Serum anti-DIDO1, anti-CPSF2, and anti-FOXJ2 antibodies as predictive risk markers for acute ischemic stroke

Background Acute ischemic stroke (AIS) is a serious cause of mortality and disability. AIS is a serious cause of mortality and disability. Early diagnosis of atherosclerosis, which is the major cause of AIS, allows therapeutic intervention before the onset, leading to prevention of AIS. Methods Serological identification by cDNA expression cDNA libraries and the protein array method were used for the screening of antigens recognized by serum IgG antibodies in patients with atherosclerosis. Recombinant proteins or synthetic peptides derived from candidate antigens were used as antigens to compare serum IgG levels between healthy donors (HDs) and patients with atherosclerosis-related disease using the amplified luminescent proximity homogeneous assay-linked immunosorbent assay. Results The first screening using the protein array method identified death-inducer obliterator 1 (DIDO1), forkhead box J2 (FOXJ2), and cleavage and polyadenylation specificity factor (CPSF2) as the target antigens of serum IgG antibodies in patients with AIS. Then, we prepared various antigens including glutathione S-transferase-fused DIDO1 protein as well as peptides of the amino acids 297–311 of DIDO1, 426–440 of FOXJ2, and 607–621 of CPSF2 to examine serum antibody levels. Compared with HDs, a significant increase in antibody levels of the DIDO1 protein and peptide in patients with AIS, transient ischemic attack (TIA), and chronic kidney disease (CKD) but not in those with acute myocardial infarction and diabetes mellitus (DM). Serum anti-FOXJ2 antibody levels were elevated in most patients with atherosclerosis-related diseases, whereas serum anti-CPSF2 antibody levels were associated with AIS, TIA, and DM. Receiver operating characteristic curves showed that serum DIDO1 antibody levels were highly associated with CKD, and correlation analysis revealed that serum anti-FOXJ2 antibody levels were associated with hypertension. A prospective case–control study on ischemic stroke verified that the serum antibody levels of the DIDO1 protein and DIDO1, FOXJ2, and CPSF2 peptides showed significantly higher odds ratios with a risk of AIS in patients with the highest quartile than in those with the lowest quartile, indicating that these antibody markers are useful as risk factors for AIS. Conclusions Serum antibody levels of DIDO1, FOXJ2, and CPSF2 are useful in predicting the onset of atherosclerosis-related AIS caused by kidney failure, hypertension, and DM, respectively. Supplementary Information The online version contains supplementary material available at 10.1186/s12916-021-02001-9.


Background
Atherosclerosis is a serious disease and a major cause of acute ischemic stroke (AIS) and acute myocardial infarction (AMI) [1]. Diabetes mellitus (DM) and chronic kidney disease (CKD) are closely related to and accompanied by atherosclerosis [2]. As atherosclerosis progresses, atherosclerotic plaques are formed on artery walls by foam cells, which are differentiated from smooth muscle cells or macrophages [3][4][5]. Diagnosing atherosclerosis is important to prevent the onset of AIS and AMI because the effectiveness of treatment and therapy is limited after their onset. Thus, to date, many risk factors and biomarkers including family history, age, obesity, smoking habit, dyslipidemia, hypertension, sleep, C-reactive protein level, interleukin-6 level, troponin level, and B-type natriuretic peptide level have been reported [6,7]; however, they are still insufficient. Genome-wide association studies on stroke have identified many genes such as NOTCH3 [8], CSTA [9], and COL3A1 [10]. However, lifestyle diseases such as stroke and atherosclerosis can be prevented by improving individuals' lifestyles.

Methods
The data that support the findings of this study are available from the corresponding author upon reasonable request.

