2.2.1 Behavioral assessment.

Somatosensory function of the affected upper limb was assessed using the Erasmus MC modified (revised) Nottingham Sensory Assessment (Em-NSA), stereognosis subscale of the original NSA, two-point discrimination test (2PD), and perceptual threshold of touch (PTT). To assure standardization, behavioral assessment was performed by one trained researcher. A detailed description of the behavioral assessment protocol can be found in our previously published work [30]. All behavioral assessment tools are reliable and valid measurements [31–34].

The Em-NSA comprises four subscales: light touch, pressure, pinprick, and sharp-dull discrimination [34]. A total score between 0 and 8 was obtained for each modality. A cut-off score of <7 out of 8 for each subscale was defined as impaired function. A total score between 0 and 32 was calculated by summing all subscores. The original NSA was used to assess the stereognosis function [31] by providing 11 commonly-used objects in the affected hand leading to a total score of 22. The cut-off score to define a stereognosis deficit was <19 out of 22 [35]. Two-point discrimination [32] tests the ability to detect two different stimuli that are simultaneously applied at the fingertip of the index finger. A cut-off score of >5mm was defined as impaired two-point discrimination function [36]. Finally, the perceptual threshold of touch [33] assesses the minimal level of detection of a touch stimulus [33] by applying transcutaneous electrical nerve stimulation (TENS) with a CEFAR Primo Pro (cefar medical AB, Sweden). The scores were compared to age- and gender-matched norm values [37].

2.2.2 Statistical analysis of the behavioral assessments.

Baseline characteristics and behavioral assessments were analyzed using SPSS version 23. Descriptive analyses were performed using median with interquartile range (IQR) and frequencies with percentages, as appropriate. The prevalence of different somatosensory deficits such as light touch or sharp-dull discrimination impairment were calculated using frequencies with percentages. Therefore, the different somatosensory variables were dichotomized according to the presence of a deficit or normal functioning, based on the above-mentioned pre-defined cut-off values.

2.2.3 Brain imaging data acquisition.

Anatomical and resting-state fMRI images of all patients were acquired on the same 3.0 Tesla Philips MR scanner (Best, The Netherlands) with an 8-channel phased-array head coil. Scan sessions started with the acquisition of the anatomical scan, followed by the resting-state fMRI scan. Anatomical images were acquired using fluid-attenuated inversion recovery imaging (FLAIR) with the following parameter settings: echo time (ET) = 350ms, repetition time (TR) = 4800ms, inversion time = 1650ms, field of view (FOV) = 250x250mm², slice thickness = 1.12mm and interslice gap = 0.56mm. The resting state fMRI images consist of a total of 30 parallel transverse orientated slices of 4mm thickness with no interslice gap. Parameter settings were: TE = 33ms, flip angle 90°, TR = 1700ms, FOV = 230mm, matrix = 64x64, duration 7 min. During scanning, participants were lying in supine position with eyes closed and were instructed to think of nothing in particular and to not fall asleep. Based on an informal survey, no patients have fallen asleep during scanning.

2.2.4 Preprocessing of resting state data.

Co-registration and delineation of the lesions, white matter and cerebrospinal fluid were acquired form the T1-structural images to implement as confounding factor in the pre-processing of the resting state fMRI data. The process of lesion segmentation was reported in detail in a previous study of our research group [38]. In short, delineation of the lesion was manually performed in several steps by an experienced rater (SK) to ensure accuracy of the lesion delineation. Linear transformation was applied to transform the individual FLAIR data and lesion segmentation into a standardized brain template in MNI space (Montreal Neurological Institute).

