“It is not the strongest of the species that survives, nor the most intelligent that survives. It is the one that is most adaptable to change”. Charles Darwin.

Animals must detect and respond to changes in external and internal environments in order to survive and reproduce. The sensory and motor systems of an animal exhibit remarkable sensitivity and plasticity in their structures and functions as needed to adjust to ever-changing environmental conditions. For example, when animals encounter predators or hazardous conditions, their sensory systems detect spatiotemporal patterns of these harmful stimuli, which are then transduced, integrated, and processed in central nervous systems to control motor systems, leading to change in behavioral programs, including locomotive maneuvers (1). Moreover, this behavioral modification is further modulated by previous experience and internal status.

Humans have a multitude of senses, including vision, audition, olfaction, gustation, and somatosensation, which are traditionally described as five senses. In the 19^th century, Scottish anatomist Sir Charles Bell first characterized “muscle sense” and referred to it as the sixth sense (2-4). In the early 20^th century, Charles Scott Sherrington studied the peripheral source of sensory afferents and their control on muscle contraction, and introduced the terms “exteroception”, “interoception”, and “proprioception” (5, 6). The exteroception senses environmental stimuli from outside the body, whereas the interoception detects internal states or signals from internal organs. Moreover, proprioception senses movement of our body, such as limbs or muscles. Early studies showed that animal locomotion is mediated by a highly coordinated sensorimotor feedback system including proprioception (7, 8) and that defects in proprioception result in motor defects, such as uncoordinated movement (8-11).

Motor control via the proprioceptive sensory system is well conserved in many motile animals, from vertebrates to invertebrates. In mammals, proprioceptive systems are well described anatomically and functionally (12-14). Invertebrate proprioceptive organs have also been well described in several species such as worms, flies and cockroaches (15, 16). Despite previous studies, the molecular mechanisms underlying proprioceptive feedback in motor control are still unclear. Here, we review recent findings about molecular and neuronal mechanisms underlying proprioception and its motor control in representative model systems, including C. elegans, Drosophila, and mice.

The nematode Caenorhabditis elegans has a relatively simple nervous system, with only 302 neurons, roughly 8,000 synapses, and 95 muscles involved in locomotion (17, 18). However, C. elegans displays a broad spectrum of locomotive behaviors, such as crawling, swimming, and head steering. Thus, the C. elegans system provides an opportunity to investigate the locomotive behavior mediated by the proprioceptive feedback system at a single-cell circuit level.

Crawling behavior

C. elegans crawls on solid substances such as a plane agar gel and moves forward and backward in the sine wave shape. During forward locomotion, C. elegans also exhibits omega turns in which to make a sharp angle turn, allowing the worms to change direction. Crawling is well coordinated to generate dorsoventral sinusoidal waves with constant speed, wavelength, wave width, and bending angle, and is controlled by neural circuits consisting of multiple proprioceptive neurons and proprioceptive receptors (19-22).

The DVA interneuron has been shown to mediate mechanical sensory integration (23, 24). Xu and coworkers identified the DVA interneuron as a type of proprioceptive neuron that regulates body bending angle (Fig. 1A) (25). This group found that trp-4 mutants exhibit an increased bending angle, referred to as “exaggerated bending” (Fig. 1B); this defect is restored by expressing the TRP-4 proteins specifically in the DVA neuron (25). The trp-4 gene encodes a transient receptor potential NOMPC-like (TRPN) channel, and TRP-4 proteins are localized throughout the axon of the DVA neuron (Fig. 1B) (25, 26). Moreover, calcium imaging data indicate that the DVA neuron is physically activated by the body stretch via the TRP-4 channel. However, laser ablation of the DVA neuron in trp-4 mutants suppressed the mutant phenotype and instead decreased the bending angle, suggesting that additional factor(s) in the DVA neuron regulate body bending (25). Together, these results indicate that the DVA neuron acts as a proprioceptive neuron-type that detects body stretch via the TRP-4 proprioceptive receptor to shape waveform in crawling. In addition, Hu et al. found that the DVA neuron secretes neuropeptide NLP-12, which mediates aldicarb-induced potentiation of cholinergic transmission (27). Moreover, aldicarb-induced muscle contraction induced NLP-12 secretion via the TRP-4 stretch receptor in the DVA neuron, supporting that the DVA neuron and TRP-4 receptor play in proprioception to couple muscle contraction to cholinergic transmission.

