Four distinct participant cohorts were recruited between December 2022 and August 2024 (Supplementary Table 1). The BF.7/BA.5.2 convalescent cohort comprised adults who recovered from SARS-CoV-2 infection in Beijing between December 2022 and January 2023. Blood samples were collected 14–28 days post-recovery (n=45) and 6 months post-recovery (n=31). The XBB convalescent cohort included individuals infected with XBB lineage variants who recovered between June and July 2023. Blood samples were collected at 7 days (n=25) and 14–28 days (n=25) post-recovery. The JN.1 convalescent cohort consisted of individuals infected with the JN.1 variant between April and August 2024, with blood samples collected 14–28 days post-recovery (n=18). The vaccination cohort comprised people of the recombinant protein vaccine (a tetravalent COVID-19 vaccine containing trimeric spike proteins from SARS-CoV-2 Alpha, Beta, Delta, and Omicron BA.1 variants (43)), with blood samples collected at baseline (Day 0) and 21 days post-vaccination (n=15). This study employed convenience sampling during consecutive peak periods of the epidemic, and the sample size was consistent with established practices for exploratory immunological research on COVID-19.

All SARS-CoV-2 infections were confirmed by quantitative reverse transcription-polymerase chain reaction (qRT-PCR) or antigen testing of throat swabs. Infecting strains were determined based on local epidemiological surveillance data during transmission peaks when specific variants achieved clear dominance (> 90% prevalence; Supplementary Figure 1). All participants reported no respiratory symptoms within 3 months prior to their most recent SARS-CoV-2 infection. The study protocol received approval from the ethics committees of the National Institute for Viral Disease Control and Prevention, Chinese Center for Disease Control and Prevention. All participants provided written informed consent.

Peripheral venous blood samples were collected at predetermined time points. For serum preparation, blood samples were allowed to clot at room temperature for 30 min, then centrifuged at 3,000 rpm for 15 min. Serum was then aliquoted and stored at −80 °C until test. For peripheral blood mononuclear cell (PBMC) isolation, blood samples were collected in heparinized tubes and processed within 4 h after blood collection. PBMCs were isolated using Ficoll density gradient centrifugation (37). Blood samples were diluted and carefully layered onto Ficoll-Paque lymphocyte separation medium, then centrifuged at 800 g for 30 min at room temperature with acceleration/deceleration setting of 3/1. The PBMC layer at the plasma-medium interface was harvested, washed 1–2 times with RPMI 1640 medium (1,500 rpm, 10 min), resuspended, and counted. Fresh PBMCs were used immediately for T cell assays or cryopreserved in liquid nitrogen using fetal bovine serum (FBS) containing 10% dimethyl sulfoxide (DMSO).

Spike protein expression plasmids were constructed for key variants, including prototype (ancestral strain, refer to Wuhan-Hu-1), BF.7, XBB.1.5, JN.1, KP.3, XEC, NB.1.8.1, LP.8.1.1 and BA.3.2. Codon-optimized spike protein ectodomain sequences were designed based on variant-specific sequences, with T4 fibritin trimerization domains fused to the C-terminus to enhance protein stability, as previously described (44, 45). Site-directed mutagenesis was performed to introduce characteristic mutations for each variant, and all constructs were verified by DNA sequencing. Spike protein expression vectors utilized mammalian cell expression systems containing CMV promoters, multiple cloning sites, and SV40 poly-A signals. Appropriate tag sequences were incorporated at N- or C-termini to facilitate downstream purification and detection.

Pseudoviruses were generated using the vesicular stomatitis virus (VSV) pseudo typing system (46). HEK-293T cells were seeded in 10 cm culture dishes to reach 70–80% confluence, then transfected with the corresponding spike protein expression plasmid using lipofectamine transfection reagent. Twenty-four hours post-transfection, cells were infected with G protein-deficient VSV (VSV-ΔG-GFP) at a multiplicity of infection (MOI) of 3–5. Two hours post-infection, cells were extensively washed with phosphate-buffered saline (PBS) to remove residual inoculum. The culture medium was then replaced with Dulbecco’s Modified Eagle Medium (DMEM) containing 10% FBS, and anti-VSV-G antibody was added to a final concentration of 10 μg/mL. After 24 h, supernatants were harvested, filtered through 0.45 μm membranes to remove cellular debris, and stored in aliquots at −80 °C. Pseudovirus titers were determined using Vero E6-ACE2 cells. Infectious units were quantified by enumerating infected cells under fluorescence microscopy, with titers expressed as transducing units (TU) per mL.

