TDP-43 protein interactome informs about perturbed canonical pathways and may help develop personalized medicine approaches for patients with TDP-43 pathology

Abstract

Transactive response DNA binding protein of 43 kDa (TDP-43) pathology is a common proteinopathy observed among a broad spectrum of patients with neurodegenerative disease, regardless of the mutation. This suggests that protein–protein interactions of TDP-43 with other proteins may in part be responsible for the pathology. To gain better insights, we investigated TDP-43-binding proteins in each domain and correlated these interactions with canonical pathways. These investigations revealed key cellular events that are involved and are important at each domain and suggested previously identified compounds to modulate key aspects of these canonical pathways. Our approach proposes that personalized medicine approaches, which focus on perturbed cellular mechanisms would be feasible in the near future.

Introduction

Proteins are one of the major building blocks of cells, and protein–protein interactions (PPIs) are in part responsible for many cellular functions and events. Even though proteins interact with each other, their interaction is precisely regulated. Most proteins have distinct domains, each with unique characteristics. Interestingly, proteins do not arbitrarily interact with any place on the other protein, but bind to specific regions and/or domains. The high-level precision in protein interactions offers mechanistic insights, especially for diseases and pathologies that stem from proteinopathies.

Transactivation DNA binding protein of 43 kDa (TDP-43) pathology is one of the most common proteinopathies detected in patients. It is present vastly in the brains of patients diagnosed with amyotrophic lateral sclerosis (ALS), ALS with frontotemporal dementia (ALS/FTLD),^1–4 Alzheimer’s disease (AD), and Parkinson’s disease (PD).⁵ Although mutations in the gene that codes for TDP-43 are detected in some patients with ALS, the TDP-43 pathology is broadly present even in the absence of mutations and in patients with sporadic ALS, who do not have a known genetic cause. This suggests that the pathology cannot be explained solely by gene mutations. We hypothesize that detailed investigation of protein interactome domains and protein binding landscape analyses may bring a mechanistic insight to proteinopathies. Therefore, it could be possible that altered or perturbed PPI dynamics are in part responsible for protein accumulation, which is the foundational problem observed in the brains of patients with ALS, ALS/FTLD, AD, and PD, among others.

TDP-43 contains 414 amino acids and is encoded by the TARDBP gene, located on chromosome 1.⁶ TDP-43 protein has numerous functions in the cell, including a multifaceted role within RNA metabolism, mRNA expression, transcription, splicing, and stability, all of which are regulated by its interaction with thousands of mRNA transcripts and many other proteins.^7–10 Additionally, significant evidence links TDP-43 activity to microRNA biogenesis and the regulation of cellular stress responses,^11–15 which have significant implications for neurodegenerative disease onset and progression when dysregulated.^16,17

The protein structure is exceptionally important for determining the mode of interaction and functionality; therefore, the structure of TDP-43 is critical in understanding its function. The TDP-43 protein has numerous well-defined domains, including a nuclear localization signal (NLS) domain, two RNA recognition motifs (RRM1 and RRM2), and a glycine-rich domain (GRD) located at the C-terminus, all of which contribute to distinct aspects of its function.¹⁸ Mutations in a given domain, would thus be expected to have implications that are mostly related to the function attributed to that particular domain.

The NLS plays a critical role in TDP-43 function as it is the site that is required for shuttling TDP-43 from the cytoplasm to nucleus and vice versa. Mutations or deletions within this domain are thought to be mainly responsible for the cytoplasmic accumulation of TDP-43, a characteristic of TDP-43 proteinopathy.^19,20 The A90V mutation, which is a prominent canonical mutation in the NLS domain, has demonstrated functional defect in both patient cases and cultured cell lines,^21,22 and in vitro site-directed mutation analyses within the NLS domain have demonstrated cytoplasmic relocalization and aggregation.²⁰

The two RRMs of TDP-43 demonstrate significant structural similarities, with both containing two five-stranded beta sheets stacked against two alpha helices. The RRMs are mostly involved in RNA/DNA molecule binding, and both also have distinct functions. RRM1, and not RRM2, is essential and sufficient for proper nucleic acid binding with its high affinity for the UG-rich sequence.^23,24 Additionally, RRM1 is uniquely related to ATP binding, and mutations in this domain have demonstrated disruption of this function.^18,25 The D169G mutation disrupts this function and prevents ATP binding, which further links TDP-43 pathology to metabolic disruptions. Additional investigation of this mutation demonstrated translocation to the cytoplasm and impairment of TDP-43 assembly into nuclear enriched abundant transcript 1- (NEAT-1) positive nuclear bodies. D169G mutation also leads to the formation of stress granules.²⁶ Currently, there are no identified patient mutations within the RRM2 domain, but there has been significant research in understanding the domain’s impact on protein function. RRM2 plays a supplementary role to RRM1 in RNA binding and a regulatory role in sequence specificity of TDP-43 substrate binding.^22,27 Therefore, local disruption of this region would still have important and consequential implications. The RRMs work together to actively bind thousands of different mRNA transcripts in 3′ untranslated regions, including TDP-43′s own mRNA as an autoregulation mechanism of cellular concentration.⁷

The GRD, which is located in the C-terminal domain, represents the largest and most critical domain for TDP-43 and its interaction with other proteins.^28,29 The exterior localization within the protein structure, which is illustrated in Figure 1, qualitatively explains the significance of this domain in modulating PPIs. The importance of this domain for TDP-43 function is further demonstrated by the analysis of mutations detected in patients. The vast majority of ALS-associated mutations are within this domain, including the most common mutations such as A382T and M337V.⁶ A382T is especially prevalent in patients with familial ALS and results in TDP-43 accumulation within the cytoplasm. This mutation and others within the GRD demonstrate disruption of nuclear function including transcriptional regulation, mRNA metabolism and transport, and microRNA biogenesis.^30–33 The mutations in this domain have broad implications with significant phenotypic heterogeneity among patients.³⁴

FIGURE 1.