Preprocessing of resting state fMRI images was performed with SPM-8 (Wellcome Department of Imaging Neuroscience, London, UK) and the CONN functional connectivity toolbox [39] both implemented in Matlab R2008a (Mathworks). Preprocessing of the data included slice timing correction through interpolation to the middle slices (reference = 17). Then, the functional images were spatially realigned and the anatomical T1-image was co-registered to the mean functional image to map functional information into anatomical space. Anatomical and functional images were then normalized to the standard EPI-template of the Montreal Neurological Institute (MNI-152), resampled into 3-mm isotropic voxels and smoothed (Gaussian smoothing; isotropic 5mm full-width-at-half-maximum).

Realignment parameters were included as regressors of no interest. The CompCor strategy [40] was applied in the CONN functional connectivity toolbox to correct for confounding factors in white matter and cerebrospinal fluid. Then, the residual time series of the resting state images were band-pass filtered (0.009< f <0.08Hz). Since even micromovements can already influence intrinsic functional connectivity [41, 42], mean frame-wise displacement (FD) was calculated for each patient and included as a nuisance covariate in the group analyses. Note that patients with excessive head motion, defined as mean FD scores exceeding 0.5 were not included in the final analyses (8 out of the initial 27 patients were excluded). Additionally, to account for lesions, region-to-region functional connectivity analyses were performed with and without regions overlapping with lesion-delineated maps.

2.2.5 Resting-state connectivity analysis.

To delineate the functional network relevant for somatosensory processing, twelve bilateral Regions of Interest (ROIs) (6 left, 6 right) were adopted from a task-based fMRI study of Chang and colleagues, exploring sensory processing in nine healthy participants[43]. The central coordinates of the predefined spherical ROIs (6 mm diameter) are reported in Table 1.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. MNI coordinates of the twelve regions-of-interests (ROIs) of the somatosensory processing network adopted to explore ROI-to-ROI resting-state functional connectivity.

https://doi.org/10.1371/journal.pone.0205693.t001

For each participant, we extracted the mean time series by averaging across all voxels in each ROI. We then computed bivariate correlation coefficients for each pair of ROIs. In the primary analysis, only ROI-to-ROI correlation coefficients were included for connections that showed no overlap with the predefined lesion [11]. MRIcron (software package version 1, 2015) was used to identify any overlap between the standardized lesion delineations and any of the predefined regions-of-interest (ROIs). Across patients, ROI-to-ROI correlation coefficients were therefore not included in the primary analysis for 26 connections due to overlap with the delineated lesions (average of 2.25 connections per patient). A group lesion overlay image is provided in Fig 1. (see S1 Table for detailed list of connections that showed overlap with the lesions of each patient). Note that secondary analyses were performed on the entire set of ROI-to-ROI correlation coefficients (i.e., also including the 26 connections that originated from ROIs in the predefined lesions) to determine the robustness of the main results of the primary analysis.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Group lesion overlay image.

Fig 1 displays the lesion location of all patients. All lesions were flipped to the right site. Color indicates the lesion locations from red (not lesioned in any patient) to yellow (lesioned in most patients).

https://doi.org/10.1371/journal.pone.0205693.g001

The resultant ROI-to-ROI correlation coefficients were Fisher z-transformed and extracted to perform group-level analyses and brain-to-behaviour correlation analyses in SPSS version 23.

Related to our a-priori hypotheses on alterations in intra- and interhemispheric connectivity, the following indices of network connectivity were calculated:

1. To obtain an index of intrahemispheric network connectivity of the ipsilesional and contralesional hemisphere, ROI-to-ROI correlation coefficients of all eligible connections (i.e., outside the lesion) in the affected (6 x 6 connections) and unaffected hemisphere (6 x 6 connections), respectively, were averaged separately for each participant.
2. To obtain an index of interhemispheric network connectivity, ROI-to-ROI correlation coefficients between homologue ROIs (total of 6 interhemispheric connections) were averaged for each participant.

