Genomic landscape of myelodysplastic/myeloproliferative neoplasm can predict response to hypomethylating agent therapy

Abstract

There are currently no known predictors of myelodysplastic syndrome (MDS)/myeloproliferative overlap neoplasm (MPN) patients’ response to hypomethylating agents (HMA). Forty-three patients with MDS/MPN who were treated with HMA during chronic phase and had next-generation sequencing using the established 63-genes panel were identified. Complete and partial remission and marrow response were assessed based on the MDS/MPN International Working Group response criteria. On univariate analysis, younger age, higher number of mutations, and mutations in SETBP1, RUNX1, or EZH2 were associated with no response. Multivariable analysis for modeling response were conducted via least absolute shrinkage and selection operator logistic regression approach, and showed that mutations in SETBP1, RUNX1, or EZH2 predict lack of HMA response. While limited by sample size, our findings suggest that genomic land-scape can potentially identify MDS/MPN patients with lower likelihood of response to HMA.

Introduction

Myelodysplastic syndrome (MDS)/myeloproliferative overlap neoplasms (MPN) are myeloid neoplasms, grouped as one category by World Health Organization [1]. These include chronic myelomonocytic leukemia (CMML), atypical chronic myeloid leukemia (aCML), MDS/MPN-unclassified (MDS/MPN-U), MDS/MPN-ring sideroblasts with thrombocytosis (MDS/MPN-RS-T) in adults. Despite recent advances in the understanding of MDS/MPN molecular biology leading to better prognostication based on their genomic analysis [2,3], treatment remains challenging due to their molecular complexity and sub-clonal evolution [2]. With the exception of MDS/MPN-RS-T, these are aggressive neoplasms with median survival of 1–3 years and allogeneic blood or marrow transplantation (BMT) is the only approach to achieve cure in a minority of patients [4].

Use of hypomethylating agents (HMA) spans all of the diagnosis included in MDS/MPN to modify disease progression, achieve cytoreduction and as bridge to BMT [5,6]. HMA result in meaningful responses but only in a limited proportion of patients [7,8]. The activation of kinase signaling and immune receptors upregulation has been associated with the development of resistance and lack of response to HMA in acute myeloid leukemia (AMLJ/MDS, and to a lesser degree in CMML [9,10], However, these molecular assays are not routinely performed in clinical practice, leaving an unmet need of clinically useful predictive markers of HMA response in MDS/MPN. Moreover, the discovery of these possible mechanisms of resistance has not yet led to the development of novel therapies that can increase the efficacy of HMA highlighting the need for further research of the biology of resistance to these treatments.

Targeted next-generation sequencing (NGS) is widely available and commonly used in these patients for prognostication; however, limited data are available on its role in predicting response to therapy. In this study, we evaluate the correlation of clinical features and molecular profile of MDS/MPN patients with HMA response.

Materials and methods

Patient selection

Patients with CMML, aCML and MDS/MPN-U as per WHO criteria [1] who were evaluated at our center (January 2010 to January 2020), and had NGS available before HMA initiation were analyzed. MDS/MPN-RS-T patients were not included due to distinct outcomes compared to other entities [2]. Complete and partial remission (CR and PR), and marrow response (MR) were evaluated based on MDS/MPN Internationa! Working Group response criteria [11].

Next-generation sequencing

NGS to identify was performed in patients before the initiation of any disease-modifying therapy by using an established panel of 63 genes at the Johns Hopkins Molecular Pathology Laboratory [12] (Supplementary Table 1).

