Abstract
Despite the increasing incidence and prevalence of amputation across the globe, individuals with acquired limb loss continue to struggle with functional recovery and chronic pain. A more complete understanding of the motor and sensory remodeling of the peripheral and central nervous system that occurs postamputation may help advance clinical interventions to improve the quality of life for individuals with acquired limb loss. The purpose of this article is to first provide background clinical context on individuals with acquired limb loss and then to provide a comprehensive review of the known motor and sensory neural adaptations from both animal models and human clinical trials. Finally, the article bridges the gap between basic science researchers and clinicians that treat individuals with limb loss by explaining how current clinical treatments may restore function and modulate phantom limb pain using the underlying neural adaptations described above. This review should encourage the further development of novel treatments with known neurological targets to improve the recovery of individuals postamputation.
Significance Statement In the United States, 1.6 million people live with limb loss; this number is expected to more than double by 2050. Improved surgical procedures enhance recovery, and new prosthetics and neural interfaces can replace missing limbs with those that communicate bidirectionally with the brain. These advances have been fairly successful, but still most patients experience persistent problems like phantom limb pain, and others discontinue prostheses instead of learning to use them daily. These problematic patient outcomes may be due in part to the lack of consensus among basic and clinical researchers regarding the plasticity mechanisms that occur in the brain after amputation injuries. Here we review results from clinical and animal model studies to bridge this clinical–basic science gap.
Clinical Phenotypes of Individuals with Limb Loss
In 2005, 1.6 million individuals were living with limb loss in the United States, with a projected increase to 3.6 million by 2050 (Ziegler-Graham et al., 2008). Global estimates in 2017 suggested that there were 57.7 million people living with acquired limb loss, with unilateral amputation occurring more frequently than bilateral amputation (McDonald et al., 2020). The etiology of amputation varies based on the affected limb (upper vs lower extremity), age, and geopolitical circumstances (e.g., conflict, endemic diseases, access to healthcare, etc.). In developed nations, the majority of amputations involve the lower extremity (65% as of data from 2005), and the most common reasons for amputation include vascular disease and diabetes mellitus, followed by trauma and then much less frequently cancer (Ziegler-Graham et al., 2008).
Recovery after limb loss requires an individual to learn new motor skills by increasing use of the intact limb(s), integrating the use of the now residual limb, and, when available, learning to use a prosthesis (Wheaton, 2017). In addition, individuals with acquired limb loss often suffer from significant psychological sequela, including depression, anxiety, and post-traumatic stress disorder (PTSD), especially if the amputation was associated with a traumatic experience (Doukas et al., 2013; Sahu et al., 2017). Various pain conditions are also associated with amputation, including postoperative pain, residual limb pain (RLP), and phantom limb pain (PLP; Schone et al., 2022). Nonpainful phantom sensations may also arise from the phantom limb or from referred sensory stimuli (Ramachandran and Hirstein, 1998). These phenomena create unique challenges for individuals to learn new motor tasks and process real and imagined external stimuli. For people living with limb loss, successful recovery requires both learning new motor skills and the reduction of chronic pain symptoms. A recent review that focused on neurorehabilitation emphasized the importance of using individualized treatment plans to enhance recovery, presumably based on the variability of patient adaptation strategies and the diverse experiences that patients encounter after amputation (Rydland et al., 2022).
Motor adaptations and prosthetic use
After amputation, individuals must learn new motor strategies. Prosthetic devices are designed to replace the normal biomechanics and functions of a lost upper or lower limb; however, even with the most advanced modern technologies, significant limitations still exist. In addition to their need for expert fitting by trained prosthetists, lower limb prosthetic users require extensive gait training by skilled therapists to achieve independent and safe use. However, even the most proficient prosthetic users will still have significantly increased energy expenditure demands during walking or functional activities when compared to individuals without amputation (Schmalz et al., 2002; Cordella et al., 2016). Prostheses for individuals with upper limb loss involve a customized socket that is suspended to the residual limb; replaced joints (e.g., shoulder, elbow, wrist, thumb, fingers, etc.); and a control system that may be “body-powered” (i.e., controlled by a cable and harness system), externally powered through actuated joints/digits, or a hybrid of mixed body-powered and externally powered components (Cordella et al., 2016). Unfortunately, successful control of an upper limb prosthesis is often not intuitive. For example, users of body-powered prostheses must learn to move their shoulder joint in a certain direction to control their prosthetic elbow or hand, while externally powered prosthetic users must learn how to contract remaining residual limb muscles to activate myoelectric sensors placed on the surface of the skin to control actuated robotic hand and finger movements. Newer techniques, such as “pattern recognition” and “sensory integration,” are novel prosthetic designs that aim to provide more intuitive control (see section 3.3.2).
Although prosthetic use may afford greater independence in mobility or activities of daily living, ∼20% of individuals with acquired upper limb amputation do not wear a prosthesis at all (Biddiss and Chau, 2007), and ∼30–50% of upper limb prosthetic users do not wear their devices regularly (Pons et al., 2005). Prosthetic abandonment is much less common for individuals with lower limb amputation; however, older individuals and those with dysvascular-related amputation are more likely to not receive or to stop using a prosthetic leg (Balk et al., 2018). In the absence of a prosthetic device, individuals with limb loss may use intact parts of the body to complete tasks. For example, amputation of the dominant hand increases the use and accuracy of the nondominant hand on complex drawing tasks (Philip and Frey, 2014). Many individuals with transradial amputation use their residual arm as a post to help complete bimanual tasks, taking advantage of the sensory feedback provided by their residuum. Additionally, patients with lower extremity amputations may rely on manual wheelchair propulsion using their hands or their intact leg for mobility.
The age at which amputation occurs also influences motor recovery strategies, with younger individuals or those with congenital unilateral upper limb loss being more likely to employ their residual arm in functional activities versus older individuals with acquired amputation being more likely to rely on their contralateral intact limb (Makin et al., 2013a). This use-dependent strategy influences brain activity, such that those who use their intact limb have more cortical space devoted to the intact limb, while those who use their residual limb have larger cortical areas of representation devoted to their residual limb (Makin et al., 2013a). In addition to the required motor learning and associated cortical reorganization, use of the residual limb with or without a prosthesis requires proper healing from the surgical intervention (Choo et al., 2022), which also has implications for pain and residual limb functional use. Research has shown that when healing is optimal, the prevalence of RLP decreases over time (Hanley et al., 2007; Ahmed et al., 2017), which may be one reason why younger individuals or those with congenital limb loss are more likely to use their residual limb; they have had a longer interval between study participation and amputation surgery. Adults with acquired amputation, however, may hesitate to use their residual limb, especially if it is painful; thus they learn motor strategies that use their contralateral intact limb.
Phantom and residual limb pain
The majority of individuals with acquired limb amputation report pain, which often affects multiple body regions and negatively effects their quality of life and function. The PLP is defined as pain in the missing limb, while RLP is defined as pain in the residuum, or the part of the limb that remains after amputation. There is variability in the existing literature regarding the prevalence of PLP and RLP; however, reports have been as high as 80% and 59–76%, respectively (Ephraim et al., 2005; Erlenwein et al., 2021; List et al., 2021).
Generally, PLP decreases in frequency, duration, and severity over the first 6 months after amputation, with only approximately 30% of individuals reporting chronic, episodic, and severe pain (Ehde et al., 2000; Ephraim et al., 2005; Erlenwein et al., 2021). Risk factors that have been implicated in the development of chronic PLP include the presence of pre-amputation pain, traumatic or oncologic etiology, upper limb loss, proximal versus distal level of amputation, and psychological comorbidities (Ephraim et al., 2005; Balakhanlou et al., 2021). The PLP is particularly challenging to study, and there are multiple theories regarding its suspected etiology. It is likely that there is a multifactorial process driving the development of PLP that may involve (1) regenerative processes and increased expression of nociceptive fibers at the peripheral nerve level; (2) sensitization and interneuronal and receptor changes at the spinal cord level; (3) neuronal plasticity, reorganization of the somatosensory cortex, unmasking of occult synapses, and receptor changes at the central level; and (4) somatic, psychological, and social factors (Alviar et al., 2016; Erlenwein et al., 2021; Schone et al., 2022).
While RLP may occur in the presence of PLP, it is generally considered a distinct syndrome resulting from a structural etiology within the residual limb (Powell et al., 2022). The RLP is more common in individuals with traumatic amputation(s), with the majority of pain being mild and episodic (Ehde et al., 2000; Ephraim et al., 2005). The causes for RLP may include nociceptive sources in the bone, muscle, nerve, soft tissue, or vasculature (Powell et al., 2022). Examples of RLP include neuromas, adventitious bursae, heterotopic ossification, seromas, soft tissue infections, improper prosthetic fit, etc. Interestingly, RLP and PLP have been found to be closely associated. More frequent RLP has been linked to a higher risk of developing PLP, and more severe RLP has been associated with more severe PLP (Erlenwein et al., 2021). Although classically thought of as distinct pain generators, the association seen between RLP and PLP may elucidate better understanding of the peripheral origins of PLP.