Patient and controls
This study was approved by the Local Ethical Review Board of the Chiba University Graduate School of Medicine (Chiba, Japan) as well as the review boards of the cooperating hospitals or institutes. Sera were collected from patients who had provided informed consent. Each serum sample was centrifuged at 3000g for 10 min, and supernatant was stored at − 80°C until use. Repeated freezing and thawing of samples was avoided.
Serum samples from patients with DM, ulcerative colitis, CTEPH, pulmonary arterial hypertension (PAH), and OSA were obtained from Chiba University Hospital, and samples collected from patients with AIS, transient ischemic attack (TIA), asymptomatic cerebral infarction (asympt-CI), chronic-phase CI (cCI), and deep and subcortical white matter hyperintensity (DSWMH) were obtained from Chiba Prefectural Sawara Hospital, Chiba Rosai Hospital, Chiba Aoba Municipal Hospital, and Chiba Medical Center. The stroke subtype of each patient was also determined according to the criteria of the Trial of Org 10172 in Acute Stroke Treatment classification system [45]. In this analysis, large-artery atherosclerosis or small-artery occlusion (lacune) were included as AIS or cerebral infarction.
Serum samples from patients with AIS used in the preliminary screening were provided by BioBank Japan. Serum samples from patients with AMI were obtained from Kyoto University Hospital [44]. Serum samples associated with AIS, TIA, and AMI were obtained within 2 weeks after disease onset. Samples collected from patients with CKD were obtained from the Kumamoto cohort [46,47], whereas those collected from patients with colorectal carcinoma, esophageal squamous cell carcinoma, gastric cancer, breast cancer, and pancreatic cancer were obtained from the Department of Frontier Surgery, Chiba University Hospital. Serum samples from patients with Sjögren's syndrome were obtained from Chiba Children's Hospital. Serum samples from patients with rheumatoid arthritis and systemic lupus erythematosus (SLE) were obtained from the National Hospital Organization, Shimoshizu Hospital, and Chiba East Hospital [48]. Serum samples from healthy donors (HDs) were obtained from Chiba University, Port Square Kashiwado Clinic, Higashi Funabashi Hospital, and Chiba Prefectural Sawara Hospital. For comparisons with TIA and AIS, serum samples from HDs were selected from patients who exhibited no abnormalities on cranial magnetic resonance imaging.

ProtoArray ® screening
The first screening was performed using ProtoArray ® Human Protein Microarrays v. 4.0 (Thermo Fisher Scientific, Waltham, MA), which were loaded with 9480 proteins species as described previously [33,38,48]. In total, 30 serum samples (15 each from HDs and patients with atherosclerosis) were used to detect antigens specifically recognized by IgG antibodies in sera. Results were analyzed using the Prospector software (Thermo Fisher Scientific), which is based on M-statistics. When comparing the two groups, a cutoff for positivity was calculated for each protein using M-statistics. For both groups, the proportion of subjects with an immune response above the cutoff value was counted, and a P value representing the significance of the difference between both groups was calculated as described [49].

Peptide array method
The epitopes in the DIDO1 protein were screened comprehensively throughout the full-length DIDO1 protein using the peptide array method, in which we designed 83 peptides of 14mer derived from the DIDO1 protein.
These peptides were synthesized onto cellulose membranes using Fmoc amino acids (Auto-Spot Robot ASP222; ABIMED Analysen-Technik GmbH, Langenfeld, Germany) as described previously [50]. The membranes were washed five times with PBS containing 1% (w/v) bovine serum albumin, 0.05% Tween 20, and 0.05% NaN 3 (PBS-T-BSA) for 30 min each and then incubated with a 1:200 dilution of sera of HDs or patients with AIS for 18 h. The membranes were subsequently washed five times with PBS-T-BSA and treated with a 1: 10,000 dilution of FITC-conjugated goat antihuman IgG (Jackson ImmunoResearch, West Grove, PA) for 1 h. After washing, the fluorescence levels of peptide spots were detected using the Typhoon 9400 Imager (GE Healthcare Life Sciences) with a 488-nm/520-nm filter, as described previously [30,48,51].

Nested case-control study
A nested case-cohort study was conducted using the abovementioned AlphaLISA detection antibody levels. This study was nested within the Japan Public Health Center (JPHC)-based Prospective Study [52,53], which involved approximately 30,000 Japanese individuals aged 40-69 years at a baseline period of 1990-1994 whose plasma samples were stored. Serum DIDO1, bDIDO1-297, bFOXJ2-426, and bCPSF2-607 antibody levels were measured in 202 cases of incidental AIS in the cohort developed between the baseline and 2008 as well as in 202 controls whose sex, age (within 2 years), date of blood sampling (within 3 months), time since last meal (within 4 h), and study location (Public Health Center area) were matched with those of the cases. We used a conditional logistic regression model to estimate odds ratios and 95% confidence intervals (CIs) for AIS with respect to serum antibody levels of the DIDO1 protein and DIDO1, FOXJ2, and CPSF2 peptides.

Statistical analysis
Mann-Whitney U test, Student's t test, and Kruskal-Wallis test were used to determine the significance of the differences between two groups or among multiple groups. Correlations were calculated using Spearman's correlation analysis. All statistical analyses were performed using GraphPad Prism 5 (GraphPad Software, La Jolla, CA). The predictive values of putative disease markers were assessed using a receiver operating characteristic (ROC) curve analysis, and cutoff values were set to maximize the sums of sensitivity and specificity. All tests were two tailed, and P values of < 0.05 were considered statistically significant.

