Identification and characterization of bone marrow plasma cells producing IFN-γ-neutralizing autoantibodies in adult-onset immunodeficiency

Summary

Adult-onset immunodeficiency (AOID) can be associated with anti-interferon (IFN)-γ autoantibodies (AIGAs). Rituximab (RTX) reduces circulating B cells but often fails to eliminate AIGAs, suggesting the presence of long-lived antibody-secreting cells (ASCs) in immune-privileged sites. Here, we identified 23 distinct IFN-γ-specific monoclonal antibodies (mAbs) from bone marrow (BM) ASCs of AOID patients using microwell array chip technology and single-cell RNA sequencing. Competitive assays demonstrated that neutralizing mAbs disrupt IFN-γ engagement with IFN-γR1 or IFN-γR2, impairing JAK-STAT1 signaling. Structural analysis of neutralizing mAb A01BM-03 revealed its binding to a quaternary epitope on dimeric IFN-γ, disrupting IFN-γR2 interaction. Notably, clonally expanded IFN-γ-specific ASCs were localized within a CD11c⁺ZEB2⁺ age-associated B cell (ABC)-like plasma cell cluster in one AOID patient, implicating its role in AOID pathogenesis. These findings provide direct evidence for IFN-γ-specific ASCs in the BM despite RTX treatment, supporting targeted ASC therapy to restore immune homeostasis in AOID.

Graphical abstract

Highlights

• •IFN-γ-specific mAbs were identified from bone marrow antibody-secreting cells
• •Neutralizing mAbs disrupt IFN-γ engagement with IFN-γR1 or IFN-γR2
• •IFN-γ-specific ASCs localize within a CD11c⁺ZEB2⁺ plasma cell cluster
• •Targeting these ASCs represents a potential novel therapeutic approach

Wang et al. identify IFN-γ-specific antibody-secreting cells (ASCs) in the bone marrow of adult-onset immunodeficiency patients, revealing functional and structural mechanisms underlying the pathogenic anti-IFN-γ autoantibody and supporting targeted ASC therapy.

Introduction

Adult-onset immunodeficiency (AOID) is a heterogeneous condition that can be associated with persistently elevated plasma levels of anti-interferon (IFN)-γ autoantibodies (AIGAs), predisposing affected individuals to a spectrum of opportunistic infections, particularly disseminated nontuberculous mycobacterial (NTM) and Talaromyces marneffei infections.^1,2,3,4,5 Among NTM species, both rapidly growing mycobacteria, such as Mycobacterium abscessus, and slowly growing species, including Mycobacterium avium–Mycobacterium intracellulare complex (MAC), are frequently isolated from AOID patients and are associated with a range of clinical manifestations.^6,7,8,9 Epidemiological studies suggest a striking racial and ethnic predisposition to AOID, with most reported cases occurring in individuals of Asian descent.^4,6,9 This observation suggests a multifactorial etiology involving both environmental and genetic factors.^{5,10,11,12,13}

The pathogenic role of AIGAs in AOID is attributed to their ability to neutralize IFN-γ by interfering with its binding to receptors IFN-γR1 and IFN-γR2, thereby blocking downstream JAK-STAT1 signaling critical for host immune responses, as well as to mediate antibody-dependent cellular cytotoxicity (ADCC) targeting IFN-γR1/2-expressing cells.¹⁴ Previous studies reported that AIGAs predominantly recognize the C-terminal region of IFN-γ (amino acids 121–131, SPAAKTGKRKR), which exhibits molecular mimicry to the ribosomal assembly protein Noc2 of Aspergillus spp.^15,16 Additionally, competition assays with murine IFN-γ-specific monoclonal antibodies (mAbs) revealed that AIGAs exhibit heterogeneous neutralizing activity, with certain clones, particularly those targeting the B27 epitope, effectively preventing IFN-γ receptor binding and signaling.^17,18 Notably, characterization of 19 IFN-γ-specific mAbs derived from peripheral memory B cells of AOID patients identified three distinct non-overlapping binding sites, each with unique mechanisms of action. These mechanisms include blocking IFN-γ binding to IFN-γR1, preventing IFN-γR1–IFN-γR2 heterodimerization, and promoting ADCC via immune complex formation.¹⁴ Collectively, these findings suggest that pathogenic AIGAs exacerbate mycobacterial infections by blocking IFN-γ and eliminating immune cells through recognition of diverse epitopes on IFN-γ.

Current treatment for AOID involves a combination of long-term multi-drug antimicrobial therapy for NTM infections and immunomodulation with glucocorticoids, immunosuppression such as cyclophosphamide, and rituximab (RTX), a CD20-targeting mAb that depletes B cells responsible for AIGA production.^9,19 While this regimen often achieves clinical remission, a subset of patients continues to exhibit high circulating autoantibody titers despite prolonged RTX therapy and sustained depletion of peripheral B cells.^9,19,20,21 Relapse following RTX therapy for systemic lupus erythematosus (SLE), immune thrombocytopenia (ITP), and rheumatoid arthritis (RA) has been associated with residual renal B cells, rituximab-resistant splenic memory B cells, and splenic and bone marrow (BM) plasma cells (PCs).^{22,23,24,25,26,27} The mechanisms underlying relapse after RTX therapy in AOID remain poorly understood.

In this study, we sought to identify and characterize 23 distinct IFN-γ-specific mAbs in the BM antibody-secreting cells (ASCs) of AOID patients using a combination of microwell array chip technology and single-cell RNA sequencing (scRNA-seq). Of these, six mAbs demonstrated potent neutralizing activity by disrupting IFN-γ binding to IFN-γR1 and IFN-γR2, thereby inhibiting IFN-γ-induced macrophage activation and polarization—an essential process for intracellular pathogen clearance, including NTM infections.^28,29 Notably, one mAb, A01BM-03, exhibited neutralizing efficacy comparable to emapalumab, a Food and Drug Administration (FDA)-approved human IFN-γ-specific mAb.³⁰ Crystal structure analysis of A01BM-03 in complex with dimeric IFN-γ at 3.02 Å resolution revealed that A01BM-03 binds a quaternary conformational epitope spanning two IFN-γ protomers and imposes steric hindrance at the IFN-γR2 binding site, providing a structural basis for its potent neutralizing activity. Furthermore, scRNA-seq analysis revealed that IFN-γ-specific ASCs in the BM of an RTX-treated patient were predominantly found within CD11c⁺ZEB2⁺ BM-PCs, which resemble a distinctive B cell subset known as age-associated B cells (ABCs).^31,32,33 ABCs have been implicated in BM homeostasis, autoimmune pathogenesis, and age-related immune dysregulation.^31,34 In conclusion, our findings provide direct evidence for the presence of IFN-γ-specific ASCs in the BM of AOID patients and offer potential mechanistic insights into the sustained presence of AIGAs despite RTX treatment.

Results

Rebound of plasma AIGA titers in an RTX-treated AOID patient

To investigate potential sources of these residual B cells, we enrolled a cohort of 22 AOID patients at Peking Union Medical College Hospital between October 2020 and June 2024. The cohort comprises 8 males and 14 females, with symptom onset ages ranging from 14 to 65 years and a duration from symptom onset to diagnosis of AIGAs ranging from less than 1 to 8 years (Table S1). Peripheral blood samples were obtained from all patients during their initial visit. As shown in Figures 1A and 1B, there was considerable variability in the binding and neutralizing activities of plasma AIGAs among patients at enrollment, consistent with previous reports and indicative of heterogeneous disease status and prognosis.^36,37 Half-maximal binding titers (ED₅₀) to recombinant IFN-γ, measured by enzyme-linked immunosorbent assay (ELISA), ranged from 737 to 35,804, which were significantly higher than those observed in patients with infectious diseases (IDs), including those with isolated pulmonary NTM infections (all less than 45) and healthy donors (all less than 57) (p < 0.001, Figure 1A). Similarly, half-maximal neutralization titers (NT₅₀) against IFN-γ, determined by assessing the inhibition of IFN-γ-induced JAK-STAT1 signaling in THP-1 cells, varied from 22 to 2,240. In contrast, neither ID patients nor healthy donors exhibited detectable IFN-γ neutralizing activity (p < 0.001, Figure 1B). Correlation analysis between binding and neutralizing activities indicated that they were positively correlated (R² = 0.77, p < 0.001; Figure 1C).

Figure 1.

Rebound of plasma IFN-γ-specific autoantibody titers in patient A01 despite RTX treatment

(A and B) Plasma IFN-γ-specific binding and neutralizing titers in a local cohort of AOID patients during their initial visit. Binding titers (A) were measured via ELISA, while neutralizing titers (B) were assessed using a THP-1 cell-based neutralization assay. Data represent mean ± SEM from two independent experiments, with red dots indicating titers for patients A01, A18, and A19. ID, infectious disease. Statistical comparisons across cohorts were performed using Mann-Whitney U test, with corresponding p values indicated. The same statistical method was consistently applied in Figures 2 and 4 for comparative analyses.

(C) Correlation analysis between binding antibody titers and neutralizing antibody titers in this patient cohort using linear regression.

(D–F) Longitudinal analysis of plasma IFN-γ-specific binding (blue) and neutralizing (green) titers, RSV-F-specific binding titer (orange), peripheral B cell counts (red), and the timing of treatments, including RTX (orange bar), plasmapheresis (purple bar), and bone marrow aspiration (magenta bar), along with ceftazidime, cyclosporin, glucocorticoids, and anti-NTM therapy in patients A01 (D), A18 (E), and A19 (F). Part of data in (D) was originally reported in Nie et al.³⁵

Patient A01, a 54-year-old female at enrollment, was diagnosed with infections caused by MAC and Burkholderia cenocepacia and subsequently received multimodal treatment that included RTX, plasmapheresis, glucocorticoids (prednisone and methylprednisolone), ceftazidime, and anti-mycobacterial agents (rifampicin, ethambutol, and clarithromycin) over a 90-week follow-up period (Figure 1D).³⁷ Clinical remission in patient A01 was initially accompanied by effective peripheral B cell depletion and a marked reduction in both binding and neutralizing autoantibody titers until approximately week 24 post-admission. The cervical abscess caused by B. cenocepacia gradually resolved and was healed by approximately week 36. However, during subsequent follow-up to 90 weeks, both titers (ED₅₀/NT₅₀) rebounded, reaching levels of 7,466/1,681 at week 52; 10,679/1,899 at week 72; 10,822/2,167 at week 76; and 16,933/5,787 at week 90, comparable to the pretreatment values (14,308/2,062). These rebounded titers ranked at the upper end within our cohort (Figures 1A and 1B). Moreover, plasma antibody binding titers against viral antigen respiratory syncytial virus (RSV) F protein exhibited a trend highly similar to that of IFN-γ-specific autoantibody titers. Concurrently, peripheral B cell counts increased from undetectable levels to 18/μL at week 74 and 119/μL at week 90, despite initiation of a second RTX course at week 50. During this time, the patient remained free of abscess recurrence, and the rebound in AIGA titers did not temporally coincide with infectious relapse. These findings suggest that RTX, while effective in reducing peripheral B cells, does not fully eliminate the B cell populations responsible for AIGA production.

Identification of IFN-γ-specific ASCs in the BM of both RTX-treated and treatment-naïve AOID patients

To identify additional sources of AIGA production, we focused on the BM, which serves as a rich reservoir of long-lived PCs that secrete and sustain antibodies present in the plasma.^38,39 BM aspirates from the iliac crest were additionally collected from three individuals (A01, A18, and A19). Notably, patient A01 was diagnosed with AOID 7 years after symptom onset and had been managed with an RTX-containing regimen, while patients A18 and A19 were diagnosed within 1 year and were receiving antimicrobial therapy for NTM. Among the three patients who consented to provide BM samples via iliac crest aspiration, patient A01 underwent BM aspiration at week 76 post-enrollment, coinciding with an autoantibody-rebound phase despite receiving a second RTX course at week 50. In contrast, BM samples from patients A18 and A19 were collected at week 4 and 1 post-enrollment, respectively, prior to RTX treatment for patient A18, while patient A19 remained untreated for any B cell depletion therapy (Figures 1E and 1F). To isolate and systematically characterize IFN-γ specific ASCs in the BM, we employed a combination of microwell array chip technology and scRNA-seq (Figure 2A). Our previously developed Modular Superhydrophobic Microwell Array (MoSMAR) Chip⁴⁰ enabled the detection of IFN-γ-specific ASCs surrounded by IFN-γ-coated microspheres, which were visualized via fluorescence microscopy and selected using a custom-designed single-cell distribution instrument (SCDI).⁴¹ The selected single ASCs were processed for single B cell antibody gene amplification, expression, and characterization.

Figure 2.

Identification of bone marrow plasma cells (BM-PCs) secreting IFN-γ-specific autoantibodies in patients A01 and A19

(A) Schematic representation of the experimental workflow for isolating and characterizing BM-PCs secreting IFN-γ-specific autoantibodies using MoSMAR-chip and scRNA-seq. Figure was created with BioRender.com. aa, amino acid.

(B) Representative fluorescence microscopy images of four BM-PCs secreting IFN-γ-specific autoantibodies from patient A01, detected using recombinant IFN-γ-coated microspheres in MoSMAR-chip. Scale bars, 10 μm.

(C and D) Variability in the binding and neutralizing activities of 21 IFN-γ-specific mAbs derived from BM-PCs and PB-MBCs of patient A01 and BM-PCs from patient A19, together with emapalumab as a positive control, as measured by ELISA and a THP-1-cell-based neutralization assay. Data are representative of two independent replicates.

(E) Binding affinity of neutralizing (NAbs, IC₅₀ < 1 μg/mL, red dots) and non-neutralizing (NNAbs, IC₅₀ > 1 μg/mL, blue dots) mAbs from patients A01 and A19 to recombinant IFN-γ, assessed via SPR. Green dot denotes emapalumab. Data are shown as mean ± SEM.

(F–L) Fold changes in IFN-γ-stimulated expression of CD80 (F), HLA-ABC (G), iNOS (H), HLA-DR (I), and CXCL10 (J) in THP-1-derived macrophages, as well as HLA-DR (K) and CXCL10 (L) in monocytes from healthy donors’ PBMCs, in the presence of NAbs and NNAbs at 50 μg/mL, relative to unstimulated controls. Data are shown as mean ± SEM from a representative experiment of two independent replicates.

From patients A01, A18, and A19, we isolated 11, 2, and 28 IFN-γ-reactive ASCs, respectively, from approximately 50,000 CD38⁺ and CD138⁺ magnetically enriched BM-PCs (Figure 2B). However, only 5 of 11 mAbs from A01 and 15 of 28 mAbs from A19 were successfully cloned and expressed, each exhibiting distinct sequences and binding activities to IFN-γ (Figure 2C), whereas no successful clones were obtained from A18. Additionally, one mAb, A01PB-01, was obtained from approximately 2,000 CD19⁺ IgD⁻ memory B cells in the peripheral blood (PB-MBCs) of A01 (at week 90) through B cell immortalization, whereas no IFN-γ-specific mAbs were detected in the PB-MBCs of patients A18 and A19.

The neutralization capacity of these mAbs was assessed based on their ability to inhibit IFN-γ-induced JAK-STAT1 signaling in THP-1 cells. Five out of 21 mAbs demonstrated >99% neutralizing activity against IFN-γ-induced STAT1 phosphorylation at 50 μg/mL, of which three were from A01 (A01PB-01, A01BM-03, and A01BM-04) and two were from A19 (A19BM-13 and A19BM-18). The estimated half-maximal inhibitory concentration (IC₅₀) values ranged from 0.04 to 0.33 μg/mL, with A01BM-03 exhibiting the highest potency (IC₅₀ = 0.04 μg/mL), surpassing the efficacy of emapalumab, an FDA-approved human anti-IFN-γ monoclonal antibody (IC₅₀ = 0.06 μg/mL) (Figure 2D). The remaining 15 mAbs were classified as non-neutralizing antibodies (NNAbs) due to either undetectable neutralizing activity or IC₅₀ values exceeding 1 μg/mL (Figure 2D). No statistically significant differences were observed in binding activities (K_D) between NAbs and NNAbs measured by surface plasmon resonance (SPR) (p > 0.05, Figure 2E), suggesting binding alone is not sufficient to determine neutralizing activities. However, the most potent mAb, A01BM-03, exhibited the strongest binding to recombinant IFN-γ (K_D = 0.017 nM), exceeding the affinity of emapalumab (K_D = 0.062 nM) and aligning with its superior neutralizing activity.

Next, we examined the impact of antibody neutralization on IFN-γ-induced activation of THP-1-derived macrophages by measuring surface expression of human leukocyte antigen (HLA)-ABC, CD80, inducible nitric oxide synthase (iNOS), HLA-DR, and CXCL10 secretion levels. As shown in Figures 2F–2J, the presence of 50 μg/mL NAbs resulted in a significantly greater reduction in median fluorescence intensity and CXCL10 concentration compared to NNAbs, suggesting inhibition of IFN-γ-induced M1 macrophage polarization, a critical process for intracellular pathogen clearance.^28,29 Similarly, NAbs could also block IFN-γ-mediated upregulation of HLA-DR and CXCL10 in monocytes derived from peripheral blood mononuclear cell (PBMCs) of healthy donor (Figures 2K and 2L). Overall, these findings indicate the presence of IFN-γ-specific ASCs in the BM of both RTX-treated (A01) and RTX-naïve (A19) patients. Neutralizing activity of these IFN-γ-specific autoantibodies can disrupt IFN-γ-induced activation of macrophages, potentially leading to impaired clearance of intracellular pathogens such as NTM.

ABC-like plasma cell clusters are shared across AOID patients, with IFN-γ-specific ASCs localized to distinct subsets

To further characterize the IFN-γ-specific ASC population in the BM, we performed 5′-based scRNA-seq of CD38⁺ and CD138⁺ magnetically enriched BM-PCs from three donors. Following stringent quality control, including the retention of only cells containing paired immunoglobulin heavy and light-chain sequences, we obtained transcriptome data from 3,500 BM-PCs (488 from A01; 2,061 from A18; and 951 from A19). Dimensional reduction and unsupervised clustering revealed minimal expression of MS4A1 (encoding CD20) across all BM-PCs (Figure 3A), indicating that RTX is unlikely to be effective against these cells. In contrast, the expression profile of canonical-PC-associated genes—including upregulated CD38, SDC1 (encoding CD138), TNFRSF17 (encoding BCMA), CXCR4, PRDM1 (encoding BLIMP1), XBP1, JCHAIN, MZB1, IRF4, and TNFRSF13B (encoding TACI), together with downregulation of CD19 and PAX5—was consistent with a typical PC transcriptional program (Figures 3A and S2K). Unsupervised clustering at a resolution of 0.5 identified eight distinct clusters (Figure 3B), demonstrating overall similarity and consistent cluster stability across the three donors (Figures 3C, S2I, and S2J).

