Edging toward Entelechy in Motor Control

Abstract

The organization and functional logic of corticospinal motor neurons and their target connections remains unclear, despite their evident influence on movement. Spinal interneurons mediate much of this influence, yet we know little about the way in which corticospinal neurons engage spinal interneurons. This is perhaps not surprising given that the principles of organization of local spinal microcircuits remain elusive—we have glimpses of an underlying order but lack a comprehensive view of their functional architecture. In this brief essay we make a case that a new focus on the intersection of cortical and spinal circuits may provide clarity to the interpretation of corticospinal motor neuron firing patterns and help specify the logic of corticospinal motor neuronal function.

“If motion is such an ultimate term, then to define it by means of anything but synonyms is willfully to choose to dwell in a realm of darkness.…”

—Sachs (2005)

From Aristotle onward, we have realized that movement defines the human condition. It is, ultimately, what shapes our relationship with the external world. Over the course of evolution, with little tolerance for sloppiness or error, motor strategies have been sculpted into the implements of will, tasked with translating decision and desire into action. The neural circuits that underlie these motor strategies face daunting demands: sensory signals in a variety of forms are channeled into the nervous system, processed, and converted into action. The job of the motor system is to interpret this signaling cacophony and elicit movements that are both cohesive and effective. And each and every circuit for movement must design and focus its activity through the lens of the motor neuron. When viewed from this perspective, the fundamental challenge of the nervous system is to organize itself so as to orchestrate appropriate motor neuron activity—a challenge the logic of which we still have not come close to comprehending.

In their task of governing behavior, the activity of motor neurons is controlled collectively by spinal, descending, and sensory inputs. Defining how movement is achieved requires an understanding of the way in which local and long-range circuits are coordinated to generate patterned motor activity. Attempts to explore this process experimentally have usually focused on separating motor modules—those found, for example, in the spinal cord, brainstem, basal ganglia, cerebellum, and cerebral cortex—and interrogating their functions individually. This separatist approach has provided considerable insight into the way in which the engagement or removal of individual neuronal populations perturbs motor behavior. But, intuitively, it seems that the problem of movement will only be understood through analysis of the unified sum of its many parts. There may be a case, then, for combining an ever-improving capacity for fine-grained dissection of individual neurons and networks with a parallel emphasis on the mechanisms through which connected motor regions interact.

In this essay we focus on the link between the motor cortex and spinal cord—two elemental threads of an interwoven motor network—indicating gaps in our understanding of their connectivity and suggesting approaches that could begin to redress this state of comparative ignorance. The intent here is to edge toward a motor systems entelechy—the dynamic purpose encoded in a system—or, as Aristotle put it, a condition of actuality as opposed to potentiality. We also consider briefly whether lessons learned from motor systems have a more general applicability to other neurons, circuits, and behaviors.

Motor Neurons First and Foremost

The neural control of movement has been pursued at many different levels, both experimental and theoretical, with the aim of explaining the stereotyped action programs associated with locomotion as well as the goal-directed challenges of skilled arm and hand movements. Yet it is worth remembering that even for the control of sophisticated limb movements, the nervous system is merely a servant, charged with supplying limb musculature with information of biomechanical utility and validity. At several levels of organization, motor neurons respond to this demand by conforming to a spatial logic that respects the biomechanical constraints of their limb targets (Jessell et al., 2011; Romanes, 1964). First, individual sets of motor neurons segregate into myocentric pools within the ventral spinal cord. Second, motor pools that supply muscles with similar biomechanical roles at a joint cluster together into higher-order columelar groups. Third, motor columels destined to control progressively more proximal muscles are located at ever more ventral positions in the spinal cord. Fourth, motor columels that innervate antagonist muscles at a given joint are segregated spatially along the mediolateral axis of the spinal cord (McHanwell and Biscoe, 1981). Such topography is thought to facilitate the formation of sensory-motor circuits that direct motor poolspecific firing patterns during behavior (Sü rmeli et al., 2011). At a molecular level, the functional organization of motor neurons has its basis in the combinatorial expression of transcription factors (Philippidou and Dasen, 2013). Thus, a window into the functional organization of motor neurons has led to an appreciation of the primacy of biomechanics in defining the architecture of spinal motor circuitry.

