Devens, Mary E.

Abstract Purpose: Methylated DNA markers can discriminate upper from lower GI neoplasms (Gastroenterology 2013;144(5):S84). It is unknown if organ-specific molecular prediction of GI tumor site can be achieved. Experimental design: An unbiased search for neoplasm markers at each major GI organ site was conducted by reduced representation bisulfite sequencing. Top marker candidates were then blindly validated by methylation specific PCR on independent tissue DNA samples from normal mucosa [17 esophagus (E), 13 stomach (S), 33 pancreas (P), 35 bile duct/liver (BD), 21 colorectal (CR)] and neoplasia [42 E, 43 S, 36 P, 48 BD, 97 CR]. Recursive partitioning (rPart) trees modeled all stepwise combinations of single markers to classify controls vs neoplasms and then neoplasms by site (CR, gastroesophageal or pancreaticobiliary). A 2nd model was designed from tissue data for blinded clinical piloting on 42 independent plasma DNA samples (14 normal and 28 cancers (14 P, 14 CR). Results: From the top 100 markers, 95 were validated and used for rPart modeling. A 3- marker panel (chr12.133484978-5047, BMP3, chr11.123301058-255) was selected for the universal detection of GI neoplasms, as it classified neoplasms and control tissues with 95% accuracy. Two markers (chr7.25896389-501, QKI) then assigned lower vs upper GI neoplasms with 94% accuracy. Finally, 3 markers (PDGFD, ELOVL2, PCBP3) called pancreaticobiliary vs gastroesophageal neoplasms with 94% accuracy. All 8 markers applied in a single model (Table) accurately predicted control vs tumor by site [88% (p <0.0001)]. In plasma, a 2-marker panel (BMP3, QKI) was 78% accurate in assigning patients to cancer or control groups (p<0.0009) and 83% accurate in assigning cancer origin to CR or P sites (p<0.0001). Predicted by ModelTissue SourceControlsColorectal neoplasiaPancreatico-biliary cancerGastro-esophageal neoplasiaControls1169102Colorectal neoplasia18617Pancreatico-biliary cancer10654Gastro-esophageal neoplasia12872 Conclusion: Panels of methylated DNA markers were identified for both universal detection of GI neoplasia and accurate prediction of tumor site. Pilot testing of these markers in plasma supports potential clinical feasibility. Citation Format: John B. Kisiel, William R. Taylor, Tracy C. Yab, Hassan M. Ghoz, Patrick H. Foote, Mary E. Devens, Douglas W. Mahoney, Thomas C. Smyrk, Lisa A. Boardman, Gloria M. Petersen, Navtej S. Buttar, Lewis R. Roberts, Graham P. Lidgard, David A. Ahlquist. Accurate site prediction of gastrointestinal cancer by novel methylated DNA markers: Discovery & validation. [abstract]. In: Proceedings of the 106th Annual Meeting of the American Association for Cancer Research; 2015 Apr 18-22; Philadelphia, PA. Philadelphia (PA): AACR; Cancer Res 2015;75(15 Suppl):Abstract nr 4252. doi:10.1158/1538-7445.AM2015-4252

