Determinants of DNMT2/TRDMT1 preference for substrates tRNA and DNA during the evolution

ABSTRACT

RNA methyltransferase DNMT2/TRDMT1 is the most conserved member of the DNMT family from bacteria to plants and mammals. In previous studies, we found some determinants for tRNA recognition of DNMT2/TRDMT1, but the preference mechanism of this enzyme for substrates tRNA and DNA remains to be explored. In the present study, CFT-containing target recognition domain (TRD) and target recognition extension domain (TRED) in DNMT2/TRDMT1 play a crucial role in the substrate DNA and RNA selection during the evolution. Moreover, the classical substrate tRNA for DNMT2/TRDMT1 had a characteristic sequence CUXXCAC in the anticodon loop. Position 35 was occupied by U, making cytosine-38 (C38) twist into the loop, whereas C, G or A was located at position 35, keeping the C38-flipping state. Hence, the substrate preference could be modulated by the easily flipped state of target cytosine in tRNA, as well as TRD and TRED. Additionally, DNMT2/TRDMT1 cancer mutant activity was collectively mediated by five enzymatic characteristics, which might impact gene expressions. Importantly, G155C, G155V and G155S mutations reduced enzymatic activities and showed significant associations with diseases using seven prediction methods. Altogether, these findings will assist in illustrating the substrate preference mechanism of DNMT2/TRDMT1 and provide a promising therapeutic strategy for cancer.

Introduction

DNA cytosine-5 methylation, modulated via DNA methyltransferases (DNMTs), is an important epigenetic modification in animals. Mammalian DNMTs can be categorized into three families: DNMT1, DNMT2, and DNMT3. DNMT1, situated at the replication fork, is in charge of maintaining the DNA methylation pattern during the DNA synthesis. DNMT3A and DNMT3B are responsible for de novo DNA methylation in the animal normal development. DNMT3L is a pivotal cofactor for recognizing lysine 4 methylation status of histone H3 and can recruit and activate DNMT3A [1,2]. DNMT2 was originally annotated as a DNA methyltransferase by sequence comparison to other methyltransferases. However, subsequent studies have shown that DNMT2 exhibits extremely weak DNA methylation activity [3]. Even in some species where DNMT2 is the only retained methyltransferase, genomic methylation is still obscure. Goll et al. have demonstrated that DNMT2, as a (cytosine-5) RNA methyltransferase, conducts tRNA^Asp-GUC cytosine-38 (C38) methylation, resulting in renaming of this enzyme to tRNA-Asp-Methyltransferase 1 (TRDMT1) [4]. Subsequent studies have confirmed that human DNMT2/TRDMT1 and some homologues in other species methylate C38 of tRNA^Gly-GCC, tRNA^Val-AAC and tRNA^Glu-CUC in vivo [4–7]. In previous studies, we have identified some determinants for tRNA recognition of DNMT2/TRDMT1 [8,9], but its DNA methylation activity remains to be explored.

DNMT2/TRDMT1 is the most conserved member of the DNMT family from bacteria to plants and mammals. Although DNMT2/TRDMT1 is devoid of the regulatory N-terminal domains present in DNMT1 and DNMT3, all the homologues contain 10 conserved catalytic motifs commonly found in DNMTs. Similar residues in motifs IV, VI and VIII of this enzyme are engaged in the catalysis reaction as confirmed in other DNMTs through site-directed mutagenesis, indicating that DNMT2/TRDMT1 uses a DNMT-like catalytic mechanism to methylate tRNA [5]. DNMT2/3 and DNMT1 are independently evolved from DNA methyltransferase in bacteria, and an early DNMT2/TRDMT1 may change its substrate preference from DNA to tRNA [10,11]. Moreover, DNMT2/TRDMT1 homologues are reliably identified from other bacterial and eukaryotic DNMTs via conserved CFT motifs in the target recognition domain (TRD, residues 278−306 in human) [10]. DNMT2/TRDMT1 in Drosophila willistoni (D. willistoni) may recognize specific DNA targets via the distinctive TRD between VIII and IX motifs [12–14]. Nevertheless, whether TRD or other domains determine the DNMT2 substrate transition from DNA to tRNA needs to be elucidated.

Low-cost and efficient bioinformatics methods have been well exploited to determine the interaction mechanism of protein and nucleic acid. Protein-nucleic acid complexes can be obtained by molecular docking programme, for instance, HDOCK [15]. Moreover, molecular dynamics (MD) simulations can optimize the protein-nucleic acid structure, to gain insights into conformational information of protein and the protein-DNA/RNA recognition mechanism. Because of cancer genome sequencing initiatives, a large number of protein mutant data are brought forth. PremPRI and mmCSM-NA servers have been developed to calculate the binding affinity change of protein-nucleic acid complex owing to the protein mutation [16,17]. Excitingly, the association between protein mutation and disease occurrence can also be predicted using FATHMM-XF, FATHMM-MKL, CScape, Cancer, CanSavPre, PROVEAN and PhD-SNP [18–21]. Reportedly, DNMT2/TRDMT1 E63K mutation promotes tRNA methylation, which may be related to the phylogenetic consensus and charge property [22]. G155V and R371H mutations almost lead to the loss of this enzyme activity, possibly because of G155 and R371 residues in catalytic motifs VIII and X [22]. Unfortunately, L257V mutant shows more than 6-fold decreased activity, which is very difficult to explain [22]. As possible reasons for mutant activity changes, the effects of DNMT2/TRDMT1 mutation on protein conformation, protein stability, binding affinity and disease have not been explored to date. Hence, it is necessary to determine the DNMT2/TRDMT1-nucleic acid structure and analyse these effects.

So far, there is no definite substrate DNA for DNMT2/TRDMT1 in vivo, but it still possesses a weak activity towards DNA in vitro. Moreover, this enzyme has a strong activity on tRNA through DNMT-like catalytic mechanisms [5]. Thus, it is intriguing that DNMT2/TRDMT1 adopts the specific substrate preference and methylation mechanism. In this study, conserved TRD and TRED emerged in mammal DNMT2/TRDMT1s during the evolution, possibly contributing to the interaction between this enzyme and DNA or tRNA. Moreover, the enzymatic preference for substrate tRNA was determined by TRD and TRED, as well as the easily flipped state of tRNA target cytosine. Additionally, DNMT2/TRDMT1 cancer mutant activity was collectively mediated by five enzymatic characteristics, which might impact gene expressions. Altogether, our findings will provide a new insight into DNMT2/TRDMT1 substrate preference and this enzyme utility as a potential anticancer target.

Results

Phylogenetic tree construction and multiple sequence alignment of DNMT2/TRDMT1 homologues

DNMT2/TRDMT1 is the most conserved member of the DNMT family, and DNMT2/3 and DNMT1 independently evolve from DNMTs in bacteria [10,11]. In this study, the DNMT2/TRDMT1 distribution was extremely extensive in different species and its evolutionary origin was consistent with previous studies (Figure S1). Both DNMT2/TRDMT1 homologues and bacterial DNMTs had 10 motifs and target recognition domains (TRDs), whose sites were corresponding in the sequence alignment (Figure 1). Bacterial DNMTs had the most conserved six motifs, namely, motifs I (F-X-G-X-G), IV (PC), VI (ENV), VIII (Q-X-R-X-R), IX (R-X-X-X-X-X-E) and X (GN), which were similar to those in DNMT2/TRDMT1s (Figure S2 and S3). However, the mean sequence identity and similarity of 35 DNMT2/TRDMT1 homologues were 32.96% and 45.28%, respectively, which were significantly enhanced compared with those of 19 bacterial DNMT homologues (17.77% and 28.57%) (Table S2). Mammalian DNMT2/TRDMT1s had the highest mean sequence identity (80.86%) and similarity (85.25%) among all the species (Table S2). A tyrosine was substituted for the canonical phenylalanine in motif I of most DNMT2/TRDMT1s, compared to that (with the consensus F-X-G-X-G) in bacterial DNMTs (Figure S2 and S3). Coincidentally, the canonical glutamine was replaced with an asparagine in motif VIII of most DNMT2s, compared to that (with consensus Q-X-R-X-R) in bacterial DNMTs, except for tyrosine in plant, methionine in G. metallireducens and serine in G. sulfurreducens (Figure S2 and S3). In brief, DNMT2/TRDMT1 was a conserved (cytosine-5) RNA methyltransferase and widely distributed in the biological world.