Results
Recognition of DIDO1, CPSF2, and FOXJ2 via serum igg antibodies of patients with atherosclerosis The first screening for AIS biomarkers was performed using SEREX and the protein array method. After the second screening using serum samples from HDs obtained from Chiba University Hospital and serum samples from patients with AIS obtained from BioBank Japan, we identified 74 antibody markers for AIS, some of which have been reported previously (Table 1). Preliminary validation tests using serum samples from patients with AIS obtained from BioBank Japan showed that the antibody levels against these three antigens, DIDO1, FOXJ2, and CPSF2, were reproducibly and significantly higher in AIS sera than control HD sera. Thus, we focused on three antibody markers highly associated with AIS.
The results of ProtoArray ® loaded with 9480 protein species showed that DIDO1 (accession no. BC000770.1) antibodies were observed in 4 out of 5 serum samples from patients with atherosclerosis and 1 out of 5 serum samples from HDs. FOXJ2 (accession no. NM_018416.2) antibodies were found to react with antibodies in 7 out of 15 serum samples from patients with atherosclerosis and none of the 15 serum samples from HDs. CPSF2 (100 kDa; accession no. NM_017437.1) antibodies reacted with antibodies in 5 out of 10 serum samples from patients with atherosclerosis and 2 out of 10 serum samples from HDs. Subsequently, GST fusion proteins that contained DIDO1 N or DIDO1 C were expressed in E. coli and purified via affinity chromatography. In addition, 5 predicted epitopes of CPSF2 and 1 of FOXJ2 were prepared, and the following preliminary experiments showed that serum bFOXJ2-426 and bCPSF2-607 antibody levels more highly reacted with serum antibodies in patients with AIS than with those in HDs. To examine epitopes in the DIDO1 protein recognized by serum antibodies, we synthesized a peptide array [30,48,51] loaded with 83 species of 14-mer peptides derived from the DIDO1 protein. bDIDO1-297, which was most closely associated with AIS, was also used as an antigen to evaluate serum antibody levels.
Presence of serum antibodies against purified proteins in patients with TIA or AIS We then confirmed the presence of antibodies against the GST fusion proteins of DIDO1 N , DIDO1 C , FOXJ2, and CPSF2 in serum samples from patients with TIA or AIS via Western blotting. GST, GST-DIDO1 N , GST-DIDO1 C , GST-FOXJ2, and GST-CPSF2 were recognized by the anti-GST antibody as reactions of 26-, 70-, 57-, 95-, and 110-kDa proteins, respectively ( Fig. 1). GST-DIDO1 N , GST-FOXJ2, and GST-CPSF2 were recognized by each specific commercial antibody. GST-DIDO1 N and GST-DIDO1 C (but not GST) reacted with antibodies in serum samples from patients with TIA-#07207, AIS-#07684, TIA-#07175, and AIS-#07115, whereas the serum antibodies of patients with AIS-#07684 and TIA-#07060 recognized GST-DIDO1 N but not GST-DIDO1 C . GST-CPSF2 reacted with antibodies in serum sample from a patient with TIA-#07175, and GST-FOXJ2 reacted with antibodies in serum sample from patients with AIS-#07115 and TIA-#07060. None of these antigenic proteins were recognized by serum IgG in patients with HD-#30017. As such, the reactivity of GST fusion antigenic proteins with serum antibodies may be primarily attributed to the antigenic protein regions but not to the GST domain. GST-DIDO1 N was   The serum levels of anti-bDIDO1-297 peptide antibodies (DIDO1pep-Abs) were also higher in patients with TIA and AIS than in HDs (Supplementary Figure S1).
Elevation of serum DIDO1, FOXJ2, and CPSF2 antibody levels in patients with AMI or DM We then examined DIDO1-Ab, FOXJ2-Ab, and CPSF2-Ab in HDs and patients with AMI and DM. Serum samples from patients with AMI were obtained from Kyoto University Hospital, those from patients with DM were obtained from Chiba University Hospital, and those from HDs were obtained from the Port Square Kashiwado Clinic. The mean age (±SD) of HDs and patients with AMI and DM was 58.29 ± 5.63, 58.20 ± 8.50, and 58.37 ± 9.11 years, respectively. A total of 128 samples each of HDs and patients with AMI and type 2 DM were assayed simultaneously using AlphaLISA on a 384-well plate. Serum DIDO1-Ab levels were not visibly different between the serum samples from HDs and those from patients with AMI or DM (Fig. 3a). However, serum FOXJ2-Ab levels were significantly higher in patients with AMI or DM than in HDs (Fig. 3d). Using cutoff values as described in the previous section, positive rates were 3.1% in HDs, 34.4% in patients with AMI, and 22.7% in those with DM (Table 3). Serum CPSF2-Ab levels were significantly higher in patients with DM (although not in those with  5) were separated through sodium dodecyl sulfate-polyacrylamide gel electrophoresis, followed by western blotting analysis using anti-GST (b), anti-DIDO1 (c), anti-FOXJ2 (d), anti-CPSF2 (e), and sera from HD-#30017 (f), patients with AIS-#07684 (g), AIS-#07581 (i), and AIS-#07115 (k), and those with TIA-#07207 (h), TIA-#07175 (j), and TIA-#07060 (l). The Coomassie brilliant blue (CBB) staining profile is also shown in a. M, molecular weight marker AMI) than in HDs (Fig. 3g). The positive rate of CPSF2-Ab in patients with DM was 13.3% ( Table 2).
Elevation of serum DIDO1, FOXJ2, and CPSF2 antibody levels in patients with CKD Next, we examined antibody levels in serum samples from patients with CKD, which is also closely related to atherosclerosis. Patients with CKD were divided into three groups: type 1 (diabetic kidney disease), type 2 (nephrosclerosis), and type 3 (glomerulonephritis). Serum samples from patients with CKD were obtained from the Kumamoto cohort and those from HDs were obtained from Chiba University. All CKD groups had significantly higher serum DIDO1-Ab and FOXJ2-Ab levels than HDs (Fig. 4a and Table 2. Receiver operating characteristic curve (ROC) analysis was performed to assess the ability of DIDO1-Abs (b, c), FOXJ2-Abs (e, f), and CPSF2-Abs (h, i) to detect TIA and AIS. The numbers in the figures indicate the cutoff values for marker levels, and the numbers in parentheses indicate the sensitivity (left) and specificity (right). The areas under the curve (AUC), and 95% confidence intervals (CIs) are also shown in Table 5 type 3 CKD were 7.3%, 43.4%, 37.5%, and 26.8%, respectively, and those of FOXJ2-Ab were 3.7%, 28.3%, 34.4%, and 13.8%, respectively (Table 4). No apparent difference was found in CPSF2-Ab levels between HDs and patients with any type of CKD (Fig. 4i, Table 4).

