Proteomic and Machine Learning Signatures of Rabies Virus Infection Reveal Stage-Specific Biomarkers

Proteomic and Machine Learning Signatures of Rabies Virus Infection Reveal Stage-Specific Biomarkers | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Proteomic and Machine Learning Signatures of Rabies Virus Infection Reveal Stage-Specific Biomarkers Iana Suly Santos Katz, Elaine Raniero Fernandes, Fernanda Guedes, and 4 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-7652543/v1 This work is licensed under a CC BY 4.0 License Status: Published Journal Publication published 24 Dec, 2025 Read the published version in Journal of NeuroVirology → Version 1 posted 7 You are reading this latest preprint version Abstract Rabies virus (RABV) is a highly neurotropic pathogen with near-uniform lethality after symptom onset, yet the molecular mechanisms underlying disease progression in the central nervous system (CNS) remain poorly defined. To address this gap, we performed time-resolved, label-free quantitative proteomics in RABV-infected mouse brains at defined clinical phases of RABV infection (asymptomatic, progressive, terminal). Differential protein expression was analyzed by clustering, enrichment, and protein–protein interaction networks. Machine learning models classified infection stages, and top biomarkers were validated by Western blot. Principal component and clustering analyses separated infection phases robustly, while GO/KEGG and PPI analyses revealed a progression from cytoskeletal/trafficking remodeling (early) to innate immune activation (intermediate) and proteostasis collapse/neurodegeneration-linked pathways (late). A Support Vector Machines classifier discriminated phases with high performance (F1 = 0.88; AUC = 0.79) and SHAP interpretation highlighted LAMP2, IL18 and SNCA among the top phase-specific predictors, and confirmed experimentally. This integrative proteomics–machine learning approach maps dynamic molecular transitions during RABV infection and nominates diagnostic biomarkers relevant to neurovirology. These findings provide mechanistic insights into viral neuropathogenesis and highlight parallels with neurodegeneration. rabies virus proteomics machine learning biomarker discovery neurovirology neurodegeneration Pathogenesis Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Introduction Rabies virus (RABV) remains one of the deadliest pathogens in human history, causing an estimated 59,000 deaths annually, with the highest burden in Africa and Asia (Hampson et al., 2015 ; WHO, 2023 ). Unlike many viral infections that can be managed with supportive care or antiviral therapies, rabies is almost invariably fatal once clinical symptoms appear. This outcome reflects the virus’s pronounced neurotropism and its capacity to evade host immune responses while advancing toward widespread central nervous system (CNS) involvement (Jackson, 2016 ; Ugolini, 2011 ; Willoughby et al., 2005 ). Classical neuropathological hallmarks of rabies include Negri bodies, neuronal degeneration, and perivascular inflammatory infiltrates (Zhao et al., 2012 ; Chienwichai et al., 2025 ). These observations, while historically significant, provide static snapshots of a disease defined by dynamic molecular interactions. Rabies neuropathogenesis involves both viral subversion of host pathways and maladaptive immune responses, yet the temporal sequence of these molecular events remains poorly defined (Zhao et al., 2012 ; Chienwichai et al., 2025 ). Recent advances in high-throughput proteomics allow direct interrogation of host protein networks, capturing dynamic alterations in signaling, trafficking, and neuroinflammation (Aebersold & Mann 2016 ; Chienwichai et al., 2025 ). Proteomic studies have begun to elucidate viral modulation of neuronal and immune pathways in CNS infections, including herpesviruses, flaviviruses, and coronaviruses (Ahmed et al., 2022 ; Bojkova et al., 2020 ; Li et al., 2020 ). However, proteomic analyses in rabies remain limited, often restricted to single time points or narrow protein subsets (Beatman et al., 2017 ; Mehta et al., 2015 ; Venugopal et al., 2013 ; Yan et al., 2019 ;). Notably, studies by Venugopal et al. 2013 and Mehta et al. 2015 highlighted synaptic and homeostatic changes in murine brain tissue, while Yan et al. 2019 reported proteomic alterations in suckling mouse models. These efforts, though informative, failed to resolve the full temporal complexity of infection. Filling this knowledge gap is critical for the field of neurovirology, as it may uncover mechanisms of viral persistence, neuronal dysfunction, and immune-driven injury within the central nervous system (CNS). Complementary computational approaches, particularly machine learning (ML), enhance biomarker discovery by identifying patterns within high-dimensional omics data (Budhraja et al., 2023 ; Dooley et al., 2025 ; Huang et al., 2025 ). ML has been applied to biomarker discovery in Alzheimer’s disease (Budhraja et al., 2023 ) and viral infections such as SARS-CoV-2 (Bojkova et al., 2020 ), but remains underexplored in rabies. Coupling ML with time-resolved proteomics may enable the identification of stage-specific molecular signatures with diagnostic and therapeutic relevance, addressing the pressing need for reliable biomarkers of disease progression. Here, we hypothesized that RABV infection progresses through distinct molecular phases, each characterized by unique proteomic signatures reflecting viral replication, immune activation, and neurodegeneration. To test this, we applied longitudinal label-free proteomics to brains of RABV-infected mice, integrated ML-driven classification, and validated candidate biomarkers. This integrative framework refines our understanding of rabies neuropathogenesis and supports biomarker discovery within neurovirology. By linking viral dynamics, host responses, and neuronal injury, our findings highlight opportunities for biomarker-guided diagnosis and therapeutic development, while contributing to broader efforts at the interface of viral pathogenesis, neuroinflammation, and neurodegeneration. Materials and Methods Animal Model All animal experiments were conducted in accordance with the ARRIVE guidelines and approved by the Institutional Animal Care and Use Committee of Instituto Pasteur (CEUA-IP Protocol no. 04/2013). Six-week-old-Webster mice (n = 6 per group) were housed under pathogen-free conditions with ad libitum access to food and water. Mice were randomly assigned to experimental groups, and outcome assessors were blinded to group allocation during clinical scoring and sample processing. Humane endpoints were established to minimize suffering, and euthanasia was performed using CO₂ asphyxiation in compliance with institutional welfare policies. Viral Infection and Experimental Design Animals were intradermally inoculated in the hind footpad with 1 × 10⁴ focus-forming units (FFU) of the V-2 strain of RABV (Hierholzer and Killington, 1996 ). Control groups received mock inoculations with sterile PBS. Following infection, animals were monitored daily for changes in clinical symptoms, which were scored using a standardized scale: 0 = asymptomatic, 1 = disordered movement, 2 = ruffled fur and hunched posture, 3 = trembling and shaking, 4 = paralysis, and 5 = death. Based on symptomatology, animals were euthanized and brains collected at three defined infection phases: Phase 1 (early/innate): 48 hours post-infection (hpi), before clinical symptoms (score 0) Phase 2 (progressive): 5–7 days post-infection (dpi), onset of neurological symptoms (score 1–3) Phase 3 (late/severe): 8–15 dpi, advanced disease with paralysis (score 4) Whole brains were immediately frozen in liquid nitrogen and stored at − 80°C until processing. Sample Collection and Protein Extraction Frozen brain tissue samples were homogenized in ice-cold 8M urea lysis buffer for mass spectrometric analyses. Lysates were clarified by centrifugation at 12,000 × g for 15 min at 4°C, and protein concentrations were measured using the BCA Protein Assay kit (Pierce - Thermo Fisher Scientific Inc.). Aliquots containing 100 µg of protein per condition were reduced with 10 mM dithiothreitol (DTT) for 30 min at room temperature and alkylated with 50 mM iodoacetamide (IAA) in the dark for 1 h at room temperature. The proteins were then digested overnight at 37 o C using mass spectrometry-grade trypsin (Trypsin Gold, Promega) at an enzyme-to-substrate ratio of 1:50 (w/w). Trypsin activity was quenched by acidification with formic acid to a final concentration of 5% (v/v). Peptide samples were desalted using in-house prepared StageTips containing SDB-XC membranes (Empore Styrene Divinylbenzene extraction disk cartridge, 3M). LC–MS/MS Acquisition The desalted peptides were resuspended in 0.1% (v/v) formic acid in water (solvent A) and separated using an EASY-nLC II nanoflow HPLC system (Thermo Fisher Scientific, Bremen, Germany) coupled online to an LTQ-Orbitrap Velos mass spectrometer (Thermo Fisher Scientific). Peptides were loaded onto a pre-column (ID 100 µm × OD 360 µm x 5 cm, packed in-house with C18 10 µm beads; Acqua, Phenomenex) and separated on an analytical column (ID 75 µm × OD 360 µm x 15 cm, packed in-house with C18 5 µm beads; Jupiter, Phenomenex). Separation was achieved using a linear gradient of 5–30% acetonitrile in 0.1% formic acid (Solvent B) over 54 min at a constant flow rate of 200 nL/min. The mass spectrometer was operated in data-dependent acquisition mode. Full-scan MS spectra (m/z 300–1800) were acquired in the Orbitrap analyzer at a resolution of 30,000. The top 10 most intense precursor ions from each scan were selected for fragmentation via collision-induced dissociation (CID) in linear ion trap, with a dynamic exclusion of 70 s. The nanospray voltage was set to 2.3 kV, and the capillary source temperature was maintained at 250°C. Protein Identification and Quantification Tandem mass spectra were processed using PEAKS Studio software (version 7; Bioinformatics Solutions, Ontario, Canada). Database searches were performed against the Rabies lyssavirus (RABV) protein database (NCBI Taxonomy ID 11292). Search parameters included cysteine carbamidomethylation as a fixed modification and methionine oxidation as a variable modification. Trypsin was specified as the protease, allowing for up to two missed cleavages. The mass error tolerances were set to 10 ppm for precursor ions and 0.5 Da for fragment ions. Label-free quantification was performed using the top-three precursor ion intensity method within the Progenesis QI software (Nonlinear Dynamics). Data Normalization and Filtering Protein intensity values were log₂-transformed and normalized using quantile normalization to correct for technical variation. Missing values were handled according to their mechanism: For proteins absent primarily in low-intensity samples, left-censored (MNAR-aware) imputation was applied using a downshifted Gaussian distribution. For sporadic missing values (< 5% of entries), k-nearest neighbors (k = 3) imputation was used. Statistical Analyses Differential abundance testing was performed using the limma-voom pipeline implemented in R. P-values were adjusted for multiple testing using the Benjamini–Hochberg false discovery rate (FDR) procedure, and proteins with FDR ≤ 0.05 were considered statistically significant. Proteins were retained for downstream analyses if they met all of the following criteria: (i) identified with ≥ 2 unique peptides, sequence coverage ≥ 20%, and score ≥ 500; (ii) quantified in ≥ 50% of samples within at least one experimental group; and (iii) passed limma-voom testing at FDR ≤ 0.05. This stepwise filtering reduced the dataset from 2,950 identified proteins to 27 high-confidence candidates (Supplementary Table S1 ) Exploratory and Functional Analyses Principal component analysis (PCA) clustering was performed using the final 27 prioritized proteins. Gene Ontology (GO) biological process and KEGG pathway enrichment analyses were performed using the clusterProfiler package with a hypergeometric test. P-values were corrected for multiple testing using the Benjamini–Hochberg false discovery rate (FDR) procedure, and pathways with q ≤ 0.05 were considered significantly enriched. Protein–protein interaction networks (PPI) were constructed with STRING v11.5, visualized in Cytoscape v3.9.1, and analyzed for topological properties. Machine Learning Classification and Interpretation Machine learning analyses were performed in Python (v3.10) using scikit-learn (v1.2), complemented by XGBoost (v1.7) and SHAP (v0.41) for model interpretation. Model training and deployment were managed through the Azure Machine Learning platform, with structured data handling via Azure SQL. Final dashboards and visualization for exploratory analyses were generated using QlikSense. Multiple algorithms were evaluated for classifying infection phases, including Support Vector Machine with radial basis function kernel (SVM-RBF), Random Forest, XGBoost, Multi-Layer Perceptron (MLP), Logistic Regression, and a Stacking Ensemble. To avoid information leakage, we adopted a nested cross-validation scheme: (i) the outer loop used leave-one-mouse-out cross-validation for unbiased performance estimation; (ii) the inner loop conducted hyperparameter optimization by grid search. Data were partitioned using stratified sampling (70% training, 30% testing) to preserve class balance. Within each training fold, Recursive Feature Elimination with Cross-Validation (RFECV) was employed to identify minimal biomarker sets, ensuring feature selection occurred independently of test data. Correlated predictors were removed (Spearman’s ρ ≥ 0.93) to minimize redundancy. The hyperparameter ranges evaluated for each algorithm are summarized in Table 1 . Table 1 Hyperparameter ranges for machine learning models Model Hyperparameters (grid search ranges) SVM-RBF C = [0.1, 1, 10, 100]; γ = [1e-3, 1e-4, 1e-5]; kernel = rbf Random Forest n_estimators = [100, 300, 500]; max_depth = [5, 10, 20] XGBoost learning_rate = [0.01, 0.1]; max_depth = [3, 6, 10]; n_estimators = [200, 500] MLP hidden_layers = [(50,), (100,), (100, 50)]; activation = [relu, tanh]; learning_rate = [1e-3, 1e-4] Logistic Regression penalty = [l2]; C = [0.1, 1, 10]; solver = [lbfgs] Stacking Ensemble Base: SVM, RF, XGBoost; Meta-learner: Logistic Regression Performance evaluation included balanced accuracy, precision, recall, F1-score, and AUC, each with 95% bootstrap confidence intervals. Calibration curves and learning curves were generated to assess model reliability. Class imbalance was addressed by stratified CV, class-weight adjustments (for SVM and Logistic Regression), and explicit reporting of balanced accuracy. Among the models tested, SVM-RBF achieved the best performance, with model interpretability assessed using SHAP (SHapley Additive exPlanations) values. SHAP values were computed with the KernelSHAP method using 100 background samples per fold, and feature importances were averaged across outer folds to ensure unbiased interpretation. Western Blot Validation Candidate biomarkers (LAMP2, IL18, SNCA) were selected based on recurrence across ML folds, SHAP importance, and pathway relevance. Validation was performed by Western blotting on independent samples (n = 6 per time point). Proteins were separated by SDS-PAGE, transferred to PVDF membranes, and probed with anti-LAMP2 (1:400, Abcam, ab13524), anti-IL18 (1:800, Abcam, ab71495), anti-SNCA (1:400, Abcam, ab138501), and anti-β-actin (1:5000, Sigma-Aldrich). Detection used HRP-conjugated secondary antibodies (1:5000, Santa Cruz, sc-2030) and chemiluminescence. Densitometry was performed in ImageJ (IMAGE QUANTTM400, Amersham Biosciences), normalized to β-actin, and correlated with proteomic data (Pearson’s r). An overview of the experimental workflow, including sample preparation, proteomic analysis, data processing, and the machine learning pipeline for phase classification and biomarker prioritization, is presented in Fig. 1 . Results Proteomic Landscape of RABV Infection Proteomic profiling of CNS tissues across three time points post-RABV infection initially identified 2,950. After applying sequential filters for detection consistency, statistical significance, and biological relevance, a