Introduction

Myelodysplastic syndromes (MDS) are a heterogeneous group of acquired clonal hematopoietic stem cell disorders characterized by ineffective hematopoiesis, dysplasia of hematopoietic cells, recurrent genetic abnormalities and an inherent risk for transformation to acute myeloid leukemia1. Clinical outcomes of patients with MDS are variable due to the multiple genetic and epigenetic alterations involving different cellular pathways that affect hematopoietic cell survival, growth, proliferation and apoptosis2,3,4,5. Large-scale high-throughput sequencing studies have shown substantial complexity in the genetic landscape of MDS4,6. Nearly 90% of MDS cases have recurrent somatic gene mutations involving the core components of one or more pathways: RNA splicing machinery (SF3B1, SRSF2, U2AF1), DNA epigenetic regulation (TET2, DNMT3A, IDH1/2); chromatin modification (ASXL1, EZH2), tyrosine kinase signaling (RUNX1, RAS, RAF) and DNA repair response (TP53)3,7,8. Somatic mutations involving genes RUNX1, SRSF2, EZH2, ETV6, TP53 and ASXL1 are independent predictors of poorer outcome in MDS patients3,7,9,10.

Despite these advances, it has been challenging to incorporate gene mutation data formally into any of the current prognostic scoring systems for risk-stratification. This is due to the high degree of complexity and permutations of a multiple variables that affect the outcome, such as combinations of co-mutations, sequence of co-mutations, variant allele frequencies (VAF) as well as the presence of additional alterations at the level of gene copy number, mRNA and protein that lead to similar downstream consequence2,6,11. For accurate prognostication and appropriate use of targeted therapies, there is not only a need for mutation detection but also characterization of multitude of alterations leading to similar clinicopathologic features, and to apply this knowledge for novel “individualized” therapeutics.

Enhancer of zeste 2 homologue 2 (EZH2) is a gene that encodes a histone methyltransferase, which is an enzymatic component of the polycomb repressive complex 2 (PRC2) complex, that is crucial for epigenetic silencing of genes involved in stem cell renewal12,13. EZH2 protein suppresses the transcription of other genes and is an essential regulator of hematopoietic stem cells12,13,14. Unlike germinal center derived B-cell lymphomas where most cases show an EZH2 Y641 hot-spot gain-of-function mutation, MDS cases show a spectrum of mutations with a loss-of-function phenotype. Multiple studies have shown that EZH2 mutations are associated with a poorer prognosis in MDS3,7,15. The presence of EZH2 mutations has been shown to induce therapeutic resistance to cytotoxic drugs in AML16.

EZH2 gene mutation is only one of several mechanisms that can alter its function. EZH2 function can be disrupted by copy number changes, mutations in the non-coding region, and mis-splicing at the mRNA level caused by other gene mutations, all of which lead to the downregulation of protein. Assessment of EZH2 protein expression can identify loss of downstream EZH2 function resulting from multiple alterations that may be missed by mutation assessment alone. Targeting EZH2 is more challenging in myeloid malignancies due to the loss-of-function phenotype, but multiple therapeutic strategies to restore EZH2 function, such as targeting HOXA9 and other genes derepressed by EZH2, small molecules to inhibit demethylases to modify methylation signatures and addition of bortezomib to cytarabine-based therapies, are currently under exploration16,17,18,19. Since most of these approaches involve restoration of EZH2 function, their application is not restricted to mutations alone, but extend to all mechanisms that lead to decreased EZH2 protein expression. Hence, there is a need for precise evaluation of EZH2 mutations as well as protein expression by immunohistochemistry, which is simple and feasible in clinical laboratories, in well-characterized clinical MDS cohorts.

