Phosphorylation enables allosteric control of a viral condensate

Abstract

In many viruses, intrinsically disordered proteins (IDPs) drive the formation of replicative organelles essential for viral production. In species A rotaviruses, the disordered protein NSP5 forms condensates in cells via liquid-liquid phase separation (LLPS). Yet the sequence diversity of NSP5 raises the question of whether condensate formation is conserved across all strains and if distinct variants employ alternative mechanisms for nucleating phase separation. Using a machine learning approach, we demonstrate that NSP5 variants differ significantly in their propensity to phase-separate. We engineered a variant incorporating amino acid signatures from strains with low LLPS tendency, which failed to phase separate in vitro yet supported the formation of replicative condensates in recombinant viruses in cells. Low-tendency LLPS strains require phosphorylation of NSP5 to nucleate phase separation, whereas high-tendency strains do not, suggesting distinct nucleation mechanisms. Furthermore, hydrogen-deuterium exchange mass spectrometry revealed a phosphorylation-driven allosteric switch between binding sites on the high-propensity variant. These findings establish that phosphorylation plays a context-dependent role in the formation of replicative organelles across diverse rotaviruses.

Figure 1.. Computational identification and rational design of an NSP5 variant with reduced LLPS propensity

A. LLPS propensity analysis of 451 unique full-length RVA NSP5 sequences using DeePhase. The violin plot shows that most NSP5 variants exhibit DeePhase scores above 0.5. The cell-culture adapted RVA strains RF and SA11 highlighted. A zoomed-in inset displays the seven low-scoring variants designated S1_low-S7_low. Sequence similarity was assessed by computing pairwise Levenshtein distances, ranging from 0 (identical sequence) to 110 (maximally divergent). Among these, NSP5 sequences from S4_low - S7_low show the greatest similarity with strain SA11 and are collectively referred to as the SA11-like cluster (SC). B. In silico design of the SC_low NSP5 variant using DeePhase predictions and NSP5 conservation analyses. A consensus sequence (SC consensus) was generated from the SA11-like cluster, incorporating 21 degenerate positions, which were substituted with corresponding residues from the S4_low sequence. The seven remaining differences between the SC consensus and S4_low gave rise to 2^7 (128) sequence permutations, each evaluated by DeePhase. Of these, 78 variants scored below 0.5. A second consensus was derived from these low-scoring variants, resolving one remaining ambiguous position by adopting the S4_low amino acid. Asterisks mark serine phosphorylation sites reported for NSP5 (Sotelo et al., 2010), all of which are retained in the final SC_low sequence.

Figure 2.. Phylogenetic and biophysical features underlying NSP5 phase separation propensity

A. Phylogenetic analysis of 451 unique NSP5 sequences alongside 128 computationally designed permutation variants derived from the SA11-like cluster. Naturally occurring low-scoring variants (S1_low-S7_low; DeePhase < 0.5) are indicated at their respective tips. Cell-culture-adapted reference strains SA11 and RF are highlighted in dark red, and the SA11-like cluster (SC) consensus with its optimised SC_low variant are marked in dark blue. Among the 128 permutation variants (SC128), tips are coloured orange for sequences with DeePhase scores > 0.5, while light blue denotes scores < 0.5. B. NSP2–NSP5 condensate formation in model RVA strains RF and SA11, observed both in cellulo and in vitro. Upper panel: MA104–NSP5–eGFP cells infected with either RF or SA11 RVAs develop cytoplasmic condensates. Lower panel: recombinantly purified and Atto488-labelled NSP5 and NSP2 from RF and SA11 strains form condensates in vitro. Scale bars, 10 μm. C. Intrinsic disorder predictions for NSP5 from RVA strains SA11 (blue) and RF (red), as determined by AUPred2A (Mészáros et al., 2018). Scores above 0.5 (dashed line) indicate disordered regions. The N-terminal (residues 1–33) and C-terminal (residues 181–198) segments (shaded) are required for NSP2 binding and NSP5 homo-oligomerisation, respectively (Arnoldi et al., 2007; Eichwald et al., 2004b; Fabbretti et al., 1999). Phosphorylated serine residues reported by Sotelo et al. (Sotelo et al., 2010) are marked.

Figure 3.. SC_low-RV supports replication, particle production, and viroplasm formation in cell culture.

