Aging and changes in complexity in the neurobehavioral system.

The issue of complexity is more and more present in numerous domains of biological research, including aging research. In the present paper, based on a selective review of literature, we propose both a conceptual and a methodological framework to address age-related changes in functional complexity of the neurobehavioral system, presumably resulting from modifications of the coupling between cognitive and sensorimotor processes. In particular, after reviewing pioneering and more recent studies on aging and complexity in the neuromusculoskeletal system, we explore the possibility that an age-induced increase in the coupling between cognitive and sensorimotor domains could be captured by a stronger covariation of high-order variables, common to both cognitive and sensorimotor functioning. Our main assumption is that these variables could behave as neurobehavioral markers of aging in the neuromusculoskeletal system. The present approach markedly differs from other traditional approaches, which focused on process-specific variable correlates of chronological age, domain-by-domain, and task-by-task. It provides a coherent conceptual framework, a terminology, and a method for studying age-related coupling of cognitive and sensorimotor processes with the use of complexity and nonlinear dynamical systems theories.


Introduction
The passage of time in the organism ineluctably leads to structural and functional alterations in its different component subsystems, which result in a decline of performance and loss of behavioral adaptability in many daily living situations. Although various age-induced phenomena occur concomitantly at multiple levels (cells, tissues, organs, functions, and behavior), across different subsystems, and in various task conditions, interactions inevitably occur between functional domains of the neurobehavioral system. Consequently, researchers are forced to adopt a clear-cut theoretical position on the way of dealing with the issue of complexity in the neuromusculoskeletal system and its change during aging. In this respect, an explicit reference to dynamical systems analysis and complexity theories has been made in numerous studies on aging of physiological and neurobehavioral systems during the last 20 years (1)(2)(3)(4). Nevertheless, a firm theoretical background remains to be established to make these approaches more attractive in aging research.
Based on a brief survey of the existing literature, the present paper aims to provide a conceptual framework for understanding age-related changes in neurobehavioral complexity, in the theoretical context of analysis of nonlinear dynamical systems.
In particular, we argue that an increased coupling of cognitive and sensorimotor processes observed during aging reflects a loss of complexity in functional organization of the neurobehavioral system, resulting from dedifferentiation of neural processes. Accordingly, new lines of a research agenda are proposed to identify neurobehavioral markers of increased coupling between cognitive and sensorimotor processing during aging.
In the first part of this paper, we present some preliminary statements about classic aging research and the different approaches of biological complexity. In the second part, we show how the study of variability and complexity of behavioral output provides meaningful information about the underlying organization of the aging neurobehavioral systems. Finally, we present new research directions for exploring age-related changes in neurobehavioral complexity by focusing on common markers to cognitive and sensorimotor domain.

Aging Research and Biological Complexity: Some Preliminary Statements Different Approaches of Complexity in the Neuromusculoskeletal System
It is commonly accepted that the neuromusculoskeletal system is a complex system composed of multiple interacting components at different levels (cells, organs, subsystems, …). Divergence exists, however, among researchers with respect to whether biological complexity must be considered as a resource or as an obstacle to understanding aging process. Typically, two main theoretical approaches of biological complexity, so-called "local" and "global," can be distinguished in aging literature.
Proponents of the "local" approach assumed that the neurobehavioral system is decomposable so that the problem of complexity can be solved by dividing it into smaller, simpler, and thus more tractable units (processes, tasks, …) in separate domains. Accordingly, they consider that age effects operate on multiple juxtaposed spaces and independent time scales, which relegate each subsystems and components to their own causality.
Numerous researchers have however argued that the human neurobehavioral system is more than a collection of static and independent components, governed by simple and linear causalities (3,5) so that aging ineluctably leads to changes in biological complexity as a consequence of alterations of the different components and their interactions (1). Proponents of this global approach argued that a challenge for aging research is to understand when, how, and why the complexity arising from the interactions of the different systems evolves over time. In this perspective, concerted efforts have been devoted since the early 1990s to establish greater linkages between the different levels of analysis (e.g., brain, information processing, and behavior) (6,7) or even cross-domain linkages (the coupling of cognitive and sensorimotor processes) (8)(9)(10). In parallel, a "physical biology approach" (4), mainly grounded on nonlinear dynamic systems analysis and complexity theories, has emerged in aging literature (1,4,11,12). The recognition is at the core of this approach that the neuromusculoskeletal system (NMSS) is a self-organizing/self-structuring, multiscale and multilayered system, whose dynamics is driven by chronological age and shaped by constraints arising from multiple sources of various origins (environment, organism, education, life habits, …), each having its own time scale.