Figure 3.

Clonal expansion of BM-PCs secreting IFN-γ-specific autoantibodies in patient A01

(A) Feature plots displaying the differential expression of B cell marker genes in BM-PCs (N = 3,500) derived from patients A01 (N = 488), A18 (N = 2,061), and A19 (N = 951). The redder the dot, the higher the log-normalized gene expression. See also Figure S2.

(B) Uniform manifold approximation and projection (UMAP) embeddings of 3,500 BM-PCs derived from patients A01, A18, and A19, with eight clusters identified via graph-based clustering with a resolution of 0.5 and represented in distinct colors.

(C) Cluster repartition of BM-PCs in each donor.

(D) UMAP plot highlighting BM-PCs with immunoglobulin sequences having a high identity to those of IFN-γ-specific autoantibodies, identified using microwell array chip and B cell immortalization. Cluster 3 contains 23 red dots representing A01-derived BM-PCs, while cluster 2 includes one dark blue dot from patient A19.

(E) Phylogenetic analysis of 45 IFN-γ-specific autoantibodies derived from patients A01 and A19, including sequences identified via scRNA-seq. The phylogenetic tree highlights clonal expansion of BM-PCs in patient A01, featuring germline heavy-chain IGHV1-46 paired with germline light-chain IGKV2-40. Each line represents paired heavy and light chains, with black lines denoting NAbs. Dot colors indicate germline gene usages: IGHV1-46 (red), IGHV3-74 (blue), IGKV2-40 (orange), and IGLV6-57 (green).

(F) Visualization of the heavy-chain CDR3 amino acid sequences in cluster 3, displayed as a word cloud, where the size of each word corresponds to the frequency of the respective sequence.

(G) Comparison of somatic hypermutation (SHM) levels in the heavy-chain variable region of IFN-γ-specific and unspecified BM-PCs within cluster 3 (N = 494).

(H) Clone abundance distribution in the BM-PC repertoire of each donor, featuring the top 150 clonotypes ranked by clone size. Statistical comparisons across donors were performed using Kolmogorov-Smirnov test.

(I and J) Clone size distribution across donors (I) and clusters (J), colored according to size category: large (red, N > 20), medium (yellow, 5 < N ≤ 20), small (blue, 1 < N ≤ 5), and single (gray, N = 1). See also Figure S3.

We further analyzed antibody repertoires from scRNA-seq data to assess clonal relationships with antibodies isolated via microwell array chip technology and B cell immortalization. Hierarchical clustering was performed by grouping amino acid sequences with identical immunoglobulin heavy-chain (IgH) V-J gene usage and CDR3 length, followed by applying a CDR3 identity threshold based on Hamming distance (Figure 2A). This clustering of IgH sequences identified 23 BM-PCs in A01 and one BM-PC in A19, exhibiting high similarity to experimentally isolated antibodies. The 23 BM-PCs from A01 (designated A01BM-bc01 to A01BM-bc23) were clonally expanded, sharing 100% heavy-chain sequence identity among themselves, 90.4% (104/115) with PB-MBC-derived A01PB-01, 63.8% (81/127) with BM-PC-derived A01BM-04, and 61.0% (75/123) with BM-PC-derived A19BM-13, all of which shared IGHV1-46 usage and exhibited potent IFN-γ-neutralizing activity (Figures 2D and 3E). The single heavy-chain sequence identified in A19 (A19BM-bc01) exhibited 95.0% (115/121) identity with A19BM-05, an IFN-γ-specific binding but non-neutralizing BM-PC-derived antibody (Figure 3E). Two representative mAbs, A01BM-bc01 and A19BM-bc01, were synthesized and confirmed to possess IFN-γ-binding activities (Table S2). Phylogenetic analysis revealed that all 23 BM-PCs within the A01BM-bc01 lineage shared identical heavy- (IGHV1-46) and light-(IGKV2-40) chain germlines, as well as identical heavy- and light-chain sequences, except for A01BM-bc12, A01BM-bc17, and A01BM-bc19, where five amino acid substitutions were identified in their light-chain sequences (Figure 3E). Furthermore, 7 out of 16 (43.8%) of BM-PC-derived mAbs of A19 exhibited the same heavy (IGHV3-74) and light (IGLV6-57) germlines, although minor sequence variations were found among them, suggesting some levels of somatic hypermutation occurred among these mAbs. Among the 21 mAbs isolated using the microwell array chip technology and B cell immortalization, the IgG1 isotype accounted for 71.4% (15/21). while IgG2 comprised 28.6% (6/21). Of the 23 mAbs identified by scBCR-seq, IgG1 represented 95.7% (22/23) and IgG2 4.3% (1/23) (Table S2). For comparison, Shih et al. reported an IgG subclass distribution of 73.3% IgG1 (11/15) and 26.7% IgG3 (4/15), highlighting substantial inter-individual variability in IFN-γ-specific antibody isotypes.¹⁴ When projected onto the transcriptomic landscape (UMAP), the 23 newly identified sequences, named as the A01BM-bc01 clonotype, clustering with A01PB-01 were primarily localized within cluster 3, whereas the single sequence named as A19BM-bc01 grouping with A19BM-05 was mapped to cluster 2 (Figure 3D). Notably, within cluster 3, the A01BM-bc01 clonotype was the dominant heavy-chain CDR3 sequence (VRDLGRYFDR, Figure 3F) and exhibited a significantly higher somatic hypermutation (SHM) rate of 13.9% compared to the remaining cluster 3 cells (Figure 3G). In contrast to A01BM-bc01, none of the other top six clonotypes by size in A01’s BM demonstrated concentrated distribution in cluster 3 (Figure S3). Global and cluster-level characterization of the BM-PC repertoires highlighted notable inter-donor and inter-cluster heterogeneity. Clone abundance distribution in the BM-PC repertoire of each donor, featuring the top 150 clonotypes ranked by clone size, revealed a markedly distinct curve in A01 compared with the other two patients (Figure 3H). Quantification of intra-patient clonotype proportions further showed a lower frequency of single clones in A01, indicating a more pronounced clonal expansion that may reflect inter-individual differences in disease course or severity (Figure 3I). Consistently, cluster 3 also displayed an exceptional degree of clonal expansion compared to other clusters (Figure 3J).

Cluster 3 exhibited a transcriptional profile distinct from previously described BM-PC subsets (Figure 3A). Differential expression analysis revealed upregulation of ITGAX (encoding CD11c) and ZEB2, hallmark genes of ABCs.^{32,33,42,43,44} These findings suggest that cluster 3 represents a distinct ABC-like PC subset. We next analyzed additional PBMCs to compare the proportions of ABCs and double-negative 2 (DN2) B cells between patients with AOID, ID, and healthy donors (Figure S4). No significant differences in the frequencies of ABCs or DN2 B cells were observed among PBMCs from AOID (N = 15), ID (N = 4), and healthy donors (N = 4) (Figure S4B). Notably, donors A01 and A18 exhibited low frequencies of ABCs in PBMCs following rituximab (RTX) treatment (Figure S4C). Unfortunately, PBMCs from donor A19 were unavailable for this analysis due to prior exhaustion of the sample.

Collectively, these findings demonstrate that both untreated (A19) and rituximab-treated (A01) AOID patients harbor BM-PCs producing AIGAs, including a clonally expanded and CD11c⁺ZEB2⁺ ABC-like PC subset within the BM of patient A01 and a single IFN-γ-specific PC in A19. However, the absence of significant differences in ABC-like cells within PBMCs among AOID, ID, and health donors suggest the peripheral and BM compartments are distinct. The origin and development trajectory of CD11c⁺ZEB2⁺ ABC-like PC subset in A01 is complex and warrant further investigation.

IFN-γ-specific NAbs disrupt IFN-γ binding to its receptors IFN-γR1 and IFN-γR2

To elucidate the mechanism of action of IFN-γ-specific NAbs, we evaluated their competitive binding activity using SPR against a panel of isolated IFN-γ-specific mAbs, including emapalumab as a positive control. Additionally, two representative mAbs, A01BM-bc01 and A19BM-bc01, were selected based on scRNA-seq analysis of BM-PCs from donors A01 and A19, respectively. These mAbs were synthesized and validated for their binding affinity and neutralizing activity before inclusion in the characterization panel (Table S2). Competitive binding analysis of the seven NAbs against the full panel of 23 IFN-γ-specific mAbs together with emapalumab generated a heatmap that distinctly categorized the mAbs into two major groups (Figure 4A). Group 1 mAbs exhibited competitive binding with NAbs A01PB-01, A01BM-04, and A01BM-bc01 but showed no competition with NAbs A01BM-03, A19BM-18, or emapalumab. Conversely, Group 2 mAbs displayed an opposite competition profile. Notably, A19BM-08 and A19BM-11 exhibited competitive binding with both group 1 and group 2 NAbs and were therefore classified as non-group-1/2 mAbs (Figure 4A) but failed to demonstrate substantial neutralizing activity against IFN-γ-induced activation (Figure 2D). These findings suggest the presence of at least two distinct neutralizing epitopes on IFN-γ, each recognized by a different subset of NAbs, indicating distinct mechanisms of neutralization.

Figure 4.

IFN-γ-specific neutralizing autoantibodies compete with IFN-γ receptors for binding to recombinant IFN-γ

(A) Classification of IFN-γ-specific mAbs into group 1 and group 2, based on competitive binding with six BM-PC-derived NAbs and emapalumab (names shown in red), measured via SPR. Results are presented as a heatmap of percent competition, defined as the relative inhibition of the secondary antibody binding to the primary antibody/IFN-γ complex.