The next candidates as proprioceptive neurons are the PVD and FLP sensory neurons (Fig. 1A). These neurons appear to sense noxious signals in a way similar to how mammalian nociceptive receptor neurons do so (28, 29). Albeg and coworkers identified new roles of the PVD and FLP neurons in modulating crawling (30). The PVD and FLP neurons have characteristic structures that directly detect body stretch (Fig. 1A) (17, 30). The dendritic branches of the PVD and FLP neurons encompass the whole body; PVD branches cover the body region from the pharynx to the tail, whereas FLP branches surround the head region (Fig. 1A) (29, 30). Moreover, their terminal branches are positioned between the body-wall muscle and hypodermis (30, 31). This group found that PVD-ablated mutants exhibit a decreased bending angle and a more extended waveform (Fig. 1B). Also, mutants lacking functional PVD and FLP neurons exhibited locomotive defects, including reduced speed, increased reversal, and pauses (30). The PVD and FLP neurons express the MEC-10 degenerin/epithelial sodium channel (DEG/ENaC), which has been identified in mechanosensation (Fig. 1B) (29, 32). The group found that mec-10 mutants exhibit a decreased bending angle (Fig. 1B) and that during locomotion, the PVD neurons are activated through the MEC-10 channel (30). Thus, these results indicate that the MEC-10 channel can function as a proprioceptor in the PVD and FLP neurons, and that proprioceptive feedback from the PVD and FLP neurons may modulate proper crawling. However, direct activation of the PVD and FLP by muscle contraction and their downstream targets needs to be further verified in order to conclude that the PVD and FLP neurons and MEC-10 are bona fide proprioceptive neurons and receptors, respectively.

Head steering

During forward movement, C. elegans exhibits head steering, moving the head left and right repeatedly. Yeon and coworkers found that the SMDD sensory/inter/motor neurons control the head steering of C. elegans (33). The SMD neurons consist of two pairs of neurons (dorsal SMDDs and ventral SMDVs), of which cell bodies are located in the head, and whose synapse-free processes extend along the body (Fig. 1A) (17). Moreover, these neurons innervate the head/neck muscles, suggesting a role in head locomotion. Previously, the SMD neurons have been shown to regulate omega turns (34). This group showed that genetic and physical ablation of the SMDD neurons causes ventral circling locomotion (Fig. 1B) and optogenetic activation of all four SMD neurons together, resulting in synchronized activation of all four cells that also results in similar locomotive defects. Moreover, a forced stretch of neck muscle activated the SMDD neurons, which in turn directly regulated dorsal head-muscle contraction. These results indicate that the SMDD neurons are the bona fide proprioceptive cells, and that these cells are both necessary and sufficient to gene-rate head-steering locomotion. This group also showed that the SMDD neurons co-express mechanosensitive TRP-1 and TRP-2 TRPC (canonical) channels, of which double mutants, but not single mutants, exhibit ventral circling locomotion similar to that of SMDD-ablated animals (Fig. 1B) (26, 33). Moreover, in trp-1 trp-2 double mutants, the Ca²⁺ activity of the SMDD neurons was synchronized with that of the SMDV neurons. Ectopic expression of TRP-1 or TRP-2 in the AWC chemosensory conferred neck bending dependent Ca²⁺ activity. These results indicate that these two TRPC channels are necessary and sufficient for proprioceptive responses, detect dorsal neck-muscle stretch and desynchronize dorsal muscles from body locomotion circuits.

Wave propagation

C. elegans exhibits distinct locomotive behaviors that are repe-titive, rhythmic, and depend on their environment. For example, they show repetitive S-shaped crawling on a solid agar plate and repetitive C-shaped swimming in a liquid (22, 35, 36). This undulatory wave appears to result from coordinating rhythmic muscle activities generated by a central pattern generator (CPG) (37, 38). These rhythmic activities from CPGs are transmitted to the whole body and are properly maintained by proprioceptive feedback (39-43). For example, the cell bodies of B-type cholinergic motor neurons are located in the ventral nerve cord, and their process extends along the ventral or dorsal region (17). Wen et al. found that these B-type motor neurons are activated by body bending, and that inactivation of the B-type motor neurons by the expression of the K⁺ channel prevents propagation of the wave from the anterior region to the posterior region (41). Thus, these results suggest that B-type motor neurons respond to body bending and provide proprioceptive feedback to the subsequent motor neurons to generate wave propagation.

