Nature versus Nurture in Ventral Visual Cortex: A Functional Magnetic Resonance Imaging Study of Twins

Participants.

We tested 13 pairs of right-handed MZ twins (nine female pairs, four male pairs, ages 18–29 with a mean age of 21.3) and 11 pairs of DZ twins (seven female pairs, four male pairs, ages 18–23, mean age 19.9). Zygosity was determined by comparing seven to eight highly variable DNA markers (D5S818, D13S317, D7S820, D16S539, vWA, TH01, TPOX, CSF1PO) from the buccal cells of twins collected by swabbing the cheek of each participant. DNA was amplified using the polymerase chain reaction technique. Twins in whom all the markers matched were classified as monozygotic and twins in whom some markers mismatched were classified as dizygotic.

Experimental design.

During functional MRI scanning, participants performed a one-back matching task (press a button with the right, middle finger if the current stimulus matches the last stimulus, press a button with the right, index finger if they do not match) on gray-scale pictures from five categories: faces, places (pictures of houses), pseudowords (pronounceable nonword letter strings), chairs, and phase-scrambled control images (see Fig. 1). Pseudowords, instead of words, were used to ensure that there were no effects of word semantics on activation patterns (and pseudowords and words are known to elicit similar activation patterns in ventral visual cortex) (Polk and Farah, 2002). In keeping with many previous studies, houses were used to activate the so-called parahippocampal place area (Epstein and Kanwisher, 1998). The control images were versions of the experimental images in which the phase information was scrambled, but spatial frequency was preserved (i.e., the power spectra were identical). There were three runs with 15 20 s blocks per run (three blocks of each of the five categories of stimuli in pseudorandom order) with a 20 s rest period preceding each run. Each block consisted of 10 items from the same category presented for 1500 ms each, followed by a 500 ms intertrial interval.

fMRI data acquisition.

Structural images were T1-weighted images collected in axial slices parallel to the anterior commissure–posterior commissure line with a resolution of 0.9375 × 0.9375 × 5.0 mm. Neural activity was estimated based on the blood oxygen level-dependent (BOLD) signal measured by a GE 3T scanner using a spiral acquisition sequence (2000 ms repetition time, 30 5 mm thick axial slices with an in-plane resolution of 3.75 × 3.75 mm, 24 cm field of view, 30 ms echo time, 90 degree flip angle).

Preprocessing.

The functional images (T2*-weighted echo planar images) for each participant underwent reconstruction, slice timing correction, and realignment as part of preprocessing. These images were then spatially smoothed with a Gaussian kernel of 8 × 8 × 8 mm and normalized into a standard MNI (the Montreal Neurological Institute) space and resampled into a resolution of 3 × 3 × 3 mm per voxel. The anatomical images (T1-weighted images) for each participant were also normalized and resampled into a resolution of 3 × 3 × 3 mm. Normalized images were resampled for computational efficiency for the calculation of effective sample size (see below), and it was verified that the similarity results were the same whether images were resampled or not.

ROI.

An anatomical mask was first constructed that included the parahippocampal, fusiform, inferior temporal, and inferior occipital gyri bilaterally using the PickAtlas software toolbox (Maldjian et al., 2003). Then, the functional data from the first run (Run1) were analyzed on a voxel-by-voxel basis using a general linear model corrected for temporal autocorrelation with regressors corresponding to each of the four experimental conditions (using the VoxBo software package, http://www.voxbo.org). A region of interest was then defined as the union of voxels within this mask that were significantly active for any of the four major contrasts (faces, places, pseudowords, or chairs vs phase-scrambled control) in a random-effects group analysis that included all participants (p < 10⁻⁶, uncorrected). The region of interest consisted of 1237 voxels.

Subregions of the region of interest (ROI) were also separately defined to test whether anatomical similarity was uniform across different functional subregions of ventral visual cortex. Voxels within the anatomical ROI that were significantly active for each of the four stimulus categories versus control in the group analysis (p < 10⁻⁶, uncorrected) were defined as the four subregions. For example, the face subregion was defined as the voxels significantly active for the contrast of faces versus phase-scrambled control, and it included 414 voxels. Likewise, the house subregion included 1139 voxels, the pseudoword subregion included 67 voxels, and the chair subregion included 764 voxels.

Functional maps of activation.

The functional data from the second and third run (Run2 and Run3) were analyzed on a voxel-by-voxel basis using a general linear model corrected for temporal autocorrelation with regressors corresponding to each of the four experimental conditions. Then, functional statistical parametric maps of t-values and β-values for each of the four major contrasts were computed. Thus, the data used to generate the functional maps was independent of the data used to generate the ROI. An initial examination of the functional maps of all participants indicated that one participant from the MZ twin group did not show any activation within the anatomical mask for the face and house contrasts and another participant from the same group did not show any activation for the house contrast, even when a liberal activation threshold (p < 0.01, uncorrected) was adopted. These participants and their siblings were therefore excluded from further analysis.

Similarity analysis.

The similarity of each twin pair for all four contrasts was assessed by computing a Pearson correlation coefficient directly on the t values (and also, β weights) across the entire region of interest (see Fig. 1). To conduct inferential tests on the functional similarities of MZ versus DZ twins, the correlation coefficient r for each pair was transformed into a z-score using Fisher's r-to-Z transformation. In addition, because the maps are spatially autocorrelated, the effective sample size is smaller than the total number of voxels in the region of interest. We therefore computed an effective sample size for a given pair of maps using the technique described by Clifford et al. (1989). A Z-statistic corresponding to the contrast between z-scores for MZ and DZ twins was then computed as follows (Cohen and Cohen, 2003): where λ_i is the contrast weight (e.g., 1 for MZ twins, −1 for DZ twins), z _i is the z-score corresponding to the similarity of the functional maps from twin pair i, and M̂ is the effective sample size for twin pair i. Statistical significance for the difference between MZ and DZ pairs' similarity measures was calculated based on a one-tailed test of the resulting Z-statistic unless otherwise noted.

The similarity of the two T1-weighted anatomical maps from each twin pair was assessed using the same method as the similarity analysis of the functional maps. A Pearson correlation coefficient between the two anatomical maps was computed to estimate the similarity, and the inferential test was based on the Z-statistics. A second analysis of anatomical similarity was conducted using T2*-weighted echo-planar scout images, rather than T1 images, in case the T2*-weighted images captured other aspects of anatomical structure not captured by the T1-weighted images. The similarity of the two T2*-weighted anatomical maps from each twin pair was assessed using the same method described previously.

Within-subject analysis.

Within-subject correlations were computed to estimate the reliability of the experiment and to estimate the potential maximum correlation values within twin pairs. For each participant, a functional activation map for each contrast was computed from the second run (Run2) and another functional map for each contrast was computed from the third run (Run3). A correlation coefficient was then computed as described previously.

Unrelated pairs.

Each participant was randomly paired with another unrelated participant to create a hypothetical group of twenty-two unrelated pairs. The functional and anatomical similarities of these unrelated pairs were computed using the same method used to compute the similarities of the related pairs.