A Predictive Signature For Pathological Response States Of Patients With Locally Advanced Rectal Cancer To Neoadjuvant Chemoradiation

Currently,we are still unable to identify patients who will be susceptible to pCRT and benefit from this therapy.Thus,there is a need to develop a predictive signature based on pretreatment data.With the wide application of high throughput gene expression profiling techniques,many classifiers based on quantitative transcriptional signatures for predicting the pathological response to nCRT in LARCs have been developed.However,clinical applications of these transcriptional signatures are scarce due to technological,mathematical and translational barriers.Besides factors such like tissue sampling and sample preparation quality,a well-known factor is that gene expression data are often “noisy” and subject to lab and batch effects introduced by the differences in laboratory conditions and personnel.In contrast,it has been found that the within-sample relative expression orderings(REOs)of gene pairs,which is also called Relative Expression Analysis(RXA),are robust against experimental batch effects and invariant to monotone data transformation.Besides,the within-sample REOs of gene pairs are robust against variations of the tumor epithelial cell proportions in tissues sampled from different sites of a tumor,partial RNA degradation in the sample preparation process and during the storage stage and amplification bias for minimum specimens even with about 15–25 cancer cells,which are also important factors leading to the failure of validation and clinical application of the quantitative transcriptional signatures.Using 42 samples as the training cohort,we searched a signature by screening the gene pairs whose within-sample relative expression orderings are significantly correlated with the pathological response.For the training dataset,17 of the 19 responders were correctly classified(sensitivity = 89.47%),and 22 of the 23 nonresponders were correctly classified(specificity = 95.65%)using the 27-GPS signature.The accuracy was 92.86%,and AUC was 0.95(95% CI =0.83-0.99).In the first validation cohort(GSE35452),the signature resulted in a prediction of nCRT responders with an 87.50%(21/24)sensitivity,a 90.90%(20/22)specificity,and an 89.13% accuracy(95% CI,76.43-96.38%).In the second cohort measured at our laboratory,a 95.24%(20/21)sensitivity,an 83.33%(10/12)specificity and a 90.91% accuracy(95% CI,75.67-98.08%)were observed.And,the AUC reached 0.95(95% CI,0.84-0.99)and 0.91(95% CI,0.76-0.98)for the two validation cohorts,respectively.These findings indicated that 27-GPS can robustly predict the response to nCRT in LARCs.Because the survival data for the cohorts of LARCs were unavailable,we turned to evaluate whether the signature could predict the survival benefit of colorectal cancers from 5Fu-based chemotherapy,given that rectal and colon cancers are genetically almost indistinguishable and the responders of LARCs will achieve greater benefits from postoperation chemotherapy than the nonresponders.Furthermore,among 285 samples,138 were predicted as responders and 147 as non-responders.The responders have an average disease-free survival of 56.4 months which is significantly longer than that of non-responders(45.5 months),after stage,age and gender were adjusted(multivariate Cox regression,p = 0.01,hazard ratio = 0.60).These findings show that 27-GPS could robustly predict the survival benefits of LARCs from nCRT before the treatment.The investigation of nCRT-response mechanism is conductive to elevating responsivity of nCRT in LARCs.And,a common approach for investigating mechanisms of nCRT response is firstly identifying differentially expressed genes between the nonresponders and responders.However,such DE genes may have various origins.Of them,some might be irrelevant to rectal oncology and whose perturbation would also involve the normal physiological function and induce adverse reaction.In contrast,some of DE genes relevant to rectal oncology will not.The within-sample relative expression ordering of genes is robust against batch effects and invariant to monotonic data transformation.Thus,the available data of normal rectal samples generated by different laboratories can work as the non-parallel control group.And on this basis,we identified tumor-related genes differentially expressed between rectal cancers and normal.Further,the response-related and tumor-related gene was identified as the tumor response gene.Then,we conducted analyses of protein-protein interaction between the tumor response gene and 113 genes known to be involved in response to nCRT.The resultant findings revealed that TFDP1 directly connected with MYC,PCNA and BIRC5 which play important roles in the chemoradiation response.Indeed,we found higher expression of TFDP1,MYC,PCNA and BIRC5 in the responders compared with the non-responders.These finding suggest that the upregulated TFDP1 boosts the sensitivity of the rectal cancer cell to cytotoxic agents via upregulating MYC,PCNA,BIRC5.In the present study,we established a signature consisting of 27 gene pairs,which can robustly predict the nCRT response for LARCs and survival benefits as well.And,the upregulated TFDP1 in responders raised the expression level of MYC,PCNA,BIRC5,and boosted DNA replication and cell cycle,and the sensitivity of nCRT in LARCs as well.