The fruit fly Drosophila melanogaster is an attractive model system to study proprioceptive feedback. A single set of Drosophila molecular genetic tools can be applied to two distinct behavioral arenas, the larva and the adult fly, which show different behavioral mechanisms in responding to sensory inputs. Moreover, both larvae and adults contain well-identified proprioceptive organs and mechanoreceptors (16, 44, 45).

Leg control in adult-fly locomotion

Proprioception of the adult legs is in part mediated by the mechanosensory apparatus such as chordotonal organs (COs) (Fig. 2A). Insect legs harbor multiple classes of exteroceptive and interoceptive mechanoreceptors (45). COs house internal mechanoreceptor neurons typically located at and between joints residing in individual limbs and body segments. The fundamental unit of COs is called the scolopidium, consisting of one to three bipolar sensory neurons and two types of accessory cells (neuron-enveloping scolopale and neuron-anchoring cap cells). The femoral chordotonal organ (FeCO) is a well- known proprioceptor in insects and is widely located in the legs (Fig. 2A). In Drosophila, FeCO is structurally well conserved, but specific functional studies are still in progress. The loss-of-function genetic study by Mendes et al. suggested that the Drosophila leg FeCO is functionally involved in locomotive coordination (46) through genetic manipulations for loss of function studies. Both deficiency of the nanchung (nan) gene, which encodes a member of TRPV (vanilloid) channel expressed in sensory cilia of COs and inhibition of the leg FeCO sensory neurons led to impaired gaits and decreased speed. These results suggest that the Drosophila leg FeCO functions in proprioceptive feedback to generate precise patterns of leg movements during walking (46). Furthermore, Drosophila genetics combined with two-photon calcium imaging allowed Mamiya and coworkers to study the anatomy and function of 152 neurons in FeCO while the legs were magnetically controlled (47). Genetically separable groups of FeCO neurons with anatomically differential innervation to the ventral nerve cord (VNC) were necessary for specific functions, such as encoding tibial position, movement direction, bidirectional movement, and vibration frequency. Based on the innervation patterns to the VNC, the three subtypes were named club, claw, and hook, although their cell bodies reside in the same FeCO. Thus, Drosophila FeCO neurons are critical for precise leg coordination for gaiting; to this end flies use functionally and anatomically distinct mechanosensors within FeCO neurons.

Despite functional studies of proprioceptive organs, the molecular mechanisms of leg proprioception are still unclear. For leg proprioception, Akitake et al. found that the TRPγ ion channel functions in Drosophila leg-motor control (48). TRPγ is a Transient Receptor Potential Canonical subfamily member in Drosophila; it was previously identified as a cation channel subunit heteromerizing with other TRPC channels, such as TRP and TRPL, for visual signal transduction. Akitake et al. found that TRPγ functions as a mechanoreceptor in the leg FeCO (48). The trpγ reporter is expressed in FeCO (neurons and scolopale cells) and macrochaetes on the dorsal thorax and legs. The trpγ alleles show decreased forward walking speed, decreased average step length, and impaired leg replacement (Fig. 2B). These postural defective behavior phenotypes indicate that trpγ is essential for proprioception-mediated fine motor control. Patch-clamp recording of TRPγ expressed in HEK293 cells showed that TRPγ was directly activated by membrane stretch (48). In the meantime, Cheng et al. found that TRP channel TRPN1/NompC is required for controlling locomotion of adult flies (49), and that nompC mutants display uncoordinated leg/wing twisting movements and reduced walking speed (Fig. 2B). The nompC is expressed in many neurons, especially proprioceptors in leg joints (50) and ciliary tips of COs and along sensory neuron dendrites (49, 51). The subcellular NompC localizations appear to be associated with ANK repeats in its N terminal cytoplasmic domain and contribute to NompC protein stability (49). These results suggest that TRPN1/NompC mechanosensitive channels may functionally be involved in leg proprioceptors to generate precise locomotive behavior in adult flies.