Nonpainful phantom and referred sensations
Phantom limb sensation is defined as nonpainful sensations perceived to be emanating from the missing limb and has been found to affect up to 76% of individuals with acquired limb loss (Ehde et al., 2000; Kooijman et al., 2000). Although nonpainful, these sensations have been identified as being associated with an increased relative risk of PLP for those with upper limb amputation (Kooijman et al., 2000). Additionally, these sensations can be frustrating or uncomfortable, commonly described as “tingling,” “itching,” “cold,” or “falling asleep” (Ehde et al., 2000; Kooijman et al., 2000). It can be difficult at times for individuals to distinguish PLP from phantom limb sensations. One commonly reported sensation is telescoping, where the phantom hand or foot is perceived to be directly attached to the end of the residual limb, rather than in its correct anatomical position (Ramachandran and Hirstein, 1998). It has been theorized that this phenomenon is potentially related to cortical magnification of the missing hand or foot in the somatosensory cortex, and as a result, its sensation outlasts the remainder of the phantom limb that has faded over time (Ramachandran and Hirstein, 1998).
Referred sensations to the phantom limb have also been noted in case reports. Somatosensory stimulation to the face and areas of the upper body has elicited reported sensation in the phantom limb of individuals with upper limb amputation (Ramachandran and Hirstein, 1998; Collins et al., 2017). Similarly, genito-pelvic stimulation, which occurs during bowel movements, has been documented to elicit phantom limb sensation in those with lower limb amputation (Sivan et al., 2010). In the authors’ (T.S. and P.P.) clinical experience, patients have similarly reported PLP and phantom limb sensations during menstrual cycles and intercourse. These referred sensations may arise from somatosensory cortical reorganization induced by amputation which contributes to PLP and sensation, as the face is near to the hand and the genitals are near to the foot on the somatosensory homunculus.
Neural Adaptations after Amputation/Denervation
After unilateral amputation, there are widespread alterations in activity in multiple brain regions. Historically, research has focused on map plasticity, where the brain repurposes now deprived sensorimotor regions to respond to other, neighboring body regions. This reorganization has been a topic of study and debate for decades (Ramachandran and Hirstein, 1998; Chen et al., 2002; Makin and Bensmaia, 2017; Makin and Flor, 2020). This review briefly discusses map plasticity before diving into circuitry, neurotransmitters, and beyond.
For the purpose of this review, we divide the brain into regions and explore commonalities and differences between what is observed in humans and in animal models (Fig. 1 and Table 1). As shown in Figure 1, unilateral loss of limb, paw, and whisker innervation produces a bilateral recruitment of blood oxygenation level-dependent (BOLD) functional magnetic resonance imaging (fMRI) signal in sensorimotor cortices during stimulation of the intact limb/paw/whiskerpad. Regions described here include the following: (1) deprived sensorimotor cortex (contralateral to the amputation), (2) intact sensorimotor cortex (contralateral to the intact hand), (3) the connections between them via the corpus callosum, (4) other cortical regions, largely related to visual-sensory integration, and (5) subcortical regions including the brainstem and thalamus.
Diagram depicting the similar BOLD fMRI responses in people after limb loss (A), rodents after paw amputation (B), and rodents after whisker denervation (C). All three instances have increased responses to intact limb/paw/whisker stimulation in bilateral sensorimotor cortices. fMRI images reused with permission from (A) Valyear et al. (2020), (B) Pelled et al. (2009), and (C) Petrus et al. (2019). BOLD fMRI, blood oxygenation level-dependent functional magnetic resonance imaging.
Commonalities and differences of brain regions between human and animal models
Sparling 2024 Figure Citations
Rodent models allow for precise descriptions of the molecular and circuitry adaptations that occur after paw amputation (Fig. 1B); however, the small and poorly delineated forepaw/hindpaw cortical regions in rodents make circuitry studies particularly challenging. In addition, rodents rely more heavily on their whiskers for sensory input and adapt locomotion strategies relatively easily with their remaining three limbs. For this reason, many circuitry studies use a whisker denervation model (Fig. 1C), where the whisker cortical region is large and easily identified (Staiger and Petersen, 2021), and the removal of this sensory input has dramatic consequences for the animal's ability to navigate its environment.
Deprived/affected somatosensory circuitry (contralateral to amputation)
Remapping of primary somatosensory and motor cortices
Reorganization of the deprived S1/M1 cortices (contralateral to amputation) has often been studied in the context of map plasticity and has been studied in humans, primates, and rodents (Fig. 1). For a more comprehensive review on remapping in humans, please see Makin and Flor (2020).
Hand amputation in primates increased face-responsive neurons in the deprived hand region (Fig. 2A; Merzenich et al., 1983; Garraghty and Kaas, 1991; Florence et al., 1998, 2000). It is hypothesized that the phantom limb sensation that occurs during facial stimulation is due to these new face-responsive neurons (Ramachandran and Hirstein, 1998; Collins et al., 2017). However, these phantom sensations do not appear to coincide with increased fMRI responses to facial stimulation (Valyear et al., 2020). More recent studies have questioned the remapping hypothesis, as the preserved representation of the amputated hand has been detected in the cortex (Kikkert et al., 2016; Makin and Bensmaia, 2017). Remapping of neuronal responses may depend on motor recovery strategies that humans or animals use after unilateral injury/amputation. The invasion of the intact hand responsive region into deprived S1/M1 is also associated with reduced prosthesis use (Karl et al., 2004), indicating that this region may have been colonized for other uses, and is no longer available to support the use of a new limb. When humans develop new strategies to use their residual limb or increase the use of their intact limb, cortical maps adapt to support this behavior (Makin et al., 2013a).
Diagram depicting the cortical adaptations after unilateral limb loss/paw amputation/whisker denervation. Deprived (green—A) and intact (red—D) sensorimotor cortices in upper and lower limb amputees have many adaptations noted after injury. Deprived (green—C) and intact (red—E) somatosensory forepaw cortex adaptations after unilateral paw amputation. Deprived (green—C) and intact (red—F) somatosensory whisker cortical adaptations after unilateral whisker denervation. Interhemispheric connections are related to observations between the two hemispheres and mediated by the corpus callosum. BOLD fMRI, blood oxygenation level-dependent functional magnetic resonance imaging; FC, functional connectivity; rs-fMRI, resting-state functional magnetic resonance imaging; S2, secondary somatosensory area; M2, secondary motor area.
Rodent injury models provide specific information on remapping after unilateral amputation (Fig. 2B,C). After rat forelimb amputation, the majority of neurons in deprived sensorimotor regions have persistent residual limb-responsive neurons, indicating that remapping may be limited (Bowlus et al., 2003). After a 2–3-week delay, shoulder-responsive and lower jaw-responsive neurons are detected in the deprived forelimb S1 (Pearson et al., 1999, 2003; Pellicer-Morata et al., 2023). Interestingly, even after a few months, there are portions of deprived cortex that remain silent, and new synapses were not detected from neighboring cortical areas (Pearson et al., 1999, 2003). However, it is likely that limb amputation studies in laboratory rodents are missing some adaptations at the synaptic or map levels because rodents typically rely more heavily on their whiskers for sensory perception and do not usually receive physical therapy training or enriched environments. These factors could reduce many plasticity programs that are observed in humans.
The debate regarding map plasticity continues; however it may be the case that dual, overlapping maps exist in the cortex: one for the phantom and one for the unmasked neighbor synapses that are now driving responses in hyperexcitable neurons. Humans can be asked to “move” their phantom; thus, the persistent representation of the now missing hand map is detectible with fMRI (Makin et al., 2013b). Regardless of how/if map plasticity contributes to PLP, one thing that is consistent between patients and animal models of amputation are the response recruitment of the deprived S1/M1 to intact limb stimulation on the ipsilateral side (Fig. 1; Lotze et al., 2001; Pelled et al., 2009; Simões et al., 2012; Yu et al., 2012).
Neurotransmitters in deprived cortex
Map plasticity is likely supported by altering the excitation/inhibition balance in the neural circuit. Increased excitation and reduced inhibition have been observed in deprived sensorimotor cortices in humans, nonhuman primates, and rodents in response to limb amputation (Fig. 2A,B). Transcranial magnetic stimulation (TMS) can be used to estimate cortical inhibition by comparing paired and single magnetic pulses over the motor cortex and recording muscle evoked potentials in the periphery. The facilitation or depression of muscle evoked potential amplitudes that are evoked with various pairings is estimated to reflect the amount of intracortical inhibition (LeDoux et al., 1985; Reis et al., 2008; Hordacre et al., 2015; Teixeira et al., 2021). Dysregulation of excitation and inhibition impacts motor learning; for example, decreased intracortical inhibition in bilateral motor cortices correlates with poor gait function in lower limb amputees (Hordacre et al., 2015).
