Capturing the Mechanism Underlying TOP mRNA Binding to LARP1

SUMMARY

The RNA-binding protein La-related protein 1 (LARP1) plays a central role in ribosome biosynthesis. Its C-terminal DM15 region binds the 7-methylguanosine (m⁷G) cap and 5’ terminal oligopyrimidine (TOP) motif characteristic of transcripts encoding ribosomal proteins and translation factors. Under the control of mammalian target of rapamycin complex 1 (mTORC1), LARP1 regulates translation of these transcripts. Characterizing the dynamics of DM15-TOP recognition is essential to understanding this fundamental biological process. We use molecular dynamics simulations, biophysical assays, and x-ray crystallography to reveal the mechanism of DM15 binding to TOP transcripts. Residues C-terminal to the m⁷G-binding site play important roles in cap recognition. Further, we show that the unusually static pocket that recognizes the +1 cytosine characteristic of TOP transcripts drives binding specificity. Finally, we demonstrate that the DM15 pockets involved in TOP-specific m⁷GpppC-motif recognition are likely druggable. Collectively, these studies suggest unique opportunities for further pharmacological development.

Graphical Abstract

eTOC Blurb

La-related protein 1 (LARP1) plays a central role in ribosome biogenesis. The DM15 region binds mRNAs that encode ribosomal proteins and translation factors. Cassidy, Lahr et al. use simulations, biochemical assays, and crystallography to reveal the mechanism of DM15/mRNA binding. These studies suggest future opportunities for pharmacological development.

INTRODUCTION

The ribosome and associated translation machinery lie at the heart of gene expression. Their biosynthesis is controlled at multiple levels (Fonseca et al., 2014). Under energy or nutritional stress, cells reduce processes associated with high energy expenditure and direct resources towards survival. Nutrient and energy deprivation are invariably accompanied by reduced ribosome production, an energetically demanding cellular process. In contrast, cells replete with nutrients and growth factors engage ribosome biosynthesis to ensure maximal growth and proliferation (Fonseca et al., 2018; Warner, 1999). Cell propagation is fueled by amplified protein synthesis, which requires the de novo synthesis and assembly of new ribosomes (Pelletier et al., 2018). The signals that control protein synthesis, protein turnover (macroautophagy), and cell propagation converge on the mammalian target of rapamycin complex 1 (mTORC1) (Saxton and Sabatini, 2017).

mTORC1 regulates protein synthesis by phosphorylating proteins that play key roles in translation control (Fonseca et al., 2014). For instance, mTORC1 stimulates cap-dependent translation initiation by phosphorylating eukaryotic initiation factor 4E (eIF4E)-binding proteins (4E-BPs) (Gingras et al., 2001). Phosphorylation causes 4E-BPs to dissociate from eIF4E (Brunn et al., 1997; Burnett et al., 1998), a 7-methylguanosine (m⁷G) cap-binding protein (Filipowicz et al., 1976; Sonenberg et al., 1978). When bound to unphosphorylated 4E-BPs, eIF4E cannot associate with the m⁷G cap or eIF4G, as required to recruit the 40S ribosome subunit to the 5’ untranslated region (5’UTR) of the mRNA. 4E-BP-bound eIF4E prohibits cap-dependent translation initiation by inhibiting the formation of the eIF4F translation initiation complex (Sonenberg and Hinnebusch, 2009). mTORC1 stimulates cap-dependent translation initiation by relieving eIF4E inhibition via 4E-BP phosphorylation (Sonenberg and Hinnebusch, 2009).

In addition to regulating 4E-BP phosphorylation and eIF4E binding, mTORC1 also regulates the phosphorylation of another recently identified substrate: La-related protein 1 (LARP1) (Hong et al., 2017; Kang et al., 2013). mTORC1 phosphorylates LARP1, causing it to release the 5’UTRs of ribosomal protein transcripts (e.g., the RPS6 transcript, per our unpublished data). In doing so, mTORC1 orchestrates the translation of transcripts that encode ribosomal proteins and some translation-associated factors (Fonseca et al., 2015). These transcripts share a motif in their 5’UTRs termed the 5’terminal oligopyrimidine (5’TOP) motif that is required for the coordinated regulation of TOP mRNA translation (Fonseca et al., 2014; Meyuhas and Kahan, 2015). The 5’TOP motif has an invariant C in the +1 position, followed by a tract of cytidines and uridines (Levy et al., 1991).

LARP1 is a translational regulator of TOP transcripts that functions downstream of mTORC1 (Fonseca et al., 2015). LARP1 is hypothesized to repress TOP translation (Aoki et al., 2013; Fonseca et al., 2015) by sequestering key mRNA transcripts from translation initiation factors (Lahr et al., 2017) and anchoring them to stress granules (Wilbertz et al., 2019). LARP1 appears to elicit its inhibitory effects on TOP mRNA translation via its C-terminal DM15 region (Philippe et al., 2017). Specifically, the C-terminal LARP1 DM15 region, comprised of three HEAT-like repeats, directly engages the 5’TOP sequence and the 7-methylguanosine (m⁷G) cap via a conserved and positively charged surface (Lahr et al., 2017; Lahr et al., 2015). By characterizing the molecular interactions between the DM15 region, the m⁷G cap, and the 5’TOP motif, we provided a molecular mechanism linking LARP1 to TOP mRNA translation repression. Recent work has demonstrated that this repressive role is mTORC1-dependent (Philippe et al., 2017).

The precise biological role of LARP1 is still debated. In addition to observations that LARP1 represses TOP mRNA translation (Fonseca et al., 2015; Lahr et al., 2017; Philippe et al., 2017), several studies have also noted that LARP1 controls mRNA decay in both plants (Merret et al., 2013) and animals (Aoki et al., 2013; Fonseca et al., 2015; Gentilella et al., 2017). In mammals, LARP1 binds and stabilizes TOP transcripts (Aoki et al., 2013; Fonseca et al., 2015; Gentilella et al., 2017), possibly helping to sustain translation. Nutritional (Damgaard and Lykke-Andersen, 2011) and oxidative stress (Wilbertz et al., 2019) halt TOP mRNA translation and lead to an accumulation of TOP transcripts in stress granules (SGs). Recent data (Wilbertz et al., 2019) indicate that LARP1 plays a seminal role in anchoring TOP transcripts to SGs, thereby averting their rapid decay. Storage of TOP transcripts in SGs by LARP1 likely allows for engagement of translation without the requirement for de novo transcription. Consistent with an important role for LARP1 in maintaining mRNA stability, LARP1 depletion has been observed to reduce the levels of some TOP-encoded proteins (Tcherkezian et al., 2014).

A number of studies suggest that, analogous to other downstream mTORC1 targets, LARP1 instigates cell proliferation (Tcherkezian et al., 2014) and thus contributes to the onset and maintenance of several cancers (Hopkins et al., 2016; Mura et al., 2015; Selcuklu et al., 2012; Xie et al., 2013). Given its potential role as both a repressor of translation and stabilizer of mRNA transcripts, some have argued that LARP1 functions as a molecular switch in cancer biology (Hong et al., 2017). Small-molecule modulators of the mTORC1-LARP1 axis thus have therapeutic potential. One approach to altering the activity of this axis is to inhibit mTORC1 itself, and potent and specific mTORC1 inhibitors exist (Feldman et al., 2009; Thoreen et al., 2009; Yu et al., 2010). But mTORC1 inhibition impacts many critical pathways, leading to side effects that complicate the long-term therapeutic use of mTORC1 inhibitors (Creel, 2009). Targeting the downstream LARP1 protein could bypass these broad effects by limiting inhibition to the mTORC1-LARP1 axis.

