Alterations of the expression of TET2 and DNA 5-hmC predict poor prognosis in Myelodysplastic Neoplasms

The study subjects included patients with a confirmed diagnosis of primary myelodysplastic neoplasms (as per WHO 2022 classification of MDS) [1] who had not received any disease-modifying treatment, and patients with de novo AML, or AML-MR. The control arm of the study included patients who had diagnosis of non-malignant conditions and a morphologically normal bone marrow (e.g., patients with peripheral blood cytopenias who were on a trial of vitamin B₁₂ due to suspected deficiency where the marrow was found to be morphologically normal at the time of bone marrow sampling, and patients with non-malignant causes of hypersplenism who presented with cytopenias but had a morphologically normal marrow). Only adult patients (≥ 18 years of age at the time of sample collection) were included. Those patients with therapy related MDS or AML, MDS/myeloproliferative neoplasm (MPN) overlap syndromes, chronic myelomonocytic leukemia (CMML), and acute promyelocytic leukemia were excluded. The study was performed in accordance with the relevant guidelines and regulations (Declaration of Helsinki) and was approved by the Institute Ethics Committee for Post Graduate Research, All India Institute of Medical Sciences, New Delhi, vide Letter No. IECPG-309/07.09.2017 dated September 14, 2017. Written informed consent was obtained from all the study subjects from whom any biological sample was collected. Up to 2.5 mL of bone marrow aspirate was collected from the study subjects in EDTA vial for obtaining bone marrow nucleated cells for DNA and RNA isolation. 4 μm sections that were cut from formalin fixed paraffin embedded bone-marrow biopsy specimens onto poly-L-lysine were also collected. The other details like clinico-hematological parameters and cytogenetics were obtained from the patients’ medical records and hospital information system.

A protocol optimised for downstream extraction of DNA and RNA was adopted while isolating BMNCs [22]. Briefly, the bone marrow aspirate was transferred to a 15 mL centrifuge tube and was centrifuged at 4 °C. After the removal of the supernatant, an equal volume of 1X RBC lysis buffer (BioLegend, San Diego, CA) was added to the tube, followed by gentle mixing and incubation at room temperature for 10 min. The tube was centrifuged, the supernatant was removed, and the same was repeated after the addition of 1 mL 1X RBC lysis buffer, this time in a 1.5 mL microcentrifuge tube. Following high-speed centrifugation, the pellet was washed with 1 mL phosphate buffered saline (PBS). The pellet was then suspended in Buffer RLT Plus (Qiagen, USA) (with β-mercapto-ethanol added). The cells in the buffer were homogenized by passing through a 20-gauge needle at least 5 times. The homogenized cells in the Buffer RLT Plus were stored at—80 °C for subsequent DNA and RNA isolation using AllPrep DNA/RNA Mini Kit (Qiagen, USA), which enabled the isolation of DNA and RNA from the same starting material in one go. The RNA isolation involved in-column DNase digestion to remove any contaminant DNA. The extracted DNA and RNA were quantified using a nano-spectrophotometer. A_260/280 and A_260/230 values of ≥ 1.8 and ≥ 2 were considered suggestive of good-quality DNA and RNA, respectively. Aliquots of the isolated DNA and RNA were also subjected to agarose gel electrophoresis to check for the integrity of the nucleic acids and detection of contamination with RNA or DNA, as the case may be. Only those DNA and RNA samples that met adequate quality standards and had sufficient quantity were subjected to further analysis.

The input amount of DNA for the 5-hmC assay (described later) was 100 ng in a volume of 4 μL – i.e., 25 ng/μL. Since the input DNA amount was critical due to the sensitive nature of the assay, the DNA concentration in the samples used for the assay was estimated using a dye-based method (QuantiFluor® ONE dsDNA System – Promega Corporation – Madison, WI), where a fluorescent double-stranded DNA-binding dye (504 nm Ex/ 531 nm Em) specific only for double-stranded DNA was used. The fluorescence after dye-binding was estimated by Quantus™ Fluorometer (Promega Corporation – Madison, WI).