The domain structure of TDP-43 protein and its 3D conformation, showing the location of each canonical domain. Additionally, a linear representation of the TDP-43 protein’s amino acid sequence is shown.

PPIs are a critical component of almost every important biological process.^35,36 Therefore, to determine which biological processes would be most affected when mutations or disruptions are present in distinct domains, we investigated the PPIs of each domain. Recent studies utilizing large data management applications in conjunction with canonical research of protein interaction assays, have provided insights into canonical pathways and downstream events that are primarily manifested by the interaction in that given domain.^37,38 In our study, we found that the proteins identified as TDP-43-binding proteins did not equally interact with each domain and that there was a level of selectivity in their binding preference, further affecting the cellular and downstream signaling cascade. These binding partners therefore represent hubs of key interactions and help determine their contribution to function.

We expanded the global PPI analysis by utilizing a domain-specific approach that computationally modeled each domain’s interaction partner from experimentally observed interaction data. The computational models provide predictive high-resolution protein complex structures, which help reveal protein interaction maps.^39,40 The variation in composition of PPIs across domains demonstrates a potential source of phenotypic heterogeneity in patients.

TDP-43 protein interaction partners

TDP-43 has a well characterized and critical role in RNA processing. The vast majority of cellular functions require the interaction of proteins with their distinct set of binding partners. Therefore, proteins do not influence overarching cellular events independently, but rather require interaction with other proteins to demonstrate their full function. Together, these interactions lead to the modulation of key canonical pathways and control different cellular events. The associated binding partners, interactome domains, and upstream regulators of a specific protein help reveal key events primarily associated with the protein and its domains.

Despite the prevalence of TDP-43 pathology in neurodegenerative disease, we still do not have a full understanding of the underlying heterogeneous mechanisms that are perturbed in different patients. Therefore, we have begun to identify important downstream cellular events and gain mechanistic insights into TDP-43 proteinopathy.

Previous publications and online public databases have reported binding partners of TDP-43 protein. Each binding partner was identified with at least one publication that confirmed direct binding and interaction (Table S1 in the supplementary material online). The initial experimentally determined binding partners for TDP-43 protein (n = 323) were subjected to further refinement to reduce the chance of false positives and to increase the association with TDP-43 pathology with respect to ALS. In that regard, we selected the proteins that bind to at least two more ALS-related proteins.³⁷ This refinement resulted in the identification of 147 proteins that were shown to bind to TDP-43 with experimental findings and that also interact with two more proteins that belong to the ALS protein landscape, as previously determined.³⁷ The comprehensive list of ALS-specific interaction partners (n = 147) was then further filtered by the feasibility of predictive docking analysis, which was the result of the availability of three-dimensional (3D) predicted protein structures. The vast majority of interaction partners (n = 143) had predicted 3D protein structures readily available and were retrieved from Alpha Fold Protein Structure database.^41,42 This database provides high accuracy predictive models for protein structures based on the amino acid sequences.

Predictive protein–protein docking

In the history of medical genetics, there have been many single mutations in a single gene identified as ‘the cause’ of a disease, although the effects of distinct genetic mutations on their encoded proteins vary significantly as a function of their location. The mutation is thought to impair the ability of the native protein to properly interact with its binding partners and thus fail to initiate or modulate the cellular events that it is responsible for or take part in.

Mutations in TDP-43 demonstrate local changes in structure and suggest the disruption of binding partner interactions. Even though there is no sufficient information on how the 3D structure of the protein changes with a given mutation, it is possible to run predictive docking analysis between TDP-43 and its binding partners. High Ambiguity Driven protein–protein DOCKing (HADDOCK v2.4) was utilized to model these interactions.^43,44 The utilization of the docking procedure was facilitated via an open access web server interface provided by the Bonvin lab (https://wenmr.science.uu.nl/haddock2.4/). Additionally, the web server allowed for easy modification of the docking parameters that drove the analysis. The overarching process used ambiguous interaction restraints with various other possible restraint parameters including specific distance restraints. The majority of the default docking parameters were kept throughout this study’s docking procedure, including the removal of buried and inaccessible residues and nonpolar hydrogen atoms to facilitate greater accuracy on potentially active residues and faster computational analysis. The modifications to the default settings customized the docking procedure to the needs of this specific study. These included preventing both the automatic definition of ‘passive residues’ and the random exclusion of ambiguous restraints. Both of these restrictions allowed for more thorough analysis of the potentially active residues within the entire protein. Additionally, the HADDOCK procedure requires the determination of active residues in the docking procedure, but with the nature of this analysis, the exact active residues are unknown. Therefore, instead of directly defining the residues, we utilized randomized patches of active residues throughout the entire 3D structure to give equal opportunity to every docking confirmation. All parameters were kept constant for the docking of TDP-43 with all 143 binding partners.

