Genetic heterogeneity is a major concern in cancer especially AML as it drives phenotypic adaptability which may contribute to clonal selection during treatment. One of the important causes of genetic heterogeneity is genomic instability. This instability leads to an increased mutation rate which can shape the evolution of the cancer genome through many mechanisms. Moreover, the diversity of genetic events in a cancer genome poses a significant challenge to identifying the oncogenic effect of specific genetic abnormalities. The advent of next generation sequencing (NGS) technology has allowed for more comprehensive analysis of somatic variants which has facilitated understanding of AML initiation and progression [8, 13–15, 18–20].
In this study, high-throughput paired-end exome sequencing of 6 matched presentation and relapse CN-AML samples was performed which enabled comparison of mutational landscape to better understand cancer evolution from presentation to relapse. The advantage of paired-end sequencing allowed both ends of a DNA fragment to be sequenced thus generating high-quality and alignable sequence data to facilitate detection of genomic variants with high confidence. However, one caveat with this approach is that the genome coverage was very low (the human exome represents ~ 1.2% of the total genome) and mutations in non-coding sequence were not identified. Despite that, with high read depth achieved in this study as a result of exonic sequence enrichment, identification of somatic variants present in minor leukemic cell clones both at presentation and relapse was achieved.
Somatic Variants at Presentation and Relapse
SNVs and indels are the most abundant type of genetic variation in the human genome with ratios of indels to SNVs were 0.19–0.22 in WGS and 0.14 in WES [21, 22]. Higher ratio of indels to SNVs were identified in this study with 0.60 at both presentation and relapse. In addition, the average number of somatic variants (SNVs and indels) identified in this study were fewer compared to average number of mutations found in previous somatic mutation studies [14, 15, 19, 20, 23]. However, some previous studies had greater coverage of the genome (whole genome sequencing) and larger sample sizes whereas SNVs and indels identified in this study were strictly limited to somatic variants that may have biological implications in AML pathogenesis. Despite this, the number of somatic mutations in AML is relatively low with an average of 13 mutations per genome as opposed to other cancers occurring in adults which often have hundreds of somatic mutations, especially solid tumors such as breast, lung or pancreas [19]. Nevertheless, accurate identification of pathogenic SNVs and indels is one of the most important challenges in cancer genome analysis and requires detailed information of their impact on disease development. The use of analysis software equipped with linked databases that provides comprehensive information of each variant is very useful to discern functionality and a role in leukemogenesis.
Coexisting Variant within Individual Patients
The two-hit model of leukemogenesis has served as a basis to understand how mutations in certain genes drive AML [16]. Based on this model, mutations were divided according to two fundamental characteristics of cancer cells, including mutations that confer cellular proliferation (class I mutations) and mutations that impair myeloid differentiation (class II mutations). The model predicts that both classes of mutation are required for AML transformation. However, genomic studies have revealed the presence of mutations in genes associated with epigenetic modification (ASXL1, DNMT3A, IDH1/2, TET2, EZH2 and MLL) in a significant proportion of AML patients [24]. Disruption of epigenetic processes including DNA methylation and histone modification can lead to altered function in genes involve in key cellular pathways such as DNA repair, RAS signaling, cell cycle and apoptosis which may cause malignant cellular transformation in cancer [25]. In addition, a genome-wide DNA methylation profiling study has demonstrated that specific methylation profiles are associated with specific AML subtypes [26], which identified oncogenic cooperativity between somatic alterations in epigenetic regulators and known AML driver mutations. However, determining oncogenic cooperativity between mutations is difficult as it requires prior knowledge of the function of the genes as well as the impact of the mutated genes on the translated protein. Therefore, given the abundant number of mutated genes with uncertain significance identified in this cohort, analysis was limited to genes most recurrently mutated in AML as reported in COSMIC and genes not previously implicated in AML pathogenesis that have important functions potentially relevant for leukemogenesis.