1 μg of the extracted RNA was used for cDNA synthesis with random hexamer priming using Verso cDNA synthesis kit (Thermo Scientific, EU) according to the manufacturer's protocol. 1 μL of the cDNA (equivalent to 50 ng input RNA) was used for the subsequent qPCR reactions. The primers used for the qPCR reactions are listed in Supplementary Table 1. All the primers were designed to span an exon-exon junction in order to nullify the inadvertent amplification of genomic DNA targets by these primers. The cDNA was amplified using DyNAmo Flash SYBR Green qPCR Kit (Thermo Scientific, EU). The reactions were performed in triplicates, with negative and -RT (without reverse transcriptase) controls, and the runs were validated by performing a melt-curve analysis. The AriaMx Real-Time PCR System (Agilent Technologies.Inc) was used for performing the runs. The fold-change for the Gene of Interest (GOI) was calculated in the test samples in comparison to the control samples using the using the ΔΔCt method, using Glyceraldehyde 3-phosphate dehydrogenase (GAPDH) as the reference gene [23]. The latter was selected from a panel of reference genes as it showed the most consistent expression in the hematopoietic cell lines and marrow aspirate samples. The results were log-transformed and were expressed as log₂-fold change with respect to the controls.

Colorimetric assay based on a one-step ELISA for quantification of global DNA hydroxymethylation was performed using MethylFlash Global DNA Hydroxymethylation (5-hmC) ELISA Easy Kit (Colorimetric) (Epigentek, USA) using manufacturer’s protocol. The input amount of DNA for the assay was 100 ng (in a volume of 4 μL – i.e., 25 ng/μL). All the samples, standards, and the negative control were assayed in duplicates and the average absorbance of the negative control was subtracted from the samples and the standards. The standard curve was generated by plotting the absorbance of the different positive control samples on the Y-axis against the known 5-hmC percentage of these samples in the X-axis. A second order polynomial curve was graphed, and the second order polynomial regression equation in the form Y = aX² + bX + c was obtained, were X = 5-hmC%, Y = absorbance, a and b are slope 1 and slope 2, respectively. The percentage 5-hmC in the test samples were calculated by the formulawhere S is the input DNA amount (100 ng in the current study)

The absorbance in test samples was compared with that of standards to obtain percentage 5-hmC levels in the test samples.

Formalin fixed paraffin embedded (FFPE) bone marrow biopsy tissue sections were deparaffinized by heat and multiple xylene washes, followed by removal of xylene by graded washes with ethanol and rehydration in de-ionized water. The sections were further dipped in 10 mM citrate buffer (pH: 6) and were heated for 40 min for antigen unmasking, followed by three washes with Tris wash-buffer. The sections were placed for peroxide block for 10 min and the Tris wash was repeated. The non-specific binding of antibodies was blocked with Protein Block followed by three washes with phosphate buffered saline (PBS). The sections were incubated with the primary antibody (TET2 – mouse monoclonal antibody – C15200179, stock: 1 μg/μL, Diagenode – USA in 1:200 dilution, or 5-hmC – rat monoclonal antibody—C15220001-50, stock: 1 μg/μL, Diagenode – USA in 1:250 dilution) for 90 min in the dark, at room temperature, followed by two washes with Tris buffer, and then with biotinylated secondary antibodies followed by washing. After incubation with streptavidin-peroxidase complex, and washing, freshly prepared di-amino benzidine (DAB) (along with peroxidase substrate) was applied on the slides. The slides were immersed in distilled water as soon as a crisp brown nuclear staining was seen on monitoring with a microscope. The slides were counterstained with hematoxylin, dehydrated in graded alcohol, were then passed through xylene, and were mounted with Dibutylphthalate Polystyrene Xylene (DPX). Nuclear positivity was assessed in mononuclear cells and the percentage nuclear positivity was calculated in the stained cells. The cell counts were repeated thrice for determining the percentage of positive cells within the cellular areas of the marrow.