ROC analysis
The results of the ROC analysis are shown in Figs. 2b, c, e, f, h, i, 3b, c, e, f, h, i, 4b, c, d, f, g, h, j, k, l, S1B, and S1C and summarized in   Table 3. Responses to DIDO1-Abs (b, c), FOXJ2-Abs (e, f), and CPSF2-Abs (h, i) were also evaluated using ROC analysis, and summarized in Table 5 AUCs of FOXJ2-Ab were > 0.65 versus TIA, AIS, AMI, DM, and CKD, among which AUC was the highest versus type 2 CKD (0.8769; Table 4). AUC versus DM was relatively low (0.6584). Thus, FOXJ2-Ab may be associated with kidney failure and atherosclerosis, but it does not primarily reflect DM. However, CPSF2-Ab was not associated with AMI or type 1/type 2/type 3 CKD. The lowest P values were observed versus AIS and DM, suggesting that CPSF2-Abs reflect diabetic AIS.
Serum DIDO1-Ab, FOXJ2-Ab, and CPSF2-Ab levels in cancer Because autologous antibodies often develop in patients with cancer [11][12][13][14], we examined serum samples from patients with colorectal carcinoma, esophageal squamous cell carcinoma, gastric cancer, breast cancer, and pancreatic cancer obtained from Chiba University Hospital. Notably, serum DIDO1-Ab and CPSF2-Ab levels were not significantly different between HDs and patients with any type of cancer (Supplementary Table S1).
However, serum FOXJ2-Ab levels were significantly higher in patients with colorectal carcinoma but not in those with other types of cancer than in HDs.
Association of serum DIDO1-Ab, FOXJ2-Ab, and CPSF2-Ab levels with autoimmune diseases Autoantibodies may have causal effects on autoimmune diseases such as Sjögren's syndrome, rheumatoid arthritis, SLE, and ulcerative colitis. Some of these autoimmunity-related characteristics are known to be involved in the development of atherosclerosis [61][62][63][64]. We examined antibody levels in serum samples from patients with Sjögren's syndrome, rheumatoid arthritis, SLE, and ulcerative colitis. Serum DIDO1-Ab and FOXJ2-Ab levels were significantly higher in patients with rheumatoid arthritis and SLE (but not in those with Sjögren's syndrome or ulcerative colitis) than in HDs (Supplementary Table S2). Serum CPSF2-Ab levels were higher in patients with rheumatoid arthritis (but not in those with Sjögren's syndrome, SLE, or ulcerative colitis) than in HDs.
Association of serum DIDO1-Ab, FOXJ2-Ab, and CPSF2-Ab levels with pulmonary diseases OSA is frequently accompanied by hypertension. Serum anti-COPE was identified by SEREX screening using serum samples from patients with atherosclerosis, and   The antigens used were purified DIDO1-GST protein and synthetic peptides, bFOXJ2-426 and bCPSF2-607. The shown numbers are as described in Table 2; P values lower than 0.05 and positive rates higher than 10% are marked in bold. Box-whisker plots of the same results are shown in Fig. 3a, d, and g its level was elevated in patients with OSA compared with in HDs [65]. Pulmonary diseases including CTEPH and PAH are distinct from hypertension but could have an inflammatory condition similar to that in hypertension (e.g., elevation of Pentraxin 3 level) [66]. Serum FOXJ2-Ab levels were higher in patients with CTEPH and PAH (but not in those with OSA) than in HDs, whereas serum DIDO1-Ab and CPSF2-Ab levels did not show any apparent difference between HDs and patients with CTEPH, PAH, or OSA (Supplementary Table S3).

