Systems Biology to Address Unmet Medical Needs in Neurological Disorders

Abstract

Neurological diseases are highly prevalent and constitute a significant cause of mortality and disability. Neurological disorders encompass a heterogeneous group of neurodegenerative conditions, broadly characterized by injury to the peripheral and/or central nervous system. Although the etiology of neurological diseases varies greatly, they share several characteristics, such as heterogeneity of clinical presentation, non-cell autonomous nature, and diversity of cellular, sub-cellular, and molecular pathways. Systems biology has emerged as a valuable platform for addressing the challenges of studying heterogeneous neurological diseases. Systems biology has manifold applications to address unmet medical needs for neurological illness, including integrating and correlating different large datasets covering the transcriptome, epigenome, proteome, and metabolome associated with a specific condition. This is particularly useful for disentangling the heterogeneity and complexity of neurological conditions. Hence, systems biology can help in uncovering pathophysiology to develop novel therapeutic targets and assessing the impact of known treatments on disease progression. Additionally, systems biology can identify early diagnostic biomarkers, to help diagnose neurological disease preceded by a long subclinical phase, as well as define the exposome, the collection of environmental toxicants that increase risk of certain neurological diseases. In addition to these current applications, there are numerous potential emergent uses, such as precision medicine.

1. Introduction

Neurological diseases constitute a significant burden of illness in the population. Worldwide, in 2016, neurological illnesses were the second most frequent cause of mortality, and first most significant contributor to disability [1]. Neurological disorders encompass a wide spectrum of neurodegenerative conditions, broadly characterized by injury to the peripheral and/or central nervous system (Figure 1). Nerve damage can occur from aging, systemic diseases like diabetes, heritable genetic mutations, environmental exposures, or mechanical trauma. Although there are broad clinical and molecular differences, both within and between neurological diseases, complexity of pathogenesis is a unifying thread. It is also the principal reason that systems biology has gained traction in the recent decade as an important and central research tool for understanding neurological disease pathophysiology.

Figure 1. Types of neurological diseases.

Neurological diseases are characterized by damage to central and peripheral nervous tissue. Broadly, some categories of neurological disease include, (A) central neurodegenerative diseases, (B) motor neuron diseases, (C) peripheral neuropathies (PN). Central neurodegenerative diseases are characterized by neuronal loss in various areas of the brain, the cortex in Alzheimer’s disease (blue shading; Ai) and substantia nigra in Parkinson’s disease (blue shading; Aii). Motor neuron diseases lead to neurodegeneration of neuromuscular junctions; Bi), which can lead to atrophy of limb and diaphragm muscles. Nerve damage (Ci) in PN usually occurs in a symmetric, length-dependent manner (blue coloring; starting in the feet and progressing to the hands upon reaching the calves), including in the most common metabolically-acquired diabetic PN. Sensory neuron degeneration also shown, distally (from axon termini) to proximally (towards cell body and dendrites). There can be overlap between these categories of neurological disease; for instance, a subset of patients with the motor neuron disease, amyotrophic lateral sclerosis, can also have central frontotemporal dementia in 15 to 20% of cases. Created, in part, with BioRender.com.

Disease complexity is evident in the heterogeneity of clinical presentation, breadth of etiology, non-cell autonomous nature, and cellular, sub-cellular, and molecular pathway diversity of each neurological illness. For example, amyotrophic lateral sclerosis (ALS) is a fatal neurodegenerative disease, characterized by motor neuron degeneration and consequent muscle wasting [2]. Regarding clinical presentation, ALS can manifest as bulbar or spinal onset, which influences the progression rate. Bulbar onset ALS presents as difficulty speaking, swallowing and breathing, and is fast progressing, whereas spinal onset initially presents as limb weakness, and is usually slow progressing [2,3]. ALS phenotypes are dictated by multiple patient clinical and genetics characteristics [4]. Additionally, although progression of symptomatic disease is relatively rapid, ALS has a long subclinical prodromal phase.

With regards to breadth of etiology, ALS is associated with close to 30 genetic mutations [5]; however, a known genetic mutation is present in only around 15% of ALS individuals. Thus, polygenic risk [6,7], environmental exposure [8–10], and potential gene-environment interactions [11] have been proposed to account for cases lacking a known monogenic cause. In terms of a non-cell autonomous nature, although ALS is widely regarded as a motor neuron disease, there is a central neurodegenerative component, frontotemporal dementia, in around 15 to 20% of cases [2]. Additionally, immune system involvement is documented in ALS [12–17], which correlates with disease progression [14] and in a sex-dependent manner [15,16].

Lastly, ALS is characterized by molecular heterogeneity in genetic [5–7], epigenetic [18–22], transcriptomic [23,24], and metabolomic [25–27] signatures. Multiple biological processes are also involved, centered on excitotoxicity [28], mitochondrial dysfunction [29], and oxidative stress [30].

Although outlined here specifically for ALS, this level of heterogeneity and complexity of disease processes occurs in most neurological diseases, especially those of non-monogenic etiology. Clearly, appropriate analytical platforms are needed to uncover and dissect the numerous aspects present in these complex, multifactorial disorders. Systems biology can be leveraged to address this need, by agnostically querying molecular pathways in ach aspect of disease pathogenesis. This can shed light on mechanisms, correlate molecular signatures, i.e. genomic, transcriptomic, epigenomic, proteomic, metabolomics, or multi-Omics, to clinical presentation to refine disease classification and diagnosis, and identify potential drug targets. This is especially essential for most neurological disorders, which remain recalcitrant to treatment. Technological advances also enable single-cell resolution, which is useful for non-cell autonomous diseases. Further, bioinformatics analysis can integrate multi-Omics datasets to gain additional insight.

This chapter will be sub-divided by disease to illustrate how systems biology has advanced our understanding of neurological disorders in the recent decade. First, each sub-section will provide an overview of the neurological disease, highlighting the complexity, which systems biology can help address. Second, each sub-section will discuss the most salient studies shedding light specifically on the points stated above, namely pathomechanisms, diagnosis/classification, and drug target identification. Rather than a comprehensive review, this chapter will highlight studies that deliver overarching messages, identify future areas of investigation, and serve as a guide to researchers leveraging or planning to leverage systems biology in their research endeavors. Additionally, the chapter will emphasize metabolically-acquired peripheral neuropathy and ALS, which are our areas of expertise, but will still illustrate systems biology examples in other neurological illnesses.