A. Representative plaque morphologies produced by SC_low-RV and and wild-type SA11-RV in MA104 cells. Box plots show plaque areas measured for 50 individual plaques. The horizontal bar indicates the mean ± standard error of the mean (SEM). **p = 0.0033, unpaired t-test. B. Replication kinetics of SC_low-RV and wild-type SA11-RV in MA104 cells infected at a multiplicity of infection (MOI) of 1. Viral titres were determined at 4, 8, 12, 16, 24, and 48 hours post infection (h.p.i.) using TCID₅₀ assays. Each data point represents the mean ± standard deviation (SD) from three independent experiments. C. Transmission electron micrographs of MA104 cells infected with SA11-RV (left) and SC_low-RV (right), fixed at 8 h.p.i. Multiple viral particles are observed within viroplasms (V). Scale bar, 200 nm. D. SC_low-RV forms viroplasms that accumulate viral transcripts. Atto647N-labelled smFISH probes specific for RV transcripts were used to visualise viral RNAs localising to EGFP-tagged viroplasms, as described in Methods.

Expanded view Figure 3 -. Design, rescue, and genetic stability of the SC_low rotavirus

A. RNA sequence design for gene segment 11 SC_low ORF using the reference RNA sequence of NSP5 SA11. The coding sequence of NSP5 from SA11 was used as a reference, and synonymous codons were selected to minimise nucleotide divergence while preserving amino acid identity wherever possible. In cases of amino acid substitution (e.g., Proline to Leucine at position 99), the target codon required the smallest possible number of nucleotide changes. A total of 35 nucleotide changes were introduced in the SC_low sequence, while untranslated regions (UTRs) were left unchanged. Positions of nucleotide substitutions are indicated. B. Schematic overview of the reverse genetics strategy used to generate SC_low-RV. Ten pT7 plasmids encoding the SA11 gene segments (GS1–GS10) and one recombinant plasmid encoding SC_low, were co-transfected into BHK-T7 cells along with expression plasmids for NSP2 and NSP5 (pcDNA3-NSP2 and pcDNA3-NSP5). At 48 h post-transfection, either WT-MA104 or NSP5-expressing MA104 (NSP5-MA104) cells were overlaid. Recombinant virus was harvested after three freeze-thaw cycles upon observation of full cytopathic effect (CPE), as described in Methods (Papa et al., 2019). C. Conservation analysis of NSP2 across rotavirus A strains. Full-length NSP2 sequences were aligned, and residue conservation was mapped onto the NSP2 octamer crystal structure (PDB: 1L9V; SA11 strain). Key NSP5-binding regions are conserved across strains (Jiang et al., 2006). D. Multiple sequence alignment of the NSP6 open reading frame encoded by gene segment 11. The NSP6 sequences of SA11, RF, and SC_low sequence were compared, confirming preservation of NSP6 in the engineered SC_low-RV genome.

Figure 4.. Delayed viroplasm formation and impaired NSP5 hyperphosphorylation in cells infected with SC_low-RV.

A. Wide-field fluorescence microscopy of MA104-NSP5-eGFP cells infected with either SA11-RV (WT-RV) or SC_low-RV at MOI of 20. NSP5-eGFP–labelled viroplasms (green) and nuclei (DAPI, blue) are shown at 4, 6, and 8 hours post-infection (h.p.i.). Insets show enlarged regions of interest. Scale bar, 10 μm. B,C. Quantification of the number (B) and size (C) of viroplasms formed in cells infected by SA11-RV (burgundy) or SC_low-RV (dark blue). Data represent mean values from three independent experiments (N > 200 cells per condition). Error bars indicate standard error of the mean (SEM). Two-way ANOVA was used to compare SA11-RV and SC^low-RV at each time point: viroplasm number at 4 h.p.i. (*p = 0.0004), 6 h.p.i. (*p = 0.0012), 8 h.p.i. (*p = 0.0071); viroplasm sizes at 4 h.p.i. (ns = 0.9244), 6 h.p.i. (*p = 0.0278), 8 h.p.i. (*p = 0.0211). D. Immunoblot analysis of NSP5 extracted from MA104 cells infected with SA11-RV or SC_low-RV at 4, 6 and 8 h.p.i. (MOI = 20). Note the absence of the strongly hyperphosphorylated 35 kDa NSP5 isoform in SSC_low-RV–infected cells at later time points.

Expanded view Figure 4.. Validation of anti-NSP5 antibodies detecting both WT-NSP5 and SC_low, as well as their phosphomimetics.

