Conserved asymmetry underpins homodimerization of Dicer-associated double-stranded RNA-binding proteins

Abstract Double-stranded RNA-binding domains (dsRBDs) are commonly found in modular proteins that interact with RNA. Two varieties of dsRBD exist: canonical Type A dsRBDs interact with dsRNA, while non-canonical Type B dsRBDs lack RNA-binding residues and instead interact with other proteins. In higher eukaryotes, the microRNA biogenesis enzyme Dicer forms a 1:1 association with a dsRNA-binding protein (dsRBP). Human Dicer associates with HIV TAR RNA-binding protein (TRBP) or protein activator of PKR (PACT), while Drosophila Dicer-1 associates with Loquacious (Loqs). In each case, the interaction involves a region of the protein that contains a Type B dsRBD. All three dsRBPs are reported to homodimerize, with the Dicer-binding region implicated in self-association. We report that these dsRBD homodimers display structural asymmetry and that this unusual self-association mechanism is conserved from flies to humans. We show that the core dsRBD is sufficient for homodimerization and that mutation of a conserved leucine residue abolishes self-association. We attribute differences in the self-association properties of Loqs, TRBP and PACT to divergence of the composition of the homodimerization interface. Modifications that make TRBP more like PACT enhance self-association. These data are examined in the context of miRNA biogenesis and the protein/protein interaction properties of Type B dsRBDs.

Chemical shift assignment of WT PACT-D3 NMR spectra To assign the NMR signals to the correct states, triple resonance backbone assignment experiments (CBCANH and CBCA(CO)NH) were first used to generate fragments of sequentially linked spin systems. Within each fragment, the two states could be distinguished by their differing Ca and Cb chemical shifts. The boundaries of the fragments were formed by prolines, or residues with similar Ca and Cb chemical shifts in the two states. To link the fragments, 15 N-and 13 C-NOESY-HSQC spectra were recorded, and NOEs between fragments identified. This allowed internally consistent state assignment for residues 256-288 and 294-313 to be assigned. For the remaining residues, no unambiguous inter-fragment NOEs could be identified, so the state could not be assigned. Note that correct state assignments are not required for calculation of the chemical shift difference between states.
Fitting of EXSY data ( 1 H, 15 N) EXSY spectra (1) were recorded with mixing times between 0.1 s and 1 s, and intensities of well resolved peaks were fit using CCPN Analysis 2.3 (2). These intensities were used to calculate a parameter X for each residue at each time point (3) This treatment removes the effects of differential relaxation between the two states up to third order in time. The calculated X values were fit to a quadratic model using R, and the confidence interval calculated using bootstrapping.

Supplementary tables
Supplementary table 1: Discarded hypotheses for the cause of the two-state behaviour of PACT-D3 observed in NMR spectra. Hypothesis Evidence Monomer and dimer exchange SEC-MALLS shows PACT-D3 is entirely dimeric at concentrations well below those used for NMR spectroscopy. NMR intensity and linewidth similar for both states. Folded and misfolded exchange TALOS-N secondary structure analysis and amide resonance dispersion indicates that both states are folded.

Bound ions from purification
The peak doubling is reproducible between samples, and remained in the presence of 10 mM EDTA. Alternate disulphide bonding Experiments were carried out in the presence of 10 mM TCEP, and side chain Ca and Cb chemical shifts suggest that all cysteines are reduced. Alternate histidine protonation states A version of PACT-D3 with an H285A mutation showed two states by NMR, so protonation of H285 cannot be responsible for the two states. H300 is solvent exposed and distant from the residues with largest chemical shift differences between states, so is also unlikely to be responsible.

Proline isomerisation
Side chain Cb and Cg chemical shifts indicate that P280 adopts the cis isomer in both states. Dimer domain swapping Exchange between the two states is too fast -previous examples of domainswapped homodimers exchange on a timescale of hours to days (4, 5)

Figure S1
The amino acid sequences of PACT, TRBP and Loqs homologues were retrieved from NCBI, and aligned using ClustalX with default settings. The alignment of domain 3 is shown. The sequences are separated into groups on the left based on whether they are most similar to human PACT, TRBP or Drosophila Loqs. Branchiostoma floridae and Crassostrea gigas have only a single PACT homologue, which shares features with both PACT and TRBP. A yellow highlight with red text indicates >70% sequence identity; residues with 100% conservation are highlighted in black. Specific residues discussed in the main text are indicated above the alignment in blue (for PACT) or green (for TRBP). The secondary structure of Loqs D3 from PDB accession 4X8W is shown below the alignment. UniProt accessions (in order shown in figure): O75569; NP_001039335; EDL27212; EDL79230; XP_006935445; XP_008116415; Q7ZYA5; XP_005172565; NP_599150; NP_001069146; XP_003988820; NP_033345; NP_001030113; XP_018100523; XP_008101858; NP_956291; XP_002587993; XP_011456094; AAF53295; EHJ63609; gnl|Amel_4.5|GB47214-PA; EFX79291. Note that the accession given for Apis mellifera Loqs is for the Amel_4.5_OGSv3.2_pep database of predicted protein sequences based on the Apis mellifera genome assembly 4.5, not for UniProt.

