Molecular Predictors of Response in Patients with Myeloid Neoplasms Treated with Lenalidomide

While patients with del(5q) MDS treated with Lenalidomide (LEN) have a response rate as high as 70%, the efficacy of this drug is lower in non-del(5q) cases. Aside from the presence of del(5q), up front identification of potentially responsive patients is difficult, particularly as the mechanistic underpinnings of LEN response have not been elucidated. Although expression signatures of responders were described in 2008, they have not yet been translated into an actionable diagnostic test. Analyses of molecular lesions including somatic mutations and chromosomal defects may predict response to LEN in MDS. We performed deep targeted DNA sequencing on 62 genes in 111 cases of myeloid neoplasms (MDS, MDS/MPN, and MPN) treated with Len for at least 3 months for whom fully annotated clinical outcomes were available. Clinical parameters, FISH, SNP array-based karyotyping and metaphase cytogenetics were also included in our analysis. We assessed response according to IWG 2006 criteria and performed analyses for responses at 3 or 6 months of therapy.

Of 111 LEN-treated patients, 77% had lower-risk MDS (IPSS Low /Int-1) and 23% higher-risk disease (IPSS Int-2/High/sAML). Regimens included either LEN alone (52%), or in various combinations (29%) LEN+azacytidine, TLK+LEN (1.8%) or high-dose chemotherapy (7+3)+LEN (0.9%). Any hematologic improvement, cytogenetic response, and complete response (BM) were achieved in 58%, 19% and 18% at 3 months and 84%, 44% and 30% at 6 months, respectively. Responders had better survival, with HR=0.55 (0.32, 0.94; P=.03). The mean age did not differ between responders and non-responders. Using IPSS scoring criteria, there was no difference in proportion of patients with lower-risk disease among responders and non-responders (73% vs. 81%). When IPSS-revised (-R) score was applied, there also was no significant difference between responder and non-responders with very low risk (4% vs. 7%), low risk (30% vs. 41%), intermediate risk (22% vs. 15%), high risk (29% vs. 22%), and very high risk (14% vs.12%). Refractory patients showed significantly lower platelet counts compared to responders (117 vs. 215 K/uL; P=.01). Responders tended to have higher reticulocyte counts prior to therapy compared to non-responders (0.5 vs. 0.3 M/uL; P=.07) and had significantly higher MCVs compared to refractory cases (99 vs. 91fL; P<.01).

Focusing on karyotype, there was no difference between responders and non-responders in the proportion of patients with +8, -7/del(7q), and those with normal cytogenetics. Del(20q) was marginally associated with treatment failure (6/8 failed; P=.07). In this highly selected cohort, among all del(5q) patients (N=38) 63% responded, compared to 53% in non-del(5q) (N=73), (P=.4). Among lower-risk del(5q) MDS (blast<5%) 75% (12/16) had a response vs. 50% (16/32) in lower-risk non-del(5q) MDS (P=.06). In del(5q) patients both interstitial and long del(5q) (including q11.1-q14.2 and/or q34-qter) showed similar response rates. TP53 mutations were found coinciding with del(5q) and marginally correlated with failure to respond to LEN (P=.07), but not precluded response.

Using multiplex amplicon panels of 62 genes commonly mutated in MDS, we confirmed 143 somatic mutations in responders vs. 137 mutations in non-responders. Mutations in RUNX1 correlated with LEN responses (OR=3.62 [0.63-20.87]). Mutations in DDX41 correlated with LEN response (8/8 responded; P=.009, odds ratio OR=Infinity), while mutations in U2AF1 correlated with failure to respond to LEN (1/9 responded; P=.01, OR = .08 [0, 0.66]) as did mutations in IDH1/2 (1/8 responded; P=.02, OR = .1 [0, 0.81]). DDX41 pooled with DDX54 had an odds ratio of OR=5.66 [0.58, 54.85] and DNMT3A, TET2 and IDH2pooled yielded OR=.12 [0.04-0.38]. Bayesian Model Averaging (BMA) was then applied to these and other covariates. BMA fits all submodels of a full model and then forms a weighted average of them wherein the weight of each model is the probability that it is correct relative to all other models in the model space. This yielded the linear predictor S=0.76 - 1.91•DNMT3A.TET2.IDH2 - 2.15•U2AF1 + 0.77•DDX41.54 + 0.06•del5q - 0.61•del20q-0.14•TP53.cmplx + 0.39•RUNX1 + 0.05•KDM6A that awaits validation using an independent set of patient data.

In conclusion, in addition to the presence of del(5q), the various molecular lesions including specific somatic mutations may help to better predict responses to LEN.