Computational modeling of brain pathologies: the case of multiple sclerosis

Abstract

The central nervous system is the most complex network of the human body. The existence and functionality of a large number of molecular species in human brain are still ambiguous and mostly unknown, thus posing a challenge to Science and Medicine. Neurological diseases inherit the same level of complexity, making effective treatments difficult to be found. Multiple sclerosis (MS) is a major neurological disease that causes severe inabilities and also a significant social burden on health care system: between 2 and 2.5 million people are affected by it, and the cost associated with it is significantly higher as compared with other neurological diseases because of the chronic nature of the disease and to the partial efficacy of current therapies. Despite difficulties in understanding and treating MS, many computational models have been developed to help neurologists. In the present work, we briefly review the main characteristics of MS and present a selection criteria of modeling approaches.

immune system, systems biology, vaccine, artificial intelligence, bioinformatics, computational models

Introduction

In vertebrates, the nervous system is the part of the body that coordinates voluntary and involuntary actions and transmits/receives signals to and from different parts of the body. A nervous system is present in most of the animals living on Earth, and according to evolutionary models, it first arose in wormlike organism about 550 to 600 million years ago. In vertebrate species, it consists of two main parts: the central nervous system (CNS), which includes brain and spinal cord; and the peripheral nervous system.

The human brain is considered as the most complicated structure in the known universe and most complex organ of the body. One can infer the complexity of the brain simply by considering the number of brain’s entities, namely, neurons, nonneuronal glial cells and synapses. It has been shown that human brain has ∼8.6×10¹⁰ neurons and almost the same number of nonneuronal glial cells. Each neuron can be connected to as much as 10⁴ other neurons; thus, connections could potentially reach up to 10¹⁴. The existence and functionality of plethora of molecules in human brain are still ambiguous and largely unknown [1], thus posing a challenge to human brilliance.

As human brains direct all the other parts of the body to react on different stimuli of the environment, the interplay between environment and genetic information is crucial for brain activity, and the role of environment is significantly important in its well-being. The interaction of both of these factors is called gene–environment interaction. One interesting and simple example of environmental role on our health maybe represented by the risk of nonmelanoma skin cancer during prolonged exposure to sunlight, which is significantly higher in white people when compared with dark skin people [2]. Seemingly, sunlight exposure promoting vitamin D synthesis may play a role in dampening neuroinflammation in autoimmune brain disorders.

Neurodegenerative diseases impair the brain functionality, and they are difficult to treat because of our limited knowledge of underlying molecular pathology. Neurodegenerative diseases are an area of major concern of health care providers, as they are burden on the social system and jeopardize patients’ independence. Many initiatives have been taken to understand brain functionality and to better tackle neurodegenerative diseases. Human Brain Project, BrainInitiative, Allen Brain Atlas, Blue Brain Project and BrainMaps are few among countless projects [The Human Brain Project – Human Brain Project (https://www.humanbrainproject.eu/); Brain Research through Advancing Innovative Neurotechnologies (BRAIN) – National Institutes of Health (NIH; http://braininitiative.nih.gov/); Allen Brain Atlas – Home (http://www.brain-map.org/); Bluebrain | EPFL (http://bluebrain.epfl.ch/); BRAINMAPS.ORG – BRAIN ATLAS, BRAIN MAPS, BRAIN STRUCTURE, NEUROINFORMATICS, BRAIN, STERE OTAXIC ATLAS, NEUROSCIENCE (http://brainmaps.org/)].

The ongoing research and publications associated with neurodegenerative diseases are growing at a higher rate (Figure 1), but unfortunately, there is a substantial unmet need for drugs in such field. Many attempts have been made to develop drugs, which can treat neurodegenerative disorders, but despite some of them were proven effective in experimental models they do not confirm their efficacy in humans.

Figure 1.