2. Peripheral neuropathies

Peripheral neuropathies (PN) are a class of neurological diseases, which incur damage to the peripheral nerves [31]. The most common clinical presentation is a distal symmetric neuropathy, beginning in both feet, progressing distally to the calves, at which point it commences in the fingers and progresses distally to encompass both hands. Alternatively, PN can manifest focally as a mononeuropathy or a plexopathy, or to the autonomic nervous system as autonomic neuropathy. This chapter will refer to distal symmetric neuropathy as PN. The most frequent PN cause is metabolically acquired, secondary to either diabetes, prediabetes, or the metabolic syndrome [32]. Other causes include genetic mutations, chemotherapy, toxin exposure, infectious disease, vasculitis, mechanical injury, vitamin deficiencies, and immune-mediated disorders [33], although the cause is unknown in around 40% of PN cases, known as idiopathic PN [32]. This subsection will focus on metabolically-acquired PN, as the most prevalent PN.

2.1. Metabolically acquired neuropathies

Diabetes, an elevated fasting blood glucose (hyperglycemia), and prediabetes, a state of impaired glucose tolerance, are extremely prevalent metabolic disorders. Diabetes itself can be sub-divided into type 1 (T1D) diabetes, constituting around 5% of patients who lose pancreatic β-cells and no longer produce insulin. Patients with the more prevalent type 2 diabetes (T2D) develop insulin resistance and can no longer regulate glucose. In 2019, 463 million individuals had T2D globally, with 374 million individuals with prediabetes [34]. The metabolic syndrome, a constellation of obesity, dyslipidemia (abnormal blood lipid profile), and hypertension [35], also constitutes a massive and rising global epidemic [36,37]. Up to 50% and 30% of T2D and prediabetes patients, respectively, develop PN [31]. PN also develops in obese individuals [38–53], even independent of hyperglycemia, and in proportion to the number of metabolic syndrome components [45,54,55,47]. Frequently, however, T2D is comorbid with obesity and the metabolic syndrome [39,56].

Metabolically-acquired PN pathophysiology is complex, encompassing abnormal glucose- and lipid-centric pathways [31,57], bioenergetics and mitochondrial defects [58,59], oxidative stress [60], and inflammatory processes [61]. Moreover, PN progression may be non-cell autonomous, through a breakdown in neurometabolic coupling and crosstalk between axons and their supporting glia cells [62,63]. Thus, PN development is highly complex, advocating a systems biology approach to gain a deeper understanding of pathophysiology to develop mechanism-based treatments.

Indeed, we have conducted extensive informatics studies of PN in both mouse and human neuropathic nerve to identity recurrent pathways and possible routes to disease-modifying drugs. We have employed several mouse models of metabolically-acquired PN [64]. Streptozotocin (STZ) destroys pancreatic β-cells, mirroring the T1D scenario. The ob/ob and db/db mice harbor spontaneous mutations to satiety regulating leptin and the leptin receptor, respectively, leading to over-eating, obesity, and a T2D phenotype. Alternatively, the high-fat diet (HFD) low-dose STZ mouse model of T2D was developed to more closely mirror diet-induced T2D with comorbid obesity in humans. Omitting STZ, and solely feeding mice HFD leads to an obese prediabetes model. Equipped with these models, we have leveraged systems biology approaches to address multiple questions.

2.1.1. What is PN pathophysiology?

Early studies employing gene expression microarray technology were conducted on human sural [65] and mouse sciatic nerve [66] (Figure 2A). The human study was of sural biopsies from both T1D and T2D participants with PN, categorized as “progressors” (decrease in myelinated fiber density [MFD] as a measure of PN) versus “non-progressors” (no MFD change) over the course of a 52-week clinical trial [65]. Progressors differed in 532 differentially expressed genes (DEGs) from non-progressors; functional enrichment of DEGs identified pathways involving inflammatory responses and lipid metabolism, centered on apolipoprotein E (APOE), leptin, peroxisome proliferator-activated receptor gamma (PPARγ), JUN, and serpin family E member 1 (SERPINE1). A follow-up human study took a closer look at a subset of “regenerator” participants, which increased MFD during one year, indicative of nerve regeneration and improvement in PN [67]. Microarray analysis found regenerator sural nerves were upregulated in genes related to cell-cycle and myelin sheath functions, and downregulated in those related to immune/inflammatory pathways.

Figure 2. Select systems biology applications to metabolically-acquired peripheral neuropathy.

Systems biology has manifold applications to address unmet medical needs for peripheral neuropathy (PN). (A) What is PN pathophysiology? Tissue samples (sural or sciatic nerve from mouse (animal model) or human with T2D, type 2 diabetes, versus WT, wild-type, in this example) are profiled by an Omics platform (transcriptomics in this example). Next, differential species are identified (DEGs, differentially expressed genes, in this example). Pathway enrichment analysis of differential species lends biological insight. Thus, systems biology can uncover pathomechanisms, which can suggest therapeutic avenues. (B) What are shared and unique pathophysiology aspects in T1D versus T2D PN? Systems biology can differentiate pathomechanisms in T1D versus T2D PN, which can lead to tailored treatment regimens. (C) How does anti-type 2 diabetic drug treatment affect nerve health? Systems biology can shed insight on why current anti-type 2 diabetic drugs (PIO, pioglitazone, in this example) do not prevent PN onset and development (worsens PN by increasing expression of DEGs related to pathology, improves PN by decreasing or expression of DEGs related to pathology). (D) What new therapeutic targets for PN can systems biology identify? Multi-Omics systems biology can identity strong candidate targets for drug development (circles Omics platform 1; squares Omics platform 2). Created, in part, with BioRender.com.

