Identification of Two DNMT3A Mutations Compromising Protein Stability and Methylation Capacity in Acute Myeloid Leukemia

Somatic mutations of DNMT3A occur in about 20% of acute myeloid leukemia (AML) patients. They mostly consist in heterozygous missense mutations targeting a hotspot site at R882 codon, which exhibit a dominant negative effect and are associated with high myeloblast count, advanced age, and poor prognosis. Other types of mutations such as truncations, insertions, or single-nucleotide deletion also affect the DNMT3A gene, though with lower frequency. The present study aimed to characterize two DNMT3A gene mutations identified by next-generation sequencing (NGS), through analysis of protein stability and DNA methylation status at CpG islands. The first mutation was a single-nucleotide variant of DNMT3A at exon 20 causing a premature STOP codon (c.2385G > A; p.Trp795∗; NM_022552.4). The DNMT3A mutation load increased from 4.5% to 38.2% during guadecitabine treatment, with a dominant negative effect on CpG methylation and on protein expression. The second mutation was a novel insertion of 35 nucleotides in exon 22 of DNMT3A (NM_022552.4) that introduced a STOP codon too, after the amino acid Glu863 caused by a frameshift insertion (c.2586_2587insTCATGAATGAGAAAGAGGACATCTTATGGTGCACT; p. Thr862_Glu863fsins). The mutation, which was associated with reduced DNMT3A expression and CpG methylation, persisted at relapse with minor changes in the methylation profile and at protein level. Our data highlight the need to better understand the consequences of DNMT3A mutations other than R882 substitutions in the leukemogenic process in order to tailor patient treatments, thus avoiding therapeutic resistance and disease relapse.


Introduction
Genetic alterations of chromatin regulators and epigenetic modifications cooperate to the pathogenesis of acute myeloid leukemia (AML) [1]. Among epigenetic modifications, DNA methylation represents a mechanism adopted by cells to regulate DNA accessibility through the conversion of 5methylcytosine (5-mC) to 5-hydroxymethylcytosine (5-hmC). It switches on/off the promoter of several target genes, thus regulating their expression levels and contributing to disease phenotype. DNA methyltransferases (DNMT3A, DNMT3B, DNMT1, and DNMT3L) are the key enzymes involved in DNA methylation. DNMT3A and DNMT3B produce de novo hemimethylated DNA and de novo symmetric DNA methylation, respectively, while DNMT1 is involved in maintenance of DNA methylated status [2,3].
DNA methylation has been identified as a specific functional category of mutated genes in AML that includes mutation in DNMT3A, DNMT3B, DNMT1, Tet methylcytosine dioxygenase 1 and 2 (TET1, TET2), and isocitrate dehydrogenase 1 and 2 (IDH1, IDH2) genes [4]. Moreover, Wilms tumor 1 (WT1) mutations were shown to reduce DNA methylation levels significantly in AML patients, conferring the same hypomethylation signature observed in patients harbouring TET2 mutations [5]. Haploinsufficiency of WT1 in preclinical models was also shown to decrease 5-hmC levels and impair TET2 function, especially in elderly animals [6].
DNMT3A mutations are among the most frequent driver mutations in AML, third for frequencies to Fms-like tyrosine kinase 3 (FLT3) and nucleophosmin 1 (NPM1) [1]. About 20% of patients with de novo AML show recurrent somatic mutations in the DNMT3A gene [7], which represents an early lesion in leukemia development. Indeed, DNMT3A genomic alterations are considered a preleukemic event in hematopoietic stem cells (HSCs) and confer a proliferation advantage [8].
ey have been identified in preleukemic stem cells of myelodysplastic syndrome (MDS) and AML patients [9]. Moreover, mutations in genes encoding epigenetic modifiers, including TET2, ASXL transcriptional regulator 1 (ASXL1), and DNMT3A have been reported in aging healthy people and are considered the first "hit" for the "clonal hematopoiesis of indeterminate potential (CHIP)" [10,11].
Genomic lesions of the DNMT3A gene mostly consist in missense mutations targeting a hotspot site at R882 codon, which exhibits a dominant negative effect [12]. Furthermore, DNMT3A R882 mutations are associated with a hypomethylated status of CpG islands, higher relapse rate, and poor prognosis [13][14][15][16][17][18], and they persist during remission [19,20]. It has been recently shown that the methylation levels are dependent on the amount of active DNMT3A and higher methylation confers a better prognosis to AML patients [21]. Among patients carrying DNMT3A mutations, 15-20% do not show the substitution at R882 codon, but they harbour truncations, insertion, or single-copy deletions [1,7]. e present study aimed to characterize two undescribed DNMT3A mutations identified by next-generation sequencing (NGS) in two different AML patients, through analysis of protein stability and DNA methylation status at CpG islands.  Cycle Sequencing Kit ( ermo Fisher Scientific) on an Applied Biosystems 3730xL genetic analyzer system ( ermo Fisher Scientific).