Herein, we present a molecular dynamics (MD) simulation of the LARP1 DM15 region. The flexibility of this cancer-associated protein, here revealed in atomic detail, suggests a molecular mechanism whereby LARP1 structural features mediate TOP mRNA binding. The simulation reveals interactions not observed in any crystal structure that may contribute to the binding mechanism. We present biochemical and biophysical evidence to support the proposed interactions. We also identify DM15 druggable hotspots, taking into account the pocket dynamics that the simulations reveal. Some of these hotspots are not evident in any crystal structure. They provide unique drug-discovery opportunities to specifically treat a variety of LARP1-implicated cancers while minimizing off-target effects.

RESULTS

The 4ZC4 crystallographic dimer captures the flexible region connecting the α7 and α8 helices (S923-L928) in two distinct conformations (Lahr et al., 2015). This α7–α8 bridge is a 3₁₀ helix in chain B, but a disordered loop in chain C. Given that the bridge abuts the m⁷G-binding pocket (Lahr et al., 2017; Lahr et al., 2015), we hypothesize that its dynamics contribute to the mechanism of cap recognition. Though the DM15 region often crystalizes as a dimer, we have shown that it is a monomer in solution (Lahr et al., 2015). We therefore used two molecular dynamics (MD) simulations of LARP1 DM15 monomers, based on the 4ZC4:B and 4ZC4:C structures, to test this hypothesis. These monomeric simulations suggest a molecular mechanism that may arbitrate LARP1/5’TOP-motif binding in vivo.

Simulation Equilibration

We first verified that each DM15 simulation had properly equilibrated by calculating the C_α RMS distance (RMSD) between each simulation frame and the respective first frame. This analysis revealed that the RMSD values had not stabilized by the beginning of the production simulations (Figure S1, blue line). As the 4ZC4 protein construct had been truncated after D946 to facilitate crystallization (Lahr et al., 2015), we recognized that the dynamics of the most C-terminal residues might be artefactual. Therefore, we recalculated the RMSD of all simulated C_α up to G939, ignoring subsequent C-terminal residues. The revised RMSD plot was much flatter (Figure 1A), confirming that the biologically relevant portion of the DM15 region had equilibrated. Out of an abundance of caution, we discarded the first 60 and 50 ns of the 4ZC4:B and 4ZC4:C simulations, respectively (Figure 1A and Figure S1, dotted lines). Subsequent analyses focused only on the remaining portions of each simulation.

Figure 1. RMSD and RMSF analyses of the 4ZC4:B and 4ZC4:C simulations.

(A) The C_α RMSD between each frame and the first, excluding residues that were C-terminal to G939. Running averages are shown in darker colors, against the lighter-colored raw data. The frames preceding the dashed lines were not used in subsequent analyses. (B) The center-of-geometry, per-residue RMSF values of the 4ZC4:B and 4ZC4:C simulations, in blue and red respectively. The inset highlights the RMSF values of the α7–α8 bridge residues. (C) The RMSF values of the 4ZC4:B simulation projected onto a ribbon representation of the LARP1 DM15 region. Helices are labelled as in ref. (Lahr et al., 2015). See also Figure S1.

The α7–α8 Bridge Is Highly Dynamic

To investigate the flexibility of the α7–α8 bridge, we calculated the root mean square fluctuation (RMSF) of the center of geometry of each residue. As expected, the C-terminal residues of the 4ZC4:B simulation (Figure 1B, in blue) had RMSF values that were substantially higher than those of the remaining construct (maximum: 8.26 Å), likely an artefact of protein construct truncation. For reference, the average RMSF of the remaining residues was only 1.09 Å.

Other, non-terminal residues with notably high RMSF values were positioned near the m⁷G pocket, suggesting that this portion of the DM15 region is particularly flexible (Figure 1B inset, in blue). These residues had RMSF values greater than 2.0 Å, with the exception of K921 (1.73 Å) and S923 (1.47 Å). K921 and Y922 are located on the α7 helix, near the m⁷G pocket (Figure 2A). S923, K924, A925, K926, N927, and L928 belong to the α7–α8 bridge (Figure 2A).

Figure 2. Specific residue interactions contribute to m⁷G-pocket dynamics.

(A) A frame from the MD simulation that illustrates key interactions. (B) Representative conformations of the α7-α8-bridge helix and loop microstates colored by secondary structure (blue: helical; orange: loop), extracted from the simulation. (C) The probability distributions of the m⁷G-pocket volumes (helical bridge, n = 117; disordered bridge, n = 883). The two conformational populations are statistically different (Levene’s Test, p = 0.021; t-test, p = 9.6 × 10⁻²⁵). (D) The probability distributions of the distance between the K924 amino and E886 carboxyl groups. The two populations are statistically different (Levene’s Test, p = 0.65; t-test, p = 2.0 × 10⁻²⁵). (E) The probability distributions of the distance between the K921 amino and Y922 hydroxyl groups. The two populations are statistically different (Levene’s Test, p = 2.7 × 10⁻³⁷; t-test, p = 8.3 × 10⁻⁴⁷). In all cases, the insets show representative helical- and disordered-bridge conformations from the simulation, taken from the labelled bins (asterisks). (F) Protein melt-curve data (n = 3). Error bars are standard deviations. **, p<0.01; *** p<0.001. See also Figure S2, Figure S3, and Movie S1.

The 4ZC4:C RMSF values were generally similar to those of the 4ZC4:B simulation, except for the same key residues of the α7 helix and α7–α8 bridge (Figure 1B inset, in red). Though these residues were notably flexible in the 4ZC4:C simulation as well, their motions were less dynamic than those observed in the 4ZC4:B simulation. Only Y922, K924, K926, and N927 had RMSF values greater than 2.0 Å. K926, the most flexible non-terminal residue, had an RMSF value of 4.12 Å and 2.76 Å in the 4ZC4:B and 4ZC4:C simulations, respectively. We chose to focus on the more dynamic 4ZC4:B simulation. The 4ZC4:C simulation is further described in the Supporting Information.

The instability of the α7–α8 bridge was surprising given that the 4ZC4:B bridge was helical in the starting conformation. Visual inspection of the trajectory revealed that the helix unfolds and refolds over the course of the simulation (Movie S1). To quantify this event, we calculated the hydrogen-bond lengths between the backbone atoms of S923 and K926, K924 and N927, and A925 and L928, per the i to i+3 pattern characteristic of a 3₁₀ helix. A given α7–α8 bridge conformation was assumed to be helical if none of these three distances exceeded 4 Å. By this metric, the bridge was helical 11.7% of the time over the course of the production simulation.

m⁷G-Pocket Dynamics

The secondary structure of the α7–α8 bridge determines the volume of the neighboring m⁷G pocket. We used the POVME 2.0 algorithm (Durrant et al., 2014) to calculate the m⁷G-pocket volumes of 1,000 equally spaces frames taken from the 4ZC4:B production simulation. The pocket volumes ranged from 0 to 393.375 ų. We separated these volumes into two populations: those associated with a helical α7–α8 bridge, and those associated with a disordered-loop bridge (Figure 2B). A probability distribution of the pocket volumes associated with each bridge conformation shows that the volume tends to be larger when the α7–α8 bridge is helical (Figure 2C).

A transient interaction between K924 and E886 appears to influence m⁷G-pocket volume by sequestering E886 away from the pocket. To investigate if this interaction correlated with the conformation of the α7–α8 bridge, we calculated the distances between the K924 sidechain nitrogen atom and the E886 side-chain carboxylate carbon atom over the course of the simulation. We divided the distances into bridge-helical and bridge-loop populations. The probability distribution of the K924-E886 distances shows that when the α7–α8 bridge is a disordered loop, these two residues participate in a tight electrostatic interaction. The distance between the K924 side-chain nitrogen atom and the E886 side-chain carboxyl carbon atom is less than 4 Å roughly 20% of the time (Figure 2D). This same interaction is present only 4.3% of the time when the bridge is a 3₁₀ helix.