These findings were echoed in sciatic nerve from db/db T2D mice with PN, which revealed dysregulation of genes responsible for lipid and carbohydrate metabolism, PPAR signaling, apoptosis, and axon guidance [66]. Promoter sequence analysis demonstrated these changes were coregulated, indicative of structural changes of axonal degeneration involving lipid metabolism. PN is progressive and evolves over time. Gene expression microarray analysis of ob/ob T2D mice at earlier 5-week and later 13-week time points found 1503 and 642 DEGs, respectively, which were overrepresented in immune/inflammatory functions, especially at 5 weeks, suggesting an early and contributory role to PN onset [68]. Analysis of multiple transcriptomic datasets underscored inflammation as a recurrent theme in diabetic PN, particularly through toll-like receptor (TLR) signaling [69]. Knocking out TLR2/4 from a prediabetes HFD model slows the onset of PN, affirming immune system involvement early in pathogenesis.

2.1.2. What are shared and unique pathophysiology aspects in T1D versus T2D PN?

PN phenotype, i.e., slowed nerve conduction velocities (NCV) in large fibers and intraepidermal nerve fiber (IENF) loss of small fibers, is similar in T1D and T2D. However, glucose control is more effective for slowing T1D versus T2D PN [70], suggesting possible pathophysiological differences (Figure 2B). Transcriptomic analysis is an ideal tool for agnostically querying pathway differences and similarities between T1D and T2D PN. Comparison of microarray results from STZ T1D versus db/db T2D sciatic nerve and kidney tissue from animals with PN and diabetic kidney disease, known as nephropathy, identified exceptionally high concordance among DEGs in diabetic nephropathy (94% of 2433 genes), but not in diabetic PN (54% of 1558 genes) [71]. These findings support the concept that distinct pathophysiology may underlie PN in T1D versus T2D, although transcriptional network analysis suggests the inflammatory Janus kinase (JAK)-signal transducer and activator of transcription (STAT) pathway is shared, regardless of diabetes type. Expanding comparison of T1D versus T2D to include human sural (progressors and non-progressors) as well as mouse sciatic nerve (STZ T1D, db/db and ob/ob T2D), a total of eight microarray datasets were used to generate a merged transcriptional network with centrality analysis, which identified top and universally shared DEGs [72]. Pathway analysis discovered these shared DEGs to be connected to pathways involving liver X receptor (LXR)/retinoid X receptor (RXR) activation, adipogenesis, and glucocorticoid receptor signaling, as well as, as anticipated, multiple cytokine and chemokine pathways. However, although pathological pathways are shared, directionality of DEGs differed in human versus mouse samples. This may be related to the time course and PN stage, source tissue location (sural, which is more distal and affected earlier in disease course in humans, versus sciatic, which is more proximal and affected later in mice), or control comparisons (progressors versus nonprogressors against neuropathic versus non-neuropathic) in humans versus mice.

Overall, however, these analyses identified highly recurrent pathways cross-species and in both T1D and T2D PN, as well as divergent pathways, which differed in T1D versus T2D.

2.1.3. How does anti-type 2 diabetic drug treatment affect nerve health?

There are several classes of anti-type 2 diabetic drugs; among them, thiazolidinediones are PPARγ agonists, which boost transcription of genes controlling glucose and lipid metabolism and improve insulin sensitivity [73]. In mice, pioglitazone and rosiglitazone, examples of thiazolidinediones, may improve some PN outcomes associated with T1D and T2D, though through an unknown mechanism. Systems biology is ideal for elucidating the mechanism of treatment-induced PN improvements (Figure 2C). In an STZ T1D model, rosiglitazone (3 mg/kg) does not reverse hyperglycemia, but does lower nerve oxidative stress [74]. Regarding PN, it does not improve function of large myelinated nerve fibers assessed by measuring sural or sciatic NCVs, but does prevent loss of small unmyelinated fibers. This preservation of function is assessed by measuring thermal hypoalgesia (loss of sensation to heat) and anatomical quantitation of small unmyelinated intraepidermal nerve fiber density (IENFD). Gene expression microarrays identified 318 DEGs between T1D versus rosiglitazone-treated T1D mouse sciatic nerve; analysis of DEGs collectively upregulated or collectively downregulated by rosiglitazone identified two transcription factor motifs linked to PN development. These transcription factor motifs were related to insulin-stimulated glucose metabolism, neurite outgrowth, growth factors, apoptosis, and survival, implicating these pathways in rosiglitazone’s mechanism of action in PN.

We have also investigated the effect of pioglitazone on PN in db/db T2D mice. In this model, pioglitazone (15 mg/kg) normalizes fasting blood glucose, glycated hemoglobin (HbA1c), and triglycerides, and lowers plasma oxidative stress, but increases body weight [75]. Pioglitazone also prevents small fiber IENFD loss, but does not affect large fiber sural or sciatic NCVs. There were 4537 DEGs in sciatic nerve between db/db versus db/db pioglitazone using microarrays, and pathway analysis revolved around adipogenesis, adipokine signaling, and lipoprotein signaling. These pathways suggest nerve lipid accumulation, a possible reason for the blunted therapeutic response on large fiber PN. Bulk RNA sequencing (RNA-seq) of nerve and kidney followed by self-organizing maps found pioglitazone reversed mitochondrial dysfunction in both tissues, but only rescued cell death and inflammation in kidney [76]. In fact, pioglitazone may have even been detrimental to the inflammatory nerve response. Pathway crosstalk perturbation network modeling of this RNA-seq nerve dataset further identified glycolysis, gluconeogenesis, and carbohydrate metabolism as contributing to the return to health upon pioglitazone treatment [77].

Overall, our data implies inflammation may underlie large fiber dysfunction, since pioglitazone does not improve NCV nor reverse nerve inflammatory pathways. On the other hand, mitochondrial dysfunction may drive small fiber dysfunction in T2D PN, since pioglitazone improves IENFD and influences DEG expression related to mitochondria. The differential impact of pioglitazone on PN and diabetic nephropathy also upholds an important tenet in diabetes complications research, which states T2D induces distinct tissue-specific metabolic changes [78]. This suggests it may also require tissue-specific therapeutic solutions, rather than a ‘one size fits all’ approach, which may have deleterious impact in one tissue, while improving another. These studies also highlight the power of systems biology to draw important and biologically relevant insight with potentially translational implications.