Correlation analysis
Comparative analysis of serum antibody levels and subject data was performed using 851 serum samples  Table 4. Responses to DIDO1-Abs (b-d), FOXJ2-Abs (f-h), and CPSF2-Abs (j-l) were also evaluated using the ROC analysis and are summarized in Table 5 obtained from Chiba Prefectural Sawara Hospital including 188 serum samples from HDs, 162 from patients with DSWMH, 18 from patients with asympt-CI, 66 from patients with TIA, 351 from patients with AIS, 66 from patients with cCI, and 66 from disease controls. Other subject information is shown in Supplementary  Table S4. Comparison using Mann-Whitney U test revealed that serum DIDO1pep-Ab, FOXJ2-Ab, and CPSF2-Ab levels were significantly higher in patients with TIA, AIS, and cCI (but not in those with DSWMH) than in HDs (Table 6, uppermost panel). Then, antibody levels were compared between males and females; those with or without DM, hypertension, CVD, and dyslipidemia; and those with or without smoking and alcohol intake habits. Hypertension was defined as a history of systolic blood pressure of > 140 mmHg, diastolic blood pressure of > 90 mmHg, or use of antihypertensive agents. Significantly higher serum DIDO1pep-Ab levels were observed in patients with hypertension, CVD, dyslipidemia, or a smoking habit (but not in those with DM) than in their control groups (Table 6, lower panels). Serum FOXJ2-Ab levels showed similar results, except that they were not correlated with dyslipidemia. Meanwhile, serum CPSF2-Ab levels were associated with DM, hypertension, and smoking habit but not with CVD or dyslipidemia. Sex and alcohol intake displayed no association with any of these three antibody levels. Spearman's rank-order correlation analysis was performed to determine the correlation between serum antibody levels of DIDO, FOXJ2, and CPSF2 peptides and subject parameters including general information such as age, body height, weight, body mass index, and degree of artery stenosis (maximum intima media thickness, max IMT). The following blood test data were also included: albumin/globulin ratio, aspartate aminotransferase, alanine amino transferase, alkaline phosphatase, lactate dehydrogenase, total bilirubin, cholinesterase, γglutamyl transpeptidase, total protein, albumin, blood CKD types 1, 2, and 3 correspond to diabetic kidney disease, nephrosclerosis, and glomerulonephritis, respectively. The upper panel indicates the numbers of all samples and samples from males and females as well as age (average ± SD). The lower panel summarizes the serum antibody levels examined by AlphaLISA using purified DIDO1-GST protein and synthetic bCPSF2 and bFOXJ2 peptides as antigens as described in the legend of Table 2. Box-whisker plots of the same results are shown in Fig. 4a, e, and i. P values lower than 0.05 and positive rates higher than 10% are marked in bold urea nitrogen, creatinine, estimated glomerular filtration rate, uric acid, amylase, total cholesterol, high-density lipoprotein cholesterol, triglyceride, sodium, potassium, chlorine, calcium, inorganic phosphate, iron, C-reactive protein, low-density lipoprotein cholesterol, white blood cells, red blood cells, hemoglobin, hematocrit, mean corpuscular volume, mean corpuscular hemoglobin, mean corpuscular hemoglobin concentration, red cell distribution width, platelets, mean platelet volume, procalcitonin, platelet distribution width, blood sugar, and glycated hemoglobin (HbA1c). All three antibody levels were correlated with age and max IMT but inversely correlated with height and weight and cholinesterase, total protein, and albumin levels. Serum DIDO1 and FOXJ2 antibody levels, but not serum CPSF2 antibody level, were correlated with alkaline phosphatase, white blood cell count, and mean corpuscular volume (Table 7). Blood sugar and HbA1c, which reflect DM, were not correlated with these antibody levels, except for a slight correlation (P = 0.195) between serum DIDO1pep-Ab level and blood sugar.

