Expression profiles of GSE96804 mRNA were obtained from the GEO database (GPL17586 platform, Affymetrix Human Transcriptome Array 2.0) (12). In total, 61 tissue biopsies, 41 tissue samples from DN tissue samples and 20 from the normal, were obtained from the National Clinical Research Center of Kidney Diseases, Jinling Hospital, Nanjing University School of Medicine. ^“Limma^” packaged (13) in R software was used to process data and the “ggplot2” (14), “Pheatmap” packages for drawing of figures. DEGs were identified with |log Fold Change | ≥1 & adj P Val < 0.05.

GO analysis was conducted using the ‘cluster Profiler’ (15), ‘GO plot’, and ‘ggplot2’ packages for up- and down-regulated DEGs with altered DN and normal kidney tissue. The KEGG pathway enrichment analysis was completed by DEGs, and the figures were generated with the packages “ggplot2” and “enrich plot”.

Gene co-expression networks of DN patients were constructed based on the GSE96804 microarray dataset by the “WGCNA” package (16). The soft-thresholding power was five when 0.9 was used as the correlation coefficient threshold, and 50 was chosen as the minimum number of genes in modules. To merge possible similar modules, we defined 0.25 as the threshold for cutting height. A heatmap between the correlation between modules and DN was drawn, and the ME-brown gene module was the most related to DN.

729 secreted genes are available for the HPA database (https://www.proteinatlas.org). Venn diagram (https://bioinfogp.cnb.csic.es/tools/venny/index.html) demonstrates the genes which are commonly associated with the 3 datasets (DEGs, WCANA, and secreted to blood genes). In the common genes, we further filtered the core secretory genes by using different machine algorithms (Lasso and SVM-RFE algorithm).

Lasso logistic regression is a machine learning process that determines covariates by seeking the λ value that minimizes the classification error (17). The “glmnet” package was utilized to structure the LASSO model. Meanwhile, With SVM-RFE, an approach for building machine training on support vector machines, we detect the optimal variables by decimating the feature vectors created by svm (18). Recursive features of differential genes were acquired and erased by running the “e1071 package”, and the research was conducted by applying the Lapply function to sort all the features of the training set. Ultimately, the error rate is minimized and the hub gene is eventually obtained.

The common genes derived from these two machine algorithms are demonstrated by the Venn diagram, heat maps, line plots, and deviation plots.

As illustrated in our previous research, the expression of biomarkers was confirmed by using the Nephroseq database (https://www.nephroseq.org/resource/main.html) (19). Meanwhile, by using the database, biomarker expression and renal function data were analyzed for correlation.

The STZ-induced DN mouse model was elucidated in detail in our previous research (19), and among them, there were 5 mice in the control group (Ctrl) and 5 mice in the diabetic nephropathy group (DN). Following the successful construction of the DN mouse model, we conduct the collection of experimental animals. The research was approved by the Ethics Committee of Qilu Hospital, Shandong University (Approval No: KYLL-2020(KS)-030).

The experimental operation of Western Blot was as described (20). The main antibodies are described as follows: APOC1(1:2000, Abcam, USA), GAPDH (1:4000; Proteintech Group, China).

Immunohistochemistry and immunofluorescence of kidney tissue sections as previously described (21) The main antibodies are described as follows: APOC1(1:200, Abcam, USA), Goat anti-Rabbit IgG Dy-Light 488 (1:500; Abbkine Scientific Company, USA).

We have collected serum specimens from DN patients and healthy. Detection of biomarkers in serum with commercial Elisa kits, ELISA method, in DN patients and healthy. Follow the experimental steps in the Elisa kit instructions to detect the expression level of the marker in the serum (Apolipoprotein CI ELISA kit, Abcam, ab108808, USA).

The “PROC” package was used to construct Receiver Operating Characteristic (ROC) curves to characterise hub gene to evaluate the diagnostic value of DN, as previously described (19).

Data are expressed as mean ± SEM. Software R4.1.2 was used to draw the research Figures. GraphPad Prism 6.01 software was used in statistical data analysis. Between the two groups, Student’s t-test was used if the data matched the normal distribution, and the Kruskal-Wallis test was used for non-normally distributed data. For statistical analysis of the correlation between the two characters, the Spearman test was applied. Statistical significance was set at P < 0.05, *P < 0.05, **P < 0.01, ***P < 0.001.