Figure 1.

Multiple sequence alignment of mammalian DNMT2/TRDMT1 homologues and bacterial DNMTs performed using ClustalW multiple sequence alignment programme. The somatic cancer mutated residues were denoted in red, while conserved residues studied in TRD and TERD were indicated in green. Ten motifs, TRED, and TRD in DNMT2/TRDMT1 homologues were marked in black, whereas these domains in bacterial DNMTs were presented in blue.

The variable region was located between motifs VIII and IX in DNMT2/TRDMT1s and bacterial DNMTs, and displayed the greatest discrepancy in sequence composition and length (Figure S2 and S3). TRDs in variable regions were characterized by CFT and TL sequences for two kinds of enzymes, respectively (Figure S2 and S3). Sequence CFT in TRD of DNMT2/TRDMT1 was conserved during the evolution, except for sequence CFI in Holophaga foetida and CVT in Dictyostelium discoideum (D. discoideum) (Figure S3). Unexpectedly, conserved motifs IX, X and TRD were missing in rice DNMT2/TRDMT1 (Figure S3). The two homologues in rice and Z. mays were in the same branch of the phylogenetic tree (Figure S1). Moreover, the high sequence identities of rice ReDNMT2a, ReDNMT2b and ReDNMT2c with Z. mays ZmDNMT2 were 52.37%, 52.80% and 53.78%, respectively (Figure S4). These results suggested that the horizontal gene transfer during the evolution might lead to the loss of important motifs in rice DNMT2/TRDMT1. Additionally, bacterial DNMT sequence showed three long gaps in the variable region, compared to that of DNMT2/TRDMT1 (Figure 1). Herein, the DNMT2/TRDMT1 sequence corresponding to the longest gap was provisionally named the target recognition extension domain (TRED, residues 189−229 in human) (Figure 1). Conserved TRED could only be found in the variable region of mammal DNMT2/TRDMT1s (Figure S3). Altogether, CFT-containing TRD first evolved from bacterial DNMTs, and then conserved TRED came out in mammal DNMT2/TRDMT1s.

Structural modeling of human DNMT2/TRDMT1

Since human DNMT2Δ47 crystal structure lacked residues 79−96 (Motif IV: residues 75−84, Motif V: residues 95−106) and 189−247 [23], we modelled the intact three-dimensional (3-D) structure of this enzyme and then performed MD simulations to optimize it. In the current study, the backbone RMSD of DNMT2/TRDMT1-modelled structure from the initial structure at 4−8 ns was stable (Figure S5 A). Then, we superimposed the DNMT2Δ47 crystal structure and DNMT2/TRDMT1-modelled structures before and after MD simulations. The result demonstrated that the backbone RMSD value (2.978 Å) between the crystal structure and the modelled structure after MD simulations was much larger than the backbone RMSD value (0.053 Å) between this crystal structure and modelled structure before MD simulations (Figure S5 B, C). Furthermore, the backbone RMSD between the DNMT2/TRDMT1-modelled structures before and after MD simulations was 3.181 Å (Figure S5 D). From the above, the backbone conformation of DNMT2/TRDMT1-modelled structure changed greatly after MD simulations.

Thus, we detailedly investigated conformational changes of the whole structure, Loop, TRED, α-Helix 1, α-Helix 2, α-Helix 3, β-Sheet 1 and β-Sheet 2 after MD simulations, in comparison to these in the initial DNMT2/TRDMT1-modelled structure (Table S3). The average backbone RMSDs of α-Helix 1, α-Helix 2, α-Helix 3, β-Sheet 1 and β-Sheet 2 during the last 4 ns were no more than 0.09 nm (Figure S5 A, Table S3). However, the average backbone RMSDs of Loop and TRED were 0.269 nm and 0.448 nm, respectively, probably causing the large backbone fluctuation for the modelled structure after MD simulations (RMSD, 0.475 nm) (Figure S5 A, D, Table S3). Interestingly, TRED, α-Helix 1, α-Helix 2 and α-Helix 3 shifted towards the structural centre after MD simulations (Figure S5 D), indicating that this structure did not get loose but compact. Moreover, two loops transformed into β-Sheet 1 and β-Sheet 2, whereas β-Sheet 3 turned into a large Loop (residues 79−96) and moves towards the structural centre (Figure S5 D). To further evaluate the rationality of these structural changes, we superimposed the crystal structure of Entamoeba histolytica (E. histolytica) DNMT2 homologue (Ehmeth) and the DNMT2-modelled structure after MD simulations. The result exhibited that β-Sheet 2 and the large Loop in human DNMT2/TRDMT1 were consistent with the corresponding parts in EhMeth (Figure S5 E).

As evaluated by ERRAT programme, the overall quality of the optimized modelled structure was 79.319 and the overall G-factor was −0.79. The VERIFY-3D evaluation result showed an average 3D-1D score of ≥ 0.2 for at least 83.33% of amino acid residues. Ramachandran diagram analysis for this DNMT2/TRDMT1 structure demonstrated that 97.7% of residues were present in the allowed region and only 2.3% were in the disallowed region (Figure S5 F). In conclusion, this optimized DNMT2/TRDMT1 structure was good and suitable for structural analysis in depth.

Structural basis of substrate tRNA and DNA methylation by DNMT2/TRDMT1

Human DNMT2/TRDMT1 has 10 conserved motifs similar to those of bacterial DNMTs and can methylate substrate tRNA through a DNMT-like catalytic mechanism. Also, the M. HhaI-DNA-SAH crystal structure has been completely resolved [24], and this short DNA sequence is widely present in human genome (Figure 2 A, Table S4). The above favourable conditions prompted us to study the binding mechanism of DNMT2/TRDMT1 to DNA. Furthermore, the similarities of conserved motifs and their functions between DNMT2/TRDMT1 and M. HhaI motivated us to analyse the tRNA binding mechanism of the two enzymes using HDOCK server. MD simulations were carried out to optimize DNMT2/TRDMT1-tRNA and DNMT2/TRDMT1-DNA complex structures, respectively.

Figure 2.

The interaction pattern between the methyltransferase and nucleic acid. A. DNA sequence in the M. Hhal-DNA-SAH crystal structure. B. RMSD changes of DNMT2/TRDMT1 backbone, tRNA^Gly and DNA during the MD simulations of DNMT2/TRDMT1-nucleic acid structure. C. The DNMT2/TRDMT1-tRNA average structure during the last 75 ns of MD simulations. D. The DNMT2/TRDMT1-DNA average structure during the last 75 ns of MD simulations. E. The crystal structure of M. Hhal-SAH-DNA complex. F. M. Hhal-tRNA^Gly complex structure. The tRNA^Gly molecule was docked to DNMT2/TRDMT1 or M. Hhal using HDOCK server. The catalytic loops in DNMT2/TRDMT1 (residues 78–97) and M. Hhal (residues 80−99) were coloured in blue. The distance between SH group of C79 and C6 position of tRNA^Gly C38 was shown in red dashed line. The hydrogen bond interaction between the enzyme and nucleic acid was denoted by yellow dashed line.

