Complex pattern of facial remapping in somatosensory cortex following congenital but not acquired hand loss

  1. Victoria Root
  2. Dollyane Muret  Is a corresponding author
  3. Maite Arribas
  4. Elena Amoruso
  5. John Thornton
  6. Aurelie Tarall-Jozwiak
  7. Irene Tracey
  8. Tamar R Makin
  1. WIN Centre, University of Oxford, United Kingdom
  2. Institute of Cognitive Neuroscience, University College London, United Kingdom
  3. Medical Research Council Cognition and Brain Sciences Unit (CBU), University of Cambridge, United Kingdom
  4. Department of Psychosis Studies, Institute of Psychiatry, Psychology & Neuroscience, King's College London, United Kingdom
  5. Wellcome Trust Centre for Neuroimaging, University College London, United Kingdom
  6. Queen Mary’s Hospital, United Kingdom

Abstract

Cortical remapping after hand loss in the primary somatosensory cortex (S1) is thought to be predominantly dictated by cortical proximity, with adjacent body parts remapping into the deprived area. Traditionally, this remapping has been characterised by changes in the lip representation, which is assumed to be the immediate neighbour of the hand based on electrophysiological research in non-human primates. However, the orientation of facial somatotopy in humans is debated, with contrasting work reporting both an inverted and upright topography. We aimed to fill this gap in the S1 homunculus by investigating the topographic organisation of the face. Using both univariate and multivariate approaches we examined the extent of face-to-hand remapping in individuals with a congenital and acquired missing hand (hereafter one-handers and amputees, respectively), relative to two-handed controls. Participants were asked to move different facial parts (forehead, nose, lips, tongue) during functional MRI (fMRI) scanning. We first confirmed an upright face organisation in all three groups, with the upper-face and not the lips bordering the hand area. We further found little evidence for remapping of both forehead and lips in amputees, with no significant relationship to the chronicity of their phantom limb pain (PLP). In contrast, we found converging evidence for a complex pattern of face remapping in congenital one-handers across multiple facial parts, where relative to controls, the location of the cortical neighbour – the forehead – is shown to shift away from the deprived hand area, which is subsequently more activated by the lips and the tongue. Together, our findings demonstrate that the face representation in humans is highly plastic, but that this plasticity is restricted by the developmental stage of input deprivation, rather than cortical proximity.

Editor's evaluation

This fundamental work substantially advances our understanding of cortical remapping in people with congenital or acquired missing hands. The evidence supporting the idea that remapping may not follow cortical proximity but instead functional rules as to how the effector is used are compelling, with rigorous univariate and multivariate analyses applied to functional Magnetic Resonance Imaging data. Importantly, the authors suggest this is mostly the case for one-handers but not for amputees for who the reorganization seems more limited in general.

https://doi.org/10.7554/eLife.76158.sa0

Introduction

Our brains capacity to adapt, known as cortical plasticity, is integral to our successful functioning in daily life, as well as rehabilitation from injury. A key model for exploring the extent, and consequences of, cortical plasticity is upper-limb loss (via amputation or congenital absence). Here, the cortical hand territory in the primary somatosensory cortex (hereafter S1), suffers an extreme loss of sensory input in tandem with dramatic alterations of motor behaviour (Makin et al., 2013a; Muret and Makin, 2021). The functional and perceptual correlates of amputation-related plasticity are currently debated (Makin and Bensmaia, 2017; Ortiz-Catalan, 2018). In particular, it is not clear whether functional cortical reorganisation is restricted to early life development or can also occur in adults.

Traditionally, research assessing cortical plasticity after upper-limb loss has followed the tenet that neighbouring body parts of the missing hand, and the lower face in particular, shift and encroach into the deprived hand area. This emphasis on the lip representation stems from early electrophysiological work in non-human primates, where numerous studies demonstrated an ‘upside-down’ facial somatotopy, with the lower face immediately neighbouring the hand (Dreyer et al., 1975; Merzenich et al., 1978; Sur et al., 1982; Cusick et al., 1986; Lin and Sessle, 1994; Manger et al., 1995; Manger et al., 1996; Jain et al., 2001; Cerkevich et al., 2014). Here, the lips and/or chin inputs have been shown to remap into the deprived hand area after sensory loss (Pons et al., 1991; Jain et al., 1997), leading to the well-accepted assumption that remapping is determined by cortical proximity (Buonomano and Merzenich, 1998; Nardone et al., 2013). Thereafter, human measurement of topographic shifts has tended to focus on that of the lips, where researchers have reported that shifted lip representation towards and into the deprived hand area is significantly associated with phantom limb pain (PLP) intensity (Flor et al., 1995; Birbaumer et al., 1997; Lotze et al., 2001; Grüsser et al., 2001; Foell et al., 2014). PLP is a neuropathic pain syndrome experienced in the missing, amputated limb by the majority of amputees (Limakatso et al., 2019). This condition is commonly thought to arise from maladaptive cortical plasticity in S1 (although see Makin, 2021), specifically from a signal mismatch between the missing hand representation and the remapped inputs of the lips in the deprived hand area (Ramachandran and Hirstein, 1998).

The research focus on lip cortical remapping in amputees is based on the assumption that the lips neighbour the hand representation. However, this assumption goes against the classical upright orientation of the face in S1 (Penfield, 1950; Schwartz et al., 2004; Roux et al., 2018; Sato et al., 2005; Willoughby et al., 2020), as first depicted in Penfield’s Homunculus and in later intracortical recordings and stimulation studies (Penfield, 1950; Schwartz et al., 2004; Roux et al., 2018; Sato et al., 2005), with the upper-face (i.e. forehead) bordering the hand area. Furthermore, neuroimaging studies in humans studying face topography provided contradictory evidence for the past 30 years. While a few neuroimaging studies provided partial evidence in support of the traditional upright face organisation (Willoughby et al., 2020), other studies supported the inverted (or ‘upside-down’) somatotopic organisation of the face, similar to that of non-human primates (Yang et al., 1993; Servos et al., 1999). Other studies suggested a segmental organisation (Moulton et al., 2009), or even a lack of somatotopic organisation (Iannetti et al., 2003; Nguyen et al., 2004; Kopietz et al., 2009), whereas many studies provided inconclusive or incomplete results (Mogilner et al., 1994; Hoshiyama et al., 1996; Disbrow et al., 2003; Nevalainen et al., 2006). Together, the available evidence does not successfully converge on face topography in humans. In line with the upright organisation originally suggested by Penfield, recent work reported that the shift in the lip representation towards the missing hand in amputees was minimal (Makin et al., 2015; Raffin et al., 2016), and likely to reside within the face area itself. Surprisingly, there is currently no research that considers the representation of other facial parts, in particular the upper-face (e.g. the forehead), in relation to plasticity or PLP. Detailed mapping of the upper and lower face is therefore needed to assess typical topography of facial sensorimotor organisation, as well as remapping after limb loss.

More recent electrophysiological studies in monkeys demonstrated that much of the face remapping observed in the primary sensorimotor cortex following upper-limb deafferentation does not result from cortico-cortical plasticity, but instead arises from plasticity in the brainstem (Florence and Kaas, 1995; Kambi et al., 2014). This important finding also highlights the limited explanatory power of local activity increase that has been historically used to infer changes to the representational features of a given brain. Does it reflect a gain modulation of input coming from the brainstem, or increase of local processing of the face input in the hand area? While it remains challenging to dissociate these two contributions, alternative analysis tools are becoming increasingly popular for mining richer information of the processing underlying activity in a given cortical region. Multivariate analyses are sensitive to more subtle changes in representational content, that are not accessible with the traditional univariate approach. In the context of facial remapping, if the deprived hand area updates its local processing to include facial information (and does not just display more facial activity), we would expect it to show greater information about representational features relevant to different facial parts.

Remapping after upper-limb loss has also been documented in individuals born without a hand (hereafter one-handers), who do not experience PLP (Makin et al., 2013b). Here, it has been shown that the representation of multiple body parts, including the residual arm, legs and lips, remapped into the missing hand territory (Hahamy et al., 2017; Hahamy and Makin, 2019). Importantly, cortical remapping in this group does not depend on cortical proximity. With regard to the lips, a recent transcranial magnetic stimulation (TMS) study has reported functionally relevant lip activity in the deprived hand area of one-handers (Amoruso et al., 2021), showcasing that reported remapping may also be functional. It was proposed that the observed remapping of various body parts could have been shaped by compensatory behaviour (Hahamy et al., 2017), as these body parts are all used by one-handers to compensate for their missing hand function, but this hypothesis awaits validation.

Here, we conducted a mapping of face cortical organisation to determine facial orientation (upright versus inverted) in the primary sensorimotor cortex in 22 two-handed controls, 17 amputees, and 21 one-handers using an active functional MRI paradigm, where participants were visually instructed to move their forehead, nose, lips or tongue. This paradigm was chosen because it enabled simultaneous bilateral activation of S1 within individual participants, providing a within-participants control design. We also explored the representation of multiple facial movements, which have not been previously studied in the context of deprivation-triggered brain plasticity. To measure the extent of cortical remapping of the upper (forehead) and lower (lips) face in relation to the deprived (or non-dominant) hand area across all groups, we used surface-based topographic comparisons. For this purpose, we employed up-to-date methodology to harvest traditional measures, that is winner-takes-all assessment of surface coverage in the hand and face areas, followed by cortical (geodesic) distances, and similarity analysis of selectivity maps. Furthermore, to go beyond the gross topographic properties of face representation, we used multivariate representational similarity analysis (RSA). This approach allows us to characterise more subtle alterations in the relationship between facial activity patterns (forehead, nose, lips, tongue) in the deprived hand and the face areas.

We found that, in line with Penfield’s original description, facial topography was arranged in an upright manner, with the forehead (i.e. upper-face) bordering the hand area across all groups. Contrary to traditional theories (Flor et al., 2006), we did not find evidence for facial remapping of either the lips or the (cortically neighbouring) forehead, into the deprived hand area of amputees. We did, however, observe significant remapping of multiple face parts (upper and lower face) in the one-handers group, validating our methodology as suitable for identifying remapping effects. Interestingly, remapping of the cortical neighbour (upper-face) within the one-handers group was away from the missing hand area, while the lips and tongue representations shifted towards the deprived cortex, hinting that the underlying mechanism of remapping is more complex than simple cortical proximity.

Results

The cortical neighbour of the hand representation is the forehead

We first visualised the average group activity resulting from active movements with each of the facial parts (versus rest), within a broad sensorimotor mask. When looking at gross facial organisation at the group-level, we found qualitatively similar activity maps across groups (see Figure 1), highlighting a robust somatotopy of the face with preserved symmetry across the two hemispheres. These facial maps also indicate an upright orientation of the face in S1, with the forehead located closest to the hand area, followed by the nose, lips, and the tongue located laterally, across all groups. The facial somatotopy presented here therefore suggests that the hand’s cortical neighbour is the forehead (or upper-face), highlighting the need to reassess the often-cited, traditional lip-to-hand marker of cortical remapping in amputees and one-handers. However, conclusions based on threshold-dependant group averages may be misleading as they ignore inter-individual differences.

Group-level activity maps for each facial movement versus rest.

Group average activity for the forehead (red), nose (yellow), lips (blue) and tongue (green) movements, contrasted to rest, in the (A) deprived/non-dominant and (B) intact/dominant hemisphere of controls (n=22), amputees (n=17), and one-handers (n=21). All clusters were created using a threshold-free cluster enhancement procedure with a sensorimotor pre-threshold mask (defined using the Harvard Cortical Atlas; outlined in darker grey), and thresholded at p<0.01. The hand and face regions of interest (ROIs) are outlined in purple and orange respectively, and the central sulcus is denoted with a white arrow.

One-handers, but not amputees, show lip remapping in the deprived cortex based on univariate topographic mapping

To account for inter-individual differences in functional topography and brain topology, we calculated for each participant a winner-takes-all map across facial parts within a (combined) hand and face S1 region of interest (ROI). Focusing on the centre of gravity (CoG) of the lips cluster, we first explored changes in the cortical (geodesic) distance between the lips and an anatomical landmark (~1 cm lateral to the hand knob) of amputees and controls. Here we found no statistically significant main effects or group x hemisphere interaction (F(1,36)=0.019, p=0.890, n2p=0.001, BF10=0.297; controlled for brain size volume; Figure 2B), indicating that the lip area in amputees is not located differently to that of controls. We also measured the proportion of the deprived hand ROI occupied by the lip-winner surface area (relative to the intact hand area; Laterality index). We did not find a significant remapping of the lips (i.e., greater surface coverage) in the missing hand ROI for amputees when compared to controls (U=141.000, p=0.197, rb = 0.246, BF10=0.579; Figure 2C) or to zero (W=91.000, p=0.245, rb=0.338, BF10=0.472; Figure 2C), though here the Bayes Factors did not provide conclusive evidence. Together, these results suggest, contrary to popular theories on brain plasticity in amputees (Flor et al., 2006), that the lips do not remap into the deprived hand area. We next compared the lips laterality index between those individuals who reported suffering from PLP (n=11) and those who no longer experienced chronic PLP (n=6) and found no significant differences (U=27.000, p=0.295, rb = 0.182, BF10=0.629).

Characterisation of lip (re)mapping in the primary somatosensory cortex.

(A) Group-level consistency map for the lips clusters resulting from the individual winner-takes-all maps in the S1 ROI (defined by combining the hand and face areas). The colour gradient represents participant agreement for the lips ‘winning’ that particular voxel, relative to other face movements. Please note that the individual-participant winner-takes-all maps are minimally thresholded, and thus produce an inherently different spatial distribution relative to the group contrast maps presented in Figure 1. The hand ROI is outlined in purple and central sulcus denoted by the white arrow. (B) Cortical geodesic distances from the lip CoG to the anatomical landmark (~1 cm lateral to the hand knob) are plotted for amputees (n=17), controls (n=22), and one-handers (n=21). Distances in the intact/dominant hemisphere are plotted in light blue, and distances in the deprived/non-dominant hemisphere are plotted in darker blue (in amputees and one-handers/controls, respectively). Positive distances indicate the lips CoG is located medial to the anatomical landmark, and negative distances indicate the lips CoG is located lateral to that landmark. The anatomical landmark itself equates to a geodesic distance of zero. For main effects of comparison between amputees and one-handers versus controls, see Figure 2—source data 1–2. (C) Laterality indices for the proportion of surface area coverage of the lips in the hand ROI for all groups (amputees, controls and one-handers). Positive values indicate greater surface area coverage in the deprived/non-dominant hemisphere relative to the intact/dominant hemisphere (in amputees and one-handers/controls, respectively), and negative values reflect greater surface area coverage in the intact/dominant hemisphere relative to the deprived/non-dominant hemisphere. Standard error bars and all individual data-points are plotted in grey and uncorrected for brain size. Amputees with PLP (yes/no) are plotted in orange. ** p<0.01; coloured asterisk’s indicate values are significantly different from zero.

Figure 2—source data 1

Main effects and interaction for comparison of geodesic distances between amputees and controls for the lips.

https://cdn.elifesciences.org/articles/76158/elife-76158-fig2-data1-v2.docx
Figure 2—source data 2

Main effects and interaction for comparison of geodesic distances of the lips between one-handers and controls.

https://cdn.elifesciences.org/articles/76158/elife-76158-fig2-data2-v2.docx
Figure 2—source data 3

Raw data for cortical geodesic distances of the lips for amputees, controls, and one-handers.

https://cdn.elifesciences.org/articles/76158/elife-76158-fig2-data3-v2.xlsx
Figure 2—source data 4

Raw data for laterality indices of the lips for amputees, controls, and one-handers.

https://cdn.elifesciences.org/articles/76158/elife-76158-fig2-data4-v2.xlsx

When visualising the average lip activity within the one-handers group, however, we did note a slight qualitative shift in the location, and spread, of the lip activity within the deprived hemisphere (Figure 1A). This is further supported by a visible shift of the one-handers lip-winner consistency map towards the deprived hand area (Figure 2A). These qualitative changes in the lip representation resulted in a significant group x hemisphere interaction for the lips cortical distance to the anatomical landmark in one-handers and controls (F(1,40)=4.352, p=0.043, n2p=0.098; controlling for brain size; Figure 2B). Confirmatory comparisons indicated no statistically significant shifts of the lip CoG in the deprived hemisphere when compared to the controls non-dominant hemisphere (t(40)=-1.178, p=0.246,~d = −0.395, ~BF10=0.148; corrected alpha = 0.025; uncorrected p-values reported; Figure 2B). Although when compared to their intact hemisphere, shorter distances from the lips to the hand area were found in the deprived hemisphere of one-handers (t(40)=-3.374, p=0.002,~d = −0.621), indicating evidence for lip remapping. These shifts in the deprived hemisphere were also reflected in significantly greater surface area coverage of the lips in the hand ROI when compared to controls (t(41)==-2.762, p=0.009, d=−0.843; Figure 2C), which was significantly different from zero (W=1098.000, p=0.006, rb = 0.426). This converging evidence of lip remapping is in line with previous work in one-handers (Hahamy et al., 2017; Amoruso et al., 2021).

One-handers, and to a lesser extent also amputees, show forehead remapping away from the hand area in the deprived cortex

As we note a qualitative upright orientation of the face (see Figure 1), the question remains as to whether the neighbour to the hand – the forehead – would reorganise after limb loss in amputees, as hypothesised by traditional theories (Flor et al., 2006). Again, we found no significant evidence for cortical remapping of the neighbouring forehead in amputees when assessing changes in cortical distances (group x hemisphere: F(1,36)=1.338, p=0.255, n2p=0.036, BF10=0.695; controlled for brain size volume; Figure 3B). But a significant difference was found for reduced forehead surface area coverage in the deprived hand ROI when compared to controls (t(37)=2.236, p=0.031, d=0.722; Figure 3C). Interestingly, the direction of this effect indicates less, not more, remapping of the forehead in the deprived hand ROI of amputees. However, note that this decrease of surface area coverage was not significantly different from zero (t(16)=-1.86, p=0.082, d=−0.451, BF10=1.012; Figure 3C). When comparing the forehead laterality index for amputees with and without PLP, no significant differences were found (t(15)=-0.729, p=0.761, d=−0.370, BF10=0.291). Taken together, these results suggest that if remapping of the cortical neighbour – the forehead – does occur, this occurs away from the hand area, and is not related to PLP.