Statistical analysis

Response rates were summarized via proportion estimates with 95% exact confidence interval (Cl). Univariate analysis for assessing the associations between response and the number of somatic mutations or high-risk mutations (NRAS, SETBP1, RUNX1, EZH2, TP53, ASXL1, 5TAG2), or other disease-specific factors at the time of the HMA initiation were conducted via logistic regression. Multivariable analysis for modeling response was conducted via least absolute shrinkage and selection operator (LASSO) logistic regression approach. In this approach, the candidate predictors included all available demographic, disease characteristics, and mutation data (as those variables in univariate model where discrete variables were grouped as binary). Predictors in final multivariable analysis of response were selected from LASSO approach based on 5-fold cross validation with turning parameter selected to minimize deviance of logistic regression model. Landmark analysis with landmark time selected at 6 months from HMA initiation was conducted to assess the differences of OS by CR/PR response status, where Kaplan-Meier curves of OS by response status and log-rank test were reported. Hazard ratio from Cox-regression analysis considering the occurrence of CR/PR response as a time-varying covariate was estimated to quantify the association of CR/PR response to the overall survival.

Results

Patients’ characteristics and outcomes

One hundred and fifty patients were seen at Johns Hopkins for the diagnosis of MDS/MPN. Forty-three were included in our study, who received HMA single-agent therapy and had NGS available from prior to HMA initiation. Of these 43 patients, 28 were men and median age at diagnosis was 67years (45–85 years). Median follow up was 1.6 years (91 days-5.2 years) (Table 1). Seventeen patients (39.5%) underwent BMT following HMA. The incidence of AML transformation was 16% at a median of 0.9 months from diagnosis. No patient achieved a CR, 24 patients (56%) had PR and 18 patients (42%) had MR. The median number of HMA cycles was 5.5 (range, 3–65) among patients with PR, 4 (range, 1–13) among patients without PR, 4 (range, 2–21) among patients with MR, and 5 (range, 1–21) among patients without MR. Patients who underwent BMT received a median of 4 (range, 1–65) HMA cycles, and patients who did not undergo BMT received a median of 6 (range, 2–21) cycles. Patients who achieved a PR were referred to as ‘responders’ for the purpose of the analysis, while those who did not achieve a CR/PR, were labeled as ‘non-responders’. Of the patients who underwent BMT, 11/17 (64.7%) had a PR, 6 had stable disease, and 7/17 (41.2%) had MR before BMT.

Table 1.

Patients’ characteristics at the initiation of HMA.

Characteristics	Number (%)
Age median (range)	67 (45 – 85)
Diagnosis
MDS/MPN-U)	15 (34.9)
CMML	25 (58.1)
aCML	3(7)
Gender
Women	15 (34.9)
Men	28 (65.1)
Treatment
Azacitidine	34 (79.1)
Decitabine	9 (20.9)
Median number of HMA cydes, range	4 (1 – 65)
Previous treatments
None	27 (62.8)
Hydroxyurea	9 (20.9)
Targeted therapy	4(93)
Chemotherapy	1 (23)
AlloBMT	2 (4.7)
Blasts percentage (%) median, range	2 (1–18)
Karyotype
Very good	1 (23)
Good	32 (74.3)
Intermediate	3(7)
Poor	3(7)
Very poor	2 (4.7)
Missing	2 (4.7)
R-IPSS, median, range	3(0–7)

MDS: myelodysplastic syndrome; MPN-U: myeloproliferative overlap neoplasms unclassified; CMML: chronic myelomonocytic: leukemia; aCML atypical chronic myeloid leukemia; HMA; hypomethylating agent; alloBMT: allogeneic blood or marrow transplantation; R-IPSS: Revised International Prognostic Scoring System.

The presence of at least one mutation in any of the SETBP1, RUNX1, or EZH2 genes is associated with worse response to HMA independent of other clinical and molecular characteristics

