Martin Hirst

Structural variants (SVs) contribute greater diversity at the nucleotide level between two human genomes than any other form of genetic variation and are three-fold more likely to correlate in genome-wide association studies (GWAS) than single nucleotide variants (SNVs). Using short-read, high-throughput sequencing technologies to uncover such variation has proven to be troublesome and the methods to detect SVs depend on indirect inferences. However, while larger (>5kb) copy number variations (CNVs) could be characterized using read-depth-based algorithms, this approach often fails for smaller and balanced events. Another fundamental problem for detection of SVs from short-read sequencing is inherent to the predominant data type and typical SV detection algorithm that is effective in unique sequences often fails within complex genomic regions, which have been proven to be highly enriched for SVs. In addition, most SV discovery methods do not indicate the haplotype-origin for a given SV and require parental sequencing for this information. For a more complete description and interpretation of human genomic information in relation to phenotypes such as e.g. cancer predisposition and response to therapies, it will, therefore, be necessary to arrange sequence data into parental haplotypes and ascertain polymorphic inversions with respect to such haplotypes. All this can be achieved using Strand-seq. Strand-seq complements other sequencing approaches by providing crucial information about the genetic make-up of individuals that cannot be obtained in any other way. To make Strand-seq available for human studies worldwide is an immense challenge. Library construction, as well as data analysis, needs to be further developed, integrated and made user-friendly to allow accurate and rapid interpretation of results. Here we present a custom bioinformatics pipeline for analyzing Strand-seq data that streamlines the workflow of raw sequence read alignment, putative variant calling, variant call refinement and haplotype assembly by integrating current available Strand-seq specific tools. In addition, relevant metric data are compiled and visualized, ensuring and reinforcing the potential of Strand-seq as a robust sequencing method for uncovering clinically significant SVs and the assembly of WGH without additional parental genomic data.

View record

Neomorphic mutations in isocitrate dehydrogenase 1/2 (IDH1/2) and inactivating mutations in ten eleven translocation dioxygenase 2 (TET2) are frequent and mutually exclusive events in de novo acute myeloid leukemia (AML). IDH1/2 mutations drive epigenomic dysfunction through production of the oncometabolite R-2-hydroxyglutarate (R-2HG), which inhibits the ability of TET2 to oxidize 5-methylcytosine (5mC) to 5-hydroxymethylcytosine (5hmC), and causes DNA hypermethylation. Vitamin C (vitC) induces epigenetic reprogramming and cellular differentiation through TET re-activation in a murine leukemia model with mutant IDH1ᴿ¹³²ᴴ. Although TET mutations in AML occur exclusively in the TET2 gene, TET2 and TET3 are expressed at similar levels in de novo AML blasts. The role of TET3 in the regulation of myeloid activation and its role in AML is currently unknown. Based on the functional redundancy among TET enzymes, I hypothesize that the vitC-induced re-establishment of methylation homeostasis and differentiation in IDH1 mutant cells is also mediated through TET3. To delineate the contributions of TET2 and TET3 to vitC-induced reprogramming, we inactivated TET2 (TET2KO) and TET3 (TET3KO) individually or in combination (DKO) in HOXA9/IDH1ᴿ¹³²ᴴ (R132H) cells using CRISPR/Cas9. After 15 and 72 hours of vitC treatment, we sequenced total RNA and 5hmC/5mC-immunoprecipitated DNA. We showed that IDH1ᴿ¹³²ᴴ inhibition of TET is incomplete and that loss of either TET2 or TET3 down-regulates a set of 136 genes related to myeloid activation. A subset of the TET2/3-dependent genes can be rescued with vitC activation of the remaining TET, from which we identified PU.1, CEBPE, and RUNX1 as putative transcription factors.We identified a hypermethylation phenotype at enhancers in the context of the IDH1ᴿ¹³²ᴴ, which can be reversed through vitC treatment in a TET2- and TET3- dependent mechanism. Pathway analyses of nearby genes and genes up-regulated by vitC suggest a common myeloid differentiation pathway that both TET2 and TET3 can activate. To verify this, TET2KO and TET3KO but not DKO cells showed reduced proliferation and increased levels of myeloid differentiation markers from vitC treatment. These findings support my hypothesis and suggest a model in which vitC activates both TET2 and TET3 to reprogram the enhancer landscape of IDH1ᴿ¹³²ᴴ leukemic cells to drive differentiation.

View record

Increasing evidence of functional and transcriptional heterogeneity in phenotypically similar single-cells has prompted interest in protocols for obtaining parallel methylome data. Despite appreciable advancements in experimental protocols for single-cell DNA methylation measurements, methods for analyzing the resulting data are still immature. To address the challenge of stochastic data loss associated with single cell measurements, current strategies average methylation in windows or region sets. However previous studies have demonstrated that single CpGs are functional and our analysis of single cell methylation measurements revealed a rapid decay in concordance neighbouring CpG states beyond 1kb. To leverage the information content of individual CpGs in the context of single cell methylation measurements we developed an analytical strategy for deriving single-cell DNA methylation states using individual CpGs, which we term PDclust. We validated PDclust on existing datasets and on data we generated from single index-sorted murine and human hematopoietic stem cells (HSCs) that are highly enriched in functionally defined stem cells. Using PDClust, we identified epigenetically distinct subpopulations within these HSC populations. Strikingly, human cord blood derived HSC populations were separable by donor specific methylation states whereas more differentiated hematopoietic cells separated solely by cell type. Interestingly, removal of methylation sites near genetic variants did not impact this separation, suggesting that these epigenetic states may be a consequence of environmental differences. Finally, through protocol optimization and deep sequencing we generated one of the most comprehensive sets of single cell methylome profiles (20% of CpGs on average) and from these were able to generate genomewide profiles from as little as 6 epigenetically related HSCs to derive subtype-specific regulatory states.

View record