Light‐Activation of DNA‐Methyltransferases

Abstract 5‐Methylcytosine (5mC), the central epigenetic mark of mammalian DNA, plays fundamental roles in chromatin regulation. 5mC is written onto genomes by DNA methyltransferases (DNMT), and perturbation of this process is an early event in carcinogenesis. However, studying 5mC functions is limited by the inability to control individual DNMTs with spatiotemporal resolution in vivo. We report light‐control of DNMT catalysis by genetically encoding a photocaged cysteine as a catalytic residue. This enables translation of inactive DNMTs, their rapid activation by light‐decaging, and subsequent monitoring of de novo DNA methylation. We provide insights into how cancer‐related DNMT mutations alter de novo methylation in vivo, and demonstrate local and tuneable cytosine methylation by light‐controlled DNMTs fused to a programmable transcription activator‐like effector domain targeting pericentromeric satellite‐3 DNA. We further study early events of transcriptome alterations upon DNMT‐catalyzed cytosine methylation. Our study sets a basis to dissect the order and kinetics of diverse chromatin‐associated events triggered by normal and aberrant DNA methylation.

Introduction 5-Methylcytosine (5mC,F igure 1a)i sadynamic regulatory element of mammalian genomes with key roles in transcription regulation, differentiation and development. [1] 5mC is introduced and maintained by DNA-methyltransferases (DNMT), and DNMT mutations are an early event in carcinogenesis. [2] Thea bility to control DNMTs with spatiotemporal resolution in vivo would enable precise kinetic insights into how cancer-related mutations alter DNMT function. Moreover,e xperimental control of 5mC levels at user-defined genomic loci is of central interest in epigenome engineering,a nd spatiotemporal resolution would greatly enhance the precision of this approach. [3] Compared to traditional small molecule effectors or transcriptional/translational control, light can offer direct control of protein functions with spatiotemporal resolution Figure 1. Light-control of DNA methyltransferase catalysis. a) Mechanism of DNMT-catalyzed cytosine methylationa tC5. SAM:S -adenosylmethionine;S AH:S -adenosylhomocysteine. b) Crystal structure of human DNMT3a active site with catalyticC 710 trapped with the cytosine analog zebularine (Z, pdb 6F57 [11] ). c) Structure and lightdeprotection of photocaged cysteine 1.d)Domain and crystal structure (pdb 6F57 [11] )o verviewso fpcDNMT3a3Lwith position of catalytic C710!1 mutation for light-activation. NLS:nuclear localizations equence. HA:hemagglutinin tag. e) Incorporation of 1 at pcDNMT3a3L amber codon (C710TAG) in HCT116 DKO cells analyzed by anti-HA immunostaining and flow cytometry (FCM). Error bars show standard deviations from three independentbiological replicates. f) Anti-5mC immunostainings and FCM analysis of HCT116 DKO cells expressing pcDNMT3a3L 24 hafter light or no light irradiation. Error bars show standard deviations from three independent biologicalr eplicates. and unparalleled tunability,given that suited light-responsive tools are available. [4] In this direction, caged 5-aza-2'-deoxycytidine (5Aza-dC) derivatives have enabled light-control of DNMT inhibition. [5] However,5Aza-dC itself is not selective for specific cytosine-directed DNMTs,a nd it acts by nonspecific incorporation into DNAand subsequent formation of DNMT-DNA-crosslinks,l eading to DNAd amage. [6] Moreover, reduction of 5mC levels occurs via passive 5mC dilution, which limits temporal resolution. Alternatively,l ight-activated recruitment of DNMT3a-cryptochrome 2p rotein (Cry2) fusions to specific target loci has been demonstrated via dimerization with the cytochrome-interacting helix-loophelix protein 1( CIB1) fused to DNAb inding domains. [7] However,t he use of such fusion constructs is restricted to methylation of specific loci, and it does not offer studying natural enzyme functions.Further, the mere recruitment of an overexpressed, constitutively active DNMT is expected to create ahigh methylation background.
We here report light-control of DNMTs by directly masking an essential catalytic residue via substitution with ag enetically encoded [8] photocaged amino acid. [9] This provides ageneral approach for the expression of transiently inactive wild type (wt) and mutant DNMT isoforms-as single enzymes and fusion constructs-followed by their activation with light. Using these tools,westudy how cancerrelated mutations alter de novo methylation by aD NMT3a construct in vivo.W ef urther demonstrate sequence-specific and tuneable methylation of satellite-3 (SATIII) DNAb y light-controlled, programmable DNMT-transcription activator-like effector (TALE) fusions,p roviding as trategy for advancing the control and precision of methylome engineering tools.F inally,w es tudy early events of transcriptome reshaping by light-activated DNMT3a methylation.

Results and Discussion
DNMTs activate the cytosine nucleobase for attack of the S-adenosylmethionine (SAM) cofactor methyl-group via 1,4addition of acysteine (C710) sulfhydryl group at the C6 atom followed by protonation of the N3 atom ( Figure 1a,b). [10] To control the nucleophilicity of C710 in al ight-dependent manner,wetargeted it for replacement with the noncanonical amino acid (ncAA) 4,5-dimethoxy-2-nitrobenzyl-L-cysteine 1 (Figure 1c). [9b,c] ncAA 1 can be incorporated into proteins in response to the amber codon (TAG)inyeast and mammalian cells by use of an Escherichia coli amber suppressor leucyl-tRNA-synthetase (LRS)/tRNA Leu pair with engineered LRS aminoacylation and editing sites.Subsequent light-irradiation of 1 leads to effective decaging to the natural cysteine residue (Figure 1c). [9b,c] We constructed av ector encoding the catalytic domain (CD,a a6 12-912) of human DNMT3a (which is responsible for de novo methylation) fused to its activating interaction partner DNMT3L [12] with aC -terminal HA-tag.T he CD contains an amber codon at position C710 (Figure 1d). For the expression of the LRS/tRNA Leu pair, we employed asecond vector that allows for amber suppression with 1 with high efficiencya nd fidelity (confirmed in aw idely used mCherry-GFP-39TAG control;s ee Figures S1 and S4 for imaging and electrospray ionization (ESI)-MS/MS characterization). We co-transfected both vectors into HCT116 colon cancer cells that lack functional DNMT1 and DNMT3b genes (HCT116 DKO). These cells exhibit < 5% of the global 5mC level found in wild type HCT116 cells. [13] After cultivation for 48 hi np resence or absence of 0.05 mM 1,w ef ixed and permeabilized the cells,a nd analyzed both populations by anti-HA immunostaining and flow cytometry (FCM). We observed expression of the full-length photocaged DNMT_C710!1 protein construct (briefly: "pcDNMT3a3L") only in presence of 1 (~50 %c ompared to non-amber DNMT), indicating high efficiencyand fidelity of its incorporation at the target position ( Figure 1e,f or characterization of amber suppression by aw estern blot, see Figure S5). To evaluate the light-responsiveness of pcDNMT3a3L, we extended the FCM analyses to 5mC immunostainings (this assay provides sufficient sensitivity to measure the presence or absence of methylation by nonamber wt or catalytically inactive mutant (E756A [14] )DNMT in the HCT116 DKOb ackground;F igure S9). When we expressed pcDNMT3a3L for 48 h, removed 1 and irradiated the cells for 5min with light (365 nm), we observed a5 mC signal only for the irradiated cell population (Figure 1f and Figure S7). This shows that pcDNMT3a3L is indeed translated in an inactive state,a nd that it can be effectively activated with light.
DNMT3a is by far the most frequently mutated DNMT in haematological malignancies,b ut the effects of such mutations on its in vivo catalytic activity are still poorly understood. [2,15] In vitro studies with synthetic DNAs as well as studies relying on mammalian expression of DNMT3a mutants and 5mC quantification have revealed ac omplex picture of cancer mutations,w ith both activity increases or decreases being possible. [2,11] However,i nv itro studies have limited relevance for enzyme functions under physiological conditions,and the expression of constitutively active mutants does not allow studying the kinetics of DNMT3a catalysis uncoupled from the kinetics of upstream processes,s uch as DNMT3a translation, folding, posttranslational modification and transport.
We employed pcDNMT3a3L as wt or bearing frequent cancer-associated DNMT3a mutations in HCT116 DKOcells using light-activation after 48 h. At this point, expression of wt and mutant DNMTs had reached aplateau of similar level (Figure S10), and 1 was removed from the medium to terminate the expression.
We chose arange of mutations including several ones that had recently been studied in vitro in the specific context of DNMT3a3L. [11] Though av ariety of differences between in vitro and in vivo experimental conditions preclude aq uantitative comparison of individual mutant activities,t he similar DNMT constructs used in the in vitro study makes it the best available reference for comparisons between the two conditions.The mutations had different locations in this complex: DNMT3a and DNMT3L form at etramer with two central DNMT3a molecules interacting via the homodimer interface, and two DNMT3L, each interacting with one DNMT3a via the opposite face. [11] TheD NMT3a catalytic domain harbors Angewandte Chemie Forschungsartikel ten highly conserved motifs (I-X, Figure 2a)that are involved in SAM binding (I and X) and catalysis (e.g.IVand VI). DNA binding of DNMT3a is mainly mediated by the target recognition domain (TRD) loop (aa 831-848), the catalytic loop (aa 707-721, containing C710), and the homodimer interface (partly constituted by aa 881-887, Figure 2a,b). [11] Them ost frequently mutated residue of DNMT3a in cancers is R882, which is located at the homodimer interface, and undergoes as tabilizing interaction with TRD loop residue S837 (Figure 2b). [16] In preliminary studies,wefound that the most prevalent mutants R882H and R882C cannot form 5mC to ad etectable level when expressed in constitutively active form ( Figure S9). We observed the same for mutations of T835 and W893 that interact with the CpGbridging phosphate and the SAM cofactor, respectively ( Figure S9). We then conducted activity measurements with light activated DNMT via assessing 5mC formation 24 hafter light irradiation. At this time,5mC formation is still dynamic, offering relative activity comparisons between the mutants studied in our experiment (see Figure 2c). We first studied the isosteric mutation S714C that is located in the catalytic loop and is expected to selectively delete ah ydrogen bond to the CpG 5'-phosphate by ad efined O!Sa tomic substitution (Figure 2b). This mutant exhibited~50% 5mC signal compared to wt pcDNMT3a3L after light activation, being in asimilar range as observed for the DNMT3a CD in vitro but different to observations in murine ES cells,w here this mutant is inactive (Figure 2c). [11,17,18] We next studied mutation R736H, involved in stabilizing the active site fold by ah elix-helix interaction (Figure 2b). Though located at the interface to DNMT3L, R736H does not strongly affect the activation by DNMT3L in vitro. [17] Conflicting results have been reported with DNMT3a CD for different substrates in vitro,ranging from 4-fold increased to completely abolished activity. [17,18] Our light activation measurements revealed ar eduction in activity to~60 % ( Figure 2c). Another frequently mutated residue of the same interface is R771, being engaged in asalt bridge to DNMT3L (Figure 2b). Thet wo cancer mutations R771Q and R771G show markedly increased in vitro methylation (7-and 3-fold higher than wt, respectively), and no or only slight reduction in DNMT3L activation. [17] Surprisingly,w eo bserved the opposite:R 771Q slightly reduced, and R771G enhanced methylation 1.5-fold compared to wt DNMT.Finally,wewere interested in the mutation R836A located in the TRD loop that interacts with the target CpG-guanine via aw atermediated hydrogen bond (Figure 2b;D NA interaction not shown). [11] R836 is involved in cooperative DNAb inding by DNMT3a multimers. [19] We observed no alterations for the R836A mutation itself,b eing in agreement with in vitro and cellular studies (Figure 2c). [11] Overall, our light activation studies afforded in several cases surprisingly different activity data than studies conducted in vitro or with constitutively active mutants in other cell contexts would suggest. We thus expect that af uture extension of our light activation technique to time-resolved studies with such mutants and using full-length DNMT3a will be instrumental to further dissect the exact nature of their malfunction.
Thea bility to selectively deposit 5mC in user-defined genomic sequences is ak ey goal in epigenome engineering. This enables establishing causalities between local 5mC and various other chromatin features,i ncluding transcriptional activity. [3] Compared to the current approach for light control of local methylation based on Cry2-recruitment of constitutively active DNMT to ap rogrammable DNAb inding domain, [7] adirect catalysis control of programmable DNMT constructs by light would enable locus-specific DNAmethylation with enhanced spatiotemporal control and lower propensity for background methylation. Moreover,t he amount of locally deposited 5mC may be precisely tuned via the light dosage.W ed esigned ac onstruct of at ranscription-activator-like effector (TALE) domain targeted to as equence of SATIII pericentromeric DNAf used to pcDNMT3a3L ("SATIII-pcDNMT", Figure 3a, [20] Figure S20 shows vector map). SATIII DNAisthe origin of nuclear stress bodies (nSB,atype of stress-induced membrane-less organelle [21] ), and exhibits aberrant methylation in several cancers. [22] We confirmed high efficiencya nd fidelity of amber suppression with 1 for this construct (Figures S2 and S3). We then evaluated SATIII-pcDNMT for correct, selective genomic localization by expressing it in HEK293T cells and subjecting the cells to heat- stress (1 h, 44 8 8C). We imaged the cells after conducting aco-immunostaining for the HA-tag of SATIII-pcDNMT and for endogenous HSF1, at ranscription factor recruited to SATIII upon heat stress.C o-localization indicated correct target binding of SATIII-pcDNMT (Figure 3b and Figure S12). In vivo measurements of DNA methylation catalysis by pcDNMT3a3L with frequent cancer-associated DNMT3a mutations using light-control. a) Domain structure of pcDNMT3a3L with conserved DNMT motifs I-X and mutation positions analyzed in this study.C atalytic loop in green, TRD loop in yellow,part of homodimer interface in orange. b) Crystal structure of human DNMT3a active site bound to DNA (pdb 6F57 [11] )with mutation positions analyzed by light activation shown as blue sticks. DNMT3L in pink, other color codes as in (a). c) Activity studies of pcDNMT3a3L cancer mutants conducted by anti-5mC immunostainings and FCM. Measurements were conducted 24 hafter light or no light irradiation.A tt his point, 5mC formation has not reached saturation for the wt (e.g.,compare to R771G reaching even higher 5mC level). Error bars show standard deviations from three independentbiological replicates.

Forschungsartikel
We next irradiated the cells with light for 5min and conducted time-resolved 5mC quantification at aS ATIII CpG by bisulfite PCR and Illumina amplicon deep sequencing (see SI for details). We observed arapid increase of 5mC over the first 10 h, followed by as lower increase up to 24 h (Figure 3c,see Figure S13 for additional controls).
This increase was strictly light-dependent, indicated by the 5mC levels of the non-irradiated samples that did not exceed the level of an untransfected control (Figure 3c). Moreover,w eo bserved as imilar result for as econd CpG in the amplicon ( Figure S14).
To evaluate the locus-specificity of methylation by SATIII-pcDNMT,w ec onducted bisulfite PCRs of SATIII and two off-target loci in the BRCA1 and p16 genes with DNAo fc ells harvested 24 ha fter light. Sanger sequencing revealed methylation of SATIII, but not BRCA1 or p16, showing locus-specific methylation by SATIII-pcDNMT (Figure 3d and Figure S15). In turn, we could effectively methylate both off-target loci with dedicated, BRCA1 and p16targeting TALE-DNMT constructs,i ndicating that the absence of off-target methylation by SATIII-pcDNMT is due to high targeting selectivity,a nd not to generally inaccessible chromatin states at these loci (Figure 3e;for an evaluation of the methylation window of ar epresentative TALE-DNMT construct, see Figure S16). Finally,wewere interested, if local perturbation of 5mC by SATIII-pcDNMT could be precisely tuned by applying different light dosages.I ndeed, we observed aclear dependence of 5mC levels on the irradiation time up to 10 min (Figure 3f;a fter this time,t he majority of 1 is decaged in in vitro reference experiments,s ee Figure S17). This dosing ability may provide aconvenient means for controlling the strength of 5mC-triggered effects in local chromatin, such as the recruitment or release of 5mCresponsive DNAb inding factors or transcriptional activity of alocus.
5mC is an important regulator of chromatin condensation, with transcriptional downregulation as an ultimate biological consequence. [1] To evaluate,ifour light-activated methylation would enable revealing early events in system-wide transcriptional downregulation, we expressed wt pcDNMT3a (CD without DNMT3L) in HEK293T cells and activated it with light. We then measured mRNAe xpression levels on the transcriptome-wide level by mRNA-Seq at zero,f our and eight hours after activation. We observed significant lightdependent changes in transcription levels of alarger number of genes both after 4a nd 8h (Figure 4s hows representative examples;see Figure S18 and SI tables for full data analyses). Thev ast majority of genes thereby showed downregulation, albeit with different time profiles.Asmall group of fifteen genes was downregulated already after 4h,i ncluding regulators involved in transcription activation, histone modification, and cell proliferation (e.g., CASZ1, SIK1, MAR-VELD1). Interestingly,the majority of these genes was again upregulated after 8h (Figure 4, top left block). Thel argest group (73 genes) however showed as omewhat slower onset, with significant downregulation observed after 8h.Strikingly, this group included an umber of factors involved in central processes of signal transduction, chromatin regulation and RNAp rocessing. Thef ormer included RABL3, DENND4C (a RAB10 activator [23] )a nd CHM (a subunit of the Rab GGTase complex [24] ). Chromatin factors with direct, general roles in transcription were MED12L that is part of the mediator complex involved in activation of most RNAPol IItranscribed genes [25] and SETX, am odulator of RNAP ol II interactions with chromatin. [26] We also observed an early downregulation of ELP4, being part of ahistone acetyltransferase complex associated with RNAP ol II. [27] Other factors had general roles in chromatin regulation, such as CHD1, an ATP-dependent chromatin remodeler,S MYD2, am ethyltransferase of H3K4me3, Rb1 (involved in maintenance of heterochromatin via recruitment of e.g.,S UV39H1), MYSM1, ad eubiquitinase (DUB) of histone H2A, RCHY1, au biquitin ligase of HDAC1, and PRKAA2, ak inase targeting histone H2B and HDAC5. [28] Strikingly,s everal of Figure 3. Light-controlled, targeted methylation of SATIII loci. a) Domain structure of SATIII-pcDNMT construct. b) Images of heatstressed HEK293T cells expressing SATIII-pcDNMT immunostained with anti-HA and anti-HSF1 antibodies. Scale bar is 5 mm. c) Kinetics of SATIII methylation with SATIII-pcDNMT after light irradiation and no light controls. 5mC was quantified by bisulfite PCR and Illumina amplicon deep sequencing. Error bars show standard deviations from three independentbiological replicates. UT = untransfected + light. d) Methylation of SATIII and two off-target loci by SATIII-pcDNMT 24 h after irradiation analyzed by Sanger sequencing of bisulfite PCR products. gDNA sequencei sshown below with Cs bold and CpG underlined. e) Methylation of the BRCA1 and p16 loci of (d) with TALE-DNMT constructs designed to specifically target the BRCA1 and p16 loci, respectively.f)Light dosage dependence of SATIII methylation in experiments conducted as in (c). 5mC was quantified by bisulfite PCR and pyrosequencing. Error bars show standard deviations from three independentbiological replicates. UT = untransfected. the factors are also direct modifiers of p53 (SMYD2, PRKAA2 and RCHY1). Factors involved in RNAb inding, processing and transport were REXO2, RBM18, PTBP2, DCP2, RANBP6, SCAF11, PA BPC4L, SREK1, and TPR ( Figure 4).
These experiments show that downstream effects of de novo methylation can be studied for our light-activatable DNMT on the system-wide level. Fort he here employed model, we obtain insights into the initial target genes of pcDNMT3a and the early regulatory output of its methylation activity on the transcriptome level. Theo bserved downregulation of alarge number of genes with general, not genespecific regulatory functions suggests that the early DNA methylation events in this model have ah igher-order role in the regulation of downstream chromatin-associated events.

Conclusion
In conclusion, we report light-control of DNMT catalysis in cells by replacing ac atalytically essential cysteine residue with agenetically encoded photocaged cysteine.This enables the expression of individual, wild type or mutant DNMTs in atransiently inactive state followed by their rapid activation. Given their highly conserved mechanism, we anticipate that this strategy is transferrable to other DNMTs,a sw ell. We reveal the impact of cancer-associated mutations on the catalytic activity of DNMT3a3L directly in an intracellular, nuclear environment. We further demonstrate light-control of ap rogrammable DNMT-TALE fusion construct, offering selective methylation of user-defined DNAs equences,a nd enabling the deposition of defined 5mC levels via tuning the light dosage.F inally,w ed eliver first insights into the early events of transcriptome reshaping by light activated de novo methylation. Future studies will focus on dissecting the rate and order of early chromatin-associated events that follow normal and aberrant DNAm ethylation by individual fulllength and genome-integrated DNMTs.T his will help to understand how the regulation and perturbation of DNMT functions ultimately control the formation of cellular phenotypes during development and malignant transformation.  Figure S18) as log2-fold change against average expression over all conditions; differential light-dependent expression at af alse discovery rate of < 0.05.