Rodent models allow identification of specific neurons and synapses that mediate this altered excitation/inhibition balance. After unilateral lower limb loss, deprived sensorimotor cortices have reduced levels of the inhibitory neurotransmitter (GABA; Chen et al., 1998), but not reduced GABA receptors (Capaday et al., 2000). This decreased inhibition is accompanied by increased excitability in the deprived/affected sensorimotor (Schwenkreis et al., 2000). In primates, excitatory AMPA receptors were shown to increase in the deprived cortex a week after forelimb denervation (Mowery and Garraghty, 2009). Blocking excitatory NMDA receptor activity halted much of the cortical reorganization observed, indicating that NMDA receptor activity is required for remapping (Garraghty and Muja, 1996). After squirrel monkey forelimb denervation, precursors for GABA production were decreased in denervated S1 within a few hours and remained low several months later (Garraghty and Kaas, 1991; Wellman et al., 2002). This reduction in GABA is hypothesized to allow nearby synapses to become unmasked, thus encouraging circuit rewiring (Orczyk and Garraghty, 2015). These results indicate that there may be an important temporal pattern in the excitatory/inhibitory balance that allows the brain to adapt to injury. The regulation of inhibition is hypothesized to be a major regulator of plasticity (Hensch and Quinlan, 2018). After digit amputation, deprived somatosensory regions had a transient increase in AMPA receptor levels between days 1 and 9 postinjury, followed by an elevation of GABA 2–7 weeks later (He et al., 2004). The hypothesized mechanism is that AMPA/NMDA increases and opens an excitatory window for remodeling, followed by an elevation in GABA to lock down the new circuit (Garraghty and Muja, 1996; He et al., 2004; Hensch and Quinlan, 2018). Forelimb amputation increases the activity of deep neurons in deprived S1 forepaw within hours (Han et al., 2013) and persists for weeks (Pelled et al., 2009) (Fig. 2B).
Similar results have been observed after whisker denervation, where deprived whisker cortex also responds to intact (ipsilateral) whisker stimulation (Figs. 1C, 2C; Yu et al., 2012; Petrus et al., 2019). As in forepaw denervation models (Pelled et al., 2007), deep layer neurons have also been shown to drive the recruitment of neuronal responses to ipsilateral stimulation (Fig. 2C; Petrus et al., 2020). These results agree with observations in human literature (Chen et al., 1998; Dettmers et al., 2001) and provide more detailed information about how this remapping may occur: by modifying the basal activity of deep neurons in both deprived S1 forepaw and S1 whisker cortex (Pelled et al., 2009; Han et al., 2013; Jouroukhin et al., 2014; Petrus et al., 2019, 2020).
Cortical excitability is dysregulated after unilateral limb loss and may play a role in PLP incidence. In a recent multicenter trial, researchers correlated decreased inhibition in deprived/affected sensorimotor cortex with increased PLP and phantom sensations in lower limb amputees (Teixeira et al., 2021). Attempts to restabilize this hyperexcitability with pharmacological or noninvasive techniques are discussed in section 3.1. For a more comprehensive review on PLP, please see Schone et al. (2022).
Intact/unaffected somatosensory circuitry (ipsilateral to amputation)
Excitation and inhibition studies conducted on individuals with acquired amputation have focused on the deprived, reorganized sensorimotor cortex and have often used the contralateral intact cortex as a control, “unaffected” cortex (Cohen et al., 1991; Fuhr et al., 1992; Kew et al., 1994). Thankfully newer studies have used separate cohorts of individuals without amputation as true controls for cortical excitability studies. After limb loss, individuals have increased intact sensorimotor cortex activation during sensation or movement of the intact limb, meaning both hemispheres have increased activation (Lotze et al., 2001; Simões et al., 2012; Valyear, 2020; Figs. 1A, 2A,D).
Circuitry studies are best performed in animal models, and as most studies focus on map plasticity occurring in the deprived cortex, there are few papers describing circuitry adaptations in the intact cortex after paw amputation. Two weeks after unilateral rat paw amputation/denervation, there are higher levels of activity in deep cortical neurons during stimulation of the intact paw (Han et al., 2013; Jouroukhin et al., 2014). Two weeks after unilateral whisker denervation, the thalamocortical synapse reactivates plasticity mechanisms usually reserved for developing systems. These mechanisms include long-term potentiation (Yu et al., 2012) and silent synapse formation (Chung et al., 2017). Interestingly these changes do not begin to unfold before 9 d after injury and remain elevated up to 3 weeks afterward (Fig. 2F).
Understanding the circuitry and altered synaptic components (i.e., more AMPA and NMDA receptors) gives important information as to how the intact cortex responds to amputation. While synaptic studies in humans are challenging, animal models likely have shared adaptation mechanisms. The reactivation of developmental plasticity mechanisms in the intact cortex is especially appealing when developing strategies to enhance learning after injury.
Corpus callosum's connection between intact and deprived cortices
The corpus callosum (CC) integrates bilateral sensory input and coordinates bilateral motor output (Bloom and Hynd, 2005; Roland et al., 2017). Importantly, the CC provides interhemispheric inhibition (Perez and Cohen, 2009; Palmer et al., 2012), which silences the “unused” cortex during a unilateral task—for example, kicking a soccer ball with only one foot. After unilateral limb loss, the interhemispheric balance is disrupted. Limb loss can lead to alterations in connectivity between the two hemispheres; for example, fMRI detects less resting-state functional connectivity between bilateral sensorimotor regions in individuals with lower limb amputation (Bramati et al., 2019) and limb-amputated rats (Pawela et al., 2010). Myelination of callosal axons directly influences bilateral connectivity, and reduced fractional anisotropy (a myelin integrity measurement performed with MRI) has been detected below and between bilateral premotor areas in both humans (Jiang et al., 2015; Li et al., 2017) and rats (Vianna-Barbosa et al., 2021). This disconnection may make coordinated bilateral movements, like walking, more difficult to learn even with the aid of prosthetic devices (Li et al., 2017).
Stimulus to the intact limb after unilateral amputation in both humans and rats yields a bilateral sensorimotor cortex response (Fig. 1; Lotze et al., 2001; Pelled et al., 2009; Simões et al., 2012). This effect is magnified when the intact cortex is transiently silenced—indicating that the lack of interhemispheric inhibition contributes to this response recruitment in deprived somatosensory paw cortex (Li et al., 2011). After unilateral rat whisker denervation, this bilateral activity disappears after ablation of the intact whisker sensory cortex (Yu and Koretsky, 2014) and is supported by a stronger synapse from the intact to deprived whisker sensory cortex (Petrus et al., 2019; Fig. 2). Both animals and humans experience changes in callosal connections between hemispheres. However, other cortical and subcortical regions of the brain are also known to be altered after unilateral amputation.
Beyond somatosensory/motor cortices
Limb loss causes dramatic changes in neural activity beyond the bilateral sensorimotor activity described above. Regions that process visual feedback, body position, and emotional regulation have been described to be altered after limb loss. For example, after unilateral lower limb amputation, less fractional anisotropy is detected in white matter tracts connecting vision–motor integration areas (Jiang et al., 2015), and less functional connectivity was detected between multisensory regions after rat forelimb amputation (Pawela et al., 2010). The use of the intact limb recruits increased activity in posterior parietal regions, which is hypothesized to support an increased reliance on visual feedback for proper use of the previously nondominant limb (Williams et al., 2016). These widespread adaptations are not universally observed, as stimulation of the intact limb produced no significantly different activity in visual cortex compared to controls (Valyear et al., 2020). Perhaps active motor movement places different demands on the brain compared to passive stimulation detection; visual recruitment may be required to “know” where the intact limb needs to go, versus vision is not needed to detect sensation.
Beyond task-activated fMRI, others have used resting-state fMRI (rs-fMRI) to detect decreased connectivity between visual areas and the deprived S1/M1 (Zhang et al., 2018). Others have demonstrated altered activity in brain regions devoted to motor planning, cognition, and emotional experience (Bao et al., 2021a). Interestingly these effects were dynamic, as the most significant observations were detected 12 months postinjury versus 2 and 6 months. The same group used rs-fMRI to detect a correlation between low frequency fluctuations in the parietal regions of the brain with depression and phantom limb sensation (Bao et al., 2021b). The right parietal lobe's altered activity correlates with distorted body images that people experience after limb loss (Bao et al., 2021b). Beyond connectivity between specific regions of the brain, whole network reorganization has been described as well. Rs-fMRI has demonstrated that limb loss decouples the deprived sensorimotor cortex from the sensorimotor network and increases its connectivity to the default mode network (Makin et al., 2015). This decoupling may make learning new motor skills to overcome the injury challenging; however, restoring touch sensation to amputees may improve/rebalance these dysregulated cortical networks (Ding et al., 2022). New advances in noninvasive imaging allow for more precise and widespread characterization of neural adaptations after limb loss.
