Neural correlates of tactile hardness intensity perception during active grasping

Participants and ethics approval

A total of twenty-five healthy volunteers (mean age = 24 ± 4 years old, age range = 19–29, nine male and 16 female), with no contraindications against MRI and no history of neurological disorders, participated in this study. All participants were right-handed and had no deficits in tactile processing. Experimental procedures were approved by the ethics committee of the Ulsan National Institute of Science and Technology (UNISTIRB-15-16-A). The study was conducted in accordance with the Declaration of Helsinki. All participants were informed of the study objectives and experimental procedures and voluntarily submitted a written consent form.

Tactile stimuli

The tactile stimuli for the current study had the same physical properties (e.g., shape and surface texture) except their hardness intensity. To satisfy this criterion, we selected a commercially available product (Eggsercizer, Magister Corp.; Chattanooga, TN, USA), primarily used in motor rehabilitation (e.g., grip strengthening). The Eggsercizer is made of synthetic rubber and silicon and has an oval shape with a diameter of 5.1 cm and a height of 7 cm (Figs. 1A and 1B). The stimulus weights were approximately identical (75.24 ± 0.4 g). The physical hardness intensities were quantitatively measured by resistance to indentation using a Durometer Hardness OO-Type device (ASKER/Shore; OO-Type) at KOPTRI (Korea Polymer Testing & Research Institute, Seoul, South Korea). For the experiment, four different stimuli with different hardness intensities (28, 36, 45, and 57 a.u.) were prepared where a higher number indicated a harder stimulus (physical hardness intensity of each stimulus is denoted as H28, H36, H45, and H57, hereafter). Although these intensity values did not map exactly to perceived hardness, in general, tactile hardness intensities of 20 and 50 were reportedly perceived similar to those of chewing gum and solid truck tires, respectively. The stimuli were kept clean after each experiment using a wet tissue until their surface was free of matter.

Experimental procedure

Before the fMRI experiment, participants performed a behavioral task in which they received a pair of stimuli with eyes closed (Fig. 1C). During the behavioral experiment, participants sat on a chair and put their right hands on a desk in front of the chair. To eliminate any visual effects, participants were blindfolded during the entire behavioral experiment. An experimenter placed the stimulus on the palm of the participants’ right hand with the thin side oriented toward the thumb. The stimulus was placed at the palm center with the major axis perpendicular to the fingers except the thumb and participants grasped each stimulus once for 1 s and released it. Participants were instructed to exert a grasping force as constantly as possible for every stimulus. Immediately after participants released the stimulus, the experimenter removed it from participants’ hands and placed the next stimulus after a 5-s inter-stimulus interval. In the instruction session, an experimenter presented to participants a pair of stimuli with the highest (H57) and lowest intensities (H28) to allow participants to perceive the greatest difference in the hardness intensity of the experimental stimuli. Then, a pair of stimuli with the same hardness intensity (randomly selected out of four levels) was presented to participants along with information about the same hardness intensity.

Participants performed four main sessions of the behavioral task in each of which a total of 10 stimulus pairs were presented to participants. This set of 10 pairs consisted of pairs of stimuli with different hardness and four additional pairs with the same hardness level (4C2 + 4 = 10 pairs). In each trial of the behavioral task, participants were presented with one pair of stimuli after the other and grasped each stimulus as described above. Then, they verbally reported a perceived difference. Each participant was allowed to report a difference in their own numerical scale, which would be normalized in the following data analyses (see Data analysis for normalization details). Each main session began with the instruction session. There was a maximum 60-s break between successive sessions. Overall, a total of 40 pairs (10 pairs × 4 repetitions) were presented to participants during the behavioral experiment.

The objective of the fMRI experiment was to identify brain regions related to tactile hardness perception and neural activities representing the hardness intensity during active exploration of presented objects. As in the behavioral experiment, stimuli with four hardness levels were presented to participants. In addition, participants were given a “sham” stimulus as a control task where they executed the same grasping and releasing motions without holding any objects. During this control task, participants were asked to curl their fingers as if they grasped a virtual object in their hands, following the visual cue. Note that participants were instructed to avoid contacting their fingers with the palm to minimize the potential effects of unnecessary tactile sensations.