Immunohistochemical analysis of antigenic proteins
Assuming that autoantibodies against DIDO1, FOXJ2, and CPSF2 peptides develop in patients with atherosclerotic diseases, these antigenic proteins should be expressed at high levels in atherosclerotic lesions. As such, we also examined the expressions of antigenic proteins in surgically resected carotid atherosclerotic plaques via immunohistochemistry. The DIDO1 and CPSF2 proteins were predominantly expressed in the intima of atherosclerotic plaques, similar to the localization of vimentin and smooth muscle actin, which are markers for smooth muscle cells (Fig. 5). DHPS, reported as an atherosclerosis marker [55], was also expressed in smooth muscle cells. The expression of FOXJ2 showed a similar pattern as that of CD31-and CD34positive vascular endothelial cells. CD68 expression in macrophages was not similar to any of the other antigen expressions (Fig. 5).

JPHC cohort analysis
We conducted a case-control study nested within the JPHC-based Prospective Study, which involved approximately 30,000 plasma samples [52,53]. The antibody level against the DIDO1 protein was positively and strongly associated with a risk of AIS: odds ratios (95% CIs) were 3.99 (1.93-8.23), 3.40 (1.62-7.13), and 4.02 (1.94-8.35) for those with the second, third, and highest quartiles of antibody levels, respectively, versus for those with the lowest quartile (Table 8). Likewise, the antibody levels of the DIDO1, FOXJ2, and CPSF2 peptides were positively correlated with a risk of cerebral infarction: odds ratios (95% CIs) of the highest quartile were 2.66 (1.43-4.95), 2.24 (1.27-3.95), and 2.41 (1.33-4.37), respectively. These results indicate that the antibody markers against the DIDO1 protein and DIDO1, FOXJ2, and CPSF2 peptides are useful in predicting the onset of AIS.