Characterisation of forehead (re)mapping in the primary somatosensory cortex.

All annotations are as in Figure 2. For main effects of cortical geodesic distance comparison between amputees and one-handers versus controls, see Figure 3—source data 1–2. Distances in the intact/dominant hemisphere are plotted in pink and deprived/non-dominant hemisphere in red. (B) # p<0.05; * p<0.025 (corrected alpha); (C) * p<0.05; ** p<0.01; *** p<0.001; coloured asterisk’s indicate values are significantly different from zero.

Figure 3—source data 1

Main effects and interaction for comparison of geodesic distances between amputees and controls for the forehead.

https://cdn.elifesciences.org/articles/76158/elife-76158-fig3-data1-v2.docx
Figure 3—source data 2

Main effects and interaction for comparison of geodesic distances between one-handers and controls for the forehead.

https://cdn.elifesciences.org/articles/76158/elife-76158-fig3-data2-v2.docx
Figure 3—source data 3

Raw data for cortical geodesic distances of the forehead for amputees, controls, and one-handers.

https://cdn.elifesciences.org/articles/76158/elife-76158-fig3-data3-v2.xlsx
Figure 3—source data 4

Raw data for laterality indices of the forehead for amputees, controls, and one-handers.

https://cdn.elifesciences.org/articles/76158/elife-76158-fig3-data4-v2.xlsx

When looking at the one-handers group we did find significant evidence of forehead remapping with a group x hemisphere interaction (F(1,40)=7.437, p=0.009, n2p=0.157; controlled for brain size volume; Figure 3B). Confirmatory comparisons indicated a positive trend for shorter distances of the foreheads’ CoG to the anatomical landmark in the deprived hemisphere when compared to their intact hemisphere (t(40)=2.085, p=0.043,~d = 0.435,~BF10=1.18; corrected alpha = 0.025; trend defined as p<.05; uncorrected p-values reported) and significantly shorter distances when compared to the controls non-dominant hemisphere (t(40)=2.580, p=0.014,~d = 0.774). As the forehead’s CoG tended to be located above the anatomical landmark (see Figure 3A), these results indicate a significant shift of forehead activity away from the deprived hand ROI. This is further supported by a significant decrease of surface area coverage for the forehead in the deprived hand ROI when compared to controls (t(41)=3.676, p<.001, d=1.122), which was significantly different from zero (t(20)=-3.57, p=0.002, d=−0.779; Figure 3C). Remapping of the cortical neighbour in one-handers, therefore, manifests in a shifting away of the upper-face from the deprived hand area, possibly due to increases in activity of other facial movements, for example lips.

Tongue movements produce different topographic maps across groups

We also assessed changes in the tongue representation, which is not an immediate neighbour to the hand in S1 (Figure 4A). We did find evidence for significant shifts in the tongue’s CoG towards the anatomical landmark in amputees when compared to controls (group x hemisphere: F(1,36)=4.859, p=0.034, n2p=0.119; controlled for brain size volume; Figure 4B). Confirmatory comparisons indicated significantly shorter distances in the deprived hemisphere of amputees when compared to their intact hemisphere (t(36)=-2.595, p=0.014,~d = 0.678) but not to the controls non-dominant hemisphere (t(36)=1.690, p=0.100,~d = 0.454,~BF10=1.211; corrected alpha = 0.025; uncorrected p-values reported). The tongue showed only a trend for greater surface area coverage in the deprived hand ROI of amputees when compared to controls (t(37)=-2.011, p=0.052, d=−0.650, BF10=1.48; Figure 4C), and tended to be different to zero (t(16)=-1.93, p=0.072, d=−0.467). As tongue remapping is not reflected consistently across analyses, and due to the lack of pre-existing hypotheses, this preliminary result should be interpreted with caution. However, it does indicate that some level of cortical remapping may occur in amputees after limb loss.

Characterisation of tongue (re)mapping in the primary somatosensory cortex.

Distances in the intact hemisphere are plotted in light green and distances in the deprived hemisphere in dark green. For main effects of cortical geodesic distance comparison between amputees and one-handers versus controls, see Figure 4—source data 1–2. (B) * p<0.025 (corrected alpha); *** p<0.001; (C) # p<0.1; * p<0.05; coloured asterisk’s indicate values are significantly different from zero. All other annotations are as in Figure 2.

Figure 4—source data 1

Main effects and interaction for comparison of geodesic distances between amputees and controls for the tongue.

https://cdn.elifesciences.org/articles/76158/elife-76158-fig4-data1-v2.docx
Figure 4—source data 2

Main effects and interaction for comparison of geodesic distances between one-handers and controls for the tongue.

https://cdn.elifesciences.org/articles/76158/elife-76158-fig4-data2-v2.docx
Figure 4—source data 3

Raw data for cortical geodesic distances of the tongue for amputees, controls, and one-handers.

https://cdn.elifesciences.org/articles/76158/elife-76158-fig4-data3-v2.xlsx
Figure 4—source data 4

Raw data for laterality indices of the tongue for amputees, controls, and one-handers.

https://cdn.elifesciences.org/articles/76158/elife-76158-fig4-data4-v2.xlsx

We next explored whether this trend for an increase in tongue activity within the deprived hand ROI, as captured by the laterality index of amputees, was related to PLP (Figure 4C), and found a non-significant difference (U=28.000, p=0.325, rb = 0.152, BF10=0.589). These results suggest, along with an inconclusive Bayes Factor, that amputees with PLP may not report greater instances of tongue remapping, when compared to amputees without PLP.

When repeating the same analysis in one-handers, we also found a significant group x hemisphere interaction (F(1,40)=8.536, p=0.006, n2p=0.176; controlled for brain size volume; Figure 4B) for the cortical distance between the tongue’s CoG and the anatomical landmark. Confirmatory comparisons indicated significantly shorter distances to the anatomical landmark in the deprived hemisphere compared to the intact hemisphere (t(40)=-3.794, p<.001, ~d = −0.677), as well as when compared to the controls’ non-dominant hemisphere (t(40)=-2.380, p=0.022, ~d = −0.751). This was also reflected in greater surface area coverage of the tongue in the deprived hand ROI (see Figure 4A) that was significantly different from zero (W=162.000, p=0.035, rb = 0.543) and from controls (t(41)=-2.534, p=0.015, d=−0.773; Figure 4C). These results suggest that cortical remapping in one-handers extends to tongue movements.

Nose movements produce similar topographic maps across groups

Similar analyses were performed to assess changes in the nose representation (see Appendix 1—figure 1). We did not find evidence for either CoG shifts (p values ≥ 0.829, BF10 ≤0.337) nor differences in surface area coverage (p values ≥ 0.174, BF10 ≤0.664) in both amputees and one-handers compared to controls (see Appendix 1—figure 1). These results suggest that the nose representation remains unaffected in both amputees and one-handers, with conclusive Bayes Factors for one-handers indicating evidence for the null.

Amputees’ topographic maps are more similar to the maps of controls than of one-handers

Finally, we wanted to investigate whether amputees’ facial maps in the deprived hemisphere were more similar to those of controls or to one-handers. To provide a summary measure of univariate facial maps, we performed a Jaccard similarity analysis. This analysis quantifies the degree of similarity (0=no overlap between maps, 1=full overlap) between the map of each amputee and those of each individual in the controls or one-handers group respectively. A linear mixed model was used to compare the complete face map (including each of the facial sub-parts, see Methods). Results showed a significant group x hemisphere interaction (F(1,240.0)=7.70, p=0.006; controlled for age; Figure 5), indicating that amputees’ maps showed different similarity values to controls’ and one-handers’ depending on the hemisphere. Post-hoc comparisons (corrected alpha = 0.025; uncorrected p-values reported) revealed significantly higher similarity to controls’ than to one-handers’ maps in the deprived hemisphere (t(240)=-3.892, p<.001). Amputees’ maps also showed higher similarity to controls’ maps in the deprived relative to the intact hemisphere (t(240)=2.991, p=0.003). Amputees, therefore, displayed greater similarity of facial somatotopy in the deprived hemisphere to controls, suggesting again fewer evidence for cortical remapping in amputees.

Figure 5 with 2 supplements see all
Jaccard similarity analysis comparing the winner-takes-all maps of amputees to the maps of controls and one-handers respectively.

Similarity values indicate greater (towards 1) or reduced (towards 0) similarity between amputees’ winner-takes-all maps (n=17) and those of controls (n=22; pink dots) or one-handers (n=21; purple dots), respectively. Results indicated significantly increased similarity to controls in the deprived hemisphere compared to one-handers’ maps in the deprived hemisphere or to controls’ maps in the intact hemisphere. For main effects and interaction, see Figure 5—source data 1. Means are plotted by the crosses, with standard error of the mean plotted along with individual data points. ** p<0.01, *** p<0.001. For reference, the intervals corresponding to the average intra-group similarities (+/-standard error) are represented by bands in the background (controls in grey, amputees in pink, and one-handers in purple). See Figure 5—figure supplements 12 for follow-up Jaccard analyses.

Figure 5—source data 1

Results from the linear mixed model comparing Jaccard similarity values of amputees’ maps relative to the ones of one-handers and controls respectively.

https://cdn.elifesciences.org/articles/76158/elife-76158-fig5-data1-v2.docx
Figure 5—source data 2

Jaccard similarity values of amputees’ maps relative to the ones of one-handers and controls, respectively.

https://cdn.elifesciences.org/articles/76158/elife-76158-fig5-data2-v2.xlsx

However, it is important to note that the high intra-group similarity observed for controls (e.g. how similar controls’ maps are to other controls’ maps; grey bands in Figure 5) in the deprived hemisphere, could inflate the enhanced similarity to controls observed for amputees (see Figure 5—figure supplement 1 to see all inter- and intra-group comparisons. Related to this we also observed that (i) one-handers also show an enhanced similarity to controls in the deprived hemisphere, and that (ii) even if lower than controls, amputees and one-handers show similar intra-group similarity in the deprived hemisphere). To account for this potential bias, we calculated the similarity between controls’ maps (highly consistent across themselves) relative to amputees and one-handers respectively. Results showed a significant group x hemisphere x face-parts interaction (F(3,315.0)=2.876, p=0.036; controlled for age; see Figure 5—figure supplement 2). Follow-up comparisons (corrected alpha = 0.006; uncorrected p-values reported) revealed that lip-winner maps of controls were significantly more similar to amputees’ than to one-handers’ lip maps in the deprived hemisphere (t(315)=2.854, p=0.005). Conversely, controls’ tongue-winner maps in the deprived hemisphere were significantly more similar to one-handers’ maps than to amputees’ maps (t(315)=-2.883, p=0.004). Finally, controls’ forehead-winner maps in the intact hemisphere were also significantly more similar to one-handers’ maps than to amputees’ maps (t(315)=-3.576, p<.001). Altogether, these results confirm our previous quantification of univariate changes, with greater remapping of the lips in one-handers and if anything, remapping of the tongue in amputees.

Brain decoding in the deprived hand area reveals stable facial representational pattern for amputees, and increased facial information in one-handers

The analyses described above focused on the topographic relationship of the four facial parts, but cortical remapping could potentially manifest subtly, without disrupting the spatial distribution of the face representation. RSA identifies statistical (dis)similarities across activity patterns, providing a more sensitive measure of representational changes (Diedrichsen et al., 2018).

When looking at face-face pairwise dissimilarity in the hand ROI across all three groups, we found a non-significant group x hemisphere x face-face interaction (F(10,627.0)=0.572, p=0.837; controlled for age; Figure 6), suggesting a similar representational structure of the face across hemispheres and groups. However, when we looked at the average amount of facial information within the hand ROI, we did find a significant group x hemisphere interaction (F(2,627.0)=14.544, p<.001), indicating potential differences in facial information content. Post-hoc comparisons (corrected alpha = 0.01; uncorrected p-values reported) exploring this effect reported significantly greater dissimilarity between facial-part representations in the deprived hemisphere of amputees (M=0.214; SE = 0.021; t(627.0)=−4.401, p<.001) and one-handers (M=0.273; SE = 0.018; t(627.0)=−5.668, p<.001), when compared to their respective intact hemisphere (amputees: M=0.152; SE = 0.0205; one-handers: M=0.202; SE = 0.018). When comparing to the controls’ non-dominant hemisphere, we only found significantly greater facial information in the one-handers' deprived hand area (t(72.8)=−3.297, p=0.002) and a non-significant effect for amputees (t(71.7)=−0.828, p=0.411). We note that the effects observed in amputees may be influenced by reduced facial information in the intact hand (M=0.152, SE = 0.021; controls: M=0.207, SE = 0.017). While facial information in the deprived hand area was increased in one-handers compared with amputees, this effect did not survive our correction for multiple comparisons (t(70.7)=−2.117, p=0.038). Similar, though weaker, results were obtained in M1 hand ROI (see Figure 6—figure supplement 1 and Figure 6—source data 2). These results are in line with our univariate analyses, which demonstrate significant cortical remapping of facial parts in the one-handers group. In addition, these results indicate that there may be inter-hemispheric changes in facial information in the intact hand ROI of amputees, although this latter result awaits further confirmation.

Figure 6 with 1 supplement see all
Representational Similarity Analysis (RSA) in the deprived/non-dominant hand area across all groups.

(A) Representational Dissimilarity Matrices (RDMs) for amputees (n=17), controls (n=22), and one-handers (n=21). Greater dissimilarity between activity patterns for the chosen pairwise comparison indicates more information for that facial part within the hand area. Smaller dissimilarity values of facial activity patterns indicate a reduced ability to discriminate between the chosen movements in the hand area. (B) Multi-dimensional scaling plots for each group, which projects the RDM distances into a lower dimensional space. Here, the distances between each marker reflects the dissimilarity, with more similar activity patterns represented closer together, and more distinct activity patterns positioned further away. Forehead movements are plotted in red, with the nose in yellow, lips blue and tongue green, and the standard error is plotted around each data point. Please note, a different scale was used compared to the face ROI (Figure 7). For main effects and interaction for face-face pairwise distances in hand ROI see Figure 6—source data 1. For a similar analysis in M1 see Figure 6—figure supplement 1 and Figure 6—source data 2.

Figure 6—source data 1

Results from the linear mixed model used to explore differences in face-face pairwise distances in the hand ROI for amputees, one-handers, and controls.

https://cdn.elifesciences.org/articles/76158/elife-76158-fig6-data1-v2.docx
Figure 6—source data 2

Results from the linear mixed model used to explore differences in face-face pairwise distances in the M1 hand ROI for amputees, one-handers, and controls.

https://cdn.elifesciences.org/articles/76158/elife-76158-fig6-data2-v2.docx
Figure 6—source data 3

Multivariate distances for face-face pairs in the hand region-of-interest for amputees, one-handers, and controls.

https://cdn.elifesciences.org/articles/76158/elife-76158-fig6-data3-v2.xlsx
Figure 6—source data 4

Multivariate distances for face-face pairs in the hand region-of-interest for amputees, one-handers, and controls in M1.

https://cdn.elifesciences.org/articles/76158/elife-76158-fig6-data4-v2.xlsx
Figure 7 with 1 supplement see all
Representational Similarity Analysis (RSA) in the deprived/non-dominant face area across all groups.

All annotations are as in Figure 6. For main effects and interaction for face-face pairwise distances in face ROI see Figure 7—source data 1. For a similar analysis in M1 see Figure 7—figure supplement 1 and Figure 7—source data 2.

Figure 7—source data 1

Results from the linear mixed model used to explore differences in face-face pairwise distances in the face ROI for amputees, one-handers, and controls.

https://cdn.elifesciences.org/articles/76158/elife-76158-fig7-data1-v2.docx
Figure 7—source data 2

Results from the linear mixed model used to explore differences in face-face pairwise distances in the M1 face ROI for amputees, one-handers, and controls.

https://cdn.elifesciences.org/articles/76158/elife-76158-fig7-data2-v2.docx
Figure 7—source data 3

Multivariate distances for face-face pairs in the face region-of-interest for amputees, one-handers, and controls.

https://cdn.elifesciences.org/articles/76158/elife-76158-fig7-data3-v2.xlsx
Figure 7—source data 4

Multivariate distances for face-face pairs in the face region-of-interest for amputees, one-handers and controls in M1.

https://cdn.elifesciences.org/articles/76158/elife-76158-fig7-data4-v2.xlsx

For completeness, we also looked at facial activity patterns (i.e. face-face pairwise dissimilarities) within the face ROI across all three groups. Here we found non-significant differences for a group x hemisphere x face-face interaction (F(10,627.0)=0.136, p=0.999) and group x hemisphere (F(2,627.0)=0.626, p=0.535), suggesting a similar representational pattern of facial activity, that is facial information content, across hemispheres and groups (Figure 7). See Figure 7—figure supplement 1 and Figure 7—source data 2 for a similar analysis performed in M1 face ROI.

