Age-associated non-mutational resistance in acute myeloid leukemia is driven by DNA damage-induced transcriptional heterogeneity

Abstract Older patients with acute myeloid leukemia (AML) have significantly worse response rates and overall survival compared to younger patients, even when matched for cytogenetic risk. This discrepancy has previously been explained by decreased tolerance of therapy and higher incidence of adverse risk disease. However, when these factors are taken into consideration there remains a higher incidence of primary refractory disease and relapse in older individuals, suggesting an underlying non-mutational resistance to therapy. Recent studies have implicated epigenomic disorganization from DNA damage repair as an important driver of cellular aging. The biological basis for age-related resistance and whether epigenomic disorganization contributes remains unclear. To investigate the impact of age on AML biology, we generated murine AML models by transducing hematopoietic stem and progenitor cells (HSPCs) from fetal (E13.5), young adult (8-10 weeks), and aged (24 months) mice with retroviruses expressing MLL-ENL and NRasG12D. All models developed monocytic AML, confirmed by morphology and immunophenotyping, and were lethal upon transplantation into syngeneic recipients. Additionally, gene expression profile differences between young and old murine models were consistent with differences seen between younger and older AML patient samples (FDR q 0.001) supporting their relevance to the human disease. We assessed in vitro sensitivity to menin inhibition, azacitidine, venetoclax, doxorubicin, and cytarabine. Whole exome sequencing (WES), bulk RNA-seq, single-cell RNA-seq (scRNA-seq), and proteomic profiling were performed to identify molecular correlates of resistance. AML cells derived from aged mice HSPCs were significantly more resistant to all tested therapies compared to fetal and young derived AML cells with marked resistance to cytarabine, doxorubicin and menin inhibition (p0.001). WES did not reveal any recurrent mutations to explain this resistance. Transcriptomic and proteomic data were analyzed by KEGG pathway analysis revealing significant down regulation of several DNA damage repair pathways in older AML cells including base excision repair, homologous recombination and mismatch repair. In contrast, older AML cells had significantly increased expression of anti-apoptotic genes including multiple BCL2 family members. RNA and proteomics data demonstrated that younger AML cells had significantly increased expression of genes/proteins involved in chromatin and epigenome maintenance as identified by Gene Ontology terms (FDR q 0.01) suggesting older AML cells may have a more disordered epigenome. Consistent with this, scRNA-seq revealed significantly increased transcriptional heterogeneity in aged AML cells compared to young and fetal models. This finding was corroborated in primary human AML samples when compared by age (p=0.0004). This suggests that AML cells derived from older HSPCs have a more disordered epigenome likely as the result of repeated rounds of DNA damage repair displacing epigenetic marks. To directly test if repeated rounds of DNA damage repair from a non-carcinogenic source would result in AML therapy resistance, we constitutively expressed the restriction endonuclease PPOI or GFP in fetal derived AML cells. After 12 weeks PPOI expressing fetal AML cells were significantly more resistant to cytarabine (p0.0001) and doxorubicin (p0.0001) compared to GFP expressing cells, mimicking the aged phenotype. They also appeared more aggressive when injected into syngeneic recipients. These findings support a unifying model of epigenetic based therapy resistance in AML, wherein aged HSPCs after multiple years of DNA damage repair, have a disordered epigenome and increased transcriptional heterogeneity. This heterogeneity is passed on to the resultant AML and enables survival of resistant subclones under therapeutic pressure, contributing to poor outcomes in older adults. This would also explain why therapy related and relapsed AML, where repeated DNA damage repair cycles have taken place, also have a worse outcome within the same cytogenetic risk category. Finally, it identifies a potential source of confounding in preclinical models that utilize enzymes that repeatedly induce DNA strand breaks like CAS9. Most importantly, it suggests that clinical trials data should be interpreted with extreme caution when considering patients outside the age range of those enrolled in the study.