www.elsevier.com/locate/ynimg NeuroImage 34 (2007) 1708 – 1722
Developmental differences in the neural bases of the face inversion effect show progressive tuning of face-selective regions to the upright orientation A.M. Passarotti, a,⁎ J. Smith, a M. DeLano, b and J. Huang b a
Department of Psychology, Cognitive Science and Neuroscience Program, Michigan State University, Psychology Bld., East Lansing, MI 48824, USA Department of Radiology, Michigan State University, East Lansing, MI 48824, USA
b
Received 12 October 2005; revised 5 July 2006; accepted 6 July 2006 Available online 22 December 2006
Face inversion hinders face processing in adults, while not affecting children in the same way. This fMRI study examines the neural underpinnings of the behavioral face inversion effect (FIE) from childhood to adulthood, and how face-selective regions in the brain may change with development. Adults, children, and teens performed a facial expression decision on upright and inverted face stimuli. In the right hemisphere (RH) all age groups showed similar profiles of neural activation for upright faces, but important developmental differences occured for inverted faces. For inverted faces, adults, and to a lesser degree teens, exhibited decreased levels of activity in the face-selective, right lateral fusiform gyrus (LFG). However, children exhibited greater activation for inverted than for upright faces in the same region. We found similar, but less robust, developmental trends in the right superior temporal sulcus (STS) and medial fusiform gyrus (MFG). Furthermore, the present study identifies the right LFG as the primary neural correlate of the behavioral FIE, and therefore of face processing expertise, by showing a significant correlation between the behavioral FIE and the neural FIE only in this region. Finally, the present findings shed some light on at least one of the possible mechanisms underlying the development of face processing expertise, by suggesting a progressive tuning of face-selective regions in the right hemisphere to the upright orientation, that extends well into adolescence. © 2006 Elsevier Inc. All rights reserved.
The ability to recognize and discriminate between different faces is one of the most important human social skills. Behavioral and neurophysiological research suggests early brain specialization for face processing (Tzourio-Mazoyer et al., 2002) and newborn preference for upright faces compared to non-face patterns (De Haan et al., 2002; Morton and Johnson, 1991) or inverted face-like stimuli (Valenza et al., 1996). Nonetheless, there is also evidence for steady neural (Passarotti et al., 2003; Aylward et al., 2005; ⁎ Corresponding author. Fax: +574 631 8883. E-mail address:
[email protected] (A.M. Passarotti). Available online on ScienceDirect (www.sciencedirect.com). 1053-8119/$ - see front matter © 2006 Elsevier Inc. All rights reserved. doi:10.1016/j.neuroimage.2006.07.045
Gathers et al., 2004) and behavioral (Carey, 1992; Diamond and Carey, 1977; Mondloch et al., 2003; Taylor et al., 2004) development of face processing expertise through the teen years. While adults are experts at processing upright faces, their performance worsens when faces are inverted. This phenomenon is called “the face inversion effect” (FIE) (Yin, 1969; Valentine, 1988; Moscovitch et al., 1997), and is much more pronounced for faces than objects (Yovel and Kanwisher, 2004). A widely accepted explanation for the FIE is that adult expertise relies mostly on configural processing (i.e., processing of the relationship between facial features, rather than processing of single facial features) and that configural properties are extracted less efficiently when faces are upside down (Carey, 1992; Diamond and Carey, 1977; Freire et al., 2000; Mondloch et al., 2003; Farah et al., 1995). In fact, experimental evidence suggests that configural information is more salient than featural information when processing faces. Studies using the “composite-face effect” paradigm suggest that recognizing that the top and bottom portions of a face in a composite actually belong to different faces is harder when the two parts are fused than when they are spatially separated (Young et al., 1987), probably because we are biased to integrate face parts into a meaningful facial configuration. Also, only for faces but not for other objects, it is easier to recognize single face features when they are within a whole face, than when the features are isolated (Tanaka and Farah, 1993). Furthermore, people are still fairly accurate at identifying blurred faces even when the internal features are not very visible, whereas the opposite is not true. These findings suggest that configural information is more salient than featural information when we try to identify faces (Sergent, 1984; Hayes, 1988). With regard to hemispheric specialization for face processing recent neuroimaging evidence suggests that the right hemisphere (RH) is specialized for configural processing (Rossion et al., 2000). Supporting this view, the RH superiority in performance for faces decreases or disappears with face inversion (Hillger and Koenig, 1991). In addition, a MEG study (Watanabe et al., 2003) suggests a RH preference for upright faces since it found that latencies in the