Discussion

It is a well-accepted notion, rooted in non-human primate electrophysiological data, that upper-limb amputation triggers cortical remapping of the assumed neighbour – the lower face – into the missing hand area. This previous work predominantly characterised remapping by investigating shifts of the lip representation (Flor et al., 1995; Birbaumer et al., 1997; Lotze et al., 2001; Grüsser et al., 2001; Foell et al., 2014; Karl et al., 2001; MacIver et al., 2008). However, by focusing on only one face part, activity elicited by other facial parts (such as the forehead) are not taken into account (see Muret and Makin, 2021). Here, we explored the relationship of face-to-hand remapping in controls and one-handed groups, and used both univariate (topographic) and multivariate (representational structure) methods to investigate in detail the information content of the face in both the deprived and intact hand and face areas. We found evidence for an upright somatotopy of the face across all groups, confirming that the cortical neighbour to the hand in humans is the upper, not lower, face. We further found little evidence for remapping of all tested facial parts in amputees, with no significant relationship to the presence of PLP. As a positive control, we also recruited individuals that were born without a hand (one-handers), who have previously shown cortical remapping across multiple body parts (Hahamy et al., 2017; Hahamy and Makin, 2019; Amoruso et al., 2021). Across multiple facial parts (forehead, lips and tongue), one-handers showed evidence for a complex pattern of face remapping in the deprived hand area, with consistent and converging evidence across analysis approaches. Finally, we demonstrate that facial representation in amputees’ deprived hemisphere is more similar to two-handed controls than to one-handers. Together, our findings demonstrate that the face representation in humans is highly plastic, but that this plasticity is restricted by the developmental stage of input deprivation, rather than cortical proximity.

Firstly, our univariate analyses at both group and individual levels provides converging and clear evidence for an upright orientation of the face in controls, amputees and one-handers. Contrary to previous neuroimaging studies reporting an inverted facial somatotopy (Servos et al., 1999; Yang et al., 1993; similar to primates Manger et al., 1996; Jain et al., 2001), or a lack of somatotopic organisation (Iannetti et al., 2003; Nguyen et al., 2004; Kopietz et al., 2009), here we found that the forehead representation borders the hand representation, followed by the nose, the lips, and the tongue located most laterally. These discrepancies may arise from the challenge to find a robust and reliable method to stimulate face parts, and thus elicit detectable cortical activation. In this context, it may be argued that it is difficult to achieve isolated execution of specific facial muscles when performing gross movements without impacting sensory processing of neighbouring facial parts. For instance, tongue movements in our paradigm (e.g. touching the roof of the mouth with the tongue), may be best considered as a holistic inner mouth movement, and forehead movements may be best considered as engaging the upper-face. While this critique is valid, it may also be relevant (although to a smaller degree) for passive paradigms, as stimulation can induce waves that propagate through the skin (Manfredi et al., 2012; Sofia and Jones, 2013; Shao et al., 2016) and Pacinian receptors were found to activate during stimulation of remote sites (Edin and Abbs, 1991; Prsa et al., 2019). Despite this caveat, both our univariate and multivariate analyses showed that we were successful in isolating sensorimotor representations of the various movements (forehead, nose, lips, and tongue) within our regions of interest (see Appendix 1—figure 2 for validation of our approach using vibrotactile stimulation). In other words, even if somatosensory information is overlapping across movements, there is still enough distinct information to separate representational patterns. This finding indicates the suitability of our motor paradigm for teasing apart facial somatotopy, allowing us to characterise the face in greater detail than previously attempted. According to the confirmed upright organisation, if cortical remapping of neighbours exists, we would expect to see the forehead shifting towards and into the hand area – not the lips.

When looking at face-to-hand remapping in amputees, where the remapping of cortical neighbours has been the prevalent explanation for PLP, we find little evidence of shifts of locality and remapping towards the deprived hand area for facial parts, including the neighbour (forehead) and hypothesised neighbour (lips). This was further confirmed by the Jaccard analysis showing that amputees’ maps were more similar to the ones of controls than of one-handers in the deprived hemisphere, as well as similar spatial representation of amputees’ phantom thumb movements relative to controls (Appendix 1—figure 3). Our univariate results are further partially supported by our multivariate analysis, where we find no significant changes in dissimilarity of activity patterns across facial parts in amputees’ deprived hand area relative to controls. These results support previous work reporting minimal cortical remapping after amputation (Makin et al., 2013b; Kikkert et al., 2018; Raffin et al., 2012), suggesting that in amputees this area might be functionally unipotent – pertaining to hand-related activity alone and lacking the ability to reorganise after hand loss. However, due to some inconclusive Bayes Factor in our key analyses, we cannot strongly conclude that remapping does not occur in this group. This could be attributed to our relatively small sample (further recruitment was prevented due to Covid-19 restrictions), and in particular, the small proportion of amputees experiencing PLP (11 out of 17). However, pain is not a necessary condition for deprivation-triggered remapping (Recanzone et al., 1992; Wang et al., 1995; Andoh et al., 2020) and vice versa, PLP has been shown to be experienced in absence of remapping (De Nunzio et al., 2018). Moreover, previous studies reporting significant difference between amputees who experienced PLP and those who do not, often employed similar (or smaller) sample sizes (Lotze et al., 2001), indicating that the expected effect of remapping should be substantial.

Our findings seem contradictory to the many previous studies reporting lip remapping in amputees (Flor et al., 1995; Birbaumer et al., 1997; Lotze et al., 2001; Grüsser et al., 2001; Foell et al., 2014; Karl et al., 2001; MacIver et al., 2008). A major difference with regard to these previous studies, which predominantly focused on a single part of the face, lies in the fact that our study was the first to assess the mapping and potential remapping of multiple facial parts at once. By focusing on the lips only, previous designs excluded other facial parts which may have elicited greater activity in certain areas of S1, resulting in a less accurate delineation of the lip-selective representation. Such down-sampling of body maps, therefore, can lead to biased results and interpretation (Muret and Makin, 2021). While our design is not exempt from this limitation, the fact that we assessed other parts of the face may explain why our results diverged from previous findings.

We did find anecdotal evidence for remapping for the tongue within the deprived hand area in amputees. This was a surprising result, as the tongue is not a cortical neighbour to the hand and was not specifically hypothesised to remap in amputees. We also found that amputees demonstrated a different amount of facial information across the two hand areas. Although this multivariate result was not significantly different to that of controls, it demonstrates the plausibility that an inter-hemisphere imbalance may exist to a certain degree, albeit any relationship to PLP is tenuous. While these latter results require further validation, they support our premise that cortical proximity of representations may not be a necessity for remapping to occur. In this context, as our tongue condition could also be classed as an ‘inner mouth’ movement, it is important to note that previous work addressing sensorimotor representations of the mouth and the larynx have demonstrated both lateral and more medial ‘hotspots’ (i.e. a ‘double’ representation Eichert et al., 2020). The potential tongue remapping in amputees, therefore, may reflect changes in the medial mouth representation, but this would need to be investigated further. Moreover, even if the remapping we observed here goes against the theory of cortical proximity, it can still arise from representational proximity at the subcortical level, in particular at the brainstem level (Florence and Kaas, 1995 ; Kambi et al., 2014). While challenging in humans, mapping both the cuneate and trigeminal nuclei would be critical to provide a more complete picture regarding the role of proximity in remapping.

We did find converging and conclusive evidence for cortical remapping of multiple facial parts, both neighbours (forehead) and non-neighbours (lips and tongue) in our congenital one-handed group. Here, the pattern of remapping is strikingly different to that of cortical neighbourhood theories. Specifically, the location of the cortical neighbour – the forehead – is shown to shift away from the deprived hand area, which is subsequently more activated by the lips and the tongue than is the intact hand area. The increase of facial activity in the deprived hand area is in turn supported by our multivariate results, whereby significantly greater information content for the face was found in the deprived hand area for one-handers when compared to controls. One-handers’ deprived hand area, therefore, seems to have increased discriminability between different facial movements. It is difficult to ascertain from our study the drivers of this remapping. It has been suggested previously that remapping within this group may be driven by functionally-relevant behaviour substituting for the loss of the limb (Hahamy et al., 2017; Amoruso et al., 2021). Together with the recent evidence that lip information content is already significant in the hand area of two-handed participants (Muret et al., 2022), compensatory behaviour since developmental stages might further uncover (and even potentiate) this underlying latent activity. Alternative explanations relate to an overall and unspecific release of inhibition (i.e. decreased GABA) in the missing hand area, allowing for latent activity of other body parts to be detected (Hahamy et al., 2017). While speculative, our results tend to support the former, as we report remapping for facial parts which have the ability to compensate for hand function, for example using the lips and/or mouth to manipulate an object, and a lack of remapping into the hand area for those that cannot (the forehead and nose). This increased activity from body parts compensating for hand function may represent a stabilising mechanism, aimed at preserving the integrity of the sensorimotor network and its function (Muret and Makin, 2021). The deprived hand area in one-handers, therefore, may reflect domain specificity – suitable for adapting to multiple body parts (Amoruso et al., 2021), which may preserve the role of the hand area by sustaining its hand-function related information content.

A limitation that should be acknowledged arises from the potential contribution to S1 from M1 activity. Since these cortical areas are neighbours, it is difficult to separate them with certainty. We minimised the contribution of M1 by taking multiple acquisition and pre-processing steps, including the use of anatomical delineation at the individual level, as well as a comprehensive analytical approach (e.g. both univariate and multivariate techniques). Perhaps most convincingly, our RSA evidence for remapping in congenital one-handers but not in amputees were qualitatively stronger in S1 relative to M1. Furthermore, it has been claimed that active movements may produce different cortical maps to those with passive stimulation (Flor et al., 2013; Andoh et al., 2018), and previous work demonstrating a relationship between cortical remapping and PLP tended to use passive stimulation (Flor et al., 1995; Birbaumer et al., 1997; Foell et al., 2014; Karl et al., 2001). However, we do not think this methodological difference underlies our contrasting results as movement-induced lip activity has been shown to demonstrate lip remapping before (Lotze et al., 2001; MacIver et al., 2008; Striem-Amit et al., 2018), indicating that an active paradigm is suitable for assessing cortical remapping (if it exists). Conversely, a recent study using passive lip stimulation in amputees did not find any evidence for remapping (Philip et al., 2017). Moreover, we recently ran a study which found that S1 topography and multivariate representational structure are similar across active and passive paradigms (Sanders et al., 2019) (see Appendix 1—figure 2 for a comparison between active and passive facial stimulation). In our view, the choice of an active paradigm is the most reflective of naturalistic tactile inputs in everyday life. Together with robust evidence for remapping in one-handers using all the methods tested here, our choice of active paradigm is clearly suitable to identify topographic organisation and remapping, and is practically accessible and translatable to fMRI designs.

To conclude, both our univariate and multivariate analyses found consistent evidence for a complex pattern of face remapping in congenital one-handers, in line with the theory suggesting remapping in this group reflects compensatory behaviour (Muret and Makin, 2021). This is in contrast to amputees, where we find little evidence for cortical remapping, indicating a relative stability of both the hand and face representation after limb loss. By and large, remapping measures were not linked to PLP. Our results call for a reassessment of traditional remapping theories based on cortical proximity, and future research into potential remapping of the inner mouth representation after limb loss.

Materials and methods

Participants

Seventeen individuals with acquired unilateral upper-limb amputation (age; M=53.71, SE = 2.69, women; n=4, missing right hand; n=9), twenty-one individuals with unilateral congenital transverse arrest (age; M=42.67, SE = 3.04, women; n=13, missing right hand; n=8) and twenty-two two-handed controls (age; M=45.55, SE = 2.02, women; n=10, left-handers; n=6) were recruited (see Table 1 for full details). Two additional amputees who were recruited for the study did not participate in the scanning session due to MRI safety concerns, and further recruitment was stalled due to Covid-19 restrictions. The proportion of participants with intact/dominant right hand, as well as gender, were matched across groups (χ2(2)=2.674, p=0.263; χ2(2)=5.593, p=0.061). While significant differences between groups were observed for age (H(2)=7.689, p=.021), post-hoc comparisons confirmed non-significant differences between amputees and one-handers relative to controls. Age covariates were therefore only included in statistical analyses when direct comparisons between amputees and one-handers were carried out. Procedures were in accordance with NHS National Research Ethics Service approval (18/LO/0474), and written informed consent was obtained.

Table 1
Demographic details for amputees (A01-17) and congenital one-handers (C01-21).

Level of limb deficiency is as follows: 1=limb loss above elbow (transhumeral), 2=limb loss below elbow (transradial); L=left, R=right; PLS & PLP frequency: 0=no sensation or pain, 1=once or less per month, 2=several times per month, 3=once a week, 4=daily, 5=all the time. *PLP intensity rating was on average. PLS = phantom limb sensations; PLP = phantom limb pain.

ParticipantsAgeGenderHandedness (prior to amputation for amputees)Affected limbLevel of limb deficiencyYears since amputationPLS intensityPLS frequencyChronic PLSPLP intensityPLP frequencyChronic PLPCause of amputation
AA0160MRR243100510060560Trauma
AA0234MRR13502.514.670*217.5*Trauma
AA0358MRR13390590100120Trauma
AA0459MRL2164018010Trauma
AA0554MAL136100510080440Trauma
AA0647FRL21880440000Electrocution
AA0840FRR11040313.3000Trauma
AA0947MRR25704351045Trauma
AA1053MRL2342000000Trauma
AA1156FLL1129059080580Tumour
AA1266MRR13860560000Trauma
AA1365FLL1109059080440Trauma
AA1466MRL13580220100225Trauma
AA1664MRR1187557565565Trauma
AA1765MRR1870570000Trauma
AA1848MRR1238558565565Trauma
AA1931MRL214305302515Trauma
CA0132FRL2
CA0232FRL2
CA0335MRL2
CA0448MRL2
CA0522FLR2
CA0654FRL2
CA0756FLR2
CA0853MRL1
CA0954FRL2
CA1058MLR2
CA1122MRL2
CA1230FRL2
CA1324MLR2
CA1433FLR2
CA1539FLR2
CA1655FRL2
CA1767FLR2
CA1830FLR2
CA1943MRL2
CA2063MRL2
CA2146FRL2

Phantom sensations rating

Request a detailed protocol

Amputees were asked to rate the frequency of PLP experience within the last year. They also rated the intensity of their worst PLP experience during the last week (or in a typical week involving PLP; 0=no pain, 100=worst pain imaginable). A chronic measure of PLP was calculated by dividing the worst PLP intensity in the last week by PLP frequency (1=all the time, 2=daily, 3=once a week, 4=several times per month, 5=once or less per month). This approach which takes into account the chronic aspect of PLP has been used successfully before (Makin et al., 2015; Makin et al., 2013b; Kikkert et al., 2018; Draganski et al., 2006; Lyu et al., 2016; Kikkert et al., 2016; Kikkert et al., 2017), and has high inter-session reliability (Kikkert et al., 2018). We also asked amputees about the vividness and frequency of non-painful phantom sensations (see Table 1).

Functional MRI sensorimotor task

Request a detailed protocol

We used a facial active motor paradigm, where participants were visually instructed to move their forehead, nose, lips or tongue. This paradigm was chosen because it enabled bilateral activation of S1 simultaneously, allowing us to directly compare activity patterns between the two hemispheres (see Appendix 1—figure 2 for validation of the active paradigm and Discussion for other considerations). Participants were also instructed to move their left and right thumb (amputees were asked to flex/extend their phantom thumb to the best of their ability; one-handers were asked to imagine such movement), resulting in 6 conditions. Since the way one-handers performed the missing thumb condition could not be controlled (e.g., visual or motor imagery), this condition was not considered for analysis. Baseline (i.e. rest) was included as a 7th condition. Specific instructions involved: raising eyebrows (forehead), flaring nostrils (nose), puckering lips (lips), tapping tongue to the roof of the mouth (tongue), flexing and extending (thumb). The protocol comprised of 8 s blocks, with each condition repeated 4 times per run (5 times for baseline), over 3 functional runs. Before entering the scanner, participants practised each movement with the experimenter to ensure that the movement could be executed and to standardise each movement across participants (e.g. specificity and pace). Performance during scanning was visually monitored online with the use of an eye-tracker camera and an experimenter dedicated to this task. Note that multiple participants reported during the experimenter briefing that they could not successfully flare their nostrils, and were therefore instructed to attempt moving their nose in the scanner.

MRI data acquisition

Request a detailed protocol

Functional and anatomical MRI data were obtained using a 3 Tesla Prisma MRI scanner (Siemens, Erlangen, Germany) with a 32-channel head coil. Anatomical data was acquired using a T1-weighted sequence (MPRAGE), with the following parameters: TR = 2530ms; TE = 3.34ms; flip angle = 7°; voxel size = 1 mm isotropic resolution. Functional data based on the blood oxygenation level dependant (BOLD) signal were acquired using a multiband T2*-weighted pulse sequence, with a between-slice acceleration factor of 4 and no in-slice acceleration (TR = 1450ms; TE = 35ms; flip angle = 70°; voxel size = 2 mm isotropic resolution; imaging matrix = 106 x 106; FOV = 212 mm). 72 slices were oriented in the transversal plane. A total of 172 whole-brain volumes for each of the three runs were collected per participant. Field-maps were acquired for field unwarping.

MRI pre-processing

Request a detailed protocol

Functional data was first pre-processed using FSL-FEAT (version 6.00). Pre-processing included motion correction using MCFLIRT (Jenkinson et al., 2002), brain extraction using BET (Smith, 2002), temporal high-pass filtering with a cut-off of 119 s and spatial smoothing using a Gaussian kernel with a FWHM of 3 mm. Field maps were used for distortion correction of functional data.

For each participant, we calculated a midspace between the three functional runs, that is the average space in which the images are minimally reorientated. Each functional run was then aligned to the midspace and registered to structural images (within-subject) using FMRIB’s Linear Image Registration Tool (FLIRT), and optimised using Boundary-Based Registration (Greve and Fischl, 2009). Where specified, functional and structural data were transformed to MNI152 space using FMRIB’s Nonlinear Registration Tool (FNRIT Andersson et al., 2010).

Functional MRI analysis

Request a detailed protocol

Time-series statistical analysis was carried out using FMRIB’s Improved Linear Model (FILM). Task-based statistical parametric maps were computed by applying a voxel-based General Linear Model (GLM), as implemented in FEAT. The design was composed of 6 explanatory variables for each movement, convolved with a double-gamma hemodynamic response function (Friston et al., 1998), and its temporal derivative. The six motion parameters were included as regressors of no interest. Motion outliers (>0.9 mm) of large movements between volumes were included as additional regressors of no interest at the individual level (of total n volumes per group: amputees: 0.36%; controls: 0.36%; one-handers: 0.42%). For our main comparisons 6 contrasts were set up, corresponding to the facial movements’ (forehead, nose, lips, tongue, and left/right thumb) relative to rest.

