Donor cohort characteristics and study design

To study the efficacy of mRNA COVID-19 vaccines in SLE, we enrolled donors who received two (Vax1 + Vax2) or three (Vax3) doses of monovalent (WA.1) BioNTech/Pfizer or mRNA-1273 (Moderna/National Institute of Allergy and Infectious Diseases (NIAID)) vaccines. The cohort consisted of patients with SLE (n = 79 patients; n = 10 prepandemic individuals, n = 69 vaccinees) and age- and sex-matched HD controls (n = 64 HDs; n = 8 prepandemic individuals, n = 56 vaccinees) collected between March 2021 and October 2022. The SLE group was enriched for female patients of Black ancestry, reflecting the race and sex bias of SLE and the demographics of the Atlanta metropolitan area (Supplementary Table 1). Owing to the restrictions imposed by the pandemic isolation requirements, the HD group had an underrepresentation of Black individuals relative to the SLE cohort. Samples were mainly cross-sectional collections, with longitudinal follow-ups as indicated in the diagrams (Supplementary Table 1 and Extended Data Fig. 1). Given that our study was designed to evaluate primary responses to mRNA vaccines, in the absence of any infection at a time when vaccine administration was erratic, longitudinal follow-up was limited to the fraction of patients who fulfilled these criteria. A total of 256 HD and 212 SLE samples were serologically evaluated for vaccine-induced antibody responses, and a total of 161 HD and 182 SLE paired peripheral blood mononuclear cell (PBMC) samples were assessed for antigen-specific B and T cell responses using high-dimensional flow cytometry combining antigen reactivity and deep immune profiling (Extended Data Fig. 2).

Lower seroconversion upon primary mRNA vaccination in patients with SLE

To assess the level of seroconversion in our two cohorts and to define the specificity, kinetics and persistence of circulating antibodies recognizing different portions of the mRNA-coded proteins, we performed an isotype-specific plasma screen28 against the following SARS-CoV-2 targets: S1 and S2 subunit domains of the spike protein, RBD of the S1 subunit, and the N-terminal domain (NTD) (Extended Data Fig. 3a). Nucleocapsid-specific IgG antibodies were also tested to exclude prior infections (Extended Data Fig. 3b). Overall, IgM responses were minor contributors to the spike reactivity in the HD cohort and were increased in the SLE cohort (Extended Data Fig. 3c). IgA-specific responses were largely induced against the spike domain and RBD (Extended Data Fig. 3d), with similar responses between the two groups. As previously reported, we observed a predominant IgG-mediated anti-spike and anti-RBD response (Extended Data Fig. 3e and Fig. 1a–g). Upon the administration of one vaccine dose (Vax1, 3–4 weeks), ~85% of HDs seroconverted, whereas patients with SLE had a lower seroconversion rate of ~58%, with significantly more negative/low responders (Fig. 1e and Extended Data Fig. 3e). Completion of the primary series of vaccines (Vax2) increased the seroconversion rate of both the SLE (88%) and HD (100%) groups, although reduced mean titers and overall increased number of negative/low responders remained among the vaccinees with SLE (Fig. 1f and Extended Data Fig. 3e). A booster dose (Vax3) normalized the mean titers of IgG RBD between the two cohorts (Fig. 1g).

Fig. 1: Serological evaluation of anti-spike vaccine-mediated antibody responses.

a–d, Luminex-based detection of RBD IgG-binding serum antibodies (net MFI values) in the HD and SLE cohorts, shown for each vaccine administration. Each dot represents a sample. Connecting lines show longitudinal collections. Comparisons between mRNA vaccines (BioNTech/Pfizer (aqua) and Moderna/NIAID (salmon)) are shown for Vax1 + Vax2 in the HD (a) and SLE (c) cohorts and for Vax3 in the HD (b) and SLE (d) cohorts. e–g, Clusters of IgG RBD titers based on binned time points for samples collected at Vax1 (e), Vax2 (f) and Vax3 (g). Statistical analysis was performed with a two-sided Mann–Whitney U test and indicated when significant. Pie charts show the distribution of seronegative (MFI 0–2,500) and seropositive (low MFI 2,500–10,000, medium MFI 10,000–100,000 and high MFI >100,000) values. The number of samples is indicated in the pies, and the percentage of responders was compared using a chi-square test. *P < 0.05; **P < 0.01; ***P < 0.001; ****P < 0.0001. MFI, mean fluorescence intensity; GMT, geometric mean titer; Pre-CoV, before the coronavirus pandemic; d, day(s); mo, month(s).

