Search

Analysis of DNA methylation is a valuable tool to understand disease progression and is increasingly being used to create diagnostic and prognostic clinical biomarkers. While conversion of cytosine to 5-methylcytosine (5mC) commonly results in transcriptional repression, further conversion to 5-hydroxymethylcytosine (5hmC) is associated with transcriptional activation. Here we perform the first study integrating whole-genome 5hmC with DNA, 5mC, and transcriptome sequencing in clinical samples of benign, localized, and advanced prostate cancer. 5hmC is shown to mark activation of cancer drivers and downstream targets. Furthermore, 5hmC sequencing revealed profoundly altered cell states throughout the disease course, characterized by increased proliferation, oncogenic signaling, dedifferentiation, and lineage plasticity to neuroendocrine and gastrointestinal lineages. Finally, 5hmC sequencing of cell-free DNA from patients with metastatic disease proved useful as a prognostic biomarker able to identify an aggressive subtype of prostate cancer using the genes TOP2A and EZH2, previously only detectable by transcriptomic analysis of solid tumor biopsies. Overall, these findings reveal that 5hmC marks epigenomic activation in prostate cancer and identify hallmarks of prostate cancer progression with potential as biomarkers of aggressive disease. Significance: In prostate cancer, 5-hydroxymethylcytosine delineates oncogene activation and stage-specific cell states and can be analyzed in liquid biopsies to detect cancer phenotypes. See related article by Wu and Attard, p. 3880

Our study employed the detection of 5-hydroxymethyl cytosine (5hmC) profiles on cell free DNA (cfDNA) from the plasma of cancer patients using a novel enrichment technology coupled with sequencing and machine learning based classification method. These classification methods were develoiped to detect the presence of disease in the plasma of cancer and control subjects. Cancer and control patient cfDNA cohorts were accrued from multiple sites consisting of 48 breast, 55 lung, 32 prostate and 53 pancreatic cancer subjects. In addition, a control cohort of 180 subjects (non-cancer) was employed to match cancer patient demographics (age, sex and smoking status) in a case-control study design.Logistic regression methods applied to each cancer case cohort individually, with a balancing non-cancer cohort, were able to classify cancer and control samples with measurably high performance. Measures of predictive performance by using 5-fold cross validation coupled with out-of-fold area under the curve (AUC) measures were established for breast, lung, pancreatic and prostate cancer to be 0.89, 0.84, 0.95 and 0.83 respectively. The genes defining each of these predictive models were enriched for pathways relevant to disease specific etiology, notably in the control of gene regulation in these same pathways. The breast cancer cohort consisted primarily of stage I and II patients, including tumors < 2 cm and these samples exhibited a high cancer probability score. This suggests that the 5hmC derived classification methodology may yield epigenomic detection of early stage disease in plasma. Same observation was made for the pancreatic dataset where >50% of cancers were stage I and II and showed the highest cancer probability score.

3044 Background: Epigenomics assays have recently become popular tools for identification of molecular biomarkers, both in tissue and in plasma. In particular 5-hydroxymethyl-cytosine (5hmC) method, has been shown to enable the epigenomic regulation of gene expression and subsequent gene activity, with different patterns, across several tumor and normal tissues types. In this study we show that 5hmC profiles enable discrete classification of tumor and normal tissue for breast, colorectal, lung ovary and pancreas. Such classification was also recapitulated in cfDNA from patient with breast, colorectal, lung, ovarian and pancreatic cancers. Methods: DNA was isolated from 176 fresh frozen tissues from breast, colorectal, lung, ovary and pancreas (44 per tumor per tissue type and up to 11 tumor tissues for each stage (I-IV)) and up to 10 normal tissues per tissue type. cfDNA was isolated from plasma from 783 non-cancer individuals and 569 cancer patients. Plasma-isolated cfDNA and tumor genomic DNA, were enriched for the 5hmC fraction using chemical labelling, sequenced, and aligned to a reference genome to construct features sets of 5hmC patterns. Results: 5hmC multinomial logistic regression analysis was employed across tumor and normal tissues and identified a set of specific and discrete tumor and normal tissue gene-based features. This indicates that we can classify samples regardless of source, with a high degree of accuracy, based on tissue of origin and also distinguish between normal and tumor status.Next, we employed a stacked ensemble machine learning algorithm combining multiple logistic regression models across diverse feature sets to the cfDNA dataset composed of 783 non cancers and 569 cancers comprising 67 breast, 118 colorectal, 210 Lung, 71 ovarian and 100 pancreatic cancers. We identified a genomic signature that enable the classification of non-cancer versus cancers with an outer fold cross validation sensitivity of 49% (CI 45%-53%) at 99% specificity. Further, individual cancer outer fold cross validation sensitivity at 99% specificity, was measured as follows: breast 30% (CI 119% -42%); colorectal 41% (CI 32%-50%); lung 49% (CI 42%-56%); ovarian 72% (CI 60-82%); pancreatic 56% (CI 46%-66%). Conclusions: This study demonstrates that 5hmC profiles can distinguish cancer and normal tissues based on their origin. Further, 5hmC changes in cfDNA enables detection of the several cancer types: breast, colorectal, lung, ovarian and pancreatic cancers. Our technology provides a non-invasive tool for cancer detection with low risk sample collection enabling improved compliance than current screening methods. Among other utilities, we believe our technology could be applied to asymptomatic high-risk individuals thus enabling enrichment for those subjects that most need a diagnostic imaging follow up.

