fNIRS detects temporal lobe response to affective touch

Abstract

Touch plays a crucial role in social–emotional development. Slow, gentle touch applied to hairy skin is processed by C-tactile (CT) nerve fibers. Furthermore, ‘social brain’ regions, such as the posterior superior temporal sulcus (pSTS) have been shown to process CT-targeted touch. Research on the development of these neural mechanisms is scant, yet such knowledge may inform our understanding of the critical role of touch in development and its dysfunction in disorders involving sensory issues, such as autism. The aim of this study was to validate the ability of functional near-infrared spectroscopy (fNIRS), an imaging technique well-suited for use with infants, to measure temporal lobe responses to CT-targeted touch. Healthy adults received brushing to the right forearm (CT) and palm (non-CT) separately, in a block design procedure. We found significant activation in right pSTS and dorsolateral prefrontal cortex to arm > palm touch. In addition, individual differences in autistic traits were related to the magnitude of peak activation within pSTS. These findings demonstrate that fNIRS can detect brain responses to CT-targeted touch and lay the foundation for future work with infant populations that will characterize the development of brain mechanisms for processing CT-targeted touch in typical and atypical populations.

INTRODUCTION

Affective touch, such as that shared between a mother and her infant, plays a critical role in social–emotional development. This caress-like touch, when applied to hairy skin, is processed by a specific type of nerve fiber called C-tactile (CT) afferents (Morrison et al., 2010). Despite an extensive behavioral literature on the importance of touch in early development, researchers have only recently begun to study the neural underpinnings of the CT system in typical adults (e.g. Olausson et al., 2002; Gordon et al., 2013; McGlone et al., 2012), and developmental research on the topic is scant. An understanding of these neural mechanisms and their maturation in typically developing infants and children will improve our understanding of the critical role of touch across the lifespan and may inform study of neurodevelopmental disorders that involve sensory issues, such as autism. Functional near-infrared spectroscopy (fNIRS) is an emerging tool in developmental cognitive neuroscience that can be used with awake and alert infants, toddlers and children (e.g. Taga and Asakawa, 2007; Grossmann and Johnson, 2010; Ichikawa et al., 2010; Lloyd-Fox et al., 2010; Gervain et al. 2011). Although functional magnetic resonance imaging (fMRI) has been used to investigate brain mechanisms that support the processing of CT-targeted affective touch (Olausson et al., 2002; Morrison et al., 2010; Gordon et al., 2013; Voos et al., 2013), no study to date has used fNIRS to examine this system. We sought to establish a paradigm to study the developmental trajectory of brain mechanisms for processing affective touch, targeting the CT system. Based on our fMRI results (Gordon et al., 2013; Voos et al., 2013), we hypothesized posterior temporal lobe involvement in processing CT-targeted touch. Furthermore, we investigated the association between the neural response to such touch and individual differences in autistic traits, based on previous findings of this relationship in healthy adults (Voos et al., 2013).

The skin is the largest and earliest developing sensory system in humans (Montagu, 1971). For example, as young as 8 weeks in utero, a fetus will pull away from an object that touches its face (Hepper, 2002). Touch plays a fundamental role in social development (Atkinson et al., 1982; Maurer and Maurer, 1988), and interpersonal touch is one of the earliest forms of parent–child communication (e.g. Frank, 1957; De Chomaso, 1971; McDaniel and Andersen, 1998). Studies with non-human primates (e.g. Harlow and Zimmermann, 1959) and human infants (e.g. Barnett, 2005) have highlighted the importance of social touch in healthy development and specifically in the development of emotion regulation (Feldman et al., 2003). Mothers exhibit a distinct pattern of slow, affective touching behavior (beginning with fingertips and expanding to full palm) in conjunction with prolonged eye contact when first interacting with their newborns, implicating a specific parental touch response (Rubin, 1963; Klaus et al., 1970). Furthermore, <4 days post-partum, mothers (with eyes and noses blindfolded) can recognize their newborn’s hand solely based on their sense of touch (Kaitz et al., 1992, 1993). Finally, tactile stimulation is soothing to newborns (Birns et al., 1966; Korner and Thoman, 1972) and improves the health of preterm infants by increasing weight gain and caloric intake (Helders et al., 1989; Scafidi et al., 1990). Taken together, these studies highlight the importance of touch in early social interactions. Although the benefits of affective touch have received extensive empirical attention (Stack, 2001), related neurodevelopmental research is sparse. An understanding of these neural mechanisms and their typical development is necessary for a complete understanding of the biological bases supporting the critical role of touch across the lifespan.

