From Exercise Physiology to Network Physiology of Exercise

Abstract

Exercise physiology (EP) and its main research directions, strongly influenced by reductionism from its origins, have progressively evolved toward Biochemistry, Molecular Biology, Genetics, and OMICS technologies. Although these technologies may be based on dynamic approaches, the dominant research methodology in EP, and recent specialties such as Molecular Exercise Physiology and Integrative Exercise Physiology, keep focused on non-dynamical bottom-up statistical inference techniques. Inspired by the new field of Network Physiology and Complex Systems Science, Network Physiology of Exercise emerges to transform the theoretical assumptions, the research program, and the practical applications of EP, with relevant consequences on health status, exercise, and sport performance. Through an interdisciplinary work with diverse disciplines such as bioinformatics, data science, applied mathematics, statistical physics, complex systems science, and nonlinear dynamics, Network Physiology of Exercise focuses the research efforts on improving the understanding of different exercise-related phenomena studying the nested dynamics of the vertical and horizontal physiological network interactions. After reviewing the EP evolution during the last decades and discussing their main theoretical and methodological limitations from the lens of Complex Networks Science, we explain the potential impact of the emerging field of Network Physiology of Exercise and the most relevant data analysis techniques and evaluation tools used until now.

1. Introduction

Exercise physiology (EP), the study of how the body adapts physiologically to the acute and chronic stress of exercise or physical activity, has evolved extensively since the beginning of the early twentieth century. Due to an increased interest in exercise and health, initially motivated by the poor physical capacity of soldiers, is today a scientifically founded branch that provides the basis of physical fitness, exercise performance, training, testing, and rehabilitation programs addressed to all types of population, including elite athletes and clinical patients. Its potential to enrich Basic Physiology and diverse fields such as Sports Medicine, Sports Rehabilitation, Sport Science, or Training Science is still undervalued and has to be rediscovered under the framework of Complex Systems and Network Science approaches [1].

In this chapter, the present and future of EP will be overseen from a historical and scientific perspective. The main limitations of the EP available evidence-based research, strongly influenced by excessively simplified theoretical and methodological assumptions, will be discussed using the example of the exercise-induced fatigue. Finally, the research approach of the new emerging field of Network Physiology of Exercise, focused on the coordination and integration among physiological systems across spatiotemporal scales (from the subcellular level to the entire organism), will be presented.

2. Evolution of Exercise Physiology

It is essential to understand the EP history when approaching the future [2]. As any scientific branch, the EP evolution has been constrained by multiple and multilevel factors acting at different timescales such as financial possibilities, organizational and ideological positions. Although historical data pertaining to EP spans more than 2000 years, first research contributions correspond to the early twentieth century, which was characterized by an increasing specialization and sub-specialization in many scientific fields. This state of affairs brought about a flood of fragmentation in science that promoted the naissance and development of the main EP research labs in the world.

First works, initiated by Scandinavian scientists, were related to metabolism and heat production during exercise and recovery. Maximal oxygen uptake was described as the upper limit of performance [3], and lactate production (from glucose metabolism) was related to fatigue [4]. Research was focused later on circulation, muscle physiology, or environmental physiology and provided the basis of exercise as medicine. While the major concern of research after the World War II was the health and fitness of soldiers, the most recent concern is the obesity epidemic and other diseases related to the food abundance and lack of physical activity.

With the development of labs and the creation of world organizations such as the American College of Sports Medicine (ACSM), and the European College of Sport Sciences (ECSS), the field of EP has become enormously specialized in the last years, and EP researchers usually work in one area (e.g., cardiovascular, muscular, etc.). This has produced a loss of the original essence of Physiology, the unique branch of Biology specifically dealing with synthesis and integration. Although technological advances have led to create more sophisticated and better equipped labs, the type of inquiry and research focus of EP has been kept in general quite immutable, and clinical exercise physiologists keep mainly directed to testing energy production (e.g., aerobic power, anaerobic thresholds, etc.) [5].

Influenced by reductionism, Biology has traditionally emphasized: the decomposition of systems responsible for a given phenomenon into component parts and processes. Identifying such components and describing their mechanisms apart during exercise have been one of the main EP endeavors. Using a range of experimental models from cells to animals and humans, main approaches have laid the description of the biological mechanisms of temporary and persistent functional changes in response to acute and repeated exercise [5, 6]. It is worth noticing that despite the evolution of biology and the technological advances of the last years, the initial theoretical assumptions of EP have kept almost intact.

Even if “why” questions, related to teleological explanations, have been traditionally avoided because the final purpose of physiological systems is assumed to be unknown or nonexistent, research questions often reflect the excessively simplistic assumptions that have characterized EP from the very beginning (see Section 3 for further detailed explanation). The linear and reductionist approach of the scientific production is reflected in the redundant expression “Effects of,” highlighted in the content analysis done on over 22.000 ECSS abstracts submitted during almost two decades [7]. This expression reflects a very specific mode of inquiry, and its consequential data acquisition tools, analysis techniques, and inductive methods, commonly and uncontroversially used in EP research.

