The Complexity of Interpreting Genomic Data in Patients with Primary and Secondary Acute Myeloid Leukemia (AML)

Background

Acute myeloid leukemia (AML) is a complex, heterogeneous neoplasm characterized by the accumulation of complex genetic alterations that are responsible for the initiation and progression of the disease. Secondary AML (sAML) represents a progression from antecedent hematologic disorders such as myelodysplastic syndromes (MDS) or myeloprolifrative neoplasms (MPN). Certain acquired mutations have been reported to be specific for sAML when compared to primary AML (pAML), but many limitations exist when cytogenetic grouping or other parameters are taken into account. In addition, some mutations have been shown to impact survival in some studies, but not others.

Methods

We performed targeted deep sequencing on samples from bone marrow and peripheral blood of pts diagnosed with sAML and pAML and treated at our institution between 1/2003-1/2013. Additional data on pAML was added from The Cancer Genome Atlas (TCGA). A panel of 62 gene mutations described as frequently recurrent mutations in myeloid malignancies were assessed. Cytogenetic grouping was defined by CALGB/Alliance criteria. Differences were compared using Fisher's exact test and the Mann-Whitney U test for categorical and continuous variables, respectively. Overall survival (OS) was calculated from the time of diagnosis to last follow up or death.

Results:

A total of 496 pts included: 273 with pAML and 223 with sAML. Comparing pAML to sAML, pts were younger (median age 59 vs. 68 years, p<.001) and had a higher WBC at diagnosis (13.5 vs. 3.9 X 10⁹/L, p<.001), respectively. Cytogenetic analysis showed significant differences: 58% of pAML pts had normal karyotype (NK) compared to 37% of sAML (p=.002), whereas 24% and 26% of sAML had intermediate risk (other than NK) and complex karyotype (> 3 abnormalities) compared to 11% and 16% for pAML (p< .001, .009), respectively. Mutations in ASXL1 (p<.001), JAK2 (p=.014), CBL (p=.05), BCOR (p=.02), STAG2 (p =.003), SF3B1 (p=.04), SRSF2 (p=.001 ), and U2AF1 (p=.03) were highly specific for the sAML phenotype, whereas mutations in NPM1 (p<.001 ), FLT3 (p< .001), DNMT3A (p<.001), and IDH2 (p=.02) were more specific for pAML. When the analysis was restricted to pts with NK cytogenetics, only ASXL1 (p<.001) remained specific for sAML and DNMT3A (p<.001) for pAML.Further, when the analysis was restricted to pts with unfavorable risk cytogenetics, only ASXL1 (p=.01) remained specific for sAML. No other mutations were specific for pAML.

We then evaluated whether the mutations that were specific to each AML phenotype had an impact on OS. We observed different mutations that impacted OS in each phenotype: DNMT3A (HR 1.81, 95% CI 1.28-2.57, p<.001), TP53 (HR 3.1, 95% 1.74-5.53, p< .001), and SUZ12 (HR 3.18, 95% CI 1.01-10, p=.05) led to worse OS in pAML, whereas mutations in EZH2 (HR 2.12, 95% CI 1.07-4.21, p =.03), PRPF8 (HR 2.32, 95% CI 1.20-4.46, p=.01), and TP53 ( HR 2.92, 95% CI 1.69-5.04, p<.001) lead to worse OS in sAML. Different mutations had a different impact on OS when cytogenetic analysis was taken into account. Mutations in FLT3 (HR 2.15, 95% CI 1.37- 3.35, p<.001) and DNMT3A (HR 2.41, 95% CI 1.57-3.70, p<.001) led to worse OS in NK pAML, whereas none of the mutations impacted OS in NK sAML. Further, in pAML with unfavorable cytogenetics, BCOR (HR 2.41, 95% CI 1.57-3.70, p<.001) and TP53 (HR 2.41, 95% CI 1.57-3.70, p<.001) had led to worse OS, whereas BOCR (HR 2.95, 95% CI 1.03-8.50, p<.001), SF3B1 (HR .19, 95% CI .05-.82, p<.001), SUZ12 (HR .12, 95% CI .01-.99, p<.001),and TP53 (HR 1.9, 95% CI 1.09-3.46, p<.001) only impacted OS in sAML.

Conclusion

Clear genomic variations exist between sAML and pAML. Although some of these genomic changes are more specific to each phenotype in general, this specificity and the impact on OS differed for each cytogenetic subgroup, highlighting the complexity of interpreting genomic information in pts with AML and the need to incorporate both cytogenetic and molecular data in prognosis-driven treatment decisions.