In this study, we evaluated the spectrum of EZH2 mutations and clinicopathologic features and outcomes. In addition, we correlated the mutation findings with EZH2 protein expression using immunohistochemistry. We show that lack of EZH2 expression correlates with the presence of an EZH2 mutation in 69% of patients, while a subset (23%) of patients, in the absence of EZH2 mutation or copy number loss, also showed loss of EZH2 expression. Both EZH2 mutations and loss-of-expression correlated with poor overall survival, independent of R-IPSS, age and gender, with loss-of-expression showing a stronger correlation than mutation. We conclude that EZH2 protein assessment by immunohistochemistry, alongside mutation analysis, is important for prognostic workup of MDS.

Materials and methods

Case selection

We performed an electronic search of our departmental archives for newly diagnosed MDS patients with available next-generation-sequencing (NGS) mutation data. The clinical and laboratory data were collected from patients’ electronic medical records with emphasis on variables that are historically known to be more important in myeloid neoplasms. This study was approved by the Institutional Review Board (IRB) and performed in accord with the Declaration of Helsinki.

Histologic evaluation

The diagnosis of MDS was confirmed based on morphologic review of bone marrow (BM) aspirate and peripheral blood smears and routinely prepared histologic sections of biopsy and/or clot specimens in all patients. For controls, 34 consecutive MDS cases with wild-type EZH2 and confirmed histologic diagnosis were included. Cases were further subcategorized into different subtypes based on criteria established by the current WHO classification1.

Immunohistochemistry

Immunohistochemical analysis for EZH2 was performed using anti‐EZH2 antibody (clone 6A10; Novocastra CAT#NCL-L-EZH2) on 4‐μm sections of decalcified formalin-fixed paraffin embedded BM biopsies using a streptavidin-biotin complex technique. Slides were first deparaffinized and rehydrated followed by heat-induced antigen retrieval using 0.01 M citrate buffer at pH 6.0 in a microwave oven. The slides were stained on a Dako Autostainer (Dako, Carpinteria, CA, USA) with EnVision+ (Dako) staining reagents. After blocking the endogenous peroxidase activity (Dual Endogenous Block, Dako for 10 minutes) and buffer wash, the sections were incubated with anti‐EZH2 antibody (clone 6A10; Novocastra CAT#NCL-L-EZH2) at a dilution of 1:5000 for 60 min. Following a buffer wash, the sections were incubated with the EnVision+ Dual Link (Dako/Agilent) detection reagent for 30 min and subsequently treated with a solution of diaminobenzidine (DAB) and hydrogen peroxide (10 min) to produce the visible brown pigment. After rinsing, the color was enhanced with DAB enhancer (Dako) and counterstained with haematoxylin, dehydrated and cover slipped with a permanent medium.

Nuclear EZH2 staining was scored independently by three hematopathologists (A.S., C.B.R and R.K-S.) by using a semi-quantitative approach (blinded to data). The mutation data were unknown during scoring. The percentage of positive BM cells was scored as follows: 0 (no staining, 0%), 1 (1–5%), 2 (6–20%), 3 (21–50%), and 4 (51–100%). The intensity of staining was scored qualitatively from 0 to 3. A multiplicative staining score (H-index) was obtained by multiplying the % positive cells and staining intensity, giving a range of 0–12. H-index of 0–1 was considered negative (EZH2NONEXP) implying “loss of expression”; ≥2 was considered expressors (EZH2EXP).

Cytogenetic studies

Conventional cytogenetic analysis was performed on unstimulated cultured BM aspirate specimens using standard GTG-banding as described previously20. At least 20 metaphases were analyzed. Results were reported using the 2016 International System for Human Cytogenetic Nomenclature (ISCN)21. Complex karyotype is defined as having at least three structural and/or numerical chromosomal abnormalities.

