Network based approach for drug target identification in early onset Parkinson’s disease

Abstract

Despite the abundance of large-scale molecular and drug-response data, current research on early-onset Parkinson’s disease (EOPD) markers often lacks mechanistic interpretations of drug-gene relationships, limiting our understanding of how drugs exert their therapeutic effects. While existing studies provide valuable EOPD markers, the mechanisms by which targeted drugs act remain poorly understood. We propose DTI-Prox, a novel workflow that identifies potentially overlooked EOPD markers and suggests relevant drug targets. DTI-Prox employs network proximity to measure how closely connected a drug and gene are within a biological network. Additionally, node similarity, which assesses the functional resemblance between network nodes, reveals meaningful drug-gene connections. DTI-Prox identifies 417 novel drug-target pairs and four previously unreported EOPD markers (PTK2B, APOA1, A2M, and BDNF), demonstrating significant pathway enrichment in neurodegenerative processes. Notably, shared pathway analysis shows that prioritized drugs such as Amantadine, Apomorphine, Atropine, Benztropine, Biperiden, Bromocriptine, Cabergoline, Carbidopa, and Citalopram, currently used for other conditions, interact with key EOPD-associated diagnostic markers, suggesting their potential for drug repurposing. The constructed functional network’s validity is reinforced by statistically significant drug-target pairs. The findings provide new insights into EOPD drug mechanisms and identify promising therapeutic candidates, potentially leading to more effective, personalized treatment approaches for EOPD patients.

Introduction

Early-onset Parkinson’s disease (EOPD), defined as neurological symptoms manifesting before age fifty, represents a distinct clinical entity within the Parkinson’s disease spectrum¹. Unlike late-onset PD, EOPD presents unique challenges in diagnosis and treatment, characterized by a more complex genetic profile and potentially different disease progression. Current treatment strategies primarily focus on symptomatic management through dopaminergic therapies, which often lead to significant motor complications over time². This approach underscores the urgent need for more targeted, disease-modifying interventions.

The genetic landscape of EOPD is complex and only partially understood. Key genetic mutations are identified in genes such as PARK2 (parkin)³, PINK1⁴, DJ-1, and LRRK2, each potentially disrupting crucial cellular processes. These mutations impact fundamental biological mechanisms including mitochondrial function, protein homeostasis, and oxidative stress responses. However, the precise molecular pathways connecting these genetic variations to disease progression remain elusive. Compounding this challenge, biomarker research is predominantly generalized across PD, with insufficient attention to the unique characteristics specific to early-onset manifestations⁵.

Existing research approaches in EOPD are fundamentally limited by their fragmented methodology. Investigators typically focus on identifying genetic markers or potential drug targets in isolation, without developing a comprehensive framework that explains how these elements interact within disease pathways. This siloed approach significantly constrains our ability to develop targeted therapeutic strategies, leaving a substantial gap in our understanding of EOPD’s complex molecular mechanisms. The advent of large-scale molecular profiling and comprehensive drug-response datasets now offers an unprecedented opportunity to explore intricate drug-gene relationships. However, these valuable resources remain largely underutilized, with current research failing to integrate diverse data streams into a cohesive mechanistic understanding of EOPD pathogenesis.

To address these critical limitations, we introduce DTI-Prox, an innovative computational workflow designed to bridge the gap between marker identification and mechanistic interpretation in EOPD research. By leveraging network proximity and sophisticated node similarity measures, DTI-Prox provides a nuanced approach to uncovering functional connections between drugs and genes. The framework’s distinctive methodology incorporates two complementary network analysis techniques: shortest path and node similarity metrics. The shortest path analysis captures direct connectivity between molecular entities, revealing potential functional relationships, while node similarity assessment evaluates structural and attributional resemblances between network components. This dual approach enables a more comprehensive and sophisticated examination of potential therapeutic interactions.

Through rigorous computational analysis, DTI-Prox not only identifies previously overlooked genetic markers such as A2M, BDNF, APOA1, and PTK2B but also prioritizes drug-target pairs based on shared functional characteristics. By establishing statistically validated functional networks of drug-target relationships, our approach offers a promising pathway for advancing EOPD research, potentially revolutionizing drug repurposing efforts and paving the way for personalized treatment strategies.