After full definition of the input as well as the docking parameters, the HADDOCK analysis was run with the input of two protein data bank files, which contained the 3D structure of TDP-43 and the respective binding partner for that PPI analysis. The initial stage of the docking procedure consisted of rigid body docking, which randomizes the starting orientations of TDP-43 and its binding partner. Following the starting position randomization, the molecules were treated as rigid bodies and underwent conformational analysis for energy minimization, which specifically analyzes the electrostatic and van der Waals potentials. By default settings, 1000 different models were analyzed with each having the energy minimization process repeated five times. Then the top 200 solutions from all of the different models of the docking were subjected to the second stage of docking analysis, semi-flexible simulated annealing (SA) in torsion angle space, which consisted of high temperature rigid body search, rigid body SA, semi-flexible SA with flexible side chains at the interface, and semi-flexible SA with a fully flexible interface (both backbone and side chains).

HADDOCK automatically defines the semi-flexible regions by considering all residues that are within 5 Å of another molecule. During this stage of the analysis, each protein complex model was optimized and ascribed a HADDOCK score. This score value represented a weighted sum of energies and scoring terms that allowed for the ranking of models. The docking procedure output the top ranked complex models, which represents the lowest energy conformation of the dimerization complex between the TDP-43 and the respective binding partner in that analysis. Each binding partner (n = 143) was subjected to this same dimerization analysis and the top ranked complex model of each binding partner was saved for further analyses.

Each of the top ranked complex model’s protein data bank files were subsequently uploaded as an input into PyMOL, which is an open source molecular visualization system. This system allowed for full visualization of the 3D structure and orientation of the protein complexes between TDP-43 and its binding partners. With the predicted conformational structure determined via the HADDOCK analysis, the individual residues on the TDP-43 protein that are interacting within each complex were identified with PyMOL’s interface residue function. This function takes the area of the entire complex and then splits it into the two respective amino acid chains. The surface area of each chain was calculated, and the difference between the complex based areas and the chain only based areas was compared with the desired cutoffs to determine the validity of an interface residue. The function highlighted all of the interacting residues within the respective complex and their location within the TDP-43 amino acid sequence. This interaction residue determination was repeated for all binding partner complexes that were output from HADDOCK, and the location active residues within TDP-43 were compiled across the entire amino acid sequence. Therefore, each amino acid in the sequence had a recorded value for the number of predicted interactions it has with the binding partners. Additionally, the identity of the binding partner proteins that showed interactions with residues within the known domains of TDP-43 were also revealed and recorded.

The PPIs and subsequent analyses of the complexes yielded the predicted active amino acid residues. Each of these ‘active’ amino acid residues represented a predicted location of a key interaction for the protein’s function. The number of interactions associated with each amino acid residue were summed together for all analyzed binding partners and therefore demonstrated the importance of different amino acids in ‘hot spots’ throughout the sequence. Therefore, mutations in patients that fall within these ‘hot spots’ may have a more drastic impact on protein function. A comparison of the number of predicted interactions across the entire amino acid sequence is graphically represented in the construction of a heat map of the TDP-43 protein structure (Figure 2). This heat map demonstrates the especially high number of interactions associated with RRM1 and the GRD. The GRD is especially important, with every amino acid residue in this domain interacting with at least one binding partner. Additionally, the GRD contains the vast majority of causative mutations in patients with ALS and patients with TDP-43 pathology.⁴⁵ The sum total of all interactions for each amino acid residue, the identity of each binding partner, was recorded. This helped determine the complete list of binding partners that are predicted to be associated with each TDP-43 domain (Figure 2c).

FIGURE 2.

(a) Graphical representation of the sum of the predicted interaction residues at each amino acid position in the TDP-43 protein. (b) Linear representation of TDP-43 amino acid sequence and distinct domains. Additionally, an overlay of previously identified patient mutations is shown. (c) List of interaction partners that show a predicted association with that respective domain.

Domain-specific protein interaction analysis

With identification of the binding partners for each domain, we utilized a domain-specific PPI approach. Large data management toolboxes such as Ingenuity Pathway Analysis (IPA) serve as excellent asset for the analysis of large data sets, such as protein binding partner interactions. The utilization of these tools requires caution to avoid false negatives or false-positive affirmations due to their inclusion of both experimental and curated findings from a wide variety of resources. Despite these challenges, numerous recent studies have demonstrated strong reputable results with their utilization.^37,38,46 To avoid bias, we utilized a stringent inclusion criterion, which limited the interactions to experimentally observe direct interactions in the neuronal systems. This strong criterion provides confidence in input parameters and in validation of the data-driven hypothesis. IPA uses an array of statistical analyses to determine whether the analyzed data set has significant coverage with any of the previously determined canonical pathways, cellular events, PPI domains, and pathways (https://www.qiagenbioinformatics.com/products/ingenuity-pathway-analysis accessed on 9 September 2022). The statistical analyses within the program mainly use Fisher’s exact tests, which outputs a significance P-value that represents the likelihood that random chance would yield the same association of the experimental group and a given pathway. Therefore, the P-value is dependent on the statistical null model, whereas the Fisher’s exact test determines the likelihood of random chance and shows the significance of the association. The experimental groups were formed by taking the list of binding partners predicted to interact with each of the known domains for the TDP-43 (RRM1, RRM2, GRD, and NLS).