A specific type of nerve fiber, CT afferents, has been implicated in processing slow, gentle touch (Morrison et al., 2010), much like the touch described earlier in parent–infant interactions. CT afferents exist only in the hairy skin of mammals and respond specifically well to slow, caress-like touch ranging from 1 to 10 cm/s (Kumazawa and Perl, 1977; Vallbo et al., 1993; Löken et al., 2009). In their pioneering work, Olausson et al. (2002) identified CT afferents in a neuropathy patient lacking myelinated A-beta nerves, which normally function in discriminative tactile sensation, and showed that stimulation of CT afferents elicited activation in insular cortex. This finding led researchers to propose the ‘skin as a social organ’ hypothesis (Morrison et al., 2010), positing that the CT system represents an evolutionarily conserved mechanism for processing affective, or ‘limbic’ touch (Olausson et al., 2002, 2008). Recent fMRI studies from our group (Gordon et al., 2013; Voos et al., 2013) support this hypothesis by demonstrating that CT-targeted touch activates key nodes of the ‘social brain’ (Brothers, 1990; Frith 2007; Adolphs, 2009) including the posterior superior temporal sulcus (pSTS), medial prefrontal cortex (mPFC), amygdala and posterior insula. In addition, the pSTS response to CT-targeted touch has been linked to individual differences in autistic traits (Voos et al., 2013). Using fMRI, Voos et al. (2013) found that neurotypical adults with more autistic traits exhibited a diminished pSTS response to CT-targeted touch. As social deficits are a core feature of autism (APA, 2000), the relationship between brain mechanisms for processing affective touch and autistic traits is of particular interest. Autism is a neurodevelopmental disorder, and therefore disrupted brain mechanisms for processing touch may arise early in life and affect the subsequent development of this system. One investigation of adults with autism demonstrated abnormal brain responses to tactile stimulation, with these adults showing decreased neural response to pleasant/neutral touch but increased neural response to aversive touch compared with healthy controls (Cascio et al., 2012). This supports the notion that individuals with autism experience and process touch differently. Affective touch plays a critical role in social–emotional development, and we propose that pSTS may represent the neural basis of this process. Thus, this study aimed to assess the ability of fNIRS to detect brain responses to CT-targeted touch in pSTS, as this imaging method is particularly suited for developmental studies of infant brain function.

Functional NIRS is well-suited for the study of infants for several reasons. Compared with fMRI, fNIRS imposes fewer safety concerns because there is no magnetic field or noise (therefore, no need for hearing protection). It is also more ecologically valid as infants can sit up naturally in the lap of a parent (Gervain et al., 2011). Compared with electroencephalography, fNIRS has better spatial localization, is unaffected by eye blinks, and is less sensitive to motion artifacts (Gervain et al., 2011). Given these advantages, and to pursue the use of fNIRS to study the development of brain mechanisms for processing affective touch, we sought to assess whether fNIRS can reliably measure brain responses to CT-targeted touch in an adult population. Although a growing number of fNIRS studies are investigating brain mechanisms for social processing in the auditory (Sakatani et al., 1999; Csibra et al., 2004; Taga and Asakawa, 2007; Bortfeld et al., 2009; Saito, 2009; Grossmann et al., 2010) and visual (Otsuka et al., 2007; Grossmann et al., 2010; Ichikawa et al., 2010; Kojima and Suzuki, 2010; Lloyd-Fox et al., 2010) domains, few studies have explored the tactile domain (but see Haensse et al., 2004; Becerra et al., 2008; Saito, 2009; Shibata et al., 2012).