A.M. Passarotti et al. / NeuroImage 34 (2007) 1708–1722
M170 (a face-selective component that possibly originates in the fusiform gyrus) are shorter in the RH for upright faces than for inverted faces. On the other hand, some evidence suggests that both configural and featural processing contribute to expert face processing and are affected by face inversion. Recent findings with normal subjects (Riesenhuber et al., 2004; Yovel and Kanwisher, 2004) and subjects with developmental prosopagnosia (Yovel and Duchaine, 2006) suggest that face-selective mechanisms process faces “holistically”, by extracting both configural and featural information as a whole, and that holistic processing is disrupted by inversion. Either way, since the FIE is considered an indicator of face processing expertise, understanding whether and how the behavioral FIE emerges with age would provide a crucial insight on the development of face processing expertise. The literature examining the behavioral FIE in childhood does not provide clear-cut results, probably because of the use of different paradigms and measures of performance (Brace et al., 2001). Some studies suggest a qualitative change in FIE with age, while others suggest only quantitative changes or no change at all with age. Carey (Carey, 1992; Carey and Diamond, 1977) reported a significant “reversed” behavioral FIE (i.e., better performance for inverted than for upright faces) in children younger than age 10, while Brace et al. (2001) reported this result only in children younger than age 5. These studies would suggest a qualitative switch from featural processing (which is less sensitive to orientation) to configural processing (which is sensitive to orientation) with age. But other studies found inversion effects in younger children (Baenninger, 1994; Chun and Thomson, 1995; Flin, 1985) and suggest that during development either the size of the FIE increases (Mondloch et al., 2002) or remains stable (Young and Bion, 1980; Pascalis et al., 2001). The neural underpinnings of the behavioral FIE have been explored almost exclusively in adults. Most fMRI studies on inverted face processing have found a small but significant reduction in activation in the face fusiform area (FFA) (Yovel and Kanwisher, 2004; Gauthier et al., 1999; Haxby et al., 1999; Kanwisher et al., 1998), sometimes accompanied by increased activation in inferior temporo-occipital regions that usually process features (Haxby et al., 1999; Kanwisher et al., 1998; Leube et al., 2003). Neuropsychological evidence also suggests that for inverted faces the FFA may need additional input from object-processing regions (Haxby et al., 1999; Leube et al., 2003; Rossion et al., 2003). In fact, a patient with object agnosia (i.e., with object processing deficits) could process upright faces but showed impaired processing of objects and inverted faces (Moscovitch et al., 1997), whereas patients with prosopagnosia (i.e., with deficits in discriminating between faces) usually exhibit the opposite pattern (Farah et al., 1995). Recent neurophysiological findings suggest that the neural correlates of the behavioral FIE undergo protracted developmental changes. The N170 is a face-specific, negative event-related potential (ERP) component, which may have its anatomical sources in the inferior occipito-temporal areas (Taylor et al., 2004; Bentin et al., 1996). In adults face inversion causes larger N170 latencies and greater amplitudes in the RH (Taylor et al., 2004). The N170 occurs as early as age 4, and even younger children show some evidence of FIE in their N170 profiles (Taylor et al., 1999). Nevertheless, different face processing strategies exhibit a different developmental timeline. Taylor et al. (2001) found that for featural face processing the N170 reaches adult-like profiles by age 11, while configural processing develops more
1709