subset of 27 proteins was retained for downstream analyses. Reproducibility analysis of this filtered set showed a Pearson correlation coefficient of 0.9397, indicating high consistency between replicates. PCA of the proteomic dataset demonstrated a clear temporal separation of samples, delineating three distinct stages of disease progression. The first two principal components (PC1 and PC2) accounted for 79.88% of the total variance, reflecting substantial shifts in protein expression profiles over the course of infection ( Fig. 2A ). PC1, explaining 57.42% of the variance, discriminated infected samples from controls and separated phase 3 from phases 1 and 2. PC2, accounting for 22.46% of the variance, differentiated early from intermediate infection stages ( Fig. 2B ). K-means clustering analysis identified k = 3 as the optimal solution, with an average silhouette score of 0.834, indicating that three clusters most robustly represent the intrinsic structure of the data, consistent with the separation observed in the PCA ( Fig. 2C). Both PCA and unsupervised clustering demonstrated clear separation of samples by clinical phase. Notably, K-means clustering independently validated this three-phase structure, reinforcing the concordance between proteomic signatures and clinical phenotypes. Differential Expression and Key Biomarkers Phase assignment was validated through gap statistical analysis and visual inspection of expression heatmaps. Hierarchical clustering analysis demonstrated clear separation between the control and experimental groups (Groups 1–3) (Fig. 3 A). Temporal expression profiling revealed smooth transitions between phases, with approximately 15% overlap in significantly expressed proteins between adjacent phases. Group 1 showed the most pronounced changes, with 66% of its associated proteins exhibiting large effects (d > 1.5), including the upregulation of YWHAQ, YWHAG, and LAMP2. Group 2 presented a more heterogeneous profile, with 34% of proteins showing large effects, particularly involving inflammatory and metabolic markers such as IL18, C3, RELA, and ALDH2. Group 3 displayed 55% of proteins with large effects, including mitochondrial-associated proteins (e.g., MRPL12, MRPS36, HSPE1, ALDH2), proteasome-related proteins (e.g., PSMD2, CTSB, CTSD), and several ribosomal proteins (RPS2, RPS15, RPS31) (Fig. 3 B). Control samples predominantly displayed lower expression levels for these proteins, contrasting with the coordinated upregulation observed in the experimental groups. Functional Enrichment of Differentially Expressed Proteins Across Infection Phases As shown in Fig. 4 A, early-phase DEPs were significantly enriched in processes related to organelle organization, protein localization, intracellular transport, cytoskeletal remodeling, and kinase-driven signaling. Strongly enriched terms included responses to endogenous and external stimuli, epithelial cell adhesion, and MAPK cascade activation. These patterns indicate early reprogramming of host cell organization and signaling to support viral entry and replication. Consistent with this, cytoplasmic components, cell junctions, and microtubule-associated complexes were prominently represented, highlighting viral exploitation of trafficking and adhesion machinery. KEGG pathway analysis revealed significant enrichment in cell cycle regulation, PI3K-Akt signaling, and viral infection–related pathways such as hepatitis B/C and EGFR tyrosine kinase inhibitor resistance. The Phase 1 PPI network ( Fig. 4 D) showed densely interconnected clusters, with hub proteins from the 14-3-3 family (Ywhaz, Ywhab, Ywhae), Csnk1e, and Haus complex members linking cytoskeletal regulation with kinase signaling. Together, these findings demonstrate extensive host cell reprogramming during early infection to facilitate viral replication. In the intermediate phase, Functional enrichment shifted toward immune and inflammatory regulation (Fig. 4 B), including activation of immune defense, cytokine production, inflammasome and NF-κB complex activity, and complement system involvement. Significant KEGG pathways included complement and coagulation cascades, NOD-like receptor, Toll-like receptor, NF-κB, and cytokine–cytokine receptor interactions, as well as viral pathways such as influenza A. The Phase 2 PPI network ( Fig. 4 E) displayed two main clusters: one centered on NF-κB subunits (Nfkb1, Rela) and proinflammatory mediators (Il1b, Il18), and another composed of metabolic enzymes (Aldh2, Acs1, Acox1). This bipartite structure reflects simultaneous activation of innate immunity and metabolic reprogramming. In the late phase, enrichment analysis indicated dominant processes involving protein degradation and cell death (Fig. 4 C), including autophagy, macroautophagy, proteasome activity, and regulation of apoptosis. KEGG pathways highlighted autophagy, apoptosis, ribosome function, and neurodegeneration-related pathways such as Alzheimer’s disease, suggesting late-stage host cell damage from viral replication. The Phase 3 PPI network (Fig. 4 F) revealed interconnected clusters of autophagy-related proteins (Atg3, Atg5, Atg7, Atg12, Lamp2), proteasome subunits (Psmd6, Psmd7, Psmd12), and ribosomal proteins (Rps2, Rps15). Hubs such as Atg5 and Psmd6 bridged autophagy and proteasome clusters, indicating crosstalk between degradation pathways. Together, these findings illustrate a sequential response: early reorganization of cellular architecture and signaling (Phase 1), escalation of immune and inflammatory defenses (Phase 2), and activation of autophagy, proteolysis, and apoptosis leading to cell death (Phase 3). Hub protein networks corroborated this transition, emphasizing the dynamic shift from host cell remodeling to immune activation and ultimately to degradation pathways. Machine learning-based phase classification and biomarker importance To classify disease phase, we trained and compared multiple classifiers. ROC curves (Fig. 5 A) demonstrated that the SVM-RBF model achieved the highest discriminative ability, followed closely by the Stacking Ensemble. Although the Stacking Ensemble (Precision = 0.87, Recall = 0.80, F1-score = 0.87) showed competitive performance, SVM-RBF was selected as the primary classifier. This choice was justified by its superior precision (0.89), recall (0.83), and F1-score (0.88), as well as its consistent performance across folds and its ability to capture complex non-linear boundaries in high-dimensional proteomic data. Random Forest achieved intermediate performance (F1-score = 0.69), whereas XGBoost (F1-score = 0.68) and MLP (F1-score = 0.64) were comparatively less effective. The ROC curves of all models were shifted away from the diagonal, confirming performance above random classification. These results established a robust foundation for subsequent SHAP analysis, which quantified the most predictive protein features across infection phases, linking computational predictions to molecular mechanisms. The Fig. 5 B presents the top 10 proteins influencing SVM-RBF predictions across all phases. Phase-specific contributions included LAMP2, YWHAG, DPYSL2, YWHAQ, ACTA 1 (Phase 1), IL18, C3, RELA (Phase 2), and SNCA, CAPN1 (Phase 3). Feature selection prioritized the top 5% of SHAP values. Further granularity is shown in Fig. 5 C, where phase-specific SHAP rankings highlight proteins exerting distinct effects depending on the infection stage. For example, LAMP2, ACTA1, IL18, YWHAQ, C3, CAPN1, and RELA displayed the strongest influence on classification outcomes. This detailed view strengthens biological interpretability and supports their candidacy as stage-specific biomarkers. Overall, the SVM-RBF model demonstrated the strongest ability to capture meaningful proteomic signatures, while SHAP analysis provided transparent biological insights into the relative contributions of protein features. These findings highlight candidate biomarkers for further validation and underscore the value of integrating machine learning with molecular analysis. Expanding sample size and validating results in independent cohorts will likely improve robustness and translational applicability. Validation of Key Biomarkers Confirms Proteomic Findings Western blot analysis of independent tissue samples validated temporal expression patterns for top biomarkers identified through SHAP analysis. LAMP2 protein levels showed significant upregulation during Phase 1 (3.2-fold vs. phase 2, p < 0.01 and 6.2-fold vs. phase 3, p < 0.01), with strong correlation to proteomic data (r = 0.89, p < 0.001). IL18 exhibited characteristic Phase 2 expression pattern (3.3-fold vs. phase 1, p < 0.001, 5.9-fold vs. phase 3, p < 0.001, r = 0.91 with proteomics). SNCA demonstrated robust Phase 3 upregulation (2.7-fold vs. phase 1, P < 0.001, 4.7-fold vs. phase 2, p < 0.001 r = 0.87 with proteomics). Discussion A major strength of this study is its time-resolved design, which captures molecular signatures across clinically defined stages of rabies rather than relying on isolated cross-sectional snapshots. To our knowledge, this represents the first proteomic analysis of rabies infection with explicit temporal resolution, providing mechanistic insight into stage-specific host responses. Within this framework, we identified 27 high-confidence proteins that robustly discriminated disease phases, among which LAMP2, IL18, and SNCA emerged as promising candidate biomarkers with diagnostic relevance Our results expand the understanding of rabies pathology in several ways. Early infection (Phase 1) was associated with innate immune activation, including modulation of autophagy-related proteins and inflammasome components. In the progressive phase (Phase 2), we observed enrichment of pathways linked to vesicular trafficking, mitochondrial dysfunction, and cytokine signaling. The late phase (Phase 3) was characterized by profound dysregulation of synaptic and neuronal proteins, converging on signatures reminiscent of neurodegenerative disorders. The upregulation of 14-3-3 proteins (YWHAQ, YWHAG, YWHAE) together with LAMP2 suggests that RABV exploits host cytoskeletal and lysosomal systems to support viral entry and replication. Similar strategies have been documented in other RNA viruses, including vesiculoviruses and coronaviruses, which manipulate cytoskeletal dynamics and endosomal trafficking to promote productive infection (Ahmad et al., 2017 ; Bojkova et al., 2020 ). In parallel, the enrichment of PI3K–Akt and MAPK signaling pathways indicates virus-driven modulation of host signaling networks, potentially enhancing neuronal survival and thereby prolonging the intracellular environment required for viral replication. Notably, 14-3-3 proteins act as multifunctional adaptors that regulate apoptosis, kinase signaling, and synaptic plasticity (Ashraf & Uversky, 2024 ; Morrison, 2009 ). Their sustained expression into later phases of infection suggests that viral subversion of these survival pathways extends beyond initial entry events and remains active throughout disease progression. The second phase was marked by pronounced activation of innate immune responses, with strong upregulation of IL18, complement C3, and NF-κB (RELA). These alterations are consistent with inflammasome activation and neuroinflammatory cascades commonly described in viral encephalitides (Fonseca et al., 2020 ; Freeman & Ting, 2022 ). While immune activation is necessary for viral clearance, excessive inflammation can exacerbate neuronal injury, a phenomenon also observed in West Nile, Japanese encephalitis, and Zika virus infections (Sullivan et al., 2021 ). The continued presence of 14-3-3 proteins during this stage suggests convergence between viral exploitation of host survival pathways and immune-driven pathology (Jiaqi et al., 2021 ; Mao et al., 2016 ). This interplay highlights rabies neuropathogenesis as a dynamic continuum, in which viral persistence and host defense mechanisms operate simultaneously rather than sequentially. Immunomodulatory strategies targeting IL18 or NF-κB may therefore attenuate pathology, but such approaches would need to balance suppression of damaging inflammation with preservation of antiviral defenses (Katz et al., 2017 ; Yuan et al., 2004 ). By the third phase, signatures of proteostasis failure, mitochondrial dysfunction, and apoptotic activation became dominant. This included accumulation of SNCA, activation of calpain proteases (CAPN1, CAPN2), and dysregulation of autophagy (ATG5, LAMP2) and proteasome subunits (PSMD2) (Beatman et al., 2017 ; Li et al., 2017 ; Kopil et al., 2012 ). These alterations parallel molecular mechanisms implicated in neurodegenerative disorders such as Parkinson’s, Alzheimer’s, and Huntington’s disease (Menzies et al., 2015 ; Schneider & Cuervo, 2014 ). The overlap between acute rabies pathology and chronic neurodegeneration suggests that rabies infection may serve as a model to investigate early molecular triggers of neuronal dysfunction. Therapeutic interventions that stabilize lysosomal or proteasomal activity, or that inhibit calpain activation and α-synuclein aggregation, could potentially provide neuroprotective benefits in late-stage disease. The integration of machine learning with proteomics strengthened the robustness of our findings. Nested cross-validation and feature selection identified minimal protein sets capable of discriminating disease phases with high accuracy, with SVM-RBF achieving the best performance. Importantly, model interpretation using SHAP values confirmed the central importance of LAMP2, IL18, and SNCA across folds, reducing the risk of model bias. These biomarkers represent promising candidates for minimally invasive diagnostics, particularly if detectable in cerebrospinal fluid (CSF) or blood-derived exosomes (Ashraf et al., 2024; Budhraja et al., 2023 ). Nevertheless, biomarker specificity remains a key concern. Elevated IL18 expression has been documented in several viral encephalitides, including HSV and JEV, while SNCA accumulation has also been observed in SARS-CoV-2–associated neuronal stress (Cheeran et al., 2005 ; Lebratti et al., 2021 ; Gemignani et al., 2025 ). By contrast, LAMP2 dysregulation appears more uniquely associated with RABV infection, suggesting it may represent a distinguishing feature of rabies-related trafficking perturbations. Beyond their diagnostic potential, these findings highlight therapeutic avenues. Interventions aimed at stabilizing autophagic and proteasomal pathways or selectively modulating inflammasome activity may help reduce neuronal damage during rabies infection. To advance translational application, future studies should test the sensitivity and specificity of these candidates across viral CNS infections and validate them in clinical cohorts. Although preliminary, biomarker-informed therapeutic stratification—spanning early antiviral treatment, intermediate immunomodulation, and late neuroprotection—emerges as a rational framework for rabies management. In conclusion, this work provides a time-resolved proteomic atlas of rabies virus infection in the CNS and nominates a small set of proteins as candidate biomarkers of disease progression. While additional validation is required before translational application, our findings advance the molecular understanding of rabies neuropathogenesis and establish a foundation for biomarker-driven approaches to diagnosis and therapeutic development. Declarations Declaration of Competing Interest All authors declare that they have no competing interests. Clinical trial number not applicable Funding This work was supported by Fundação de Amparo à Pesquisa do Estado de São Paulo (FAPESP –Process No. 2017/08215-7; 2017/17943-6, 2016/04000-3, and 2013/07467-1) and Instituto Pasteur/São Paulo/Brazil (Process No. IP02/17). Author Contribution ISSK designed the study, coordinated the research efforts, supervised the study, performed experiments, analyzed all data, and drafted the manuscript. ERF, FG and SRS, and OGR contributed intellectually to the study design and participated in discussions to refine the research approach. APS and KK conducted the machine learning analyses, including the development and implementation of the Support Vector Machines classifier and SHAP interpretation;LKI provided critical input