The genomic landscape of individual patients, as non-responders and responders, is shown in Figure 1. Overall, ASXL1 mutation was reported in 58% of patients. The frequency of mutations in different genes in responders and nonresponders is depicted in Figure 2 and SETBP1, RUNX1, EZH2, and STAG2 mutations appear notably more frequent among nonresponders (Figure 2). Since SETBP1, RUNX1, and EZH2 were the three genes most prominently different between responders and nonresponders, we combined them as a variable for the univariate and multivariate analysis. On univariate analysis for PR, the absence of response was significantly associated with the presence of ≥2 hig-hrisk mutations (OR 0.19, 95% Cl: 0.05–0.67), SETBP1 mutation (OR 0.16, 95%: Cl 0.02–0.76), RUNX1 mutation (OR 0.1, 95% Cl: 0.01–0.48), and a mutation in at least one out of the SETBP1, RUNX1, and EZH2 genes (OR 0.05, 95% Cl: 0.01–0.21). Details of univariate analysis are shown in Supplementary Table 2. Older age as a continuous variable (OR 2.41, 95% Cl: 1.08–5.35, per decade) was associated with a higher likelihood of response. As for MR analysis (Supplementary Table 3), ≥2 high-risk mutations (OR 0.23, 95% Cl: 0.05–0.9), and the presence of a mutation in one of the SETBP1, RUNX1, and EZH2 genes (OR 0.16, 95% Cl: 0.03–0.62) were associated with the absence of MR on univariate analysis. Older age, as a continuous variable (OR 3.17, 95% Cl: 1.27–7.91, per decade), was associated with higher likelihood of MR. A forest plot of the univariate analysis is shown in Figure 3.

Figure 1.

The genomic landscape with the combination of mutations identified in MDS/MPN patients without partial remission (nonresponders) and MDS/MPN patients with partial remission (responders).

Figure 2.

The frequency of mutations in different genes in responders and nonresponders: nonresponders present significantly higher frequency of RUNX1 and SETBP1 mutations.

Figure 3.

Forest plot of the univariate analysis for the association of response with a number of variables. *Missingness in the factor.

Next, we performed multivariable analysis for modeling response to identify independent predictors of response to HMA. Among all the predictor candidates from LASSO approach, the presence of a mutation in one of the SETBP1, RUNX1, and EZH2 genes was selected along with patients’ age based on the results from univariate analysis. In multivariable analysis, the presence of a mutation in one of the SETBP1, RUNX1, and EZH2 genes was associated with the absence or response (OR 0.05, 95% Cl: 0.01–0.27 for PR and OR 0.19, 95% Cl: 0.04–0.91 for MR, respectively) (Table 2).

Table 2.

Generalized linear model results with selected predictors identified from LASSO.

Variable	Partial remission			Marrow response
	OR	95% Cl	p value	OR	95% Cl	p value
SETBP 1/RUNX 1/EZH2 mutation
No	1			1
Yes	0.05	0.01–0.27	<0.001	0.19	0.04–0.91	0.037
Age at treatment (per 10y)	2.42	0.88–6.64	0.087	2.75	1.05–7.18	0.039

Altogether, these results support that the presence of a mutation in any of the SETBP1, RUNX1, or EZH2 genes is associated with worse response to HMA among MDS/MPN patients independent of other clinical and molecular characteristics.

Worse response to HMA by 6 months is associated with worse overall survival

Finally, we performed Kaplan-Meier analysis to evaluate if worse response to HMA is associated with different overall survival in this MDS/MPN cohort. In the landmark analysis, responders at 6 months had superior overall survival compared to nonresponders (median 18 vs 56 months, p = 0.010, Figure 4). Cox-regression analysis also revealed that responders had superior overall survival compared to nonresponders (HR 0.26, 95% CI: 0.9–0.13, p = 0,010).

Figure 4.

Kaplan-Meier estimates of OS by PR status at 6 months in landmark analysis, p value was based on the log-rank test. OR and the corresponding p values were estimated from logistic regression models.

Discussion

HMA remain the most commonly used agents in MDS/MPN patients [7] with a response rate usually below 50% in the majority of studied cohorts [13–15]. The analysis of the genomic profile of these neoplasms has improved their classification and has added important prognostic information [2,16]. However, its role in predicting the response of MDS/MPN patients to HMA remains limited in clinical practice.