The analyses also output an overlap ratio value, which represents the number of ALS-specific binding partners’ genes divided by the total number of genes in the canonical pathways. Therefore, the ratio is useful in determining the pathways that overlap with the predicted binding partners of each TDP-43 domain. The P-value measures the likelihood that the association between the binding partners and a given pathway is not due to a random chance. P < 0.05 is considered significant and demonstrates that the association cannot be explained due to randomness or ‘chance’. The calculation of the P-value is completed by considering: (1) the number of pathways/functions participating in the cellular event; (2) the total number of molecules associated with a given pathways; and (3) the number of binding partners in each experimental group for each domain.

GRD

The GRD was the interaction site of 95.1% of all PPIs. The majority of GRD-binding partners were localized in the nucleus, but there was also a large number of proteins found within the cytoplasm (Figure 3a). The circular representation of these proteins also demonstrated the key involvement of GRD interaction partners with translational and transcriptional regulation (Figure 3a). This strong connection to translation and transcription was furthered with the canonical pathways identified in this analysis (Figure 3b) such as the eukaryotic initiation factor 2 (eIF2) signaling pathway (P = 2.62E-18, 20/226; Figure 2c), cold shock domain E1 (CSDE1) signaling pathway (P = 3.06E-11, 9/56; Figure 5c), regulation of the eIF4 and p70S6K signaling pathways (P = 7.47E-09, 11/180; Figure 4c), and the microRNA biogenesis signaling pathway (P = 1.30E-07, 10/187; Figure 6c, Figure S1 in the supplementary material online). Interestingly, all of these identified canonical pathways have a direct relationship with RNA metabolism.

FIGURE 3.

Glycine rich domain (GRD) interaction partner analysis. (a) Circular representation of GRD binding partner’s association between type of location. (b) Bar graphs of canonical pathways associated with GRD interaction partners [−P(log value)] and their overlap (ratio, yellow line). (c) Image of eIF2 signaling pathway, representing the extent of GRD protein involvement.

FIGURE 5.

RNA recognition motif 2 (RRM2) interaction partner analysis. (a) Circular representation of RRM2 binding partner’s association between type of location. (b) Bar graphs of canonical pathways associated with RRM2 interaction partners [−P(log value)] and their overlap (ratio, yellow line). (c) Image of the CSDE1 signaling pathway, representing the extent of RRM2 protein involvement.

FIGURE 4.

RNA recognition motif 1 (RRM1) interaction partner analysis. (a) Circular representation of RRM1 binding partner’s association between type of location. (b) Bar graphs of canonical pathways associated with RRM1 interaction partners [−P(log value)] and their overlap (ratio, yellow line). (c) Image of regulation of the eIF4 and p70S6K signaling pathways, representing the extent of RRM1 protein involvement.

FIGURE 6.

Nuclear localization signal (NLS) domain interaction partner analysis. (a) Circular representation of NLS binding partner’s association between type of location. (b) Bar graphs of canonical pathways associated with NLS interaction partners [−P(log value)] and their overlap (ratio, yellow line). (c) Image of microRNA biogenesis signaling pathway, representing the extent of NLS protein involvement.

RRM1

The RRM1 domain showed a high level of association with TDP-43 interaction partners, with 91 of 143 (63.6%) proteins having direct interactions in the RRM1 domain. The localization of RRM1 binding partners was almost evenly distributed between the nucleus and the cytoplasm (Figure 4a). The type of protein interacting with the RRM1 varied, and the major categories included transcription and translation regulators, as well as numerous enzymatic reactions (Figure 4a). The RRM1 binding partners and the canonical pathways significantly associated with them were analyzed (Figure 4b). The most significant canonical pathways were the eIF2 signaling pathway (P = 1.14E-09, 11/226; Figure 3c), regulation of the eIF4 and p70S6K signaling pathways (P = 3.20E-08, 9/180; Figure 4c), the CSDE1 signaling pathway (P = 6.46E-05, 9/56; Figure 5c), the microRNA biogenesis signaling pathway (P = 8.60E-07, 6/187; Figure 6c), and the oxidized GTP and dGTP detoxification pathways (P = 8.74, 2/4; Supplemental Figure S1).

RRM2

The RRM2 had interactions with 56 of 143 (39.2%) TDP-43 binding partners. Circular representation of the type and location of these proteins suggested that these interactions were different from other domains. They were mainly in the nucleus and cytoplasm (Figure 5a). The type of protein interacting with this domain had a very similar distribution to the RRM1 domain, which included a significant portion of enzymes as well as transcription and translation regulators (Figure 5a). The canonical pathways (Figure 5b) associated with the proteins interacting with this domain were mainly the eIF2 signaling pathway (P = 4.76E-08, 8/226; Figure 3c), CSDE1 signaling pathway (P = 1.89E-07, 5/56; Figure 5c), microRNA biogenesis signaling (P = 4.52E-06, 6/187; Figure 6c), regulation of eIF4 and p70S6K signaling (P = 5.95E-05, 5/180; Figure 4c), and cell cycle control and chromosomal replication pathways (P = 2.96, 3/56; Figure S1 in the supplementary material online).