Publication growth record of neurodegenerative diseases, created using GoPubMed (Doms and Schroeder, 2005) queried with ‘Neurodegenerative diseases [mesh]’ on 4 January 2016, http://www.gopubmed.com/.

Even though there is a remarkable amount of scientific literature available on neurodegenerative diseases, we are still struggling to treat patients, and this concept may be well summarized in Poland’s statement ‘We're drowning in a sea of data while dying of thirst for knowledge’ [3]. The quest of transforming ‘data into knowledge and wisdom’ can be obtained by changing the ways we look into and interpret it.

Multiple sclerosis

Multiple sclerosis (MS) is one of the most common chronic inflammatory diseases affecting CNS. MS is one of the most common disorders affecting CNS, and affects 2.3–2.5 million people worldwide (Atlas of MS [http://www.msif.org/about-us/advocacy/atlas-of-ms/] link verified on 04.01.2016, last updated 13th April 2015). MS commonly starts at an early age and makes people disabled through the years up to a wheelchair or bedridden status. The typical course of MS is relapsing–remitting one, eventually changing to a chronic progressive type with accumulation of disability. However, the evolution over time of MS course is not predicable and can only be diagnose based on clinical features. It causes significant disability and has a considerable influence on patients’ life, with respect to social impact, cost and quality of life deterioration.

MS usually begins between 20s and 50s, and the ratio between male and females is 1:2 [4]. Neuropathological hallmarks of MS are demyelinating lesions (plaques), which mostly occur in the white matter of brain and spinal cord. However, besides demyelination, other MS features are inflammation and neurodegeneration, and the end stage of this complex scenario is brain atrophy [5].

The etiology of MS is still unknown, diagnosis is not always simple, disease course is unpredictable and therapeutic response may be subject to change from patient to patient [6]. There are many factors potentially concurring in MS pathogenesis including viral infections [7], epigenetical changes [8], mitochondrial defects [9], vitamin D [10] and genetic predisposition [11]. The diagnosis is also challenging, as physicians have to rule out other ailments (as many other conditions may mimics the symptoms), and even the most experienced experts can reach a correct diagnosis in only 90–95% of cases, as there is no single diagnostic test specific for MS. The course of the disease at onset maybe relapsing–remitting or relentlessly progressive, but, as years go by, a secondary progression may replace the relapsing course. This change in disease course is still unpredictable, and there are no indicators of such an event. Although >15 drugs are now available for the treatment of MS, the therapeutic response varies among patients, and there are still difficulties in selecting drugs according to patient features and timely forecasting the therapeutic response.

The etiology of MS is still unanswered; however, a number of factors are now associated with the disease and considered relevant in its pathogenesis. Among these factors, few gene polymorphisms, Epstein–Barr Virus (EBV) infection, vitamin D levels and cigarettes smoking are those most strongly associated with MS appearance [12]. Despite there are no genes clearly predisposing to MS, the disease may recur within the same family, and the risk of developing MS depends on the closeness of the relation of the diseased person, being 30% in identical twins, 5% in nonidentical twins and 2.5% in siblings [4, 13].

Furthermore, sex and geographical factors may also play a role, as MS is more common in women and in people who live farther from equator [14].

There are also factors that could halt or trigger a relapse: pregnancy, specially the last trimester, is associated with a reduced risk of relapse [15], whereas infections, mostly viral ones, may promote relapses [16–19].

The clinical forms of MS are classified according to disease course:

• Relapsing–remitting multiple sclerosis (RRMS) is the most common disease course and is characterized by clearly defined attacks of worsening neurologic function. These attacks—called relapses, flare-ups or exacerbations—are followed by partial or complete recovery periods (remissions), during which symptoms improve partially or completely, and there is no apparent progression of disease. Approximately 85% of people with MS are initially diagnosed with RRMS.

• Secondary progressive MS. This course usually follows the initial relapsing–remitting course with or without superimposed relapses.