The conformation of the α7–α8 bridge also appears to influence an interaction between K921 and Y922 that stabilizes the m⁷G pocket in a state preferential for ligand binding. We again divided the K921-sidechain-nitrogen/Y922-sidechain-oxygen distances into bridge-helical and bridge-loop populations. The probability distribution of the K921-Y922 distances shows that the K921–Y922 interaction—which prevents Y922 from occluding the m⁷G pocket—is much more frequent when the α7–α8 bridge is a 3₁₀ helix than when it is not (28.2% vs. 12.7% of the time, respectively; Figure 2E).

The Supporting Information describes a possible bridge-dependent interaction between S923 and E886 that may also displace E886, just as the K924–E886 interaction does (Figure S2). The S923-E886 interaction has never been captured in any crystal structure. The Supporting Information also describes the interaction between R817 and E857, two residues that are distant from the m⁷G pocket. This interaction is not correlated with the conformation of the α7–α8 bridge and so serves as a negative control (Figure S3).

Mutagenesis Studies Support the MD-Predicted Interactions

If K924 and K921 do in fact participate in the key interactions observed in the apo simulations, these residues should contribute to the stability of the apo DM15 region. To test this hypothesis, we generated K924A and K924D mutants to disrupt the MD-predicted K924–E886 interaction, and a K921A mutant to disrupt the K921–Y922 interaction. The stability of each recombinantly expressed and purified mutant was assayed using melting temperature (T_m) as a readout, where a decrease in melting temperature corresponds to a decrease in thermal stability. Compared to wild type (WT) DM15, these mutants exhibited a significant average decrease in T_m of at least 3.2°C (Figure 2F). The K924A and K924D mutants both resulted in an equal loss of protein stability (p = 7.0 × 10⁻³ and p = 7.1 × 10⁻³, respectively). The T_m of the K921A mutant was also significantly lower than that of wild-type DM15 and K924A (p = 2.6 × 10⁻⁴ and p = 4.1 × 10⁻³, respectively). In contrast, the melting temperature of the K815A control mutant was not significantly different from wild type (Figure 2F); K815 is distant from the m⁷G pocket and does not form substantial interactions with other residues, per our simulation.

Key Interactions Gate TOP-mRNA Access

To better understand how the conformation of the α7–α8 bridge impacts pocket dynamics, we identified maximally open (Figure 3A and 3B, in green) and collapsed (Figure 3B, in purple) m⁷G-pocket conformations. As expected, in the selected collapsed state the α7–α8 bridge is a disordered loop. This conformation allows amino acids such as Y922 and K924 to occlude mRNA binding (Figure 3). In contrast, the maximally open conformation approaches what we call the “ligand-ready state” (Figure 3, in green). In this state, the α7–α8 bridge is helical, as seen in the 4ZC4:B crystal structure (Lahr et al., 2015). The simulations show that K924 is a key pocket-occluding residue. In the collapsed state, it participates in a transient electrostatic interaction with E886, the residue primarily responsible for recognizing the Watson-Crick face of the m⁷G moiety (Figure 3B, purple). This interaction displaces E886 from the mRNA-bound orientation observed crystallographically (Lahr et al., 2017; Lahr et al., 2015). In the ligand-ready state, K924 is more distant from E886. E886 is thus more prone to adopt a position that can accommodate cap binding, as seen in the 4ZC4:B crystal structure and some of our simulation frames.

Figure 3. The largest and smallest m⁷G-pocket conformations, by volume.

(A) A surface representation of the frame with the largest pocket volume. (B) Large and small pocket conformations, superimposed. (C) An example of a small-pocket conformation. In orange, the m⁷GpppC ligand from PDB ID:5V87 (Lahr et al., 2017) is superimposed for reference. See also Figure S4 and Figure S5.

Y922 also plays a prominent role in determining the m⁷G-pocket volume. Our simulations show that in the collapsed state, the Y922 sidechain flips in and out of the m⁷G pocket, potentially occluding mRNA binding. In contrast, in the ligand-ready state, the helical α7–α8 bridge pulls Y922 away from the binding pocket (Figure 3B, in green). K921 appears to help maintain Y922 in this mRNA-compatible conformation. A single bridging water molecule forms hydrogen bonds with both K921 and Y922, helping to mediate the K921–Y922 interaction. The 4ZC4:B structure includes this water molecule, and the 4ZC4:A electron density also suggests its presence. Using in-house scripts, we verified that the same location (relative to K921 and Y922) is often water occupied over the course of our MD simulations (Figure S5).

Ligand Specificity by Exploiting the TSS Pocket

To further future drug-discovery efforts, we also studied the TSS (transcription start site) pocket, which binds the cap-adjacent cytosine that is characteristic of 5’TOP-motif-containing transcripts. Based on published crystal structures and our simulations, we hypothesized that this pocket determines TOP-transcript specificity. If true, better characterizing the TSS pocket may suggest new drug-discovery approaches that achieve small-molecule/DM15 specificity through a similar mechanism.

We predicted that R847 is primarily responsible for +1 nucleotide recognition and that it determines LARP1–5’TOP-motif specificity through Watson-crick recognition of the +1 nucleotide. We further surmised that the aromatic residues stacking the +1 nucleotide, Y883 and F844, impart stability and align the nucleotide in the TSS pocket via π-stacking interactions. Our simulations confirm that these residues are notably stable. Even in the apo 4ZC4:B simulation, Y883 and F844 have RMSF values of only 1.12 and 0.95 (Figure 1B).

To test these hypotheses, we used mutagenesis to alter the amino acids in the cytosine-specific TSS pocket. We tested R847E, F844Y, and F844W mutants (individually and in combination) for RNA binding using electrophoretic mobility shift assays (EMSA). We predicted that R847E would change sequence recognition from m⁷GpppC to m⁷GpppG. We further predicted that changing F844 to a larger aromatic residue would increase the stability of RNA binding. As expected, the F844Y/R847E double mutant (which we call FYRE) had a higher affinity for m⁷GpppG (non-TOP) than m⁷GpppC (TOP), indicating a shift of RNA substrate. It is important to note that the affinities reported here are weaker than those reported previously (Lahr et al., 2017) due to substrate length; here, we tested the binding of DM15 to a 20-mer representing RPS6 rather than the 42-mer RNA sequence used previously (Lahr et al., 2017).

Notably, the FYRE mutant did not merely weaken specificity for m⁷GpppC relative to wild-type DM15; rather, it switched specificity to m⁷GpppG. The FYRE variant had a single discrete shift on the EMSA even at higher protein concentrations. FYRE apparently establishes a single register of binding, similar to the single resister of wild-type DM15 binding to m⁷GpppC (Lahr et al., 2017) (Figure 4A, B).

Figure 4. Mutations that affect the TSS pocket alter TOP-motif recognition.

(A) A representative EMSA of FYRE DM15 binding to m⁷Gppp-TOP RNA and m⁷Gppp-non-TOP RNA, with the indicated protein titrations. (B) Quantification of triplicate binding assays of the FYRE mutant with the indicated RNAs, analyzed by EMSA. (C) Co-crystal structure of the FYRE mutant bound to m⁷GpppG (in green) resolved to 2.34 Å resolution. α-helices are represented as cylinders, and amino acids involved in dinucleotide recognition are represented as sticks. The electron density of the composite omit map contoured at 2.0 sigma, carved around the m⁷GpppG dinucleotide, is represented in blue mesh.