Next-generation sequencing

NGS-based mutation analysis was performed using 28-gene or 81-gene panels at CLIA-certified molecular laboratory as previously described22,23,24. The complete gene lists provided in Supplemental Table S1. Briefly, sequencing libraries were prepared from 250 ng of gDNA followed by amplicon-based targeted NGS (llumina Miseq, San Diego CA, USA). Variant calling required at least 250x bidirectional coverage and 2% variant allele frequency (lower limit of detection; reference genome: GRCh37/hg19). Somatic nature was inferred from literature and online databases. SNPs in ExAC, dbSNP 137/138, and 1000 Genomes were excluded.

Statistical analysis

Overall survival (OS) was defined as the time from diagnosis to death/last follow-up. Patients alive at their last follow-up were censored. The median OS was evaluated by the Kaplan-Meier method. Univariate Cox proportional hazards analyses were used to identify associations between risk factors and survival followed by multivariate Cox analyses. Logistic regression (for continuous) and Fisher’s exact test (for categorical variables) were used to assess associations. Statistical analysis was performed using R version 3.5.1.

Results

Study group characteristics

Forty (n = 40) patients with MDS with a confirmed somatic mutation in EZH2 gene (EZH2MUT) were identified from a total of 454 newly diagnosed MDS patients (40/454, 8.8%) in the archives of the MD Anderson Cancer Center between years January 1, 2014 and December 31, 2017. There were 33 men and seven women with a median age of 74 years (range, 55–90) at diagnosis. The median BM blast percentage was 4% (range, 0–16%); 24 (60%) cases presented with <5% blasts. Within the sub-categories of the WHO 2016 classification, EZH2 mutations were preferentially observed in MDS with multilineage dysplasia (MDS-MLD, n = 16, 40%), followed by MDS with excess blasts-1 (MDS-EB-1, n = 9, 23%), MDS with excess blasts-2 (MDS-EB-2, n = 6, 15%), therapy-related MDS (t-MDS, n = 3, 8%) and MDS with single lineage dysplasia (MDS-SLD, n = 2, 5%), MDS with MLD and ring sideroblasts (MDS-MLD-RS, n = 2, 5%) and MDS with single lineage dysplasia and ring sideroblasts (MDS-SLD-RS, n = 2, 5%). All three cases of t-MDS had multilineage dysplasia. Using IPSS-R criteria for cytopenias, anemia was most common (n = 31, 78%) followed by neutropenia (n = 27, 68%) and thrombocytopenia (n = 26, 78%). Pancytopenia was present in 18 (45%) patients followed by bi-cytopenia in 10 (25%) patients. Twelve (30%) patients had monocytopenia.

Of 39 patients with available karyotype by conventional cytogenetic analysis, 12 (31%) were diploid, 19 (49%) showed a non-complex karyotype, and eight (20%) revealed a complex karyotype (defined here as three or more chromosomal abnormalities). Fourteen (36%) cases showed concurrent abnormality(ies) in chromosome 7 alone or in combination with other changes; a subset of these were confirmed by FISH. These cases included 11 (27.5%) with deletion of whole chromosome 7 (−7; n = 10) or long arm of chromosome 7 (−7q; n = 1). Three cases showed other abnormalities of chromosome 7 that did not lead to deletion of the EZH2 locus at band 7q36.1 locus including; r(7)(p12q11.2); r(7)(p11.2q22); and d(1.7)(q10;p10).

A control cohort of 38 consecutive newly diagnosed treatment naïve MDS patients with wild-type EZH2 (MDSEZH2WT) over a 5-month time period (April 1, 2017 to August 31, 2017) who underwent BM exam was selected. There was no difference in the distribution of BM blast percentage or R-IPSS scores. Compared to EZH2 wild-type MDS, EZH2 mutated MDS showed male predominance (54% vs. 83%; p = 0.014) and frequent loss in chromosome 7 (8% vs. 36%; p = 0.001). There were no appreciable differences with respect to the types of morphologic dysplasia or lineages involved by dysplasia between EZH2 mutated and EZH2 wild-type MDS. The clinicopathologic characteristics of MDS patients with and without EZH2 mutations are summarized in Table 1.