Subcortical reorganization
Sensation and motor control signals are modulated as they flow between the peripheral and central nervous systems. In fact, each synaptic connection represents an opportunity for the central nervous system to shape perception of incoming stimuli or refine outgoing motor signals. Thalamic neurons adapt during learning and after injury. Like cortical neurons, thalamic neurons shift responses to face/trunk regions after upper limb denervation in macaques (Jones and Pons, 1998; Florence et al., 2000). Human imaging studies demonstrate that although sensorimotor thalamus loses grey matter after injury (Fig. 3A; Draganski et al., 2006), thalamic rewiring occurs with increasing use and fine motor control of an upper limb prosthesis (Whatley et al., 2018). Dual mapping of phantom and new regions may overlap in the thalamus as well. Imagined and real motor movements in able-bodied subjects produced indiscernible thalamic activity (Anderson et al., 2011). Stimulation of the thalamus can yield phantom sensations (Davis et al., 1998), suggesting the phantom still resides in its proper thalamic place. These results indicate that the now missing limb is still represented in this area, even if new maps have appeared to respond to nearby brain regions.
Diagram depicting the subcortical brain regions where adaptations are observed in models of unilateral limb/paw/whisker amputation. A, Deprived (green) sensorimotor thalamus in upper and lower limb amputees has reduced grey matter volume. D, Cerebellum also has reduced grey matter volume after amputation. B, Deprived (green) thalamus and E, brainstem experience some adaptations after unilateral paw amputation in rodents. C, Unilateral whisker denervation drives changes in the thalamus and brainstem F, VPL, ventral posterolateral nucleus; VPM, ventral posteromedial nucleus (sensory thalamus); PrV, principal nucleus of the fifth trigeminal nerve.
Rodent models of unilateral paw amputation and whisker denervation in adults elucidate the postcritical period synaptic mechanisms underlying subcortical map plasticity. Rat forelimb amputation produces brainstem reorganization that is restricted to a small subregion, making the brainstem alone an unlikely candidate to support the extensive rewiring of the cortex (Fig. 3E; Li et al., 2013). Subsequent experiments detected rewiring along the thalamocortical pathway, indicating that this connection may be more malleable to change in this model of amputation (Fig. 3B; Li et al., 2014). After unilateral whisker denervation (but not whisker plucking), neurons in the brainstem make new connections with the thalamus (Fig. 3C,F; Takeuchi et al., 2012). These new inputs bring ectopic sensation from the other two spared branches of the trigeminal nerve and underlie mechanical hypersensitivity in nearby facial regions (Takeuchi et al., 2017). This rewiring is thought to be mediated by the hyperexcitability of the brainstem and requires both microglia (Ueta and Miyata, 2021) and tonic inhibitory tone (Nagumo et al., 2020) to enable this rewiring. The invasion of nearby orofacial regions into the deprived thalamus from the brainstem may drive hyperalgesia. This mechanism may be how spared body regions become more sensitive or even painful after unilateral denervation/amputation.
The cerebellum refines motor movement and is important for balance. After unilateral lower limb loss, the ability of the cerebellum to adapt is imperative for recovery, with or without a prosthesis. Rs-fMRI detects decreased functional connectivity between bilateral somatomotor cortex/thalamus with the cerebellum and basal ganglia (Zhang et al., 2018). The cerebellum is also subject to grey matter loss after unilateral amputation, which can be restored with regular prosthetic use (Fig. 3D; Di Vita et al., 2018).
Subcortical regions modulate ascending sensory and descending motor information—both of which are disrupted after unilateral limb loss and may provide additional targets for modulation to enhance recovery. There remains, however, a significant lack in knowledge regarding how the cerebellum and basal ganglia circuits, both important for motor control, are changed after amputation/denervation; thus more animal studies are needed to elucidate these relationships.
How Current Treatments Harness Known Brain Changes?
Basic and clinical research have identified several phenomena that are preserved across species, which present targets for modulation to enhance functional recovery after unilateral amputation/denervation and reduce the burden of PLP. These include the following: (1) dysregulation of excitation and inhibition via specific receptors, after unilateral limb loss. These changes occur in a spatially and temporally precise manner, supporting the targeted use of noninvasive neuro-activity modulation or pharmacological treatments, and (2) the loss of unilateral sensory input and motor output unbalances the bilateral somatomotor system, and reestablishing input could help the brain to restore function. While there are no “best” strategies to enhance motor recovery after amputation or to prevent/treat PLP (Aternali and Katz, 2022; Schone et al., 2022), the following sections describe how some treatments may restore the underlying neural adaptations described above.
Modulation of excitation/inhibition
As described in section 2.2, hyperexcitability in deprived cortex after unilateral amputation/denervation may be mediated by an increase in the excitatory neurotransmitter glutamate and some of its receptors: AMPA and NMDA. A decrease in GABA receptors/neurotransmitters is also often observed. These dynamics shift the excitation/inhibition balance to favor excitation, which is thought to be a maladaptive consequence of limb loss. The mechanisms underlying this hyperexcitability may be one (or more) of the following: a homeostatic adaptation to the loss of feedforward sensory drive (Orczyk and Garraghty, 2015), an increase in callosal synaptic strength from the intact hemisphere (Petrus et al., 2019), or a decrease in interhemispheric inhibition, which effectively removes the inhibition from this cortex (Li et al., 2011). Reducing excitability in specific groups of neurons or restricted brain regions remains a difficult prospect for individuals with acquired limb loss; however, targeted, noninvasive modulation of neuronal activity is possible. In addition, reducing excitability with pharmacological agents is often effective in reducing PLP. Ideally, treatments would recalibrate the excitatory/inhibitory imbalance in the brain to both reduce pain and enhance motor recovery after amputation/denervation.
Minimally invasive central neuromodulation
If the excitation/inhibition balance is disturbed after unilateral amputation, it stands to reason that real-time modulation of activity could ameliorate this dysfunction. In animal models, this has been achieved with optogenetics and/or chemogenetics. Both strategies can be tailored to increase or decrease activity in genetically identified, spatially precise neurons, in a temporally restricted manner. Unfortunately, these tools require intracranial injection and transfection of brain tissue with viral vectors and light or drug delivery. While these applications are not practical for clinical treatment, they offer insight on the underlying neurobiological processes involved in pain transmission and provide a theoretical basis for the application of various safe forms of central nervous system neuromodulation, such as transcranial stimulation.
Repetitive transcranial magnetic stimulation
TMS is delivered by an external device that delivers pulses of magnetic fields that penetrate the skull and modulate the neural activity of superficial layers of the cortex (Wasserman, 1998; Wagner et al., 2007). This has been demonstrated in instances of cortical hyper excitability. For example, tinnitus is hypothesized to result from auditory cortex hyperexcitability, and repetitive transcranial magnetic stimulation (rTMS) has been demonstrated to effectively reduce symptoms (Müller et al., 2013). The rTMS has also been used to reduce hyperexcitability in deprived sensorimotor cortex after amputation, where it has been shown to reduce PLP (Grammer et al., 2015; Garcia-Pallero et al., 2022). Unfortunately, despite daily rTMS sessions for several weeks over deprived sensorimotor cortex, the benefits are short lived, with PLP returning after minutes (Töpper et al., 2003) or weeks (Malavera et al., 2016). A case study found similar results with reducing excitability in intact S1/M1 (Di Rollo and Pallanti, 2011). The rTMS over motor cortex can prevent the progression of acute to chronic pain in healthy controls (Cavaleri et al., 2019); however, the short-term benefits of rTMS in individuals with amputation may be related to the release of β-endorphins (Ahmed et al., 2011) rather than a recalibration of excitation/inhibition. Using rTMS to prevent or remedy PLP may have potential applications; however, this treatment strategy is not widely used (Schone et al., 2022) and does not appear to be a long-term treatment solution.
With regard to motor learning after amputation, rTMS has been used to enhance motor learning in healthy individuals (Kim et al., 2004) and to improve recovery after stroke (Hoyer and Celnik, 2011). After forepaw amputation, rats who underwent rTMS exhibited reduced anxiety and increased plasticity-related genes which could be recruited to support motor recovery (Cywiak et al., 2020). Although not a motor task, rTMS did benefit humans and macaques learning to use a visual prosthesis (Najarpour Foroushani et al., 2018). The rTMS paired with motor learning could potentially replicate these results for individuals learning how to use a prosthesis; however, to our knowledge, this has not yet been attempted. Although modulating cortical excitability with rTMS to either reduce PLP or to enhance motor recovery after amputation is an appealing theory, more research is needed to better understand if and how rTMS could modulate dysfunctional cortical regions in a temporally and spatially precise manner. New methods of targeting TMS in patient-specific ways by identifying connected brain regions with rs-fMRI neural activity has yielded promising results for depression treatments (Cole et al., 2022), which could be helpful in tailoring treatments to patients after limb loss as well. One limitation of rTMS is that its penetration into the brain is limited to superficial layers of the cortex, while animal studies demonstrate that the deepest layers in both intact and deprived sensory cortices are the most affected by injury. This means that the superficial modulation provided by rTMS may not be effectively targeting the areas that need it most.