Aging as an Integrated Process Leading to Changes in Behavioral Adaptability
A characteristic feature of healthy biological systems is that they lose their internal robustness and complexity with aging and disease (1,13). In this perspective, aging can be considered as a dynamic process, reflected in a sequence of (more or less) transient functional states over time. A challenge for researchers is to describe and understand general principles of changes in typical states of neurobehavioral and physiological systems over time. In this perspective, assessing behavioral adaptability, instead of mean performance, is crucial. Behavioral adaptability refers to the capacity of preserving stability and flexibility facing task constraints. These two seemingly conflicting properties may be altered by age-induced alterations of the system components and their interactions. Behavioral adaptability is thus an indirect indicator of underlying changes in functional complexity of the system for researchers and a meaningful property for clinicians (2).
During the last 20 years, the physical biology approach has produced convincing findings and has permitted to identify some general principles of age-related changes in the complexity of the outputs of physiological subsystems (e.g., cardiovascu- It is composed of multiple elements, at different levels, which interact to produce adaptive (i.e., stable and flexible) behavior. The local approach focuses on the study of the components independently. In the physical biology approach, the analysis focuses on the spatiotemporal complexity of behavioral outputs arising from multiple interactions between components.

NMSSS Physical Biology Approach
Behavior Local Approach lar system [1]) and task-specific action systems (e.g., grip force control, posture, locomotion, … see [12] and [14] for extensive reviews). These findings are briefly reviewed in the following chapter.

Aging and Complexity of Physiological and Neurobehavioral Systems Functional Meaning of Variability of Behavioral Outputs in Complex Systems
During the last two decades, researchers and clinicians have been progressively aware of the potential applications of dynamical system analysis and complexity theories in the domains of aging and disease (1). It has led to increasing recognition that what appears as white, unstructured, uncorrelated noise in data may possibly be structured in the form of correlation functions in the time and frequency domains (i.e., colored noise). Accordingly, researchers have devoted most efforts to quantification of fluctuations of behavioral outputs by using measures of regularity, randomness, and historicity of time series capturing system behavior. Fluctuations of behavioral outputs represent a coarse grain signature of how the multiple elements interact depending on task-context, timescales, and spaces. In this respect, nonlinear analyses of variability may reveal structures in fluctuations that remain undetected by more traditional measures, such as mean performance changes or standard deviation. Struc-tured fluctuations possess a great theoretical and clinical significance since they reveal the presence/ absence of coupling and characteristic temporal scale that may confer more or less adaptability to the system (1,15). Thus, the challenge for complexity approaches to aging is to understand the meaning of the modification of structured variability with regard to the functional organization of physiological and neurobehavioral system.