(B–E) Schematic representation and percent competition of group 1, group 2, and non-group-1/2 mAbs with either IFN-γR1 for binding to IFN-γ (B, C) or with IFN-γR2 for binding to the IFN-γ/IFN-γR1 complex (D, E). NA, not applicable. Data are shown as mean ± SEM.

Since IFN-γ exerts its function by interacting with its heterodimeric receptors, IFN-γR1 and IFN-γR2, we next evaluated the ability of mAbs from group 1 (N = 11) and group 2 (N = 11) to compete with these receptors for IFN-γ binding using surface plasmon resonance (SPR) (Figure 4B). Competition analysis revealed that group 1 NAbs (N = 4) exhibited average 93.2% ± 3.9% competition with IFN-γR1, whereas NNAbs in this group (N = 7) showed variable but consistently lower competition levels, averaging about 49.0% ± 23.9% (Figure 4C). In contrast, all of group 2 (N = 11) and non-group-1/2 mAbs (N = 2) demonstrated only weak competition (less than 40%) with IFN-γR1. The weak binding affinity between the wild-type IFN-γR1/IFN-γ complex and IFN-γR2 previously precluded competition assays between antibodies and IFN-γR2. The screening of the IFN-γR1 F05 mutant by Menzoda et al. enabled these critical studies.⁴⁵ Similarly, group 2 NAbs (N = 3), including the positive control emapalumab, exhibited average 97.3% ± 6.5% competition with IFN-γR2, while NNAbs in this group (N = 8) displayed significantly lower competition activity, averaging around 73.5% ± 5.2% (Figures 4D and 4E). Since most group 1 mAbs strongly block the binding of IFN-γ to IFN-γR1, their competition with IFN-γR2 for binding to the pre-formed IFN-γ/IFN-γR1 complex could not be accurately measured. Among the non-group-1/2 mAbs, only A19BM-11 demonstrated detectable competition (74.6%) with IFN-γR2, while A19BM-08 shows minimal interference with either receptor. Although certain NNAbs partially block receptor binding, these distinct receptor-binding competition patterns suggest that group 1 NAbs neutralize IFN-γ by disrupting IFN-γR1 engagement, whereas group 2 NAbs achieve neutralization by interfering with IFN-γR2 interaction.

Structural insights into IFN-γ neutralization by A01BM-03

To elucidate the structural basis of IFN-γ neutralization, we resolved high-resolution crystal structures of the Fab fragment of the potent NAb A01BM-03 of group 2 in complex with dimeric IFN-γ (PDB ID: 9VNP) at a resolution of 3.02 Å in space group P2₁2₁2 (Table S3). Structural analysis reveals that each IFN-γ homodimer engages two A01BM-03 Fab molecules (Figure 5A). The A01BM-03 Fab binds IFN-γ via a quaternary conformational epitope spanning two IFN-γ protomers, involving residues E75, D76, N78, V79, N83, S84, N85, K86, K87, R89, and D90 from one protomer and K34, N35, W36, and K37 from the other (Figure 5B). The corresponding paratope on A01BM-03 consists of seven heavy-chain residues, R31 (HCDR1); D54 and K57 (HCDR2); and S101, V102, Y103, and I104 (HCDR3), along with eight light-chain residues, including Y28, A30, S31, Y32, and Y33 (LCDR1); E51 and D52 (LCDR2); and T69 from the light-chain framework region 3 (LFR3) (Figure 5C). Notably, only three of the seven heavy-chain residues form hydrogen bonds or salt bridges with IFN-γ (Figure 5D; Table S4), whereas six of the eight light-chain residues do (Figure 5E; Table S4). We conducted alanine-scanning mutagenesis experiments on IFN-γ to identify key residues involved in antibody recognition. Among the mutations tested, only K86A and K87A—both forming hydrogen bonds with the antibody light chain—reduced A01BM-03 binding to IFN-γ by more than 10-fold (based on K_D values, Figure S5). These findings highlight the dominant role of the A01BM-03 light chain in IFN-γ recognition.

Figure 5.

Structural basis for IFN-γ neutralization by the IFN-γ-specific autoantibody A01BM-03

(A) Structure of the IFN-γ/A01BM-03 Fab complex (PDB ID: 9VNP), superimposed on the IFN-γ/IFN-γR1/IFN-γR2 complex (PDB ID: 6E3K). A01BM-03 Fab heavy and light chains are shown in orange and blue, respectively. IFN-γ protomers are displayed in green and cyan and IFN-γR1 and IFN-γR2 in yellow and pink, respectively. This colors scheme is consistent across panels.

(B and C) Epitope and paratope of A01BM-03, with paratope residues within 4 Å of IFN-γ shown as stick models and individually labeled. Epitope across two IFN-γ protomers are highlighted in white. The interactive surface between IFN-γ and IFN-γR2 is outlined in pink. Residues K86 and K87, whose critical role in A01BM-03 binding was confirmed by mutagenesis, are labeled in bold. See also Figure S5.

(D and E) Key residues mediating molecular interactions between the heavy- (orange) and light-chain (blue) variable regions of A01BM-03 and IFN-γ protomers are shown. Hydrogen bonds and salt bridges are depicted as magenta dashed lines, determined via the PISA program.

See also Table S4.

Structural comparison with the recently resolved IFN-γ/IFN-γR1/IFN-γR2 complex (PDB ID: 6E3K, 3.25 Å resolution) reveals that A01BM-03 binding induces a steric clash, predominantly between the A01BM-03 heavy chain and IFN-γR2 (Figure 5A). The overlapping residues recognized by A01BM-03 and IFN-γR2 include E75, D76, V79, N83, S84, and K86 (Figure 5B). This structural overlap suggests that A01BM-03 directly competes with IFN-γR2 for IFN-γ binding, consistent with competition with IFN-γR2 measured by SPR (Figure 4E). Collectively, these insights demonstrate that A01BM-03 neutralizes IFN-γ signaling by sterically blocking IFN-γR2 interaction, thereby preventing receptor-mediated signal transduction.

Discussion

The presence of AIGAs in AOID patients despite RTX treatment poses a significant challenge for therapeutic efficacy and prognosis. Using a combination of microwell array chip technology and scRNA-seq, we identified IFN-γ-specific ASCs in the BM of both RTX-treated (A01) and RTX-naïve (A19) AOID patients. Molecular cloning and characterization of these IFN-γ-specific mAbs revealed their strong binding and neutralizing activities against IFN-γ. Notably, the NAbs effectively disrupted IFN-γ binding to its receptors, IFN-γR1 and IFN-γR2, thereby inhibiting IFN-γ-induced activation. This interference may impair M1 macrophage polarization, a critical process for intracellular pathogen clearance, such as the control of NTM infections.^28,29

We do acknowledge that the small number of patients and the BM samples analyzed here limits the ability to fully capture inter-patient heterogeneity and to generalize the transcriptional and clonal features of IFN-γ-specific ASCs across the broader AOID population, particularly related to the CD11c⁺ZEB2⁺ ABC-like PC subset. Nonetheless, the identification of IFN-γ-specific ASCs in both A01 and A19 patients supports the existence of shared pathogenic mechanisms that warrant further validation in larger cohorts. The precise type and number of B cells responsible for the observed rebound in autoantibody titers in patient A01 remain uncertain. While both CD20⁺ memory B cells in peripheral blood and CD20⁻ BM-PCs could contribute, the latter likely play a dominant role. While it is expected that some of the IFN-γ-specific memory B cells such as A01PB-01 may differentiate into PCs upon activation; their depletion following rituximab treatment, however, likely limits their contribution to relapse compared to the CD20⁻ BM-PCs, which can continue secreting IFN-γ-specific autoantibodies despite treatment.

Consistent with previously published dataset, the light-chain V-gene IGLV6-57 was frequently utilized by IFN-γ-specific mAbs in our study (Table S2). We compared the antibody germline usage between group 1 and group 2 mAbs defined in our study and the three categories described by Shih et al.¹⁴ The epitope targeted by site II mAbs (residues 37–62 and 87–98) overlaps with that of group 2 mAb A01BM-03. Specifically, both site II and group 2 mAbs exclusively employ the IGLV6-57 light-chain V gene, suggesting they likely recognize similar epitopes. Notably, whereas site II mAbs reported by Shih et al. were derived from the peripheral memory B cells, our group 2 mAbs originated from BM-PCs, indicating that B cells encoding pathogenic autoantibody genes reside in both compartments. However, without direct competition assays, we remain uncertain whether our group 1 mAbs overlap with site I or site III epitopes defined by Shih et al.¹⁴

The scRNA-seq analysis of BM-PCs further revealed that a clonally expanded subset of AIGA-secreting cells was predominantly localized within the ABC-like PC cluster, which is characterized by elevated CD11c and ZEB2 expression.^42,46 This unique transcriptional profile provides a partial explanation for the sustained presence of AIGAs in the plasma despite RTX treatment. ABC-like cells have been implicated in BM dynamics and autoimmune pathogenesis and are known to accumulate with age.^42,47 Their expansion has been reported in several autoimmune diseases, including SLE,^46,48,49 RA,^42,50 and multiple sclerosis (MS).⁵¹ Additionally, increased ABC-like cell frequencies have been observed in patients with acquired aplastic anemia, correlating with elevated IFN-γ levels, underscoring their role in disease immunopathology.⁵² Further investigation into the ontogeny and functional regulation of ABCs in BM physiology and autoimmunity will be critical for developing targeted therapies to restore immune homeostasis and mitigate autoimmune disease progression.