These insights pose the question of the extent to which premotor circuits—those networks that provide key instructive input to motor neurons—are arranged similarly in a manner that respects limb axes. At present, the sole motor circuitry that has been defined in any significant detail is that of sensory feedback from limb muscles. From this sensory perspective group Ia proprioceptive afferents exhibit predictable and well-defined patterns of connectivity in which the innervation of homonymous motor pools, those supplying the muscle of sensory origin, is accompanied by the engagement of inhibitory interneurons that target antagonist motor pools—an anatomical design that underlies reciprocal inhibition in the stretch reflex circuit (Baldissera et al., 1981). Local central pattern-generating circuits presumably achieve a similar precision in coordinating the activation of flexor and extensor motor neurons—although here the fundamental features of organization of local spinal interneurons, and the principles at work in the selection of motor neuron targets, are far from clear.

Yet buried in the weeds of spinal interneuronal circuitry lies the ability of the motor system to respect or override specific motor programs in a goal- or task-dependent manner. The simple act of reaching, for example, requires a transition from alternation to synchrony in the activation of motor neurons controlling muscles at a single limb joint (Hyland and Jordan, 1997). To achieve this state switch, the reciprocal inhibitory constraints that are thought to ensure alternation of motor pool firing during the early phases of limb extension need to be overridden to permit the co-contraction of erstwhile antagonist motor neurons and muscles, helping to stiffen and stabilize the arm after its extension.

How is state switching achieved? Spinal inhibitory microcircuits appear to facilitate this flexibility (Nielsen and Kagamihara, 1992, 1993). But left to their own rhythmic devices, spinal interneuronal circuits appear to lack the capacity for transition between different motor states (Grillner, 2006). Descending motor control systems may then be critical arbiters of motor flexibility (Lundberg, 1967). If this is indeed the case, insight into motor task selection will emerge only when there is greater clarity about the way in which descending pathways interface with spinal interneuronal circuits. In this case study we therefore examine the general issue of connectivity between motor modules with a focus on corticospinal motor neurons (CSMNs) as an illustrative descending system, examining the links between the engagement of spinal interneurons, transitions in motor strategy, and the emergence of behavior.

The Palimpsest of Motor Cortical Activity

Early microstimulation studies established the sufficiency of motor cortical activity in directing movement and further suggested that semidiscrete subregions control the movement of distinct body parts (Penfield and Boldrey, 1937). Such topographic structure, however, says little about the precise operations performed by motor cortical networks. Moreover, more recent findings using longer-duration stimulation in monkeys and mice have raised the possibility that motor cortex may be more accurately subdivided on the basis of involvement in different categories of behavior—defensive postures or movements of the hand to the mouth as just two examples from the monkey (Graziano, 2006; Harrison et al., 2012). The behaviors on which these newer maps are based rely on limb trajectories that are characterized by coordinated movement across multiple joints—a feature that is likely to be reflected in the functional diversity of cortical neurons contributing to particular behaviors. Indeed, a number of distinct conceptual frameworks have been used to interpret motor cortical activity, and implicit in each framework are assumptions about the nature and function of motor cortical output.

The extent of the motor cortical conundrum is illustrated by the fact that even the simplest idea about the function of CSMNs—that they directly determine muscle activation via motor neurons—has been hard to validate or refute with any certainty. EMG patterns measured during movement can be well fit by summing the firing rates of motor cortical neurons (Morrow and Miller, 2003), including subsets that appear to target directly corresponding motor pools (Schieber and Rivlis, 2007). However, such fits are best achieved when a substantial delay (~50 ms) between firing and muscle activation is assumed. In addition, the activity of muscles whose motor pools appear directly innervated by a particular CSMN can show negative correlation or lack any discernible correlation with its firing (Kalaska, 2009). Other results suggest that during certain movements the firing of motor cortical neurons does not obviously track muscle activation (Shalit et al., 2012). Such disparities between CSMN and muscle activity may reflect the fact that muscles are driven primarily by descending inputs subject to significant transformation by spinal interneuronal networks.