Western Blot
Analysis. Protein extracts were separated by sodium dodecyl sulphate-polyacrylamide gel electrophoresis (SDS-PAGE, Bio-Rad) and transferred onto nitrocellulose membranes. e following antibodies were used: rabbit anti-DNMT3A (D23G1; Cell Signaling Technologies) and goat anti-β-actin (Santa Cruz Biotechnology) as control; horseradish peroxidase (HRP)-conjugated antirabbit immunoglobulin (Ig)G (GE Healthcare) and anti-goat IgG (Santa Cruz) as secondary antibodies. ECL Prime (GE Healthcare) reagent was used for detection using ChemiDoc XRS+ System (Bio-Rad). Signal quantification was performed using Image J software.

DNA Methylation Assay.
Methylation was quantified on total DNA using the MethylFlash Methylated DNA 5-mC Quantification Kit (Epigentek) using triplicates of 100 ng of DNA from each sample. e absorbance (OD) was read at 450 nm using the Multiskan EX automatic microplate reader ( ermo Fisher Scientific). To quantify the absolute amount of methylated DNA, we generated a standard curve using a negative control (ME3) and 5 dilutions of positive control (10.0; 5.0; 2.0; 1.0; 0.5 ng/μl); next we determined the slope (OD/ng) using linear regression. According to manufacturer's instructions, we used the following formulas to calculate the amount and percentage of 5-mC: Where * 2 is a factor to normalize the positive control because it contains only 50% of 5 mC.