Table 1 Baseline clinicopathologic characteristics of patients with myelodysplastic syndrome with and without EZH2 mutations.
Full size table

EZH2 gene mutation characteristics

Using NGS myeloid panel that covers the entire coding region (exons 2–20) of EZH2, we identified the study cohort of 40 patients with 45 EZH2 mutations: 35 patients with a single mutation and five patients with double mutations. Of the mutations, 28 (62%) were missense, five (11%) were nonsense and 12 (27%) were frameshift types. EZH2 mutations spanned the entire coding region: 16 (36%) involved exons 18 and 19 (Fig. 1A). Nearly half (n = 21, 47%) mutations were within the catalytic “SET” domain. Based on the support from evidence in the literature and online databases, all of these mutations are considered to be somatic.

Fig. 1: EZH2 mutation characteristics of this cohort.
figure 1

A Distribution of mutations along different the coding region of EZH2. B Highlights the distribution of EZH2 mutations in different functionally important sections of the gene. Mostly mutations occur in the catalytic “SET” domain. C Mutational heatmap in the EZH2 mutated (study) vs. EZH2 wild-type (control) cohorts showing epigenetic modifier and related genes: DNA methylation regulators: DNMT3A, TET2, and IDH2, histone modifiers: ASXL1 and EZH2 (including the core components of PRC2 pathway: EED and SUZ12) and RUNX1. D Clonal relationship between EZH2 and additional mutations in RUNX1 (sub-clonal), ASXL1 (sub-clonal and co-dominant) and TET2 (co-dominant) based on variant allele frequencies gene exons.

Full size image

Among EZH2 mutated MDS, 33 (83%) cases harbored at least one additional mutation (Fig. 1B). The most common concurrently mutated genes in the decreasing order of frequency were: ASXL1 (n = 22, 55%), RUNX1 (n = 17, 43%), TET2 (n = 15, 38%), SF3B1 (25%), TP53 (14%), and DNMT3A (10%) (Table 2). EZH2 mutated MDS had a significantly higher frequency of concurrent mutations in ASXL1 (55% vs. 26%; p = 0.012) whereas IDH2 mutations were absent (0 vs. 11%; p = 0.05). The mutational frequencies of RUNX1 and TET2 were higher but not significantly different from wild-type. Sixteen (40%) patients showed mutations in at least two of three genes: ASXL1, RUNX1 or TET2, including seven patients with all three genes mutated in addition to EZH2. Of note, many patients showed a second sub-clonal mutation within the same gene (double mutations), including EZH2 (n = 6), RUNX1 (n = 4), TET2 (n = 4), and ASXL1 (n = 2). Figure 1C illustrates the spectrum of concomitant epigenetic modifier gene mutations in EZH2 mutated and wild-type groups: DNA methylation regulators (DNMT3A, TET2, IDH1, IDH2), histone modifiers (ASXL1) and genes in PRC2 complex (EED, SUZ12).

Table 2 Multivariate analysis showing that the presence of EZH2 mutation and EZH2 loss of protein correlated with poor survival, independent of R-IPSS.
Full size table

The median VAF of EZH2 mutation was 30% (range, 1.3–92). Twenty-one (53%) were greater than the median suggesting that EZH2 was a dominant mutant clone. Eleven (24%) mutations showed a VAF of >60%, suggesting either a homozygous/ biallelic EZH2 mutation or a concurrent loss of heterozygosity due to chromosome 7 alterations. Twelve (27%) mutations were a minor clone (VAF < 20%), although, in two of these cases the percentage of non-myeloid cells (lymphocytes and plasma cells) was high (48% and 49%, respectively) due to hemodilution. These findings suggest that EZH2 mutation is a dominant mutant clone in MDS and likely to be a primary driver. By comparing the VAFs of EZH2 with co-mutated genes, we determined that TET2 mutations were frequently co-dominant (n = 11, 85%) and that RUNX1 mutations were frequently sub-clonal [n = 10, 77%]. ASXL1 mutations were both sub-clonal [n = 8, 47%] and as co-dominant mutations [n = 9, 53%] (Fig. 1D).