While the RBD of the S1 subunit has been described as a main target of broad nAbs, other non-RBD structural proteins of the spike, namely S2 and NTD, can harbor neutralizing epitopes. The S2 domain, harboring more conserved epitopes and some level of cross-reactivity with other coronaviruses, was targeted by IgG similarly in HD and SLE samples (Extended Data Fig. 3e), whereas the IgG reactivity toward the NTD was lower in the SLE cohort, potentially compromising the control of viral escape29 (Extended Data Fig. 3e). Preexisting immunity to the four seasonal common-cold coronaviruses (CCCs)—alpha coronavirus strains ‘HKU1’ and ‘229E’ and beta coronavirus strains ‘OC43’ and ‘NL63’—was similar at baseline, with the only detectable difference being the higher anti-OC43 IgG titers in SLE (Extended Data Fig. 3f). Whether this observation could be explained by either more frequent or prolonged seasonal infections in patients with SLE remains to be elucidated. Overall, the evaluation of serological responses to mRNA vaccination revealed a defective primary response in SLE that requires vaccine boosters for full seroconversion.

Reduced RBD-specific antibody competitiveness and neutralization in SLE

The functional activity of circulating RBD-specific antibodies was determined by a competitive ELISA to assess their efficiency in blocking the interaction of recombinant human RBD with its receptor, recombinant human angiotensin-converting enzyme 2 (ACE2) (Fig. 2a,b)30. Interestingly, while the overall anti-RBD titers were only mildly reduced in patients with SLE (Fig. 1), this group displayed significantly impaired ability to block ACE2 binding across most time points (Fig. 2a,b). Despite medium/high RBD IgG titers, SLE samples were more frequently enriched for either non- or low-competitive antibodies (Fig. 2c), which are usually low-avidity IgG antibodies30,31. To test the hypothesis that impaired ACE2-blocking activity could result from defective affinity maturation of B cell responses in SLE, we tested antibody avidity using surface plasmon resonance (SPR)32. Comparisons of off-rate values confirmed that patients with SLE were enriched for anti-RBD immunoglobulins with medium/low avidity, with the greatest impairment observed after Vax2; this was only partially rescued after the booster dose (Fig. 2d). Of note, at Vax2, the absence of detectable RBD binding (Fig. 2d, nonbinder) was detected exclusively in a fraction of patients with SLE treated with belimumab.

Fig. 2: Reduced neutralization and breadth in the cohort of vaccinated patients with SLE.

a, ELISA determination of antibody-mediated inhibition of SARS-CoV-2 RBD binding to solid-phase ACE2. The graph shows the reciprocal plasma or serum dilution that blocks 80% binding (BD80) of RBD to human ACE2. log(BD80) values are shown as negative (0–1), low (1–2) and high (>2). Box plots represent the minimum to maximum values, showing all points as individual serum samples: HD (day 0, n = 5; 1 week pre-2nd, n = 14; Vax2 1–3 months, n = 19; Vax2 4 months–before Vax3, n = 23; Vax3, n = 48) and SLE (day 0, n = 8; 1 week pre-2nd, n = 16; Vax2 1–3 months, n = 47; Vax2 4 months–before Vax3, n = 59; Vax3, n = 51) Statistical analysis was performed using a two-sided Mann–Whitney U test. b, Pie graphs showing the frequency and statistical comparison of competitive immunoglobulins in the two cohorts. A chi-square with Fisher’s test was used for comparisons. The number in the circles indicates the total number of samples tested, whereas the numbers in the pies show the relative percentages of the negative (black), low (gray) and high (white) values. c, Graphs showing the linear correlation between the blocking of RBD binding to ACE2 (BD80) and the total RBD immunoglobulin-binding antibodies in the same sample, tested from vaccinated individuals from both the HD (left graph) and SLE (right graph) cohorts. d, Polyclonal antibody avidity (as measured by the dissociation off-rate per second) to the SARS-CoV-2 RBD protein at ~2–5 months after the second vaccination (Vax2) or ~3–5 months after the third mRNA vaccination (Vax3) for serum samples analyzed by SPR. Off-rate constants were determined from two independent SPR runs. The table shows the frequency of responders for each cohort and time point analyzed. An unpaired t test was applied. e–g, Pseudoviral neutralization in vitro assay performed on plasma samples isolated from vaccinated individuals after Vax2 and Vax3. The graphs show the neutralizing titers inhibiting 50% of the viral growth (NT50) tested for the SARS-CoV-2 WA.1 wild-type (e), Delta (B.1.617.2) (f) and Omicron (B.1.1.529 BA.1) (g) strains. Each dot in the box plots represents an individual sample tested. Horizontal lines indicate the median. The pie charts show the comparison of negative, low and high neutralizers. Statistical comparison was performed using a chi-square with Fisher’s test. The number in the circles indicates the total number of individual samples tested, whereas the numbers in the pies show the relative percentages of the negative (black), low (gray) and high (white) values. Not significant, P > 0.05; *P < 0.05; **P < 0.01; ***P < 0.001; ****P < 0.0001. HRP, horseradish peroxidase; amIgG, anti-mouse IgG; mFc, monomeric Fc; rec-hRBD, recombinant human RBD; rec-hACE2, recombinant human ACE2; d0, day 0; wk, week(s); pre-2nd/3rd, before the second or third dose.

