Evolutionary plasticity of the NHL domain underlies distinct solutions to RNA recognition

Abstract

RNA-binding proteins regulate all aspects of RNA metabolism. Their association with RNA is mediated by RNA-binding domains, of which many remain uncharacterized. A recently reported example is the NHL domain, found in prominent regulators of cellular plasticity like the C. elegans LIN-41. Here we employ an integrative approach to dissect the RNA specificity of LIN-41. Using computational analysis, structural biology, and in vivo studies in worms and human cells, we find that a positively charged pocket, specific to the NHL domain of LIN-41 and its homologs (collectively LIN41), recognizes a stem-loop RNA element, whose shape determines the binding specificity. Surprisingly, the mechanism of RNA recognition by LIN41 is drastically different from that of its more distant relative, the fly Brat. Our phylogenetic analysis suggests that this reflects a rapid evolution of the domain, presenting an interesting example of a conserved protein fold that acquired completely different solutions to RNA recognition.

Fig. 1

Structure vs. sequence preference of RNA-binding proteins. a Schematics describing the meta-analysis of 260 RNAcompete experiments. The top 2% of RNAs enriched in pulldowns were considered as bound fraction. All possible 7-mer sequence and 11-mer structure motifs were counted in the bound and unbound fractions to calculate enrichment of each motif. The enrichment values were then Z-value transformed and average Z-values of the top 10 motifs were calculated. The schematics shows these calculations for one of the 260 RNAcompete experiments as an example. b Average Z-values of the top 10 sequence motifs were plotted against the top 10 structure motifs, for each RNA-binding experiment, comparing preference for sequence vs. structure. The top three outliers with high structure preference, LIN-41, TRIM71, and Wech, all related TRIM-NHL proteins, are shown in red. Other TRIM-NHL proteins included in the meta-analysis are shown in blue (NCL-1, Brat, Mei-P26, TRIM56; Brat is marked twice, as it was analyzed in two separate studies)

Fig. 2

CeLIN41 recognizes structured RNA. a Enrichment for each 11-mer structure motif in the CeLIN41 RNAcompete experiment was plotted against its overall occurrence in the RNA pool. Top 30 enriched 11-mer structures are highlighted in red and the corresponding dot-bracket strings are indicated. The dot-bracket strings correspond to a stem-loop motif, with exactly three nucleotides in the loop, schematically represented above. As controls, all structure 11-mers containing a loop with four, five or six nucleotides, highlighted in cyan, blue and purple, did not show any enrichment. b CeLIN41 binding to parts of mab-10 3′-UTR. Top: schematics of the mab-10 3′-UTR divided into six overlapping RNA probes (part 1 to 6), each of 200 nucleotides. Tri-loop SLs are present in parts 2 and 4. Single red asterisk on one side of the stem indicates a stem disrupting mutation (mut) and two red asterisks, one on each side of the stem, indicate a stem disrupting and the compensatory stem-restoring mutations (mut-res). Respective sequences are listed in Supplementary Table 1. Bottom: in gel-shift experiments, CeLIN41 bound to parts 2 and 4 of mab-10 3′-UTR. Mutations disrupting the stems in part 2 and part 4 nearly abolished CeLIN41 binding. Compensatory mutations in the stems (mut-res) restored the binding to wild-type levels. c Top: the “wt” RNA corresponds to a synthetic “condensed” mab-10 3′-UTR reporter construct containing five stem-loops (I to V). Asterisks denote mutations as in b. Respective sequences are listed in Supplementary Table 1. Bottom: in gel-shift experiments, CeLIN41 bound the synthetic mab-10 condensed 3′-UTR. Mutations disrupting the stem (mut) abolished binding, whereas compensatory mutations (mut-res) restored binding. d Micrographs of early L3-stage C. elegans larvae, treated with either lin-41 or mock RNAi, showing reporter GFP in hypodermal nuclei (white circles demarcate nuclei), expressed from the lin-29A promoter under the control of unregulated act-1 5′-UTR and unc-54 (ctrl) 3′-UTR. The 3′-UTR insert corresponds to the constructs in c. The mab-10 condensed 3′-UTR imposed CeLIN41-mediated repression on the GFP reporter. Mutations disrupting the stem abolished this regulation (white arrowheads point to GFP-expressing nuclei), while compensatory mutations reinstated the repression. Scale bars, 10 µm