Immunohistochemical EZH2 protein expression in EZH2 mutated and wild-type MDS

For immunohistochemical evaluation, we evaluated EZH2 protein expression using decalcified BM sections from MDS with mutated and wild-type EZH2 using a semi-quantitative staining score described under methods. In order to assess the baseline expression of EZH2, we first evaluated EZH2 expression in 11 apparently healthy individuals with no prior/ concurrent diagnosis of malignancy or clonal hematopoiesis (negative for mutations using the same comprehensive NGS panel). EZH2 showed nuclear positivity in most erythroid and myeloid precursors and a subset of megakaryocytes; within myeloid precursors, maturation was associated with loss of EZH2 expression, with segmented neutrophils showing virtually absent protein. This was established as the normal staining pattern. The median H-score was 9 (range, 6–12).

Next, we scored the nuclear EZH2 expression in 29 MDS with EZH2MUT (those who had adequate quality decalcified biopsy cores) and 34 MDS with wild-type EZH2. Twenty of 29 (69%) EZH2MUT MDS showed loss of EZH2 protein (EZH2NONEXP) whereas 9 (31%) retained protein expression (EZH2EXP). Among EZH2-wild-type MDS, 9 (27%) cases showed loss and 25 (74%) cases retained protein expression. Based on these data, EZH2 expression was lost more frequently in EZH2MUT MDS compared to EZH2WT (p = 0.001), and hence correlated with absence of EZH2 mutation. Representative images of different patterns of staining in all three groups are shown in Fig. 2.

Fig. 2: Representative examples of EZH2 protein expression using immunohistochemistry.
figure 2

A EZH2 stain on bone marrow biopsy of apparently healthy individual showing nuclear positivity in erythroid and myeloid precursors, and megakaryocytes. B At higher magnification, EZH2 protein is lost with maturation of myeloid precursors: segmented neutrophils show virtually absent protein expression. C EZH2 mutated MDS with loss of EZH2 protein (EZH2 deficient). D EZH2 wild-type MDS with retained EZH2 protein (EZH2 expressor MDS). E EZH2 wild-type MDS with loss of EZH2 protein (EZH2 deficient). F Violin plot showing significantly decreased EZH2 protein expression (measured by H-score) in EZH2 mutated MDS compared to wild-type.

Full size image

We were particularly interested in the nine patients with EZH2WT MDS that showed loss of EZH2 protein expression (EZH2NONEXP). These included two patients with deletions of chromosome 7 [monosomy 7, and del(7q31)] and five patients who had concurrent mutations in other related genes: four with coexisting SRSF2 mutations and one patient with mutations in ASXL1, U2AF1, and KMT2A. In the latter five patients, mutations in SRSF2 and U2AF1 could have possibly led to mis-splicing of wild-type EZH2 and/or EZH2 protein disruption. The remaining two patients did not show any other concurrent genomic alterations leading to protein loss. On the contrary, nine patients with EZH2MUT MDS had preserved EZH2 expression (EZH2EXP), six had mutations located in the SET domain and two had loss of entire chromosome 7. The negative predictive value of a preserved EZH2 expression for lack of EZH2 mutation was 76% (Fisher’s exact test).

Due to differences in EZH2 mutation vs. EZH2 expression, we also compared the differences in clinicopathologic features and outcomes between EZH2EXP and EZH2NONEXP MDS, in addition to EZH2MUT/EZH2WT subgroups. EZH2NONEXP MDS showed frequent loss of chromosome 7 (35.7% vs. 5.9%, p = 0.004) compared to EZH2EXP MDS, similar to EZH2MUT MDS. In addition, patients with EZH2NONEXP MDS showed significantly lower median platelet count (68 vs. 112, p = 0.041) and a trend for higher R-IPSS scores and BM blast% compared with EZH2EXP MDS. There were no significant differences in the types of morphologic dysplasia or lineages involved by dysplasia between EZH2 non-expressors and expressors. There were no significant differences in mutations of other genes between EZH2 non-expressors and expressors.

