Intrinsic Disorder in the Host Proteins Entrapped in Rabies Virus Particles

Abstract

A proteomics analysis of purified rabies virus (RABV) revealed 47 entrapped host proteins within the viral particles. Out of these, 11 proteins were highly disordered. Our study was particularly focused on five of the RABV-entrapped mouse proteins with the highest levels of disorder: Neuromodulin, Chmp4b, DnaJB6, Vps37B, and Wasl. We extensively utilized bioinformatics tools, such as FuzDrop, D²P², UniProt, RIDAO, STRING, AlphaFold, and ELM, for a comprehensive analysis of the intrinsic disorder propensity of these proteins. Our analysis suggested that these disordered host proteins might play a significant role in facilitating the rabies virus pathogenicity, immune system evasion, and the development of antiviral drug resistance. Our study highlighted the complex interaction of the virus with its host, with a focus on how the intrinsic disorder can play a crucial role in virus pathogenic processes, and suggested that these intrinsically disordered proteins (IDPs) and disorder-related host interactions can also be a potential target for therapeutic strategies.

1. Introduction

The rabies virus (RABV), also known as Rhabdovirus, causes rabies, which is a preventable (through the prompt administration of post-exposure prophylaxis (PEP) to victims of bites by rabid animals [1]) but rarely curable disease [2]. Once the symptoms start manifesting, the disease is nearly 100% fatal [3]. It was reported that an RABV infection causes more than 55,000 deaths worldwide [4].

The rabies virus affects the central nervous system, causing acute infection [5]. The transmission of the virus usually happens through the bite of a rabid animal [2,3]. The virus has a rod- or bullet-like shape, and its genome is a single-stranded, negative-sense, linear non-segmented enveloped RNA [6]. The RABV belongs to the Rhabdoviridae family and genus Lyssavirus, hence, the name rhabdovirus [6,7].

The genome encodes for five different proteins named N (nucleoprotein), P (phosphoprotein), M (matrix protein), G (glycoprotein), and L (polymerase) [6]. The bullet-shaped virus is enclosed in a lipid envelope covered by glycoproteins that facilitates the attachment of the virus to the host cell receptors and thus ensures viral entry. The helical ribonucleocapsid core is composed of the viral genome and nucleoprotein [8].

Most often, exposure to the RABV happens due to the bite or scratches of a rabid animal [2,6]. At the site of injury, the muscle cells of the new host become exposed to the rabid animal saliva, which contains the particles of the rabies virus [9,10]. The RABV initially replicates in the muscle cells, but its next destination is the peripheral nervous system [6,9,10]. The virus binds to the receptors on the nerve endings of the peripheral nervous system near the site of infection [11,12]. From here on, the RABV moves along the nerves through axonal transport to enter the peripheral nervous system [11]. Then, it moves to the main target, the central nervous system [2]. When the RABV is in the central nervous system of the host, it starts to replicate rapidly, spreading to the spinal cord and different parts of the brain, causing inflammation of the brain (encephalitis) [2].

The lifecycle of the rabies virus as it enters the host cell can be divided into the following steps:

-

Attachment/adsorption: At first, glycoprotein G of the virus interacts with the specific cell surface receptors [11];

-

Endocytosis/penetration: Then, the virus enters the host cell through receptor-mediated endocytosis [6,11];

-

Fusion: Upon binding of the glycoprotein to a host cell receptor, the pH-triggered fusion between the viral and host membranes is mediated [13,14];

-

Uncoating (envelope removal): The fusion of the viral and endosomal membranes leads to the release of the viral ribonucleoprotein (RNP) complex into the cytoplasm [6,11];

-

As the viral genome is tightly encapsidated by the viral nucleoprotein N, phosphoprotein P, and large protein L (or RNA-dependent RNA polymerase (RdRp)), upon its release into the cytoplasm, this RNP acts as the template for the transcription and replication processes catalyzed by the L-P polymerase complex [15];

-

Negri body (inclusion body or viral factory) formation: An RABV infection induces the formation of cytoplasmic inclusion bodies (Negri bodies [16]), the biogenesis of which is driven by liquid–liquid phase separation [17,18], which serve as viral factories, i.e., functional structures, where viral transcription and replication take place [15];

-

Transcription (or primary transcription): Since the genome of the RABV represents a linear, single-negative-stranded RNA, a viral-encoded RdRp (L protein) transcribes the viral antigenome RNA to mRNA in the cytoplasm [6,11]. Transcription leads to the synthesis of a positive-stranded leader RNA and five monocistronic capped and polyadenylated mRNAs;

-

Translation: A viral mRNA strand is used for the translation of five major proteins (N, P, M, G, and L);

-

Replication: RdRp replicates the progeny genome through a complementary replicative intermediate, the antigenome [6,11]. Here, “the RABV RdRp ignores the signals for mRNA synthesis on the genome to copy it into the positive-strand antigenome” [19]. After its antigenome is assembled into the RNP complex via its association with N, this replicative intermediate antigenome acts as a template for further rounds of replication to generate genomic RNA for progeny virions (antigenome is always always encapsidated by the N protein). Replication requires the newly synthesized N, P, and L proteins and a set of host factors;

-

Secondary transcription: New rounds of transcription (secondary transcription), translation, and replication take place following primary replication;

-

Assembly: All these viral particles (genome and proteins) assemble into new virions [11];

-

Budding: Assembled virions bud off from the cell surfaces of host cells, acquiring their envelope from the host cell membrane [20];

-

Release: The mature rabies virus normally releases from the cells through cell lysis and spreads through the central nervous system and brain to infect healthy cells [20].

During the assembly of viral progeny, some host proteins become integrated into the mature virion particles, which may help the virus to camouflage as host cells to escape the immune system [21]. In this article, we will focus on the analysis of the intrinsic disorder of such host proteins entrapped in the virus particles. Knowing more about the intrinsic disorder properties of these proteins will help us understand the interactions of viruses with host cells, because intrinsically disordered proteins (IDPs) and intrinsically disordered regions (IDRs) are highly flexible and can change their structure and function in response to different environments [22]. Therefore, intrinsic disorder of proteins can help viruses to become more adaptable and flexible. We can also learn the strategies of viruses in evading the immune system to help us understand the pathogenesis of the rabies virus in greater depth.

In this context, Yan Zhang and colleagues published a paper discussing the host proteins that are incorporated into RABV particles when they are released from the host cells [23]. The authors purified the viral particles to perform the proteome profiling of the RABV. They found out that along with 5 main viral proteins, 49 host proteins are also integrated into viral particles, and 24 of these directly take part in viral replication, suggesting that the virus hijacks the host cellular machinery and interacts with host proteins for efficient replication [23]. An illustrative example is given by the integration of heat shock protein (HSP70) into a matured RABV virion. Decreasing the expression of HSP70 leads to a substantial reduction in the levels of viral RNAs, proteins, and virions [24]. This suggests that the enveloped viruses utilized the host proteins specifically to carry out their replication [23].

Rabies viruses that belong to the Rhabdoviruses family bud out of host cells using the host endosomal sorting complex required for transport (ESCRT) machinery [25,26]. The hijacking of the host ESCRT machinery plays a vital role in integrating the host proteins into the virus particles [25,26]. Two important proteins in this respect are charged multivesicular body protein 4b (Chmp4b) and Vacuolar protein sorting-associated protein 37B (Vps37b); both play crucial roles in the budding process during the virus life cycle [23]. Chmp4b is an essential component of ESCRT III complex, which is responsible for the final stages of budding [23]. Thus, the protein is involved in the final detachment of the newly formed virions from the cell membrane of the host cells. On the other hand, Vps37b is involved with ESCRT I and takes part in the initial step of the viral budding process [23]. Therefore, these two proteins can serve as potential therapeutic targets.

The protocol utilized by Zhang and colleagues in this important study [23] is outlined in Supplementary Materials S1. Zhang et al. also performed a functional characterization of the 49 incorporated host proteins found in the virus particles through the gene ontology database [23]. They were aiming to achieve a deep understanding of the complex interaction of host cells and the RABV and the functional implications of these proteins in the virion particles, like the involvement in viral processes such as budding [23]. A protein–protein interaction network analysis was carried out, which also strongly suggests that many of these host proteins are involved in viral budding, especially through ESCRT machinery [23]. This implies the possibility that the virus might be exploiting these host proteins, mainly the ones involved in ESCRT machinery, to exit the host cells, further assisting the viral pathogenesis [23]. One important aspect was left unexplored by the authors, namely the intrinsic disorder status of the host proteins entrapped in RABV particles.

