In this work, we designed and prepared an AID microarray that enables to detect 125 autoantibodies that are related to different organs and systematic autoimmune diseases, which is shown in Fig. 1a and 1b (Figure S1, Table S1). The detection workflow is shown in Fig. 1c and in the Methods.
Fig. 1Schematic illustration of AID microarray preparation and serological autoantibody detection. a Autoimmune diseases that are related to the autoantibodies targeted by AID microarray. b Layout of autoantigens printed in AID microarray. c Workflow of AID microarray detection process. d Representative results of serological detection using AID microarray: fluorescence signals of IgG autoantibody in serum of healthy controls, systemic lupus erythematosus (SLE), rheumatoid arthritis (RA) patients, sjogren’s syndrome, autoimmune hepatitis and primary bile cholangitis. The markers circled in the red box are representative autoantibodies corresponding to the disease
To evaluate the feasibility of the AID microarray, we analyzed pooled serum samples from patients with autoimmune diseases and healthy controls. Each pooled sample was prepared from 10 randomly selected individuals within each group. The results demonstrated significantly elevated levels of disease-related autoantibodies in the patient groups. Specifically, anti-RO60 and anti-RO52 antibodies were markedly increased in SLE patients, while rheumatoid factor (RF) and anti-cyclic citrullinated peptide (CCP) antibodies were significantly elevated in RA patients. These findings demonstrate the feasibility and potential applicability of AID microarray in the diagnosis of autoimmune diseases.
Given the critical importance of reliability and clinical applicability in biomarker development, we evaluated the reproducibility of AID microarray and its consistency with clinical chemiluminescent immunoassays. To address this concern, four batches of the protein microarray were printed, with 100 slides per batch. Two slides were selected from batch #1, and one slide from batches #2, #3, and #4. The arrays were incubated with the same serum pool sample and the signals were detected using an anti-human IgG secondary antibody. AID microarray demonstrated high reproducibility, with Pearson correlation coefficients of 0.99 (intra-batch) and 0.97 (inter-batch) (Fig. 2a).
Fig. 2Reproducibility and correlation of AID microarray with chemiluminescent immunoassays. a Reproducibility of AID microarray for serological autoantibody detection. The diagonal indicates the SNR distribution of the sample, the lower left indicates the bivariate scatter plot with a fitted line, and the upper right indicates the correlation coefficient and the significance (***p < 0.001). b Correlation between AID microarray and chemiluminescent immunoassays (Median R2 = 0.86). c Correlation analysis of representative results between AID microarray and chemiluminescent immunoassays. The horizontal axis represents the results of chemiluminescence detection, and the vertical axis represents the SNR values from microarray detection. Each scatter point represents the detection data of a sample, and different sub-graphs correspond to different autoantibodies, including SSB, RO52, RO60, and AMA-M2. SNR, Signal-to-noise ratio
To assess concordance with clinical methods, we analyzed 130 ANA-positive patient samples, a total of 12 IgG autoantibodies, including AMA-M2, Nuc (NAP1L4), and P0 (RPLP0), were detected using both chemiluminescent immunoassays and AID microarray. The two platforms exhibited a strong correlation (median R2 = 0.86, Fig. 2b), with four representative autoantibody comparisons shown in Fig. 2c. These results confirm that AID microarray is highly reliable and consistent with existing clinical technologies, supporting its potential clinical application in biomarker discovery.
Characteristics of lung cancer irAEs cohortThe baseline characteristics of 83 lung cancer patients are shown in Table 1. At the data cut-off (14 December 2024), the median duration of follow-up was 12.5 months (95% CI: 10.6–14.4) in the discovery cohort. The median overall survival (OS) for all patients was 34.5 months (95% CI: 30.4–38.7). Figure 3a shows the type and severity of irAEs in 83 NSCLC patients. We identified 8 types of irAEs in 72 patients developed irAEs. Hematotoxicity was the most common type of irAEs (51.79%), followed by gastrointestinal toxicity (20.54%) and hepatotoxicity (13.39%) (Fig. 3b). Among the patients, 33 cases (45.83%) developed two or more types of irAEs (Fig. 3c).
Fig. 3Identification of autoantibodies related to irAEs in lung cancer patients. a The irAEs identified in lung cancer patients with ICIs treatment. (b, c) Doughnut chart based on the type of irAEs and the number of irAEs types in patients. d Autoantibodies associated with irAEs. Volcano plots showed baseline differential autoantibodies associated with irAEs for G1-G3 groups versus G0 group (without irAEs). The selection of autoantibodies was performed using samr-nonparametric test (p < 0.05). Blue and red dots represent down-regulated and up-regulated autoantibodies
Changes of autoantibodies pre- and post-treatmentNext, to explore the effect of immunotherapy on autoantibody levels, we performed a differential analysis of the changes in autoantibodies before and after immunotherapy in 83 patients in the training cohort (samr-nonparametric test, p < 0.05). Compared with the pre-treatment serum, we found that 6 autoantibodies were up-regulated and 15 autoantibodies were down-regulated in the post-treatment group (Figure S2). It is worth noting that autoantibodies such as anti-CGA IgG, anti-FSH IgG, and anti-AMH IgA are related to reproduction and infertility; anti-Mi-2 IgG and anti-NARS1 IgA autoantibodies are related to polymyositis/dermatomyositis (PM/DM); anti-LCN2 IgG autoantibodies are related to neuropsychiatric systemic lupus erythematosus (Fig. 1b and S1, Table S1). This shows that immunotherapy has a significant effect on the patient's autoantibody spectrum, and some of the affected autoantibodies are closely related to specific diseases, which provides important clues for a deeper understanding of the immune-related effects and potential clinical significance of immunotherapy.
Identification of baseline autoantibodies associated with irAEsPrevious studies have suggested that baseline autoantibody levels are associated with the risk of developing irAEs, but their relationship with the severity of irAEs remains unclear [14, 31]. To investigate this, we analyzed baseline serum antibody profiles in patients with and without irAEs. Compared to G0 (without irAEs) group, we identified 20, 29, and 37 differential IgG autoantibodies, and 24, 23, and 40 differential IgA autoantibodies in G1, G2, and G3 groups, respectively (samr-nonparametric, p < 0.05) (Fig. 3d; Table S2). Interestingly, we found the number of differential autoantibodies gradually increased with the elevation of irAEs severity (Fig. 4a), suggesting that patients had a progressive immune imbalance before immunotherapy.
Fig. 4Autoantibodies associated with irAEs. a Number of differential autoantibodies identified in lung cancer patients with different irAEs severity (G0-G3). b Venn diagram showed the overlap of differential autoantibodies among groups. c Box plot showed abundance of overlapping autoantibodies across four severity groups (G0, G1, G2, G3). d Tissue- and cell-specific enrichment analysis based on PaGenBase. e Disease association analysis using DisGeNET, where bar length and color intensity represent the statistical significance (p value) of differential autoantibodies within pathways. f GO biological process enrichment analysis, with circle size and color intensity indicating the number of associated autoantibodies and pathway significance (p value)
Venn diagram showed four overlapping autoantibodies among the groups (Fig. 4b), with their abundance visualized in box plots (Fig. 4c). To comprehensively characterize the functions of all differential autoantibodies, we performed bioinformatics analysis, which revealed their specific enrichment in lung and bronchial epithelial cells (Fig. 4d). Moreover, disease-related enrichment analysis demonstrated that the top-ranked pathways were primarily associated with lung diseases (Fig. 4e). Biological process analysis indicated that the differential autoantibodies were mainly involved in cellular amino acid metabolism, tRNA metabolism, and chemical synaptic transmission (Fig. 4f). The result suggests that these pathways related to autoantibodies in patients may have been altered and directly or indirectly participated in the occurrence of irAEs.
Identification of baseline autoantibodies associated with irAEs severityTo further investigate the relationship between autoantibodies and irAEs severity grading, we performed hierarchical clustering analysis of shared differential autoantibodies across the three irAEs severity groups (IgG: n = 13; IgA: n = 12; Figure S3, Table S3 and S4). This analysis identified 7 IgG autoantibodies (anti-Ku, anti-NAP1L4, anti-NARS1, anti-PLP1, anti-SmD3, anti-SNRPC, anti-TARS1) and 2 IgA autoantibodies (anti-AARS1, anti-PLP1) that progressively increased with irAEs severity (Fig. 5a and c). Box plots depicting the abundance of these 9 autoantibodies are shown in Fig. 5b and 5d.
Fig. 5Autoantibodies associated with irAEs severity and clinical prognosis. (a, c) Autoantibodies (IgG: 7, IgA: 2) positively correlated with irAEs severity. (b, d) Box plots showed abundance of these autoantibodies across four severity groups (G0, G1, G2, G3). e Performance of biomarker panels in distinguishing G0 vs. G1&G2&G3 and G0 vs. G3, evaluated by ROC curve analysis (AUC, area under the curve). f Cox regression forest plot illustrated the association of autoantibodies and baseline characteristics with prognosis
Next, we constructed and evaluated a 9-autoantibody predictive model (9-panel) for irAEs risk assessment. The 9-panel achieved an AUC of 0.854 in predicting irAEs occurrence (Fig. 5e). Additionally, its predictive accuracy for G1 and G2 irAEs reached AUC = 0.897 and 0.856, respectively (Figure S4). Notably, based on CTCAE v5.0, ICIs therapy is typically suspended when irAEs reach grade 3 (G3) severity. The 9-panel model demonstrated robust predictive performance for G3 irAEs, achieving an AUC of 0.934, underscoring its potential clinical utility in guiding treatment decisions (Fig. 5e).