Three novel antibody markers for atherosclerosis
We performed large-scale screening using SEREX and the protein microarray method and identified 69 candidate antigenic proteins related to atherosclerosis (Table  1). In the present study, we focused on three antigens-DIDO1, FOXJ2, and CPSF2-that appeared to be of Area under the curve (AUC), 95% CI, cutoff value, sensitivity (%), specificity (%), and P value of the ROC analysis are shown. Purified GST-DIDO1 protein and synthetic peptides-bDIDO1-297 (DIDO1pep), bFOXJ2-426, and bCPSF2-607-were used as antigens. P values lower than 0.05 and AUCs higher than 0.7 are marked in bold  The subjects were divided as follows: sex (male and female); presence (+) or absence (−) of complication of DM, hypertension (HT), cardiovascular disease (CVD), or dyslipidemia, and lifestyle factors (smoking and alcohol intake habits). Antibody levels (Alpha counts) were compared using the Kruskal-Wallis test (upper panel) and the Mann-Whitney U test (lower panels). Sample numbers, averages, and SDs of counts as well as P values are shown. Significant correlations (P < 0.05) are marked in bold much interest in relation to AIS. The presence of antibodies against these proteins was confirmed by Western blotting (Fig. 1). We then examined epitopes and selected bDIDO1-297, bFOXJ2-426, and bCPSF2-607 as useful antigenic peptides to measure serum antibody levels. The amino-terminal half of DIDO1 was also used as an antigen. Serum antibody levels of these antigens were more elevated in patients with AIS and TIA than in HDs (Fig. 2, Supplementary Figure S1). All of bDIDO1-297, bFOXJ2-426, and bCPSF2-607 were closely correlated with max IMT (Table 7), which is a typical index of the development of atherosclerosis leading to AIS and CVD [67][68][69][70]. Thus, these serum antibodies can be markers for atherosclerosis. A casecontrol study nested within the JPHC-based Prospective Study showed that the three antibody markers are associated with the risk of cerebral infarction and indicated that these markers are useful in predicting the onset of cerebral infarction (Table 8). However, they had distinct characteristics. The DIDO1 protein was first identified as a regulator of apoptosis [71]. Serum DIDO1pep-Ab levels were elevated in patients with TIA, AIS, cCI, CKD, rheumatoid arthritis, and SLE but not in those with AMI, DM, any type of cancer, or ulcerative colitis (Figs. 2, 3, and 4; Tables 2, 3 and 4; Supplementary Tables S2 and S3). In particular, the AUC values of DIDO1 versus CKD were > 0.8 ( Table 5), suggesting that DIDO1-Ab reflects kidney failure basically and atherosclerosis indirectly.
FOXJ2 is a member of the forkhead family of transcription factors [72]. Serum FOXJ2-Ab levels were elevated in patients with TIA, AIS, cCI, AMI, DM, CKD, colorectal carcinoma, rheumatoid arthritis, and SLE compared with in HDs (Figs. 2, 3, and 4; Tables 2, 3, and 4; Supplementary Tables S1 and S2). Serum FOXJ2-Ab levels correlated well with hypertension (Table 6) and were elevated in patients with CTEPH and PAH (Supplementary Table S3), suggesting that these levels reflect systemic arterial hypertension and can differentiate hypertension-related diseases. In fact, hypertension is also a risk factor for colorectal carcinoma [73], and SLE is frequently associated with PAH [74].
Collagen diseases such as rheumatoid arthritis and SLE are high-risk groups of AIS and AMI [61,62]. Serum DIDO1-Ab and FOXJ2-Ab levels were significantly associated with rheumatoid arthritis and SLE but not with Sjögren's syndrome or ulcerative colitis (Supplementary Table  S2). It is possible that DIDO1-Ab and FOXJ2-Ab are discriminant in the case of AIS of which one of the causes is a collagen disease. That is, each marker may correspond to a different cause of atherosclerosis.
CPSF2 encodes the 100-kD subunit of CPSF, which plays a central role in the 3′ processing of pre-mRNA [75]. Serum CPSF2-Ab levels were associated primarily with AIS and DM and partly with TIA, cCI, esophageal squamous cell carcinoma, and rheumatoid arthritis but not with AMI, CKD, CVD, colorectal carcinoma, gastric cancer, breast cancer, pancreatic cancer, Sjögren's syndrome, SLE, or ulcerative colitis (Figs. 2, 3, 4; Tables 2, 3, and 4; Supplementary Tables S1 and S2). Serum CPSF2-Ab levels were correlated with aortic hypertension (Table 6) but not with pulmonary hypertension such as CTEPH and PAH (Supplementary Table S3). Moreover, the levels correlated most closely with max IMT (Table 7), indicating that CPSF2-Ab can mainly detect DM-caused atherosclerosis leading to AIS.

Relationship between BMP/TGF-β and atherosclerosis
Bone morphogenetic proteins (BMPs) are involved in the transforming growth factor-β (TGF-β) superfamily. It is well documented that BMP signals play important roles in the development of atherosclerosis [76,77]. BMP-2 and BMP-4 expressions were elevated in atherosclerotic endothelium [78,79], and plasma BMP-2 levels are elevated in patients with type 2 DM [80]. Chronic infusion of BMP-4 induces endothelial dysfunction and hypertension [81], and treatment with the BMP antagonist, matrix Gla protein, and BMP inhibitors prevents the development of ATS [82,83]. On the other hand, the knockdown of the BMP type II receptor BMPRII accelerates ATS [84]. Therefore, BMP family members may play a subtle regulatory role in the development of ATS. It should be noted that DIDO1 is the target gene of BMP and promotes cell attachment, migration, invasion, and apoptosis resistance in melanoma [85].
CPSF proteins interact with Smad via Smicl and potentiate TGF-β/BMP-stimulated Smad-dependent transcriptional responses [86,87]. We previously reported the elevation of autoantibodies against SOSTDC1 and NBL1/ DAN, which are the antagonists of BMP, in patients with AIS [48] and OSA [36], respectively. As such, it is possible that some, if not all, autoantibodies against TGF-β/BMPrelated proteins play causal or suppressive roles in the development of atherosclerosis-related diseases.

Involvement of marker genes in development and differentiation
DIDO1 is the target gene of Oct4, Sox2, and Nanog; in reverse, Nanog and Oct4 are the target genes of DIDO1 [88]. Thus, DIDO1 plays a key role in the self-renewal of embryonic stem cells. Futterer suggested that DIDO1 is a switchboard that regulates embryonic stem cell transition from pluripotency maintenance to differentiation [89]. During the development of atherosclerosis, smooth muscle cells differentiate into foam cells to form atheroma [90]. Highly expressed DIDO1 in intimal smooth muscle cells (Fig. 5) may have an important role in their differentiation into foam cells. FOXJ2 expression is also regulated by Oct4 and involved in oocyte development [91]. Transient FOXJ2 transgenesis experiments have shown that FOXJ2 overexpression has a lethal effect on embryonic development from E10.5 [92]. FOXJ2 is also involved in differentiation and inhibits TGF-β1-induced epithelial-mesenchymal transition [93]. Thus, high FOXJ2 expression (Fig. 5) may affect otherwise normally functioning vascular endothelial cells.