This fNIRS study aimed to identify a pSTS response to CT-targeted touch in typical adults to establish a method for future research investigating early development of cortical mechanisms for processing affective touch. Although other brain areas, such as the insula (Olausson et al., 2002), are known to be involved in processing CT-targeted touch, these regions were not examined in this study due to cortical depth beyond fNIRS measurement capabilities and limited coverage of the optode lattice, respectively. We hypothesized that fNIRS would be sensitive to cortical brain activity to CT vs non-CT touch in right posterior temporal brain regions found in an fMRI study of CT-targeted touch. We utilized an identical paradigm to our fMRI study (Gordon et al., 2013), which included alternating blocks of gentle touch to the arm (CT) and palm (non-CT). This similarity allows us to compare data collected with fMRI and fNIRS. We also examined individual differences in the response to CT-targeted touch as a function of participants’ autistic traits. We hypothesized that individuals with more autistic traits would exhibit a diminished response to this socially relevant touch. Hence, we sought to establish fNIRS as a neuroimaging technique to distinguish between posterior temporal lobe neural responses to CT- and non-CT-targeted touch in healthy adults. A secondary goal was to determine whether fNIRS could capture a relationship between individual differences in autistic traits and the neural response to CT-targeted affective touch. Successfully achieving these goals will allow us to downward extend this paradigm to infants to further understand how the neural systems for processing CT-targeted affective touch develop in typical and atypical populations.

EXPERIMENTAL PROCEDURES

Participants

Participants included 30 healthy, right-handed adults (18 females, 24.2 years ± 3.47). Four participants were excluded from analyses because two or more of their 52 recording channels did not collect data, which prohibited spatial interpolation of the results. Four additional participants were excluded due to lack of channel position measurements, as this prohibited spatial coregistration of data into standard space. Therefore, 22 participants were included in subsequent analyses (13 female, 24.1 years ± 3.80). Written informed consent was obtained for each participant according to a protocol approved by the Yale School of Medicine Human Investigations Committee. Participants received $25 for their participation.

Pre-experiment self-report ratings

Before the fNIRS session, an experimenter brushed participants on their right arm and palm. Participants rated the pleasantness of each type of touch on a 1–5 Likert scale (1 = ‘not at all’, 2 = ‘slightly’, 3 = ‘moderately’, 4 = ‘very’ and 5 = ‘extremely’). Each participant completed the Autism-Spectrum Quotient (AQ; Baron-Cohen et al., 2001), Social Responsiveness Scale (SRS; Constantino and Todd, 2003) and Social Touch Questionnaire (STQ; Wilhelm et al., 2001). The AQ is a self-report measure of autistic traits. Scores range from 0 to 50; higher scores indicate more autistic traits. The SRS measures social responsiveness and is completed by a friend or family member of the participant. Scores range from 0 to 195; higher scores indicate less social responsiveness. The STQ is a self-report measure that assesses participants’ attitudes toward social touch. Scores range from 0 to 80; higher scores indicate an aversion to giving, receiving and witnessing social touch. After completing these ratings and questionnaires, we measured 8 cm on the right arm (beginning at the wrist) and 4 cm on the right palm (beginning at the base of the hand) of each participant, to demarcate the brushing area.

Experimental design

Participants received continuous brushing back and forth (proximal–distal orientation) to the right forearm and palm separately, in a block design procedure. There were two alternating blocks of each condition. Each block contained eight repetitions of 6 s periods of touch (arm or palm) followed by 12 s of rest (no touch). Between blocks, there were six additional seconds of rest to allow the experimenter to prepare for the next block of touch (Gordon et al., 2013). Tactile stimuli were slow strokes (8 cm/s) performed with a 7-cm wide watercolor brush administered by one of two trained female experimenters. The brushing velocity of 8 cm/s was chosen because this speed is within the optimal range for targeting CT afferents (Löken et al., 2009). Before data acquisition, participants were instructed to close their eyes throughout the procedure and to focus on the touch. The experimenter watched and confirmed that all participants kept their eyes closed throughout the duration of the experiment. In total, the procedure lasted 10.03 min.