Changing F844 in isolation did not alter the affinity of the mutant for m⁷GpppG RNA (data not shown). This suggests that hydrogen bonding with R847 is important for recognition of the +1C nucleotide. It also implies that the multiple shifted bands that we previously observed with uncapped or capped +1G RNA are most likely nonspecific binding events, with each band representing increasing stoichiometries of protein:RNA (Lahr et al., 2017). Locking in the first and second positions of the RNA restricts these non-specific binding events observed at high protein concentrations via EMSA.

To visualize the specificity switch, we resolved a cocrystal structure of m7GpppG with the FYRE mutant to 2.34 Å resolution (Table 1). The structure reveals that E847 recognizes the +1G of the co-crystallized dinucleotide (Figure 4C), as predicted. Though LARP1 is a monomer in solution (Lahr et al., 2015), the wild-type DM15 region bound to m⁷GpppC most often crystallizes with a two-fold non-crystallographic symmetry (NCS) axis orthogonal to the phosphate linkage between the bases (Lahr et al., 2017). The crystal form of FYRE bound to m⁷GpppG is similar. The occupancy of the dinucleotide is approximately evenly divided between the conformation shown in Figure 4C and a second conformation splayed across the NCS (Lahr et al., 2017).

Table 1.

Data collection and refinement statistics for the FYRE DM15 (+1 C pocket) mutant.

Data collection
Space group	P21
a, b, c (Å)	58.58, 87.36, 72.89
β	93.37
Resolution (Å)	29.24–2.34
Rmerge (%)	0.05 (0.586)
I/σ(I)	12.6 (1.6)
Completeness (%)	99.7 (98.1)
Redundancy	3.9 (3.7)
Refinement
Resolution (Å)	29.24–2.34
No. reflections	121549
Rwork/Rfree	0.220/0.267
RMSD bond angle (°)	0.6
RMSD bond length (Å)	0.03
Average B-factor	74.0
PDB ID	6PW3

Druggability Assessment

Given the potential benefits of pharmaceutically targeting the LARP1 m⁷G and TSS pockets, we assessed the DM15 region for druggability. We used affinity propagation clustering (Frey and Dueck, 2007) to extract representative ensembles of distinct LARP1-DM15 microstates from the simulations. The 4ZC4:B simulation yielded an ensemble of 12 representative conformations. We next used the FTMap server (Brenke et al., 2009) to flood the surface of each ensemble conformation with virtual organic probes. A custom Python script identified consistent druggable hotspots (across all ensemble members) where probes tended to congregate (Figure 5A). Our ensemble-based druggability analysis of the 4ZC4:B simulation confirmed that the m⁷G and TSS pockets are druggable (Figure 5A, hotspots “a” and “b” in red). An unexpected druggable hotspot was also identified beneath the α7–α8 bridge (Figure 5A, hotspot “c”).

Figure 5. LARP1 DM15 druggable hotspots.

(A) The locations of persistent druggable hotspots—across all twelve 4ZC4:B ensemble conformations—are shown as silver surfaces. These surfaces encompass regions where FTMap positioned small organic probes. One of the 4ZC4:B-ensemble protein conformations is shown in yellow ribbon. (B) The same FTMap analysis applied to the 4ZC4:B simulation frame with the largest m⁷G-pocket volume, per POVME.

To further study the druggability of the m⁷G pocket, we repeated the FTMap analysis using the largest m⁷G-pocket conformation (by volume) sampled over the course of the 4ZC4:B simulation. Though the TSS pocket is collapsed in this conformation, the m⁷G pocket is much expanded. The m⁷G-pocket hotspot (Figure 5B, hotspot “b”) is contiguous with two adjacent druggable hotspots (Figure 5B, “a” and “c”) that could also be pharmacologically exploited.

DISCUSSION

LARP1 occupies a central node in mTORC1 signaling (Fonseca et al., 2015; Hong et al., 2017; Philippe et al., 2017; Tcherkezian et al., 2014), and LARP1 protein levels are altered in several cancers (Hopkins et al., 2016; Mura et al., 2015; Stavraka and Blagden, 2015; Ye et al., 2016). Small molecules that alter LARP1 activity could thus be developed into novel anti-cancer therapies. Crystal structures of the LARP1 DM15 region bound to TOP mRNA and m⁷GpppC reveal two potentially druggable pockets that bind the m⁷G cap and first nucleotide, respectively (Lahr et al., 2017; Philippe et al., 2017). The m⁷G pocket of these structures strongly resembles that of other cap-binding proteins. However, unlike other cap-binding proteins, LARP1 preferentially binds TOP transcripts with m⁷GpppC motifs at the 5’ end. Understanding the structural, dynamic, and biochemical factors that enable this selectivity may be key to designing new small-molecule therapeutics that bind specifically to the LARP1 DM15.

The LARP1-DM15 m⁷G Pocket

Remarkable Pocket Flexibility

Published crystal structures provide some evidence that the cap-binding region of LARP1 DM15 is highly flexible. The pocket-adjacent linker that connects the α7 and α8 helices (the α7–α8 bridge) has been captured in several distinct conformations (Lahr et al., 2015), including a helical conformation (4ZC4:B), a loop conformation (4ZC4:C), and a loop conformation so disordered that it could not be resolved crystallographically (4ZC4:D).

The computational and experimental analyses described here build on this previous work. Our simulations suggest that the secondary structure of the α7–α8 bridge may play a critical role in promoting or discouraging TOP mRNA binding. By capturing the transitions between helical- and disordered-bridge conformations, the simulations reveal m⁷G-pocket flexibility beyond what has been observed crystallographically. These transitions alter the volume of the m⁷G pocket and so likely impact TOP mRNA binding.

The Collapsed State

Some of our simulation frames captured the m⁷G pocket in a transient collapsed state that has not been seen in any crystal structure. TOP mRNA binding is unlikely in this state for three reasons. First, the disordered α7–α8 bridge allows the Y922 side chain to flip in and out of m⁷G pocket, sterically hindering TOP mRNA binding. Second, the bridge residue K924 transiently interacts with and displaces E886, a key residue that directly mediates TOP mRNA binding via hydrogen bonds with the m⁷G-cap Watson-Crick face (Lahr et al., 2017). This interaction may sequester E886 away from the optimal m⁷G-binding conformation.

Third, the K924–E886 interaction may also contribute to the free energy of binding in ways that indirectly promote LARP1/mRNA dissociation. Binding affinity is determined by the difference in molar Gibbs free energy between the ligand-bound and unbound states. Put another way, ligand affinity is proportional to the extent to which the binding event increases the stability of the system. When TOP mRNA is bound, the positively-charged m⁷G guanine moiety forms energetically favorable interactions with the pocket (e.g., cation-π interactions with both Y922 and Y883), enhancing the stability of the complex (Lahr et al., 2017). But a similarly favorable K924–E886 interaction occurs exclusively in the absence of bound mRNA, reducing the energetic/stability difference between the mRNA-bound and collapsed states.

The K924–E886 interaction forms and breaks multiple times over the course of the MD simulation. Although some apo crystal structures hint at this interaction (e.g., 4ZC4:C), the simulations suggest it is far more prominent. To confirm the role that K924 plays in stabilizing the mRNA-unbound (apo) protein, we used a protein melting temperature assay to evaluate K924A and K924D DM15 mutants. The melting temperatures of both were reduced by 3.2°C, showing that K924 contributes to apo protein stability as expected (Figure 2F). Interestingly, mutation of K924 did not change the affinity of the DM15 region for capped TOP sequence (data not shown); this result, however, was not unexpected for an assay conducted at equilibrium. We anticipate that the mutation affected the kinetics of binding, which were not explicitly tested.

