Impaired behavioral and neural representation of scenes in Williams syndrome

Abstract

Boundaries are crucial to our representation of the geometric shape of scenes, which can be used to reorient in space. Behavioral research has shown that children and adults share exquisite sensitivity to a defining feature of a boundary: its vertical extent. Imaging studies have shown that this boundary property is represented in the parahippocampal place area (PPA) among typically developed (TD) adults. Here, we show that sensitivity to the vertical extent of scene boundaries is impaired at both the behavioral and neural level in people with Williams syndrome (WS), a genetic deficit that results in severely impaired spatial functions. Behavioral reorientation was tested in three boundary conditions: a flat Mat, a 5 cm high Curb, and full Walls. Adults with WS could reorient in a rectangular space defined by Wall boundaries, but not Curb or Mat boundaries. In contrast, TD age-matched controls could reorient by all three boundary types and TD 4-year-olds could reorient by either Wall or Curb boundaries. Using fMRI, we find that the WS behavioral deficit is echoed in their neural representation of boundaries. While TD age-matched controls showed distinct neural responses to scenes depicting Mat, Curb, and Wall boundaries in the PPA, people with WS showed only a distinction between the Wall and Mat or Curb, but no distinction between the Mat and Curb. Taken together, these results reveal a close coupling between the representation of boundaries as they are used in behavioral reorientation and neural encoding, suggesting that damage to this key element of spatial representation may have a genetic foundation.

Introduction

Boundaries are crucial to our representation of the geometric shape of different types of scenes, such as outdoor landscapes, city streets, and interior rooms. When human adults, children, and animals are disoriented, they often reorient themselves by re-establishing the relationship between the direction of their own heading and the geometry of the external environment, searching for a hidden target at the correct corner or its (erroneous) geometric equivalent (Cheng, 1986; Cheng & Gallistel, 1984; Hermer & Spelke, 1994). These results demonstrate that the geometric shape of the environment, often defined by its boundaries, plays a primary role in reorientation by humans as well as many other species (see Cheng & Newcombe, 2005, for review).

However, recent studies indicate that not all boundaries are equally effective in defining geometry. Most studies of reorientation have used enclosures with complete walls to define the boundaries of the array, and even 2-year-olds use the geometry of such boundaries to reorient themselves (Learmonth, Newcombe, & Huttenlocher, 2001). Lee and Spelke (2008; 2011) found that effective boundaries must include some degree of vertical extent, however slight: 4-year-old children can reorient themselves when array boundaries are only 2 cm in height, but not when the boundaries are replaced by a flat mat on the floor (see Figure 1). Just like children, chicks will spontaneously reorient by a 2 cm curb boundary, but will fail to reorient by boundaries defined by 2D brightness contours (Lee, Spelke, & Vallortigara, 2012). These findings demonstrate that vertical extensions of boundary structure are crucial input for the reorientation mechanism; sensitivity to this property is evidenced early in human life and across species. Moreover, in human adults, studies show that a part of the visual scene network, the parahippocampal place area (PPA, Aguirre et al., 1996; Epstein & Kanwisher, 1998) distinguishes between “open” scenes (those with little to no boundary structure, such as a field) and “closed” scenes (those with high amounts of boundary structure, such as an indoor room, Kravitz, Peng, & Baker, 2011; Park et al., 2011). Recent research has also found that the PPA is sensitive to even minimal amounts of boundary structure (Ferrara & Park, 2016), echoing children’s use of a small curb boundary to guide their behavioral reorientation.

Figure 1.

The three boundary conditions of Experiment 1: Mat (1.2 m × 1.8 m), Curb (1.2 m × 1.8 m × 5 cm), and Wall (1.2 m × 1.8 m × 2 m) (following Lee & Spelke, 2008).

Here, we present evidence that this sensitivity to boundary structure is disrupted in Williams syndrome (WS), a genetic disorder. WS is characterized by a deletion of approximately 25 genes on chromosome 7q11.23 (Morris, 2006) and presents with a cognitive profile of intellectual disability and severe impairment in a range of spatial functions (Atkinson et al., 2001; Landau & Ferrara, 2013; Lakusta et al., 2010; Mervis et al., 2000; see Landau & Hoffman, 2012, for review). For example, Ferrara and Landau (2015) found that individuals with WS require especially salient presentation of geometric information in order to reorient geometrically (i.e., a chamber with 4 walls above eye-level in height, where all surface junctures and hiding locations are clearly visible). Impairments are also found in real-world and virtual reality navigation tasks that test route learning (Farran et al. 2010; 2012), as well as use of spatial frames of reference to localize objects (Nardini et al., 2008; Paul et al., 2002; Vicari et al., 2005, 2006). Corresponding to these known spatial impairments, individuals with WS show anatomical and functional atypicalities in the hippocampus (Meyer-Lindenburg et al., 2005; 2006), a region that is crucial for navigation and is highly connected with the PPA (Baldassano et al., 2016). These behavioral and neural atypicalities raise the question of whether parallel deficits exist in the representation of one specific environmental cue that is crucial to both geometric reorientation and visual scene encoding: boundary structure.

In the present study we investigate this link by testing both behavioral and neural sensitivity to boundaries in individuals with WS. In Experiment 1, participants were disoriented and then recovered a hidden target in three different arrays that varied in terms of boundary structure: a Mat, a Curb, and full Walls. In Experiment 2, fMRI was used to measure the neural response in the same participants to scene images that depicted the boundary arrays used in Experiment 1: a Mat, a Curb, and a Wall. We hypothesized that people with WS would show behavioral deficits in using some types of boundaries for reorientation when compared to TD age-matched controls. We further hypothesized that people with WS would show a deficit in neural sensitivity to particular boundary types, possibly those which they cannot use to accomplish reorientation.