Results

EOPD possess similar characteristics with PD. However, they present differently. EOPD may caused by specific genetic mutations. To understand the molecular mechanism that differentiate EOPD from PD and their therapeutic approaches towards neurodegenerative diseases, a frame is proposed in Fig. 1.

Fig. 1.

Schematic representation of the DTI-Prox workflow, illustrating network construction, proximity measurement, and drug-target identification steps.

Identification of candidate biomarkers for EOPD

After preprocessing of the curated datasets 55 disease-specific genes and 806 drug targets are identified as an input of DTI-Prox framework. Based on the input gene the integrated proximity-based approach in the DTI-Prox framework identified six candidate biomarkers strongly associated with early-onset Parkinson’s disease (EOPD). These biomarkers are A2M (Alpha-2-macroglobulin), BDNF (Brain-derived neurotrophic factor), LRRK2 (Leucine-rich repeat kinase 2), APOA1 (Apolipoprotein A1), PTK2B (Protein tyrosine kinase 2 beta), and SNCA (Alpha-synuclein). These biomarkers are selected based on their high proximity scores shown in the expanded network established during the study. These markers indicate their critical roles within the EOPD-specific network and their significant associations with potential drug targets. Moreover, identified biomarkers are demonstrated significantly elevated expression levels in EOPD patients (Fig. 2).

Fig. 2.

Differential Gene Expression Profile in Early-Onset Parkinson’s Disease. Comparative analysis of gene expression levels between healthy controls and EOPD patients reveals significant molecular alterations. The heatmap illustrates the expression patterns of six key genes: A2M, BDNF, SNCA, LRRK2, PTK2B, and APOA1.

Drug repurposing and target identification

Subsequently, 1803 drug-disease pairs exhibiting high proximity have been identified. Leveraging drug repurposing, we have further predicted 417 novel drug-target pairs. By prioritizing pairs, the framework identified top drugs Amantadine⁶, Apomorphine⁷, Atropine, Benztropine⁸, Biperiden⁹, Cabergoline¹⁰, and Carbidopa¹¹ identified as potential candidates for EOPD therapy through their strong connectivity to the identified biomarkers. For prioritization of drug-target pairs, we have integrated multiple criteria: statistical significance, overlapping pathways, and minimal offsite shown in Fig. 3 using a barplot. Statistical significance has been assessed by calculating p-values for each pair, with values below 0.05 indicating robust associations (Fig. 4). These statistically significant associations highlight the efficacy of network-based approaches in unravelling intricate disease mechanisms and identifying promising therapeutic targets.

Fig. 3.

Prioritization of drug-target pairs based on proximity analysis within the biological network. The ranking highlights drug-gene pairs with significant proximity values, emphasizing their potential therapeutic impact on the studied disease. Each pair is evaluated to minimize off-target effects, identifying the most promising candidates for targeted interventions.

Fig. 4.

Statistical significance is evaluated by calculating p-values for each identified pair, with values below 0.05 indicating strong associations. These statistically significant results underscore the effectiveness of network-based approaches in uncovering complex disease mechanisms and identifying potential therapeutic targets.

Furthermore, a pathway enrichment analysis has been performed on these genes, identifying several enriched pathways by using KEGG¹² and Reactome¹³ databases. Pathways analysis further explicates the functional relationships between drugs and genes (Fig. 5), emphasizing the importance of shared biological processes. Functional analysis of the candidate biomarkers revealed significant enrichment in Wnt signaling and MAPK signaling pathways, which are known to play pivotal roles in neurodegenerative processes, including synaptic plasticity, neuroinflammation, and oxidative stress. The identification of these pathways underscores the potential repurposing of drugs like Cabergoline and Carbidopa, which have shown promise in modulating Wnt and MAPK signaling in other neurodegenerative diseases.

Fig. 5.

Comprehensive mapping of (A) gene pathways and (B) repurposed drug pathways (B) reveals critical molecular interactions across key neurological and cellular signaling networks. The bubble plot visualizes the systemic interactions of identified drug-target pairs, highlighting their significant involvement in multiple pathological pathways. Color intensity and bubble size represent the relative significance and interaction strength of molecular entities, providing a nuanced view of potential therapeutic interventions and disease mechanism interactions.