The result of MD simulations exhibited that RMSD changes of the protein backbone and tRNA in the enzyme-tRNA structure were slightly higher than those in the enzyme-DNA structure (Figure 2 B). This implied that the enzyme-DNA structure was more stable during MD simulations. The RMSD changes of the DNMT2/TRDMT1 backbone in two complexes tended to be stable in the range of 125–200 ns (Figure 2 B). Consequently, the average structures of two complexes were calculated during the last 75 ns. Later, we perform the energy minimization with the steepest descent method and conjugate gradient method for two average structures. Next, the binding affinity of protein-nucleic acid was predicted using PredPRBA and PreDBA servers. The predicted binding affinity of DNA to DNMT2/TRDMT1 (−11.89 kcal/mol) was significantly stronger than that of tRNA to this enzyme (−7.67 kcal/mol), which supported the above speculation that the enzyme-DNA structure was more stable. Conformably, DNMT2/TRDMT1 forms two denaturant-resistant and probably covalent complexes with both 5-fluoro-2′-deoxycytosine-containing DNA and control oligonucleotides, unlike other DNMTs [23]. Similarly, M. HhaI displayed a stronger predicted affinity with DNA (−11.95 kcal/mol) than tRNA (−6.95 kcal/mol). In brief, DNMT2/TRDMT1 and M. HhaI were more likely to bind DNA than tRNA.

In the predicted complex structure, residues in motifs VI, VIII, TRED and TRD formed hydrogen bonds with DNA or tRNA, which might play an important role in the tight binding of nucleic acid to DNMT2/TRDMT1 (Figure 2 C, D, E, and Table S5). From another perspective, bases in the anticodon-stem loop (C35, C38 and G39) and D-stem loop (C13, G15 and A23) of tRNA formed hydrogen bonds with DNMT2/TRDMT1, presumably facilitating the direct interactions between the enzyme and tRNA (Figure 2 C, Table S5). E119, K122 and R162 formed hydrogen bonds with target C38 in the DNMT2/TRDMT1-tRNA structure (Figure 2 C), while Y10, E119 and R162 formed hydrogen bonds with target DC427 to stabilize it in the enzyme-DNA structure (Figure 2 D, Table S5). Additionally, sequence alignment results show that residues E119 and R162 in DNMT2/TRDMT1 corresponded to residues E119 and R165 in M. HhaI, respectively (Figure 1). Fortunately, F79, S85, E119 and R165 formed hydrogen bonds with DC427 to immobilize it in the M. HhaI-DNA-SAH crystal complex (Figure 2 F, Table S5). However, the target C38 was oriented in the opposite direction to the active centre in the M. HhaI-tRNA structure, implying that tRNA C38 might be not methylated by M. HhaI (Figure 2 G).

Compared to that in the M. HhaI-SAM crystal structure (PDB ID: 1HMY), the catalytic loop (residues 80−99) moved about 26 Å in the M. HhaI-DNA-SAH crystal structure and formed a closed state [24,25]. The loop movement not only made DC427 form hydrogen bonding with F79 and S85 but also caused the SH group of C81 in motif IV to approach the C6 position of DC427, leading to the target methylation initiation by a nucleophilic attack (Figure 2 F). Surprisingly, the catalytic loop (residues 78−97) did not significantly move towards the nucleic acid in the DNMT2/TRDMT1-tRNA or DNA structure, which did not make C79 very close to the C6 position of target C38 or DC427 (Figure 2 C, D). This suggested that DNMT2/TRDMT1, like M. HhaI, might need the synergistic effect of this enzyme, nucleic acid and methyl donor SAM to drive the catalytic loop to a closed state.

In a previous study, conserved CUXXCAC sequence in the anticodon loop plays an important role in stabilizing the C38 flipping and contributing to the DNMT2/TRDMT1 activation [9]. Moreover, it has been confirmed that this sequence exists in the classical substrate tRNAs of DNMT2/TRDMT1s (Figure 3 A). Therefore, it can be regarded as a characteristic sequence of substrate tRNA. Due to intra-loop base interactions, C38 bases in Homo sapiens (H. sapiens) tRNA^Gly-GCC and Thermus thermophilus (T. thermophilus) tRNA^Val-CAC structures flipped out of the anticodon loop (Figure 3 B, C), whereas C38s in E. coli tRNA^Asp-GUC and tRNA^Glu-UUC structures inverted into the loop (Figure 3 D, E). These were consistent with our previous results that DNMT2/TRDMT1 activities on tRNA^Gly-GCC and tRNA^Val-CAC were enhanced in vitro, compared to those of tRNA^Asp-GUC and tRNA^{Glu-CUC 9}. Besides, all three crystallized tRNAs had the characteristic sequence in the anticodon loop (Figure 3 F). So, we speculated that, in addition to the characteristic sequence, other bases in the anticodon loop might play a decisive role in maintaining the C38-flipping. We then modelled tertiary structures of human tRNA^Gly-GCC G34 or C35 mutants. The result showed that base mutations at position 34 or 35 had minimal effects on the tRNA conformation, other than the anticodon loop (Figure S6 and S7). Additionally, when the position 35 was occupied by U, the anticodon-loop conformation changed greatly, making C38 twist into the loop; when C, G or A was located at position 35, the loop conformation underwent minor changes, keeping the C38-flipping state (Figure S7). In conclusion, the classical substrate tRNA for DNMT2/TRDMT1 had a characteristic sequence CUXXCAC, whose base type at the position 35 could determine the C38 orientation.

Figure 3.

The target C38-flipping determinants. A. The classical characteristic sequence of substrate tRNA for DNMT2/TRDMT1. B. Tertiary L-shaped structure of H. sapiens tRNA^Gly-GCC. C. D and E. The crystal structures of T. thermophilus tRNA^Val-CAC (PDB ID: 1IVS), E. coli tRNA^Asp-GUC (PDB ID: 6UGG) and E. coli tRNA^Glu-UUC (PDB ID: 2DER). The hydrogen bonding between C38 and nearby base was denoted by yellow dashed line. F. Base sequence alignment between H. sapiens tRNAs and three crystallized tRNAs.

Determinants of substrate tRNA and DNA methylation by DNMT2/TRDMT1

Based on the structural analysis of DNMT2/TRDMT1-nucleic acid interactions, we found that TRED and TRD might play important roles in interacting with DNA or tRNA. The 11 residues (including P194, A198, S215, S219, Q221, S223, G224, I228, L229, Y297 and G305), except S215, were conserved in TRD and TRED of mammalian DNMT2/TRDMT1s, two residues (Y297 and G305) of which were conserved in all the species (Figure S3). Therefore, wild-type DNMT2/TRDMT1, point and truncation mutant proteins were purified, and their activities were detected to investigate the determinants of tRNA and DNA methylation. These proteins were in the concentration range of 90−400 μM and at > 90% pure as determined by Coomassie staining (Figure S8 A), indicating that the two protein parameters were propitious to activity characterization. Additionally, as outlined above, the anticodon-stem loop of tRNA was the most critical binding domain for DNMT2/TRDMT1. To minimize the impact on L-shaped structure of tRNA, only two base pairs (28UC-41GG) were deleted from the anticodon stem of tRNA^Gly (Figure 4 A). Despite this, the truncated anticodon-stem mutation affected the spatial conformation of anticodon loop (Figure 4 B). Compared with the controls, the bright bands appeared in the assays of DNA template syntheses and tRNA transcriptions, revealing that tRNA^Gly-GCC and tRNA^GlyDelta-ACS transcripts were successfully synthesized (Figure 4 C). Subsequently, the effects of these proteins and tRNA mutations on DNMT2 activities were determined as follows.

Figure 4.

Methylation activities of wild-type DNMT2/TRDMT1, deletion mutants, and M. Hhal on tRNA^Gly or DNA. A. The cloverleaf structure of tRNA^Gly-GCC. B. The tertiary structure of tRNA^Gly-GCC and tRNA^GlyDelta-ACS modelled using tRNAscan-SE 2.0 and RNAComposer. Two base pairs (28UC-41GG) of tRNA^Gly-GCC were coloured in cyan. C. The enzymatic syntheses of DNA templates and tRNA transcripts for tRNA^Gly-GCC and tRNA^GlyDelta-ACS. D. Bar graph of the incorporated ³H into tRNA^Gly-GCC and tRNA^GlyDelta-ACS by wild-type DNMT2/TRDMT1 and deletion mutants (ΔTRD, ΔTRED, and ΔTRD-TRED) after 70 min. E. Bar graph of the incorporated ³H into DNA by wild-type DNMT2/TRDMT1, deletion mutants, and DNMT2-DNMT3L after 70 min. F. Bar graph of the incorporated ³H into tRNA^Gly-GCC and DNA by M. Hhal after 70 min. Data were indicated as mean value ± SD from three independent trials.

