The study participants for cytokine microarray and validation were recruited between July 2014 and August 2019 from the Keimyung University Dongsan Hospital, Daegu, Korea. The study was approved by the Keimyung University Dongsan Hospital Institutional Review Board, with the protocol numbers being: 2013-09-003, 2015-03-010, 2016-03-028, 2016-05-018, and 2018-05-058. All subjects voluntarily provided informed consent. Subjects were divided into groups of normoglycemia, prediabetes, and T2DM according to the ADA guidelines [2].

Blood samples were obtained from each patient using ethylenediaminetetraacetic acid tubes (Becton Drive, Franklin Lakes, NJ, USA). Plasma was isolated after centrifuging whole blood at 3,000 ×g for 3 minutes and stored at −80°C for further analysis. Proteins were extracted from plasma using protein extraction buffer (Full Moon BioSystems, Sunnyvale, CA, USA). After extraction, the protein solution was purified using a gel matrix column that was included in the antibody array assay kit (Full Moon BioSystems). The column was vortex-mixed for 5 seconds and hydrated for 60 minutes at room temperature. Following centrifugation at 750 ×g for 2 minutes, the column was placed into a collecting tube, and 100 μL of the protein sample was added followed by the same centrifugation condition. The concentration of the purified sample was measured using a Pierce BCA protein assay kit (Thermo Fisher Scientific, Waltham, MA, USA) using a NanoPhotometer (Implen, Munich, Germany), and its purity was confirmed using UV spectroscopy.

Each protein sample of 50 μL was brought up to a volume of 75 μL using labeling buffer and treated with 3 μL of 10 μg/μL biotin/dimethylformamide solution. The samples were incubated for 90 minutes with gentle shaking. Samples were then treated with 35 μL of stop reagent and incubated for 30 minutes while gently shaking. The antibody microarray slide (Full Moon BioSystems) was incubated with 30 mL of blocking solution with shaking at 60 rpm for 30 minutes and washed with distilled water. This step was repeated thrice. After rinsing, each labeled sample was mixed with 6 mL of coupling solution. Then, the blocked array slide was incubated in a coupling dish with the coupling mixture on a shaker for 2 hours at 60 rpm. After coupling, the slide was soaked in 30 mL of washing solution in a Petri dish and washed six times on a shaker at 60 rpm for 5 minutes. The slides were rinsed with distilled water. The detection mixture was prepared by adding 30 μL of 0.5 mg/mL Cy3-streptavidin (GE Healthcare, Chalfont St. Giles, UK) to 30 mL of detection buffer. The coupled array slide was treated with the detection mixture in a Petri dish on a shaker at 60 rpm for 20 minutes. After this step, the slide was washed six times with 30 mL of washing solution with shaking at 60 rpm for 5 minutes. The slides were rinsed with distilled water. The whole experimental procedure was carried out at room temperature. All solutions used in this analysis without manufacturer specified were purchased from Full Moon BioSystems.

Slide scanning was performed using the GenePix 4100A scanner (Axon Instruments, Berkeley, CA, USA). The slides were fully dried and scanned within 24 to 48 hours at a resolution of 10 μm using optimal laser power and photomultiplier tube. The obtained the scans were gridded and quantified using the GenePix 7.0 software (Axon Instruments). The numerical data were analyzed using the Genowiz 4.0 software (Ocimum Biosolutions, Indianapolis, IN, USA). After analysis, the data on protein information were annotated using the UniProt database (www.uniprot.org).

Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway analyses were performed using the Database for Annotation, Visualization and Integrated Discovery (DAVID) v6.8 [13,14]. Functional network analysis was conducted with cytokines that showed at least a 1.5-fold difference in patients with prediabetes compared to subjects with normoglycemia, using the Search Tool for the Retrieval of Interacting Genes/Proteins (STRING) database in Cytoscape [15], requiring a minimum confidence score of 0.4.

The differential expression of cytokines in the plasma from patients with prediabetes and T2DM was confirmed using enzyme-linked immunosorbent assay (ELISA). PROZ levels were measured using the human PROZ Quantikine ELISA kit (MyBioSource, San Diego, CA, USA) according to the manufacturer’s protocol.

All statistical analyses were performed using the SPSS version 25.0 (IBM Co., Armonk, NY, USA). Cytokine microarray and ELISA data were expressed as mean±standard error of the mean obtained from two independent experiments. Statistically significant differences between groups were determined using the one-way analysis of variance (ANOVA) and Tukey’s post hoc test or two-tailed Student’s t test. Clinical parameters of the study groups were expressed as the mean±standard deviation and were analyzed using the chi-square test and one-way ANOVA. Correlations between PROZ and clinical parameters were analyzed using the Pearson correlation coefficient. A receiver operating characteristic (ROC) plot was drawn and area under the curve (AUC) was calculated using ‘pROC’ and ‘EPI’ packages of R version 4.0.3 (R Foundation for Statistical Computing, Vienna, Austria). Results were considered statistically significant at P<0.05.