Experiment 1: Behavioral Reorientation

Methods

In this section, we report how we determined our sample size, all data exclusions, all inclusion/exclusion criteria, whether inclusion/exclusion criteria were established prior to data analysis, all manipulations, and all measures in the study.

Participants.ⁱ

Two main participant groups were tested in Experiment 1: 15 individuals with WS (Mean age = 22.36 years; Range = 17.13–33.74 years; SE = 1.79 years), and 15 typically developing (TD) individuals matched to the WS participants for chronological age and gender, henceforth referred to as TD age-matched controls (Mean age = 22.25 years; Range = 17.35–33.92 years; SE = 1.82 years). We also tested 18 TD 4-year-olds (Mean age = 4.56 years; Range = 4.0–4.86 years; SE = 0.11 years), because previous research has shown that TD children of this age can reorient according to the geometry of both Wall and Curb boundaries, but not 2D Mat boundaries (Lee & Spelke, 2011). Inclusion of this group ensured that our methods and procedures could replicate this pattern. Sample size was determined based on prior reorientation studies of individuals with WS (Landau & Lakusta, 2010) and young TD children (Lee & Spelke, 2008).

All participants with WS had the characteristic genetic deletion on the long arm of chromosome 7, as determined by FISH (fluorescent in situ hybridization) test (Ewart, 1993). Chronological age matches were made such that the participant with WS and the TD individual had birthdates that were within 3 months of one another. All participants had normal or corrected-to-normal vision. The study protocol was approved by the Institutional Review Board of Johns Hopkins University. Prior to the start of the study, written informed consent was provided by TD age-matched controls and by the legal guardians of WS individuals and TD children. WS individuals and TD children provided verbal assent and/or written consent. WS individuals were recruited through the Williams Syndrome Association. TD age-matched controls were recruited via on-campus flyers and daily email announcements from the University. TD children were recruited via local day cares and preschools. At the end of the study, TD children chose a small toy to take home. WS participants and TD age-matched controls received monetary compensation for their time.

To assess the overall cognitive profile of participants, the Kaufman Brief Intelligence test (K-BIT II; Kaufman & Kaufman, 2004) was administered. The K-BIT II is a standardized measure normed for TD individuals ages 4 to 90 years and yields an overall composite IQ score. WS participants in the present study had an average IQ of 84 (SD = 9.69, range = 67–109). The Pattern Construction subtest of the Differential Abilities Scale (DAS-II; Elliott, 1990) was also administered. It requires participants to copy the pattern of a model by assembling sets of blocks. It is a key diagnostic of the spatial deficit in WS (Mervis et al., 2000). WS participants had an average ability score of 234.93 (SD = 44.20, Range = 104–276) and an average TD age equivalent of 8.43 years (SD = 2.24 years, Range = 3.33–11.25 years), which is much younger than their chronological age (Mean = 22.25 years). These scores are consistent with those observed among individuals with WS in previous research (Mervis et al., 2000; Landau & Hoffman, 2012; Pitts & Mervis, 2016).

Specific subtests of the Differential Abilities Scale (DAS II; Elliott, 1990) were also administered to assess working memory ability: Digit Span (Forward and Backward), and Recall of Sequential Order. The Digit Span (Forward) subtest requires participants to repeat ordered sequences of numbers that have been dictated to them by an experimenter. The Digit Span (Backward) subtest requires the participant to reverse the order of the numbers. On the Forward subset, WS participants had an average ability score of 152.33 (SD = 20.77, Range = 120 – 184), and an average TD age equivalent of 9.15 years (SD = 3.3 years, Range = 5.33 – 15.75 years). On the Backward subset, WS participants had an average ability score of 114.0 (SD = 25.33, Range = 73 – 151), and an average TD age equivalent of 11.25 years (SD = 4.04 years, Range = 6.08 – 17.75 years). In the Recall of Sequential Order subtest, the experimenter dictates a list of body part names, to which the participant responds by naming the body parts from the spatially highest to the lowest. WS participants had an average ability score of 95.93 (SD = 44.20; Range = 58 – 122), and an average TD age equivalent of 7.63 years (SD = 1.37 years; Range = 5.58 – 8.83 years). Standardized IQ measures were not administered to TD age-matched controls or TD 4-year-olds.

Design.

Experiment 1 was conducted in a room with a uniform white circular hanging curtain (2.3 m in diameter). The curtain could be parted at a single point to allow for entry and exit, and this opening was visibly concealed once the participant entered the space. The floor was covered by a solid gray rug. Ceiling lights were symmetrically spaced in a circle around the edge of the curtain so as to prevent them from being used as polarizing cues that could help to orient oneself within the room. The room was protected from outside noise by four equally spaced white noise machines that were positioned beyond the barrier of the curtain.

Three within-subject experimental conditions tested reorientation within three different boundary arrays that varied in terms of amount vertical structure: Mat, Curb, and Wall (Figure 1). Each boundary array measured 1.8 m width × 1.2 m length and was situated at the center of the circular space. For the Mat condition, a flat black mat was placed on the floor. For the Curb condition, connected wooden beams that stood 5 cm high were placed around the perimeter of the mat. For the Wall condition, the curb structure was removed and was replaced by black felt panels that stood 2 m high. For all three conditions, circular black metal tins (3 cm high, 8 cm diameter) served as hiding containers, with one placed at each of the four corners of the array. Condition order was counterbalanced within each participant group. The hiding location of the stickers remained constant across the test trials and conditions for each participant (as in Ferrara & Landau, 2015; Hermer & Spelke, 1996; Lakusta et al., 2010; Lee & Spelke, 2008; 2011) and was randomized across participants.

