Integrative structural analysis of NF45-NF90 heterodimers reveals architectural rearrangements and oligomerization on binding dsRNA

Abstract

Complexes of nuclear factors 45 and 90 (NF45-NF90) play a multitude of roles in co- and post-transcriptional RNA processing, including regulating adenosine-to-inosine editing, cassette exon and back splicing, and splicing fidelity. NF45-NF90 complexes recognize double-stranded RNA (dsRNA) and, in human cells, primarily interact with Alu inverted repeats (AluIRs) that are commonly inserted into introns and other non-coding RNA regions. Intronic AluIRs of ∼300 bp can regulate splicing outcomes, such as generation of circular RNAs. We examined domain reorganization of NF45-NF90 domains on dsRNAs exceeding 50 bp to gain insight into its RNA recognition properties on longer dsRNAs. Using a combination of phylogenetic analysis, solution methods (including small angle X-ray scattering and quantitative cross-linking mass spectrometry), machine learning, and negative stain electron microscopy, we generated a model of NF45-NF90 complex formation on dsRNA. Our data reveal that different interactions of NF45-NF90 complexes allow these proteins to coat long stretches of dsRNA. This property of the NF45-NF90 complex has important implications for how long, nuclear dsRNAs are recognized in the nucleus and how this might promote (co)-regulation of specific RNA splicing and editing events that shape the mammalian transcriptome.

Graphical Abstract

Figure 1.

Folded domains and functional sites are conserved and depleted of benign population variants. (A) Primary structures of NF110 and NF90. Folded domains are indicated by rectangles along a horizontal line representing the length of the protein from N to C terminus. Fractional values associated with domains are missense depletion per domain over depletion over the whole protein (V_d/V_p) ratios. Lollipops are phosphorylation sites, RGG sequences are marked with a vertical line. Three tracks show conservation (from ConSurf, using a gradient of 9 conservations scores, as indicated in the key); the positions of population missense variants extracted from gnomAD (vertical lines); and positions of secondary structure elements as observed in published structures. (B) Similar analysis with equivalent tracks for NF45, which has only one domain.

Figure 2.

NF45–NF90 complexes in solution show compaction of domains. (A) Overview of constructs used for structural analyses. (B) SEC–SAXS profiles of three constructs. (C) Scattering curves and Guinier analysis (inset) of samples from panel (B). (D) P(r) functions derived from panel (C). (E) Bead models for NF45–NF90 constructs with D_max and calculated MW values (from amino acid composition) or estimated MW from SAXS analysis.

Figure 3.

Solution analysis of NF45–NF90 binding to dsRNA of increasing lengths. Eight SAXS measurements were carried out with no dsRNA or with 25-, 36-, and 54-bp RNA at increasing molar ratios of protein:dsRNA. (A) Intensity profiles of samples as eluted from SEC. Sample identities for all graphs are given in the inset key. (B) Real space profiles of all SAXS samples with dotted lines showing calculated D_max on the curve. Associated bead models are shown with D_max values indicated under each model.

Figure 4.

NF45–NF90 complexes undergo conformational change on binding to dsRNA. CLMS on NF45–NF90 complexes in solution without dsRNA (A) and with dsRNA (B). Intramolecular cross-links within NF90 and NF45 (arcs) are distinguished from intermolecular cross-links (lines). Domain limits are shown as in Fig. 1. Quantitative differences in cross-linking comparing free NF45–NF90 (C) and dsRNA-bound NF45–NF90 (D). Cross-links with at least two-fold differences are shown while unquantified cross-links are not displayed. (E) Mapping of cross-links on to NF90_dsRBDs. The asymmetric unit of the crystal structure is shown along with the closest symmetry equivalent of the NF90–dsRNA complex. DsRBD domains are cyan and purple with symmetry mates in lighter colours. dsRNA is shown in grey and as sticks. Shortest possible cross-links are shown in dark blue and Cα–Cα distances <24 Å are annotated. (F) A rotated view of the same structure is shown. Cα–Cα distances >24 Å are annotated.

Figure 5.

NF45–NF90_long complexes oligomerize on long stretches of dsRNA. (A) RNA-dependent cross-links between NF90_DZF and NF45_DZF displayed on a single heterodimer are longer than the distance constraint for EDC cross-linkers. (B) Surface of an NF45_DZF–NF90_DZF heterodimer showing a gradient of likelihood of RNA binding, based on pyRBDome analyses (none-to-high, value range is given below gradient bar). (C) Combination of two NF45_DZF–NF90_DZF heterodimers into a higher order oligomer model, showing a possible binding site for dsRNA. (D) Re-analysis of cross-links in panel (A) in the context of an open-ended oligomer model. (E) Negative stain electron micrograph of NF45–NF90_long protein alone; dsRNA with a maximum length of 410 bp; 10 mg/ml NF45–NF90_long mixed with dsRNA with a maximum length of 310 bp; and 10 mg/ml NF45–NF90_long mixed with dsRNA with a maximum length of 410 bp. (F) Zoomed-in image of box in panel (E). (G) Zoomed-in images of box indicated in panel (E). (H) Quantitation of interparticle, length, and width measurements from negative stain micrographs.