A century of reductionist research has produced a lot of information and descriptive knowledge, some obtained through very well-designed experiments, but has provided only a partial understanding of exercise-related phenomena, and led to several controversial findings. In fact, some of the main questions still remain without clear responses. For instance, which are the limits of performance, what limits VO₂max, what causes fatigue, why are there responders and non-responders to exercise, etc. The extant controversies seem strongly affected by the excessively simplified theoretical and methodological assumptions that characterize the field.

2.1 Molecular Exercise Physiology. Are explanations only in the cell?

Due to the lack of clear responses to the main topics obtained investigating at system and organ levels (e.g., cardiovascular system, muscle), the reductionist rationale led EP investigation, and medical investigation in general, in which disease is increasingly understood in molecular terms, toward microscopic levels (Molecular Biology, Genetics, and OMICS technologies). The acknowledgment of the role of exercise on health status and the pathway toward personalized medicine has also reinforced the micro-level research focus of EP [8, 9, 10, 11].

While the 1970s were the decade of Biochemistry, the 1980s represented the Molecular Biology era. Technological advances played a fundamental role in this evolution. The introduction of DNA microarrays, a fast technology to study thousands of DNA and protein molecules simultaneously, supposed a revolution in biological research. Coupled with computational methods, pushed the development of Systems Biology (e.g., [12]), a branch that focuses on complex interactions within biological systems, and enabled to investigate the behavior of the genes of an organism under different conditions [13]. The identification of new biomarkers, the improved sensitivity and specificity of the existing ones, and new insights into the personalized therapeutic strategies to improve athletic performance and human health through precision exercise medicine is the main aim [14].

Research in Molecular Exercise Physiology and “sportomics” [15, 16, 17] is mostly focused on omics data collection and analysis efforts to catalog exercise-regulated pathways. Although Molecular Biology dwells on dynamic principles, the dominant research methodology in Molecular Exercise Physiology, and Integrative Physiology [10, 18] keep focused on non-dynamic bottom-up group-pooled statistical inference modes of inquiry. Main properties of CAS as synergies, established not only horizontally (e.g., among molecules) but vertically (among molecular, cellular, tissular, organ, and system levels), are neglected. The embeddedness of lower levels in upper levels, the circular causality (bottom-up, top-down) relationship among the levels, the different timescales of activity, the nonlinear dynamic processes that suffer qualitative changes through self-organization are also some of the main neglected properties.

Many component processes can lose or gain on significance during exercise; for instance, as fatigue develops (see Section 2.2). Physiological and psychobiological synergies compensate critical quantitative values registered at micro-level keeping stable behavioral variables registered at action level. In biological systems, the same effect at microscopic level can be produced by many different macroscopic phenomena. In addition, genes are dominantly pleiotropic, that is, the same gene can be involved in different physiological effects and states. Hence, the informativeness of the microscopic (Molecular Biology levels) may be at best an initial point in a much more elaborate study of the organism-environment interaction to conclude on the real macroscopic phenomenon that is involved in a health or performance problem.

Personal health and performance cannot be reduced to molecular and genetic levels as medicine cannot be geneticized. In addition, the network relations do not operate only bottom-up, but also top-down, that is, from the entire person to the genes level following the circular causality property of complex systems [19].

The view that everything can be explained at microscopic (molecular) level directly implies that health can be intervened and repaired only at this level, that is, through pharmacological substances. However, a person-environment approach [19] implies that health and performance are products of the interaction of many different levels, and health can be also improved intervening at environment or psychological level. In fact, interventions at macroscopic (social, psychological) level have been proven as crucial in changing the processes at subcellular level [20, 21]. Results seems to show that in a sick society, where more often than not, the competition for socially imposed success becomes a goal for itself, there is an increased likelihood of cell aging and poor personal health [22]. In multilevel complex networks, the macroscopic ambience strongly constrains its embedded components [23, 24]. While the use of pharmacological substances may be promoted somewhat by the big financial benefits that lie behind, mental interventions, usually cheaper and requiring only the development of self-knowledge and self-discipline [20], receive in general less scientific attention. In particular, exercise is a privileged type of intervention because it may affect all personal levels in a correlated cascade way [7, 25].

Systems Biology and Integrative Physiology strive for the same goal: to understand Biology whole-istically [18]. Systems Biology is focused on systems operating at a cellular level and has evolved over the past decade (called the “omics era”) as a direct result of advances in high-throughput molecular biology platforms and associated bioinformatics. In fact, Systems Biology has been described as: “the study of an organism, viewed as an integrated and interacting network of genes, proteins and biochemical reactions which give rise to life.” Instead of analyzing individual components or aspects of the organism, such as sugar metabolism or a cell nucleus, systems biologists focus on all the components and the interactions among them, all as part of one system. These interactions are ultimately responsible for an organism’s form and function [26].

Physical exercise dictates the magnitude and pattern of how networks of genes, proteins, and biochemical reactions will integrate and interact. This is an important point, because to the exercise physiologist most, if not all, cellular-network-based change will be secondary to the physiological stimulus causing that change, e.g., muscle contraction, rather than originating at the level of the network per se. This in essence is one important difference between the molecular biology focus at the core of Systems Biology and functional feedback approach of Integrative Physiology, a difference eloquently described by Noble [27]. In conclusion, although there is a tremendous potential for omics approaches to fill critical gaps in our understanding of the integrative networks underlying the health benefits of exercise [28] and allow going beyond the one-size-fits-all model of prescription [29], the complexity and interconnectedness of exercise biological networks cannot be unraveled and understood by studying single tissues or molecular targets alone. They require dynamic, multilevel, and global approaches. For instance, whether molecular processes can inform about the level of stress, the macroscopic phenomenon of stress cannot be explained only at microscopic level.