During the scanning sessions, participants laid comfortably on the MRI bed, leaving their right palm facing up. A total of two or three foam cushions were placed in the space between the head and the head coil to fix the participant’s head, minimizing movement effects. Participants were instructed to grasp the stimuli with constant force, and to focus on stimuli’s hardness during the grasping action. The procedure of each trial is illustrated in Fig. 2. An experimenter stood at the entrance of the magnet bore and presented stimuli along the screen’s instruction. After a 15-s baseline period, participants performed the grasping task by following the instructions shown on a visual display. Each trial comprised five 3-s grasping periods, taking 15 s total (Fig. 2). In each grasping period, the cue “Grab” was presented for 2 s. At the same time, the experimenter placed a stimulus on participants’ hands. Each time participants viewed the ‘Grab’ instruction, they were asked to grasp the stimulus for 1 s and released it afterwards. After 15 s, the stimulus was collected by the experimenter and participants held a resting position for 9 s until the next trial. The same stimulus was grasped repeatedly by participants in a single trial. A total of five stimuli (four hardness levels plus one sham stimulus) were presented five times each in a fully randomized order over 25 trials per session. Participants conducted two sessions overall. Each session took 618 s on average.

MRI Data acquisition and preprocessing

The blood oxygen-level-dependent (BOLD) signal of participants during the tactile hardness perception task was measured using a 3-T fMRI scanner (Magnetom TrioTim, Siemens Healthineers; Erlangen, Germany). Three-dimensional (3D) functional images constructed from 48 slices covered the whole cerebrum (T2*-weighted gradient echo planar imaging, covering the whole depth of brain area, repetition time [TR] = 3,000 ms, echo time [TE] = 30 ms, flip angle = 90°, field of view [FOV] = 192 mm, slice thickness = 3 mm, and voxel size = 2.0 × 2.0 × 3.0 mm³). Anatomical high-resolution images were also acquired (T1-weighted 3D magnetization prepared rapid gradient echo sequence, TR = 2,300 ms, TE = 2.28 ms, flip angle = 8°, FOV = 256 mm, voxel size = 1.0 × 1.0 × 1.0 mm³). Functional images were preprocessed using SPM12 software (Wellcome Department of Imaging Neuroscience; London, UK), following the sequence: slice-timing correction, realignment, co-registration, segmentation, spatial normalization to the Montreal Neurological Institute (MNI) template and smoothing with a 6-mm full-width-half-maximum isotropic Gaussian kernel.

Data analysis

We excluded four participants from all data analyses because of excessive head movements over the maximal translation (2, 2, and 3 mm in x, y, z, respectively). Consequently, behavioral and functional data analyses were performed on 21 participants. To estimate the perceptual magnitude of hardness intensity, participants’ responses were first normalized into a scale from 0 to 10 using the unity-based normalization given by

(1) $$x^{\mathrm{\prime }}={\displaystyle \frac{x-\mathrm{m}\mathrm{i}\mathrm{n}\left(x\right)}{max\left(x\right)-\mathrm{m}\mathrm{i}\mathrm{n}\left(x\right)}\times 10}$$

where x is the original value of participants’ responses and x′ is the normalized value between 0 to 10. This normalization process was performed on each participant’s data.

Then, we examined a possible order effect of stimulus presentation within stimulus pairs (see Behavioral experiment). Specifically, we used a paired t-test to compare the normalized responses between the trials where the same set of stimuli were presented in the opposite order (e.g., [H28, H57] vs. [H57, H28]). In addition, we used a repeated-measures analysis of variance (rmANOVA) to compare the normalized responses among trials involving different levels of tactile hardness.

We performed event-related fMRI analyses of participants’ BOLD responses to tactile hardness perception. Two separate design matrices were used for first-level general linear model (GLM) analyses: one design matrix to find brain regions involved in tactile hardness perception, and the other to find the brain regions that track hardness intensity. In the first design matrix, all stimulus types (four hardness levels plus one sham) were modeled as separate linear regressors. We specified tactile perception events as the time at which participants grasped the stimuli with their hand. All events in each regressor were convolved with the canonical hemodynamic response function. At first-level, contrast images that reflected BOLD response differences between all actual versus sham tactile perceptions were generated for each participant. The second-level model was constructed as a one-sample t-test using the contrast images from the first-level. Where indicated, family-wise error (FWE) was controlled to correct for multiple comparisons.