slowly and improves through adolescence (Taylor et al., 2001, 2004). Moreover, different from the findings for featural processing, findings for inverted faces showed that the typical adult N170 latencies (with greater right hemisphere amplitudes) emerged only in mid-adolescence, suggesting that children did not just use feature processing with inverted faces, but probably used a combination of both feature and configural processing (Taylor et al., 2001, 2004). The view of gradual, quantitative changes in the neural substrates of face processing (Taylor et al., 2001, 2004) is confirmed by developmental fMRI studies on upright face processing which found more bilateral activation and widespread right fusiform activation in children than in adults (Passarotti et al., 2003), with differences in activation until late adolescence (Passarotti et al., 2001), as well as a progressive increase in right fusiform selectivity for faces compared to objects (Gathers et al., 2004; Aylward et al., 2005). Whereas these studies did not examine the neural underpinnings of the behavioral FIE, the present fMRI study is the first, to our knowledge, to establish the neural segregation and organization of the neural correlates of the behavioral FIE, from childhood to adulthood. We wished to address the crucial question of whether the effects of inversion on the functioning of faceselective regions may differ with development, and if so, in what fashion. By using fMRI during processing of upright and inverted faces we wanted to clearly localize the anatomical underpinnings of the FIE in children and adults. Furthermore, we tested children in the 8–11 age range and adolescents to address the existing debate on whether there is an abrupt, qualitative change in face processing skills around age 10 (Carey, 1992; Carey and Diamond, 1977) or whether development of face processing expertise is more gradual and quantitative in nature (Baenninger, 1994; Chun and Thomson, 1995; Taylor et al., 2004). With regard to the adult group, for inverted face processing we predicted a decrease in performance accompanied by a decrease in activation of faceselective regions in the right hemisphere, for inverted compared to upright faces (Yovel and Kanwisher, 2004; Leube et al., 2003). We expected teens to exhibit similar but somewhat weaker profiles of activation in face-selective regions as the adults, since expert configural processing is not reached until late adolescence (Taylor et al., 2004). As for the younger children, we predicted that their functional activation in face-selective regions may not differ significantly for upright and inverted faces, because children may use face-specific strategies less efficiently than adults, or they may use more general processing strategies that are not as sensitive to orientation (Baenninger, 1994; Taylor et al., 2001). To test our predictions we collected behavioral and fMRI data from adults, teens and children during a face-emotion processing task with inverted and upright face presentations. Face emotion processing, like face identity processing, relies mostly on configural or holistic strategies and shows comparable face inversion effects (Calder et al., 2000; McKelvie, 1995). We used angry and happy expressions because the ability to recognize them is fully developed by age 6 (Markham and Adams, 1992). Methods Participants Thirteen adults (aged 20–30 years), ten teens (aged 13– 17 years), and fourteen children (aged 8–11 years) were recruited
1710
A.M. Passarotti et al. / NeuroImage 34 (2007) 1708–1722
from the local community and Michigan State University. All participants were right-handed as assessed by a handedness questionnaire (Oldfield, 1971) and had normal or corrected to normal vision. They had no known-neurological or cognitive impairment. Adult participants were either volunteering or were given class credit. Children and teens received a monetary compensation ($10) for their 2-h participation. Before testing, adult participants and child parents/guardians signed an informed consent, while children and teens signed an assent form. A medical clearance questionnaire was administered to each participant to ensure scanning safety. All our forms and research protocols were approved by the Michigan State University Committee for Research with Human Subjects. fMRI data from 2 adults and 1 child were discarded because of technical problems, whereas data from 1 adult, 1 teenager, and 2 children were excluded because of low performance levels or excessive motion. The final fMRI dataset included 10 adults (5 F, 5 M; mean 25.3 ± 4 years), 9 teens (7 F, 2 M; mean 14.9 ± 1.6 years), and 11 children (5 F, 6 M; mean 10 ± 1.6 years). Due to technical difficulties with button press recordings, we did not collect behavioral data from one male child. Therefore our behavioral dataset included 10 children.
emotions (i.e., happy or angry). Each 5-s trial began with a 1000 ms fixation cross, followed by central presentation of a face picture for 2000 ms and a response time of 2000 ms. We decided to give participants a whole 2 s to process stimuli, to ensure that even the younger children would perform well. In addition, depending on the run the face stimuli were presented either upright or upside-down (see Fig. 1a for an example of an inverted face trial). On each trial of the control task participants gave a simple motor response (i.e., a key press) when the face stimulus (i.e., neutral face) appeared. These control trials were blocked and no cognitive decision was required. Order of task conditions (inverted, upright faces) and target emotion (happy, angry) was counterbalanced across participants.