Correlation of EZH2 mutation and EZH2 protein loss with survival

Over the course of the study, 34 (85%) patients received therapy with hypomethylating agent(s). Over a median follow-up of 14 months (range, 1–45), 17 (42.5%) patients died and 12 (30%) patients transformed to acute myeloid leukemia (AML).

EZH2MUT MDS patients had a significantly worse overall survival (OS) compared to EZH2WT [20 months vs. not reached (NR), HR 2.953 (1.352–6.449), p = 0.0066] (Fig. 3A). EZH2NONEXP MDS patients (with loss or deficient EZH2 expression), irrespective of the EZH2 mutation status, had a significantly worse overall survival compared to EZH2EXP MDS patients (NR vs. 20 months, HR 5.231 (2.205–12.41), p = 0.0002) (Fig. 3B). There was no significant survival difference between MDS patients with EZH2 mutations located in the catalytic “SET” domain compared to those with EZH2 mutations located in other domains. (Supplemental Fig. S1A). There was no significant survival difference in EZH2 mutated MDS patients between those with or without deletion of chromosome 7q36 locus (Supplemental Fig. S1B).

Fig. 3: Survival analysis.
figure 3

A comparison of overall survival (OS) between MDS patients segregated by (A) EZH2 mutation status and (B) EZH2 protein expression status. EZH2 protein expression status was able to segregate these patients better than mutation status.

Full size image

By multivariate analysis that also included age, gender and R-IPSS, both EZH2 mutation (p = 0.027) and loss of protein expression (0.0063) correlated with poor survival, independent of R-IPSS [Table 2. When mutation and protein expression were analyzed together, protein expression [HR 0.30 95% CI: 0.085–1.057; p = 0.061) showed a stronger correlation with survival than mutation (HR: 1.66 95% CI: 0.47–5.82; p = 0.43).

EZH2 immunohistochemistry in MDS at the time of AML transformation

To explore the value of immunohistochemical assessment of EZH2 protein expression in MDS cases transformed to AML (AML-MRC), we performed EZH2 immunohistochemical staining on six cases of EZH2MUT MDS and four cases of EZH2WT MDS, at the time of MDS diagnosis and AML transformation. All cases had the same EZH2 mutations detected at both time points. The median blast count was 35% (range, 21–89%). Two EZH2MUT patients had concurrent del(7q23). We also included three cases of EZH2MUT de novo AML without a history of MDS (but meeting the criteria for AML-MRC based on dysplastic morphology in >50% of cells in at least two lineages).

Among six EZH2MUT MDS cases that transformed to AML, four (67%) showed loss-of-expression (EZH2NONEXP) whereas two (33%) retained protein expression (EZH2EXP). The immunohistochemical EZH2 protein expression results were similar at the time of MDS and AML in all cases. The two cases with discordant EZH2EXP included the following: one patient with 89% blasts and mutations in EZH2 p.N693K (VAF 30%) and NRAS (VAF 79%); second patient with 21% blasts with multiple mutations involving EZH2 p.G628A (VAF 33%), ASXL1 p.G644fs (32%), RUNX1 p.R320* (28%), TET2 p.Q916* (35%) and TET2 p.S1671fs (31%). All four EZH2WT cases retained protein expression (EZH2EXP) at the time of MDS and AML. All three cases of EZH2MUT de novo AML showed loss-of-expression. The details are provided in Supplemental Table S2. These preliminary findings suggest that data could be extrapolated to AML cases. The findings need to be confirmed in a larger number of AML cases and the clinical significance needs to be evaluated within the context of prognostic attributes pertinent to AML, which is beyond the scope of the current study.