Drosophila melanogaster Loqs(4X8W)
Drosophila melanogaster Loqs(4X8W) Figure S2 SEC-MALLS of A) FL PACT and FL PACT L273R; and B) FL PACT, PACT-D12 and PACT-D3. In these cases, the column was pre-equilibrated with 20 mM phosphate pH 6.5, 500 mM NaCl, 5 mM DTT and 5 % glycerol, and data was processed using a dn/dc value of 0.165. C) Electrophoretic mobility shift assays testing for binding between PACT constructs and pre-miR-155.

Figure S3
secondary structure prediction, and the black line shows the predicted S 2 order parameter (lower values indicate greater flexibility). D) Histograms of the relative population of state A (IA/(IA + IB)) for well resolved residues under different conditions. Regardless of temperature, pressure, or low concentration of denaturants, the distribution of populations remains centred around 0.5, showing that the populations of states A and B remain equal. Deviations from 0.5 are likely due to differential relaxation between the two states. Except for the condition being varied, all spectra were recorded at 25˚C at atmospheric pressure.

Figure S6
A) NOESY experiments detect magnetisation transfer between protons separated by less than about 5 Å. 13 C-filtered NOESY-HSQC experiments filter out some of these signals, leaving only cases where magnetisation was transferred from a proton attached to 12 C, to a proton attached to 13 C. In some cases, signals arising from protons attached 13 C are not fully suppressed. However, these cases can be recognised because no decoupling is applied during evolution of this proton, resulting in the signal appearing as a doublet. B) In a 50:50 mixture of 13 C-labelled and unlabelled PACT-D3, a mixture of 13 C/ 13 C-, 13 C/ 12 C-and 12 C/ 12 C-labelled homodimers will be present. The dimer interface of the 13 C / 12 C-labelled homodimers is the only region in which 13 C and 12 C are in close proximity. Therefore, only intermolecular NOEs from the dimer interface are observed in 13 C-filtered NOESY-HSQC experiments.

T283B H γ2
V283A H γa from V283 in state A and the g 2 methyl group of T282 in state B. The assigned ( 1 H, 13 C)-HSQC is shown in grey. The ( 1 H, 13 C) planes are shown to demonstrate that these NOEs can be assigned despite the large peak width in the 13 C dimension of the 13 C -filtered NOESY. D) Strips from a 13 C-filtered NOESY-HSQC of PACT-D3 showing intermolecular contacts between L273 H d2 in state A, and Q304 H b2 and Y305 H a and H b2 in state B.

Figure S7:
Structural comparison of the two subunits of the Loqs-D3 dimer (chains A and D/F from PDB accession 4X8W). A) Superposition of the two subunits of the homodimer reveals only minor differences in structure between the two protomers, with a C a RMSD of approximately 1 Å. B) However, if the entire homodimer is rotated and superposed so that two protomers (orange, red) are optimally aligned, the other subunits (blue, green) are displaced from one another by approximately 8 Å. Figure S8: A-C) ( 1 H, 15 N)-HSQC spectra of A) wild-type TRBP-Ext-D3, B) TRBP-Ext-D3 R353H R354N, and C) TRBP-Ext-D3 L326R. The observed and predicted numbers of peaks are shown in the top left of each spectrum. The HSQC of TRBP-Ext-D3 shows fewer peaks than predicted, likely due to intermediate exchange between monomeric and asymmetric dimer states on the millisecond timescale.  Figure S9: A) Sequence alignment of PACT, TRBP and Loqs domain 3. The secondary structure of Loqs domain 3 (from 4X8W) is shown above, while circles below indicate residues at the Loqs homodimer interface (green) and the interface between TRBP and Dicer (purple). Open circles show residues that are only buried in one half of the Loqs-D3 homodimer. Surface residues were determined using PISA v1.51 (6). B) Comparison of Loqs-D3 dimer and TRBP-D3/Dicer-Ins structures.