Transcranial direct current stimulation
Transcranial direct current stimulation (tDCS) delivers a weak electrical current between target and reference electrodes that pass through the brain: a positive, anodal current is thought to facilitate and a negative, cathodal current is thought to inhibit neural activity via a polarity-dependent shift in a resting membrane potential (Nitsche and Paulus, 2000; Thair et al., 2017). Anodal tDCS of deprived motor cortex has been shown to upregulate motor cortex excitability and provide short-lasting (up to 1 week) reduction of PLP without affecting RLP (Bolognini et al., 2013, 2015; Kikkert et al., 2019). Conversely, inhibitory cathodal tDCS on the posterior parietal cortex has been shown to produce immediate and short-lasting reduction in phantom limb sensation and increased phantom motor control (Bolognini et al., 2013). These findings suggest that PLP and phantom limb sensations are associated with cortical excitability shifts in differently involved brain regions. Trials to date have been limited by small sample sizes and its unknown duration of effect.
Pharmacology
A recent meta-analysis comparing pharmacological treatments for PLP reported that only ketamine, gabapentin, and morphine had sufficient evidence to support their use (Alviar et al., 2016). As previously mentioned, cortical hyperexcitability may be attributed to increased NMDA (excitation) and/or reduced GABA (inhibition) in the sensorimotor system (see section 2.1.2). The efficacy of ketamine and gabapentin may be related to their ability to regulate excitability by acting on these receptors.
Ketamine
Ketamine is an NMDA antagonist that is effective in treating depression, PTSD, and chronic pain (Niesters et al., 2014; Liriano et al., 2019; Oliver et al., 2022). The maladaptive hyperactivity in chronic pain conditions (Woolf and Salter, 2000) could be reduced by blocking NMDA receptors. Referred hyperalgesia after tail amputation in mice is reduced with NMDA antagonism (Zhou, 1998), which indicates it could be effective for humans experiencing hyperalgesia or PLP. At least one study demonstrated that the use of ketamine before planned amputation surgery is effective at reducing postamputation pain (Hayes et al., 2004); however, this is not an option for individuals that experience trauma-related limb loss. Because ketamine has psychoactive properties, an alternative NMDA antagonist, memantine, can be used to treat individuals with chronic pain. A study by Schwenkreis et al. (2003) demonstrated that memantine treatment for 3 weeks increased intracortical inhibition and decreased intracortical facilitation but did not reduce incidence of PLP. Interestingly, an additional week of memantine was effective at treating PLP (Maier et al., 2003). When memantine treatment was combined with brachial plexus blockade for individuals with upper limb amputation, PLP was further reduced; however, the authors reported that the effect persisted for <12 months (Schley et al., 2007). These results indicate that NMDA antagonism can reduce cortical excitability; however, PLP reduction outcomes are variable. Individuals learning to use a prosthesis or new motor adaptations require plasticity to support learning. NMDA receptors are an integral part of potentiating synaptic strength to support this learning. The use of NMDA antagonists over an 8 d period blocked cortical motor map plasticity in healthy controls (Schwenkreis et al., 2005), indicating that while chronic treatment with an NMDA antagonist may reduce pain, it may be counterproductive to learning new motor skills.
Gabapentin
Gabapentin is a derivative of the GABA neurotransmitter that blocks voltage-gated calcium channels and thereby reduces the release of glutamate and substance P, a nociceptive transmitter (Smith et al., 2005). Gabapentin is FDA approved as an anticonvulsant and is typically used to treat PLP (Levendoǧlu et al., 2004; Abbass, 2012; Schone et al., 2022). Presumably, its effectiveness is due to its ability to increase inhibition and reduce hyperexcitability; however, its effectiveness is not universal (Smith et al., 2005; Flor et al., 2006). This lack of effect is likely because cortical hyperexcitability is restricted to specific cortical loci or occurs in a temporally precise manner; thus, global or improperly timed pharmacological treatments might be ineffective.
Although pharmacological agents can be used for treating pain conditions, their influence on motor recovery is largely untested. However, their nonspecific region and temporal actions may make them less useful for motor learning.
Vagus nerve stimulation
Vagus nerve stimulation (VNS) uses a peripherally implanted device that targets the vagus nerve to activate the central nervous system (George et al., 2000). It is used to treat epilepsy, depression, and stroke (Hays et al., 2013; Khodaparast et al., 2013; Engineer et al., 2019). The VNS targets noradrenergic, cholinergic, and serotonergic circuitry in the brain (George and Aston-Jones, 2010) and produces increased expression of plasticity genes and proteins in the hippocampus and cortex (Follesa et al., 2007). Because the synapses releasing noradrenaline, acetylcholine, and serotonin arrive in more superficial layers of the cortex, it is hypothesized that VNS stimulation alters how the brain processes feedforward sensory stimuli. Superficial cortical layers modulate inhibitory and disinhibitory circuitry, allowing the brain to control the signal's gain (Takahashi et al., 2020). After unilateral limb loss, it is likely that sensory gain is dysregulated, so targeting VNS may not only enhance learning but may also support synaptic mechanisms that normalize incoming (painful) stimuli (Ness et al., 2000). These mechanisms of action may enhance learning, for example, by enabling amputees to learn new motor skills. The VNS paired with upper limb rehabilitation improved recovery after stroke (Kimberley et al., 2018), spinal cord injury (Ganzer et al., 2018), and traumatic brain injury (Pruitt et al., 2016).
Basic research in animal models of forelimb denervation have yielded promising results. The VNS may specifically improve motor learning by enlarging the cortical areas devoted to learning a new task (Porter et al., 2012). Recovery of motor function and sensory discrimination tasks after partial forelimb denervation in rats was enhanced by VNS when paired with motor task learning (Meyers et al., 2019). Unfortunately, when animals displayed behaviors associated with chronic pain, VNS was ineffective at enhancing motor recovery (Adcock et al., 2022). These animal studies of peripheral denervation demonstrate exciting central plasticity and motor learning recovery when paired with VNS. These results combined with the effectiveness of VNS on motor rehabilitation in humans with other central nervous system injuries indicate that this could be an area worth pursuing in amputees learning new motor skills.
Rebalance the somatomotor and visual-motor networks
Combining visual feedback, sensory input, and motorized output with new prosthetic designs enhances functional recovery, independence, and quality of life after acquired limb loss. As prosthetic technologies continue to improve and as integration of motor control and sensory feedback become more naturalistic, the brain may also have the information it needs to remap the sensorimotor cortices to not only optimize prosthetic use and function but also to improve PLP. While fully integrated motor and sensory prosthetic limbs are developed, current strategies, such as novel surgical techniques, biofeedback prosthetics or transplants, graded motor imagery, and virtual/augmented reality (VR/AR), exist to help rebalance visual, sensory, and motor signaling for individuals with acquired limb amputation.
Surgical procedures
Novel surgical techniques, such as targeted muscle reinnervation (TMR), agonist–antagonist myoneural interface (AMI) surgery, and osseointegration (OI), improve the outcomes of individuals with limb loss by augmenting motor control, enhancing sensory feedback, and lowering the incidence/severity of PLP. Strategies for managing the residual peripheral nerves after amputation have significantly evolved over the past decade. Historically, peripheral nerves were cut while on traction and allowed to retract proximally for relocation to a more proximal, protected space within the residual limb (Pet et al., 2015) or were sometimes sutured into neighboring muscle (Eftekari et al., 2022). This technique of traction neurectomy has traditionally had a high failure rate, with a high incidence of individuals developing significant postoperative neuroma pain (Pet et al., 2015). More recently developed surgical techniques provide the severed nerve with a new distal target, effectively giving the nerve “somewhere to go and something to do” (Vernadakis et al., 2003; Bogdasarian et al., 2021). These include physiologic target reassignment via targeted muscle or sensory reinnervation (TMR, TSR; Hebert et al., 2014; Souza et al., 2014; Bowen et al., 2017; Valerio, et al., 2019) or regenerative peripheral nerve interfaces (RPNI; Woo et al., 2016; Kubiak et al., 2019); nonphysiologic target reassignment via single nerve coapts (Belcher and Pandya, 2000) or neurorrhaphy (Economides et al., 2016); or end closures with nerve caps (McKee, 2020; Tork et al., 2020).