Since TET2 is a relatively large gene with exons 3–11 of the gene coding for 2002 amino acids, individually amplifying each of these exons and performing individual Sanger sequencing was technically cumbersome. Hence, exome sequencing (NGS) with an average depth of 100X was adopted to test for TET2 variants. For this 50 ng of DNA from each sample was used for Whole Exome Sequencing (WES) library preparation using Twist Library Preparation Kit (Twist Biosciences) followed by enrichment of the exome using ‘Twist Fast Hybridization Target Enrichment Protocol’ which was done in 3 pooled samples. Paired end Illumina sequencing using Illumina Hiseq 4000 NGS platform was carried out to generate 2 × 150 bp reads at an average sequencing depth of 100X. The general protocol for data analysis was obtained from Scaria V, et al. [24], where the fastq files from the sequencer were checked for quality using FASTQC, followed by trimming of the adapters and the low quality reads using Trimmomatic-0.36. An average Phred score of 30 and a Phred score of 20 in a sliding window of 5 were maintained. Stampy with Burrows-Wheeler-Alignment was used for mapping of the reads to the reference genome hg38. The aligned reads were sorted using SAMtools and the alignment quality was verified using Qualimap. Reads aligning to multiple loci were removed using Picard, and variant calling was performed with Platypus. BCFTools was used to screen the vcf files for the depth of sequencing (minimum depth of 20). The reads were visualized with Integrated Genome Viewer. Annotation of the variants was performed using ANNOVAR for genomic co-ordinates, chromosome location, population databases, in silico predictions, and known disease databases. The variants were first filtered by selecting those having splicing altering potential based on dbscSNV-ADA score and RF-score > 0.6 [25] and those located in the exons. Exonic synonymous variants were excluded, and frequency filter was applied with a cut-off of minor-allele frequency < 10% using gnomAD-Exome, gnomAD-Genome, ExAc and 1000 genome project databases. In-silico pathogenicity assessment of the mis-sense variants and indels was performed by SIFT (‘Deleterious’ and ‘Unknown’), Polyphen2-HDIV (‘Probably Damaging’, ‘Damaging’, and ‘Unknown’), and a CADD-Phred Score ≥ 15 (with at least 2 of these databases giving a concordant result). Frame-shift variants leading to premature termination codons were also considered pathogenic. The datasets [aligned bam files of the WES reads to the TET2 locus] generated and analysed during the current study are available in the NCBI-SRA repository [Accession no. SRP441583].

The TET2 protein domain architecture is shown in Fig. 4a for clarity. The 3D structure of the TET2 catalytic domain (TET2-CD) was predicted in order to model the Low Complexity Insert (LCI), which is missing in the crystal structure (PDBID:5D9Y) formed by residues 1481–1844 [26]. The structure of the LCI was modeled using RoseTTAFold on the Robetta Server, [27] followed by modeling of TET2 from 1127–1938 residues along with DNA using PDBID:5D9Y as a template. The model was validated using the Ramachandran plot and other metrics provided by the MolProbablity server [28]. The modeled structure was further refined through molecular dynamics simulations. The deleterious mutation, H1778R, was modeled using Pymol.

The modeled structure of TET2 and its variant forms were subjected to molecular dynamics simulations to understand the influence of the H1778R mutation on its local and global structure. Simulations were carried out using the GROMACS software suite (version 2021.5) with AMBER99SB-ILDN force field (Abraham, SoftwareX 2015). The proteins and protein-DNA complexes were solvated in a dodecahedron box with a TIP3P water model followed by neutralization with an appropriate number of sodium or chloride ions. The neutralized system was energy minimized using the steepest descent algorithm followed by equilibration under NVT [constant number (N), constant-volume (V), and constant-temperature (T)] and NPT [as for NVT, but pressure (P) is regulated] ensemble sequentially. Production simulation was run using a leapfrog dynamic integrator with a step size of 2 fs for a timescale of 200 ns, considering periodic boundary conditions in all three dimensions. Post simulation analysis was done after eliminating periodic boundary conditions using modules available in GROMAC and in-house python scripts. Simulations were carried out for wild type and H1778R variant TET2 in both apo state and DNA bound form.