The Ligand-Ready State

Some of our simulation frames captured the m⁷G pocket in a conformation more amenable to TOP mRNA binding. In this state, the α7–α8 bridge is in a helical conformation similar to that of 4ZC4:B. The more rigid helical α7–α8 bridge rotates K924 away from E886, such that E886 more readily adopts the orientation required for mRNA-cap binding. Finally, the helical structure also pulls Y922 away from the m⁷G pocket, reducing the steric hindrance seen in the collapsed state (Figure 3B, in green). The simulations suggest that interactions with K921—possibly mediated by hydrogen bonding with a common water molecule (Figure S5)—maintain Y922 in this open-pocket position (Figure S5).

We did not expect K921 to play so prominent a role in preventing Y922-mediated pocket occlusion. Several crystallographic conformations show that the side chains of these two amino acids are physically adjacent, but visual inspection of the simulations suggests a fairly robust interaction between the two. To provide experimental evidence in favor of this interaction, we used a protein melting temperature assay to evaluate the K921A mutant. This mutant had a melting temperature 4.0°C below that o f the wild type (Figure 2F), showing that K921 does in fact contribute substantially to apo protein stability.

The Collapsed and Ligand-Ready States May Interconvert in vivo

Given that our 4ZC4:B simulation started from the 4ZC4:B conformation, it is not surprising that it sampled the α7–α8 bridge in a helical conformation. What is remarkable is that over the course of our simulation, the 3₁₀ helix unfolded and then refolded (Movie S1). This unfolding-and-refolding event caused DM15 to transition between the collapsed and ligand-ready states, suggesting that the energetic difference between the two is small. LARP1 DM15 may interconvert rapidly between both states in vivo.

Since these two states appear to be nearly isoenergetic, one would expect that even small perturbations to the protein in this region could bias the conformational ensemble in favor of one or the other. Indeed, mutations near or in the α7–α8 bridge have been observed in several cancers. I930V and I930T mutations have been implicated in lung adenocarcinoma and breast cancer, respectively (COSMIC (Forbes et al., 2017) Study IDs COSU417 and COSU652). Y922D has also been implicated in colon adenocarcinoma (COSMIC (Forbes et al., 2017) Study ID COSU376).

These amino-acid changes were identified in tumors, but post-translational modifications of bridge and bridge-adjacent residues may also serve to regulate LARP1/5’TOP-motif binding in a non-pathogenic context. Though LARP1 associates with mTORC1, there are no known phosphorylation sites within the simulated DM15 region (Hong et al., 2017; Philippe et al., 2017). However, a recent proteome-wide study identified the DM15 bridge residue K926 as a ubiquitination site (Hornbeck et al., 2015; Udeshi et al., 2013). The role of ubiquitination is not limited to targeting proteins for proteasome-mediated proteolysis (Mukhopadhyay and Riezman, 2007; Schnell and Hicke, 2003), so it is reasonable—albeit speculative—to hypothesize that ubiquitination here serves to alter TOP mRNA binding via changes in m⁷G-pocket dynamics. While mTORC1 plays a role in overall ubiquitin-mediated degradation (Zhao and Goldberg, 2016; Zhao et al., 2015), ubiquitination signaling (unrelated to degradation) is also known to modulate mTORC1-pathway activity (Deng et al., 2015).

Future Avenues for Drug Discovery: Targeting the m⁷G Pocket

The proposed role that m⁷G-pocket flexibility plays in regulating TOP mRNA binding provides novel opportunities for drug discovery. The DM15 conformations captured crystallographically show a cap-binding pocket that resembles those of other cap-binding proteins (Lahr et al., 2017). These similarities suggest that LARP1-DM15 ligands targeting the crystallographic m⁷G-pocket conformation may be promiscuous, leading to unacceptable side effects.

To assess the druggability of alternate, non-crystallographic m⁷G-pocket conformations, we applied FTMap (Kozakov et al., 2015; Ngan et al., 2012) to the 4ZC4:B-simulation conformation with the largest m⁷G pocket (per POVME 2.0 (Durrant et al., 2014), Figure 5B). In this conformation, a cryptic druggable pocket opens near the key α7–α8 bridge (Figure 5B, hotspot “a” in red), connected to the m⁷G pocket via a narrow channel. The bridge pocket is not directly involved in m⁷G binding and so is less likely to be conserved among cap-binding proteins (Lahr et al., 2017; Lahr et al., 2015).

The optimal pharmacological strategy for targeting the m⁷G pocket will depend on the prevailing cellular role of LARP1 (e.g., sustaining vs. repressing TOP mRNA translation). LARP1 helps regulate both TOP mRNA stability and TOP mRNA translation. We and others have observed that LARP1 represses the translation of TOP mRNAs (Fonseca et al., 2015; Philippe et al., 2017) while simultaneously protecting these (and other) mRNAs from degradation (Aoki et al., 2013; Fonseca et al., 2015; Gentilella et al., 2017). If in protecting TOP mRNAs from degradation LARP1 predominantly promotes the production of ribosomes, then compounds that disrupt TOP mRNA binding could have anti-cancer properties. DM15 ligands that bind the m⁷G pocket and extend chemical moieties into the bridge pocket could be both potent and specific direct competitors.

On the other hand, some evidence suggests that LARP1 predominantly represses TOP mRNA translation (Fonseca et al., 2015; Philippe et al., 2017), likely by preventing the assembly of the eIF4F complex on the 5’UTR of TOP transcripts (Lahr et al., 2017; Philippe et al., 2017) and sequestering those transcripts in stress granules (Wilbertz et al., 2019). If repression of TOP mRNA translation dominates, compounds that encourage LARP1 to adopt the ligand-ready conformation could have anti-cancer properties. Such compounds could block TOP mRNA translation, promoting transcript shuttling to stress granules where they are kept in a translation-repressed state. Allosteric DM15 ligands that bind the bridge pocket alone could encourage the ligand-ready conformation, perhaps by stabilizing the α7–α8 bridge in the mRNA-amenable helical conformation. Aside from serving as anti-cancer drug leads, future ligands will also be useful chemical probes for better understanding these complex translational regulatory mechanisms.

The LARP1-DM15 TSS Pocket

The TSS Pocket Governs TOP mRNA Specificity

The TSS pocket, which binds the cap-adjacent cytosine that is characteristic of TOP mRNAs (Lahr et al., 2017; Lahr et al., 2015), was far more stable in our simulations. The amino acids comprising this pocket moved little over the course of the 4ZC4:B simulation, with RMSF values of 0.85 (R847), 0.95 (F844), 1.12 (Y883), and 1.15 (R879). R847, the amino acid that recognizes the +1C Watson-Crick face of TOP RNAs, was particularly stable.

Despite this stability, the TSS pocket presents an important opportunity for LARP1-specific drug discovery. LARP1 is unique among cap-binding proteins in its ability to recognize the m⁷GpppC sequence at the 5’ ends of TOP transcripts (Lahr et al., 2017; Philippe et al., 2017). It is the +1 nucleotide, or the first nucleotide after the cap—the cytosine—that apparently determines specificity.

To test this hypothesis, we used mutagenesis to change RNA specificity from a capped-TOP transcript to a capped-non-TOP transcript (m⁷GpppC to m⁷GpppG). We specifically selected mutations that would make the TSS pocket resemble the m⁷G pocket, changing F840 to tyrosine and R847 to glutamic acid. Comparison of this so-called FYRE mutant with mutants of each individual residue shows that the conserved arginine does, indeed, determine DM15 specificity for the +1 nucleotide.

