www.alzheimers.emory.edu

DiseasediagnosisforADusingcell-freeDNA

Cell-freeDNA(cfDNA)

• ShortDNAfragmentsinplasmathatdon’tbelongtoanycell.

• Releasedfrombloodcells/deadtissuecells/externalcells.

• ExamplesofcfDNA:o Incancerpatients,circulatingtumorDNA

(ctDNA).

o Inpregnantwomen,cell-freefetalDNA(cffDNA).

MotivationtostudycfDNA

• Hasgreatpotentialstobediagnosticbiomarker:o Noninvasive.

o Earlydetection.

o Cheapandeasy.

“liquidbiopsy”.

Cancerdetectionandassessment

cfDNAasdiagnosticbiomarker

• Theessenceistodiagnosebasedon“abnormal”cfDNA.

• Currentlytheinformationof“abnormality”isbasedonmutations.o cfDNAwithunusuallyhighnumberofmutationsaredeemed

abnormal,andcanbelinkedtodisease.

• SofartheapplicationofcfDNAaremostlylimitedto“mutation-rich”diseasessuchascancer,NIPT,etc.

• Theprincipalcannotbedirectlyappliedformutation-poordiseases,suchasAD.

cfDNAepigenetics

• Itisknownthatepigeneticsignaturesarehighlytissue-specific.

• ItispossibletoexploretheepigeneticinformationoncfDNA,andthenlinkthosetodiseases.

• Existingworks:o DNAmethylation(Sunetal.2015PNAS).o Nucleosomeposition(Snyderetal.2016Cell).

cfDNAepigeneticbiomarker

• cfDNAisamixtureofDNAfromdifferenttissues.

• ADleadstochangeofmixingproportions(greaterproportionofcfDNAfrombraininADcases),thusthemarginalepigeneticprofiles.

• Useepigeneticprofilesatselectedgenomicregionsaspredictorsfordisease.

Predictionmethod

• Weinvestigatedandcomparedmethodsto– UsecfDNAmarkers

– Useestimatedmixingproportions

Disease prediction by cell-free DNA methylationHao Feng, Peng Jin and Hao WuCorresponding author: Hao Wu, Department of Biostatistics and Bioinformatics, Emory University Rollins School of Public Health, Atlanta, GA 30322, USA.Tel.: þ1 404-727-8633. E-mail: hao.wu@emory.edu

AbstractDisease diagnosis using cell-free DNA (cfDNA) has been an active research field recently. Most existing approaches performdiagnosis based on the detection of sequence variants on cfDNA; thus, their applications are limited to diseases associatedwith high mutation rate such as cancer. Recent developments start to exploit the epigenetic information on cfDNA, whichcould have substantially wider applications. In this work, we provide thorough reviews and discussions on the statisticalmethod developments and data analysis strategies for using cfDNA epigenetic profiles, in particular DNA methylation, toconstruct disease diagnostic models. We focus on two important aspects: marker selection and prediction model construc-tion, under different scenarios. We perform simulations and real data analysis to compare different approaches, and pro-vide recommendations for data analysis.

Key words: cell-free DNA; epigenetics; DNA methylation; liquid biopsy; marker selection; predictive modeling

IntroductionPrognosis and diagnosis play vital roles in the prevention andtreatment of diseases. Traditionally, various types of surgicalbiopsies such as bone marrow or needle biopsies are performedin clinical setting, especially for cancer diagnosis [1]. However,because of the invasive nature of the procedure and the poten-tial sampling bias of tumor biopsy, surgical biopsy is often not apreferred choice. As an alternative to surgical biopsy, research-ers and clinicians have been looking for molecular biomarkersfor disease diagnosis. These biomarkers, either genetic or epige-netic, carry various indicative features for biological or diseasestates. They help achieve disease state detection, subtypes clas-sification, progression prediction and response-to-treatmentcharacterization [2]. The scope of molecular biomarker discov-ery has been greatly expanded during the past two decades,because of the advances of high-throughput genomics technol-ogies such as microarray and next-generation sequencing. Forexample, based on gene expression microarray data, PredictionAnalysis for Microarrays (PAM) identified a subset of gene bio-markers for cancer class prediction [3]. The PAM50 panel, which

tests a group of 50 selected genes, has become the de facto goldstandard for breast cancer subtype classification and metastasisprediction [4]. Zilliox et al. [5] created the ‘gene expressionbarcode’, which was trained on public gene expression microar-ray data, and can predict for a number of diseases given a newmicroarray data set.

In recent years, disease diagnosis based on molecular bio-markers in specimen, including blood, urine and cerebral spinalfluid, has gain tremendous attention. For example, the practiceto look for traces of cancer DNA by interrogating biomarkers onplasma-isolated cell-free DNA (cfDNA) or circulating tumor DNA(ctDNA) is known as ‘liquid biopsy’ [6]. As a safer, cheaper andquicker alternative to surgical biopsy, the liquid biopsy hasgreat potential in clinical practice. cfDNAs are short DNA frag-ments ("160–180 base pairs) existing in plasma. When normalcells undergo apoptosis in a healthy individual, DNA fragmentsfrom the cells are shredded and released to blood stream. Thus,cfDNA is a mixture of DNA fragments from different cell types.In cancer patients, the cfDNA includes some ctDNA, which areDNA fragments released from cancer cells [7]. As a hallmark of

Hao Feng is a PhD Student in the Department of Biostatistics and Bioinformatics at Emory University, Rollins School of Public Health. His research interestis to develop methods and tools for high-throughput epigenetics data.Peng Jin is a Professor and Vice Chair in the Department of Human Genetics at Emory University, School of Medicine. His research interests are the func-tions and mechanisms of epigenetics and noncoding RNAs in neural development and brain disorders.Hao Wu is an Associate Professor in the Department of Biostatistics and Bioinformatics at Emory University, Rollins School of Public Health. His mainresearch interest is to develop statistical methods and computational tools for analyzing large-scale biomedical data, in particular different types of high-throughput omics data.Submitted: 22 December 2017; Received (in revised form): 6 March 2018

VC The Author(s) 2018. Published by Oxford University Press. All rights reserved.For permissions, please email: journals.permissions@oup.com

1

Briefings in Bioinformatics, 2018, 1–13

doi: 10.1093/bib/bby029Review Article

Downloaded from https://academic.oup.com/bib/advance-article-abstract/doi/10.1093/bib/bby029/4973009by gueston 17 April 2018

cfDNA 5hmCdataforAD

• Sample:– 10healthyyoungpeople– 10healthyoldpeople– 10ADpatients

• Samplesource:plasma• Sequencingtechnology:hmC-Sealsequencing• Measurement:genome-wide5-hydroxymethylcytosine

(5hmC)level

Dataprocessing

• Genomewascutintobinswith5kblengtheach

• 5hmClevelwascalculatedforallthebinsforeachsample

• Findthedifferentiallyhydroxymethylation regions(DhMR)between– 10oldsamplesvs10youngsamples(209DhMRs)

– 10oldsamplesvs10ADsamples(296DhMRs)

• UseDhMR aspredictorsforADprediction– Markerselectionisthekey

Preliminarypredictionresults

85%accuracy

Pred AD Pred normal

TrueAD 9 1

True normal 2 8

Futureplan

• Largersamplesize.

• 5mCsequencingofcfDNA.

• Newstatisticalmethoddevelopment:– Featureselection.

– Sampledeconvolution.

– Integrated5mC/5hmC/nucleosomepositioningmodel.

Acknowledgement

• EmoryADRC

• PengJin(EmoryHumanGenetics)

• HarryFeng(EmoryBiostatisticsandBioinformatics)