Supplemental Table S1 shows the baseline characteristics of subjects in the cytokine microarray analysis. There were no significant differences in clinical parameters among normoglycemia, prediabetes, and T2DM groups, except for those seen with FPG, HbA1c, homeostasis model assessment for insulin resistance (HOMA-IR), sex, age, and creatinine levels. To identify cytokines differentially secreted in subjects with prediabetes and T2DM compared to those with normoglycemia, we collected blood samples from each group and analyzed respective cytokine profiles using a cytokine microarray analysis. The expression patterns of the 41 cytokines that differed by 1.5-fold or more in at least one comparison between any two groups were shown in the heat map image (Fig. 1). We separately sorted out the numbers of up-regulated proteins and down-regulated proteins by comparison between groups using Venn diagram (Supplemental Fig. S1). We also plotted cytokines showing significant changes in subjects with normoglycemia, prediabetes, and T2DM using principal component analysis (PCA) analysis. PCA indicates that T2DM group seems to be scattered; however, it showed distinguishable pattern of separated three groups which are normoglycemia, prediabetes and T2DM (Supplemental Fig. S2). Table 1 displays the representative cytokines that showed at least a 2-fold difference in each comparison between the two groups. We identified carbohydrate antigen (CA) 15-3 and nerve growth factor (NGF)-β, whose expression levels were significantly decreased in the prediabetes group, but not in the T2DM group, when compared to the normoglycemia group. C-reactive protein (CRP) and B lymphoid kinase levels were significantly increased in both the prediabetes and T2DM groups compared with those in the normoglycemia group. In contrast, PROZ, interleukin (IL) 15, signal transducer and activator of transcription 6, insulin-like growth factor II, interferon-lambda 2, thyroglobulin, alpha-fetoprotein (AFP), and carcinoembryonic antigen (CEA) were significantly decreased in both the prediabetes and T2DM groups compared with those in the normoglycemia group.

We carried out a bioinformatic analysis to investigate the associated biological activity of cytokines and screened 33 cytokines that differed by at least 1.5-fold in the prediabetes group compared with the normoglycemia group (Supplemental Table S2). We determined the predicted enriched biological processes and pathways associated with these 33 cytokines using DAVID, the results of which are presented in Supplemental Fig. S3. We used STRING to identify the protein–protein interactions (PPIs) for the 33 cytokines. It revealed strong enrichment of biological processes such as immune response, inflammatory response, and cell proliferation regulation (Fig. 2). However, PROZ, follicle stimulating hormone, CEA, and CA15-3 were not involved in these biological processes.

In order to find a promising biomarker for prediabetes from the listed cytokines, we decided to validate PROZ. We selected PROZ because its expression showed a marked decrease in patients with prediabetes compared to those with normoglycemia (Fig. 3A), and it has not been identified as a factor associated with diabetes in any previous study. We performed ELISA on the samples from all the groups to determine PROZ levels. Table 2 shows the baseline characteristics of the subjects of the different groups that underwent validation. Through a validation study, the average plasma levels of PROZ in 35 subjects with prediabetes and 36 subjects with T2DM were determined and compared with those in the 71 subjects with normoglycemia. Plasma PROZ levels were decreased in subjects with prediabetes (1,490.32±367.19 pg/mL) and T2DM (1,583.34±465.43 pg/mL) compared to those in subjects with normoglycemia (1,864.07±450.83 pg/mL) (P<0.001) (Fig. 3B).

The correlations between PROZ and the other clinical parameters are shown in Table 3. There were significant negative correlations between PROZ and FPG (Pearson R=–0.284, Pearson P=0.001) as well as between PROZ and HbA1c (Pearson R= −0.216, Pearson P=0.010). PROZ had negative correlations with insulin and HOMA-IR, whereas it had a positive correlation with homeostasis model assessment of β-cell function (HOMA-β); however, these were not statistically significant. Other parameters did not show a significant correlation with PROZ.

Fig. 4 shows the ROC curves of PROZ levels for prediabetes and T2DM compared with those in normoglycemia. The AUC of PROZ between prediabetes and normoglycemia was 0.751 (0.650 to 0.852), sensitivity was 62.9%, and specificity was 80.3%. The AUC of PROZ between T2DM and normoglycemia was 0.686 (0.572 to 0.800), sensitivity was 50.0%, and specificity was 84.5%.