In the second design matrix, the subjective hardness intensity obtained from the behavioral experiment was added as a parametric modulator to identify the brain regions that track tactile hardness intensity. At first-level, contrast images that reflected whole-brain responses positively and negatively correlated with hardness intensity were generated for each participant. The same strategy as with the first design matrix (contrast analysis described above) was followed for second-level analysis and multiple comparisons correction. For both contrast and parametric modulation analyses, we labeled statistically significant clusters using the automated anatomical labeling toolbox (Tzourio-Mazoyer et al., 2002).

Furthermore, we calculated and depicted the percent signal change (PSC) within the brain regions identified by parametric modulation analysis to better illustrate the region’s responses tracking the stimuli hardness information. PSC was calculated as the BOLD signal ratio in response to stimuli over that without stimulus. This analysis aimed to supplement the finding of brain regions above by confirming the correlation between the PSC in the identified regions and hardness intensity. We set the activated regions determined by the parametric modulation analysis as regions of interest (ROIs) and estimated the PSC of each single ROI in response to each hardness stimulus using the Marsbar toolbox (Brett et al., 2016).

To examine relationships between physical intensities, perceptual responses, and neural responses regarding tactile hardness, we further constructed the dissimilarity matrices of each data type. The PSC in the ROIs above was used as representational neural responses to tactile hardness. As the perceptual responses were normalized in a scale between 0 and 10, neural and physical intensity data were similarly normalized between 0 and 10. A dissimilarity matrix of physical hardness was constructed using the normalized differences between every pair of the four tactile stimuli, resulting in a 4 × 4 symmetric matrix with diagonal entries equal to zero. We constructed the dissimilarity matrix of perceptual responses in the following manner. First, participants’ perceptual responses to each stimulus were normalized within a scale from 0 to 10 using the unity-based normalization given in Eq. (1). Second, the difference of the normalized perceptual responses between every possible pair of stimuli was calculated. Third, the difference values for each pair were averaged over all participants. Fourth, a dissimilarity matrix of perceptual responses was constructed from each averaged difference. The dissimilarity matrix of neural responses of each ROI was also constructed using the normalized PSC difference values averaged across participants but resulting in a 4 × 4 symmetric matrix.

A Mantel test, often used for analyzing functional connectivity in the brain (Glerean et al., 2016), was employed to find statistically significant correlations (Pearson’s correlation) between dissimilarity matrices (Mantel, 1967).

Lastly, we applied multidimensional scaling (MDS) to the dissimilarity matrices to find low-dimensional representations of physical, perceptual, and neural responses. These low-dimensional representations of hardness intensity were meant to visualize similarities between the three different domains of tactile hardness: physical, perceptual, and neural representation of object hardness.

Behavioral responses to tactile stimuli with different hardness intensities

We first confirmed the lack of order effects in perceived difference between the stimuli set. Thus, we did not further consider presentation order in subsequent analyses of behavioral responses. It also led us to define the absolute perceptual differences between a pair of stimuli as the perceived distance of tactile hardness. Then, we examined differences in perceived distance among all possible pairs of tactile stimuli using rmANOVA and observed a significant difference of mean perceived distance among those pairs (F ((9,180)) = 144.62, p < 0.0001) (Fig. 3). A post-hoc Bonferroni test further revealed significant differences in perceived distance between pairs with dissimilar hardness differences (all p < 0.0001; Fig. 3) but no difference between pairs with similar hardness differences. As an additional explorative analysis, we examined potential gender differences in the perceived distance among all possible pairs of tactile stimuli using two sample t-tests, but none of the pairs showed significant difference between male and female participants (Table S1). These results show that participants could accurately perceive objective differences in hardness between pairs of tactile stimuli.

Neural responses to tactile stimuli with different hardness intensities

We examined the BOLD signal evoked by tactile hardness by contrasting each hardness intensity against the “Sham” using GLM analysis. However, we found no commonly activated regions across the whole brain. To examine the brain regions that track the physical properties (hardness intensity) of stimuli, we ran a whole-brain analysis using a design matrix including physical hardness intensity as a parametric modulator. We found a significant positive correlation between hardness intensity and neural responses in the posterior insula in the right hemisphere (rpIns) (MNI space coordinates of x = 40, y = −26, z = 18; FWE corrected p < 0.05, cluster size > 10, cluster defining threshold p < 0.001) (Fig. 4A). A set of brain regions negatively tracked physical hardness intensity (Fig. 5A). And we found a significant negative correlation in the right posterior lobe of the cerebellum (rpCerebellum) (MNI space coordinates of x = 42, y = −74, z = −40; FWE corrected p < 0.05, cluster size > 10, cluster defining threshold p < 0.001) (Fig. 5A).

