Cuneate nucleus: The somatosensory gateway to the brain

Abstract

Much remains unknown about the transformation of proprioceptive afferent input from the periphery to the cortex. Until recently, the only recordings from neurons in the cuneate nucleus (CN) were from anesthetized animals. We are beginning to learn more about how the sense of proprioception is transformed as it propagates centrally. Recent recordings from microelectrode arrays chronically implanted in CN have revealed that CN neurons with muscle-like properties have a greater sensitivity to active reaching movements than to passive limb displacement, and we find that these neurons have receptive fields that resemble single muscles. In this review, we focus on the varied uses of proprioceptive input and the possible role of CN in processing this information.

Introduction:

Proprioception is generated by a variety of receptors that encode mechanical strain and deformation caused by the movement of all parts of the body, including the trunk, head and limbs. Chief among these receptors are muscle spindles that encode muscle length and the speed of length change (Proske & Gandevia, 2012). Golgi tendon organs that respond to active muscle force, joint receptors responding to loads and extreme positions, and skin receptors activated by movement-related stretch of the skin (J. Houk & Simon, 1967). This diverse set of receptors supplies information throughout the cerebral cortex and cerebellum and underlies all aspects of proprioception, from simple spinal reflexes to complex supraspinal reflexes as well as the planning and execution of voluntary movements. Information from these same receptors is also necessary for the conscious perception of the position and motion of our limbs, a perception that remains largely in the background (Proske & Gandevia, 2012) causing it to be referred to colloquially as the “hidden” sixth sense.

A significant portion of afferents from these receptors project directly or indirectly to a caudal brainstem region referred to as the dorsal column nuclei (DCN) complex (Loutit, Vickery, & Potas, 2021; Mountcastle, 2011). This complex of nuclei is in an ideal position to regulate these inputs. Early work examining their structure and function was primarily conducted on cat models, almost always while under sedation (Andersen, Eccles, Oshima, & Schmidt, 1964; Andersen, Eccles, Schmidt, & Yokota, 1964; Cooke, Larson, Oscarsson, & Sjölund, 1971; Rosén & Sjölund, 1973), with notable exceptions (Ghez & Pisa, 1972).

We have recently begun to record in awake monkeys from the cuneate nucleus (CN) (Suresh et al., 2017), the portion of the DCN that carries signals from the arms to the thalamus (Rosén & Sjölund, 1973). Such recordings now allow us to make observations that were previously impossible under sedation. For example, our results show that the sensitivity of many CN neurons differs for actively generated reaches and passive limb displacements of the arm. Those neurons that appear to receive input from muscle spindles are typically more sensitive during active movement. Furthermore, we found that the tuning of CN neurons for movements in different directions is quite similar to what we would expect from receptors of a single muscle, matching the results of previous studies using single muscle stretches (Hummelsheim & Wiesendanger, 1985; Rosén & Sjölund, 1973) but contrasting with a study using electrical stimulation of peripheral nerves (Witham & Baker, 2011). In this review, we will attempt to reconcile the apparent inconsistencies in the previous literature, focusing on two major areas: proprioceptive gain modulation and convergence of afferent input in DCN. In doing so, we hope to provide a perspective from which to examine previous DCN research and to design new studies to illuminate how proprioceptive information is processed as it moves from the periphery to the brain.

Proprioceptive gain modulation in the cuneate nucleus:

Sensory gating, or the attenuation of afferent input, is a feature of many sensory systems (Azim & Seki, 2019). During saccadic eye movements, visual information is attenuated to avoid blurred images caused by the movement of the eye (Binda & Morrone, 2018; Bremmer, Kubischik, Hoffmann, & Krekelberg, 2009; Crevecoeur & Kording, 2017; Holt, 1903). Similarly, tactile sensations arising during active touch are significantly weaker than the same stimuli presented passively (Cohen & Starr, 1987; Schmidt, Schady, & Torebjörk, 1990). These observations have led to the hypothesis that the nervous system turns down the gain on sensory receptors when the information they are transmitting is likely to be noisy (Ghez & Pisa, 1972). As the somatosensory gateway to the brain, neurons in the DCN complex are a logical site of proprioceptive gating.