The mutational spectrum exhibited by each patient in this study was totally different which demonstrates the heterogeneous nature and complexity of the genomic landscape in AML. Despite this variation, all patients had mutations in at least one known AML driver genes that is reported as recurrently mutated in AML in COSMIC. NPM1, FLT3, NRAS, CEBPA and IDH2 were the mutated genes identified in these patients at presentation which are well reported for their importance and contribution to AML pathogenesis. However, co-occurrence of class I and II mutations was only observed in one patient at presentation (Patient 1). FLT3-ITD and NPM1 p.W288fs (TCTG insertion) mutations harbored by Patient 1 are inarguably defined as AML driver mutations since numerous studies has showed the co-occurrence of these mutations in AML patients [15, 18, 27, 28]. In Patient 3, CEBPA mutation appeared to be a driver mutation as it is recurrently mutated in AML patients [29, 30] and plays a key role in AML initiation as demonstrated in murine models [31, 32]. Although no additional mutations in known AML driver genes were detected in this patient, mutation in MUTYH gene appeared as a candidate mutation to cooperate with CEBPA mutation at presentation since its involvement in oxidative DNA damage repair [33, 34] compared to another accompanying mutation in CFTR gene which was commonly reported in patients with cystic fibrosis [35, 36]. WRN gene was recognized as tumor-suppressor gene and evidence suggests a role in promoting oncogenic proliferation [37, 38] which may likely contribute to AML transformation in Patient 5 in addition to NPM1 and IDH2 mutations.
Collectively, these findings emphasize the necessity of further investigation of mutation cooperativity, especially for mutations in genes that does not belong to either class I or class II mutations. Indeed, it is particularly important to identify mutation cooperativity in patients that lack known recurrent combinations of mutations as this might identify novel mechanisms of AML leukemogenesis.
Clonal Evolution from Presentation to Relapse
One major advantage of NGS approaches is that it can be used to quantitate the proportion of variant reads for any given mutation, also known as the variant allele fraction (VAF), which indicates the percentage of tumor cells that harbor a specific mutation assuming a relatively pure leukemic sample. Using this VAF score, each mutation at presentation can be tracked to map clones that persisted, those that resolved from presentation to relapse as well as novel mutations that emerged at relapse in order to delineate clonal evolution from presentation to relapse.
In most cases, mutations detected at presentation were also present at relapse with additional mutations, except for one case (Patient 3) which had no additional mutation acquired at relapse (Fig. 1). However, almost all of these additional mutations acquired at relapse are mutations in genes that have yet unknown function in leukemogenesis. By excluding these mutations, variants were analyzed to define the effect of chemotherapy on the prevalence of somatic variants at relapse specifically in known AML driver genes based on comparison between VAF score at presentation and relapse (Fig. 4). Except for NPM1 p.W288fs and FLT3-ITD mutations in which no VAF scores were available (not detected by exome sequencing), FLT3 p.T526M, CEBPA p.Q346fs*10 and CEBPA p.V343fs*11 mutations were found to be persist after treatment by having similar values at presentation and relapse (less than 10% difference). NRAS p.G12D mutation is predicted to confer relative resistance to standard combination chemotherapy treatment based on their increased VAF (greater than 10% difference) at relapse, whereas CEBPA p.Q312dup mutation was predicted to confer sensitivity to chemotherapy based on a reduced VAF score (greater than 10% difference) at relapse. On the other hand, ATM c.497-1G > T is predicted to be induced by or selected by chemotherapy since this mutation was not detected at presentation.
Further analysis was performed to determine the clonal evolution pattern from presentation to relapse in this cohort. Based on the models of clonal evolution of AML [39], a linear evolution pattern was clearly observed in one patient (epitomized by Patient 6; Fig. 4) which is characterized by the persistent of major clone at relapse with additional mutations. However, in some cases, relapse might be driven by the same set of mutations acquired at presentation, suggesting these mutations confer selective advantage during AML treatment, resulting in clonal expansion and eventually leading to relapse (epitomized by patient 1, 4 and 5; Fig. 4). In addition, despite no evidence of additional mutations at relapse, the VAF score for CEBPA p.Q312dup was greatly reduced, suggesting a major clone with a CEBPA p.Q312dup mutation failed to survived chemotherapy (epitomized by patient 3; Fig. 4).
Collectively, the findings from this study suggests two possible mechanisms of relapse: (1) a major clone at disease presentation survived chemotherapy and re-emerged at relapse after acquired additional mutations; (2) a major clone survived chemotherapy and re-emerged at relapse with the same set of mutations without additional mutations. The former showed that relapse is driven by the acquisition of additional somatic mutations, consistent with other previous studies investigating clonal evolution of AML [8, 13–15, 40], while the latter is associated with clonal evolution with stable mutations only [39], suggesting the presence of non-genetic alterations, such as epigenetic alterations, that might confer chemotherapy resistance. Sequence analysis of a remission sample is therefore essential to determine the VAF of these mutations at remission in order to confidently identify the pattern of clonal evolution.