The estimates from the three functional runs were then averaged voxel-wise using a fixed effects model in participants structural space, with a cluster forming z-threshold of 2.3 and family-wise error corrected cluster significance threshold of p<0.05. Each estimates’ average was masked prior to cluster formation with a sensorimotor mask, defined as the precentral and postcentral gyrus from the Harvard Cortical Atlas. The sensorimotor mask was registered to the individuals structural scan using an inversion of the nonlinear registration by FNIRT. All functional MRI analysis was carried in individual’s native anatomical space.

Regions of Interest (ROI) definition

Request a detailed protocol

Facial topography and remapping were studied using anatomical ROIs for the hand and face areas in S1. Although the primary motor cortex (M1) and S1 are expected to activate during facial movement we primarily focused on S1 remapping due to the traditional focus in the maladaptive plasticity literature on S1 representational shifts (Flor et al., 2006). Furthermore, M1 topography tends to be less well-defined (Schieber, 2001; Graziano and Aflalo, 2007), and so characterisation of typical facial topography may be more apparent in S1. Nevertheless, we wish to note that due to the proximity of S1 to M1, it is possible that marginal contribution from M1 may have affected our S1 activity profiles.

Firstly, S1 was defined on the average surface using probabilistic cytoarchitectonic maps, by selecting nodes for Brodmann Areas (BAs) 1, 2, 3 a, and 3b (Wiestler and Diedrichsen, 2013). The S1 hand ROI (hereafter hand ROI) was defined by selecting the nodes approximately ~1 cm below and ~2.5 cm above the anatomical hand knob. In contrast to earlier work (Wesselink and Ejaz, 2019), we defined a conservative lateral boundary of the hand ROI (~1 cm below the hand knob) to ensure there was limited facial activity captured. From the remaining parts of S1, the medial region was discarded and the lateral region was selected as a first approximation of the S1 face ROI (hereafter face ROI; Figure 8A).

Regions of interest and winner-takes-all analysis in the primary somatosensory cortex for an example participant.

(A) Regions of interest (ROIs) used for univariate analyses are outlined in purple for the hand, and in orange for the face. Shaded areas of each region of interest denote the trimmed ROIs used for multivariate analyses. ROI overlap with the secondary somatosensory cortex (S2) is highlighted in red, and was removed from the face ROI in order to minimise somatotopic contribution from that region. (B) A typical winner-takes-all map from an example participant, with forehead activity in red, nose activity in yelllow, lip activity in blue, and tongue activity in green. The centre-of-gravity for each movement is denoted by a coloured dot outlined in white. The anatomical landmark (used as an anchor for the CoG analysis) is outlined in black, with the midpoint denoted by a grey dot. Cortical geodesic distances were measured from each facial parts CoG to the anatomical landmark midpoint.

Structural T1-weighted images were then used to reconstruct pial and white-grey matter surfaces using Freesurfer (version 7.1.1) at the individual level. The hand and lateral ROIs were then projected into individual brains via the reconstructed individual anatomical surfaces. As the secondary somatosensory cortex (S2) contains a crude somatotopy (Ruben et al., 2001), the lateral ROI was further trimmed in participant’s structural space by removing the overlap with S2. S2 was defined in MNI152 space using the Juelich Histological Atlas (Wiech et al., 2014). The S2 ROI was registered to participants’ structural space using an inversion of the nonlinear registration carried about by FNIRT. The remaining lateral ROI with the overlap from S2 removed was used as the face ROI for all univariate analyses. We note that due to the probabilistic nature of these masks, there could be some marginal contribution from S2 in our estimated face area.

Winner-takes-all approach

Request a detailed protocol

To characterise S1 facial topography, the hand and face ROIs were combined to produce an overall S1 ROI (minus the medial region), and a winner-takes-all approach was used (Figure 8B). For each participant, thresholded z-statistics averaged across the three functional runs were assigned to one of four face parts (forehead, nose, lips, tongue), dependent on which facial movement relatively showed maximal activity within the S1 ROI. Face-winners (i.e. the output of the winner-takes-all) were then projected to the individual’s anatomical surface. Note that we excluded the thumb, which covered ~66% of the deprived hand ROI surface area in amputees and controls (see Appendix 1—figure 3). This allowed us to align our analysis with previous research, and to draw comparisons of facial somatotopy across all groups (one-handers do not have a phantom limb, and therefore we cannot probe the ‘missing’ hand representation directly). All subsequent analyses at the individual’s anatomical surface level were computed using Connectome Workbench (v1.4.2).

Cortical distance analysis

Request a detailed protocol

To assess possible shifts in facial representations towards the hand area, the centre-of-gravity (CoG) of each face-winner map was calculated in each hemisphere. The CoG was weighted by cluster size meaning that in the event of multiple clusters contributing to the calculation of a single CoG for a face-winner map, the voxels in the larger cluster are overweighted relative to those in the smaller clusters. The geodesic cortical distance between each movement’s CoG and a predefined cortical anchor was computed. The cortical anchor was defined as the midpoint of the lateral border of the hand ROI (see Figure 8B). This anatomical landmark was drawn manually for each participant, the midpoint calculated, and both were visually confirmed by a second experimenter. The geodesic distance was assigned a negative value if the movement’s CoG was located below the hand border (i.e. laterally).

Surface area calculation

Request a detailed protocol

To assess possible remapping into the hand area, a secondary winner-takes-all analysis was restricted to the hand ROI only. The surface area coverage (mm2) for each face-winner were computed on the individual anatomical inflated surface. We next calculated the proportion of the hand ROI occupied by each face part by dividing each face-winner’s surface area by the total hand ROI surface area for each individual. From the resulting percentages, we produced a laterality index for each movement with the following formula:

(1) laterality index= (deprivedm-intactm)(deprivedm , intactm)

whereby deprivedm and intactm represent the percentage of surface area coverage for the facial movement m, respectively in the deprived and intact hemisphere. A subsequent laterality index of +1 indicates greater surface area coverage of that movement within the deprived hemisphere (or the hemisphere contralateral to the non-dominant hand in controls), whereas a value of 0 represents an equal balance of surface area coverage across both hemispheres. Note that this approach characterises cortical remapping in relation to the intact hemisphere and has been used in numerous previous studies on amputees (Flor et al., 1995; Birbaumer et al., 1997; Grüsser et al., 2001; Foell et al., 2014). It assumes that the intact hemisphere reflects baseline (i.e., that it is truly ‘intact’), which may not be the case due to inter-hemisphere plasticity and/or homeostatic mechanisms (Muret and Makin, 2021; Valyear et al., 2020; Philip and Frey, 2014) and so also we compared our results to the control group.

Jaccard analysis of similarity

Request a detailed protocol

To quantify the degree of similarity (0=no overlap between maps; 1=full overlap) between winner-takes-all maps across groups, we performed a Jaccard analysis of similarity between amputees’ maps and those of controls and one-handers, respectively. For each face part winner map, the degree of similarity was calculated as follows (illustrated between amputees and controls):

(2) Jaccard similarity= |amputeexm controlym||amputeexm controlym|

whereby amputeexm and controlym represent the winner-takes-all maps of the facial movement m of a given participant amputee x and control y. For each amputee, the similarity across the 22 controls and 21 one-handers respectively was averaged. The same approach was used to compare the other groups. For intra-group similarity, the participant analysed was excluded from the rest of its group to avoid comparing it to itself.

Group-level visualisations

Request a detailed protocol

Prior to group-level visualisations, participant information regarding hand dominance (controls) and deprived hemisphere (one-handed participants) were used to sagittal-flip raw pre-processed data, such that the brain activity corresponding to the non-dominant/missing hand is always represented in the left hemisphere (note that a similar proportion of participants were flipped across groups). Group-level statistical parameter maps were then created with a threshold-free cluster enhancement (TFCE) approach using FSL’s Randomise tool (Winkler et al., 2014). TFCE is a nonparametric, permutation-based method for cluster formation, and has been shown to demonstrate improved sensitivity when compared to typical thresholding methods (Smith and Nichols, 2009). Group activity mixed-effect maps were calculated for each fixed-effect (i.e. averaged across the three functional runs) parameter estimate of a face movement (forehead, nose, lips, tongue) contrasted to baseline. Prior to permutation (n=5000), parameter estimates were masked with a sensorimotor mask, defined as the precentral and postcentral gyrus from the Harvard Cortical Atlas. A family-wise error correction of p<0.05 and variance smoothing of 5 mm (as recommended for datasets with less than 20 participants) were used. Resulting clusters were thresholded at p<0.01 and projected to a group cortical surface (Glasser et al., 2016) using Connectome Workbench (v1.4.2), and activity is visualised in Brodmann Areas 1, 2, 3 a, 3b, and 4.

As well as activity maps, we also visualised the winner-takes-all output at the group-level. Here the ‘winners’ for each face movement within the S1 ROI (the hand and face region combined) in MNI152 space were concatenated into a single volume per group to produce a consistency map for the individual movements (i.e. how many participants maximally activated the same voxel when moving a given facial part). Resulting consistency maps were then projected to a group cortical surface (Glasser et al., 2016) using Connectome Workbench (v1.4.2) for visualisation only.

Multivariate representational analysis

Request a detailed protocol

Representational Similarity Analysis (RSA Nili et al., 2014) was used to assess the multivariate relationship between activity patterns generated by each face part. RSA was conducted in the hand and face ROIs to explore possible remapping across representational features between groups. To ensure the selectivity of the hand and face areas, the ROIs used for univariate analyses were each further trimmed medially by ~1 cm, creating a 1 cm gap between the hand and face ROIs. For each participant, parameter estimates for the four facial movements (forehead, nose, lips and tongue) and the contralateral thumb (for controls and amputees only) were extracted from all voxels within the chosen ROI, as well as residuals from each runs’ first-level analysis (three runs in total). Multidimensional noise normalisation was used to increase reliability of distance estimates (noisier voxels are down-weighted), based on the voxel’s covariance matrix calculated from the GLM residuals. Dissimilarity between resulting facial activity patterns were then measured pairwise using cross-validated Mahalanobis distances (Walther et al., 2016). Due to cross-validation, the expected value of the distance is zero if two patterns are not statistically different from each other. Distances significantly different from zero indicate the two representational patterns are different; negative distances indicate noise. Larger distances for movement pairs therefore suggest greater discriminative ability for the chosen ROI. The resulting six unique inter-facial representational distances (10 unique distances when including the thumb for controls and amputees only) were characterised in a representation dissimilarity matrix (RDM). Multidimensional scaling (MDS) was also used to project the higher-dimensional RDM into lower-dimensional space, whilst preserving inter-facial dissimilarity, for visualisation purposes only. Analysis was conducted on an adapted version of the RSA Toolbox in MATLAB (Nili et al., 2014), customised for FSL (Wesselink and Maimon-Mor, 2018).

Statistical analyses

Request a detailed protocol

All statistical analyses were carried out using JASP (Version 0.14). Outliers were classified as +/-3 standard deviations to the mean. We chose to not remove outliers in our analyses and checked that the significance and direction of the results did not change if identified outliers were removed (see below for number of outliers). When appropriate, univariate analyses was compared using parametric statistics. To assess normality for parametric tests, Shapiro-Wilk tests were run on residuals in combination with inspection of Q-Q plots and reporting of Levene’s Test for Equality of Variances. Where stated, non-parametric test statistics are reported where the assumption of normality has been violated. Analysis of Variance (ANOVA) was used to explore group differences to controls in cortical distances (n=1 outlier for forehead in controls; n=1 outlier for tongue in controls). Each mixed ANOVA had a between-subject factor of Group (Controls x Amputees; Controls x One-handers) and a repeated-measures factor of Hemisphere (Intact/Dominant x Deprived/Non-dominant), and was run separately for each facial movement (forehead, nose, lips and tongue). We controlled for brain size volume when comparing cortical distances between groups. Post-hoc comparisons were conducted with a Bonferroni correction for multiple comparisons (reported corrected alpha and uncorrected p-values in text). Due to issues relating to integration of the correction for brain size in follow-up analyses, both the effect size (noted as ~d) and Bayes factor (noted as ~BF10) in the post-hoc t-tests were not accounting for brain size. If assumptions of normality were violated, the difference between cortical distances in the intact and deprived hemisphere were calculated, and the group difference between one-handed groups and controls was computed using a Mann-Whitney U test. Resulting statistics are reported alongside the mixed ANOVA output. Independent t-tests were used to calculate laterality indices group differences (no outliers identified across groups). We reported the corresponding Bayes Factor (BF10), defined as the relative support for the alternative hypothesis, for non-significant interactions and post-hoc comparisons. While it is generally agreed that it is difficult to establish a cut-off for what consists sufficient evidence, we used the threshold of BF <1/3 as sufficient evidence in support of the null, consistent with others in the field (Wetzels et al., 2011; Dienes, 2014) (though see Kass and Raftery, 1995). The Cauchy prior width was set at 0.707 (JASP’s default). To investigate whether remapping measures were related to PLP, we used a one-tailed Mann-Whitney U tests to compare laterality indices of amputees with (n=11) and without PLP (n=6) for relevant facial parts, under the hypothesis that PLP should result in greater remapping (see Appendix 1—figure 4 for the analogous analysis for the cortical geodesic distances).

For Jaccard analysis of similarity, a linear mixed model (LMM) analysis was used to compare the similarity of amputees’ maps relative to those of controls and one-handers (no outliers), containing fixed factors of group (controls and one-handers), hemisphere (intact, deprived) and facial movement (forehead, nose, lips and tongue). A random effect of participant, as well as covariates of age, were also included in the model. A similar analysis was used to compare the similarity of controls’ maps relative to those of amputees and one-handers respectively (n=1 outlier for the nose in controls). For multivariate analyses, the same analysis was used but with the fixed factors of group (controls, amputees, and one-handers), hemisphere (intact, deprived) and face-face pairs (6 unique representational distances; n=1 outlier in hand and face ROI for controls). All LMM’s were carried out in Jamovi (version 1.6.15) under restricted maximum likelihood (REML) conditions with Satterthwaite adjustment for the degrees of freedom.

Appendix 1

Supplementary Results

Nose movements produce similar topographic maps across groups

We also assessed changes in the nose representation (Appendix 1—figure 1). We did not find evidence for shifts in the nose’s CoG towards the anatomical landmark in amputees when compared to controls (group x hemisphere: F(1,36)=0.047, p=0.829, n2p=0.001, BF10=0.337, controlled for brain size volume; Appendix 1—figure 1B). Similarly, no significant differences in surface area coverage were found in the deprived hand ROI of amputees when compared to controls (t(37)=-1.385, p=0.174, d=−0.447, BF10=0.664; Appendix 1—figure 1C), which however tended to be different from zero (t(16)=-2.09, p=0.052, d=0.506).

Appendix 1—figure 1
Nose remapping in amputees and one-handers in the primary somatosensory cortex.

Distances in the intact hemisphere are plotted in light yellow and distances in the deprived hemisphere in dark yellow. All other annotations are as in Figure 2. # P<.1.

Similarly to amputees, we did not find any evidence for shifts in the nose’s CoG towards the anatomical landmark in one-handers when compared to controls (F(1,40)=0.033, p=0.857, n2p=0.001, BF10=0.294; controlled for brain size volume; Appendix 1—figure 1B), nor differences in surface area coverage compared to controls (U=203.000, p=0.498, rb = 0.121, BF10=0.332; Appendix 1—figure 1C). Taken together, these results suggest that the nose representation remains unaffected in both amputees and one-handers, with conclusive Bayes Factors for one-handers indicating evidence for the null.

Passive validation of our active paradigm

In order to validate our active paradigm, two two-handed controls underwent the active paradigm twice, as well as a passive version of it involving a tactile stimulation (See Supplementary Methods). A significant positive relationship between the two active sessions was found for both participants (see Appendix 1—figure 2B), indicating a relatively stable representational pattern of facial activity in the face region across time. We also find a trend for a positive correlation between the second active session and the passive paradigm in one participant, that is significant in the second participant (see Appendix 1—figure 2B). This indicates a broadly similar representational structure of the face when evoked using both movements and passive stimulation. Note, however, smaller dissimilarity values for the passive paradigm, indicating a reduction in facial information available overall (see Appendix 1—figure 2A).

Appendix 1—figure 2
Comparison of active versus passive stimulation of the face for two control participants in the non-dominant hemisphere.

(A–B) Representational Dissimilarity Matrices (RDMs) for two control participants in the face region of interest (see Methods; Multivariate representational analysis) who completed two active motor paradigm (see Methods; Functional MRI sensorimotor task) sessions ~12 months apart and one passive session (see Supplementary Methods; Validation using passive stimulation). Greater dissimilarity between activity patterns for the pairwise comparison indicates an increased ability to discriminate between the two facial movements/stimulations in the face region, that is there is a greater amount of facial information content. Smaller dissimilarity values indicated a reduced ability to discriminate between the two face parts. (C–D) Pearson’s correlations examining the relationship between face-face and face-thumb dissimilarity values for both the first and second active motor paradigm sessions, and the second active session and passive paradigm.

Appendix 1—figure 3
Phantom and non-dominant thumb representation in the deprived hemisphere of amputees and controls.

Group-level consistency map for the phantom/non-dominant thumb in the hand ROI for amputees (n=17) and controls (n=22). Note the percentage of surface area coverage of the deprived/non-dominant hand ROI was not significantly different for the phantom (M=69.448%; SE = 3.752%) compared to the non-dominant thumb of controls (M=63.989%; SE = 3.594%; U=225.000, p=0.292, d=0.203, BF10=0.579). The colour gradient represents participant agreement for maximally activating that particular voxel, relative to the face movements (winner-takes-all approach). The hand ROI is outlined in purple and central sulcus denoted by the white arrow.