throughout, and performed the experimental proteomics work. All authors reviewed, edited, and approved the final version of the manuscript. Data Availability All the raw data were uploaded and stored at the Center for Computational Mass Spectrometry of the University of California, San Diego, MassIVE website. They can be downloaded from: ftp:// [email protected] (temporary editor access password: RABV). References Aebersold R, Mann M (2016) Mass-spectrometric exploration of proteome structure and function. Nature 537:347–355. https://doi.org/10.1038/nature19949 Ahmad W, Li Y, Guo Y, Wang X, Duan M, Guan Z, Liu Z, Zhang M (2017) Rabies virus co-localizes with early (Rab5) and late (Rab7) endosomal proteins in neuronal and SH-SY5Y cells. Virol Sin 32:207–215. https://doi.org/10.1007/s12250-017-3968-9 Ahmed S, van Zalm P, Rudmann EA, Leone M, Keller K, Branda JA, Steen J, Mukerji SS, Steen H (2022) Using CSF proteomics to investigate herpesvirus infections of the central nervous system. Viruses 14:2757. https://doi.org/10.3390/v14122757 Ashraf HN, Uversky VN (2024) Intrinsic disorder in the host proteins entrapped in rabies virus particles. Viruses 16:916. https://doi.org/10.3390/v16060916 Beatman EL, Massey A, Shives KD, et al (2017) Alpha-synuclein expression restricts RNA viral infections in the brain. J Virol 91:e02140-16. https://doi.org/10.1128/JVI.02140-16 Bojkova D, Klann K, Koch B, Widera M, Krause D, Ciesek S, Cinatl J, Münch C (2020) Proteomics of SARS-CoV-2-infected host cells reveals therapy targets. Nature 583:469–472. https://doi.org/10.1038/s41586-020-2332-7 Budhraja S, Doborjeh M, Singh B, Tan S, Doborjeh Z, Lai E, Merkin A, Lee J, Goh W, Kasabov N (2023) Filter and wrapper stacking ensemble (FWSE): a robust approach for reliable biomarker discovery in high-dimensional omics data. Brief Bioinform 24:bbad382. https://doi.org/10.1093/bib/bbad382 Cheeran MCJ, Hu SX, Sheng WS, et al (2005) Differential responses of human brain cells to West Nile virus infection. J Neurovirol 11:512–524. https://doi.org/10.1080/13550280500384982 Chienwichai P, Ketsuwan K, Lumlertdacha B, et al (2025) Proteomic and phosphoproteomic analysis of rabies pathogenesis in the clinical canine brain and identification of a kinase inhibitor as a potential repurposed antiviral agent. PLoS One 20:e0323931. https://doi.org/10.1371/journal.pone.0323931 Dooley KE, Morimoto M, Kaszuba P, et al (2025) Evidence generation for a host-response biosignature of respiratory disease. Viruses 17:943. https://doi.org/10.3390/v17070943 Fonseca MI, Chu SH, Hernandez MX, et al (2020) Contribution of complement activation pathways to viral clearance and neuroinflammation in viral encephalitis. J Neuroinflammation 17:82. https://doi.org/10.1186/s12974-020-01755-8 Freeman LR, Ting JP (2022) The NLRP3 inflammasome: an overview of mechanisms of activation and regulation during viral infection. Immunol Rev 309:122–132. https://doi.org/10.1111/imr.13120 Gemignani JF, Netz PA, Izecksohn D, et al (2025) An amyloidogenic fragment of the spike protein from SARS-CoV-2 virus stimulates the aggregation and toxicity of Parkinson’s disease protein alpha-synuclein. ACS Chem Neurosci . https://doi.org/10.1021/acschemneuro.5c00478 Hampson K, Coudeville L, Lembo T, et al (2015) Estimating the global burden of endemic canine rabies. PLoS Negl Trop Dis 9:e0003709. https://doi.org/10.1371/journal.pntd.0003709 Hierholzer JC, Killington RA (1996) Suspension assay method. In: Mahy BWJ, Kangro HO (eds) Virology methods manual . Academic Press, San Diego, pp 39–40 Huang YN, Liu S, Park T, et al (2025) Plasma proteomic analysis identifies proteins and pathways related to Alzheimer’s risk. Alzheimers Dement 21:e70579. https://doi.org/10.1002/alz.70579 Jackson AC (2016) Human rabies: a 2016 update. Curr Infect Dis Rep 18:38. https://doi.org/10.1007/s11908-016-0540-y Jiaqi L, Shengliang C, Guofei D, et al (2021) The role of 14-3-3 proteins in cell signalling pathways and virus infection. J Cell Mol Med 25:4173–4182. https://doi.org/10.1111/jcmm.16490 Katz ISS, Guedes F, Fernandes ER, Dos Ramos Silva S (2017) Immunological aspects of rabies: a literature review. Arch Virol 162:3251–3268. https://doi.org/10.1007/s00705-017-3484-0 Kopil CM, Siebert AP, Foskett JK, Neumar RW (2012) Calpain-cleaved type 1 inositol 1,4,5-trisphosphate receptor impairs ER Ca²⁺ buffering and causes neurodegeneration in primary cortical neurons. J Neurochem 123:147–158. https://doi.org/10.1111/j.1471-4159.2012.07859.x Lebratti T, Lim YS, Cofie A, et al (2021) A sustained type I IFN–neutrophil–IL-18 axis drives pathology during mucosal viral infection. Elife 10:e65762. https://doi.org/10.7554/eLife.65762 Li L, Jin H, Wang H, et al (2017) Autophagy is highly targeted among host comparative proteomes during infection with different virulent RABV strains. Oncotarget 8:21336–21350. https://doi.org/10.18632/oncotarget.15184 Li M, Ramage H, Cherry S (2020) Deciphering flavivirus-host interactions using quantitative proteomics. Curr Opin Immunol 66:90–97. https://doi.org/10.1016/j.coi.2020.06.002 Mao X, Phanavanh B, Hamdan H, Moerman-Herzog AM, Barger SW (2016) NFκB-inducing kinase inhibits NFκB activity specifically in neurons of the CNS. J Neurochem 137:154–163. https://doi.org/10.1111/jnc.13526 Mehta S, Sreenivasamurthy S, Banerjee S, Mukherjee S, Prasad K, Chowdhary A (2015) Pathway analysis of proteomics profiles in rabies infection: towards future biomarkers? OMICS 20:97–109. https://doi.org/10.1089/omi.2015.0137 Menzies FM, Fleming A, Rubinsztein DC (2015) Compromised autophagy and neurodegenerative diseases. Nat Rev Neurosci 16:345–357. https://doi.org/10.1038/nrn3961 Morrison DK (2009) The 14-3-3 proteins: regulators of signaling proteins at multiple levels. Nat Rev Mol Cell Biol 10:789–802. https://doi.org/10.1038/nrm2789 Schneider JL, Cuervo AM (2014) Autophagy and human disease: emerging themes. Curr Opin Genet Dev 26:16–23. https://doi.org/10.1016/j.gde.2014.04.003 Sullivan C, Soos BL, Millard PJ, Kim CH, King BL (2021) Modeling virus-induced inflammation in zebrafish: a balance between infection control and excessive inflammation. Front Immunol 12:636623. https://doi.org/10.3389/fimmu.2021.636623 Ugolini G (2011) Rabies virus as a transneuronal tracer of neuronal connections. Adv Virus Res 79:165–202. https://doi.org/10.1016/B978-0-12-387040-7.00008-1 Venugopal AK, Ghantasala SS, Selvan LD, et al (2013) Quantitative proteomics for identifying biomarkers for rabies. Clin Proteomics 10:3. https://doi.org/10.1186/1559-0275-10-3 WHO (2023) Rabies fact sheet. World Health Organization. https://www.who.int/news-room/fact-sheets/detail/rabies (accessed 2025) Willoughby RE Jr, Tieves KS, Hoffman GM, et al (2005) Survival after treatment of rabies with induction of coma. N Engl J Med 352:2508–2514. https://doi.org/10.1056/NEJMoa050382 Yan W, Xiang R, Chen J, et al (2019) Proteomics analysis of suckling mouse brain infected with attenuated rabies virus strain SRV9. Acta Virol 63:423–432. https://doi.org/10.4149/av_2019_406 Yuan HJ, Hu RL, Bao SJ, Zhang SF (2004) Cloning and immunological characteristics of canine IL-18 cDNA. Xi Bao Yu Fen Zi Mian Yi Xue Za Zhi 20:526–529 Zhao P, Zhao L, Zhang K, et al (2012) Infection with street strain rabies virus induces modulation of the microRNA profile of the mouse brain. Virol J 9:159. https://doi.org/10.1186/1743-422X-9-159 Additional Declarations No competing interests reported. Supplementary Files SupplementaryTableS1.docx Cite Share Download PDF Status: Published Journal Publication published 24 Dec, 2025 Read the published version in Journal of NeuroVirology → Version 1 posted Editorial decision: Revision requested 03 Nov, 2025 Reviews received at journal 24 Oct, 2025 Reviewers agreed at journal 06 Oct, 2025 Reviewers invited by journal 01 Oct, 2025 Editor assigned by journal 23 Sep, 2025 Submission checks completed at journal 23 Sep, 2025 First submitted to journal 18 Sep, 2025 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-7652543","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":528154136,"identity":"45c1753b-4514-4ee8-8b00-b2744f74c451","order_by":0,"name":"Iana Suly Santos Katz","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAAyklEQVRIiWNgGAWjYBACNghlwcAPohIKiNciwSDZANJiQLxlEgwGB0A0MVr42M8e/PizTSJx8/nViR8eGDDI84sdIOAwnrxkaV6glm033m6WADrMcObsBAJaGHIMpBnBWs5uAGlJMLhNSAv/G+OfYIfNOLv5B3FaJHLMJEAO28Dfu41IWyTemFnznJMwnnGDd5tFgoEEYb/I9+cY3/xRZiPb3392880fFTby/NIEtMCAY4MEWKUEccpBwJ6B/wDxqkfBKBgFo2BkAQD70UDH8CznsgAAAABJRU5ErkJggg==","orcid":"","institution":"Pasteur Institute","correspondingAuthor":true,"prefix":"","firstName":"Iana","middleName":"Suly Santos","lastName":"Katz","suffix":""},{"id":528154137,"identity":"fdbbe23d-32f1-4266-ae80-b0f612d6c3c6","order_by":1,"name":"Elaine Raniero Fernandes","email":"","orcid":"","institution":"Pasteur Institute","correspondingAuthor":false,"prefix":"","firstName":"Elaine","middleName":"Raniero","lastName":"Fernandes","suffix":""},{"id":528154138,"identity":"edf5d17e-1c5b-4f8c-94a8-89d4b6bb8ca1","order_by":2,"name":"Fernanda Guedes","email":"","orcid":"","institution":"Pasteur Institute","correspondingAuthor":false,"prefix":"","firstName":"Fernanda","middleName":"","lastName":"Guedes","suffix":""},{"id":528154139,"identity":"f6cd9f8b-3429-49a6-b5e5-108d7bdd7593","order_by":3,"name":"Sandriana Dos Ramos Silva","email":"","orcid":"","institution":"Pasteur Institute","correspondingAuthor":false,"prefix":"","firstName":"Sandriana","middleName":"Dos Ramos","lastName":"Silva","suffix":""},{"id":528154140,"identity":"5ff71842-1b71-40d5-ba55-d6dda7dcc632","order_by":4,"name":"Alfonso Portabales Souza","email":"","orcid":"","institution":"Draper \u0026 Dash","correspondingAuthor":false,"prefix":"","firstName":"Alfonso","middleName":"Portabales","lastName":"Souza","suffix":""},{"id":528154141,"identity":"fe2f237d-6d74-473d-b6e4-e64dcf02326a","order_by":5,"name":"Koushik Vuttha","email":"","orcid":"","institution":"Draper \u0026 Dash","correspondingAuthor":false,"prefix":"","firstName":"Koushik","middleName":"","lastName":"Vuttha","suffix":""},{"id":528154142,"identity":"2b847133-8131-4fec-9994-a0f2236161c5","order_by":6,"name":"Leo Kei Iwai","email":"","orcid":"","institution":"Butantan Institute","correspondingAuthor":false,"prefix":"","firstName":"Leo","middleName":"Kei","lastName":"Iwai","suffix":""}],"badges":[],"createdAt":"2025-09-18 19:38:19","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-7652543/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-7652543/v1","draftVersion":[],"editorialEvents":[{"content":"https://doi.org/10.1007/s13365-025-01294-3","type":"published","date":"2025-12-24T15:57:12+00:00"}],"editorialNote":"","failedWorkflow":false,"files":[{"id":93568349,"identity":"848c073d-4f53-4254-a3bb-e79ebee54bfa","added_by":"auto","created_at":"2025-10-15 08:50:45","extension":"docx","order_by":0,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":946055,"visible":true,"origin":"","legend":"","description":"","filename":"Figure1.docx","url":"https://assets-eu.researchsquare.com/files/rs-7652543/v1/7310f21b7842703707ff59f4.docx"},{"id":93566627,"identity":"bc62407c-6402-4a93-8c56-f2faf978dbed","added_by":"auto","created_at":"2025-10-15 08:34:45","extension":"docx","order_by":1,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":52119,"visible":true,"origin":"","legend":"","description":"","filename":"Manuscript.docx","url":"https://assets-eu.researchsquare.com/files/rs-7652543/v1/47eae684eb3f2d88f13f8a36.docx"},{"id":93566638,"identity":"2c198ff9-38ea-47cc-99df-2f844feb9994","added_by":"auto","created_at":"2025-10-15 08:34:45","extension":"docx","order_by":2,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":89446,"visible":true,"origin":"","legend":"","description":"","filename":"Figure2.docx","url":"https://assets-eu.researchsquare.com/files/rs-7652543/v1/e56b2d100d6a7b2d89f4af70.docx"},{"id":93566631,"identity":"4d2c2fbe-010b-4a1a-917f-ef0860d76233","added_by":"auto","created_at":"2025-10-15 08:34:45","extension":"docx","order_by":3,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":15076,"visible":true,"origin":"","legend":"","description":"","filename":"table1.docx","url":"https://assets-eu.researchsquare.com/files/rs-7652543/v1/bc668ce7fd0e895c9146b297.docx"},{"id":93568075,"identity":"3182128b-05c4-46ea-8f73-71bea4f46acc","added_by":"auto","created_at":"2025-10-15 08:42:45","extension":"docx","order_by":4,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":246308,"visible":true,"origin":"","legend":"","description":"","filename":"Figure3.docx","url":"https://assets-eu.researchsquare.com/files/rs-7652543/v1/7c792b46530882c5a6653ba8.docx"},{"id":93571964,"identity":"7d4766d7-20aa-4764-adf2-5068beaf026b","added_by":"auto","created_at":"2025-10-15 09:06:45","extension":"docx","order_by":5,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":713537,"visible":true,"origin":"","legend":"","description":"","filename":"Figure4.docx","url":"https://assets-eu.researchsquare.com/files/rs-7652543/v1/8e141a1d7f35b79629419ccf.docx"},{"id":93566646,"identity":"c2b17a98-b031-45bc-882b-dd04cb5dea85","added_by":"auto","created_at":"2025-10-15 08:34:45","extension":"docx","order_by":6,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":117166,"visible":true,"origin":"","legend":"","description":"","filename":"Figure5.docx","url":"https://assets-eu.researchsquare.com/files/rs-7652543/v1/47d4b0b73798b1cb09a5b196.docx"},{"id":93566635,"identity":"c5c6d48a-c2bc-42f5-97f6-c9d1f838be4a","added_by":"auto","created_at":"2025-10-15 08:34:45","extension":"docx","order_by":7,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":67919,"visible":true,"origin":"","legend":"","description":"","filename":"Figure6.docx","url":"https://assets-eu.researchsquare.com/files/rs-7652543/v1/2c4a7fc48db868be53c0e7cb.docx"},{"id":93568350,"identity":"14b91a2c-0d01-4870-8c3b-b9b54c664a47","added_by":"auto","created_at":"2025-10-15 08:50:45","extension":"json","order_by":8,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":8657,"visible":true,"origin":"","legend":"","description":"","filename":"d89935cd89c34cd6bcd4fdc7bc808e1f.json","url":"https://assets-eu.researchsquare.com/files/rs-7652543/v1/11330ffb398fb8e515e1873f.json"},{"id":93566641,"identity":"b8796824-2f10-4743-8279-041c203ce99c","added_by":"auto","created_at":"2025-10-15 08:34:45","extension":"docx","order_by":9,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":15746,"visible":true,"origin":"","legend":"","description":"","filename":"SupplementaryTableS1.docx","url":"https://assets-eu.researchsquare.com/files/rs-7652543/v1/487f9df61b72adda724f1834.docx"},{"id":93568081,"identity":"4837a0db-618a-4e3d-9f92-7d6456f9f07c","added_by":"auto","created_at":"2025-10-15 08:42:46","extension":"xml","order_by":10,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":108784,"visible":true,"origin":"","legend":"","description":"","filename":"d89935cd89c34cd6bcd4fdc7bc808e1f1enriched.xml","url":"https://assets-eu.researchsquare.com/files/rs-7652543/v1/8de8649375533dc7ef095c44.xml"},{"id":93566640,"identity":"8ff563e3-fbab-4063-8bf4-60b26fe8aa22","added_by":"auto","created_at":"2025-10-15 08:34:45","extension":"eps","order_by":11,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":48633,"visible":true,"origin":"","legend":"","description":"","filename":"drawingimage1.eps","url":"https://assets-eu.researchsquare.com/files/rs-7652543/v1/06ea6da2bb298594febe9c5e.eps"},{"id":93566686,"identity":"b75fbc0b-5f95-474f-b83b-6a91df47bf9e","added_by":"auto","created_at":"2025-10-15 08:34:47","extension":"eps","order_by":12,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":48585,"visible":true,"origin":"","legend":"","description":"","filename":"drawingimage2.eps","url":"https://assets-eu.researchsquare.com/files/rs-7652543/v1/9414dac059d49c54e5745bee.eps"},{"id":93568080,"identity":"084d1d59-c570-428a-9338-362321d5c168","added_by":"auto","created_at":"2025-10-15 08:42:46","extension":"eps","order_by":13,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":48773,"visible":true,"origin":"","legend":"","description":"","filename":"drawingimage3.eps","url":"https://assets-eu.researchsquare.com/files/rs-7652543/v1/40a678c9bf6fdbe07f7ea7b1.eps"},{"id":93568082,"identity":"36d26963-aca4-4095-b8db-5e3e7a7266e4","added_by":"auto","created_at":"2025-10-15 08:42:46","extension":"eps","order_by":14,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":48633,"visible":true,"origin":"","legend":"","description":"","filename":"drawingimage1.eps","url":"https://assets-eu.researchsquare.com/files/rs-7652543/v1/286233a6b9b25bfab4a32a27.eps"},{"id":93566663,"identity":"05b36956-8efc-459d-a3e6-c611fcb03ff6","added_by":"auto","created_at":"2025-10-15 08:34:46","extension":"eps","order_by":15,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":48632,"visible":true,"origin":"","legend":"","description":"","filename":"drawingimage5.eps","url":"https://assets-eu.researchsquare.com/files/rs-7652543/v1/018e207acdf4fc005c450813.eps"},{"id":93566654,"identity":"64e5eb52-0851-4b45-8ee9-4f591d1b7d17","added_by":"auto","created_at":"2025-10-15 08:34:46","extension":"eps","order_by":16,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":106861,"visible":true,"origin":"","legend":"","description":"","filename":"drawingimage6.eps","url":"https://assets-eu.researchsquare.com/files/rs-7652543/v1/c2e093ff1a1a69ae425ed7d7.eps"},{"id":93566661,"identity":"b95a3616-7d91-4301-8fc5-3f08a475a9b9","added_by":"auto","created_at":"2025-10-15 08:34:46","extension":"eps","order_by":17,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":48593,"visible":true,"origin":"","legend":"","description":"","filename":"drawingimage7.eps","url":"https://assets-eu.researchsquare.com/files/rs-7652543/v1/d6b7d96b290b34e8de0ef1aa.eps"},{"id":93568088,"identity":"2858c227-54dd-44ce-bfa5-53f9f6f773e5","added_by":"auto","created_at":"2025-10-15 08:42:46","extension":"jpeg","order_by":18,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":15621,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage1.