Brain (re)organisation following amputation: Implications for phantom limb pain

1Three decades ago it was first demonstrated that the sensory and motor maps in the adult primate brain can change as a consequence of injury as well as in response to training and stimulation, and that these organisational changes are not limited to early brain development. This remapping has been attributed to an unmasking of normally inhibited connections between representational areas, changes in subcortical projections to cortex and even structural changes such as axonal sprouting (though see on changes after spinal cord injury). Subsequently, alterations in the organisation of sensory maps have been associated with a number of perceptual and behavioural changes, with many of them viewed as maladaptive, ranging from tinnitus to focal dystonia and phantom limb pain (PLP). However, these map changes have also been related to adaptive behaviors, such as improved sensory discrimination, advanced musical training and recovery from stroke. Here we review recent evidence on map changes in the primary somatosensory cortex (S1), and its association to PLP, based on recent neuroimaging studies in humans.

2Topographic body representation is one of the foundational organising principles in the brain. Multiple reports from the late 19th century observed that localised electric stimulation in animals evokes individualised movements of specific body parts. This work was refined by Penfield and colleagues, who reported a body-part map (somatotopy) along the human primary motor cortex (M1). Benefiting from verbal reports of their awake patients, Penfield and colleagues were able to infer that a second body map, relating to sensory perception, existed adjacently in the postcentral gyrus. These classical results, later elaborated in both animal and human cortical stimulation studies, established the role of S1 in eliciting key aspects of sensory bodily perception (e.g. stimulus modality, location, frequency and amplitude; note that S1 stimulation did not elicit painful sensations). More recent work, specifically focused on the causal role of S1, indicates that somatosensory perception might not be as heavily reliant on its processing. Instead, S1 functioning may be particularly important for the consolidation of sensory and motor learning.

3The consistent organisation of the sensorimotor maps was summarized in the canonical illustration of the human homunculus by Penfield and colleagues. However, despite this simplified illustration, the authors emphasised that the body map contains fuzzy boundaries, due to overlapping representations across body parts. Further work over the course of the 20th century revealed that although the motor map is best defined as crudely organised, S1 contains detailed representations of specific body parts characterised by clearer boundaries. The most striking, fine-grained organisation exists for the hand, where each of the digits and digit pads are represented separately and adjacently, resulting in a detailed hand map.

4The rise of neuroimaging allowed the investigation of somatotopic maps in the healthy human brain using PET, MEG, task-based fMRI, resting-state functional connectivity and even structural neuroimaging techniques (e.g. DTI tractography). Unlike direct brain stimulation and recordings, human neuroimaging provided opportunities to identify body and hand representation beyond S1/M1 – namely in the cerebellum, basal ganglia, operculum and insula, supplementary motor cortex, occipitotemporal cortex, parietal cortex and more. Despite these findings indicating that body representation should not be studied in S1 in isolation, in the rest of this paper we will primarily focus on S1 body representation as identified with task-based fMRI (see Fig. 1 for an illustrating example). We will not consider the literature on M1 (re)mapping nor will we discuss the interactions of S1 changes with peripheral factors, although we acknowledge their importance in understanding S1 organisation and PLP.

5There are several methodological considerations that have constrained the study of S1 somatotopy, which should be taken into consideration before interpreting this wealth of literature. First, due to the need for multiple repetitions of each condition, and the limited time-frame of a typical neuroimaging session, very few studies attempted to reconstruct the full homunculus. With few exceptions (e.g.), most studies focused on the relative layout of several distinct body parts, and most prominently the foot, hand and mouth. This leaves potential gaps in the layout of the human somatotopy (e.g., it is still debated whether the hand representation neighbours the upper or lower face representation). Other studies have focused on detailed investigations of one body part (most commonly the hand). Here the classical studies were restricted by limited resolution, making the dissociation of individual digits relatively noisy. Recent advances in fMRI, and in particular techniques that provide increased signal to noise ratio and spatial resolution (multiband sequences and 7T MRI), offer new opportunities to overcome some of these technical issues. These new advancements allow for characterisation of the S1 hand maps with unprecedented detail: 7T fMRI permits identification of individual-digit finger maps with high intra- and inter-subject consistency. Structural techniques provide a first glimpse into the anatomical constraints of this organisation, for example, the hand-face border.