In this study, ΔTRD, ΔTRED and ΔTRD-TRED activities on tRNA^Gly were significantly reduced, compared to wild-type DNMT2/TRDMT1 activity (Figure 4 D and S8 B). However, ΔTRD, ΔTRED and ΔTRD-TRED activities on DNA were about twice stronger than wild-type protein activity (Figure 4 E and S8 C). Reportedly, DNMT3L can activate the DNMT3A activity on DNA [2,26]. Unfortunately, DNMT3L supplement, instead of increasing wild-type enzyme activity on DNA, attenuated the activity by half (Figure 4 E and S8 C). As positive and negative controls, M. Hhal showed the strongest activity on DNA, but no activity on tRNA^Gly (Figure 4 F and S8 D), suggesting that it was a specific DNA methyltransferase and consistent with the analysis result for M. Hhal-DNA/tRNA structure. Perhaps, DNMT2/TRDMT1 retained a weak DNA methylation activity following the legacy of its ancestor, but the increasing richness of TRD and TRED sequences made this enzyme prefer substrate tRNA during the evolution.

As expected, DNMT2/TRDMT1 activity showed the 18-fold reduction on tRNA^GlyDelta-ACS compared to wild-type tRNA^Gly (Figure 4 D and S8 B). Additionally, P194A, A198L and S215A mutations slightly increased DNMT2/TRDMT1 activities on tRNA^Gly, compared with the wild-type (Figure 5 A and S8 E). In contrast, the activities of S219A, S223A and G224A mutants were reduced to 32.72%, 29.08% and 36.54% of wild-type DNMT2/TRDMT1 activity, respectively (Figure 5 A and S8 E). Strikingly, Q221A, I228A, L229A, Y297 and G305A mutations attenuated this activity to 7.74−16.11% of the wild-type (Figure 5 A and S8 E). Next, the effects of 11 mutated residues on protein stabilities were predicted by five methods. I228A, L229A, Y297A and G305A mutations destabilized the DNMT2/TRDMT1 protein (ΔΔG <0 Kcal/mol), as predicted by all the methods (Table S6). Interestingly, the activities of these mutants on tRNA^Gly were also impaired. Furthermore, DNMT2/TRDMT1-tRNA binding affinity change induced by the protein mutation was predicted via mmCSM-NA server. The result showed that A198L mutation improved the binding affinity of tRNA to this enzyme, enhancing the DNMT2/TRDMT1 activity, whereas S219A, Q221A, S223A, G224A, I228A, L229A and Y297A mutations decreased binding affinities, diminishing the activities of these mutants (Table S7). This indicated that the aforementioned mutated residues might be the binding groups in the active centre of this enzyme. In summary, the anticodon stem of tRNA was essential for DNMT2/TRDMT1 activity, while residues Q221, I228, L229, Y297 and G305 in TRD and TRED played important roles in promoting the substrate tRNA selection.

Figure 5.

Methylation activities and local conformation changes of DNMT2/TRDMT1 mutants in TRD and TRED. A. Bar graph of the incorporated ³H into tRNA^Gly-GCC by wild-type DNMT2/TRDMT1 and its mutants after 70 min. Data were indicated as mean value ± SD from three independent trials. B. DNMT2/TRDMT1 mutants in TRD and TRED were shown in this enzyme structure. C. Backbone RMSD changes of mutants during the MD simulations. Hydrogen bond patterns before and after mutations: I228A (D and E), L229A (F and G) and G305A (H and I).

Eleven mutated residues mentioned above were displayed in the DNMT2/TRDMT1-modelled structure (Figure 5 B). Hydrogen bond plays an important role in maintaining the secondary structure of protein, as well as keeping the tertiary and quaternary structures stable. Hence, hydrogen bond changes between these residues and their surrounding residues can reflect the local conformational changes of this enzyme. MD simulations were carried out for about 30 ns to refine the mutant structures. The backbone RMSD changes of I228A, L229A and G305A mutants tended to be steady during MD simulations (Figure 5 C). The results showed that, I228 and L229 formed hydrogen bonds with the surrounding residues in the wild-type enzyme, but A228 and A229 did not have hydrogen bond interactions in their mutants (Figure 5 D, E, F, G). On the contrary, G305 could not form hydrogen bonds with surrounding residues in the wild-type, but A305 was hydrogen bonded to nearby D281 in G305A mutant (Figure 5 H, I). Taken together, I228A, L229A and G305A mutations could change the local spatial conformations of protein, which might be related to the reduced stability of protein.

The somatic cancer mutation affected five enzymatic characteristics, modulating the DNMT2/TRDMT1 activity and disease occurrence

Besides somatic cancer mutation S223A described above, 15 other mutations (E63K, G155C, G155S, G155V, E185D, E185K, E202V, E202Q, D226Y, D226H, L257I, L257V, E317D, E317G and R371H) were selected from COSMIC database. The purity and concentration of these mutant proteins were similar to those of point mutants in TRD and TRED, which were suitable for enzymatic studies (Figure S9A). Subsequently, methylation activities of somatic cancer mutants on tRNA^Gly were detected. DNMT2/TRDMT1 mutant activities on tRNA^Gly was closely correlated with the reported activities of nine mutants on tRNA^Asp (Pearson Correlation: 0.837, P value: 0.005) (Figure 6 A, B) [22], indicating that this enzyme adopted a similar catalytic mechanism for different tRNAs.

Figure 6.

Methylation activities and local conformation changes of DNMT2/TRDMT1 cancer mutants. A. Bar graph of the incorporated ³H into tRNA^Gly-GCC by wild-type DNMT2/TRDMT1 and its cancer mutants after 70 min. Data were indicated as mean value ± SD from three independent trials. B. Correlation analysis of DNMT2/TRDMT1 mutant activities on tRNA^Gly-GCC and tRNA^Asp-GUC. The activity data of DNMT2/TRDMT1 mutants on tRNA^Asp-GUC were obtained from the literature [22]. C. Somatic cancer mutants were displayed in the DNMT2/TRDMT1 structure. D. Backbone RMSD changes of DNMT2/TRDMT1 mutants during the MD simulations. Hydrogen bond patterns before and after cancer mutations: E63K (E and F), G155C (G and H), G155S (G and I), G155V (G and J), E185D (K and L), and R371H (M and N).

Some mutated residues occurred in conserved motifs, such as E63 in motif III, G155 in motif VIII, R371 in motif X, and E202, S223, and D226 in TRED (Figure S3). G155, L257 and R371 were relatively conserved in all the species, whereas E63, E185 and E317 were strongly conserved in mammals (Figure S3). In the study, DNMT2/TRDMT1 activity was generally reduced owing to the mutation from relatively conserved residue to non-conserved residue, such as G155C, G155S, G155V, E185K, D226H, D226Y, L257V, E317D, E317G and R371H (Figure 6 A, S3, and S9 B, C). R371H mutant had only 1.71% of this wild-type enzyme activity, which almost made it inactive (Figure 6 A and S9 C). G155C, D226H, and L257V mutations reduced the activity by about 80%, while G155S and D226Y mutations lowered the activity by about 90% (Figure 6 A and S9 C). Additionally, the activities of G155V, E317G and E317D mutants were weakened to 38.31%, 48.42% and 82.34% of the wild-type enzyme activity, respectively (Figure 6 A and S9 B, C). Also, residues tended to mutate into amino acids that emerged during the evolution, such as E63K, E202Q, E202V and L257I, generally improving the enzymatic activities. The activities of E63K, E202Q, E202V and L257I mutants were enhanced by 55.30%, 28.03%, 42.34% and 15.13%, respectively (Figure 6 A, and S9 B). Furthermore, when the negatively charged residue mutated into positively charged or uncharged residue, such as E63K, E202Q and E202V, the enzymatic activity was usually increased (Figure 6 A and S9 B). In short, DNMT2/TRDMT1 mutant activity was mediated by the conservation and the charge property of mutated residue.