Limitations of the study

First, we acknowledge that mechanistic insights into the sustained presence of AIGAs despite RTX treatment, along with the transcriptional and clonal features of IFN-γ-specific ASCs, are derived from the limited number of BM samples analyzed. The sequences identified through our clustering approach in single-cell analysis are intrinsically related to, and defined by, the antigen-specific sequences identified through experimental screenings. This analytical focus implies the resulting inferred lineages represent a targeted pool of the overall antigen-specific repertoire, rather than a completely comprehensive census. Therefore, the transcriptomic characteristics we observed (e.g., CD11c and ZEB2 expression) should be interpreted as features of the sampled population, not universal signatures of antigen-specific cells. Further validation in larger cohorts will be important to strengthen and generalize these conclusions.

Second, it should be noted that our binding assays did not fully account for why NNAbs could compete with the IFN-γ/IFN-γR1 interaction yet fail to exhibit neutralizing activity. This discrepancy may reflect mechanistic differences between the recombinant-protein-based competition assay, which measures interactions among soluble IFN-γ, IFN-γR1, and IFN-γR2, and the cell-based neutralization assay, in which membrane-bound receptor complexes on THP-1 cells mediate functional signaling.

Resource availability

Lead contact

Further information and requests for resources and reagents should be directed to and will be fulfilled by the lead contact, Linqi Zhang (zhanglinqi@tsinghua.edu.cn).

Materials availability

Plasmids encoding recombinant proteins generated in this paper are available from the lead contact with a completed materials transfer agreement.

Data and code availability

• •The raw data of scRNA-seq in this paper have been deposited at the Genome Sequence Archive for Human (https://ngdc.cncb.ac.cn/gsa-human/) under accession number HRA016193. They are available upon request if access is granted. To request access, contact the lead contact.
• •This paper does not report original code.
• •The coordinates and structure factor files for the A01BM-03 Fab/IFN-γ complex have been deposited at the Protein DataBank (http://www.rcsb.org) under accession number 9VNP. Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request.

Acknowledgments

We thank all of the participants who made this research possible through their generous blood and bone marrow donations. We also thank Bin Yu and Pengcheng Jiao (Core Facility, Center of Biomedical Analysis, Tsinghua University) for technical support with flow cytometry analysis. This study was funded by Prevention and Control of Emerging and Major Infectious Diseases - National Science and Technology Major Project (2025ZD01904200), National Key Plan for Scientific Research and Development of China (2022YFF1203100, 2021YFC0864500, 2022YFC2604100, 2022YFC2303400, and 2023YFC3043300), National Natural Science Foundation of China (92169205, 92469204, 82241072, 82150205, 32270983, 32171202, and 82160635), the Wanke Scientific Research Program (20221080056).

Author contributions

L.Z., J.F., Xinquan Wang, and P.L. conceived and designed the study. H.W., T.Z., Y.L., and Y.Y. performed most experiments with assistance from J.Y., J.H., Z.W., X.F., J.L., and Xin Wang. Q.Y. conducted scRNA-seq data analysis. J.Y. refined the structural model. S.P., Y.W., J.W., and J.F. obtained ethical approvals and collected clinical specimens. H.W., Q.Y., and L.Z. accessed all data, generated figures/tables and ensured data integrity. H.W., Q.Y., J.Y., and L.Z. wrote the manuscript with contributions from P.Z., H.D., Q.Z., and X.S. All authors reviewed and approved the final manuscript.

Declaration of interests

Patent application has been filed on IFN-γ-specific neutralizing antibody. L.Z., H.W., Q.Y., Y.L., Y.Y., and X.S. are the inventors.

STAR★Methods

Key resources table

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
Alexa Fluor 568 anti-human IgG (H + L), cross-adsorbed	Invitrogen	Cat# A21090; RRID: AB_2535746
HRP anti-human IgG(H + L)	Promega	Cat# W4031; RRID: AB_430835
FITC anti-human IgD	BioLegend	Cat# 348205; RRID: AB_10613638
PE anti-human CD21	BioLegend	Cat# 354903; RRID: AB_2561406
BUV563 anti-human CD19	BD Biosciences	Cat# 612916; RRID: AB_2870201
PE/Dazzle 594 anti-human CD11c	BioLegend	Cat# 301641; RRID: AB_2564082
PerCP/Cyanine5.5 anti-human CD185 (CXCR5) Antibody	BioLegend	Cat# 356909; RRID: AB_2561818
Brilliant Violet 421 anti-human CD3	BioLegend	Cat# 344833; RRID: AB_2565674
Brilliant Violet 510 anti-human CD14	BD Biosciences	Cat# 563079; RRID: AB_2737993
Brilliant Violet 480 anti-human CD16	BD Biosciences	Cat# 566108; RRID: AB_2739510
PE/Cyanine7 anti-human CD19	BioLegend	Cat# 302215; RRID: AB_314245
APC anti-human CD307e (FcRL5)	BioLegend	Cat# 340305; RRID: AB_2564326
Alexa Fluor 700 anti-human CD45	BioLegend	Cat# 304023; RRID: AB_493760
APC/Cyanine7 anti-human CD38	BioLegend	Cat# 303533; RRID: AB_2561604
Brilliant Violet 421 anti-human CD3	BioLegend	Cat# 344833; RRID: AB_2565674
Brilliant Violet 421 anti-human CD14	BioLegend	Cat# 301829; RRID: AB_10899407
Brilliant Violet 421 anti-human CD16	BioLegend	Cat# 302037; RRID: AB_10898112
Pacific Blue anti-human IgD	BioLegend	Cat# 348223; RRID: AB_2561596
Brilliant Violet 605 anti-human CD27	BioLegend	Cat# 302829; RRID: AB_11204431
FITC anti-human IgG	BioLegend	Cat# 410719; RRID: AB_2721575
PerCP/Cyanine5.5 anti-human CD80	BioLegend	Cat# 305231; RRID: AB_2566490
Pacific Blue anti-human HLA-ABC	BioLegend	Cat# 311417; RRID: AB_493668
FITC anti-human HLA-DR	BioLegend	Cat# 307604; RRID: AB_314682
Alexa Fluor 700 anti-human CD11b	BioLegend	Cat# 301355; RRID: AB_2750074
CoraLite Plus 647 anti-iNOS	Proteintech	Cat# CL647-18985; RRID: AB_2934933
PE/Cyanine7 anti-human CD14	BioLegend	Cat# 301813; RRID: AB_389352
Alexa Fluor 647 anti-STAT1 Phospho (Tyr701)	BioLegend	Cat# 666410; RRID: AB_2814503
Biological samples
Peripheral blood samples from AOID, ID and healthy donors	This paper	N/A
Bone marrow aspirates from 3 AOID donors	This paper	N/A
Chemicals, peptides, and recombinant proteins
Ficoll-Paque PLUS	Cytiva	Cat# 17144002
Recombinant human IFN-γ	R&D	Cat# 285-IF/CF
SPHERO Streptavidin Coated Particles	Spherotech	Cat# SVP-60-5
Zombie UV Fixable Viability Dye	BioLegend	Cat# 423107
DAPI	Beyotime	Cat# C1006
Recombinant human IL-2 protein	Sino Biological	Cat# 11848-HNAH1-E
Recombinant human IL-21 protein	Sino Biological	Cat# 10584-HNAE
Lipopolysaccharide (LPS)	Beyotime	Cat# S1735
Phorbol 12-myristate 13-acetate (PMA)	Beyotime	Cat# S1819
Polyvinyl pyrrolidone (PVP) K-30	Solarbio	Cat# P8060
Endoproteinase Lys-C	Roche	Cat# 11047825001
Critical commercial assays
BD Cytofix/Cytoperm Fixation and Permeabilization Solution	BD Biosciences	Cat# 554722; RRID: AB_2869010
Dead Cell Removal Kit	Miltenyi	Cat# 130-090-101
Plasma Cell Isolation Kit II, human	Miltenyi	Cat# 130-093-628
CD138 MicroBeads	Miltenyi	Cat# 130-097-614
Human IP-10 Precoated ELISA Kit	Dakewe	Cat# 1117452
SeekOne DD Single Cell 5′ Library Preparation Kit	SeekGene	Cat# K00501
SeekOne DD Single Cell V(D)J Enrichment Kit (human BCR)	SeekGene	Cat# K00701
VAHTS DNA Clean Beads	Vazyme	Cat# N411-01
His Capture Kit	Cytiva	Cat# 28995056
Series S Sensor Chip CM5	Cytiva	Cat# 29104988
Deposited data
Raw data for human scRNA-seq dataset	This paper	GSA-Human ID: HRA016193
Crystal structures of A01BM-03 Fab/IFN-γ complex	This paper	PDB: 9VNP
Experimental models: Cell lines
THP-1 cells	ATCC	ATCC TIB-202; RRID: CVCL_0006
FreeStyle 293F cells	Thermo Fisher Scientific	Cat# R79007; RRID: CVCL_D603
NIH/3T3-huCD40L cells	Provided by Dr. Hai Qi	N/A
Recombinant DNA
IFN-γ	GenScript	GenBank: AY255837.1
Single-chain IFN-γ dimer with a 2× G4S linker	GenScript	Mendoza et al.45
IFN-γR1	GenScript	GenBank: J03143.1
IFN-γR2	GenScript	GenBank: U05877.1
IFN-γR1 F05 mutant	GenScript	Mendoza et al.45
RSV F glycoprotein from DS2 designer	Yang et al.53	Joyce et al.54
Heavy and light chains of emapalumab	GenScript	Patent: US 7.700.098 B2
Software and algorithms
GraphPad Prism 9	GraphPad	www.graphpad.com
FlowJo 10 software	FlowJo	https://www.flowjo.com/
Biacore Insight Evaluation Software	Cytiva	https://www.cytivalifesciences.com/en/us/products/items/biacore-insight-evaluation-software-p-23528
fastp v0.23.1	Chen et al.55	https://github.com/OpenGene/fastp
SeekSoulTools v1.2.1	Sang et al.56	http://seeksoul.seekgene.com/en/v1.2.1/0.download.html
R 4.3.3	R Core Team	https://www.r-project.org/
IMGT/HighV-QUEST 1.9.5	IMGT57,58	https://www.imgt.org/HighV-QUEST/
scRepertoire 2.5.7	Yang et al.59	https://www.borch.dev/uploads/screpertoire/
sciCSR 0.3.2	Ng et al.60	https://github.com/Fraternalilab/sciCSR
Seurat 5.2.1	Hao et al.61	https://satijalab.org/seurat
SCTransform	Stuart et al.62	https://github.com/satijalab/sctransform
UMAP	Becht et al.63	https://github.com/lmcinnes/umap
Celda 1.18.2	Yang et al.64	https://github.com/campbio/celda
Python 3.10	Python Software Foundation	https://www.python.org/
NumPy 2.3.4	Harris et al.65	https://numpy.org/
Jellyfish 1.1.3	Turk	https://pypi.org/project/jellyfish/1.1.3/
Hamming distance	Hamming66	https://jamesturk.github.io/jellyfish/functions/#hamming-distance
SciPy 1.16.3	Virtanen et al.67	https://scipy.org/
Pandas 2.3.3	The pandas development team	https://pandas.pydata.org/
wordcloud2 0.2.1	Lang	https://github.com/Lchiffon/wordcloud2
MAFFT 7.490	Katoh et al.68,69	https://mafft.cbrc.jp/alignment/software/
ggtree 3.4.4	Xu et al.70	https://github.com/YuLab-SMU/ggtree
mclust 6.1.1	Scrucca et al.	https://mclust-org.github.io/mclust/
pheatmap 1.0.12	Kolde	https://www.rdocumentation.org/packages/pheatmap/versions/1.0.12
PHASER	McCoy et al.71	http://www.phaser.cimr.cam.ac.uk/index.php/Phaser_Crystallographic_Software
COOT	Emsley et al.72	http://www2.mrc-lmb.cam.ac.uk/Personal/pemsley/coot/
PHENIX	Adams et al.73	http://www.phenix-online.org/
MolProbity	Chen et al.74	http://molprobity.biochem.duke.edu/
Other
Superdex 200 High-Performance column	Cytiva	N/A
Biacore 8K+	Cytiva	N/A
BD FACSDiscoverS8 Cell Sorter	BD Biosciences	N/A
Moflo Astrios EQ flow cytometer	Beckman	N/A
BD LSRFortassa	BD Biosciences	N/A
iMark Microplate Absorbance Reader	Bio-Rad	N/A
Qubit	Thermo Fisher Scientific	N/A
Bio-Fragment Analyzer	Bioptic	N/A
NovaSeq 6000	Illumina	N/A

