Cortical Plasticity in Phantom Limb Pain: a fMRI study on the neural correlates of behavioral clinical manifestations.

Abstract

The neural mechanism of phantom limb pain (PLP) is related to the intense brain reorganization process implicating plasticity after deafferentation mostly in sensorimotor system. There is a limited understanding of the association between the sensorimotor system and PLP. We used a novel task-based functional magnetic resonance imaging (fMRI) approach to (1) assess neural activation within a-priori selected regions-of-interested (motor cortex [M1], somatosensory cortex [S1], and visual cortex [V1]), (2) quantify the cortical representation shift in the affected M1, and (3) correlate these changes with baseline clinical characteristics. In a sample of 18 participants, we found a significantly increased activity in M1 and S1 as well as a shift in motor cortex representation that was not related to PLP intensity. In an exploratory analyses (not corrected for multiple comparisons), they were directly correlated with time since amputation; and there was an association between increased activity in M1 with a lack of itching sensation and V1 activation was negatively correlated with PLP. Longer periods of amputation lead to compensatory changes in sensory-motor areas; and itching seems to be a protective marker for less signal changes. We confirmed that PLP intensity is not associated with signal changes in M1 and S1 but in V1.

1. INTRODUCTION

Phantom Limb pain is a subtype of neuropathic pain, characterized by the presence of painful sensation in the area that corresponds to the missing limb. However, the underlying pathophysiologic mechanisms related to the development of pain in amputees are not well established, particularly because of controversial results seen in amputees with and without PLP. In the last decade, compelling evidence has been published supporting the theory in which reorganization processes - maladaptive cortical reorganization - in central control areas (i.e motor and somatosensory system), at cortical ^1–4 and thalamic levels ⁵ are associated with the development phantom limb pain. However, larger studies evaluating neurophysiologic changes post-amputation are needed.

In fact, task-based fMRI using a combination of motor execution and motor imagery (e.g. imagination of the image observed as being the subject’s own missing limb) is widely adopted as to determine changes in blood oxygen level-dependent (BOLD) signal in networks involved in PLP. These areas include the sensory-motor region and their interplay with visual networks^6–8. In patients experiencing PLP, fMRI can quantify patterns of functional brain activity as well as maladaptive neural reorganization through the shifts in cortical representation caused by the amputation and therefore better unravel phantom pain mechanisms and identify neurological markers as predictors of treatment success^9,10.

Current evidence concerning the mechanisms of PLP remains contradictory, studies performed in upper limb amputees with fMRI by Flor and collaborators revealed that amputation leads to changes at a central level. This was characterized by neighboring cortical regions invading the corresponding areas of the missing limbs, most likely as a consequence of a lack of afferent inputs post-amputation¹. In the same way, Lotze et al. (2001)¹¹ and Foell et al (2013)¹² revealed that in hand amputees, there was an invasion of the lip representation over the deafferented hand area. The former found that amount of displacement was correlated with higher phantom pain and the latter revealed a reversal of dysfunctional cortical reorganization in primary somatosensory cortex in patients with phantom pain who responded to mirror therapy. Additionally, the stimulation of the lip generated neural activation in the primary somatosensory and motor cortices of the deafferented hand positive correlated with PLP intensity ^1,11,13. Likewise, after behavioral therapies such as mirror therapy, functional MRI studies have shown a return of the cortical reorganization towards a congruent somatotopic pattern of activity, and that has been associated with relief of PLP symptoms.

However, opposing results were also described in upper limb amputees by Makin and collaborators¹⁴, as they demonstrated a preserved functional and structural organization after amputation. Reductions in interregional functional connectivity were hypothesized to be the most likely potential mechanism underlying the emergence of PLP¹⁴. Some other studies replicated these findings but found no statistical significance between network-level reorganization and phantom pain or sensation^15,16. Most of the data investigating the neural mechanisms of PLP with fMRI included subjects with upper limb amputation, and our sample has only unilateral lower limb amputation due to traumatic cause.

This investigation aimed at characterizing functional alterations in leg movements and phantom leg movements in traumatic lower limb amputees that experience PLP to further investigate the possible links between different activations patterns with pain intensity and PLP-related characteristics. We examined task-related neural activity separately in the primary somatosensory (S1), motor (M1), and visual (V1) cortices of all amputees. We also explored (no correction for multiple comparisons) the relationship between the shift in neural activity and phantom pain intensity and its clinical characteristics using cortical distance measures.

2. MATERIALS AND METHODS

2.1. Study Design and Settings

This is a cross-sectional analysis of baseline fMRI results and clinical characteristics of participants that are currently enrolled in an ongoing double-blinded randomized multicenter clinical trial evaluating the effects of the combined intervention with transcranial direct current stimulation (tDCS) and mirror therapy (MT) for the relief of PLP in traumatic lower limb amputees (NCT02487966)¹⁷. Ethics approval was obtained from the Partner Healthcare Institutional Review Board. All participants included in this analysis signed the approved consent form before undergoing any trial procedures. The study PI and/or a co-investigator obtained informed consent.

