香京julia种子在线播放

Thirteen subjects (9 female, 11 right-handed, mean age 26.54 ± 4.31 years) participated in this study. All subjects were native speakers of French, without musical training. None reported history of neurological or psychiatric illness or hearing deficits and all had hearing thresholds within normal limits. Prior to the imaging session, each subject completed six questionnaires on their health status, handedness [Edinburgh Handedness Inventory, (Oldfield, 1971)], anxiety and depression state [Hospital Anxiety and Depression, HAD, scale; (Zigmond and Snaith, 1983)], personality traits [Big-Five Inventory, (Courtois et al., 2018)], and a musical aptitude questionnaire developed in the lab. These questionnaires revealed no significant differences in personality traits nor in the presence of mood disorders between our subjects and normal population. The experimental procedures were approved by the Ethics Committee of the Canton de Vaud; all subjects gave written, informed consent.

The experimental design consisted of two fMRI sessions (~55–60 min each) during which auditory stimuli were presented while the subjects listened passively to the stimuli with eyes closed. In total, each subject performed two runs of tonotopy mappings, one run of voice localizer, and 12 runs of “emotions&space” runs. Each of the latter consisted of 20s of silent rest (with no auditory stimuli except the scanner noise), followed by nine 36 s-blocks of 11 sounds of the same condition (22 s sounds and 14 s of silent rest), and again 20 s of silent rest. Each block was composed of 11 different sounds from the same category (human vocalizations or other environmental sounds), all of which had the same emotional valence (positive, neutral or negative) and the same lateralization (left, center, right). Finally, blocks and their sequence order were pseudo-randomized within runs and across subjects.

Sounds (16 bits, stereo, sampling rate of 41 kHz) presented binaurally at 80 ± 8 dB SPL via MRI-compatible headphones (SensiMetrics S14, SensiMetrics, United States), with a prior filtering with the SensiMetrics filters to obtain a flat frequency transmission, using MATLAB (R2015b, The MathWorks, Inc., Natick, Massachusetts, United States) and the Psychophysics Toolbox¹. The auditory stimuli were the same as the battery used in previous studies (Aeschlimann et al., 2008; Grisendi et al., 2019), the total 66 different emotional sound files were 2 s-long and were equally distributed in the six categories: Human Vocalizations Positive (HVP; e.g., baby or adult laughing; erotic vocalizations by man or woman), Human Vocalizations Neutral (HV0; vowels or consonant-vowels without significance), Human Vocalizations Negative (HVN; e.g., frightened scream; vomiting; brawl), Non-Vocalizations Positive (NVP; e.g., applause; opening beer can and pouring into a glass; river), Non-Vocalizations Neutral (NV0; e.g., running car engine; wind blowing; train), and Non-Vocalizations Negative (NVN; e.g., ticking and exploding bomb; tire skids; breaking glass). Sounds were lateralized by creating artificially a temporal shift of 0.3 s between the left and right channel (corresponding to ~60°), using the software Audacity (Audacity Team²), and were either perceived as presented on the left, the center or the right auditory space. Thus, the combination of all the different characteristics resulted in a total of 18 conditions (2 Categories x 3 Valences x 3 Lateralizations).

As previously, using a specific software, PRAAT³, and MATLAB scripts, the sound acoustic characteristics (spectrograms, mean fundamental frequency, mean intensity, harmonics to noise ratio, power, center of gravity, mean Wiener entropy and spectral structure variation) were controlled for each category: first, the significant differences between the mean spectrogram of pairs of sounds of different categories were maintained <1% to avoid bias toward a specific category (as in De Meo et al., 2015); second, all the sounds characteristics were tested with a two-way repeated measures ANOVA with the factors Category (Human-Vocalizations, Non-Vocalizations) x Valence (Positive, Neutral, Negative) to compare the effect of each acoustic feature on the sound categories. As already reported in our previous study (Grisendi et al., 2019), the analysis on mean Wiener entropy showed a main effect of Category [F(1,64) = 18.68, p = 0.0015], a main effect of Valence [F(2,63) = 21.14, p = 1.17E-5] and an interaction Category x Valence [F(2,63) = 8.28, p = 0.002]; while the same analysis on the center of gravity revealed a main effect of Valence [F(2,63) = 10.51, p = 0.0007]. The analysis of the harmonics-to-noise ratios highlighted a main effect of Category [F(1,64) = 134.23, p = 4.06E-7], a main effect of Valence [F(2,63) = 69,61, p = 9.78E-10] and an interaction of Category x Valence [F(2,63) = 17.91, p = 3.48E-5], and these of the power showed an interaction of Category x Valence on the mean intensity [F(2,63) = 12.47, p = 0.0003] and on the power [F(2,63) = 14.77, p = 0.0001].