Moreover, minimizing the potential for offsites guided the prioritization process by considering drugs with favorable safety profiles. Figure 6 presents a network visualization of the interactions between the top ranked drug target pairs of EOPD. Interestingly, these drugs are currently used in the management of Parkinson’s disease and related neurological disorders and have mechanisms of action that align with dysregulated pathways in EOPD.

Fig. 6.

Comprehensive Network Visualization of Drug-Gene Interactions in EOPD. The network illustrates the intricate molecular relationships between repurposed drugs (orange nodes), directly connected genes (blue nodes), and genes of interest (red nodes). This visualization reveals the complex interconnectedness of pharmacological targets and molecular pathways, highlighting potential therapeutic interactions EOPD mechanisms.

Network significance and validation

The shortest-path and Jaccard similarity analyses revealed that the proximity scores of identified biomarkers and drug targets are significantly higher than expected by random chance (empirical p-value< 0.05). These results are validated across three independent datasets, including two early-stage Parkinson’s disease datasets and curated protein information from UniProt, ensuring robustness and reproducibility.

Based on these PPI networks, subnetwork-disease and subnetwork-drug associations have been established. The MCL algorithm has been employed to identify maximal gene clusters within these subnetworks. We have selected subnetworks containing the maximum number of disease-specific and drug-target genes to ensure a comprehensive coverage of relevant biological interactions. Furthermore, a pathway enrichment analysis, this analysis indicated the impact of established subnetworks in neurodegenerative diseases as well as signalling pathways involved in disease progression. Subsequently, we have merged the subnetworks to construct an integrated network, carefully preserving overlapping proteins and interactions. This integrated network comprises 1592 nodes and 6450 edges. To account for indirect interactions, we have expanded the network by including two layers of neighbouring nodes and edges, resulting in a network with 3180 nodes and 13,550 edges.

Discussion

The DTI-Prox framework has enabled us to uncover critical insights into the molecular landscape of early-onset Parkinson’s disease (EOPD). By identifying 1,803 drug-target pairs with highly overlapping neighboring nodes, we have prioritized six EOPD-related markers that demonstrate key roles in disease onset and progression. The autosomal dominant genes LRRK2 and SNCA have long been recognized as central players in EOPD pathogenesis¹⁴. SNCA, encoding -synuclein, is particularly significant, as its pathological aggregation and inhibition of neurotransmission represent a critical early intervention point¹⁵. Similarly, the dual GTPase and kinase activity of LRRK2 makes it a vital therapeutic target, especially given the contribution of its genetic variants to earlier disease onset. Importantly, our analysis revealed four additional markers with significant implications for EOPD. A2M (Alpha-2-Macroglobulin) exhibited elevated expression in EOPD patients and is known to influence age of onset, suggesting its potential as an early diagnostic biomarker¹⁶. The overlap between A2M and Alzheimer’s disease mechanisms indicates shared neurodegenerative pathways that could be targeted therapeutically.

BDNF (Brain-Derived Neurotrophic Factor) emerged as a particularly promising target due to its dual neuroprotective and neuromodulatory functions, especially in dopaminergic neurons. The interaction between BDNF and LRRK2 suggests a potential mechanism for early disease modification¹⁷. This finding could have profound therapeutic implications, as targeting BDNF pathways may not only preserve neuronal integrity but also delay disease progression. Future therapeutic strategies could explore the potential of BDNF agonists or modulators in neuroprotection, specifically tailored for early-onset patients who may benefit most from preserving dopaminergic function.

APOA1 (Apolipoprotein A1), with its decreased levels in early-stage PD and comparable diagnostic potential to -synuclein, highlights its value as an early biomarker¹⁸. Additionally, APOA1’s role in lipid transport and inflammation suggests that it could serve as a therapeutic target to modulate neuroinflammatory processes, which are increasingly implicated in EOPD progression.

Finally, PTK2B (Protein Tyrosine Kinase 2 Beta) was found to correlate with cognitive function in early PD stages, suggesting its potential use in monitoring disease progression and cognitive decline. Pathway analysis revealed that PTK2B is involved in cellular stress responses and synaptic plasticity, making it a compelling target for therapeutic interventions aimed at mitigating cognitive impairments in EOPD patients¹⁹.