2.2. Participants

The present study included 18 participants with unilateral traumatic lower limb amputation (baseline characteristics are described in table 1). Participants were considered eligible if they fulfilled the following inclusion criteria and none of the exclusion criteria: (1) age 18 years and older; (2) able to provide informed consent; (3) unilateral lower limb amputation caused by traumatic event; (4) experiencing a minimum of 3 months of PLP after the amputated limb has completely healed post-resection procedures; (5) PLP must occur at least once a week; (6) with an average pain of at least 4 on a numeric rating scale (NRS), ranging from 0 to 10 in the prior week to enrolling in the study; (6) dosage of any medication must have keep the same/unchanged for the past two weeks. Participants were excluded if: (1) there was confirmed pregnancy or trying to become pregnant in the next following 2 months; (2) history of alcohol or drug abuse within the past 6 months as self-reported; (3) presence of any contraindications to tDCS and/or TMS (e.g., plates or pins, bullets, shrapnel, cochlear implants, vagal nerve stimulator in the head or neck); (4) prior head injury resulting in permanent neurological deficits, such as cognitive and motor impairments; (5) unstable medical conditions (e.g. uncontrolled diabetes, uncompensated cardiac issues, heart failure or chronic obstructive pulmonary disease); (6) history of uncontrolled epilepsy or prior seizures within past year prior to enrollment; (7) self-reported history of unexplained fainting spells or loss of consciousness within the last two years; (8) prior neurosurgical history; (9) subjects that underwent mirror therapy in the prior 3 months to enrollment.

Table 1.

Demographics and clinical characteristics.

Subject	Age	Level of amputation	Gender	Side of amputation	BAI	BDI	PLP	PLS	RLP	Time (months)
1	72	BKA	M	R	3	5	6	8	3	7
2	56	BKA	M	L	9	11	8,5	9,5	9	21
3	37	BKA	F	R	17	12	2,5	4	3,5	49
4	49	AKA	F	L	7	5	2	3	2	4
5	45	AKA	M	R	22	29	3,5	5	4,5	396
6	46	BKA	F	R	24	7	4	2	2	50
7	47	AKA	F	R	26	32	6	9,5	5	23
8	52	BKA	M	R	4	0	6,5	7	5	163
9	59	AKA	M	L	6	9	5	8	1	468
10	21	BKA	M	L	1	5	5	3	5	160
11	64	BKA	F	R	0	0	5,5	2	0	21
12	63	AKA	M	R	3	0	5	2	0	213
13	69	BKA	M	L	0	0	6	0	0	6
14	59	BKA	F	L	14	13	7	8	5	5
15	60	AKA	F	L	29	18	5	10	4,5	7
16	32	AKA	M	L	20	5	6	10	0	4
17	61	AKA	M	L	7	1	4	4	0	6
18	57	BKA	F	L	9	8	4	3	0	456
Mean (SD)	52.72 (13.13)				11.16 (9.54 )	8.8 (9.3)	5.0 (1.59)	5.4 (3.2)	2.75 (2.60)	114.38 (163.11)
Median (IQR)	56.5 (15)				8 (17)	6 (11)	5 (2)	4.5 (5)	2.5 (5)	22 (157)

Legend: PLP: phantom limb pain; PLS: phantom limb sensation; RLP: residual limb pain; BDI: Beck depression inventory; BAI: Beck anxiety inventory; AKA: above the knee; BKA: below the knee; M: male; F: female; R: right; L: left.

2.3. Clinical Assessments

Demographics:

Demographics characteristics and medical history were assessed during the consent visit and are included in the present analysis.

Amputation-related variables:

The adapted Groningen Questionnaire for lower limb amputation was applied to collect data with regard to PLP. This questionnaire was adapted for lower limb amputation this in order to extract the data on the side of amputation, level of amputation, time since amputation, pain prior to the amputation, opioid use, and previous treatment approaches and whether they have been effective or not; in addition, we extracted PLP/Residual Limb Pain (RLP)/ Phantom Limb Sensations (PLS) frequencies, PLS quality, and types.

Visual/Numeric Analogue Scale for Pain:

Pain was evaluated through the application of the visual analogue scale assessment tool. This is a self-reported rating that addresses the presence and intensity of pain related to PLP and ranging from 0 (no pain at all) to 10 (worst imaginable pain). This scale is also colored, from green (at 0) to red (at 10), as a visual indicator of pain. In addition, similar scales were applied to assess phantom limb sensation –non-painful sensation- and residual limb pain –stump pain.