NLS

The NLS had the fewest number of interactions with only 17 of 143 (11.9%) TDP-43 binding partners, possibly due to the small size of this region. The location of proteins associated with this domain differed from other domains because the vast majority were nuclear proteins and none were localized to the plasma membrane. Additionally, a much smaller proportion was from the cytoplasm compared with the other domains (Figure 6a), and the type of protein interacting with this domain also varied in comparison with other domains. Unlike other domains, the NLS interaction partners did not include translation regulators and had a significantly larger proportion of transcription regulators (Figure 6a). We observed numerous significant canonical pathways associated with these binding partners (Figure 6b). The most significantly associated canonical pathways were the microRNA biogenesis signaling pathway (P = 8.68E-06, 4/187; Figure 6c), CSDE1 signaling pathway (P = 7.84E-04, 2/56; Figure 5c), DNA double-strand break (DSB) repair by homologous recombination (P = 1.01E-02, 1/14; Figure S1 in the supplementary material online), Ran signaling pathway (P = 1.23E-02, 1/17; Figure S1), and differential regulation of cytokine production in macrophages and T helper cells by interleukin 17A (IL-17A) and IL-17F (P = 1.30E-02, 1/18; Figure S1).

Overall, our analysis represents a novel computational approach, which provides a potential understanding of cellular events that are primarily associated with distinct domains and TDP-43 binding partners. We utilized a domain specific analysis of TDP-43, which included the GRD, RRM1 and RRM2, and an NLS. All domains were analyzed independently to gain insights into the canonical pathways associated with the predicted binding partners for each region.

There were numerous shared canonical pathways because some binding partners interact with multiple domains. There were also canonical pathways that are selectively involved with a given domain. We observed a large variation in the proportions of type and location for each domain’s interaction partners. For example, the NLS domain shows only a limited number of cytoplasmic proteins and has the majority of nuclear proteins, whereas other domains have a significantly higher proportion of cytoplasmic proteins. This variation highlights the fact that, despite canonical pathways being shared, the list of interaction partners for each pathway may differ. This is important because the outcome of a nonfunctional canonical pathway may be the same, but the reason why that canonical pathway becomes inactive may be different. This information is especially important as we begin to build personalized medicine approaches. Patients may show similar clinical outcomes, but the underlying causes and reasons for their disease-causing defects may be different, even within the same canonical pathway, and consequently one drug or one treatment strategy may not be effective for all patients. Therefore, knowing the gene that is mutated may not be enough and the assumption that all patients with mutations in the same gene have the same underlying problems may not be valid. Detailed and mechanism-focused drug discovery efforts for each patient is vital.

Canonical pathways that are common and unique

A logical correlation between canonical pathways identified in the analysis was the strong relationship to the cellular process of RNA metabolism mostly because of TDP-43′s known involvement in RNA metabolism. This observation was demonstrated well with the eIF2, CSDE1, microRNA biogenesis, and eIF4 and p70S6K signaling pathways, all of which are related to RNA processing. Translational dysregulation is a well-documented aspect of numerous ALS pathologies including TDP-43 with canonical reports of disruption in mRNA stabilization, transport, and translation.^47,48

The eIF2 signaling pathway is of special interest because it is the most significantly associated canonical pathway in the GRD, RRM1 and RRM2 domains. This pathway is a critical component of eukaryotic translation, specifically start codon recognition and selection.⁴⁹ The exact functionality of the eIF2 is dependent on the phosphorylation status of its alpha subunit and the state of the bound GTP molecule. Phosphorylation of the alpha complex inhibits the exchange of GDP to GTP in eIF2, which reduces formation of the eIF2 ternary complex and therefore reduces global translation.^50–53 The pathway is especially important for cellular stress response as the kinases responsible for eIF2 phosphorylation are stress-activated.⁵⁴ In addition, despite a global reduction in translation, a specific subset of stress-specific mRNA transcripts are enhanced as a result of the phosphorylation.

Another aspect of eIF2 signaling pathway’s functionality is its involvement in stress granule formation, which further demonstrates the pathway’s strong association with the cellular stress response.^54–56 Studies have demonstrated the direct relationship of TDP-43 protein with stress granules and recently showed that under stressful cellular conditions, including oxidative insult, TDP-43 localizes to stress granules.¹⁶ One analysis of ALS-causing mutations, A315T and M337V, located in the GRD, specifically demonstrated the modulation of stress granule dynamics by TDP-43.¹⁶ The clear association of the mutations in the GRD of TDP-43 and the modulation of stress granule formation, which is a key functional outcome of the GRD’s most significantly associated canonical pathway, suggests the accuracy of our analysis. Overall, the highly significant association of this canonical pathway across multiple domains, suggests the importance of the eIF2 signaling pathway, especially within the context of TDP-43 pathology.

The regulation of eIF4 and p70S6K signaling pathways was also identified as one of the key canonical pathways activated by all domains, except the NLS. This canonical pathway is similar to the eIF2 signaling pathway in that it involves a eukaryotic initiation factor and plays a key role in translation initiation, but distinguishes itself with its specific role in both cell proliferation and cell death.^57,58 eIF4 initially stabilizes the mRNA and small ribosomal subunit complex, which then recruits the eIF2 ternary complex.^59,60 The activity of eIF4 is modulated by its phosphorylation state, which is controlled by p70S6K.⁵⁷ Interestingly, numerous ALS-causing mutations within the TARDP gene have demonstrated alterations to their cell proliferation activity and are linked to p53-mediated apoptosis, which ultimately induce neurodegeneration.⁶¹ Therefore, the eIF4 and p70S6K signaling pathways represent the second canonical pathway that is suggested to be significantly related to TDP-43 function and potentially pathology when it becomes dysfunctional in patients.