Future Avenues for Drug Discovery: Targeting the TSS Pocket

Compounds that exploit the unique, LARP1-specific TSS pocket may be less likely to bind promiscuously to other cap-binding proteins. The TSS pocket is itself fairly shallow; ligands that bind this pocket alone are unlikely to have high affinities. But m⁷G-pocket ligands that extend moieties into the TSS pocket, as do endogenous TOP mRNAs, could be developed into therapeutics with fewer toxic side effects in the clinical setting. Our ensemble-based druggability assessment (Figure 5A), which suggests that both the m⁷G and TSS pockets are druggable, supports this strategy.

DATA AVAILABILITY

Data Sharing

We will share data, materials, and computer code with the community for academic and non-commercial use upon reasonable request, within a reasonable amount of time. We will provide relevant reagent samples, as requested.

Atomic coordinates and structure factors for the reported crystal structures have been deposited with the Protein Data bank under accession number (PDBID: 6PW3).

STAR★Methods

Lead Contact and Materials Availability

Further information and requests for resources and reagents should be directed to and will be fulfilled by the Lead Contact, Jacob D. Durrant.

Experimental Models and Subject Details

NEB DH5α competent E. coli (cat# C2987) cells were used for mutagenesis and plasmid preparation of DM15. Recombinant protein expression was performed with BL21(DE3) competent E. coli cells (cat# C2527). Cells were cultured according to the manufacturer’s protocols, and protein expression was performed as detailed below.

Method Details

Model Building and Parameterization

We considered chains B and C of the 4ZC4 LARP1 DM15 structure (Lahr et al., 2015) separately. We selected these chains because they capture S923-L928, a region near the 7-methylguanosine (m⁷G) pocket, in two distinct conformations. To prepare each chain for simulation, we added hydrogen atoms using the PDB2PQR 2.1.1 (Dolinsky et al., 2007; Dolinsky et al., 2004) implementation of the PROPKA algorithm (Olsson et al., 2011), with the pH set to 7.0. PDB2PQR also optimized the hydrogen-bond network. We then used the Ambertools18 tleap program (Case et al., 2017) to add a water box extending 10 Å beyond the protein in all three dimensions; Cl⁻ counterions as required to bring the system to electrical neutrality; and Na⁺ and Cl⁻ counterions as required to achieve a 150 mM concentration.

We used Ambertools18 to parameterize each system. The protein and counterions were parameterized according to the Amber ff14SB force field (Maier et al., 2015). The water molecules were parameterized according to the TIP3P forcefield (Jorgensen et al., 1983).

Minimization, Equilibration, and Production Simulation

We minimized and equilibrated both systems (4ZC4:B and 4ZC4:C) using the NAMD 2.9 molecular dynamics simulation package (Kale et al., 1999; Phillips et al., 2005). The minimization protocol included four phases of 5,000 minimization steps each. We first relaxed hydrogen atoms; then hydrogen atoms and water molecules; then hydrogen atoms, water molecules, and protein side chains; and finally all atoms. Following minimization, four 0.25-ns isothermal–isobaric (NPT) simulations were used to equilibrate each system (310 K). We applied harmonic constraints to the protein backbone atoms. The associated restraining forces were gradually weakened: 1.0, 0.75, 0.5, and 0.25 kcal/mol/Ų for each phase, respectively. Following these equilibration steps, the systems were next subjected to extended NPT production runs. We simulated 4ZC4:B and 4ZC4:C for 500 ns each. All figures showing simulation-derived DM15 structures (Figures 1C, 2A, 2B, 3, and 5) were generated using BlendMol (Durrant, 2018).

Root Mean Square Distances and Fluctuations

We used MDAnalysis 0.16.2 (Michaud-Agrawal et al., 2011) to align each trajectory by its C_α. We extracted frames spaced 100 ps apart and calculated the RMS distances (RMSDs) between each and the corresponding first frame. Early frames were discarded because their RMSDs had not yet sufficiently equilibrated. Subsequent analysis focused on the last 88% and 90% of the 4ZC4:B and 4ZC4:C simulations, respectively.

We used frames spaced 10 ps apart to calculate RMSF values for each residue center of geometry. The RMSF values for the 4ZC4:B and 4ZC4:C simulations were calculated separately using custom Python scripts built on MDAnalysis.

Hotspot Analysis

We used the affinity propagation algorithm implemented in MDAnalysis to cluster each trajectory (AffinityPropagationNative, preference = −35). A representative conformation (centroid) was selected from each cluster. The set of centroids associated with each simulation constitutes a conformational ensemble. We analyzed each 4ZC4:B and 4ZC4:C cluster centroid with FTMap (Kozakov et al., 2015; Ngan et al., 2012) to identify druggable hotspots.

For a given protein conformation, FTMap outputs a PDB file that includes both the protein structure and multiple small organic probes docked into the identified hotspots. To more easily identify persistently druggable regions, we superimposed the FTMap outputs associated with each ensemble conformation. We wrote a Python script that converted all docked probes across all ensemble members into a single density map. For each probe atom, we generated a set of 1,000 points distributed according to a 3D Gaussian function centered on the corresponding atom, with a standard deviation of 0.5 Å. We then used MDAnalysis to bin those points into cubic voxels (0.5 Å × 0.5 Å × 0.5 Å) and to output the corresponding density map in DX format. Programs such as VMD (Humphrey et al., 1996) can visualize this density as a surface at a given isovalue. Selecting different isovalues allows one to easily identify regions with varying degrees of persistent (ensemble-wide) druggability.

POVME Analysis

We used the POVME 2.0 algorithm (Durrant et al., 2014) to track the volume of both the m⁷G and TSS (transcription start site) pockets over the course of the 4ZC4:B and 4ZC4:C simulations. One thousand aligned, regularly spaced frames from each simulation were considered.

To measure pocket shapes and volumes, POVME requires users to first identify an inclusion region that encompasses the many sampled pocket conformations. We used carefully chosen spheres to define these regions. For the m⁷G pocket, we found that a single inclusion sphere with radius 9 Å encompassed all conformations. The shape of the TSS pocket was more complex; the associated inclusion region was taken to be the union of four carefully chosen spheres of radii 5, 5, 4, and 3 Å, respectively. Each inclusion region was filled with probe points spaced 0.5 Å apart in the X, Y, and Z directions.

For each pocket conformation (simulation frame), probe points outside the pocket itself were removed. Specifically, POVME first removed all points within 1.09 Å of any protein atom. Second, POVME removed points outside the convex hull defined by the receptor atoms. POVME also removed any points that were not contiguous with a user-defined “seed region.” This region identifies portions of the pocket that are persistently open (i.e., open in all simulation frames). For the m⁷G pocket, we defined this region using a single sphere of radius 3 Å. For the TSS pocket, the region was taken to be the union of three spheres of radii 3, 3, and 2 Å, respectively.

POVME calculates the volume of each pocket conformation by counting pocket-occupying points. Each point corresponds to a cubic volume of 0.125 ų (0.5 Å × 0.5 Å × 0.5 Å). POVME also tracks how often each point is pocket occupying over the course of the simulation. It outputs a density map that can be used to determine which pocket regions are most persistently open.

Protein Purification

We cloned amino acids 796–946 of LARP1 isoform 2 from cDNA (OpenBioSystems, now ThermoFisher [BC033856]) into a modified pET28a+ vector as described previously (Lahr et al., 2015). Point mutants were generated using this vector as template for site-directed mutagenesis, using the mutagenic primers listed in the key resource table. The resulting constructs produced wild-type or mutant DM15 with an N-terminal His₆-MBP tag, followed by a tobacco etch virus (TEV) protease cleavage site and a glycine₆ linker.