2.2 From microscopic to systemic hypotheses in Exercise Physiology. Example of exercise-induced fatigue research

The research performed to respond to the question about the causes of exercise-induced fatigue and spontaneous task failure is a good example of the current tendencies in EP research.

Over the last century, physiologists have tried to find the etiology and underlying mechanisms of exercise-induced fatigue [30]. Despite a wealth of knowledge about individual components intervening in the fatigue process and their adaptation to different types of exercise, they have failed to detect a single component or process responsible of the phenomenon and the limits of exercise tolerance in general [31]. The questions of “what causes exercise-induced fatigue” or “are limiting mechanisms central or peripheral, are there in the brain or in the muscle [32, 33]?” are clear examples of the type of inquiry searching for cause-effect relationships and the fragmentation tendencies derived from the reductionist models applied to the EP research.

The research investigating central and peripheral mechanisms of exercise-induced fatigue has not provided either a clear response to the question [34, 35, 36, 37]. The impaired action potential propagation, the inhibition of reflex mechanisms, the stimulation of chemical and nociceptive afferent signals, the corticospinal stimulation changes, the increase in extracellular serotonin, the cytokines liberation, the muscle acidosis, the accumulation of NH₄, H+, Mg₂+, Pi, the hyperthermia, the inhibition of Ca₂+ liberation, the glycogen reduction, the increase of K+ and free radicals are all associated processes to the fatigue development but cannot explain it.

In a similar way, assumed cause-effect or dose–response relationships among biochemical and performance variables have been proven to be often wrong. For instance, lactic acid, initially thought to be the consequence of oxygen lack in contracting skeletal muscle and related to the limits of high-intensity short-duration exercise, now it is recognized as being formed under fully aerobic conditions and associated to ergogenic and antifatigue properties [38, 39].

Fatigue is a macroscopic phenomenon and reflects itself in macroscopic behavior of performers [40]. Microscopic processes associated to it are not linearly independent (as it is usually tacitly assumed), and their total effects cannot be treated as a sum of individual effects. In particular, when knowing that there is a circular causality spread over the levels in all complex systems. Instead of focusing on isolated central and peripheral processes, the exercise-induced fatigue and task failure can be studied at behavioral coordination level, which integrates all network levels [41]. Using a macroscopic kinematic variable extracted at action level, the authors studied the time-variability properties of the elbow angle, considered as order parameter, during an effort performed until exhaustion. Critical behavior such as critical slowing down, enhancement of fluctuations, and correlation enhancement in interlimb coordination was reproducibly observed [42]. In this way, fatigue was understood as a process that leads to a system-level phase transition (spontaneous task disengagement), due to the circular causality mechanism that spreads over the levels in CAS.

In this way, it was hypothesized that the spontaneous task failure consists of a percolation process, produced by the impaired ability of the psychobiological network system to make the necessary short-term adjustments for negotiating the imposed external workload. In this scenario, the spontaneous task failure/disengagement, represent a giant (at systemic level), protective inhibitory fluctuation that causes a temporary abrupt switch to a lower energy expenditure level, a critical phenomenon prominent in complex systems, such as human psychobiological networks. This means that the loss of stability at systemic level is the cause of the task disengagement and not a singular process at singular level that can be pinpointed.

3. Limitations of Exercise Physiology. Contrasting approaches from complex network science

In contrast to decomposition, EP has paid much less attention to re-composition of physiological mechanisms [43]. The greatest challenge today, not just in Biology but in all of science, is then to reassemble such decomposed mechanisms capturing the key properties of the entire ensembles [44]. If composing and closing the sequence of the Krebs cycle and/or describing the sequence of reactions of the glycolysis was years ago a key innovation, the challenge today is defining the nonlinear dynamics of embedded network processes under constraints.

New technologies and interdisciplinary work have promoted the introduction of complex systems thinking on EP, but there is still a long way to go. The sophistication and capacity of modern technology, able to shift the landscape of basic life sciences research from that of traditional biological reductionism to a much more integrative, holistic systems approach [45], are not enough. In fact, a change from a reductionist to holistic paradigm cannot be achieved only via the technical world. Together with new techniques and technologies, the development of new theoretical assumptions, conceptual frameworks, and analysis tools is necessary. New ways of doing and understanding based on complex systems, proven as successful in other disciplines, should be also implemented in EP [46]. As pointed by Greenhaff and Hargreaves [26], “perhaps the tools of Systems Biology should be viewed increasingly as a valuable addition to the arsenal that exercise scientists can use to interrogate physiological function and adaptation” (Table 1).