Consistent with the sensory gating hypothesis, CN receives descending signals from the somatosensory and motor cortices (Andersen, Eccles, Schmidt, et al., 1964; Leiras, Velo, Martín-Cora, & Canedo, 2010; Loutit et al., 2021). Their effect on afferent transmission has been the subject of experiments conducted mostly in anesthetized cats. Stimulation of these cortical areas leads to both excitatory and inhibitory effects, though early studies focused primarily on the inhibitory ones (Aguilar, Rivadulla, Soto, & Canedo, 2003; Andersen, Eccles, Oshima, et al., 1964; Andersen, Eccles, Schmidt, et al., 1964). Much like the effect of cortical stimulation, the afferent volley from stimulating the second of two peripheral nerves in close succession is markedly attenuated, suggesting that inhibitory circuitry within CN also contributes to the attenuation of afferent signals (Andersen, Eccles, Oshima, et al., 1964).

The potential functional role of this afferent attenuation was studied more directly by recording medial lemniscus field potentials evoked by stimulation of the tactile superficial radial nerve in cats (Ghez & Pisa, 1972). The resulting afferent volleys, which would have been generated by axons supplying RFs throughout the forearm and paw were attenuated during stepping. Without finer spatial resolution, it would have been difficult to see combined enhancement and attenuation of the effects, if it were there. As a means to determine the mechanism giving rise to the attenuation they applied Wall’s technique (Wall, 1958), which measures the amplitude of the antidromic potential in the peripheral nerve in response to CN stimulation. This amplitude is correlated with the extent of depolarization in the presynaptic terminal, called primary afferent depolarization (PAD), itself an indirect measure of presynaptic inhibition. In these experiments, PAD increased in a velocity-dependent manner throughout a step, suggesting that presynaptic effects on the inputs to CN mediate at least some of the sensory gating of tactile signals.

The problem of the blurring of retinal images during rapid eye movements was recognized already in the 11^th century by the Persian scholar Alhazen (Saliba & Sabra, 1992). Over 100 years ago, Holt proposed that vision is simply suppressed during saccades (Holt, 1903), but we now know that a more selective filtering of visual input occurs (Binda & Morrone, 2018). The sensation of a shirtsleeve sliding over the skin during reaching may be analogous to blurred vision during a saccade, contributing noise that the somatosensory system might appropriately attenuate. However, uniformly gating all somatosensory signals during movement, including muscle length changes or unexpected object contact, could cause blindness to critical sources of feedback. In CN, perhaps as in the visual system, there is evidence of gain modulation that is more complex than simple gating (Leiras et al., 2010; Palmeri, Bellomo, Giuffrida, & Sapienza, 1999). The experiments described above that yielded predominantly inhibitory effects in CN (Andersen, Eccles, Oshima, et al., 1964) relied on broadly distributed cortical stimulation. In other experiments that matched the receptive field of the stimulated cortical area to that of the CN neuron, the effect was typically excitatory. As the receptive fields became more dissimilar, the effect of stimulation was more likely to be inhibitory (Palmeri et al., 1999), leading to a “spotlighting” effect.

While these results were for CN neurons with cutaneous receptive fields, the idea of more flexible gain modulation might well apply broadly across the somatosensory system. We investigated this question with extracellular recordings from implanted electrode arrays that allowed us to record single CN neurons from awake, behaving monkeys. We compared the movement sensitivity of CN neurons during reaching to that of passive limb perturbations. Fig 1A shows the response of one example neuron that appeared to receive input from the anterior deltoid. We fit sinusoidal tuning curves to the responses and found the neuron’s preferred direction (PD) using simple linear models (Georgopoulos, Kalaska, Caminiti, & Massey, 1982).