Intrinsically disordered proteins (IDPs) are a class of biologically active proteins without unique structures [22,27]. Contrary to traditional ordered proteins, IDPs and intrinsically disordered regions (IDRs) lack well-defined, three-dimensional structures and exist as highly dynamic conformational ensembles [22,27]. Intrinsic disorder is highly prevalent, and almost 70% of PDB structures have disordered regions [28]. IDPs are multifunctional proteins that can have multiple binding partners and are characterized by their high sensitivity to subtle changes in local environmental conditions like the pH and temperature, being capable of rapid change of their structures in response to the external environment [22,27]. IDPs/IDRs have a large interface area with a dominance of hydrophobic–hydrophobic contact. Unlike ordered proteins, IDPs have a weak hydrophobic core (if any), as their amino acid sequences have a low content of hydrophobic and aromatic residues and contain large numbers of charged and polar residues [22,29]. All these properties make intrinsically disordered proteins an integral part of the protein universe, with important biological functions that complement the functionality of ordered proteins. The flexibility and adaptability of IPDs make them suitable candidates to take part in diverse cellular functions like cell signaling, molecular recognition, and protein–protein interactions [30]. At the same time, the adaptable and flexible nature of IDPs also makes them important players in the pathogenesis of various diseases like cancer and neurodegenerative diseases [31,32,33].

In this study, to analyze the intrinsic disorder status of the host proteins entrapped in the RABV, we used the data on the 47 high-confidence host proteins reported by Zhang et al. [23]. These entrapped proteins were subjected to a multifactorial disorder analysis using a set of commonly used disorder predictors. Then, we conducted a more detailed bioinformatics characterization of the five entrapped proteins with highest levels of predicted disorder.

2. Materials and Methods

2.1. Protein Datasets

The UniProt IDs of all mouse proteins analyzed in this study were retrieved from

Table 1. Functional enrichment of the intra-set PPI network of the 47 mouse proteins entrapped in RABV particles.

ID	Description	p-Value
Biological Process (Gene Ontology)
GO:0051128	Regulation of cellular component organization	4.73 × 10−14
GO:0051049	Regulation of transport	1.28 × 10−12
GO:0032879	Regulation of localization	4.85 × 10−12
GO:0051050	Positive regulation of transport	8.84 × 10−12
GO:0008104	Protein localization	8.84 × 10−12
Molecular Function (Gene Ontology)
GO:0044877	Protein-containing complex binding	1.53 × 10−07
GO:0005515	Protein binding	3.76 × 10−07
GO:0003925	G protein activity	6.11 × 10−05
GO:0005488	Binding	6.56 × 10−05
GO:0019904	Protein domain-specific binding	0.00014
Cellular Component (Gene Ontology)
GO:0031982	Vesicle	8.04 × 10−13
GO:0042470	Melanosome	1.16 × 10−12
GO:0031410	Cytoplasmic vesicle	5.80 × 10−12
GO:0005829	Cytosol	1.83 × 10−11
GO:0005768	Endosome	2.11 × 10−11

of the Zhang et al. research article [23]. These IDs were used to collect the amino acid sequences (in FASTA format) of these proteins from the UniProt database, which are listed in Supplementary File S1. We subjected all these proteins to a bioinformatics analysis and selected the most disordered proteins for in-depth research. The selected proteins are neuromodulin (also known as growth-associated protein 43 (Gap43), calmodulin-binding protein P-57, or axonal membrane protein GAP-43; UniProt ID: P06837), a charged multivesicular body protein 4b (Chmp4b, UniProt ID: Q9D8B3), Dnaj homolog superfamily B member 6 (Dnajb6, UniProt ID: O54946), a vacuolar protein sorting-associated protein 37B (Vps37b, UniProt ID; Q8R0J7), and a Neural Wiskott–Aldrich syndrome protein (also known as actin nucleation-promoting factor WASL; UniProt ID: Q91YD9). The analysis of proteins using various bioinformatics tools discussed below was performed by submitting their amino acid sequences in FASTA format to corresponding computational platforms.

2.2. Exploration of the Intrinsic Disorder Predisposition

The susceptibility of our protein dataset to intrinsic disorder was evaluated through the RIDAO web platform, which is a convenient bioinformatics tool to generate the disorder profiles of query proteins. RIDAO combines the outputs of six commonly used per-residue disorder predictors, such as PONDR^® FIT, PONDR^® VSL2, PONDR^® VL3, PONDR^® VLXT, IUPred Short, and IUPred Long to generate the integral disorder profile of an individual query protein or to provide the global disorder characterization of a protein dataset [34]. The disorder score was assigned to each residue, with a residue with disorder score equal to or above 0.5 being considered as disordered and a residue with disorder score below 0.5 being predicted as ordered. Residues/regions with disorder scores between 0.15 and 0.5 were considered as ordered but flexible. For each protein, RIDAO also calculated the percent of predicted intrinsically disordered residues (PPIDRs), which was used for the classification of proteins as ordered (PPIDR < 10%), moderately disordered (10% ≤ PPIDR < 30%), and highly disordered (PPIDR ≥ 30%).

In this and other studies conducted by our group, we utilize multiple disorder predictors mostly for illustrative purposes, i.e., to show the similarities and differences between different predictors in the per-residue disorder propensity plots generated for individual proteins. This is in line with the accepted practice in the field to use multiple tools, as they are sensitive to different disorder-related sequence features. On the other hand, while conducting global disorder predisposition analyses of various protein datasets, we are usually ranking proteins based on the PONDR^® VSL2 outputs, as the effectiveness and accuracy of this tool has been proven in the Critical Assessment of protein Intrinsic Disorder (CAID) [35]. In the second CAID round, PONDR^® VSL2 was not listed among the top 10 predictors, being ranked #20 and #18 based on the AUC (area under the receiver operating characteristic (ROC) curve) values derived from the analysis of a 1000-residue-long sequence for the Disorder—NOX and Disorder—PDB reference datasets [36]. However, this tool was one of the fastest disorder predictors tested in CAID2, being ranked #5 based on its prediction time of 0.4 s for a sequence of 1000 residues in length. Furthermore, based on its AUC values, PONDR^® VSL2 was ranked #2 and #3 (for the Disorder—NOX and Disorder—PDB reference datasets) among the five fastest disorder predictors [36]. These observations indicated that PONDR^® VSL2 continues to be a competitive tool characterized by a short execution time and reasonably high accuracy. Therefore, we selected it for our analyses.

2.3. ELMs: Eukaryotic Linear Motifs

The ELM (eukaryotic linear motif) database is a platform used to recognize the SLiMs (short linear motifs) in the proteins [37,38,39,40,41,42,43]. The motifs recognized are special in a way that if the information on the 3D organization of a functional protein is absent, SLiMs still provide a way to evaluate the potential functionality of protein, since these functional motifs are linear, which is a unique property because of the intrinsic disorder nature of these motifs [44]. The identification of these motifs helps in the understanding of the functionality of the protein, as SLiMs are involved in important interactions and perform regulatory roles [42]. In this study, we found the eukaryotic linear motifs in the aggregation hotspots, droplet-promoting regions, multiple binding-mode regions, and molecular recognition feature (MoRF) regions of our selected proteins. The goal was to map the identified ELMs/SLiMs onto these IDRs. By identifying ELMs, the goal was to deepen our understanding of the functionality of our proteins and how they interact and play a role within the cellular environment.

2.4. Functional Annotation Derived from Disorder

D²P² is a special Database of Disordered Protein Prediction designed to facilitate the statistical comparison among different prediction methods to facilitate the analysis of IDPs [45]. Along with disorder predictions, D²P² also shows the localization of MoRF regions, unique disordered binding sites that become ordered following interaction with specific partners, and are found through the ANCHOR algorithm, PTMs, and also list the SUPERFAMILY domains from evolutionary studies [45].

2.5. FuzDrop Analysis: Identifying LLPS Promoters

We used FuzDrop [46] to predict the likelihood of proteins taking part in spontaneous liquid–liquid phase separation and generate a scoring system based on the sequence of proteins to identify the regions that promote this process. Protein with a pLLPS (probability of liquid–liquid phase separation) score of 0.60 or higher are identified as promoters of droplet formation and participants of liquid–liquid phase separation, which leads to droplet formation and generates membrane-less organelles that are important for several cellular functions such as stress response and regulation [47].

2.6. Protein–Protein Interaction Network

The STRING database strives to incorporate all established and predicted connections among proteins, comprising both the physical and functional associations [48,49,50]. Users get to analyze network visualizations, predicted connections, and functional annotations for the analysis of proteins. PPI networks of proteins were retrieved by using the STRING database (https://string-db.org, accessed on 10 March 2024). For the analysis of protein interactions through STRING, we used a medium confidence level and 500 interactors in the 1st shell to generate the PPI network. For the global interactions network, the 11 most disordered proteins were used to generate a PPI network, using the same settings mentioned above. The functional enrichment data of these proteins can be found in Supplementary Tables S1–S3.