Psychological variables:

Depression was evaluated by Beck Depression Index (BDI) and anxiety was evaluated by Beck Anxiety Index (BAI).

2.4. FMRI scanning session and design

Functional MRI assessments were performed at baseline and at the end of the primary endpoint (final intervention period) of the main clinical trial¹⁷. The current analysis included only fMRI data from the baseline assessment visit. The participants were trained and practiced the tasks accordingly before the fMRI scans (see tasks below).

We based our fMRI design on a prior experiment investigating task-dependent activation of motor networks associated with observation and imagination of the movements of the lower limbs (motor imagery) ⁶. The initial task consisted of the movement of the lower limb (dorsiflexion and plantar-flexion of the ankle) with the eyes closed, followed by the same exact movement with their eyes open to allow observation of his/her leg movement in a mirror through online real-time video projection inside of the scanner and finally, imagining this leg movement as being participant’s amputated leg (detailed methodology of the fMRI design has been shown in an article published elsewhere¹⁸.

The following three conditions were investigated (each participant was instructed to follow a determined pace during the task performance):

1. Movement of the leg (Leg): The participants performed movements of the intact leg with the eyes closed.

2. Movement of the leg observing the mirror (Mirror): The participants performed the same movements from the previous condition now with the eyes open, observing at the mirrored image of the intact leg in an online video projection and imagining it is the amputated leg moving.

3. Rest condition (Rest): The participants rest during this period. Subjects were instructed to not perform any kind of movements during this period and maintain their eyes closed.

For each task/condition, the length was 20 seconds. The complete fMRI session had four runs that contained 6 blocks (each condition was repeated 6 times). For accuracy, the participant wore headphones with an audio recording that started with the sequence Leg, Mirror and Rest (these words were spoken with a recorded voice). Each of these words inaugurated the task followed by 20 seconds in which the participant listened a continuous “click” sound spaced by one second that dictated the pace of the leg movement. In the Rest condition, the participant listened to the same “click” sound but stayed at rest with the eyes closed. When it was time to switch tasks, the audio again started with the spoken word of the subsequent task. Additionally, the study staffs made sure that the participants understood the commands correctly from the audio and could execute the movements accordingly. A prior practice run inside of a mock scanner identical to the real MRI machine was performed at the imaging center for consistency.

At the scan set up, a real-time video system set up by Saleh Velez et al. 2019¹⁸ allowed the transmission of the intact leg movements being reflected by the mirror located between the participants lower limbs and conveyed the images towards a screen on the back of the scanner bore that the patient can see through a system of mirrors attached to the MRI head coil (figure 2). The recordings were transmitted in real time online to the screen, so the participants had the true experience of performing a mirror therapy task somewhat similar to the intervention of the main clinical trial¹⁷.

Figure 2:

Schematic figure depicting the online task-based fMRI scan. In this set up, participants were able to observe a real-time video of the intact leg movements and its reflection by the mirror located between the participants lower limbs using a mirror placed over the MRI head coil. This is a reproducible and easy to build setup, more details can be found in Saleh Velez et al. 2019¹⁸.

2.5. FMRI acquisition

The neuroimaging scans were performed with a 3 Tesla Philips Achieva Scanner (Philips Medical Systems, Eindhoven, The Netherlands) located at the Center for Biomedical Imaging (CBI) at the Boston University Medical Center. The anatomical T1-weighted images were acquired through a turbo spin echo sequence (TE = 3.1 ms, TR = 6.8 ms, flip angle = 9°, voxel size 0.98 × 0.98 × 1.20 mm, no slice gap, acquisition matrix 256 × 254) as stated in our previously published study protocol¹⁷. Each functional acquisition run provided 182 volumes with a voxel size of 3mm (isotropic resolution) recorded with TR=2s, TE=28ms and a flip angle of 90º. Additionally, functional images were measured with T2*-weighted echo-planar images (EPIs) using blood-oxygenation-level-dependent (BOLD) contrast¹⁷.

2.6. FMRI preprocessing

Preprocessing and data analyses were completed using FSL Version 5.0.9 (FMRIB, Oxford, UK¹⁹) with the following workflow: Processing and analysis of imaging data were performed using FSL FEAT (“FEAT,” n.d.²⁰). Preprocessing included brain extraction using FSL’s BET (Brain Extraction Tool), motion correction using FSL’s MCFLIRT (intra-modal motion correction tool)²¹. The remaining fMRI signals were spatially blurred with a 6 mm full-width-athalf-maximum (FWHM) Gaussian kernel²². A subject-dependent number of individual nuisance regressors for removing outlier timepoints were created using the fsl_motion_outliers tool, employing the Derivative of root mean square VARiance over voxels (DVARS) metric and framewise displacement (FD) of 0.5 mm to identify problematic time points during the scan and to prevent potential motion artifacts²³. The threshold used was the standard boxplot outlier threshold set in FSL²⁴. Individual runs were excluded if they had >20% of movement outliers or >2mm mean displacement, and participants were excluded if they had more than one excluded run. Based on this criteria three participants were excluded.