The statistical analysis was performed using GraphPad Prism 8.0. The data were expressed in median and inter-quartile range or mean ± SD. Testing for normality was done by Kolmogorov–Smirnov test. Most of the data in the current study were non-parametric. These data were compared across different groups using Mann–Whitney U-test or Kruskal–Wallis test (when number of groups were > 2), since the effect of outliers on such analyses is minimal [29]. Quantitative variables were compared between each other using Spearman’s rank correlation coefficient. A p-value of < 0.05 was considered statistically significant.

In the current study, samples were collected from a total of 42 patients with MDS, 13 patients with AML-MR, 26 patients with de-novo AML, and 11 subjects with morphologically normal bone marrow. Cytogenetics data was available for 31 of the 42 MDS patients, and this, along with other laboratory parameters like hemoglobin, absolute neutrophil count, platelet count and bone marrow blast percentage, was used to obtain the Revised International Prognostic Scoring System (IPSS-R) score [20]. The MDS patients were also subtyped based on marrow morphology and cytogenetics as per the 2022 WHO Classification [1]. The baseline demographics of the study subjects from this study are given in Table 1. The age of the study subjects was comparable (Supplementary Figure 1a), and the median age was ≤ 50 years across all the groups. This is in contradiction to the data from the US and elsewhere where nearly 86% of MDS patients belonged to the age-group of 60 years or more, where the median age at diagnosis was found to be 76 years [30]; similarly, the median age at diagnosis for AML in the US was 65 years and 60% of AML patients were aged ≥ 60 years [31]. However, lower age at presentation in this study (Supplementary Figure 1b) was similar to that reported in the previous studies from our center [21, 32‐34] and elsewhere in India on MDS [35, 36] and AML, [37] though the reason for such a varied observation is yet to be elucidated. The age distribution across the IPSS-R subcategories as well as the WHO subtypes of MDS remained nearly uniform (Supplementary Figure 1c and d).

On quantification of the mRNA expression using qPCR, the patients in general showed a significantly lower expression of TET2 when compared to controls (log2-fold change: -0.5, p-value: < 0.05; Mann–Whitney U test) (Fig. 1a). The expression below a cut-off log₂-fold change of -1.1 was observed only among the patients. As for the individual groups, the expression was significantly lower in AML-MR compared to controls (log2-fold change: -0.8, p-value: 0.03; Mann–Whitney U test) (Fig. 1b); the expression was lower but not statistically significant in MDS and AML, compared to controls. Further, when the expression was analyzed across different IPSS-R risk categories, it was found to be significantly lower in the higher IPSS-R risk categories (p-value: < 0.05; Mann–Whitney U test) and this reduction in the expression was similar to that in AML-MR (Fig. 1c). Though certain MDS sub-types like MDS-LB, MDS-LB-RS, and MDS-IB showed a decrease in TET2 expression with respect to controls and other subtypes, these were not statistically significant (Fig. 1d). The expression of TET2 obtained in the qPCR was further validated by performing immunohistochemistry for TET2 protein in the FFPE sections of bone marrow biopsy from two of the subjects studied with high and low TET2 expression in qPCR, and the results of the IHC corroborated with the expression values obtained in qPCR (Fig. 1e–h).

