Abstract 111: Accurate multi-cancer detection using methylated DNA markers and proteins in plasma

Abstract Purpose: Screening for high prevalence cancers (breast, colorectum and uterine cervix) reduces mortality; however, most cancer deaths arise in unscreened organs. An ideal solution would accurately detect multiple currently unscreened cancers in a single round of testing. We aimed to assess the feasibility of a targeted assay panel of methylated DNA markers (MDMs) and proteins for detection of highly lethal cancers in a multi-cancer case-control study. Experimental Procedures: Prospectively collected plasmas from 180 cases of 6 cancer types (stage A-D liver (n=36), and TNM stage II-IV esophageal (n=18), lung (n=36), ovarian (n=30), pancreatic (n=30), and stomach (n=30)) and 257 smoking status, age-, and gender-matched asymptomatic controls were tested in blinded fashion. To quantify 26 MDMs previously found to be common among many cancers, multiplex PCR followed by LQAS (Long probe Quantitative Amplified Signal) assay was performed on bisulfite converted DNA extracted from 3 mL of plasma. Additionally, 5 proteins were tested from paired serum aliquots and combined with MDMs for a multi-analyte analysis. Two-thirds of the cases and controls were used to develop prediction algorithms, both logistic regression and random forest models, and the remaining 1/3 were used to validate the models. Results: Using stepwise logistic regression, a combination of 3 proteins (CEA, CA-125, CA19-9) and 5 MDMs (ZNF671, GRIN2D, NDGR4, SHOX2, B3GALT6) resulted in an AUC of 0.95 and an overall sensitivity of 87% for all cancers at 95% specificity. The logistic model on the validation set of the samples resulted in an AUC of 0.96 and a sensitivity of 83% with an observed specificity of 94%. The cancer-specific sensitivities ranged from 78% for lung cancer to 90% for ovarian and pancreatic cancer (Table). Random forest and logistic analyses were in agreement. Conclusion: Targeted assay of select MDMs and proteins show high discrimination for multiple cancers in a case control study, which included early stage patients. Performance with 95% confidence intervals of multi-target assay for multi-cancer detectionSensitivitySpecificityModelOverallBy cancer typeLungEsophagealGastricPancreaticLiverOvarianLogisticTraining set87%(79-92%)79%83%90%95%88%85%95%(90-97%)Test set83%(72-91%)75%100%80%80%75%100%94%(87%-98%)Training + Test186%(80-90%)78%89%87%90%83%90%95%(91-97%)Training + Test275%(68-81%)50%89%83%76%81%80%98%(96-99%)Random forest377%(70-83%)58%78%83%77%81%87%95%(92-97%)195% specificity goal, 5 MDMs and 3 proteins.298% specificity goal, 5 MDMs and 3 proteins.3With 500-fold in silico cross-validation on all samples and all markers (26 MDMs + 5 proteins) Citation Format: Slava Katerov, Abram Vaccaro, Jacquelyn Hennek, Jennifer Carlson, William R. Taylor, Douglas Mahoney, John B. Kisiel, Hatim T. Allawi. Accurate multi-cancer detection using methylated DNA markers and proteins in plasma [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2021; 2021 Apr 10-15 and May 17-21. Philadelphia (PA): AACR; Cancer Res 2021;81(13_Suppl):Abstract nr 111.