2.7. fMRI analysis and experimental design

Primary neuroimaging analysis was restricted to the primary motor (M1) (further subdivided into anterior and posterior areas)^25,26, primary sensory (S1) and primary visual (V1) areas as a priori regions of interest (ROIs) based on the frequent involvement of these regions in Phantom Limb Pain⁹. We analyzed the relation of M1, S1, V1 activation for the contrasts ‘Leg > Rest, Mirror > Rest’ and ‘Mirror > Leg’ with clinical variables. The hemispheres of the right amputees were flipped at the vertical axis before any analysis was performed to make all participants virtually left amputees and facilitate group analysis. The affected hemisphere always refers to the right hemisphere and unaffected hemisphere always to the left hemisphere. Driven by the results, we explored in-depth the role of these regions running a post-hoc analysis with secondary and tertiary cortices.

2.8. A priori Analyses

Analyses were conducted within the a priori ROIs (figure 3). For this, we used a mask created from the corresponding primary cortices as defined by the Jülich histological (cyto- and myelo-architectonic) ^27,28with a threshold at 85% probability to remove any non-sensory-motor voxels and avoid overlapping of voxels of adjacent brain areas²⁹. By using the same histological atlas we further created masks for anterior and posterior M1 areas. The contrasts for ‘Leg > Rest, Mirror > Rest and Mirror > Leg were calculated separately for each run of each participant. Linear-modeling was used to analyze the blood-oxygen-level-dependent (BOLD) contrast from preprocessed fMRI data and included three levels. At the first level, the contrasts for ‘Leg > Rest, Mirror > Rest, and Mirror > Leg’ were calculated separately for each run of each participant. Then, at the second level, one fixed-effects analysis was conducted, including all runs of each participant. A third level Mixed Effect Analysis of Variance test was performed to get brain activity at the group level to further investigate the relationship between clinical variables and peak activity cluster within the ROI. Both on the lower-level and higher-level analyses, Z (Gaussianised T/F) statistic images were thresholded using small volume correction and clusters determined with Z>2.3 voxel-wise thresholding and a family-wise error-corrected (FWE) cluster significance threshold of p=0.05 ^30,31 and included age and gender as covariates. This thresholding occurs in two steps, one at the z-stats level (z> 2.3) and cluster-level correction looking for contiguous voxels that are activated and compared using a cluster probability test (p<0.05). Where brain activity survived thresholding, this third level yielded for each contrast of interest a significant cluster (table 2) that was then used as a mask to extract the percent signal change from that cluster by obtaining the product of the event block height times and the parameter estimate for that event block from each participant’s using FSL’s Featquery tool.

Figure 3.

A priori regions of interest (ROIs). Shown are masks created from the corresponding primary cortices as defined by the Jülich histological (cyto- and myelo-architectonic) atlas²⁷ with a threshold at 85% probability to remove any non-sensory-motor voxels²⁹. Visual cortex (blue), motor cortex (yellow), somatosensory cortex (red).

Table 2.

Peak voxel activation for all experiments

Contrast a b	ROI	Cluster size	MNI Peak coordinate (mm) X, Y, Z c	Z-Scored
Mirror>Rest
	M1 – Affected	129	2, −38, 72	2.88
	M1 – Unaffected	528	−8, −24, 66	3.48
	V1 – Affected	643	22, −104, 0	3.79
	V1 – Unaffected	451	−8, −104, 0	3.69
Mirror>Leg
	M1 – Affected	633	24, −34, 72	3.05
	M1 – Unaffected	721	−2, −26, 72	2.83
	S1 – Affected	1425	8, −38, 74	3.98
	S1 – Unaffected	1408	−54, −28, 50	2.82
	V1 – Affected	207	2, −102, −6	2.98
	V1 – Unaffected	108	0, −104, −6	3.04
Leg>Rest
	M1 – Unaffected	52	−14, −26, 70	3.70

Legend: Peak activations reported as percentage BOLD signal change for Mirror and Movement of the Leg analysis conditions. PSG = percent signal change; M1 = Brodmann area 4 (BA4); S1 = BA1; V1 = BA17; Affected= right hemisphere and unaffected = left hemisphere.

^a= cluster corrected with threshold z>2.3, p<0.05;

^b= small-volume-correction using family-wise error correction (FWE) with a p<0.05 as implemented in FSL.

^cMNI coordinates of the voxel of maximal statistical significance within each region.

^dZ-score of the peak voxel of the significant cluster of maximal statistical significance within each region of interest after family-wise error correction for multiple comparisons (p < 0.05, voxel level) circumscribed using the small volume correction approach.