We then collected an independent validation cohort (n = 31) and validated the ability of the 9-panel to predict the occurrence of irAEs using a validation cohort. The model maintained strong predictive power in the validation cohort, achieving an AUC of 0.826 (Figure S5). This result further confirms the reliability and generalizability of our 9-autoantibody predictive model across different patient populations, reinforcing its value in clinical practice for early identification of patients at risk of developing irAEs.
To evaluate the prognostic significance of differential autoantibodies and baseline clinical characteristics (e.g., age, sex), we analyzed their correlation with patient survival outcomes. Among the identified autoantibodies, anti-NAP1L4 IgG (p = 0.006, HR = 13.59) and anti-Ku IgG (p = 0.004, HR = 3.52) were found to be significant risk factors for poor prognosis (Fig. 5f). The corresponding autoantigens and their functions are detailed in Table S5.
Identification of baseline autoantibodies associated with immunotherapy efficacyTo assess the potential of autoantibodies in predicting immunotherapy response, we divided the baseline serum samples into R and NR groups. Compared with the NR group, differential analysis (samr-nonparametric, p < 0.05) identified 5 up-regulated autoantibodies (anti-MBP IgA, anti-GLRA2 IgA, anti-KRT20 IgA, anti-SAE1 IgA, anti-BIN1 IgA) and 3 down-regulated autoantibodies (anti-PLA2R1 IgG, anti-GAD2 IgG, anti-SSB IgA) in the R groups (Fig. 6a). Their secretion levels are shown in Fig. 6b.
Fig. 6Autoantibodies associated with immunotherapy efficacy. a Volcano plot displayed baseline IgG and IgA autoantibodies from AID microarray, comparing NR vs. R. Autoantibodies were selected using samr-nonparametric test (p < 0.05), with blue and red dots representing down-regulated and up-regulated autoantibodies, respectively. b Box plot showed the secretion levels of identified autoantibodies in NR and R groups. (c, d) ROC curves evaluated the predictive performance of the 8-autoantibody panel (8-panel) in the training cohort and validation cohort. AUC, area under the curve; irAEs, immune-related adverse events; ROC, receiver operating characteristic. e Cox regression forest plot illustrated the association of efficacy markers and baseline characteristics with immunotherapy outcomes
Using univariate logistic regression, we further evaluated the predictive ability of characteristics, hematology index and differential autoantibodies for immunotherapy efficacy (Table S6). Among study variables, age, anti-PLA2R1 IgG, anti-SSB IgA, anti-MBP IgA, anti-GLRA2 IgA, anti-KRT20 IgA, and anti-SAE1 IgA exhibited p values less than 0.05, indicating a significant impact on treatment efficacy. Specifically, anti-PLA2R1 IgG showed an odds ratio (OR) < 1, suggesting that higher levels of it may reduce the risk of efficacy-related events. In contrast, anti-MBP IgA, anti-GLRA2 IgA, anti-KRT20 IgA, and anti-SAE1 IgA had OR > 1, implying that elevated levels of these markers may increase the risk of efficacy-related events. The majority of the remaining variables, with p > 0.05, did not show a significant influence on treatment efficacy (Table S7).
Then we constructed a prediction model (8-panel) using these 8 autoantibodies. In the training cohort, the model achieved an AUC of 0.855 for predicting immunotherapy response (Fig. 6c). Moreover, when validated in the independent validation cohort (n = 31), the model maintained robust performance, achieving an AUC of 0.746 (Fig. 6d). These results demonstrated the model's strong generalizability and discriminative power in identifying treatment responders across different datasets.
To assess the impact of clinical characteristics (such as sex and age) and eight autoantibodies on patient prognosis, we performed a multivariate Cox regression analysis. We found that anti-SSB IgA, anti-GLRA2 IgA, and anti-KRT20 IgA significantly affected patient outcomes. Among them, anti-SSB IgA was identified as a risk factor (p = 0.096; HR = 7.41), whereas anti-GLRA2 IgA (p = 0.02; HR = 0.14) and anti-KRT20 IgA (p = 0.023; HR = 0.52) were protective factors (Fig. 6e). The findings suggest that these autoantibodies may serve as potential prognostic biomarkers, aiding in the development of personalized treatment strategies. The eight autoantigens and their biological functions are summarized in Table S8.
Comments (0)