HSV-1 US3 Hijacks Conserved Actin Regulatory Complexes to Drive F-Actin Remodeling

Figures

• 

  Figure 1. US3 is required for dynamic F-actin remodeling during HSV-1 infection. (a) Flow-cytometry gating strategy. N2A cells were gated on FSC-A vs. SSC-A (cells), FSC-A vs. FSC-H (singlets), viability dye (live cells), and Alexa Fluor 647-phalloidin (F-actin GeoMean). (b) Quantification of F-actin GeoMean across time (0, 6, 12, 24 hpi) from three independent experiments. Note that 0 hpi refers to the time immediately after the attachment step. Data are mean ± SEM. Two-way ANOVA with Tukey multiple comparisons test; * p < 0.05, ** p < 0.01, ns = not significant.

• 

  Figure 2. Confocal imaging shows US3-dependent loss of filamentous actin at late infection. N2A cells were infected with WT HSV-1, ΔUS3 HSV-1, or mock-infected, fixed at 0 or 24 hpi, and stained with Alexa Fluor 488–phalloidin (F-actin, green) and DAPI (nuclei, blue). Scale bars, 10 μm. (a) Mock-infected F-actin stress fiber morphology at 0 hpi. (b) Mock-infected F-actin stress fiber morphology at 24 hpi. (c) WT HSV-1-infected F-actin stress fiber morphology at 0 hpi. (d) WT HSV-1-infected F-actin stress fiber morphology at 24 hpi. (e) ΔUS3 HSV-1-infected F-actin stress fiber morphology at 0 hpi. (f) ΔUS3 HSV-1 infected F-actin stress fiber morphology at 24 hpi. (g) Quantification of total F-actin content (log10 mean fluorescence intensity, MFI) and (h) F-actin area fraction (log10%) from confocal images in (a–f), measured in Fiji/ImageJ. Cells were segmented and background-subtracted, and images were coded and analyzed blind to condition. Bars represent mean ± SEM of three independent biological replicates. Statistical significance was assessed by two-way ANOVA with Tukey’s multiple-comparisons test. ns, not significant; * p < 0.05; *** p < 0.001; **** p < 0.0001.

• 

  Figure 3. ClueGO network analysis reveals US3 interactors enriched in cortical actin-remodeling pathways. (a) Functional network of GO terms enriched among high-confidence US3-interacting host proteins (ClueGO analysis) from US3 precipitated mass spec analysis. Node size reflects enrichment significance; node color denotes functional modules. Cytoskeleton-related processes (green) form a dominant cluster centered on cortical cytoskeleton organization. (b) Actin-focused subnetwork highlighting terms directly linked to F-actin regulation, including cortical cytoskeleton organization, positive regulation of stress fiber assembly, Arp2/3 complex-mediated actin nucleation, and regulation of actin filament polymerization.

• 

  Figure 4. US3 interactors cluster within conserved CORUM protein complexes leading actin dynamics and transport. (a) Network representation of CORUM complexes containing at least one US3 interactor. Nodes represent protein complexes grouped by functional category (F-actin regulators, cytoskeleton, mitochondrial transport, nuclear transport, vesicle transport); edges connect complexes that share US3-bound subunits. (b) Heatmap of “coverage” values for each complex, defined as the fraction of its subunits detected as US3 interactors. High coverage of Arp2/3, kinesin-1, COPI, TIM/TOM, nuclear pore, and V-ATPase V1 complexes indicates that US3 interfaces with intact multiprotein machines rather than with isolated factors. (c) Reciprocal co-immunoprecipitation of US3 and β-catenin. β-Catenin was detected following US3 immunoprecipitation, and US3 was detected following β-catenin immunoprecipitation; IgG and mock-lysate controls showed no specific signal.

• 

  Figure 5. Machine-learning prioritization of US3 interactors using label-independent mass-spectrometry features. (a) Receiver-operating-characteristic and (b) precision–recall curves for logistic regression, random forest, and gradient-boosted trees under protein-group-aware cross-validation. Labels derive from curated actin-regulator annotations; features are independent of these labels. Logistic regression performed best (ROC-AUC 0.66; average precision 0.24 vs. a 0.05 baseline). Modest performance reflects the imbalanced task and indicates the analysis is suited to prioritization rather than definitive classification.

• 

  Figure 6. Model comparison under protein-group-aware cross-validation. Box-and-whisker plots summarizing performance across 10 random seeds for logistic regression (LR), random forest (RF), and gradient-boosted trees (GB). (a) Average precision (AP). (b) Area under the ROC curve (ROC-AUC). (c) Matthews correlation coefficient (MCC). (d) F1 score. (e) Balanced accuracy. Logistic regression significantly outperformed both tree-based models on average precision (Friedman χ2 = 29.6, p < 0.0001; Wilcoxon p < 0.0001 for LR vs. RF and LR vs. GB), whereas random forest and gradient boosting did not differ (p = 0.65). The dashed line indicates the class-prevalence baseline (AP = 0.05).

• 

  Figure 7. Pharmacological validation and inferred model of US3-mediated actin remodeling. (a) Schematic of the machine-learning workflow used to prioritize candidate F-actin regulators. Label-independent quantitative mass-spectrometry features were used for model training, whereas curated actin-regulator annotations were used only to define class labels. Prioritized candidates were subsequently selected for biological interpretation and experimental validation. (b) Flow-cytometric quantification of F-actin levels in N2A cells infected with WT HSV-1 or ΔUS3 HSV-1 and treated with the formin inhibitor SMIFH2, the Arp2/3 inhibitor CK-666, or both inhibitors. F-actin abundance was measured as the GeoMean MFI of Alexa Fluor 647–phalloidin at 0 and 24 hpi. Inhibition of either pathway partially attenuated the WT-associated reduction in F-actin, whereas combined inhibition prevented a statistically detectable reduction under the tested conditions. Data represent the mean ± SEM from three independent biological experiments. Statistical significance was assessed using two-way ANOVA followed by Šidák’s multiple-comparisons test. ns, not significant; * p < 0.05; **** p < 0.0001. (c) Proposed model…