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-7652543/v1/1489fbef5c697d6eee8498d1.jpeg"},{"id":93568078,"identity":"37ed6a4d-3dd9-4e6b-b123-23609878cba7","added_by":"auto","created_at":"2025-10-15 08:42:45","extension":"png","order_by":19,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":45312,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage10.png","url":"https://assets-eu.researchsquare.com/files/rs-7652543/v1/87c07783efedea8a8b28e1df.png"},{"id":93566660,"identity":"1dae0180-33dd-4aef-b4c3-78d81d8ae690","added_by":"auto","created_at":"2025-10-15 08:34:46","extension":"jpeg","order_by":20,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":36337,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage11.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-7652543/v1/76f8c46ad21c8ab0ace3d813.jpeg"},{"id":93566673,"identity":"5895f60b-d3a0-431a-b5e2-afa94415ca29","added_by":"auto","created_at":"2025-10-15 08:34:47","extension":"jpeg","order_by":21,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":1074,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage12.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-7652543/v1/bb5724a941551a06344e710a.jpeg"},{"id":93566666,"identity":"1094ed78-8972-4f33-993a-1ffaf5999820","added_by":"auto","created_at":"2025-10-15 08:34:46","extension":"jpeg","order_by":22,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":255169,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage13.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-7652543/v1/997c8dbcf58bc2b46a35a6b3.jpeg"},{"id":93568356,"identity":"68d2c6f6-f21e-4f60-a8e1-417499278bb9","added_by":"auto","created_at":"2025-10-15 08:50:47","extension":"png","order_by":23,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":59623,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage14.png","url":"https://assets-eu.researchsquare.com/files/rs-7652543/v1/b277bdea7df9a906979099bc.png"},{"id":93566647,"identity":"e9fa844d-6cf3-4490-9c26-acb796c0b241","added_by":"auto","created_at":"2025-10-15 08:34:46","extension":"jpeg","order_by":24,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":359625,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage15.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-7652543/v1/d7991e5eddc092c0f1d2d00c.jpeg"},{"id":93568098,"identity":"0d4c8651-af59-4780-9219-9f27ea39e2cd","added_by":"auto","created_at":"2025-10-15 08:42:48","extension":"jpeg","order_by":25,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":187842,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage16.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-7652543/v1/d67067e27967a83f43957363.jpeg"},{"id":93568090,"identity":"90daaf05-3cdf-4638-8ef1-57ad02805bea","added_by":"auto","created_at":"2025-10-15 08:42:47","extension":"jpeg","order_by":26,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":160413,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage17.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-7652543/v1/10cfbe5edbe5ad707db245f0.jpeg"},{"id":93566662,"identity":"e18a828f-fc71-4cf1-bcc9-8d4cd26fb1a8","added_by":"auto","created_at":"2025-10-15 08:34:46","extension":"jpeg","order_by":27,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":447418,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage2.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-7652543/v1/3fa0700a0b2fbe4babfb5e05.jpeg"},{"id":93566655,"identity":"0ac51207-053c-4625-81a4-becf334f988e","added_by":"auto","created_at":"2025-10-15 08:34:46","extension":"jpeg","order_by":28,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":888803,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage3.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-7652543/v1/e2d24b4c4577d253e6b9a078.jpeg"},{"id":93566645,"identity":"bcc33cf1-dc3b-4610-b5e6-ac7e9c45a2b1","added_by":"auto","created_at":"2025-10-15 08:34:45","extension":"jpeg","order_by":29,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":14733,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage4.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-7652543/v1/cbb1f9715eb9c97ec1537b35.jpeg"},{"id":93568083,"identity":"3f679aa3-1895-48cc-a0a0-6a107a9a774b","added_by":"auto","created_at":"2025-10-15 08:42:46","extension":"jpeg","order_by":30,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":6242,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage5.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-7652543/v1/2af8a9304b1c580f58a5b72c.jpeg"},{"id":93566651,"identity":"0a25fba2-c3a9-48d9-a05b-52cb1508a953","added_by":"auto","created_at":"2025-10-15 08:34:46","extension":"jpeg","order_by":31,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":1074,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage12.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-7652543/v1/d0156d5c9913aeca27b2e10c.jpeg"},{"id":93566683,"identity":"779863a6-e012-4224-9598-4f10c61be69d","added_by":"auto","created_at":"2025-10-15 08:34:47","extension":"jpeg","order_by":32,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":678944,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage7.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-7652543/v1/6565294d83f4544a5aa4a4cb.jpeg"},{"id":93568079,"identity":"8426c8d0-af6e-43d2-8eee-c71e0d371d81","added_by":"auto","created_at":"2025-10-15 08:42:46","extension":"png","order_by":33,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":37542,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage8.png","url":"https://assets-eu.researchsquare.com/files/rs-7652543/v1/3b44e2da8da1877c88bb8692.png"},{"id":93566657,"identity":"996e0097-cfa3-48ad-a255-ef5ccdcd85b3","added_by":"auto","created_at":"2025-10-15 08:34:46","extension":"png","order_by":34,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":44961,"visible":true,"origin":"","legend":"","description":"","filename":"floatimage9.png","url":"https://assets-eu.researchsquare.com/files/rs-7652543/v1/2ef766d55b4e71b25f934bd6.png"},{"id":93568092,"identity":"993650ce-6d50-4b99-a3f9-39ccf6ef3c84","added_by":"auto","created_at":"2025-10-15 08:42:47","extension":"jpeg","order_by":35,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":53669,"visible":true,"origin":"","legend":"","description":"","filename":"groupimage1.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-7652543/v1/f51470870c13dd08377cd8b3.jpeg"},{"id":93566682,"identity":"d3885faa-f6f3-42ef-9f16-d978fee61b29","added_by":"auto","created_at":"2025-10-15 08:34:47","extension":"jpeg","order_by":36,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":26601,"visible":true,"origin":"","legend":"","description":"","filename":"groupimage2.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-7652543/v1/a433eb9c4f8f2b0eff25f102.jpeg"},{"id":93568355,"identity":"59c0b178-c72b-400f-9ead-b16b56c389dd","added_by":"auto","created_at":"2025-10-15 08:50:46","extension":"jpeg","order_by":37,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":23608,"visible":true,"origin":"","legend":"","description":"","filename":"groupimage3.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-7652543/v1/dad7efd49a6a7aaccd6f432a.jpeg"},{"id":93566652,"identity":"e5a58fe7-b3ff-43ca-a875-aa7daf7489b9","added_by":"auto","created_at":"2025-10-15 08:34:46","extension":"jpeg","order_by":38,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":56870,"visible":true,"origin":"","legend":"","description":"","filename":"groupimage4.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-7652543/v1/c2b8a463c42356bb16cb89da.jpeg"},{"id":93566650,"identity":"9fb445bc-ec7f-44b4-8cb8-015982188ef8","added_by":"auto","created_at":"2025-10-15 08:34:46","extension":"png","order_by":39,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":2841,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-7652543/v1/8ff66eed07b48e2bec15c88c.png"},{"id":93568095,"identity":"1b00bbfe-3a79-4859-9128-581fe4a1d5c1","added_by":"auto","created_at":"2025-10-15 08:42:47","extension":"png","order_by":40,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":19554,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage10.png","url":"https://assets-eu.researchsquare.com/files/rs-7652543/v1/dcf927328f0878f594fc97a3.png"},{"id":93566684,"identity":"2db349e8-5b66-4223-8157-018af52048ca","added_by":"auto","created_at":"2025-10-15 08:34:47","extension":"png","order_by":41,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":4229,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage11.png","url":"https://assets-eu.researchsquare.com/files/rs-7652543/v1/c7e6675642bfb724c9ef567d.png"},{"id":93568354,"identity":"8c08aa59-173f-423f-afb6-d34dce587a2c","added_by":"auto","created_at":"2025-10-15 08:50:46","extension":"png","order_by":42,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":935,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage12.png","url":"https://assets-eu.researchsquare.com/files/rs-7652543/v1/d163c2717c2d04112d1cb0d9.png"},{"id":93568096,"identity":"a4c461bd-0b92-4f09-ab04-c8417e0956f6","added_by":"auto","created_at":"2025-10-15 08:42:47","extension":"png","order_by":43,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":40974,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage13.png","url":"https://assets-eu.researchsquare.com/files/rs-7652543/v1/5a2fd380fe5c74ef714fa237.png"},{"id":93566664,"identity":"c2c7184f-3ef5-4540-a128-46c8e63f4d53","added_by":"auto","created_at":"2025-10-15 08:34:46","extension":"png","order_by":44,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":26987,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage14.png","url":"https://assets-eu.researchsquare.com/files/rs-7652543/v1/8227dfcb9c70fc9b2e2ccf82.png"},{"id":93566659,"identity":"404a2342-4c34-4df4-a047-5ac9ec1e2805","added_by":"auto","created_at":"2025-10-15 08:34:46","extension":"png","order_by":45,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":70497,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage15.png","url":"https://assets-eu.researchsquare.com/files/rs-7652543/v1/19ceaa9c68c5f5c15dd99685.png"},{"id":93568094,"identity":"a9f6e7ba-0125-4419-8213-67db0f1eb777","added_by":"auto","created_at":"2025-10-15 08:42:47","extension":"png","order_by":46,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":126882,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage16.png","url":"https://assets-eu.researchsquare.com/files/rs-7652543/v1/68a382fcd7817f9321fc8c4b.png"},{"id":93568089,"identity":"02e39bb8-9791-4a72-b570-cdbf0f005336","added_by":"auto","created_at":"2025-10-15 08:42:47","extension":"png","order_by":47,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":23920,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage17.png","url":"https://assets-eu.researchsquare.com/files/rs-7652543/v1/a41b6cefe095ac2055969ee7.png"},{"id":93568097,"identity":"40cadc20-a25a-4c2f-bace-06a571bd7c4e","added_by":"auto","created_at":"2025-10-15 08:42:47","extension":"png","order_by":48,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":72210,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-7652543/v1/13d151e4a53b82460a033816.png"},{"id":93566653,"identity":"a5d76fd4-f450-4ef5-8898-75db368a7b1e","added_by":"auto","created_at":"2025-10-15 08:34:46","extension":"png","order_by":49,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":179187,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage3.png","url":"https://assets-eu.researchsquare.com/files/rs-7652543/v1/e54228e4d9a54a6ba3b70e21.png"},{"id":93568093,"identity":"cad151e8-248f-457d-a7f4-8a6db33e039a","added_by":"auto","created_at":"2025-10-15 08:42:47","extension":"png","order_by":50,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":3601,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage4.png","url":"https://assets-eu.researchsquare.com/files/rs-7652543/v1/d096d84675726e8aea7df1c9.png"},{"id":93566676,"identity":"8a1001a8-101b-4a46-ac00-a06bb1516f02","added_by":"auto","created_at":"2025-10-15 08:34:47","extension":"png","order_by":51,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":2003,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage5.png","url":"https://assets-eu.researchsquare.com/files/rs-7652543/v1/7c165d7486f47231abecdf88.png"},{"id":93566677,"identity":"593bdf5c-e479-4445-85f4-bacec369e040","added_by":"auto","created_at":"2025-10-15 08:34:47","extension":"png","order_by":52,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":935,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage12.png","url":"https://assets-eu.researchsquare.com/files/rs-7652543/v1/97b1cb881b4019b479088e88.png"},{"id":93566689,"identity":"9af77179-6117-496d-a949-1dc49f08bbb3","added_by":"auto","created_at":"2025-10-15 08:34:47","extension":"png","order_by":53,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":91678,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage7.png","url":"https://assets-eu.researchsquare.com/files/rs-7652543/v1/a56b7e637cf58690baa5a0fa.png"},{"id":93566687,"identity":"eaf15170-6331-4753-8b30-b7c4a8b500ff","added_by":"auto","created_at":"2025-10-15 08:34:47","extension":"png","order_by":54,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":16940,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage8.png","url":"https://assets-eu.researchsquare.com/files/rs-7652543/v1/b7a213b9a19232b60d04f3d1.png"},{"id":93566674,"identity":"da060805-8172-4f04-ad85-2e9f376d3590","added_by":"auto","created_at":"2025-10-15 08:34:47","extension":"png","order_by":55,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":19522,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage9.png","url":"https://assets-eu.researchsquare.com/files/rs-7652543/v1/32027279d9e99027df43fd50.png"},{"id":93568086,"identity":"58b046ee-7436-4172-84b8-8d969c07efe4","added_by":"auto","created_at":"2025-10-15 08:42:46","extension":"png","order_by":56,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":16215,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinegroupimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-7652543/v1/727a66b43fc3774af19019af.png"},{"id":93568084,"identity":"5248f162-f14f-48d8-9fad-ba2043a0ab44","added_by":"auto","created_at":"2025-10-15 08:42:46","extension":"png","order_by":57,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":7086,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinegroupimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-7652543/v1/82c4b3c53a29e0ef8bb8f944.png"},{"id":93566694,"identity":"49bebc4e-804f-4094-935c-8b4b865f3c18","added_by":"auto","created_at":"2025-10-15 