Procedure.ⁱⁱ

Following a well-established reorientation procedure (Hermer & Spelke, 1994; 1996), participants observed a sticker being hidden in a particular container located at one of the four corners of the array. Participants were then led to the center of the array and were blindfolded. The experimenter guided participants as they turned around in place for 10 s, switching the direction of rotation two times. The blindfold was then removed and disorientation was checked by asking the participant to point to the place where they had entered the chamber. Participants who accurately pointed to the entry (N = 5, across all groups) underwent the disorientation procedure once more and their disorientation was checked again. All participants were disoriented before proceeding to the test phase. Four test trials were included per boundary condition, following the established procedure in previous reorientation studies of individuals with WS (Landau & Lakusta, 2010) and young TD children (Lee & Spelke, 2008). On each of the test trials, participants ended their rotation by facing one of the 4 walls. Participants then removed their masks and were asked to pick the container that they thought held the sticker. If participants were incorrect in their first choice, they were encouraged to choose another location. If they were incorrect on their second choice, the experimenter led them to the correct container. The entire procedure (sticker hiding, disorientation, test) was repeated for a total of 4 trials in each boundary condition (12 trials total).

Resultsⁱⁱⁱ

Figure 2 shows the mean proportions of search at each of the four corners of the arrays for the WS participants (2.A), TD age-matched controls (2.B), and TD 4-year-olds (2.C). (See Supplemental Figure S1 for graphs illustrating the geometric performance of participants in each condition.) Reorientation using the geometry of these rectangular arrays would result in more frequent searches at the correct corner (C) and its rotational equivalent (R) compared to searches at the other two corners, those that were near (N) or far (F) from the correct corner.

Figure 2.

Average proportions of search at each corner (C = correct, R = rotationally equivalent, N = near, and F = far) for the A) WS participants, B) TD age-matched controls, and C) TD 4-year-olds. Because hiding location varied across participants, all data have been rotated into alignment.

Following Lakusta et al. (2010), performance was first analyzed using nonparametric statistics.^iv The number of participants who searched most often at the geometrically correct corners (C, R) were compared to those who searched most often at the non-geometric corners (N, F), and those who searched the two corner types an equal number of times (i.e., two searches to each corner type). A chi-square goodness of fit test was used to determine whether these observed patterns differed significantly from what would be expected by chance (50%).

In the Mat condition, 5 of the WS participants searched more often at the geometrically appropriate corners (C, R), 3 participants searched more often at the N and F corners, and 7 searched at the two corners types an equal number of times. In this condition, the number of WS participants who searched more often at the geometrically appropriate corners (C, R), did not differ from chance (χ² (2, n = 15) of 1.6, p = .45), demonstrating a random pattern of search. In the Curb condition, 3 of the WS participants searched most often at the geometrically appropriate corners, 7 participants showed the opposite pattern, and 5 searched at the two corners types an equal number of times. The group of WS participants again did not reliably use the geometry of the array (χ² (2, n = 15) of 1.6, p = .45). In the Wall condition, 12 of the WS participants searched most often at C and R, 0 participants showed the opposite pattern, and 3 searched at the two corners types an equal number of times. The number of WS participants who searched more often at the geometrically appropriate corners did significantly differ from chance (χ² (2, n = 15) of 15.6, p < .001), showing consistent use of geometry to reorient by the full Wall boundaries.

By comparison, in the Mat condition, 14 of the TD age-matched controls searched most often at the geometrically appropriate corners, 0 participants showed the opposite pattern, and 1 searched at the two corners types an equal number of times. These numbers significantly differed from chance (χ²(2, n = 15) of 24.4, p < .001), indicating that TD age-matched controls were able to reorient according to the geometry of the Mat. In the Curb and Wall conditions, all 15 of the TD age-matched controls searched most often at the geometrically appropriate corners, which significantly differed from chance (χ² (2, n = 15) of 30, p < .001).

In the Mat condition, 4 of the TD 4-year-olds searched most often at the geometrically appropriate corners, 7 participants showed the opposite pattern, and 7 searched at the two corners types (geometric and non-geometric) an equal number of times. The number of TD 4-year-olds who searched at the geometrically appropriate corners in the Mat condition did not differ from chance (χ² (2, n = 18) of 1.0, p = .61), indicating a failure to use geometry in this condition. In the Curb condition however, 11 of the 4-year-olds searched most often at the geometrically appropriate corners, 2 participants showed the opposite pattern, and 5 searched at the two corners types an equal number of times. This significantly differed from chance (χ² (2, n = 12) of 7.0, p = .03), indicating that the 4-year-olds reoriented according to the geometric boundary structure of the small Curb. This pattern was also found for the Wall condition, where 14 of the 4-year-olds searched most often at the geometrically appropriate corners, 1 participant showed the opposite pattern, and 3 searched at the two corners types an equal number of times (χ² (2, n = 12) of 16.33, p < .0001), demonstrating a consistent use of the geometry of the Wall boundary to reorient themselves.