Since the first stable product from TET2 mediated catalysis is 5-hmC, we quantified the levels of 5-hmC in the DNA of the study subjects (except de novo AML) using a colorimetry-based immuno-assay. The median percentage 5-hmC levels in AML-MR (4.6 × 10^–3) differed significantly from that of controls (9.6 × 10^–3) and MDS (7.2 × 10^–3) (p-value: 0.02 and < 0.05 respectively, Mann–Whitney U test) (Fig. 2a). The levels were lower, but not statistically significant, in the IPSS-R higher risk categories compared to controls and IPSS-R lower risk categories (Fig. 2b). Also, there was no significant difference in the 5-hmC levels across different MDS subtypes (Fig. 2c). The pattern of reduction in the 5-hmC levels in the higher risk IPSS-R categories and AML-MR was similar to that of the TET2 expression shown in Fig. 1. The mRNA expression of TET2 showed a positive significant positive correlation with the percentage 5-hmC levels in the DNA (p-value: 0.03; Spearman correlation) (Fig. 2d). The reduction in 5-hmC levels in AML-MR and higher risk categories of MDS was also confirmed by the examination of FFPE tissue sections from bone marrow biopsy using immunohistochemistry for 5-hmC. IHC was carried out for 23 samples (AML-MR:3, IPSS-R Very-high risk MDS: 2, IPSS-R High risk MDS: 2, IPSS-R Intermediate risk MDS: 6, IPSS-R Low risk MDS: 4, and Controls: 5). Control samples and lower IPSS-R risk categories showed higher expression of 5-hmC while the higher risk categories and AML-MR samples had much lower 5-hmC expression (Fig. 2e-h and Supplementary Figure 2).

DNA from BMNC samples of 24 patients (17 MDS and 7 AML-MR) were assessed for the presence of pathogenic variants of TET2 using exome sequencing, where 7 of the 24 samples (5 MDS and 2 AML-MR) tested positive (Supplementary Table 2). Of these, one patient with AML-MR had multiple aberrations in the TET2 gene including 2 frameshift deletions (Fig. 3a); the only other frameshift variant was found in a patient with high-risk MDS (Supplementary Table 2). To validate the results of the exome sequencing, the presence of two of the variants was further confirmed using Sanger sequencing (Fig. 3b-c). A number of these variants were localised to the catalytic domain of TET2 (Fig. 3d). Next the effect of the pathogenic variants on the levels of 5-hmC was checked for. Most of the samples with a pathogenic variant in TET2 showed a lower level of 5-hmC; however, this could not be considered causal, since majority of these samples also had a lower level of TET2 mRNA expression (Fig. 3e). Further, since some of the samples also showed a lower 5-hmC levels despite a high TET2 mRNA expression even in the absence of TET2 mutations, the expression of TET1 which is a paralog of TET2 [38] was studied in the study subjects using quantitative real time PCR, although TET1 is not considered as a major DNA hydroxy-methylating enzyme in the hematopoietic system. The expression of TET1 however did not vary significantly across the patient groups (Fig. 3f) or IPSS-R categories, giving rise to speculations about other mechanisms that can alter the 5-hmC levels in the BMNCs, a likely candidate being the 3^rd paralog of TET proteins – TET3. Yet, analysis of publicly available AML-MR and MDS datasets (GSE5881 and GSE145733) from NCBI-GEO using GEO2R did not show any significant alteration of TET3 expression in AML-MR and MDS when compared to healthy controls (Fig. 3g).