The pathway enrichment analysis further underscores the significance of these markers in EOPD. Key pathways such as MAPK signaling, which regulates cellular stress responses, and Wnt signaling, crucial for neuroprotection, present actionable targets for therapeutic intervention^20,21. The involvement of these pathways suggests that repurposed drugs like Cabergoline and Carbidopa, which modulate these pathways in other conditions, could be repositioned to address EOPD-specific mechanisms.

The identification of potential therapeutic candidates, such as Amantadine, Apomorphine, and Benztropine, highlights the opportunity to move beyond symptomatic management and toward disease-modifying treatments. These drugs, currently used for other conditions, align with the dysregulated pathways identified in EOPD, emphasizing their repurposing potential. For example, Amantadine, which modulates NMDA receptor activity, may offer neuroprotective benefits by addressing excitotoxicity, a hallmark of neurodegeneration. By emphasizing the clinical implications of identified biomarkers and drug candidates, this study highlights a path forward for developing targeted therapies that address the unique molecular mechanisms of EOPD.

Methods

The proteins responsible for Parkinson’s disease have been curated from different publicly available datasets. We have considered three datasets consisting of two different stages of Parkinson’s disease. 27 early-stage differentially expressed proteins are downloaded from peripheral blood samples of 41 patients and 40 healthy controls are evaluated. Similarly, the protein expression data containing early onset Parkinson’s disease at the mild cognitive impairment (MCI) stage has been used: GSE74763. The authors in the study considered the sera²², obtained from a pre-defined age and gender-matched threshold value, of control samples. The 25 ADNI MCI samples are compared with 25 control samples from 9486 human protein microarrays using prospector analysis software²³. In both datasets, the patients are suffering from the early stage of Parkinson’s disease. The third dataset is prepared based on the information available regarding proteins responsible for Parkinson’s disease at the UniProt database www.uniprot.org. Subsequently, we have curated the drug list used to treat Parkinson’s disease along with their targets²⁴. The framework of our proposed DTI-Prox is shown through a flowchart in Fig. 1.

Dataset processing and preparation

Differential protein expression analysis is performed using the limma package²⁵ in R, which employs a linear modeling approach combined with empirical Bayes statistics. Samples are considered significantly differentially expressed if they meet the following criteria: absolute log2 fold change and adjusted p-value . Multiple testing correction is performed using the Benjamini-Hochberg procedure to control the false discovery rate (FDR). The moderated t-statistics computed by limma are used to assess the statistical significance of differences between groups, as this approach is more robust than traditional t-tests, particularly for datasets with small sample sizes. The moderated t-test implemented in limma provides improved statistical power by borrowing information across genes to estimate sample variances, making it particularly suitable for high-throughput protein expression data analysis. Let for each sample, group 1 consist of diseased sample with mean , standard deviation , and as the population mean, and group 2 consists of control sample with mean , standard deviation and population mean . The t-value is defined as follows:

1

The degree of freedom can be expressed as:

2

The t-test applied to the normalized data of the three disease datasets is used to identify the DE in each case. A threshold of 0.05 is considered for the rejection of the null hypothesis. If the p-value of a sample is less than 0.05, it signifies the protein is differentially functioned. For the UniProt dataset, we have used their Rest API to retrieve disease-specific proteins. The list of differentially expressed genes for three datasets are considered as early-onset PD-specific genes or EOPD.

Constructing disease-specific and drug-target PPI networks for EOPD prognostic marker identification