Prior reports suggest a lower likelihood of HMA response in CMML in the presence of ASXL1, RUNX1, and CBL mutations [17]. Since HMA is the common modality of treatment for all aggressive MDS/MPN, we expand this analysis to the overall MDS/MPN category. Genomic analysis of CD34+ cells from MDS and CMML patients revealed that azacitidine responders have higher percentage of their progenitor cells in cell cycle, while progenitors from nonresponders exhibit features of quiescence [18]. Similarly, azacitidine induces immune signaling and suppresses metabolic pathways in AML cells [19]. On the contrary, it has been more recently reported that resistance to azacitidine in AML cell lines is associated with deregulation of several cancer-related pathways such as the phosphatidylinosito-3 kinase signaling [10]. These results support that specific molecular alterations may promote resistance to HMAs in AML/MDS but remain unexplored in MDS/MPN. Furthermore, these have not been translated to clinically useful predictive markers for everyday practice.

SETPB1 mutations occur in 9% of MDS/MPN patients, with a higher incidence among patients with aCML and are associated with adverse karyotypic abnormalities and worse outcomes [20]. SETBP1 mutations are also related to worse azacitidine response, increased incidence of transformation to AML [21], and poorer survival among high-risk MDS patients [22], It has been recently reported that SETBP1 mutations lead to the upregulation of mitogen-activated protein kinase signaling and the downregulation of differentiation pathways [23] suggesting a possible rationale of its negative impact in the response to HMA. Further understanding of the molecular implications of these mutations is required to shed light into this hypothesis and evaluate the role of MEK inhibitors in combination with HMA for patients with MDS/MPN.

Mutations in RUNX1 gene are identified in 15% of patients with CMML and 10–15% of patients with MDS/MPN-U and aCML [24]. They are associated with worse survival and higher incidence of AML transformation among MDS/MPN patients [16]. Of note, the co-occurrence of RUNX1 and ASXL1 mutations have been associated with worse response to HMA and shorter survival among MDS patients [25]. These findings are consistent with our results highlighting the adverse effect of mutations in RUNX1 among MDS/MPN patients particularly under HMA therapy.

EZH2 gene is mutated in 10–13% of patients with MDS/MPN and these mutations frequently co-occur with ASXL1 mutations [26]. Interestingly, EZH2 mutations tend to be mostly early events in leukemogenesis and show a complex mutational hierarchy [26]. Our group has recently highlighted the male predilection of EZH2 mutations and their negative impact in the outcomes of MDS/MPN patients [12]. Despite that demethylation of EZH2 has been identified as a biomarker of response to azacitidine [27] it remains unclear if EZH2 mutations alter the sensitivity of patients with chronic myeloid neoplasms to HMA. Our data support that these mutations have probably a negative impact on the response of MDS/MPN patients to HMA but further study is required for the confirmation of these associations.

Our study further underscores the role of molecular profiling in MDS/MPN, which provides not only prognostic information [2,12] but can also be predictive of HMA response. Using a multivariable prediction risk assessment tool, we describe the presence of a mutation in any of the SETBP1, RUNX1, and EZH2 genes as a predictor of poor response to HMA as single agents. Our study was limited with a small sample size and the number of events for a rare, yet consequential diagnosis of MDS/MPN. A larger cohort study is usually difficult to conduct in rare cancers including MDS/MPN, but can possibly clarify some of the statistical uncertainty posed by a limited sample size. Mutations in ASXL1, are universally associated with high risk and poor prognosis [28], was unexpectedly not associated with differences in response in our study. That is possibly explained by the usual ancestral origin of ASXL1 while the sub-clones further dictate probability of response to HMA [2], Nevertheless, these results emphasize the limitations of HMA as single agent and the urgent need to develop novel treatment strategies in MDS/MPN especially in the presence of hig-hrisk mutations.

In conclusion, the analysis of 43 individuals with MDS/MPN revealed that the presence of at least one mutation in SETBP1, RUNX1, or EZH2 gene is associated with worse response to HMA independent of other clinical and molecular features. Confirmation of these results in bigger cohorts is required to strengthen these conclusions and potentially evaluate their utilization for better prognostication in clinical practice.