6A further constraint for S1 mapping relates to the means of stimulation available, particularly when considering the restricted environment of MRI scanners, safety considerations, limited physical space and magnetic field distortions. Additionally, the peripheral nervous system has been shown to display high adaptation rates to repeated tactile stimuli. For these reasons, many researchers opted to using active paradigms, where participants are cued to move different body parts resulting in robust S1 activity. However, this approach raises its own confounds, in particular unreliable delineation of S1 from M1 with standard acquisition resolution and pre-processing techniques. Moreover, movement-related S1 activity differs from passive sensory stimulation S1 activity, in that it recruits multiple additional inputs, including efferent signals from the motor system, deep cutaneous and proprioceptive peripheral signals and even cognitive or multisensory top-down modulatory inputs, though see and for similarities across multiple representational features of the hand and for similarities in homunculus remapping between active and passive paradigms.

7Non-invasive research in humans has also opened up new opportunities to study the role of S1 in pain processing, and specifically what are the organising principles of nociception in S1. Combined with electrophysiological research in both animals and humans, the evidence is inconclusive. While some studies demonstrated that painful stimuli are being processed in S1, other studies suggested that the observed activity does not reflect the nociceptive input per se, but rather other aspects relating to pain such as attention, salience or expectation. The spatiotemporal attributes of pain versus touch representation in S1 have also been debated, with some studies showing overlap in topographic finger representation, while others identified differential organisation. In the context of our review, it is important to note that nerve injury such as following amputation is known to trigger central sensitisation, whereby painful stimulation can activate S1 indirectly, due to plasticity in the dorsal pathway in the spinal cord.

8As highlighted above, the organising principles of the S1 somatotopy beyond nociception are highly ubiquitous. As such, S1 somatotopy provides an ideal model to address the question of brain remapping – can the properties of the map change, and in particular the boundaries between distinct body parts, once these have been established? Previous electrophysiology research in monkeys has identified extensive changes to the map features following amputation of a single digit, deafferentation of a nerve or the entire arm. Here it was found that once neurons are deprived of their primary input, they become responsive to stimulation that activates the cortical neighbours of the deprived area. This activity change results in shifted boundaries of the body part map, termed cortical reorganisation. However, considering the key contribution of unmasking of already existing inputs in driving the shifted boundaries of the body map, this description might be misleading. Perhaps the most dramatic demonstration of remapping occurs following arm deafferentation. Here the hand area of monkeys becomes responsive to inputs from the lower face – whose representation neighbours the hand area (though note that the activity in the deprived cortex is considerably smaller in amplitude to native face activity; see Figure 5 in for an example). Since hand and face inputs are segregated throughout the somatosensory hierarchy leading to S1, and since only sparse connections normally exist across the hand-face boundary, this process might reflect more profound changes to organising features of the body map.

9In humans, research characterising lower face representation in unilateral arm amputees has not identified clear facial activity in the missing hand cortex (Fig. 3C). Instead, multiple studies examining activity associated with stimulation of the lower face (both actively and passively have found that the centre (or spatial extent) of the lip cluster becomes medially shifted in the deprived hemisphere, particularly with respect to the opposite (intact) hemisphere. Critically, the extent of this lip activity shift associates with PLP intensity (;; Fig. 2). PLP is a sub-class of phantom sensations that amputees report as arising from their missing limb. While these sensations range in their specific characteristics (spanning an extensive range of tactile, proprioceptive and kinesthetic sensations) they are often also experienced as bothersome and painful, usually developing into a chronic condition, which is difficult to treat. This is due to the fact that we still do not have a complete understanding of its neural basis. As such, the observed correlation between S1 remapping and PLP intensity has opened up new avenues for exploring the mechanisms behind, and novel treatments for, PLP.

10In particular, it has been proposed that the displaced facial inputs caused by the deprivation-triggered remapping prompt aberrant processing in the S1 hand area, which may in turn be interpreted as phantom sensation or pain arising from the missing hand. As mentioned above, it was found that people who experience worse PLP also show a greater shift of their lower face representation towards the missing hand cortex. A similar association was never established for non-painful phantom sensations, which have been in turn linked with frontal and parietal activity. The finding of a positive association between S1 remapping and PLP has been replicated using multiple paradigms and participant groups. It provided first evidence that reorganisation of the S1 map bears functional consequences, maladaptive at that, with new opportunities for PLP treatment. However, it also opens up a host of questions about the potential relationship between brain remapping and PLP, in particular – how would displaced input to S1 interact with other pain mechanisms to manifest the pain sensation? What is the time-course of cortical remapping with respect to PLP? Is this relationship between remapping and PLP also found following remapping of other body parts into the missing hand cortex? (see for further considerations). Some of these questions are starting to unravel by neuroimaging studies that focus on neural changes following various treatment approaches to PLP, as described below.