The subdivision of the early-stage auditory areas was carried out in individual subjects as described previously (Da Costa et al., 2015, 2018). The subjects listened to two runs (one ascending and one descending) of a tonotopic mapping paradigm, which consisted of progressions of 2 s-bursts of pure tones (14 frequencies, between 88 and 8,000 Hz, in half octave steps) presented in 12 identical cycles of 28 s followed by a 12-s silent pause for a total duration of 8 min (as in previous studies Da Costa et al., 2011, 2013, 2015, 2018). Then, briefly, based on the resulting individual frequency reversals and anatomical landmarks, each early-stage auditory area was localized and defined in each subject as the primary auditory cortex, A1 and R, as well as the lateral (L1, L2, L3, L4) and medial non-primary areas (M1, M2, M3, M4). The coordinates of these regions were in accordance with previously published values (Table 1; Viceic et al., 2006; van der Zwaag et al., 2011; Da Costa et al., 2015, 2018).

Finally, the position of VA was defined using a specific voice localizer (Belin et al., 2002; Pernet et al., 2015). Briefly, human vocalizations (vowels, words, syllables laughs, sighs, cries, coughs, etc.) and environmental sounds (falls, wind, animals sounds, etc.) were presented in a 10-min run, which consisted of forty 20s-long blocks (with 8 s of sounds followed by a silent pause of 12 s). This localizer was developed to easily and consistently identify the individual voice area along the lateral side of temporal plane, by displaying the results of the general linear model (GLM) contrast Human vocalizations vs. Environmental sounds. In this study, the same approach was used in BrainVoyager (BrainVoyager 20.6 for Windows, Brain Innovation, Maastricht, Netherlands). After initial preprocessing, the functional run was first aligned with the subject anatomical, and analyzed with a general linear model using a boxcar design for the two conditions. Second, the results of the contrast Human vocalization vs. Environmental sounds was projects on the individual 3D volume rendering with a p value of p < 0.005 (uncorrected) in order to cover the same extend in each subject. Finally, the activated region within the bilateral lateral borders of the STS/STG was manually selected as a patch of interest using the manual drawing tools from BrainVoyager and projected back into the MNI space and saved as the individual region of interest. The coordinates of the VA were also in accordance with those of previous studies (Belin et al., 2002; Pernet et al., 2015).

Brain imaging was acquired on a 7-Tesla MRI scanner (Siemens MAGNETOM scanner, Siemens Medical Solutions, Germany) with an 32-channel head RF-coil (Nova Medical Inc., MA, United States). Functional datasets were obtained with a 2D-EPI sinusoidal simultaneous multi-slice sequence (1.5 × 1.5 mm in-plane resolution, slice thickness = 1.5 mm, TR = 2000 ms, TE = 23 ms, flip angle = 90°, slice gap = 0 mm, matrix size = 146 × 146, field of view = 222 × 222, with 40 oblique slices covering the superior temporal plane). T1-weigthed 3D structural images were obtained with a MP2RAGE sequence [resolution = 0.6 × 0.6 × 0.6 mm³, TR = 6,000 ms, TE = 4.94 ms, TI1/TI2 = 800/2700 ms, flip angle 1/flip angle 2 = 7/5, slice gap = 0 mm, matrix size = 320 × 320, field of view = 192 × 192 (Marques et al., 2010)]. Finally, the physiological noise (respiration and heart beat) was recorded during the experiment using a plethysmograph and respiratory belt provided from the MRI scanner vendor.