We next evaluated performance using independent samples two-tailed t-tests to compare the number of geometric searches (C+R) made by a) TD age-matched controls vs. WS participants and b) TD 4-years-olds vs. WS participants. In the Mat condition, TD adults made significantly more searches to the geometric corners than WS participants (t(28) = 5.16, p < .001, Cohen’s d = 1.89). TD 4-year-olds and WS participants did not differ from one another, as both showed a random pattern of search in the Mat condition (t(31) = 0.71, p = .48, Cohen’s d = 0.25). In the Curb condition, both TD controls (t(28) = 5.56, p < .001, Cohen’s d = 2.07) and TD 4-year-olds (t(31) = 2.23, p = .03, Cohen’s d = 0.78) made significantly more searches to the geometric corners in comparison to individuals with WS. In the Wall condition, neither TD age-matched controls (t(28) = 1.38, p = .18, Cohen’s d = 0.50) nor TD 4-year-olds (t(31) = 0.42, p = .68, Cohen’s d = 0.15) differed from WS participants, consistent with the observation that all groups reoriented according to the geometry of the array in this condition.

Finally, we asked whether the behavioral performance of WS individuals was related to their scores on standardized IQ measures; the K-BIT, which yields an overall IQ composite score, the Pattern construction subtest of the DAS (a strong diagnostic of the WS spatial deficit, Mervis et al., 2000), or measures of working memory and recall (Digits Forward and Backward, and Recall of Sequential Order subtests of the DAS). Partial correlations controlling for age were conducted between the number of geometric searches in each condition and each participant’s assessment score. No relationships emerged as significant (all rs < .35, ps > .34).

Overall, WS participants showed a qualitatively different pattern from both TD age-matched controls and TD children, in particular, failing to use a small amount of vertical boundary structure in the Curb condition to reorient themselves in space. We hypothesized that this observable impairment at the behavioral level may be linked to atypical patterns of neural processing in areas that are known to be crucially important for the representation of boundaries in scenes. We tested this hypothesis in Experiment 2.

Experiment 2: Neural Representation of Scenes

In Experiment 2, we focused on a brain area that is known to be selectively involved in the representation of scenes: The parahippocampal place area (PPA) (Aguirre et al., 1996; Epstein & Kanwisher, 1998). This area responds strongly during passive viewing of navigationally relevant visual stimuli, such as scenes and buildings (Aguirre et al., 1998; Epstein & Kanwisher, 1998; Nakamura et al., 2000). The literature indicates that the PPA is involved in representation of local physical scene structure (Epstein et al., 2003; Park & Chun, 2009; Park et al., 2011). Furthermore, our previous research has shown that the PPA in TD adults is sensitive to even a small amount of 3D grounded boundary structure (Ferrara & Park, 2016). Thus in Experiment 2, we sought to functionally localize and probe the representation of scene boundaries in the WS PPA, hypothesizing that this may reveal atypical patterns of processing that mirror the behavioral impairment in using boundary structure to accomplish reorientation. Additional scene-selective regions of interest (ROIs) included retrosplenial complex (RSC) (Epstein 2008; Maguire et al., 1997; 1998) and the occipital place area (OPA, Dilks et al., 2013; Julian et al., 2016), also referred to as TOS (a region near the transverse occipital sulcus, Dilks et al., 2011; MacEvoy & Epstein, 2007; Nakamura et al., 2000). These areas are also functionally localized and show selective responses to scenes. Lastly, we focused on primary visual area V1 to determine whether there existed group differences in the initial stages of visual processing. If so, this could indicate that identified WS impairments may not be due to atypical representation of boundaries in scene areas per se, but rather due to impaired processing of low-level visual information that is then passed on to later stages in the visual processing hierarchy.

Methods

Participants.

Twelve of the WS participants from Experiment 1 and 12 of their respective TD age-matched controls participated in Experiment 2. Three of the WS participants who had completed Experiment 1 were not scanned due to the presence of metal in the body.

Visual Scene Stimuli.^v

Full-color artificial images were created using Autodesk Sketchbook Designer (2012) and Adobe Photoshop CS6 (as in Ferrara & Park, 2016). Mirroring the design of Experiment 1, there were three different boundary conditions: Mat, Curb, and Wall (Figure 3). Twenty-four different textures (e.g., wood, stone) were used to maximize variability in the set of images. The complete stimulus set included 72 different images (3 boundary conditions × 24 textures). Images were 740 × 590 pixel resolution (6° × 4.5° visual angle), and were presented in the scanner using an Epson PowerLite 7350 projector (type: XGA, brightness: 1600 ANSI Lumens).

Figure 3.

The three boundary conditions of Experiment 2 (Mat, Curb, and Wall).

fMRI Experimental Design.

Ten images from each of the three boundary conditions were presented in blocks of 12 s each. Each image was displayed for 1 s within a block, followed by a 200 ms blank. An 8 s fixation period followed each block. Three blocks per condition were acquired within a run (188 s, 94 TRs) and the order of these blocks was randomized within each run. Participants performed a one-back repetition detection task in which they pressed a button whenever there was an immediate repeat of an image. All participants completed 8 runs of the experiment, for a total of 24 blocks per condition.

Visual Discrimination of Boundary Types.

To establish that WS individuals were able to visually detect the boundary differences in the stimuli, a separate behavioral experiment was conducted on a different day from the scanning session using an independent set of boundary stimuli that matched the Mat, Curb, and Wall dimensions of the fMRI stimulus set, but differed in color and texture. Participants were seated before a laptop computer at a comfortable viewing distance. They were instructed to clap their hands whenever they saw the boundary change in height (e.g., whenever a Mat image was followed by a Curb, or a Curb image was followed by a Wall, etc., but not when a Curb image was followed by another Curb). The timing of presentation of images and the duration of blanks between images matched the parameters of the fMRI experiment (image duration = 1 s, time between images = 200 ms). All WS participants accurately detected the change in boundary height on 100% of the trials (as did all TD age-matched controls).

fMRI Data Acquisition and Analysis.

