Phenotypic heterogeneity and remodeling drive resistance to venetoclax treatment in acute myeloid leukemia

Abstract While venetoclax (VEN) combined with hypomethylating agents (HMA) has greatly advanced outcomes for acute myeloid leukemia (AML), resistance remains a major clinical challenge. Prior studies revealed several resistance mechanisms, including monocytic differentiation and RAS/TP53 mutations, but the full spectrum of mechanisms is still elusive. To assess the cellular mechanisms underlying VEN resistance, we performed an integrative analysis using bulk RNA sequencing (RNA-seq) and single-cell multi-omics in samples collected from the AML patients treated with 10-day decitabine and venetoclax (DAC10+VEN) trial (NCT03404193). First, we performed RNA-seq on 81 baseline and 10 relapse samples. K-means clustering of baseline samples identified a cluster enriched with non-responders, marked by a higher frequency of KRAS mutations (31%, p=0.012) and the absence of NPM1 and IDH1/2 mutations. Using LASSO logistic regression, we developed a 25-gene VEN-resistance (“VEN-R”) score comprising genes associated with MAPK, NF-κB, and apoptosis pathways. The VEN-R score significantly distinguished treatment response to DAC10+VEN (p<0.0001) and was also elevated in relapse samples (p=0.00018), supporting its association with resistance. We further validated VEN-R in two external, ex vivo datasets, finding significantly higher VEN-R scores in VEN-resistant samples compared to VEN-sensitive samples (BeatAML: p<0.0001 and Hashimoto et al. (Nature Cancer 2021): p=0.0002). To study the association between cell composition and DAC10+VEN response, we performed computational deconvolution of the RNA-seq data using CIBERSORTx. Baseline samples had higher fractions of erythroid (p=0.026) and CD8 effector (p=0.034) cells in nonresponders, while hematopoietic stem/progenitor (p=0.0006) and late erythroid (p=0.032) cells were more abundant in responders. Furthermore, compared with baseline samples, relapse samples had decreased hematopoietic stem (p=0.018), lymphoid primed multipotent progenitor (p=0.028), and naïve B (p=0.013) cells, and increased early erythroid (p=0.048), late erythroid (p=0.020), and granulocyte monocyte progenitor (p=0.036) cells. To assess the relationship between cell composition and VEN resistance at the cellular level, we performed single-cell RNA sequencing (scRNA-seq) on 22 samples (117,540 cells; 6 baseline, 16 relapse) treated with DAC10+VEN. scRNA-seq results corroborated the enrichment of mature cells at relapse; specifically, we identified the enrichment of erythroid, monocytic, and cDC (“EMD”) populations and the loss of progenitor cells. BCL2-expressing cells were depleted at relapse, whereas cells expressing MCL1, BCL2L1, or BCL2A1 were enriched in monocytes, erythroid cells, and cDCs, respectively. The EMD cells also had one of the highest expression levels of the VEN-R score at the single-cell level. To examine whether phenotypic remodeling is driven by genetic changes, we then performed single-cell DNA-antibody sequencing (DAb-seq) on 75 longitudinal samples, including six baseline-relapse pairs. DAb-seq analysis substantiated the findings from scRNA-seq that relapse is marked by the loss of stem/progenitor cells and the enrichment of erythroid/monocytic cells. Phenotypic remodeling occurred in five of the six pairs. One monocytic shift was accompanied by the de novo acquisition of FLT3 and KRAS mutations, one erythroid shift by expansion of an NRAS-mutant subclone, and a second erythroid shift by loss of a GATA2-mutant subclone. In the remaining two cases, remodeling occurred without any detectable genetic alterations. Integrating scRNA-seq and DAb-seq from 22 relapse samples revealed four phenotypic patterns: erythroid-predominant (6 cases), monocytic-predominant (4), mixed (erythroid±monocytic±cDC; 8), and inconclusive (4). Finally, we examined whether baseline phenotypic diversity predicts relapse after DAC10+VEN. In the DAb-seq cohort, a higher Shannon diversity index was associated with a higher risk of not only relapse (p=0.043) but also poor response (p=0.057). In summary, DAC10+VEN resistance is frequently observed with the enrichment of mature AML cells. To our knowledge, this is the first study to implicate erythroid cells and cDCs as major contributors to VEN resistance in patient samples, extending prior ex vivo studies (Kuusanmäki et al., Blood 2023) to a clinical setting. Monitoring these phenotypes and combining VEN with agents that target both immature and mature clones may improve patient outcomes.