Brain imaging procedures Participants learned the experimental tasks during a training session preceding scanning. In addition, children and teens listened to audio recordings of scanner noises to become more familiar and comfortable with them. Then, the scanning session started and lasted approximately 40 min. We adopted a standard block design to maximize signal-to-noise ratio for short imaging sessions. We alternated six 40-s blocks of each task and control in 4 min and 8 s. Eight trials (each lasting 5 s) were presented for each block, for a total of 48 trials in each run. Visual stimuli were projected directly on a small MR-compatible LCD screen (view angle: 12° vertically and 16° horizontally) positioned on the head coil inside the MRI scanner. Our visual stimuli were presented through the IFIS-SA System (Psychology Software Tools Inc., Milwaukee, WI). Button presses were performed with a special mouse, part of the IFIS-SA system, and were recorded at millisecond accuracy. BOLD images were acquired on a 3 T magnet (G.E. Medical Systems, Pittsburgh, PA) using a standard GE quadrature head coil. After an initial coronal localizer and manual shimming, T2*weighted single-shot gradient-echo planar images (EPI) were acquired with TR = 2 s, TE = 40 s, FOV = 220 mm, flip angle = 80°, 25 contiguous 5.5 mm axial slices, 120 images, 64 × 64 matrix, in-plane resolution, 3.4 × 3.4 × 5.5 mm. For each participant, these images were registered to high-resolution T1weighted structural images (116 sagittal fast SPGR images, FOV 256, 1.5 mm slices). We used padding materials to stabilize participants’ heads, and earplugs to attenuate the scanner noise. We showed to children and teens cartoon videos during the initial localizer protocols and during anatomical acquisitions, to keep them relaxed and still. Behavioral task During scanning, on each trial participants saw a color picture of a female face presented centrally on a black background, and, depending on the run, responded selectively to one of two face
Fig. 1. Example trial and behavioral results. (a) An example of the visual display and trial timeline. Here we present an inverted face trial. Our face pictures were taken from the NimStim Face Stimulus Set. (b) Mean proportion correct for upright and inverted face trials in children (N = 10), teens (N = 9), and adults (N = 10). (c) Mean RT (ms) for performance on upright and inverted face trials in children, teens, and adults. Asterisk indicates significance at p < .05. Error bars in panels b, c represent the SEM.
A.M. Passarotti et al. / NeuroImage 34 (2007) 1708–1722
fMRI image processing Anatomical and functional images were processed in AFNI (Cox, 1996). To control for initial magnetic field inhomogeneities,
1711
the first four time points of all functional runs were discarded a priori. In-plane (slice-wise or 2D) and volumetric (3D) motion correction was performed on the dataset using the AFNI programs 2dImReg and 3dvolreg, respectively, using the third acquisition as
Table 1 Significant clusters of activation, their laterality, and their Talairach coordinates (Talairach and Tourneaux, 1988) for the upright and inverted face conditions, in children (N = 11), teens (N = 9), and adults (N = 10) Upright faces