Although gene expression microarrays and bulk RNA-seq have been very instrumental to elucidating PN pathophysiology, the heterogeneity of cells in peripheral nerves presents limitations to these studies. To further unravel the cell-specific transcriptome and the cellular communications, single-cell RNA-seq (scRNA-seq) and spatial transcriptomics should be performed in future studies.

2.1.4. What new therapeutic targets for PN can systems biology identify?

Despite intensive research, PN remains untreatable. Effective anti-type 2 diabetic drugs improve glucose handling and systemic metabolic health; but none to date prevent PN onset and progression clinically in humans. Systems biology may aid in the discovery of potential PN therapeutics through agnostic query. Moreover, adopting a multi-Omics approach, by considering dysregulation on multiple levels, can strengthen identified candidates (Figure 2D). Systems biology can also aid in the development of new drugs or repurposing of approved drugs by integrating gene expression datasets to develop connectivity maps, which correlates dysregulated transcriptome with drug databases [79]. Lipidomics of sciatic nerve of HFD-STZ T2D and HFD prediabetes mice found triacylglycerol and, to a lesser extent, diacylglycerol accumulation in neuropathic nerve, which was reversed upon a switch back to a regular diet [80]. Transcriptomics revealed “fat digestion and absorption” and “glycerophospholipid metabolism” as important pathways; integrated lipidomics-transcriptomics centered especially on three candidates, CD36 (lipid transport), LPL (lipoprotein hydrolysis), and DGAT2 (triacylglycerol synthesis). The study also confirmed DGAT2 was elevated in sural nerve biopsies from hyperlipidemic versus non-hyperlipidemic T2D participants. Although not in development for PN, inhibiting DGAT2 have been considered for treating non-alcoholic fatty liver disease [81], a common comorbid condition of obesity.

An epigenetic (DNA methylation) and transcriptomic analysis of human sural nerve biopsies from T2D PN participants by high versus low HbA1c found overlap between DEGs and differentially methylated genes, which integrated functional and network analysis found were related to immune response, extracellular matrix regulation, and PI3K-Akt signaling [82]. This study, for the first time, demonstrated that DNA methylation could constitute a mechanism regulating gene expression in PN, revealing a gene-environment interaction by integrating epigenomics with transcriptomics. Gene-environment interaction can be also investigated by integrating scRNA-seq with the assay for transposase-accessible chromatin-seq [83].

Both studies share the same weakness, namely that causality cannot be inferred from these cross-sectional analyses. Thus, longitudinal bioinformatics analysis of nerve coupled with validation by knockout in mouse models will be required to select the best therapeutic candidates. Additionally, any developed drugs will require optimal pharmacological profiles to penetrate the nerve. Collectively, however, these studies demonstrate the power of systems biology to identify targets for therapeutic development.

2.1.5. What new PN research avenues can systems biology open?

In addition to the aforementioned established avenues systems biology has investigated in metabolically-acquired PN, it is poised to shed light on emergent and future novel avenues. Although PN is primarily considered metabolically-acquired in T2D, prediabetes, and obesity, a growing number of genome-wide association studies are identifying risk loci and generating polygenic risk scores to predict the chance of developing T2D PN [84,85] (Figure 3A). A systems biology approach of the microbiome through metagenome-wide association studies found some dysbiosis in T2D participants, which correlated with decreased universal butyrate-producing bacterial abundance and increased opportunistic pathogens [86]. Although far less investigated for PN (Figure 3B) [87], butyrate is also key in obesity-driven PN [88]. The gut microbiome may modulate PN in STZ T1D rats [89] and enteric nerves in HFD prediabetic mice [90].

Figure 3. Emergent and future systems biology research avenues in metabolically-acquired peripheral neuropathy.

Systems biology is poised to shed light on emergent and future novel research avenues in PN. (A) Genome-wide association studies to identify risk loci and generate polygenic risk scores (PRS) for PN. Genome sequencing a population and correlating to individuals with PN (dark grey figures) versus without (light grey) identifies risk loci and single nucleotide polymorphisms (SNPs). Low PRS (few risk SNPs) means a low chance to develop PN; high PRS (several risk SNPs) means a high chance to develop PN. (B) Metagenome-wide association studies to evaluate the contribution of the microbiome to PN. Microbiome can be sequenced by 16S profiling to identify microorganisms. Microorganisms and the gut secrete metabolites, which might affect the brain through the microbiome-gut-brain axis, or peripheral organs through the gut-organ axis. (C) Pharmacogenetics to identify patients that will respond to pain medications for painful PN. Genome sequencing a population can match identified SNPs with specific drugs. Created, in part, with BioRender.com and ACD/ChemSketch.

An additional possible avenue for systems biology is in the search for effective treatments for painful PN (Figure 3C), which occurs in a subset of T2D patients, who experience oftentimes debilitating pain over the course of neuropathy progression. Currently, only 1 in 7 individuals get relief from the current standard treatments available for painful PN [91,32]. One approach is to match PN sensory profiles with specific drug mechanism of actions, but this process is empirical and time consuming [32]. An unexplored avenue is pharmacogenetics, the intersection of genetic profiling with drug response [92]. In T2D, certain single nucleotide polymorphisms modulate anti-diabetic drug efficacy [93,94], and this may be the scenario in diabetic PN or painful PN. Indeed, in a small pilot study of amitriptyline, a tricyclic antidepressant, first-line painful PN treatment, participants with normal or ultra-rapid metabolizer phenotypes had fewer side effects versus individuals with lower cytochrome p450 2D6 (CYP2D6) activity [95]. Thus, this is an unexplored and potentially valuable research avenue for selecting drugs in a precision systems biology-driven approach, which might bring relief to patients with painful PN.