NIRS data acquisition

Preceding data collection, each participant’s head dimensions were measured as per the international 10–20 system (Jasper, 1958) to assess variability in participants’ head sizes and shapes. The average measured distances from nasion to inion, left ear to right ear and nasion to right ear were 29.82 cm (± 3.31), 35.98 cm (± 1.56) and 16.23 cm (± 1.10), respectively. Importantly, as cap placement was standardized to the right ear in each participant, the relatively low variability in distances from the nasion to the right ear validated the use of this landmark for consistent optode location. Measurements of neural activation were taken using a 50 two-channel NIRS machine (ETG-4000, Hitachi Medical) with 33 optodes separated by 3 cm configured in a 3 × 11 lattice. This spacing of optodes allowed for measurement of hemoglobin changes at a depth of up to ∼2 cm, thus occluding the measurement of deeper brain structures, such as the insula (Hock et al., 1997). Changes in oxygenated (oxy-Hb) and deoxygenated (deoxy-Hb) hemoglobin were measured using two wavelengths of infrared light (695 and 830 nm). Analyses in this study focused predominantly on oxy-Hb, although our region of interest (ROI) analysis of both oxy-Hb and deoxy-Hb showed that the two measures were inversely related, as would be expected (Figure 2). Data were collected at a frequency of 10 Hz. Placement of the optode lattice was standardized across subjects by positioning source probe number 25 directly above each participant’s right ear. Following the acquisition of functional data, all optodes were removed from the lattice, and a 3D digitizer system (Polhemus, VT, USA) was used to localize the placement of each optode in relation to reference points on the participant’s head (nasion, inion, left and right ears, top and back of the head). The locations of these reference points were used to coregister each recording channel into MNI space on a single-participant level using NIRS-SPM (Jang et al., 2009; Ye et al., 2009; Tak et al., 2010a,b) to perform subsequent group-level general linear model (GLM) analyses and ROI analyses in standard space.

Fig. 2.

Activation to arm vs palm touch within the pSTS ROI. (a) Visualization of the peak voxel used for ROI analysis, with the extent of the sphere depicting the average distance from the peak voxel of interest that included the four recording channels used for the individualized ROI analysis (average radius = 2.3 cm). (b) Waveforms for the four-channel pSTS group ROI analysis (arm touch > palm touch), with stimulus from 2 to 8 s. (c) Peak amplitude of pSTS response to arm touch in the high (N = 5) and low (N = 5) AQ groups, restricted within the time window of 5–12 s post-stimulus onset. Range of AQ scores for each group are shown in parentheses (low-medium and high-medium groups not shown here).

Data analysis

GLM-based analyses

Before GLM analyses, data were low-pass filtered at a frequency of 0.2 Hz, and a moving average of 1 s was applied to decrease high-frequency noise. Using NIRS-SPM version 3.2 (Jang et al., 2009; Ye et al., 2009; Tak et al., 2010a,b), global trends were removed from each single participant’s measurement data using a wavelet-minimum description length detrending algorithm. For GLM analyses, 6-s blocks of arm and palm touch were modeled separately with boxcar functions. We then performed single-participant GLM analyses by convolving the two task functions (arm touch and palm touch) with a double-gamma hemodynamic response curve to model the hypothesized oxy-Hb response during each experimental condition. On a single-participant level, channel placement was registered to MNI space, and statistical values calculated for each recording channel were interpolated to increase spatial resolution (Ye et al., 2009). Three of the 22 participants were missing data from 1 of their 52 channels. Because data from all 52 channels are needed for spatial interpolation, data points for these channels were estimated by averaging non-processed measurement data from the four adjacent channels for each time point of the experiment. None of these estimated channels were used in the ROI analysis. For the four main contrasts of interest (arm touch > palm touch, palm touch > arm touch, arm touch > baseline and palm touch > baseline), single-participant activation maps were combined in group-level mixed-effect GLM-based analyses, analyzing any pixel in which 15 of the 22 participants had overlapping functional data (see outline on Figure 1). All analyses were assessed at an uncorrected statistical threshold of P < 0.05.