The effect of DNMT2/TRDMT1 cancer mutation on the protein stability was predicted. The result showed that mutations G155V, E185D, L257V, E317D and R371H decreased the protein stabilities (ΔΔG <0 Kcal/mol), while G155C, G155S and E317G significantly destabilized proteins (ΔΔG <-0.80 Kcal/mol) (Table S6). Apparently, all the above mutations, except E185D, reduced the enzymatic activities to some extent (Figure 6 A). In contrast, E202V mutation stabilized the protein (ΔΔG >0 Kcal/mol), perhaps enhancing the DNMT2/TRDMT1 activity (Figure 6 A, Table S6). Next, binding affinity changes of tRNA to DNMT2/TRDMT1 caused by mutations were predicted. The further result demonstrated that, mutations E63K, E185D and E185K increased the enzyme-tRNA binding affinity (Table S7). Coherently, the activities of E63K and E185D mutants were increased to 155.30% and 177.06% of the wild-type DNMT2/TRDMT1 activity, separately, although E185K mutant activity was fine-tuned to 99.91% of the wild-type (Figure 6 A). Conversely, G155C, G155V, D226Y, L257V and E317G decreased the enzyme-tRNA binding affinity, probably reducing the enzymatic activity (Figure 6 A, Table S7). In brief, DNMT2/TRDMT1 mutant activity could be modulated by the protein stability and the binding affinity of tRNA to this enzyme.

Based on the electrostatic interaction between protein and tRNA phosphodiester backbone, a defined tRNA binding cleft is found by lysine and arginine mutations in DNMT2/TRDMT1 [27]. In the DNMT2/TRDMT1 structure, G155, E185, E202, D226, L257, E317 and R371, except E63, were located in the tRNA binding pocket (Figure 2 C, 6 C). CD spectra of mutants did not alter in the Far-UV region (190–250 nm), compared to those of the wild-type, indicating that these mutations had little effect on the overall secondary structure of the enzyme (Figure S9 D).

Next, we made the point mutant in the DNMT2/TRDMT1 structure and then performed MD simulations for this mutated structure. Backbone RMSD changes of mutants were inclined to be stable during the MD simulations (Figure 6 D). The hydrogen bond between the backbone carbonyl group of E/K63 and the backbone nitrogen atom of R67 was formed before and after the residue 63 mutation (Figure 6 E, F). However, E63K mutation converted the positive charge into the negative charge at position 63, perhaps contributing to the binding of this enzyme to tRNA. Although the backbone nitrogen atom of G/C155 formed hydrogen bonds with the backbone carbonyl group of T152, the presence of the sulphhydryl group of C155 made it more polar than G155 (Figure 6 G, H). Additionally, S155 formed hydrogen bonds with S191 and M252 in G155S mutant (Figure 6 I); V155 formed hydrogen bonds with P151 in motif VII and M252 in G155V mutant (Figure 6 J). Consequently, three G155 mutations might affect the pivotal conformation of motif VIII and destabilize the protein, leading to the decreased enzymatic activity. E185 formed hydrogen bonds with L148, S153 and K188 in the wild-type (Figure 6 K), whereas D185 formed hydrogen bonds with L148 and K188 in E185D mutant (Figure 6 L). According to the result predicted above, this change induced by E185D mutation might reduce the protein stability, but increased the binding of tRNA to DNMT2/TRDMT1, promoting the enzymatic activity. Hydrogen bonds between R/H371 and N375 and between R/H371 and K367 remained unchanged (Figure 6 M, N). Nevertheless, the hydrogen bond between two side chains (amino group of R371 and carbonyl group of Q368) made the side chain of R371 parallel to the tRNA binding pocket wall, whereas the side chain of H371 lied almost perpendicular to α-helix in motif X and the groove wall (Figure 6 M, N). This change probably hindered the binding of tRNA to protein, decreasing the enzymatic activity. In conclusion, DNMT2/TRDMT1 activity was closely related to the local spatial conformation of protein.

Cancer is characterized by global hypomethylation and local hypermethylation in human genome. Methyltransferase, responsible for genome methylation, is frequently mutated and plays an important role in the development of cancer. Mutations in six methyltransferase genes were analysed using the COSMIC project in COSMIC v96. Missense substitution was the most common mutation type for all the methyltransferase genes in cancer (Table S8). However, missense substitution had a smaller number of samples in RNA methyltransferase, especially DNMT2/TRDMT1, than those in DNA methyltransferase (DNMT1, DNMT3A, DNMT3B and DNMT3L) (Table S8). Among 157 missense substitution samples, 124 different CDS mutations were found in DNMT2/TRDMT1 (Table S8). The relationship between these mutations and diseases was strictly predicted using seven methods. The result exhibited that mutations E63K and E185D were not related to the disease occurrence, predicted by at least five methods (Table S9). In contrast, R371H mutation was associated with the disease occurrence by five methods (Table S9). Excitingly, G155C, G155S and G155V mutations were dramatically associated with the disease predicted by all the methods (Table S9).

DNMT2/TRDMT1-mediated tRNA methylation diversified physiological functions of this enzyme

The C38 methylation of tRNA^Asp by DNMT2/TRDMT1 enhances its aminoacylation efficiency, thereby promoting the synthesis of proteins containing poly-Asp, such as Phosphatase 2A inhibitor I2PP2A (protein-SET), transcription factor DP-1 (TFDP1), TATA-box binding protein-associated factor 9 (TAF9) and Enhancer of zeste homologue 2 (EZH2) in mouse embryonic fibroblast (MEF) cells [28]. Moreover, DNMT2/TRDMT1 plays an important role in reducing the mistranslation of non-synonymous near-cognate codon, thus promoting the accurate polypeptide synthesis [29]. Furthermore, DNMT2/TRDMT1-mediated C38 methylation can protect tRNA^Asp-GUC, tRNA^Gly-GCC and tRNA^Val-AAC from the angiogenin-induced cleavage in flies and mice [6]. Indeed, this enzyme still prevents tRNA^Asp-GUC and tRNA^Gly-GCC cleavages in MEF cells and mouse under no external stress [7,29]. Hence, it was speculated that DNMT2/TRDMT1 not only could enhance the expression of poly-Asp-containing protein but also might contribute to the expression of protein containing poly-Gly or poly-Val.

In the previous study, differentially expressed proteins in DNMT2/TRDMT1-deficient mice have been detected through two different proteomic methods [29]. The results have exhibited that, protein expression levels of a large number of genes are downregulated, but are not correlated with their mRNA levels. To verify the above speculation, we dissected sequence compositions of these proteins based on the published proteomic data. As speculated, all three kinds of proteins could be found in the downregulated proteins. That is, poly-Asp-containing proteins are as follows: Calsequestrin-1 (CASQ1), Calsequestrin-2 (CASQ2), Histidine-rich Ca²⁺ binding protein (HRC), Asporin (ASPN), Eukaryotic translation initiation factor 3 subunit J-A (EIF3J1), Methionine aminopeptidase 2 (METAP2), Elongation factor 1-beta (EEF1B), Prothymosin alpha (PTMA) and Prostaglandin E synthase 3 (PTGES3) (List 1 A); poly-Gly-containing proteins are as follows: WIP (encoded by gene WASP/WASL interacting protein family member 1, WIPF1), Family with sequence similarity 98 member B (FAM98B), Keratin-9 (KRT9), Striatin-4 (STRN4), Alpha-actinin skeletal muscle isoform 3 (ACTN3), Plasminogen activator inhibitor 1 RNA-binding protein (SERBP1), Heterogeneous nuclear ribonucleoprotein D0 (HNRNPD), RNA and export factor-binding protein 2 (REFBP2) and 60 kDa heat shock protein, mitochondrial (HSPD1) (List 1 B); Val-rich proteins are as follows: Guanylate-binding protein 9 (GBP9), Collagen alpha-1(VI) chain (COI6A1), Putative RNA-binding protein 3 (RBM3), Leucine aminopeptidase 3 (LAP3), Proteasome subunit beta type-9 (PSMB9), Phosphoglycolate phosphatase (PGP), Carboxymethylenebutenolidase homologue (CMBI) and Collagen alpha-1(XIV) chain (COI14A1) (List 1 C). Of note, synthesis rates of CASQ1, CASQ2, HRC, ASPN, PTGES3, KRT9, STRN4, ACTN3, COI6A1, CMBI and COI14A1 proteins were reduced by more than twofold (Table S10).