Complete case analysis was used to handle missing data and median and interquartile range to deal with outliers when the data was not normally distributed. The statistical analyses were performed with Stata version 15.1. to evaluate normality using Shapiro-Wilkś test, skewness and kurtosis. Baseline characteristics were reported using descriptive statistics (e.g., mean and SD for continuous variables, frequency tabulations for categorical variables). For group comparisons comparing behavioral and demographic data Mann-Whitney-U-Tests and for correlations Spearman Rank and Pearson‟s Correlation Tests were used. Linear regression models were built to test our main hypothesis, and Q-Q plots and residuals analysis were performed to review assumptions of the model.

2.9. Shift in cortical representation

A shift in the cortical representation was calculated using the formula for the Euclidean distance described at Foell et al., 2013. The Euclidean distance is the “ordinary” (i.e. straight-line) distance between two points in Euclidean space. With this distance, Euclidean space becomes a metric space. The shift in cortical organization caused by the amputation was defined as the Euclidean distance between the peak activity locations under the assumption that both locations lie in the same hemisphere. This was performed by the peak activation coordinates for ‘Leg > Rest’ on the Left hemisphere (e.g. PLP004 = −16, −32,70) these are in MNI coordinates, followed by the peak activation for ‘Mirror > Rest’ on the Right hemisphere (e.g. 52, −6. 42) and flipping the X coordinate for the Right hemisphere (e.g. −52, −6, 42) multiplying the x-coordinate of the right HM peak with −1. We then could calculate the Euclidean distance using the Euclidean distance formula: $\sqrt{}\left[{(x1-x0)}^2+{(y1-y0)}^2+{(z1-z0)}^2\right]$ within SciPy scientific computing library for the Python³² for peak activation for M1 area.

3. RESULTS

3.1. General characteristics of the sample

Between February 2016 and July 2019, 18 patients with lower limb amputation experiencing phantom limb pain were included in this analysis (table 1). A total of 10 men and 8 women participants fulfilled the criteria, with age ranging from 21 to 72 years old (median = 56.5; IQR= 15). The mean PLP intensity was 5 (SD 1.59). The level of amputation was either above the knee (8 subjects), and below the knee (10 subjects), 7 subjects were receiving opioids as treatment by the time of enrollment (38.8%) in the study, 8 subjects reported pain on the amputated leg before the amputation (44.4%), 13 subjects received previous treatment for PLP (72.2%), and 5 subjects reported that the treatment produced some relief on PLP (27.7%). Further clinical data related to the amputation is provided in Table 1. Interestingly, most common non-painful sensations were movement and electric sensations (figure 4).

Figure 4.

Incidence of phantom limb sensation demonstrating the occurrence of more than one type of sensation per participant.

3.2. Task-related functional activation

We included 18 participants with a total of 67 runs (13 participants with four runs and 5 participants with three runs) in the final analysis. Table 1 in supplement depicts the magnitude of motion and the excluded runs per individual.

The group results of the Mirror and Leg tasks were contrasted separately with the Rest condition to evaluate the task-related effects. In addition, the Mirror condition was contrasted with the Leg conditions, in order to reveal the specific effect of observing the intact leg movement and imagining that this is the amputated leg (observation and imagery) while subtracting the interference of leg movement per se (motor execution with eyes closed).

The significant cluster are presented in Table 2 and Figure 5, as well as the coordinates for the peak of activation, cluster size and z scores across each contrast.

Figure 5.

Group analysis for all patients. The normalized fMRI data (thresholded using small volume correction and clusters determined with Z>2.3 voxel-wise thresholding and a family-wise error-corrected cluster significance threshold of p=0.05) are projected on the standard Montreal Neurological Institute (MNI) brain. Shown are contrasts for Mirror>Rest, Mirror>Leg and Leg>rest for somatosensory cortex (S1), motor cortex (M1) and Visual cortex (V1). The right hemisphere always shows contralateral to the amputation side (i.e. R = affected hemisphere). Therefore, right-sided amputations were x-flipped before preprocessing.

a) In the mirror task we showed bilaterally activity in M1 and V1, but not in S1. By subdividing M1 into anterior and posterior areas we found bilaterally activity in the anterior area of M1 but not in the posterior: anterior affected (cluster size: 97; coordinate: 4, −38, 72; z-score: 2.30) and anterior unaffected (cluster 282; coordinates −8, −4, 66; z-score 2.32). In the Mirror>Leg contrast, M1 revealed activation bilaterally in both sub-areas: anterior affected (cluster size: 89; coordinate: 44, −14, 36; z-score: 2.31) and posterior affected (cluster 115; coordinates 36, −26, 58; z-score 2.30); anterior unaffected (cluster size: 167; coordinate: −36, −18, 56; z-score: 2.4) and posterior unaffected (cluster 144; coordinates −38, −20, 50; z-score 2.52). This pattern of activation reveals a critical engagement of the anterior portion of M1 in the mirror tasks compared with an additive engagement of the posterior portion of M1 when it comes to the Mirror>Leg contrast.