Quantification Time Series Regularity of Behavioral Outputs
Meaningful changes in behavioral variability in healthy aging systems, as well as in clinically relevant syndromes (falling, neurodegenerative disease, …), are quantifiable by regularity measures. Classically, one considers that healthy systems are those that produce the lower levels of noise (i.e., of stochastic inputs). Thus, variability is primarily considered as reflecting undesirable random noise of varying magnitude, which would be superimposed on an invariant deterministic signal (16). It is noticeable, however, that magnitude and complexity of fluctuations are relatively independent: magnitude of variability may be unaltered whereas complex structure changed.
Instead, when the regulating systems are disturbed within a complex network, as a result of aging or disease, behavioral control may be impaired, Since complexity of times series is associated with structural richness, to assess it, one quantifies "irregularity" and time dependence of fluctuations in the time series. In this respect, an important methodological challenge is to detect and quantify irregularities, scaling and correlation properties of physiologic and neurobehavioral time series, which are typically nonstationary (see 17 and 18 for detailed descriptions of the different methods). In this aim, multiple nonlinear methods can be used, each having their own interest and limits (e.g., approximate entropy, detrended fluctuation analysis, spectral analysis, or Lyapunov exponent) to capture relative contributions of stochastic and deterministic influences to the change in the system output (19)(20)(21)(22).
According to these general principles, two types of measure have been predominantly used in aging research for investigating changes in complexity of physiologic and neurobehavioral system outputs (1,23). On the one hand, it is approximate entropy (ApEn), which is a measure of the amount of information needed to predict the future state of the system: the larger the entropy, the more irregular the fluctuations, the less predictable the system (1,22,24,25). Typically, the ApEn algorithm returns a value tending toward 2 for highly irregular signals and 0 for highly regular signals. It does not identify, however, the contribution of deterministic and stochastic processes to the observed regularity, and surrogate tests are necessary to rule out the possibility that irregularity reflects random noise. On the other hand, it is fractal scaling exponent calculation (α), which provides an estimation of the "roughness" of the time series and permits to quantify the reliance on history through the presence of time sequential and frequency-dependent properties of time series (self-similarity). Specifically, one considers that the larger the value of the fractal exponent, the smoother the time series. In this context, a value of α equal to 1 (so-called 1/f noise) can be interpreted as the best compromise between complete unpredictability associated with white noise (α=0.5) and the much smoother profile of Brownian noise (α=1.5). A breakdown in optimal long-range correlation properties in time series is in keeping with the loss of complexity in the system output. Presumably, such breakdown is associated with functional impairment in information transmission and processing in the neurobehavioral system (17).

Aging and the Loss of Complexity Hypothesis
During the last 20 years, knowledge gleaned from seminal studies of physiologic systems (1) has stimulated dynamical investigations of age-related variability in neurobehavioral systems (12,(26)(27)(28). A central assumption for theorizing complexity and aging holds that alterations in the structural and functional properties of the organism lead to a loss of complexity in system outputs and, consequently, in the capacity of the elderly to adapt to various forms of constraints (1, 3, 17, 29, see 23 for a review). This hypothesis was supported by pioneering studies based on the analysis of physiological and neurobehavioral systems (cardiac rhythm, postural control, …) (1,24). They showed that healthy systems possess the highest adaptability (healthy) and generate the most physiologically irregular signals. For instance, analysis of scaling behavior in a variety of cardiac pathologies and in healthy elderly people indicated significant alterations in short-and long-range correlation properties (for reviews, see 17,23). It was noticed, however, that alterations of scaling behavior associated with physiologic aging were different compared with those characterizing heart failure. These findings were consistent with those observed by Lipsitz and Goldberger (1992) using approximate entropy analysis (i.e., a decrease in ApEn in older adults), which suggested a loss of complexity of fluctuations in the time series. Thus, loss of complexity, which is a consequence of ageinduced aberrations in the temporal organization of the evolving dynamics, might be a generic feature of aging and a subset of dynamical disease (1,11).
Since the seminal work done by Lipsitz and Goldberger (1992), approximate entropy and fractal scaling analyses have been used in complex time series of a large variety of subsystems outputs: systolic blood pressure (30), gait (20), electroencephalography (31), center of pressure trajectory in postural control (32), grip force control (33), aiming (34), or even human cognition (35)(36)(37). Results confirmed the presence of structured, long-range correlated fluctuations in healthy organisms and supported the hypothesis of systematic breakdown of long-range correlations in behavioral outputs as a result of aging and disease. However, whether or not aging led to a decrease in complexity seemed to depend on a particular dynamic property of the system under investigation. It has led Vaillancourt and Newell (23, see also 4) to propose the "loss of adaptability hypothesis." They assumed that aging might result in either a decrease or an increase in complexity, as a function of task-specific constraints, both leading to impairment in behavioral adaptability.