Fig. 1.

Activation to arm > palm touch. The lighter region encompasses the pixels analyzed in the group-level GLM analyses. Activations indicate regions with a greater response to arm touch relative to palm touch (P < 0.05). Activation is presented on the right cortical surface, in Montreal Neurological Institute space.

ROI analyses

Based on our a priori hypothesis that pSTS is involved in processing CT-targeted touch (Gordon et al., 2013; Voos et al., 2013), we implemented an ROI analysis focusing on this brain area. Our ROI was based on a pSTS region that was significantly active to arm vs palm touch in an fMRI study of identical design to the current fNIRS investigation (Gordon et al. 2013). Using the voxel of peak activation in the pSTS region reported in the aforementioned fMRI study (Talairach coordinates: 57, −55, 13), and converting each participant’s NIRS channel location coordinates from MNI to Talairach space, we identified the four fNIRS recording channels whose approximated cortical locations were closest to this peak coordinate for each participant. Maximum distances from the peak voxel coordinate to the farthest channel included in the ROI for each participant ranged from 1.8 to 2.8 cm. ROI analyses were performed by integrating oxy-Hb signals over each 6 s block of arm touch (16 blocks) and palm touch (16 blocks) on a single-participant level in each channel. These integrated signals in the four channels identified as being within each participant’s individually defined pSTS ROI were then averaged for each time point of data acquisition in the 6-s stimulus block, as well as for the 2 s pre-stimulus onset and 10 s post-stimulus offset. A moving average of 1 s was also applied to decrease high-frequency noise, and integrated data were baseline corrected using a pre-block period of 2 s and a post-block period of 2 s (beginning after a recovery time of 8 s following block completion). The same was done for deoxy-Hb signals. To take advantage of the within-participant experimental design, we conducted paired-sample t-tests to examine the differential pSTS ROI activation to arm and palm touch ranging from 2 s pre-stimulus onset to 10 s post-stimulus offset for each participant. This difference waveform was than averaged on a group level, providing a visualization of the oxy-Hb and deoxy-Hb responses to arm > palm touch in the pSTS.

Based on the GLM results, we conducted a post hoc ROI analysis of the dlPFC with equivalent parameters to the pSTS analysis described earlier. The dlPFC region was based on that identified as active to CT-targeted touch in Voos et al. (2013) (Talairach coordinates: 39, 44, 13). Maximum distances from the peak voxel coordinate to the farthest channel included in this ROI for each participant ranged from 1.5 to 4.5 cm.

Relationship between autistic traits and brain response to CT-targeted touch

Exploratory analyses examined the relationship between autistic traits and peak neural responses to CT-targeted touch in the pSTS ROI to inform future work on the relation of these brain responses to autistic symptoms. For each participant, the peak magnitude of oxy-Hb during integrated arm (CT-targeted) and palm (non-CT-targeted) trials were identified within the time window of 5–12 s after stimulus onset. This was chosen to correspond to the previously demonstrated time window of the hemodynamic response to CT touch (Gordon et al., 2013). After removing one participant whose peak oxy-Hb value for arm touch was greater than two standard deviations from the mean, 21 participants remained in the analysis. We performed a correlation analysis between participants’ AQ scores and their peak response to CT touch within their four-channel pSTS ROI. Given our limited AQ range, we also split participants into quartiles based on their AQ scores and examined the relationship between peak response to CT touch and AQ scores.

RESULTS

Self-report measures

Self-report measures based on the experience of arm and palm touch prior to the fNIRS experiment revealed a mean pleasantness rating for arm of 3.7 (± 0.7) and for palm of 3.2 (± 0.9). Arm touch was rated significantly more pleasant than palm touch [t(22) = 2.5, P = 0.04]. The average scores on the remaining self-report measures were as follows: STQ 27.1 (± 10.31), AQ 13.09 (± 5.79) and SRS 15.32 (± 11.67).