KEY RESOURCES TABLE

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Bacterial and Virus Strains
NEB DH5α competent E.coli	NEB	cat#C2987
BL21(DE3) competent E.coli	NEB	cat#C2527
Chemicals, Peptides, and Recombinant Proteins
m7GpppG	NEB	cat#S1405S
Sypro Orange q	Thermo Fischer Scientific	cat#S6650
T4 PNK	NEB	cat#M0201
vaccinia capping enzyme	NEB	cat#M2080
α-32P-GTP	Perkin Elmer	cat#BLU506H250UC
His-Pur NiNTA	Thermo Fischer Scientific	cat#88222
HiTrap Q	GE	cat#17115401
HiTrap SP	GE	cat#17115201
HiTrap Butyl	GE	cat#28411005
10K MWCO	Millipore	cat#C7715
calf intestinal phosphatase	Roche	cat#11097075001
Deposited Data
Apo X-ray crystal structure of DM15	(Lahr et al., 2015)	PDBID: 4ZC4
F844YR847E DM15 X-ray crystal structure	This study	PDBID: 6PW3
Oligonucleotides
Forward and reverse primers for K924D Mutation of LARP1 DM15 5’CTGGGCCTTCTTGAAA TATTCCGACGCCAAAAAT TTGGAC3’ 5’GTCCAAATTTTTGGCGTC GGAATATTTCAAGAAGGC CCAG3’	Sigma	N/A
Forward and reverse primers for K924A Mutation of LARP1 DM15 5’CTGGGCCTTCTTGAAATATTCCGCAGCCAAAAAT TTGGAC3’ 5’GTCCAAATTTTTGGCTGC GGAATATTTCAAGAAGGC CCAG3’	Sigma	N/A
Forward and reverse primers for K921A Mutation of LARP1 DM15 5’caaatttttggctttggaatatgccaagaaggcccagaacttctcc 3’ 5’ggagaagttctgggccttcttggcatattccaaagccaaaaatttg3’	Sigma	N/A
Forward and reverse primers for K815A Mutation of LARP1 DM15 5’ cacaacacgtctaccatgcgtatcgtaggcgctgcc 3’ 5’ ggcagcgcctacgatacgcatggtagacgtgttgtg 3’	Sigma	N/A
Forward and reverse primers for F844YR847E Mutation of LARP1 DM15 R847E 5’ catctttttgttgaagtgatcctcgaggaagaaggaccagaagcg 3’ 5’ cgcttctggtccttcttcctcgaggatcacttcaacaaaaagatg 3’ F844YR847E 5’catacatctttttgttgaagtgatcctcgaggaaataggaccagaagcggaagagtgtgt 3’ 5’acacactcttccgcttctggtcctatttcctcgaggatcacttcaacaaaaagatgtatg 3’	Sigma	N/A
Splint adapter 5’ CTTGAAGCAGCTGAACGCCTCCGAGGCGCCACGGAAAAGAGG 3’	Sigma	N/A
Recombinant DNA
human LARP1 isoform 2	ThermoFisher	BC033856
Software and Algorithms
PDB2PQR 2.1.1	(Dolinsky et al., 2007; Dolinsky et al., 2004)	http://nbcr-222.ucsd.edu/pdb2pqr_2.1.1/
Ambertools18	(Case et al., 2017)	http://ambermd.org/AmberTools.php
NAMD 2.9	(Kale et al., 1999; Phillips et al., 2005)	http://www.ks.uiuc.edu/Research/namd/
MDAnalysis 0.16.2	(Michaud-Agrawal et al., 2011)	https://www.mdanalysis.org/
FTMap	(Kozakov et al., 2015; Ngan et al., 2012)	https://ftmap.bu.edu/
VMD	(Humphrey et al., 1996)	https://www.ks.uiuc.edu/Research/vmd/
Phenix 1.13–2998	(Adams et al., 2010)	https://www.phenix-online.org/
POVME 2.0	(Durrant et al., 2014)	rocce-vm0.ucsd.edu/data/sw/hosted/POVME/
QuantStudio™ Design & Analysis Software V 8.1.0.0	ThermoFisher Scientific	https://www.thermofisher.com/us/en/home/technical-resources/software-downloads/ab-quantstudio-3-and-5-real-time-pcr-system.html
PyMOL Molecular Graphics System, Version 1.8.6.0.	(DeLano, 2002)	https://pymol.org/2/
GraphPad Prism 7.0	GraphPad Software	Prism - graphpad.com
SPSS Statistics V.25	IBM	https://www.ibm.com/products/spss-statistics

Expression plasmids were transformed into BL21(DE3) E. coli and grown overnight on LB agar plates supplemented with 30 g/ml kanamycin. The His₆-MBP-DM15 fusion protein was expressed by autoinduction (Studier, 2005) for 3 hours at 37°C, and then for 18 hours at 18°C. Cells were collected by centrifugation, flash frozen in liquid nitrogen, and stored at −80°C until used.

Cells (~2g) were resuspended at 4°C by gentle stirr ing in 50mL NiNTA lysis buffer (50 mM Tris-HCl, pH 7.5 or 8, 400 mM NaCl, 10 mM imidazole, 10% glycerol). Protease inhibitors PMSF (1 μM final concentration), leupeptin (0.1 μM final), and aprotinin (0.1 mM final) were added. Cells were lysed by homogenization, followed by clarification via centrifugation (12,000 RPM) at 4°C for 30 minutes. The soluble fraction was nutated with 4 mL equilibrated HisPur Ni-NTA Resin (ThermoFisher) for 2 hours at 4°C. The beads were washed two times in 50 mL lysis buffer and three times with wash buffer (50 mM Tris-HCl, pH 7.5, 400 mM NaCl, 35 mM imidazole, 10% glycerol). His₆-MBP-DM15 fusion protein was eluted from beads in 30 mL elution buffer (50 mM Tris-HCl, pH 7.5, 400 mM NaCl, 250 mM imidazole, 10% glycerol). The N-terminal His₆-MBP tag was removed by the addition of 2 mg TEV protease for cleavage overnight in 10K MWCO SnakeSkin dialysis tubing (ThermoFisher) in 2 L of dialysis buffer (50 mM Tris-HCl, pH 8, 150 mM NaCl, 1 mM EDTA, 10% glycerol, 1 mM DTT) at 4°C.

Cleaved DM15 protein was further purified by tandem HiTrap Q and HiTrap SP columns (GE Lifesciences). DM15 protein free of nucleic acid contaminants was eluted off the HiTrap SP column with gradient from 150 mM NaCl to 1 M NaCl over 50 mL. MBP flowed through both columns, untagged DM15 eluted at 33 mS/cm, and uncleaved fusion protein eluted at 20 mS/cm, allowing for efficient separation of target DM15 construct. Fractions containing DM15 were pooled and brought to 1 M ammonium sulfate by the dropwise addition of 3 M ammonium sulfate with gentle swirling. The protein was diluted to 40 mL in 50 mM Tris-HCl, pH 7, 1 M ammonium sulfate and loaded onto a 5 mL Butyl HP column (GE Lifesciences) at 0.5 mL/min. The Butyl HP column was eluted over a 10 CV gradient to 50 mM Tris-HCl, pH 7, 2 mM DTT. Fractions containing DM15 were collected and concentrated to 20 mg/mL using a 10K MWCO spin concentrator (Millipore) in 50 mM Hepes, pH 7, 50 mM NaCl, 2 mM DTT for crystallographic experiments, or to ~2.0 mg/mL in 50 mM Tris-HCl, pH 7.5, 250 mM NaCl, 25% glycerol, 2 mM DTT for biochemical experiments. Protein was flash frozen in 10 μL aliquots and stored at −80°C for further use.

