Grade determines prognosis and treatment. Tumor grade is the most important prognostic factor after staging in endometrial cancer. Five-year distant recurrence rates are 3%, 8%, and 25% for grades 1, 2, and 3, respectively, while 5-year disease-specific survival rates are 97%, 94%, and 76%. Only grade 3 disease drives inclusion in high-risk adjuvant therapy protocols.
Interobserver variability is a critical problem. Histomorphological grading suffers from significant inter-observer disagreement, particularly for high-grade EC. Studies report that only 62.5% of high-grade EC cases achieve diagnostic consensus among three reviewers, meaning up to 37.5% of cases may be misclassified by conventional pathology alone.
miRNAs as grading biomarkers. MicroRNAs are 22-nucleotide small non-coding RNAs that regulate gene expression via RNA-induced silencing complexes, controlling proliferation, differentiation, and apoptosis. They are relatively stable molecules, are dysregulated in cancer, and can serve as objective, reproducible molecular biomarkers.
NGS advantage over microarray. Prior miRNA studies in EC predominantly used qPCR or microarray, which are limited to examining predefined miRNA sets. Next generation sequencing (NGS) provides unbiased whole-miRNome profiling, enabling discovery of novel biomarkers without a priori hypothesis.
In-house cohort. 24 EC patients (grade 1: n=10, grade 2: n=7, grade 3: n=7) were recruited from Jagiellonian University Hospital in 2016-2017. Tumor samples from surgical hysterectomy specimens underwent total RNA extraction using the mirVana kit, with RNA integrity numbers (median RIN 7.3) assessed by TapeStation.
Small RNA library construction. Libraries were prepared using New England Biolabs chemistry from 100 ng total RNA. After 12 PCR cycles and BluePippin size selection (141 bp), samples were pooled (6 per run) at 12 pM on Illumina MiSeq v2 cartridges. PhiX was added at 20% to increase library diversity.
Bioinformatic pipeline. Raw reads were trimmed with Cutadapt and aligned to miRBase 22.1 using miRDeep2. Differential expression analysis used edgeR in R with Benjamini-Hochberg FDR correction (threshold FDR below 0.05). Multidimensional scaling and heatmap visualization confirmed data quality. One outlier sample (grade 3 with undifferentiated carcinoma at stage III) was removed from analysis.
TCGA external validation. Results were validated in 407 TCGA-UCEC samples (grade 1: n=97, grade 2: n=119, grade 3: n=191) downloaded via TCGAbiolinks. The same grade grouping strategy (low: G1+G2 vs. high: G3) was applied, and overlapping differentially expressed miRNAs were identified. qPCR with SYBR Green and the SNORD48 reference gene confirmed top candidates.
Differential expression landscape. The analysis identified 50 differentially expressed miRNAs between grade 2 and grade 1, 40 between grade 3 and grade 1, and only 3 between grade 3 and grade 2. Most miRNAs were downregulated with increasing grade. The three G3-vs-G2 transcripts were miR-375, miR-214-3p, and miR-190b.
TCGA validation success. External TCGA validation yielded 60% gene overlap for the low vs. high grade comparison and 31% for the G1 vs. G2+G3 comparison - reasonable overlap given the small in-house cohort size. Both miR-375 and miR-190b were confirmed as differentially expressed in the external dataset.
qPCR confirmation. In grade 3 vs. grade 2 patients, log fold-changes by qPCR were -3.04 for miR-190b and -4.39 for miR-375 (both P below 0.003), closely matching the RNAseq values of -3.33 and -4.34, confirming technical accuracy of the discovery findings.
Random forest grade classification. Starting from 72 differentially expressed miRNAs, cross-validation estimated that 10 features minimized classification error. A random forest trained on the top 10 miRNAs (including miR-375, miR-190b, miR-652-3p, miR-421, and miR-941) achieved classification errors of 10% for grade 1, 14.3% for grade 2, and 14.3% for grade 3 - superior to the estimated 37.5% misclassification rate of conventional histopathology for grade 3.
Grade 1 cluster coherence. Unsupervised clustering identified a coherent cluster of 6 out of 10 grade 1 samples (60% homogeneity) based on differentially expressed miRNAs from the G1 vs. G2+G3 comparison. This suggests that grade 1 EC has a relatively distinct miRNA signature.
Grade 3 cluster coherence. The G1+G2 vs. G3 discrimination analysis produced a clear coherent cluster of 5 grade 3 samples out of 6 (84% homogeneity), demonstrating that high-grade EC has a more distinct and homogeneous miRNA profile than lower grades.
Grade 2 heterogeneity. Grade 2 samples failed to form a distinct cluster in either analysis, showing significant molecular heterogeneity. This finding reinforces clinical observations that grade 2 EC is a heterogeneous category that may not represent a biologically distinct entity, and supports using molecular subtyping to better resolve grade 2 prognosis.
Gene ontology of miRNA targets. GO analysis of the 72 differentially expressed miRNA target genes identified three top biological processes: positive regulation of transcription by RNA polymerase II, negative regulation of cell differentiation, and modulation by virus of host morphology or physiology - the latter being an unexpected finding suggesting possible viral mechanisms in high-grade EC pathogenesis.
miR-375 tumor suppressor role. miR-375 was first identified as a pancreatic islet-specific miRNA but is broadly downregulated across many cancer types including hepatocellular carcinoma, esophageal carcinoma, gastric cancer, head and neck cancer, melanoma, and glioma. It acts as a tumor suppressor by repressing oncogenes PDK1, JAK2, IGF1R, and AEG-1. Its reduced expression in grade 3 EC is consistent with its established tumor-suppressive role.
miR-190b dual roles. miR-190b has been described as promoting tumor progression in bladder carcinoma and colorectal cancer, but has also been identified as a novel tumor suppressor in lung cancer via in vivo miRNA knockout screening. Its consistent downregulation in grade 3 EC in both the in-house cohort and TCGA confirms its diagnostic relevance independent of tumor type-specific paradoxes.
Comparison to survival-based miRNA panels. Tang et al. previously identified a 7-miRNA panel for overall survival prediction in EC, which included both miR-190b (differentially expressed between G1+G2 and G3) and miR-let-7b (differentially expressed between G1 and G2+G3) - both independently confirmed in this in-house cohort - linking grading biomarkers to prognostic signatures.
Viral mechanism hypothesis. GO analysis unexpectedly flagged modulation by virus of host morphology or physiology as significantly enriched. Although HPV is detected in EC at 16-24%, it is not considered a causative agent. The findings suggest investigating other oncoviruses such as human mammary tumor virus (HMTV) in high-grade EC pathogenesis.
Key conclusions. NGS miRNA profiling identified miR-375 and miR-190b as the most reliable molecular markers distinguishing grade 3 from grade 1-2 EC, validated across two independent cohorts totaling 431 patients. A 10-miRNA random forest model achieved classification accuracy superior to histomorphological examination.
Grade 2 reclassification potential. The high molecular heterogeneity of grade 2 EC observed in unsupervised clustering suggests that miRNA profiling may help reclassify grade 2 cases into clinically meaningful molecular subgroups, potentially identifying which grade 2 patients need adjuvant therapy traditionally reserved for grade 3.
Future direction. The small cohort size is acknowledged as a primary limitation. Future studies should prospectively validate this 10-miRNA classifier in larger independent cohorts, integrate miRNA grading with POLE mutation and MMR status for comprehensive molecular subtyping, and explore miRNA-based liquid biopsy approaches using plasma.