Appendix 1—figure 4
Comparison of cortical distances in the deprived hemisphere of amputees.

Cortical (geodesic) distances were compared between amputees who reported the presence of PLP (n=11; orange) and amputees without PLP (n=6; grey) using a one-tailed Mann-Whitney test. Non-significant differences were found for the lips (t(15)=-0.068, p=0.527, d=−0.035, BF10=0.414), forehead (t(15)=-1.720, p=0.947, d=−0.873, BF10=0.203) and tongue (t(15)=-3.018, p=0.996, d=−1.532, BF10=0.156). Contrary to popular theories of brain plasticity and phantom limb pain (PLP; see Introduction), these results demonstrate that individuals with PLP do not exhibit greater instances of cortical remapping in the deprived hemisphere of the tested facial parts (including both the traditional marker of plasticity – the lips – and the cortical neighbour – the forehead).

Supplementary Methods

Validation using passive stimulation

Procedure

Two two-handed individuals took part in this validation procedure (aged 33 and 29, 2 women, 1 left-handed). In addition to the passive stimulation task described below (hereafter Passive), these individuals underwent two sessions of the sensorimotor active task used in the main analyses (hereafter Active1 and Active2). This allowed us to (i) assess the consistency of our data between the two active sessions and (ii) account for the fact that a different scanner was used for the passive task (and thus one of the active sessions, namely Active2, to be compared to the Passive session).

Functional MRI passive task

Soft pneumatic actuators (SPA-skin Sonar et al., 2021) were placed on the participants’ forehead, nose, lips, tongue on the body midline, and on the thenar eminence of each thumb. Once the actuators had been attached and the participant was inside the scanner bore, a short thresholding procedure was carried out, whereby the pressure (KPa) of the actuators was adjusted to reach above-threshold subjective equality across body parts. The procedure began by defining the pressure for the least sensitive body part – the nose. Here pressure began at 35 KPa and was increased or decreased in step-sizes of 5 KPa. Stimulation length was 8 seconds and participants had 3 seconds to make their response, for example pressing a button to either increase/decrease the pressure or state that the pressure was ‘ideal’ (clearly perceived). The nose was then used as a reference body part, whereby 8 seconds of stimulation was administered to the nose followed by 8 seconds of stimulation to one of the other body parts (forehead, lips, tongue, left and right thumb), that is the target body part. The participant then had 3 seconds to respond via button-press in order to adjust the pressure (increasing or decreasing by 5 KPa) of the target body part in order to generate a similar tactile experience to that of the nose. Throughout thresholding, if the participant responded with ‘ideal’ twice, or if 5 trials had been administered, the resulting pressure was chosen for that body part. During the functional MRI task, each body part was stimulated using the thresholded KPa and a stimulation frequency of 5 Hz. The same block design (with similar time-course) as for the Active sessions was used (i.e., 8 s stimulation blocks, each condition repeated 4 times per run, 5 times for baseline), but over 4 functional runs instead of 3.

Functional MRI data acquisition and analysis

Functional and anatomical MRI data for the Passive and Active2 sessions were obtained with the same MRI parameters as for the Active1 session (collected on the same scanner as the data collected for the main analyses), but on a different 3T Prisma scanner. Functional data was pre-processed and analysed as described for the main analyses.

Data availability

The data generated and analysed during this study is available to the public on Open Science Framework (https://osf.io/xq3am/).

The following data sets were generated
    1. Root V
    2. Muret D
    3. Arribas M
    4. Amoruso E
    5. Thornton J
    6. Tarall-Jozwiak A
    7. Tracey I
    8. Makin TR
    (2021) Open Science Framework
    ID 10.17605/OSF.IO/XQ3AM. Complex pattern of facial remapping in somatosensory cortex following congenital but not acquired hand loss.

References

  1. Software
    1. Andersson JLR
    2. Jenkinson M
    3. Smith S
    (2010)
    Non-linear registration, aka spatial normalization
    FMRIB Technical Report TR07JA2.
    1. Kass RE
    2. Raftery AE
    (1995) Bayes factors
    Journal of the American Statistical Association 90:773–795.
    https://doi.org/10.1080/01621459.1995.10476572
  2. Software
    1. Wesselink DB
    2. Maimon-Mor RO
    (2018)
    RSA toolbox extension for FSL
    RSA.

Decision letter

  1. Olivier Collignon
    Reviewing Editor; Université Catholique de Louvain, Belgium
  2. Christian Büchel
    Senior Editor; University Medical Center Hamburg-Eppendorf, Germany
  3. Moritz Wurm
    Reviewer; University of Trento, Italy
  4. Sliman J Bensmaia
    Reviewer; University of Chicago, United States
  5. Olivier Collignon
    Reviewer; Université Catholique de Louvain, Belgium

Our editorial process produces two outputs: i) public reviews designed to be posted alongside the preprint for the benefit of readers; ii) feedback on the manuscript for the authors, including requests for revisions, shown below. We also include an acceptance summary that explains what the editors found interesting or important about the work.

Decision letter after peer review:

Thank you for submitting your article "Reassessing face topography in primary somatosensory cortex and remapping following hand loss" for consideration by eLife. Your article has been reviewed by 3 peer reviewers, and the evaluation has been overseen by a Reviewing Editor and Christian Büchel as the Senior Editor. The following individuals involved in the review of your submission have agreed to reveal their identity: Moritz Wurm (Reviewer #1); Sliman J Bensmaia (Reviewer #2); Olivier Collignon (Reviewer #3).

The reviewers have discussed their reviews with one another, and the Reviewing Editor has drafted this to help you prepare a revised submission.

Essential revisions:

After consultation with the other reviewers, we believe that this is an interesting and potentially important study but that significant clarifications are needed to fully convince us that the claims are indeed supported by the data. This opinion was reached because ultimately we feel additional analyses are required to address some of the key concerns raised.

All the points raised in the separate reviews should be addressed but here follows what are the key elements that were discussed in the consultation phase.

One concern links to the claim that it was unsolved whether the face representation in S1 is upside-down in humans as it has been shown in non-human primates. To us, it seems there is already ample evidence that an up-right organization of the face would be observed in the human S1. Even if the fMRI data in humans triggered some debates it seems it did not capture sufficient momentum to revisit the original upright face homunculus as described by Penfield itself, since most textbook representations continue to include upright face representations.

The originality of the current study compared to previous work done by the authors themselves should be clarified.

Given that a motor task was used it seems relevant to not only limit analyses to S1 but also include M1; the claim the M1 maps are less clear than those in S1 is not fully convincing.

The authors should provide additional data to support some of the claims made in the paper. In particular, the authors should be coherent in how they treat the nose condition across uni and multivariate analyses. Our feeling is that the nose should be reintroduced in the analytical pipeline and distance/overlap (both uni and multi-variate) should be calculated including the nose. If the authors decide to remove this condition, they should provide empirical support for doing so, and do it for all analytical steps if the reason is that the activity triggered by this condition was unreliable. Similarly, the reason to not include the thumb condition in some analyses and sometimes selectively in one-handers is unclear. This seems actually an interesting condition to fully explore, we guess for the same reasons the authors included the condition in the design. It seems the rationale relates to the fact one-handers have no fantom hand sensation but this was known beforehand, so if this is the reason, why include this condition?

The difference between the winner-takes-all analysis and the group consistency maps should be clarified.

Finally, there are concerns about the claim that one-handers differ from amputees when no direct statistical data support such a claim.

Reviewer #1 (Recommendations for the authors):

Using fMRI-based univariate and multivariate analyses, Root, Muret, et al. investigated the topography of face representation in the somatosensory cortex of typically developed two-handed individuals and individuals with a congenital and acquired missing hand. They provide clear evidence for an upright face topography in the somatosensory cortex in all three groups. Moreover, they find that one-handers, but not amputees, show shorter distances from lip representations to the hand area, suggesting a remapping of the lips. They also find a shift away of the upper face from the deprived hand area in one-handers, and significantly greater dissimilarity between face part representations in amputees and one-handers. The authors argue that this pattern of remapping is different to that of cortical neighborhood theories and points toward a remapping of face parts which have the ability to compensate for hand function, e.g., using the lips/mouth to manipulate an object.

These findings provide interesting insights into the topographic organization of face parts and the principles of cortical (re)organization. The authors use several analytical approaches, including distance measures between hand- and face-part-responsive regions and representational similarity analysis (RSA). Particularly commendable is the rigorous statistical analysis, such as the use of Bayesian comparisons, and careful interpretation of absent group differences.

There are only a few aspects that could be clarified or added:

Only in the discussion does it become clear that an active execution paradigm is used to map the somatotopy of facial parts. This could be clarified earlier, for example at the beginning of the Results section. The use of an execution paradigm seems to be a reasonable choice, which is supported by the robust topographic organization they find and the comparison between active and passive paradigms. However, the topographic organization is not restricted to S1; a similar organization can be found in the adjacent motor cortex. This could be mentioned in the Results, and perhaps a few words on these findings and/or the relationship between somatotopy and mototopy with regard to remapping might be informative.

It is a bit unclear why the nose is not included in the main winner-takes-all analysis (Figure 1) and only shown in the Supplementary Methods. The authors argue in the Methods section that they excluded the nose in the univariate analysis due to weak activation, whereas they include it in the RSA due to the increased sensitivity to subtle changes in activity patterns. This appears a bit inconsistent because also the RSA should suffer from weak voxelwise activations. In Supplementary Figure 1, the activation does not look particularly weak, and the winner-takes-all approach indicates that there are voxels that are most strongly activated for the nose compared to all other facial parts. This could be explained in more detail. Related to this point, did the authors test for nose remapping in amputees and one-handers?

What happened with the thumb movement imagery in the one-handers? Suppl. Figure 2 shows the respective effects for amputees and controls. Did the one-handers fail to show significant effects? It might be interesting to compare the one-handers' imagery effects with the execution effects in amputees and controls, even if putative differences might be less straightforward to interpret (might be due to representational differences or different paradigms).

Related to this point, the face-thumb distance values in Figure 5 are only shown for the controls and amputees. For the reader, the reason is not immediately clear, so perhaps this could be clarified in the figure caption.

There are differences between the forehead clusters found in the winner-takes-all analysis (Figure 1) and the group consistency maps; the latter showing a strong overlap with the hand ROI. For the lips and tongue, the maps appear much more consistent. What could be the reason for this discrepancy, and how can the overlap between hand ROI and forehead be interpreted?

When reporting the weak but significant result of reduced forehead coverage in amputees, the authors could also refer to the BF=1.5, which suggests only anecdotal evidence.

The authors did not remove outliers, which is justified, and the authors checked that the significance and direction of the results did not change if identified outliers were removed. However, could the authors indicate the number of outliers?

Reviewer #2 (Recommendations for the authors):

After amputation, the deafferented limb representation in the somatosensory cortex is activated by stimulation of other body parts. A common belief is that the lower face, including the lips, preferentially "invades" deafferented cortex due to its proximity to cortex. In the present study, this hypothesis is tested by mapping the somatosensory cortex using fMRI as amputees, congenital one-handers, and controls moved their forehead, nose, lips or tongue. First, they found that, unlike its counterpart in monkeys, the representation of the face in the somatosensory cortex is right-side up, with the forehead most medial (and abutting the hand) and the lips most lateral. Second, there was little evidence of "reorganization" of the deafferented cortex in amputees, even when tested with movements across the entire face rather than only the lips. Third, congenital one-handers showed significant reorganization of deafferented cortex, characterized principally by the invasion of the lower face, in contrast to predictions from the hypothesis that proximity was the driving factor. Fourth, there was no relationship between phantom limb pain reports and reorganization.

As a non-expert in fMRI, I cannot evaluate the methodology. That being said, I am not convinced that the current consensus is that the representation of the face in humans is flipped compared to that of monkeys. Indeed, the overwhelming majority of somatosensory homunculi I have seen for humans has the face right side up. My sense is that the fMRI studies that found an inverted (monkey-like) face representation contradict the consensus. Similarly, it is not clear to me how the observations (1) of limited reorganization in amputees, (2) of significant reorganization in congenital one-handers, and (3) of the lack of relationship between PLP and reorganization is novel given the previous work by this group. Perhaps the authors could more clearly articulate the novelty of these results compared to their previous findings. Finally, Jon Kaas and colleagues (notably Niraj Jain) have provided evidence in experiments with monkeys that much of the observed reorganization in the somatosensory cortex is inherited from plasticity in the brain stem. Jain did not find an increased propensity for axons to cross the septum between face and hand representations after (simulated) amputation. From this perspective, the relevant proximity would be that of the cuneate and trigeminal nuclei and it would be critical to map out the somatotopic organization of the trigeminal and cuneate nuclei to test hypotheses about the role of proximity in this remapping.

A few comments:

Which cortical fields do the authors refer to as primary somatosensory cortex? Brodmann's areas 3a, 3b, 1 and 2?

On line 322, I suggest using "completeness" rather than "completion."

On line 372, I suggest using "critique" rather than "limitation."

Reviewer #3 (Recommendations for the authors):

In their study, the authors set up to challenge the long-held claim that cortical remapping in the somatosensory cortex in hand deprived cortical territories follows somatotopic proximity (the hand region gets invaded by cortical neighbors) as classically assumed. In contrast to this claim, the authors suggest that remapping may not follow cortical proximity but instead functional rules as to how the effector is used. Their data indeed suggest that the deprived hand area is not invaded by the forefront which is the cortical neighbor but instead by the lips which may compensate for hand loss in manipulating objects. Interestingly the authors suggest this is mostly the case for one-handers but not in amputees for who the reorganization seems more limited in general (but see my comments below on this last point).

This is a remarkably ambitious study that has been skilfully executed on a strong number of participants in each group. The complementarity of state-of-the-art uni- and multi-variate analyses are in the service of the research question, and the paper is clearly written. The main contribution of this paper, relative to previous studies including those of the same group, resides in the mapping of multiple face parts all at once in the three groups.

In the winner takes all approach, the authors only include 3 face parts but exclude from the analyses the nose and the thumb. I am not fully convinced by the rationale for not including nose in univariate analyses – because it does not trigger reliable activity – while keeping it for representational similarity analyses. I think it would be better to include the nose in all analyses or demonstrate this condition is indeed "noisy" and then remove it from all the analyses. Indeed, if the activity triggered by nose movement is unreliable, it should also affect multivariate.

The rationale for not including the hand is maybe more convincing as it seems to induce activity in both controls and amputees but not in one-handers. First, it would be great to visualize this effect, at least as supplemental material to support the decision. Then, this brings the interesting possibility that enhanced invasion of hand territory by lips in one-handers might link to the possibility to observe hand-related activity in the presupposed hand region in this population. Maybe the authors may consider linking these.

The use of the geodesic distance between the center of gravity in the Winner Take All (WTA) maps between each movement and a predefined cortical anchor is clever. More details about how the Center Of Gravity (COG) was computed on spatially disparate regions might deserve more explanations, however. Moreover, imagine that for some reason the forefront region extends both dorsally and ventrally in a specific population (eg amputees), the COG would stay unaffected but the overlap between hand and forefront would increase. The analyses on the surface area within hand ROI for lips and forehead nicely complement the WTA analyses and suggest higher overlap for lips and lower overlap for forehead but none of the maps or graphs presented clearly show those results – maybe the authors could consider adding a figure clearly highlighting that there is indeed more lip activity IN the hand region.

In addition to overlap analyses between hand and other body parts, the authors may also want to consider doing some Jaccard similarity analyses between the maps of the 3 groups to support the idea that amputees are more alike controls than one-handers in their topographic activity, which again does not appear clear from the figures.

This brings to another concern I have related to the claim that the change in the cortical organization they observe is mostly observed in one-handers. It seems that most of this conclusion relies on the fact that some effects are observed in one-handers but not in amputees when compared to controls, however, no direct comparisons are done between amputees and one-handers so we may be in an erroneous inference about the interaction when this is actually not tested (Nieuwenhuis, 11). For instance, the shift away from the hand/face border of the forehead is also (mildly) significant in amputees (as observed more strongly in one-handers) so the conclusion (eg from the subtitle of the Results section) that it is specific to one-hander might not fully be supported by the data. Similar to the invasion of the hand territory from the lips which is significant in amputees in terms of surface area. All together this calls for toning down the idea that plasticity is restricted to congenital deprivation (eg last sentence of the abstract). Even if numerically stronger, if I am not wrong, there are no stats showing remapping is indeed stronger in one-handers than in amputees and actually, amputees show significant effects when compared to controls along the lines as those shown (even if more strongly) in one-handers. Also, maybe the authors could explore whether there is actually a link between the number of years without hand and the remapping effects.

One hypothesis generated by the data is that lips remap in the deprived hand area because lips serve compensatory functions. Actually, also in controls, lips and hands can be used to manipulate objects, in contrast to the forehead. One may thus wonder if the preferential presence of lips in the hand region is not latent even in controls as they both link in functions?

1) The authors should re-assess their conclusion that remapping is mostly present in one-handers than amputees in absence of a convincing statistical demonstration that it is actually the case. The amputees show similar effects or trends for similar remapping as those observed in one-handers. maybe one source of this lower reliability in the effects of amputees links to the lower number of participants and their higher variability in ethology (eg age of amputation etc…). Maybe the authors could explore whether there is actually a link between the number of years without hand and the remapping effects.

2) The authors should be coherent in the body part they include for their univariate and multivariate analyses. I let them decide whether they think the nose condition was too noisy (eg because people could not follow instructions to move the nose) or not. If they think the condition however did not elicit reliable activity, this should affect both uni and multivariate analyses.

3) I recommend the authors show the map elicited by hand movement in the three groups, support their claim that this is only reliably observed in controls and amputees but not one-handers; and maybe discuss (or even better do some analyses on that, eg correlation) the possibility that the preservation of the hand map is potentially a factor influencing remapping (eg do the amputees with the most salient remaining hand maps are those showing less remapping?).