Inverted faces
x
y
z
BA
Hemisphere
Region
x
y
z
BA
Hemisphere
Region
Adults (N = 10)
− 34 39 − 58 63 − 47 − 63 57 − 33 58 − 59 64 −9 11
45 40 55 44 − 42 − 30 37 84 66 60 28 24 24
− 18 − 19 − 12 − 12 17 3 0 −7 −2 6 9 35 33
20/37 20/37 20/21 20/21 9 20/22 20/22 18/19/37 18/19/37 22 22 20 20
L R L R L L R L R L R L R
LFG LFG IT gyrus IT gyrus MF gyrus MT gyrus MT gyrus IOG IOG STS STS MFG MFG
Teens (N = 9)
− 40 38 − 21 14 45 52 − 56 58 − 58 − 27 42 − 42 64
− 45 − 41 36 31 43 48 32 − 16 32 − 84 − 60 − 43 −9
− 15 − 13 30 33 40 7 − 15 − 20 − 13 −7 −6 −4 11
37 37 20 20 40 20/22 20/21 20/21 20/22 18/19/37 18/19/37 22 22
L R L R R R L R L L R L R
LFG LFG MFG MFG IP lobule MT gyrus IT gyrus IT gyrus MT gyrus IFG IFG STS STS
− 45 36 −7 15 18 52 22 − 64 56 − 45 45 − 20 57 − 42 48
41 42 54 55 96 − 14 42 − 16 −9 − 49 − 48 81 − 67 − 47 − 51
−9 − 14 −3 4 −5 45 62 − 15 − 20 2 2 −5 −3 −1 3
37 37 20 20 18 9 7 20/21 20/21 20/22 20/22 18/19/37 18/19/37 22 22
L R L R R R R L R L R L R L R
LFG LFG MFG MFG Ling. gyrus MF gyrus SP lobule IT gyrus IT gyrus MT gyrus MT gyrus IOG IOG STS STS
− 44 35 −9 11 56 − 48 − 32 − 55 4 − 10 55 − 40 − 59 58 − 63 57 − 39 58 − 57 56 19 − 45 40 − 21 14 − 52 56 − 59 67 − 19 46 − 54 48 − 44 36 −7 15 − 27 − 35 −6 − 30 − 52 56 − 47 49 − 29 40 − 51 47
49 40 24 24 − 16 10 −2 −9 72 87 54 − 20 49 29 − 30 37 86 60 − 39 42 38 − 47 − 38 36 31 − 53 − 33 − 49 − 46 − 92 73 46 − 51 44 42 54 55 − 23 − 48 49 68 43 39 −8 25 81 72 50 49
20 − 21 35 33 13 − 28 − 38 − 18 41 43 13 − 20 − 13 − 14 3 0 −7 −2 8 9 28 − 20 − 21 30 33 − 13 − 15 −1 2 −6 −8 −2 4 − 11 − 12 −3 4 −5 −8 9 58 − 13 − 13 − 35 − 14 −4 −4 −1 5
37 37 20 20 47 21 20/22 20/22 7 7 22 22 20/21 20/21 20/22 20/22 18/19/37 18/19/37 22 22 24 37 37 20 20 20/21 20/21 20/22 20/22 18/19/37 18/19/37 22 22 37 37 20 20 47 9 23 7 20/21 20/21 20/22 20/22 18/19/37 18/19/37 22 22
L R L R R L L L R L R L L R L R L R L R R L R R L L R L R L R L R L R L R L L L L L R L R L R L R
LFG LFG MFG MFG IF gyrus IT gyrus MT gyrus MT gyrus Precuneus Precuneus ST gyrus ST gyrus IT gyrus IT gyrus MT gyrus MT gyrus IOG IOG STS STS Cing. gyrus LFG LFG MFG MFG IT gyrus IT gyrus MT gyrus MT gyrus IOG IOG STS STS LFG LFG MFG MFG IF gyrus MF gyrus Post. Cing. SP lobule IT gyrus IT gyrus MT gyrus MT gyrus IOG IOG STS STS
Children (N = 11)
The Talairach coordinates (in the x, y, z axes) for our axial images indicate the center of mass of each significant cluster of activity. X = right to left; Y = anterior to posterior; Z = superior to inferior. Note that in our X coordinates the right (R) hemisphere has a positive sign and the left (L) hemisphere has a negative sign. Abbreviations: lateral fusiform gyrus = LFG; medial fusiform gyrus = MFG; inferior temporal gyrus = IT gyrus; middle temporal gyrus = MT gyrus; inferior occipital gyrus = IOG; superior temporal sulcus = STS; inferior parietal lobule = IP lobule; superior parietal lobule = SP lobule; inferior frontal gyrus = IFG; middle frontal gyrus = MF gyrus; Ligual gyrus = Ling. gyrus; cingulate gyrus = Cing. gyrus; posterior cingulate = Post. Cing.