In the V1, clusters are located in the middle of area 17 - primary visual cortex (and parts of the BA18 - V2 secondary visual cortex and BA19 - V3 tertiary visual cortex) for both hemispheres overlapping with extrastriate areas. These areas border with inferior parietal cortex known to integrate information from different sensory modalities. Post-hoc analyzes revealed that associative motor and visual regions [M2 (BA6), V2 (BA 18) and V3 (BA19)] were bilaterally activated in the mirror task (Supplement table 2).

b) The contrast of leg execution-only (eyes closed) is shown as an experimental control. A significant cluster is observed within the unaffected M1 (contralateral to the intact leg). This cluster solely appeared in the anterior portion of the unaffected M1 (cluster size: 36; coordinate: −14, −26, 70; z-score: 2.33). More importantly, peak coordinate activation was located at the cluster in the mesial part of the primary motor cortex (BA4) surrounding the leg area (JuBrain Cytoarchitectonic Atlas Viewer³³). This pattern of activation was also seen in unaffected M2 (Supplement table 2). No significant clusters survived to the threshold in the remaining areas of interest.

Phantom leg movements and shift in cortical representation

We focused on the pattern of activation resulting from the phantom leg movement observation and imagination - contrast Mirror > Leg - which highlights the observation of the leg movement in the mirror as being the amputated leg contrasting with the brain effect of the intact leg movement. We showed a widespread pattern of activation bilaterally within the 3 ROIs (M1, S1 and V1), with higher percentage of signal change in the affected primary motor and somatosensory regions compared to the unaffected hemisphere.

For M1, we observed that the significant clusters are located in the precentral gyrus, more specifically in the primary motor cortex (BA4) towards the pre-motor cortex (BA6) for the affected hemisphere that encompass the somato-sensory cortex; and for the unaffected HM the location is more mesial towards the leg area in BA4 neighboring with pre-motor area. Besides that, the measurement of the peak of activation in M1 region of the affected HM compared to the M1 of the unaffected HM revealed a shift of 12.2 mm.

Collectively, there has been an evident pattern across contrasts from Leg>Rest circumscribed to the unaffected leg area and the Mirror>Leg revealing a broader engagement of brain regions (including all regions analyzed). Consequently, the Mirror>Rest contrast having the higher z-score in the unaffected M1 compared to the unaffected M1 of the Mirror>Leg congruent with the presence of leg movement in the latter.

3.3. Neural correlates of clinical symptoms.

The significant percentage of BOLD signal change (PSC) from clusters of voxels within the regions were correlated with behavioral markers of phantom pain. Table 3 detail the results. One important result to highlight is the lack of correlation between PLP intensity and PSC in the affected sensory-motor areas (M1, p= 0.47 and S1, p=0.09).

Table 3.

Percentage BOLD signal change (PSC) and clinical presentation.

ROI	Contrast	Hemisphere	Result	Significance
M1	Mirror>Rest	Affected	PSC is directly correlated with time since amputation.	Rho: 0.606* p = 0.007
	Mirror>Leg	Affected	Patients who manifested itching had lower PSC than who did not.	p = 0.042**
	Mirror>Leg	Unaffected	Patients who manifested itching had lower PSC than who did not.	p = 0.002**
S1	Mirror > Leg	Affected	PSC is directly correlated with time since amputation.	Rho = 0.522* p = 0.026
V1	Mirror>Leg	Affected	PSC is inversely correlated with PLP	Rho = −0.58* p = 0.013
	Mirror>Leg	Unaffected	PSC is inversely correlated with PLP	Rho = −0.503* p = 0.033
	Mirror>Leg	Unaffected	PSC is inversely correlated with PLS	Rho= −0.529* p = 0.024

Legend: Affected hemisphere: right; unaffected hemisphere: left; percentage BOLD signal change: PSC;

^*Spearman correlation;

^**Wilcoxon rank test.

In the primary motor cortex the percent of signal change in M1 area was lower in the subset of patients who had itching sensation compared to those who did not report itching. This association was found in the contrast Mirror>Leg in the affected (p=0.042) and unaffected HM (p=0.002). By subdividing M1, the activation in the posterior area of the affected hemisphere accounted for this association (p=0.003). In the unaffected hemisphere this association was significant in the anterior (p=0.013) and posterior area of M1 (p=0.011). Yet the post-hoc analysis with the mask over M2 (BA 6 - encompassing supplementary motor area and pre-motor cortex) did not yield significant correlation with itching sensation.