A quite different approach to the problem of interpreting motor cortical activity has relied on encoding models in which neuronal firing encodes specific kinematic (e.g., joint angle or joint angular velocity) or kinetic (e.g., joint torque) features of movement, as distinct from muscle activation (Kalaska, 2009). One product of this approach was the demonstration that reach direction could be decoded from the firing of a population of motor cortical neurons using a vector sum (the “population vector”) of the preferred reach directions of each neuron (i.e., the direction of movement evoking maximal firing) weighted by their firing rate during the reach (Georgopoulos et al., 1982). But these and other related frameworks have thus far failed to yield general models that indicate how to map CSMN firing onto movement (Kalaska, 2009; Todorov, 2000). Instead, as new data have accumulated, models have become ever more convoluted—somewhat reminiscent of the way in which models of celestial mechanics became increasingly complex in attempting to account for movements of stars before the advent of the heliocentric theory. In such encoding frameworks, the job of translating movement parameters into muscle activation is left up to the spinal cord. But because we do not know how spinal circuits themselves perform such transformations, the issue of how motor cortical output is interpreted at the spinal level remains unresolved.

Yet another view of motor cortical activity has emerged more recently. Here, rather than fitting encoding models to firing rates, the focus has been on characterizing prominent collective patterns in firing across motor cortical neurons that can be captured by dynamical models (Shenoy et al., 2013). In this dynamical view, relevant patterns of collective firing may not bear much resemblance to the activity of any one motor cortical neuron. Collective firing patterns are presumed to arise from interactions among neurons, such that individual neurons can best be viewed as functioning in concert to generate output patterns needed to drive movement. Some components of collective firing may arise as a residue of pattern generation, while a separate subset reflects relevant output. This dynamical approach remains agnostic about what, if anything, motor cortical firing represents about movement. Models fit to firing data can generate sufficient structure to reconstruct EMG activity patterns (Churchland et al., 2012). However, sufficiency does not imply that the spinal cord is without a role in transforming descending input into motor pool activation patterns. All in all, we are left to conclude that relevant aspects of CSMN function need not be obvious from the scrutiny of single neurons and may emerge only from the collective behavior of the population.

One of the problems in trying to divine the basic units of CSMN function from the analysis of motor cortex per se is that the role of spinal circuits in mediating CSMN function remains ambiguous at best. Spinal interneurons indubitably intervene between CSMNs and motor neurons, and thus the way CSMN firing patterns influence movement will depend critically on the nature and organization of spinal circuits and the ways in which CSMNs engage them. Any comprehensive characterization of CSMN function, we would argue, will need to account for this dependence.

Cortical Motor Circuitry Viewed through the Lens of Spinal Interneuron Organization

Most mammalian CSMN axons, and seemingly all of them in nonprimates, synapse not onto motor neurons, but onto interneurons located in the intermediate and dorsal zones of the spinal cord (Kalaska, 2009). Thus, evolutionarily conserved polysynaptic corticospinal pathways, channeled through spinal interneurons, are likely of crucial relevance to the translation of cortical motor output. Because spinal interneurons are tasked with integrating CSMN input, along with information from sensory afferents and other descending pathways, the link between CSMN activity and motor behavior is likely to represent only one element of a larger logic of spinal motor circuitry.

Here, we consider two potentially informative ways of probing the organization of spinal interneuron classes and motor networks, with a view to clarifying the contribution of cortical commands (Figure 1). The first is the “degree of separation” factor: the question of how many synapses removed from direct contact with motor neurons are different spinal interneuron subtypes. The second is the issue of how local interneurons assemble themselves with respect to their motor neuron targets: do some interneuron subtypes function as motor pool “specifists” and others as deliberate “generalists”?

Figure 1. Strategies for Spinal Motor Control.

Distinct motor neuron pools innervate limb muscles with different biomechanical functions at specific joints. The logic of engagement of motor pools by spinal interneurons and descending inputs, including those from motor cortex, remain unclear. This diagram attempts to capture some of the many unresolved issues about interneuron and motor neuron engagement, with an eventual emphasis on the way in which descending inputs engage spinal interneurons. (A) Two possible modes of engagement of motor pools by zero-order premotor interneurons. Specifist engagement indicates the selection of motor pools in a manner that respects basic biomechanical function—flexor or extensor functions at a joint. Group Ia inhibitory interneurons, and possibly Renshaw interneurons, represent examples of this interneuron category. Generalist engagement links pools of differing function, as may underlie the simultaneous activation of muscle groups during particular behaviors. Cervical propriospinal interneurons involved in reaching movements represent one example of this class, and V0_C interneurons may represent a second example.

