Liquid biomarkers associate with TGF-beta Type I receptor and hypoxia in kidney cancer

<p dir="ltr">Clear cell renal cell carcinoma (ccRCC) is an aggressive kidney cancer subtype frequently associated with poor prognosis. Most ccRCC cases are asymptomatic in early stages and symptomatic mostly in advanced stages. Furthermore, the heterogeneity of ccRCC presents a challenge to design new treatments. In this study, using proximity extension assay (PEA), we analyzed blood samples from 134 patients with ccRCC and from 111 age- and gender-matched healthy donors. We identified a panel of seven proteins (ANXA1, ESM1, FGFBP1, MDK, METAP2, SDC1, and TFPI2) that are associated with clinicopathological parameters and patient survival. These biomarkers can differentiate patients with ccRCC from the control individuals with high diagnostic sensitivity and specificity. Moreover, by studying protein expression in solid tumors from the same ccRCC patients, we revealed associations between the panel biomarkers and proteins in the TGF-β and VHL-HIF signaling pathways. We found that most tumor promoting biomarkers were positively associated with TGF-β signaling and HIF-2α, and negatively associated with pVHL and HIF-1α. We also found that most tumor suppressing biomarkers were positively associated with pVHL and HIF-1α and negatively associated with TGF-β signaling and HIF-2α. For ccRCC patients, the blood protein biomarkers that were connected to poor prognosis and TGF-β/HIF-2α signaling, as identified in this study, are potentially important assets in personalized medicine.</p><p dir="ltr">We used an Olink panel to measure protein levels in clear cell renal cell carcinoma (ccRCC) patients (N=134) and healthy controls (N=111). 92 oncology-related protein levels are measured across all samples (Supplementary Data 1), and the dataset is corrected for patient age (Supplementary Data 2). 80 proteins are significantly altered in ccRCC patients compared to controls (Supplementary Data 3). Using the top 50 most significantly altered proteins, we trained a random forest (RF) model, with cross-validation (Supplementary Data 4 and 5). The top seven significantly altered proteins are sufficient to perfectly (AUC=1) classify patients and healthy controls (Supplementary Data 6). We further trained an elastic-net penalized logistic regression (ENLR) model using the top seven proteins, which also resulted in a perfect classifier. Use of random sets of seven proteins and their combinations are not as significant (Supplementary Data 7-10). We explored the correlations between transforming growth factor-β (TGF-β), VHL, and hypoxia signaling pathway protein expressions (TGFBR1-Full length receptor (FL), TGFBR1-intracellular domain (ICD), HIF-1A, HIF-2A, pVHL, pSMAD2/3) in solid tumors and the plasma protein levels from the same cohort (Supplementary Data 11-12). The names and accession numbers (UniProt) for the Olink proteins are listed in Supplementary Data 13. Levels of the TGFB and VHL pathway proteins in solid tumors are given in Supplementary Data 14 and the antibodies used in immunoblotting (IB) are listed in Supplementary Data 15. Lists of protein names measured in solid tumor samples vs protein names measured in plasma are in Supplementary Data 16.</p>