Targeted muscle reinnervation
Of the above listed techniques, TMR has the evidence supporting its prevention of neuroma and PLP for individuals after amputation (Valerio et al., 2019; Bogdasarian et al., 2021). Targeted reinnervation via TMR or TSR is a mechanism by which sensory input and motor output can be restored. In this procedure, residual sensorimotor nerves are surgically connected to nearby skin and muscle (Kuiken et al., 1995). The re-routed nerves form new connections in this residual tissue and allow the brain to bidirectionally communicate with new body regions to enhance motor or sensory control (Hijjawi et al., 2006; Kuiken et al., 2007). Surgical guidance for nerve reinnervation supports the bidirectional sensorimotor information flow between the periphery and deprived brain regions, allowing motor map plasticity to recalibrate when observed with fMRI (Chen et al., 2013; Serino et al., 2017). Although now more commonly used to prevent or treat neuroma formation for individuals with upper and lower limb loss, TMR was originally developed (and continues to be used today) as a means to enhance myoelectric signals and improve prosthetic control for individuals with proximal upper limb amputation (Kuiken et al., 2009).
Agonist–antagonist myoneural interfaces
Another novel surgical technique that aims to improve bidirectional neural connections is the Ewing amputation, which utilizes AMIs to maintain communication of musculotendinous proprioceptive information to the central nervous system (Clites et al., 2017, 2018b). Modified amputations using AMI constructs have been performed at the transtibial (the Ewing amputation), transfemoral, transradial, and transhumeral levels (Clites et al., 2018b; Srinivasan et al., 2021) and have been performed at initial elective amputation or as a surgical revision [the regenerative AMI (Srinivasan et al., 2019)]. Outcome studies and clinical experience to date have revealed decreased PLP but increased phantom limb sensation in patients with this procedure (Srinivasan et al., 2020). It is hypothesized that this is due to improved proprioceptive sensation secondary to the maintained relationships of muscle spindles and Golgi tendon organs in AMI constructs, and these promote realistic phantom motor imagery (MI) and increased perceived range of motion (Clites et al., 2018b; Srinivasan et al., 2020, 2021). The fMRI studies in individuals with AMI constructs have revealed BOLD activity in BA3a (the proprioceptive center in the brain) more similar to activity seen in able-bodied individuals as opposed to activity seen in those with standard amputations, confirming improved proprioception within this patient population (Srinivasan et al., 2020). Additionally, individuals with AMI constructs demonstrated increased functional connectivity (FC) of the frontal medial cortex to other brain regions and decreased coupling of the visual and sensorimotor networks, both of which were associated with increased reported phantom limb sensations in subjects.
Osseointegration
The OI involves the distal skeletal fixation of a prosthetic component to the residual limb and is another surgical technique that is currently being used in individuals with limb loss. It was developed in the 1990s and has recently been approved by the FDA for individuals with transfemoral amputation in the United States (Hebert et al., 2017; Hoellwarth et al., 2020). Individuals with osseointegration have reported the ability to identify sensations transmitted through their prosthesis, a phenomenon called osseoperception (Brånemark et al., 2001). Mechanical stimulation of the bone-anchored prosthesis leads to awareness of sensorimotor position and function. There is limited data, however, on the underlying physiological mechanisms of osseoperception and the associated cortical changes associated with OI. Dental studies have demonstrated that osseoperception is transmitted by mechanoreceptors in the surrounding muscle, joint, and periosteal tissues and is associated with activation of bilateral primary and secondary somatosensory cortices (Klineberg et al., 2005; Habre-Hallage et al., 2012).
Biofeedback—prosthetics and transplants
Prosthetics
As discussed in section 1.1, individuals with upper limb loss have a high rate of prosthetic abandonment due to discomfort, heavy weight, and poor function (Østlie et al., 2012). One major improvement in the field of prosthetics is the incorporation of biofeedback (Raspopovic et al., 2014), which allows peripheral sensation to restore “normal” brain activity while individuals use the prosthesis (Granata et al., 2020). Although sensory feedback neuroprostheses have been around for decades, new surgical and technological developments have advanced these devices to improve user experiences. Implantable peripheral interfaces have been used to evoke sensory control of upper extremity prosthetic devices. Epineural interfaces [cuff and flat interface nerve electrodes (FINEs; Tan et al., 2015)], penetrating interfaces [Utah array (Wendelken et al., 2017)], and intraneural interfaces [longitudinal intrafascicular electrodes (LIFEs; Horch et al., 2011), transverse intrafascicular multichannel electrodes (TIMES; Petrini et al., 2019)] have been studied in humans to date. Electrical signals generated by sensory feedback can be tailored in amplitude and frequency so that individuals can experience naturalistic touch (Valle et al., 2018), and interestingly, saturation of this signal can be detected as painful (Raspopovic et al., 2014). Additionally, skin-like materials provide high-fidelity mechanoreception and can be fitted to gloves that wrap around a prosthetic hand (Tee et al., 2015).
Restoring peripheral sensation has the potential to improve the quality of life and function for individuals living with limb loss. For example, holding hands with a loved one is more pleasant when it is a bidirectional sensory/motor activity. Sensory feedback for upper limb prosthetic devices increases embodiment (Graczyk et al., 2019) and improves tolerance and gait in lower limb amputees (Petrini et al., 2019; Preatoni et al., 2021). Furthermore, sensorized prostheses reduce both PLP and the cortical hyperexcitability that is so often seen in patients (Rossini et al., 2010). Both embodiment and improved tolerance contribute to increased prosthetic use, opening an opportunity for enhanced motor learning for motorized prosthetics. Ongoing developments in prosthetic technology include the combination of osseointegration and implanted electrodes (the eOPRA system) to afford bidirectional (sensory/motor) prosthetic control (Ortiz-Catalan et al., 2014, 2020) and low-cost, lightweight, modular bidirectional prosthetic hands made of soft materials to further mimic a true hand (Gu et al., 2021).
Motorized prosthetic devices use signals generated by the motor cortex (Flesher et al., 2021) or muscles and/or nerves in the residuum (Raspopovic et al., 2014; Fang et al., 2020; Engdahl et al., 2022) to direct their movements. For example, the bidirectional neural connections generated by AMI constructs afford individuals the ability to volitionally control myoelectric prosthetic joints (Clites et al., 2018a). Functionally, these individuals outperform controls during proprioceptive tasks. In addition, sensory feedback enhances motor learning. For example, the firm but gentle grasp required to crack open an egg requires sensory grip feedback during the motor action (Fishel and Loeb, 2012). Sensory feedback from sensitized leg/feet prosthetic components has also been shown to improve walking and reduce PLP (Petrini et al., 2019). Restoring the bidirectional sensorimotor connection between the brain and prosthesis induces cortical reorganization similar to individuals without amputation (Lotze et al., 1999). The use of a sensorized prosthesis also has the potential to overcome visual dependence while completing motor tasks (Tan et al., 2014), meaning people no longer heavily rely on visual feedback to functionally use an upper limb prosthesis.
Transplants
Although prostheses are advancing in quality, hand transplants are another viable option to provide sensory feedback (Shores et al., 2015). The first successful hand transplant was performed in 1998, and at least 148 procedures have been performed on 96 patients as of 2022 (Wells et al., 2022). Both transhumeral and transradial transplants have been associated with decreased disability scores, increased functional benefits, and high allograft survival rates (Wells et al., 2022). However, the procedure is not without medical risks, particularly the development of hyperglycemia, diabetes, cytomegalovirus infections, and renal insufficiency. It is therefore important to target appropriate patients (i.e., those individuals with bilateral upper extremity amputation or those individuals with unilateral upper extremity amputation and blindness) when considering transplantation as a viable therapy. Increased practice and skill with transplanted hands is correlated with neural activity mimicking those of two-handed subjects (Valyear et al., 2019); however, the restoration of cortical map representations is not always detected (Philip et al., 2022).
These findings indicate that in addition to potentially improving functional independence after amputation, the restoration of sensation from either a prosthesis or transplants may also reduce PLP and RLP. Gate control theory postulates that the presence of tactile sensation in a painful region reduces pain perception (Melzack and Wall, 1965; Mancini et al., 2015). The lack of touch may exacerbate the perception of pain from the residuum or phantom; thus, biofeedback from transplanted limbs or prostheses may restore beneficial touch-mediated pain modulation. Limb loss disrupts ascending sensation and motor output between the brain and the periphery. The appearance of a phantom may be reverberations of this disrupted communication, and an affected patient has little control over movement or pain modulation in the phantom. In addition, nearby touch is known to reduce perception of noxious stimuli (Melzack and Wall, 1965; Zampino et al., 2018), and this touch-mediated analgesia may be mediated by unique groups of cells in the brain (Lu et al., 2022; Osaki et al., 2022). The lack of touch after limb loss is another possible mechanism for how PLP is established and makes it difficult to treat. New advances in prosthetics allow for bidirectional somatomotor communication between the brain and periphery, which may restore balance to the brain (Gupta et al., 2023), enhance learning (Raspopovic et al., 2021), and promote embodiment (Segil et al., 2022).