The data was processed with BrainVoyager with the following steps: scan time correction (except for tonotopic mappings runs), temporal filtering, motion correction, segmentation and normalization into the MNI space. Individual frequency preferences were extracted with a linear cross-correlation analysis, resulting correlation maps were averaged together (ascending and descending correlation map) to define the best frequency value for each voxel in the volumetric space, and then the average map was projected onto the cortical surface meshes for the ROIs definition (Da Costa et al., 2011, 2013, 2015, 2018). For the VA localizer and the emotion&space runs, a random effects (RFX) analysis was performed at the group level, with movement and respiration parameters as regressor, and then we tested for the contrast ‘Sounds vs. Silence’ with an FDR correction at q < 0.05 (p < 0.05). The GLM results for the VA localizer was used to outline the VA in the left and in the right hemisphere of each individual brain, while the GLM results for the emotion&space runs were used to verify that our ROIs were activated by the paradigm. The scope of this paper was to evaluate the effects of spatial origin on the encoding of emotional sounds, therefore the remaining analysis focused on the BOLD responses extracted from all the ROIs.

Functional individual BOLD time courses were processed as the following: first, they were extracted using BrainVoyager, imported into MATLAB. Second, they were normalized by their own mean signal, and divided according to their condition. Third, they were averaged spatially (across all voxels within each ROI), temporally (over blocks and runs), and across the 13 subjects. The resulting time course consisted of 18 time points for each ROI and condition. Finally, these time courses were analyzed with a time-point-by-time-point Three-Way repeated measure ANOVA, two Category (Human-Vocalizations, Non-Vocalizations) x 3 Valence (Positive, Neutral, Negative) x 3 Lateralization (Left, Center, Right) according to Da Costa et al. (2015, 2018) and Grisendi et al. (2019). This three-way ANOVA was further decomposed for each vocalization category onto a two-way repeated measure ANOVA, 3 Valence (Positive, Neutral, Negative) x 3 Lateralization (Left, Center, Right). For each ANOVA, and each pair of condition, post hoc time-point-by-time-point paired t-tests were performed to evaluate the causality of the effects. Finally, results were restricted temporally by only considering at least three consecutive time points with significant p-values lower or equal to 0.05.

Heartbeat and respiration recordings were processed with an open-source toolbox for Matlab, TAPAS PhysIO (Kasper et al., 2017). The cardiac rates were further analyzed with the same pipeline as the BOLD responses to obtain a pulse time course for each condition, while the respiration rates were used within the GLM model as motion regressor. The effect of space and emotional contents of the sounds on the individual cardiac rhythm was evaluated by computing the heart rate variability as reported in previous studies by others (Goedhart et al., 2007) and by us (Grisendi et al., 2019).

A significant main effect of Lateralization was present in the left hemisphere in A1 (during the 10–14 s and 22–26 s time periods); in R (10–14 s and 18–36 s); and in M1 (10–14 s; Figure 1A). The effect was driven by greater activation for contra- than ipsilateral stimuli (Figure 1B).

Significant interaction of Category x Valence was present in either hemisphere. In the left hemisphere this was the case in A1 (12–26 s); R (14–18 s); L1 (10–18 s and 22–26 s); L2 (12–20 s and 24–28 s); M2 (22–26 s); and VA (4–16 s and 20–28 s; Figure 1A). In the right hemisphere this was the case in A1 (14–18 s); R (12–28 s); L1 (14–24 s); L2 (10–26 s); L3 (12–18 s); L4 (14–18 s); M1 (14–18 s and 22–26 s); M3 (30–36 s); M4 (32–36 s); and VA (8–26 s; Figure 2A). In A1, R, L1 and L2 the interactions appeared to be driven by the predominance of positive vocalizations and/or neutral non-vocalizations (Figures 1B, 2B).

A significant main effect of Valence was present in several areas of either hemisphere. In the left hemisphere this was the case A1 (18–30 s); R (20–36 s); L1 (24–36 s); L2 (12–14 s and 18–36 s); L3 (6–12 s and 16–36 s); M1 (16–24 s and 28–36 s); M2 (28–36 s); M4 (16–24 s and 28–36 s); and VA (6–28 s; Figure 1A). In the right hemisphere it was the case in A1 (20–24 s); R (24–36 s); L2 (28–32 s); L3 (6–12 s and 24–36 s); M1 (20–24 s); M2 (18–22 s); M4 (16–24 s and 28–36 s); and VA (8–20 s; Figure 2A). The effect tended to be driven by greater activation by vocalizations with positive rather than negative or neutral valence and by non-vocalizations with neutral rather than positive valence (Figures 1B, 2B).