Identification of a Premature Stop Codon in Exon 20
of the DNMT3A Gene in AML. NGS on a primary AML sample identified a single-nucleotide variant of DNMT3A at exon 20 causing a premature STOP codon (c.2385G > A; p.Trp795Ter) that was confirmed by Sanger sequencing ( Figure 1(a)). Mutations at codon 795 were previously reported in angioimmunoblastic T-cell lymphoma (p.Trp795_Gly796ins3) [22] and in refractory anemia evolving to secondary AML (Trp795Cys) [23]. A STOP codon at position 795 has been reported in the Leiden Open Variation Database (c.2384G > A, https://www.lovd.nl/) and the c.2385G > A mutation has been found annotated in the database of single-nucleotide polymorphisms (dbSNP, rs1395575712). Moreover, missense variants affecting the codon 795 (chr2:25462024:A > C and chr2:25462024: A > G) were also annotated in the Genome Aggregation Database (gnomAD) with an allelic frequency of 0.000003979 and 0.000007957, respectively (https:// gnomad.broadinstitute.org/). We detected the mutation (VAF � 4.5%) in the BM sample of a 74-year-old woman (AML#1), with 70% of AML blasts at diagnosis, normal karyotype, wt-FLT3, NPM1, and tumor protein p53 (TP53) and intermediate cytogenetic risk according to ELN 2017 [24]. Mutations of IDH2 (c.515G > A; p.Arg172Lys), BCL6 corepressor (BCOR) (c.2915_2916insA; p.Tyr972Ter), and TET2 (c.3641G > A; p.Arg1214Gln) were also detected in the patient, with a VAF of 6.1%, 4.8%, and 3.2%, respectively (Figure 1(b) and Table 1). Previous evidence reported the co-occurrence of DNMT3A lesions with mutations of NPM1, IDH2, and TP53 and with those affecting genes involved in chromatin and splicing in AML cases [25]. e patient received an induction therapy with the DNMT inhibitor guadecitabine [26], with persistence of disease at the bone marrow evaluation after 4 courses of therapy (40% of blasts and stable cytogenetic risk). e DNMT3A mutation load increased to 38.2% in the sample evaluated 4 months after treatment. NGS analysis revealed an increasing VAF during treatment of the IDH2 (6.1% to 39.8%) and BCOR mutations (4.8% to 42.1%) detected at diagnosis, along with the emergence of a TP53 mutation (c.607G > A; p.Val203Met, VAF 5.3%, Figure 1(b) and Table 1), which was not detectable at diagnosis. Nowadays, no predictive marker of response to guadecitabine has been defined for newly diagnosed AML. However, Chung et al. reported no significant association between gene mutations and complete remission in a cohort of 128 relapsed/refractory AML [27]. Trends were observed for TET2-mutated cases (higher CR rate) and IDH1/2-mutated and TP53-mutated AML (resistance). Moreover, TET2 mutations have been associated with increased response to hypomethylating agents in MDS [28,29] and AML with low blast count [28]. We believe that our results suggest the persistence during treatment of a minor clone harbouring DNMT3A, IDH2, and BCOR mutations, which was positively selected and progressively expanded, along with the acquisition of a novel TP53-mutated subclone. In parallel, TET2-mutated blasts were killed by the treatment.  Figure 2(a)). e variant, which has never been reported before, was detected in primary leukemic cells isolated from a 63-year-old woman (AML#2), with 80% of blasts in the BM at diagnosis and 90% at relapse. e patient had normal karyotype, low FLT3-ITD allelic burden (c.1747_1748ins57: GCTCCTCAGA-TAATGAGTACTTCTACGTTGATTTCAGAGAATATG-AATATGATCCAA; VAF: 11.7% at diagnosis), and     [24]. Molecular analysis also detected mutations of NPM1 (c.863_864insCTTG; p.Trp288fs; VAF 37.4%) and TET2 (c.395delA; p.Asn132fs; VAF 38.6% and c.5504delG; p.Gly1835fs; VAF 43.6%, Figure 2(b) and Table 2). e patient started fludarabine, arabinosyl cytosine, and idarubicin (FLAI-5) induction regimen [30,31] and obtained complete hematological remission one month later, with undetectable FLT3 mutations. After 7 months of complete remission, the patient relapsed. Mutational analysis performed on the diagnosis and relapse samples showed that the DNMT3A mutation loads were, respectively, 35.8% and 42.5%. e relapse sample presented an increase of 5.5% for VAF of NPM1 mutation, along with an expanded FLT3-ITD clone (11.7% to 40.1%), the persistence of TET2 mutations with a slight VAF increase (c.395delA; p.Asn132fs; VAF 48.4% and c.5504delG; p.Gly1835fs; VAF 49.7%), acquisition of a WT1 mutation (c.1109G > C; p.Arg370Pro; VAF: 48.3%; Figure 2(b) and Table 2). In this patient, the treatment was not able to eradicate the leukemic clone harbouring TET2, DNMT3A, and NPM1 mutations, which seems to have acquired an additional WT1 mutation at relapse. Moreover, AML relapse was characterized by the expansion of the FLT3-ITD clone that was reduced but not eradicated by the treatment, with an increase of allelic ratio from 0.09 (diagnosis) to 0.21 (relapse). A recent study showed that the persistence of DNMT3A mutation at a VAF ≥2% at first remission is a common event in AML, occurring in 65% of cases with mutation at diagnosis and is associated with older age and inferior relapse-free survival [32].