Background: Our feasibility study employed a novel genomic detection methodology that enriches 5-hydroxymethylcytosine (5hmC) loci in cell free DNA (cfDNA) from the plasma of cancer patients using click chemistry coupled with sequencing and machine learning based classification methods. These classification methods were developed to detect the presence of disease in the plasma of cancer and control subjects. Cancer and control patient cfDNA cohorts were accrued from multiple sites consisting of 48 breast, 55 lung, 32 prostate and 2 pancreatic datasets consisting of 41 and 53 cancer subjects (Set 1 and 2). In addition, a control cohort of 260 subjects (non-cancer) was employed to match cancer patient demographics (age, sex and smoking status) in a case-control study design. Methods: Machine learning methods, applied to each cancer case cohort individually, with a balancing non-cancer cohort, were able to classify cancer and control samples. Measures of predictive performance using 5-fold cross validation coupled with out-of-fold Area Under the Receiver Operating Characteristic Curve (AUROC) measures were employed. Gene sets selected as part of biomarker discovery were further analyzed for disease relevance using pathway analysis tools (GSEA, mSigDB). Results: 260 controls and 229 cancers from four disease types (breast, lung, pancreas and prostate) were analyzed; more than 60% of cancer patients had early stage disease (I or II). Predictive performance, employing AUROC measures, was established for breast (0.89), lung (0.84), pancreas (set 1 - 0.95 and 2 - 0.93) and prostate (0.83). The genes defining each of these predictive models were enriched for pathways relevant to disease specific etiology, notably in the control of gene regulation in these same pathways. The breast cancer cohort consisted primarily of stage I and II patients including tumors < 2 cm and these samples exhibited a higher prediction probability score. The prostate cancer cohort consisted of both indolent and aggressive disease sample and prediction performance was equally high for both (AUROC for indolent vs aggressive was 0.81 and 0.77, respectively). Conclusions: These findings suggest that 5hmC changes in cfDNA enable non-invasive detection of early stage breast, pancreatic, prostate, and lung cancers. Furthermore, 5hmC profiling in cfDNA may enable the prediction of clinically relevant features such as tumor size in breast adenocarcinoma or indolent disease in prostate cancer. Finally, this study identified a suite of 5hmC biomarkers that may be further validated in larger, and more diverse, patient cohorts. Citation Format: Anna Bergamaschi, Jeremy Ku, Yuhong Ning, Francois Collin, Chris Ellison, Tierney Phillips, Erin McCarthy, Wendy Wang, Michael Antoine, David Haan, Aaron Scott, Paul Lloyd, Gulfem Guler, Alan Ashworth, Stephen Quake, Samuel Levy. Epigenomic detection of multiple cancers in plasma derived cell free DNA [abstract]. In: Proceedings of the Annual Meeting of the American Association for Cancer Research 2020; 2020 Apr 27-28 and Jun 22-24. Philadelphia (PA): AACR; Cancer Res 2020;80(16 Suppl):Abstract nr 783.