In summary, three different studies of leg proprioception indicate that the Drosophila leg mechanosensory organ FeCO is pivotal for the coordination of fly gaiting. Such joint-moving-dependent FeCO activation can be initiated by stretch-sensitive TRPγ and TRPN1/NompC channels. Further studies will be required to unravel the precise roles of these and other mechanoreceptor molecules in mechanosensitive organs as well as FeCO.

Visuomotor gain

The generation of precise motor control is mediated by multisensory integrations, such as vision, hearing, and touching. In Drosophila, flight is controlled by the halteres, which are vestigial wings that function as gyroscopes. Interestingly, motoneurons innervating the haltere muscles were identified as a target of excitatory visual interneurons, detailing a possible mechanism of visually regulated flight (52). Recently, Bartussek et al. suggested that the integration of visual signals and proprioceptive feedback from wings and halteres generates precise wing steering muscles (WSM) activation (53). Object fixation, optomotor altitude control, and saccadic escape reflexes were examined with wing kinematics under haltere immobilization or wing-nerve treatment, revealing antagonistic signalings between wings and halteres. Decreased wing-steering ranges resulted from haltere immobilization, whereas that range was increased by suppression of wing feedback. Thus, Bartussek et al. suggested that two different proprioceptive feedbacks regulate visuomotor gain to control the muscle spiking phase to enable precise flight (53).

Larval locomotion

As in adults, sensory feedback is essential for larval crawling, but the specific roles of neurons and muscles in crawling remain to be fully understood (54, 55). Genetic manipulations on two classes of multidendritic (md) neurons, bipolar dendrites (bd), and the class I mds showed that they are essential in normal larval crawling as proprioceptors (Fig. 2A) (56). Further studies revealed morphologies and positions of the diverse multipledendritc (md) neurons (57, 58), suggesting that each of six md cell types may show functional differences in proprioceptive feedback circuits. To identify the functional difference of each md cell type, Vaadia et al. performed in vivo 3D imaging of the dendrites in freely moving Drosophila larvae to observe their deformation and neuronal calcium dynamics during crawling using high-speed volumetric SCAPE microscopy (59). Six md neurons?ventral posterior dendritic arborization neuron (vpda), dorsal dendritic arborization neuron E (ddaE), dorsal dendritic arborization neuron D (ddaD), dorsal multidendritic neuron 1 (dmd1), and dorsal and ventral bipolar dendrite md neurons (dbd and vbd)?responded at different times during peristaltic waves. Five neurons (vpda, ddaE, ddaD, dmd1, and vbd) were sequentially activated during segment contraction. In contrast, the dbd neuron was excited during segment stretch. These findings indicate that the six md neurons may function as a proprioceptor during larval locomotion and be functionally synchronized. Moreover, ddaD and ddaE neurons also showed excitation in head turning and retraction. In this work and a previous study of the synaptic connections of these neurons (60), Vaadia et al. hypothesized that in generating larval forward crawling, ddaD, vbd, and dmd1 neurons may activate inhibitory premotor neurons to mediate segment relaxation and anterior wave propagation (59). On the other hand, the vpda neuron provides input into the excitatory premotor neuron A27h to activate GABAergic dorsolateral (GDL) interneurons to inhibit the neighboring contraction of anterior segments from preventing premature wave propagation (61).