Experimental model and study participant details

Study approval and patient samples

The study was approved by the Research Ethics Committee of Peking Union Medical College Hospital (ZS-3065). A cohort of 22 adult-onset immunodeficiency (AOID) patients with anti-IFN-γ autoantibodies was enrolled at Peking Union Medical College Hospital between October 2020 and June 2024, comprising 8 males and 14 females. Among these, 21 patients were diagnosed with disseminated nontuberculous mycobacterial (NTM) infections, with only one patient (A12) having disseminated tuberculosis (TB) infection. Peripheral blood samples were collected from all patients at enrollment. Additionally, a total of 24 longitudinal peripheral blood samples were obtained from patient A01 over a 90-week follow-up period. Bone marrow aspirates from the iliac crest were collected from three patients, A01, A18, and A19, at week 50, week 4 and week 1 post-enrollment, respectively. For comparison, blood samples were also obtained from 23 patients with localized NTM infections and 20 healthy donors. Peripheral blood samples and bone marrow aspirates were collected using Vacutainer EDTA Tubes (BD). Peripheral blood mononuclear cells (PBMCs) and bone marrow mononuclear cells (BMMCs) were isolated by density gradient centrifugation using Ficoll-Paque PLUS. Plasma samples were heat-inactivated at 56°C for 30 min and stored at −80°C while PBMCs and BMMCs were preserved in cryoprotective agents and stored in liquid nitrogen until further use. All participants had provided written informed consents for sample collection and subsequent analysis.

Cell lines

FreeStyle 293F cells (Thermo Fisher Scientific) were maintained in SMM 293-TII expression medium (Sino Biological) at 37°C in a humidified incubator with 5% CO₂. THP-1 cells were maintained in RPMI-1640 medium (Gibco) supplemented with 10% (v/v) fetal bovine serum (FBS, Gibco), 50 μM of 2-Mercaptoethanol (Gibco), and 100 U/mL of penicillin–streptomycin (Gibco) at 37°C under 5% CO₂. The NIH/3T3 cell line stably transfected with the human CD40 ligand gene (NIH/3T3-huCD40L), kindly provided by Dr. Hai Qi from Tsinghua University, was cultured in Dulbecco’s Modified Eagle Medium (DMEM, Gibco) supplemented with 10% (v/v) FBS and 100 U/mL penicillin–streptomycin at 37°C in a 5% CO₂ incubator. All cell lines were authenticated by morphology and growth rate. All cell lines were confirmed free of mycoplasma contamination.

Method details

Expression and purification of recombinant IFN-γ and IFN-γR1/2 protein

The nucleotide sequences encoding IFN-γ (amino acids 1–143) with C-terminal 8× His, Twin-Strep, and Flag tags; wild-type IFN-γR1 extracellular domain (amino acids 1–227) with C-terminal Twin-Strep and Flag tags; IFN-γR1 F05 mutant (containing T149I, M161K, Q167K, K174N, Q182R and H205N substitutions to enhance IFN-γR1/IFN-γ complex affinity for IFN-γR2) with C-terminal 8× His and Avi tags; and IFN-γR2 extracellular domain (amino acids 1–226) with C-terminal Twin-Strep and Flag tags were synthesized by GenScript and cloned into the pVRC8400 vector. To generate biotinylated IFN-γ for antigen-specific PC isolation, initial attempts involved expression IFN-γ with C-terminal 8× His and Avi tags, but expression levels were suboptimal. Based on the structural design of a single-chain IFN-γ dimer,⁴⁵ an optimized construct was designed by fusing two IFN-γ monomers in tandem with a 2× G₄S linker, followed by C-terminal 8× His and Avi tags. This construct exhibited robust expression. Fifteen alanine-substituted mutants of IFN-γ were constructed using the KOD One PCR Master Mix (Toyobo) and verified by sequencing. All recombinant proteins were expressed in HEK293F cells following transient transfection. For biotinylated IFN-γ production, plasmids encoding the single-chain IFN-γ dimer and the biotin-protein ligase BirA were co-transfected into HEK293F cells at a mass ratio of 2:1. After harvest of the culture supernatant, wild-type and alanine-substituted IFN-γ, wild-type IFN-γR1, IFN-γR1 F05 mutant, and IFN-γR2 proteins were purified using Strep-Tactin XT resin, while recombinant single-chain IFN-γ dimer was purified using Ni-NTA resin. Further purification was performed via gel filtration chromatography using a Superdex 200 Increase column (Cytiva). Expression and purification of recombinant RSV F glycoprotein from DS2 designer have been described previously.^53,54 Protein concentrations were determined using a Nanodrop 2000 Spectrophotometer (Thermo Fisher Scientific). Purified proteins were flash-frozen in liquid nitrogen and stored at −80°C until use.

ELISA

96-well ELISA plates (Corning) were coated with 1 μg/mL of recombinant human IFN-γ or RSV F glycoprotein in PBS (Gibco) and incubated overnight at 4°C. Plates were washed three times between each step with PBS containing 0.05% Tween 20 (PBS-T). Following washing, the plates were blocked with PBS-T containing 0.5% PVP K-30 for 1 h at 37°C. Serially diluted plasma samples (starting at 1:12 dilution) or mAbs (starting at 50 μg/mL) were added and incubated for 1 h at 37°C. Subsequently, plates were then incubated with anti-human IgG conjugated with horseradish peroxidase (HRP, Promega) at 1:5,000 dilution for 30 min at 37°C. Detection was performed using 3,3′,5,5′-tetramethylbenzidine (TMB, CWBio), and the reaction was stopped by adding 1M H₂SO₄. Absorbance was measured at 450 nm using an iMark Microplate Absorbance Reader (Bio-Rad). Half-maximum titers and concentrations were calculated using a dose-response model with an asymmetric five-parameter equation in GraphPad Prism 9. Plasma samples exhibiting binding antibody activity with a half-maximal effective concentration greater than twice that of healthy plasma samples were considered positive for binding antibodies.