Next, for all indices of functional network connectivity, Shapiro-Wilk normality tests were performed to assess normality of distribution. Normality was not confirmed for the functional connectivity index of the ipsilesional intrahemispheric network due to one extreme outlier. Particularly, one patient showed an extreme ipsilesional intrahemispheric connectivity score (defined as smaller than Q1 ± 3(Q3-Q1), with Q1 and Q3 being the first and third quartile). All primary analyses were performed after removal of the outlier data point, however, for completeness; secondary analyses were performed with inclusion of the outlier to assess the robustness of the primary results. Two groups were created based on the median value of the total score on Em-NSA, to explore whether indices of functional network connectivity were significantly different between patients with severe somatosensory impairments (n = 9) (Em-NSA score of <13 out of 32) and patients with mild to moderate somatosensory impairments (n = 10) (Em-NSA ≥13 out of 32). First, mean and standard deviations of functional connectivity values of each index were calculated for the total group and for each subgroup (mild to moderate and severe). Further, to explore differences in functional connectivity between subgroups, a general linear model was constructed with the index of network connectivity as dependent variable; ‘group’ (severe, mild to moderate) as between-subject factor, and age, centre (stroke unit where data were collected) and mean framewise displacement (FD) as covariates of no interest (nuisance regressors). Additionally, to investigate potential brain-behavioral associations between the indices of functional network connectivity and scores of the behavioral somatosensory assessments, non-parametric partial correlations were calculated also including age, center and mean FD as nuisance regressors. Non-parametric spearman rho correlation tests were adopted due to ordinal scoring of the behavioral somatosensory measures. Brain-behavior associations were assessed with the following behavioral somatosensory assessments; (i) light touch, (ii) pressure, (iii) pinprick, (iv) sharp-dull discrimination, (v) stereognosis, (vi) two-point discrimination, and (vii) perceptual threshold of touch. Residuals of the partial correlations were extracted and plotted against the behavioral assessments, when significant.

3.2.1 Group differences in somatosensory network connectivity between patients with severe and mild to moderate somatosensory impairments.

Patients with severe somatosensory impairments (n = 9) showed significantly lower intrinsic functional connectivity compared to patients with mild to moderate somatosensory impairments (n = 10) in terms of (i) intrahemispheric connectivity of the ipsilesional hemisphere (F(1,18) = 5.667; p = 0.035); and (ii) interhemispheric connectivity between homologue regions (F(1,19) = 17.088; p = 0.001). Group differences were also found in terms of intrahemispheric connectivity of the contralesional hemisphere; lower connectivity scores were revealed in the patient group with severe compared to mild impairment, albeit not significant (F(1,19) = 2.011; p = 0.180) (Fig 3).

When analyses were performed across all connections (i.e., ‘whole-network’ functional connectivity combining all inter- and intra-hemispheric connections (12 x 12 connections)) a significant reduction in functional connectivity of the somatosensory network was revealed indicating lower connectivity in the severe, compared to the mild affected patient group (F(1,19) = 9.945; p = 0.008).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. Group differences in somatosensory network connectivity between patients with severe and mild to moderate somatosensory impairment.

Fig 3 shows the differences in functional connectivity between subgroups, investigated by a general linear model with the index of network connectivity as dependent variable; ‘group’ (severe, mild to moderate) as between-subject factor, and age, centre (stroke unit where data were collected) and mean framewise displacement (FD) as covariates of no interest (nuisance regressors). Patients with severe somatosensory impairments (Em-NSA <13/32) displayed significantly lower network intrinsic connectivity in the ipsilesional intrahemispheric and interhemispheric network compared to patients with mild to moderate somatosensory impairments (Em-NSA ≥13/32). Networks are displayed in frontal plane in back view; Error bars display the Standard Error of mean; *p< 0.05; ** p<0.01; ***p<0.001 (gray scale, 2-collum fitting).

https://doi.org/10.1371/journal.pone.0205693.g003

3.2.2 Associations between connectivity of the somatosensory network and somatosensory impairments.

Brain-behavioral relationships were assessed between the indices of network connectivity and behavioral assessments of somatosensory impairment. Similar to the categorical group-related analysis, a general pattern emerged, indicating that—on a dimensional scale—patients with more severe somatosensory impairments showed reduced functional connectivity of the somatosensory network. For intrahemispheric connectivity of the contralesional hemisphere, low, non-significant associations were found (r = 0.23–0.42).