Next, substrate tRNAs of DNMT2/TRDMT1 were analysed using human tRNA gene in Genomic tRNA Database. The results showed that the amounts of substrate tRNA^Asp-GUC, tRNA^Gly-GCC and tRNA^Val-AAC accounted for 80.00%, 36.59% and 31.11% of the total amount of tRNAs transporting the same amino acids, respectively. Furthermore, all the tRNA^Asp-GUC sequences have high scores, compared to other two kinds of tRNAs. This indicated that poly-Asp-containing protein expression could be most strongly affected by DNMT2/TRDMT1. Then, we searched for proteins containing poly-Asp, Gly, and Val, separately, using Ensembl Human database in Scansite 4.0 server. The result further revealed that 11 poly-6-Val-containing proteins were coded by four genes, while 186 proteins containing poly-6-Asp were expressed by 60 genes. Consistent with those reported in mice [28,29], poly-6-Asp sequence was also present in human protein homologues, including SET similar protein (SETSIP), TAF9, EZH2, CASQ1, CASQ2, HRC and ASPN (List 2). Interestingly, four genes encoding eight proteins, ATPase family AAA domain-containing protein 2 (ATAD2), Unconventional prefoldin RPB5 interactor 1 (URI1, URI or RMP), HRC and ASPN, contained at least 13-Asp-aggregated sequence (List 2). Besides, there was a relatively large amount of proteins containing poly-6-Gly, of which 193 proteins containing poly-9-Gly were expressed by 74 genes. Strikingly, both human WIP and FAM98B proteins contained no fewer than 10-Gly-enriched peptide fragments, and their homologues were reported to be downregulated in DNMT2/TRDMT1-deficient mice (List 2) [29]. In brief, tRNA methylation by DNMT2/TRDMT1 could regulate expressions of proteins containing poly-Asp, Gly and Val.

GO term enrichment analysis was performed for a total of 138 genes using GOnet server, according to three GO term categories: cellular component, biological process and molecular function. In the cellular component category, five term clusters were formed and seven most prominent terms belonged to a cluster starting from ‘nucleus’ (Figure 7 A, B). In the molecular function class, three clusters originated from ‘organic cyclic compound binding’, ‘transcription regulator activity’ and ‘chromatin binding’, respectively (Figure 7 C). Ten most pronounced items were assigned to ‘organic cyclic compound binding’ and the ‘transcription regulator activity’ clusters containing most of the genes (Figure 7 C, D). In the category of biological processes, four clusters stemmed from ‘regulation of metabolic process’, ‘developmental process’, ‘nucleic acid metabolic process’ and ‘chromosome organization’, severally (Figure 7 E). Ten most outstanding items were allocated to ‘regulation of metabolic process’ cluster containing the vast majority of genes (Figure 7 E, F). From the above, a large percentage of proteins, located in the nucleus, mainly performed organic cyclic compound and heterocyclic compound binding functions and then participated in the regulations of gene expression and nucleobase-containing compound metabolic process.

Figure 7.

GO term enrichment analysis of 138 genes (encoding proteins containing poly-Asp, Gly or Val sequence) using GOnet server. (A, C and E) GO trees were built by cellular component, molecular function and biological process terms, respectively. (B, D and F) The top 20 most enriched GO terms were listed by cellular component, molecular function and biological process categories, respectively.

Discussion

DNMT2/TRDMT1s and bacterial DNMTs contain 10 conserved catalytic motifs, of which motifs I, IV, VI, VIII, IX and X have minimal sequence changes. However, primary motifs IX and X cannot be explicitly verified in M. Sssl and M. cviJI [30]. Coincidentally, some DNMT2/TRDMT1s lack conserved motifs IX and X, such as DNMT2a in Glycine max and DNMT2a and DNMT2c in Lotus japonicas [31]. In the present study, rice DNMT2/TRDMT1 was short of TRD, motif IX and motif X. Reportedly, DNMT2/TRDMT1 expression in rice roots and shoots was dramatically enhanced under salt stress [32]. Similarly, Physcomitrella patens DNMT2/TRDMT1 overexpression under salt stress promoted the tRNA^Asp-GUC stability and weakened its sensitivity to stress [33]. This implies that DNMT2/TRDMT1 overexpression may play a similar role in rice. The speculation is also consistent with our result that TRD-deleted mutant still has the activity on tRNA. Human DNMT2/TRDMT1 crystal structure has serious drawbacks, lacking the vital catalytic loop in motif IV and TRED sequences. Interestingly, the intact DNMT2/TRDMT1 modelling structure shrank and became compact after MD simulations, which might be due to the effect of TRED and catalytic Loop supplement. In addition, the paired U28C29-G41G42 in the anticodon stem of tRNA^Gly-GCC was not conserved among all the substrates, but it played a crucial role in exerting the DNMT2/TRDMT1 activity. Thus, the conserved spatial structure of tRNA can sometimes be as important as the conserved sequence of DNMT2/TRDMT1.

Structural features of M. HhaI interacting with DNA and S-adenosy-L-methionine (SAM) have been revealed, which provide instructive insights into the catalytic mechanism of DNMT2/TRDMT1. Three conformational changes are indispensable for M. HhaI activation. After the binding of M. Hhal to DNA, the target cytosine-flipping and the closed conformation of the catalytic loop occurred in a direct coupling manner. The closed catalytic loop stabilized the cytosine and guided SAM into the cofactor-binding pocket through hydrophobic interactions (Figure S10 A, B) [24,25]. S-Adenosyl-L-homocysteine (SAH), as a SAM product in the catalysis reaction, only lacks a methyl group. Except for E. histolytica DNMT2/TRDMT1-SAH complex, three complexes in human, Schizosaccharomyces pombe (S. pombe) and Spodoptera frugiperda (S. frugiperda) are crystallized without RNA supplement [23,34–36]. Even so, SAH in each crystal structure is located in an active orientation (Figure S10 C, D, E, F). Thus, probably unlike M. Hhal, SAM enters the cofactor-binding pocket of DNMT2/TRDMT1, which does not require the guidance from RNA or DNA-induced conformational change. However, the base-flipping and the closed catalytic loop are required for RNA or DNA methylation by DNMT2/TRDMT1. Due to base interactions in the anticodon loop, C38 has been already located at the flipping state or easy to twist compared with the paired cytosine in DNA. This may be why DNMT2/TRDMT1 is more likely to methylate tRNA than DNA. Residues F84, V121 and F124 in M. HhaI adopt van der Waals contact to the target cytosine, which is essential for inducing the cytosine flipping [37]. In this study, V121 and F124 in M. HhaI were extremely conserved among DNMT2/TRDMT1 homologues, while F82 in human DNMT2/TRDMT1, corresponding to F84 in M. HhaI, was present in 20 of 35 homologues (Figure 1, S3). In brief, DNMT2/TRDMT1 has the structural basis of DNA binding and methylation.