2.7. CH-CDF Analysis

CH-CDF graph combined the results of two plots: charge–hydropathy (CH) and cumulative distribution function (CDF). The CH graph is plotted based on the net charge and hydropathy of proteins; disordered proteins tend to have high net charge and low hydropathy, and they are found to be clustered in the specific area of the plot [51,52]. A linear line is placed to separate these disordered proteins from the ordered [51,52]. A CDF plot is based on PONDR scores, plotting PONDR scores to their frequency. PONDR scores tell us about the disorder associated with the protein sequence. For the CH plot, a protein that appears above the linear boundary is considered disordered, and the one that appears below the boundary is considered as ordered [51,52].

For the CDF plot, the CDF curve for ordered proteins is plotted below the order–disorder line when it is considered to be disordered, and if it appears above this boundary, it is labeled as an ordered protein [51].

The CH-CDF plot classified proteins effectively in two categories, ordered and disordered, by plotting the average distance of the protein from the order–disorder boundary (CDF) and the scores obtained through the CH plot [53].

2.8. 3D Structures of Proteins

Alpha Fold, a protein structure database developed by DeepMind exploits an AI system to predict the 3D structures of proteins based on the amino acid sequences with a high accuracy [54].

3. Results and Discussion

3.1. Global Disorder Analysis of Host Proteins Entrapped in RABV Particles

First, to get an overview of the overall disorder status of the host (mouse) proteins entrapped in RABV particles, we analyzed these proteins using a set of commonly used per-residue disorder predictors, such as PONDR^® VSL2, PONDR^® VL3, PONDR^® VLXT, and PONDR^® FIT, IUPred Short, and IUPred Long. These predictors were accessed through the Rapid Intrinsic Disorder Analysis Online (RIDAO) platform (available at https://RIDAO.app; accessed on 10 March 2024) [34]. The average disorder scores (ADSs) and percentages of predicted disordered residues (PPDRs) were computed for each protein, employing the outputs of these per-residue predictors. The ADS is a measure of the average disorder for a protein, and the PPDR is a measure of the proportion of amino acids within a protein that have a predicted disorder score above 0.5.

The results of these analyses are summarized in Supplementary Table S4. These data were used to classify each protein by its disorder status. Of note, since the ADS does not share a direct relationship with the PPDR, we defined proteins as highly ordered if they had a PPDR of less than 10% and/or an ADS of less than 0.15. Proteins with 10% ≤ PPIDR < 30% and/or 0.15 ≤ ADS < 0.5 were considered moderately disordered. Proteins with a PPDR ≥ 30% and an ADS of 0.5 or more were labeled as highly disordered. These categorizations are consistent with the standards set in our previous publications and are in line with the accepted practice in the field [55]. This approach provides the means for a more detailed study of protein structures by clearly identifying varying levels of their structural (dis)organization.

Since the effectiveness and accuracy of PONDR^® VSL2 has been proven in the Critical Assessment of protein Intrinsic Disorder [35], we used the outputs of this tool to generate an illustrative representation of global disorder distribution in mouse proteins entrapped in the RABV particles. The results of this analysis are shown in Figure 1A, which indicates that most of the host proteins are predicted as moderately or highly disordered.

In fact, approximately 27.7% of entrapped host proteins are in the red zone (highly disordered), and an additional 27.7% are in the light pink zone (i.e., proteins with PPDR_VSL2 ≥ 30% but 0.15 ≤ ADS_VSL2 < 0.5). Furthermore, 40.4% of proteins are predicted as moderately disordered; they are located within the dark pink area and are therefore characterized by 10% ≤ PPIDR_VSL2 < 30% or 0.15 ≤ ADS_VSL2 < 0.5. None of these proteins was predicted as highly ordered based on their PPIDR_VSL2 and ADS_VSL2 data, and only two were placed in the light cyan area, being characterized by PPDR_VSL2 < 10% but ADS_VSL2 > 0.15. Figure 1A also shows that neuromodulin (UniProt ID: P06837) represents a noticeable exception, being located at the top corner of the red zone and being notably separated from other data points. These observations suggest that neuromodulin has a much higher disorder propensity than the rest of the dataset. The detailed characterization of neuromodulin as a highly disordered protein could be of particular interest for further investigation in relation to its unique functional implications in a wide range of biological processes, as well as its disease associations.

To gain further insight into the structural organization of the entrapped host proteins, we combined the outputs of two binary disorder predictors to their outputs using the charge–hydropathy (CH) plot, which classified proteins based on the distribution of charged amino acids, and the cumulative distribution function analysis. Compared to ordered proteins, disordered proteins often have a lower hydrophobicity and higher net charge [51,52]. The CDF describes the cumulative frequency of disordered proteins along the length of a given protein. If the CDF curve of a given protein is below the order–disorder boundary, this protein is considered to be disordered and is considered ordered if the CDF curve is located above this boundary [51]. The outputs of these binary predictors were used to generate the ∆CH-∆CDF plot, presenting us with the global disorder analysis for our sets of proteins [53,56]. With this technique, we were able to classify proteins based on where they fell on the plot. Quadrant 1 (Q1, bottom right) encompasses proteins that are likely structured. Quadrant 2 (Q2, bottom left) comprises proteins that are either molten globular or hybrid, i.e., proteins that are compact yet lack a distinctive 3D structure or contain noticeable levels of ordered and disordered residues. Quadrant 3 (Q3, top left) includes highly disordered proteins, whereas Quadrant 4 (Q4, top right) captures proteins that are predicted to be disordered according to the CH plot yet ordered according to the CDF plot [53,56]. Therefore, based on their position within the ∆CH-∆CDF phase space, proteins can be classified into ordered with a stable structure, molten and globule-like (not completely ordered and disordered, with a flexible structure), and highly disordered proteins lacking a stable 3D structure.

Figure 1B represents the results of the global disorder analysis of the entrapped host proteins in the form of the ∆CH-∆CDF graph. The top left quadrant is designated as Quadrant 3; it is where both binary predictors agree that the protein is unstructured and called the disorder quadrant. Neuromodulin is again acting as an outlier in the ∆CH-∆CDF plot, occupying the top-most position in Q3. In addition to neuromodulin, this quadrant contains four more highly disordered proteins. Furthermore, eight entrapped mouse proteins are classified as molten and globular or hybrid, whereas all the remaining proteins in this dataset (34 or 72.34%) are placed in Q1, indicating that they are expected to be mostly ordered. There are no proteins in the upper right quadrant (Q4). Some proteins are located at the boundaries between two quadrants, suggesting they may have mixed characteristics attributed to both adjacent quadrants, indicating that these proteins may have flexible structures.

Next, we analyzed the intra-set interactivity of mouse proteins entrapped in RABV particles. To this end, we utilized the STRING platform, which generates a protein–protein interaction (PPI) network of predicted associations based on predicted and experimentally validated information on the interaction partners of a protein of interest [50]. Surprisingly, Figure 2 shows that all 47 proteins analyzed in this study were involved in the formation of a rather dense PPI network, which is characterized by an average node degree of 10.3 and an average local clustering coefficient of 0.651. Proteins in this network are involved in 243 PPIs, which significantly exceeds the expected number of interactions (69) for a random set of proteins of the same size and degree distribution drawn from the genome. Table 1 represents the most enriched biological processes, molecular functions, and cellular components (as per Gene Ontology annotations) of the members of this network.

Note that Table 1 represents GO terms corresponding to the five biological processes, molecular functions, and cellular components characterized by the lowest false discovery rates (a measure that describes the enrichment significance evaluated as p-values corrected for multiple testing within each category using the Benjamini–Hochberg procedure). However, the complete lists of the GO terms found in this STRING-based analysis include 324 biological processes, 35 molecular functions, and 106 cellular components. In agreement with Zhang et al. [23], who indicated that based on the associated biological processes, virion-packed mouse proteins can be grouped into 12 functional categories, such as cell adhesion, cytoskeleton organization, endocytosis, exosomal secretion, morphogenesis, protein localization, protein ubiquitination, regulation of gene expression, transcription, translation, transport, and viral processes; our analysis also found all these functional categories. Some of the viral life cycle-related biological processes ascribed to the virion-entrapped mouse proteins included viral budding (Vsp4b, Tsg101, Chmp2a, and Pdcd6ip), viral budding from the plasma membrane (Vsp4b and Pdcd6ip), viral budding via the host ESCRT complex (Vsp4b, Chmp2a, and Pdcd6ip), the viral life cycle (Cd81, Chmp2a, Hsp90ab1, Pcbp1, Pdcd6ip, Rab7, Slc3a2, Tsg101, Vps37b, and Vps4b), viral release from the host cell (Vsp4b, Chmp2a, Vps37b, Tsg101, and Rab7), the regulation of the viral life cycle (Ddx3x, Ifitm2, Lgals1, Ppia, Tsg101, and Vps37b), the regulation of viral genome replication (Ddx3x, Ppia, and Ifitm2,), the regulation of viral process (Ddx3x, Ifitm2, Lgals1, Ppia, Rab7, Tsg101, and Vps37b), and positive regulation by the host of the viral process (Cfl1 and Hspa8). For the complete lists of biological processes, molecular functions, and cellular components ascribed by STRING to 47 mouse proteins entrapped in RABV particles, see Supplementary Tables S5–S7.