Data were acquired with a 3-Tesla Phillips fMRI scanner with a 32-channel phased-array head coil. Structural T1-weighted images were acquired using a magnetization-prepared rapid-acquisition gradient echo (MPRAGE) with 1 × 1 × 1 mm voxels. Functional images were acquired with a gradient echo-planar T2* sequence (2.5 × 2.5 × 2.5 mm voxels; TR 2 s; TE 30 ms; flip angle = 70°), 36 axial 2.5 mm slices (.5 mm gap), parallel to the anterior commissure-posterior commissure line.

Functional data were preprocessed using Brain Voyager QX software (Brain Innovation, Maastricht, Netherlands). Preprocessing included slice scan-time correction, linear trend removal, and 3D motion correction. For univariate and multi-voxel pattern analyses, no additional spatial or temporal smoothing was performed and data were analyzed in each individual’s ACPC space.

Regions of Interest.

ROIs were defined for each participant in individual ACPC space. The scene-selective ROIs of PPA, RSC, and OPA were defined using a functional localizer. The functional localizer run presented blocks of images of scenes, faces (half female, half male), real-world objects, and scrambled objects. Scrambled object images were created by dividing intact object images into a 16 × 16 square grid and then scrambling positions of the resulting squares based on eccentricity (Kourtzi & Kanwisher, 2001). There were 4 blocks per each image condition (scenes, faces, objects, and scrambled objects), presented for 16 s with 10 s rest periods in between blocks. Within each block, each image was presented for 600 ms followed by a 200 ms fixation screen, and there were 20 images per block. Participants performed a one-back repetition detection task during the localizer run.

For all functionally-defined scene ROIs, the entire contiguous cluster of active voxels that passed the threshold (p < .0001, cluster threshold of 4 voxels) was used. The left and right PPA were defined separately for individual participants by contrasting brain activity of scene blocks – object blocks and identifying clusters between the posterior parahippocampal gyrus and anterior lingual gyrus. This contrast also defined left and right RSC, near the posterior cingulate cortex, and left and right OPA, near the transverse occipital sulcus. In TD age-matched controls, PPA was defined in 12 individuals at the significance threshold, RSC in 12, and OPA in 11. In WS participants, PPA was defined in 12 individuals, RSC in 8, and OPA in 11.

To define primary visual area V1, a functional retinotopic localizer was used. This localizer presented vertical and horizontal visual field meridians to delineate the borders of retinotopic areas (Spiridon & Kanwisher, 2002). Previous work has successfully used fMRI-based retinotopic mapping to define V1 in individuals with WS (Olsen et al., 2009). Triangular wedges of black and white checkerboards were presented either vertically (upper or lower vertical meridians) or horizontally (left or right horizontal meridians) in 12 s blocks, alternating with 12 s blanks. During these blocks, participants fixated on a small central dot. For retinotopic analysis, the cortical surface of each subject was reconstructed from the high-resolution T1-weighted anatomical scan, acquired with a 3D MPRAGE protocol. These 3D brains were inflated using the BV surface module, and the obtained retinotopic functional maps (p < .0001, cluster threshold of 4 voxels) were superimposed on the surface-rendered cortex. The retinotopic borders of left and right V1 were defined by a contrast between vertical and horizontal meridians from the retinotopic localizer. Using these methods V1 was defined in 11 TD age-matched controls and 9 WS participants.

No differences in size (number of voxels) were found between the WS and TD groups for any of the ROIs. Despite this, we additionally performed all analyses with ROIs defined in such a way that the number of voxels was the same for all participants to ensure that differences in ROI size did not impact our findings. A maximally scene-selective voxel was localized for each hemisphere within the functionally and anatomically defined features for the particular ROI (following Park & Chun, 2009). These coordinates were used to create spherical ROIs around the maximum voxel (4 mm radius). No differences for WS participants or TD age-matched controls were found between results based upon analyses of these spherically-defined ROIs or ROIs defined as the entire cluster of voxels that passed the threshold of p < .0001.

Multi-voxel Pattern Analysis.

Patterns of activity were extracted from all the voxels within an ROI. The MRI signal intensity from each voxel within an ROI across all time points was transformed into z-scores by run so that the mean activity was set to 0 and the SD was set to 1 (Kamitani & Tong, 2005). The activity level for each block of each individual voxel was labeled with its respective condition, which spanned 12 s (6 TR), with a 4 s (2 TR) offset to account for the hemodynamic delay of the blood oxygenation level-dependent (BOLD) response. These time points were averaged to generate a pattern across voxels within an ROI for each stimulus block.

A linear support vector machine classifier (LIBSVM, http://sourceforge.net/projects/svm)^vi was trained to assign the correct condition label to the voxel activation patterns of each ROI for each participant. We employed a leave-one-out cross validation method in which one of the blocks was left out of the training sample. The data from the left-out run were then submitted for testing to the classifier, which generated predictions for the condition labels. This was repeated so that each block of the participant’s dataset played a role in training and testing (Walther et al., 2009). Three pair-wise (2-way) classifiers were used to classify the condition pairings of Mat vs. Curb, Curb vs. Wall, and Mat vs. Wall.