A significant main effect of Category was present in either hemisphere. In the left hemisphere this was the case in L1 (6–22 s); L2 (6–26 s and 32–36 s); L3 (6–28 s and 32–36 s); M1 (8–12 s); and VA (6–28 s and 32–36 s; Figure 1A). In the right hemisphere this was the case in L2 (6–28 s); L3 (6–26 s); and VA (6–28 s and 32–36 s; Figure 2A). The effect was driven by greater activation by vocalizations than non-vocalizations by overall greater activation by vocalizations than non-vocalizations (Figures 1B, 2B).

Auditory stimuli presented within the left space elicit stronger responses in A1 and R of the left and right hemisphere when positive vocalizations are used (Figure 3). In both hemispheres neural activity elicited by positive vocalizations presented on the left was higher than neural activity elicited by (i) neutral or negative vocalizations presented at any of the three positions; or (ii) non-vocalizations of any valence at any of the three positions. The involvement of left A1 and R in favor of the ipsilateral and that of right A1 and R in favor of the contralateral, left space speaks against a mere effect of contralateral space or a classical hemispheric dominance.

The stronger encoding of positive vocalizations presented on the left side suggests that they may be more salient than when presented at other positions. The pre-eminence of the left auditory space, which we describe here, is reminiscent of the left-ear advantage, which was reported for emotional dichotic listening tasks in two studies (Erhan et al., 1998; Jäncke et al., 2001). Both studies compared emotional vs. neutral vocalizations, but did not discriminate between positive and negative valence. Their results have been interpreted in terms of right hemispheric competence for emotional processing (see also Gadea et al., 2011). Another series of studies used emotional valence of spoken words for spatial orienting of attention. Emotional word cues presented on the right side introduced spatial attentional bias for the following neutral sound (beep; Bertels et al., 2010). The interpretation of these results was influenced by the assumption that (i) one-sided presentation of auditory stimuli is preferentially treated by the contralateral hemisphere and (ii) the nature of the stimuli – verbal vs. emotional – tends to activate one hemisphere. Thus, the right side bias introduced by emotional words was eventually interpreted as prevailing influence of verbal content (Bertels et al., 2010). The nature of stimuli used in these studies, all verbal vocalizations, and the fact that they were presented mono-aurally, and not lateralized with interaural time (as here) or intensity differences, precludes their interpretation in terms of the emotional value of space.

The left-space preference, which we observed bilaterally in A1 and R, is greater for positive vocalizations than other stimuli. The phenomenon we describe here, the pre-eminent encoding of emotional vocalizations when presented in the left space in left and right R and A1, differs from previously described principles of auditory encoding. First, our results cannot be simply interpreted in terms of the well documented preference of the early-stage auditory areas for the contralateral space. This has been demonstrated for auditory stimuli in general (Deouell et al., 2007; Da Costa et al., 2015; Stecker et al., 2015; McLaughlin et al., 2016; Derey et al., 2017; Higgins et al., 2017) and more recently for auditory stimuli with positive emotional valence, which yielded strong contralateral activity when presented on the left side (Kryklywy et al., 2013). Second, our results do not show lateralization for a given type of stimuli, i.e., a preferential encoding within the left or the right auditory cortex, such as shown for stimuli with rapid formant transition in left auditory cortex (Charest et al., 2009); for varying rates of stimuli in the left and increasing spectral information in the right auditory cortex (Warrier et al., 2009); or more generally the asymmetry of the auditory regions in terms of temporal selectivity (Nourski and Brugge, 2011).

We did not investigate in this first study, whether the pre-eminent encoding of positive vocalizations when presented on the left side differs between male and female subjects, as do parts of the networks controlling speech production (de Lima Xavier et al., 2019).