08:34:48","extension":"png","order_by":58,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":6496,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinegroupimage3.png","url":"https://assets-eu.researchsquare.com/files/rs-7652543/v1/317911c183e7e42612d397ce.png"},{"id":93566669,"identity":"2ff1665f-f938-4e20-bca3-51d1e1a4dca0","added_by":"auto","created_at":"2025-10-15 08:34:46","extension":"png","order_by":59,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":25369,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinegroupimage4.png","url":"https://assets-eu.researchsquare.com/files/rs-7652543/v1/198d6142175da175daa11dca.png"},{"id":93566679,"identity":"8e89482b-67dd-481c-ba3d-b3172df765ed","added_by":"auto","created_at":"2025-10-15 08:34:47","extension":"xml","order_by":60,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":106331,"visible":true,"origin":"","legend":"","description":"","filename":"d89935cd89c34cd6bcd4fdc7bc808e1f1structuring.xml","url":"https://assets-eu.researchsquare.com/files/rs-7652543/v1/b3a13683e24512f422cf502c.xml"},{"id":93566691,"identity":"9eb034bd-ca98-4e3c-9589-02fbf4d6d451","added_by":"auto","created_at":"2025-10-15 08:34:48","extension":"html","order_by":61,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":118387,"visible":true,"origin":"","legend":"","description":"","filename":"earlyproof.html","url":"https://assets-eu.researchsquare.com/files/rs-7652543/v1/12fd7d6bbd60ee8bfc144bd4.html"},{"id":93566633,"identity":"25483d84-69e6-4f9d-97ba-404d90cfa068","added_by":"auto","created_at":"2025-10-15 08:34:45","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":166625,"visible":true,"origin":"","legend":"\u003cp\u003eSchematic representation of the experimental workflow. (\u003cstrong\u003eA\u003c/strong\u003e) Workflow of experimental procedures: from sample preparation (left) to mass spectrometry-based analysis (center) and subsequent interpretation of results (right). (\u003cstrong\u003eB\u003c/strong\u003e) Machine Learning Pipeline Flow Diagram. Classifiers (SVM-RBF, Random Forest, and others) were trained and evaluated with nested cross-validation. Feature importance was assessed using SHAP to identify candidate biomarkers, which were subsequently validated by Western blot\u003c/p\u003e","description":"","filename":"1.png","url":"https://assets-eu.researchsquare.com/files/rs-7652543/v1/9a68a27c1b70c9877476c294.png"},{"id":93566626,"identity":"496a7430-854e-4e34-937a-86cf796fe198","added_by":"auto","created_at":"2025-10-15 08:34:45","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":121255,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003ePrincipal component analysis (PCA) of proteomic profiles during RABV infection.\u003c/strong\u003e (A) Variance distribution across principal components, with the first three PCs explaining 79.88% of the overall variance. (B) Two-dimensional projection (PC1 vs. PC2) demonstrates clear temporal segregation of infection stages. Distinct proteomic signatures were observed across infection phases (n = 6 per group), while PBS-treated controls remained clearly separated. (C) Ninety-five percent confidence ellipses highlight the trajectory of disease progression, transitioning from the asymptomatic phase (Phase 1, blue), to the inflammatory phase (Phase 2, yellow), and ultimately to the terminal phase (Phase 3, red). This progression reflects a significant and stepwise increase in proteomic divergence (ANOVA for PC1: F = 237.33, p \u0026lt; 0.0001; ANOVA for PC2: F = 110.67, p \u0026lt; 0.0001). These findings confirm that quantile normalization effectively minimized technical variability while retaining biologically meaningful differences.\u003c/p\u003e","description":"","filename":"2.png","url":"https://assets-eu.researchsquare.com/files/rs-7652543/v1/06573e2892d1af4ed12f01c1.png"},{"id":93568069,"identity":"6606fdc5-5abd-4c76-8d62-bd6354d540a3","added_by":"auto","created_at":"2025-10-15 08:42:45","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":152824,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eProteomic signatures, clustering patterns, and effect size distribution during RABV infection. (A\u003c/strong\u003e) Hierarchical heatmap of the 27 most significantly regulated proteins, normalized by log₂ transformation, illustrating distinct clustering patterns that separate infection phases and control samples. Color scale represents standardized expression values (Z-scores), with red indicating upregulation and blue indicating downregulation. (\u003cstrong\u003eB\u003c/strong\u003e) Distribution of effect sizes (Cohen’s d) for significantly regulated proteins within each experimental group. Bars indicate the proportion of proteins exhibiting large (d \u0026gt; 1.5), moderate (0.5 \u0026lt; d \u0026lt; 1.5), and small (d \u0026lt; 0.5) effects. Data represent group means (n = 6 per group).\u003c/p\u003e","description":"","filename":"3.png","url":"https://assets-eu.researchsquare.com/files/rs-7652543/v1/2ce91be6fe2e5745396fa869.png"},{"id":93568348,"identity":"235e3d53-8c9f-4cf6-b483-dae300d8765f","added_by":"auto","created_at":"2025-10-15 08:50:45","extension":"jpg","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":282965,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eFunctional enrichment and protein–protein interaction (PPI) networks of differentially expressed proteins (DEPs) across three phases of RABV infection. \u003c/strong\u003e(\u003cstrong\u003eA–C\u003c/strong\u003e) Bubble plots showing Gene Ontology (GO) enrichment (Biological Process, Molecular Function, Cellular Component) and KEGG pathway analysis for each infection phase. The x-axis represents the gene ratio, bubble size indicates the number of proteins, and color reflects significance (–log10 p-value). (\u003cstrong\u003eA\u003c/strong\u003e) Phase 1 (early infection): enrichment in organelle organization, intracellular transport, and cytoskeleton organization, with KEGG pathways. The PPI network (\u003cstrong\u003eD\u003c/strong\u003e) identifies a core cluster of 14-3-3 proteins. (\u003cstrong\u003eB\u003c/strong\u003e) Phase 2 (intermediate infection): enrichment in defense response and cytokine production; KEGG pathways include complement/coagulation cascades, NF-κB signaling, and cytokine–cytokine receptor interactions. The PPI network (\u003cstrong\u003eE\u003c/strong\u003e) reveals clusters of inflammatory mediators (IL18, RELA) and metabolic enzymes. (\u003cstrong\u003eC\u003c/strong\u003e) Phase 3 (late infection): enrichment in autophagy, proteolysis, and apoptosis, with KEGG pathways including autophagy and Alzheimer’s disease. The PPI network (\u003cstrong\u003eF\u003c/strong\u003e) highlights clusters of autophagy-related proteins (ATG5, ATG7) and proteasome components (PSMD6). Data are expressed as means (n=6 per group).\u003c/p\u003e","description":"","filename":"4.jpg","url":"https://assets-eu.researchsquare.com/files/rs-7652543/v1/96f0b063e9b917765fd49b70.jpg"},{"id":93568072,"identity":"19dc9cd8-8ffe-456e-8d76-f34fdbea3233","added_by":"auto","created_at":"2025-10-15 08:42:45","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":124508,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eMachine learning–based classification and biomarker prioritization.\u003c/strong\u003e (\u003cstrong\u003eA\u003c/strong\u003e) Receiver operating characteristic (ROC) curves comparing five classifiers (SVM-RBF, Stacking Ensemble, Random Forest, XGBoost, and MLP) for discrimination of infection phases. (\u003cstrong\u003eB\u003c/strong\u003e) Global SHAP importance plot for the SVM-RBF model showing the ten most predictive proteins across all phases. Each point represents a protein feature, ranked by mean absolute SHAP value, with color indicating direction of effect. (\u003cstrong\u003eC)\u003c/strong\u003e Phase-specific SHAP beeswarm plots based on 18 infected samples, highlighting proteins with the greatest discriminative power in each infection phase. Distinct predictors emerged: LAMP2 and YWHAG in Phase 1, IL18 and C3 in Phase 2, and SNCA and CAPN1 in Phase 3. These analyses provide interpretable links between classifier outputs and underlying biological signatures.\u003cstrong\u003e\u003cbr\u003e\n(A)\u003c/strong\u003e\u003c/p\u003e","description":"","filename":"5.png","url":"https://assets-eu.researchsquare.com/files/rs-7652543/v1/8f0c5e556f21d99087cc4bde.png"},{"id":93568071,"identity":"4b9b6f15-74e0-4835-a29e-440a099ac2d9","added_by":"auto","created_at":"2025-10-15 08:42:45","extension":"png","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":64841,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eWestern blot validation of candidate biomarkers. \u003c/strong\u003eRepresentative blots from independent brain samples confirmed stage-specific expression of LAMP2, IL18, and SNCA, with β-actin as loading control. Densitometric quantification showed strong correlation with proteomic data, validating these proteins as biomarkers of rabies disease progression. Expression levels were normalized to β-actin and presented as relative densitometry units. Data are expressed as means ± SEM (n = 6). *p \u0026lt; 0.05 indicates significant differences between groups.\u003c/p\u003e","description":"","filename":"6.png","url":"https://assets-eu.researchsquare.com/files/rs-7652543/v1/64c65f65a6e942afa1124cd3.png"},{"id":99172335,"identity":"f4964860-ae04-40eb-ae1e-a83018dd2a97","added_by":"auto","created_at":"2025-12-29 16:08:00","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1919578,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-7652543/v1/578d6db2-e9b1-491c-8f2c-9aa288ac35ab.pdf"},{"id":93566628,"identity":"21fefc1c-f105-46de-a410-01fba71accbd","added_by":"auto","created_at":"2025-10-15 08:34:45","extension":"docx","order_by":1,"title":"","display":"","copyAsset":false,"role":"supplement","size":15746,"visible":true,"origin":"","legend":"","description":"","filename":"SupplementaryTableS1.docx","url":"https://assets-eu.researchsquare.com/files/rs-7652543/v1/28ea46a2c5f157f572501c38.docx"}],"financialInterests":"No competing interests reported.","formattedTitle":"Proteomic and Machine Learning Signatures of Rabies Virus Infection Reveal Stage-Specific Biomarkers","fulltext":[{"header":"Introduction","content":"\u003cp\u003eRabies virus (RABV) remains one of the deadliest pathogens in human history, causing an estimated 59,000 deaths annually, with the highest burden in Africa and Asia (Hampson et al., \u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e2015\u003c/span\u003e; WHO, \u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). Unlike many viral infections that can be managed with supportive care or antiviral therapies, rabies is almost invariably fatal once clinical symptoms appear. This outcome reflects the virus\u0026rsquo;s pronounced neurotropism and its capacity to evade host immune responses while advancing toward widespread central nervous system (CNS) involvement (Jackson, \u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e2016\u003c/span\u003e; Ugolini, \u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e2011\u003c/span\u003e; Willoughby et al., \u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e2005\u003c/span\u003e).\u003c/p\u003e\u003cp\u003eClassical neuropathological hallmarks of rabies include Negri bodies, neuronal degeneration, and perivascular inflammatory infiltrates (Zhao et al., \u003cspan citationid=\"CR36\" class=\"CitationRef\"\u003e2012\u003c/span\u003e; Chienwichai et al., \u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e2025\u003c/span\u003e). These observations, while historically significant, provide static snapshots of a disease defined by dynamic molecular interactions. Rabies neuropathogenesis involves both viral subversion of host pathways and maladaptive immune responses, yet the temporal sequence of these molecular events remains poorly defined (Zhao et al., \u003cspan citationid=\"CR36\" class=\"CitationRef\"\u003e2012\u003c/span\u003e; Chienwichai et al., \u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e2025\u003c/span\u003e).\u003c/p\u003e\u003cp\u003eRecent advances in high-throughput proteomics allow direct interrogation of host protein networks, capturing dynamic alterations in signaling, trafficking, and neuroinflammation (Aebersold \u0026amp; Mann \u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e2016\u003c/span\u003e; Chienwichai et al., \u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e2025\u003c/span\u003e). Proteomic studies have begun to elucidate viral modulation of neuronal and immune pathways in CNS infections, including herpesviruses, flaviviruses, and coronaviruses (Ahmed et al., \u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e2022\u003c/span\u003e; Bojkova et al., \u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e2020\u003c/span\u003e; Li et al., \u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e2020\u003c/span\u003e). However, proteomic analyses in rabies remain limited, often restricted to single time points or narrow protein subsets (Beatman et al., \u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e2017\u003c/span\u003e; Mehta et al., \u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e2015\u003c/span\u003e; Venugopal et al., \u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e2013\u003c/span\u003e; Yan et al., \u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e2019\u003c/span\u003e;). Notably, studies by Venugopal et al. \u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e2013\u003c/span\u003e and Mehta et al. \u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e2015\u003c/span\u003e highlighted synaptic and homeostatic changes in murine brain tissue, while Yan et al. \u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e2019\u003c/span\u003e reported proteomic alterations in suckling mouse models. These efforts, though informative, failed to resolve the full temporal complexity of infection. Filling this knowledge gap is critical for the field of neurovirology, as it may uncover mechanisms of viral persistence, neuronal dysfunction, and immune-driven injury within the central nervous system (CNS).\u003c/p\u003e\u003cp\u003eComplementary computational approaches, particularly machine learning (ML), enhance biomarker discovery by identifying patterns within high-dimensional omics data (Budhraja et al., \u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e2023\u003c/span\u003e; Dooley et al., \u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e2025\u003c/span\u003e; Huang et al., \u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e2025\u003c/span\u003e). ML has been applied to biomarker discovery in Alzheimer\u0026rsquo;s disease (Budhraja et al., \u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e2023\u003c/span\u003e) and viral infections such as SARS-CoV-2 (Bojkova et al., \u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e2020\u003c/span\u003e), but remains underexplored in rabies. Coupling ML with time-resolved proteomics may enable the identification of stage-specific molecular signatures with diagnostic and therapeutic relevance, addressing the pressing need for reliable biomarkers of disease progression.\u003c/p\u003e\u003cp\u003eHere, we hypothesized that RABV infection progresses through distinct molecular phases, each characterized by unique proteomic signatures reflecting viral replication, immune activation, and neurodegeneration. To test this, we applied longitudinal label-free proteomics to brains of RABV-infected mice, integrated ML-driven classification, and validated candidate biomarkers.\u003c/p\u003e\u003cp\u003eThis integrative framework refines our understanding of rabies neuropathogenesis and supports biomarker discovery within neurovirology. By linking viral dynamics, host responses, and neuronal injury, our findings highlight opportunities for biomarker-guided diagnosis and therapeutic development, while contributing to broader efforts at the interface of viral pathogenesis, neuroinflammation, and neurodegeneration.