Results

Neural sensitivity to boundary structure was tested in two ways: we evaluated the overall average activation of all the voxels within an ROI (univariate analyses) and also analyzed the pattern of activation across voxels within an ROI (multi-voxel pattern analysis, MVPA). This was done for all ROIs (PPA, RSC, OPA, and V1).

Univariate analyses.

First we tested whether the univariate response for each boundary condition differed between WS and TD age-matched control participants. A general linear model (GLM) was computed for the time courses obtained from each ROI to extract beta values that provide an estimated effect size of the univariate response for each condition. Each condition block was separately estimated by the hemodynamic response function, and entered as predictors in the GLM. A 2 (Group: WS, TD) × 3 (Boundary condition: Mat, Curb, Wall) mixed analysis of variance (ANOVA) of the resulting beta values obtained for the PPA revealed a significant main effect of Boundary condition (F(2,44) = 81.74, p < .001, η_p² = .788) and an interaction of Group and Boundary condition (F(2,44) = 4.87, p = .024, η_p² = 24). The main effect of Group was not significant (F(1,22) = 0.14, p = .717, η_p² = .006).

Planned within-group comparisons revealed that the PPA of TD age-matched controls showed significantly different levels of activation for all three boundary conditions (replicating the findings of Ferrara & Park, 2016): activation for the Wall was greater than the Curb (two-tailed t(11) = 3.77, p = .003 (all reported t-tests are corrected for multiple comparisons using the Bonferroni correction), Cohen’s d = 0.75) and the Mat (t(11) = 5.82, p < .001, Cohen’s d = 1.68). Activation for the Curb was greater than the Mat (t(11) = 4.87, p < .001, Cohen’s d = 0.66. In contrast, activation of the PPA of WS participants did not distinguish between the Mat and the Curb (Figure 4). For the PPA of WS participants, activation levels differed between the Wall and the Curb conditions (t(11) = 9.56, p < .001, Cohen’s d = 1.43) and the Wall and the Mat conditions (t(11) = 8.96, p < .001, Cohen’s d = 2.58). However, there was no difference between the Curb and Mat conditions (t(11) = 0.76, p = .46, Cohen’s d = 0.07).

Figure 4.

Beta weights for the PPA for each of the three boundary conditions in A) WS participants and B) TD age-matched controls. Error bars represent +/− standard error of the mean. Asterisks indicate significant within-group difference between conditions, determined by t-test, p <.003. For between-group comparisons (not depicted), TD age-matched controls were found to have significantly higher activation for the Curb condition in comparison to WS participants.

Additional planned comparisons showed that for the Mat and Wall conditions, activation levels of WS participants and TD age-matched controls did not differ from one another (Mat: t(22) = −.16, p = .87, Cohen’s d = 0.09; Wall: t(22) = 0.40, p = .69, Cohen’s d = 0.17). However for the Curb condition, TD age-matched controls had significantly greater activation than WS participants (t(22) = −3.48, p = .002, Cohen’s d = 0.89). For both groups, Pearson product-moment correlation coefficients confirmed that there were no relationships between age and PPA activation for the three conditions (see Supporting Information). Collectively, these results indicate that the scene-specific response of the PPA of WS individuals is characterized by a lack of a neural distinction between boundaries that have no vertical structure (i.e., the Mat condition) compared to those that have a minimal amount of such structure (i.e., the Curb condition).

To determine whether the observed differences in PPA activation between groups might be influenced by differences in low-level visual processing, we also analyzed activation levels in primary visual area V1. An ANOVA of the beta weights obtained from the GLM of V1 revealed a significant main effect of Boundary (F(2,36) = 93.01, p < .001, η_p² = .838). The main effect of Group was not significant (F(1,18) = 0.06, p = .818, η_p² = 003), and the Boundary by Group interaction was not significant (F(2,36) = 0.31, p = .614, η_p² = .017). V1 of TD age-matched controls showed a significant difference between the Mat and Curb (t(10) = −2.98, p = .014, Cohen’s d = 0.66) and also between the Curb and Wall (t(10) = −6.42, p < .001, Cohen’s d = 1.95). V1 of WS participants also showed a significant difference between both the Mat and Curb (t(8) = −2.98, p = .016, Cohen’s d = 0.44), and the Curb and Wall (t(8) = −9.89, p < .001, Cohen’s d = 2.76). Just as had been found for the PPA, no significant correlations between age and activation were found for V1 for either of the participant groups (rs < .21, ps > .45). This is consistent with previous findings of grossly normal recruitment of primary visual cortex in WS individuals (Olsen et al., 2009) and suggests that the WS impairment in boundary representation is specific to higher-level visual areas. We performed the same analyses in the scene-selective ROIs of RSC and OPA. Neither of these areas distinguished the Curb boundary from the Mat for either TD or WS participants (see Supporting Information). Pearson product-moment correlation coefficients confirmed that there were no relationships between age and activation in any ROIs (see Supporting Information).

Whole-Brain Analyses.

We additionally conducted exploratory whole-brain analyses to determine whether areas other than the PPA might show sensitivity to the difference between the Mat and Curb conditions. Participant data were first normalized to the MNI EPI template (NeuroElf toolbox, Jochen Weber, http://neuroelf.net/). Contrasts specified a priori were: 1) Mat < Curb and 2) Curb < Wall. These within-subject contrasts were combined into group randomeffects analyses with a threshold of p < 0.001 (corrected for serial correlations, cluster threshold = 4 voxels). This threshold follows prior convention for exploratory whole-brain analyses (Epstein et al., 2007a; Johnson et al., 2007; Yi et al., 2008). Fixed-effects whole brain analyses were performed with a threshold of p < 0.0001 (corrected for serial correlations, cluster threshold = 4 voxels) in individual participants to confirm whether the regions identified in the group-level random-effects analyses replicated at the individual level. For WS participants (Figure 5), these analyses did not reveal any significant clusters for the Mat < Curb contrast (Table S1). This suggests that it is unlikely that boundary sensitivity is represented in an atypical area in the WS brain; rather it is consistent with the hypothesis that there is an overall lack of neural sensitivity to the Curb boundary among these individuals. The Curb < Wall contrast revealed significant clusters of activity in parahippocampal and retrosplenial areas, as well as early visual areas of the occipital cortex and middle occipital gyrus.