4) The authors should clarify in their figure where do they observe that the lip increases activity WITHIN the hand region in one-handers.

5) The fact that the univariate results of Figure 1 do not follow well those presented in the WTA maps deserves some consideration. Indeed, what seems striking in Figure 1 is that the forehead activity extends more dorsally in one-handers, approaching the hand region, which contradicts conclusions from WTA. Also, it might be informative to have an outline of the brain mask (in all figures) of the S1 ROI used for constraining spatially the univariate analyses.

6) In Figure 5 the thumb is not represented in one-handers. It would be interesting for the reader to see that indeed the instruction to move the thumb did not elicit a reliable activity map and what was the criteria to selectively remove it in one-handers aside from the fact they don't experience fantom sensation. It is also unclear why the thumb is not displayed in the MDS while it's computed in the RDM of controls and amputees.

7) Maybe discuss further the potential (latent) link between hand and lips as they both can be used to manipulate objects even in controls.

https://doi.org/10.7554/eLife.76158.sa1

Author response

Essential revisions:

After consultation with the other reviewers, we believe that this is an interesting and potentially important study but that significant clarifications are needed to fully convince us that the claims are indeed supported by the data. This opinion was reached because ultimately we feel additional analyses are required to address some of the key concerns raised.

All the points raised in the separate reviews should be addressed but here follows what are the key elements that were discussed in the consultation phase.

One concern links to the claim that it was unsolved whether the face representation in S1 is upside-down in humans as it has been shown in non-human primates. To us, it seems there is already ample evidence that an up-right organization of the face would be observed in the human S1. Even if the fMRI data in humans triggered some debates it seems it did not capture sufficient momentum to revisit the original upright face homunculus as described by Penfield itself, since most textbook representations continue to include upright face representations.

We agree with the reviewers that the human’s neuroimaging data (not just using fMRI but also MEG) did not trigger an overwriting of the original findings from Penfield. But we think this reflects the disproportional impact of Penfield’s cartooning of the homunculus, despite the scarcity and inherent limitations of the data supporting his work and the (over-)simplified depiction of the Homunculus (as we detail below). We believe that the silent consensus is that sensorimotor body representation is more nuanced then revealed by this textbook cartoon, and as such, the idea that the hand area is neighbouring the lower face in humans have been strongly embedded in contemporary research in particular relating to brain reorganisation. As an illustrating example, the first paper to describe the lower face (lips and cheek) shifting towards the hand area (Flor et al., 1995), has been cited over 2,000 times. This is one of many related examples, and when put together there is no doubt that this misrepresentation is highly popular and has influenced not only development of treatment for phantom limb pain, but also other related condition, such as complex regional pain syndrome (Maihöfner et al., 2004, 2006), dystonia (Burman et al., 2009; Elbert et al., 1998) and spinal cord injury (Wrigley et al., 2009). This empirical ambiguity is due to the fact that the vast majority of human studies is focused on lip representation, combined with the methodological difficulties of stimulating the face in an MRI environment. Consequently, our study is among the first to replicate Penfield’s observations with converging neuroimaging analyses in individual participants.

Having said all that, to appease the reviewers, we have re-written the introduction to reflect the inclusive neuroimaging results, and nuanced our claim throughout the manuscript (starting with the title and abstract) to reflect our findings confirm the original observations made by Penfield and colleagues.

The originality of the current study compared to previous work done by the authors themselves should be clarified.

We are happy to clarify on the conceptual, methodological and empirical innovation our study provides. In short:

  1. Conceptually, it is crucial for us to understand if deprivation-triggered plasticity is constrained by the local neighbourhood, because this can give us clues regarding the mechanisms driving the remapping. We provide strong topographic evidence about the face orientation in controls, amputees and one-handers.

  2. The vast majority of previous research on brain plasticity following hand loss (both congenital and acquired) in humans has exclusively focused on the lower face, and lips in particular. We provide systematic evidence for stable organisation and remapping of the neighbouring upper face, as well as the lower face. We also study topographic representation of the tongue (and nose) for the first time.

  3. The vast majority of previous research on brain remapping following hand loss (both congenital and acquired, neuroimaging and electrophysiological) was focused on univariate activity measures, such as the spatial spread of units showing a similar feature preference, or the average activity level across individual units. We are going beyond remapping by using RSA, which allows us to ask not only if new information is available in the deprived cortex (as well as the native face area), but also whether this new information is structured consistently across individuals and groups. We show that representational content is enhanced in the deprived cortex one-handers whereas it is stable in amputees relative to controls (and to their intact hand region).

  4. Based on previous studies, the assumption was that reorganisation in congenital one-handers was relatively unspecific, affecting all tested body parts. Here, we provide evidence for a more complex pattern of remapping, with the forehead representation seemingly moving out of the missing hand region (and the nose representation being tentatively similar to controls). That is, we show not just “invasion” but also a shift of the neighbour away from the hand area which has never been documented (or in fact suggested).

  5. Using Bayesian analyses we provide definitive evidence against a relationship between PLP and forehead remapping, providing first and conclusive evidence against the remapping hypothesis, based on cortical neighbourhood.

Given that a motor task was used it seems relevant to not only limit analyses to S1 but also include M1; the claim the M1 maps are less clear than those in S1 is not fully convincing.

To explore M1 face representation, without having to make assumption about its underlying topography, we conducted Representational Similarity Analysis (RSA) in M1 (BA4) hand and face ROI’s (see resulting statistics below). RSA allows us to quantify and characterise functional organisation beyond the spatial attributes of selectivity maps. As detailed below and summarised in the new supplementary figures (Figure 6 and 7 —figure supplement 1), we find very similar representational structure for both face organisation and “reorganisation” across groups to what we reported originally in S1. The statistical evidence for remapping in the hand ROI were similar, though qualitatively stronger in S1 (i.e., significant difference between one-handers and two-handed controls only observed in S1).

The authors should provide additional data to support some of the claims made in the paper. In particular, the authors should be coherent in how they treat the nose condition across uni and multivariate analyses. Our feeling is that the nose should be reintroduced in the analytical pipeline and distance/overlap (both uni and multi-variate) should be calculated including the nose. If the authors decide to remove this condition, they should provide empirical support for doing so, and do it for all analytical steps if the reason is that the activity triggered by this condition was unreliable.

Following this comment, we re-ran all univariate analyses to include the nose, and updated throughout the main text and supplemental results and related figures. In short, adding the nose did not change the univariate results, apart from a now significant group x hemisphere interaction for the CoG of the tongue when comparing amputees and controls, better matching the trends for greater surface coverage in the deprived hand ROI of amputees. As such, we did not need to update our main interpretation based on this (extensive) reanalysis.

Similarly, the reason to not include the thumb condition in some analyses and sometimes selectively in one-handers is unclear. This seems actually an interesting condition to fully explore, we guess for the same reasons the authors included the condition in the design. It seems the rationale relates to the fact one-handers have no fantom hand sensation but this was known beforehand, so if this is the reason, why include this condition?

We did not intent the thumb condition in one-handers for analysis, as the task given to one-handers (imagine moving a body part you never had before) is inherently different to that given to the other groups (move – or at least attempt to move – your (phantom) hand). This condition was included solely to fill the experimental gap so that the protocols were matched as closely as possible across groups, as per our general practice (Makin et al., 2013 Nat. Comm.; Makin et al., 2013 eLife; Hahamy et al., 2017). As we note in our original manuscript, activity elicited by thumb movements dominates most of the hand area (66%) in both amputees and controls. To reduce the discrepancy in our analyses across all three groups, we cannot include this condition. For this reason, we decided to remove the hand-face dissimilarity analysis which we included in our original manuscript. Upon reflection we agreed that this specific analysis does not directly relate to the question of remapping (but rather of shared representation).

The difference between the winner-takes-all analysis and the group consistency maps should be clarified.

The individual-participant winner-takes-all maps are minimally thresholded, and thus produce an inherently different spatial distribution relative to the group contrast maps presented in Figure 1. Note that Figure 1 does not involve a winner-takes-all procedure, but rather random effect group contrasts (for each condition versus baseline separately). We made sure to clarify these differences.

Finally, there are concerns about the claim that one-handers differ from amputees when no direct statistical data support such a claim.

We are particularly grateful for this comment, we now provide clear evidence to demonstrate that amputees topographic face mapping is more similar to controls than to one-handers.

Reviewer #1 (Recommendations for the authors):

Using fMRI-based univariate and multivariate analyses, Root, Muret, et al. investigated the topography of face representation in the somatosensory cortex of typically developed two-handed individuals and individuals with a congenital and acquired missing hand. They provide clear evidence for an upright face topography in the somatosensory cortex in all three groups. Moreover, they find that one-handers, but not amputees, show shorter distances from lip representations to the hand area, suggesting a remapping of the lips. They also find a shift away of the upper face from the deprived hand area in one-handers, and significantly greater dissimilarity between face part representations in amputees and one-handers. The authors argue that this pattern of remapping is different to that of cortical neighborhood theories and points toward a remapping of face parts which have the ability to compensate for hand function, e.g., using the lips/mouth to manipulate an object.

These findings provide interesting insights into the topographic organization of face parts and the principles of cortical (re)organization. The authors use several analytical approaches, including distance measures between hand- and face-part-responsive regions and representational similarity analysis (RSA). Particularly commendable is the rigorous statistical analysis, such as the use of Bayesian comparisons, and careful interpretation of absent group differences.

We thank the reviewer for their positive and constructive feedback.

There are only a few aspects that could be clarified or added:

Only in the discussion does it become clear that an active execution paradigm is used to map the somatotopy of facial parts. This could be clarified earlier, for example at the beginning of the Results section.

Thank you for pointing this out. We agree with the reviewer’s recommendation and have added the detail regarding active somatotopic mapping at the beginning of the Results section (line 154), as advised, and at the end of the Introduction (line 126).

The use of an execution paradigm seems to be a reasonable choice, which is supported by the robust topographic organization they find and the comparison between active and passive paradigms. However, the topographic organization is not restricted to S1; a similar organization can be found in the adjacent motor cortex. This could be mentioned in the Results, and perhaps a few words on these findings and/or the relationship between somatotopy and mototopy with regard to remapping might be informative.

We agree with the reviewer that the primary motor cortex (M1) is indeed relevant and informative for our results. To explore M1 face representation, we conducted Representational Similarity Analysis (RSA) in M1 (BA4) hand and face ROI’s (see resulting statistics below). RSA allows us to quantify and characterise functional organisation beyond the spatial attributes of selectivity maps. Since M1 is considered to be more broadly topographically organised relative to S1, we believe this approach is the most appropriate to compare the results across these two brain areas. As detailed below and summarised in the new supplementary figures (Figure 6 and 7 —figure supplement 1), we find very similar representational structure for both face organisation and “reorganisation” across groups to what we reported originally in S1. Interestingly, the statistical evidence for remapping in the hand ROI were similar, though qualitatively stronger in S1 (i.e., significant difference between one-handers and two-handed controls only observed in S1).

When looking at face-face pairwise dissimilarity in the M1 hand ROI (Figure 6 —figure supplement 1), we found a non-significant three-way interaction of Group (Amputees x Controls x One-handers) x Hemisphere (Intact x Deprived) x Face-Face pair (including the nose) (F(10,627.0)=0.398, p=0.948; controlled for age) and a significant Group x Hemisphere interaction (F(2,627.0)=7.553, p<.001). Post-hoc comparisons (corrected α=0.0125; uncorrected p-values reported) indicated a similar pattern of results to S1, with significantly greater dissimilarity between facial-part representations in the deprived hemisphere of amputees (M=0.214; SE=0.0193 ; t(627.0)=3.9633, p<.001) and one-handers (M=0.243; SE=0.0170; t(627.0)=3.8525, p<.001), when compared to their respective intact hemisphere (amputees: M=0.157; SE=0.0193; one-handers: M=0.193; SE=0.0170; corrected α=0.0125; uncorrected p-values reported). When comparing to controls non-dominant hemisphere (M=0.192; SE=0.0163), we did not find any significant differences in both one-handers (t(77.4)=2.1656, p=0.033) and amputees (t(75.9)=0.8371, p=0.405).These results are in line with both our multivariate and univariate results in S1, which demonstrate extensive cortical remapping of facial parts in our one-handers group, and tentatively suggests again inter-hemispheric changes in the intact hand ROI of amputees (amputees: M=0.157; SE=0.0193; controls: M=0.201; SE=0.0163; t(75.9)=-1.7382, p=0.086).

When looking at face-face pairwise dissimilarity in the M1 face ROI we note minor differences to our S1 multivariate results. Here, despite a non-significant Group x Hemisphere x Face-Face interaction (F(10,627.0)=0.2969, p=0.982; controlled for age; see Figure 7 —figure supplement 1), we found a significant Group x Hemisphere interaction (F(2,627.0)=4.3429, p=0.013), arising from lower facial information content in the intact (M=0.557; SE=0.0305) compared to the deprived (M=0.597; SE=0.0305) face ROI in amputees, but this difference did not survive our correction for multiple comparisons (t(627.0)=2.22199, p=0.027; corrected α=0.0125; trend defined as p<0.025; uncorrected p-values reported). Between-hemisphere differences were non-significant when looking at the one-hander group (deprived: M=0.536; SE=0.0268; intact: M=0.567; SE=0.0268; t(627.0)=-1.93153, p=0.054). Non-significant differences were also found when comparing the deprived face ROI in amputees (t(67.2)=-0.00525, p=0.996) and one-handers (t(68.0)=-1.64705, p=0.104) to the controls non-dominant hemisphere (M=0.597; SE=0.0257). These results therefore tentatively suggest that the reported interaction is likely driven by decreased facial information content in the intact face ROI when compared to the deprived face region in amputees (although note no significant differences were reported when comparing the intact face ROI in amputees to controls (amputees: M=0.557; SE=0.0305; controls: M=0.593; SE=0.0257; t(67.2)=-0.90806, p=0.367)), suggesting again the presence of inter-hemispheric plasticity within this group (akin to results in S1). We also surprisingly found a significant Group x Face-Face (F(10,627.0)=0.1934, p=0.038) interaction when looking at face-face pairwise distances in the M1 face ROI, suggesting that the information content for each movement (regardless of hemisphere) differed across groups. This interaction arose from significantly smaller forehead-lips distances for one-handers compared to controls (t(125)=-2.625, p=0.010), but this effect did not survive correction for multiple comparisons (corrected α=0.004).

These additional analyses are specified in lines 426-427, 442-444, in the legends of Figure 6 and Figure 7 of the main text, with their respective Figure supplements 1 and data source 2, and in the Discussion (line 574-576).

It is a bit unclear why the nose is not included in the main winner-takes-all analysis (Figure 1) and only shown in the Supplementary Methods. The authors argue in the Methods section that they excluded the nose in the univariate analysis due to weak activation, whereas they include it in the RSA due to the increased sensitivity to subtle changes in activity patterns. This appears a bit inconsistent because also the RSA should suffer from weak voxelwise activations. In Supplementary Figure 1, the activation does not look particularly weak, and the winner-takes-all approach indicates that there are voxels that are most strongly activated for the nose compared to all other facial parts. This could be explained in more detail. Related to this point, did the authors test for nose remapping in amputees and one-handers?

Following this comment, we have re-ran all univariate analyses to include the nose condition, and updated all results throughout the main text, supplemental results and related figures. We initially chose to exclude the nose from univariate analyses because many participants could not move it. RSA is less sensitive to noise (due to cross-validation) and to selectivity/topography (two issues we had with the nose condition) and for these reasons we initially decided to keep the nose in the RSA analysis and exclude it from the univariate analyses. But we agree that being consistent avoids unnecessary confusion, and adding the nose condition into all analyses did not change the univariate results, apart from a now significant group x hemisphere interaction for the CoG of the tongue when comparing amputees and controls, better matching the trends for greater surface coverage in the deprived hand ROI of amputees.

We also performed and added in the Supplementary section the univariate results for the nose (Appendix Figure 1). We only shortly mention this result in the main text (lines 324-330) since we feel like the Results section is already pretty dense with results, and this additional result does not change our interpretation and conclusions.

What happened with the thumb movement imagery in the one-handers? Suppl. Figure 2 shows the respective effects for amputees and controls. Did the one-handers fail to show significant effects? It might be interesting to compare the one-handers' imagery effects with the execution effects in amputees and controls, even if putative differences might be less straightforward to interpret (might be due to representational differences or different paradigms).

In short – we cannot answer this question. To clarify, this condition was present solely to fill the experimental gap so that the protocols were matched as closely as possible across groups and we never intended to analyse this condition in one-handers. While we agree it could have been interesting to study how one-handers relate to their missing hand, there are different kinds of motor imagery (e.g., visual, kinaesthetic) and without dedicated research tools it would have been tricky for us to monitor individual one-handers strategies for imagining movement of a limb they never had. Consequently, we gave only broad instructions for this condition, that one-handers could interpret in various ways (visual, motor imagery or even prosthesis movements). It thus does not make much sense for us to look at it, and will not be meaningful to compare to the amputees group where we very explicitly instructed participants to move their phantom hand. Appendix Figure 3 addresses the question of persistent representation in amputees, which is highly relevant for this paper. To reduce the discrepancy, we removed the analysis including the phantom hand (see next comment). And we added this explanation in the methods (lines 645-647).

Related to this point, the face-thumb distance values in Figure 5 are only shown for the controls and amputees. For the reader, the reason is not immediately clear, so perhaps this could be clarified in the figure caption.

Upon reflection (and thanks to the reviewers’ comments), we decided to remove entirely the hand-face dissimilarity analysis, which does not directly relate to the question of remapping (but rather of shared representation), in addition to making the paper unbalanced (as explained above, we were unable to run this analysis in one-handers). We will now feature this analysis in another paper that appears more appropriate in the context of referred sensations in amputees (Amoruso et al., 2022 MedRxiv).