A molecular dynamics simulation was performed to study the effect of one of the pathogenic amino-acid variants in TET2 – p.H1778R – on the protein structure and stability. The TET2 protein domain architecture is shown in Fig. 4a. The crystal structure of the minimally active TET2 protein (residues 1128–1936 Δ1481-1843) includes a cysteine-rich domain followed by the double-stranded β-helix (DSBH) domain which is formed by a core of double-stranded beta helix structure also known as a jelly roll motif. DNA binds to the L1 and L2 loops of the Cys-Rich domain above the DSBH domain (Fig. 4b). The pathogenic variant H1778R lies in the Low Complexity Insert (LCI) within the DSBH domain of the TET2 gene. To assess the impact of H1778R on the structure of TET2, we modelled the structure of the catalytic domain (CD) of TET2 along with the low complexity insert since the existing crystal structures of TET2 are devoid of the LCI region (DOI for the model: https://​www.​modelarchive.​org/​doi/​10.​5452/​ma-9k1ka). The structure of the LCI (residues 1481–1846) was modelled using RosettaFold on the Robetta server. The predicted structure shows a globular protein having an α/β fold with a significant unstructured region between the well-formed secondary structures (Fig. 4c). The structure of the TET2-CD, in complex with DNA, was then modelled using comparative modelling on the Robetta Server. The modelled structure shows a distinct exterior domain for the LCI in concordance with the previously reported predictions based on multiple sequence alignment [39]. Further, the catalytic domain of TET2-CD with H1778R variant was also built through comparative modelling using Pymol.

Molecular dynamics simulations were performed to understand the structural influence of the H1778R variant on the catalytic domain of TET2-CD. We subjected the modelled structures of TET2-CD harboring the LCI region, in the native and variant forms, for 200 ns simulations in the presence of DNA and absence of DNA. The TET2-CD proteins and protein-DNA complexes were stable throughout the 200 ns simulations but displayed significant differences in their global dynamics, as shown in Fig. 4d to g. The LCI region demonstrated significantly more rotational and translational movement compared to the DSBH domain. The variant, H1778R, seemed to alter the global and local structure of the TET2-CD, as shown in Fig. 4e and g. More residue fluctuations were observed in the DNA-binding region of TET2-CD in the absence of DNA. The LCI region was more compact in the variant owing to increased intramolecular interactions (Fig. 4e). The TET2-CD_H1778R was more stable and compact, compared to the wild-type TET2-CD, as evident from the Root Mean Squared Deviation (RMSD) and Radius of gyration (Rg), respectively, as shown in Fig. 5a-d. In the case of the variant, the DNA was surrounded by the Cys-rich domain and the LCI domain, while in the wild type, interactions of LCI with DNA were diminished (Fig. 4d).

The N-terminal region of the LCI (residues 1462–1481), harbouring positively charged residues, showed interactions with DNA in both wild-type and variant TET2-CD. While in the wild type, these interactions were limited to the DNA backbone, the variant formed extensive base-specific interactions along with the DNA backbone (Fig. 4h). Also, the helix-turn-helix formed by the residues 1772 to 1807 showed interactions with the DNA backbone in the case of variant TET2-CD. This helix-turn-helix was very far from the DNA in the case of wild-type TET2-CD. In the absence of DNA, the positively charged N-terminus of LCI formed interactions with this helix-turn-helix (Fig. 4i). Also, interactions between the Cys-rich domain and LCI were observed in the variant TET2-CD, which were absent in the wild-type TET2-CD. This interface was formed by the N-terminus region of LCI and the residues R1253, K1254, Y1255, P1278, R1279, and D1314 of the Cys-C subdomain. The H1778 residue is part of the helix formed by residues 1768–1780. During the simulation, the hydrophobic H1778 in the wild type limited solvent exposure by forming intramolecular interactions, unlike the R1778 in the variant, which was predominantly solvent exposed.

In our patient samples, though two of the samples harboured the TET2 p.H1778R variant, one of them had co-occurring variants in TET2 including frameshifts, and the effect of the p.H1778R was hence not exclusive (Supplementary Table 2). The alternate sample, from a patient with IPSS-R intermediate risk and a normal bone marrow cytogenetics, showed the pathogenic variant in 58% of the reads (Supplementary Figure 3a). This patient showed a relatively low 5-hmC levels compared to the rest of the patient cohort, which is partly explained by the low mRNA expression levels of TET2 (Supplementary Figure 3b-c). This patient also was found to have a low TET2 protein expression and 5-hmC immunostaining on IHC (Supplementary Figure 3d-e), corroborating with the possibility of altered catalytic activity of TET2, as elucidated by molecular dynamics simulation.