LNM had an incidence of 18.1% in EC, which was quite difficult to predict in clinical practice. (3) In our institution, patients with early stage EC with MI ≥ 1/2, G3 or non-endometrioid histology, or with advanced stage EC, used to undergo a staging surgery including pelvic and para-aortic lymphadenectomy, with sentinel lymphadenectomy (SLND) optional for patients with relatively low risk. However, non-therapeutic systemic lymphadenectomy still existed in most cases (71.4%, unpublished data), resulting in an increase in the complication morbidity, e.g. blood loss, lymphatic cyst and lower limb edema (29). Meanwhile, variables used to determine lymphadenectomy were mostly those significantly related to LNM. The molecular subtype gained wide attention due to its stronger correlation with LNM (3), but was not widely adopted for lymphadenectomy yet. LNM is a complex biological process where EMT plays a central role in the invasion of lymph vessel and survival in LNs (30, 31), literatures focusing on the role of EMT in LNM in EC remains few (18). Our team is committed to explain and predict LNM at the mechanism level. Therefore, we designed this first TCGA-based study combining transcriptomic data with conventional parameters to explore EMT-related molecular characteristics of LNM in EC.
We applied a well-validated EMT quantification method to describe a tumor sample on the EMT continuum. (23) Though unable to discriminate hybrid E/M samples from mixture of E&M ones, this model did strongly suggested samples from LNM cases tended to have a more mesenchymal ones, consistent with the association between EMT and LNM in literatures, which exhibited the feasibility of further exploration of this process in EC. Intriguingly, this difference in EMT score were more significant in G3 and MI ≥ 1/2 cases, suggesting an association between invasive phenotypes and the role of EMT in LNM. In all the molecular subtypes, LNM cases tended to have higher EMT scores, with POLE showing a statistical significance. This indicated a possible variation between subtypes regarding the extent of dependence on EMT during LNM. There is a blank of study on EMT in different molecular subtypes, which seems promising considering the abundant pathways and molecules in the EMT process. (32)
Among the 1027 ERGs included in this study, over 1/4 showed a differential expression between LNM and non-LNM cases, with multiple pathways being enriched, where TGF-β signalling pathway has been proven a potent inducer of EMT in EC (33). However, no published literature discussed the role of TGF-β pathway in LNM in EC. In the GO analysis based on DEERGs, various pathways related to TGF-β were enriched in the BP group. A similar result was reached by KEGG analysis, with other EMT-related pathways significantly enriched as well, members of which were mostly upregulated (Fig. 3). Hub genes suggested by PPI analysis, e.g. Myc, ESR1 and KRAS, were reported to participate in EMT pathways, i.e. Wnt, TGF-β, MAPK signalling pathways, etc. (34–39), and they also had most interactions with other DEERGs. These hub genes might be associated with the core mechanism and containment strategy of LNM in EC in the future.
Data from this study showed different LNM rate between molecular subtypes, especially for a significantly higher rate in CNH cases, consistent with previous reports (3). DEERGs shared between subtypes were relatively few, suggesting possibly different landscapes of LNM between molecular subtypes in EC, as stated above. Detailed pathway analyses were omitted due to a small sample size in subgroup analysis, and a prospective study on EMT in LNM and its relationship with molecular subtype in EC is being carried out in our institution (NSFC81972426).
As mentioned earlier, an accurate method calculating the LNM probability will contribute to a precise strategy for lymphadenectomy and adjuvant therapy. Several predictive models with good performances have been created, mostly based on clinicopathological parameters. (40, 41) But before the definitive surgery, molecular features of EC can be acquired through hysteroscopy sample as well (42), providing a chance to calculate the risk of LNM based on both clinical and molecular features. With logistic regression model, we finally reached a gene signature composed of 7 DEERGs, the new model based on which exhibited a good discrimination and calibration. It is noteworthy that according to the Nomogram, the ERS accounted for the majority of LNM risk, while traditional clinicopathological parameters used to guide lymphadenectomy in our institution accounted for only a small proportion of the risk. This partly explained the difficulty to predict LNM in clinical practice, i.e. a prediction model based on pathogenesis would be more precise. The value of the gene signature was further validated in a new EC database from CPTAC, suggesting that to determine this gene signature by tissue microarray (TMA) or genechip in hysteroscopy biopsies might contribute to the development of a most potent clinical prediction model for LNM in EC.
There are several limitations in our research. One is that due to the heterogeneity of transcriptome data in different databases, our study did not develop a clinical prediction model able to be adopted universally. But the 7-ERGs based gene signature did exhibited a diagnostic value independent of clinicopathological parameters across databases, supporting future applications. The other is that this study is limited to bioinformatics analysis and not verified by specimens from our institution. The above limitations can be remedied through developing a TMA or genechip with a unified standard. Meanwhile, there is a huge gap for laboratory research to fill on mechanisms of EMT to induce lymphatic spread in EC.