11Human neuroimaging studies have also examined the neural representation of the missing hand itself. As mentioned above, phantom sensations can manifest in the form of kinaesthesia – the sense of movement that amputees experience when volitionally trying to move their phantom hand. Previous research has demonstrated that when amputees are instructed to move their phantom hand, this results in motor output to the denervated motor neurons as evidenced with EMG, demonstrating engagement of the motor system. fMRI studies taking advantage of this simple manipulation have shown that phantom movements activate both M1 and S1 (; Fig. 3A), albeit these original studies could not reliably differentiate activity profiles of these two neighbouring regions. It has been demonstrated that activity elicited by phantom hand movements in the sensorimotor missing hand cortex positively correlates with PLP – people who experience more chronic PLP also exhibit greater activity when moving their phantom hands. This finding highlights that mechanisms pertaining to persistent hand representation may also be relevant for PLP, involving either bottom up (peripheral) or top down processes (e.g., from the motor system). These associations, however, still await a mechanistic framework by which phantom movements relate to PLP and it is therefore currently not clear whether there is a direct causal relationship between persistent representation and PLP, see further discussion below).

12The research on S1 remapping after amputation has yielded novel interventions that target the presumed maladaptive brain remapping. In mirror treatment, the intact hand is moved while the patient views it in the mirror and perceives the mirror image as the phantom hand, while the phantom hand moves along or while no phantom hand movement was present. observed that not all amputees activated the cortical representation of the phantom hand during observation of phantom movements through mirror training, and those that did, experienced less PLP. In a later study, it was found that successful reduction of PLP by several sessions of mirror treatment normalised lip activity, as evidenced by the degree of representational shift away from the missing hand area in comparison to the intact hemisphere. Normalisation of lip activity was found to significantly correlate with treatment success. Interestingly, patients with telescoping – where the phantom hand moves towards the residual limb – did not significantly benefit from mirror treatment. This was potentially due to the disjunct of the intact hand viewed in the mirror and the telescoped phantom percept, creating a perceptual mismatch. Augmented reality applications can overcome some limitations of mirror treatment, and have shown that patients with a telescoped phantom can also benefit from mirror training, as demonstrated by treatment-related reorganisation. In addition, activity of a region in the inferior parietal cortex, which has been associated with body image maintenance, was predictive of reduced PLP. The latter data suggest that the non-painful perception of the phantom also modulates brain changes and PLP in amputees. A further study used a combination of phantom limb movement and sensory feedback in a small sample of amputees, and observed a trend towards a change in lip representation after treatment. Similar changes were reported for motor imagery. These studies show that S1 changes and treatment-induced reductions in PLP covary, but the direction of the relationship, i.e. if the brain changes are a cause or consequence of the change in PLP cannot be determined from these studies. Studies that can assess causal relationships are needed.

13Further studies using brain stimulation techniques, such as transcranial magnetic stimulation (TMS) and direct current stimulation (tDCS), can begin to address the question of causality by inducing cortical change and potential alterations to PLP. TMS studies (as reviewed in) show that the phantom limb representation in motor cortex is shifted in amputees with PLP and that the representation of neighbouring body parts enlarged, in line with findings from S1. A recent tDCS study investigated the consequences of successful PLP reduction and preserved phantom hand representation after stimulation over the S1/M1 missing hand area, involving a functional task (phantom hand movements). PLP reduction was found to correlate with reduced phantom hand activity, in contrast to the findings on mirror treatment. However, this could be attributed to different activation paradigms (active phantom movement versus mirrored movements), as well as use-dependent changes related to the intact hand which is engaged in the mirror paradigms (as discussed below). Importantly, activity of the posterior insula and additional regions involved in pain processing emerged as an important mediator of pain relief and activity changes relating to phantom hand representation. These pain-related regions have not yet been as actively studied in amputees with PLP, compared to patients with other types of chronic pain. For example, research of chronic musculoskeletal pain has been focused on widespread brain alterations and less on S1. In this context, the above mentioned activity could be due to a greater association of PLP with changes in sensory rather than affective processing.

14Other research studying the neural correlates of PLP relief has focused on peripheral contributions to PLP. For example, targeted muscle and sensory reinnervation (TMR) of the amputated limb, which involves connecting nerves leading to the amputated limb to specific muscles that can drive a prosthetic limb has been shown to reduce PLP (; though see) and there is similar initial evidence for intraneural stimulation. Finally, brain computer interfaces (BCI) have been explored as modulators of brain activity and PLP. BCI’s are promising avenues for treatment as they can provide new insights by up- and downregulating brain activity in the phantom cortex and related areas, whilst examining associated pain changes. Here it is worth considering more global consequences of neuromodulation on somatosensation. To date, the relationship between amputation-triggered changes to S1 and altered somatosensation is far from clear, since no systematic studies exist that have related various sensory modalities to brain changes. Studies on psychophysical properties have found both enhanced and deficient sensory processing.