1712
A.M. Passarotti et al. / NeuroImage 34 (2007) 1708–1722
the fiducial volume. The signal was then orthogonalized to the obtained parameters of rigid body rotation and global drift was removed on a voxel-wise basis over each functional run. For volumetric motion correction, an iterated, weighted linear least squares algorithm with Fourier interpolation (Cox and Jesmanowicz, 1999) was utilized to register all dataset volumes to the third acquired T2*-weighted functional volume and time shifted to the fiducial volume in order to account for slice acquisition offset. We considered the algorithm iterations “convergent” when maximum movement was less than .02 voxels and maximum rotation was less than .03°. There were no time points at which these constraints were exceeded; angular and linear drift along and around the x, y, and z axes of the dataset were quantified as vector components and used as regressors in further analyses. High- and low-bandpass filters were applied to the data (using the AFNI program 3dFourier) at .20 Hz and .60 Hz, respectively, in order to suppress predictable pulsatile physiologic noise such as cardiac and respiratory rhythms. Stimulus-related activation was detected by examining the time-course correlation of each voxel’s signal intensity with a family of idealized hemodynamic response waveforms (Bandettini et al., 1993). High- and low-bandpass filters were applied to the data (using the AFNI program 3dFourier) at .20 Hz and .60 Hz, respectively, in order to suppress predictable pulsatile physiologic noise such as cardiac and respiratory rhythms. Stimulus-related activation was detected by examining the time-course correlation of each voxel’s signal intensity with a family of idealized hemodynamic response waveforms (Bandettini et al., 1993). Voxels correlating at a coefficient threshold of .23 (p < .001, uncorrected) were retained. For each individual, statistical maps were then overlaid on the coregistered high-resolution normalized anatomical dataset (Talairach and Tourneaux, 1988). Child brain normalization is an accepted protocol in developmental fMRI studies, since total brain size does not increase significantly between ages 9 and 18 (Klingberg et al., 2002). Therefore data from 8 to 9 year olds and adults can be effectively transformed into the same stereotactic space (Schlagger et al., 2002; Kang et al., 2003) with minimal agerelated difference. The average motion (roll, pitch, yaw) for children was .049°, for adults .020°, and for teens .017°. The average motion for children was .055 mm, for adults .017 mm, and for teens .022 mm. Student’s t-tests revealed that the child group differed significantly (p < .05) from the other two groups. However, tolerance for motion correction in AFNI was .07 mm × .03° between slices and only subjects in any age cohort who did not exceeded these tolerance parameters were kept in our analyses. For each subject we obtained mean volume and % signal separately for upright and inverted face emotion trials using a cross-correlation analysis (r > .23) in which we identified contiguous voxels that correlated to a modified waveform with threshold at p < .01. We followed the same procedure for neutral trials. Our analyses focus on upright and inverted face emotion trials, since these are the trials that required a cognitive decision and therefore ensured accurate processing. Then, Cluster analyses were carried out in AFNI. Minimum alpha (confidence) levels were computed using Monte Carlo noise simulations in the AFNI program AlphaSim (Ward, 1997). In this step, adjacent voxels with p values falling below .001 had to form connected regions of at least 195 mm3 for significant activation. In addition to individual cluster analyses, an additional cluster analysis was performed on each group’s statistical maps.