Discussion

EZH2 encodes a histone methyltransferase with an important role in chromatin modification and epigenetic changes12,13. The frequency of EZH2 mutation in newly diagnosed MDS in this study was ~5%, within the reported ranges from other studies (range, 2–13%)3,4,13. EZH2 mutations in MDS patients included missense, nonsense and frameshift subtypes distributed throughout the entire coding region without a hot-spot location. A third of the mutations clustered in exons 18 and 19 within the SET domain, the main catalytic site of the methyl transferase enzyme, in accordance with other studies12,25. This is in contrast to EZH2 mutations in low-grade B-cell lymphomas and diffuse large B-cell lymphomas, which are often a gain-of-function mutation in tyrosine 641 (Y641), also located in SET domain26,27,28. Others have shown that EZH2 mutation is a poor prognostic biomarker, which we confirm in this study. Therefore, knowledge of the EZH2 mutation spectrum is important to design clinical NGS panels that include the whole coding region for workup of myeloid neoplasms.

The observed distribution of EZH2 mutations, with a substantial proportion of nonsense and frameshift mutations, together with a strong association with concurrent disruption of the other EZH2 allele, frequently by chromosomal changes (monosomy or segmental deletions), supports the loss-of-function phenotype for EZH2 alterations in MDS and other myeloid malignancies29,30. This statement is further corroborated by this study showing an association of EZH2 mutations with loss of protein expression in MDS. Frequent loss of chromosome 7 or del(7q) also points to EZH2 as a tumor suppressor gene13,31, underscoring the importance of additional testing, beyond NGS-based mutational analysis, to look for these concurrent alterations in other EZH2 allele using copy number analysis such as single nucleotide polymorphism (SNP) arrays to identify cytogenetically cryptic aberrations and/or copy-neutral loss of heterozygosity.

About 90% of cases of EZH2 mutated MDS had concurrent mutations, most frequently of another chromatin modifier ASXL1 (~45%). Based on the comparison of mutant allele burdens of EZH2 and ASXL1 genes, ASXL1 appears to be a secondary sub-clone following an EZH2 initiating event4,32,33. In contrast, TET2 mutation consistently showed a parallel clonal size with a similar or higher allelic frequency than EZH2. Six (15%) patients showed an additional mutation in TP53, half of which had abnormalities of chromosome 5 as previously recognized34. These interesting clonal relationships based on bulk NGS need to be confirmed using single-cell mutation studies. Nevertheless, MDS patients with both EZH2 and TET2 mutations had a worse survival compared to MDS patients without TET2 mutations, supporting the synergistic effects of this combinatorial mutational change as suggested by others35,36,37. We found a similar prognostic effect of additional RUNX1 mutations, but not with ASXL1 or TP53 mutations (data not shown). This result may reflect EZH2 and ASXL1 being part of the same chromatin modification system of polycomb repressive complex 2 (PRC2) and PRC1, respectively, both of which directly or indirectly result in methylation of lysine 27 on histone H3 of DNA30,38,39,40. Loss of EZH2 promotes MDS by expansion of MDS-initiation cells in RUNX1 S291fs mutated mice15.

In addition to mutations in the coding region, loss of EZH2 protein function could be disrupted by multiple upstream alterations that are beyond the detection capacity of targeted NGS± copy number analysis, such as mutations in non-coding intronic or splice-site locations, mis-splicing of mRNA due to mutations in other splicing factor gene mutations among others. Bulk RNA sequencing of CD34+ cells showed EZH2 mRNA downregulation in MDS patients even without monosomy 7/del(7q), suggesting alternate mechanisms of downregulation of these genes41,42. Hence, quantification of EZH2 protein expression can be a surrogate measure of EZH2 function. In this study, we evaluated the functional disruption of EZH2 by immunohistochemistry. Due to loss-of-function phenotype of EZH2 mutations in MDS, the expression levels correlated with mutation status in 73% of MDS patients, unlike diffuse large B cell lymphoma cases, where EZH2 expression was independent of mutation status43,44. The possible reasons for the incomplete correlation in the remaining (27%) MDS patients are elaborated below.