To corroborate the ELISA results further, we performed pseudoviral neutralization assays33 against the SARS-CoV-2 WA.1 (wild-type), Delta (B.1.617.2) and Omicron (B.1.1.529 BA.1) strains (Extended Data Fig. 4a,b). Patients with SLE displayed impaired neutralization activity against those viruses, with significantly lower neutralization titers and reduced breadth, relative to their HD counterparts (Fig. 2e–g and Extended Data Fig. 4b). Interestingly, neutralizing titers decayed significantly faster in HDs relative to patients with SLE, with significant differences for the WA.1 wild-type (half-life of 38 days for the HD group versus 73 days for the SLE group; P = 0.03) and Delta (half-life of 39 versus 68 days; P = 0.03) strains after Vax3 (day 42 after vaccination and onwards)34. These data identify qualitatively defective and lower-avidity circulating antibodies resulting in reduced breadth of neutralization in patients with SLE.

Defective and distinct anti-spike B cell responses in SLE

To evaluate the magnitude of cellular responses and the identity of B cells induced upon mRNA vaccination, we performed immunophenotypic profiling with a high-dimensional 28-color flow panel that provided tetramer-based detection of WA.1 spike- and RBD-reactive B cells (Extended Data Fig. 2a and Fig. 3a,b). During the post-Vax1 priming phase, 62% of the HDs mounted an early anti-spike B cell response (Fig. 3c), with ~35% of anti-Spike B cells also binding to the RBD tetramer (Fig. 3d). Overall, the HD group displayed a higher and persistent anti-spike and anti-RBD recall response to Vax2 and Vax3 (Fig. 3c,d). The SLE group had a significantly lower proportion of responders after priming and within memory recall responses, with ~10–30% of patients failing to generate a detectable B cell response at any time (Fig. 3c,d). B cell lymphopenia is common in SLE owing to both disease activity and therapy. To account for this variable, we compared the frequencies of the total CD19+ B cells between the HD and SLE groups (Extended Data Fig. 4c). Both total and antigen-specific circulating B cells were reduced in patients with SLE, with enrichment of patients with very low B cell frequency (<1%) and low spike-reactive B cells (<0.0022%) (Extended Data Fig. 4c,d). Normalization of antigen-specific reactivity to B cell numbers confirmed that the SLE cohort carried more negative responders than the HD cohort (Extended Data Fig. 4e,f). Preexisting immunity to CCCs in the B cell memory compartment did not significantly influence the anti-SARS-CoV-2 B cell response, as measured by the level of cross-reactive WA.1+/CCC+ spike B cells (Extended Data Fig. 5a–c). Furthermore, this response was not notably different between the two cohorts (Extended Data Fig. 5c). Contrary to the greater durability of their antibody response, the frequency of spike-reactive B cells declined more rapidly in the SLE group after Vax2 (half-life; 95% confidence interval, calculated decay P = 0.009) but was similar to that in the HD group after Vax3. Together, these results suggest that B cell defects in SLE are responsible for reduced mRNA vaccine efficacy.