ROI definition We examined three core regions in extrastriate cortex that have been defined as part of a distributed face processing system (Haxby et al., 2000). Although the functional role of these regions and how they interact with each other is still not completely understood, several studies suggest that the lateral fusiform gyrus (LFG) is a face-selective region that processes invariant aspects of faces, such as identity (Haxby et al., 1999; Haxby et al., 2002; Kanwisher et al., 1997), whereas the superior temporal sulcus (STS) processes changeable aspects of faces, such as gaze orientation or face expressions (Hoffman and Haxby, 2000; Puce et al., 1998; Rolls, 1984) and may be one of the sources of the N170 (Puce et al., 1998). The inferior occipital gyrus (IOG) processes single face features (Haxby et al., 1999) and may either feed featural information onto the STS and LFG (Hoffman and Haxby, 2000; Haxby et al., 1999), or receive feed-back signals from the LFG to guide fine-grained feature analyses (Rossion et al., 2003). In addition, we examined the medial fusiform gyrus region (MFG), which is involved in object processing in adults (Ishai et al., 1999), but has been found to contribute considerably to face processing in children (Passarotti et al., 2003). All our participants showed bilateral clusters of activation in these regions (see Table 1). Our ROIs were defined anatomically, using anterior and posterior boundaries as defined in previous studies (Gauthier et al., 1999). Since there is evidence of minimal age-related differences in spatial transformation after age 8 (e.g., Kang et al., 2003; Schlagger et al., 2002) we did not trace the ROIs for each subject individually, rather we defined each ROI anatomically on a template (Talairach and Tourneaux, 1988), and applied it to the normalized brains within each group, as several developmental fMRI studies did (e.g., Nelson et al., 2003; Passarotti et al., 2003). For the LFG ROI we first isolated in each subject a significant continuous cluster of activation with center of mass coordinates that were similar to the ones given by Grill-Spector and Kanwisher (2005) and Haxby et al. (2002). We then placed an eight-voxel bounding box around the individuals’ cluster center of mass, which corresponded to the spatial extent of the FFA given in Rossion et al. (2003) and Kanwisher et al. (1997), and averaged the masked individual clusters across subjects within each of the age groups. For each group we had therefore an averaged cluster, which we found to overlap with the anatomically defined lateral fusiform region (Haxby et al., 2002; Grill-Spector and Kanwisher, 2005; Rossion et al., 2003). Therefore we adopted the anatomically defined LFG as our ROI. The IOG, STS and medial fusiform gyrus (MFG) were also derived from functional clusters and ultimately anatomically defined (Talairach and Tourneaux, 1988). Separate ROIs were obtained for the left and the right hemisphere. Results Behavioral performance and behavioral FIE Separate ANOVAs were carried out for Reaction time (RT) and Accuracy data. Incorrect button presses were excluded from the RT analyses but were included in the Accuracy analyses. Face orientation (upright, inverted) was the within-subjects factor whereas Age (children, teens, and adults) was the between-subjects factor. When an Age effect was found separate ANOVAs by Age were also carried out to further examine performance within each age group.
A.M. Passarotti et al. / NeuroImage 34 (2007) 1708–1722
Accuracy The Accuracy results are summarized in Fig. 1b. Accuracy levels were above 90% in each age group. The teens’ accuracy did not differ from that of children and adults (F values < 1), but adults had a significantly higher accuracy than children [F(1,18) = 5.34; p = .03]. In an ANOVA with Age (children, teens, adults) and Orientation (upright, inverted) as factors a significant Face orientation effect [F(1,26) = 5.56; p = .03] demonstrated that overall Accuracy was significantly higher for upright (.97) than for inverted faces (.94). Nevertheless, the interaction of Age × Orientation was not significant (F < 1), suggesting that there were no robust age differences in FIE for the accuracy data. Separate ANOVAs by Age confirmed that accuracy for upright and inverted faces did not differ significantly in either children [F(1,9) = 1.95. p = .19], teens [F(1,8) = 2.61 p = .14], or adults (F < 1). Since participants had quite a long time to process stimuli and to respond (i.e., 2 s), it is possible that some ceiling effects occurred and weakened Age × Orientation effects in the accuracy data. Therefore these data need to be considered with caution. We
1713
turn now our attention to the RT data, which have been found to be more useful in limiting ceiling effects (Ellis, 1990). Reaction time The RT results are summarized in Fig. 1c. As expected, a significant main effect of Age indicated that RT improved steadily with age for both upright and inverted face presentations [F(2,26) = 19.28; p = .00007]. Planned comparisons showed that the three groups differed significantly from each other (p < .001). An inspection of the individual data revealed that whereas in 80% of the adults RT was elongated for inverted faces, in children and teens RT was more variable and did not show a consistent cost for face inversion (there was a RT cost for inversion only in about 40% of children and 44% of teens). The majority of children (60%) exhibited a trend for faster RT with inverted than with upright faces. In spite of these trends, the interaction of Age × Orientation was not significant (F < 1). Separate ANOVAs by Age confirmed that the adult RT was significantly higher for inverted faces (858 ms) than for upright
Fig. 2. fMRI maps of significant (p < .01, corrected) group cluster activation in children (N = 11), teens (N = 9), and adults (N = 10) for the upright face task (left side) and the inverted face task (right side). See also Table 1. The particular axial slice represented in this figure shows the right and left Fusiform Gyrus. Significant functional clusters of activation were overlaid on a T1 anatomical image from one participant. The Z coordinate (Z = − 11) indicates the distance in mm of the axial slices (i.e., according to a superior to inferior axis) from the intercommissural plane (anterior commissure − posterior commissure). The color bar on the top right side represents the scale of percent signal change increases in the experimental tasks compared to the control task. Note that brain images follow the radiological convention [i.e., the left side of the brain picture represents the right hemisphere (R), the right side of the brain picture represents the left hemisphere (L)].