Among the 9 EZH2WT MDS cases with loss of EZH2 protein expression, two patients had deletion of 7q36.1 locus which might have led a dose dependent decrease in EZH2 expression45. EZH2 overlaps one of the three commonly deleted regions (CDRs) in MDS with del(7q)41,42. Four patients had concurrent mutations in SRSF2 which causes mis-splicing of EZH2 mRNA leading to nonsense mediated decay and consequently lower protein levels46. One additional patient had mutation in U2AF1, another splicing factor gene that might lead to similar effect. Four patients were co-mutated with ASXL1, also a component of PRC1/2 complex, leading to reduction of EZH2 expression. In the remaining two EZH2WT MDS patients, the reasons for EZH2 protein loss are unclear. There was no EZH2 deletion by karyotyping or copy number evaluation by targeted NGS. Copy-neutral loss-of-heterozygosity of 7q was not assessed but is rare in MDS. Cabrero et al. have shown that ~50% of EZH2WT MDS with diploid karyotype had downregulation of EZH2 mRNA in CD34+ cells29. There is a possibility of mutations in other major proteins of the PRC2 complex that are closely related with EZH247. Only two genes [embryonic ectoderm development (EED)1 and suppressor of zeste (SUZ)12] were included in the panel and tested wild-type. The status of the other components of PCR2, such as JARID2, AEBP2, EZH1 etc are unknown48.

The reasons for preserved EZH2 expression in nine EZH2MUT MDS cases include: first, the possibility of residual normal/non-neoplastic clones in the bone marrow. Second, we cannot completely exclude the possibility of these EZH2 variants being rare germline polymorphisms, although these specific variants have been shown to be somatic alterations in the literature and COSMIC databases. These variants were not identified in gNOMAD or EXAC databases. Eight of these patients had disruption of SET domain with median VAF of 68% (range, 1–91). Nevertheless, this finding brings to light the challenges and importance of standardization of NGS interpretation using functional studies that is partly addressed by protein correlation. It is noteworthy that evaluating the mutational status of EZH2 in myeloid malignancies is not as straightforward as lymphoid neoplasms26,49.

Finally, we correlated both EZH2 mutations and immunohistochemistry expression levels with clinicopathologic findings and outcomes. Although not formally incorporated into risk-stratification models, studies have shown that EZH2 is an independent predictor of worse survival in MDS patients3,7,15,50. While both EZH2 mutation and protein loss correlated with poor survival in MDS, independent of R-IPSS, age and gender, protein expression showed a stronger correlation with survival than mutation (p = 0.061 vs. p = 0.43). Remarkably, even within this small cohort of EZH2MUT MDS, the OS of patients with preserved and loss of EZH2 expression was significantly different (Supplemental Fig. S1C). A similar observation was noted among EZH2WT MDS (Supplemental Fig. S1D). This is important for therapy. EZH2 alterations are dominant clones which occur early in the disease course, making this an important and attractive targetable epigenetic regulator. Currently, multiple therapeutic approaches to restore the function of EZH2 are underway and show encouraging results19, which are not restricted to MDS cases with EZH2 mutations alone. Hence, there is a need for development and correlation of assays such as immunohistochemistry that are feasible for routine clinical workup for baseline and response assessment. The findings from this study highlight the prognostic value of EZH2 protein expression in MDS, regardless of the presence of mutation. Validation of the current findings in larger clinical cohorts of MDS patients, in the context of EZH2 targeted therapeutic strategies is needed. Overall, this study highlights the limitation of using NGS data “in silo” for prognostication of MDS and confirms that the epigenetic and genetic complexity of MDS extends beyond mutations.