Aging and the Loss of Adaptability Hypothesis
In an extensive review of literature, Vaillancourt and Newell (4,23) examined the universality of the principle of loss of complexity with aging. They suggested that the direction of changes in output complexity could be system-or task-dependent. More specifically, they showed that in some systems (e.g., the cardiovascular system), a loss of complexity revealed an impaired capacity, and in other systems (e.g., gait, grip force control), an increase in complexity signaled an impairment of control. These findings were counter to the loss of complexity hypothesis, which assumed that the predictability in physiological output should increase with age. Rather, they suggested that there is not a universal principle of directional change in the complexity of behavioral outputs. Instead, change in complexity might depend on the nature of adaptive change required to realize particular system's environment demands. Newell and collaborators (4, 23) distinguished between two types of intrinsic dynamics (resulting from fixed-point and rhythmical attractors), which presumably determine different routes for effective adaptation to a task-specific constraint. On the one hand, fixed-point attractor tasks (attractor dimension =0) when the system operates around a steady state as for instance in physiological homeostasis (heart rate, hormone secretion, postural control). In this zero dimension attractor, an increase of the system's active degrees of freedom is needed to maintain optimal performance resulting in an increase of the output complexity (higher ApEn). In the case of aging or disease, the impossibility to increase the system's degrees of freedom should result in a reduction in adaptability with a reduced output complexity (lower ApEn). In this case, there is a loss of complexity with aging. On the other hand, Vaillancourt and Newell (23) postulated that for limit-cycle attractor tasks (oscillatory dynamics with attractor dimension =1) where a decrease in degrees of freedom is required to perform the task, one should observe an increase in complexity with aging and disease (i.e., an increase in ApEn). Indeed, in these tasks, the elderly are unable to decrease the number of active degrees of freedom to perform the task and thus fluctuates more around the intrinsic dynamic resulting in a more complex behavioral or physiological output. Gait cycle or diurnal rhythm of adrenocorticotropic hormone concentration was a typical example of oscillatory dynamics (20,38). Vaillancourt and Newell (12) tested this hypothesis in an isometric force production task performed in either constant or periodic (sine wave) conditions. Results showed that structured fluctuations of force output were less complex in old and older-old adults, in the constant force production task, and that they were more complex in the sine wave task. These results confirmed that the directional change of the complexity of physiologic and neurobehavioral system outputs covaries with the dimension of the intrinsic dynamic that organize the system (23).