Pseudovirus neutralization assays were performed following established protocols. Serum samples were heat-inactivated at 56 °C for 30 min to inactivate complement, then serially diluted 2-fold starting from 1:20. Equal volumes of diluted sera were mixed with standardized pseudovirus (approximately 1,000 TU) in 96-well plates and incubated at 37 °C with 5% CO₂ for 1 h to allow antibody-virus binding. Subsequently, 100 μL of serum-virus mixtures were transferred to 96-well plates pre-seeded with Vero E6-ACE2 cells. Following approximately 15 h of culture, TU were quantified using a CQ1 confocal imaging cytometer. The 50% pseudovirus neutralization titer (pVNT₅₀) was calculated by fitting the dose-response curve using four-parameter logistic regression analysis. Samples with pVNT₅₀ values below the detection threshold (< 20) were classified as seronegative and assigned a value of 10 for geometric mean titer (GMT) calculation. Immune escape fold change was calculated as the ratio of reference variant pVNT₅₀ (typically the homologous infection strain) to target variant pVNT₅₀.

Overlapping peptide pools covering S1 (101 peptides) and RBD (30 peptides) regions were synthesized based on variant-specific spike protein sequences. Peptides were 15–18 amino acids in length with 10 amino acid overlaps, adjusted according to the mutation patterns in different Omicron variants relative to the prototype sequence. Peptide purity exceeded 95%. Peptides were dissolved in DMSO to prepare 20 mg/mL stock solutions and diluted to working concentrations in culture medium before use. Convalescent PBMCs were cultured for 9 days using S1 peptide pools corresponding to their infecting variants, i.e., BF.7/BA.5.2 convalescents with BF.7/BA.5.2-S1 peptide pools, XBB convalescents with XBB-S1 peptide pools, and vaccine recipients with BA.5.2-S1 peptide pools (reflecting the vaccine’s early Omicron components). Subsequently, cultured cells were used to test specific T cells against S1 and RBD peptide pools from representative variants, including homologous strains.

Antigen-specific T cell responses were assessed using IFN-γ ELISpot kits (BD Biosciences, Franklin Lakes, NJ, USA). Pretreated 96-well ELISpot plates were coated with anti-human IFN-γ monoclonal antibody overnight at 4 °C. Following PBS washed, cells were blocked with RPMI 1640 complete medium containing 10% FBS. Cultured PBMCs (10⁵ cells/well) were seeded into pre-coated plates with corresponding peptide pools (final concentration 2 μg/mL). Negative control (medium only), positive control (phorbol 12-myristate 13-acetate, PMA, 5 μg/mL), and variant peptide pool stimulation groups were established. After incubation at 37 °C with 5% CO₂ for 18–24 h, cells were discarded and plates washed with water-Tween solution. Biotinylated detection antibodies were added and incubated at room temperature for 2 h. After PBS-Tween washes, streptavidin was added and incubated at room temperature for 1 h. Color development was performed using a substrate solution and terminated with deionized water. Following plate drying, spots were enumerated using an ELISpot reader (CTL Corp, Ohio, USA). Results were expressed as spot-forming cells per million PBMCs (SFCs/10⁶ PBMCs).

PBMCs were adjusted to 1×10⁶ cells/mL and stimulated with corresponding peptide pools (2 μg/mL) at 37 °C with 5% CO₂ for 1 h. Protein transport inhibitors brefeldin A (10 μg/mL) and monensin (2 μM) were then added, with incubation continued for 9–12 h. Post-stimulation, cells were harvested and stained with Live/Dead APC-Cy7 dye for viability assessment. Surface staining was performed using anti-CD3-FITC, anti-CD4-PerCP-Cy5.5, and anti-CD8-BV510 antibodies with incubation at 4 °C for 30 min. Cells were then fixed and permeabilized, followed by intracellular staining with anti-IFN-γ-PE-Cy7, anti-TNF-α-PE, and anti-IL-2-APC at 4 °C for 30 min. Following washes, cells were analyzed using a flow cytometer (FACSAria II, BD Biosciences), with at least 10,000 lymphocyte events collected per sample. Data analysis was performed using FlowJo software. Results were expressed as percentages of cytokine-positive cells within corresponding T cell subsets following the gating strategy as described previously (Supplementary Figure 2).