Quantitative Western Blot analysis of recombinantly produced NSP5 (SA11 and SC_low) and their corresponding hyperphosphorylation mimetics. Four distinct proteins (NSP5, SC_low, NSP5 HP, and SC_low HP) were recombinantly produced and purified, as described in Methods. Equal amounts of each protein (400 ng per lane) were loaded in technical duplicates, resolved by SDS–PAGE, and analysed by Western blot using anti-NSP5 antibodies (Geiger et al., 2021).

Figure 5 -. Hyperphosphorylation of SClow restores its ability to phase separate in vitro

A. In vitro phase separation assays of NSP5 variants in the presence of NSP2. Recombinant Atto-488-labelled NSP2 (25 μM) was mixed with equimolar amounts (25 μM) of unlabelled NSP5 variants, including wild-type SA11 (NSP5 SA11), its hyperphosphomimetic (NSP5 SA11 HP), bovine strain RF (NSP5 RF), NSP5 RF HP, SC_low, SC_low HP, and the intrinsically disordered region-only construct of RF NSP5 (NSP5 RF IDR and IDR HP), as described in Methods. Fluorescence microscopy images show representative fields of view. Scale bar, 10 μm. B. Quantification of average condensate size for each NSP5 variant mixed with NSP2. Condensates were imaged within a 500 μm × 800 μm region of interest across three biological replicates (N = 3). Data were analysed by one-way ANOVA with Bonferroni and Sidak multiple comparisons test (α = 0.05). ****P < 0.0001.

Figure 6 -. Phase behaviour of NSP5 variants in the presence of NSP2 reveals phosphorylation-dependent coacervation.

PhaseScan-derived phase diagrams showing coacervation between NSP2 and different NSP5 variants. Combinatorial microfluidics was used to generate high-resolution phase diagrams for Atto-488-labelled NSP2 and Alexa-647-labelled NSP5 variants across a concentration matrix (2–25 μM for each protein), as detailed in Methods. LLPS probability heatmaps were reconstructed from the following number of droplets: NSP5 SA11 (n = 28,280), NSP5 SA11 HP (n = 87,053), SClow (n = 91,057), SClow HP (n = 101,659), NSP5 RF (n = 18,813), NSP5 RF HP (n = 17,449), NSP5 IDR (n = 36,511), and NSP5 IDR HP (n = 66,455). LLPS probability is colour-coded from low (blue) to high (red), with black dotted lines indicating inferred phase boundaries.

Figure 7 -. Hyperphosphorylation enables oligomerisation of SC_low

A. AlphaFold2 structural models for NSP5 SA11, NSP5 SA11 HP, SC_low and SC_low HP. Models are coloured by per-residue confidence scores (pLDDT), with blue (pLDDT > 90) indicating high confidence and orange-red (pLDDT < 50) marking regions predicted to be intrinsically disordered. B. Oligomeric size distributions of NSP5 variants determined by mass photometry. Recombinant NSP5 SA11, NSP5 SA11 HP, SC_low, SC_low HP, NSP5 RF, NSP5 RF HP, or NSP5 RF IDR were analysed at 100 nM each, as described in Methods. Representative molecular weight histograms are shown, with Gaussian fits highlighting major oligomeric species. Approximate molecular weights (in kDa) and corresponding standard deviations (σ) are indicated.

Expanded View Figure 7 –. The C-terminal region (CTR) of NSP5 alone forms large aggregates

A. Hydrodynamic diameters (D, in nm) of 1 μM NSP5-RF CTR peptide measured in PBS (pH 7.4) using dynamic light scattering (DLS). B. Summary of the oligomeric states and liquid–liquid phase separation (LLPS) capacity of various NSP5 constructs used in this study. Schematic representations of the protein constructs are shown; phosphomimetic variants are indicated with a red asterisk.

Figure 8 –. The SC_low S67A-RV mutant fails to form viroplasms in infected cells despite active viral transcription.

A. Schematic representation of the S67A mutation introduced into gene segment 11 of SC_low, with corresponding Sanger sequencing chromatograms confirming the mutation. B. Representative plaque assays in MA104-NSP5 cells infected with SC_low-RV or SC_low S67A-RV. C. Wide-field fluorescence microscopy of NSP5-eGFP-tagged viroplasms (green) in cells infected with SC_low-RV, SA11 S67A-RV or SC_low S67A-RV at an MOI of 10. Images were acquired at 24 h.p.i. Scale bars: overview = 30 μm; zoomed regions = 10 μm. D. RNA FISH detection of viral transcripts (gene segment 1) in MA104 WT and MA104-NSP5 cells infected with SC_low S67A-RV, using Atto647N-labelled probes. Viral RNA accumulates in both cell types, indicating transcriptional activity. Scale bars: overview = 30 μm; zoomed regions = 10 μm.