Reportedly, DNA oligonucleotide methylation by DNMT2/TRDMT1 can be detected via continuously changing tRNA contents into DNA [38]. Substrate DNA for DNMT2/TRDMT1 can also be sought based on these DNMTs with shared functional motifs and known DNA substrates, for instance, M. HhaI. In the current study, human DNMT2/TRDMT1 had a larger 3-D structure than M. HhaI, spanning 13 nucleotide bases of DNA. Moreover, it methylated the DNA fragment derived from M. HhaI substrates and containing two contiguous CpG sites, which was not yet promoted by human DNMT3L. As the report goes, DNMT2/TRDMT1 in Plasmodium falciparum (P. falciparum) cannot methylate a DNA fragment containing 50 CpG-rich sites, but mouse DNMT3A can [39]. Because of the stronger bond energy for the paired G-C, the superfluous CpG repeats make DNA more rigid, perhaps hampering the phosphodilipid backbone distortion and base-flipping in the DNMT2/TRDMT1-mediated catalysis. Intriguingly, whether the DNA fragment can be methylated by human DNMT2/TRDMT1 or DNMT3A needs to be further identified in vivo. Nevertheless, we preliminarily explore the catalytic mechanism of this enzyme on DNA.

Currently, it is well known that DNMT2/TRDMT1 evolves from bacterial DNMTs. DNMT2/TRDMT1 homologues not only retain conserved sequences from the legacy of their ancestors but also have some special sequences needed for the survival under the evolutionary selection. DNMT2/TRDMT1 sequence alignment exhibits some great distinctions for TRD and TRED in D. willistoni and D. melanogaster, leading to an outstanding difference of surface electrostatic distribution [14]. Interestingly, the differential cleavage pattern occurrence between sexes represents a DNA methylation phenomenon in D. willistoni, which does not arise in D. melanogaster [12,13]. Ten conserved motifs can act as a stable framework, determining the methylation activities of DNMTs. However, the changed residues between conserved motifs restrict the substrate recognition specificity. In this study, the greatest different region was located between motifs VIII and IX, and was divided into TRED and TRD. Importantly, TRED and TRD were responsible for substrate tRNA preference, especially Q221, I228, L229, Y297 and G305.

The mutation is a double-edged sword for DNMT2/TRDMT1 activity on tRNA. In the present study, all the mutations at one site, such as G155C, G155S, G155V, D226H, D226Y, E317G and E317D, reduced the enzymatic activities, while two mutations E202V and E202Q increased the activities. However, L257I and L257V mutations cause opposite changes for the enzymatic activities. Additionally, DNMT2/TRDMT1 gene mutation is closely related to the occurrence and development of diseases. Reportedly, three intronic single-nucleotide polymorphisms (SNPs) (rs7085709, rs11254397 and rs12241572) in DNMT2/TRDMT1 are discovered to be closely related to the total plasma folate level considered as risk factors of congenital heart defects (CHD) [40]. The intronic rs2295809 in this gene is bound up with the upregulated vitamin B12 and red blood cell folate levels, diminishing the risk of neural tube defects (NTD) [41,42]. An exonic SNP (rs11254413, H101Y) in DNMT2/TRDMT1 is in close relationship with the gastric cancer susceptibility [43]. According to the COSMIC database, mutations G155C, G155V and G155S occur in lung adenocarcinoma, ovarian adenocarcinoma and breast cancer, respectively. In the study, all the three mutations at G155 were predicted to show significant associations with diseases.

During the evolution, DNA methylation induced by DNMT2/TRDMT1 may not be so important, but this enzyme-mediated RNA methylation is more and more shining, due to its role in regulating the protein expression. In the published data, poly-Leu-containing protein (Notch receptor 4 and Interleukin 21 receptor), poly-Cys-containing protein (G protein-coupled receptor 37 like 1 and TXK tyrosine kinase) and poly-Lys-containing protein (Family with sequence similarity 133 member A and Transcription termination factor 1) could not be found in the downregulated proteins (List 3) [29]. However, many poly-Asp, Gly or Val-containing proteins relevant to substrate tRNAs of DNMT2/TRDMT1 were revealed, of which ATAD2, EZH2, WIP, CASQ2, TAF9, ASPN, URI and FAM98B functioned as oncogenic proteins. Except ATAD2 and URI, the expressions of other six proteins in mice have been confirmed to be regulated by DNMT2/TRDMT1 [28,29]. As oncogenes, ATAD2 and EZH2 have been extensively studied in lung, ovarian and breast cancers, which can promote the proliferation, migration and invasion of tumour cells [44–47]. WIP, encoded by gene WIPF1, promotes the tumour proliferation and invasion in breast cancer and lung adenocarcinoma, and can be consequently used as a potential diagnostic marker [48,49]. CASQ2 can enhance the tumorigenesis and metastasis through mediating the tumour microenvironment in breast cancer [50], while TAF9 overexpression contributes to the radiation resistance in breast cancer radiotherapy [51]. Although ASPN can activate the invasion in gastric cancer, it has been reported to play the dual role in breast cancer [52–54]. URI can strengthen the invasion, metastasis and chemotherapeutic resistance in cervical cancer and ovarian carcinomas [55,56]. FAM98B overexpression can reinforce the proliferation capacity and colony formation of colorectal cancer cells [57]. Thus, it can be speculated that three mutations at G155 reduce DNMT2/TRDMT1 activities, probably exerting anticancer effects through downregulating oncogene expressions.

Conclusions

Taken together, CFT-containing TRD in DNMT2/TRDMT1 homologues first evolved from bacterial DNMTs and then conserved TRED emerged in mammal DNMT2/TRDMT1s. TRD and TRED played a crucial role in the substrate DNA and RNA selection during the evolution. Moreover, the classical substrate tRNA for DNMT2/TRDMT1 possessed a characteristic sequence CUXXCAC, whose base type at position 35 mediated the C38 orientation. Hence, the substrate preference of DNMT2/TRDMT1 was determined by the easily flipped state of tRNA target cytosine, as well as TRD and TRED, especially residues Q221, I228, L229, Y297 and G305. Additionally, DNMT2/TRDMT1 mutant activity was collectively modulated by the conservation and charge property of mutated residue, the protein stability, the binding affinity and the local conformation. Furthermore, DNMT2/TRDMT1 could regulate gene expressions at the translational level through tRNA methylation, and these genes, in turn, might regulate the downstream gene expressions at the transcription level via GO term enrichment analysis. More importantly, G155C, G155V and G155S mutations significantly reduced enzymatic activities and were closely related to the disease occurrence using seven prediction methods. Thus, these findings will assist in illustrating the substrate preference mechanism of DNMT2/TRDMT1 and provide a promising therapeutic strategy for cancer (Figure 8).

Figure 8.

The preferred substrate tRNA for DNMT2/TRDMT1 and consequent association between somatic mutation and disease occurrence.

Materials and methods

Multiple sequence alignment and phylogenetic tree construction

The protein sequences of 35 DNMT2/TRDMT1 homologues were searched from NCBI database. The multiple sequence alignment was performed using the identity matrix of ClustalW programme in CLUSTAL X (2.0.12) window interface. Alignment parameters were set under default. The sequence identity and similarity of aligned sequences were calculated by Ident and Sim programmes in Sequence Manipulation Suite server [58],and their mean values were calculated in Microsoft Excel 2010. We used the Bayesian inference method (BI) and the GTR replacing model to construct the phylogenetic tree of DNMT2/TRDMT1 homologues via MrBays 3.1 software. The phylogenetic tree was beautified through Evolview server.

Structural modeling of human tRNA and DNMT2/TRDMT1

According to our previous methods [9], the tertiary structures of human wild-type tRNA^Gly-GCC, G34 or C35 point mutant and two base pair (28UC-41GG)-deleted mutant were modelled using tRNAscan-SE 2.0 and RNAComposer. Due to human DNMT2Δ47 crystal structure lacking residues 79−96 and 189−247 (PDB ID: 1G55), the intact DNMT2/TRDMT1 structure was modelled by the protein homology modelling Swiss modeller server. Later, this predicted structure was subjected to MD simulations about 8 ns through GROMACS 4.6 package. Finally, the optimized DNMT2/TRDMT1 structure was validated using SAVES v4.0 server, including ERRAT, VERIFY 3D and PROCHECK programmes.