Fig. 3: Lower magnitude of antigen-specific memory B cells in the vaccinated SLE cohort.

a, Cartoon showing the ex vivo tetramer-based detection of spike- and RBD-reactive B cells and high-dimensional flow immunoprofiling of B cells from PBMCs. b, Representative fluorescence-activated cell sorting (FACS) plots showing the gating strategy applied to characterize the total CD19+CD20+ B cells (excluding the CD20−CD38hi plasma cells) binding to dual-tetrameric spike probes and tetrameric RBD probes. c,d, Quantification of the total spike-specific (c) and RBD-specific (d) B cells shown as the frequency of CD20+ B cells in the HD and SLE cohorts. Each dot represents an individual sample tested at baseline (day 0) and after receiving one (Vax1), two (Vax2) or three (Vax3) vaccine doses. Differences among groups were analyzed using multiple-group comparisons by nonparametric Kruskal–Wallis statistical testing using Dunn’s post hoc analysis in GraphPad Prism. Comparisons using pie charts and a chi-square with Fisher’s test are shown. *P < 0.05; **P < 0.01; ***P < 0.001; ****P < 0.0001. S, spike; bio–SA, biotin–streptavidin; PC, plasma cell.

Anti-spike CD27− populations persist during the memory phase and expand in SLE

Despite the increasingly recognized complexity of these compartments, previous studies have largely concentrated on plasmablast and CD27+ memory responses. As different effector and memory pathways and the participation of IgD−CD27− double-negative (DN) B cells may be induced in different SLE types24,35, we sought to interrogate further the expanded pool of antigen-reactive B cells and define their dynamics. The distribution of DN subsets (DN1–DN4) is an indicator of the origin and function of the corresponding B cells. Specifically, we have previously associated DN1 B cells with conventional CD27+ memory pathways36 and shown that they represent a large majority of DN cells in healthy individuals. In contrast, DN1 cells represent a much lower fraction in acute SLE and severe COVID-19 infections, in which DN2 and DN3 B cells are dominant. In these acute situations, DN2 and DN3 cells are considered extrafollicular naive-derived effector cells23,24,25,37,38. However, little is known about the contribution of DN subsets to effector and memory vaccination responses and specifically to mRNA vaccination.

Using unsupervised PaCMAP (pairwise controlled manifold approximation) and FlowSOM tools, we determined the complexity of the global B cell compartment and defined the participation of multiple B cell clusters in the anti-spike/RBD response (Extended Data Fig. 6a and Fig. 4a). In SLE, the major differences were driven by populations of B cells typically associated with the disease (for example, activated B naive, DN2, DN3 and 9G4-expressing cells24,35; Extended Data Fig. 6a and Fig. 4a). In particular, within the DN B cells, RBD−spike+ B cells in the SLE group were significantly enriched for DN2/DN3 clusters (#4, 5, 25) over DN1 (cluster #11) (Extended Data Fig. 6a). Increased DN2 and spike++ DN2 populations were shared among patients with SLE, with greater frequency in those of Black ancestry (Extended Data Fig. 6b).

Fig. 4: Greater DN2 expansion in the vaccinated SLE cohort.