Figure 1:

CN neurons with muscle-like inputs tend to respond more strongly to reaching movements than to passive arm perturbations. A) Responses recorded from a single CN neuron that appeared to receive input from muscle spindles in the anterior deltoid. The monkey grasped the handle of a planar manipulandum and made “center-out” movements in eight directions (left group of eight responses). We applied force perturbations in the same eight directions when the hand was at rest in the center-hold position prior to 50% of the movements (right group of responses). Raster plots (above) and trial-averaged firing rate histograms (below) are shown for each movement direction, positioned relative to the center of each group of plots. In the center of the plots is the tuning curve of the neuron. Overlaid on the raster plots are trial averaged hand speed traces for each direction. B) Scatter plot relating the firing rate of the example neuron in A to the hand speed in the PD. Each data point represents a single 10 ms time bin, color coded by condition. Blue and orange lines represent the best linear model fit from hand velocity to firing rate. C) Summary of the active and passive sensitivity of spindle-receiving CN neurons (filled circles, three monkeys,48 neurons) and somatosensory cortical area 2 (open circles, two monkeys, 86 neurons) neurons. D) Percentage of neuros with’ sensitivity that was significantly enhanced (+), was unchanged (0), or was attenuated (−) in the active case compared to the passive case.

In addition to deriving the PD of each neuron, these linear models allowed us to infer the sensitivity of each neuron’s firing rate to the speed of movement. We compared these inferred sensitivities between the active and passive conditions. The slope of the fitted lines in Fig 1B represents the sensitivity for both active (blue) and passive (orange) limb movements. In this example, the sensitivity of the active movements was larger (1.3 Hz/(cm/s)) than that of the of the passive condition (0.8 Hz/(cm/s)). Across all muscle-like CN neurons (those that had receptive fields that resembled muscles with no tactile response), the active sensitivity tended to be greater than the passive sensitivity (Fig 1C, filled circles). To make statistical comparisons between active and passive sensitivity, we used bootstrapping to estimate the confidence interval of the sensitivity difference for each neuron (Efron & Tibshirani, 1986). We then counted those neurons with significantly enhanced or attenuated sensitivity. Across three monkeys, the sensitivity of muscle-like CN neurons was more than twice as likely to be enhanced during active reaching than attenuated (Fig 1D, black bars). There was no such bias in CN neurons with tactile receptive fields (Fig 1D, gray bars). We also performed this analysis for neurons recorded under the same conditions from area 2, a mixed cutaneous and proprioceptive area of cerebral cortex. We found that unlike CN, area 2 neuron sensitivities were somewhat more likely to be attenuated during active movement than enhanced (open symbols and bars, Fig 1 C, D), in contrast with an earlier study whose methods didn’t take into account differences in kinematics and found no significant difference across conditions (London & Miller, 2013). This may reflect additional attenuation that occurs after signals pass through CN, for which there is some evidence (Chapman, Jiang, & Lamarre, 1988; Dale & Cullen, 2017). The functional role of this added inhibition is not clear.

Gain modulation in CN could arise from multiple sources including descending modulatory input to CN, altered gamma drive to muscle spindles, and altered transmission of the afferent input through spinal interneurons. Muscle spindles receive descending gamma drive that directly modulates their sensitivity (Prochazka, Hulliger, Zangger, & Appenteng, 1985). During locomotion, gamma drive is modulated substantially, particularly so during less stereotypic gait (Bennett, De Serres, & Stein, 1996; Ellaway, Taylor, & Durbaba, 2015). Although gamma drive has the potential to explain the context-dependence that we observe in CN, its modulation during reaching has not been well studied and extrapolating to reaching from quadrupedal locomotion in cats is problematic (Jones, Wessberg, & Vallbo, 2001). Experiments using methods insensitive to gamma drive, such as measurement of PAD, have found similar enhancement in proprioceptive spinal interneurons (Confais, Kim, Tomatsu, Takei, & Seki, 2017), evidence that reach-related enhancement is likely not wholly due to alterations in gamma drive. Experiments designed to further identify the site or sites of proprioceptive gain modulation would make an important contribution to our understanding of this system.