e Identified DNMT3A Mutations Alter DNA CpG Islands Methylation and Protein Stability.
To investigate the functional consequences of the identified mutations, we performed methylation analysis of CpG islands on DNA extracted from primary leukemic samples of both patients. Our cases showed DNA hypomethylation in comparison with primary AML samples with wt-DNMT3A gene or R882H mutation (VAF 37.0%) (Figure 3(a)). In the sample carrying Trp795 * mutation, CpG islands methylation shifted from 47.1% at diagnosis to 24.4% at follow-up (p � 0.020), in accordance with the increase in the mutations allelic burden (from 4.5% at diagnosis to 38.2% at follow-up). At follow-up, CpG methylathion was significantly reduced compared with the wt-DNMT3A sample (p � 0.017). However, the observed hypomethylation status may be associated both with hypomethylating agent treatment and with the expansion of the DNMT3A-mutated clone during treatment.
In the sample carrying the insertion at r862_Glu863, CpG island methylation at diagnosis was significantly reduced compared with the wt-DNMT3A sample (p � 0.047). Moreover, its levels were similar to those observed in the R882H sample. No major changes occurred between disease diagnosis and relapse in terms of CpG methylation level and DNMT3A (c.2586_2587ins35; p. r862_Glu863fsins) VAF. e results suggest that the identified mutations induce DNMT3A loss of function, similarly to the R882H alteration. It has been recently shown that hypomethylation is an initiating feature of AML with DNMT3A R882 [33] and that demethylator phenotypes, which are partially related to DNMT3A mutational status, have a prognostic role, independent of age and cytogenetic abnormalities [34]. Future analyses of the aberrant DNA methylation pattern may help define specificities compared with DNMT3A R882 AML and novel potential silenced or activated enhancers involved in leukemogenesis [35].
To understand whether alterations in DNMT3A protein expression were responsible for the observed changes in CpG methylation, we performed western blot on BM samples at different time points. Protein analysis revealed that despite the low DNMT3A mutation burden in the diagnosis sample of the case AML#1 (Trp795 * ), DNMT3A protein levels were reduced to 72% and 21% compared with wt-DNMT3A and R882H samples, respectively (Figure 3(b)). Furthermore, the increased VAF of truncated DNMT3A was associated with no detectable protein in the follow-up sample (Figure 3(b)). Despite the heterozygous mutational status, the wt protein isoform became undetectable. We speculated that the wt protein was unable to form a heterodimer with the truncated DNMT3A form,   which may also be unstable, thus causing a premature degradation. is hypothesis is supported by a recent study, which demonstrates that DNMT3A truncation mutations have a dominant negative effect with loss of function and haploinsufficiency in AML [12]. On the contrary, DNMT3A protein was detectable in the analyzed samples from AML#2 (p. r862_Glu863fsins). In this patient, we observed that DNMT3A levels were reduced in the diagnosis sample compared with the wt one and were similar to those obtained in the R882H-mutated case (Figure 3(b)). DNMT3A expression slightly increased in the relapse sample, showing a 30% reduction in protein level compared with the wt sample. e data suggest that the mutation, which is predicted to introduce a premature STOP codon, interferes with protein expression but does not alter the stability of the wt protein. Future investigation will be useful to demonstrate if the presence of a portion of the catalytic domain in this truncated protein allows it to bind to wt-DNMT3A. Taken together, the protein and methylation analyses indicate that this mutation results in decreased DNMT3A expression and function.

Conclusions
DNMT3A mutations are becoming highly relevant in hematological malignancies, thanks to NGS technologies which allow us to better characterize myeloid disorders. In this study, we presented two mutations of the DNMT3A gene never described in AML and investigated their consequences on protein expression and function. Both mutations localized in the catalytic domain of the DNMT3A protein and were predicted to cause loss of function. e first one is responsible for a truncated, nondetectable protein associated with hypomethylation of CpG islands, which expanded under the pressure of hypomethylating agent treatment.
e second mutation was an insertion of 35 nucleotides, with a hypomethylation pattern, suggesting a negative effect on CpG methylation mediated by mutant DNMT3A. e DNMT3A-mutated clone escaped therapy selection and likely acquired a WT1 mutation. Both patients showed evidence of clonal evolution, resulting from the selective pressure induced by treatment. Our strategy, based on bulk sequencing, allows us to draw a picture of AMLrelated mutations, along with their frequency and to speculate on mutation co-occurrence at disease diagnosis and progression. However, a single-cell sequencing approach is needed to precisely evaluate the clonal complexity of AML across the different disease stages, drive conclusions on the clonal and subclonal architecture, and uncover genomic trajectories of leukemia evolution [36][37][38][39]. e presence of cooperating mutations and their contribution to the leukemic phenotype along with the identified DNMT3A mutations deserve further investigation. Our data highlight the need to characterize and monitor patient-specific genomic alterations in AML, in order to tailor treatments and allow early detection of expanding subclonal population. Moreover, future studies are needed to define the hypomethylation pattern resulting from the described genomic lesions and its cooperation with differentiation stage-specific histone modification in the regulation of the leukemogenic transcriptional program.

Data Availability
e clinical and molecular data used to support the findings of this study are included within the article.

Conflicts of Interest
Giovanni Martinelli receives compensation as a consultant for ARIAD/INCYTE, Pfizer, Celgene, Amgen, J&J, and Roche. e other authors declare no conflicts of interest.