A Two-Gene Classifier for Chronic Myelomonocytic Leukemia (CMML) Patients Treated with Hypomethylating Agents (HMA): A Report By the GFM

Context: Mutations in epigenetic regulators such as TET2 or ASXL1 predict response but do not appear to impact overall survival (OS) in MDS patients (pts) treated with HMA. Somatic mutations are highly frequent in CMML, where ASXL1 confers independent poor prognosis (GFM score; JCO,2013;31:2428). Both azacitidine (AZA) and decitabine (DAC) are active in CMML. The prognostic value of gene mutations in CMML pts treated with HMA has not been analyzed so far.

Methods: Retrospective analysis of advanced CMML pts treated with AZA (within EMA label) or DAC (in GFM trials) in GFM centers and Dresden University. Screening for mutations was based on Sanger sequencing (JCO, 2013;31:2428). All statistical analyses were stratified on HMA.

Results: Mutation status was available for 79 pts (M/F: 57/22, median age 71y), prior to onset of AZA (n= 46) or DAC (n= 33). Median time from diagnosis to HMA onset was 8 months. At HMA onset, WHO diagnosis was CMML-1, CMML-2 and CMML having progressed to AML in 47%, 39% and 14%, respectively (resp). Pts received a median of 9 cycles for both drugs (range: 1-55). Median follow-up was 50 months.

Compared to AZA pts, DAC pts had poorer cytogenetic risk according to CMML Prognostic Scoring System (CPSS, Such Blood 2013: DAC: fav/int/poor in 49/21/30% vs AZA: 78/9/13%, p=0.009), and higher WBC (median 20.6 vs 13.4 G/L p=0.06).

The most frequently mutated genes were ASXL1 (47%), SRSF2 (47%), TET2 (40%), NRAS (20%), RUNX1 (19%), KRAS (10%) and CBL (10%). The mutation spectrum was similar in AZA and DAC groups.

Overall Response Rate (ORR) was 58% (AZA: 61%, DAC: 55%, p=NS), including CR in 26% and 15%, resp. (p=NS) according to IWG 2006 criteria. In univariate analysis, no single mutation significantly affected ORR, but KRAS-mutated pts tended to have higher ORR than KRAS-wildtype (wt) pts (88 vs 55%, p=0.08). Mutation in either KRAS or NRAS conferred significantly higher CR rate (36%, vs 14% in pts with NRAS/KRAS-wt, p=0.04).

Median OS was 27.5 months. In univariate analysis, ASXL1 (HR=2.10, p=0.008), SRSF2 (HR=2.97, p=0.008), RUNX1 (HR=2.40, p=0.01) and to a lesser extent CBL (HR=2.22, p=0.054) mutations conferred poorer OS. TET2 mutations had no prognostic impact. The higher CR rate in pts with KRAS/NRAS mutations did not translate into prolonged OS. There was a significant interaction between ASXL1 and SRSF2 status (p=0.0009). A two-gene classifier could stratify patients with significantly different 3y-OS estimates of 80.8%, 44.1%, and 23.6% in pts with neither, either, or both ASXL1/SRSF2 mutated (Figure; p<0.0001).

Other variables associated with worse OS in univariate analysis were age, WBC, peripheral immature myeloid cells (IMC), peripheral blasts, and high risk cytogenetics by CPSS. Splenomegaly, Hb and Plt levels, transfusion dependency (TD) or marrow blast % (hence WHO category) had no impact.

CPSS, which includes cytogenetics , WBC ,TD and WHO, had thus significant but relatively limited prognostic value, with worse outcome in high vs low risk pts (p=.04), but no difference between low and int-1 or int-2 pts (both p>.2).

GFM risk system which includes ASXL1 and WBC, Hb and Plt, also had prognostic value restricted to high vs low risk comparison (p=0.04).

In multivariate analysis, the two-gene classifier retained significant prognostic impact (1 mutation: HR=3.5, p=.05; 2 mutations: HR=9.4, p=.001), together with increasing IMC (HR=1.05, p=.04), independently of age, WBC, peripheral blasts and cytogenetic risk. The classifier also had prognostic value independent of CPSS (p=.001).

Median OS was 18.4 vs 32.3 months in DAC vs AZA treated pts (p=0.01), but this difference was not significant after adjusting for CPSS cytogenetic risk (p=0.19).  Heterogeneity analyses did not identify any significant interaction with the type of HMA independent of cytogenetics or WBC, suggesting no genotype predicted better outcome with one HMA over the other.

Conclusion: In this retrospective cohort of CMML pts treated with AZA or DAC, response to HMA could not be predicted by frequently mutated genes, except for a higher CR rate in RAS-mutated pts. Treatment with HMA did not abrogate the poor prognostic value of ASXL1 mutations in CMML. A two-gene classifier based on ASXL1 and SRSF2 could stratify the prognosis of CMML treated with HMA independently of clinical variables. Finally, there was no evidence that genotype should influence the choice of HMA (AZA vs DAC) in CMML.

Figure: