Development of m⁶A Readers for Enhanced Recognition In Vitro via Genetic Code Expansion

Abstract

The YTH domain was recently demonstrated as a promising tool for recognizing and regulating m⁶A-modified RNAs. However, the deficiency in binding affinity (micromolar range) and selectivity (off-target binding) for m⁶A of the wild-type YTH domain limit its sensitivity and accuracy for m⁶A recognition. Engineering strategies for improving the affinity and selectivity of such epitranscriptomic readers remain lacking. Here, we demonstrate that site-specific introduction of noncanonical tryptophan derivatives within the aromatic m⁶A-binding pocket of the YTH domain modulates the binding affinity and selectivity. Specifically, we report two YTH variants, with 5-bromotryptophan and 5-hydroxytryptophan incorporated at the W465 of the YTH domain of YTHDF1, which show improved binding affinity and selectivity against m⁶A-modified RNAs. DFT-based energy decomposition analysis indicates that the binding affinity can be modulated through the substitution group’s effect on the dispersion force in the van der Waals recognition complex, suggesting a critical role of the CH–π interaction in m⁶A-reader recognition. This work reports new reader domains with improved m⁶A recognition and demonstrates the potential of m⁶A reader engineering via genetic code expansion.

Graphical Abstract

INTRODUCTION

N⁶-methyladenosine (m⁶A) has been discovered as an abundant modification in mammalian messenger RNAs (mRNAs) since the 1970s;^1,2 however, the physiological relevance of m⁶A in mRNAs was not fully appreciated until the development of the transcriptome-wide mapping technology of m⁶A in 2012 (“m⁶A-seq”³ and “MeRIP-Seq”⁴), which revealed the conserved occurrence of m⁶A in the consensus sequence DRm⁶ACH (D = A/G/U, R = A/G, and H = A/C/U) in transcriptomes of mice and humans. Importantly, Jia et al.⁵ and Zheng et al.⁶ showed that the m⁶A in mRNAs can be removed by eraser proteins FTO and ALKBH5. These findings suggest that m⁶A in mRNAs is reversible and can be regulated upon altered activities of erasers.^7,8 Growing evidence suggests m⁶A in mRNAs and non-coding RNAs (ncRNAs) plays vital roles in regulating RNA stability,^9,10 translation,^11,12 and transcription;^13,14 dysregulations of m⁶A effector proteins have profound impacts in numerous human diseases, including cancers,^15–18 neurological disorders,¹⁹ and viral infections.²⁰

Chemical biology tools that selectively recognize m⁶A play critical roles in unraveling m⁶A’s occurrence and biological functions. For example, anti-m⁶A antibodies were used to enrich m⁶A-modified RNA fragments in vitro by immunoprecipitation and served as key reagents in the transcriptome-wide m⁶A mapping methods such as m⁶A-seq,³ MeRIP-seq,⁴ and miCLIP.²¹ The YTH domain, an endogenous reader protein for recognizing m⁶A-modified RNA, was recently used as a synthetic biology building block and fused with other functional domains such as the APOBEC1 protein to recognize and modulate m⁶A-modified transcripts inside cells.^22–24 Despite the fact that the YTH domain is promising as a tool for m⁶A recognition, wild-type YTH domains (i.e., five homologous YTH domains found in mammalian cells) exhibit suboptimal binding affinities and selectivities for m⁶A. For instance, the affinities between YTH domains and m⁶A-modified RNA range from K_d = 0.2 to 1.7 μM,^25,26 which are significantly weaker compared to typical RNA-binding proteins used for biological applications.^27,28 Furthermore, the wild-type YTH domains can be subjected to off-target binding to unmodified transcripts. For example, PAR-CLIP experiments performed on endogenous YTHDF2 and YTHDF1 proteins enriched a large number of RNA sequences that do not overlap with m⁶A-containing transcripts.^9,11 The weak binding affinity and limited selectivity of the wild-type YTH domains can hinder their utility as readers when used to recognize m⁶A-modified RNAs with diverse sequence contexts in the transcriptome.

Protein engineering provides powerful strategies for developing specific binding proteins with improved binding affinities and selectivities for specific substrates, including RNAs^27,29 and proteins carrying post-translational chemical modifications such as methylated lysine in histones.^30,31 However, strategies for engineering readers for m⁶A-modified RNAs remain lacking. The YTH domain recognizes the N⁶-methyl group of m⁶A through a conserved aromatic binding pocket enriched in tryptophan residues.^25,32–34 The high-resolution crystal structure of the YTH domain from the human YTHDF1 shows that the m⁶A residue flips out of the stacked bases in the single-stranded RNA and is closely surrounded by three tryptophan residues W411, W465, and W470 in the YTH domain (Figure 1A).²⁵ Mutation of any one of the three tryptophans to an alanine abolishes the binding to m⁶A RNA.²⁵ Improving the selective recognition of a methyl group-modified nucleobase within a large and poly-anionic RNA molecule can be challenging. The recognition complex of the YTH domain and m⁶A RNA involves favorable interactions, including hydrogen bonding, electrostatic interactions, van der Waals (vdWs) interactions such as π–π stacking and CH–π interaction, and the hydrophobic effect.^25,33 Notably, most of the favorable interactions between the YTH domain and modified RNA are not specific to the N⁶-methyl group, except for the CH-π interaction. We reason that tuning the strength of the N⁶-methyl-specific interaction is a promising avenue to modulate the recognition of the methyl group by the YTH domain. However, the tryptophan residues within the binding pocket represent the largest and most electron-rich aromatic system among the natural amino acids;³⁵ to further strengthen the CH–π interactions, we pursued the incorporation of tryptophan derivatives into the YTH domain using the genetic code expansion (GCE) technology.^36–38 GCE uses engineered aminoacyl-tRNA synthetase/tRNA pairs to enable co-translational, site-specific incorporation of diverse unnatural amino acids (UAAs) into proteins in living cells in response to a repurposed nonsense codon.^37,39,40 Interestingly, the incorporation of tryptophan derivatives such as 6-methoxytryptophan and other UAAs has been reported to improve the binding affinity and selectivity of the chromodomain proteins for recognizing methylated lysine residues by tuning the cation-π interaction.^31,41 To the best of our knowledge, analogous tuning of CH–π interactions in protein-RNA recognition using UAAs has not been explored before. Here, we report the site-specific incorporation of three tryptophan derivatives into the m⁶A-binding pocket of the YTH domain of human YTHDF1 (i.e., “wtYTH”) using GCE (Figure 1A). Incorporation of 5-bromotryptophan (5BrW) and 5-hydroxytryptophan (5HTP) at the W465 position significantly increased the binding affinity and selectivity against multiple m⁶A-modified RNA oligonucleotides compared to the wtYTH. Interestingly, a tandem-YTH domain design further improved the binding affinity to m⁶A-modified RNA, but not the selectivity. We further used DFT-based energy decomposition analysis to explore the energetic contributions by different tryptophan derivatives in a simplified recognition complex formed by the m⁶A and a tryptophan (or a tryptophan derivative), which suggested dispersion force as a dominating contributor to the stabilized interaction energy between W465 and m⁶A by the 5HTP and 5BrW substitutions.

Figure 1.

Recognition between wtYTH and m⁶A-modified RNA. (A) Shown are conserved tryptophan residues in the aromatic m⁶A-binding pocket of wtYTH (left) and the description of reader development by replacing the conserved tryptophan residues with noncanonical tryptophan derivatives (5HTP, 5MTP, and 5BrW) via the genetic code expansion technology (right). (B) Validation of wtYTH as a reader for m⁶A in vitro via the FP binding assay. Representative binding curves are shown for the binding of wtYTH with m⁶A-RNA (in red) or A-RNA (in gray). The measured dissociation constants (K_d) of binding are labeled on the plot, with the averaged value and standard deviations from n = 24 (for m⁶A-RNA) or n = 8 (for A-RNA) total number of replicates (Figures S4a and S5a).

RESULTS

To establish a robust measurement of binding between the YTH domain and m⁶A-modified or unmodified RNA, we expressed the wtYTH with an N-terminal polyhistidine (His) tag in BL21(DE3) cells, followed by purification using Ni²⁺-NTA affinity chromatography.^25,42 The molecular weight and purity of the purified wtYTH were confirmed by LC-MS and SDS-PAGE (Figure S1A,B and Materials and Methods section). We assessed the binding affinities of wtYTH against both m⁶A-modified and unmodified RNA oligonucleotides using the fluorescence polarization (FP) binding assay⁴² (Materials and Methods section). Briefly, the purified wtYTH protein was titrated against a constant concentration of fluorescein (i.e., FAM)-labeled m⁶A-RNA, yielding a specific binding curve with K_d = 0.4 ± 0.1 μM, consistent with reported affinities in the literature²⁵ (Figure 1B). Meanwhile, the wtYTH did not show specific binding to the FAM-labeled A-RNA within the range of titrated protein (0.009–20 μM); therefore, we noted K_d > 20 μM for the wtYTH to the unmodified A-RNA (Figure 1B). We quantified the selectivity of the wtYTH protein to be greater than 50 by taking the ratio of binding affinities against m⁶A-RNA versus A-RNA (i.e., $\text{selectivity}={K_{\text{d}}}^{\text{wtYTH: A-RNA}}/{K_{\text{d}}}^{\text{wtYTH: m6A-RNA}}$). We further explored two YTH variants (W465Y and W465F) by mutating W465 within the pocket to Y or F. Both variants show comparable binding affinities to m⁶A-RNA and A-RNA with the wtYTH (Figure S1c,d), which suggests the binding pocket of wtYTH can tolerate different aromatic residues and reveals limited tunability of affinities within the realm of natural amino acids.

We then pursued site-specific incorporation of the tryptophan derivatives into the “aromatic cage” of the wtYTH using an Escherichia coli-derived tryptophanyl-tRNA synthetase (EcTrpRS)/tRNA^Trp pair in the engineered ATMW-BL21 strain of E. coli^43,44 (Figure 1A). The ATMW-BL21 strain was engineered from BL21 (DE3), which is widely used for recombinant protein expression, by replacing its endogenous EcTrpRS/tRNA^Trp pair with a yeast-derived counterpart. The ‘liberated’ EcTrpRS/tRNA^Trp pair was then re-introduced into the resulting strain as an orthogonal TGA-suppressor, and engineered for charging various tryptophan derivatives at the TGA (opal) nonsense codon^43,44 (Figure 2A). We first individually mutated two of the tryptophan residues (W465 or W470) within the “aromatic cage” of the wtYTH to a TGA codon (Table S1). We then transformed the YTH-465-TGA or YTH-470-TGA plasmid into the ATMW-BL21 cells and induced overexpression of the YTH protein with 1 mM IPTG and 1 mM tryptophan derivative (Figure 2A). We tested three 5-substituted tryptophan analogs that were demonstrated to be compatible with the engineered EcTrpRS/tRNA^Trp_UCA pair, including 5HTP, 5BrW, and 5-methoxytryptophan (5MTP) (Figure 1A).^43,44 We successfully expressed and purified six YTH proteins with a tryptophan derivative incorporated at either the W465 or W470 position and denoted them based on the type of UAA and the position of incorporation: 5BrW465, 5MTP465, and 5HTP465 refer to the wtYTH proteins of which the W465 is replaced by one of the three tryptophan derivatives (Figures 2B and S2). We obtained appreciable amounts of the purified UAA-containing YTH variants, which are sufficient for in vitro characterizations and binding studies. The yields of purified proteins varied among different UAA incorporations, ranging within 2–4 mg for 5BrW variants, 8–32 mg for 5MTP variants, and 8–22 mg for 5HTP variants per liter of cell culture (Table S2). We used whole-protein LC-MS to confirm the incorporation of the targeted UAA in the six purified proteins (Figures 2C and S3). Despite the success in expressing UAA-incorporated YTH domains at W465 and W470 positions, our attempts to introduce tryptophan derivatives at the W411 position resulted in the production of wtYTH, for reasons that remain unclear (Discussion section).

Figure 2.

Expression of the UAA-containing YTH variants. (A) Schematics of using GCE technology for incorporating tryptophan derivatives into wtYTH using the engineered EcTrpRS/tRNA^Trp pair in ATMW-BL21 cells. (B) Shown is the SDS-PAGE gel of purified YTH variants with site-selective incorporation of 5HTP, 5BrW, or 5MTP at 465 or 470 residue positions. (C) Characterization of UAA-containing YTH variants by whole-protein mass spectrometry, where the observed or theoretical mass of a protein is labeled with “o” or “t”, respectively; full MS data of the six proteins are shown in Figure S3.

Next, we assessed the binding affinities of the UAA-incorporated YTH proteins against m⁶A-RNA using the FP binding assay (Materials and Methods section). To ensure reproducibility, we performed binding measurements with at least 2 biological replicates and 2 technical replicates (Materials and Methods section, Figures 3A, S4 and Table S3). Remarkably, 5BrW465 and 5HTP465 exhibited an 8-fold and 2-fold increase in binding affinities toward m⁶A-RNA compared to wtYTH, respectively (Figures 3A and S4). In contrast, the same substitutions at position 470 (5BrW470 and 5HTP470) showed binding affinities comparable with the wtYTH. Notably, the incorporation of 5MTP (5MTP465 and 5MTP470) showed reduced binding affinities (Figure 3A). We then measured the binding of all YTH proteins against the unmodified A-RNA to quantify their binding selectivities (Figure S5). Interestingly, 5BrW465 does not show increased affinity to A-RNA within the range of titrated protein concentration, and thus improves the selectivity for m⁶A by 8-fold compared to the wtYTH (Figure 3B). Meanwhile, 5HTP465 showed approximately 2-fold increased selectivity compared to wtYTH. In contrast, 5BrW470 and 5HTP470 showed similar selectivities to wtYTH (Figure 3B). Notably, 5MTP465 and 5MTP470 not only showed decreased affinities for m⁶A-RNA but also bound unmodified A-RNA with significantly higher affinities compared with wtYTH (Figure S5f), which further reduced their selectivities for m⁶A-RNA (Figure 3B).

Figure 3.

Binding affinity and selectivity of UAA-incorporated readers for m⁶A. (A) Summary of measured K_d values for wtYTH and UAA-containing YTH domains with m⁶A-RNA is shown on the left. Listed are the averaged K_d values with standard deviations measured from n = 4–24 replicates for each binding pair (Figure S4 and Table S3). Binding curves of 5BrW465 (in orange) and wtYTH (in black) against m⁶A-RNA (one replicate) measured by the FP binding assay are shown on the right. (B) Binding selectivities for m⁶A are plotted for wtYTH and each UAA-containing YTH variant. Errors of selectivity are propagated from standard deviations from n = 4–24 replicates (Figures S4, S5 and Table S3). (C) Shown are binding curves of wtYTH (in black) and 5BrW465 (in orange) against m⁶A-RNA-13mer or m⁶A-RNA-27mer measured by the FP binding assay. Averaged K_d with standard deviation measured from n = 4–8 replicates is labeled on each plot (Figure S6 and Table S4). (D) Binding selectivities for m⁶A of wtYTH (gray bars) or 5BrW465 (orange bars) are plotted for three RNA sequence contexts: m⁶A-RNA, m⁶A-RNA-13mer, and m⁶A-RNA-27mer. Errors of selectivity are propagated from standard deviations from n = 4–22 replicates (Figures S6–S8 and Table S4).

Given the significantly improved binding affinity and selectivity of 5BrW465 compared to wtYTH, we further evaluated the binding affinities of both wtYTH and 5BrW465 to m⁶A-modified RNA oligonucleotides with variable lengths and sequence contexts. We measured the binding of YTH domains against a 27-mer and a 13-mer RNA with m⁶A introduced within the GGACA and GGAGG contexts, respectively (Table S1). The m⁶A-RNA-27mer contains a sequence motif (GGACA) within the consensus DRACH motif for m⁶A occurrence in the mammalian transcriptome^3,4 and extended length relative to the 16-mer m⁶A-RNA, whereas the m⁶A-RNA-13mer represents a sequence context (GGAGG) different from DRACH. The wtYTH binds to the m⁶A-RNA-27mer or the m⁶A-RNA-13mer with K_d = 0.22 ± 0.04 or 0.4 ± 0.2 μM, respectively (Figures 3C, S6a, and Table S4), comparable with its affinity to the 16-mer m⁶A-RNA (Figure 3A). This observation is consistent with previous reports that the wtYTH binds to single-stranded m⁶A-modified RNAs in a sequence-independent manner.¹¹ Excitingly, the 5BrW465 shows robustly improved affinities to these two m⁶A-modified RNAs compared to wtYTH, with K_d = 0.07 ± 0.01 μM for m⁶A-RNA-27mer and K_d = 0.07 ± 0.02 μM for m⁶A-RNA-13mer (Figure 3C and Table S4). Both wtYTH and 5BrW465 show weak and nonspecific binding to the A-RNA-13mer, comparable with the 16mer A-RNA (Figures S5a and S7). Notably, both proteins bind to the A-RNA-27mer with improved affinities, with K_d ~ 3 μM for wtYTH and K_d ~ 2 μM for 5BrW465 (Figure S8 and Table S4). As a result, the recognition selectivity of the wtYTH is relatively reduced for the m⁶A-RNA-27mer compared to the 13-mer and 16-mer RNAs. These results suggest that the selectivity of the YTH domain for m⁶A-modified RNAs is sequence-context dependent (Figure 3D). Despite the sequence dependence of the wtYTH’s selectivity, 5BrW465 consistently shows higher selectivity compared to the wtYTH in all three tested RNA sequence contexts, suggesting the incorporation of 5BrW at the W465 position robustly improves the recognition of m⁶A (Figure 3D).

Inspired by the previous work by Zhao et al. showing that the fusion of multiple chromodomains (i.e., reader domain for histone lysine methylation) enhances binding affinities against the methylated lysine,³¹ we constructed a “tandem YTH-465-TGA” vector for expressing the “tandem 5BrW465” protein, where two 5BrW465 domains are connected by a flexible linker (Figure 4A and Table S1). The tandem 5BrW465 was expressed in the ATMW-BL21 cells under conditions similar to those used for 5BrW465 (Materials and Methods section). Not surprisingly, the overall expression level for the tandem 5BrW465 is lower than the single-domain 5BrW465 protein, and the Ni-NTA affinity chromatography alone was not able to produce protein with comparable purity with the single-domain YTH variants; therefore, we added an additional purification step of size-exclusion chromatography to further improve the purity of the protein (Figures S9a, 4B and Materials and Methods section). The final yield of the tandem 5BrW465 protein was 1.2 mg per liter of cell culture (Table S2), which was sufficient for subsequent characterization and binding studies. We validated the mass of the purified tandem 5BrW465 protein via LC-MS (Figures 4B and S9b). Interestingly, the tandem 5BrW465 binds m⁶A-RNA with K_d = 0.03 ± 0.01 μM, ~2-fold higher affinity compared to 5BrW465 and ~13-fold tighter compared to the wtYTH (Figures 4C, S10a and Table S3). However, the tandem 5BrW465 binds the unmodified A-RNA with elevated affinity (K_d = 2 ± 1 μM) compared to the single YTH domain (Figure S10b and Table S3), suggesting that the nonspecific interaction between the YTH domain and RNA can be promoted upon the tandem construct design. Though the tandem 5BrW465 shows the highest binding affinity to the m⁶A-RNA, the selectivity is not significantly better than the wtYTH and is weaker compared to the single 5BrW465 domain reader (Figure 4D). While tandem protein design can enhance binding affinity to m⁶A-RNA, careful examination is essential to avoid compromised selectivity. Future engineering efforts for the YTH domain will be required for reducing binding to unmodified RNAs in broad sequence contexts.

Figure 4.

Design, purification, and binding of tandem 5BrW465. (A) Shown are the design of the tandem YTH-465-TGA plasmid construct and the expression of the tandem 5BrW465 reader. (B) Characterizations of the purified tandem 5BrW465 by 12% SDS-PAGE (left; loaded together with the single-domain 5BrW465 as a control) and by whole-protein mass spectrometry (right). The observed or theoretical mass of a protein is labeled with “o” or “t”, respectively; full MS data of the tandem 5BrW465 are shown in Figure S9b. (C) Shown are the binding curves of wtYTH (in black), 5BrW465 (in orange), or tandem 5BrW465 (in magenta) against m⁶A-RNA measured by the FP binding assay. The averaged K_d values with standard deviations measured from n = 4–24 replicates are labeled on each plot (Figure S10 and Table S3). (D) Binding selectivities for m⁶A of wtYTH, 5BrW465, or tandem 5BrW465 are shown, which are measured on m⁶A-RNA and A-RNA. Errors of selectivity are propagated from standard deviations from n = 6–8 technical measurements (Figure S10 and Table S3).

Our results demonstrate that substituting the key tryptophan residues in the binding pocket with their chemical derivatives can modulate m⁶A recognition, possibly by impacting the energetics of the CH–π interaction within the recognition complex. CH–π interactions occur frequently as a stabilizing force in protein folding,^45–47 protein recognitions of glycans³⁵ and drug molecules,^48,49 in nucleic acids,^50–52 and in nucleic acid-protein complexes.^53–55 Modulations of the strength of CH–π interactions were applied in the rational drug design⁴⁹ and proposed to be critical for developing carbohydrate recognition proteins via protein engineering.⁵⁶ For delving deeper into the energetic nature of interactions between the m⁶A and the different π systems within the recognition pocket of the current YTH variants with diverse tryptophan derivatives, we performed energy decomposition analysis on a simplified model - a vdW “A-B” complex formed by W465 (“A”) and m⁶A (“B”) isolated from the crystal structure of m⁶A RNA and YTH recognition complex (PDB: 4RCJ) (Figure 5A), using quantum mechanical symmetry-adapted perturbation theory (SAPT) (Materials and Methods section). The intermolecular noncovalent interactions in SAPT theory were decomposed into electrostatics, dispersion, induction, and the Pauli exchange terms. To compare the contribution of each energy term to the overall vdWs interaction energy (ΔE_vdW), we used the absolute value of each component’s energy divided by the sum of the absolute values of all components. For the A-B vdW complex, dispersion is the dominant contributor, accounting for approximately 62% of the interaction energy, followed by 22% from the electrostatic component, 9% from exchange, and 7% from induction (Figure 5B). This is consistent with the prior energy decomposition analysis on CH–π interaction, where dispersion is the dominant component,⁵⁷ suggesting the critical role of CH–π interaction in the A-B vdW complex.

Figure 5.

Energy decomposition analysis of m⁶A molecular recognition by W465, 5HTP465, or 5BrW465. (A) Visualization of the optimized molecular geometry of the A-B vdW complex between W465 (noted as “A”) and m⁶A (noted as “B”), isolated from the YTH-m⁶A RNA recognition structure. (B) Shown is the percentage contribution of each energy component in the A-B vdW complex calculated by the energy decomposition analysis. (C) For each energy component, correlative analysis is shown for the calculated relative energy for three vdW complexes where A is W465, 5HTP465, or 5BrW465 (Table S5), against the experimentally measured binding energy for proteins wtYTH, 5HTP465, or 5BrW465 with m⁶A-RNA (ΔG_exp).

To further understand the energetic contributions of the tryptophan derivatives, we prepared two additional “A-B” vdW complexes where “A” is replaced with 5BrW or 5HTP, and “B” remains m⁶A. The 5MTP substitution at W465 was not included in this analysis due to a spotted steric clash between 5MTP465 and R404 upon manual addition of a methoxy group at C5 of W465 in PyMOL (Figure S11a); consistently, the 5MTP465 residue exhibited significant displacement to alleviate the clash with R404 upon energy minimization (Materials and Methods section and Figure S11b). Characterizations of 5BrW465 and 5HTP465 proteins via circular dichroism (CD) spectroscopy and thermal shift assays revealed minimal perturbations of the 5BrW465 and 5HTP465 in the overall secondary structures and stabilities relative to wtYTH. Molecular dynamics simulations of the modeled 5BrW465 or 5HTP465 in complex with m⁶A RNA suggested minimal deviations of the binding core compared to the energy-minimized wtYTH complex (Figure S11c). These results support the relevance of the structure-based calculations (Figure S12). We performed correlation analysis between each computed decomposed energy term for an “A-B” vdW complex where “A” is 5BrW or 5HTP, against the measured binding energy (ΔG_exp) by FP for the corresponding protein binding with m⁶A-RNA (Materials and Methods section). Interestingly, ΔG_exp shows a strong linear correlation with the dispersion component, rather than the other three energy components, suggesting the dispersion component is potentially critical for the enhanced binding by the 5HTP and 5BrW substitutions (Figure 5C). To further understand the dispersion component, we computed the dispersion energy (ΔE_cal) based on the isotropic polarizability and ionization potential of A and B (Materials and Methods section). We observed a strong linear correlation between the SAPT-computed dispersion energy component (ΔE_disp) and ΔE_cal with the slope = 0.77 (Figure S13), suggesting that polarizability and ionization potential are reliable indicators of the dispersion energy in this vdW complex.

DISCUSSION

Despite the availability of a few tools using YTH domains for m⁶A recognition,^22,24,58 the weak binding affinities and limited selectivities of the YTH domains for m⁶A-modified RNAs hinder their applications. Here we report the development of enhanced YTH domains as readers through site-specific incorporation of 5BrW or 5HTP into the W465 of the recognition pocket. The 5BrW465 YTH domain shows 8-fold improved binding affinity and selectivity compared to the wtYTH. We recognize that the unsuccessful incorporation of UAA at the 411 position limits the generality of the current strategy. When expressing 5HTP411 and 5BrW411, we obtained wtYTH rather than UAA-containing proteins; the yields of the resulting “wtYTH” proteins are around 3–5 mg/L of cell culture, which are significantly lower compared to regular wtYTH expression (39–48 mg/L of culture). Indeed, near-cognate suppression of UGA by the tRNA^Trp_CCA was previously reported to be present when incorporating BocK with PylRS/tRNA^Pyl_UCA at the opal stop codon.⁵⁹ It is possible that the different outcomes of TGA suppression at W411 versus W465 and W470 stem from distinct local mRNA sequence context⁶⁰ and/or an interplay between the stop codon position and the rates of transcription and translation.⁶¹ It is also possible that the UAA-incorporated products at this site are unstable and rapidly degraded, leaving behind low levels of Trp-misincorporation products. In the future, synonymous optimization of the mRNA context and/or optimization of the expression conditions could help us overcome this challenge and enable evaluation of UAA-substituted products at the promising W411 site.

Unlike the W465 position, the 5BrW and 5HTP substitutions at the W470 did not improve binding affinities to m⁶A-RNA; rather, both the 5BrW470 and 5HTP470 show comparable binding affinities as the wtYTH (Figure 3A). Examination of the crystal structure reveals differences in the orientation and distance for W465 and W470 relative to the bound m⁶A (Figure 1A). First, the aromatic ring of m⁶A orients in a “T-shape” relative to the indole group of W465, and in parallel to W470. Next, the distance from the carbon of the N⁶-methyl group of m⁶A and the center of the W465 indole (3.6 Å) is significantly shorter than the distance to W470 (5.1 Å) (Figure 1A). There can be an increasing π–π stacking contribution from W470 to the stabilization of the RNA-protein interaction, compared to W465, whereas W465 may contribute more to the CH–π interaction. Interestingly, the energy decomposition analysis of the vdW complex formed by isolated W470 (“C”) and m⁶A (“B”) revealed increased stabilization of both electrostatics and dispersion components compared to the A-B complex (Figure S14a,b), likely due to the additional π–π stacking interaction for C-B. The C-B complex shows a decreased percentage contribution from dispersion and increased contribution from electrostatics to the overall interaction energy, compared to A-B (Figure S14c). This analysis reveals the fine differences in the energy components of the vdW interaction when the tryptophan residue is positioned at different distances and orientations to the m⁶A (i.e., A-B versus C-B complex). Therefore, the substitution effect on energetics by introducing the 5-hydroxy or 5-bromo group at W470 can be different from that at W465.

In addition, the energy minimized structures of 5HTP470, 5BrW470, and 5MTP470 show minor perturbations of the loop structure surrounding the 470 residue compared to the wtYTH. Though such perturbations may not be significant considering thermal fluctuations, they were not observed in the energy-minimized structures of the 465-substituted YTH domains (5HTP465, 5BrW465, and 5MTP465) (Figure S15). On the other hand, we observed significant changes in the CD spectra of 5BrW470 and 5MTP470 relative to the wtYTH, which indicate secondary structure changes of the proteins upon introducing 5BrW or 5MTP at W470 (Figure S16). Therefore, structural perturbations can contribute to the different substitution effects at position 470 compared to position 465.

Despite the remarkably improved properties of 5BrW465 for binding m⁶A in shorter RNA sequences, the reduced selectivity observed for longer RNA substrates presents an unresolved challenge. Future engineering of these UAA-incorporated readers through directed evolution should enable further optimization of their affinities and selectivities. In addition, it should be possible to extend this technology to mammalian cells, given that the incorporation of 5BrW and 5HTP in these cells using the EcTrpRS/tRNA^EcTrp pair is well-established.^37,44,62

CONCLUSIONS

In summary, our study reports two YTH variants (5BrW465 and 5HTP465) with improved affinities and selectivities for the m⁶A-RNA, highlighting the potential of engineering m⁶A reader proteins via the incorporation of UAAs. The incorporation of the same tryptophan derivative (5BrW or 5HTP) at different residue positions within the pocket can result in different impacts on m⁶A recognition. The energy decomposition analysis reveals differences in the energetic components for the interaction of W465 and W470 to m⁶A, and the enhanced binding trend of 5HTP465 and 5BrW465 to m⁶A is consistent with the dispersion component modulated by the 5-Br or 5-OH substitution in the model complex, suggesting the importance of tuning the weak interaction for m⁶A recognition.

MATERIALS AND METHODS

RNA Oligonucleotides

RNA oligonucleotides m⁶A-RNA-27mer, A-RNA-27mer, m⁶A-RNA-13mer, and A-RNA-13mer were purchased from Integrated DNA Technologies (IDT), whereas m⁶A-RNA and A-RNA were gifts from the He lab at the University of Chicago. All oligonucleotides were dissolved in water for RNA work (Fisher Scientific, BP561-1) prior to use.

Plasmids for Expressing wtYTH and UAA-Containing wtYTH Variants

We used the previously reported pET vector containing the YTH domain of YTHDF1 (residues 375–552) and an N-terminal 6×His tag for overexpression of wtYTH in BL21 cells.⁶³ Site-directed mutations were introduced into the YTH gene using standard Gibson Assembly cloning to prepare the YTH-465-TGA, YTH-470-TGA, and the tandem YTH-465-TGA plasmids (Table S1).

Protein Expression and Purification

We expressed and purified wtYTH based on the reported protocol.⁶³ For UAA-containing YTH variants, the plasmid was transformed into ATMW-BL21 competent cells; the transformed cells were plated on an agar plate containing an antibiotic cocktail, including spectinomycin (95 μg/mL), chloramphenicol (20 μg/mL), gentamicin (10 μg/mL), zeocin (15 μg/mL), and kanamycin (50 μg/mL). A single colony was picked from the plate and cultured in a 10 mL LB medium containing the same antibiotic cocktail. The 10 mL overnight culture was used to inoculate 1 L of LB medium supplemented with spectinomycin (95 μg/mL), chloramphenicol (20 μg/mL), and kanamycin (50 μg/mL). The inoculated culture was incubated at 37 °C with shaking at 250 rpm. 1 mM Isopropyl β-D-1-thiogalactopyranoside (IPTG) and 1 mM tryptophan derivative (5BrW, 5HTP, or 5MTP) were added to induce protein expression when the OD₆₀₀ of the culture reached 0.7. Following induction, the culture was incubated at 30 °C for 16 h, and cells were subsequently harvested by centrifugation at 2500 rpm for 20 min at 4 °C with the Thermo Scientific Sorvall X4R Pro-MD centrifuge. Cell pellets were resuspended in the lysis buffer containing 20 mM HEPES (pH 7.5), 150 mM NaCl, and 1 mM dithiothreitol (DTT), and cells were lysed by sonication. The resulting lysate was centrifuged at 11,000 rpm for 15 min at 4 °C, and the supernatant was collected. 1% (w/v) streptomycin sulfate (Acros Organics) was added to the supernatant to assist precipitation of genomic DNA, and we performed a second centrifugation at 11,000 rpm for 15 min at 4 °C to remove any precipitates. The supernatant was incubated with pre-equilibrated His60 Ni Superflow Resin (Takara Bio) at 4 °C for 1 h before being loaded onto an empty column. The resin was extensively washed with over 20 mL of lysis buffer containing 25 mM imidazole, and the bound proteins were subsequently eluted using a gradient of imidazole concentrations (50, 100, 150, 250, and 500 mM) prepared in the lysis buffer. Eluted fractions were analyzed by 12% SDS-PAGE, and those containing the purified protein were pooled and concentrated to approximately 3 mL using a centrifugal concentrator (Sigma-Aldrich, amicon ultra-15). The concentrated protein was then desalted and buffer exchanged into the storage buffer composed of 25 mM HEPES (pH 7.5), 300 mM NaCl, 5% glycerol, 0.04% Triton X-100, and 1 mM DTT using a PD-10 desalting column (GE Healthcare). After desalting, the protein was further concentrated to around 5 mg/mL, aliquoted, flash-frozen in liquid nitrogen, and stored at −80 °C. Protein concentrations were measured via absorbance at 280 nm with 1 absorbance unit approximated to 1 mg/mL protein using a Nanodrop spectrophotometer (Thermo Scientific, NanoDrop 2000C).

For the tandem 5BrW465 protein, we performed an additional purification step via size exclusion chromatography (SEC). Elution fractions from the Ni-affinity column were buffer exchanged into the SEC buffer in ~1 mL. The samples were subjected onto Superdex 200 Increase 10/200 GL column and ran on the ÄKTA go FPLC with a flow rate of 0.38 mL/min at 4 °C. All elutions were characterized by SDS-PAGE; fractions containing the purified protein were collected and concentrated to ~1 mL with concentration measured, and 50 μL of glycerol was added to the protein. Proteins were then aliquoted and flash frozen for storage.

LC-MS Analysis of Purified Proteins

Purified proteins were analyzed using an Agilent 1260 Infinity HPLC coupled with the 6230 ESI time-of-flight (ESI-TOF) mass spectrometer (MS). The protein sample (~5 μM) was subjected onto a C8 column (100 × 4.5 mm Phenomenex Aeris 3.6-μm Widepore XB-C8) and run through the column under an increasing gradient (5–95%) of buffer B in buffer A (buffer A: 95% water, 5% acetonitrile, 0.1% formic acid; buffer B: 5% water, 95% acetonitrile, 0.1% formic acid). The eluted protein was injected into the MS and analyzed under the positive-ion mode. Total protein masses for the purified YTH domains were calculated through deconvolution using MagTran (Amgen).

FP Binding Assay

The FP binding assay was performed by incubating 25 nM FAM-labeled RNA and a variable amount of protein in a binding buffer composed of 20 mM HEPES (pH 7.5), 150 mM NaCl, and RNase inhibitor. For any tight binding with an observed affinity in the low nanomolar range, we used 5 nM FAM-labeled RNA instead of 25 nM in the binding assay. To prepare the binding assay, FAM-labeled RNA samples were diluted from a concentrated stock to 50 nM using the binding buffer. A YTH protein sample (wtYTH or any UAA-containing YTH variant) was thawed and buffer exchanged into the binding buffer at 4 °C using 10 kDa-cutoff concentrators (Sigma-Aldrich) with a final concentration above 40 μM, unless otherwise specified. Next, we prepared a 40 μM, 60 μL protein solution with the concentrated protein, 5 μL of SUPERaseIn RNase inhibitor (Invitrogen), and the binding buffer. Eight protein solutions (40 μL each) were subsequently prepared by serial dilution, yielding a range of protein concentrations from 18 nM to 40 μM. 40 μL of the 50 nM RNA was added into each 40 μL protein solution. The mixed protein and RNA solutions were covered with aluminum foil and incubated at room temperature for 30 min. For each solution (~80 μL in total), samples were transferred to two wells with 37.5 μL each well on a black polypropylene 384-well round-bottom plate (Cellvis). These two wells serve as technical replicates for FP measurements. A biological replicate is defined as replicates of which each binding component was independently pipetted into the binding solution and mixed. Among different biological replicates, the same batch or different batches of the purified proteins can be used.

FP was measured using a Synergy Neo2 Multimode Microplate Reader with excitation at 485/20 nm and emission at 528/20 nm at 24 °C. The FP signal was calculated using the following eq 1:

$$\text{FP}=\frac{I_{\parallel }-G\times I_{\perp }}{I_{\parallel }+G\times I_{\perp }}$$ (1)

where I_∥ and I_⊥ are the intensities of the parallel- and perpendicular-polarized light, respectively, and a G-factor of 0.75 (determined from instrument calibration) was used.

The measured FP values at varying protein concentrations (x) were fitted by a quadratic binding eq 2 that models single-site specific binding using the Levenberg-Marquardt nonlinear curve fitting algorithm in OriginPro 2018.

$$\text{FP}=\text{B}+\text{C}\times \left(\text{D}+K_{\text{d}}+x-\sqrt{{\left(\text{D}+K_{\text{d}}+x\right)}^2-4\text{D}x}\right)$$ (2)

In this equation, D represents the constant FAM-labeled RNA concentration (25 or 5 nM), and K_d is the dissociation constant. We calculated the binding free energy based on eq 3,

$$\Delta G_{\text{exp}}=RT\text{In}\left(K_{\text{d}}\right)$$ (3)

where T is the temperature for FP measurement (24 °C), and K_d is the dissociation constant.

Energy Decomposition Analysis

The “A-B” or “C-B” vdW complexes were prepared by extracting the W465 (as an amino acid) or W470 and m⁶A (base-only) from the crystal structure (PDB ID: 4RCJ). Geometric constraints were imposed by fixing the positions of nitrogen atoms in both the N-terminus of tryptophan and the five-membered ring of N⁶-methyladenine to their crystallographic coordinates to maintain structural integrity during optimization. Geometry optimizations of the monomers and vdW complexes were performed using unrestricted M06-2X density functional theory⁶⁴ with the aug-cc-pVTZ basis set. Convergence criteria for geometry optimizations were set to a maximum force threshold of 0.0045 Hartrees/Bohr. Zero-point vibrational energies (ZPEs) were obtained via frequency calculations on optimized structures and corrected using a scaling factor⁶⁵ of 0.971 to account for anharmonicity and partially rectify errors in electronic structure calculations. Implicit solvation effects mimicking the protein environment were accounted for using the electron-density-based solvation model – SMD,⁶⁶ with an effective dielectric constant of ε_r = 8, which was optimized to yield physical binding free energies and was representative of the residues in the pocket.⁶⁷ All density-functional theory (DFT) calculations, including geometry optimizations, vibrational frequency calculations, and single-point energies, were conducted with Gaussian16 software.⁶⁸ Interaction energies within the vdW complexes were then decomposed into physically meaningful components—electrostatics, dispersion, induction, and Pauli exchange—using the SAPT.⁶⁹ SAPT calculations employed the ωB97X density functional⁷⁰ and the aug-cc-pVTZ basis set, executed with the Psi4 program package.

For future structural engineering, easily estimable molecular properties will be highly valuable for design guidance, eliminating the need for complex quantum chemistry calculations. To establish a correlation between simple molecular descriptors—specifically, the ionization potential of a monomer and its polarizability—and the dispersion component of vdW interactions, we computed the molecular descriptor-based dispersion interaction using the following eq 4,

$$\Delta E_{\text{disp}}\propto -\left(\frac{3}{32{\pi }^2{\epsilon }_0^2}\right)\left(\frac{I_{\text{A}}{I}_{\text{B}}}{I_{\text{A}}+I_{\text{B}}}\right)\left(\frac{{\alpha }_{\text{A}}{\alpha }_{\text{B}}}{R_{\text{AB}}^6}\right)$$ (4)

in which I_A and I_B are the ionization potentials of the monomers A and B, respectively, which are simply taken as the orbital energy of the highest occupied molecular orbital (HOMO) at the DFT level. The zero-frequency isotropic polarizabilities, α_A and α_B, are deduced from their electron densities. The center-of-mass separation between the fragments in the vdW complex is denoted as R_AB. ε₀ is the vacuum permittivity.

Energy Minimization and MD Simulations

The initial structures of UAA-containing YTH domains were generated from the crystal structure of the YTH domain in complex with m⁶A RNA (PDB ID: 4RCJ) by manual modification of W465 or W470 in PyMOL (The PyMOL Molecular Graphics System, Version 2.5 Schrödinger, LLC). Specifically, the indole ring of W465 or W470 was modified at the 5-position to introduce bromo, hydroxy, or methoxy substituents, generating the six variants 5BrW465, 5HTP465, 5MTP465, 5BrW470, 5HTP470, and 5MTP470. Energy minimizations were performed with Amber.^71,72 The prepared PDB structures of wtYTH and the six tryptophan-derivative variants were used as starting coordinates. Topologies were generated with tleap using Amber ff19SB for the protein,⁷³ RNA.OL3 for RNA,⁷⁴ and GAFF2-derived terms where needed for noncanonical substituent parameters.^71,75 The m⁶A residue “6MZ” was treated as an internal RNA nucleotide derived from the AmberTools^71,72 modified-RNA 6MA parameter set, while preserving the RNA 2’-hydroxyl and phosphodiester connectivity.⁷⁶ The 5-bromotryptophan, 5-hydroxytryptophan, and 5-methoxytryptophan residues were modeled as internal peptide residues based on the ff19SB tryptophan template. Their partial charges were assigned from AM1-BCC calculations using antechamber on neutral ACE-UAA-NME-capped models and normalized to zero net charge for the internal residue.^77–79 Each complex was solvated in a truncated octahedral TIP3P water box with a 12 Å buffer⁸⁰ and neutralized with Na⁺ ions using TIP3P-compatible Joung-Cheatham monovalent ion parameters.⁸¹ Energy minimization was then performed in four stages: solvent/ion relax with solute heavy atoms restrained, local side-chain/base relaxation with protein/RNA backbone restraints, weakly restrained solute relaxation, and final unrestrained full-complex relaxation. Each minimization stage began with steepest-descent minimization, followed by conjugate-gradient minimization. For the first three relaxation stages, 5000 minimization cycles were used, with the first 2500 cycles assigned to steepest descent and the remaining cycles to conjugate-gradient minimization. The final unrestrained stage used 10,000 cycles, split into 5000 steepest-descent cycles followed by 5000 conjugate-gradient cycles. The initial minimizer step length was 0.01 Å, and the RMS Cartesian energy-gradient tolerance was 10⁻⁴ kcal/(mol × Å). The structures used for comparison were taken from the final unrestrained minimization stage.

Starting from the final unrestrained minimized structures, short MD relaxation simulations were performed for wtYTH and each of the six UAA variants. For each system, three independent replicas were initiated with different random velocity seeds. Each replica was gradually heated from 0 to 300 K over 250 ps under NVT conditions with a 10 kcal/(mol × Å²) harmonic restraint on solute heavy atoms, followed by 200 ps of NVT equilibration at 300 K with the solute heavy-atom restraint reduced to 2 kcal/(mol × Å²). The systems were then equilibrated under NPT conditions for 200 ps with a 1 kcal/(mol × Å²) solute heavy-atom restraint, followed by 300 ps of unrestrained NPT equilibration. Finally, each replica was subjected to 1 ns of unrestrained NPT production simulation, giving 3 ns of short-relaxation sampling per system.

All MD simulations used a 2 fs timestep with bonds involving hydrogen constrained by SHAKE.⁸² Temperature was controlled at 300 K using Langevin dynamics with a collision frequency of 2 ps⁻¹.⁸³ Pressure was maintained at 1 atm using isotropic pressure coupling with the Amber Berendsen barostat.⁸⁴ A 10 Å real-space cutoff was used for nonbonded interactions under periodic boundary conditions.⁸⁵ Production coordinates were saved every 10 ps.

CD Measurements

Purified proteins were buffer exchanged into the binding buffer with 2 mM dithiothreitol (DTT) added. 25 μM 300 μL protein sample was transferred to a quartz cuvette (JASCO, Rectangular Quartz Cell – 1 mm path length) and the CD of the sample was measured from 180 to 260 nm at 20 °C using a JASCO J-1100 CD Spectrometer. CD spectra from 200 to 260 nm were used for comparing overall secondary structure change between different proteins.

Thermal Shift Assay

Thermal shift assays were performed following the reported assay using SYPRO Orange dye, known to exhibit enhanced fluorescence in hydrophobic environments associated with protein unfolding.⁸⁶ 4 μM of protein was incubated with 5x SYPRO Orange dye (Invitrogen) in the binding buffer at room temperature for 15 min. After incubation, samples were heated from 25 to 75 °C at a rate of 0.2 °C per second, with fluorescence intensities (λ_excitation = 470 nm and λ_emission = 570 nm) recorded every 0.3 °C by a QuantStudio 3 Real-Time PCR System (Thermo Fisher). Fluorescence data were normalized by defining the highest and lowest intensities recorded during each melting run as 100 and 0%, respectively. Melting temperatures (T_m) were determined by fitting normalized data to the Boltzmann Sigmoid equation using GraphPad Prism 6.0 software.

Supplementary Material

The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acschembio.6c00476.

Experimental materials, methods, supporting figures including proteins expression data, LC-MS data, CD data, FP binding curves with replicates, and additional DFT calculation results, and supporting tables for oligonucleotides and plasmids sequences, yields of purified proteins, tabulated binding affinity results, and tabulated energy decomposition analysis results (DOCX)