Dissimilarity analysis

We constructed three dissimilarity matrices of physical hardness intensities, perceived responses, and neural responses to hardness intensities (Fig. 6). For the neural dissimilarity matrix, PSC differences between the rpIns and rpCerebellum were used as a neural representation of tactile hardness.

MDS analysis of the dissimilarity matrices produced two-dimensional (2D) representations of physical, neural, and perceptual hardness intensity (Fig. 7). This 2D depiction clearly visualized the response clustering into four groups according to hardness intensity across all three types of representations (physical, perceptual, and neural). Moreover, it appeared that only one dimension (MDS1) was sufficient to effectively represent hardness intensity, not only for physical stimuli properties but also for neural and perceptual responses.

Perceptual discrimination of tactile hardness

The behavioral results showed that participants could distinguish different levels of tactile hardness when they grasped the presented stimuli. Self-reported perceptual differences between stimuli well-reflected the relative physical difference of objects’ hardness. In our data, stimulus presentation order did not affect the recognition of hardness differences. This might be due to the fact that we only used four different types of stimuli (i.e., the four stimuli resulted in six distinct pairs); participants might have easily learned the minimum and maximum hardness at the beginning of each session which might helped with accurate estimation. Nonetheless, we prepared stimuli with clearly separated hardness for this study because we wanted to explore neural substrates of well explained tactile perception.

Brain regions representing tactile hardness intensity

By comparing brain responses to stimuli against sham, we attempted to find brain regions involved in hardness intensity perception. Because all stimuli had the same shape and surface texture, we expected to identify the brain regions commonly involved in different hardness intensity perception by contrasting each hardness intensity against the sham trials. However, there was no such commonly responsive brain region, including the SI or SII. A possibility is that the SI might have been involved in sham trials due to participants’ imaginary tactile perception. Previous rodent studies showed similar prominent activation in both the MI and SI during motor control (Diamond et al., 2008; Matyas et al., 2010). In our task, hand movements were dominant in all trials because participants perceived hardness via active exploration, leading to SI activations in response to both sham and tactile stimulations.

We found that neural responses in the rpIns increased as a function of perceived tactile hardness. Conversely, neural activity in the rpCerebellum decreased with increased perceived tactile hardness. The rpIns has been reported to be involved in tactile pressure and proprioceptive perception. Previous studies reported bilateral posterior insula activation during a tactile hardness perception task (Lederman et al., 2001) and the involvement of the posterior insula in tactile and body displacement perception and bodily self-consciousness (Richer et al., 1993; Blefari et al., 2017; Findlater et al., 2018), as well as activation in response to tactile pressure stimulation (Chung et al., 2015). Based on previous lesion studies, the posterior insula is known to be involved in self-awareness of actions (Karnath & Baier, 2010). Previous research backs up our finding that rpIns might be engaged in the perception of hardness intensity during active grasping. Our results provide additional evidence that the human posterior insula might play a role in the tactile perception of hardness intensity. While we found activation only in the rpIns, a more liberal statistical threshold (p < 0.001, uncorrected) revealed involvement of the bilateral posterior insula. Future studies should examine the functional connectivity between the bilateral posterior insula and other brain regions, such as the SI, to understand why rpIns showed stronger responses than the contralateral hemisphere.

One of the most noticeable findings in the current study is that the cerebellum was activated during active perception of tactile hardness. A number of human neuroimaging studies has shown that the cerebellum is functionally connected to the sensorimotor network combining the sensorimotor cortex, premotor area, and supplementary motor area. In addition, the cerebellum is known to participate in motor control, motor learning, and motor planning (Habas et al., 2009; Krienen & Buckner, 2009; O’Reilly et al., 2010; Zeeuw & Brinke, 2015; Gao et al., 2018). In particular, the rpCerebellum is known to be engaged in self-produced tactile perception (Blakemore, Wolper & Frith, 1999), which may be necessary in to recognize hardness intensity. According to a recent study, functional connectivity between the primary somatosensory cortex and the cerebellum was strengthened when participants reported perceptual attenuations of self-generated touch (Kilteni & Ehrsson, 2020). These results suggest that the ipsilateral cerebellum has a crucial role in predicting self-generated touch. Our results showed that cerebellar activity is associated with perceived tactile hardness during active grasping, and are in line with the aforementioned previous findings.