Visuomotor adaptations after amputation/denervation
Successful recovery from limb loss requires learning new motor strategies and reducing pain conditions. After limb loss, connectivity between visual and motor planning regions of the brain are altered with bilateral sensorimotor cortical regions (see section 2.4). The increased reliance of patients on visual feedback to move their intact limb or prosthesis either drives these activity changes or is a result of plasticity among these connections. Regardless, the interaction of visual and motor regions provides opportunities for treatments by harnessing these plastic connections to enhance recovery. Enhancing motor learning can occur by imagining the motor movements (MI) and watching another person perform a task (action observation, AO). Furthermore, training on a task with one hand can enhance performance with the nontrained hand (intermanual transfer). Mirror therapy and VR/AR can both enhance motor learning and reduce pain by “tricking” the brain into seeing limbs that are no longer present. Many of these strategies go beyond the traditional repetition-based practice and provide opportunities for patients missing one limb with which to learn new motor skills.
Motor imagery and action observation
MI can be used to improve performance on motor tasks by imagining oneself successfully completing the task (Decety, 1996; Di Rienzo et al., 2016). AO is achieved by watching another person perform a task and is used to improve motor deficits after stroke (Ertelt et al., 2007; Ryan et al., 2021). Both MI and AO activate regions of the brain involved in performing motor tasks and often occur simultaneously (Eaves et al., 2016).
These types of training may support learning for people with limb loss by enhancing their agility with new prostheses or adapting motor skills to effectively use residual or intact limbs. Crucially, people living with limb loss can distinguish between motor execution and MI of the phantom (Raffin et al., 2012). The MI improved walking and “timed up and go” tasks in lower limb amputees (Saimpont et al., 2021). The AO improved prosthesis use (Cusack et al., 2012) and depended on the vantage point of the observer (Lawson et al., 2016). As with most treatment programs for patients with limb loss, improved recovery depends on reduced incidence of pain. Graded MI can both reduce PLP (Tung et al., 2014) and improve functionality of the residuum (Priganc and Stralka, 2011). One potential problem with MI in amputees is the distorted body image, for example, telescoping (Malouin et al., 2009). However, therapies could be tailored to account for how accurate (or not) the patient's body image is and either remove MI from the treatment plan and/or increase the use of AO.
Motor learning (further supported by MI and/or AO) transiently increases corticospinal excitability (Clark et al., 2004), but after a motor sequence is learned, the corticospinal system should shift toward efficiency or decrease in excitability (Higuchi et al., 2012; Eaves et al., 2016). This increased activity with MI or AO begs the question: is this necessarily a good thing for amputees? The MI evokes larger amplitudes and areas of cortical activity in amputees than in able-bodied individuals (Hunter et al., 2023), consistent with the increased responsiveness to touch and movement of intact limbs (Lotze et al., 2001; Valyear et al., 2020). While others use pharmacology or noninvasive stimulation techniques to reduce this increased cortical excitability, these therapies are designed to increase it. There should be a sweet spot where MI-/AO-induced increased excitability can be used to support learning, while decreased excitability during nonlearning time periods could be coordinated to bimodally modulate the brain's activity. Alternatively, constraint therapy has been shown to induce corticomotor depression, which can be recovered with AO (Bassolino et al., 2014). One strategy could involve restraining the intact limb to stabilize the cortical excitability and concurrently train the prosthetic or residual limb. Restraint therapy may also force the patient to use new motor strategies and increase motivation and attention for those tasks.
Intermanual transfer
Learning a skill with one hand “trains” bilateral motor cortices to learn the task and has been shown to improve performance in the untrained hand (Gabitov et al., 2016). This intermanual transfer skill enhancement has been shown in learning to use a prosthesis with both two-handed individuals (Romkema et al., 2013) and in people after limb loss (de Boer et al., 2016). In a finger tapping task, an amputee with good phantom control trained their phantom and improved performance on the intact hand, while another amputee with poor phantom control did not experience such improvements (Garbarini et al., 2018). Patients with higher incidence of PLP have worse phantom motor control (Kikkert et al., 2017) and more negative body image (Beisheim-Ryan et al., 2023). In addition, the presence of distorted phantoms or poor body image can impair motor control of the phantom (Lyu et al., 2016). Despite these issues, intermanual transfer remains an interesting avenue for improving use of the intact limb by training with a phantom or using the intact limb to enhance dexterity/proficiency with a prosthesis. Patients who would benefit the most from this therapy would have accurate body image and low levels of PLP—both of which may interfere with successful intermanual transfer.
Mirror therapy
Mirror therapy is a form of graded MI and was developed to provide individuals with a mechanism by which visual perception of the lost limb (via a mirror placed in the sagittal plane in front of the patient). Mirror therapy training improves affected limb recovery after stroke (Gandhi et al., 2020) and enhances motor control of phantom movements in amputees (Brodie et al., 2007). Mirror therapy has only modest improvements for motor function (Barbin et al., 2016); however, it is more commonly used to provide relief from phantom limb sensations or pain (Rajendram et al., 2022). There remains controversy regarding its effectiveness (Aternali and Katz, 2022); however, it is endorsed by 80% of clinicians (Schone et al., 2022). To be most effective, mirror therapy requires the presence of visual feedback of what is perceived as the phantom limb, rather than the intact limb (Chan et al., 2007). It has been demonstrated that individuals with lower limb amputation have abnormally large responses in sensorimotor cortices to pictures of feet, which correlates to how well they respond to mirror therapy (Chan et al., 2019). Individuals with more severe PLP often require more sessions of mirror therapy to experience beneficial outcomes (Griffin et al., 2017). Importantly, regular mirror therapy appears to return the way the brain perceives the missing limb to a baseline configuration similar to two-limbed individuals (Foell et al., 2014). The mixed effectiveness of mirror therapy likely stems from vast heterogeneity among individuals with limb loss, as well as the frequency of its use, as many individuals may give up just after trying one or two sessions. Lessons learned from mirror therapy, however, support the theory that integrating visual feedback and motor output likely helps to rebalance the brain after limb loss and opens the door for developing and incorporating new technology, such as VR and AR into rehabilitation strategies.
Unilateral limb loss has widespread effects on brain activity and connectivity that are not restricted to sensorimotor networks, and abnormal visual-motor integration likely contributes to PLP (Collins et al., 2018). The integration of visual and sensorimotor experiences is important for sensation perception; for example, visual attention improves performance on tactile discrimination tasks (Cardini et al., 2011; Longo et al., 2011). In contrast, when visual feedback is discordant with motor movements in healthy individuals, discomfort is reported as pain, tingling, and numbness (McCabe et al., 2005). Treatments such as mirror therapy and VR/AR likely hijack visual feedback to compensate for this disruption; for example, if a phantom is twisted in an uncomfortable position, the visual feedback of untwisting the intact hand (perceived as the phantom/missing hand) may provide relief. The combination of vision and sensation feedback can increase embodiment and decrease PLP (Rognini et al., 2019).
Virtual reality/augmented reality
Amputation produces a lack of motor output and visual/sensory feedback from the missing limb. The neural plasticity described in section 2 is in response to this lack of feedback and may hamper learning to use a prosthesis or contribute to PLP. One strategy that reactivates visual and motor signals is VR or AR training (Dunn et al., 2017). VR is an immersive experience where avatars or other body representations can be part of a created environment. AR superimposes virtual elements into the subject's real environment, for example, giving an amputee a virtual limb superimposed on their residuum. VR/AR can improve learning; for example, healthy controls can use myoelectric output to power virtual limbs/players in video games or in VR (Pons et al., 2005; Hauschild et al., 2007; Perry et al., 2018). People with limb loss can learn to control a prosthesis to perform bimanual motor tasks (bow tying) when trained with VR (Yoshimura et al., 2020). These studies show exciting potential toward using residual limb muscle signals to power prosthetics (Van Dijk et al., 2016).