2.2. Other neuropathies

Although impaired metabolism is the most frequent PN cause, there are multiple other etiologies as outlined in the Section 2 introduction. In a sciatic transection model of nerve damage in rats, scRNA-seq disclosed that the main source of nerve factors following injury are Schwann cells and, unexpectedly, nerve mesenchymal cells, including from the endoneurium [96]. Integrated scRNA-seq-proteomic systems biology modeling predicted novel nerve mesenchymal cell-derived factors, which could potentially stimulate peripheral axon growth. In vitro validation of predicted factors in cultured sympathetic axons identified three factors, angiopoietin 1 (ANGPT1), C-C motif chemokine 11 (CCL11), and vascular endothelial growth factor C (VEGFC), which effectively stimulated outgrowth. This approach could make important discoveries for potential neuroregenerative therapies.

In a model of chronic autoimmune neuritis, an inflammation of the peripheral nerves, scRNA-seq was used to characterize immune cell populations [97]. Under homeostatic conditions in control mouse sciatic nerve, immune populations comprised nerve-resident homeostatic myeloid cells, which were transcriptionally distinct from central nervous system microglia. In contrast, scRNA-seq profiling of autoimmune neuritis sciatic nerve found that homeostatic myeloid cells were outnumbered by infiltrating lymphocytes, which restructured the local immune cell-to-cell interactome rather than single immune cell types. This discovery suggests a potential treatment targeting peripheral rather than resident lymphocytes or a therapeutic approach disrupting the dysregulated immune network, rather than specific immune populations.

Thus, systems biology can lend insight into the pathogenesis of various peripheral neuropathies, as attested by these two examples, unlocking possible therapeutic avenues.

3. Central neurodegenerative diseases

The most common neurodegenerative disorders of the central nervous system can be classified based on neuropathological protein aggregates, which cause nerve damage and neuronal loss. Broadly, they encompass amyloidoses, tauopathies, synucleinopathies, and TDP-43 proteinopathies, which occur in various brain regions and lead to neuronal loss and subsequent loss of nervous system function [98]. Most are incurable and treatment entails symptom management and palliative care, although there are new candidates in the preclinical and clinical pipeline [99,100], such as gene therapy [101].

3.1. Alzheimer’s disease

Alzheimer’s disease (AD) is the most common dementia, affecting 1 in 10 individuals 65 years and older [102]. In 2020, around 5.8 million people were living with AD in the United States, a number projected to increase to 13.8 million by 2050. It is familial and heritable in around 1–3% of cases, and sporadic in the other >95% [103]. AD is slowly progressive, with a long prodromal phase, sequentially followed by mild cognitive impairment, before frank mild, moderate, and severe dementia [102]. Monogenic and polygenic AD risks have been identified [104], as well as numerous modifiable [105,106] and potential exposome [107,108] risks. AD histopathology is defined by insoluble deposits of extracellular amyloid-β and intracellular hyperphosphorylated tau protein with oligomeric neurotoxic forms [109,110]. Additionally, AD pathomechanisms include metabolic [111–113] and mitochondrial [114] derangements, protein aggregates [110], autophagy [115], neurotransmission breakdown [116], inflammation [117,118], and oxidative stress [119]. There is also a non-cell autonomous component, and the brains’ resident immune cells, microglia, may actively participate in AD pathogenesis [117,118].

Systems biology has been widely used in AD research, for instance through precision medicine by leveraging genetic variants linked to neuroinflammation [120]. Herein, we will discuss four vignettes from transcriptomic, proteomic, metabolomic, and metagenomic perspectives, which uncovered exciting research avenues. Microglia and neuroinflammation have long been considered AD hallmarks; however, microglia appear to adopt a protective role, which ultimately fails, leading to neurodegeneration [118]. To shed deeper insight, Keren-Shaul et al. leveraged scRNA-seq to investigate microglia from transgenic AD mice with versus without knockout of an immune cell receptor triggering receptor expressed on myeloid cells 2 (TREM2) [121]. TREM2 mutations significantly increase the risk of AD [122]. Indeed, scRNA-seq identified a microglia phenotype the authors referred to as “disease-associated microglia” (DAM), which are activated in a sequential two-step TREM2-independent and TREM2-dependent process [121]. DAMs localize near and phagocytose amyloid plaques, slowing disease progression. Thus, this analysis revealed heterogeneity in microglia phenotypes [123], in addition to evolution of their roles over time.

Bai et al. employed proteomics and phosphoproteomics of autopsy brain samples from AD patients along a spectrum of disease progression [124]. Mass spectrometry profiled 14,513 proteins and 34,173 phosphoproteins, of which 173 candidates in 17 pathways correlated with AD progression. These hits were validated in two independent cohorts, and comparison with cerebrospinal fluid suggested possible biomarker candidates. A similar metabolomic analysis in AD participants over time found correlations between specific lipid species with disease progression, amyloid burden in cerebrospinal fluid, and magnetic resonance imaging parameters [125]. Finally, gut microbiome dysbiosis has also been linked to neurodegenerative disease [126], although confounding parameters, such as diet, poses challenges. However, novel correlations between metagenome and ketogenic diet with mild cognitive impairment, a phase preceding frank dementia, suggests possible lifestyle and dietary interventions for slowing cognitive decline in AD [127,128]

Thus, this overview exemplifies how systems biology can be capitalized to generate pathophysiological insights, generate biomarker panels, and unlock novel and paradigm-shifting therapeutic approaches in AD. The topic is discussed in greater detail in these recent reviews [129,130,126,120,131].

3.2. Parkinson’s disease

Parkinson’s disease (PD) is a neurodegenerative disease characterized by dopaminergic neurons loss in the substantia nigra and a classical motor deficit phenotype [132]. It is the second most common neurodegenerative illness after AD, with 10 to 1500 prevalence per 100,000. However, clinical presentation is highly heterogeneous, and can also involve cognitive impairment, sleep, mood, and psychiatric disorders, autonomic dysfunction, pain, and fatigue. Monogenic [132] and polygenic [133–135] PD risks have been identified, as well as numerous modifiable and potential toxic environmental exposure risks [132,108]. Histologically, Lewy bodies of α-synuclein deposits are present in various areas of the nervous system, which spread over the course of this slowly progressive disease [136]. As with AD, and even with peripheral neuropathies, such as metabolically-acquired PN, mitochondrial dysfunction, inflammation, and oxidative stress [132,137], as well as protein aggregates and autophagy [115] are major pathological aspects. Additionally, PD pathogenesis progresses in a non-cell autonomous manner via astroglia [138].