Functionally, gain modulation serves at least two purposes. First, it can enhance or attenuate the intensity of the conscious experience of a sensation, as demonstrated in previous psychophysical studies (Juravle, Binsted, & Spence, 2017; Schmidt et al., 1990). Perhaps more importantly, sensory gain must be optimized for motor control. For example, the gain of the stretch reflex is reduced in muscles that would otherwise oppose the generation of fast movements (Adams & Hicks, 2005). Throughout the gait cycle of normal walking, the stretch reflex is maximal during stance and completely suppress in the transition from stance to swing (Sinkjær, Andersen, & Larsen, 1996). Recently, groups have begun to investigate the consequences of disrupting these gain-modulating pathways, leading to profound motor deficits, including oscillatory movements that are consistent with an underdamped feedback control system (Fink et al., 2014). Gain modulation at every level of the somatosensory neuraxis (including fusimotor drive to the spindles) likely underlies the flexibility of multiple hierarchical feedback control loops (Kurtzer, Pruszynski, & Scott, 2008; Nashed, Crevecoeur, & Scott, 2014; Pruszynski et al., 2011; Scott, 2004, 2016; Scott, Cluff, Lowrey, & Takei, 2015; Weiler, Gribble, & Pruszynski, 2019).

Convergence properties in CN and area 2

In addition to selective modulation of gain, sensory afferent pathways may also combine inputs across space and differing modalities. The evidence for such convergence in CN is mixed. In one study, 87% of CN neurons responded to electrical stimulation of more than one peripheral nerve, even across modalities (Witham & Baker, 2011). Other experiments, in which individual muscles were stretched, found very little convergence (Hummelsheim & Wiesendanger, 1985; Rosén & Sjölund, 1973).

We estimated the extent of convergence in CN with two complementary methods: mapping receptive fields using vibratory stimuli and examining the spatial tuning of single CN neurons during passive arm movements. A good fraction (~50%) of neurons in CN that appeared to have muscle-like receptive fields from manual testing responded robustly to ~100 Hz muscle vibration, a stimulus that strongly activates muscle spindles (Fig 2A). Fig 2B shows a neuron with a phase-locked response to vibration with a lag of ~8 ms from the peak voltage driving the vibrator. Despite these strong responses from individual muscles, it was quite rare that a given CN neuron could be driven by vibration of more than one muscle. Attempts to evoke similar responses in area 2 were uniformly unsuccessful. We speculated that the inability to drive area 2 neurons may be due to their receiving convergent input not only from multiple muscles but also cutaneous afferents, thereby diluting the effect of the spindle input from a single vibrated muscle. It would be informative to repeat this experiment in thalamus and somatosensory cortical area 3a, as both regions have neurons which receive exclusively muscle inputs.

Figure 2:

Neurons in CN are strongly activated by 100 Hz sinusoidal vibration. A) Example CN response to vibration of brachialis. During 100 Hz vibration (grey box), firing rates increased to ~100 Hz, and returned to baseline immediately after stimulation ended. B) Time-dependent probability of the occurrence of the first spike after peak indentation suggests that this example CN neuron was phase-locked to the vibration.

We found a striking nonuniformity in the distribution of CN PDs (Fig. 3A) and asked whether it might also be evidence of limited convergence. To this end, we used DeepLabCut, a markerless motion tracking system (A. Mathis et al., 2018), an OpenSim musculoskeletal model (Chan & Moran, 2006; Delp et al., 2007), and a simple model of the spindle response to muscle length change (a one-half power law mapping muscle lengthening to firing rate (J. C. Houk, Rymer, & Crago, 1981)) to simulate the activity of muscle spindles throughout the 18 major muscles of the arm during the passive limb movements. These simulated muscle spindle PDs were also highly nonuniform, falling primarily along the axis towards and away from the body, qualitatively like that of CN (Fig 3B). We reasoned that convergence of multiple muscles would cause a significantly more uniform distribution.

Figure 3:

Both CN and area 2 appear to inherit strongly bimodal distributions of preferred direction from the biomechanics of the arm during passive limb displacement. A) Distribution of CN preferred directions during passive arm movements. B) PD distribution from a population of simulated proximal arm muscle spindles. C) Distribution of PDs for area 2 neurons for passive arm movements. D) Convergence of simulated muscle spindle afferents from multiple muscles slightly decreases mean absolute deviation from uniformity. Inset polar histograms show the PD distribution for simulated spindles from different numbers of muscles. Deviation from nonuniformity for actual CN and area 2 distributions plotted at the extreme of the plot. Shaded areas indicate one standard deviation of the mean across bootstrap iterations.