\u003c/p\u003e"},{"header":"Materials and Methods","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e\u003ch2\u003eAnimal Model\u003c/h2\u003e\u003cp\u003e All animal experiments were conducted in accordance with the ARRIVE guidelines and approved by the Institutional Animal Care and Use Committee of Instituto Pasteur (CEUA-IP Protocol no. 04/2013). Six-week-old-Webster mice (n\u0026thinsp;=\u0026thinsp;6 per group) were housed under pathogen-free conditions with ad libitum access to food and water. Mice were randomly assigned to experimental groups, and outcome assessors were blinded to group allocation during clinical scoring and sample processing. Humane endpoints were established to minimize suffering, and euthanasia was performed using CO₂ asphyxiation in compliance with institutional welfare policies.\u003c/p\u003e\u003c/div\u003e\n\u003ch3\u003eViral Infection and Experimental Design\u003c/h3\u003e\n\u003cp\u003eAnimals were intradermally inoculated in the hind footpad with 1 \u0026times; 10⁴ focus-forming units (FFU) of the V-2 strain of RABV (Hierholzer and Killington, \u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e1996\u003c/span\u003e). Control groups received mock inoculations with sterile PBS. Following infection, animals were monitored daily for changes in clinical symptoms, which were scored using a standardized scale: 0\u0026thinsp;=\u0026thinsp;asymptomatic, 1\u0026thinsp;=\u0026thinsp;disordered movement, 2\u0026thinsp;=\u0026thinsp;ruffled fur and hunched posture, 3\u0026thinsp;=\u0026thinsp;trembling and shaking, 4\u0026thinsp;=\u0026thinsp;paralysis, and 5\u0026thinsp;=\u0026thinsp;death. Based on symptomatology, animals were euthanized and brains collected at three defined infection phases:\u003c/p\u003e\u003cp\u003ePhase 1 (early/innate): 48 hours post-infection (hpi), before clinical symptoms (score 0)\u003c/p\u003e\u003cp\u003ePhase 2 (progressive): 5\u0026ndash;7 days post-infection (dpi), onset of neurological symptoms (score 1\u0026ndash;3)\u003c/p\u003e\u003cp\u003ePhase 3 (late/severe): 8\u0026ndash;15 dpi, advanced disease with paralysis (score 4)\u003c/p\u003e\u003cp\u003eWhole brains were immediately frozen in liquid nitrogen and stored at \u0026minus;\u0026thinsp;80\u0026deg;C until processing.\u003c/p\u003e\n\u003ch3\u003eSample Collection and Protein Extraction\u003c/h3\u003e\n\u003cp\u003eFrozen brain tissue samples were homogenized in ice-cold 8M urea lysis buffer for mass spectrometric analyses. Lysates were clarified by centrifugation at 12,000 \u0026times; g for 15 min at 4\u0026deg;C, and protein concentrations were measured using the BCA Protein Assay kit (Pierce - Thermo Fisher Scientific Inc.). Aliquots containing 100 \u0026micro;g of protein per condition were reduced with 10 mM dithiothreitol (DTT) for 30 min at room temperature and alkylated with 50 mM iodoacetamide (IAA) in the dark for 1 h at room temperature. The proteins were then digested overnight at 37 \u003csup\u003eo\u003c/sup\u003eC using mass spectrometry-grade trypsin (Trypsin Gold, Promega) at an enzyme-to-substrate ratio of 1:50 (w/w). Trypsin activity was quenched by acidification with formic acid to a final concentration of 5% (v/v). Peptide samples were desalted using in-house prepared StageTips containing SDB-XC membranes (Empore Styrene Divinylbenzene extraction disk cartridge, 3M).\u003c/p\u003e\n\u003ch3\u003eLC–MS/MS Acquisition\u003c/h3\u003e\n\u003cp\u003eThe desalted peptides were resuspended in 0.1% (v/v) formic acid in water (solvent A) and separated using an EASY-nLC II nanoflow HPLC system (Thermo Fisher Scientific, Bremen, Germany) coupled online to an LTQ-Orbitrap Velos mass spectrometer (Thermo Fisher Scientific). Peptides were loaded onto a pre-column (ID 100 \u0026micro;m \u0026times; OD 360 \u0026micro;m x 5 cm, packed in-house with C18 10 \u0026micro;m beads; Acqua, Phenomenex) and separated on an analytical column (ID 75 \u0026micro;m \u0026times; OD 360 \u0026micro;m x 15 cm, packed in-house with C18 5 \u0026micro;m beads; Jupiter, Phenomenex). Separation was achieved using a linear gradient of 5\u0026ndash;30% acetonitrile in 0.1% formic acid (Solvent B) over 54 min at a constant flow rate of 200 nL/min.\u003c/p\u003e\u003cp\u003eThe mass spectrometer was operated in data-dependent acquisition mode. Full-scan MS spectra (m/z 300\u0026ndash;1800) were acquired in the Orbitrap analyzer at a resolution of 30,000. The top 10 most intense precursor ions from each scan were selected for fragmentation via collision-induced dissociation (CID) in linear ion trap, with a dynamic exclusion of 70 s. The nanospray voltage was set to 2.3 kV, and the capillary source temperature was maintained at 250\u0026deg;C.\u003c/p\u003e\n\u003ch3\u003eProtein Identification and Quantification\u003c/h3\u003e\n\u003cp\u003eTandem mass spectra were processed using PEAKS Studio software (version 7; Bioinformatics Solutions, Ontario, Canada). Database searches were performed against the \u003cem\u003eRabies lyssavirus\u003c/em\u003e (RABV) protein database (NCBI Taxonomy ID 11292). Search parameters included cysteine carbamidomethylation as a fixed modification and methionine oxidation as a variable modification. Trypsin was specified as the protease, allowing for up to two missed cleavages. The mass error tolerances were set to 10 ppm for precursor ions and 0.5 Da for fragment ions. Label-free quantification was performed using the top-three precursor ion intensity method within the Progenesis QI software (Nonlinear Dynamics).\u003c/p\u003e\u003cdiv id=\"Sec8\" class=\"Section2\"\u003e\u003ch2\u003eData Normalization and Filtering\u003c/h2\u003e\u003cp\u003eProtein intensity values were log₂-transformed and normalized using quantile normalization to correct for technical variation. Missing values were handled according to their mechanism: For proteins absent primarily in low-intensity samples, left-censored (MNAR-aware) imputation was applied using a downshifted Gaussian distribution. For sporadic missing values (\u0026lt;\u0026thinsp;5% of entries), k-nearest neighbors (k\u0026thinsp;=\u0026thinsp;3) imputation was used.\u003c/p\u003e\u003c/div\u003e\n\u003ch3\u003eStatistical Analyses\u003c/h3\u003e\n\u003cp\u003eDifferential abundance testing was performed using the limma-voom pipeline implemented in R. P-values were adjusted for multiple testing using the Benjamini\u0026ndash;Hochberg false discovery rate (FDR) procedure, and proteins with FDR\u0026thinsp;\u0026le;\u0026thinsp;0.05 were considered statistically significant. Proteins were retained for downstream analyses if they met all of the following criteria: (i) identified with \u0026ge;\u0026thinsp;2 unique peptides, sequence coverage\u0026thinsp;\u0026ge;\u0026thinsp;20%, and score\u0026thinsp;\u0026ge;\u0026thinsp;500; (ii) quantified in \u0026ge;\u0026thinsp;50% of samples within at least one experimental group; and (iii) passed limma-voom testing at FDR\u0026thinsp;\u0026le;\u0026thinsp;0.05. This stepwise filtering reduced the dataset from 2,950 identified proteins to 27 high-confidence candidates (Supplementary Table \u003cspan refid=\"MOESM1\" class=\"InternalRef\"\u003eS1\u003c/span\u003e)\u003c/p\u003e\n\u003ch3\u003eExploratory and Functional Analyses\u003c/h3\u003e\n\u003cp\u003ePrincipal component analysis (PCA) clustering was performed using the final 27 prioritized proteins. Gene Ontology (GO) biological process and KEGG pathway enrichment analyses were performed using the clusterProfiler package with a hypergeometric test. P-values were corrected for multiple testing using the Benjamini\u0026ndash;Hochberg false discovery rate (FDR) procedure, and pathways with q\u0026thinsp;\u0026le;\u0026thinsp;0.05 were considered significantly enriched. Protein\u0026ndash;protein interaction networks (PPI) were constructed with STRING v11.5, visualized in Cytoscape v3.9.1, and analyzed for topological properties.\u003c/p\u003e\u003cdiv id=\"Sec11\" class=\"Section2\"\u003e\u003ch2\u003eMachine Learning Classification and Interpretation\u003c/h2\u003e\u003cp\u003eMachine learning analyses were performed in Python (v3.10) using scikit-learn (v1.2), complemented by XGBoost (v1.7) and SHAP (v0.41) for model interpretation. Model training and deployment were managed through the Azure Machine Learning platform, with structured data handling via Azure SQL. Final dashboards and visualization for exploratory analyses were generated using QlikSense.\u003c/p\u003e\u003cp\u003eMultiple algorithms were evaluated for classifying infection phases, including Support Vector Machine with radial basis function kernel (SVM-RBF), Random Forest, XGBoost, Multi-Layer Perceptron (MLP), Logistic Regression, and a Stacking Ensemble.\u003c/p\u003e\u003cp\u003eTo avoid information leakage, we adopted a nested cross-validation scheme: (i) the outer loop used leave-one-mouse-out cross-validation for unbiased performance estimation; (ii) the inner loop conducted hyperparameter optimization by grid search. Data were partitioned using stratified sampling (70% training, 30% testing) to preserve class balance.\u003c/p\u003e\u003cp\u003eWithin each training fold, Recursive Feature Elimination with Cross-Validation (RFECV) was employed to identify minimal biomarker sets, ensuring feature selection occurred independently of test data. Correlated predictors were removed (Spearman\u0026rsquo;s ρ\u0026thinsp;\u0026ge;\u0026thinsp;0.93) to minimize redundancy. The hyperparameter ranges evaluated for each algorithm are summarized in Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e.\u003c/p\u003e\u003cp\u003e\u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e\u003ccaption language=\"En\"\u003e\u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e\u003cdiv class=\"CaptionContent\"\u003e\u003cp\u003eHyperparameter ranges for machine learning models\u003c/p\u003e\u003c/div\u003e\u003c/caption\u003e\u003ccolgroup cols=\"2\"\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e\u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e\u003cthead\u003e\u003ctr\u003e\u003cth align=\"left\" colname=\"c1\"\u003e\u003cp\u003eModel\u003c/p\u003e\u003c/th\u003e\u003cth align=\"left\" colname=\"c2\"\u003e\u003cp\u003eHyperparameters (grid search ranges)\u003c/p\u003e\u003c/th\u003e\u003c/tr\u003e\u003c/thead\u003e\u003ctbody\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eSVM-RBF\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eC = [0.1, 1, 10, 100]; γ = [1e-3, 1e-4, 1e-5]; kernel\u0026thinsp;=\u0026thinsp;rbf\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eRandom Forest\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003en_estimators = [100, 300, 500]; max_depth = [5, 10, 20]\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eXGBoost\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003elearning_rate = [0.01, 0.1]; max_depth = [3, 6, 10]; n_estimators = [200, 500]\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eMLP\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003ehidden_layers = [(50,), (100,), (100, 50)]; activation = [relu, tanh]; learning_rate = [1e-3, 1e-4]\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eLogistic Regression\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003epenalty = [l2]; C = [0.1, 1, 10]; solver = [lbfgs]\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003ctr\u003e\u003ctd align=\"left\" colname=\"c1\"\u003e\u003cp\u003eStacking Ensemble\u003c/p\u003e\u003c/td\u003e\u003ctd align=\"left\" colname=\"c2\"\u003e\u003cp\u003eBase: SVM, RF, XGBoost; Meta-learner: Logistic Regression\u003c/p\u003e\u003c/td\u003e\u003c/tr\u003e\u003c/tbody\u003e\u003c/colgroup\u003e\u003c/table\u003e\u003c/div\u003e\u003c/p\u003e\u003cp\u003ePerformance evaluation included balanced accuracy, precision, recall, F1-score, and AUC, each with 95% bootstrap confidence intervals. Calibration curves and learning curves were generated to assess model reliability. Class imbalance was addressed by stratified CV, class-weight adjustments (for SVM and Logistic Regression), and explicit reporting of balanced accuracy.\u003c/p\u003e\u003cp\u003eAmong the models tested, SVM-RBF achieved the best performance, with model interpretability assessed using SHAP (SHapley Additive exPlanations) values. SHAP values were computed with the KernelSHAP method using 100 background samples per fold, and feature importances were averaged across outer folds to ensure unbiased interpretation.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec12\" class=\"Section2\"\u003e\u003ch2\u003eWestern Blot Validation\u003c/h2\u003e\u003cp\u003eCandidate biomarkers (LAMP2, IL18, SNCA) were selected based on recurrence across ML folds, SHAP importance, and pathway relevance. Validation was performed by Western blotting on independent samples (n\u0026thinsp;=\u0026thinsp;6 per time point). Proteins were separated by SDS-PAGE, transferred to PVDF membranes, and probed with anti-LAMP2 (1:400, Abcam, ab13524), anti-IL18 (1:800, Abcam, ab71495), anti-SNCA (1:400, Abcam, ab138501), and anti-β-actin (1:5000, Sigma-Aldrich). Detection used HRP-conjugated secondary antibodies (1:5000, Santa Cruz, sc-2030) and chemiluminescence. Densitometry was performed in ImageJ (IMAGE QUANTTM400, Amersham Biosciences), normalized to β-actin, and correlated with proteomic data (Pearson\u0026rsquo;s r).\u003c/p\u003e\u003cp\u003eAn overview of the experimental workflow, including sample preparation, proteomic analysis, data processing, and the machine learning pipeline for phase classification and biomarker prioritization, is presented in Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e.\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003c/div\u003e"},{"header":"Results","content":"\u003cdiv id=\"Sec14\" class=\"Section2\"\u003e\u003ch2\u003eProteomic Landscape of RABV Infection\u003c/h2\u003e\u003cp\u003eProteomic profiling of CNS tissues across three time points post-RABV infection initially identified 2,950. After applying sequential filters for detection consistency, statistical significance, and biological relevance, a subset of 27 proteins was retained for downstream analyses. Reproducibility analysis of this filtered set showed a Pearson correlation coefficient of 0.9397, indicating high consistency between replicates.\u003c/p\u003e\u003cp\u003ePCA of the proteomic dataset demonstrated a clear temporal separation of samples, delineating three distinct stages of disease progression. The first two principal components (PC1 and PC2) accounted for 79.88% of the total variance, reflecting substantial shifts in protein expression profiles over the course of infection (\u003cb\u003eFig.\u0026nbsp;2A\u003c/b\u003e). PC1, explaining 57.42% of the variance, discriminated infected samples from controls and separated phase 3 from phases 1 and 2. PC2, accounting for 22.46% of the variance, differentiated early from intermediate infection stages (\u003cb\u003eFig.\u0026nbsp;2B\u003c/b\u003e).\u003c/p\u003e\u003cp\u003eK-means clustering analysis identified k\u0026thinsp;=\u0026thinsp;3 as the optimal solution, with an average silhouette score of 0.834, indicating that three clusters most robustly represent the intrinsic structure of the data, consistent with the separation observed in the PCA (\u003cb\u003eFig.\u0026nbsp;2C).\u003c/b\u003e Both PCA and unsupervised clustering demonstrated clear separation of samples by clinical phase. Notably, K-means clustering independently validated this three-phase structure, reinforcing the concordance between proteomic signatures and clinical phenotypes.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec15\" class=\"Section2\"\u003e\u003ch2\u003eDifferential Expression and Key Biomarkers\u003c/h2\u003e\u003cp\u003ePhase assignment was validated through gap statistical analysis and visual inspection of expression heatmaps. Hierarchical clustering analysis demonstrated clear separation between the control and experimental groups (Groups 1\u0026ndash;3) (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e3\u003c/span\u003eA). Temporal expression profiling revealed smooth transitions between phases, with approximately 15% overlap in significantly expressed proteins between adjacent phases.\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003eGroup 1 showed the most pronounced changes, with 66% of its associated proteins exhibiting large effects (d\u0026thinsp;\u0026gt;\u0026thinsp;1.5), including the upregulation of YWHAQ, YWHAG, and LAMP2. Group 2 presented a more heterogeneous profile, with 34% of proteins showing large effects, particularly involving inflammatory and metabolic markers such as IL18, C3, RELA, and ALDH2. Group 3 displayed 55% of proteins with large effects, including mitochondrial-associated proteins (e.g., MRPL12, MRPS36, HSPE1, ALDH2), proteasome-related proteins (e.g., PSMD2, CTSB, CTSD), and several ribosomal proteins (RPS2, RPS15, RPS31) (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e3\u003c/span\u003eB).