a, PaCMAP and FlowSOM representations of spike++CD20+ B cells from HDs (n = 126) and SLE donors (n = 161). Samples were combined from Vax1 + Vax2 + Vax3. b, Representative FACS plots showing the characterization of spike-reactive CD20+ B cell subsets based on the expression of IgD and CD27. CD21 and CD11c markers are used to define the DN subsets further. Individual samples from the HD (Vax1, n = 23; Vax2 1–3 months, n = 22; Vax2 >3 months, n = 34; Vax3 1–3 months, n = 37; Vax3 >3 months, n = 13) and SLE (Vax1, n = 20; Vax2 1–3 months, n = 45; Vax2 >3 months, n = 51; Vax3 1–3 months, n = 24; Vax3 >3 months, n = 20) cohorts. c, Bar graphs showing the relative frequency of spike++ B cell subsets based on IgD and CD27 expression in the HD and SLE cohorts. Vertical lines indicate the s.e.m. A two-sided Mann–Whitney U test was used to calculate the significance of the SLE group compared to the HD group. d, Relative frequency of spike++ B DN cell subsets in vaccinees from the HD and SLE groups. Vertical lines indicate the s.e.m. A two-sided Mann–Whitney U test was used to calculate the significance of the SLE group compared to the HD group. e, Pie charts showing comparisons of the average sum for DN1 versus non-DN1 (DN2 + DN3 + DN4) spike++ B DN cells. A chi-square with Fisher’s test was used for significance testing. f, Reactivity of DN subsets among nonresponders and responders. A chi-square test was used for statistical comparisons. The LOS was based on median values of baseline + 2 × s.d. g, PaCMAP and FlowSOM data representing the level of expression of CXCR3 on clusters of spike++CD20+ B cells as in a. h, Dot plots representative of the CCR6 and CXCR3 expression of the total and spike++ B cells. The bar graphs show the distribution of CCR6- and CXCR3-expressing spike-reactive B cells of the HD and SLE cohorts. Individual samples from the HD (Vax1, n = 23; Vax2 1–3 months, n = 22; Vax2 >3 months, n = 34; Vax3 1–3 months, n = 37; Vax3 >3 months, n = 13) and SLE (Vax1, n = 20; Vax2 1–3 months, n = 45; Vax2 >3 months, n = 51; Vax3 1–3 months, n = 24; Vax3 >3 months, n = 20) cohorts. Vertical lines indicate the s.e.m. A two-sided Mann–Whitney U test was used to calculate the statistical significance of the B cell subset populations in the SLE cohort compared to HD frequencies, as shown in the SLE graphs. i, Pie charts showing the comparison of the total CXCR3+spike++ and CXCR3−spike++ B cells, as well as relative frequencies. A chi-square with Fisher’s test was used for comparisons. When indicated, the LOD was set to logarithmic 0.001 for B cells and 0.003 for T cells. The LOS was based on median values of baseline + 2 × s.d. *P < 0.05; **P < 0.01; ***P < 0.001; ****P < 0.0001. Pac blue, Pacific blue; Unsw, unswitched; B mem, B memory.

We also used a supervised two-dimensional flow cytometry approach to measure the distribution of spike-reactive B cells among previously identified parental populations defined by the expression of IgD and CD27: naive, CD27+ unswitched memory, CD27+ isotype-switched memory and IgD−CD27− DN cells. DN B cells were further fractioned into four specific subsets (DN1–DN4)23,24,39 determined by the expression of CD21 and CD11c (full gating strategy; Extended Data Fig. 2a and Fig. 4b). The CD20+spike+ B cell responses to Vax1 included in both cohorts similarly large fractions of naive B cells (~40%) that rapidly contracted over subsequent vaccinations (Fig. 4c). The IgD+ unswitched memory cells also contributed to a small portion of the total antigen-induced population that rapidly contracted (Fig. 4c). The initial priming responses included significant fractions of early CD27+ switched memory cells and DN cells, the latter representing naive-derived effector responses as we had previously reported for autoreactive B cells in SLE and severe COVID-19 infections25,35,37. Notably, DN cells dominated the spike response early in SLE and remained dominant in both groups before booster vaccination, presumably reflecting memory cells induced by Vax1 and effector/activated memory cells induced by Vax2 (ref. 40). In both groups, while conventional CD27+ memory cells dominated the response following booster vaccination, DN cells represented a large fraction of the memory responses (Fig. 4c,d). In this study, DN1 strongly dominated the spike-specific responses in HDs across all time points. In contrast, in patients with SLE, DN2 cells dominated the early response to Vax1 and Vax2 and remained significantly higher relative to the HD cohort in the post-Vax3 period (Fig. 4d,e). Moreover, DN2 cells were enriched in low spike responders (below the limit of sensitivity (LOS)) (Fig. 4f). DN3 cells also contributed to different phases of the response and were significantly expanded at late time points after Vax2 and Vax3 in the SLE group (Fig. 4d).