1714
A.M. Passarotti et al. / NeuroImage 34 (2007) 1708–1722
faces (767 ms) [F(1,9) = 7.48, p = .02]. The size of the adult behavioral FIE (about 12%) is comparable to that found in other studies (Aguirre et al., 1999; Diamond and Carey, 1986; Itier and Taylor, 2004; Valentine, 1988). Conversely, in children [F(1,9) = .11, p = .74] and teens [F(1,8) = .12, p = .74] RT did not differ significantly for upright and inverted faces. ROI analyses and neural FIE As expected, significant clusters of activity were found in the face-selective region LFG (BA 37), in STS (BA 22), in IOG (BA 19), and MFG (BA 20) (see Fig. 2, and Table 1). We report below percent signal change analyses separately for each ROI. An initial ANOVA was run with Age group (children, teens, adults) as a between-subjects factor and Hemisphere (left, right) and Orientation (inverse, upright) as within-subjects factors. When the factor of Hemisphere interacted significantly with other factors we carried out separate analyses by hemisphere. Lateral fusiform gyrus (LFG) ROI In an initial ANOVA the three-way interaction of Age × Hemisphere × Orientation was significant [F(2,27) = 5.22, p = .01]. Planned comparisons on this interaction indicated that only children [F(1,10) = 8.83, p = .006] but not adults [F(1,9) = 1.92, p = .18] and teens (F < 1) demonstrated significant hemispheric differences in profiles of activation for upright and inverted faces. Moreover, it was only in the right LFG that the child FIE profile differed significantly from that of teens [F(1,27) = 10.77, p = .003] and adults [F(1,27) = 15.71, p = .0005], who did not differ from each other (F < 1). Given these findings, we proceeded to examine LFG data separately for each hemisphere. Importantly for our hypotheses, for the right LFG, we found a significant interaction of Age × Face Orientation [F(2,27) = 9.19, p < .0009]. Planned comparisons revealed that only for inverted faces did children show significantly higher percent signal change than adults [F(1,27) = 13.54, p < .001] and teens [F(1,27) = 9.59, p < .005], who in turn did not differ from each other (F < 1). No significant age differences in levels of activation were found for upright faces (F < 1). With regard to the neural FIE (i.e., the difference between % signal change for upright faces and % signal change for inverted faces), in accord with the RT data adults exhibited the expected FIE, with a higher percent signal change for upright (.28%) than for inverted (.09%) faces [F(1,27) = 4.81, p < .04]. This adult pattern was similar to the one found in other studies (Yovel and Kanwisher, 2004, 2005) and is usually attributed to hindrance of face-specific processes, defined either as configural (Haxby et al., 1999; Leube et al., 2003) or holistic (Yovel and Kanwisher, 2004; Riesenhuber et al., 2004) processes. Like in adults, also in teens activation was higher for upright (.26%) than for inverted (.15%) faces, although not significantly [F(1,8) = 4.81, p < .20]. On the contrary, in children there was a “reversed” neural FIE, in that children actually showed a higher percent signal (.49%) for processing of inverted faces than of upright faces (.22%) [F(1,10) = 11.86, p < .002] (Fig. 3a). The child neural FIE differed significantly from that of adults [F(1,27) = 15.71, p = .0005] and teens [F(1,27) = 10.77, p = .003], who did not differ from each other (F < 1). With regard to the left LFG, a significant Age effect [F(2,27)=4.02, p < .03] revealed that children exhibited significantly more activation than teens (p
