerase kit (Qiagen). Raw sequencing data from Sanger sequencing was filtered for quality using the R/Bioconductor package scifer (v 1.0.0) using default settings as described in Arcoverde Cerveira et al.63 Good quality sequences were aligned using IgDiscover (v 0.15.1)64, as implemented in the in‐house pipeline IgXplore,65 to the OGRDB AIRR‐C germline gene database.66, 67 Using 0 iterations and settings for single‐end reads, clonally related sequences were defined using the IgDiscover clonotypes model as those with the same V and J alleles, identical HCDR3 length, minimal 80% aa identity of HCDR3, and at least one junction without a mismatch.

IgG and IgM libraries for high‐throughput sequencing (HTS) by Illumina MiSeq 2 × 300 bp paired‐end sequencing were based on a 5‐prime multiplex PCR method. The RNA extraction was performed using Rneasy Mini Kit (Qiagen) following the manufacturer's instructions. The cDNA synthesis, multiplex PCR and Index PCR steps were performed following the previously described protocol and primers.68 The library pool and final sample preparation for the sequencing were performed using MiSeq Reagent Kit v3, 600 cycles (Illumina) following the protocol described in Vazquez‐Bernat et al.68

Reads processing was performed using the IgDiscover program (v. 0.15.1)64, with default settings, except for UMI length of 21 nucleotides and 0 iterations, and using the OGRDB AIRR‐C germline reference database.66, 67 To integrate the antigen‐specific Sanger sequences with the bulk repertoire, we used the antigen‐specific sequences to query the bulk IgG repertoire sequencing to trace their clonal relatives. To achieve that, we used the IgDiscover clonoquery and clonotypes modules.69 We first identified the clonotypes present among all antigen‐specific sequences identified through Sanger sequencing, using the IgDiscover clonotypes function with the same clonotype definition as mentioned previously. Subsequently, we used the IgDiscover clonotypes module outputs containing one representative sequence for each clonotype as our reference database of SARS‐CoV‐2 specific sequences. We queried clonal relatives of this database in bulk IgG/M HTS dataset using the clonoquery function. Total number of bulk BCR sequences, clonotypes, and clonal size identified in bulk sequencing libraries per individual are shown in Supplementary table 2 (S3).

The estimation of sample coverage interpolation and extrapolation was calculated using the vegan R package based on the clonal sizes of the merged datasets. The species richness estimation was calculated using the Chao1 and ACE1 algorithms with the vegan R package.70 The estimation was performed per individual study participant.70 To normalise for sequencing depth, clonal sizes were subsampled 100× to the lowest number of sequences per individual. The average per individual was used as input to estimate Hill numbers, and the Chao1 index was used for plotting.70

Antigen‐specific T cells were assessed by peptide stimulation followed by intracellular cytokine staining (ICS) as previously described.58 Briefly, PBMCs were thawed as described for memory B‐cell assessment. One to two million PBMCs were stimulated overnight at 37°C, 5% CO₂, with either 2 mg mL⁻¹ overlapping peptides (15‐mers overlapping by 11) covering the SARS‐CoV‐2 Spike protein (JPT), 1 mg mL⁻¹ Staphylococcal Enterotoxin B (Sigma‐Aldrich) as a positive control, or R10 with 0.8% DMSO as a negative control. Brefeldin A (Thermo Fisher) was added to all conditions for a final concentration of 10 mg mL⁻¹. Following stimulation, cells were washed twice with PBS, stained first with Live/Dead Fixable Blue or Live/Dead Fixable Aqua viability dyes (Invitrogen) for 5 min at 4°C followed by anti‐human CCR7‐BV421 (G043H7; Biolegend), CD8a⁻BV711 (RPA‐T8, Biolegend), CD4⁻PE‐Cy55 (S3.5; Invitrogen), CD45RA‐BV650 (5H9; Biolegend) for 20 min at 4°C. Following surface staining, cells were washed twice with PBS and permeabilised using the BD Cytofix/Cytoperm kit (BD Biosciences) for 20 min at RT, washed twice and stained intracellularly with anti‐human IL‐21‐AF647 (3A3‐N2.1; BD Biosciences), IL‐13‐PE (JES10‐5A2; BD Biosciences), IL‐2‐BV605 (MQ1‐17H12; BD Biosciences), IL‐17A‐BV785 (BL168; Biolegend), CD69⁻ECD (TP1.55.3; Beckman‐Coulter), CD3⁻APC‐Cy7 (SP34.2; BD Biosciences), IFNg‐AF700 (B27; Biolegend), and TNFa‐AF488 (Mab11; BD Biosciences) for 20 min at 4°C. Following ICS, cells were washed with BD 1× Fix/Perm buffer, resuspended in 1% PFA and acquired on an LSRFortessa flow cytometer (BD Biosciences). Data was analysed using Flowjo version 10 (FlowJo Inc.). Representative gatings are shown in Supplementary figure 14. The value for the unstimulated control was subtracted from the peptide stimulated condition and the result used as the final readout.

Group comparisons were performed using the Kruskal–Wallis test with Dunn's multiple comparisons post hoc with P value adjustment using Graphpad Prism v. 10.0 default settings (where using R, P‐value was adjusted for multiple comparisons using the Benjamini & Hochberg method for false‐discovery rate) when testing differences between three or more groups. The Mann–Whitney U test was used when comparing only two groups, or the Wilcoxon signed‐rank test for paired samples. Two‐tailed P‐value was used in all statistical tests comparing study groups. Spearman correlation with two‐tailed P‐value was used to analyse correlations between different parameters. Nonparametric tests were used as the low sample size did not permit the assumption of normal distribution.

Statistical analyses were performed using Graphpad Prism version 10.0 (La Jolla California, USA) or R (4.4.0), unless otherwise indicated in methods sections above. A P‐value < 0.05 was considered significant for all comparisons.

Multivariable linear regression was used to assess the association between transitional B (TrB) cell frequency and vaccine‐induced antibody responses. Log‐transformed antibody titres ED50 values were used to better align with linear regression assumptions. Multicollinearity among predictors was evaluated using variance inflation factors (VIF), with no evidence of collinearity observed. Given the small sample size and residual variability, heteroskedasticity‐consistent HC3 robust standard errors were applied.

The final model was specified as:

where time_days represents the time elapsed since transplantation (in days), immunosupp is a binary indicator of immunosuppressive treatment, and TrB denotes the percentage of transitional B cells among total CD19⁺ B cells.

The sequencing BCR data are deposited in the Swedish National Data Service (SND): https://doi.org/10.48723/bbag‐t656↗. Characteristics of the data are reported in Supplementary table 2 (S3). The data set contains sensitive personal data. Access to data can be requested via the SND research data catalogue. The request needs to be approved by the research principal and might need to undergo a confidentiality assessment before access can be released.