Behavioral Output Complexity and Functional Organization of the Neurobehavioral System
Though numerous studies supported the hypothesis that change in complexity of behavioral outputs is a generic feature of aging in physiologic and neurobehavioral systems, the question remains how this coarse grain level of analysis of behavioral outputs reflects underlying mechanisms that occur within the neurobehavioral system and gives rise to changes in complexity. In this respect, we contend that more detailed model systems are needed to better establish the relationship between mechanisms that operate within the system and the observed complexity of behavioral outputs (see 39 for a convergent point of view).
According to the current definitions of biological complexity (23,40), whatever the direction of change in the structured fluctuations of behavioral outputs, change in complexity is hypothesized to result from four (non mutually exclusive) primary mechanisms induced by aging: 1) impairment and/or loss in the number of individual subcomponents of the system; 2) change in the strength of couplings between the components; 3) change in stochastic inputs (e.g., neural noise); and 4) increase in time delays of information transmission within the system (slowing).
However, classic studies on aging were predominantly interested in investigating separately the functional consequences of these different factors on behavioral performance rather than considering them in the perspective of age-related changes in complexity. For instance, impairment of the different individual component subsystems (neuromuscular, cognitive, energetic, …) and/or loss in the number of components (e.g., loss of neural cells, muscle fibers, white matter, force decrement, …) have been widely studied in isolation (41)(42)(43)(44)(45). The role of stochastic inputs (i.e., neural noise) has also recently been the subject of increased interest in cognitive aging literature through the study of intraindividual variability of performance (7,46). The issue of behavioral consequences of age-induced increase in time delays of information processes (i.e., the general slowing hypothesis) has been addressed for a long time in cognitive literature (47,48). However, how changes in these factors affected the complexity of the neurobehavioral system was scarcely addressed in the literature, and consequently, nonlinear measures of structured fluctuations of behavioral performance were rarely considered as meaningful indicators of age-related changes in cognitive or sensorimotor functioning. As noticeable exceptions, recent studies by Delignières and coworkers showed, however, that measures of complexity fruitfully added to the understanding of underlying mechanisms of timing in finger tapping and (loss of) stability of bimanual coordination patterns, which were classically assessed by the magnitude of variability (49, 50, see 39 for an overview).
The question remains, however, how complexity approaches can be applied to the study of interactions between subcomponents of the neurobehavioral system that cannot be modeled as nonlinear oscillators as, for instance, the coupling between cognitive and sensorimotor processes. This issue has recently emerged as an important preoccupation in aging research (51,52). Specifically, it has been suggested that aging was accompanied by a dedifferentiation of cognitive and sensorimotor processes at both brain and behavior, which leads to a stronger coupling between cognitive and sensorimotor abilities (53, see 51 for a review). According to the dedifferentiation hypothesis, aging may be considered an integrated process affecting the whole functional organization of the neurobehavioral system. Consequently, the occurrence of increase in the coupling among the different subsystem components is critical to understanding age-induced changes in the complexity of the neurobehavioral system. We would like to draw some theoretical working hypotheses and methodological procedures in this perspective.

Neurobehavioral Markers of Dedifferentiation of Cognitive and Sensorimotor Processes
Ideally, to explore the dynamics of changes in the coupling of two a priori independent systems (e.g., cognitive and sensorimotor systems), one should be able to collect time series of the signals that represent behavioral outputs of the two systems simultaneously and/or to capture their relationship either by a "collective" variable of the coupling/synchronization pattern (for a general introduction, see 54; see 55 for a detailed example in interlimb coordination) or thanks to nonlinear tools as cross-entropy analysis (56). However, in the case of coupling between cognitive and sensorimotor domains, it is technically difficult since cognitive and sensorimotor systems cannot be assimilated to self-sustained oscillators and does not possess obvious collective variables. Thus, in most experimental paradigms aiming at investigation cognitive-sensorimotor coupling (including double-task paradigms), behavioral time series cannot be simultaneously and continuously recorded.
We argue that an alternative strategy permitting to account for cognitive-sensorimotor coupling lies in the search of common neurobehavioral markers to cognitive and sensorimotor processes, which might permit to assess their time (more and more correlated) coevolution. The underlying neurophysiological justification of this approach is the widely accepted "common cause hypothesis," which assumed that some of the mechanisms that are at origin of age-related decline in cognitive and sensorimotor performance might be shared among different nervous structures in the neurobehavioral system (8,9,57). These common structures could mediate the so-called "coupling" between cognitive and sensorimotor processes in a wide range of tasks. Accordingly, if cognitive and sensorimotor processes become more and more intertwined until forming a single, almost undifferentiated entity in the neurobehavioral system, then it should be possible to identify common markers permitting to assess (to infer) the degree of coupling between cognitive and sensorimotor processes.