• Primary progressive MS (PPMS) is characterized by steadily worsening neurologic function from the beginning. Although the rate of progression may vary over time with occasional plateaus and temporary, minor improvements, there are no distinct relapses or remissions. About 10% of people with MS are diagnosed with PPMS.

• Progressive-relapsing MS (PRMS) is characterized by steadily progressing disease from the beginning and occasional exacerbations along the way. People with this form of MS may or may not experience some recovery following these attacks; the disease continues to progress without remissions. PRMS is the least common of the four disease courses.

The neurological damage is mostly because of three combined mechanisms: demyelination, neurodegeneration and autoimmunity.

Myelin damage (or demyelination) is a term used to define a condition where myelin sheaths of axons get damaged, and electrical impulse is lost during transportation. Different patterns of myelin damage have been documented, e.g. myelin stripping, dissolution of myelin sheath by invasion of macrophages [20] and binding of myelin fragments to vesicles of macrophages [21]. Demyelination causes signal deterioration and, as a consequence, malfunctioning of organs that are supposed to be innervated by injured nerve fibers. Demyelination is most likely the main mechanism of neurological damage in the early phase of the disease. Neurodegeneration in MS is represented by loss of axons and, to a minor extent, also of nerve cells. Neurodegenerative processes in MS occur as direct damage by autoimmunity and inflammation or as indirect effect of myelin loss, oxidative reactions and energy exhaustion [22–29].

Autoimmunity in MS is considered to be driven by two different subsets of T helper cells, TH1 and TH17 [30], whereas the other T helper subset represented by TH2 cells maybe relevant in the remission phases. This polarization is also supported by the beneficial effect of interferon beta and glatiramer acetate whose mechanism of action promote T cell shifting to TH2 phenotype [31, 32]. In addition to TH cell populations, macrophages also have a similar pattern, which results in different cellular phenotypes [33] based on the population of different cell type. M1 macrophages are considered to have a pro-inflammatory role, and on the other hand, M2 macrophages have an anti-inflammatory role and they contribute to tissue repair. Interestingly, glatiramer acetate can induce a population shift toward M2 phenotype in MS [34] implying that a shift toward certain cell types is possible therapeutically.

MS is mainly considered as an autoimmune disease, as axon injuries, demyelination and neurodegeneration are triggered by autoimmune response particularly by T cells. One of the strong arguments to support that autoimmunity plays a crucial role in MS is that the therapeutic agents that reduce the disease severity act by suppressing autoimmune reactions [35]. For example, natalizumab, fingolimod and interferon are immune-modulating drugs. Natalizumab is a selective adhesion molecule inhibitor that binds to a4b1-integrin and blocks its interaction with vascular cell adhesion protein 1 molecule expressed by activated lymphocytes. As a result, migration of autoreactive lymphocytes into brain tissue is blocked [36].

Fingolimod (FTY720) undergoes rapid phosphorylation in vivo by sphingosine kinase 2 to produce fingolimod phosphate, which binds to four of the five S1P receptors with high affinity and exerts its therapeutic effect through modulation of S1P receptor. Its mechanism of action is thought to provide therapeutic benefit in MS by preventing normal lymphocyte egress from lymphoid tissues, thus reducing the infiltration of autoaggressive lymphocytes into the CNS [37]. Interferon beta (IFNß) was the first therapeutic agent discovered for the treatment of MS, but its mode of action is only partly elucidated [38].

Supplementary evidence for the major role of autoimmunity in MS comes from the association of the disease with selected HLA genes [39]. Over the years [40–42], there have been identified Class II risk alleles (HLA-DRB1*15:01, HLA-DRB1*13:03, HLA-DRB1*03:01, HLA-DRB1*08:01 and HLA-DQB1*03:02) and Class I protective alleles (HLA-A*02:01, HLA-B*44:02, HLA-B*38:01 and HLA-B*55:01). In addition, the analysis of high-density single-nucleotide polymorphism data on several controls from 11 cohorts of European ancestry provided evidence for two interactions involving pairs of Class II alleles: HLA-DQA1*01:01–HLA-DRB1*15:01 and HLA-DQB1*03:01–HLA-DQB1*03:02 [43].

