Identifying the DNA binding specificity of mixed lineage leukaemia in leukaemia

Thesis

Chromosomal translocations of the KMT2A gene generate MLL fusion proteins that drive aberrant gene expression and define an aggressive, poor-prognosis subset of acute leukaemia. Although MLL binds unmethylated CpG-island promoters through its CXXC domain, only a subset of these potential sites is occupied in the fusion context. This thesis investigates the determinants of that selectivity across three regulatory layers: DNA sequence, co-factors and chromatin environment, and DNA methylation, including 5-methylcytosine (5mC) and 5-hydroxymethylcytosine (5hmC).

Attribution analyses from DNA-sequence models revealed a shared CpG baseline across cell types, characterised by strong positive correlations with CG-rich motifs such as VEZF1, KLFs, and MAZ. In the non-fusion MLL context, this CpG signal was refined by negatively attributed motifs, including MYC and KLF6, restricting binding to a selective subset of CpG islands. In MLL-fusion cells, CpG density contributed less strongly to prediction, and negatively attributed lineage-specific motifs such as RUNX and SP-family transcription factors highlighted cell-type specificity, suggesting that transcription-factor context along with CpG content, determines occupancy.

Models based on co-binding proteins and histone marks revealed two regimes: MLL co-localised with Menin, LEDGF, and initiation machinery at promoters, whereas MLL-AF4 was enriched along active transcription units with elongation-associated complexes, including PAF1, the super-elongation complex, BRD proteins, RNA polymerase II, and DOT1L with H3K79me2. These co-binding patterns indicate that chromatin context directs selectivity within CpG-rich regions.

Integrating DNA methylation with hydroxymethylation showed that unmethylated CpG provides a baseline requirement for binding, but differential cytosine modification states sharpen site selection. Across cell types, MLL-AF4 cells displayed elevated promoter 5mC and globally reduced 5hmC. Inclusion of 5hmC features improved prediction of MLL occupancy. Local 5hmC at the +2 nucleosome emerged as a salient predictor in a non-fusion context. In MLL-AF4 correlations with TET2 binding suggest that TET2-driven demethylation facilitates recruitment at specific elongation-linked sites.

Together, these results support a model in which CpG recognition defines baseline potential, while transcription-factor context and chromatin environment tune site selection. TET2-mediated modulation of methylation state adds a further regulatory layer that determines which CpG-rich regions are ultimately bound.

Authors: Chahrour, C

• DOI: 10.5287/ora-o0nojqoxr
• Type of award: DPhil
• Level of award: Doctoral
• Awarding institution: University of Oxford
• Language: English
• Keywords: DNA binding specificity; Mixed Lineage Leukaemia (MLL/KMT2A)
• Subjects: Bioinformatics / Computational Biology; Epigenetics