Mechanosensitive ion channels have been associated with the coordination of larval crawling. Cheng et al. found that the Drosophila mechanosensitive channel TRPN1/NompC plays a role in that function (49). The expression of nompC was observed in several neurons, such as class I da neurons (ddaD, ddaE), bd neuron (dbd), chordotonal neurons (lch1, lch5, vchA, and vchB), ventral bd neuron (vbd), and class I da neuron (vpda). Null mutations of nompC in larvae showed prolonged stride duration with normal stride size and decreased crawling speed (Fig. 2B). These phenotypes have similarly occurred when bd and class I da neurons were silenced (56). The calcium activities of both bd and class I da neurons in nompC null mutants were reduced more than were those of wild-type larvae during crawling. These results suggest that NompC is required for peristaltic muscle contraction in larval crawling by bd and class I da neurons (49, 56, 59). He et al. found that the Drosophila transmembrane channel-like (TMC) gene functions in larval crawling via class I da neurons (Fig. 2B) (62). This group also used high-speed confocal microscopy to observe deformation and calcium activity of proprioceptors, ddaE, and ddaD during crawling. The ddaE dendrites were deformed by muscle contraction, exhibiting increased calcium activities during forward locomotion. However, the ddaE dendrites showed relatively unmoved and low calcium activities during backward locomotion. On the other hand, the ddaD dendrites were deformed by muscle contraction, showing increased calcium activities during backward locomotion, but were relatively less moved with low calcium activities during forward locomotion. As the molecular mechanism of these relative proprioceptive responses, He et al. presented behavioral gene-tics data associated with the Tmc gene, which is expressed in class I dendrites and is a well-known candidate as a gene encoding a mechanoreceptor (Fig. 2B). In Tmc mutants, larvae showed enhanced head curl behavior and increased backward locomotion (Fig. 2B) (63). Moreover, ddaD and ddaE in the mutants had decreased calcium activities and dendrite curvature in both forward and backward locomotion. Thus, Drosophila larval forward and backward locomotions are mediated by different mechanosensitive neurons but are likely mediated by a single mechanosensitive channel TMC (62). One recent physiological study of bipolar dendritic (dbd) neurons also presented pharmacological and genetic lines of evidence linking the mechanosensitive Piezo channel to the stretch sensitivity of dbd neurons (64).

As in C. elegans and Drosophila, proprioception plays a crucial role in the movement regulation of mammals. For fine coordination of movement, proprioception functions through multisensory integration in concert with other sensory modalities, including vision, touch sensation, and vestibular function.

Anatomical and genetic identity of proprioceptive sensory neurons

There are two types of major proprioceptive organs in mammals: muscle spindles (MS) and Golgi tendon organs (GTO) (Fig. 3A) (8, 65). The MS is located in the middle of the muscle fibers, with its sensory afferent ending innervating the intrafusal muscle fiber. In contrast to extrafusal muscle fibers that contract upon an alpha motor-neuron impulse to produce major muscular power, intrafusal fibers are located inside the fusiform (spindle-like) capsule and are innervated by surrounding type Ia or II proprioceptive sensory afferents. When intrafusal fibers are stretched by movement, the type Ia afferent triggers an action potential corresponding to the change in muscle length, whereas the firing rate of the type II afferent encodes the length of muscle (8, 65). The GTO is located in the junction between tendon and muscle. Type Ib sensory nerve endings innervate the distal ending of the tendon, which is ensheathed in the capsule. The contraction of the muscle elicits a stretch of the tendon linked to the muscle, thereby triggering the action potential of the GTO afferents. The GTO also detects the force imposed upon the tendon, allowing the sensation of isometric exercise. The cell bodies of both MS and GTO reside in the dorsal root ganglion, which contains a cluster of cell bodies enriched with mechanosensitive, chemosensitive, and temperaturesensitive peripheral sensory neurons and bilaterally neighbors the spinal cord (8, 65).

Mammalian joints also contain sensory organs of low-threshold mechanosensitivity, such as Ruffini endings and Pacinian corpuscles (65). However, in contrast to chordotonal neurons in insects, the joint sensation does not seem to play a critical role except in detecting a movement threshold, because joint replacement surgery can spare the proprioceptive control of fine movement (8).

The soma of both MS and GTO resides in the dorsal root ganglion (DRG) with other peripheral sensory neurons, such as mechanosensitive touch-sensing neurons and thermosensory neurons (65, 66). In the DRG, peripheral sensory neurons are not topographically segregated according to their function but are distributed in a salt-and-pepper pattern. It is the molecular composition and projection pattern that distinguish proprioceptors from other peripheral sensory neurons (67-69). These proprioceptive DRG neurons are derived from neural crest progenitors and characterized by the expression of parvalbumin (PV), TrkC, and Runx3 (67, 68). However, either PV or Runx3 expression does not exclusively coincide with proprioceptors, even among DRG sensory neurons, because their expression is also found in the cutaneous mechanosensitive receptors (67, 69).