Importantly, based on the results of this analysis, almost none of the proteins were found to be unifunctional, and instead, most of the proteins had numerous functions and were classified in multiple functional categories. This observation is illustrated in Figure 3, showing the dependence of the number of biological processes, molecular functions, and cellular components ascribed by STRING to 47 mouse proteins entrapped in RABV particles on their levels of intrinsic disorder. Figure 3 shows that the number of biological processes ascribed to each mouse protein analyzed in this study was not correlated with their level of protein disorder. On the other hand, the number of molecular functions and cellular components showed some negative and positive correlations with the protein disorder level.

This STRING-based analysis revealed that the number of biological processes attributed to a single protein ranged from 167 to 6 for the transforming protein RhoA and polyubiquitin-C (Ubc), respectively. Fourteen proteins were shown to be involved in more than one hundred biological processes each: Rhoa (167), Rac1 (151), Cfl1 (132), Ezr (117), Rack1 (117), Hspa8, (116), Tsg101 (114), Ddx3x (113), Hsp90ab1 (109), Ppia (107), Arf6 (106), Vps4b (104), Cd81 (104), and Actb (103). Eight of these proteins were predicted as highly disordered (Ezr (56.14%), Tsg101 (53.45%), Hsp90ab1 (45.17%), Ddx3x (40.63%), Rhoa (39.90%), Hspa8 (38.70%), Vps4b (36.04%), and Cfl1 (33.73%)), as their PPIDRs ≥ 30%. Five proteins (Ppia (26.83%), Actb (16.27%), Rac1 (15.10%), Arf6 (14.86%), and Rack1 (11.67%)) were characterized by the 10% ≤ PPIDR < 30% and were therefore classified as moderately disordered. Finally, Cd81 had a PPIDR of 5.93% and was identified as mostly ordered.

The number of STRING-identified molecular functions of the 47 mouse proteins entrapped in RABV particles ranged from 21 (Hsp90ab1) to 1 (Pcbp1). There were 15 proteins, each associated with at least 10 functions: Hsp90ab1 (21), Hspa8 (19), Rhoa (17), Rac1 (16), Dnaja1 (16), Arf6 (15), Rab7 (15), Tubb5 (14), Ddx3x 12), Actb (12), Rab5c (12), Vps4b (11), Sdcbp (10), Ube2n (10), and Arf3 (10). Six of these proteins were predicted as highly disordered: Dnaja1 (61.46%), Hsp90ab1 (45.17%), Ddx3x (30.63%), Rhoa (39.90%), Hspa8 (38.7%), and Vps4b (36.04%), whereas nine proteins were classified as moderately disordered: Rab5c (24.07%), Tubb5 (18.92%), Rab7 (18.36%), Sdcbp (17.06%), Actb (16.27%), Ube2n (15.79%), Rac1 (15.10%), Arf3 (14.92%), and Arf6 (14.86%).

Based on the outputs of our STRING analysis, the largest number of cellular components (55) was ascribed to Rac1, whereas Cd151 was shown to be characterized by the least number of cellular components, with 4. There were 17 proteins associated with at least 30 cellular components each: Rac1 (55), Hspa8 (53), Ezr (48), Cfl1 (44), Rhoa (40), Arf6 (38), Actb (38), Hsp90ab1 (36), Pacsin2 (36), Chmp2a (36), Tsg101 (34), Rab7 (33), Vamp3 (44), Snx18 (32), Gapdh (31), Wasl (31), and Vps4b (30). Twelve of these proteins were identified as highly disordered: Chmp2a (84.68%), Wasl (70.46%), Vamp3 (60.02%), Pacsin2 (62.14%), Ezr (56.14%), Tsg101 (53.45%), Hsp90ab1 (45.17%), Snx18 (44.95%), Rhoa (39.90%), Hspa8 (38.70%), Vps4b (36.04%), and Cfl1 (33.73%), with the remaining five proteins being moderately disordered: Rab7 (18.36%), Actb (16.27%), Rac1 (15.10%), Arf6 (14.86%), and Gapdh (12.01%).

It was indicated earlier that many host proteins found in RABV particles were also identified in other viruses from 11 viral families [23]. That study also emphasized that 15 host proteins were most frequently recruited by different viruses: Actb (IAV, HIV, VSV, RVFV, IBV, HSV, and KSHV), Cd9 (IAV, HAV, MEV, HIV, and ASFV), Cd81 (IAV, HCV, MEV, HIV, and VV), Cfl1 (IAV, RSV, HIV, and HSV), Eno1b (IAV, HAV, MEV, HIV, VSV, IBV, ASFV, and KSHV), Gapdh (IAV, RSV, HIV, RVFV, IBV, ASFV, and KSHV), Hspa8 (RSV, HIV, VSV, RVFV, HSV, ASFV, and JUNV), Hsp90ab1 (RSV, HAV, HIV, VSV, RVFV, IBV, KSHV, and SARS), Pdcd6ip (HAV, HIV, VSV, HSV, and JUNV), Ppia (IAV, MEV, HIV, HSV, and KSHV), Rab5c (HAV, HIV, HIV, ASFV, and HSV), Rab7a (HAV, HIV, RVFV, HSV, KSHV, and JUNV), Tubb5 (IAV, RSV, HIV, VSV, and ASFV), Ubc (IAV, RSV, HIV, VSV, and JUNV), and Ywhaz (HIV, HSV, KSHV, and SARS) [23]. These observations indicated that such “multiviral” host proteins (i.e., those widely recruited by different viruses) might endow viruses with some benefits for their replication cycles [23]. We looked at the intrinsic disorder predispositions of these proteins and found that six of them (Ywhaz (53.88%), Hsp90ab1 (45.17%), Pdcd6ip (44.53%), Hspa8 (38.70%), Cfl1 (33.73%), and Ubc (30.93%)) are highly disordered, eight (Ppia (26.83%), Rab5c (24.17%), Tubb5 (18.92%), Rab7a (18.36%), Actb (16.27%), Eno1b (14.52%), Cd9 (14.15%), and Gapdh (12.01%)) are moderately disordered, whereas Cd81 (5.93%) is mostly ordered. Therefore, the results of this analysis indicate that intrinsic disorder might also contribute to the “multiviral” functionality of these proteins.

Next, we looked for the presence of a correlation between the level of intrinsic disorder in a given protein and its interactivity within the intra-set PPI network (i.e., its node degree). The results of this analysis are shown in Figure 4A, illustrating that such a correlation is almost absent.