1. Theoretical assumptions

Theoretical assumptions	Exercise Physiology	Complex Networks Science
Systems	Dominated by their components	Dominated by the interactions among components
Theory	Cybernetic Control Theory	Dynamic Systems Theory
Control	CNS as the main regulator/programmer	Parametrically regulated system
Mechanisms	Homeostatic	Homeodynamic
Methodological traits	Exercise Physiology	Networks Science
Variables	Isolated variables	Networked collective variables
Measures	Means and max values of variables	Connectivity/Synergy/Coordination dynamics
Data acquisition	Group-pooled data	Intra-individual time series
Analysis	Population to individual generalization	Individual to population generalization
Relations	Bottom-up (from micro to macro levels) statistical inferences	Bottom-up and top-down (circular causality) multilevel dynamic interactions

Table 1.

Contrast of some limiting theoretical and methodological assumptions of EP research with assumptions based on Networks Science.

Instead of fragmenting and studying separately the functions of different physiological components and processes, the focus of Networks Science is put on the interaction dynamics among such components and processes. Classical cybernetics, inspiring the basic biological control system model of EP, is replaced by Dynamical Systems Theory (DST), which provides concepts and tools to describe and study the coordinative changes occurring in the physiological network over time.

According to classical cybernetics, different components and processes operate through feedback loops to maintain physical or chemical physiological parameters constant (homeostasis). The predictions of this “engineering” approach are linear, i.e., proportional between inputs and outputs and are displayed through descriptive block diagrams, commonly used in EP to represent how organic structures and processes interact. The basic assumption of these diagrams is that of time-invariant encapsulated modules, processes, and regulation profiles. While the concept of feedback works fine in simple systems that have only two parts to be joined, each of which affects the other, when a few more parts are interlaced together, the system very quickly becomes impossible to treat in terms of explicit feedback circuits.

In complex systems, there is no reference state with which feedback can be compared and no place where comparison operations are performed. Nonequilibrium steady states emerge from the nonlinear interactions among the system’s components, but there are no feedback-regulated set points or reference values as in a thermostat. For instance, it is not possible to explain through feedback loops phenomena such as the fatigue-induced task disengagement [47], the overtraining syndrome [47, 48], or the macroscopic emergence of noncontact injuries [49].

Feedback homeostatic mechanisms are replaced from a Networks Science point of view by the concept of homeodynamics or dynamic stability, i.e., a constantly changing interrelatedness of body components and processes while an overall equilibrium is maintained [50].

It is common among exercise physiologists to propose conceptual models where the main regulator or programmer is the Central Nervous System (CNS) (see, e.g., [45, 51]). Integrative Physiology also neglects that CAS does not need any internal or external programmer to regulate their functions [52]. Properties of such functions (i.e., stability, instability, switches among states, etc.) are parametrically regulated, and the CNS is also a regulated subsystem. This means that physiological states emerge from the interaction among multilevel system components (the CNS being another component) through a self-organized process. The search for the ultimate high-level regulator would end in infinite regress (who regulates the regulator that regulates…?) and represents a loan on understanding exercise-related physiological phenomena.

1. Methodological traits

Instead of isolated variables, the use of macroscopic collective variables is proposed because they behave as order parameters integrating all network levels and capturing the system organization. The dynamics of such variables, reflected in their time variability properties, may inform about the interactions among system components and may help detecting different states and anticipate qualitative changes.

Instead of using only molecular data to establish bottom-up statistical timeless inferences from micro to macroscopic phenomena, the study of the time-variability properties of behavioral macroscopic variables, extracted at action level (e.g., the elbow angle in Section 2.2) during exercise, may inform about the vicinity of qualitative changes. This approach helps to detect dynamic features such as stable and unstable states, critical behavior (phase transitions, bifurcations, critical slowing down, enhancement of fluctuations). It is worth to remark that such behavior can be produced at different quantitative values of physiological parameters or set points [41].

Specific proposals of Network Physiology are developed in sections 4 and 5.

Most of the available research on EP, based on inductive analytical research, infers intra-individual phenomena from the analysis of inter-individual variations obtained through group data means and comparison designs. This approach has some basic debatable methodological issues that should be discussed.

The main aim of physiological (systemic, biochemical, genetic, epigenetic) research is to find the mechanisms of regulation and causal changes that occur at intra-individual (i.e., organism) level as an effect of various internal and external factors. In other words, the intra-organismic processes and not the population are the explanatory target. It is important to note here that intra-individual variability and co-variability unfold in time and hence, need to be measured through time series analytical tools. While the problem of sample to population generalization has been much discussed, investigated, and used, in inferential statistics, much less attention has been focused on the question of sample or population to individual generalization. A tacit assumption has been that the results obtained at sample and generalized to population level are representative of the changes of a “typical” (i.e., average) individual [53, 54, 55]. In other words, the group-pooled data merely would enhance the typical phenomenon that already exists in every and each individual. However, in order for this assumption to be correct, there are some strict conditions to be fulfilled. These conditions are the non-violation of ergodicity¹ assumption. On the other hand, pooling over group subjects is the predominant research practice in exercise and health-related research. Even the state-of-the-art analytical software packages for time series analysis [56] are based on pooling-over-subjects approaches.

The correctness of the tacit assumption of ergodicity conditions, to our knowledge, has never been explicitly tested EP. Hence, the generalization of results from population to individual (or between clusters of individuals) may be typically not valid for developmental biological systems. This means that the structure of causal changes may drastically vary from individual to individual, and these differences are not detectable at the group-pooled level. Some approaches to overcome these serious problems have been proposed [57, 58, 59, 60].