As with AD, numerous system biology techniques have been applied to investigate PD and usher in precision medicine [139]. Of interest, lipid dysregulation has emerged as an important PD facet. Although lipid dysregulation is linked to neurodegenerative disease broadly, it has very direct links in PD through mutations to genes involved in lipid metabolism, such as glucosylceramidase beta (GBA), sphingomyelin phosphodiesterase 1 (SMPD1), galactosylceramidase (GALC), phospholipase A2 group VI (PLA2G6), and sterol regulatory element binding transcription factor 1 (SREBF1) [140]. Indeed, integrated proteomics/metabolomics and metabolomics analysis of plasma from PD participants indicate lipid dysregulation may also be key in sporadic cases [141,142].

Metagenomic studies of PD have also been launched, due to the presence of α-synuclein fibril accumulation in the gastrointestinal tract [143], suggesting a possible causal relationship with microbiome dysbiosis. 16S ribosomal RNA profiling found associations of PD with increased Akkermansia, an intestinal mucin layer-degrading species, and decreased Roseburia and Faecalibacterium, short-chain fatty acid-producing species.

Though not comprehensive, these sample studies illustrate some applications of systems biology in PD, which are detailed extensively in these recent reviews [144–147,126].

4. Motor neuron diseases

Motor neuron diseases are a broad class of disorders secondary to loss of motor neuron function in the brain and spinal cord. There are multiple well-known monogenic motor neuron diseases, such as spinal muscle atrophy; however, some, such as ALS, exhibit more complex genetic architectures, with risk factors from interactions with the exposome [148]. Like other neurodegenerative diseases of the central nervous system, motor neuron diseases are heterogeneous in clinical presentation. Additionally, there is overlap between motor neuron diseases and central neurodegenerative disease, such as ALS with frontotemporal dementia. Motor neuron diseases are also mostly lacking effective therapies, although many novel genetic approaches have recently emerged for monogenic cases [149].

The most common motor neuron disease is ALS. The clinical and molecular phenotype of ALS were outlined in the introduction to this chapter. It is a relatively rare disease, with an incidence of 1 to 2 per 100,000 per year, but is becoming increasingly prevalent as the population ages; ALS incidence is projected to rise by approx. 70% by 2040 [150]. It has a long prodromal period, but is relatively rapidly progressive upon symptom onset and diagnosis, leading to death within 2 to 4 years. Thus, this highly heterogeneous fatal disease lacks early diagnostics and is without effective treatments. Numerous systems biology studies have been launched to address this unmet medical need. These studies are uncovering molecular pathways, to pinpoint actionable drug targets [151] and seek early plasma diagnostic biomarkers [152] and modifiable environmental risk factors [148], since certain environmental exposures increase ALS risk.

4.1. What is ALS pathophysiology?

Targeted molecular studies have identified important pathological aspects in ALS, which are highly shared with central neurodegenerative diseases. In ALS, this includes altered TDP-43 protein aggregates, autophagy, excitotoxicity, impaired metabolism, dysfunctional mitochondria, inflammation, and oxidative stress [30,153,154,27]. Many studies have employed the mutant superoxide dismutase 1 (SOD1) mouse model (SOD1^G93A). However, mutant SOD1 is only present in around 12% of familial and 1–2% of sporadic ALS patients [5]. Thus, although the SOD1^G93A mouse recapitulates many ALS features, it is limited, as are other genetic models, since only 15% of ALS cases have a known genetic etiology. Therefore, systems biology approaches can help agnostically query pathophysiology in sporadic ALS, in addition to genetic models (Figure 4A).

Figure 4. Select systems biology applications to amyotrophic lateral sclerosis.

Systems biology has manifold applications to address unmet medical needs for amyotrophic lateral sclerosis (ALS). Biosamples that can be analyzed include (left to right) blood, plasma, cerebrospinal fluid from consented participants, as well as postmortem spinal cord tissue. Biosamples from animal models can also be analyzed. Omics platforms include genomics (mutations, monogenic or polygenic, which correlate with ALS risk), epigenomics (DNA methylation, microRNA), transcriptomics (mRNA, long non-coding RNAs), proteomics (including phospho-proteins), and metabolomics and lipidomics. (A) What is ALS pathophysiology? Systems biology, by Omics analysis of ALS biosamples, can uncover pathomechanisms, which can suggest therapeutic avenues. In ALS, this includes altered mitochondrial and lipid metabolism (shown; AC, acylcarnitines; DAGs, diacylglycerols as examples), among other pathways (not shown). (B) Can we identify early ALS diagnostic biomarkers? ALS diagnoses are preceded by a long subclinical prodromal phase. Treatment may be more effective if initiated early; thus, systems biology, by Omics analysis of ALS biosamples, can help by identifying early disease biomarkers using classifiers, shown for a receiver operating characteristic curve. (C) Can we define the ALS exposome? Systems biology can uncover environmental toxicants, which increase ALS risk, suggesting possible modifiable avenues. Examples include air pollution (PM_2.5, particulate matter 2.5 μm), metals, microbiome, POPs (persistent organic pollutants), pesticides, and fertilizer. This cumulative exposome over time interacts with genetic predisposition (polygenic risk), to alter ALS patient epigenome, transcriptome, proteome, and metabolome/lipidome, leading to disease onset and progression. Created, in part, with BioRender.com.

Despite the breadth of identified genetic mutations (ca. 40 known mutations) in 15% of ALS cases and the fact that 85% cases are sporadic, TAR DNA-binding protein 43 (TDP-43) inclusion bodies are an almost universal finding in ALS histopathology [3]. TDP-43 regulates transcription, pre-mRNA splicing, and mRNA translation, as well as microRNA (miRNA) biosynthesis [155,156]. Therefore, many Omics studies have concentrated on dysregulation of the epigenome and transcriptome in ALS, including miRNAs [18], which have been a significant focus. miRNAs are short ~22 nucleotide-long non-coding RNAs, which degrade target mRNAs, blocking their expression and downstream effects in ALS, such as neuromuscular junction structure and function, neurogenesis, and inflammation [157].