15As mentioned above, remapping following amputation has been considered to be primarily triggered by the loss of sensory input, driving unmasking of weak or silent inputs, and potentially promoting long-term potentiation mechanisms. According to this account, inputs that already have access to the deprived cortex, such as topographic neighbours, are more likely to be expressed in it following deprivation, resulting in a shift or expansion of an existing boundary. However, recent evidence in neuroimaging demonstrates violations of this rationale. For example, it was found that the residual arm, which is immediately neighbouring the missing hand cortex, does not remap into the deprived sensorimotor cortex in amputees, relative to controls. Converseley, multiple studies have documented increased activity in the sensorimotor missing hand cortex for the intact hand. Considering that the deprived cortex is normally inhibited by the intact hand cortex, it is difficult to resolve this finding with classical interhemispheric inhibition mechanisms (e.g. as suggested for stroke). Instead, it has been proposed that the remapping of the intact hand into the missing hand hemisphere relates to compensatory usage.

16It has long been observed that the receptive field (RF) properties of the remapped cortex gradually change over a long time-scale. It has been suggested that the initial remapping triggered by deprivation will become refined by inputs due to daily hand usage involving compensatory behaviours. Since altered input is known to drive and shape brain organisation, it is reasonable to expect that adapted behaviour substituting for the missing hand function can shape remapping. This is akin to recent studies in individuals with congenital hand loss, who exhibit remapping of multiple body parts, which are used for compensatory behaviours, including the feet, residual arm and lips into the missing hand cortex. Considering that deprived cortex has been demonstrated to undergo network-level reorganisation in functional connectivity, and that the connectome of the deprived cortex is wired to support hand function, inputs relating to typical hand function (e.g. manipulating objects) might consolidate more favourably. While still awaiting causal validation for a relationship between changed habitual behaviour and brain remapping in humans with congenital or acquired amputation, this potential process sheds new light on the classical findings. For example, consider the original studies in monkeys showing remapping of the mouth into the hand cortex. If these monkeys tended to use their mouth to substitute for their injured hand function, this could have contributed to the resulting remapping of the mouth into the hand area. It also provides new context to the aforementioned mirror-treatment studies which used intact hand movements to probe representations in the missing hand cortex. Currently, there is no systematic evidence that intact hand remapping associates with PLP, though few studies explored this potential link.

17Prosthetic limb usage was also demonstrated to associate with cortical remapping and PLP, but here again causality needs to be inferred with caution. It is known that individuals suffering from PLP are less likely to wear a prosthetic limb due to discomfort. It has also been demonstrated that individuals that use their prosthetic limb less (or more generally their residual arm), tend to use their intact hand more, presumably due to the need for alternative compensatory strategies. Any of these factors could potentially directly impact S1 organisation and beyond, or serve as an indirect moderating factor for a relationship between S1 remapping, PLP and persistent representation.

18As already noted in section 4 on clinical studies, contextual variables also influence S1 remapping. For example, telescoping, which is associated with PLP and greater S1 remapping, was related to less PLP reduction and less changes in the mouth to hand cortical distance following mirror treatment (Foell et al., 2014). Other evidence for the instability of the S1 finger maps provide further considerations, which could impact the evidence for remapping observed in amputees. For example, acute pain (which will differ across amputees during the scanning session) has been shown to increase cortical distances in S1. Furthermore, Braun and colleagues observed that the hand map can change depending on task requirements and attentional focus. These results confirm earlier suggestions that the fine features of the body maps in S1 are under-determined to an extent, and as such, could dynamically adapt to internal and external conditions within short time periods. Consequently, the context of the experimental paradigm and the details of the analysis could massively impact the boundaries of the body map being described by the researcher. This does not preclude high stability of finger maps under similar conditions. Fig. 4 depicts some of the factors that can impact on remapping in S1.

19Recent multivoxel analysis techniques offer a new lens through which somatotopic (re)mapping can be characterised by providing us with the means to quantify selectivity in terms of how distinct the representation of one body part is to another (for example, linear classifiers or the Mahalanobis distances observed between two representations in a given brain area). Moreover, these techniques provide the means to consider the type and extent of available information underlying the body map. The simple assumption here is that a brain area representing a given body part should contain distinct information about the functional features most relevant to the representation of that body part. For example, in S1 there are distinct neural representations of specific frequencies of tactile stimulus properties, which might not be easily separable based on net activity levels, but result in distinct multivariate representational motifs. Representational similarity analysis (RSA) permits the consideration of functional attributes beyond selectivity of response. For example, it allows characterisation of inter-digit similarity or overlap. Using this approach, it has been elegantly demonstrated that the canonical inter-digit representational structure in the S1 hand area can be best described in terms of daily hand use: those digits that form kinematic synergies to afford our daily interactions with objects and tools also show greater similarity in their representation.