Moreover, the functional cortical shift observed in the M1 leg areas has a moderate inverse correlation with time since amputation (rho = −0.66; p=0.002) indicating a possible enhanced cortical organization of the affected hemisphere among patients with the longer period since amputation. Accordingly, the longer timespan since amputation was correlated with higher activation in M1 (contrast Mirror>Rest; rho = 0.606; p=0.007) circumscribed to the anterior area of M1 (rho = 0.679; p=0.002) and S1 (contrast Mirror>Leg; rho = 0.622; p=0.026) regions in the affected hemisphere corroborating the association between a more organized and effective network (optimized plasticity) in patients who had longer recovery period. Interestingly, the activation in the affected M2 (BA6) region presented similar direct correlation with time since amputation (contrast Mirror>Rest; rho = 0.531; p=0.023). Neither the M1 (and M2) activation nor the shift in cortical representation was correlated with intensity of PLP.

Similarly, increase in percentage signal change in the S1 region was correlated with longer time since amputation (contrast Mirror>Leg Rho = 0.522; p = 0.026). This was the only significant correlation for S1 area.

Finally, we found significant correlations within V1. Through the contrast Mirror>Leg, the percent of signal change in V1 was inversely correlated with PLP intensity (measured by VAS) for the affected (Rho = −0.58; p = 0.013) and unaffected (Rho = −0.503; p = 0.033) hemispheres. These variables were further explored using univariate linear regression looking at V1 activation as potential predictor for PLP. Although on the edge they were non-significant for affected V1 (p=0.051) and unaffected (p=0.065) HM. Besides the correlation with PLP, the V1 percent of signal change in the unaffected HM (Rho = −0.529; p = 0.024) was inversely correlated with the intensity of phantom limb sensation (measured by VAS). Post-hoc tests explored the correlation between V2 (BA 18) and V3 (BA19) with PLP and PLS. The contrast Mirror>Leg revealed a percent of signal change in V2 and V3 inversely correlated with PLP intensity for the affected (V2: Rho = −0.665; p = 0.002; V3: Rho = −0.511; p = 0.030) and V3 for unaffected (Rho = −0.505; p = 0.034) hemisphere. These results replicated the findings for V1 and PLP, however no significant correlation has arisen with PLS.

4. DISCUSSION

The present study of baseline data from a major full-factorial clinical trial on PLP examined neural activation (percentage BOLD signal change: PSC) patterns elicited by motor execution and motor observation/imagery. The effects of these conditions on brain function were explored within three prior hypothesized ROIs (M1,S1 and V1) bilaterally. M1 area was further subdivided into anterior and posterior areas following the rationale that these sub-areas have different quantitative cytoarchitecture and functionality^25,26. The results confirmed our hypotheses: a) sensory-motor and visual areas have been significantly activated in a task-dependent pattern; b) there is a shift in functional cortical representation in M1, and c) the changes in “activation” and the shift in representation were correlated with clinical characteristics of patients with PLP. The results for the clinical analyses are exploratory since they were not corrected for multiple comparison.

4.1. Correlation between PLP intensity and BOLD signal changes

One important negative result is the lack of correlation between PLP intensity and PSC in the sensory-motor areas. Even thought, presence of PLP has been associated with motor and sensory cortex reorganization^1,11 as well as with the preservation of phantom cortex organization and function¹⁴ this is not a surprising result. In fact, several studies (Makin¹⁶ and Kikkert and colleagues³⁴) and a recent systematic-review also showed that pain intensity does not seem to be correlated with sensory-motor cortex reorganization³⁵.

Our results show that there might be a dissociation between the role of sensory-motor cortex reorganization and pain intensity, what could indicate that functional reorganization changes in sensory-motor areas are not necessarily associated with perceptional changes. This suggests that other central mechanisms might be involved in the degree of pain intensity in PLP patients. Interestingly, recent findings showed that pain threshold and pain sensitivity are not correlated with the presence or absence of PLP. However, there is a strong correlation between higher pain intensity and lower heat pain threshold in PLP patients that is not observed in amputees without PLP³⁶. In other words, reorganization may be needed for PLP to develop; however, the intensity of pain seems to be associated with other factors such as activation of sensory-processing and affective-emotional neural networks^37–39.

Notably, we showed that patients who presented higher activation in the visual cortex had lower PLP and PLS intensity. These findings pinpoint that the recruitment of the visual network is related to the individual’s capability of generating kinesthetic motor imagery of own body movement⁴⁰ . Accordingly, our post-hoc analysis showed similar correlation in the extrastriate cortex (BA 18 and BA19) outlining a broader engagement of visual associative areas exerting multimodal integrating functions. Indeed, increased activity in visual cortex demonstrates ones efficiency of incorporating visual inputs in tasks using similar approach and have been extensively explored for motor impairment^41–43, and phantom pain using mirror therapy^44,45 and using graded motor imagery^46,47 Likewise, this concept suggests the important role of self-perception and corporeal awareness of body integrity that could be optimized with therapies using mirrored imagine of intact limb^12,48–50.