The expression of C–C chemokine receptor type 6 (CCR6) and C–X–C motif chemokine receptor 3 (CXCR3) on spike-reactive B cells was also qualitatively different between the two cohorts (Fig. 4g–i). In the HD cohort, we observed a predominance of CCR6+spike+ B cells that expanded within subsequent vaccinations (Fig. 4g–i). In contrast, in patients with SLE, spike+ B cells were highly enriched for CXCR3+ cells (Fig. 4g–i). We further analyzed the contribution of several B cell isotypes to the anti-spike responses by examining their immunoglobulin surface expression with flow cytometry (Extended Data Fig. 6c–g). IgG responses dominated the spike reactivity (Extended Data Fig. 6c,d,g), whereas IgM and IgA memory B cells represented a smaller portion of this response (Extended Data Fig. 6c–f). Interestingly, IgM+ memory B cells were increased in SLE, particularly in the early post-Vax2 and Vax3 phases, accounting for >20% of all antigen-reactive CD20+ B cells in several patients (Extended Data Fig. 6e). Additionally, in the SLE cohort, spike reactivity was associated with a smaller and delayed expansion of IgG+ B cells upon recall doses (Extended Data Fig. 6g). Collectively, these data indicate that patients with SLE mount a diverse anti-B cell immunity upon mRNA vaccination, with greater expansion of DN2/DN3 B cells also persisting during the establishment of memory responses.

Impaired activation and persistence of anti-spike lupus T cells

We also sought to define the magnitude, kinetics and differentiation of CD8+ and CD4+ T cells upon mRNA vaccination. To this end, we used an in vitro system suitable for scoring the frequency of antigen-reactive T cells and that relies on the incubation of cells with megapools of peptides and allows their quantification using the activation-induced marker (AIM) assay6,41 (Fig. 5a and Extended Data Fig. 7a,b). Cells were stimulated with megapools of spike overlapping peptides spanning the entire protein (WA.1, spike aa 5–1,273) or with peptides from the hemagglutinin (HA) H1N1 (A/California/04/2009) protein, an unrelated control for viral T cell reactivity. Twenty-four hours after incubation, samples were analyzed to score the magnitude of antigen-reactive AIM+ cells using a combination of two surface markers (CD69 and 41BB for CD8+ T cells and CD40L and OX40 for CD4+ T cells). T helper CD4+ cells were classified based on their expression of chemokine receptors, and their polarization was further investigated in both the CXCR5− or CXCR5+ (referred to as ‘AIM+ circulating T follicular helper (cTFH)’) compartments (Fig. 5b and Extended Data Fig. 7b).

Fig. 5: Lower T cell reactivity in patients with SLE receiving SARS-CoV-2 mRNA vaccines.

a, Schematic showing the 24-h AIM assay-based detection of antigen-reactive T cells upon incubation of PBMCs with a megapool of spike-derived peptides and the flow cytometric analysis of surface-expressed markers of activation and immune profiling. b, Representative FACS plots showing the gating strategy applied to characterize the AIM+ spike-reactive CD8+ (41BB+CD69+) T cells or AIM+ spike-reactive CD4+ (OX40+CD40L+) T cells and AIM+ spike-reactive cTFH (CXCR5+ of AIM+CD4+) cells among the CD3+ T cells. c–e, Scatter plots showing the frequency of spike-specific AIM+ T cells quantified at each indicated time point in the HD and SLE cohorts for AIM+CD8+ T cells (c), AIM+CD4+ T cells (d) and AIM+CD4+ cTFH (e) cells. The vaccination time points in c–e indicate the following binned time points (T), as indicated in Extended Data Fig. 1a: 0 (T0, baseline), 1 (T1–T3), 2 (T4–T5), 3 (T6–T7), 4 (T8–T9), 5 (B1–B5) and 6 (B6–B9). The number of samples is indicated as ‘Total (n)’. f, PaCMAP and FlowSOM representations of AIM+CD8+ T cells (HD, n = 137; SLE, n = 163) and AIM+CD4+ T cells (HD, n = 136; SLE, n = 169) from the HD (n = 126) and SLE (n = 161) cohorts from combined Vax1 + Vax2 + Vax3. A total of 15 clusters are indicated in the plots, and the relative marker expression and classification of the clusters are shown in Extended Data Fig. 7f,g. g, Representative dot plots showing the differentiation of AIM+ T cells using CD45RA and CCR7 expression. h, Bar plots showing the distribution of the four subsets (TN/TSCM, TCM, TEM and TEMRA) of AIM+CD8+ T cells. Individual samples from the HD (Vax1, n = 17; Vax2, n = 47; Vax3, n = 43) and SLE (Vax1, n = 12; Vax2, n = 65, Vax3, n = 38) cohorts. i, Bar plots showing the distribution of the four subsets (TN/TSCM, TCM, TEM and TEMRA) of AIM+CD4+ T cells. Individual samples from the HD (Vax1, n = 19; Vax2, n = 54; Vax3, n = 48) and SLE (Vax1, n = 20; Vax2, n = 81, Vax3, n = 45) cohorts. A two-sided Mann–Whitney U test was applied to compare each subset of T cells between the SLE and HD groups, and significance is shown in the SLE bars. When indicated, the LOD was set to logarithmic 0.001 for B cells and 0.003 for T cells. The LOS was based on median values of baseline + 2 × s.d. Vertical lines indicate the s.e.m. *P < 0.05; **P < 0.01; ***P < 0.001; ****P < 0.0001. MP, megapool; TN, naive T cells; TSCM, stem cell-like memory T cells; TCM, central memory T cells; TEMRA, TEM cells reexpressing CD45RA.