207T fMRI and multivariate analysis approaches have been recently used to investigate S1 hand representation in amputees. With respect to persistent hand representation, 7T fMRI allowed us to uncover S1 digit maps, even up to three decades after amputation (; Fig. 3B). A complete digit map was also found in the missing hand cortex of an amputee with a brachial plexus avulsion (causing peripheral deafferentation), and was abolished when the participant was asked to imagine moving the phantom hand, rather than attempting to move it. As such, it appears that these maps reflect inputs from the motor system which are somatotopically structured. Multivoxel analysis techniques allow us to confirm that the representation of the missing hand, as probed using phantom hand movements (or at least attempted movements), shares the same representational features as a normal hand. In particular, we recently found that the unique configuration of inter-finger representational structure (previously linked with everyday experience), is entirely preserved after an average of 18 years since amputation. We found that across participants there was a significant relationship between the typicality of phantom hand representation and the sense of phantom kinaesthesia, specifically the number of digits that amputees perceived to be moving during the task. However, we found no significant correlations with either acute or chronic PLP, or the vividness of non-painful phantom sensations (as revealed in a linear regression model). In fact, typical hand representation was also found in the few amputees who did not experience any phantom sensations, suggesting it is a canonical organising principle, rather than a neural correlate of the phantom hand. A further study by Bruurmijn and colleagues, who used brain decoding to separate representations of multiple phantom hand gestures, was able to successfully decode the different gestures well above chance. Although decoding accuracy was reduced when compared with controls in the earlier sub-divisions of S1, this group difference was resolved in the higher sub-divisions (BA 1 and 2), indicating that the preserved representation of the phantom hand is at least partially maintained by higher order processing. Here again, no significant correlations were found with PLP (though considering the small sample size of n ​= ​8, this null result should be regarded with caution).

21Similar techniques have also been used to study the intact hand, specifically the increased ipsilateral activity that has been previously observed in the missing hand cortex. Previous research in two-handed adults demonstrated that the ipsilateral hand area contains finger-specific information which mirrors the contralateral representation of the same finger (Diedrichsen et al., 2013; Diedrichsen et al., 2018). This approach allows us to ask whether information content is greater for the ipsilateral representation of amputees’ intact hand relative to controls. In a recent study we found that ipsilateral inter-finger dissimilarity, underlying digit selectivity, was not significantly different between amputees and controls. This result indicates that the increased intact hand activity reported in amputees might not underlie increased functional processing. Similar evidence from stroke and focal dystonia patients (Ejaz et al., 2016), also showed no changes in inter-finger representational features using RSA. This is in contrast to previous studies demonstrating that hand somatotopy is altered in these groups of interest, as mentioned in our introduction. Since this recent evidence demonstrates that information processing of inter-finger representation is unchanged, it is therefore possible that the previously documented map changes, which were based on measures of net activity changes, do not entail reorganisation of representations.

22Our review shows that there is no simple relationship between somatosensory map reorganisation, PLP and preserved hand representation. This perspective converges with recent studies in related clinical conditions, for example complex regional pain syndrome and peripheral neuropathy, showing that the previously assumed relationship between S1 remapping and pain requires further consideration. Although injury-related plasticity mechanisms may be the original driver of map changes, over time a number of additional factors such as use of a prosthesis, intact hand use, alterations in body representation as a consequence of amputation and processing in other regions, particularly with regards to pain processing, may interact with injury-related remapping. Furthermore, the close and bidirectional connections of somatosensory and motor cortex are also likely to impact S1 organisation. These interactions need to be examined and related to functional changes. Thirdly, the traditional focus on S1 remapping due to changed selectivity to particular body parts is probably incomplete, considering net activity changes might not adequetly reveal the underlying functional processing. Moreover, related research in monkeys and humans indicate that sensorimotor cortical remapping might reflect changes in sub-cortical terminals (e.g. brainstem, basal ganglia) or cerebellum. Future research should aim to study PLP within this broader context, while taking into consideration the role of multiple brain networks and other contextual factors that can alter or stabilise the S1 body map at muliple time scales.

23Tamar R. Makin: Conceptualization, Writing - original draft. Herta Flor: Conceptualization, Writing - original draft.