(B) Two possible modes of organization of first-order, nonpremotor interneurons. Specifist neurons obey the rules of basic motor pool biomechanics, whereas generalist neurons supercede them. Some generalists may form recurrent connections with themselves or other interneuronal populations, eroding the clarity of this hierarchical organization. Hb9⁺ interneurons and GAD2⁺ GABApre neurons appear to be examples of first order arrangement, but their roles as specifists or generalists have not been resolved.

(C) Engagement of spinal interneurons by sensory (S) or corticospinal (CS) projection neurons. The left-hand diagram shows one example of specifist engagement—the ability of group Ia proprioceptive afferents to capture a given motor pool and concurrently to activate reciprocal inhibitory interneurons that target an antagonist motor pool. The right-hand diagram indicates a generalist CSMN that innervates zero-order premotor interneurons. Such an arrangement could conceivably contribute to the task-dependent transition from motor pool alternation to co-contraction. Mixing and matching descending and local interneuron rules of engagement offers numerous possibilities for motor flexibility.

For details see text.

Resolving these two questions first demands an appreciation of just how many different interneuron subtypes exist. From developmental studies we know that spinal interneurons have a positional provenance, with four cardinal progenitor domains arranged along the dorsoventral axis of the ventral cord giving rise to the V0, V1, V2, and V3 interneuron classes, each with its own distinctive molecular identities and axonal projection patterns (Grillner and Jessell, 2009). These cardinal subdivisions, while shown to be of relevance in constraining connectivity, appear only to scratch the surface of interneuron diversity. Molecularly, we already know of vanishingly small interneuron subsets that have measurable roles in motor control—the V0_C and Hb9 interneuron subtypes, for example, represent only 2%–3% of their parental populations (Wilson et al., 2005; Zagoraiou et al., 2009). By extrapolation, these and other studies indicate the existence of many dozens of molecularly, anatomically, and perhaps functionally different interneuron subtypes relevant to motor control. At the very least, the expression of defining molecular markers for many of these subtypes offers a way of examining their organization and function in a systematic and objective manner.

In some instances it has been possible to fit defined interneuron subtype within the “degree of separation” framework. V1 and V2a interneurons include zero-order populations that target motor neurons (Kiehn, 2011) (Figure 1A), permitting direct inhibitory or excitatory control of motor neuron activity. But it is still unclear whether all neurons within a cardinal division serve as zero-order premotor neurons. And the degree to which premotor interneurons are motor pool specifists or generalists remains unclear (Figure 1A). So-called group Ia interneurons that mediate reciprocal inhibition demand stringent targeting of specific motor pools (Eccles and Lundberg, 1958) and thus represent specifists. In contrast, other interneuron classes have been shown to coordinate the activity of multiple motor pools dedicated to the control of individual limb segments (Takei and Seki, 2010), or even segments across multiple joints (Tantisira et al., 1996) and thus may be generalists. Recent advances in genetically restricted transsynaptic tracing provide hope that some of the details of premotor interneuron organization will soon fall into place (Arber, 2012).

For first-order interneurons—those that are one interneuron removed from motor neurons—the picture is inevitably more complex (Figure 1B). A few interneuron classes of relevance to motor control have been shown to shun contact with motor neurons—notably, GAD2⁺ presynaptic inhibitory neurons, and rhythmogenic Hb9⁺ interneurons (Betley et al., 2009; Wilson et al., 2005)—but the target specificity of these neurons with respect to motor pool organization is far from clear. Moreover, closely related and molecularly coherent interneuron classes need not necessarily respect equivalent degrees of separation—V0_C and V0_G interneurons are derived from the same Pitx2⁺ subset of V0 neurons, yet differ in neurotransmitter phenotype and occupy different premotor positions—cholinergic V0_C interneurons prominently target motor neurons whereas V0_G interneurons appear instead to target interneurons (Zagoraiou et al., 2009).