Differentially expressed sample sets of disease-specific genes and drug targets are separately considered for further research. The objective of the study is to identify overlooked prognostic markers of EOPD and their corresponding targeted drugs. Therefore, to understand the impact of the key proteins that play central roles in maintaining the disease network and the complex biological mechanism of EOPD, we established two protein-protein interaction networks for disease and drugs respectively, from the STRING database. Let be the PPI network, where V is the set of proteins (nodes) and E is the set of interactions (edges). Edges are weighted based on interaction confidence scores from the STRING database. Moreover, the network complexity is reduced by retaining only high-confidence interactions. Furthermore, to focus on specific, functionally related groups of proteins within the broader network, subnetworks or densely connected clusters are identified from the established PPI networks. This targeted approach helps to simplify the complexity of large-scale PPI networks, making it easier to identify meaningful biological patterns and interactions that are relevant to EOPD. Additionally, these subnetworks allow for more precise identification of potential drug targets and biomarkers within specific biological contexts. The subnetworks (subnetwork-disease and subnetwork-drug) are established by using Markov clustering. Constructing subnetworks from the PPI network using Markov clustering (MCL) brings a novel approach to understanding the modular structure within complex biological systems. The novelty lies in MCL’s ability to identify densely connected clusters by simulating random walks, which effectively capture groups of proteins that function together within biological pathways. This clustering method is particularly advantageous as it considers the stochastic flow within the network, enabling the detection of biologically meaningful modules that might not be apparent through simpler clustering methods. By applying MCL, we uncovered functional subunits within the broader PPI networks, aiding in the discovery of key functional groups, disease-associated modules, and potential drug target clusters. The MCL process can be described by the following iterative equations:

3

Where M is the transition matrix of the random walk, and r is the inflation parameter that controls the granularity of the clusters.

Merging subnetworks to Identify EOPD drug targets

Furthermore, the subnetworks are merged to identify the drug targets for EOPD. In this regard, we merged and . The combined network is formed such that:

4

where: is the set of all unique nodes from and . is the set of all edges from , , and the cross edges connecting nodes between and . We ensured that any overlapping genes and interactions in the two subnetworks are preserved, maintaining the integrity of shared nodes and edges. To further refine this network, we optionally expanded it by including two hops of neighboring nodes and edges, and respectively. This expanded network is defined as:

5

Here, is the set of nodes connected to any node by edges in the global PPI network , expressed as:

6

is the set of edges connecting nodes in to nodes in .

Integrated proximity analysis for identifying EOPD drug targets through shortest path and jaccard similarity

The proposed method for calculating proximity between disease-specific genes and drugs follows two main steps:

Step 1: Shortest path identification

The first step involves identifying the shortest paths between disease-specific genes and drugs through their respective drug targets within the expanded network . For each pair , the shortest path length is computed based on Eq. (7):

7

where represents the set of all possible paths from (a disease-specific gene) to (a drug target), and denotes the number of edges in path p. This step generates a matrix where each entry corresponds to the shortest path length.

Step 2: Jaccard similarity for node importance

After identifying the shortest paths, the nodes involved in these paths are evaluated to understand their importance within the entire expanded network . The importance of a node is assessed using the Jaccard similarity of its neighbors.

For nodes and , the Jaccard similarity is calculated in Eq. (8):

8

where and represent the sets of neighboring nodes of and within . The Jaccard similarity provides a measure of how similar the nodes are based on their shared neighbors.

The final proximity score is derived by combining the shortest path length and the Jaccard similarity . This score is used to identify potential drug targets by evaluating both the connectivity and the importance of the nodes in the context of the expanded network in Eq. (9):

9

where is a function that integrates the shortest path distance and Jaccard similarity to assess the proximity between disease-specific genes and drugs through drug targets effectively.

To evaluate the significance of the proximity scores and similarity measures obtained from the integrated network , we computed the empirical -value. The null hypothesis was defined as follows: there is no significant difference in the proximity scores or similarity measures between drug targets and EOPD-associated genes beyond what would be expected by random chance. To test , we generated randomized networks while preserving the node degree distribution and overall structure of . The proximity scores and similarity measures are recalculated for each randomized network using the same approach applied to . The empirical -value was then determined by comparing the observed scores from to the distribution of scores from the randomized networks as defined in Eq. (10):

10

If the -value is less than the chosen significance level (typically 0.05), we rejected and considered the observed proximity scores or similarity measures as statistically significant.

Author contributions

A.D. and MC conceived and experimented. A.D. and M.C. analysed the results and drafted the manuscript. U.M. and S.B. supervised the work. All authors reviewed the manuscript.

Data availability

All the data used in this manuscript is publicly available.

Declarations

Competing interests

The authors declare no competing interests.