Figure 4A shows that in the intra-set PPI network analyzed in this study, almost half of the mouse proteins entrapped in the RABV particles are engaged in more than 12 interactions (i.e., serve as hubs of this network, with a hub being defined here as a node, with the number of interactions exceeding the average node degree of this network, which is 10.3). However, there is no clear disorder enrichment among hubs. These observations suggest that this intra-set PPI network is almost disorder neutral. This is a rather interesting and unexpected observation, as typically, there is a strong positive correlation between the protein interactivity and its intrinsic disorder predisposition. In fact, it is indicated in many studies that one of the remarkable functional features of IDPs and IDRs is their extraordinary binding promiscuity [33,57,58,59,60,61]. Therefore, IDPs/IDRs are considered as binding “professionals”, which continuously interact with various partners via multiple binding modes [33,57,58,59] and form static, semi-static, dynamic, or fuzzy complexes [60,61]; as well, they can be engaged in polyvalent interactions, where multiple binding sites of one protein are simultaneously bound to multiple receptors on another protein [62]. Often, disorder-based interactions are characterized by a combination of high specificity and low affinity [63], and many IDPs/IDRs can fold (at least partially) as a result of binding to their partners [64,65,66]. The degree of such binding-induced folding can be different in various systems, thereby forming complexes with broad structural and functional heterogeneity [60,61]. Furthermore, some IDPs/IDRs are capable of adopting different structures while forming complexes with different partners, thereby acting as morphing shape changers [58,66,67,68,69,70,71,72,73,74,75]. Often, significant levels of disorder are retained by IDPs/IDRs in their bound state (at least outside the binding interface), resulting in the formation of so-called fuzzy complexes [76,77,78,79,80,81,82,83]. Therefore, it is not surprising that many IDPs/IDRs serve as hub proteins: nodes in complex PPI networks that have a very large number of connections to other nodes [71,84,85,86,87,88,89]. As is shown in Figure 1A and Figure 4A, only two mouse proteins entrapped in the RABV particles are classified as mostly ordered (Galectin-1, UniProt ID: P16045 and CD81 antigen, UniProt ID: P35762), whereas all other proteins contain noticeable levels of disorder. It is therefore very likely that the IDRs found in all these moderately and highly disordered proteins are related to their interactability. Furthermore, considered here that the PPI network characterizes only the intra-set connectivity, it does not describe the overall interactivity of these proteins. In fact, as it follows from our comprehensive analyses of the most disordered proteins (see below), all of them are expected to be highly promiscuous binders. For example, STRING-generated PPI networks centered at the mouse neuromodulin (UniProt ID: P06837; PPIDR_VSL2 = 100.0%), Chmp4b (UniProt ID: Q9D8B3; PPIDR_VSL2 = 84.7%), DnaJB6 (UniProt ID: O54946; PPIDR_VSL2 = 96.4%), Vps37B (UniProt ID: Q8R0J7; PPIDR_VSL2 = 80.4%), and Wasl (UniProt ID: Q91YD9; PPIDR_VSL2 = 70.5%) contain 145, 100, 68, 42, and 232 nodes, respectively (see below). This is in striking contrast to their intra-set node degrees of 3, 9, 3, 9, and 8, respectively (see Figure 4A). Finally, one should keep in mind that although a positive correlation between the protein interactivity and intrinsic disorder predisposition is typically observed, ordered proteins can serve as hubs as well, but in this case, partners of such ordered hubs are mostly IDPs or proteins with IDRs [71,75].

Next, we analyzed the predisposition of mouse proteins entrapped in the RABV particles to serve as drivers of liquid–liquid phase separation (LLPS) using the FuzDrop platform [46]. The results of this analysis are summarized in Figure 4B, showing dependence of the probability of analyzed proteins for spontaneous liquid–liquid phase separation, p_LLPS, on their intrinsic disorder status. This analysis revealed that there is a strong positive correlation between PPIDR_VSL2 and p_LLPS, and all seven proteins predicted as droplet drivers (i.e., proteins characterized by p_LLPS ≥ 0.60) are also predicted to be highly disordered. It is recognized now that a significant part of cellular processes is determined by the functioning of liquid droplet-like condensates: membrane-less organelles (MLOs) [90,91]. In fact, MLOs are very diverse and commonly found in the cytoplasm, nucleus, mitochondria of various eukaryotic cells, chloroplasts of plant cells, as well as in bacterial cells. Biogenesis of MLOs is driven by the intracellular LLPS processes, which are also known as liquid–liquid demixing phase separation [92,93] and are strongly dependent on IDPs and IDRs [94,95]. In fact, many of the MLO resident proteins are IDPs or contain IDRs, and the formation of all the MLOs analyzed so far relies on IDPs/IDRs, indicating that intrinsic disorder is important for MLO biogenesis [92].

After subjecting all 47 mouse proteins found in the rabies virus to the intrinsic disorder analysis, we selected the 11 most disordered proteins for a comprehensive analysis, with 5 of these highly disordered proteins being discussed in detail (see below for discussion of neuromodulin and Appendix A for the detailed discussions of Chmp4b, DnaJB6, Vps37B, and Wasl). The information about the remaining highly disordered proteins (Pascin2, Ddx3x, Snx18, Tsg101, and Ezr) can be found in the Supplementary Materials (see Supplementary Figures S1–S5).

3.2. Functional Intrinsic Disorder in the Most Disordered Mouse Proteins Found in the Rabies Virus

Neuromodulin (UniProt ID: P06837)

Neuromodulin is a protein encoded by the gene Gap43. This protein is involved in neuron growth acting as a crucial component of the growth cones present at the tips of elongating axons (https://www.uniprot.org/uniprotkb/P06837/entry; accessed on 10 March 2024).

In mice, neuromodulin is a peripheral membrane protein that is not entirely embedded in the membrane but associated with it, which allows for its dynamic interaction with other membrane proteins. Neuromodulin is transported to the growth cones of neurons. These growth cones are present at the tips of the axons and are essential for guiding the direction of neuronal growth during development and regeneration. Several studies have been conducted to elucidate the process by which protein is transported to the growth cones. Zuber et al. suggested that the N-terminal, ten-amino acid sequence is sufficient to target the protein to these growth cones [96]. However, later, an experiment conducted with a fusion protein combining neuromodulin and β-galactosidase, which is an enzyme used as a marker in an experiment, revealed that the N-terminal, ten-amino acid sequence only is not sufficient to transport a protein to its target, and the protein’s ability to attach to the membrane through palmitoylation at cysteines 3 and 4 is also essential for assembling the protein at the growth cones [97,98]. This also signifies the importance of post-translational modification in the protein.

The mouse neuromodulin is a 227-residue long, highly disordered protein of 23.6 kDa, whose interactions with calmodulin along with neurogranin are crucial for learning and memory formation in the nervous system [99]. This protein, which is also designated as GAP-43 or P-57 neuromodulin, is one of the main presynaptic substrates of protein Kinase C [99,100,101]. The phosphorylation of neuromodulin leads to a decreased affinity for calmodulin [99]. Under low-calcium-ion conditions, the protein binds to calmodulin through a highly unstructured IQ motif (I/L/V) QXXXRXXXX(R/K), which adopts an α-helical confirmation upon binding with calmodulin [99]. Phosphorylation through protein Kinase at serine residues modulates this interaction, influencing the behavior of F actin in the growth cones of neurons [100].

Along with this, this protein consists of a “Gap junction protein N-terminal region” (residues 2–31) and IQ motif (residues 31–60). Phosphorylation occurs at Ser41, Ser86, Serine96, Thr88, Thr89, Thr89, Thr95, Ser96, Ser103, Thr138, Ser142, Ser144, Ser145, Thr172, Ser192, and Ser 193. Palmitoylation at cysteine residues at positions 3 and 4 (more specifically, S-palmitoyl cysteine modification) is important for protein association with the cellular membrane and its location. The loss of these modifications at these sites are mutations associated with PTM and can prevent the protein from properly being lipidated and lead to changes in the protein function and location (https://www.uniprot.org/uniprotkb/P06837/entry; accessed on 10 March 2024).

Figure 5 represents the results of the functional disorder analysis of this protein. The per-residue disorder profile generated using RIDAO indicates that neuromodulin is predicted as a highly disordered protein (see Figure 5A). In fact, the PPIDR scores determined using the disorder predictors PONDR^® FIT, PONDR^® VSL2, PONDR^® VL3, PONDR^® VLXT, IUPred Short, and IUPred Long were 100%, 100%, 100%, 90.75%, 96.68%, and 99.56%, respectively. The mean disorder profile (MDP) was 100%, signifying that the protein is highly disordered. The residues are predicted to be disordered above the 0.5 threshold, and an MDP value of 100% implies that neuromodulin in its entirety is likely to be intrinsically disordered [34].

The D²P² platform was used to generate a functional disorder profile for neuromodulin (see Figure 5B). The top section of the image is showing colored bars that represent the disordered regions predicted by each predictor, such as IUPred-L, IUPred-S, PV2, PrDOS, VSL2b, VLXT, Espritz-D, Espritz-X, and Espritz-N [45]. Below these colored bars of predicted disorder is the domain prediction bar exhibiting three domains, with one of these domains marked as number 3, being the IQ domain of neuromodulin we discussed above. It ranges from residue 31 to 50 and is known as an IQ calmodulin-binding motif.

The consensus bar of a green color is the predicted disorder agreement between all predictors. According to D²P² platform, all the predictors agree that the disorder regions are found at residues 2–227. This protein is highly unstructured, being the most disordered among all the 47 host proteins analyzed in this study. Moving on with the D²P² results, the yellow zigzagged lines represent MoRF regions. MoRF regions is short for molecular recognition features, which are disordered protein regions that become ordered upon binding to the respective protein partners [102]. Multiple MoRF regions are found at the ranges 1–9, 32–52, 58–81, 102–109, and 116–227, identified through the ANCHOR algorithm and also named as the disorder-based binding sites, indicating that neuromodulin has a tendency to engage in disorder-to-order transition-based interactions. Below these MoRF region predictions are the differently colored circles with letters representing the PTMs sites along the length of the protein. Other than this, D²P² also included the superfamily annotation and Pfam domains, indicating the large family the protein belongs to and the shared structural and functional domains within the family, giving insight into the role of the protein and its functional profile.