Molecular dynamics (MD) simulations

The point mutations were made in the DNMT2/TRDMT1 structure by Moe 2008.10. Subsequently, the mutant structure was placed in the truncated cubic box centre and solvated in SPC/E water. A total of 4−7 Na⁺ ions were also added to neutralize the negative charges in the system using the genion programme. In the dynamics protocol, all hydrogen atoms, ions and water molecules were first subjected to the energy minimization about 1000.0 kJ mol⁻¹ nm⁻¹ by the steepest descent, to remove van der Waals short contacts. After that, the position restriction associated with the NVT and NPT equilibrium was carried out. In the NVT equilibrium, the constant temperature was set to 300 K, the coupling time constant was 0.1 ps and the duration time was 100 ps. The speed of each step conformed to the distribution arrangement of Max well-boltzmann. Following this, NPT equilibrium was performed using the constant pressure of 1.0 bar, the coupling time constant of 2.0 ps and the duration time of 100 ps. In the NVT and NPT equilibrium, the Berendsen algorithm was utilized for temperature coupling. In NPT equilibrium, Parrinello-Rahman was employed for pressure coupling. The Lincs algorithm was used for covalent bond restriction, while the particle mesh Ewald (PME) method was applied for the long-range electrostatic analysis. MD simulations and result analysis were performed on a quad-core 3.30 GHz Xeon (E3–1230 V2) desktop computer. The simulative convergence was analysed in terms of RMSD from the initial structure. GROMACS utilities and Visual Molecular Dynamics (VMD) programme were utilized to visualize and analyse structural files and trajectories.

The tRNA^Gly-GCC molecule was docked to the DNMT2/TRDMT1 structure by HDOCK server [15]. To accelerate the MD computation, the sequence 5’ ACCA 3’ was deleted from the acceptor-arm end of tRNA. To obtain the DNMT2/TRDMT1-DNA stucture, we superimposed the modelled structure with the M. Hhal-DNA-SAH crystal structure (PDB ID: 1MHT), and then deleted M. Hhal and SAH. Following these, the DNMT2/TRDMT1-tRNA or DNMT2/TRDMT1-DNA structure was subjected to MD simulations using GROMACS version 4.6 with the Charm27 force field for about 200 ns. The complex structure, placed in the truncated cubic box centre, was solvated with TIP3P water and 100 mM NaCl. Na⁺ ions were also added using the genion programme to neutralize the negative charge density in this system. We needed to restrain the ligand (tRNA or DNA) whenever the protein was restrained. So, an index group was created for the ligand that contained only non-hydrogen atoms. DNA or tRNA was grouped with the protein for the purpose of temperature coupling. In the same way, Na⁺ and Cl⁻ ions were considered as parts of the solvent. Finally, MD protocol was performed as described in DNMT2/TRDMT1 mutant simulations.

Prediction of physicochemical characteristics of DNMT2/TRDMT1 and its mutants

The binding affinity of tRNA to DNMT2/TRDMT1 or M. HhaI was predicted using PredPRBA web server [59]. DNMT2/TRDMT1-tRNA binding affinity changes after enzyme mutations were predicted by mmCSM-NA web server [16]. The binding affinity of DNA to DNMT2/TRDMT1 or M.HhaI was predicted via PreDBA web server [60]. Protein stability changes after DNMT2/TRDMT1 mutations were predicted through mCSM, SDM and DUET in DUET web server [61], and INPS-3D and INPS sequence in INPS-MD server [62]. The relationship between these mutations and disease development was strictly predicted using FATHMM-XF, FATHMM-MKL, CScape, Cancer, CanSavPre, PROVEAN and PhD-SNP [18–21].

Gene Ontology (GO) term enrichment analysis

Human proteins with poly-Asp, Gly or Val sequence were retrieved via Scansite 4.0 web server under the ‘sequence pattern (s)’ search method using ‘Ensembl Human’ protein data source. GO term enrichment analysis of genes (encoding proteins containing poly-Asp, Gly or Val sequence) was performed by GOnet web server [63]. The GO namespace parameter was set to cellular component, molecular function and biological process. GO term enrichment was selected as the analysis type. For enrichment analysis options, the q-value threshold was set ≤ 0.05 and all the annotated genes were as the background.

Expression and purification of DNMT2/TRDMT1, DNMT3L and T7 RNA polymerase

DNMT2/TRDMT1 and T7 RNA polymerases were expressed and purified as described in our previous procedures [9]. The QuikChange site-directed mutagenesis kit (No.200516, Stratagene, USA) was used to make amino acid mutations in pET28a-DNMT2/TRDMT1. The DNMT3L DNA sequence was amplified from pOTB7-hDNMT3L plasmid (No.BC002560, Proteintech, USA) and then was inserted between NheI and XhoI sites of plasmid pET28a. Subsequently, point and truncation mutants of DNMT2/TRDMT1 and pET28a-DNMT3L were verified by DNA sequencing. The primer sequences used were exhibited in Table S1. Finally, the expressions and extractions of DNMT2/TRDMT1 mutants and DNMT3L were performed according to the previous methods [9].

Circular dichroism (CD) measurements

To determine secondary structure changes of wild-type DNMT2/TRDMT1 and its mutants, CD spectra in the Far-UV region (190–250 nm) were collected at 25°C using the J-1500 spectrometer (Jasco, Japan). A total of 300 μL 0.20 mg/mL protein sample in 10 mM phosphate buffer (pH 7.4) was added to a quartz cuvette with 0.1 cm optical path, and measured at the scan rate of 100 nm/min and the bandwidth of 1.0 nm. After baseline correction using the phosphate buffer, each spectrum was expressed by the average from five successive accumulations and then was smoothed via Origin 7.5.

Methyltransferase activity assay

Primers used for synthesizing human tRNA^Gly-GCC and tRNA^GlyDelta-ACS DNA template were displayed in Table S1. According to our previous methods [9], DNA template and tRNA transcript were prepared and validated, followed by their concentration measurements. Based on the preceding procedure [9], the activities of 1 μM DNMT2/TRDMT1, and its point and deletion mutants on 2 μM tRNA were, respectively, determined in the 1× methylation buffer, containing 100 mM Tris-HCl (PH 7.5), 100 mM KCl, 4 mM DTT, 6 mM MgCl₂ and 0.1 mM EDTA. To investigate DNA methylation induced by DNMT2/TRDMT1, 8 μM DNA was catalysed by this 4 μM enzyme or its deletion mutants in the above buffer. Then, 1 μM DNMT3L was supplemented into this system, to determine whether it could promote the DNA methylation activity of DNMT2/TRDMT1. Finally, 2 μM DNA or tRNA was methylated by 1 μL M.HhaI (25000 units/mL, No. M0217S, NEB, USA), which was considered as a positive or negative control. In the reaction system, tRNA transcript or DNA was displaced by TE buffer, which was regarded as trial control. The background value was determined by the scintillation solution. Trial values were calibrated by control and background values. Three independent trial data were indicated as mean value ± standard deviation (SD) using GraphPadPrism.v8.0.

Funding Statement

The enzyme assay was supported by the National Natural Science Foundation of China (Grant number: 31370801).

Disclosure statement

No potential conflict of interest was reported by the author(s).

Author contributions

H.L. conceived and supervised the project. H.L., D.Z., Y.Y., Y.M., Y.C., P.X., J.C., M.Q., D.X., C.C. and H.C. undertook the validation and the investigation. H.L. and D.Z. took charge of the data curation and analysis. H.L. wrote and revised the paper. All authors substantially contributed to and approved the final manuscript.

Data availability statement

The data presented in this study were not publicly available but could be requested from the corresponding author.

Supplementary material

Supplemental data for this article can be accessed online at https://doi.org/10.1080/15476286.2023.2272473