Distinct Features of Chip-Derived and De Novo MDS

Subclinical clonal expansions, referred to as clonal hematopoiesis of indeterminate potential (CHIP), are present in blood of otherwise healthy individuals and their frequency increases with age. While many CHIP-associated mutations are present in MDS, only a small proportion of asymptomatic individuals with CHIP progress to MDS. We assumed that: i) a proportion of CHIP mutations will eventually serve as ancestral hits that manifest as MDS upon acquisitions of additional genetic alterations, and ii) MDS from antecedent CHIP may be a MDS disease subtype that is distinct from de novo MDS characterized by more penetrant primary hits. Separating ancestral vs. secondary hits in MDS patients and comparing by meta-analyses their frequencies to those in CHIP may enable molecular and clinical characterizations of CHIP-related MDS.

Our study cohort consisted of 1,809 clinically annotated MDS patients. Deep targeted NGS was conducted for a panel of the 36 most frequently mutated "myeloid" genes, which revealed 3,971 somatic mutations in MDS patients after removing SNPs and errors. To discriminate between dominant and subsequent rising secondary mutations, we used a stringent binominal distribution algorithm to define VAFs confidence intervals via loci read counts. Sample maximal VAF mutations were defined as "dominant" and those with overlapping VAF 95% CI were defined as "co-dominant"; mutations with lower non-overlapping 95% CIs were "secondary". These definitions are consistent with those of other methods such as Pyclone, with 95% concordance (1,253/1,317). They yield 1,474 (36%) dominant and 1,372 (35%) secondary mutations.

We compared a meta-analysis of frequently mutated genes in 1,693 healthy CHIP individuals with somatic mutations from the CHIP meta-analysis to dominant mutations in MDS patients. Mutations in DNMT3A, TET2, ASXL1, and JAK2 were more frequent in CHIP than MDS [e.g.,DNMT3A; 52% (888/1,693) vs. 6% (110/1,809), p<.001]. Hence MDS patients with dominant mutations in these 4 genes were defined as CHIP-derived MDS (C-MDS). Other dominant mutations such as U2AF1, RUNX1 and STAG2 which were not identified in individuals with CHIP were deemed not CHIP-derived MDS (NC-MDS). And in between, mutations in TP53, SF3B1 and SRSF2, were identified in both cohorts [e.g., TP53; 4% (71/1,693) vs. 5% (97/1,809), p=.11], and such cases will be denoted as C/NC-MDS patients.

We set out to compare clinical, molecular and demographic features of C- vs. NC-MDS. There were 459 (25%) C-MDS and 498 (28%) NC-MDS cases out of 1809 MDS patients. 95% of patients with C-MDS (437/459) had at least one TET2 (51%, 234/459), DNMT3A (24%, 110/459) or ASXL1 (20%, 93/459) dominant mutation. The top 5 dominant mutations in patients with de novo MDS were U2AF1 (15%, 75/498), ZRSR2 (10%, 52/498), RUNX1 (9%, 47/498), STAG2 (9%, 43/498), and EZH2 (8%, 40/498). 52% (257/498) of patients with de novo MDS had at least one of these mutations. Focusing on secondary mutations, patients with C-MDS had a significantly higher frequency of secondary TET2 and ZRSR2 mutations than those with de novo [e.g., TET2, 17% (77/459) vs. 8% (41/498), p<.001]. In contrast, secondary ASXL1 mutations were more frequent in NC-MDS [7% (31/459) vs. 18% (88/498), p<.001]. C-MDS cases were older and had more low risk subtypes than NC-MDS cases [(average Age) 72 vs. 69, p<.001, (low risk subtypes) 56% (256/459) vs. 45% (222/498), p<.001]. Patients with C-MDS had better prognosis than those with NC-MDS [hazard ratio .75 (.63-.91), p=.003].

Prospective sequencing of serial samples of CHIP to development of MDS may be needed to fully reveal the landscape of C-MDS. This may not be practical, as over a million healthy donors would need to be followed for years to obtain adequate numbers of C-MDS cases. If our hypothesis that there exist clinical and molecular characteristics of C-MDS is correct, alternative methods that begin with MDS cases could then be employed. Regardless, detecting and monitoring CHIP is warranted, as this will lead to biomarkers that predict risks and thus preventative interventions.