Environmental exposure directly affects the cellular epigenome reflected in altered DNA methylation and histone acetylation. We investigated epigenetic regulation through DNA methylation in postmortem spinal cord tissue from sporadic ALS participants [158]. Global methylated (5mC) and hydroxymethylated (5hmC) cytosine were elevated in ALS spinal cord versus controls, indicating epigenome dysregulation. When we examined DEGs and differentially methylated genes by microarray, we found 251 shared hits, of which ~70% were hypermethylated, as aligned with global 5mC. Of 251, 112 were concordant, i.e., hypomethylated/upregulated (51 genes) or hypermethylated/downregulated (61 genes), and were enriched in biological pathways related to immune response, defense response, neuron adhesion, and plasma membrane part. Importantly, of the 112 candidates, 53 genes were cited at least once in PubMed, demonstrating the power of systems biology in one experiment to identify candidates from multiple publications. Our results additionally suggested myeloid or natural killer cell influx into ALS spinal cord, aligned with our findings in ALS participant blood samples [16,13,15].

We also analyzed miRNAs in sporadic ALS spinal cord by array profiling [20]. Globally, we saw reduced mature species levels, but no differences in immature transcripts, in ALS versus control, indicating impaired miRNA processing, which may be linked to TDP-43 lesions. Indeed, TDP-43 mis-localization to the cytoplasm alters miRNA profiles [19]. In sum, there were 90 differential miRNAs in ALS spinal cord versus control, 88 down- and 2 upregulated, which are annotated for pathways related to cell death, immune response, and brain development [20]. Enrichment analysis of biological functions of target mRNAs, both known and putative, corroborated immune and defense response, highlighting immune involvement in ALS [12].

In an exciting transcriptomic application, Maniatis et al. conducted spatiotemporal RNA profiling in spinal cord from SOD1^G93A versus wild-type SOD1 mice at presymptomatic, onset, symptomatic, and end-stage time points to investigate the mechanism of ‘spread’ in ALS neurodegeneration [23]. They found that microglial dysfunction preceded symptom onset and astroglial dysfunction in ALS, which occurred proximally to motor neurons. To complement their mouse work, they conducted a parallel analysis in human cervical and lumbar spinal cord tissue from sporadic ALS patients with bulbar (n=4) and lower limb (n=3) onset disease. As in mice, transcriptomic dysregulation was more pronounced near the site of symptom onset. Pathway analysis identified numerous biological processes, among them “ECM (extracellular matrix)-receptor interaction”, “cell adhesion molecules”, “axon guidance”, and multiple immune “cytokine-cytokine receptor interaction”, “chemokine signaling pathway”, and “complement and coagulation cascades” in mouse and/or human, in alignment with our transcriptomics findings. Additionally, several metabolic pathways emerged, including “sphingolipid signaling pathway”, “cholesterol metabolism”, and “phosphatidylinositol signaling system”. Indeed, impaired metabolism is an ALS hallmark and correlates with changes in basal metabolic rates in ALS cases [27].

Thus, we have also conducted a metabolomics analysis of ALS participant plasma versus controls [25]. Metabolites represent the cumulative effect of genetics, epigenetics, transcriptomics, and proteomics regulation, and also lend insight on potential environmental exposure through xenobiotics. Pathway analysis demonstrated that impaired lipid metabolism was a strong undercurrent in ALS, especially in complex “sphingomyelins”, “ceramides”, and “hexosylceramides” and β-oxidation intermediate “fatty acid metabolism (acyl carnitine, polyunsaturated)” species. Additionally, “creatine metabolism” was a top pathway, but is likely secondary to muscle wasting in ALS, as was xenobiotics “benzoate metabolism”.

Moving forward, these studies will need validation in independent ALS cohorts and longitudinal profiling. However, they underscore the ability of systems biology for putting into focus possible pathways, which may lead to therapeutic developments. Additionally, corroboration in model systems will be required to establish causality of any putative candidates.

4.2. Can we identify early ALS diagnostic biomarkers?

Although ALS lacks effective treatment, earlier intervention may help outcomes [159]. Unfortunately, ALS patients are generally diagnosed after symptom onset, sometimes even months following the initial symptoms. Therefore, earlier diagnosis could benefit patients if they can access treatment earlier. Along these lines, numerous studies have assessed potential ALS biomarkers in accessible biofluids, either cerebrospinal fluid or blood/plasma (Figure 4B) [152]. Among candidates are neurofilament proteins, inflammatory molecules, and cystatin C, as well as molecules related to protein aggregates (TDP-43, SOD1) and genetic mutations (C9orf72 dipeptide repeats). Omics can also be employed to identify miRNA biomarkers; however, there is no consensus on a diagnostic miRNA panel, although manifold investigations have identified differential miRNAs in heredity and sporadic ALS versus healthy controls [22]. Metabolomics has similarly been proposed as a diagnostic toll; yet, again there is no consensus on a diagnostic metabolite panel, although altered lipid metabolism is a recurrent theme [25].

Most Omics investigations of ALS biofluids have shared the same weakness, namely, that analyses were performed on samples from patients that had already developed symptoms and been diagnosed with ALS. One notable exception is a plasma metabolomics investigation by Bjornevik et al. of five large cohorts comprising over 318,000 participants with banked blood samples [160]. Participants that developed ALS after their blood sample had been banked were identified (n=275), consented, and enrolled in the metabolomics analysis against matched controls (n=549). The study found 31 differential metabolites in ALS versus controls, including many lipid species spanning diacylglycerols, triacylglycerols, phosphatidylcholines, cholesteryl ester, and sphingomyelin, as we had observed [25]. When participants were stratified by time of blood draw, there were 63 and 41 differential metabolites in samples collected less or more than 5 years, respectively, from the time of ALS diagnosis [160]. However, none of these metabolites remained significant after accounting for multiple comparisons, although penalized regression methods (lasso and elastic net) identified several metabolites, frequently lipids, which predicted ALS with moderate areas under the curve values ranging from 0.58 to 0.74. The authors suggested several study weaknesses, among them the large number of detected metabolites (n=404) versus the relatively smaller sample size (n=275), which limited statistical power. However, the study does illustrate a way forward for leveraging Omics to identify early ALS biomarkers, though the rarity of ALS poses significant challenges.