Modeling applications in MS

Modeling is the human activity consisting of representing, manipulating and communicating the world we observe in daily life, for a personal purpose. Modeling in science follows the same rules of modeling in daily life, but the approach is rationale and techniques are different.

Scientific interest is in modeling systems: a collection of interrelated elemental units (objects) whose internal structure is either unknown or does not exists. There are a lot of techniques used to represent a system, in this case a biological system, based on computational and mathematical approaches [44–46]. Typically, the choice for a specific technique depends on the problem scenario we have to deal with. MS represents a biological scenario in which many biological subsystems interact each other, and, unfortunately, two of them are the most complex ones, i.e. the nervous system and the immune system. Nonetheless, many attempts have been done to date trying to deal with different aspects of the disease, ranging from pathogenesis dynamics to response to a particular treatment, from immune system changes in the course of the disease to models including environmental and genetic factors into the dynamics of the MS. In what follows we present a scenario of these modeling approaches.

Current view about MS pathogenesis includes both genetic predisposition and environmental risk elements. Three consecutive factors are associated in the environmental risk. The first acts near birth, the second acts during childhood and the third acts long thereafter. Two candidate elements (vitamin D deficiency and EBV infection) appear well matched to the first two environmental events. Ascherio et al. [47] elaborated a mathematical model for MS pathogenesis where environmental and genetic factors interact into a causal scheme that can clarify some of the current variations in MS epidemiology. The work suggests that the causal pathway that leads to adult MS involves sequential environmental events or factors. So acting in this sequence could, potentially and theoretically, totally prevent the disease. Such a therapeutic strategy is difficult to be undertaken for a number of reasons, ranging from low incidence of MS to the requirements of thousands of individuals for many (>30) years and others.

Transcription profiling studies show important understandings in regard to molecular events, which may lead to phenotypic results such as response to drug therapy. Development of computational models that may be helpful to predict therapy response is merely possible when detailed data measurements, robust feature/gene collection and cutting-edge computational modeling approaches are merged with rigorous statistical proof and large-scale confirmation of results. Owing to the huge number of gene expression measurements in transcriptional profiling studies, feature selection denotes a bottleneck when creating computational models. The trade-off between selection of the ideal feature set and computational efficiency results in many choices for candidate gene sets, which leads to a wide range of classification accuracies.

Mostafavi and co-workers [48] present OSeMA, a fast, robust and accurate gene selection–classification framework that results in construction of classification models that are highly predictive of the recombinant IFNβ (rIFNβ) therapy response in MS patients. The authors followed three phases during the development and the validation of their model. First, they used a data set obtained by quantitative reverse transcription polymerase chain reaction from peripheral blood mononuclear cells containing the expression levels of 77 genes in 52 RRMS patients treated with rIFNβ. Then, OSeMA was applied to the gene expression data of RRMS patients at time zero; finally, they classified validated and identified minimal gene groups, whose combinatorial transcriptional profiles distinguish poor- and good-responding RRMS patients to the therapy.

Despite the advent of genome-wide association studies (an approach to look for associations between hundreds of thousands genetic variations and diseases), many questions remain about the causative mechanisms of multifactorial diseases. In particular, the effect size of the genetic variants identified through these studies explains relatively little of the heritability (the proportion of total variance in a population for a particular measurement, taken at a particular time or age, that is attributable to variation in genetic values) of most complex traits. MS is one of the diseases that best exemplify this key problem. By observing the course of this disease in its relapsing–remitting phase, Bordi and collaborators [49] presented a mechanistic model with a random forcing that was able to describe the main features of the disease course. They concluded that if the model is applied to the events that precede the disease (i.e. those that derive from the effects of the genetic and environmental risk factors), it may explain how relatively small effects may be amplified by comparably small random perturbations.