Besides that, previous studies showed that pain sensitization is not restricted to mechanisms acting on nociceptive pathways and may involve cortical mechanisms (likely thalamic-related) affecting other sensory modalities (auditory, proprioceptive and visual) particularly if the information conveyed by that sensory system is relevant for the injured body part (e.g. visual input predicting imminent contact of an object with that body part)⁵¹. Supporting this idea, patients with chronic pain often report misperception to a broad range of stimuli, including non-somatosensory stimuli^52,53. The visual network activation could represent a surrogate marker of the thalamic alteration associated with chronic sensitization. Then, optimizing multimodal sensory processing in brain areas of the pain matrix could ultimately inhibit phantom limb pain and sensations.

4.2. Itching: a protective mechanism against PLP?

An interesting result in our exploratory analyses is that patients with itching had lower activation bilaterally in the unaffected primary motor cortex and posteriorly in the affected as compared with subjects without itching. The posterior M1 connects mainly with nonmotor regions (e.g. sensory, attentional, and higher order regions). This fact suggests that perhaps itching is a compensatory mechanism for the sensory deafferentation as to decrease neuropathic pain. In a previous trial, we showed that transcranial direct current stimulation (tDCS) over the primary motor cortex seems to disrupt the compensatory itching mechanism in a group of subjects with severe burn⁵⁴. In another study, we showed that decreased beta power in frontal areas in subjects with itching as compared to healthy controls⁵⁵. It seems that patients who are unable to enhance cortical processing rely predominantly on compensatory peripheral afferent mechanisms such as itching sensation. Notwithstanding, one could also hypothesize that the itching sensation has different neural mechanisms as opposed to other sensations and also to pain itself as supported by Mochizuki and colleagues⁵⁶.

4.3. Brain functional changes and time since amputation

Although the shift in the cortical representation of the affected hemisphere (compared to the unaffected) was not correlated with phantom pain intensity we did find an inverse correlation with time since amputation in our exploratory analyses. This means that patients who present longer time span since amputation have an affected anterior region more organized as also showed by our results observed in the M1. The anterior portion of M1 has more network with motor-related areas. Moreover, in a recent study from our group we showed that changes in structural volume of M1 in the “hand” and “leg” representation are also inversely correlated with time since amputation in PLP patients corroborating the functional and Euclidian distance findings here⁵⁷. Thus, it is implicit that changes in cortical reorganization and plasticity in our PLP patients were indeed time-dependent and seem not to be related to pain intensity.

Supporting this idea, Foell et al 2014 investigating maladaptative plasticity in upper limp amputees with PLP found no significant correlation between time since amputation and treatment effects of mirror therapy and no correlation between the amount of reduction in the shift of the lip representation into the primary motor cortex and the pain reduction. Conversely, while Makin and colleagues found small shifts in lip representation contralateral to the missing hand towards, but not invading, the hand area they showed no statistical relationship between cortical reorganization and phantom sensations or pain^15,16.

In fact, emerging evidence suggests that other central (e.g. thalamic alterations, sensory-processing and affective-emotional networks) and peripheral factors (e.g. ectopic activity from neuroma) beyond the shift in brain somatotopic map may be associated to the severity of phantom pain^2,4,10,58. This broader understanding of PLP mechanisms is utmost relevant to enhance neurorehabilitation strategies for affected patients.

4.4. Strengths and Limitations:

This study has a relatively large sample for fMRI studies and we also developed a novel method to perform the mirror task in the scanner. Because of the challenges of working with the amputee population and challenges related to online mirror observation, we chose not to follow a Latin squares design for task randomization to increase task performed reliability. Although intensively addressed, the magnitude of motion during the scans could have increased the noise and somehow influenced the signal of the analysis. The voxel-wise approach by small volume correction might have increased variability in the data (different cluster sizes and activation) that makes it difficult to compare between conditions. Given the exploratory nature of the study and to serve as hypotheses generating we did not correct the clinical correlations for multiple comparisons. This approach could have increased the type 1 error and some results could have been by chance. Though, due to small sample size and relatively low testing we believe these issues do not hinder our findings.

Figure 1:

Description of the experimental tasks performed during fMRI. All tasks were performed for a total of 20 sec and repeated 6 times, for a total of 6 min run. The runs were repeated 4 times.

Highlights.

• Novel task-based functional magnetic resonance imaging approach for phantom pain;

• Assess neural activation within motor, somatosensory and visual cortex

• Activation in M1/S1 and shift in cortex representation not related to PLP.

• V1 activation is negatively correlated with PLP

• Time since amputation and itching are potentially protective factors for PLP.