Lower T cell responses were observed in the SLE cohort, with impaired priming of AIM+CD8+ T cells (Fig. 5c) and lower levels of both AIM+CD8+ and AIM+CD4+ T cells at the late memory and recall phases (Fig. 5c–e and Extended Data Fig. 7c–e). The overall fold reduction in the SLE group compared to the HD group was 6.71 (AIM+CD8+ T cells) and 3.15 (AIM+CD4+ T cells) at Vax2 and 3.85 (AIM+CD8+ T cells) and 2.92 (AIM+CD4+ T cells) at Vax3.

Unsupervised PaCMAP and FlowSOM cluster analyses were informative of significant differences in terms of T cell polarization, with T cells in the SLE cohort characterized by a reduction in effector memory T (TEM) cell subsets, also expressing CCR4 and/or CCR6 (CD8+ T cell clusters #8 and 12–14, CD4+ T cell clusters #1–3, 7 and 8) (Fig. 5f and Extended Data Fig. 7f,g). Two-dimensional flow cytometry comparisons confirmed that the SLE anti-spike T cell responses were less represented in the TEM pool over three doses of vaccine (Fig. 5g–i) and were distinguished in their SLE cTFH compartment, with an initial skewing of primed AIM+ cTFH cells into CCR6+ TFH17 cells (Extended Data Fig. 8a) and a general reduction in the magnitude of the SLE AIM+ cTFH pool upon memory responses, especially after Vax2 (Fig. 5e).

Notably, HA-reactive memory T cells were detected at normal levels in patients with SLE (Extended Data Fig. 8b), although the AIM+ HA CD8+ pool still failed to generate normal TEM cells (Extended Data Fig. 8c), suggesting that CD8+ T cell defects might be mediated by associated lupus-disease defects42. These results suggest that anti-spike T cell defects in SLE might be largely a result of inefficient mRNA activity and/or a property of vaccine- versus infection-induced T cell immunity.

Poor vaccine-mediated responses associate with an SLE extrafollicular immune signature

To investigate the influence of B cell endotypes on different vaccine responses, we used the total RBD-reactive IgG levels to classify all vaccinees into three groups of responders: negative/low (VNL), medium (VM) or high (VH) vaccine responders (Fig. 6a). As expected, RBD IgG titers directly correlated with the potency of WA.1 neutralization (Fig. 6b). Vaccinees with SLE who received one or two vaccine doses were enriched for negative/low responders, and their neutralizing IgG potency upon Vax2 was still significantly lower than that in HDs. While the booster dose improved the responses in the SLE cohort, only 78% reached high antibody titers relative to 97% in the HD cohort (Fig. 6a). We then tested the hypothesis that, in our SLE cohort, a predisposition to strong extrafollicular immune responses might be responsible for lower immunogenicity of the mRNA vaccine. To address this question, we assessed patients for circulating cellular surrogates of extrafollicular activity identified in previous studies of SLE and acute COVID-19 infections25,35,37. Based on these studies, we created an ‘extrafollicular activity score’ that included markers proposed to represent decreased GC activity (low B DN1 frequency) and low activated cTFH (act cTFH, CXCR5+PD-1+CD38+) frequency, together with markers of extrafollicular activation, including increase in B DN2 and circulating plasmablasts, as well as expansions of CXCR5− peripheral activated T helper (act TPH, CXCR5−PD-1+CD38+) cells43,