There are differences between the forehead clusters found in the winner-takes-all analysis (Figure 1) and the group consistency maps; the latter showing a strong overlap with the hand ROI. For the lips and tongue, the maps appear much more consistent. What could be the reason for this discrepancy, and how can the overlap between hand ROI and forehead be interpreted?

We apologise for any confusion. Figure 1 represents the thresholded group-level maps (vs rest, not winner-takes-all) within our initial (and very broad) sensorimotor mask, whereas the winner-takes-all inter-subject consistency maps are displayed in the following figures (2 to 4). The strong overlap with the hand ROI observed for the forehead is due to the use of an ROI including both the face and hand regions, where hand activity (not available for all groups) was not included. The logic behind this choice was that including the hand ROI would ease the detection of remapping (into the hand ROI). Given the winner-takes-all procedure (which included minimal thresholding), and the fact that the forehead is the immediate neighbour of the hand area, the slightly stronger forehead activity dominated the hand area. But when more stringent thresholding is introduced (as in Figure 1), the forehead activity is insufficient to pass the threshold. To add clarity to the activity presented in Figure 1, we have visualised the pre-thresholding mask of the pre- and post-central gyrus (defined using the Harvard Cortical Atlas in dark grey; see line 172), as well as the hand and face ROIs used in the subsequent winner-takes-all analysis, and amended Figure legends to state that the group-level activity visualised is versus rest (line 167) and winner-takes-all maps done within the combined hand+face ROI (line 199-200). We have also updated the figure legend of Figure 2 (lines 202-204) to explicitly acknowledge these differences: “Please note that the individual-participant winner-takes-all maps are minimally thresholded, and thus produce an inherently different spatial distribution relative to the group contrast maps presented in Figure 1”.

When reporting the weak but significant result of reduced forehead coverage in amputees, the authors could also refer to the BF=1.5, which suggests only anecdotal evidence.

We agree with the reviewer’s above recommendation, however we initially decided to provide Bayes Factors only in the absence of effects, to show whether we had enough evidence in support for the null hypothesis (see lines 877-879: “We reported the corresponding Bayes Factor (BF10), defined as the relative support for the alternative hypothesis, for non-significant interactions and post-hoc comparisons.”).

The authors did not remove outliers, which is justified, and the authors checked that the significance and direction of the results did not change if identified outliers were removed. However, could the authors indicate the number of outliers?

We have added the number of outliers to the ‘Statistical Analyses’ section (lines 863-864, 877, 889, 894, 896) of the Materials and methods. As detailed, there were very few outliers (up to one per condition/group).

Reviewer #2 (Recommendations for the authors):

After amputation, the deafferented limb representation in the somatosensory cortex is activated by stimulation of other body parts. A common belief is that the lower face, including the lips, preferentially "invades" deafferented cortex due to its proximity to cortex. In the present study, this hypothesis is tested by mapping the somatosensory cortex using fMRI as amputees, congenital one-handers, and controls moved their forehead, nose, lips or tongue. First, they found that, unlike its counterpart in monkeys, the representation of the face in the somatosensory cortex is right-side up, with the forehead most medial (and abutting the hand) and the lips most lateral. Second, there was little evidence of "reorganization" of the deafferented cortex in amputees, even when tested with movements across the entire face rather than only the lips. Third, congenital one-handers showed significant reorganization of deafferented cortex, characterized principally by the invasion of the lower face, in contrast to predictions from the hypothesis that proximity was the driving factor. Fourth, there was no relationship between phantom limb pain reports and reorganization.

As a non-expert in fMRI, I cannot evaluate the methodology. That being said, I am not convinced that the current consensus is that the representation of the face in humans is flipped compared to that of monkeys. Indeed, the overwhelming majority of somatosensory homunculi I have seen for humans has the face right side up. My sense is that the fMRI studies that found an inverted (monkey-like) face representation contradict the consensus.

Thank you for point this out. As we tried to emphasise in the introduction, very few neuroimaging studies actually investigated face somatotopy in humans, with inconsistent results. We agree the default consensus tends to be dominated by the up-right depiction of Penfield’s homunculus (recently replicated by Roux et al., 2018). However, due to methodological and practical constraints, alignment across subjects in the case of intracortical recordings is usually difficult to achieve, and thus makes it difficult to assess the consistency in topographical organisation. Moreover, previous imaging studies did not manage to convincingly support Penfield’s homunculus. For these two key reasons, the spatial orientation of the human facial homunculus is still debated. A further limiting factor of previous studies in humans is that the vast majority of human studies investigating face (re)mapping in humans focused solely on the lip representation, using the cortical proximity hypothesis to interpret their results. Consequently, as we highlight above in our response to the Editor, there is a wide-spread and false representation in the human literature of the lips neighbouring the hand area.

To account for the reviewer’s critic and convey some of this context, we changed our title from: Reassessing face topography in primary somatosensory cortex and remapping following hand loss; to: Complex pattern of facial remapping in somatosensory cortex following congenital but not acquired hand loss. This was done to de-emphasise the novelty of face topography relative to our other findings.

We also rewrote our introduction (lines 79-94) as follows:

“The research focus on lip cortical remapping in amputees is based on the assumption that the lips neighbour the hand representation. However, this assumption goes against the classical upright orientation of the face in S126–30, as first depicted in Penfield’s Homunculus and in later intracortical recordings and stimulation studies26–29, with the upper-face (i.e., forehead) bordering the hand area. In contrast, neuroimaging studies in humans studying face topography provided contradictory evidence for the past 30 years. While a few neuroimaging studies provided partial evidence in support of the traditional upright face organisation31, other studies supported the inverted (or ‘upside-down’) somatotopic organisation of the face, similar to that of non-human primates32,33. Other studies suggested a segmental organisation34, or even a lack of somatotopic organisation35–37, whereas some studies provided inconclusive or incomplete results38–41. Together, the available evidence does not successfully converge on face topography in humans. In line with the upright organisation originally suggested by Penfield, recent work reported that the shift in the lip representation towards the missing hand in amputees was minimal42,43, and likely to reside within the face area itself. Surprisingly, there is currently no research that considers the representation of other facial parts, in particular the upper-face (e.g., the forehead), in relation to plasticity or PLP.”

We also updated the discussion accordingly (lines 456, 468-476, 489-491).

Similarly, it is not clear to me how the observations (1) of limited reorganization in amputees, (2) of significant reorganization in congenital one-handers, and (3) of the lack of relationship between PLP and reorganization is novel given the previous work by this group. Perhaps the authors could more clearly articulate the novelty of these results compared to their previous findings.

Thank you for giving us the opportunity to clarify on this important point. The novelty of these results can be summarised as follow:

  1. Conceptually, it is crucial for us to understand if deprivation-triggered plasticity is constrained by the local neighbourhood, because this can give us clues regarding the mechanisms driving the remapping. We provide strong topographic evidence about the face orientation in controls, amputees and one-handers.

  2. The vast majority of previous research on brain plasticity following hand loss (both congenital and acquired) in humans has exclusively focused on the lower face, and lips in particular. We provide systematic evidence for stable organisation and remapping of the neighbouring upper face, as well as the lower face. We also study topographic representation of the tongue (and nose) for the first time.

  3. The vast majority of previous research on brain remapping following hand loss (both congenital and acquired, neuroimaging and electrophysiological) was focused on univariate activity measures, such as the spatial spread of units showing a similar feature preference, or the average activity level across individual units. We are going beyond remapping by using RSA, which allows us to ask not only if new information is available in the deprived cortex (as well as the native face area), but also whether this new information is structured consistently across individuals and groups. We show that representational content is enhanced in the deprived cortex one-handers whereas it is stable in amputees relative to controls (and to their intact hand region).

  4. Based on previous studies, the assumption was that reorganisation in congenital one-handers was relatively unspecific, affecting all tested body parts. Here, we provide evidence for a more complex pattern of remapping, with the forehead representation seemingly moving out of the missing hand region (and the nose representation being tentatively similar to controls). That is, we show not just “invasion” but also a shift of the neighbour away from the hand area which has never been documented (or in fact suggested).

  5. Using Bayesian analyses we provide definitive evidence against a relationship between PLP and forehead remapping, providing first and conclusive evidence against the remapping hypothesis, based on cortical neighbourhood.

Our inclination is not to add a summary paragraph of these points in our discussion, as it feels too promotional. Instead, we have re-written large sections of the introduction and discussion to better emphasise each of these points separately throughout the text, where the context is most appropriate. Given the public review strategy taken by eLife, the novelty summary provided above will be available for any interested reader, as part of the public review process. However, should the reviewer feel that a novelty summary paragraph is required (or an emphasis on any of the points summarised above), we will be happy to revise the manuscript accordingly.

Finally, Jon Kaas and colleagues (notably Niraj Jain) have provided evidence in experiments with monkeys that much of the observed reorganization in the somatosensory cortex is inherited from plasticity in the brain stem. Jain did not find an increased propensity for axons to cross the septum between face and hand representations after (simulated) amputation. From this perspective, the relevant proximity would be that of the cuneate and trigeminal nuclei and it would be critical to map out the somatotopic organization of the trigeminal and cuneate nuclei to test hypotheses about the role of proximity in this remapping.

Thank you for highlighting this very relevant point, which we are well aware of. We fully agree with the reviewer that this is an important goal for future study, but functional imaging of the brainstem in humans is particularly challenging and would require ultra high field imaging (7T) and specialised equipment. We have encountered much local resistance due to hypothetical issues for MRI safety for scanning amputees in this higher field strength, meaning we are unable to carry out this research ourselves. Our former lab member Sanne Kikkert, who is now running her independent research programme in Zurich, has been working towards this goal for the past 4 years. So we can say with confidence that this aim is well beyond the scope of the current study. In response to your comment, we mentioned this potential mechanism in the introduction (lines 98-101), we ensured that we only referred to “cortical proximity” throughout our manuscript, and we circle back to this important point in the discussion.

Lines 538-542: “Moreover, even if the remapping we observed here goes against the theory of cortical proximity, it can still arise from representational proximity at the subcortical level, in particular at the brainstem level44,45. While challenging in humans, mapping both the cuneate and trigeminal nuclei would be critical to provide a more complete picture regarding the role of proximity in remapping.”

A few comments:

Which cortical fields do the authors refer to as primary somatosensory cortex? Brodmann's areas 3a, 3b, 1 and 2?

On line 322, I suggest using "completeness" rather than "completion."

On line 372, I suggest using "critique" rather than "limitation."

We thank the reviewer for the above minor comments and can confirm that our primary somatosensory cortex refers to Brodmann’s areas 1, 2, 3a and 3b (stated in Regions of Interest (ROI) definition section, line 714, ‘Materials and methods’). We have also changed the suggested words now on line 438 and line 480 based on the reviewer’s recommendation.

Reviewer #3 (Recommendations for the authors):

In their study, the authors set up to challenge the long-held claim that cortical remapping in the somatosensory cortex in hand deprived cortical territories follows somatotopic proximity (the hand region gets invaded by cortical neighbors) as classically assumed. In contrast to this claim, the authors suggest that remapping may not follow cortical proximity but instead functional rules as to how the effector is used. Their data indeed suggest that the deprived hand area is not invaded by the forefront which is the cortical neighbor but instead by the lips which may compensate for hand loss in manipulating objects. Interestingly the authors suggest this is mostly the case for one-handers but not in amputees for who the reorganization seems more limited in general (but see my comments below on this last point).

This is a remarkably ambitious study that has been skilfully executed on a strong number of participants in each group. The complementarity of state-of-the-art uni- and multi-variate analyses are in the service of the research question, and the paper is clearly written. The main contribution of this paper, relative to previous studies including those of the same group, resides in the mapping of multiple face parts all at once in the three groups.

We are grateful to the reviewer for appreciating the immense effort that this study involved.

In the winner takes all approach, the authors only include 3 face parts but exclude from the analyses the nose and the thumb. I am not fully convinced by the rationale for not including nose in univariate analyses – because it does not trigger reliable activity – while keeping it for representational similarity analyses. I think it would be better to include the nose in all analyses or demonstrate this condition is indeed "noisy" and then remove it from all the analyses. Indeed, if the activity triggered by nose movement is unreliable, it should also affect multivariate.

Following this comment, we re-ran all univariate analyses to include the nose, and updated throughout the main text and supplemental results and related figures. In short, adding the nose did not change the univariate results, apart from a now significant group x hemisphere interaction for the CoG of the tongue when comparing amputees and controls, matching better the trends for greater surface coverage in the deprived hand ROI of amputees. Full details are provided in our response to Reviewer 1 above.

The rationale for not including the hand is maybe more convincing as it seems to induce activity in both controls and amputees but not in one-handers. First, it would be great to visualize this effect, at least as supplemental material to support the decision. Then, this brings the interesting possibility that enhanced invasion of hand territory by lips in one-handers might link to the possibility to observe hand-related activity in the presupposed hand region in this population. Maybe the authors may consider linking these.

Thank you for this comment. As we explain in our response to Reviewer 1 above, we did not intent the thumb condition in one-handers for analysis, as the task given to one-handers (imagine moving a body part you never had before) is inherently different to that given to the other groups (move – or at least attempt to move – your (phantom) hand). As such, we could not pursuit the analysis suggested by the reviewer here. To reduce the discrepancy and following Reviewer 1’s advice, we decided to remove the hand-face dissimilarity analysis which we included in our original manuscript, and might have sparked some of this interest. Upon reflection we agreed that this specific analysis does not directly relate to the question of remapping (but rather of shared representation), in addition to making the paper unbalanced. We will now feature this analysis in another paper that appears more appropriate in the context of referred sensations in amputees (Amoruso et al., 2022 MedRxiv).

The use of the geodesic distance between the center of gravity in the Winner Take All (WTA) maps between each movement and a predefined cortical anchor is clever. More details about how the Center Of Gravity (COG) was computed on spatially disparate regions might deserve more explanations, however.

We are happy to provide more detail on this analysis, which weights the CoG based on the clusters size (using the workbench command -metric-weighted-stats). Let’s consider the example in Author response image 1 for a single control participant, where each CoG is measured either without weighting (yellow vertices) or with cluster weighting (forehead CoG=red, lip CoG=dark blue, tongue CoG=dark red). When the movement produces a single cluster of activity (the lips in the non-dominant hemisphere, shown in blue), the CoG’s location was identical for both weighted (red) and unweighted (yellow) calculations. But other movements, such as the tongue (green), produced one large cluster (at the lateral end), with a few more disparate smaller clusters more medially. In this case, the larger cluster of maximal activity is weighted to a greater extent than the smaller clusters in the CoG calculation, meaning the CoG is slightly skewed towards it (dark red), relative to the smaller clusters.

Author response image 1
Centre-of-gravity calculation, weighted and unweighted by cluster size, in an example control participant.

Here the winner-takes-all output for each facial movement (forehead=red, lips=blue, tongue=green) was used to calculate the centre-of-gravity (CoG) at the individual-level in both the dominant (left-hand side) and non-dominant (right-hand side) hemisphere, weighted by cluster size (forehead CoG=red, lip CoG=dark blue, tongue CoG=dark red), compared to an unweighted calculation (denoted by yellow dots within each movements’ winner-takes-all output).

This is now explained in the methods (lines 759-764) as follows:

“To assess possible shifts in facial representations towards the hand area, the centre-of-gravity (CoG) of each face-winner map was calculated in each hemisphere. The CoG was weighted by cluster size meaning that in the event of multiple clusters contributing to the calculation of a single CoG for a face-winner map, the voxels in the larger cluster are overweighted relative to those in the smaller clusters. The geodesic cortical distance between each movement’s CoG and a predefined cortical anchor was computed.”

Moreover, imagine that for some reason the forefront region extends both dorsally and ventrally in a specific population (eg amputees), the COG would stay unaffected but the overlap between hand and forefront would increase. The analyses on the surface area within hand ROI for lips and forehead nicely complement the WTA analyses and suggest higher overlap for lips and lower overlap for forehead but none of the maps or graphs presented clearly show those results – maybe the authors could consider adding a figure clearly highlighting that there is indeed more lip activity IN the hand region.

We agree with you on this limitation of the CoG and this is why we interpret all cortical distances analyses in tandem with the laterality indices. The laterality indices correspond to the proportion of surface area in the hand region for a given face part in the winner-maps.

Nevertheless, to further convince the Reviewer, we extracted activity levels (β values) within the hand region of congenitals and controls, and we ran (as for CoGs) a mixed ANOVA with the factors Hemisphere (deprived x intact) and Group (controls x one-handers).

As expected from the laterality indices obtained for the Lips, we found a significant group x hemisphere interaction (F(1,41)=4.52, p=0.040, n2p=0.099), arising from enhanced activity in the deprived hand region in one-handers compared to the non-dominant hand region in controls (t(41)=-2.674, p=0.011) and to the intact hand region in one-handers (t(41)=-3.028, p=0.004).

Author response image 2

Since this kind of analysis was the focus of previous studies (from which we are trying to get away) and since it is redundant with the proportion of face-winner surface coverage in the hand region, we decided not to include it in the paper. But we could add it as a Supplementary result if the Reviewer believes this strengthens our interpretation.

In addition to overlap analyses between hand and other body parts, the authors may also want to consider doing some Jaccard similarity analyses between the maps of the 3 groups to support the idea that amputees are more alike controls than one-handers in their topographic activity, which again does not appear clear from the figures.

We thank the reviewers for this clever suggestion. We now include the Jaccard similarity analysis, which quantified the degree of similarity (0=no overlap between maps; 1=fully overlapping) between winner-takes-all maps (which included the nose; akin to the revised univariate results) across groups. For each face part/amputee, the similarity with the 22 controls and 21 one-handers respectively was averaged. We utilised a linear mixed model which included fixed factors of Group (One-handers x Controls), Movement (Forehead x Nose x Lips x Tongue) and Hemisphere (Intact x Deprived) on Jaccard similarity values (similar to what we used for the RSA analysis). A random effect of participant, as well as covariates of ages, were also included in the model.