\u003c/p\u003e\u003cp\u003eControl samples predominantly displayed lower expression levels for these proteins, contrasting with the coordinated upregulation observed in the experimental groups.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec16\" class=\"Section2\"\u003e\u003ch2\u003eFunctional Enrichment of Differentially Expressed Proteins Across Infection Phases\u003c/h2\u003e\u003cp\u003eAs shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e4\u003c/span\u003eA, early-phase DEPs were significantly enriched in processes related to organelle organization, protein localization, intracellular transport, cytoskeletal remodeling, and kinase-driven signaling. Strongly enriched terms included responses to endogenous and external stimuli, epithelial cell adhesion, and MAPK cascade activation. These patterns indicate early reprogramming of host cell organization and signaling to support viral entry and replication. Consistent with this, cytoplasmic components, cell junctions, and microtubule-associated complexes were prominently represented, highlighting viral exploitation of trafficking and adhesion machinery. KEGG pathway analysis revealed significant enrichment in cell cycle regulation, PI3K-Akt signaling, and viral infection\u0026ndash;related pathways such as hepatitis B/C and EGFR tyrosine kinase inhibitor resistance. The Phase 1 PPI network \u003cb\u003e(\u003c/b\u003eFig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e4\u003c/span\u003eD) showed densely interconnected clusters, with hub proteins from the 14-3-3 family (Ywhaz, Ywhab, Ywhae), Csnk1e, and Haus complex members linking cytoskeletal regulation with kinase signaling. Together, these findings demonstrate extensive host cell reprogramming during early infection to facilitate viral replication.\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003eIn the intermediate phase, Functional enrichment shifted toward immune and inflammatory regulation (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e4\u003c/span\u003eB), including activation of immune defense, cytokine production, inflammasome and NF-κB complex activity, and complement system involvement. Significant KEGG pathways included complement and coagulation cascades, NOD-like receptor, Toll-like receptor, NF-κB, and cytokine\u0026ndash;cytokine receptor interactions, as well as viral pathways such as influenza A. The Phase 2 PPI network \u003cb\u003e(\u003c/b\u003eFig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e4\u003c/span\u003eE) displayed two main clusters: one centered on NF-κB subunits (Nfkb1, Rela) and proinflammatory mediators (Il1b, Il18), and another composed of metabolic enzymes (Aldh2, Acs1, Acox1). This bipartite structure reflects simultaneous activation of innate immunity and metabolic reprogramming.\u003c/p\u003e\u003cp\u003eIn the late phase, enrichment analysis indicated dominant processes involving protein degradation and cell death (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e4\u003c/span\u003eC), including autophagy, macroautophagy, proteasome activity, and regulation of apoptosis. KEGG pathways highlighted autophagy, apoptosis, ribosome function, and neurodegeneration-related pathways such as Alzheimer\u0026rsquo;s disease, suggesting late-stage host cell damage from viral replication. The Phase 3 PPI network (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e4\u003c/span\u003eF) revealed interconnected clusters of autophagy-related proteins (Atg3, Atg5, Atg7, Atg12, Lamp2), proteasome subunits (Psmd6, Psmd7, Psmd12), and ribosomal proteins (Rps2, Rps15). Hubs such as Atg5 and Psmd6 bridged autophagy and proteasome clusters, indicating crosstalk between degradation pathways.\u003c/p\u003e\u003cp\u003eTogether, these findings illustrate a sequential response: early reorganization of cellular architecture and signaling (Phase 1), escalation of immune and inflammatory defenses (Phase 2), and activation of autophagy, proteolysis, and apoptosis leading to cell death (Phase 3). Hub protein networks corroborated this transition, emphasizing the dynamic shift from host cell remodeling to immune activation and ultimately to degradation pathways.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec17\" class=\"Section2\"\u003e\u003ch2\u003eMachine learning-based phase classification and biomarker importance\u003c/h2\u003e\u003cp\u003eTo classify disease phase, we trained and compared multiple classifiers. ROC curves (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e5\u003c/span\u003eA) demonstrated that the SVM-RBF model achieved the highest discriminative ability, followed closely by the Stacking Ensemble. Although the Stacking Ensemble (Precision\u0026thinsp;=\u0026thinsp;0.87, Recall\u0026thinsp;=\u0026thinsp;0.80, F1-score\u0026thinsp;=\u0026thinsp;0.87) showed competitive performance, SVM-RBF was selected as the primary classifier. This choice was justified by its superior precision (0.89), recall (0.83), and F1-score (0.88), as well as its consistent performance across folds and its ability to capture complex non-linear boundaries in high-dimensional proteomic data. Random Forest achieved intermediate performance (F1-score\u0026thinsp;=\u0026thinsp;0.69), whereas XGBoost (F1-score\u0026thinsp;=\u0026thinsp;0.68) and MLP (F1-score\u0026thinsp;=\u0026thinsp;0.64) were comparatively less effective. The ROC curves of all models were shifted away from the diagonal, confirming performance above random classification.\u003c/p\u003e\u003cp\u003e\u003c/p\u003e\u003cp\u003eThese results established a robust foundation for subsequent SHAP analysis, which quantified the most predictive protein features across infection phases, linking computational predictions to molecular mechanisms. The Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e5\u003c/span\u003eB presents the top 10 proteins influencing SVM-RBF predictions across all phases. Phase-specific contributions included LAMP2, YWHAG, DPYSL2, YWHAQ, ACTA 1 (Phase 1), IL18, C3, RELA (Phase 2), and SNCA, CAPN1 (Phase 3). Feature selection prioritized the top 5% of SHAP values.\u003c/p\u003e\u003cp\u003eFurther granularity is shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e5\u003c/span\u003eC, where phase-specific SHAP rankings highlight proteins exerting distinct effects depending on the infection stage. For example, LAMP2, ACTA1, IL18, YWHAQ, C3, CAPN1, and RELA displayed the strongest influence on classification outcomes. This detailed view strengthens biological interpretability and supports their candidacy as stage-specific biomarkers.\u003c/p\u003e\u003cp\u003eOverall, the SVM-RBF model demonstrated the strongest ability to capture meaningful proteomic signatures, while SHAP analysis provided transparent biological insights into the relative contributions of protein features. These findings highlight candidate biomarkers for further validation and underscore the value of integrating machine learning with molecular analysis. Expanding sample size and validating results in independent cohorts will likely improve robustness and translational applicability.\u003c/p\u003e\u003c/div\u003e\u003cdiv id=\"Sec18\" class=\"Section2\"\u003e\u003ch2\u003eValidation of Key Biomarkers Confirms Proteomic Findings\u003c/h2\u003e\u003cp\u003eWestern blot analysis of independent tissue samples validated temporal expression patterns for top biomarkers identified through SHAP analysis. LAMP2 protein levels showed significant upregulation during Phase 1 (3.2-fold vs. phase 2, p\u0026thinsp;\u0026lt;\u0026thinsp;0.01 and 6.2-fold vs. phase 3, p\u0026thinsp;\u0026lt;\u0026thinsp;0.01), with strong correlation to proteomic data (r\u0026thinsp;=\u0026thinsp;0.89, p\u0026thinsp;\u0026lt;\u0026thinsp;0.001). IL18 exhibited characteristic Phase 2 expression pattern (3.3-fold vs. phase 1, p\u0026thinsp;\u0026lt;\u0026thinsp;0.001, 5.9-fold vs. phase 3, p\u0026thinsp;\u0026lt;\u0026thinsp;0.001, r\u0026thinsp;=\u0026thinsp;0.91 with proteomics). SNCA demonstrated robust Phase 3 upregulation (2.7-fold vs. phase 1, P\u0026thinsp;\u0026lt;\u0026thinsp;0.001, 4.7-fold vs. phase 2, p\u0026thinsp;\u0026lt;\u0026thinsp;0.001 r\u0026thinsp;=\u0026thinsp;0.87 with proteomics).\u003c/p\u003e\u003c/div\u003e"},{"header":"Discussion","content":"\u003cp\u003eA major strength of this study is its time-resolved design, which captures molecular signatures across clinically defined stages of rabies rather than relying on isolated cross-sectional snapshots. To our knowledge, this represents the first proteomic analysis of rabies infection with explicit temporal resolution, providing mechanistic insight into stage-specific host responses. Within this framework, we identified 27 high-confidence proteins that robustly discriminated disease phases, among which LAMP2, IL18, and SNCA emerged as promising candidate biomarkers with diagnostic relevance\u003c/p\u003e\u003cp\u003eOur results expand the understanding of rabies pathology in several ways. Early infection (Phase 1) was associated with innate immune activation, including modulation of autophagy-related proteins and inflammasome components. In the progressive phase (Phase 2), we observed enrichment of pathways linked to vesicular trafficking, mitochondrial dysfunction, and cytokine signaling. The late phase (Phase 3) was characterized by profound dysregulation of synaptic and neuronal proteins, converging on signatures reminiscent of neurodegenerative disorders.\u003c/p\u003e\u003cp\u003eThe upregulation of 14-3-3 proteins (YWHAQ, YWHAG, YWHAE) together with LAMP2 suggests that RABV exploits host cytoskeletal and lysosomal systems to support viral entry and replication. Similar strategies have been documented in other RNA viruses, including vesiculoviruses and coronaviruses, which manipulate cytoskeletal dynamics and endosomal trafficking to promote productive infection (Ahmad et al., \u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2017\u003c/span\u003e; Bojkova et al., \u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e2020\u003c/span\u003e).\u003c/p\u003e\u003cp\u003eIn parallel, the enrichment of PI3K\u0026ndash;Akt and MAPK signaling pathways indicates virus-driven modulation of host signaling networks, potentially enhancing neuronal survival and thereby prolonging the intracellular environment required for viral replication. Notably, 14-3-3 proteins act as multifunctional adaptors that regulate apoptosis, kinase signaling, and synaptic plasticity (Ashraf \u0026amp; Uversky, \u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e2024\u003c/span\u003e; Morrison, \u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e2009\u003c/span\u003e). Their sustained expression into later phases of infection suggests that viral subversion of these survival pathways extends beyond initial entry events and remains active throughout disease progression.\u003c/p\u003e\u003cp\u003eThe second phase was marked by pronounced activation of innate immune responses, with strong upregulation of IL18, complement C3, and NF-κB (RELA). These alterations are consistent with inflammasome activation and neuroinflammatory cascades commonly described in viral encephalitides (Fonseca et al., \u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e2020\u003c/span\u003e; Freeman \u0026amp; Ting, \u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e2022\u003c/span\u003e). While immune activation is necessary for viral clearance, excessive inflammation can exacerbate neuronal injury, a phenomenon also observed in West Nile, Japanese encephalitis, and Zika virus infections (Sullivan et al., \u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e2021\u003c/span\u003e).\u003c/p\u003e\u003cp\u003eThe continued presence of 14-3-3 proteins during this stage suggests convergence between viral exploitation of host survival pathways and immune-driven pathology (Jiaqi et al., \u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e2021\u003c/span\u003e; Mao et al., \u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e2016\u003c/span\u003e). This interplay highlights rabies neuropathogenesis as a dynamic continuum, in which viral persistence and host defense mechanisms operate simultaneously rather than sequentially. Immunomodulatory strategies targeting IL18 or NF-κB may therefore attenuate pathology, but such approaches would need to balance suppression of damaging inflammation with preservation of antiviral defenses (Katz et al., \u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e2017\u003c/span\u003e; Yuan et al., \u003cspan citationid=\"CR35\" class=\"CitationRef\"\u003e2004\u003c/span\u003e).\u003c/p\u003e\u003cp\u003eBy the third phase, signatures of proteostasis failure, mitochondrial dysfunction, and apoptotic activation became dominant. This included accumulation of SNCA, activation of calpain proteases (CAPN1, CAPN2), and dysregulation of autophagy (ATG5, LAMP2) and proteasome subunits (PSMD2) (Beatman et al., \u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e2017\u003c/span\u003e; Li et al., \u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e2017\u003c/span\u003e; Kopil et al., \u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e2012\u003c/span\u003e). These alterations parallel molecular mechanisms implicated in neurodegenerative disorders such as Parkinson\u0026rsquo;s, Alzheimer\u0026rsquo;s, and Huntington\u0026rsquo;s disease (Menzies et al., \u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e2015\u003c/span\u003e; Schneider \u0026amp; Cuervo, \u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e2014\u003c/span\u003e).\u003c/p\u003e\u003cp\u003eThe overlap between acute rabies pathology and chronic neurodegeneration suggests that rabies infection may serve as a model to investigate early molecular triggers of neuronal dysfunction. Therapeutic interventions that stabilize lysosomal or proteasomal activity, or that inhibit calpain activation and α-synuclein aggregation, could potentially provide neuroprotective benefits in late-stage disease.\u003c/p\u003e\u003cp\u003eThe integration of machine learning with proteomics strengthened the robustness of our findings. Nested cross-validation and feature selection identified minimal protein sets capable of discriminating disease phases with high accuracy, with SVM-RBF achieving the best performance. Importantly, model interpretation using SHAP values confirmed the central importance of LAMP2, IL18, and SNCA across folds, reducing the risk of model bias.\u003c/p\u003e\u003cp\u003eThese biomarkers represent promising candidates for minimally invasive diagnostics, particularly if detectable in cerebrospinal fluid (CSF) or blood-derived exosomes (Ashraf et al., 2024; Budhraja et al., \u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e2023\u003c/span\u003e). Nevertheless, biomarker specificity remains a key concern. Elevated IL18 expression has been documented in several viral encephalitides, including HSV and JEV, while SNCA accumulation has also been observed in SARS-CoV-2\u0026ndash;associated neuronal stress (Cheeran et al., \u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e2005\u003c/span\u003e; Lebratti et al., \u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e2021\u003c/span\u003e; Gemignani et al., \u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e2025\u003c/span\u003e). By contrast, LAMP2 dysregulation appears more uniquely associated with RABV infection, suggesting it may represent a distinguishing feature of rabies-related trafficking perturbations. Beyond their diagnostic potential, these findings highlight therapeutic avenues. Interventions aimed at stabilizing autophagic and proteasomal pathways or selectively modulating inflammasome activity may help reduce neuronal damage during rabies infection. To advance translational application, future studies should test the sensitivity and specificity of these candidates across viral CNS infections and validate them in clinical cohorts. Although preliminary, biomarker-informed therapeutic stratification\u0026mdash;spanning early antiviral treatment, intermediate immunomodulation, and late neuroprotection\u0026mdash;emerges as a rational framework for rabies management.