Identification Of Qualitative Transcriptional Signature For The Early Diagnosis And Risk Assessment Of Colorectal Cancer

Colorectal cancer(CRC)is one of the most common malignancies with a high mortality rate.Pathological examination using colonoscopy combined with biopsy samples is the gold standard for the diagnosis of CRC.However,patients with early stage usually with small lesions,and there are usually exist some problems such as minimum biopsy specimen and sampling inaccurately(CRC adjacent normal),leading to a higher miss rate of CRC.Therefore,it is necessary to develop a CRC specific molecular signature to provide an auxiliary diagnosis for colonoscopy,improving the accuracy of the diagnosis of CRC.Currently,studies have proposed some diagnostic signatures of CRC,but they generally considered CRC adjacent normal samples as normal samples for analysis.However,the adjacent non-tumor colorectal tissues of CRC patients might have some molecular characteristics of CRC.Therefore,we use the common molecular features of CRC tissue and its adjacent normal tissues to develop the signature that distinguish CRC(including CRC adjacent normal)and non-cancer(including non-CRC adjacent normal and inflammatory bowel diseases(IBD)).In addition,for the non-cancer patients with precancerous lesion(including ulcerative colitis and adenoma),whether we could assess the cancer incidence risk of this sample is a meaningful research,which might provide the potential guidance for cancer prevention.Due to the existence of batch effects,the application of a quantitative transcriptional signature usually requires standardization for samples and hardly could be applied for the samples measured by different laboratories and individualized diagnoses,which is not suitable for common clinical scenarios.Compared with the quantitative transcriptional signatures,our recently studies have demonstrated that the qualitative transcriptional signatures,namely the relative expression orderings(REOs)of gene pairs within individual sample,are robust against experimental batch effects,partial RNA degradation,amplification bias for minimum specimens and the specimens with different proportions of the tumor epithelial cell.Based on qualitative transcriptional features,we have developed some robust diagnostic or prognostic signatures for non-small cell lung cancer,liver cancer and other types of cancer.Considering the above problems and the unique merits of qualitative transcriptional features,our research will focus on the following three aspects: 1 The uncertainty of quantitative transcriptional featureFirstly,using the data of technical replicate samples(sample A and sample B)measured by two low-throughput PCR-based technologies,Sta RT-PCR? Assays and Taq Man? Assays,from the Micro Array Quality Control(MAQC)project,we demonstrated that the quantitative measurements based on low-throughput PCR-based technologies also exist large variations.Then,based on the high-throughput microarray and RNA_seq data from public databases,we systematically demonstrated that there exists greater uncertainty for individualized application of the signature based on quantitative transcriptional feature through a case study using the support vector machine(SVM)and Na?ve Bayes classifier to construct the diagnostic signature of CRC.2 The early diagnostic signature of CRC based on qualitative transcriptional featureBecause sampling inaccurately is one of the common causes of misdiagnosis of CRC,the identified CRC signature should be able to identify those samples as cancer.Thus,we use the common molecular features of CRC tissue and its adjacent normal tissues to develop the early diagnostic signature of CRC,which is suitable for the inaccurately sampled specimens and should reduce the false negatives of clinical pathological diagnosis.Based on the qualitative transcriptional feature of a total of 692 CRC samples(including CRC adjacent normal)and 168 non-cancer samples(non-CRC adjacent normal and IBD)from training set,a signature consisting of seven gene pairs was identified.It was well validated in both biopsy and surgical resection specimens from multiple datasets measured by different platforms.For a total of 977 CRC and 163 non-cancer validation samples from public databases,the sensitivity,specificity and geometric mean of the signature was 99.7%,94.5% and 97.1%,respectively,and the AUC value was 0.9589(95% CI = 0.9521-0.9657).Moreover,using Affymetrix and RNA_seq platform,we additionally measured the gene expression profiles of 33 CRC biopsy specimens and 13 CRC surgical resection specimens with different proportion of the tumor epithelial cell(ranging from 40% to 100%).Based on our early diagnostic signature of CRC,all the CRC samples measured by our laboratory were correctly identified as cancer,which further demonstrate the robustness of our signature.3 The early-warning signature of CRC based on qualitative transcriptional featureConsidering the carcinogenesis process of CRC,whose carcinogenesis is a continuous,multi-step process from normal colorectal tissue,and the unique merits of the qualitative transcriptional signature.Based on normal colorectal and CRC tissue samples,the reversal gene pairs between them were identified as CRC early-warning signature,predicting the cancer incidence risk of non-cancer patient with precancerous lesion.Then,the performance of this signature was evaluated in samples from patients with precancerous lesion with different stages from multiple datasets.For the dataset GSE13367 with 16 active Ulcerative colitis(UC)and 18 inactive UC samples,the CRC early-warning score(median = 0.5614)in the active UC samples was significantly higher than that(median= 0.3114)in the inactive UC samples(Wilcoxon rank sum test,p = 5.9687e-05).For the active UC samples and the inactive UC samples from dataset GSE53306 and adenoma samples with different stages from dataset GSE37364,we obtained the similar results.The above results preliminarily proved that our early-warning signature could well predict the cancer incidence risk of non-cancer patient with precancerous lesion.