Fig. 3

HsLIN41 recognizes C. elegans RNA stem-loops. a Schematics describing the luciferase assay and reporters used in this assay. 3′-UTR inserts transplanted into the RL (Renilla luciferase) construct are shown. b HsLIN41 specifically downregulated Renilla luciferase (RL) activity when sequences containing the SLs (from either mab-10 3′-UTR (151-650 nt) or mab-10 condensed SL 3′-UTR) were transplanted into a control, unregulated 3′-UTR. The mab-10 condensed 3′-UTR, with stem-disrupting mutations (mut), was not repressed by HsLIN41. Mutations that restored the SLs (mut-res) also restored the repression. Bars in the graph represent the mean between three biological replicates

Fig. 4

Crystal structure of the C-terminal part of D. rerio LIN41. a The crystal structure of the DrLIN41 filamin-NHL domains is displayed in a cartoon mode, with a transparent grey surface in two orientations rotated by 90°. The molecule is colored from blue (N terminus) to red (C terminus) to indicate the topology. Protein domains, termini, and β-propeller blades are labeled for better clarity. b Top view of the RNA-binding site of the DrLIN41 NHL domain with the electrostatic surface potential mapped onto the molecular surface. Surface potential is computed by using the APBS plugin implemented in PyMOL (www.pymol.org) and is displayed from – 5.0 kT/e (red, acidic) to + 5.0 kT/e (blue, basic). c DrLIN41 filamin-NHL domains in complex with the lin-29A stem-loop RNA. The filamin and NHL domains are shown in cartoon mode in blue, with a white transparent surface. The lin-29A RNA fragment, forming a hairpin, is displayed as a cartoon with nucleotides in different colors (guanine: green, adenine: blue, cytosine: orange, uracil: cyan). d Magnified view of the lin-29A RNA stem loop bound to the DrLIN41 NHL surface (colors as in c). A diagram detailing the nucleotide composition of the stem loop is shown above

Fig. 5

Molecular interactions underlying LIN41 binding to RNA SLs. a, b Detailed views of interactions between the lin-29A RNA and the DrLIN41 NHL propeller. Nucleotides and protein side chains are highlighted and their directly interacting residues are shown as sticks; the remaining parts are shown as lines (RNA) or ribbons (protein). Hydrogen bonds are presented as dotted lines and hydrophobic interactions as solid lines. Nucleotides are colored as in Fig. 4d, whereas protein side chains are colored according to the mutational analysis. c Schematic representation of the lin-29A RNA hairpin and of its interactions with DrLIN41 NHL residues (type of interaction and color code as in a, b). d Expression of mutant HsLIN41 proteins did not severely down-regulate Renilla luciferase (RL) reporter expression unlike the wild-type HsLIN41, when a fragment corresponding to the mab-10 condensed 3′-UTR was transplanted into an unregulated 3′-UTR of the reporter construct. Bars in the graph represent the mean between three biological replicates

Fig. 6

The LIN41 response element. a Schematics depicting RNA features used to build the LIN41 Response Element (LRE) model. Considered were all possible bases in the three loop positions (I, II, and III) and all possible base pairs at the stem position 1 (– 1/ + 1). The pairing probability of stem position 1 was determined by the relative occurrence of all possible structures that a particular RNA sequence can acquire. The pairing probabilities were grouped into seven bins on a log2 scale. Combining the sequence and structure features resulted in 2688 (6 × 64 × 7) RNA motif variants. b A heat map showing the average CeLIN41-binding scores from the RNAcompete experiment for all RNAs containing any particular motif variant as described in a. Pairing probability and base pairs at stem position 1 are shown on the left and the right of the heat map respectively. The loop (I, II, and III) sequences are shown in two rows, for clarity, at the bottom of the heat map. The data were clustered based on the CeLIN41-binding score. The overall distribution of pairing probabilities is shown on top of the pairing probability scale. Bottom right: the drawing represents a stem-loop motif, based on the model, referred to as the LIN41 Response Element (LRE). Yellow: data not available (RNA motif variants supported by < 20 oligo sequences)