Do some spinal interneurons exhibit higher degrees of separation—residing two or more interneurons removed from motor neurons? Perhaps not. It seems unlikely that interneuron organization is strictly hierarchical, as recurrent interconnectivity could position all interneurons within a couple of synapses of motor neurons. Moreover, the shortest route to a motor neuron may not be the only functionally relevant route, as it may ignore other critical recurrent or feedforward connectivity within spinal circuits. Indeed, in the absence of recurrence, spinal circuits would be reduced to a feedforward architecture that would have trouble accounting for pattern generation (Grillner, 2006). It follows then that individual interneurons could exist many different synaptic distances away from motor neurons. One severe limitation in resolving the principles of spinal motor microcircuitry is the paucity of data that speaks to the interconnectivity among interneuron subtypes. Instances of identified interneuron interconnectivity have been established, notably between V2a interneurons and commissural interneurons (Crone et al., 2008), but the connectivity matrix for most subtypes has not been determined. And in the likelihood that there are dozens of motor-relevant interneuron classes, the task of delineating their connections, while daunting, cannot be ignored.

The perturbation of molecularly defined interneuron classes has revealed disruption of elements of locomotor output (Goulding, 2009). The encoding of left-right alternation seems particularly fragile, implying that the circuits that enable the opposition of contralateral limb movement are more susceptible to the loss of individual interneuron types. In contrast, the neurons that control flexor-extensor alternation and rhythm generation have been harder to pinpoint or disrupt, suggesting that emergent circuit functionality might be robust enough that no single subtype is indispensable for these tasks. Thus, while some features of motor systems appear assigned to individual interneuron types, many functions are likely to be more widely distributed among populations (Briggman and Kristan, 2008). Thus, interneuronal diversity likely contributes to functional flexibility in spinal circuits both by enabling discrete subtypes to play distinct roles and by enabling diverse interactions among interneurons that support a broader array of emergent behaviors.

Where the Rubber Meets the Road

How do descending systems exploit the anatomical and functional features of spinal circuit organization in the selection of task-specific motor output? As emphasized, the ambiguities of spinal circuit architecture limit any understanding of the way in which CSMN inputs engage spinal circuits.

Nevertheless, questions about descending cortical organization and its translation can be posed in the dialect of spinal interneuronal networks. Do some CSMNs serve as specifists and target particular interneuron subtypes, while others serve as generalists and engage a range of subtypes (Figure 1C)? Do some interneuron subtypes simply evade CSMN input? Do CSMNs target spinal interneurons selectively, on the basis of their “degree of separation” status, or their role in pattern generation? Do distinct CSMNs target zero-, first-, and even second-order interneurons? And just how many different subtypes of CSMNs are there? The more complete the molecular definition of spinal interneuron subtype, the easier it will be to resolve these questions. Practically, methods for retrograde transsynaptic tracing from, and anterograde synaptic mapping onto, defined interneuron subtypes are now working (Arber, 2012), and the major limitation may simply be to define better the molecular grain of interneuron diversity.

At a functional level, the engagement of spinal interneurons by CSMNs could provide insight into the mechanisms of motor state switching. The coordination of muscle contraction with reduced activation of antagonist muscles via reciprocal inhibitory pathways has long been established (Sherrington, 1897). However, antagonist muscle pairs are known to co-contract at certain times during movement (Smith, 1981; Tilney and Pike, 1925). Antagonist co-contraction is observed in humans during voluntary elbow rotations (Patton and Mortensen, 1971), isometric clasping of the hand (Long et al., 1970), and walking along a balance beam (Llewellyn et al., 1990). Co-contraction will stiffen and stabilize joints, which may aid in the performance of new motor tasks, or those subject to unpredictable perturbations. Spinal pathways have been implicated in suppressing reciprocal inhibition mediated by inhibitory group Ia interneurons in order to promote co-contraction. During voluntary co-contraction of antagonist ankle muscles, this suppression has been shown to involve enhanced recurrent inhibition of Ia interneurons as well as an increase in presynaptic inhibition of group Ia afferents that excite Ia interneurons, though the mechanisms underlying co-contraction at the wrist appear distinct (Pierrot-Deseilligny and Burke, 2006).