As displayed in Table 3, it can be seen that reductions in (i) intrinsic functional connectivity of the ipsilesional hemisphere (r = -0.69; p<0.01), and (ii) interhemispheric connectivity (r = -0.51; p<0.05) were significantly related to somatosensory impairment—in terms of a higher perceptual threshold of touch. Furthermore, reductions in interhemispheric connectivity were also significantly associated with somatosensory impairments in terms of light touch (r = 0.55; p<0.05) and stereognosis (r = 0.64; p<0.01), indicating that patients with more severe impairments showed larger reductions in interhemispheric connectivity. Scatterplots were created to visualize the individual brain-behavior correlations. (Fig 4).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 4. Brain-behaviour associations between specific somatosensory modalities.

Correlation plots between behavioural assessment and residuals of the partial correlations with the index inter- or intrahemispheric functional connectivity. Correlation plot between the perceptual threshold of touch and ipsilesional intrahemispheric network (A) and interhemispheric network (B); correlation plot between interhemispheric network and light touch (C) and stereognosis (D). (gray scale; 2-collum fitting).

https://doi.org/10.1371/journal.pone.0205693.g004

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 3. Partial correlations (non-parametric) between network indices and somatosensory modalities.

https://doi.org/10.1371/journal.pone.0205693.t003

Note however that the dimensional brain-behavior association with perceptual threshold of touch seemed to be largely driven by a categorical clustering of patients with ‘more severe’ or ‘more mild/moderate’ impairment on this behavioral score. Indeed, the dimensional relationship between perceptual threshold of touch and the indices of ipsilesional intrahemispheric connectivity or interhemispheric functional connectivity failed to reach significance when the factor ‘group’ (severe vs mild/moderate) was inserted as an additional nuisance regressor, indicating that the majority of the observed inter-individual variance in functional connectivity was already explained by the categorization of the patients in ‘severe’ versus ‘mild to moderate’ subdivisions.

3.2.3 Secondary analyses.

Secondary analyses were performed to determine whether the main findings of the primary analyses were replicated when connectivity analyses are performed on the entire set of ROI-to-ROI correlation values (i.e., also including connections that originated from ROIs in the predefined lesions). Overall, a similar pattern of results was shown, indicating that more severe somatosensory impairment was significantly associated with reductions in functional connectivity of the somatosensory network (most pronounced for interhemispheric connectivity and intrahemispheric connectivity of the ipsilesional hemisphere). Furthermore, in the primary analysis, the data point of one ‘outlier’ patient was excluded in the analysis of intrahemispheric connectivity of the ipsilesional hemisphere. Secondary analysis with inclusion of this data point replicated all primary results, i.e., indicating that reduced ipsilesional connectivity was associated with more severe somatosensory impairment.

4.1.1 Decreased interhemispheric functional connectivity.

In literature, decreased interhemispheric functional connectivity is described in rats and humans after stroke to be associated with impaired sensorimotor function [6, 47, 50, 52], and our study adds to this concept for the domain of arm and hand somatosensory deficits. Carter et al. [48] presented an association between a disruption in interhemispheric functional connectivity of the sensorimotor network and upper extremity motor impairments in the acute phase post stroke. Furthermore, Bannister et al. [24] explored the association of alterations in functional connectivity with somatosensory impairments; a relationship was reported between decreased interhemispheric functional connectivity between ipsilesional and contralesional primary somatosensory cortices and thalamus and more impaired touch discrimination at one and six months post stroke. Our results extend these prior findings by showing that already at one week after stroke, the extent of altered (reduced) inter-hemispheric functional connectivity of the somatosensory network is associated with more severe impairments in somatosensory function. Together, these observations highlight the behavioural relevance of the integrity of resting-state inter-hemispheric connectivity for both motor and somatosensory function and therefore provide further support to the notion that assessments of restorations/normalizations of these altered interhemispheric connectivity patterns may form a reliable index for evaluating the effectiveness of rehabilitation therapies at the neural level [6, 17, 52].