Figure 5.

Regions identified via whole-brain analysis of WS participants (random effects analysis, p < .001, cluster threshold = 4 voxels) shown on the MNIEPI template. Names of regions are marked with arrows below each figure. Regions of significance for the Curb < Wall contrast are shown (none identified for the Mat < Curb contrast for WS participants).

For TD age-matched controls (Figure 6), the Mat < Curb contrast revealed two significant clusters in the right and left lingual gryus (Table S2). The lingual gyrus is part of the functionally defined PPA, which is found in the posterior part of the parahippocampal gyrus and extends to the lingual gyrus. Thus, this is consistent with the univariate results, in which the PPA in TD age-matched controls showed significant differences in activity between the Mat and the Curb. The Curb < Wall contrast revealed significant clusters of activity in parahippocampal and retrosplenial areas, as well as the occipital gyrus and occipital cortex.

Figure 6.

Regions identified via whole-brain analysis of TD age-matched controls (random effects analysis, p < .001, cluster threshold = 4 voxels). A) Regions showing significance for the Mat < Curb contrast. B) Regions showing significance for the Curb < Wall contrast.

Multi-voxel Pattern Analysis.

Although not captured by the univariate analyses described above, it is possible that the activation pattern across all the voxels within the WS PPA contains discriminatory information about the Curb boundary. To address this, we used the more finegrained approach of MVPA, directly testing whether patterns of neural activity in scene-selective ROIs (PPA, RSC, OPA) and V1 could classify between the following condition pairs: Mat vs. Curb, Curb vs. Wall, and Mat vs. Wall. For each comparison, a linear support-vector-machine (SVM) classifier was trained and tested in a leave-one-out cross validation procedure.

Given the visual differences across the three boundary conditions, it is not surprising that classification accuracy for all three pairwise classifications was significantly above chance throughout the ventral visual cortex. This included V1 (TD: all ts >7.2, all ps <.0001; WS: all ts > 7.4, all ps <.0001), PPA (TD: all ts > 6.6, all ps < .001, WS: all ts > 2.7, all ps < .02), RSC (TD: all ts >4.1, all ps <.01; WS: all ts > 2.5, all ps <.05), and OPA (TD: all ts >4.1, all ps <.01; WS: all ts >4.1, all ps <.01; see Supporting Information, Table S3). Critically however, classification accuracy in the PPA for the Mat vs. Curb comparison was significantly lower for WS participants compared to TD age-matched controls (t(11) = 2.4, p < .04, Cohen’s d = 0.80) (Figure 7). This reduction was specific to the Mat vs. Curb classification; no differences were found between WS and TD age-matched controls for the Curb vs. Wall classification (t(11) =0.3, p = .78, Cohen’s d = 0.12) or the Mat vs. Wall classification (t(11) = 1.7, p = .10, Cohen’s d = 0.73). No group differences were found for any other ROIs, in line with the univariate results. These results further indicate that the impoverished sensitivity in the WS PPA is specific to the encoding of the minimal vertical boundary presented in the Curb condition, and not to boundary structure in general.

Figure 7.

Classification accuracy for the three pairwise boundary comparisons. Error bars represent +/− standard error of the mean. Asterisks indicate significant difference between groups, determined by t-test, p < .05.

Discussion

Our findings identify a behavioral and neural deficit in the representation of minimal amounts of vertical boundary structure among individuals with WS. This ability appears early in typical development in humans and in animal species, with no formal instruction or specific experience (Brown, Spetch & Hurd, 2007; Chiandetti & Vallortigara, 2008; Gray, Bloomfield, Ferrey, Spetch, & Sturdy, 2005; Lee & Spelke, 2008, 2011). These studies suggest that reorientation depends on mechanisms that are finely attuned to the extrapolation and representation of the geometric arrangement of the surfaces of boundaries in the environment, even if they are quite small, such as the curb used in Experiment 1. This aspect of boundary representation is known to be crucially important for reorientation in the real world and likely impacts the daily lives of individuals with WS, particularly as they are finding their way around (Atkinson et al., 2001) and attempting to remember or recall the layout of different environments.

In Experiment 1, we found that individuals with WS do not reorient geometrically using a small vertical Curb boundary (5 cm in height), although they do so when the array is bounded by a full Wall (2 m in height). Importantly, this deficit in spatial reorientation was not related to age or measures of IQ or working memory. In Experiment 2, we found a neural pattern that echoes this behavioral result: In people with WS, the PPA is not as sensitive as it is in neurotypical adults to the presence of a small vertical curb boundary within a scene. Univariate analyses revealed that the WS PPA was sensitive to the obvious differences between the Mat and Curb boundaries in comparison the Wall, but not the small difference between the Mat and the Curb. By contrast, the PPA of TD age-matched controls was sensitive to differences among all the boundary conditions. Moreover, the levels of activation for the Curb (but neither Mat nor Wall) was lower for WS participants than TD age-matched controls. Whole-brain analyses confirmed that the WS lack of sensitivity to the Curb boundary is not the result of activation in a brain area other than the a priori hypothesized and traditionally defined scene-selective ROIs.

