Abstract
In literate adults, an area along the left posterior fusiform gyrus that is often referred to as the “visual word form area” (VWFA) responds particularly strongly to written characters compared with other visually similar stimuli. Theoretical accounts differ in whether they attribute the strong left-lateralization of the VWFA to a left-hemisphere (LH) bias toward visual features used in script, to competition of visual word form processing with that of other visual stimuli processed in the same general cortical territory (especially faces), or to the well established left-lateralization of the language system. Here we used functional magnetic resonance imaging to test the last hypothesis by investigating lateralization of the VWFA in participants (male and female) who have right-hemisphere language due to a large LH perinatal stroke. Demographically matched controls were included for comparison. All participants had intact language skills and were proficient readers; age at testing ranged from 9.75 years to early adulthood. Activation maps contrasting activation during rapid presentation of pseudowords and pictures of places revealed left-lateralized fusiform activation in controls, as expected. In participants with left-hemisphere perinatal stroke and right-lateralized language, the VWFA was instead found in the right fusiform gyrus, despite the fact that the left-hemisphere tissue normally occupied by the VWFA was intact and responded normally to pictures of places. Region-of-interest analyses confirmed right-lateralization for visual word form processing, both relative to place stimuli and relative to a resting baseline. This provides compelling evidence that the lateralization of the VWFA indeed follows that of the frontotemporal language system.
Significance Statement
The visual word form area (VWFA) in healthy adults is almost always lateralized to the left hemisphere (LH). One hypothesis is that this is due to colocalization with the LH language network. This study provides support for this hypothesis from a rare participant group with right-hemisphere (RH) language; we find that their VWFA is also right-lateralized. Our findings also support the notion that the hemispheres are equipotential early in life, with a fully functional language system—including the VWFA—able to develop in the RH if the LH is damaged. These findings contribute to our understanding of functional brain organization and plasticity, with potential implications for rehabilitation approaches for adults with reading impairments due to acquired brain injury.
Introduction
The notion that a portion of the left ventral occipitotemporal cortex (vOTC) along the fusiform gyrus responds strongly to familiar script and may constitute a “visual word form area” (VWFA) was introduced by Cohen and colleagues (Cohen et al., 2000). Since then, the nature and selectivity of this area has been debated, with several studies casting doubt on its visual nature (Yoncheva et al., 2010; Reich et al., 2011; Planton et al., 2019) and others questioning how specific its activation is to word forms versus other visual stimuli (Song et al., 2012; Neudorf et al., 2022). Theoretical accounts differ in whether they consider the VWFA to be a part of the visual object recognition system that becomes particularly tuned to visual word forms due to the acquisition of literacy (McCandliss et al., 2003; Dehaene et al., 2015; Dehaene-Lambertz et al., 2018) in a mostly bottom–up, hierarchical manner or rather as part of the speech processing system whose activation during reading is at least partly driven by its involvement in phonological processing (Dębska et al., 2016; Conant et al., 2020) and/or reflects integration of bottom–up visual information with top–down linguistic information (Price and Devlin, 2003; Twomey et al., 2011).
One key argument in favor of the latter account is the VWFA's lateralization to the left hemisphere (LH), which is the language-dominant hemisphere in most adults (Broca, 1861; Wernicke, 1874; Branch et al., 1964), as well as its connectivity with areas of the left frontotemporal language cortex (Bouhali et al., 2014; Chen et al., 2019). This connectivity precedes the acquisition of literacy (Saygin et al., 2016; Li et al., 2020) and is correlated with reading skill especially in early reading (Morken et al., 2017; Vanderauwera et al., 2018). This raises the question of what happens to the lateralization of the VWFA when the language network is not left-lateralized. Right-hemisphere (RH) language dominance happens spontaneously in a small proportion of the healthy population and is more common among left handers (Knecht, 2000). Several studies have shown that, in cases of spontaneous RH language dominance, the VWFA develops in right rather than left vOTC (Cai et al., 2010; Van der Haegen et al., 2012; Gerrits et al., 2019). However, it is possible that in these cases the lateralization of the brain is reversed more generally (e.g., due to genetic influences; Pinel et al., 2015) rather than only with respect to language dominance. If so, this finding need not imply a causal connection between the lateralization of language and the VWFA; rather, the lateralization of both could be determined by other factors.
The present investigation uses functional magnetic resonance imaging (fMRI) to probe lateralization of the VWFA in another population with atypical RH language dominance: people who had a large perinatal arterial ischemic stroke affecting the left perisylvian cortex and therefore developed RH dominance for spoken language processing. In the absence of a strong familial history of left-handedness, these individuals would presumably have been left-dominant for language if not for their early LH damage. Crucially, perinatal strokes affecting the middle cerebral artery (MCA; and therefore the perisylvian cortex) usually spare the portion of vOTC in which the VWFA develops, since the vOTC is fed by the posterior cerebral artery rather than the MCA. Thus our participants could have developed a left-lateralized VWFA despite being right-dominant for spoken language processing. They therefore are an excellent population for testing whether the VWFA develops in the same hemisphere where spoken language processing has already developed or whether differential hemispheric biases for visual object processing [e.g., a LH preference for visual stimuli with high spatial frequency (Woodhead et al., 2011) or for abstract stimuli (Dien, 2009)] lead the VWFA to consistently develop in left vOTC, regardless of language lateralization.