Recent progress in single-cell RNA sequencing offers unprecedently detailed genetic insight into the molecular signatures of the proprioceptors. Usoskin et al. investigated the molecular details of 500 DRG sensory neurons (70). Sharma et al. cataloged the developmental landscape of DRG neurons and identified the developmental trajectory of proprioceptive neurons in relation to other DRG sensory neurons (71). Based on intersectional labeling of proprioceptive neurons, Oliver et al. could almost exclusively sort proprioceptive neurons and sequence the proprioceptors with enough depth to cover the subtypes (72). They identified five clusters in adult proprioceptive neurons, which correspond to types Ia, II MS, and GTO.

In line with these molecular signatures, Cre drive lines that specifically expressed Cre DNA recombinase in either PV+ or Runx3+ cells provided efficient genetic accessibility to proprioceptors (72-74). However, these marker genes are expressed not only in proprioceptors but also in certain types of cutaneous mechanosensory neurons, raising cell-type validation issues for Cre driver-based studies. To genetically label proprioceptive neurons exclusively and efficiently, advanced genetic techniques are required in addition to proprioceptive-specific genetic-profile information. For example, an intersectional double genetic switch using both Cre and Flp driven by PV and Runx3, respectively, is reportedly efficient and exclusive in specifically labeling proprioceptors (72). Since the genetic toolbox controlled by Flp is rather limited compared to Cre-based toolbox, further development of intersectional switches would facilitate the specific genetic manipulation of proprioceptors. Because the proprioceptive identity is postnatally established (71), inducible systems such as CreERT2 would be required to finely delineate the cell-type specificity of proprioceptive neurons (75).

Physiological properties of proprioceptors and their molecular principles

Proprioceptors respond to mechanical deformation of afferent endings by eliciting action potentials with notably high fidelity and low adaptation. This property makes proprioceptors exceptionally well adapted to ceaselessly monitor the position and movement of our body, where the cognate sensory stimuli are constantly present within a relatively limited range, in contrast to the evanescent sensory stimuli that stimulate our “five senses”. Accordingly, the physiological response of the proprioceptive system provides classic evidence supporting a fundamental concept of neuroscience, that the sensory stimuli generate the action potential of a fixed intensity in the sensory neuron and that the frequency of the generated action potential correlates with the strength of given stimuli (76).

Recent experimental approaches have used the fact that the DRG proprioceptors can also be excited by mechanical stimulation of the proprioceptor soma by micrometer-level indentation by a blunt-end glass needle (72-74, 77). Although the subcellular distribution of molecular receptors that transduce mechanical stimuli into a change in membrane potential is not clearly understood, the membrane-potential changes of proprioceptors in response to mechanical force imposed upon the soma of these neurons suggest that the distribution of the mechanosensitive proprioceptive channels is not limited to the sensory afferent endings. As we will explore later, although proprioception provides important information in feedback control of motor regulation, defective motor control is not necessarily attributable to the deficit of proprioception. Therefore, electrophysiological analyses of proprioceptors in response to mechanical stimuli are vital in identifying proprioceptive ion channels. A recent key discovery of mammalian proprioceptors is firmly rooted in the defective mechanosensitive current response in the DRG neurons. For example, electrophysiological properties of DRG PV neurons that genetically lack Piezo2 did not exhibit a mechanically activated rapid inactivating current, which represents the main response type of DRG PV neurons (73). A physiological study of mechanosensory neurons located in the mesenphalic trigeminal nucleus (MTN), responsible for the proprioception of the head, found them to be dependent on Piezo2 (74). In addition to a rapidly adapting current, DRG proprioceptive neurons also exhibit an intermediate and slow adapting current dependent on Tentonin3 (77). Notably, the intermediate adapting current is not impaired in either Piezo2^?/? or Tentonin3^?/? DRG, indicating that other mechanosensitive channels may function as a novel yet unappreciated molecular receptor of proprioception. Because current knowledge about the mechanosensitive proprioceptive channels cannot explain the in vivo physiology of proprioceptive neurons, such as the different mechanisms differentiating types Ia, Ib, and II fibers, advances in genetic handles to proprioceptive neurons, combined with electrophysiological analysis, will provide detailed insight into the differential transduction mechanisms of proprioceptor subtypes.