Cortical output during voluntary co-contraction is unlikely simply to reflect the combination of separate drives for activating two antagonist muscles. Recordings from motor cortex have detected units specifically active during co-contraction (Humphrey and Reed, 1983). Some CSMNs facilitate activation of certain wrist muscles but suppress their antagonists—and these have been shown to fire during flexion and extension movements but can cease during isometric clasping (Fetz and Cheney, 1987). Moreover, the suppression of group Ia inhibition during the co-contraction of ankle antagonists is far greater than that expected based on the inhibitory activity observed during activation of either muscle alone (Nielsen and Kagamihara, 1992). Lastly, measurements of cerebral blood flow (Johannsen et al., 2001) and EEG-EMG coherence (Hansen et al., 2002) suggest that distinct corticospinal pathways may be active during co-contraction of ankle antagonist muscles compared to the separate activation of either muscle alone.

If parallel descending pathways exist, how do they engage spinal circuits? A pathway involved in co-contraction could directly target interneurons mediating recurrent and presynaptic inhibition. Exploiting genetic access to measure and perturb activity in CSMNs targeting these interneurons could implicate the involvement of particular spinal targets in a co-contraction pathway. It is also possible that the generation of appropriate motor neuron drive during co-contraction involves indirect pathways through other spinal interneurons or descending relay systems. Intriguingly, measurements of forelimb EMG in rats during a reach-to-target task show distinct movement phases in which antagonist muscles either alternate activation or co-contract (Hyland and Jordan, 1997). Nevertheless, it is still possible that there is substantial overlap in the CSMNs active during co-contraction and flexion-extension movements and that temporal patterning of CSMN output is critical to differential recruitment of motor neurons. Cellular resolution functional imaging of CSMN populations during different behaviors (Dombeck et al., 2007) could reveal spatiotemporal activity patterns in different motor tasks and strategies.

The issue of how CSMN output is structured with respect to spinal pattern generating circuits is also crucial to resolve. At one extreme, CSMN input might simply bypass pattern generating circuitry during voluntary movement, targeting short feed-forward pathways in order to elicit appropriate patterns of excitatory and inhibitory input onto motor pools. The demonstration of CSMNs whose firing drives monosynaptic excitation and sometimes disynaptic inhibition of motor pools suggests that this can occur (Lemon et al., 2004). But the extent of monosynaptic motor neuron connections by CSMNs is limited, even in primates, and analysis of these direct connections may not be particularly informative about the other command roles of CSMNs. In addition, CSMN input may engage the pattern-generating capacity of spinal circuits in guiding a broad range of voluntary movements most of which bear little resemblance to locomotion. Sensory feedback, extrinsic drive, and neuromodulation regulate the rhythmic locomotor firing patterns that spinal circuits generate (Guertin, 2009), manifesting a flexibility that could be critical for the production of more complex movements. CSMN inputs could, for instance, target particular spinal interneurons and provide an input that fluctuates over time so as to elicit movements that differ from locomotion but leverage interactions among spinal interneurons that otherwise support locomotion.

As an example of CSMN engagement of spinal circuits, we consider a voluntary reaching movement involving flexion and extension at forelimb joints, as when a cat reaches out to swat a toy. Much of the output of CSMNs that guides such movements may simply be fed forward through spinal interneurons without eliciting interactions among interneurons that sustain pattern generation. Alternatively, CSMN input may drive pattern-generating circuits so as to elicit a modified version of a step forward equating to the reach. By patterning CSMN input with a particular time course onto select interneurons, interactions among spinal interneurons could be harnessed to shape idiosyncratically the drive to motor neurons that will elicit the reach. Though the generation of locomotor activity can be self-sustaining, descending input could in theory be patterned onto interneuronal circuits so as to elicit motor outputs of variable duration.

How does this view mesh with other notions of spinal organization? It has been proposed that spinal circuitry comprises behavioral modules—circuits that generate specific elementary motor outputs, sometimes called motor primitives—which can be combined together to produce coherent movement (Bizzi et al., 2008). This concept has been useful in trying to make sense of the strong but variable correlations in activation across muscles during different behaviors. But if these modules comprise pattern-generating circuits, each subject to temporally and spatially patterned extrinsic drive, their outputs need not be discrete and stereotyped but can vary according to the structure of the extrinsic drive. In this view, CSMN drive to spinal circuits does not simply initiate module output, but rather the pattern of this drive determines the dynamic behavior of the module.