NIRS GLM-based analyses

Group-level analyses were performed using a mixed-effects GLM. Group results were first assessed in the contrast arm touch > palm touch. All activations reported are in the right hemisphere of the brain. Distinct regions of the posterior temporal lobe and the dlPFC showed significant activation to arm touch > palm touch (P < 0.05; Figure 1). Secondary group-level analyses revealed that no regions showed significant activation to palm touch > arm touch, arm touch > baseline or palm touch > baseline at P < 0.05, or at a more lenient threshold of P < 0.1.

ROI analyses

The ROI analysis served to confirm and expand upon the results of our GLM analysis. Activations to arm and palm touch were calculated for each participant based on the average of his or her four recording channels located closest to the peak coordinate of activation in the pSTS region identified by Gordon et al. (2013) as being preferentially active to CT-targeted touch (Figure 2a). Paired-sample t-tests were conducted at each time point, as an ad hoc method of delineating the time window of maximum significance between brain responses to arm and palm touch. This comparison revealed that oxy-Hb activation to arm touch was significantly greater than activation to palm touch from 8.4 to 9.7 s post-stimulus onset (P’s < 0.05; Figure 2b). In addition, the subtraction of arm–palm touch at each time point for each participant allowed us to visualize the average time course of the oxy-Hb and deoxy-Hb response to arm > palm touch (Figure 2b).

In addition, activations to arm and palm touch were calculated for each participant based on the average of his or her four recording channels located closest to the peak coordinate of activation in the dlPFC region previously identified (Voos et al., 2013) to process CT-targeted touch. As described in the pSTS ROI analyses, we also performed paired-sample t-tests at each time point to examine the oxy-Hb time course response to arm > palm touch, which revealed that activation to arm touch was significantly greater than activation to palm touch from 10.8 to 11.1 s post-stimulus onset in dlPFC (P’s < 0.05).

Relationship between autistic traits and brain response to CT-touch

A correlation analysis of peak pSTS response to arm touch and AQ scores was not significant. Thus, in an attempt to elucidate how these factors might be related, participants were divided into quartiles based on their AQ scores: low AQ (M = 5.60; s.d. = 1.14), low-medium AQ (M = 9.75; s.d. = 1.71), high-medium AQ (M = 14.14; s.d. = 1.10) and high AQ (M = 20.80; s.d. = 2.59), with five, four, seven and five participants in each group, respectively. Next, an independent-sample t-test conducted in the low and high AQ quartile groups revealed a significant difference in peak oxy-Hb concentration [t(8) = 2.69, P = 0.03]: individuals with lower AQ scores (fewer autistic traits) had higher peak pSTS responses to arm touch relative to the individuals with higher AQ scores (more autistic traits) (Figure 2c). There was no difference in pSTS response to arm > palm touch or to palm touch in the low and high AQ groups. It should be noted that the Levene’s test for equal variance was non-significant; therefore, t-tests for samples with equal variance were used. Similar analyses were conducted by parsing participants into groups based on SRS and STQ scores, however, no significant differences in peak pSTS activation to arm or palm touch based on these scores were detected.

DISCUSSION

The aim of this study was to validate fNIRS as a neuroimaging technique to measure temporal lobe responses to CT-targeted touch in a sample of healthy adults. Our main objectives were to verify the ability of fNIRS to detect the cortical neural responses to gentle touch to the arm and palm by replicating results found using fMRI (Gordon et al., 2013) and to establish an fNIRS paradigm for use in future studies of the development of these neural mechanisms in infants.