The experimental design was illustrated in Figure 1 . Compared to transcripts of controls, 504 DEGs were identified by patients, respectively. Our analysis of the results is summarized in the volcano plots, which reflect that 257 genes are up-regulated in DN and 247 genes are down-regulated in DN ( Figure 2A ). In the illustration, red represents up-regulated and green indicates down-regulated genes. Results demonstrated two clusterings of this data, namely the clusters Control and DN which represents the control group and the DN patients in the heatmap ( Figure 2B ). Analysis of GO in DEGs determined shared GO terms linked to organic acid catabolic processes, and extracellular matrix organization ( Figure 2C ). Enrichment pathways to KEGG are associated with the following: Arginine and proline metabolism, Glycine, serine and threonine metabolism, and Protein digestion and absorption ( Figure 2D ).

The network topologies for the analysis of various soft threshold powers were identified and the choice of 11 to structure the joint expression network was considered reasonable ( Figure 3A ). The similarity in gene expression is ascertained by pair-weighting correlation metrics, and clustering is performed using topological overlapping metrics. Gene modules are marked with color at the bottom ( Figure 3B ). Pearson correlation coefficients for ME and disease were calculated for all modules demonstrating the intimate characteristics of the modules with DN. ME-brown (R = 0.53, P = 1e-05) potentially represented particular features of DN patients ( Figure 3C ). Furthermore, we observed that the correlation coefficient between the GS of DN and the module members was high in brown modules (R = 0.47, = 3.1e-21, Figure 3D ). There was potential biological relevance to heightened co-expression of the genes in the ME-brown module.

Venn diagram illustrating common genes across algorithms, filtering for 17 potential secretory genes that may be functionally essential in DN ( Figure 4A ). By using 2 machine algorithms, Lasso and SVM-RFE, to recognize the characteristic genes of DN. The 17 secreted genes are displayed in Figure 4B . By using 2 machine algorithms, LASSO and SVM-RFE, characteristic genes of the DN were identified again. The Lasso algorithm filtered out 9 potential hub genes ( Figures 4C, D ), while the SVM algorithm filtered out 7 potential hub genes ( Figures 4E, F ).

The Venn diagram illustrates 2 machine algorithms obtained common 6 hub secretory genes (APOC1, CCL21, INHBA, RNASE6, TGFBI, VEGFC, Figure 5A ). Furthermore, the expression of the six genes in the GSE96804 cohort is illustrated by heatmap, line graphs, and deviation plots ( Figures 5B–D ). The results revealed that 6 secretory gene generators screened for the research were significantly more over-expressed in the diabetic nephropathy population.

APOC1 expression is elevated in patients with diabetic nephropathy through multiple cohorts of the experimental GEO database (GSE96804, GSE47185, GSE30122, and the ERCB Nephrotic Syndrome Tublnt cohorts in Nephroseq database, Figures 6A–D ). ApoC1 expression was elevated in the kidney tissue of mice with DN by Western blot (P < 0.05, Figure 6E ). What’s more, we revealed that APOC1 was expressed predominantly in the glomerulus by immunohistochemistry of mouse kidney tissue ( Figure 6F ). These measurements were confirmed by tissue immunofluorescence ( Figure 6G ).

Correlations between APOC1 and the clinical information were validated by employing multiple cohorts from the Nephroseq database. Outcomes demonstrated the APOC1 expression was positively correlated with proteinuria in Schmid diabetes tubint cohorts (R² = 0.515, P = 0.013, Figure 7A ). However, associations of APOC1 expression are negatively correlated with GFR in Woroniecka Diabetes Tublnt cohorts (R² = 0.552, P = 0.014, Figure 7B ). Additionally, in ERCB Nephrotic Syndrome Tublnt cohorts, APOC1 expression was positively correlated with proteinuria (R² = 0.632, P = 0.018, Figure 7C ).

Altogether 20 healthy and 20 DN patients were enrolled in the research, and the Baseline details were presented in Table 1 . Significantly, Elisa results demonstrated that APOC1 expression in the serum of DN patients was 1.358±0.1292μg/ml, compared to 0.3683±0.08119μg/ml in the healthy population ( Figure 8A ). APOC1 was significantly elevated in the sera of DN patients and the difference was statistical significant (P > 0.001). Furthermore, APOC1 diagnostic effectiveness for DN as demonstrated by ROC curves (AUC = 92.5%, sensitivity = 95%, and specificity = 97%, P < 0.001, Figure 8B ).