The CSDE1 signaling pathway was the third canonical pathway shared among all domains, albeit different proteins were involved. This pathway plays a critical role in translational reprogramming and therefore helps determine the fate of numerous RNA during cellular activities.⁶² Some of the broad cellular functions implicated in this pathway include cell cycle, apoptosis, differentiation, and dosage compensation.⁶² CSDE1 signaling is associated with a broad spectrum of cellular functions, but the detailed aspects of its association are not fully understood. Interestingly, the CSDE1 signaling pathway has been associated with mutations in the fused in sarcoma (FUS) gene, an ALS-linked mutation, and is significantly reduced in the presence of mutated FUS.^63,64 CSDE1 is also involved in mRNA translation initiation and RNA stability and abundance, similar to the eIF2 pathway.⁶⁵ The relationship between TDP-43 and perturbed RNA stability has been previously explored.^47,66 Mutations in TDP-43 increase RNA stability, which also supports the functional overlap of TDP-43 and the CSDE1 signaling pathway.⁶⁷ Therefore, the significant association of the CSDE1 signaling pathway across numerous domains implicates the canonical pathway as being associated with TDP-43 function and proteinopathy, and suggests further experimental exploration to reveal the details of their relationship.^63,64

The microRNA biogenesis signaling pathway is significantly associated with every TDP-43 domain and represents a critical factor in neuronal health, development, and activity.^{11,22,30,68,69} MicroRNAs are small noncoding RNA sequences that negatively regulate gene expression by targeting specific mRNA sequences to silence their expression.^15,70 The mechanism of the canonical pathway is well documented in previous studies and shows exceptionally dynamic activity across many cell types.¹³ This broad functionality results in microRNA being implicated in numerous pathologies including cancer, autoimmune, motor neuron, and neurodegenerative diseases.^12,71,72 Interestingly, there is a direct canonical relationship between TDP-43 and microRNA, in which TDP-43 directly associates with pre-miRNA, pri-miRNA, Drosha, and Dicer to regulate the biogenesis of specific microRNAs.³⁰ The depletion and dysregulation of this association have exhibited detrimental effects on neuronal health contributing to neurodegeneration.^73,74 The direct relationship between TDP-43 and microRNA biogenesis is furthered by the alterations in microRNA levels with patients with TDP-43 pathology.²² Therefore, our analyses showing the high association of microRNA biogenesis across all functional domains of TDP-43 adds to previous research demonstrating that microRNA biogenesis plays a critical role in both TDP-43′s normal functionality and its proteinopathy.

It would have been remarkable if we were able to include the mRNA binding properties of TDP-43 in our analyses and investigate the list of RNAs that bind to different regions of TDP-43 and how their interaction changes with mutations in each domain. Unfortunately, such studies and proper modeling were not possible due to lack of sufficient information. However, we believe that these studies will be possible in the near future and would complement our studies.

PPI-based analysis revealed several significant canonical pathways that are primarily modulated due to interactions via specific TDP-43 domains. The unique canonical pathways included the oxidized GTP and dGTP detoxification pathway in the RRM1 domain, and the cell cycle control and chromosomal replication pathway in the RRM2 domain. The NLS domain showed multiple unique pathways including DNA DSB repair by homologous recombination, Ran signaling pathway, and the differential regulation of cytokine production in macrophages and T helper cells by IL-17A and IL-17F. The power of some of these domain-specific canonical pathways are limited due to low overlap ratios and the numbers of previously identified interaction partners for that specific domain. Therefore, the unique domain-specific pathways warrant careful further investigation in the future.

Therapeutic intervention analysis

The current therapeutic landscape for diseases associated with the TDP-43 proteinopathy is limited. There is an urgent need to develop effective and long-term treatment strategies for patients with TDP-43 pathology. Therefore, the identification of domain-specific canonical pathways that are potentially dysregulated in patients provides a novel avenue for personalized medicine that addresses the phenotypic heterogeneity.

Large-scale data platforms, such as IPA, provide information on therapeutics associated with molecules in each of the significantly associated canonical pathways. Therefore, each identified pathway is associated with various drug targets. Table S2 in the supplementary material online contains the list of the top drugs with the most interactions or regulatory molecules within each significant canonical pathway. Table 1 demonstrates the top drugs from the four most common canonical pathways: eIF2, CSDE1, microRNA bio genesis signaling pathways, and regulation of eIF4 and p70S6K signaling. Each of the canonical pathways that are shared among different domains showed unique druggable targets, and two pathways showed significant overlap. For example, drugs or compounds tested to become drugs, such as empesertib, neomycin, zotatifin, ISIS 183750, and SM1 71, were highly associated with the regulation of eIF2, eIF4 and p70S6K signaling pathways. However, more detailed and cell-based assays are needed to investigate their impact on modulation of these signaling cascade of events and how that contributes to the health of neurons that are diseased due to TDP-43 pathology.