However, when we analyzed area 2 similarly, we found those PD distributions to be only slightly more uniform than CN, but not statistically so (Fig 3C). This was unexpected, given out intuiton about convergence and an earlier report of a PD distribution in area 2 that was by eye, more nearly uniform (Prud’homme & Kalaska, 1994). Accordingly, we simulated the PD distributions neurons receiving convergent excitatory and inhibitory inputs from the spindles of different numbers of muscles, examining the changes in distribution with increasing convergence. While this slightly increased the distribution uniformity, the change was considerably less than we anticipated (Fig 3D) indicating that this tool is too crude to address the question of convergence with any precision.

Proprioceptive neuroscience is in need of better tools to precisely measure and control the relevant movement-related variables. Unlike vision or touch, which offer the means to activate receptors with nearly arbitrary spatial and temporal patterns (Killebrew et al., 2007; Korenberg & Naka, 1988), the mechanics of the muscles of the limb cause virtually unavoidable correlations during natural movements (Mollazadeh, Aggarwal, Thakor, & Schieber, 2014; Santello, Flanders, & Soechting, 1998). Opto- and chemogenetic methods are promising, potentially allowing for fine-grained experimental circuit dissection and control of afferent signals during behavior (M. W. Mathis, Mathis, & Uchida, 2017; Sauerbrei et al., 2018; Smith, Bucci, Luikart, & Mahler, 2016; Tashima et al., 2018), including targeted activation of muscle spindle afferents (Kubota et al., 2019).

Proprioceptive streams and their relevance to motor control

For a “hidden” sense, proprioception plays several vital roles. Proprioceptive inputs to the anterior and posterior parietal cortices, as well as to the secondary somatosensory cortex in the superior bank of the Sylvian fissure, contribute individually to a variety of disparate functions, including movement planning, online movement correction, as well as the conscious perception of limb state (Pavlides, Miyashita, & Asanuma, 1993; Rushworth, Nixon, & Passingham, 1997; Wolpert, Goodbody, & Husain, 1998). The ideal location of the dorsal column nuclei to combine and modulate these inputs for the diverse function they subserve (Loutit et al., 2021) is the final topic of this review.

Area 2 is the earliest cortical area with a large proportion of neurons having combined cutaneous and muscle inputs. For this reason, some consider it not to belong with areas 3a, 3b, and 1 as part of S1. The confluence of these inputs within hand area 2 is thought to be important for stereognosis, for which a knowledge of hand conformation combined with object contact points is critical (Gardner, Babu, Ghosh, Sherwood, & Chen, 2007; Rincon-Gonzalez, Warren, Meller, & Helms Tillery, 2011; Yau, Kim, Thakur, & Bensmaia, 2016). The role of arm area 2 is less obvious, but its conjunction of tactile and proprioceptive information may be important in localizing the limb relative to nearby objects in the environment. Its strong connections to area 5 in the posterior parietal cortex reinforce this possibility (Padberg, Cooke, Cerkevich, Kaas, & Krubitzer, 2018).

The posterior parietal cortex (PPC), including area 5, is considered “multimodal association cortex”, neither strictly sensory nor motor, and related to multiple interoceptive and exteroceptive sensory modalities. Interestingly, area 2 neurons retain a prominent force component (Prud’homme & Kalaska, 1994) which is eliminated in area 5, perhaps to accommodate the multimodal convergence with vision in area 7 (Hamel-Pâquet, Sergio, & Kalaska, 2006). In humans, a stroke causing a lesion in the right PPC can cause a profound neglect of the left side of the body. More precise ablations in area 5 of monkeys impair reaching in darkness but not in light, while area 7 lesions have the opposite effect: reaching in the light is impaired, but not in darkness (Rushworth et al., 1997). A human patient with a lesion in area 5 “loses” her arm when it leaves her view for more than a few seconds, but it returns when it becomes visible again (Wolpert et al., 1998). This apparent role of PPC in updating limb position is closely related to its contribution to movement planning and may also involve the secondary somatosensory cortex.