Figure 9.. Both the NSP5 CTR and intrinsically disordered region (IDR) are critical for mediating LLPS

A. Experimental setup to assess the individual contributions of the NSP5 IDR and CTR to phase separation. Wild-type NSP5 or its phosphomimetic variant (NSP5 HP) was mixed with increasing ratios of either the isolated IDR or CTR, followed by addition of NSP2 to induce LLPS. Schematics illustrate NSP5 domains: IDR (black), CTR (green), and phosphomimetic sites (red ‘P’). Phase-separated condensates were imaged by wide-field microscopy following mixing of Atto-488-labelled NSP2 (25 μM) with total NSP5 (25 μM) at the indicated full-length:CTR or full-length:IDR ratios. Scale bar, 10 μm. B. Quantification of condensate area (μm²) formed by NSP2 with wild-type or HP NSP5 in the presence or absence of IDR or CTR fragments. Nine regions of interest (50 μm × 80 μm) were analysed per condition, as described in Methods. Statistical analysis was performed using one-way ANOVA with Dunnett’s multiple comparisons test (α = 0.05): **P = 0.0082, ***P = 0.0003, ****P < 0.0001.

Expanded View Figure 9 –. Both phosphorylated and non-phosphorylated NSP5 IDRs undergo phase separation with poly-arginine

A. Phase separation of NSP5 IDR and its phosphomimetic variant (IDR HP) in the presence of poly-L-arginine. DyLight-488-labelled IDR or IDR HP (25 μM) was mixed with 5 μM poly-L-arginine (schematically illustrated), and resulting condensates were imaged by wide-field microscopy. Phosphomimetic sites are indicated in red (‘P’). Scale bar, 10 μm. B. Quantification of condensate area (μm²) formed by NSP5 IDR or IDR HP with poly-L-arginine. Nine regions of interest (50 μm × 80 μm each) were analysed as described in Methods. For comparison, condensate sizes formed by NSP2 with NSP5 IDR or IDR HP (as shown in Figure 9) are also included.

Figure 10 –. Distinct NSP5–NSP2 interaction interfaces revealed by HDX-MS upon hyperphosphorylation of NSP5

A–C. Hydrogen–deuterium exchange mass spectrometry (HDX-MS) analysis of wild-type (WT) and hyperphosphorylation-mimetic (HP) NSP5 in the absence and presence of NSP2. Cumulative Woods plots show peptides of NSP5 HP (A, B) and WT NSP5 (A, C) that exhibit significant changes in deuterium uptake upon incubation with deuterium alone or with NSP2, analysed using Deuteros with hybrid significance testing (p < 0.02; (Lau et al., 2019)). Regions showing increased protection (blue) or deprotection (red) are indicated. The C-terminal region (CTR) is marked by a purple box. D. AlphaFold2 model of monomeric NSP5 showing mapped HDX-MS data: protected regions (blue), deprotected regions (red), peptides with no significant change (light grey), and regions with no coverage (dark grey). Ser-to-Asp phosphomimetic mutations are indicated.

Figure 11.. Phosphorylation-induced transition from non-saturating to saturating interactions in NSP5-NSP2 condensates

A. Homotypic interactions of NSP5. NSP5 molecules with low LLPS scores (black IDRs and green CTRs) are unable to self-associate effectively. In contrast, NSP5 variants with high LLPS scores or those containing phosphorylated IDRs can form condensates, with LLPS favoured by interactions between the IDR and CTR. Phosphorylation of the IDR enhances these IDR–CTR interactions, facilitating homo-oligomerisation and condensate formation. B. Heterotypic interactions between NSP5 and NSP2. At early stages of infection, positively charged NSP2 octamers (light blue doughnuts) bind NSP5 via stoichiometric interactions with its CTR. In NSP5 variants with high LLPS propensity, the flexible, unstructured IDRs form a non-stoichiometric interaction network that drives phase separation. In contrast, low-LLPS NSP5 variants can still bind NSP2 via the CTR but fail to phase separate due to insufficient IDR–IDR interactions. During late infection, phosphorylation of NSP5 increases the negative charge of the IDR, promoting interactions with positively charged regions of NSP2. These phosphorylation-dependent interactions enable even low-LLPS NSP5 variants to engage in saturable binding and phase separation, facilitating condensate assembly.