Fig. 7

LIN41 binds to LREs both in vitro and in vivo. a Fluorescence polarization (FP) assays determining binding constants of CeLIN41 to LRE variants in the position III of the loop. Raw FP data of CeLIN41, interacting with a wild-type LRE (SL I in Supplementary Fig.1b), a control stem-loop RNA with five nucleotides in the loop, and LREs mutated at loop position III, are shown in units of millipolarization (mP). The equilibrium dissociation constant (K_D) is shown for the WT LRE. Each data point is a mean of three experiments and the error bars represent the standard deviation. b The FP assays determining binding constants of CeLIN41 to LRE variants in the position 1 of the stem. Raw FP data of CeLIN41, interacting with WT LRE and LREs mutated at stem position 1, are shown in units of millipolarization (mP). Each data point is a mean of three experiments and the error bars represent the standard deviation. The WT LRE data is the same as in a. It is replotted for easy comparison with the mutants. c Contribution of LREs of varying strengths, predicted by the model in Fig. 6b, present in 5′-UTRs, coding sequences (CDS) and 3′-UTRs, to CeLIN41 binding as determined by linear regression. RNA binding was assayed by co-precipitation with CeLIN41, followed by RNA sequencing (RIP-seq). The error bars represent SEs for the coefficients obtained from the linear regression

Fig. 8

RNA binding preferences of LIN41 and Brat. a Left: crystal structure of the DmBrat NHL domain in a complex with a single-stranded linear RNA (PDB 4ZLR). The protein surface is colored by the electrostatic surface potential from – 8 kT/e (red, acidic) to + 8 kT/e (blue, basic) and the RNA is shown as a cartoon. The approximate footprint of the RNA interaction on the protein surface is shown as a dotted red line. Right: magnified view of the RNA-binding site. A fragment of the interacting protein surface is shown and colored as on the left. The RNA is shown in surface mode, with carbon atoms in green and other atoms in standard colors. Nucleotide positions are labeled as in the RNA sequence displayed below. b Left: crystal structure of DrLIN41 in complex with the lin-29A RNA stem loop. The protein surface is colored as in a. Right: magnified view of the RNA-binding site in an orientation rotated by 90°. The RNA stem loop is shown in surface mode, with carbon atoms in gold and other atoms in standard colors. The corresponding RNA sequence is on the right

Fig. 9

Evolutionary relationships between NHL domains. Left: a simplified phylogenetic tree of NHL domains, with a focus on putative RNA-binding proteins. For details, see Supplementary Fig. 5. Branches comprising multiple sequences from one subfamily (named after a representative protein) have been collapsed and are illustrated as triangles. The apex of a triangle indicates the start-point for an individual branch, whereas the width of the base indicates the number of the members that constitute the branch. The tree is drawn to scale, with branch lengths in the same units as evolutionary distances used to build the phylogenetic tree. Bootstrap values are shown for all nodes. Right: structural characteristics of the selected representatives of the seven subfamilies; from top to bottom: DrLIN41, CeLIN41, DmWech, CeNHL-1, HsTRIM3, DmMei-P26, and DmBrat. All structures are shown in surface representation. The left column depicts either experimentally determined crystal structures (DrLIN41 and DmBrat), or computationally modeled 3D structures of the proteins, with the mapped electrostatic potential on the solvent accessible surface from – 5 kT/e (red, acidic) to + 5 kT/e (blue, basic). The middle column depicts the sequence conservation within the TRIM71 subfamily (top image) and between the TRIM71 subfamily and each of the other subfamilies, mapped on the DrLIN41 crystal structure. The right column depicts the sequence conservation within the Brat subfamily (bottom image) and between the Brat subfamily and each of the other subfamilies, mapped on the DmBrat crystal structure. The middle and right panel use the same coloring scheme: red for invariant residues (within the reference subfamily, i.e., TRIM71 or Brat, respectively, or between the reference family and the compared subfamily), orange and yellow for partially conserved residues, green for weakly conserved residues, and blue for non-conserved residues