This study replicated previous fMRI findings of pSTS involvement in processing CT-targeted touch. We targeted CT-afferent nerves located in hairy skin because they respond specifically well to slow, gentle touch (Olausson et al., 2002), which is reminiscent of that shared in early social interactions, such as a parent’s caress of their child. This type of touch is known to acutely impact social–emotional development (Stack, 2001; Barnett, 2005). To study early development of brain mechanisms for processing such affective touch, we must first establish the brain response to affective touch in healthy adults as measured with fNIRS. To this end, we implemented group-level GLM and individualized ROI analyses, two independent strategies, both of which revealed complimentary results of pSTS activation to CT-targeted touch. Thus, the GLM and ROI analyses do not represent distinct findings. Instead, the ROI analysis confirms that the activation visualized in the GLM analysis is anatomically comparable with the locus of activation determined in a previous fMRI study of identical experimental procedure (Gordon et al., 2013). In addition, exploratory analyses showed that individual differences in autistic traits were related to peak pSTS activation to CT-targeted touch. Individuals with more autistic traits had lower peak pSTS activation to arm touch than those with fewer autistic traits. This finding is concordant with our recent fMRI result of a negative correlation between pSTS response to CT-targeted touch and autistic traits (Voos et al., 2013). Thus, the current results validate fNIRS as a neuroimaging method for studying the neural mechanisms of CT-targeted affective touch and lay the foundation for studying the development of this system in the first years of life. The implications of these findings are discussed below in the context of social neuroscience, typical development and developmental disorders, such as autism.

To our knowledge, this is the first fNIRS study to examine the brain mechanisms for processing CT-targeted affective touch. Here, we replicate fMRI findings of pSTS response to this touch, expanding upon previous work showing concordance between brain responses measured with fMRI and fNIRS (Steinbrink et al., 2006). As, compared with fMRI, fNIRS has reduced spatial resolution and decreased signal-to-noise ratio, we implemented two methods to account for these limitations. We digitally localized optode placement to normalize the location of recording channels to standard space in each participant. Using this normalized placement, we applied a hemodynamic response function to model the oxy-Hb response to arm and palm touch using a GLM. Together, these analysis strategies enhanced spatial resolution and statistical power in the identification of true hemodynamic responses recorded with fNIRS. In addition, we implemented an ROI analysis that utilized the peak voxel of activation in pSTS to CT-targeted touch identified in an fMRI study of identical design (Gordon et al., 2013). The four recording channels nearest to this peak coordinate for each participant were combined to form ROIs that were individualized based on each participant’s optode placement. This novel analysis method allowed us to hone in on a specific brain region involved in processing CT-targeted touch. The results from these two independent analyses converge to support the hypothesis that the pSTS is involved in the neural processing of CT-targeted, affective touch. While it is possible that differential stroking distance influenced the results, previous fMRI findings of pSTS activity in response to CT-targeted touch (i.e. from a paradigm in which stroking distance was held constant) (Voos et al., 2013) suggest that this is not the case. Our findings validate the use of fNIRS to measure brain responses to affective touch, establishing the efficacy of our paradigm for future studies of the development of the CT system, an area ripe for future research.

Our finding of pSTS activation to CT-targeted touch is noteworthy given this region’s role as a key node of the ‘social brain’ (Brothers, 1990). The pSTS responds to social stimuli in the visual (Allison et al., 2000; Pelphrey et al., 2005), auditory (Belin et al., 2000; Shultz et al., 2012) and tactile (Gordon et al., 2013; Voos et al., 2013) domains. The diverse nature of these studies emphasizes the ‘multimodal’ role of this region in social processing (Barraclough et al., 2005; Beauchamp 2005). Future research on the development of this region’s response to social stimuli is especially important in the tactile domain, as this sensory system is the earliest to develop (Montagu, 1971) and plays an important role in social function in both primates (Harlow and Zimmermann, 1959; Bowlby, 1969) and human infants (Stack, 2001; Barnett, 2005).

