Poster #94: Muhammed Erguder PhD

Long-Read Sequencing and T2T-CHM13 Alignment Improve Structural Variant Discovery in an Expanding Alzheimer's Disease Cohort

Muhammed F. Erguder, PhD (Columbia University); Dolly Reyes-Dumeyer (Columbia University); Angel L. Piriz (Columbia University); Diones Rivera Mejia, MD (CEDIMAT, Santo Domingo, Dominican Republic); Martin Medrano, MD (Pontificia Universidad Católica Madre y Maestra, Santiago, Dominican Republic); Rafael A. Lantigua (Columbia University); Clifton L. Dalgard, PhD, CLS (Uniformed Services University of the Health Sciences, Bethesda, MD); Richard Mayeux, MD, MSc (Columbia University); Michael C. Zody, PhD (New York Genome Center); and Badri N. Vardarajan, PhD (Columbia University)

Structural variants (SVs) are increasingly recognized as contributors to Alzheimer's disease (AD) risk, but short-read sequencing (SRS) has limited power to resolve them. Long-read sequencing (LRS) and more complete genome references such as the telomere-to-telomere (T2T-CHM13) assembly offer improved SV detection and discovery of biologically relevant variation missed by short reads. We performed LRS on 422 individuals using Oxford Nanopore Technology, with completed analysis for an initial 35 and an additional 387 fastq files now being processed through the same pipeline; results from this expanded cohort will be available by the time of the conference. Base-calling was completed with Dorado v2.1.0 in super-accuracy mode, and reads were aligned to GRCh38 and T2T-CHM13. SVs were called with CuteSV v2.1.1. All individuals also had SVs called from matched Illumina NovaSeq 150bp short-read data, enabling direct cross-platform comparison. SV callsets were further benchmarked against the gnomAD-SV database to characterize known versus novel variant discovery by platform. At ~30x depth, LRS achieved a mean read length of 7.3kb and N50 of 11.8kb. Long-read alignment to GRCh38, and especially to T2T-CHM13, substantially increased SV yield relative to short-read calling, with insertions and deletions comprising the majority of variants across all three callsets, and only moderate overlap between platforms, reflecting the added sensitivity of long-read and T2T-based analysis. LRS additionally revealed SVs in genes at AD GWAS loci, including SORL1, ABCA7, and APP, that were undetected by short-read sequencing. These findings demonstrate that LRS combined with T2T-CHM13 alignment substantially improves SV discovery and uncovers biologically relevant variation in AD-associated genes that short-read approaches miss, underscoring the value of long-read approaches for identifying novel structural contributors to AD risk and building toward a long-read reference resource to support imputation in population-scale AD cohorts.