The PPA is a high-level visual area that is known to represent aspects of scene geometry in neurotypical adults (Ferrara & Park, 2016; Kravitz, Peng, & Baker, 2011; Park et al., 2011).To our knowledge, the present work is the first to functionally define this area in WS individuals. Functional localization of the PPA at the individual level was possible for all participants with WS, and its size (number of voxels) did not in fact differ from TD age-matched controls. Despite this, through the use of stimuli that were designed to exhibit scene boundaries of different types, we identified an atypical neural response that indicates that the WS PPA does not represent minimal boundary structure (the Curb condition) in the same way as TD age-matched controls.

There are strong connections between parahippocampal areas, the hippocampus, and parietal cortex (Kravitz, Salem, Baker, & Mishkin, 2011), which are all crucial for spatial navigation (Aguirre & D’Esposito, 1999; Burgess et al., 1999; Epstein & Kanwisher, 1998; Squire et al., 2004) and are known sites of impairment in WS (Meyer-Lindenburg et al., 2005; 2006; O’Hearn et al., 2011; Sarpal et al., 2008). The atypical representation of boundaries in the WS PPA may be tied to impairments in other parts of the navigation system, for example boundary vector cells (BVCs) in the subiculum of the hippocampal formation (Lever et al., 2009; Sharp, 2006). Lever and colleagues have found that low-ridge boundaries (similar to the Curb boundary used in the present study) anchor BVC firing in rats (Lever et al., 2009). If the atypical encoding of boundaries by BVCs is then passed on as input to place cells in the hippocampus proper (Burgess et al., 2000; Hartley et al., 2000; O’Keefe & Burgess, 1996) and then to the rest of the navigation neural network, this could ultimately contribute to the WS behavioral impairment observed in the present study and in a variety of related navigation tasks (Farran et al., 2010; 2012; Ferrara & Landau, 2015; Nardini et al., 2008). Although research has yet to determine whether the specific function of BVCs is impaired in WS, genetically-altered mice who have the WS deletion show structural abnormalities in hippocampal neurons and impaired behavior in navigation tasks (Meng et al., 2002). Animals with direct hippocampal lesions also show impairments in using the geometric layout of the environment to navigate (Jones et al., 2007; McGregor, Hayward, Pearce, & Good, 2004; Tommasi, Gagliardo, Andrew, & Vallortigara, 2003; Vargas et al., 2004).

We conclude by issuing a word of caution. Our findings suggest parallel behavioral and neural deficits in the representation of boundaries among individuals with WS, but the tasks used in Experiment 1 and Experiment 2 differ from one another in many ways. Most critically, Experiment 1 relies upon representation of boundary properties for use in real-world navigation (i.e., the reorientation task). In contrast, Experiment 2 relies upon representation of boundary properties for visual detection of repeated images in the scanner (i.e., the one-back repetition detection task). Despite the differences in tasks between the two experiments, individuals with WS showed a consistent deficit in the representation of boundary information as presented in the Curb condition, suggesting that atypical boundary representation in WS individuals is likely to be independent of task requirements. Virtual reality versions of the reorientation task have been used in previous research (Sutton, Joanisse, & Newcombe, 2010; Sutton et al., 2012). These studies are likely a reasonable proxy for the real-world task, as they elicit activation in key areas known to be involved in real-world navigation (e.g., hippocampus, parahippocampal cortex, and entorhinal cortex). Future work could aim to adapt this task for the WS population to further explore the neural correlates of their behavioral deficit.

Furthermore, we are far from understanding the mechanistic and causal relationship between neural representation and behavior. On several counts, a direct mapping between neural activation and behavior would have predicted that people showing substantial neural distinction between the Mat and Curb, and less distinction between the Curb and Wall, would have been more capable of using both the Curb and Wall for reorientation. However, such transparent relationships between neural activation and behavior were not demonstrated in our study. For example, although people with WS and TD age-matched controls did not differ in terms of their neural response to the Mat condition, TD age-matched controls were able to use the geometry of this boundary to reorient, while WS participants were not.^vii Relatedly, MVPA revealed that the WS PPA did in fact show above-chance classification for the Mat vs. the Curb comparison (although significantly lower than TD age-matched controls), but our group of WS participants did not use the Curb to reorient geometrically, while TD age-matched controls did. Three WS individuals did demonstrate some sensitivity to the geometry of the Curb in behavior, but examination of their PPA activation for the Curb condition did not reveal differences from those participants who demonstrated a random pattern of behavioral search in the Curb boundary array.^viii Generally, these examples of failing to find direct relationships between neural activation and behavioral competence highlight the as yet unknown relationship between the two kinds of measures. As noted by Kanwisher and colleagues, the fact that differential multi-voxel patterns that can be read-out from a particular cortical area does not necessarily mean that that information will be associated with, or indicative of, behavioral performance (Kanwisher, 2017; Williams, Dang, & Kanwisher, 2007).

Despite our inability to completely specify the means by which the brain’s responses to different boundary types are translated into behavior, our findings suggest a surprisingly specific deficit in both the neural representation and behavioral use of certain types of boundaries. The fact that this deficit is found in individuals with a genetic syndrome raises intriguing questions about its developmental trajectory, the degree to which a similar deficit might be found in other genetic and developmental disorders, and its role in the brain’s navigation system as a whole.