A multiplex platform for the identification of ovarian cancer biomarkers

This is a relatively strong paper for Olink in the Oncology field, with positive statements from the authors about our technology, good correlations with standard assays and Olink data that improves the predictive score for the existing clinical assay. Despite the high incidence and mortality of ovarian cancer, there are no protein biomarker assays with sufficient sensitivity and specificity for use as diagnostic screening tools in the general population. There are clinical assays for HE4 and CA125 approved by the FDA, but only for monitoring the recurrence of ovarian cancer. In this study, the Onc I, v2 panel was used to look for potentially useful signatures in a patient cohort of women that included controls (n=21), benign disease (n=18), and both early (n=21) and late (n=21) stage ovarian cancer. (The controls were actually patients with benign non-gynecological health conditions, so “healthy” from the ovarian cancer perspective!).

Using PCA of all 92 proteins, the patients could be seen to segregate clearly into healthy/benign and early/late cancer groups. When the PCA was applied specifically to the markers that are currently approved for the clinic, HE4 & CA125, this segregation was very apparent for the individual proteins. When the same analysis was applied to CA125 levels measured with the FDA-approved ELISA and using the defined cut-off level to distinguish them, the PCA plots for the ELISA and Olink measurements were extremely similar. They went on to make a direct comparison of ELISA/Olink measurements for CA125 and found a very good correlation (0.89). Linear regression was performed to identify proteins that were differentially expressed between the different patient groups , and significant trends (p < 0.001) across the 4 groups from controls to late stage cancer were observed for 38 of the 92 proteins, many of which were novel as candidate serum biomarkers for ovarian cancer. The 12 most significant proteins were CA125, HE4, MK, KLK6, hK11, CXCL13, FR-alpha, IL-6, TNFSF14, FADD, PRSS8, FUR) and of these, CXCL13, FADD, and TNFSF14 have never been previously reported in connection with ovarian cancer.ROC curve analyses (plotting sensitivity vs specificity) were performed for all 92 proteins to determine those that could discriminate between the sera from the different groups within the study and area under the curve (AUC) values were calculated to determine the classification accuracy for each protein. The results using the Olink-determined values for CA125 & HE4 confirmed previous data using the approved clinical assays, and many additional proteins from the Onc panel also showed significant AUC values for the different comparisons – many of them with established associations with ovarian cancer, but also including several highly novel findings. For late stage cancer vs controls, 51 proteins generated an AUC of >0.7. In the critical comparison of early stage ovarian cancer vs controls, 23 proteins had an AUC >0.7. When the top 12 proteins were analyzed in combination using machine learning techniques, the specificity of this “profile” improved the specificity of early-stage ovarian cancer detection from 93% to 95.2% compared to using CA125 alone. Although this seems like a relatively incremental increase compared to the existing test, if applied to the clinical setting, this would have great benefits in the scenario of wide-scale preventive screening. The authors acknowledge the limitations of this small study and that larger cohorts would be required to validate and develop their findings.

The authors conclude that Olink technology can both replicate the results established by conventional clinical assays and identify new candidate biomarkers for ovarian cancer. They also point out the potential of using more powerful multi-protein signatures to improve the sensitivity and specificity of clinical tests based on single biomarkers