e21505 Background: Liquid biopsies are gaining prominence for not only cancer diagnosis but also patient monitoring. Mutational signals derived from cell-free DNA (cfDNA) show promise to assess response to cancer treatment, including immunotherapy. However, reliance of these methods on mutational data from tissue biopsies limit their applicability when a tumor biopsy is unavailable, or when mutational landscape of tumor changes under the selective pressures of cancer drug treatment. Epigenomic approaches have the potential to address these shortcomings. Methods: Blood draws were obtained from a cohort of non-small cell lung cancer (NSCLC) patients (n = 19) who went on to anti-PD1 treatment prior to therapy start and while on therapy. cfDNA was isolated from plasma and was subsequently processed to generate 5hmC genome-wide profiles. Results: We analyzed cfDNA from NSCLC patients undergoing anti-PD1 therapy to investigate whether immunotherapy response can be detected from plasma. Using a predictive model trained on lung cancer and non-cancer samples, we were able to detect changes in prediction scores in patient treated with immunotherapy that were consistent with RECIST. Patients with progressive disease (n = 3), determined by RECIST, had prediction scores that increased while they received treatment. On the other hand, majority of the patients that exhibited partial response to treatment (n = 12) had predictive scores that decreased with treatment, again consistent with RECIST. Furthermore, score changes consistent with RECIST was observed one cycle prior to the RECIST timepoint in all except one patient, where an extra blood draw after baseline was available (n = 7). Annotation of the regions that account for differential scoring identified enhancer, 5’UTR and promoter regions. Comparison of partial responders to patients with progressive disease revealed genes involved in metastasis, oncogenes and tumor suppressors that change in opposing directions between these patient groups, consistent with the underlying biology. Conclusions: Our results suggest that 5hmC profiles from cfDNA can be used to determine immunotherapy response in non-small cell lung cancer patients. Compared with mutation based liquid biopsy methods to assess response, epigenomics-based methods have the advantage of being agnostic to starting tumor mutations, and not relying on a mutational analysis from tumor biopsy. Future work will help determine applicability of this method to other kinds of therapies and cancer types.

1539 Background: Epigenomic changes in DNA methylation patterns are more precisely delineated by active demethylation events as marked by 5-hydroxymethylation (5hmC) of cytosine residues. 5hmC appears to be dynamically modulated in tumor tissues and can be employed as a cancer biomarker. Strategies which interrogate 5hmC genome-wide patterns in a liquid biopsy context may provide efficient and precise technology for early cancer screening and detection. In this study we identified genome-wide 5hmC changes in plasma based circulating free DNA (cfDNA) from breast, colorectal, lung, pancreatic and prostate cancer patients versus non-cancer individuals. Methods: cfDNA was isolated from plasma, enriched for the 5hmC fraction using novel click-chemistry protocol for labelling followed by sequencing and alignment to a reference genome to construct features sets of 5hmC patterns. Regularized classification models were constructed to classify cancer samples apart from non-cancer. Results: > 500 non-cancer individuals and > 500 cancer patients across five cancer types (breast, colorectal, lung, pancreas and prostate) were included in this study. About 60% of the cancer samples were early stage disease (I or II). The ability to classify non-cancer versus cancer patients was evaluated by 5-fold cross validation of our trained prediction models. Our models were able to classify all breast cancer with a test auROC of 0.86 while prediction model classification for ER negative samples had an auROC of 0.92. Colorectal performance auROC was 0.9; lung auROC = 0.92, pancreatic auROC = 0.97 and prostate auROC = 0.91. Overall sensitivity range, when allowing 2% false positive, was between 85% and 52%. Further using 5hmC signal in blood we were able to identify several signaling pathways specifically relevant to the biology of the cancers investigated. Conclusions: These findings further demonstrate that 5hmC changes in cfDNA enable non-invasive detection of breast, colorectal, lung pancreatic, and prostate cancers. Further, 5hmC signals enabled the identification of a suite of cancer signaling pathways differentially enriched in cancers versus non-cancers. These data suggest that dynamic changes in tumor cell methylation, detectable through 5-hydroxymethylation, are contained in the circulating blood and signal active disease biology.

Pancreatic cancers are typically diagnosed at late stage where disease prognosis is poor as exemplified by a 5-year survival rate of 8.2%. Earlier diagnosis would be beneficial by enabling surgical resection or earlier application of therapeutic regimens. We investigated the detection of pancreatic ductal adenocarcinoma (PDAC) in a non-invasive manner by interrogating changes in 5-hydroxymethylation cytosine status (5hmC) of circulating cell free DNA in the plasma of a PDAC cohort (n=51) in comparison with a non-cancer cohort (n=41). We found that 5hmC sites are enriched in a disease and stage specific manner in exons, 3’UTRs and transcription termination sites. Our data show that 5hmC density is reduced in promoters and histone H3K4me3-associated sites with progressive disease suggesting increased transcriptional activity. 5hmC density is differentially represented in thousands of genes, and a stringently filtered set of the most significant genes points to biology related to pancreas (GATA4, GATA6, PROX1, ONECUT1) and/or cancer development (YAP1, TEAD1, PROX1, ONECUT1, ONECUT2, IGF1 and IGF2). Regularized regression models were built using 5hmC densities in statistically filtered genes or a comprehensive set of highly variable 5hmC counts in genes and performed with an AUC = 0.94-0.96 on training data. We were able to test the ability to classify PDAC and non-cancer samples with the Elastic net and Lasso models on two external pancreatic cancer 5hmC data sets and found validation performance to be AUC = 0.74-0.97. The findings suggest that 5hmC changes enable classification of PDAC patients with high fidelity and are worthy of further investigation on larger cohorts of patient samples.