Crystallization and Structure Solution

The F884Y/R847E (FYRE) mutant was concentrated to ~10 mg/mL in 25 mM Hepes, pH 7.0, and 75 mM NaCl. FYRE and m⁷GpppG (NEB cat#S1405S) were incubated at a 1:1.2 ratio at room temperature (final concentrations of 480 μM (9 mg/mL) DM15 and 576 μM m⁷GpppG). Crystals were set up using hanging drop vapor diffusion with a 1:1 ratio of complex to mother liquor in a final drop volume of 3 μL. The initial crystals grown in 100 mM Hepes, pH 7.5, 0.15 M NaCl, and 15% PEG 3350 diffracted poorly. To optimize crystals for publication-quality diffraction, crystals were slowly equilibrated into 100 mM Hepes 7.5, 0.08 M NaCl, and 36% PEG 3350 using the following scheme: crystals grew overnight and were allowed to equilibrate for 24 hours in the original hit condition. Every 24 hours, 0.5 L of a new mother liquor solution was added to the drop, and the coverslip was moved to a well containing the new mother liquor. The salt was decreased by 5 mM and the PEG 3350 was increased by 1.5% per transfer, for a final well solution of 100 mM Hepes 7.5, 0.08 M NaCl, and 36% PEG 3350.

X-ray diffraction data were collected at NLS-II 17-ID-1 (AMX) at wavelength 0.92 Å using an Eigen 9M detector. An initial map was generated by molecular replacement using a 3 Å trimmed model of chain A of 4ZC4 (Lahr et al., 2015) as a search model. To remove bias from the initial search model, simulated annealing composite omit maps were used to confirm amino acid register [Phenix (Adams et al., 2010), CCP4 (Winn et al., 2011)]. Iterative building and refinement using xyz coordinates, real-space, occupancies, and individual B-factor parameters were performed in Coot (Emsley et al., 2010) and Phenix (Adams et al., 2010), respectively. Figure 4C was generated with the PyMOL Molecular Graphics System (Schrödinger, LLC).

Protein Melting Temperature Assays

Three technical triplicates were performed using an 80 μL reaction mixture composed of 20 μL (4x) 40% glycerol/4 mM β-mercaptoethanol mix; 4 μL 1M Tris, pH 8; 1.33 μL 3 M NaCl; and 8 μL 100x SYPRO orange (Thermo Fisher Scientific, cat#S6650). Protein was added to a final concentration of 5 μM. Reactions were incubated at room temperature for 10 min. The three 20 μL technical replicates were performed using the respective reaction mixes in separate wells of a 96-well real-time PCR plate. The final concentration of the buffer components varied slightly depending on the volume of protein required to reach 5 μM final concentration. All reactions contained a final concentration of 10x SYPRO orange and 1 mM β-mercaptoethanol, but the final concentrations of the other buffer components were 50 mM-53 mM Tris pH 8; 55 mM-67mM NaCl; 37 μM-130 μM DTT; and 11%–12% glycerol.

Protein unfolding was measured by monitoring the fluorescence of SYPRO orange using the x3-m3 peak channel of the QuantStudio™ 3 System (Thermo Fisher Scientific) during a temperature ramp from 30–90°C. To calculate melting temperatures, fluorescence data were analyzed using the melt temp feature of the QuantStudio™ Design & Analysis Software, per the QuantStudio™ 3 and 5 Real-Time PCR Systems Installation, Use, and Maintenance Guide (Pub. No. MAN0010407, Rev. C.0). Three independent replicates were performed, each consisting of three averaged technical replicates.

RNA Preparation

The 5’-triphosphorylated 20-mer RPS6 TOP-RNA sequence (CCUCUUUUCCGUGGCGCCUC) was synthesized on a 1 μmol scale using a MerMade4 DNA synthesizer (Bioautomation, Irving, TX) with ultramild RNA and 2′-OMe RNA phosphoramidites with phenoxyacetyl (PAC), acetyl (Ac), or isopropyl-phenoxyacetyl (iPrPAC) protecting groups; CPG supports; and standard solid-phase synthesis reagents (Glen Research, Sterling, VA and Chemgenes, Wilmington, MA). The 5’ triphosphate was installed on the RNA on solid support using freshly prepared tributylammonium pyrophosphate according to the procedures and protocol for 5’-phosphitylation, hydrolysis, oxidation, and substitution (with the tributylammonium pyrophosphate) in ref. (Zlatev et al., 2012). Following synthesis, cleavage from the CPG beads and deprotection was performed using standard protocols. After deprotection, RNA was analyzed for purity using reverse phase HPLC. The HPLC system consisted of a Waters 1525 pump system, a Waters 2998 photodiode array detector, a Waters XBridge OST C18 Column 2.5 um (4.6 × 50 mm) in 0.1 M triethylamine acetate, and 80:20 acetonitrile:water in 0.1 M triethylamine acetate at 25°C. Mass spectrometry of the RNA sequence was performed on an Applied Biosystems Voyager DE-STR MALDI-TOF instrument in positive mode using a 3-hydroxypicolinic acid matrix. Mass calculated: 6445; mass found: 6443 [M-2H]. Both the TOP and non-TOP (+1G) 5’-triphosphate RNAs were capped using NEB vaccinia capping enzyme (M2080) and α−³²P-GTP, and then gel extracted.

Electrophoretic Mobility Shift Assays

Each binding reaction contained the indicated final concentration of recombinant human LARP1 DM15, 500 counts of radiolabeled RNA (<2 nM final concentration), binding buffer (20 mM Tris–HCl, pH 8, 150 mM NaCl, 10% glycerol, 1 mM DTT), 1 g bovine serum albumin (BSA), and 0.5 g tRNA. Reactions were incubated on ice for 30 minutes and analyzed on an 7% polyacrylamide (29:1) native 0.5× TBE native gel. Gels were run at 4°C before being dried and exposed overnight to phosphor screens and scanned with a Fuji plate reader. Pixel density of shifted (complexed) RNA over total counts was quantified using Image Quant and graphed using GraphPad Prism.

Quantification and Statistical Analysis

We used IBM® SPSS® Statistics 25 for statistical analyses. To compare the means of two populations, we first used Levene’s Test of Equal Variances to determine whether equal population variances could be assumed. We then used an independent-samples two-tailed t-test to assess the difference in populations means, selecting the appropriate p-value given the conclusion of Levene’s test.

For protein melting temperature assays, three 20 μL technical replicates were performed (n = 3).

For the α7-α8-bridge probability distribution analyses, we compared helical and loop datasets (helical bridge, n = 117; disordered bridge, n = 883).

EMSAs (n = 4) where analyzed with ImageQuantTL by quantifying the free RNA (unbound) pixel density and bound RNA (complex) pixel density. The ratio of bound over total pixel density was plotted as a function of protein quantity using GraphPad Prism 7 (Prism version 7.00 for Windows, GraphPad Software, La Jolla California USA, www.graphpad.com). Curve fitting and affinity calculations were performed using the model representing one site-specific binding to saturation.

Data and Software Availability

Coordinates of the structure described in this article have been deposited in the PDB with accession number PDBID: 6PW3.

Supplementary Material

2

Movie S1. The 4ZC4:B simulation. The crystallographic conformation of the α7–α8 bridge (helical) is shown in yellow ribbon. The simulated protein is shown in green ribbon. The same key residues highlighted in Figure 3 are shown in sticks representation. Related to Figure 2.

Highlights.

• Simulations and experiments provide insight into how LARP1 binds TOP mRNA.

• Key residues contribute to cap-recognition dynamics and mRNA specificity.

• Druggable LARP1 pockets suggest avenues for anti-cancer drug discovery.