What is a Neurobehavioral Marker of Aging?
A neurobehavioral marker of aging is a parameter intended as a quantitative measure of the rate of aging in the neuromusculoskeletal system that represents a more accurate index than chronological age. Moreover, in the absence of disease, it must better predict functional capacity at some later age than will chronological age (58). Accordingly, to be considered as a neurobehavioral marker, any measured variable at brain or behavioral level should tap into the fundamental capacities of the central nervous system (CNS) and summarize its general state (59). According to Deary and Der (60, see also 61,62), other general characteristics, which might be possessed by the chosen variables to meet the criteria of a neurobehavioral marker aging, are as follows: 1) stable individual differences across several years in large adult samples; 2) a clear pattern of age-related deterioration from young adulthood to old age; and 3) changes from baseline measurement, which are related to mortality and morbidity in longitudinal studies even after adjustment for age.
Examples of behavioral measures that have been used as biomarkers of aging in cognitive and sensorimotor domains were grip strength, blood pressure, visual acuity, or mental processing speed (63). The classic procedure consisted of studying longitudinal correlations of two or more cognitive and sensorimotor measures in order to infer dedifferentiation between the two domains. We contend, however, that to reflect age-related increase in the coupling between cognitive and sensorimotor processes, the most important constraint to the definition of neurobehavioral markers is that they should measure "high order" common phenomena that are observable in both cognitive and sensorimotor domains.

Potential Common Neurobehavioral Markers to Cognitive and Sensorimotor Processes
Even if a general decrease in information processing capacity is not the unique cause of age-related declines in performance of the neurobehavioral system, it presumably explains most of behavioral manifestations of age-related sensorimotor and cognitive changes. Consequently, the following premises can be followed for identifying variables measuring overt behavioral manifestations resulting from changes in neural resources and information processing capacities in both cognitive and motor domains: 1) the CNS is an information-processing system whose limited central capacity affects both cognitive and sensorimotor functioning, and 2) speed of cognitive and motor behavior are highly dependant on information processing capacity; 3) both cognitive and motor functioning become more reliant on central mechanisms during aging by virtue of increased sharing of neural resources (64,65), and 4) organization and coupling between components of the CNS change over time, thereby modifying their interactions (23); 5) a decline in general information processing capacities depends on neural processes that are (at least in part) shared by both domains; 6) age-related changes that occur at both neurochemical and neuroanatomical levels of the brain decrease functional information processing capacities in both specific and nonspecific global fashion (47,64); 7) neural decline in information processing might then "enslave" the evolution of both cognitive and sensorimotor domains, thereby causing correlated behavioral outcomes; 8) dedifferentiation of information processing capacities in cognitive and sensorimotor domains might be reflected in similar overt manifestations, which can be consider as common neurobehavioral markers of their functioning (66-69); and 9) neural noise increases in the information processing system during aging resulting in more inconsistent behavior.
Taken together, the above considerations suggest that (at least) three categories of neurobehavioral markers could capture the coupling between cognitive and sensorimotor coupling: 1) complexity measures of fluctuations of behavioral performance of cognitive and sensorimotor systems; 2) an intraindividual variability measure of behavioral performance; and 3) response speed. Most importantly, since these three makers are supposed to reflect a common cause in the neurobehavioral system, their evolution over age should be increasingly correlated.
Changes in Complexity of Cognitive and Sensorimotor Performance. If change in complexity is a generic feature of aging (4,23), then changes in structured behavioral outcomes in both cognitive and sensorimotor domains of functioning should be correlated (1). Adequate markers could then be found in entropy and fractal dimension analyses to quantify the complexity of time series of behavioral output in both cognitive and sensorimotor tasks. Correlated changes in complexity should permit to infer underlying modifications of coupling interactions between the components of the system (23). One predicts that directional change in complexity reflecting a loss of adaptability should be similar in both cognitive and sensorimotor tasks.
Time Measures. It is currently admitted that speed is critical to the performance of information processing systems (70). Since rapid responses require an efficient system of communication within the CNS in both cognitive and sensorimotor tasks, speed of processing is commonly considered as an indicator of basic neurocognitive resources (71) and age-related neurological status (48). Accordingly, speed of processing in both cognitive and motor tasks could be a useful neurobehavioral marker of cognitive aging.
Intraindividual Variability Measure. Because neural noise strongly depends on the central command-signal resulting from information transmission and processing, an intraindividual variability of behavioral outcome is also commonly considered as an indicator of basic neurocognitive resources and age-related neurological status (72).
It is noticeable that response speed and intraindividual variability are currently considered as presenting the formal characteristics of biomarkers of aging in cognitive literature (60)(61)(62). It remains to be determined whether they are common biomarkers to sensorimotor processes and whether they can reflect the increased coupling between these two functional domains.