4.3. Can we define the ALS exposome?

There are well-documented ALS mutations [5]; however, the vast majority of ALS cases lack a known genetic etiology, despite numerous genome-wide association studies, whole genome studies, and exome sequencing studies. This has led to the emergence of the gene-time-environment hypothesis of ALS, which posits that cumulative environmental toxic exposures over time superimposes on genetic susceptibility to trigger disease onset and progression [11]. Thus, there has been significant interest in defining the ALS exposome, the collective of environmental exposures, which increases risk of disease (Figure 4C) [148]. Indeed, our targeted environmental studies of potential environmental toxicants indicated a risk of ALS from pesticide, fertilizer, and persistent environmental pollutant exposures [8–10]. In our analysis of metals in teeth from ALS versus controls, we used laser ablation-inductively coupled plasma-mass spectrometry to assess early exposure to metals [161]. Metal levels were elevated in cases versus controls, after adjusting for sex, smoking, occupational exposures, and ALS family history, findings corroborated for copper in teeth from SOD1^G93A mice [162].

Exposome studies of ALS remain in the nascent stages. However, a high-profile metagenomic study in SOD1^G93A versus wild-type SOD1 mice provided important insight on a possible causative role of gut microbiome on disease progression [163]. Longitudinal analysis revealed early gut dysbiosis occurred in ALS versus healthy mice, which centered around multiple species. Honing in on 11 candidates, Blacher et al. decolonized the gastrointestinal tract of SOD1^G93A mice with antibiotics and inoculated mice individually with each of these focused species. Remarkably, inoculating SOD1^G93A mice with Akkermansia muciniphila slowed disease progression and significantly increased survival. In addition to pinpointing the microbial species associated with disease severity, Omics (metabolomics) also elucidated a possible mechanism by identifying pathways centered on nicotinamide metabolism. Similar findings were corroborated in human ALS participants, whose microbiome could be differentiated from healthy controls by principal coordinate analysis, as were serum and cerebrospinal fluid analyses of nicotinamide levels. We have also conducted longitudinal microbiome investigations of SOD1^G93A mice, similarly observing early gut dysbiosis, including of Akkermansia muciniphila [164]. Furthermore, gut dysbiosis correlated with immune cell infiltration into the brain and spinal cord.

Thus, systems biology can uncover how the exposome can exert an influence on ALS progression, with examples through the microbiome. However, broader studies involving additional candidate environmental pollutants, e.g., air pollution, untargeted pollutant detection, are needed to fully define the ALS exposome.

5. Conclusions and future directions

In this chapter, we illuminated several recent studies, which employed systems biology in neurological disease for multifactorial goals, spanning pathophysiology, treatment response, therapeutic candidate identification, biomarker discovery, and exposome research. These studies demonstrate the ability of systems biology for advancing our understanding of neurologic diseases and suggest prospects for drug development, which is especially crucial since most diseases lack effective treatments.

Critically, more longitudinal studies are needed to identify the earliest pathological changes, which would allow therapeutic targeting of upstream events for disease-modifying treatments, rather than downstream events (palliative care) (Figure 5A). This is challenging in human studies, especially during the prodromal phase [160], but can be readily accomplished in animal models [164,163,23]. Although the field of neurology is adopting these powerful Omics platforms, it lags behind the field of oncology, for example; however, it also ushers in the opportunity to learn from cancer research, clinical trials, and precision medicine for adoption in neurology. For example, Omics profiling has become well-entrenched in cancer research, where individuals with cancer are assigned specific treatments based on identified tumor mutations, and also participate in multiple novel clinical trials [165], a practice which could be adopted in neurological diseases, including the use of basket and umbrella trials (Figure 5B). Indeed, investigators are implementing newer clinical trial designs, such as platform trials, for neurological diseases [166,167], though this is lagging behind oncology trials.

Figure 5. Emergent and future systems biology research avenues in neurological diseases to address unmet medical needs.

(A) Longitudinal studies to identify the earliest pathological changes, prior to symptom onset, to therapeutically target upstream (disease-modifying) rather than downstream (palliative care) events. (B) Omics/multi-Omics profiling to stratify participants for clinical trials of targeted, mechanism-based drug candidates. Basket trials comprise participants with multiple diseases harboring the same actionable mutation, and testing a single targeted drug candidate. Umbrella trials comprise participants with a single disease harboring multiple actionable mutations; participants are profiled and matched to a targeted drug candidate out of number of possible candidates. (C) Omics/multi-Omics profiling to develop molecular-based, rather than phenotype-based, diagnostic criteria and treatment selection for neurological illnesses. (D) Development of biomarkers or biomarker panels for early diagnostics of neurological illnesses. (E) Precision medicine to match patient Omics/multi-Omics profile with approved drugs. Created, in part, with BioRender.com.

Omics/multi-Omics profiling can be employed to develop molecular-based, rather than phenotype-based, diagnostic criteria and treatment selection for neurological illnesses (Figure 5C). Although there are robust molecular tests for inherited PN, molecular tests for other neurological diseases, for example C9orf72 expansion in ALS, could also have diagnostic and treatment related implications. For instance, rather than bulbar versus spinal onset ALS, a molecular classification might additionally provide guidelines for future targeted treatment. This also includes generating molecular insight into idiopathic and sporadic neurological diseases. Finally, earlier studies during prodromal phases are needed to develop biomarkers or biomarker panels for early diagnostics of neurological illnesses (Figure 5D). Omics/multi-Omics could bring the promise of precision medicine to bear in neurological diseases by matching patient profiles to approved drugs likely to be effective, e.g., pain medications for painful PN (Figure 5E). Thus, systems biology applications in neurological illnesses has a track record of success and a bright future for novel upcoming directions.