Sensitive detection of novel structural variants and 3D chromosome conformation reveals likely novel drivers including enhancer hijacking in AML Free

Introduction: Extensive characterization of the genetic landscape in AML has revealed a set of driver mutations sufficient to initiate leukemic transformation. However, some cases do not contain any of these known drivers and thus pose a clinical challenge for prognostication and targeted therapy. To address this, we employed a novel, proximity ligation based, linked-read whole genome NGS assay to identify single nucleotide variants (SNVs), short Insertions/Deletions (InDels), copy number variation (CNV), structural variants (SVs) and 3D genome contact maps in AML. By employing this approach, we investigate whether comprehensive detection of genetic alterations in conjunction with insights into the epigenetic state of leukemic cells can reveal important biological information in cases lacking canonical driver mutations.

Methods: We curated a cohort of patient samples where concomitant targeted sequencing by NGS and conventional karyotyping showed no detectable driver mutations. Further, we included control specimens with either known AML drivers as well as remission specimens after allogeneic transplantation. We utilized a multi-purpose NGS assay (Dovetail® LinkPrep™) to detect SNVs, InDels, CNVs, SVs and 3D genome contact maps. Peripheral blood or bone marrow cells were thawed and immediately fixed, subjected to crosslinking and in situ tagmentation, followed by library prep. Libraries were sequenced on Illumina sequencers with a target of 400M reads per library. Data analysis was performed using both published and custom tools to call SNVs/InDels (DeepVariant), Copy Number estimates (Purple), Structural Variants (HiC-Breakfinder, DTG-SELVA) and 3D contact maps (Juicebox).

Results: We sequenced whole genome libraries to achieve an average coverage of 50.4x (range 41.1x - 58.1x) showing a long-range linkage percentage of 26.2% (range 17.9% - 31.8%). SNVs and short InDels, such as likely pre-leukemic mutations in DNMT3A, TET2, and mutations in GATA2 were detected reliably. Of note, the linked-read nature of this assay allowed us to perform phasing in a case with two TET2 mutations and confirm the compound heterozygous nature of these two TET2 mutations at a distance of 24kb. Further, known AML drivers in control specimens in genes such as NPM1, FLT3 and KRAS were reliably detected with variant allele frequencies as detected by targeted sequencing. Surprisingly, however, we found new structural variants in half of our cohort, including the diagnostic specimens with no detectable driver mutations, despite all specimens having no chromosomal abnormalities by conventional karyotyping. These new SVs included large intrachromosomal inversions and unbalanced translocations. We validated the presence of all SVs in sorted leukemic blasts and absence in sorted T cell fractions by PCR and Sanger sequencing. We subsequently examined the 3D genome contact maps and performed RNA-Sequencing to determine the epigenetic state and potential biological consequences across these novel breakpoints. We found several neoloops connecting enhancers with promoters across these breakpoints in 3D contact maps, and gene expression dysregulation proximal to the SVs by RNA-Seq. When we evaluated the expression of these putative driver genes across a cohort of healthy donors and AML patients, we found lack of expression in healthy cells but significantly increased expression in 9-10% of patients with AML (n=777 patients). Finally, we also identified an undetected structural variant in engrafted donor cells from a patient with myelodysplastic neoplasms (MDS) in remission after allogeneic transplantation, highlighting the potential to assay SVs in the context of non-leukemic hematopoiesis as well.

Conclusion: In this exploratory cohort, we find several otherwise undetected structural variants in AML cases without known driver lesions using a proximity ligation based, linked-read whole genome NGS assay while maintaining our ability to detect recurrent, known driver lesions. This approach allowed us to interrogate the biological consequences of these genomic alterations by leveraging the ability to not only detect genomic variants but also analyze the epigenetic state through 3D genome contact maps. Our results imply that our current understanding of the genomic landscape in AML might be underestimating a sizable number of structural variants that are difficult to detect using standard assays, implying putative novel driver genes in AML pathogenesis.