Compared with other neurological diseases, the diagnosis of MS is complex. Indeed, a computerized method that helps in its diagnosis can be useful. A software that helps in making correct decisions in different activities is called decision support system (DSS). When correctly developed, a DSS can be thought as an interactive software-based system planned to assist decision-makers assembling useful information from raw data, documents, personal knowledge and/or business models to classify and unravel problems and make decisions. A clinical decision support system (CDSS) is a health information technology system that is designed to provide physicians and other health professionals with clinical decision support, that is assistance with clinical decision-making tasks. A working definition has been proposed by Robert Hayward of the Centre for Health Evidence: ‘Clinical Decision Support systems link health observations with health knowledge to influence health choices by clinicians for improved health care’. Linder and co-workers [50] propose a CDSS that aims to provide a support based on cerebrospinal fluid and blood findings to improve their value and ease the diagnostic procedure of chronic inflammatory diseases of CNS in which MS is included. The proposed system uses a combination of univariate analysis, multiple logistic regression and artificial neural network methodologies.

Biochemical systems theory (BST) is a mathematical and computational framework for analyzing and simulating systems [51]. It was originally developed for biochemical pathways, but it has become much more widely applied to systems throughout biology and beyond. BST is called ‘canonical’, which means that model construction, diagnosis and analysis follow stringent rules. The key ingredient of BST is the power-law representation of all processes in a system. Broome and Coleman [52] developed a mathematical model for MS using BST. Even if this framework uses specific literature results, it has a limited focus on the action of reactive oxygen and nitrogen species, the permeability transition pore, apoptotic factors and the eventual cell death in the oligodendrocyte. It can then be used to analyze, even if only theoretically, potential drug therapies and possible trigger points for the MS disease.

The main histopathological hallmarks of MS are represented by lesions including inflammatory cells, myelin sheath damage and axonal loss. In their work, Mohan and co-workers [53] proposed a model made of a pathological process that causes cellular damage and programmed cell death initiated through an intercellular signaling component. They simulated the CNS network in two-dimensional using an undirected, fixed radius random graph G(n,r), with n nodes (vertices) and radius of connectivity; fixed radius implies that nodes are connected only if they are within a distance of r. Biologically speaking, the nodes of the graph can be viewed as representing cell bodies or functional units, and the edges (bonds) of the graph can be viewed as axons or the interconnections between functional units. The pathological and regeneration processes are driven by probabilistic events. When signals are propagated, the signals received on a node are summed, and they will be propagated further when the summed signal strength reaches a predetermined threshold. The amplitude of the signal propagated along an edge is equal to its edge weight. The spread of the pathologic process was driven by the success rate in causing cellular damage. The simulations demonstrate that the spread of the pathologic process can be arrested by programmed cell death in the periphery of the lesions.

In a recent work by Pennisi and colleagues [54], the authors describe for the first time an agent-based model able to simulate the main dynamics of MS. In particular, they consider one of the more common forms of MS, i.e. the RRMS taking into account the cross-regulation between the two cells populations, coupled with an external agent (EBV) supposed as the cause of chronic inflammation. They want to analyze whether such system is able to show stable oscillatory behaviors in healthy patients, and presence of unrecoverable neural damage in patients with a malfunction in the cross-regulation mechanisms between effector T cells and regulatory T cells. The model is rather simple, as it includes only three classes of entities. However, agent-based models are flexible, and the model shown by Pennisi and co-workers can be expanded to include a more detailed description of the immune system entities and functions or environmental factors, such as, for example vitamin D (in a recent work, Pappalardo and collaborators proposed a model that analyze the role of vitamin D in MS dynamics [55]). The model allows to grasp how the role of regulatory versus effector cells and their internal dynamics is crucial in the understanding the evolution of neural damage in the progression of the pathology.