Using molecular data, Integrative Exercise Physiology and OMICS techniques are focused on establishing non-dynamical bottom-up statistical inferences between micro- and macro-level states, ignoring one of the main properties of CAS: the tendency to form multilevel synergies. Synergies acting at different levels (e.g., molecular, cellular, tissue, organ, etc.) allow reciprocal compensations among physiological components and processes to satisfy a task goal during exercise. Such synergies are flexible assembled patterns of coordination, which form emergent structures and functions responding to the exercise requisites. Without them, life would not, and could not, exist. Through circular causality relations, components form new synergies, which govern, in turn, the components’ behavior [61, 62]. While computer scientists build programs that tell circuits what to do, nature builds synergies [63, 64].

Synergetic is also manifested through the CAS property of degeneracy: different components produce the same function, and different synergies may be activated to attain the same task goal [65, 66]. For instance, different motor units cooperate and reciprocally compensate their activation over several timescales to perform an effective or functional motor action over time during a running competition. The self-assembled, adaptive interactions of CAS underpin also another robustness enabling property: pleiotropy or multifunctionality, that is, the same components may be assembled to produce multiple functions. For instance, the skeletal muscle, with genuine/primordial contractile functions, may exert as well immunological and endocrine functions [67, 68]. Such properties enable CAS to switch between diverse coordinative states and maintain a metastable dynamic [69].

4. Network physiology of exercise: A paradigm shift

Dynamic models, initially rejected by biologists, initiated some 40 years ago a paradigm shift in general biology [70, 71], molecular and cell biology [72], genomics [73], and all the “omic”-based approaches [74], which are now at the forefront of science. Such interaction-based approaches have started to spread in EP and relevant fields of medical research such as cancer [75]. However, Physiology, and in particular EP, should do a substantial effort for reassembling biological processes and focusing not only on horizontal interactions at molecular level and establishing non-dynamical statistical inferences to the entire person (e.g., performance or health status, see Figure 1, left), but integrate all vertical network levels (e.g., molecular, cellular, tissue, organ, systems, see Figure 1, right). That is, avoiding the gap between micro and macro structures and functions and considering the multiple vertical synergies that may act among them.

Figure 1.

Contrast between Molecular Exercise Physiology (left), focused on non-dynamical bottom-up statistical inference techniques, and Network Physiology of Exercise (right), focused on the nested dynamics of the vertical and horizontal physiological network interactions.

Network Physiology of Exercise (NPE) emerged inspired by the field of Network Physiology [76, 77, 78, 79, 80, 81, 82, 83, 84] and Networks Science [1]. Network Physiology addresses the fundamental question of how physiological systems and subsystems coordinate, synchronize, and integrate their dynamics to optimize functions at the organism level and to maintain health. It aims at uncovering the biological dynamic mechanisms [85, 86, 87, 88] since it satisfies both the mechanistic requirement of structure and localization (e.g., nodes and edges/links in dynamic networks may represent localized integrated organ systems, subsystems, localized components or processes, and interactions among them across various levels in the human organism) and the requirement of dynamical invariance and generality that is enabled by dynamical systems approach [89].

In the context of exercise, NPE aims to transform the theoretical assumptions, the research program, and the current practical issues of current EP. It focuses the research efforts on improving the knowledge of the nested dynamics of the vertical (among levels) and horizontal (among organs and components) network interactions to understand how physiological states and functions emerge under different constraints and contexts.

Studying the organism as a dynamical system means studying a set of variables that interact over time, that is, their time series, that may exhibit various patterns. DST comprises a highly general set of mathematical concepts and techniques for modeling, analyzing, and interpreting these patterns in time series data. Therefore, DST is not applied exclusively to the area of biomedical sciences, it can also be used to describe social and psychological phenomena, among others [90, 91, 92].

Many physiological mechanisms exhibit oscillations or more complex dynamical behavior, which is crucial for orchestrating operations within the mechanism. Such complex behavior is non-sequential, because some of the interactions in the mechanism are nonlinear, and the system is open to energy. Initial positive adaptations of physiological functions are followed by stagnation or decrease of such functions when workload increases further (e.g. overtraining syndrome, see [47, 48, 93]).

Interactions, generate novel information that determines the future of elements, and thus of the system itself [94]. The interaction-dominant dynamics of humans, in contrast with the typical component-dominant dynamics of machines [95], has been emphasized in the EP literature [41, 96, 97]. This means that the behavior of CAS cannot be simply explained through linearly independent variability sources, processes, or local mechanisms. For instance, exercise physiologists cannot rely on critical quantitative endpoints in cardiovascular, respiratory, metabolic, or neuromuscular systems to explain the limits of performance [31, 98, 99] and should reformulate their research hypothesis on the basis of CAS properties.