Antigenic cartography was constructed based on neutralization antibody data. Multidimensional scaling (MDS) was employed to reduce the high-dimensional neutralization data to a two-dimensional space, where antigenic distances correspond to immunological distances. Antigenic distance was defined as log₂ (relative neutralization titer), with relative neutralization titer calculated as the ratio of test strain pVNT₅₀ to reference strain pVNT₅₀. pVNT₅₀ values were first converted to antigenic distances, then MDS was used to project viral variants and serum samples into two-dimensional antigenic maps such that Euclidean distances between points optimally reflected actual antigenic relationships. Analysis was performed using the Racmacs package (https://acorg.github.io/Racmacs/) in R. Serum clustering in maps represents samples with similar antibody recognition profiles, while viral strain positions reflect their antigenic similarities. Radar plot analysis visualized overall neutralization patterns of different sera against variant panels. Neutralization profiles were plotted using GMT values for each variant as coordinate axis. Overall neutralization strength was quantified by area calculation, with cross-reactive response pattern assessed through shape analysis.

All variant sequences used for phylogenetic and antigenic analyses were obtained from NCBI Virus database (https://www.ncbi.nlm.nih.gov/labs/virus/) (Supplementary Table 2). Multiple sequence alignment and phylogenetic analysis were performed using the Clustal W algorithm. Phylogenetic trees were constructed based on full-length spike protein amino acid sequence using MEGA X.

All analyses were conducted using R software (version 4.3.0) and GraphPad Prism software (version 9.5). Neutralizing antibody titers were expressed as GMT with 95% confidence intervals (CI), and T cell responses were expressed as medians with interquartile ranges (IQR). For cross-sectional comparisons among multiple groups or variants, Friedman tests (non-parametric test for repeated measures or matched data) were used, followed by Dunn’s multiple comparison test for pairwise comparisons. For longitudinal paired data (e.g., 14–28 days vs. 6 months in the BF.7/BA.5.2 cohort), the Wilcoxon matched-pairs signed-rank test was employed. Escape fold changes were calculated as ratios of reference strain GMT to test strain GMT. Significance levels were denoted as “ns” (non-significant); *P < 0.05; **P < 0.01; ***P < 0.001; ****P < 0.0001. Specific statistical tests used for each analysis are indicated in the corresponding figure legends.

To characterize the evolutionary relationships among emerging SARS-CoV-2 Omicron variants, we conducted maximum likelihood phylogenetic reconstruction based on full-length spike protein sequences, which revealed the complex evolutionary trajectory from prototype to contemporary Omicron variants (Figure 1A). The evolutionary trajectory diagram demonstrated a detailed evolutionary trajectory step-wise progression of Omicron variants (Figure 1B). BF.7/BA.5.2, as descendants of BA.4, dominated initial Omicron epidemic wave in China during late 2022. This was followed by the successive emergence and spread of recombinant XBB lineages and the JN.1 series. Driven by the progressive accumulation of mutations in the spike protein, JN.1 and its descendant variants (e.g., NB.1.8.1, LP.8.1.1, and XEC) now achieve global ascendancy, a trend that underscores their rapid emergence and expansion under evolutionary selective pressure. The evolutionary diagram revealed that BA.2.86 harbored 34 spike protein mutations compared with BA.2, while BA.3.2 carried 52 mutations relative to BA.3, underscoring the substantial genetic divergence of these variants. Mutational analysis revealed striking patterns of convergent evolution, whereby phylogenetically distinct lineages independently acquired identical key mutations. Multiple variants convergently evolved the R346T mutation, appearing across BF.7, XBB, and KP.3 variants. Similarly, mutations at positions L452 (L452R or L452Q) and F486 (F486P or F486V) were repeatedly selected in later-emerging variants, including BF.7, BQ.1, and JN.1, indicating strong positive selection for these amino acid changes. The systematic accumulation of mutations, particularly in antigenic regions of the spike protein, established the basis for evaluating their impact on neutralizing antibody responses and T cell cross-reactivity in different convalescent populations.