Pancreatic cancers are typically diagnosed at late stage where disease prognosis is poor as exemplified by a 5-year survival rate of 8.2%. Earlier diagnosis would be beneficial by enabling surgical resection or earlier application of therapeutic regimens. We investigated the detection of pancreatic ductal adenocarcinoma (PDAC) in a non-invasive manner by interrogating changes in 5-hydroxymethylated cytosines (5hmC) in circulating cell free DNA in the plasma of a PDAC cohort (n=51) in comparison with a non-cancer cohort (n=41). 5hmC profiles from PDAC and non-cancer samples were generated using a previously published modified hMe-Seal protocol that utilizes chemical labeling of 5hmC by β-glucosyltransferase and allows detection of cell free 5hmC from small amounts of cfDNA (1). We found that 5hmC sites are enriched in a disease and stage specific manner in exons, 3’UTRs and transcription termination sites. Our data show that 5hmC density is reduced in promoters and histone H3K4me3 associated sites with progressive disease suggesting increased transcriptional activity. 5hmC density is differentially represented in thousands of genes, and a stringently filtered set of the most significant genes points to biology related to pancreas (GATA4, GATA6, PROX1, ONECUT1) and/or cancer development (YAP1, TEAD1, PROX1, ONECUT1, ONECUT2, IGF1 and IGF2). Regularized regression models were built using 5hmC densities in a comprehensive set of genes with the most variable 5hmC counts and performed with an AUC = 0.94 - 0.96 on training data. We tested the ability to classify PDAC and non-cancer samples with the Elastic net and Lasso models on three independent pancreatic cancer 5hmC data sets (n = 26, 23 and 7) compared with corresponding independent non-cancer cohorts (n =103, 53 and 10), and found validation performance to be AUC = 0.74 - 0.97. The findings suggest that 5hmC changes enable classification of PDAC patients with high fidelity and are worthy of further investigation on larger cohorts of patient samples. Reference: 1. Song, C. - X. et al. 5 - Hydroxymethylcytosine signatures in cell-free DNA provide information about tumor types and stages. Cell Res 27, 1231 (2017). Citation Format: Francois Collin, Yuhong Ning, Gulfem D. Guler, Tierney Phillips, Erin McCarthy, Aaron Scott, Chris Ellison, Chin-Jen Ku, Kim Chau, Alan Ashworth, Stephen R. Quake, Samuel Levy. Detection of early stage pancreatic cancer using 5–hydroxymethylcytosine signatures in circulating cell free DNA [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2019; 2019 Mar 29-Apr 3; Atlanta, GA. Philadelphia (PA): AACR; Cancer Res 2019;79(13 Suppl):Abstract nr 1372.

3058 Background: Methylation and hydroxymethylation of cytosines enable the epigenomic regulation of gene suppression and activation. 5-hydroxymethyl-cytosine (5hmC) is globally decreased in tumor tissue. However, genome-wide analysis using precise 5hmC labelling techniques reveals more nuanced changes upon tumorigenesis and raises the possibility that this loss could be exploited for developing a cancer biomarker. This suggests that 5hmC profiles might enable discrete classification of not only tumor tissue but also of tumor cell-free DNA (cfDNA). We sought to identify genome-wide 5hmC changes in plasma based cfDNA from cancer patients representing multiple disease types, stages and clinical characteristics in comparison with non-cancer patients. Methods: cfDNA was isolated from plasma, enriched for the 5hmC fraction using chemical labelling, sequenced, and aligned to the genome to determine 5hmC counts per genomic feature. Regularized regression models were constructed to classify cancer samples (age matched or corrected for smoking status) on non-overlapping training (80% of all samples) and test sample sets (20% of all samples). Results: 226 non-cancer patients and 278 cancers across four cancer types (breast, colorectal, lung-squamous and pancreas) were included in this study, where more than 60% of cancer samples were early stage disease (I or II). Upon comparison with non-cancer samples, 5hmC peaks have reduced enrichment in exons in breast, colorectal and lung cancer but not in pancreatic cancer. Further, 5hmC peaks in pancreas show different patterns of enrichment in 3’UTR, translational termination sites, promoters and LTR. Overall 5hmC signal density was reduced in late stage cancers across all four diseases. The ability to classify non-cancer versus cancer patients was evaluated via cross-validation of out of fold prediction in the training set with AUC > 0.84 for all four cancer types. Further, test set sensitivity across all four cancer types was found to be > 66% with 98% specificity. Conclusions: These findings suggest that 5hmC changes in plasma cfDNA enable classification of early stages of breast, colorectal, lung-squamous and pancreas cancer and are promising biomarkers for disease detection.