\u003c/p\u003e\u003cp\u003eIn conclusion, this work provides a time-resolved proteomic atlas of rabies virus infection in the CNS and nominates a small set of proteins as candidate biomarkers of disease progression. While additional validation is required before translational application, our findings advance the molecular understanding of rabies neuropathogenesis and establish a foundation for biomarker-driven approaches to diagnosis and therapeutic development.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003ch2\u003eDeclaration of Competing Interest\u003c/h2\u003e\u003cp\u003eAll authors declare that they have no competing interests.\u003c/p\u003e\u003c/p\u003e\u003cp\u003e\u003ch2\u003eClinical trial number\u003c/h2\u003e\u003cp\u003enot applicable\u003c/p\u003e\u003c/p\u003e\u003ch2\u003eFunding\u003c/h2\u003e\u003cp\u003eThis work was supported by Funda\u0026ccedil;\u0026atilde;o de Amparo \u0026agrave; Pesquisa do Estado de S\u0026atilde;o Paulo (FAPESP \u0026ndash;Process No. 2017/08215-7; 2017/17943-6, 2016/04000-3, and 2013/07467-1) and Instituto Pasteur/S\u0026atilde;o Paulo/Brazil (Process No. IP02/17).\u003c/p\u003e\u003ch2\u003eAuthor Contribution\u003c/h2\u003e\u003cp\u003eISSK designed the study, coordinated the research efforts, supervised the study, performed experiments, analyzed all data, and drafted the manuscript. ERF, FG and SRS, and OGR contributed intellectually to the study design and participated in discussions to refine the research approach. APS and KK conducted the machine learning analyses, including the development and implementation of the Support Vector Machines classifier and SHAP interpretation;LKI provided critical input throughout, and performed the experimental proteomics work. All authors reviewed, edited, and approved the final version of the manuscript.\u003c/p\u003e\u003ch2\u003eData Availability\u003c/h2\u003e\u003cp\u003eAll the raw data were uploaded and stored at the Center for Computational Mass Spectrometry of the University of California, San Diego, MassIVE website. They can be downloaded from: ftp://[email protected] (temporary editor access password: RABV).\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eAebersold R, Mann M (2016) Mass-spectrometric exploration of proteome structure and function. \u003cem\u003eNature\u003c/em\u003e 537:347\u0026ndash;355. https://doi.org/10.1038/nature19949\u003c/li\u003e\n\u003cli\u003eAhmad W, Li Y, Guo Y, Wang X, Duan M, Guan Z, Liu Z, Zhang M (2017) Rabies virus co-localizes with early (Rab5) and late (Rab7) endosomal proteins in neuronal and SH-SY5Y cells. \u003cem\u003eVirol Sin\u003c/em\u003e 32:207\u0026ndash;215. https://doi.org/10.1007/s12250-017-3968-9\u003c/li\u003e\n\u003cli\u003eAhmed S, van Zalm P, Rudmann EA, Leone M, Keller K, Branda JA, Steen J, Mukerji SS, Steen H (2022) Using CSF proteomics to investigate herpesvirus infections of the central nervous system. \u003cem\u003eViruses\u003c/em\u003e 14:2757. https://doi.org/10.3390/v14122757\u003c/li\u003e\n\u003cli\u003eAshraf HN, Uversky VN (2024) Intrinsic disorder in the host proteins entrapped in rabies virus particles. \u003cem\u003eViruses\u003c/em\u003e 16:916. https://doi.org/10.3390/v16060916\u003c/li\u003e\n\u003cli\u003eBeatman EL, Massey A, Shives KD, et al (2017) Alpha-synuclein expression restricts RNA viral infections in the brain. \u003cem\u003eJ Virol\u003c/em\u003e 91:e02140-16. https://doi.org/10.1128/JVI.02140-16\u003c/li\u003e\n\u003cli\u003eBojkova D, Klann K, Koch B, Widera M, Krause D, Ciesek S, Cinatl J, M\u0026uuml;nch C (2020) Proteomics of SARS-CoV-2-infected host cells reveals therapy targets. \u003cem\u003eNature\u003c/em\u003e 583:469\u0026ndash;472. https://doi.org/10.1038/s41586-020-2332-7\u003c/li\u003e\n\u003cli\u003eBudhraja S, Doborjeh M, Singh B, Tan S, Doborjeh Z, Lai E, Merkin A, Lee J, Goh W, Kasabov N (2023) Filter and wrapper stacking ensemble (FWSE): a robust approach for reliable biomarker discovery in high-dimensional omics data. \u003cem\u003eBrief Bioinform\u003c/em\u003e 24:bbad382. https://doi.org/10.1093/bib/bbad382\u003c/li\u003e\n\u003cli\u003eCheeran MCJ, Hu SX, Sheng WS, et al (2005) Differential responses of human brain cells to West Nile virus infection. \u003cem\u003eJ Neurovirol\u003c/em\u003e 11:512\u0026ndash;524. https://doi.org/10.1080/13550280500384982\u003c/li\u003e\n\u003cli\u003eChienwichai P, Ketsuwan K, Lumlertdacha B, et al (2025) Proteomic and phosphoproteomic analysis of rabies pathogenesis in the clinical canine brain and identification of a kinase inhibitor as a potential repurposed antiviral agent. \u003cem\u003ePLoS One\u003c/em\u003e 20:e0323931. https://doi.org/10.1371/journal.pone.0323931\u003c/li\u003e\n\u003cli\u003eDooley KE, Morimoto M, Kaszuba P, et al (2025) Evidence generation for a host-response biosignature of respiratory disease. \u003cem\u003eViruses\u003c/em\u003e 17:943. https://doi.org/10.3390/v17070943\u003c/li\u003e\n\u003cli\u003eFonseca MI, Chu SH, Hernandez MX, et al (2020) Contribution of complement activation pathways to viral clearance and neuroinflammation in viral encephalitis. \u003cem\u003eJ Neuroinflammation\u003c/em\u003e 17:82. https://doi.org/10.1186/s12974-020-01755-8\u003c/li\u003e\n\u003cli\u003eFreeman LR, Ting JP (2022) The NLRP3 inflammasome: an overview of mechanisms of activation and regulation during viral infection. \u003cem\u003eImmunol Rev\u003c/em\u003e 309:122\u0026ndash;132. https://doi.org/10.1111/imr.13120\u003c/li\u003e\n\u003cli\u003eGemignani JF, Netz PA, Izecksohn D, et al (2025) An amyloidogenic fragment of the spike protein from SARS-CoV-2 virus stimulates the aggregation and toxicity of Parkinson\u0026rsquo;s disease protein alpha-synuclein. \u003cem\u003eACS Chem Neurosci\u003c/em\u003e. https://doi.org/10.1021/acschemneuro.5c00478\u003c/li\u003e\n\u003cli\u003eHampson K, Coudeville L, Lembo T, et al (2015) Estimating the global burden of endemic canine rabies. \u003cem\u003ePLoS Negl Trop Dis\u003c/em\u003e 9:e0003709. https://doi.org/10.1371/journal.pntd.0003709\u003c/li\u003e\n\u003cli\u003eHierholzer JC, Killington RA (1996) Suspension assay method. In: Mahy BWJ, Kangro HO (eds) \u003cem\u003eVirology methods manual\u003c/em\u003e. Academic Press, San Diego, pp 39\u0026ndash;40\u003c/li\u003e\n\u003cli\u003eHuang YN, Liu S, Park T, et al (2025) Plasma proteomic analysis identifies proteins and pathways related to Alzheimer\u0026rsquo;s risk. \u003cem\u003eAlzheimers Dement\u003c/em\u003e 21:e70579. https://doi.org/10.1002/alz.70579\u003c/li\u003e\n\u003cli\u003eJackson AC (2016) Human rabies: a 2016 update. \u003cem\u003eCurr Infect Dis Rep\u003c/em\u003e 18:38. https://doi.org/10.1007/s11908-016-0540-y\u003c/li\u003e\n\u003cli\u003eJiaqi L, Shengliang C, Guofei D, et al (2021) The role of 14-3-3 proteins in cell signalling pathways and virus infection. \u003cem\u003eJ Cell Mol Med\u003c/em\u003e 25:4173\u0026ndash;4182. https://doi.org/10.1111/jcmm.16490\u003c/li\u003e\n\u003cli\u003eKatz ISS, Guedes F, Fernandes ER, Dos Ramos Silva S (2017) Immunological aspects of rabies: a literature review. \u003cem\u003eArch Virol\u003c/em\u003e 162:3251\u0026ndash;3268. https://doi.org/10.1007/s00705-017-3484-0\u003c/li\u003e\n\u003cli\u003eKopil CM, Siebert AP, Foskett JK, Neumar RW (2012) Calpain-cleaved type 1 inositol 1,4,5-trisphosphate receptor impairs ER Ca\u0026sup2;⁺ buffering and causes neurodegeneration in primary cortical neurons. \u003cem\u003eJ Neurochem\u003c/em\u003e 123:147\u0026ndash;158. https://doi.org/10.1111/j.1471-4159.2012.07859.x\u003c/li\u003e\n\u003cli\u003eLebratti T, Lim YS, Cofie A, et al (2021) A sustained type I IFN\u0026ndash;neutrophil\u0026ndash;IL-18 axis drives pathology during mucosal viral infection. \u003cem\u003eElife\u003c/em\u003e 10:e65762. https://doi.org/10.7554/eLife.65762\u003c/li\u003e\n\u003cli\u003eLi L, Jin H, Wang H, et al (2017) Autophagy is highly targeted among host comparative proteomes during infection with different virulent RABV strains. \u003cem\u003eOncotarget\u003c/em\u003e 8:21336\u0026ndash;21350. https://doi.org/10.18632/oncotarget.15184\u003c/li\u003e\n\u003cli\u003eLi M, Ramage H, Cherry S (2020) Deciphering flavivirus-host interactions using quantitative proteomics. \u003cem\u003eCurr Opin Immunol\u003c/em\u003e 66:90\u0026ndash;97. https://doi.org/10.1016/j.coi.2020.06.002\u003c/li\u003e\n\u003cli\u003eMao X, Phanavanh B, Hamdan H, Moerman-Herzog AM, Barger SW (2016) NF\u0026kappa;B-inducing kinase inhibits NF\u0026kappa;B activity specifically in neurons of the CNS. \u003cem\u003eJ Neurochem\u003c/em\u003e 137:154\u0026ndash;163. https://doi.org/10.1111/jnc.13526\u003c/li\u003e\n\u003cli\u003eMehta S, Sreenivasamurthy S, Banerjee S, Mukherjee S, Prasad K, Chowdhary A (2015) Pathway analysis of proteomics profiles in rabies infection: towards future biomarkers? \u003cem\u003eOMICS\u003c/em\u003e 20:97\u0026ndash;109. https://doi.org/10.1089/omi.2015.0137\u003c/li\u003e\n\u003cli\u003eMenzies FM, Fleming A, Rubinsztein DC (2015) Compromised autophagy and neurodegenerative diseases. \u003cem\u003eNat Rev Neurosci\u003c/em\u003e 16:345\u0026ndash;357. https://doi.org/10.1038/nrn3961\u003c/li\u003e\n\u003cli\u003eMorrison DK (2009) The 14-3-3 proteins: regulators of signaling proteins at multiple levels. \u003cem\u003eNat Rev Mol Cell Biol\u003c/em\u003e 10:789\u0026ndash;802. https://doi.org/10.1038/nrm2789\u003c/li\u003e\n\u003cli\u003eSchneider JL, Cuervo AM (2014) Autophagy and human disease: emerging themes. \u003cem\u003eCurr Opin Genet Dev\u003c/em\u003e 26:16\u0026ndash;23. https://doi.org/10.1016/j.gde.2014.04.003\u003c/li\u003e\n\u003cli\u003eSullivan C, Soos BL, Millard PJ, Kim CH, King BL (2021) Modeling virus-induced inflammation in zebrafish: a balance between infection control and excessive inflammation. \u003cem\u003eFront Immunol\u003c/em\u003e 12:636623. https://doi.org/10.3389/fimmu.2021.636623\u003c/li\u003e\n\u003cli\u003eUgolini G (2011) Rabies virus as a transneuronal tracer of neuronal connections. \u003cem\u003eAdv Virus Res\u003c/em\u003e 79:165\u0026ndash;202. https://doi.org/10.1016/B978-0-12-387040-7.00008-1\u003c/li\u003e\n\u003cli\u003eVenugopal AK, Ghantasala SS, Selvan LD, et al (2013) Quantitative proteomics for identifying biomarkers for rabies. \u003cem\u003eClin Proteomics\u003c/em\u003e 10:3. https://doi.org/10.1186/1559-0275-10-3\u003c/li\u003e\n\u003cli\u003eWHO (2023) Rabies fact sheet. World Health Organization. https://www.who.int/news-room/fact-sheets/detail/rabies (accessed 2025)\u003c/li\u003e\n\u003cli\u003eWilloughby RE Jr, Tieves KS, Hoffman GM, et al (2005) Survival after treatment of rabies with induction of coma. \u003cem\u003eN Engl J Med\u003c/em\u003e 352:2508\u0026ndash;2514. https://doi.org/10.1056/NEJMoa050382\u003c/li\u003e\n\u003cli\u003eYan W, Xiang R, Chen J, et al (2019) Proteomics analysis of suckling mouse brain infected with attenuated rabies virus strain SRV9. \u003cem\u003eActa Virol\u003c/em\u003e 63:423\u0026ndash;432. https://doi.org/10.4149/av_2019_406\u003c/li\u003e\n\u003cli\u003eYuan HJ, Hu RL, Bao SJ, Zhang SF (2004) Cloning and immunological characteristics of canine IL-18 cDNA. \u003cem\u003eXi Bao Yu Fen Zi Mian Yi Xue Za Zhi\u003c/em\u003e 20:526\u0026ndash;529\u003c/li\u003e\n\u003cli\u003eZhao P, Zhao L, Zhang K, et al (2012) Infection with street strain rabies virus induces modulation of the microRNA profile of the mouse brain. \u003cem\u003eVirol J\u003c/em\u003e 9:159. https://doi.org/10.1186/1743-422X-9-159\u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":true,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"journal-of-neurovirology","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"njiv","sideBox":"Learn more about [Journal of NeuroVirology](http://link.springer.com/journal/13365)","snPcode":"13365","submissionUrl":"https://submission.nature.com/new-submission/13365/3","title":"Journal of NeuroVirology","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Springer Hybrid","inReviewEnabled":true,"inReviewRevisionsEnabled":false},"keywords":"rabies virus, proteomics, machine learning, biomarker discovery, neurovirology, neurodegeneration, Pathogenesis","lastPublishedDoi":"10.21203/rs.3.rs-7652543/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-7652543/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eRabies virus (RABV) is a highly neurotropic pathogen with near-uniform lethality after symptom onset, yet the molecular mechanisms underlying disease progression in the central nervous system (CNS) remain poorly defined. To address this gap, we performed time-resolved, label-free quantitative proteomics in RABV-infected mouse brains at defined clinical phases of RABV infection (asymptomatic, progressive, terminal). Differential protein expression was analyzed by clustering, enrichment, and protein\u0026ndash;protein interaction networks. Machine learning models classified infection stages, and top biomarkers were validated by Western blot. Principal component and clustering analyses separated infection phases robustly, while GO/KEGG and PPI analyses revealed a progression from cytoskeletal/trafficking remodeling (early) to innate immune activation (intermediate) and proteostasis collapse/neurodegeneration-linked pathways (late). A Support Vector Machines classifier discriminated phases with high performance (F1\u0026thinsp;=\u0026thinsp;0.88; AUC\u0026thinsp;=\u0026thinsp;0.79) and SHAP interpretation highlighted LAMP2, IL18 and SNCA among the top phase-specific predictors, and confirmed experimentally. This integrative proteomics\u0026ndash;machine learning approach maps dynamic molecular transitions during RABV infection and nominates diagnostic biomarkers relevant to neurovirology. These findings provide mechanistic insights into viral neuropathogenesis and highlight parallels with neurodegeneration.\u003c/p\u003e","manuscriptTitle":"Proteomic and Machine Learning Signatures of Rabies Virus Infection Reveal Stage-Specific Biomarkers","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-10-15 08:34:40","doi":"10.21203/rs.3.rs-7652543/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"decision","content":"Revision requested","date":"2025-11-03T21:17:00+00:00","index":"","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2025-10-24T22:12:32+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"288938404726687108285436228879625431400","date":"2025-10-06T18:00:42+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2025-10-01T17:32:43+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2025-09-23T07:03:46+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2025-09-23T07:02:31+00:00","index":"","fulltext":""},{"type":"submitted","content":"Journal of NeuroVirology","date":"2025-09-18T19:24:58+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"journal-of-neurovirology","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"njiv","sideBox":"Learn more about [Journal of NeuroVirology](http://link.springer.com/journal/13365)","snPcode":"13365","submissionUrl":"https://submission.nature.com/new-submission/13365/3","title":"Journal of NeuroVirology","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Springer Hybrid","inReviewEnabled":true,"inReviewRevisionsEnabled":false}}],"origin":"","ownerIdentity":"e9ce05d4-c442-4548-9efe-7d98f7b4174b","owner":[],"postedDate":"October 15th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"published-in-journal","subjectAreas":[],"tags":[],"updatedAt":"2025-12-29T16:01:49+00:00","versionOfRecord":{"articleIdentity":"rs-7652543","link":"https://doi.org/10.1007/s13365-025-01294-3","journal":{"identity":"journal-of-neurovirology","isVorOnly":false,"title":"Journal of NeuroVirology"},"publishedOn":"2025-12-24 15:57:12","publishedOnDateReadable":"December 24th, 2025"},"versionCreatedAt":"2025-10-15 08:34:40","video":"","vorDoi":"10.1007/s13365-025-01294-3","vorDoiUrl":"https://doi.org/10.1007/s13365-025-01294-3","workflowStages":[]},"version":"v1","identity":"rs-7652543","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-7652543","identity":"rs-7652543","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}