In addition to pSTS, we found dlPFC sensitivity to CT-targeted touch. Notably, this region likely does not correspond to the mPFC activation previously found to process affective touch (Gordon et al., 2013) because fNIRS only measures a depth of ∼2 cm from the scalp (Cui et al., 2011). However, dlPFC activation in this study shows similarity to an area of dlPFC found to activate to CT touch in another fMRI study by our group (Voos et al., 2013). Based on the peak voxel of this activation, we conducted a post hoc ROI analysis to determine whether dlPFC activation was anatomically analogous to that previously reported. Indeed, we found that activation to arm touch in dlPFC measured with fNIRS was significantly greater than palm touch for a brief period (10.8–11.1 s) post-stimulus onset. We hypothesize the role of the dlPFC in processing CT touch to be related to reward, given that in this study and others (Gordon et al., 2013; Morrison et al., 2011; Voos et al., 2013), participants rated CT-targeted touch as more pleasant than non-CT-targeted touch. The dlPFC has been implicated in reward processing in a variety of tasks in both humans and primates (Inoue et al., 1985; Leon and Shadlen, 1999; Hornak et al., 2004). Furthermore, this region responds to pleasant vs unpleasant touch to the leg (Hua et al., 2008), and pleasant vs unpleasant words (Herrington et al., 2005). While this study did not aim to investigate this region’s role in processing social touch, our GLM and ROI analyses converge to suggest its importance in processing CT-targeted touch and future work may clarify this region’s functional involvement.

In addition, we found that individual differences in autistic traits were related to the magnitude of peak activation to CT-touch in pSTS. Individuals with more autistic traits, as measured by the AQ, had significantly lower peak activation in pSTS to gentle arm touch relative to individuals with fewer autistic traits. This supports previous findings that individuals with more autistic traits exhibit dampened pSTS responses to CT-targeted touch (Voos et al., 2013). Although these findings are within a healthy sample, it is possible that pSTS activation to CT-targeted affective touch may be disrupted in individuals with autism. Indeed, Cascio et al. (2012) recently demonstrated that adults with autism show decreased neural response to pleasant touch and increased neural response to aversive touch.

While novel, this study incurred some limitations. An inherent limitation to fNIRS methodology is the difficulty in standardizing cap placement due to variability in head size and shape. We mitigated this problem by implementing digital localization to normalize all subjects’ recording channels to standard space. In addition, the depth to which the infrared light can penetrate within the brain limits the spatial resolution of fNIRS (up to ∼2 cm from the skull). Thus, we were unable to measure deeper regions involved in CT processing, such as the insula and the orbitofrontal cortex (McGlone et al., 2012). In addition, because our fNIRS device can simultaneously record from a maximum of 33 optodes, we chose to retain spatial specificity by placing these optodes at a traditionally used distance of 3 cm apart from each other in our 3 × 11 lattice. Due to this spacing, we were only able to image one side of the brain. We focused on the right hemisphere because previous studies from our group have shown ipsilateral (right hemisphere) temporal lobe (pSTS) activation to right-lateralized touch in response to CT-targeted touch (Gordon et al., 2013; Voos et al., 2013). Also, due to the criteria of only analyzing pixels where 2/3 or greater of the participants contributed data (to avoid reporting spurious results from a small subset of the sample), we were unable to explore the somatosensory cortex (see Figure 1 for extent of analyzed pixels) although it may be of interest (but see Olausson et al., 2002). Finally, because we had a small range of AQ scores, a larger range of AQ scores and a larger sample size might reveal a more robust relationship between autistic traits and brain responses to touch.

This study lays the foundation for future research aimed at exploring the typical and atypical development of neural processing of gentle touch processed by the CT system. We successfully replicated a tactile paradigm using fNIRS, originally implemented in fMRI, and found pSTS and dlPFC activation to CT-targeted, affective touch. Concordant with our previous fMRI study, individuals with more autistic traits showed a diminished pSTS response to CT-targeted touch. Future work will explore cortical brain mechanisms for processing CT-targeted touch in individuals with autism, as well as infants both at high- and low-risk for developmental disorders such as autism, where hypersensitivity to sensory information, especially touch, is often present (Blakemore et al., 2006). Neuroimaging studies of the development of affective touch processing hold great promise for illuminating the biological mechanisms underlying the robust influence of touch on social–emotional development.

FUNDING

This work was funded by a Harris Professorship (to Dr. Kevin Pelphrey), Autism Speaks (to Dr. Martha Kaiser).