Revisiting the transcriptional analysis of primary tumours and associated nodal metastases with enhanced biological and statistical controls: application to thyroid cancer

Abstract

Background:Transcriptome profiling has helped characterise nodal spread. The interpretation of these data, however, is not without ambiguities.Methods:We profiled the transcriptomes of papillary thyroid cancer nodal metastases, associated primary tumours and primary tumours from N0 patients. We also included patient-matched non-cancerous thyroid and lymph node samples as controls to address some limits of previous studies.Results:The transcriptomes of patient-matched primary tumours and metastases were more similar than those of unrelated metastases/primary pairs, as previously reported in other organ systems. This similarity partly reflected patient background. Lymphoid tissues in the metastases confounded the comparison of patient-matched primary tumours and metastases. We circumvented this with an original data adjustment, revealing a differential expression of stroma-related gene signatures also regulated in other organs. The comparison of N0 vs N+ primary tumours uncovered a signal irreproducible across independent data sets. This signal was also detectable when comparing the non-cancerous thyroid tissues adjacent to N0 and N+ tumours, suggesting a cohort-specific bias also likely present in previous similarly sized studies. Classification of N0 vs N+ yielded an accuracy of 63%, but additional statistical controls absent in previous studies revealed that this is explainable by chance alone. We used large data sets from The Cancer Genome Atlas: N0 vs N+ classification was not better than random for most cancers. Yet, it was significant, but of limited accuracy (<70%) for thyroid, breast and head and neck cancers.Conclusions:The clinical potential of gene expression to predict nodal metastases seems limited for most cancers.

Prediction of Lymph Node Metastasis in Breast Cancer by Gene Expression and Clinicopathological Models: Development and Validation within a Population-Based Cohort

“…Previous studies investigating prediction of nodal metastasis in breast cancer have reported diverse performances of GEX-based predictors, with AUCs between chance level to near perfect separation (23)(24)(25)(26)(27)(28), likely to be due to differences in patient characteristics, cohort sizes, definition of nodal disease, gene expression analysis platforms, and feature-selection strategies. In other malignancies, nodal prediction accuracies between 50% to 60% have been reported based on TCGA data (28). Our predictors containing gene expression data (GEX and MIXED models) revealed AUCs of 0.58 to 0.72 during validation and were consistent with a midrange performance as defined in previous publications (23)(24)(25)(26)(27)(28).…”

“…However, risk assessment of metastatic axillary spread based on these clinicopathological variables is considered imperfect and consequently SLNB remains the standard axillary staging procedure (18,21,22). Previous studies of gene expression patterns from the primary breast tumor for prediction of nodal metastasis have shown inconsistent results, ranging from being nearlyperfect to almost inadequate in small, selected cohorts (23)(24)(25)(26)(27)(28). The classifiers developed to date are not sufficient to support relevant clinical conclusions (29).…”

Identification of hub genes in papillary thyroid carcinoma: robust rank aggregation and weighted gene co-expression network analysis

“…The 6 unprocessed gene expression profiles that met the inclusion criteria of this study were downloaded from the Gene Expression Omnibus (GEO, http://www.ncbi. nlm.nih.gov/geo/) database: GSE6004 [22], GSE58545 [23], GSE27155 [24,25], GSE53157 [26], GSE60542 [27] and GSE33630 [28,29]. The relevant data are shown in Table 1.…”

BRAFwild papillary thyroid carcinoma has two distinct mRNA expression patterns with different clinical behaviors

“…Selected gene signature was used to stratify patients in a validation cohort of GSE60542. 22 Briefly, expression data for 10 cluster-specific genes (RAS-signature) in the training set were combined to form a classifier according to the Bayesian Compound Covariate Predictor (BCCP). 23 The BCCP classifier estimated the likelihood that an individual patient was included in 1 of 2 clusters according to a BCCP P value of .5.…”

“…These 10 gene signatures (see Figure 7) were used to stratify patients from a validation cohort of GSE60542. 22 Results of the prediction for 27 nonmutated, RAS-mutated, and RETmutated PTCs are shown in Table 4. In Table 4, the probability in the right column means the BCCP score.…”