4.1 Network Physiology of Exercise. Data analysis techniques

Novel data analysis techniques have been successfully applied in the context of Network Physiology to explore how physiological systems dynamically integrate as a network to produce distinct physiologic functions [80, 85, 86, 100]. The goal of such tools is to develop a general theoretical framework and a computational instrumentarium tailored to infer and quantify interactions among diverse dynamical systems—specifically, (i) systems of oscillatory, stochastic, or mixed type; (ii) systems with noisy, nonstationary, and nonlinear output signals; (iii) systems acting on widely different timescales from milliseconds to hours; (iv) systems coupled through multiple coexisting forms of interaction. Some of the most relevant data analysis techniques to infer couplings among several physiological systems, with potential to be utilized under exercise settings, are the following:

4.1.1 Time delay stability (TDS) method

Integrated physiologic systems are coupled by feedback and/or feed-forward loops with a broad range of time delays. To probe the network of physiologic coupling, a novel concept has been introduced, time delay stability, and a new TDS method has been developed to study the time delay with which modulations/bursts in the output dynamics of a given system are consistently followed by corresponding modulations in the signal output of other systems. Periods with constant time delay indicate stable interactions, and stronger coupling between systems results in longer periods of TDS (Figure 2) [80, 101]. Thus, the strength of the links in the physiologic network is determined by the percentage of time when TDS is observed: higher percentage of TDS corresponds to stronger links. To identify physiologically relevant interactions, represented as links in the physiologic network, we determine a significance threshold level for the TDS based on comparison with surrogate data: only interactions characterized by TDS values above the significance threshold are considered. The TDS method is robust and can track in fine temporal detail how the network of connections between organ systems changes in time. The method is general and can be applied to diverse systems.

Figure 2.

Schematic presentation of the TDS method: Segments of (a) heart rate (HR) and (b) respiratory rate (Resp) in 60 sec time windows (I), (II), (III) and (IV). Synchronous bursts in HR and Resp lead to pronounced cross-correlation (c) within each time window in (a) and (b), and to a stable time delay characterized by segments of constant τ0 as shown in (d)— four red dots high- lighted by a blue box in panel (d) represent the time delay for the 4 time windows. Note the transition from strongly fluctuating behavior in τ0 to a stable time delay regime at the transition from deep sleep to light sleep at ∼9400 sec and inversely from light sleep back to deep sleep at ∼10,100 sec (shaded areas) in panel (d). The TDS analysis is performed on overlapping moving windows with a step of 30 sec. Long periods of constant τ0 indicate strong TDS coupling.

4.1.2 Phase synchrogram algorithm (PSA)

Nonlinear oscillatory systems are characterized by nonidentical eigenfrequencies and highly irregular signal output can synchronize even when their coupling is weak—i.e., their respective frequencies and phases “lock” at a particular ratio. Despite the significant difference in the periodicity of the cardiac and respiratory rhythms represented by the heartbeat and respiratory intervals, and despite the complex noisy variability in the cardiac and respiratory signals, previous work found that episodes of heartbeat-respiration phase synchronization emerge. Previous authors developed a synchrogram algorithm able to identify segments of cardiorespiratory phase synchronization and to track how the degree of this nonlinear form of coupling changes in time and across physiologic states ([85]; Figure 3). The PSA synchrogram algorithm is robust and can identify interrelations between output signals of nonlinear coupled systems even when these signals are not cross-correlated [86]. Thus, the PSA can quantify the degree of coupling between nonlinear systems when other conventional methods (such as cross-correlation or cross-coherence analysis) cannot.

Figure 3.

Schematic presentation of the PSA method: (A) three consecutive breathing cycles and (B) a simultaneously recorded ECG signal. (C) Demonstration of phase synchronization between the heartbeats and respiratory cycles shown in (A) and (B). For each breathing cycle, all first heartbeats occur at the same respiratory phase φ1r(t), and all second and third heartbeats within each breathing cycle occur at φ2r (t) and φ3r (t), respectively (symbols collapse), indicating robust phase synchronization. (D) each heartbeat in the ECG signal (B) is shown with its phase φr (t) relative to the beginning of the breathing cycle in which it occurs. Different symbols represent heartbeats in different breathing cycles as in (A) and (C), and vertical dashed lines show the beginning of each breathing cycle. Three horizontal lines formed respectively by the first, second, and third heartbeats in the three breathing cycles indicate robust 3:1 phase synchronization despite noisy heart rate and respiratory variability.

4.1.3 Cross-correlation of instantaneous phases (CCIP) method

Due to the nonstationary trends embedded in physiologic signals, traditional cross-correlation and cross-coherence analyses fail to accurately quantify the interrelation between physiological systems. This approach based on the cross-correlation between the instantaneous phase increments of the output signals of nonlinear coupled systems is not affected by the nonstationarity of the signals. Chen et al. [86] successfully applied this new approach to study cerebral autoregulation in healthy subjects and in stroke patients. The approach is sensitive to uncover previously unknown differences in the coupling between cerebral blood flow velocity and peripheral blood pressure in the limbs for healthy and post-stroke subjects (Figure 4). In contrast, linear cross-correlations and other traditional methods cannot identify changes in cerebral autoregulation after stroke.

Figure 4.