Relationship between atherosclerosis and cancer
BMP-induced DIDO1 promotes cell attachment, migration, invasion, and apoptosis resistance in melanoma [85]. Serum FOXJ2-Ab levels, which correlated well with hypertension, were elevated in patients with colorectal carcinoma (P < 0.001) but not in those with esophageal squamous cell carcinoma, gastric cancer, breast cancer, or pancreatic cancer (Supplementary Table S1). This is consistent with the report that hypertension is also a risk  factor for colorectal carcinoma but not for esophageal squamous cell carcinoma or gastric cancer [73,94]. FOXJ2 overexpression is associated with poor prognosis, progression, and metastasis in nasopharyngeal carcinoma [95]. FOXQ1, a member of the FOX family, is overexpressed in colorectal cancer, and it enhances tumorigenicity and tumor growth [96]. However, it has been reported that FOXJ2 suppresses migration and invasion in extrahepatic cholangiocarcinoma [97], hepatocellular carcinoma [98], glioma [99], and breast cancer [100]. Thus, FOXJ2 can promote or suppress malignancy depending on cancer type, which may account for the colorectal carcinoma-selective association of FOXJ2-Abs (Supplementary Table S1).
CPSF2 has a suppressive role in cell invasion in thyroid cancer and cancer stem cell population [101]. It is involved in the 6-gene prognostic signature for hepatocellular carcinoma overall survival prediction [102]. Our results showed only a slight association of CPSF2-Abs with esophageal squamous cell carcinoma (P < 0.01) but not with other types of cancer (Supplementary Table  S1). CPSF2-Ab may reflect DM-caused atherosclerosis as described above, and the causes of cancer and atherosclerosis overlap with each other. Thus, CPSF2-Abs may be associated indirectly with some types of cancer.

Characteristics of antibody biomarkers
Atherosclerosis progresses slowly over many years, finally leading to the onset of AIS or AMI. The prodromal stages of AIS and AMI may be accompanied by tissue destruction in arteries. The development of autoantibodies may be caused by high expressions of antigenic proteins in arteries followed by tissue destruction-induced exposure of antigens to immune cells. Repeated destruction/exposure can considerably increase antibody levels while keeping the antigen level low. Thus, antibody markers are much more sensitive than antigen markers. In addition, serum IgG proteins are highly stable and not easily degraded. As such, antibody markers are highly suitable for detecting trivial alterations caused by early-stage lesions. This is consistent with results that in this study, serum DIDO1-Ab, FOXJ2-Ab, and CPSF2-Ab were elevated not only in patients with AIS but also in those with TIA, a prodromal lesion of AIS (Fig. 2).
AIS is a severe disease that often leads to death. Once it occurs, even without death, affected patients require a long rehabilitation period, with this disease also being the first cause of being bedridden. However, if the onset of AIS is predicted, most patients can avoid it via an appropriate treatment. Therefore, the development of highly sensitive and predictive biomarkers is eagerly expected. We discovered that serum DIDO1-Ab, FOXJ2-Ab, and CPSF2-Ab are useful in predicting the onset of AIS (Table 8), although these three markers may not be sufficient to diagnose all AIS types. AIS is a multifactorial disease, and each marker may associated with a different cause. The more biomarkers are identified, the more precise predictions can be achieved. Further investigations may be necessary for practical use.

Limitation
Although the present study suggests kidney failureassociated DIDO1-Ab, hypertension-related FOXJ2-Ab, and DM-related CPSF2-Ab markers as risk factors of AIS, further study using the increasing number of specimens is needed to verify the suggestion. Because our present study was carried out using specimens obtained from hospitals and universities in Japan, it is obscure whether our conclusion is generalized in other population. Further international collaborative research using the specimens from many countries is necessary for the practical use in the world.

Conclusions
Serum DIDO1-Ab, FOXJ2-Ab, and CPSF2-Ab appear to be useful for diagnosing AIS and may originate from kidney disease, hypertension, and DM, respectively.  The odds ratios (ORs) and 95% CI of the 2nd, 3rd, and the highest (max) quartiles versus the lowest quartile are shown for AIS with respect to the antibody levels of DIDO1 protein, DIDO1 peptide, FOXJ2 peptide, and CPSF2 peptide