Figure 5C represents the FuzDrop-generated plot, showing the sequence distribution of the residue-based, droplet-promoting probabilities, p_DP. Residues with p_DP values above 0.6 are capable of promoting liquid–liquid phase separation. In neuromodulin, most of the residues have p_DP values above the indicated threshold. Therefore, most of the neuromodulin residues have a high probability of promoting droplet formation. Peaks in the graph indicate the regions that can promote the formation of membrane-less organelles in the cells through liquid–liquid phase separation. Membrane-less organelles are liquid compartments within the cell involved in specific biological functions, like in gene regulation, that are not enclosed by traditional lipid membranes [103]. In neuromodulin, the droplet-promoting region, i.e., a region that is particularly susceptible to phase separation, is located at the residues 2–127. Furthermore, neuromodulin contains one aggregation hotspot (residues 52–66), which is a region with high probability of promoting droplet formation that is predicted to exhibit a multiplicity of binding modes, enabling the adaptability of interactions to the cellular context. Furthermore, the p_LLPS value was predicted to be 0.9949 for neuromodulin. Since the proteins with p_LLPS ≥ 0.60 are designated as droplet drivers, with a tendency to undergo spontaneous liquid–liquid phase separation, mouse neuromodulin is predicted as a protein with a very high droplet-driving potential.

Figure 5D shows a FuzDrop-generated multiplicity of binding modes (MBM) plot, indicating that the protein can bind to multiple partners, behaving differently in terms of its structure and function, either as an ordered or disordered state depending on the type of interaction and its environment. Values of MBM ≥ 0.65 suggest that the residues/regions are context dependent and are prone to engage in multiple interactions. The bar graph shows positions of MBM regions (residues 9–16 and 40–66) that have the potential to be engaged in multiple binding modes, assisting the phase separation.

The interactability of neuromodulin was evaluated using the STRING database. Figure 5E reveals that this protein is acting as the central node in the complex PPI network. We used a medium confidence threshold and a maximum limit of 500 interactors to generate this PPI network, which contains 145 nodes, with each node representing a protein, including neuromodulin, and 2925 edges (protein–protein interactions). This number of edges in the neuromodulin-centered network significantly exceeds the number of edges expected for a random set of proteins of the same size and degree distribution drawn from the genome (which is 616). The average node degree of this network is observed to be 39, indicating that the average connectivity of each protein in the network is very high, which is further supported by the average local clustering coefficient of 0.659, indicating a high tendency of nodes to cluster together. Finally, the observed p-value of <1.0 × 10⁻¹⁶ is indicative of the high significance of the generated data, suggesting that the PPI network is unlikely to be produced by chance.

Table 2. Functional enrichment of the PPI networks centered at the most disordered proteins: neuromodulin, Chmp4b, DnaJB6, Vps37B, and Wasl.

Protein	ID	Description	p-Value
Neuromodulin	Biological Process (Gene Ontology)
	GO:0022008	Neurogenesis	2.34 × 10−52
	GO:0007399	Nervous system development	2.34 × 10−52
	GO:0048699	Generation of neurons	1.74 × 10−48
	GO:0030182	Neuron differentiation	7.50 × 10−47
	GO:0048666	Neuron development	2.92 × 10−43
	Molecular Function (Gene Ontology)
	GO:0005515	Protein binding	1.35 × 10−25
	GO:0005488	Binding	6.32 × 10−18
	GO:0042802	Identical protein binding	1.38 × 10−10
	GO:0005102	Signaling receptor binding	5.78 × 10−10
	GO:0044877	Protein-containing complex binding	9.88 × 10−10
	Cellular Component (Gene Ontology)
	GO:0030424	Axon	2.24 × 10−59
	GO:0036477	Somatodendritic compartment	3.44 × 10−50
	GO:0043005	Neuron projection	3.44 × 10−50
	GO:0044297	Cell body	2.06 × 10−48
	GO:0043025	Neuronal cell body	2.87 × 10−46
Chmp4b	Biological Process (Gene Ontology)
	GO:0043162	Ubiquitin-dependent protein catabolic process via the multivesicular body sorting pathway	6.73 × 10−43
	GO:0071985	Multivesicular body sorting pathway	1.62 × 10−33
	GO:0007034	Vacuolar transport	2.43 × 10−33
	GO:0032509	Endosome transport via multivesicular body sorting pathway	8.11 × 10−33
	GO:0045324	Late endosome to vacuole transport	5.55 × 10−32
	Molecular Function (Gene Ontology)
	GO:0005212	Structural constituent of eye lens	4.36 × 10−20
	GO:0005198	Structural molecule activity	1.15 × 10−18
	GO:0005200	Structural constituent of cytoskeleton	8.50 × 10−15
	GO:0005525	GTP binding	3.97 × 10−09
	GO:0044389	Ubiquitin-like protein ligase binding	4.28 × 10−07
	Cellular Component (Gene Ontology)
	GO:0036452	ESCRT complex	3.59 × 10−45
	GO:0005768	Endosome	7.27 × 10−29
	GO:0010008	Endosome membrane	3.37 × 10−28
	GO:0005770	Late endosome	5.21 × 10−27
	GO:0031902	Late endosome membrane	1.09 × 10−26
DnaJB6	Biological Process (Gene Ontology)
	GO:0006457	Protein folding	6.61 × 10−43
	GO:0061077	Chaperone-mediated protein folding	8.46 × 10−27
	GO:0042026	Protein refolding	1.27 × 10−25
	GO:0035966	Response to topologically incorrect protein	1.27 × 10−20
	GO:0006986	Response to unfolded protein	2.60 × 10−20
	Molecular Function (Gene Ontology)
	GO:0051082	Unfolded protein binding	1.35 × 10−40
	GO:0044183	Protein folding chaperone	3.88 × 10−38
	GO:0140662	ATP-dependent protein folding chaperone	2.63 × 10−31
	GO:0031072	Heat shock protein binding	6.85 × 10−24
	GO:0051087	Chaperone binding	1.15 × 10−18
	Cellular Component (Gene Ontology)
	GO:0005737	Cytoplasm	8.10 × 10−14
	GO:0101031	Chaperone complex	1.59 × 10−13
	GO:0005829	Cytosol	1.45 × 10−11
	GO:0005759	Mitochondrial matrix	2.49 × 10−10
	GO:0005739	Mitochondrion	4.02 × 10−09
Vps37B	Biological Process (Gene Ontology)
	GO:0043162	Ubiquitin-dependent protein catabolic process via the multivesicular body sorting pathway	1.09 × 10−52
	GO:0007034	Vacuolar transport	2.94 × 10−45
	GO:0071985	Multivesicular body sorting pathway	3.42 × 10−41
	GO:0032509	Endosome transport via multivesicular body sorting pathway	9.64 × 10−40
	GO:0045324	Late endosome to vacuole transport	2.43 × 10−38
	Molecular Function (Gene Ontology)
	GO:0043130	Ubiquitin binding	3.22 × 10−11
	GO:0005515	Protein binding	2.21 × 10−06
	GO:0031386	Protein tag	4.31 × 10−05
	GO:0019904	Protein domain-specific binding	4.37 × 10−05
	GO:0090541	MIT domain binding	0.00019
	Cellular Component (Gene Ontology)
	GO:0036452	ESCRT complex	9.78 × 10−58
	GO:0010008	Endosome membrane	8.71 × 10−42
	GO:0031902	Late endosome membrane	7.17 × 10−40
	GO:0005770	Late endosome	3.41 × 10−39
	GO:0005768	Endosome	1.41 × 10−37
Wasl	Biological Process (Gene Ontology)
	GO:0030029	Actin filament-based process	6.73 × 10−119
	GO:0030036	Actin cytoskeleton organization	1.35 × 10−117
	GO:0007010	Cytoskeleton organization	2.58 × 10−96
	GO:0032956	Regulation of actin cytoskeleton organization	1.33 × 10−91
	GO:0032970	Regulation of actin filament-based process	2.26 × 10−90
	Molecular Function (Gene Ontology)
	GO:0003779	Actin binding	1.01 × 10−71
	GO:0008092	Cytoskeletal protein binding	1.68 × 10−71
	GO:0005515	Protein binding	7.97 × 10−48
	GO:0051015	Actin filament binding	3.00 × 10−40
	GO:0044877	Protein-containing complex binding	3.18 × 10−34
	Cellular Component (Gene Ontology)
	GO:0005856	Cytoskeleton	4.13 × 10−75
	GO:0031252	Cell leading edge	2.82 × 10−74
	GO:0015629	Actin cytoskeleton	3.67 × 10−73
	GO:0042995	Cell projection	2.16 × 10−63
	GO:0030027	Lamellipodium	1.97 × 10−55

lists the most enriched biological processes, molecular functions, and cellular components of the members of the neuromodulin-centered PPI network.