BF.7/BA.5.2 convalescents represented the primary infected population during China’s early Omicron epidemic phase. To evaluate neutralization breadth and durability following BF.7/BA.5.2 infection, we collected sera from convalescents at 14–28 days and 6 months post-recovery for pseudovirus neutralization testing against multiple SARS-CoV-2 variants. Neutralizing antibody analysis demonstrated that 14–28 days post-recovery, sera exhibited robust neutralizing activity against the homologous BF.7 variant (GMT = 285) and responded effectively to prototype (GMT = 956). However, sera demonstrated substantial immune evasion by heterologous variants, with escape fold-changes ranging from 4.57-fold (LP.8.1.1) to 13.36-fold (KP.3). Neutralizing activity against all heterologous variants was significantly reduced compared with homologous strains (Figure 2A). Critically, 6-month follow-up data revealed more pronounced immune attenuation, with exacerbated immune evasion by heterologous variants. NB.1.8.1, LP.8.1.1, and BA.3.2 exhibited the strongest immune evasion at 6 months, with fold-changes exceeding 20 (Figure 2B). This indicated an accelerated decline in cross-variant neutralization capacity over time. Radar plot analysis visually demonstrated shifts in neutralization response patterns (Figure 2C). The GMT against heterologous strains was significantly lower in convalescent sera (14–28 days post-recovery), despite a broader neutralization breadth. Conversely, 6-month sera exhibited markedly narrowed neutralization spectra with generally reduced intensity, indicating limited neutralization capacity against Omicron variants.

Concurrently, we analyzed PBMCs from convalescents at 14–28 days and 6 months post-recovery using ELISpot techniques. Results demonstrated that 14–28 days post-recovery, all convalescents generated robust antigen-specific T cell responses (median > 2,000 SFCs/10⁶ PBMCs) upon stimulation with homologous strain (BF.7 or BA.5.2) S1 peptide pools. Importantly, stimulation with S1 peptide pools of other Omicron variants, including BA.2.75, BQ.1, BQ.1.1, and XBB, also triggered high-level cross-reactivity of comparable intensity (Figure 2D). Compared with mock controls, spot-forming cell counts stimulated by all viral antigens showed a significantly higher level, indicating establishment of extensive S protein-specific T cell immunity in convalescents, with no significant inter-variant differences. Six-month ELISpot results further confirmed T cell response durability. In contrast to substantial neutralizing antibody attenuation, T cell immune responses demonstrated a persistent trend. ELISpot analysis showed that S1 peptide pool responses remained elevated (2,500–3,500 SFCs/10⁶ PBMCs) with maintained cross-reactivity against different heterologous strains (including BA.2.75, CH.1.1, BQ.1, BQ.1.1, XBB, and XBB.1.5) (Figure 2E). Additionally, RBD peptide pool stimulation triggered significant responses (median > 1,000 SFCs/10⁶ PBMCs) (Figure 2F), indicating multiple conserved T cell epitopes within the RBD region capable of inducing long-term immune memory.

XBB convalescents represented a second ascent wave of Omicron-infected population in China during mid-2023. We employed identical approaches to assess adaptive immune responses induced by XBB breakthrough infection. Seven days post-recovery, sera demonstrated high neutralizing activity against the homologous strain XBB.1.5 (GMT = 479) while maintaining robust responses against BF.7 (GMT = 1,513) and the prototype (GMT = 509) (Figure 3A). Later-emerging variants exhibited obvious immune evasion of neutralizing antibodies, with fold-changes ranging from 3.82–12.56 (P < 0.05), among which NB.1.8.1 variant showed the highest escape fold-change (12.56). The 14–28 days follow-up data revealed that although XBB.1.5 titers decreased slightly (GMT = 353). Escape patterns paralleled 7-day data, with sera continuing to exhibit substantial evasion of JN.1 lineage strains, especially NB.1.8.1 (Figure 3B). Radar chart analysis intuitively demonstrated near-complete overlap between the two timepoints. Serum neutralizing antibody titers remained relatively stable during short-term recovery but still provided inadequate protection against newly emerging variants (Figure 3C).

Functional T cell response analysis revealed cross-reactive response characteristics similar to BF.7/BA.5.2 convalescents. ICS analysis following S1 peptide pool stimulation with representative strains (prototype, XBB.1.5, EG.5.1, and BA.2.86) demonstrated that multifunctional cytokine responses (IFN-γ, TNF-α, and IL-2) could be induced in CD4⁺ T cells, with heterologous strain response intensities not significantly different from homologous strain XBB.1.5 (Figure 4A). Cross-reactivity was also observed in CD8⁺ T cells. All tested strains effectively activated CD8⁺ T cells and enabled stable cytokine secretion (Figures 4B, C). PBMCs stimulated with the S1 peptide pools of the four strains all produced significant secretion of three cytokines, with IFN-γ and TNF-α as the predominant cytokines. These results again confirmed the broad spectrum and cross-reactivity of T cell immunity. T cell functional responses following XBB infection remained relatively stable under different stimulation conditions, indicating that activated T cell epitopes may be relatively conserved across different Omicron lineages.