Implications of proprioception for mammalian behavior

The most well-known and most widely exercised example of a motor program that depends on proprioception is the spinal monosynaptic stretch reflex, also known as a knee-jerk reflex. Brief hitting of the patellar ligament stretches the quadricep muscle and the muscle spindle therein. The stretch of the muscle spindle, in turn, induces the firing of the proprioceptive neurons, which then communicate with downstream spinal motor neurons to extend the leg. Normal proprioception is a crucial component in successful knee jerk reflexes, whereas the deficit in the reflex indicates not only abnormal proprioception but also damage in the reflex arc, either interneuron or motor neuron or muscle function. Specific examination of proprioception requires a more explicit experimental design.

More than 30 different tests have been suggested to exam the proprioceptive functions of humans (78). There is no single gold-standard test to evaluate all the proprioceptive functions of the subject. Instead, each test evaluates a specific proprioceptive function of each location: perception of the static position of a body part or of a body movement. The degradation of proprioception results in the loss of acuity in movement control, which is worsened by the deprivation of complementary sensory modality, and poor novel motor learning. In the human genetic studies, genome sequence analysis identified PIEZO2 as a molecular cause that leads to proprioception deficits (79, 80). The patients carrying mutant PIEZO2 suffered from lack of proprioception, impaired motor coordination, electrophysiological phenotypes, and various degrees of joint malformation, without any compromised cognitive functions. Retarded initiation of walking is reported in infants (79, 80).

Mouse models have been a preferred mammalian model for studying proprioceptive functions in mammals because of their genetic accessibility. Early studies investigating the development of proprioceptive neurons from their progenitors produced several mutants with developmentally abnormal proprioceptors that later lacked proper proprioceptive functions as adults. Genetic models with problems in synaptic connection between proprioceptive neurons and motor neurons or in the functional development of the muscle spindle further validate that proprioception crucially coordinates posture as well as walking and swimming in mice (Fig. 3B) (81, 82).

Another line of evidence about how proprioception contributes to motor coordination in mice originates from mutants that lack the mechanosensitive ion channels responsible for proprioception. Piezo2 is abundantly expressed in the DRG PV neurons, a predominant marker of proprioceptive sensory neurons. Patapoutian and colleagues reported that conditional knockout of Piezo2 in the DRG PV neurons ablated the proprioceptive function of the mouse (73). Atypical limb coordination in tail-suspended posture was observed in mutants, along with abnormal and less-fluent walking. These results were recapitulated in the HoxB8-Cre-dependent conditional knockout of Piezo2. PV neuron-specific knockout of Piezo2 also elicited behavioral deficits in several balance and movement tests, including gait analysis, balance-beam walking, a two-limb wire-hanging test, and a rotarod test (74). Similar deficits were also reported in Tentonin3 knockouts (77).

Thus, locomotion in model animals requires further studies for us to gain a more wholistic understanding of how proprioceptive computation accurately accommodates mechanosensory inputs from components of locomotion, as we start to glimpse a mechanistic insight into the molecular and neuronal substrates for the sensory levels of proprioception (Table 1). Con-sidering that human proprioception contributes to both con-scious and unconscious perception of limb and trunk position and movement, the current behavioral assays in model animals are limited, in that they do not require conscious processing of proprioceptive information. Whereas most behavioral assays focus on motor coordination, the engagement of proprioception in motor learning and rehabilitation may provide an additional layer of insight (83). Recent progress in machine-vision technologies, such as DeepLabCut and MoSeq, will aid a better understanding of proprioception in motor control by facilitating machine-vision-based kinesthetic analysis (84, 85). Furthermore, there is currently no study investigating the causal relationships of proprioceptive neurons in motor control at millisecond precision with reversibility, warranting an optogenetic study to understand the dynamic contribution of proprioception to motor control.

This study was supported by the National Research Foundation of Korea (NRF-2020R1A4A1019436).

The authors have no conflicting interests.