Figure 5F illustrates the 3D structure of the protein predicted by AlphaFold. Since disordered proteins or protein regions do not have single structures but represent highly dynamic conformational ensembles, they cannot be predicted by AlphaFold and are characterized by very low p_LDDT scores. In fact, based on the results of the CAID2 experiment, it has been concluded that AlphaFold2-based disorder predictors are better at detecting absence of order rather than detecting disordered regions [36]. Most of the predicted structure of our protein has low confidence scores and would be in disordered form when not interacting with the partners. In short, most of the protein would be unstructured in isolation, as the average per-residue model confidence score p_LDDT is 55.78 for this protein. The only high-confidence structural element of this protein is the blue α-helical region (residues 27–52). However, single α-helix cannot exist in isolation and is likely to be induced by binding to specific partner(s). In line with these considerations, this helical region corresponds to the IQ motif responsible for calmodulin binding.

Finally, we looked at the localization of ELMs (short functional motifs) within the various regions found in neuromodulin. The results of this analysis are summarized in

Table 3. Localization of ELMs (eukaryotic linear motifs) within the droplet-promoting regions, aggregation hotspots and MoRFs of mouse neuromodulin (UniProt ID: P06837). For additional information, see Supplementary Table S8.

Region Type	Region Range	ELM ID	Position
Droplet-promoting region	52–277	LIG_PDZ_Class_3	222–227
		LIG_WD40_WDR5_VDV_2	219–222
			218–222
			215–222
			155–161
			154–161
			133–137
			132–137
			131–137
			96–99
			95–99
			64–66
			58–64
		MOD_GlcNHglycan	209–212
			132–135
			127–130
			85–88
			84–88
		MOD_SUMO_rev_2	203–207
			200–207
			198–207
			196–201
			193–201
			192–201
			191–201
			154–159
			149–159
			122–126
			118–126
		CLV_C14_Caspase3-7	197–201
		DOC_USP7_MATH_1	207–211
			190–194
			119–123
		MOD_CK2_1	190–196
			142–148
		MOD_GSK3_1	186–193
			135–142
		MOD_PIKK_1	190–196
		LIG_TRAF6_MATH_1	184–192
		DOC_WW_Pin1_4	169–174
			139–144
			93–98
		MOD_ProDKin_1	169–175
			139–145
			93–99
		DOC_USP7_UBL2_3	153–157
		MOD_SUMO_for_1	152–155
			97–100
			25–28
		MOD_CK1_1	142–148
			128–134
			86–92
		LIG_BIR_III_2	118–122
		MOD_Plk_2-3	107–113
		MOD_CDK_SPK_2	93–98
MoRF	102–109	MOD_Plk_2-3	107–113
MoRF	58–81	LIG_WD40_WDR5_VDV_2	58–64
			63–66
Aggregation hotspot	52–66	LIG_WD40_WDR5_VDV_2	58–64
			63–66
MoRF	1–9	LIG_UBA3_1	1–9
		LIG_FHA_1	6–12
		MOD_PKA_2	5–11
		CLV_NRD_NRD_1	6–8
		CLV_PCSK_KEX2_1	6–8
		TRG_ER_diArg_1	5–7
		DEG_Nend_Nbox_1	1–3

.

The data reported in this section indicate that neuromodulin is characterized by a high level of intrinsic disorder with strong functional potential.

3.3. Global PPI Networks Analysis of the Most Disordered Mouse Proteins Found in the Rabies Virus

Next, we looked at the interconnectivity of the members of a group of the 11 most disordered mouse proteins found in RABV particles. The results of this analysis are shown in Figure 6. When this set was analyzed using STRING, using a medium confidence of 0.4 for the minimum required interaction score, these proteins were not linked in a single network but formed two disconnected networks consisting of six and three proteins, with two proteins, vesicle-associated membrane protein 3 (Vamp3) and neuromodulin (Gap43), being the loners (see Figure 6A). Although 11 proteins were connected through 8 interactions within this disjoined network (defining the low node degree of 1.45), they still had more interactions among themselves than what would be expected for a random set of proteins of the same size and degree distribution drawn from the genome (1). When the confidence of the minimum required interaction score was decreased to 0.15 (low confidence), all 11 proteins became engaged in interactions and formed a single PPI network with 25 edges and an average node degree of 4.55 (see Figure 6B).

Table 4. Functional enrichment of the intra-set PPI network of the 11 most disordered mouse proteins found in RABV particles.

ID	Description	p-Value
Biological Process (Gene Ontology)
GO:0008104	Protein localization	0.00017
GO:0051641	Cellular localization	0.00017
GO:0019076	Viral release from host cell	0.0011
GO:0043162	Ubiquitin-dependent protein catabolic process via the multivesicular body sorting pathway	0.0013
GO:0016192	Vesicle-mediated transport	0.0014
Molecular Function (Gene Ontology)
GO:0005515	Protein binding	0.0470
Cellular Component (Gene Ontology)
GO:0036452	ESCRT complex	0.00071
GO:0031410	Cytoplasmic vesicle	0.0011
GO:0005768	Endosome	0.0014
GO:0005829	Cytosol	0.0048
GO:0000813	ESCRT I complex	0.0052

lists the most enriched biological processes, molecular functions, and cellular components of the members of this PPI network.

We also checked the set-centered interactivity of these 11 most disordered mouse proteins found in RABV particles. To this end, we used the multiple proteins search option on the STRING platform and selected a custom value of 500 maximum first-shell interactions (note that the number of interactors in STRING is limited to 500) and high confidence level (minimum required interaction score of 0.7). Using these settings resulted in the generation of a well-connected PPI network containing 281 proteins involved in 3918 interactions (see Figure 7). The average node degree of this network is 20.6, and its average local clustering coefficient is 0.618.

The application of k-means clustering (which is an unsupervised machine learning algorithm designed to group the unlabeled datasets into different clusters, thereby dividing a set of data into a number of groups depending on how similar and different they are to one another) to this PPI network centered at the 11 most disordered mouse proteins found in RABV particles revealed that the set of 381 proteins can be split into 3 clusters.

The biggest cluster includes 263 proteins involved in 2312 interactions (see red circles in Figure 7). This sub-network includes many proteins from the regulation of the actin cytoskeleton pathway (KEGG pathway ID: mmu04810, p-value = 1.98 × 10⁻³⁹). The average node degree of this network is 17.6, and its average local clustering coefficient is 0.587.

The second cluster includes 60 proteins involved in 481 interactions (see green circles in Figure 7) and is mostly related to the endocytosis pathway (KEGG pathway ID: mmu04144; p-value = 1.50 × 10⁻⁴⁴). This sub-network is characterized by an average node degree of 16 and average local clustering coefficient of 0.813.

In the third cluster, there are 58 proteins connected by 875 interactions (see blue circles in Figure 7). Most of the proteins in this cluster are related to the SNARE interactions in the vesicular transport pathway (KEGG pathway ID: mmu04130l p-value = 6.94 × 10⁻⁶²). This sub-network is characterized by an average node degree of 30.2, and the average local clustering coefficient of 0.868.

Table 5. Functional enrichment of the PPI network centered at the 11 most disordered mouse proteins found in the RABV particle, as well as its three clusters.