Abstract We have demonstrated that colorectal cancer (CRC) and advanced pre-cancers can be detected non-invasively by a manual multi-target stool DNA-based test (sDNA-MT) comprising exfoliated DNA markers (methylated BMP3 and NDRG4, mutant KRAS (7 mutations, codons 12, 13), plus β-actin) and fecal hemoglobin (Hb) (Lidgard, Gastroenterology 2012;142(5);S-770). We now report the clinical performance of this sDNA-MT test using an optimized automated analytic platform and logistic algorithm. This platform could facilitate the routine performance of sDNA-MT for CRC screening by molecular diagnostics capable clinical laboratories. Method: Stool samples were collected from 1003 subjects at 36 study sites after informed consent and prior to colonoscopy bowel preparation from those presenting for average risk CRC screening (283) or surveillance (176) at 2 sites. From referred subjects with CRC, Advanced Adenoma (AA) or Sessile Serrate Adenoma ≥ 1 cm, (SSA) stool was collected at least 7 days post-colonoscopy and prior to surgery or chemo-radiation (135; 21 sites) and similarly for subjects with no neoplastic findings on colonoscopy (Neg) (409; 13 sites). The study population included: cases (207), 58% male, median age 65 yrs. (38-87), CRC (93), AA (84), SSA ≥ 1 cm (30) and controls (796), 42% male, median age 65 yrs. (50-84), Neg (641) and non-advanced adenomas (NA) (155). Stool sample collection and DNA isolation were previously described. Automated methylation, mutation and actin assays were performed with a Hamilton STARlet fluid handler (Hamilton Robotics, Reno NV), and QuARTS (Quantitative Allele-specific Real-time Target and Signal amplification) run on an ABI 7500 FastDx real time thermal cycler (Applied Biosystems, Foster City, CA). Fecal Hb (ng/ml buffer) analysis was performed by automated sandwich ELISA. A “Positive” or “Negative” result was determined with an algorithm that included the methylation and mutation results and a logistic regression score, which combines DNA marker results with Hb and actin results. Algorithm results exceeding a threshold were called “Positive”. The algorithm provided good discriminative ability, stability, sensitivity and specificity. Robustness was tested with computer simulations and statistical techniques (leave-one-out and 10-fold cross validation). Results: At a 90% nominal specificity, sDNA-MT sensitivity was 98% for CRC (91/93) [Stage: I, 95% (20/21), II, 100% (23/23), III 96% (26/27), IV 100% (7/7) and I-III combined 97% (69/71)], 57% (65/114) for precursors ≥1 cm (AA, SSA), and 86% (12/14) for precursors with high grade dysplasia. CRC patients were typically referred to colonoscopy for symptoms and test sensitivity may be elevated relative to that seen with screening. Conclusion: With this study using a novel automated sDNA-MT analytic platform with logistic algorithm, we corroborate our earlier findings using a manual process and demonstrate a platform that allows testing to be performed routinely by molecular diagnostic capable laboratories. The high sensitivity of sDNA-MT for CRC across all stages and for advanced precursors with high-grade dysplasia could lead to improved non-invasive CRC screening performance with wide accessibility to patients. A large multi-site pivotal CRC screening study (DeeP-C study clinicaltrials.gov, NCT01397747) to support such use is underway. Citation Format: Graham P. Lidgard, Michael J. Domanico, Janelle J. Bruinsma, James Light, Zubin D. Gagrat, Rebecca L. Oldham-Haltom, Keith D. Fourrier, Hatim Allawi, Tracy C. Yab, Julie A. Simonson, Mary Devens, Russell I. Heigh, David A. Ahlquist, Barry M. Berger. An optimized molecular stool test for colorectal cancer screening: Evaluation of an automated analytic platform and logistic algorithm. [abstract]. In: Proceedings of the Eleventh Annual AACR International Conference on Frontiers in Cancer Prevention Research; 2012 Oct 16-19; Anaheim, CA. Philadelphia (PA): AACR; Cancer Prev Res 2012;5(11 Suppl):Abstract nr B12.

Abstract Aberrantly methylated genes represent attractive and broadly informative markers for colorectal cancer (CRC). As demographics, numerous ingestants, environmental exposures, and obesity variably influence both gene methylation rates and CRC incidence, it is important to understand the effects of such covariates on the clinical performance of CRC screening tests that incorporate assay of methylated gene markers. Based on a recent multicenter study (Ahlquist et al. Gastroenterology 2012 (in press)), stool assay of carefully selected methylated gene marker candidates (NDRG4, BMP3, vimentin, TFPI2), alone or in combination, yield high detection rates of both CRC and large adenomas. Aim: To assess the impact of demographic, exposure, body mass, and other patient variables on stool levels of the above 4 methylated gene markers. Methods: We studied buffered stools collected within 3 years of a normal colonoscopy from 500 patients undergoing average-risk screening or polyp surveillance (median age 64 (range 44-85); 53% women). Upon receipt, stools were promptly homogenized, aliquoted, and frozen at –80°C. On supernatants from thawed/spun aliquots, target genes were purified by hybrid capture, bisulfite treated, and assayed using the analytically-sensitive QuARTS method (quantitative allele-specific real-time target and signal amplification), as described (above citation). The reference gene β-actin was assayed along with methylation of NDRG4, BMP3, vimentin, TFPI2. Log-converted data were normalized to allow comparison of effects between markers, and effect size was expressed as % change relative to standard deviation for each marker (standardized relative change (SRβ)). Results: The only patient characteristic that significantly influenced all methylated marker levels in stool was age (p<0.0001 for each marker); SRΔ was greatest with TFPI2 at +91.3 (p<0.001 vs other markers), least with BMP3 at +29.7 (p<0.001 vs others), and intermediate with vimentin at +46.0 and NDRG4 at +45.1. In contrast to methylation markers, levels of β-actin did not change across age. The other demographic variables of sex, race, and geographic residence had no effect on methylation markers. Marker levels were also not affected by exposure variables (smoking, alcohol consumption, or analgesic use), by family history of CRC or polyps, or by personal history of polyps. Finally, neither overweight status (n=176) nor obesity (n=148) had an effect on marker levels. Conclusions: Age significantly affects stool levels of tumor-associated methylated gene markers in colonoscopy-normal patients, and the degree of this age effect differs across markers. In contrast, other key patient variables had no effect on stool marker levels. Comment: Optimal use of these methylation markers for stool screening of CRC may require age-adjustment of normal cutoff levels to maximize test specificity. Citation Format: {Authors}. {Abstract title} [abstract]. In: Proceedings of the 103rd Annual Meeting of the American Association for Cancer Research; 2012 Mar 31-Apr 4; Chicago, IL. Philadelphia (PA): AACR; Cancer Res 2012;72(8 Suppl):Abstract nr 3572. doi:1538-7445.AM2012-3572