Endometrial cancer (EC) is the most frequent gynaecological cancer in developed countries, with its rates increasing. Its fast growth and spread are enabled by the angiogenic switch in the early stages of cancerogenesis through the release of pro-angiogenic and suppression of anti-angiogenic factors (AFs). Therefore, screening patients’ plasma biomarkers might enable a more precise diagnosis of EC and a tailored treatment approach. Furthermore, confirming AF gene expression directly in the tumour tissue can generate new knowledge on cancerogenesis and help identify biomarkers for the diagnosis and prognosis of EC.
In our study, the investigation of angiogenesis in EC was divided into three sequentially related phases. Our first prospective case-control monocentric pilot study included 76 postmenopausal women (38 endometrioid EC patients and 38 control patients with benign gynaecological conditions), and 37 angiogenic factors (AFs) were investigated as potential biomarkers for EC. AF concentrations in pre-operative plasma samples were measured using Luminex xMAP〢 multiplexing technology. The plasma levels of sTie-2 and G-CSF were significantly lower in EC compared to control patients, whereas the plasma levels of leptin were significantly higher in EC patients. Neuropilin-1 plasma levels were significantly higher in patients with type 2 EC (grade 3) than those with lower-grade cancer or controls. Follistatin levels were significantly higher in patients with lymphovascular invasion, and IL-8 plasma levels were significantly higher in patients with metastases.
In the validation study, we analysed 202 patients, of whom 91 were diagnosed with EC, and 111 were control patients with benign gynaecological disease. We used Luminex xMAP〢 multiplexing technology to measure the pre-operative plasma concentrations of six previously selected angiogenic factors – leptin, IL-8, sTie-2, follistatin, neuropilin-1, and G-CSF. The plasma levels of leptin were significantly higher in EC patients than in control patients. Leptin was higher in type 1 EC patients, and IL-8 was higher in type 2 EC, particularly in poorly differentiated endometrioid EC grade 3. In addition, IL-8 plasma levels were significantly higher in EC patients with lymphovascular or myometrial invasion. Besides basic statistical methods, we used a machine-learning algorithm to create a robust diagnostic model based on the plasma concentration of tested angiogenic factors. Among univariate models, the model based on leptin reached the best results on both training and test datasets. A combination of age, IL-8, leptin and G-CSF was determined as the essential feature for the multivariate model, with ROC AUC 0.94 on training and 0.81 on the test dataset. The model utilizing a combination of all six AFs, BMI and age reached a ROC AUC of 0.89 on both the training and test dataset, strongly indicating the capability for predicting the risk of EC even on unseen data.
Additionally, we evaluated publicly available datasets for the expression of angiogenesis-associated genes and proteins in EC tissues (T) compared to tumour-adjacent control tissue (TA). Nine genes with more than a 3-fold significant difference in gene expression in concert with more than a 2-fold significant difference in protein levels between T and TA tissue, together with six AF genes preselected in our previous plasma-based research (CSF3, IL8, LEP, NRP1, TEK, FST), were selected for validation on EC tissue, using the qPCR method on a cohort of 36 EC patients. By combining TCGA data and data from our study, we applied machine learning modelling to create the EC grade prediction model.
In our clinical cohort, IL8 and LEP were significantly upregulated, and CXCL12, ENPP2, FBLN5, FGF2, LYVE1, PDGFRB, SERPINF1, TIMP2, TIMP3, NRP1 and TEK were significantly downregulated in T vs TA tissue. In early stages and lower grades of EC, but not in more advanced or aggressive forms of EC, genes for AFs were differentially expressed between T and TA tissue. Genes were differentially expressed only in endometrial tissue from patients without deep myometrial (DMI) or lymphovascular invasion (LVI). We identified stronger gene co-expressions within T than TA tissue; correlations were particularly strong when LVI was present. In addition, we detected broader angiogenesis-related gene involvement in postmenopausal women with EC than in women of reproductive age. Finally, machine learning modelling created a relatively robust EC model based on the T gene expressions, differentiating between low and high-grade EC. According to our results, measuring plasma concentrations of AFs could represent an important supplementary diagnostic tool for early detection and prognostic characterization of EC, which could guide the decision-making regarding the extent of treatment. Our data suggest that angiogenesis in EC is promoted mainly by decreased gene expression of anti-angiogenic factors. In EC with prognostically less favourable characteristics, the regulation of AF genes is altered in T tissue as well as in morphologically normal TA tissue.
The findings from our studies suggest that the plasma levels of 6 AFs could serve as promising biomarkers, offering a valuable diagnostic and prognostic tool for early detection and characterization of EC. Particularly when incorporated into a machine learning statistical algorithm model, these AFs show great potential. Additionally, our data highlights that the regulation of angiogenesis-related genes in EC, specifically in tissues surrounding the endometrium, also plays a role in the prognosis of the disease. To establish a more precise diagnostic and prognostic machine learning model based on AF gene expression in EC tissue, further research collaboration among multiple centers is necessary.
|