The descending connections to CN from the sensorimotor cortex suggest that CN has a key role in flexibly modulating the gain of somatosensory input, although this would not rule out other lower-level mechanisms. Such gain modulation could serve to focus attention on a class of receptors or a portion of the limb under different behavioral contexts, and may also be necessary to generate complex, context-dependent reflex activity. Experiments to monitor the inputs to CN, for example in the dorsal root ganglia, under similar behavioral conditions, will be an important next step in understanding this processing.

Our own results, including sensory mappings of proprioceptive CN neurons and recording of their activity during behavior, point to a CN that receives potent connections from only a small number of afferent inputs. This finding is not altogether surprising—recent evidence shows that only a small number of synapses (4–8 for cutaneous receptors) dominate the firing of CN neurons despite far larger numbers of synapses observed anatomically on these neurons (Bengtsson, Brasselet, Johansson, Arleo, & Jörntell, 2013). This anatomical rather than physiological observation may also underlie the discrepancy with the much broader convergence estimates based on electrical stimulation that might synchronously recruit more of these afferents (Witham & Baker, 2011). The purpose of this apparently broad, yet weak convergence from the periphery is still an open question. One possible answer is that it may enable greater plasticity in sensory processing. Much like the analogous pruning process in the cerebral cortex, CN has many inputs that are lost late in development as descending corticobulbar fibers invade the dorsal column nuclei (Fisher & Clowry, 2009). Furthermore, recent studies have shown that the change in cortical representation observed after loss of peripheral input (Jain, Qi, Collins, & Kaas, 2008) has its origin in dorsal column remapping (Kambi et al., 2014). These observations suggest that both in development and in recovery from injury, CN may optimize the strength of its diverse peripheral inputs for the proprioceptive functions carried out by more central brain structures.

Summary

Figure 4 presents a high-level summary of the convergence and sensitivity properties of somatic sensation presented in this review. In the periphery, the sensitivity of muscle spindles, the main receptor considered in this review, is modulated in a complex, behavior-dependent manner by descending gamma drive. Golgi tendon organs, which signal force, and cutaneous receptors lack this descending control. Additional mechanisms within the spinal cord allow the sensitivity to all somatosensory modalities to be modulated. As a general rule, muscle afferent input is primarily enhanced during active movement, while cutaneous input is attenuated. Gain modulation in the spinal cord is critically important for spinal motor circuitry to produce controlled movements. Within the main cuneate nucleus, this modality-dependent movement sensitivity is largely maintained. At this first site for convergence between afferents in the brain, there appears to be quite low behaviorally-relevant convergence between muscle inputs during typical movements, although there is some evidence that multiple cutaneous submodalities (i.e., rapidly adapting and slowly adapting receptors) may converge on single CN neurons (Suresh et al., 2017). There is also evidence of a larger number of latent synapses during development and recruited in response to injury that may only contribute meaningfully to firing when they are activated with an unusually high intensity, such as by electrical stimulation of the peripheral nerve. Finally, neurons within area 2 of the somatosensory cortex are the first neurons in the cortical somatosensory pathway for which there is clear evidence of broad convergence across muscle and tactile modalities. Its anatomical position between the single-modality primary somatosensory areas and the even more broadly convergent receptive fields of posterior parietal cortex suggest an early role in the development of an internal body map for planning and controlling movement. How the more uniform (relative to CN) attenuation of input during movement might relate to such a functional role remains unclear.

Figure 4:

Overview of convergence and sensitivity properties along the somatosensory neuraxis A) Diagram of somatosensory areas discussed in this review. From top to bottom, in dashed circles: somatosensory cortex (adapted from a Scalable Brain Atlas of the macaque brain (Bakker, Tiesinga, & Kötter, 2015), dorsal column nuclei, spinal cord, and muscle receptors. Those areas most relevant to this review are expanded in B) and highlighted in light red. C) Summaries of the convergence and sensitivity in each of these areas. For brevity, we condensed a complex literature to the primary direction of sensitivity modulation (“attenuated” or “enhanced”), together with a similarly high-level overview of the convergence at each region. See text and references for more nuanced detail.