TABLE 1.

Drug screen results demonstrated the most highly associated therapeutics within each of the common significant canonical pathways.

Canonical pathways	Associated therapeutics	Mechanism of action	FDA approval status	Molecular interactions
eIF2 signaling	Empesertib	MPS1 inhibitor	Phase 1	8
	Neomycin	RPS12 inhibitor	Approved	8
	Zotatifin	EIF4A inhibitor	Phase 2	7
	ISIS 183750	eIF4E inhibitor	Phase 2	7
	SM1-71	Multi-targeted kinase inhibitor	N/A	6
	Chelerythrine	PKC inhibitor	Phase 4	3
CSDE1 signaling	Supinoxin	p68 RNA helicase inhibitor	Phase 2	2
	Mimosine	SHMT1 and CCL2 Inhibitor	N/A	1
	CNTO 888	CCL2 antibody	Phase 2	1
Regulation of eIF4 and p70S6K signaling	ISIS 183750	eIF4E inhibitor	Phase 2	12
	Zotatifin	EIF4A inhibitor	Phase 2	10
	Empesertib	MPS1 inhibitor	Phase 1	6
	Neomycin	RPS12 inhibitor	Approved	6
	SM1-71	Multitargeted kinase inhibitor	N/A	4
	Binimetinib	MEK inhibitor	Approved	3
MicroRNA biogenesis signaling	Retaspimycin	HSP90 inhibitor	Phase 3	4
	Pimitespib	HSP90 inhibitor	Phase 1	4
	Cisplatin	DNA cross-linking/alkylation	Approved	4
	SHetA2	HSPA binder	Phase 1	4
	Alvespimycin	HSP90 inhibitor	Phase 2	4
	Luminespib	HSP90 inhibitor	Phase 2	4

The drug target analysis of each significant canonical pathway suggested several potential therapeutic interventions. The majority of suggested potential therapeutics were anticancer drugs, which could be due to TDP-43′s association with RNA metabolism, a function commonly dysregulated in patients with cancer. Cisplatin is a prevalent anticancer therapeutic associated with the microRNA biogenesis pathway and was identified by our screen. Previous studies have shown an association of cisplatin with superoxide dismutase 1 (SOD1), such that cisplatin stabilizes misfolded SOD1 by inhibiting heterodimerization via reduction of zinc affinity.⁷⁵

We also found that heat shock protein 90 (HSP90) inhibitors, including retaspimycin, pimitespib, luminespib, and alvespimycin, were associated with the canonical pathways shared among different domains. These four compounds are associated with the microRNA biogenesis pathways and are canonically related to ALS, as recent studies suggest that HSP90 modulation has an impact on reducing TDP-43 toxicity.⁷⁶ Additionally, several therapeutics have been suggested to modulate the activity of key components of some of the canonical pathways shared among domains. For example, empesertib, neomycin, zotatifin, ISIS 183750, and SM1–71 were observed in both the eIF2 signaling pathway and the regulation of eIF4 and p70S6K signaling pathways.

The CSDE1 signaling pathway is associated with several therapeutics including Supinoxin, mimosine, and CNTO-888, all of which have unique mechanisms of action. Supinoxin is a phosphorylated p68 RNA helicase (p-p68) inhibitor that plays a key role in RNA metabolism, which makes its association with TDP-43 especially promising.⁷⁷ Mimosine is a naturally occurring iron and amino acid chelating agent that has been shown to arrest DNA replication in mammalian cells.⁷⁸ Additionally, mimosine has a potential connection to ALS through its role in upregulating hypoxia inducible factor-1 (HIF-1), which consequently activates vascular endothelial growth factor (VEGF). VEGF has a well-documented association with a number of ALS pathologies, such as misfolded SOD1 toxicity, and has been shown to be beneficial for general motor neuron health.^79,80 CNTO-888 is a human monoclonal antibody against chemokine (C–C-motif) ligand 2 (CCL2), which shows a promising association to ALS due to CCL2 being canonically upregulated in the T-cells of patients with ALS compared with healthy controls.^81–83 Several of the compounds suggested by our domain-specific assays are not currently United States Food and Drug Administration (FDA) approved, but their mechanisms of action are of great interest, aligns with the canonical pathways that are altered and thus require further investigation especially within the context of TDP-43 pathology. We believe that the domain specific canonical pathway analyses may form the basis of therapeutic screening and suggest the potential efficacy of therapeutics based on the location of the patient mutation.

Concluding remarks

Protein interactions determine a key aspect of TDP-43 function, and therefore bringing mechanistic insights into these interactions would provide a better understanding of the underlying mechanisms that lead to TDP-43 pathology. Investigation of PPIs reveals canonical pathways that are key contributors to different diseases such as ALS, ALS/FTLD, AD, and PD. Additionally, modeling the interactions with location specificity allows prediction of ‘mutation-effect’ on different domains and regions. Domain-specific analysis demonstrated both common and unique canonical pathways associated with distinct domains. The most common pathways included eIF2 signaling, CSDE1 signaling, microRNA biogenesis signaling, and the regulation of eIF4 and p70S6K signaling pathways, all of which have functional relationships with modulation of RNA metabolism, a well-characterized overall function of TDP-43.^6,47,84,85

In addition to the common pathways, each domain demonstrated at least one novel significant pathway, suggesting a unique importance for each domain and their interactions. We found that not all proteins bound equally to each domain, and some binding partners preferentially interacted with distinct regions. Therefore, mutations at these sites will not affect PPIs in other regions, but will result in alterations of canonical pathways associated with the mutated site. Therefore, domain-specific PPIs gives mechanistic insights into the distinct cellular events perturbed by location-specific mutations. Since drugs are developed for a specific mode of action and to modulate a given cellular event, being able to correlate the location of the mutation and the perturbed cellular function is of great importance.

