We realize that most of the samples in our study had mild to moderate arteritis (v1) and were found in early indication biopsies within the first month post-transplant. g, glomerulitis; HLA, human leukocyte antigen; i, interstitial inflammation; mm, mesangial matrix expansion; MMF, mycophenolate mofetil; ptc, peritubular capillaritis; PRA, panel reactive antibody; t, tubulitis; TAC, tacrolimus; ti, total interstitial inflammation; v, intimal arteritis. For a validation set, 12 eIV and 8 TCMRV biopsy samples from patients transplanted between 2010 and 2016 were retrospectively identified and validated by RT-qPCR (Table 2). The Institutional Review Board (IRB) of IKEM approved the study protocol (G09-12-20), and all patients provided informed consent to participate in the study. Table 2 Characteristics of patients in the validation set (%)0.339?Diabetes1 (12.5)5 (41.7)?Glomerulonephritis2 (25)2 (16.7)?Polycystosis1 (12.5)2 (16.7)?TIN2 (25)0?Hypertension1 (12.5)2 (16.7)?Ischemic nephropathy1 (12.5)0?Other01 (8.3)method of the relative quantification (RQ) Manager Software v?1.2.1 (Applied Biosystems) with normalization to an endogenous control (HPRT1). The endogenous Montelukast control was chosen from three candidate housekeeping genes (GAPDH-Hs99999905_m1, PGK1-Hs99999906_m1, HPRT1- Hs01003267_m1) using NormFinder (www.mdL.dk) as the gene with the most stable expression (HPRT1 with a stability value of 0.003). As a calibrator, one of the samples with a good expression profile on all of the target Montelukast genes was used. All investigated mRNAs were measured in triplicate for each sample. Risk of overfitting In our study, we deal with the well-known problem (the large number Montelukast of variables and the small number of samples) that represents a specific case of ill-posed problem and may result in overfitting [26,27]. This risk is minimized by careful handling with the train, test and validation datasets. First, we employ LOOCV to split between train and test sets. Both gene selection and classifier construction are performed solely on train sets, while the corresponding test sets serve for their evaluation. In particular, the SVMCRFE procedure for gene selection was re-performed with each iteration of the LOOCV procedure, so that the features are selected from each train set and applied independently to each test set. In general, this train-test split allows us to detect overfitting and avoid complex biomarkers that heavily overfit the data used for model construction. It enables to propose simple biomarkers and to smoothly distinguish between them in terms of their performance. Second, we work with the?independent RT-qPCR?data set that serves to validate the selected biomarkers, remove the selection bias and get an unbiased estimate of their classification accuracy (expressed in terms of AUC to compensate for unbalanced classes) [27,28]. Statistical methods Normality of the data was tested using the KolmogorovCSmirnov test. Nonparametric values are presented as median and interquartile range. Two groups were compared by the two-tailed MannCWhitney U-test and three groups by the KruskalCWallis test with adjustment by the Bonferroni correction for multiple tests. For comparison of categorical data, the 2 2 Fisher exact test was Montelukast used. Two-sided and compared with eIV (Figure 6). The validated genes are significantly involved in regulation of immune system process, T-cell differentiation, activation, proliferation, B-cell activation, overall lymphocyte and leukocyte activation, immune response-regulating cell signal transduction, and apoptosis. Open in a separate window Figure 6 Validation of microarray analysis by RT-qPCR of early indication biopsy samplesScatter plots show top 10 10 deregulated genes between TCMRV and eIV. Agreement between microarray and RT-qPCR data Validation of reference genes in the validation set was defined as both qualitative (direction) and quantitative agreement between microarray and RT-qPCR Rabbit Polyclonal to Synaptophysin measurements. The direction of RT-qPCR gene expressions agreed with the microarray technique in 100% of validated genes. Quantitative agreement between microarray and RT-qPCR was confirmed by a significant correlation of normalized data (Pearson = 0.663, Montelukast em P /em =0.00006) (Supplementary Figure S2). To further validate differences in the transcriptome of the study groups, the SVMCRFE classifiers were trained on RT-qPCR data. LOOCV confirmed that the genes selected for validation from microarray data showed around 80% accuracy (ACC) and a 0.75 area under the curve (AUC) (Supplementary Figure S3) thus confirming reasonable gene selection for external RT-qPCR validation. Discussion In the present study, we investigated the transcriptome of eIV with paucity of TI and TCMRV with rich TI. Our main results are that the transcriptome of eIV revealed a weak immunologic signature compared with TCMRV and showed similarity with non-rejection 3-month protocol biopsy. Based on our results, eIV may feature a non-rejection phenotype and reflect peritransplant injury. As the current Banff histopathological criteria consider intimal arteritis (after exclusion of ABMR) to be at least type II of TCMR irrespective of TI, our results agree with calls for reassessment of the current approach in histology interpretation. Furthermore, difference in non-rejection.