The CCIP approach: Identifies breakdown of coupling mechanisms of cerebral autoregulation after stroke. Signals of peripheral blood pressure (BP) in the limbs and blood flow velocity (BFV) in the brain for (a) healthy and (b) post-stroke subject during a quasi-steady supine state without external perturbations. (c-d) Traditional cross-correlation function C(τ) for the BP and BFV signals for the same subjects shown in (a) and (b) does not identify differences in the BP-BFV coupling between healthy and post-stroke subjects. (e-f) In contrast, cross-synchronization function S(τ) obtained from the phase increments of BP and BFV signals using our CCIP method shows a significant difference in the BP-BFV coupling between healthy and post-stroke subjects, even though patterns in the pairs of BP-BFV signals are visually similar (a, b) for both healthy and stoke subjects.

4.1.4 Principal components analysis (PCA)

PCA has been recently conducted on the time series of several cardiorespiratory parameters during maximal exercise (Figure 5). The PCA pinpoints and quantifies whether the increment and decrement of time patterns from different physiological processes are statistically correlated. In this way, the magnitude to which time patterns of physiological responses covary in time is reflected. The covariation of several (two or more) cardiorespiratory parameters shows the mutual information that they share. This common variance, in turn, enables time patterns of single cardiorespiratory outcomes to be represented through fewer principal components (PCs). The PCs are obtained in decreasing order of importance and reflect the highest possible fraction of the variability from the original dataset. Thus, the total number of PCs indicates the level of coordination among the initial cardiovascular and respiratory parameters. More concretely, a dimensionality reduction is indicative of the creation of new coordinative patterns [102]; therefore, the reduction in the quantity of PCs suggests an enhancement in the efficiency of cardiorespiratory system [103].

Figure 5.

Typical example of the reduction of cardiorespiratory variables to time series of cardiorespiratory coordination variables (PCs) in two consecutive maximal cardiorespiratory tests interspersed by 10-min resting: Test 1 and test 2. Top graphs: Original time series of the six selected cardiorespiratory variables in test 1 and test 2. Bottom graphs: Time series of PC scores (standardized z-values in the space spanned by PCs) in both tests. The six time series are collapsed to one time series (test 1) or two time series (test 2) as a consequence of the PC dimension reduction. The black and the red lines show the average trend of both processes as calculated by weighted least squares method. Data points of the x-axis of both graphs refer to the number of measurements recorded along the cardiorespiratory test.

4.2 Evaluation tools based on Network Physiology: Network-based biomarkers

The common testing variables used in Exercise Physiology (e.g., VO₂max, ventilatory thresholds, etc.) do not provide sufficient information about the dynamic interactions among physiological systems and their common role in an integrated network. In this line, previously published works have shown a lower sensitivity of gold standards such as VO₂max compared with other coordinative variables able to determine dynamic interactions among physiological systems [104, 105, 106, 107]. The different data analysis techniques described in the previous section have the potential to be used as novel evaluation tools to investigate interactions among physiological systems under exercise settings. More specifically, these techniques can lead to the development of new network-based biomarkers able to quantify how different key organ systems (e.g., brain, heart, skeletal muscles) coordinate and synchronize as a network during exercise and track how these network interactions change in response to fatigue and training. The use of new network-based biomarkers will break new ground in the study of multilevel inter-organ interactions and will provide new understanding of Basic Physiology and diverse exercise-related phenomena such as sports performance, fatigue, overtraining, or muscle-skeletal injuries.

4.2.1 Inter-muscular interactions

Inter-muscular coordination is defined as a distribution of muscle activation or force among individual muscles to produce a given combination of joint moments [108]. Therefore, neuromuscular control during exercise or activities of daily living is not limited to switching muscles on or off but includes fine-tuned control to select the appropriate muscle fiber types with precise timing and activation [109, 110, 111]. Techniques based on the frequency domain of the surface EMG [112, 113] are the most suitable to infer information on motor unit recruitment and muscle fiber since (i) the average conduction velocity of the active motor unit is related to fiber-type proportions, and (ii) the changes in the spectral properties are linked to the changes in the average conduction velocity. Inter-muscular coherence (IMC) is one of the most utilized methods to investigate inter-muscular interactions in the frequency domain—it estimates the amount of common neural input between two muscles during voluntary motor tasks [114]. Despite its clinical relevance to evaluate inter-muscular coordination, IMC has been recently questioned for its lack of potential to identify nonlinear dynamic coupling across frequencies [115] and, thus, ignore the interactions between distinct types of muscle fibers across muscles. Therefore, new data analysis approaches are needed to investigate the physiological mechanisms underlying cross-frequency network communication among distinct muscle fiber types across muscles during exercise.

4.2.2 Cortico-muscular interactions

Skeletal muscle activity is continuously modulated across physiologic states to provide coordination, flexibility, and responsiveness to body tasks and external inputs. Despite the central role the muscular system plays in facilitating vital body functions, the network of brain-muscle interactions required to control hundreds of muscles and synchronize their activation in relation to distinct physiologic states has not been sufficiently investigated. In this line, to identify and quantify the cortico-muscular interaction network and uncover basic features of neuro-autonomic control of muscle function, a recently published work [116] has investigated the coupling between synchronous bursts in cortical rhythms and peripheral muscle activation during sleep and wake. The findings demonstrate previously unrecognized basic principles of brain-muscle network communication and control and provide new perspectives on the regulatory mechanisms of brain dynamics and locomotor activation, with potential clinical implications for neurodegenerative, movement, and sleep disorders and for developing efficient treatment strategies. Further research is warranted to investigate cortico-muscular interactions during exercise and their changes in response to fatigue and different training methodologies.