In the dynamics of MS, an important aspect is covered by the role of the blood–brain barrier (BBB), a diffusion barrier necessary for the normal function of the CNS and it is made of endothelial cells, astrocyte end-feet and pericytes. Moreover, tight junctions situated between cerebral endothelial cells selectively exclude most blood-borne substances from entering the brain [56]. During brain diseases such as MS, the disorganization and consequently the impermeability of this brain solute barrier are compromised. The mechanisms of breakdown of BBB are not yet clear, but the involvement of inflammatory cytokines released from CD4+ T cells is likely, as a collateral effect of autoimmune attack to the myelin–oligodendrocyte complex [57].

In a recent work [58], the authors extended their previously developed agent-based model to evaluate the effects of potential treatments that may counteract mechanisms involved in the BBB breakdown and promote its functional recovery.

Conclusions

Brain is one of the most complex structures of the universe: to model that level of complexity, a framework, which could handle it is needed. Systems biology provide multilevel platforms to integrate, analyze and simulate models/data sets from different omics studies. By recent times, neuroscientists have worked only with reductionist approaches by analyzing brain based on functionality, cellular composition and parts. Even though reductionist methods yielded some success, it does not appear the best way to approach multi-factorial neurological diseases as MS. A parallel example would be the attempt to understand memory, learning and behavior by only observing neurons or any other individual cell type of the brain. Many disciplines such as neuroinformatics, computational neuroscience and neurophysiology aim to decode brain functionality by shifting toward systemic approaches gradually. The ultimate objective of systems biology in neurodegenerative diseases is to identify pathways involved in disease pathogenesis by analyzing networks constructed on different data sets, e.g. genomics, transcriptomics, proteomics and neurobiological experiments so to help discovering potential sites of interventions for new drugs.

Quantitative models require an extensive knowledge about their components and mathematical equations. Quantitative modeling is based on statistical/mathematical methods; its main activity is to interpret numerical data and develop mathematical models on which simulation and hypothetical prediction can be generated. It also makes use of measurements and manipulation of variables to see the global effect of numerical value change. Often quantitative part of modeling comes later than qualitative part, as we need to have a system, which specifies the objects, their basics, qualitative interrelationships and underlying hypotheses before enriching them with numbers and equations. Quantitative models are efficient for modeling system dynamics and providing accurate prediction if sufficient data are available, which is unfortunately not the case of neurological disorders. However, computational disease modeling may become an important method to understand systems biology and may reveal helpful for the struggle of pharmaceutical companies to find solutions for unmet needs and develop new treatment options for patients affected by complex diseases such as MS.

Key Points

• MS represents a major neurological disease that affects a lot of people.

• Computational modeling can help in finding novel therapeutic strategies for MS.

• This work reports both main MS characteristics and a presentation of the main modeling approaches.

Francesco Pappalardo is Associate Professor at the University of Catania. He was a visiting research scientist at the Dana-Farber Cancer Institute in Boston and at the Molecular Immunogenetics Laboratories, IMGT in Montpellier. His major research area is computational biomedicine.

Abdul-Mateen Rajput is a PhD student in Life Science Informatics Department at the Bonn University. He has done several research projects, including constructing disease models to find a cure of a neurodegenerative disease, developed several methodologies for automating complicated data mining tasks and published many papers during his work. His research interests focus on disease modeling, knowledge engineering and semantic Web technologies.

Santo Motta, Laurea in Physics (University of Catania,1970) and MSc in Applied Mathematics (University of London, Queen Mary College, 1971). Associate professor of Mathematical Physics (Faculty of Pharmacy, Department of Mathematics and Computer Science, University of Catania). His present scientific interests are BioMaths, BioComputing and BioInformatics.

Acknowledgements

The authors would like to thank Dr Luca Toldo for his useful comments and suggestion. Moreover, the authors would like to thank the reviewers’ suggestions that allowed to greatly improve this work.