4.1.2 Intrahemispheric functional connectivity.

In general, decreased ipsilesional and increased contralesional activity is described and associated with impaired function [12, 25, 53–56]. Albeit, several different effects are described for the acute phase. A paper of Rehme et al [14] included increased intrahemispheric functional connectivity in their criteria for classification of upper limb function. In contrast, several studies showed the association between decreased ipsilesional intrahemispheric functional connectivity and impaired sensorimotor function [16, 57]. For example, a longitudinal study of Nijboer and colleagues [57] reported a significant lower functional connectivity between motor areas within the lesioned hemisphere compared to the contralesional hemisphere at five weeks post stroke. Also Park et al. [16] reported a decrease in ipsilesional intrahemispheric functional connectivity between M1 and the sensorimotor cortex at one week post stroke. However, this group also reported increased intrahemispheric functional connectivity between ipsilesional M1 and ipsilesional cerebellum and thalamus [16]. In our results and as anticipated, higher ipsilesional intrahemispheric connectivity was found for the group with mild to moderate somatosensory impairments. On the other hand, and perhaps surprising, functional connectivity within the contralesional hemisphere did not show any significant associations with somatosensory function, neither in the analysis between severe and mild to moderate patients, nor in the nonparametric correlations with different somatosensory modalities. Explanations for this non-conforming finding can be our sample size or lesion location or size of our subjects. For example, in a rat model, only increased functional connectivity was found in rats with extensive lesions [50]. So far, in contrast to knowledge about motor function, literature did not report explicit findings for somatosensory function and associated changes in intrahemispheric functional connectivity. Only the study of Bannister et al. [24] reported alterations in intrahemispheric connectivity in patients with stroke, but it was not specified whether this was different from healthy people or in which direction the alterations were seen. Our results of ipsilesional intrahemispheric functional connectivity were in line with previous studies for motor function. However, we are not aware of studies explicitly reporting associations of somatosensory function with intrahemispheric connectivity.

4.1.3 Associations with somatosensory function.

As hypothesized, we found strong associations between interhemispheric network connectivity indices and somatosensory assessments with significant correlations for PTT, light touch and stereognosis.

As shown in Fig 4, the associations with PTT and light touch were mainly driven by group (mild to moderate vs severe impaired somatosensory function) in contrast to stereognosis, for which correlations remain significant after correction for group. Thus, for PTT and light touch there were rather two groups of impairment levels than a continuum of impairment severity. First, moderate significant correlations were found between the perceptual threshold of touch (PTT) and the index of ipsilesional intra- and interhemispheric functional connectivity. PTT is a standardized measure of light touch performance, by applying transcutaneous electrical nerve stimulation (TENS) on the light touch receptors of the skin, and activating the large myelinated Aβ fibres [33]. In previous work of our research group, higher prevalence of light touch deficits was reported based on PTT compared to other assessment methods [38], highlighting the further relevance of this assessment method for future research when investigating upper limb somatosensory deficits.

The association between stereognosis and interhemispheric connectivity can be explained by the integrative properties of stereognosis. Somatosensory information from the arm and hand will be integrated to obtain information of different properties of the object, such as size, shape, and weight. [58]. Thus, the integration of these properties combined with proprioceptive information could explain the significant correlation between stereognosis and functional connectivity [59]. Our findings therefore extend previous literature by Borstad and colleagues [60], who reported the association between haptic performance and the integrity of interhemispheric tracts connecting bilateral frontoparietal white matter and M1 in chronic stroke patients. As we found strong associations between interhemispheric network connectivity indices and stereognosis.