Neutralizing assay using THP-1 cells

Human IFN-γ Recombinant Protein was prepared at a concentration of 200 ng/mL in 50 μL of RPMI-1640 (Gibco) was incubated with 50 μL of 3-fold serially diluted plasma samples (starting at 1:20 dilution) or monoclonal antibodies (starting at 50 μg/mL) at 37°C for 1 h. The mixture was then added to 1× 10⁵ THP-1 cells suspended in 100 μL of RPMI-1640. Cells were washed twice between each step with PBS containing 2% FBS. After a further 15-min incubation at 37°C, cells were fixed by adding 200 μL of Cytofix/Cytoperm Fixation/Permeabilization Solution (BD Biosciences) and incubated at 4°C for 20 min. Subsequently, cells were permeabilized with 300 μL of ice-cold absolute methanol on ice for 15 min. Following washing steps, AF647-conjugated phospho-STAT1 (pY701) antibody was added at a 1:50 dilution and incubated at 4°C for 30 min. Cells were then washed twice and resuspended in PBS. Data acquisition was performed using a BD LSRFortessa flow cytometer and analyzed with FlowJo. Half-maximal titers were calculated using a dose-response model with an asymmetric five-parameter equation in GraphPad Prism 9.

Neutralizing assay using THP-1-derived macrophages and monocytes in PBMCs

THP-1 cells were cultured to a density of approximately 0.6×10⁶/mL and induced to differentiate by treatment with 50 ng/mL phorbol 12-myristate 13-acetate (PMA) for 24 h. The medium was then replaced, and cells were cultured for an additional 2 days at 37°C with 5% CO₂. THP-1-derived macrophages were harvested and seeded into 96-well flat-bottom plates at 5×10⁴ cells per well for 24 h. Antibodies were added at a final concentration of 50 μg/mL, followed by stimulation with 10 ng/mL lipopolysaccharide (LPS) and 100 ng/mL IFN-γ. Unstimulated controls, LPS alone without antibody or IFN-γ, were included. Cells were washed twice between each step with PBS containing 2% FBS. After 48 h of stimulation, cells were treated with 50 μL of trypsin for 2 min and stained with DAPI (1:50) along with a mixture of fluorescence-labeled antibodies, including PerCP-Cy5.5-conjugated CD80 (1:100), Pacific Blue-conjugated HLA-ABC (1:100), FITC-conjugated HLA-DR (1:100), and AF700-conjugated CD11b (1:100), at 4°C for 20 min. Following surface staining, cells were then fixed and permeabilized with 100 μL Cytofix/Cytoperm Fixation/Permeabilization Solution for 20 min. Intracellular staining was performed using CoraLite647-conjugated iNOS (1:100) in 50 μL of staining buffer at 4°C for 20 min. Data acquisition was conducted using an LSRFortessa flow cytometer (gating strategy shown in Figure S1A), and analysis was performed using FlowJo software to calculate the fold change in median fluorescence intensity (MFI) of HLA-ABC, HLA-DR, CD80, and iNOS relative to the unstimulated control. CXCL10 concentration was measured using Human IP-10 Precoated ELISA Kit (Dakewe). Cell culture supernatant was diluted 1:50 prior to measurement, and all remaining procedures were performed in accordance with the manufacturer’s instructions.

Frozen PBMCs from healthy donors were thawed at 37°C and incubated with IFN-γ, LPS, and mAbs following the same procedure described above. After 48 h of stimulation, cells were stained with DAPI (1:50), PE-Cy7-conjugated CD14 (1:100) and FITC-conjugated HLA-DR (1:100), for 20 min at 4°C. The fold changes in median fluorescence intensity (MFI) of HLA-DR and the concentration of CXCL10 were measured as previously described (gating strategy shown in Figure S1C).

Isolation of IFN-γ–specific memory B cells from peripheral blood

IFN-γ–specific memory B cells (MBCs) were isolated via B cell immortalization.⁷⁵ Frozen PBMCs were thawed in 37°C and stained with DAPI (1:50) along with a mixture of fluorescence-labeled antibodies, including PE-Cy7-conjugated CD19 (1:100), BV421-conjugated CD3 (1:100), BV421-conjugated CD14 (1:100), BV421-conjugated CD16 (1:100), Pacific Blue-conjugated IgD (1:100), BV605-conjugated CD27 (1:100), FITC-conjugated IgG (1:100). Live CD3⁻CD14⁻CD16⁻CD19⁺IgD⁻ B cells were sorted using the Moflo Astrios EQ flow cytometer (Beckman) into 96-well plates (10 cells/well) containing 150 μL of RPMI-1640 supplemented with 10% (v/v) FBS, 10 ng/mL recombinant human IL-2 protein (Sino Biological), 10 ng/mL recombinant human IL-21 protein (Sino Biological), and 5×10⁴ irradiated NIH/3T3-huCD40L cells. After 7–10 days of culture, culture supernatants were screened for IFN-γ binding using ELISA. Wells exhibiting optical density at 450 nm values at least three times higher than negative controls were selected. Cell from these wells were pelleted by centrifugation, resuspended in 10 μL of lysis buffer (DNase/RNase-free distilled water containing 0.2% Triton X-100 and 5% (v/v) RNase inhibitor), and stored at −80°C until subsequent reverse transcription and antibody gene amplification.

Isolation of IFN-γ–specific ASCs from bone marrow

IFN-γ–specific ASCs were isolated using the MoSMAR platform.⁴⁰ Frozen BMMCs were thawed in 37°C, and live BM-PCs were enriched using Dead Cell Removal Kit and Human Plasma Cell Isolation Kit II (Miltenyi) with modifications to the manufacturer’s protocol. The process involved the depletion of dead and non-plasma cells, followed by the enrichment of PCs using human CD38 and CD138 magnetic beads simultaneously. BM-PCs were then incubated with biotinylated IFN-γ–coated polystyrene particles (Spherotech) and AF568-conjugated anti-human IgG antibody (Invitrogen) at 37°C for 2 hours on the MoSMAR-chip. IFN-γ–specific BM-PCs were identified by the circular red fluorescence signal observed under a fluorescence microscope (Olympus). Individual IFN-γ–specific BM-PCs were then isolated using custom-designed SCDI, transferred into 10 μL of lysis buffer, and stored at −80°C until subsequent reverse transcription and antibody gene amplification.

Cloning and expression of monoclonal antibodies

The IgG heavy and light chain variable genes were amplified by nested PCR and cloned into linear expression cassettes or expression vectors to produce full-length human IgG1 antibodies as previously described.^76,77 The nucleotide sequences encoding the heavy and light chains of emapalumab (patent# US7.700.098B2), A01BM-bc01, and A19BM-bc01 were synthesized by GenScript and cloned into expression vector. Large-scale production of monoclonal antibodies was achieved by transfecting HEK293F cells with equal amounts of paired heavy and light chain expression plasmids. Secreted antibodies were purified from the culture supernatant using Protein A agarose resin (GenScript) and eluted by 1 M glycine (pH 3.0). Antibody concentration was determined using a Nanodrop 2000 Spectrophotometer (Thermo Fisher Scientific).

Single cell RNA sequencing (scRNA-seq) for BM-PCs

The enrichment procedure for BM-PCs is described in the preceding section. These enriched cells were used for single-cell RNA sequencing libraries preparation, which was performed using the SeekOne DD Single Cell 5′ Library Preparation Kit (SeekGene, K00501) and SeekOne DD Single Cell V(D)J Enrichment Kit (human BCR, SeekGene, K00701) according to the manufacturer’s instruction. Briefly, the cell suspension was mixed with the reverse transcription reagent and loaded into the sample well of the SeekOne DD Chip S3. Barcoded hydrogel beads and partitioning oil were dispensed into their respective wells, enabling single-cell encapsulation and reverse transcription (42°C for 90 min and 85°C for 5 min). Following this step, cDNA was purified from disrupted droplets and amplified via PCR. The amplified products were then fragmented, end-repaired, A-tailed, and ligated to sequencing adapters. Indexed PCR was performed to enrich DNA fragments incorporating cell barcodes and unique molecular identifiers (UMIs). Finally, sequencing libraries were cleaned up with VAHTS DNA Clean Beads (Vazyme), quantified by Qubit (Thermo Fisher Scientific) and assessed for quality using Bio-Fragment Analyzer (Bioptic), and then sequenced on an Illumina NovaSeq 6000 platform with PE150 read length.

Data preprocessing and quality control

Single-cell V(D)J-seq data for B cell receptors (BCRs) were reanalyzed using IMGT/HighV-QUEST (version: 1.9.5)^57,58 for V(D)J gene segment annotation and somatic hypermutation frequency calculation, retaining only productive rearrangements. Cells were then assigned paired sequences using the scRepertoire (version 2.5.7)⁵⁹ and sciCSR (version: 0.3.2)⁶⁰ frameworks, based on single-cell barcodes. Strict filtering criteria were applied: barcodes with NA values in any chain were removed, only the top two expressed chains were retained in each barcode containing multiple chains, and CollapseBCR in sciCSR was used to ensure at most one heavy-light chain pair per cell.