Results showed a significant group x hemisphere interaction (F(240.0)=7.70, p=0.006; controlled for age; Figure 5), indicating that amputees’ maps showed different similarity values to controls’ and one-handers’ depending on the hemisphere. Post-hoc comparisons (corrected α=0.025; uncorrected p-values reported) revealed significantly higher similarity to controls’ than to one-handers’ maps in the deprived hemisphere (t(240)=-3.892, p<.001). Amputees’ maps also showed higher similarity to controls’ maps in the deprived relative to the intact hemisphere (t(240)=2.991, p=0.003). Amputees, therefore, displayed greater similarity of facial somatotopy in the deprived hemisphere to controls, suggesting again fewer evidence for cortical remapping in amputees.

We added these results at the end of the univariate analyses (lines 334-350) and in the discussion (lines 463-464 and 496-499).

This brings to another concern I have related to the claim that the change in the cortical organization they observe is mostly observed in one-handers. It seems that most of this conclusion relies on the fact that some effects are observed in one-handers but not in amputees when compared to controls, however, no direct comparisons are done between amputees and one-handers so we may be in an erroneous inference about the interaction when this is actually not tested (Nieuwenhuis, 11). For instance, the shift away from the hand/face border of the forehead is also (mildly) significant in amputees (as observed more strongly in one-handers) so the conclusion (eg from the subtitle of the Results section) that it is specific to one-hander might not fully be supported by the data. Similar to the invasion of the hand territory from the lips which is significant in amputees in terms of surface area. All together this calls for toning down the idea that plasticity is restricted to congenital deprivation (eg last sentence of the abstract). Even if numerically stronger, if I am not wrong, there are no stats showing remapping is indeed stronger in one-handers than in amputees and actually, amputees show significant effects when compared to controls along the lines as those shown (even if more strongly) in one-handers.

Thank you for this very important comment. We fully agree – the RSA across-groups comparison is highly informative but insufficient to support our claims. We did not compare the groups directly to avoid multiple comparisons (both for statistical reasons and to manage the size of the Results section). But the reviewer’s suggestion to perform a Jaccard similarity analysis complements very nicely the univariate and multivariate results and allows for a direct (and statistically lean) comparison between groups, to assess whether amputees are more similar to controls or to congenital one-handers, taking into account all aspects of their maps (both spatial location/CoG and surface coverage). We added the Jaccard analysis to the main text, at the end of the univariate results (lines 334-384). The Jaccard analysis suggests that amputees’ maps in the deprived hemisphere were more similar to the maps of controls than to the ones of congenital one-handers. This allowed us to obtain significant statistical results to support the claim that remapping is indeed stronger in one-handers than in amputees (lines 345-350). We also compared both amputees and one-handers to the control group. In line with our univariate results, this revealed that the only face part for which controls were more similar to one-handers than to amputees was the tongue (lines 378-380). And that the forehead remapping observed at the univariate level in amputees (surface area), is likely to arise from differences in the intact hemisphere (lines 380-382).

Finally, we also added the post-hoc statistics comparing amputees to congenitals in the RSA analysis (lines 424-426): “While facial information in the deprived hand area was increased in one-handers compared with amputees, this effect did not survive our correction for multiple comparisons (t(70.7)=-2.117, p=0.038).”

Regarding the univariate results mentioned by the reviewer, we would like to emphasise that we had no significant effect for the lips in amputees, though we agree the surface area appears in between controls and one-handers. But this laterality index was not different from zero. This test is now added lines 189-190. Regarding the forehead, we fully agree with the Reviewer, and we adjusted the subtitle accordingly (lines 240-241). For consistency, we also added the t-test vs zero for the forehead surface area (non-significant, lines 250-252).

Also, maybe the authors could explore whether there is actually a link between the number of years without hand and the remapping effects.

To address this question, we explored our data using a correlation analysis. The only body part who showed some suggestive remapping effects was the tongue, and so we explored whether we could find a relationship (Pearson’s correlation) between years since amputation and the laterality index of the Tongue in amputees (r = 0.007, p=0.980, 95% CI [-0.475, 0.475]). We also explored amputees’ global Jaccard similarity values to controls in the deprived hemisphere (r = -0.010, p=0.970, 95% CI [-0.488, 0.473]), and could not find any relationship. Considering there was no strong remapping effect to explain, we find this result too exploratory to include in our manuscript.

One hypothesis generated by the data is that lips remap in the deprived hand area because lips serve compensatory functions. Actually, also in controls, lips and hands can be used to manipulate objects, in contrast to the forehead. One may thus wonder if the preferential presence of lips in the hand region is not latent even in controls as they both link in functions?

We agree with the reviewer’s reasoning, and we think that the distributed representational content we recently found in two-handers (Muret et al., 2022) provides a first hint in this direction. It is worth noting that in that previous publication we did not find differences across face parts in the activity levels obtained in the hand region, except for slightly more negative values for the tongue. But we do think that such latent information is likely to provide a “scaffolding” for remapping. While the design of our face task does not allow to assess information content for each face part (as done for the lips in Muret et al., 2022), this should be further investigated in follow-up studies.

We added a sentence in the discussion to highlight this interesting notion:

Lines 555-558: “Together with the recent evidence that lip information content is already significant in the hand area of two-handed participants (Muret et al., 2022), compensatory behaviour since developmental stages might further uncover (and even potentiate) this underlying latent activity.”

1) The authors should re-assess their conclusion that remapping is mostly present in one-handers than amputees in absence of a convincing statistical demonstration that it is actually the case. The amputees show similar effects or trends for similar remapping as those observed in one-handers. maybe one source of this lower reliability in the effects of amputees links to the lower number of participants and their higher variability in ethology (eg age of amputation etc…). Maybe the authors could explore whether there is actually a link between the number of years without hand and the remapping effects.

We would like to emphasise that our Bayesian statistics help address this issue when supportive of the null hypothesis (i.e., lack of interaction for the lip CoG: BF10=0.297). But we also fully agree that a more direct comparison between amputees and one-handers was necessary. As we detailed above, we now provide analysis to address both univariate and multivariate group differences, performed with the two most sensitive analyses, Jaccard analysis and RSA.

In the Jaccard analysis, amputees showed significantly higher similarity to controls, and their global similarity to controls was not related to covariates such as years since amputation. We also show in Figure 5 —figure supplement 1 comparable intra-group consistency in amputees relative to one-handers. We believe this rules out the reviewer’s concern that the amputees data is more noisy/variable due to amputation-related covariates.

2) The authors should be coherent in the body part they include for their univariate and multivariate analyses. I let them decide whether they think the nose condition was too noisy (eg because people could not follow instructions to move the nose) or not. If they think the condition however did not elicit reliable activity, this should affect both uni and multivariate analyses.

We have addressed this comment in the most straight forward way we could – re-running all analyses to include the nose and update all of the study results accordingly. Despite slight quantitative changes, the overall pattern of the results was relatively unchanged in comparison to our original analysis.

3) I recommend the authors show the map elicited by hand movement in the three groups, support their claim that this is only reliably observed in controls and amputees but not one-handers; and maybe discuss (or even better do some analyses on that, eg correlation) the possibility that the preservation of the hand map is potentially a factor influencing remapping (eg do the amputees with the most salient remaining hand maps are those showing less remapping?).

We thank the reviewer for their interesting suggestion that the preservation of the hand map may influence reported (lack of) remapping. To explore this within our data, and from the recommendation of the reviewer, we looked at the correlation between the percentage of surface area coverage for maximal phantom hand activity in the hand ROI (similar to that shown in Appendix Figure 3 in standard space) to the laterality indices of the tongue (the one facial part in amputees to tentatively indicate facial remapping). When doing so we find a non-significant correlation (kendall's tau=0.121, p=0.281, BF10=0.366), along with an inconclusive Bayes Factor, suggesting that preservation of phantom hand activity may not be a deciding factor of facial remapping within this dataset. Considering there was no strong remapping effect to explain, we find this result too exploratory to include in our manuscript. But we would happily add it to the Supplements if advised by the Reviewer.

We therefore continued with an alternative approach that might make more intuitive sense – do we have more stability, thus more control-like maps, if amputees display more (preserved) hand representation? To address this question, we ran a Pearson correlation between the global Jaccard similarity values of amputees relative to controls and the percentage of surface area coverage for maximal phantom hand activity in the hand ROI (similar to that shown in Appendix Figure 3 in standard space). We found here again a non-significant correlation (r = 0.317, p=0.215, 95% CI [-0.193, 0.692]).

Author response image 3

4) The authors should clarify in their figure where do they observe that the lip increases activity WITHIN the hand region in one-handers.

The laterality indices are clearly showing this, since it corresponds to the proportion of surface area coverage IN the hand region, for the winner-maps of each face part (thus winning over the other face parts). See above for a similar analysis on activity levels (β values) in the hand region.

5) The fact that the univariate results of Figure 1 do not follow well those presented in the WTA maps deserves some consideration. Indeed, what seems striking in Figure 1 is that the forehead activity extends more dorsally in one-handers, approaching the hand region, which contradicts conclusions from WTA. Also, it might be informative to have an outline of the brain mask (in all figures) of the S1 ROI used for constraining spatially the univariate analyses.

We apologise for any confusion. As we detail in our response to Reviewer 1 above, the activity presented in Figure 1 reflects each facial movement contrasted to rest, mainly localised outside, and posterior to, the hand ROI (outlined in Figure 1 in purple). The WTA maps, summarised in the consistency maps in Figures 2-4, reflect how consistent are the (minimally thresholded) WTA maps. If we used a higher threshold, much of the upper face WTA map would disappear.

6) In Figure 5 the thumb is not represented in one-handers. It would be interesting for the reader to see that indeed the instruction to move the thumb did not elicit a reliable activity map and what was the criteria to selectively remove it in one-handers aside from the fact they don't experience fantom sensation. It is also unclear why the thumb is not displayed in the MDS while it's computed in the RDM of controls and amputees.

Upon reflection (and thanks to the reviewers’ comments), we decided to remove entirely the hand-face dissimilarity analysis, which does not directly relate to the question of remapping (but rather of shared representation), in addition to making the paper unbalanced. We will now feature this analysis in another paper that appears more appropriate in the context of referred sensations in amputees (Amoruso et al., 2022 MedRxiv).

7) Maybe discuss further the potential (latent) link between hand and lips as they both can be used to manipulate objects even in controls.

We added a sentence in the discussion to that extent:

Lines 555-558: “Together with the recent evidence that lip information content is already significant in the hand area of two-handed participants (Muret et al., 2022), compensatory behaviour since developmental stages might further uncover (and even potentiate) this underlying latent activity.”

https://doi.org/10.7554/eLife.76158.sa2

Article and author information

Author details

  1. Victoria Root

    1. WIN Centre, University of Oxford, Oxford, United Kingdom
    2. Institute of Cognitive Neuroscience, University College London, London, United Kingdom
    3. Medical Research Council Cognition and Brain Sciences Unit (CBU), University of Cambridge, Cambridge, United Kingdom
    Contribution
    Conceptualization, Data curation, Software, Formal analysis, Validation, Investigation, Visualization, Methodology, Writing – original draft, Writing – review and editing
    Contributed equally with
    Dollyane Muret
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0002-0500-3206
  2. Dollyane Muret

    Institute of Cognitive Neuroscience, University College London, London, United Kingdom
    Contribution
    Conceptualization, Data curation, Software, Formal analysis, Supervision, Validation, Investigation, Visualization, Methodology, Writing – original draft, Project administration, Writing – review and editing
    Contributed equally with
    Victoria Root
    For correspondence
    dollyane.muret@inserm.fr
    Competing interests
    No competing interests declared
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0002-2626-654X
  3. Maite Arribas

    1. Institute of Cognitive Neuroscience, University College London, London, United Kingdom
    2. Department of Psychosis Studies, Institute of Psychiatry, Psychology & Neuroscience, King's College London, London, United Kingdom
    Contribution
    Data curation, Formal analysis, Writing – original draft
    Competing interests
    No competing interests declared
  4. Elena Amoruso

    1. Institute of Cognitive Neuroscience, University College London, London, United Kingdom
    2. Medical Research Council Cognition and Brain Sciences Unit (CBU), University of Cambridge, Cambridge, United Kingdom
    Contribution
    Investigation, Writing – original draft, Project administration
    Competing interests
    No competing interests declared
  5. John Thornton

    Wellcome Trust Centre for Neuroimaging, University College London, London, United Kingdom
    Contribution
    Resources, Writing – review and editing
    Competing interests
    No competing interests declared
  6. Aurelie Tarall-Jozwiak

    Queen Mary’s Hospital, London, United Kingdom
    Contribution
    Resources, Writing – review and editing
    Competing interests
    No competing interests declared
  7. Irene Tracey

    WIN Centre, University of Oxford, Oxford, United Kingdom
    Contribution
    Supervision, Writing – review and editing
    Competing interests
    No competing interests declared
  8. Tamar R Makin

    1. Institute of Cognitive Neuroscience, University College London, London, United Kingdom
    2. Medical Research Council Cognition and Brain Sciences Unit (CBU), University of Cambridge, Cambridge, United Kingdom
    3. Wellcome Trust Centre for Neuroimaging, University College London, London, United Kingdom
    Contribution
    Conceptualization, Resources, Supervision, Funding acquisition, Methodology, Writing – original draft, Project administration, Writing – review and editing
    Competing interests
    Senior editor, eLife
    ORCID icon "This ORCID iD identifies the author of this article:" 0000-0002-5816-8979

Funding

European Research Council (715022)

  • Tamar R Makin

Wellcome Trust (215575/Z/19/Z)

  • Tamar R Makin

The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication. For the purpose of Open Access, the authors have applied a CC BY public copyright license to any Author Accepted Manuscript version arising from this submission.

Acknowledgements

We thank Arabella Bouzigues and Maria Kromm for their substantial help in terms of recruitment and data collection, we also thank Adriana Zainurin, Esther Teo, Christine Tan, Raffaele Tucciarelli and Mathew Kollamkulam for help with data collection. We thank Opcare for their help with participants recruitment, and our participants and their families for their ongoing support to our research. This work was supported by an ERC Starting Grant (715022 EmbodiedTech) and a Wellcome Trust Senior Research Fellowship (215575/Z/19/Z), awarded to TRM.

Ethics

Written Informed consent, and consent to publish, was obtained from all participants. Ethical approval was obtained from the NHS National Research Ethics Service approval (18/LO/0474).

Senior Editor

  1. Christian Büchel, University Medical Center Hamburg-Eppendorf, Germany

Reviewing Editor

  1. Olivier Collignon, Université Catholique de Louvain, Belgium

Reviewers

  1. Moritz Wurm, University of Trento, Italy
  2. Sliman J Bensmaia, University of Chicago, United States
  3. Olivier Collignon, Université Catholique de Louvain, Belgium

Publication history

  1. Preprint posted: July 5, 2021 (view preprint)
  2. Received: December 6, 2021
  3. Accepted: December 29, 2022
  4. Accepted Manuscript published: December 30, 2022 (version 1)
  5. Version of Record published: January 19, 2023 (version 2)

Copyright

© 2022, Root, Muret et al.

This article is distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use and redistribution provided that the original author and source are credited.

Metrics

  • 272
    Page views
  • 57
    Downloads
  • 0
    Citations

Article citation count generated by polling the highest count across the following sources: Crossref, PubMed Central, Scopus.

Download links

A two-part list of links to download the article, or parts of the article, in various formats.

Downloads (link to download the article as PDF)

Open citations (links to open the citations from this article in various online reference manager services)

Cite this article (links to download the citations from this article in formats compatible with various reference manager tools)

  1. Victoria Root
  2. Dollyane Muret
  3. Maite Arribas
  4. Elena Amoruso
  5. John Thornton
  6. Aurelie Tarall-Jozwiak
  7. Irene Tracey
  8. Tamar R Makin
(2022)
Complex pattern of facial remapping in somatosensory cortex following congenital but not acquired hand loss
eLife 11:e76158.
https://doi.org/10.7554/eLife.76158

Further reading

    1. Neuroscience
    2. Physics of Living Systems
    Sabrina A Jones, Jacob H Barfield ... Woodrow L Shew
    Research Article

    Naturally occurring body movements and collective neural activity both exhibit complex dynamics, often with scale-free, fractal spatiotemporal structure. Scale-free dynamics of both brain and behavior are important because each is associated with functional benefits to the organism. Despite their similarities, scale-free brain activity and scale-free behavior have been studied separately, without a unified explanation. Here we show that scale-free dynamics of mouse behavior and neurons in visual cortex are strongly related. Surprisingly, the scale-free neural activity is limited to specific subsets of neurons, and these scale-free subsets exhibit stochastic winner-take-all competition with other neural subsets. This observation is inconsistent with prevailing theories of scale-free dynamics in neural systems, which stem from the criticality hypothesis. We develop a computational model which incorporates known cell-type-specific circuit structure, explaining our findings with a new type of critical dynamics. Our results establish neural underpinnings of scale-free behavior and clear behavioral relevance of scale-free neural activity.

    1. Neuroscience
    Barna Zajzon, David Dahmen ... Renato Duarte
    Research Article

    Information from the sensory periphery is conveyed to the cortex via structured projection pathways that spatially segregate stimulus features, providing a robust and efficient encoding strategy. Beyond sensory encoding, this prominent anatomical feature extends throughout the neocortex. However, the extent to which it influences cortical processing is unclear. In this study, we combine cortical circuit modeling with network theory to demonstrate that the sharpness of topographic projections acts as a bifurcation parameter, controlling the macroscopic dynamics and representational precision across a modular network. By shifting the balance of excitation and inhibition, topographic modularity gradually increases task performance and improves the signal-to-noise ratio across the system. We demonstrate that in biologically constrained networks, such a denoising behavior is contingent on recurrent inhibition. We show that this is a robust and generic structural feature that enables a broad range of behaviorally-relevant operating regimes, and provide an in-depth theoretical analysis unravelling the dynamical principles underlying the mechanism.