How to test Neurobehavioral Markers of Cognitive-Sensorimotor Coupling?
The above-developed theoretical framework leads to a specific research agenda, which is quite different from classic studies in aging research. Indeed, instead of analyzing the coevolution of two specific variables, each measuring behavioral performance in cognitive and sensorimotor domains, it consists of analyzing the correlated changes in common measures to cognitive and sensorimotor processes in different tasks (i.e., performance output complexity, response speed, and intraindividual variability of behavioral performance). The coevolution of these markers is hypothesized to express underlying common causes of age-related changes in behavioral performance. Specifically, if the measured values of each marker evolve in the same direction and are more and more strongly related to age, then it would indicate that the coupling between cognitive and sensorimotor processes increased so that functional capacities tend to be more and more undifferentiated.
To test the coevolution of the values of the different markers, we have to choose representative tasks of cognitive and sensorimotor central information processing capacities. In this respect, reaction time tasks (Hicks' tasks) and unimanual aiming tasks (Fitts' task) are adequate. Indeed, reaction time tasks permit a measure of the amount of information processed to choose a response (i.e., Hick's law) and have been considered for a long time as representative tasks in the cognitive domain (73). Similarly, Fitts' aiming tasks permit a measure of the amount of information processed to accurately stop the limb on a target (i.e., Fitts' law) and have been considered as representative tasks in the sensorimotor domain (67,74). Thus, it should possible to use a common metrics to quantify information processing rates in both cognitive and sensorimotor tasks. Consequently, one could determine whether processing capacities evolve with the same trend during aging, that is, if they become more and more correlated. If affirmative, one could affirm that change in structured fluctuations (i.e., complexity), slowing of motor responses and increases in intraindividual variability were high order markers of the dedifferentiation process. These markers would presumably capture the efficiency of the neurobehavioral system, as a result of increased coupling between cognitive and sensorimotor processes.

Concluding Remarks and Perspectives
Our conviction is that changes in complexity in the neurobehavioral system are the expression of subtle changes, which take place in active (though hidden) dimensions within the system (i.e., freezing/releasing of dynamical degrees of freedom). They lead to modifications of the coupling scheme and its implementation within the neurobehavioral system and, accordingly, its complexity. Searching for common markers of cognitive and sensorimotor aging and exploring their coupled evolution is a new mean for exploring changes in neurobehavioral complexity.
Identification of neurobehavioral markers of aging might have potential clinical applications for simplifying assessment of frailty and cognitive/sensorimotor disease. For instance, correlated changes in these markers might have a predictive value to predict an individual path to a loss of adaptability, frailty, and disease based on their actual values.
A caveat is in order however. At the moment, the complexity paradigm in aging research consists more of a unified way of thinking rather than a finished body of knowledge. Until a general theory appears, it is impossible to tell whether all the pieces of the aging puzzle are at hand that unify the facts well. We hope that the conceptual framework exposed in the present paper will contribute to provide a proper metatheory for aging and firmly establish the complexity approach as attractive and fruitful resources in the field of cognitive and sensorimotor aging.