Regardless of the way in which CSMNs negotiate spinal circuits, it is clear that both the identity of their postsynaptic partners and their temporal pattern of activity are critical. Thus the functional organization of CSMNs may be resolved only by combining measurements of CSMN activity and target connectivity. In describing this functional organization, the most instructive elemental components will likely involve some marriage of both spatial (postsynaptic partner) and temporal (activity) information. Such components may arise from distinct subsets of CSMNs or in a more distributed, combinatorial fashion from many or all CSMNs in different proportions. Methods for decomposing connectivity and activity will have to allow for both possibilities.

Lessons from the Path to Motor Entelechy

What can we learn from the intersection of CSMNs and their spinal targets? This brief account was intended to convey two main messages. First, that our current understanding of CSMN functional organization remains starkly limited. And second, that despite the present impasse, analysis of the intersection of cortex and spinal cord through a judicious combination of genetic manipulation, connectivity mapping, activity measurement and perturbation, behavioral quantification, and network modeling offers considerable promise for progress. Traditionally, the physiology of motor cortex and spinal cord has been examined in quite different behavioral contexts. The study of motor cortex has focused on isolated forelimb movements, whereas examination of spinal motor circuits has tended to focus on locomotion. Comparative approaches that probe general principles of motor circuit function, transcending specific muscle and task, may provide a richer seam of information. In this context, we consider that a genetically tractable mammalian organism like the mouse can have its place alongside primates in the analysis of motor cortex, even though concerns can be raised about the variable design of motor systems across mammals (Lemon, 2008).

Many other supraspinal centers of immediate relevance to motor control connect with spinal targets, and some even engage motor neurons directly with different target specificities. Understanding how CSMNs engage spinal interneuronal circuits may shed light on how other descending pathways do so. Distinguishing the logic by which other projection pathways connect to spinal interneurons may help reveal further rules of spinal circuit engagement. More generally, though, there is a need to find ways of defining the coherent strategies through which descending systems act coordinately to influence motor neuron output.

Do the interim lessons drawn from the study of motor system circuitry and function have a broader relevance—to the challenges inherent in linking neural organization to encoded behavior? Several thoughts suggest themselves. First and foremost, motor systems offer the singular virtue of a rather direct link between the organization of a neural circuit and its behavioral output—in this case, patterned muscle contraction. In the case of the motor neuron, its muscle target soon becomes a fixed and inseparable component of the “motor unit,” such that much of the neural computation inherent in the CNS is involved with the planning and execution of spinal motor programs. Understanding how the behaviors encoded by other CNS circuits impinge on core motor routines could lead to more objective and quantitative ways of evaluating the world of complex behavior.

Studies of spinal motor neurons have also served to remind us of the primacy of limb biomechanics in assigning functional order to motor circuits. Along the way, these studies have revealed that the location of a motor neuron or interneuron in the spinal cord constrains many of its potential connections, permitting some and excluding others. It may be worthwhile considering whether this positional principle extends beyond the spinal cord, and beyond the motor system. The prominence of nuclear organization as a means of positioning neurons throughout the subcortical CNS, together with the critical influence of neuronal settling position in defining patterns of sensory input connectivity, suggests that position may be a crucial determinant of connectivity throughout the vertebrate CNS. The trick in testing this assertion is the accumulation of sufficient molecular information on neuronal subtype to alter settling position without eroding core identity, and examine the subsequent impact on connectivity and behavior.

In the motor system as elsewhere, neuronal circuit models commonly suffer the weakness of being poorly constrained by existing information on connectivity within and between neuronal populations. When pursued alone, even the most contemporary methods for inferring circuit architecture from activity measurements fail to specify unambiguously the underlying circuit mechanisms that biology implements. In the same way that methodological advances in structural biology have helped to trim a seeming infinity of plausible protein models, we anticipate that increasingly detailed circuit mapping will produce constraints on neuronal circuit models that sharpen our understanding of their functional architecture.

A final inference to be drawn from this motor system precedent is that there may be considerable mileage to be gained from studies of the intersection of anatomically separable regions devoted to the control of a given behavior. Evolution and development have met the challenge of interconnecting the dozen or more brain regions conscripted to the challenge of motor control. Perhaps the interface offers an opportunity to learn about the logic of two behaviorally related systems for the price of one interrogation.