Fastp (version 0.23.1) was utilized to trim primer sequences and remove low quality bases from the raw reads.⁵⁵ Subsequently, the cleaned reads were processed using SeekSoulTools (version 1.2.1) to generate the transcript expression matrix for further analysis.⁵⁶ Gene expression matrices from all samples were merged and uniformly processed in Seurat (version 5.2.1).⁶¹ The raw data from each sample was read using Read10X and converted into individual Seurat objects, with an initial filtering criterion of retaining features detected in at least 3 cells (min.cells = 3). Quality control (QC) filtering was applied to the merged object by calculating the percentage of mitochondrial (mt), ribosomal (rb), hemoglobin (hb) and immunoglobulin (Ig) genes. Cells were stringently filtered based on multiple metrics: feature counts (200 < nFeature_RNA <4000), UMI counts (500 < nCount_RNA <20000), mitochondrial content (<10%), ribosomal content (<40%), and hemoglobin content (<1%). After filtering, 22,777 cells were remained (2,889 from A01, 8,981 from A18, and 10,907 from A19). To mitigate the effect of ambient RNA, the filtered Seurat object was split by sample, and each sub-object was decontaminated using the celda (version 1.18.2) package’s decontX function,⁶⁴ after which the objects were merged and normalized. Then, we excluded contaminated cells expressing diagnostic cell markers, e.g., CD3D, CD3E (T cell), CD16 (encoded by FCGR3A) and CD14 (Monocytes), NKG7, GNLY (Natural killer cells), CD34 (Progenitor cells), CD20 (encoded by MS4A1), PAX5, IRF8 (B cells). Stringent filtering was subsequently applied to retain only those cells possessing paired heavy and light chain B cell receptor (BCR) sequence information and an Immunoglobulin (Ig) gene expression percentage exceeding 5% (percent.Ig > 5%). This rigorous selection enabled the integrated analysis of both clonotype and transcriptome data, ultimately yielding 3,500 B/plasma cells (488 from A01, 2,061 from A18, and 951 from A19) for downstream analyses. Somatic Hypermutation (SHM) rates for heavy and light chains were calculated as H_shm = 1 - IGH_{v_identity}/100 and L_shm = 1 - IGL_{v_identity}/100, respectively, and stored in the merged Seurat object. This process yielded a total of 3,500 B/plasma cells (488 from A01, 2,061 from A18, and 951 from A19) for downstream analyses.

Normalization and cell cluster detection

For each sample, gene expression counts were normalized using SCTransform,⁶² which employs regularized negative binomial regression to correct for sequencing depth and technical variance. The top 3,000 highly variable features (HVFs) were identified per dataset based on standardized Pearson residuals, after which immunoglobulin genes were systematically excluded to prevent clonotype-driven expression bias. Following feature selection, a common set of 2,000 integration features was selected, and datasets were integrated using Canonical Correlation Analysis (CCA).⁶² Integration success was validated by uniform sample mixing on UMAP⁶³ and balanced contribution from each individual.

To evaluate the robustness of clustering under different parameter settings, we systematically compared clustering results obtained using varying combinations of principal components (30–70) and resolution parameters (0.1–1.0). Pairwise comparisons between clustering results were quantified using the Adjusted Rand Index (ARI) implemented in mclust package (version 6.1.1, https://mclust-org.github.io/mclust/), which measures the similarity between two clustering solutions while adjusting for agreement expected by random chance. Higher ARI values (closer to 1) indicate greater consistency and stability across parameter settings.

A two-way hierarchical clustering heatmap was generated to visualize the concordance of clustering results across all parameter combinations (Figure S2L, left). Parameter combinations are indicated in the middle panel, and the average ARI for each row of the heatmap is shown in the right-hand scatterplot. The parameter set showing the highest mean ARI—corresponding to the most stable clustering—was selected for downstream analysis (40 PCs, resolution = 0.5; Figure S2L).

Gene expression analysis

Differentially expressed genes (DEGs) were identified between all cell clusters using the Wilcoxon rank-sum test implemented in the Seurat package with default parameters (Table S5). It’s worth noting that the HVF list with excluded Ig genes was used to detect cell clusters, the whole gene expression list which included Ig genes was used to detect cell markers and differentially expressed genes.

Lineage clustering and phylogenetic analysis

IgH sequences were classified into lineage clusters to assess potential genetic relationships among antibody clonotypes. Hierarchical clustering of CDR3 amino acid sequences was performed using a single-linkage method with a normalized Hamming distance matrix⁶⁶ and a similarity threshold of 0.85 to define clonotypes (Table S6).^78,79,80,81 Prior to clustering, amino acid sequences were grouped by identical V gene, J gene, and CDR3 length, and lineage cluster labels were assigned with prefixes corresponding to the respective V–J combinations and CDR3 lengths. All analyses were performed in Python 3.10 using NumPy,⁶⁵ SciPy,⁶⁷ pandas (https://pandas.pydata.org), and jellyfish (https://codeberg.org/jpt/jellyfish). The visualization of heavy-chain CDR3 frequencies in cluster 3 was generated using the wordcloud2 R package (version 0.2.1). For phylogenetic analysis, IGH V-D-J and IGK/L V-J region amino acid sequences were aligned separately using MAFFT (version 7.490)^68,69 with BLOSUM62 substitution matrix, a 1.53 gap opening penalty, and 0.123 offset value. A Neighbor-Joining (NJ) tree was then constructed using Jukes-Cantor genetic distance model, followed by tree visualization and germline association analysis using the R package ggtree (version 3.4.4).⁷⁰

Characterization of ABCs and DN2 B cells in PBMCs

Frozen PBMCs were thawed in 37°C and stained with Zombie UV Fixable Viability Dye (1:100), followed by a mixture of fluorescence-labeled antibodies, including FITC-conjugated IgD (1:100), PE-conjugated CD21 (1:100), BUV563-conjugated CD19 (1:100), PE-Dazzle594-conjugated CD11c (1:100), PerCP-Cy5.5-conjugated CXCR5 (1:100), BV421-conjugated CD3 (1:100), BV510-conjugated CD14 (1:100), BV480-conjugated CD16 (1:100), PE-Cy7-conjugated CD27 (1:100), APC-conjugated CD307e (FcRL5) (1:100), AF700-conjugated CD45 (1:100), and APC-Cy7-conjugated CD38 (1:100). Data acquisition was conducted using BD FACSDiscover S8 Cell Sorter (gating strategy shown in Figure S4A), and analysis was performed using FlowJo software to calculate the frequencies of ABCs (CD19⁺CD38^lowCD27^lowCD21⁻CD11c⁺) or DN2 B cells (CD19⁺CD38^lowCD27⁻IgD^-CD21⁻CD11c⁺).

Kinetic analysis by SPR

The binding kinetics and affinity of antibodies for IFN-γ were analyzed using the Biacore 8 K+ SPR system (Cytiva). Anti-his tag antibody (Cytiva) was covalently immobilized onto a CM5 sensor chip (Cytiva). Recombinant wild-type or alanine-substituted IFN-γ containing 8× His, Twin-strep and Flag tags were captured via anti-His tag antibody. 2-fold serially diluted antibodies (starting at 640 nM) were then injected, and the resulting sensograms were fit to a 1:1 binding model using the Biacore Insight Evaluation Software (Cytiva).

Competitive analysis by SPR

Competitive SPR assays were performed to evaluate antibody-antibody competition and antibody-IFN-γ receptor competition using a Biacore 8K+ SPR system. Recombinant IFN-γ with His, Twin-strep, and Flag tags was captured on a CM5 sensor chip via anti-his tag antibody. For antibody-antibody competition, primary antibodies or an isotype-matched control antibody were injected for 100 s to achieve binding equilibrium, followed by the injections of secondary antibodies or control antibody for an additional 100 s. Binding responses were recorded to assess competitive inhibition, and the competition percentage was calculated using the following formula: Competition (%) = [1 - (RU of the secondary antibody in the presence of primary antibody - RU of the irrelevant antibody in the presence of the primary antibody)/(RU of the secondary antibody in the presence of irrelevant antibody - RU of the irrelevant antibody in the presence of the irrelevant antibody)] × 100%, where RU denotes response units. For competition with IFN-γR1, the secondary antibodies were replaced with wild-type IFN-γR1 protein. For competition with IFN-γR2, the captured antigen was substituted with the IFN-γR1 F05 mutant. Next, IFN-γ lacking tags (R&D) was incubated on the chip, followed by injection of antibodies as the primary analyte and IFN-γR2 as the secondary analyte.

Crystallization and structural determination

Purified antibody A01BM-03 was cleaved using Lys-C protease (Roche) at an IgG to Lys-C ratio of 4000:1 (w/w). The Fc fragments were removed using Protein A agarose, and the Fab fragments were further purified by gel filtration chromatography. Recombinant IFN-γ containing 8× His, Twin-strep and Flag tags, and Fab fragments were mixed at a 1:1.2 molar ratio, incubated at 4°C for 2 hours, and further purified by gel filtration chromatography. Purified complex was concentrated to >10 mg/mL in HBS buffer (10 mM HEPES, pH 7.2, 150 mM NaCl) for crystallization. Crystallization trials were performed at 18°C using the sitting-drop vapor-diffusion method, in which 0.2 μL of protein was mixed with 0.2 μL of reservoir solution. Crystals of the A01BM-03 Fab-IFN-γ complex were obtained in 0.1 M Bis-Tris pH 7.0 and 22% (w/v) Polyethylene glycol 1500. Crystals were collected, soaked briefly in mother liquid supplemented with 20% glycerol, and flash-frozen in liquid nitrogen. Diffraction data were collected at a wavelength of 0.9792 Å on the BL02U1 beamline of the Shanghai Synchrotron Research Facility and processed using HKL2000. The structure was determined by molecular replacement with the PHASER.⁷¹ Modal building was performed using COOT⁷² and refinement was performed using PHENIX.⁷³ Structure validation was performed using MolProbity.⁷⁴ All data collection and structure refinement statistics are listed in Table S3.

Quantification and statistical analysis

All statistical analyses were performed using GraphPad Prism 9 software. Differences between two independent groups were assessed using the Mann-Whitney U test for nonparametric data. Correlation analysis was conducted using simple linear regression.

Published: March 9, 2026