Relationship between neural responses, physical, and perceptual tactile hardness

The Mantel test and MDS analysis were conducted to intuitively illustrate the relationship among physical, perceptual, and neural representations of tactile hardness intensity. The Mantel test results suggested similar perception of different levels of hardness intensity to neural responses, which were marginally correlated with the physical hardness of stimuli, suggesting that the brain tracks the subjective perception of objects’ hardness rather than their physical hardness. The MDS analysis showed similar clustering of physical, perceptual, and neural representations of each hardness level. This confirms that neural responses in both posterior insula and cerebellar regions represented perceived hardness.

Limitations

This study has a few limitations. First, we assumed that individuals show stable perceptual responses. In the behavioral experiments, we could identify the relationship between the physical and the perceived hardness for each participant. During the scanning phase, we did not directly collect participants’ subjective responses because we intended for participants to fully concentrate on the grasping motions. Under our experimental environment, it was difficult to obtain a direct response from participants during MRI scanning. During neural image acquisition, participants explored stimuli with their right hand and held a squeeze ball for emergency with their left hand, which prohibited them from making behavioral responses with button pressing. Alternatively, we attempted to receive verbal responses directly during scanning, but the scanner noise was too loud. As such, we decided to separate behavioral and neural imaging experiments while ensuring that participants maintained the same active grasping across both experiments. Therefore, the question whether the neural activations measured in the scanner would be the same as those outside remains unanswered. Similarly, albeit very unlikely, it is still unclear whether participants experienced the same hardness perception outside the scanner as inside. To minimize this difference, we trained participants to grasp the stimuli with constant force before conducting the main experiments. Future studies could use MRI-compatible electromyography (EMG) or force sensors that can measure individuals’ grasping force during the scanning phase.

Second, our findings may be limited by the fact that only four hardness intensities were used. We observed that neural responses in the cerebellum and posterior insula were correlated with perceived hardness. However, due to the small number of stimulus types, our data cannot distinguish subtle differences between models that assume linear versus non-linear (e.g., logarithmic, quadratic) associations of neural and perceptual responses.

Third, there is a chance that the neural responses measured in this study might also reflect the changing reaction force applied to the hand with hardness. To minimize this confounding effect, we intended to ensure that participants grasped the stimuli with a constant force. However, it is still impossible to determine whether participants exerted exactly the same force during the experiment. In an effort to bypass this factor, we used subjective ratings of hardness levels rather than physical hardness, so that the self-reported rates should already take the participants’ own perception of exerted force into account. Still, to further optimize the task design, future studies could use a visual feedback that shows an EMG-based gage to inform participants of their exerted force in real-time. In addition, in an experiment with real-time EMG, we could observe neural responses during tactile information processing using EEG or MEG with finer time resolution (Golnaz et al., 2020; Hagiwara et al., 2020), which would help confirm the relationship between muscle strength and neural responses in real-time.

Fourth, our experimental design solely focused on active exploration of tactile stimuli. Thus, potential individual differences in passive tactile perception may exist. Note that we controlled for such a confounding factor by examining neural responses parametrically modulated as a function of perceived stimulus hardness.

Fifth, and lastly, it is possible that the marginal association between neural activity and physical hardness could be partially due to insufficient statistical power (the sample size in this study was pre-determine based on that used in similar studies using functional neuroimaging in tactile perception (Case et al., 2016)). Our task design and instruction do not rule out the possibility that neural responses would encode physical hardness, but our results suggest that the neural responses elicited in our active grasping task are more associated with perceptual object discrimination rather than tracking of physical hardness per se.

Also, from our explorative analyses addressing potential gender differences, inconsistent with a previous report (Abdouni et al., 2018), we did not find any differences in either behavioral or neural results. Still, as this is an exploratory approach, the interpretation of this result should be cautious. Such an inconsistency could be due to lack of power, given our small sample size for each gender group, and thus, future studies would be necessary to further examine the potential effects of gender differences in tactile sensation.