A recent meta-analysis and systematic review revealed that mirror therapy and VR therapy are equally efficacious in treating PLP in the amputee population (Rajendram et al., 2022). Key differences between VR/AR and mirror therapy may elucidate why VR/AR could be a more appropriate alternative for treating PLP in certain cases. Mirror therapy is limited to unilateral amputees, as it requires an intact limb be reflected for functional therapy. Individuals with bilateral limb loss who are not candidates for mirror therapy can participate in VR/AR. Additionally, the muscle movements that are required to “power” the avatar/virtual limbs provide an added signal pathway for the brain to manipulate, in contrast to the mostly visual experience of mirror therapy (Ortiz-Catalan, 2018). Indeed, imagined versus volitional movement of the phantom have different neural activity patterns (Raffin et al., 2012) and the inability to “move” the phantom positively correlates with PLP levels (Raffin et al., 2016; Kikkert et al., 2017). It may be that the paralysis of the phantom is what causes some discomfort; and using virtual/augmented reality can provide relief to amputees by activating motor pathways to unparalyze the phantom. In addition, the activation of muscle movements in the residuum increases participation/focus on the VR/AR and may incur enhanced PLP reduction. When individuals merely watched videos of limb movements and were told to align their phantom movements to match the videos, only a modest, transient relief in PLP was recorded (Mercier and Sirigu, 2009). The use of VR/AR has been shown to be beneficial for individuals with intractable PLP in case studies and in small patient cohorts (Ortiz-Catalan et al., 2016).
Multicenter, international, double-blind randomized controlled clinical trials are in progress testing take-home VR treatments, and while preliminary results are encouraging, it appears the benefits largely depend on regular home use and skilled myoelectrode placement by the user (Lendaro et al., 2020). An additional benefit of VR/AR is that it can be customized to an individual's perception of their own phantom. For example, amputees with limb telescoping used AR to recalibrate the size of the phantom to a normal size, which corresponded with normalized cortical activity in regions for the deprived/phantom limb (Thøgersen et al., 2020). VR/AR can be “fun” for participants, so using this technology to learn to power future prostheses or reduce PLP certainly holds great promise.
Neurofeedback
Surgical improvements, innovative biofeedback prostheses, mirror therapy, and VR/AR all likely allow the brain to recalibrate activity by providing peripheral feedback. New advances in imaging could also allow patients to modulate their brain activity without relying on peripheral inputs at all. Real-time fMRI neurofeedback is a mechanism by which people can self-modulate brain activity in patient-specific ways (Sulzer et al., 2013). The fMRI neurofeedback benefits from full brain coverage, high spatial and temporal resolution, and inclusion of connectivity patterns in analysis and readout to patients (Zotev et al., 2011; Paret et al., 2019). Individuals with limb loss suffering from chronic or acute pain have been able to modulate brain regions devoted to pain processing and reduced pain sensation (DeCharms et al., 2005). Able-bodied individuals were also able to modulate motor cortex to shorten motor reaction times in a finger tapping task but required significant clinician guidance on how to modulate their cortical activity (Scharnowski et al., 2015). Successful recovery from amputation requires both reduction in pain and increased motor skill acquisition, which involve distinct brain regions to be modulated in unique ways. Because each individual with limb loss likely has different brain reactions/reorganization after injury, it seems likely that having a tailored intervention strategy such as neurofeedback could benefit recovery.
Conclusions
Understanding the mechanisms underlying the brain's remapping after injury and again during recovery is important if clinicians hope to modulate this process. Consistent results from human and animal studies indicate a few overarching themes. (1) The increased excitability and decreased inhibition in deprived cortex occur as a result of temporally precise changes in the distributions and functions of excitatory (NMDA, AMPA) and inhibitory (GABA) receptors. This hyperexcitability and corresponding map plasticity/phantom responses are correlated with PLP but also leave room for exploitation when considering how to modulate neural activity to facilitate recovery. Facilitating recovery must include both reduction in PLP and also supporting motor learning. For example, if it is determined that remapping after injury depends on deep-layer neurons driving this adaptation, then reducing their activity could slow down remapping until a prosthesis could replace the lost limb. This could look like reducing hyperexcitability in deprived cortex via known and effective NMDA antagonists or GABA agonists. Once a prosthesis is fitted and an individual is learning to use it, modulating plasticity with vagus nerve stimulation, neurofeedback, and/or VR/AR could enhance training protocols to speed up learning and recovery.
Critical gaps in knowledge and translational opportunities
A common refrain among clinicians is “once you know one amputee, you know one amputee.” Patient etiology after limb loss is incredibly varied. Although much has been learned about neuronal and synaptic adaptations that occur after limb loss, basic research models do not (and perhaps should not) attempt to capture this heterogeneity. Instead, a clinical objective may be to find ways to incorporate the many facets of patient experiences into a treatment plan. These should include the following: age and fitness of the patient at injury time, ability to obtain or desire to operate a prosthesis, type of prosthesis, incidence of PLP/phantom limb sensation, and neural activity dysregulation as detected by fMRI or EEG. A practical hurdle for clinicians and individuals with limb loss is the temporal delay between surgery and prosthetic fit. This delay may allow the brain to activate maladaptive plasticity programs that could interfere with recovery. Animal models here are helpful as they support identification of the precise timeline of neural adaptations after injury. These results can support more precise therapeutic timelines for individuals with limb loss.
Although basic researchers can describe neural adaptations in excruciating detail, training rodents to use prostheses or limb transplants is challenging. This limits studies to consider only the adaptations that occur in the brain after the injury, and not necessarily those that could support prosthetic use. However, rodents do learn new motor strategies to overcome their injury, for example, using intact body parts to compensate for the loss of whiskers or limbs (Yu et al., 2010). A goal of basic researchers could be to find neural elements that support general motor learning after injury which could then be adapted by clinicians for human patients. Rodents display hyperalgesia behaviors after injury, but phantom limb symptoms have not yet been described. Basic researchers can support clinicians by describing the neural activity patterns in rodents that relate specifically to hyperalgesia with an eye toward modulating activity to reduce pain-related behaviors in rodent amputation models.
Finally, the brains of rodents are smaller and simpler than those of humans. Adaptations related to neurotransmitters or between individual synapses may be preserved; however at the systems level, likely there are major differences that impair basic research's ability to support clinical research. For example, after limb loss humans increase reliance on visual feedback to control intact or prosthetic limbs. It is unlikely that rodents (with poor vision) would use this adaptation. Studying systems-level adaptations in rodents seems unlikely to support clinical research, but further work with animal models allows precise descriptions of the following: (1) the temporal nature of injury adaptation, (2) identification of the molecular mechanisms of neuronal adaptations, and (3) identifying and targeting the underlying behavioral mechanisms (i.e., hyperalgesia vs successful motor learning). Finally, if basic researchers find that the same injury produces different behavioral outcomes, clinicians may have a more satisfying answer as to why some patients experience debilitating pain and others adapt successfully—it may not be related to the injury event but may instead be something fixable in the patient.
Promising ways forward
Limb loss deprives the central nervous system of sensory input, motor output, and visual feedback that help orient us to our position in the world. Therapies that restore these pathways are likely effective because they allow the brain to recalibrate to as natural a life experience as possible after injury. For example, MI and AO activate sensorimotor pathways that could benefit patients learn new motor skills after injury. Mirror therapy temporarily restores visual feedback, while surgical techniques give peripheral nerves new sensory/motor connections. With prostheses increasing in dexterity, and intuitive input/output designs, individuals may be able to restore all three of these feedback pathways in the brain.
New strategies to enhance motor learning and reduce pain are being developed. Although many have been pioneered in isolation, some therapies may work well together. For example, pairing AO with peripheral nerve stimulation induces motor cortex activity aligned with motor learning at equal levels as pure motor training (Bisio et al., 2017). This is a unique opportunity for amputees who lack a limb with which to train but could use AO plus nerve stimulation to improve prosthetic or residual limb use. It appears that adding additional strategies does not necessarily translate into additional gains. For example, performing motor training before and after AO plus nerve stimulation did not provide additive effects—instead it is hypothesized that the strategies use similar circuitry, and the first training session occluded gains in the second (Bisio et al., 2017). In other instances, combining mirror therapy and/or MI with intermanual transfer does not enhance task performance in able-bodied controls learning to use a prosthesis (Romkema et al., 2018). The authors hypothesized that the tasks were too difficult and the training period was too short (5 d), compared to the weeks typically used for mirror therapy or MI. These studies highlight a few necessities: (1) identify which neural networks support motor learning, and if two interventions stimulate the same one, they may not be appropriate to use together. (2) Interventions should be prescribed for a duration that has been shown to be effective. Enhancing motor learning does not occur in isolation of pain. Training must consider patient discomfort and attempt to tailor training that does not cause additional pain.
As new technologies allow better characterization of neural adaptations in humans and animal models, we are poised to make some exciting progress in improving the quality of life in people living with limb loss. Therapies that will work best will capitalize on what is known about these adaptations and can be tailored to each individual.
Footnotes
We thank Alan Korestky for his helpful commentary. E.P. is supported by NIH R00 NS112612-03.
The authors declare no competing financial interests. The opinions and assertions expressed herein are those of the author(s) and do not reflect the official policy or position of the Uniformed Services University of the Health Sciences or the Department of Defense.
- Correspondence should be addressed to Tawnee Sparling at tawnee.sparling{at}usuhs.edu or Emily Petrus at emily.petrus{at}usuhs.edu.