Further experiments need to clarify whether the preference of R and A1 for positive vocalizations when presented in the left space can be modulated by context and/or attention. The sequence in which auditory stimuli are presented was shown to influence their encoding; the auditory cortex was shown to respond more strongly to pulsed noise stimuli when they are presented to the contra- than ipsilateral ear; this contralateral advantage is no longer present when the same type of monoaural stimuli is interspersed with binaural moving stimuli (Schönwiesner et al., 2007). The right ear advantage in dichotic listening tasks decreases when attention is oriented toward the left ear; this change in performance was shown to be accompanied with decreases in neural activity demonstrated by fMRI (Kompus et al., 2012) and with MEG recordings (Alho et al., 2012).

Although compatible with evidence from previous studies, our results give a different picture of the emotional auditory space and its encoding within the early-stage auditory areas. We have documented a genuine pro-eminence of the left space for positive vocalizations and not simply a right hemispheric or contralateral dominance, the key observation being that left-sided positive vocalizations stand out within the primary auditory cortex of both hemispheres. Several aspects need to be investigated in future studies. There is no current evidence on the behavioral relevance of the emotional pro-eminence of the left auditory space. It is unclear when it emerges in human development; indirect evidence comes from studies that reported left-ear preference for emotional sounds in children (Saxby and Bryden, 1984; Obrzut et al., 2001). The emotional pro-eminence of the left auditory space may not be an exclusively human characteristic. Although not explored as such in non-human primates, the reported right-hemispheric dominance for the processing of emotional sounds may be a correlate of the emotional pro-eminence of the left auditory space [for review (Gainotti, 2022)].

Two of our observations suggest that spatial cues render emotional vocalizations more salient. First, positive vocalizations presented on the right or the left were prominent in right L3 (Table 2). Second, the use of spatial cues appeared to enhance the salience of emotional valence in several early-stage areas. In a previous study, the same set of stimuli (human vocalizations and non-vocalizations of positive, neutral and negative valence), the same paradigm and an ANOVA based statistical analysis were used, albeit without lateralization (Grisendi et al., 2019). The juxtaposition of the distribution of significant interactions and significant main effects in early-stage areas and in VA highlights striking differences, which concern almost exclusively the factor Valence (and not Category; Figure 4). Main effect of Category highlighted in both studies a very similar set of areas, with vocalizations yielding greater activation than non-vocalizations. Main effect of Valence was strikingly dissimilar, being significant in many more areas when spatial cues were used. The same was observed for the interaction Category x Valence, with many more areas being significant when spatial cues were used; it is to be noted that in both studies the interaction was driven by greater responses to positive vocalizations. This increased saliency when spatial cues are used is not due to a modulation of emotional valence by lateralization; this interaction was only significant in VA but not in any of the early-stage areas.

The mechanisms by which spatial cues confer greater salience to emotional vocalizations is currently unknown. Interaural interactions during first cortical processing stages may enhance emotional stimuli, as does increasing intensity (Bach et al., 2008, 2009). Further studies are needed to investigate whether the effect is associated uniquely with interaural time differences (used here) or whether interaural intensity differences or more complex spatial cues have the same effect.

Our analysis clearly showed that within VA the encoding of vocalizations is modulated by emotional valence, as did a series of previous studies (Belin et al., 2002; Grandjean et al., 2005; Ethofer et al., 2006, 2008, 2009, 2012; Beaucousin et al., 2007; Obleser et al., 2007, 2008; Bestelmeyer et al., 2017; Grisendi et al., 2019). The new finding is that this clear modulation of vocalizations by emotional valence is not paralleled by a modulation by the spatial origin of the sound. This is reminiscent of the findings of Kryklywy et al. (2013), who reported that emotional valence, but not spatial attributes, impacts the processing within the ventral stream on the temporal convexity. Their stimuli consisted to 75% of human vocalizations and may have driven the effect they observed.

In our study spatial information did not modulate significantly the encoding of vocalizations within VA. However, the spatial origin impacted the activity elicited by sound objects in general. Thus, positive and neutral sounds; i.e., vocalizations and non-vocalizations taken together, yielded stronger response than negative ones when presented on the left or on the right, as compared to a presentation at the center. This preference for positive and neutral sounds when presented in lateral space was present in both hemispheres.