Since TDP-43 is an RNA-binding protein, it is important to investigate the mRNA and other RNA molecules it binds to, and how different gene mutations affect its ability to bind to RNA and protein. Our current study focused on the protein interactions, mostly because the direct interaction maps of TDP-43 with the mRNA and other RNA molecules have not been fully reported, and we currently do not have sufficient data to extrapolate the impact of the mutations on RNA-binding ability of TDP-43 protein. Here, we focused on the PPIs; future studies are needed to investigate the RNA interactions and how they are altered with the mutations detected in patients.

Even though the overall goal is to identify novel compounds and drugs that would eliminate TDP-43 pathology and improve neuron health, there are numerous FDA-approved drugs with known mechanism of action and proven lack of toxicity. Therefore, repurposing drugs and findings ways to incorporate them beyond their original target population has been an emerging idea and quest. Here, we identified canonical pathways that are potentially involved and that take place in distinct regions and domains of the TDP-43 protein. The common pathways are related to previously identified functions of TDP-43. However, the unique ones for each domain, such as the oxidized GTP and dGTP detoxification pathway in the RRM1, cell cycle control and chromosomal replication pathway in RRM2, and DNA DSB repair by homologous recombination or the RAN signaling pathway within the NLS, shed light on the potential differences among patients with TDP-43 pathology, and the major underlying causes of the disease. Since there is immense heterogeneity among patients, it is important to understand why and how each patient begins to develop disease-causing pathologies and how we can develop treatments that target the cellular events perturbed in each patient. Our analyses revealed a correlation between the site of the mutation and the cellular events that may be mostly affected in patients with TDP-43 pathology. We also realized the presence of drugs and compounds that act on a given canonical pathway, which are either approved by FDA or are currently under clinical investigation for other diseases and conditions.

As we better understand the canonical pathways, cellular events, and mechanisms that become dysfunctional in each patient diagnosed with different diseases, we begin to realize the commonalities among diseases and that each patient represents a unique signature. Therefore, finding drugs for diseases becomes less relevant and finding drugs for dysregulated cellular mechanisms emerges as a novel alternative. As we better understand which cellular events are perturbed in which patient, we may also begin to appreciate the possibility of developing therapies that focus on mechanisms and not on diseases per se. It may then be possible to develop drugs that would have implications in a broad spectrum of patients diagnosed with different disease names, but in fact suffer from the same cellular problems and dysfunctional canonical pathways.

Therefore, rather than trying to find drugs that would cure all patients with such immense level of heterogeneity in one disease group, it may be more feasible to develop therapies for patients who share common problems but are diagnosed with different diseases. This line of thinking also opens the gates for repurposing current drugs, which were identified based on a given mechanisms of action for a distinct disease, and yet the same mechanism of action may be perturbed in another patient diagnosed with a different disease. The mechanistic insights in drug discovery overcome many aspects of patient heterogeneity and emphasize personalized medicine approaches that mostly focus on the cellular events and canonical pathways that are selectively perturbed in each patient.

As we move forward in our drug discovery efforts for complex and heterogeneous neurodegenerative diseases, it is time to develop more mechanistic insights in our approach so that we can begin to lay the foundation for true and effective personalized medicine approaches of the future. Here, we studied TDP-43 protein and investigated its interaction with other proteins in each region, which informed us of the importance and function of each domain. We found that based on the location of the mutation, distinct aspects of TDP-43 function would be impaired, which may in part, explain the heterogeneity among patients with TDP-43 mutations. The correlative analyses between the location of the mutation and the disrupted protein activity due to altered PPI, will not only pave the way for a mechanism-focused drugs discovery but also help develop personalized medicine approaches for each patient. We believe that the idea of mechanism-focused drug discovery for personalized medicine is the way forward, especially for complex and heterogeneous neurodegenerative diseases such as ALS and ALS/FTLD.

Biographies

Benjamin Helmold earned his undergraduate degree from Northwestern University Weinberg College of Arts and Sciences in 2023, with emphasis on Economics. He is a premedical student and will start his MD training in 2024.

Kate Pauss graduated from the University of Texas at Dallas with a BS in Neuroscience in 2021. She is currently a PhD student at the Neuroscience program of the University of Kentucky College of Medicine.

Hande Ozdinler is an Associate Professor at the Department of Neurology, Northwestern University, Feinberg School of Medicine. She is also a faculty member at the Chemistry Life Sciences Processes Institute, Les Turner ALS Center, Mesulam Center for Cognitive Neurology and Alzheimer’s Disease, and Robert H. Lurie Comprehensive Cancer Research Centers. She is the founding director of the Ozdinler Upper Motor Neuron Lab at Northwestern and the Head of Scientific Board at AKAVA Therapeutics.

Data availability

Data will be made available on request.