Network	ID	Description	p-Value
Global network	Biological Process (Gene Ontology)
	GO:0016192	Vesicle-mediated transport	3.41 × 10−80
	GO:0051641	Cellular localization	3.41 × 10−80
	GO:0051128	Regulation of cellular component organization	3.22 × 10−76
	GO:0008104	Protein localization	2.85 × 10−71
	GO:0051649	Establishment of localization in cell	1.38 × 10−65
	Molecular Function (Gene Ontology)
	GO:0005515	Protein binding	1.34 × 10−77
	GO:0019904	Protein domain-specific binding	8.56 × 10−46
	GO:0008092	Cytoskeletal protein binding	1.06 × 10−39
	GO:0005484	SNAP binding	2.99 × 10−38
	GO:0005488	Binding	5.52 × 10−36
	Cellular Component (Gene Ontology)
	GO:0031982	Vesicle	1.74 × 10−73
	GO:0031410	Cytoplasmic vesicle	6.16 × 10−66
	GO:0030054	Cell junction	6.16 × 10−66
	GO:0042995	Cell projection	3.74 × 10−64
	GO:0005488	Cytoplasm	3.74 × 10−64
Cluster 1	Biological Process (Gene Ontology)
	GO:0005515	Protein binding	5.60 × 10−58
	GO:0008092	Cytoskeletal protein binding	7.30 × 10−44
	GO:0003779	Actin binding	6.12 × 10−43
	GO:0019904	Protein domain-specific binding	1.31 × 10−40
	GO:0019899	Enzyme binding	1.22 × 10−38
	Molecular Function (Gene Ontology)
	GO:0051128	Regulation of cellular component organization	1.80 × 10−65
	GO:0030029	Actin filament-based process	1.10 × 10−53
	GO:0044087	Regulation of cellular component biogenesis	1.35 × 10−52
	GO:0030036	Actin cytoskeleton organization	5.31 × 10−50
	GO:0007010	Cytoskeleton organization	5.51 × 10−48
	Cellular Component (Gene Ontology)
	GO:0042995	Cell projection	1.61 × 10−62
	GO:0120025	Plasma membrane bounded cell projection	2.96 × 10−59
	GO:0005829	Cytosol	1.36 × 10−51
	GO:0030054	Cell junction	2.26 × 10−50
	GO:0005856	Cytoskeleton	6.03 × 10−48
Cluster 2	Biological Process (Gene Ontology)
	GO:0043162	Ubiquitin-dependent protein catabolic process via the multivesicular body sorting pathway	5.74 × 10−45
	GO:0007034	Vacuolar transport	4.66 × 10−41
	GO:0071985	Multivesicular body sorting pathway	4.84 × 10−37
	GO:0046755	Viral budding	2.25 × 10−36
	GO:0032509	Endosome transport via multivesicular body sorting pathway	5.26 × 10−36
	Molecular Function (Gene Ontology)
	GO:0043130	Ubiquitin binding	6.09 × 10−08
	GO:0005515	Protein binding	2.69 × 10−06
	GO:0004459	L-lactate dehydrogenase activity	4.33 × 10−06
	GO:0048306	Calcium-dependent protein binding	9.28 × 10−06
	GO:0005543	Phospholipid binding	5.52 × 10−05
	Cellular Component (Gene Ontology)
	GO:0036452	ESCRT complex	1.23 × 10−49
	GO:0010008	Endosome membrane	6.18 × 10−43
	GO:0031902	Late endosome membrane	1.39 × 10−38
	GO:0005768	Endosome	8.24 × 10−38
	GO:0030659	Cytoplasmic vesicle membrane	1.80 × 10−34
Cluster 3	Biological Process (Gene Ontology)
	GO:0016192	Vesicle-mediated transport	4.17 × 10−59
	GO:0061025	Membrane fusion	6.79 × 10−59
	GO:0006906	Vesicle fusion	1.63 × 10−53
	GO:0051649	Establishment of localization in cell	5.79 × 10−46
	GO:0016050	Vesicle organization	7.23 × 10−44
	Molecular Function (Gene Ontology)
	GO:0000149	SNARE binding	1.15 × 10−72
	GO:0005484	SNAP receptor activity	5.11 × 10−70
	GO:0030674	Protein-macromolecule adaptor activity	6.13 × 10−46
	GO:0019905	Syntaxin binding	4.45 × 10−43
	GO:0017075	Syntaxin-1 binding	7.64 × 10−16
	Cellular Component (Gene Ontology)
	GO:0031201	SNARE complex	2.90 × 10−91
	GO:0030133	Transport vesicle	4.17 × 10−43
	GO:0070382	Exocytic vesicle	3.76 × 10−41
	GO:0008021	Synaptic vesicle	2.50 × 10−40
	GO:0098796	Membrane protein complex	4.14 × 10−40

represents the most enriched biological processes, biological functions, and cellular components of the members of this PPI network and of each of its clusters.

To get a hint on the prevalence of intrinsic disorder in host interactors of mouse proteins entrapped in the RABV particle, we applied the RIDAO platform to proteins in the aforementioned clusters. The results of this analysis are summarized in Figure 8, which clearly shows that all these protein sets are characterized by the presence of significant levels of intrinsic disorder. In fact, in all these clusters, proteins classified as disordered based on their PPIDR values exceeding the 30% threshold constitute the vast majority, and 41% to 55% are expected to be highly disordered (based on their positions within the red segment of Figure 8A). Furthermore, from 47.5% to 65.5% of proteins in these clusters are located outside the quadrant Q1 and therefore contain significant levels of intrinsic disorder (see Figure 8B).

3.4. The Roles of Intrinsically Disordered Host Proteins in Viral Immune Evasion and Pathogenesis Enhancement

Now, we are going to focus on how the rabies virus exploited the structural chaos associated with the entrapped host proteins (i.e., their high intrinsic disorder status) to its own benefit. The incorporation of host proteins within viral particles helps them evade immunity and antiviral resistance, and eventually results in the enhancement of viral pathogenesis [21]. These functions can be associated with the intrinsic disorder present in these entrapped host proteins. The viruses with incorporated host proteins are less recognizable by the host immune system, and the antigens of the incorporated host proteins can mask the viral antigens normally recognizable through the immune system [21,104]. Because of the masking of the viral antigens by host proteins, host antibodies cannot efficiently detect the viral particles to successfully eliminate them [21,105]. Furthermore, because of this mimicry, the immune system can get confused, and the effort of finding the mimicking viral particles can sometimes trigger autoimmunity, where the host immune system cells start attacking their own healthy cells, leading to the tissue damage [105,106,107].

Viruses can also exploit host receptors to enter various host cells effectively, which not only enhances their transmission rate but also increases the range of cell types a virus can infect [108], and the addition of the host proteins to the viral particles can enhance this ability of the virus [109]. Molecular mimicry also helps viruses to evade antiviral drugs, making the development of antiviral drugs more complicated. Because these drugs are designed to attack unique viral particles without harming healthy host cells [110], the incorporation of host proteins in viral particles can make it difficult for the antiviral drugs to distinguish between the host cells and viral particles, leading to the increased toxicity and side effects and less effective therapeutic targeting.

In the context of our study, when we add the intrinsic disorder of these entrapped proteins to the picture, we can say that the scenario becomes even more complicated. As we have mentioned in the Introduction, IDPs/IDRs lack a 3D structure and are highly flexible and adaptable. They can bind to a variety of partners [111] and can facilitate the interaction of viral particles with a wide array of host cells, facilitating viral entry, replication, and overall pathogenesis. The flexible nature of IDPs can also assist viruses to evolve and become more adaptable to their environment. Viruses can manipulate the properties of intrinsically disordered host proteins to escape the environmental pressure created by the host immune system, also making therapeutic strategies more complex. We can hypothesize that these IDPs/IDRs are providing numerous additional functional and evolutionary benefits to the virus.

In short, we can target the interactions of host IDPs/IDRs with the virus to disrupt the viral life cycle. Understanding the roles of host IDPs/IDRs in the life cycle of viruses can open new lines of research to develop more effective antiviral therapeutic strategies.

4. Conclusions

The bioinformatics analysis performed on the host proteins incorporated within the rabies viruses offers significant findings regarding the role of host intrinsic disorder in the life cycles of rabies viruses.

Out of 47 host proteins that are entrapped in the viral particles, most were predicted as noticeably disordered. In fact, 40.4% of these proteins were predicted as moderately disordered (are characterized by 10% ≤ PPIDR_VSL2 < 30% and/or 0.15 ≤ ADS_VSL2 < 0.5), whereas 55.4% of the 47 host proteins were anticipated to be highly disordered (PPDR _VSL2 ≥ 30% and/or ADS _VSL2 ≥ 0.5). Based on the results of the PONDR^® VSL2-based disorder analysis, 11 proteins were predicted to be mostly disordered, since they were shown to have PPIDR values exceeding 50% and ADS values exceeding 0.5. A detailed computational analysis of the five most disordered host proteins entrapped in the RABV particles, Neuromodulin, Chmp4b, DnaJB6, Vps37B, and Wasl, revealed several important roles that intrinsic disorder can play in the functionality of these proteins. It is also very likely that intrinsic disorder of the host proteins entrapped in the viral particles could be playing essential roles in the pathogenicity of the viruses, modulating their mechanisms of immune evasion, promoting the development of antiviral drug resistance, and thereby contributing to viral adaptability and evolution.

This study has several obvious limitations. For example, it is not clear at the moment if all the RABV particles produced by an infected cell contain similar quantities of entrapped host proteins or whether the arsenal of the host proteins incorporated into the RABV particles can be influenced by the type of infected cells, where the virions are produced. It is also not clear how different sets of the virion-entrapped host proteins would be in different species infected by the RABV. Another important question is related to understanding the roles of such entrapped host proteins in the inter-species RABV infectivity (e.g., how the dog proteins entrapped in the RABV virions produced in an infected dog would affect a human bitten by said RABV-infected dog). Furthermore, it is not clear how one can use the virion-associated host proteins to experimentally infer the virus–host protein interactions in infected cells. Therefore, subsequent analyses are required to better understand the roles of intrinsic disorder of the host proteins entrapped in the RABV in the virus’s life cycle and pathogenicity.