We subsequently evaluated humoral immune response characteristics in individuals recently recovered from JN.1 infection. Cross-reactivity analysis against new variants revealed that JN.1 convalescent sera exhibited relatively mild immune evasion patterns against descendant lineage variants. Escape fold-changes ranged from only 1.10–2.08 (Figure 3D), significantly lower than levels observed in BF.7/BA.5.2 and XBB groups. Particularly, sera maintained robust neutralizing activity against several JN.1 descendant variants, including KP.3, XEC, and LP.8.1.1 (P > 0.05), with the descendant variant NB.1.8.1 exhibiting the highest escape degree (P < 0.05). This suggested that infection with recent variants may provide some degree of cross-reactivity against currently circulating variants (Figure 3D). Nevertheless, sera demonstrated limited cross-neutralization capability against the heterologous variant BA.3.2, still exhibiting obvious immune evasion (4.51-fold, P < 0.001), consistent with the results from other groups and further confirming the special status of BA.3.2 as the most antigenically distant variant in the study. Notably, serum neutralization titers against phylogenetically earlier variants decreased significantly, with GMT of only 162 against the prototype. However, the strongest neutralization response was observed against BF.7 (GMT = 614), potentially reflecting immune imprinting effects or mixed infection backgrounds.

Beyond natural infection convalescents, we also evaluated immune levels in recipients of a recombinant protein vaccine and systematically analyzed serum neutralizing antibodies and antigen-specific T cell responses in subjects on Day 0 and Day 21 post-vaccination. Pseudovirus neutralization assays revealed that 21 days post-vaccination, vaccine-induced neutralizing antibodies produced detectable responses against all tested strains, with GMTs significantly increased compared with pre-vaccination levels (Figure 5A). Serum exhibited the strongest neutralizing activity against prototype (GMT = 2,279) and BF.7 (GMT = 1,688), with GMTs exceeding 1,000, indicating that vaccines can induce robust humoral immunity against homologous or antigenically similar strains. Overall titers against various subsequent and currently prevalent Omicron variants remained retained (GMT range 50–150), and this demonstrated relatively balanced cross-reaction patterns without obvious immune evasion.

ELISpot detection of vaccine-induced T cell responses demonstrated that compared with mock controls, all vaccinated individuals generated robust IFN-γ responses upon stimulation by Omicron variant peptide pools (including BA.2.75, BA.5.2, BF.7, BQ.1, BQ.1.1, and XBB), indicating effective cellular immunity activation by vaccination. For the pre-vaccination (Figures 5B, C) and post-vaccination (Figures 5D, E), ELISpot comparisons demonstrated that stimulation with both S1 and RBD peptide pools could induce significant T cell response enhancement. Particularly, testing 21 days post-vaccination using S1 peptide pools showed that the median T cell response intensity was comparable to that in natural infection convalescents, reaching 3000–4000 SFCs/10⁶ PBMCs, with no significant differences among variants (Figure 5D). This further confirmed high conservation of T cell epitopes within the Omicron lineage. Vaccination successfully activated T cell immunity capable of broadly recognizing various variants.

To systematically evaluate humoral immune response characteristics induced by natural infection with different Omicron variants, we conducted comprehensive antigenic cartography analysis to quantify antigenic relationships between sera and viral variants. Radar chart analysis clearly revealed differences in strain specificity and cross-reactivity. Neutralization levels among the three convalescent groups followed the JN.1 group > XBB group > BF.7/BA.5.2 group (Figure 6A). This gradient was highly consistent with the relationship between infection timing and variant evolution. The BA.3.2 variant exhibited the most pronounced immune evasion across all groups, with consistently low neutralizing antibody levels.

Antigenic cartography displayed these relationships in two-dimensional space (Figure 6B). Clustering distribution patterns clearly showed evolutionary relationships, with prototype, BF.7, and XBB.1.5 in a central position, while JN.1 and its descendants in a relatively independent antigenic cluster, clearly separated from early BF.7 and XBB lineages. BA.3.2 variant was positioned most distantly from all other variants. Correspondingly, convalescent sera clustered according to their infection backgrounds. BF.7/BA.5.2 sera concentrated near prototype and BF.7, indicating narrow antigenic specificity with limited cross-reactivity against JN.1 lineage variants. XBB sera were intermediately distributed, while JN.1 sera clustered near both JN.1 lineage variants and early antigens, reflecting immune imprinting effects. Notably, BA.3.2 variant was located in marginal areas distant from all serum cluster centers. This indicates substantial antigenic divergence from currently circulating Omicron variants. This consistent escape pattern across all convalescent populations suggests that BA.3.2 possesses unique antigenic characteristics warranting continued surveillance.