4.2.3 Cardiorespiratory interactions

Previous research has demonstrated that the cardiac and respiratory systems exhibit three distinct forms of coupling: respiratory sinus arrhythmia (RSA), cardiorespiratory phase synchronization (CRPS), and time-delay stability (TDS) [76, 85, 101]. While RSA is a measure of amplitude modulation of the heart rate during the breathing cycle, CRPS and TDS characterize the temporal coordination between the cardiac and respiratory systems. Specifically, the CRPS reflects the degree of clustering of heartbeats at specific relative phases within each breathing cycle (despite continuous fluctuations in heart rate and in breathing intervals), and the TDS quantifies the stability of the time delay with which bursts in the activity in one system are consistently followed by corresponding bursts in the other system. The findings indicate that these three distinct and independent forms of cardiorespiratory coupling are of transient nature, with nonlinear temporal organization of intermittent “on” and “off” periods, even during the same episode of any given physiologic state (sleep stage), and that these coupling forms can simultaneously coexist.

In the context of exercise, cardiorespiratory coordination has been investigated through a Principal Components Analysis (PCA) performed on time series of cardiovascular and respiratory variables registered during cardiorespiratory exercise testing (expired fraction of O₂, expired fraction of CO₂, ventilation, systolic blood pressure, diastolic blood pressure, and heart rate). Cardiorespiratory coordination has been utilized to assess changes produced by different training programs [103, 105], testing manipulations [104, 106, 117, 118], nutritional interventions [107], and pathological conditions [119]. The main findings of this set of studies point toward a higher sensitivity and responsiveness of cardiorespiratory coordination to exercise effects compared with isolated cardiorespiratory parameters, such as VO₂max and other gold standard markers of aerobic fitness.

It should be noted that the aforementioned network-based biomarkers (inter-muscular interactions, cortico-muscular interactions and cardiorespiratory interactions) can provide relevant information about how different key organ systems coordinate and synchronize as a network during exercise and track how these network interactions reorganize with accumulation of fatigue and in response to different training programs. However, these network-based biomarkers can only provide information at the organic (macroscopic) level by using equipment capable of recording continuous high-frequency physiological signals (time series of EEG, ECG, EMG). Therefore, the development of adequate technology able to register continuous and synchronous data extracted from different levels is needed to investigate the dynamics of physiological network interactions (i) not only at a macroscopic level, but also at lower levels of integration (i.e., cellular and subcellular); and (ii) among multilevel systems components—that is, capturing the synergies, embeddednes, and circular causality (bottom-up, top-down) between lower and upper levels.

5. Conclusions

Despite the fundamental discoveries, vast progress and achievements in the field of EP for over a century, the reductionist framework that has traditionally dominated research in the field has imposed limitations to the exploration and understanding of the regulatory mechanisms underlying complex exercise-related phenomena.

EP research, characterized by an inductive analytic mode of inquiry, has progressively evolved toward Biochemistry, Molecular Biology, Genetics, and OMICS technologies. Although such biology branches can be subjected to dynamical approaches, Molecular Exercise Physiology and Integrative Physiology keep focused on qüestionable non-dynamic bottom-up group-pooled statistical inferences.

Inspired by the field of Network Physiology and Complex Systems Science, Network Physiology of Exercise emerges to transform the theoretical assumptions, the research program and the practical applications of EP. The cybernetic Control Theory is replaced by Dynamic Systems Theory (DST), the centralized control of the CNS by a multilevel self-organization of body functions, and the static regulatory mechanisms by dynamic mechanisms with synergetic properties. The inductive analytical research, generalizing from group inter-individual inferences to intra-individual phenomena, is replaced by an inductive/deductive research based on intra-individual time series analysis techniques. Furthermore, it fills the gap of current research in Molecular Exercise Physiology, which is almost exclusively based on establishing bottom-up static statistical inferences from molecular data to the physiology of the entire person.

Network Physiology of Exercise focuses the research efforts on investigating the nested dynamics of the vertical (among levels) and horizontal physiological network interactions. The embeddedness of lower network levels in upper levels, the circular causality (bottom-up, top-down) among levels acting at different timescales and the emergence of nonlinear network phenomena are some of its genuine expected contributions. Network Physiology provides a wide range of data analysis techniques that have the potential to be utilized as novel evaluation tools to investigate interactions among physiological systems under exercise settings. These techniques can lead to the development of new network-based biomarkers (e.g., cardiorespiratory interactions, inter-muscular interactions, and cortico-muscular interactions) able to identify how different key organ systems coordinate and synchronize as a network during exercise and track how these network interactions change in response to different physiological states and exercise interventions. The use of new network-based biomarkers will open new and exciting horizons on exercise testing, will enrich Basic Physiology and diverse fields such as Exercise Physiology, Sports Medicine, Sports Rehabilitation, Sport Science, or Training Science, and will improve the understanding of diverse exercise-related phenomena such as sports performance, fatigue, overtraining, or sport injuries.

Acknowledgments

It includes funding information.