Materials and Methods
Study procedures followed the ethical guidelines for human subjects research laid out in the Declaration of Helsinki and were approved by the Georgetown University Medical Center Institutional Review Board. All participants provided informed consent or, in the case of minors, parental consent in conjunction with written consent (ages 12+) or assent (for children younger than 12) from the participating minor.
Participants
Data presented here are from 15 adolescents and young adults (nine males, six females) with a history of perinatal or presumed perinatal arterial ischemic stroke to the LH (LH perinatal stroke; LHPS) whose lesions encompassed at least one-third of the MCA territory (Fig. 1) and whose fMRI response to an auditory sentence comprehension task was strongly right-lateralized. Neurologically healthy siblings served as controls (n = 14, 8 males). Most of these participants were part of a previously reported study on language lateralization after perinatal stroke (Newport et al., 2022); we here use the same participant labels as in this previous publication. Relative to this previous sample, the present sample includes three additional control participants (C13, C14, C15) and two additional participants with LHPS (L16, L17) whose data were collected more recently. The present sample excludes Participant C9, who did not complete the fMRI task reported on here, Participant L1, whose lesion directly affected left vOTC, and Participant L14, who did not have right-lateralized language activation [see supplementary Figs S2 and S3 in Newport et al. (2022)]. Thus, by design, in the sample reported here, all participants with LHPS showed right-lateralized activation during auditory sentence comprehension, whereas all control participants showed the typical left-lateralized activation pattern. Language lateralization for the sentence comprehension task was quantified by computing a lateralization index across the supratentorial cortex using the LI toolbox (Wilke and Schmithorst, 2006; Wilke and Lidzba, 2007), which compares activation between the LH and RH such that values near 1 indicate left-lateralization, values near −1 indicate right-lateralization, and values near 0 indicate roughly equal activation in both hemispheres. These LIs are illustrated in Fig. 2 along with axial views of language activation maps for all participants included here.
Lesion overview. Horizontal and sagittal views of native-space MPRAGEs for all participants in the LHPS group. The horizontal slice was chosen to best show the lesion and is displayed in neurological convention (left side of the brain on the left side of the image); the sagittal view was chosen to illustrate that in all participants it spared the posterior left vOTC.
Lateralization of auditory sentence comprehension. Axial views of individual activation maps contrasting activation during auditory sentence comprehension with activation during listening to (incomprehensible) reverse speech, overlaid on an MNI template brain. In all CTRL participants (blue, top), these activations showed the expected left-lateralization, whereas all LHPS participants (orange, bottom) showed a roughly mirror-symmetric right–lateralized activation pattern. The insert on the right quantifies this using a lateralization index, with each dot representing the LI of an individual participant, horizontal bars representing the group mean, and vertical bars representing the standard error of the mean. Adapted from Newport et al. (2022), Figs S2 and S3, with inclusion of new Participants C13, C14, C15, L16, and L17 (who were run more recently) and removal of Participant C9 (who did not complete the fMRI task reported on here), Participant L1 (whose lesion directly affected left vOTC), and Participant L14 (who did not have right-lateralized language activation).
All participants were native speakers of English and fluent readers. As reported previously for a mostly overlapping sample (Newport et al., 2022), the CTRL and LHPS groups performed comparably well on various assessments of language skills, including tasks in which both groups' performance was far from ceiling (see Fig. 3 for details). The groups also did not differ significantly in age (CTRL, mean = 16.20 years; SD = 4.93 years; range = 9.75–29.50 years; LHPS, mean = 19.42 years; SD = 4.69 years; range = 10.00–26.67 years) or in gender distribution.
Language skills. All participants completed the sentence comprehension and word structure subtests of the Clinical Evaluation of Language Fundamentals (CELF; Wiig et al., 2013), and all participants except for one CTRL participant completed the Active–Passive Test [developed based on Dennis and Kohn (1975)] and the Test for Reception of Grammar (TROG-2; Bishop, 2003). As is evident, performance levels for the two groups were quite similar across all measures and did not reveal the large impact on language skills that one would expect following a large stroke to the LH if it occurred in adulthood. The only measure that showed a significant group difference in a Student’s t test was the CELF Sentence Comprehension subtest (CTRL mean, 99.7%; LHPS mean, 97.7%; t(27) = 2.89; p = 0.007); however, due to the large number of participants performing at ceiling in this test, this statistical comparison should be interpreted with caution (Liu and Wang, 2021). Importantly, there were no significant group differences on measures without ceiling effects, such as the TROG-2 center-embedded items (CTRL mean, 71.1%; LHPS mean, 71.7%; t(26) = −0.05; p = 0.960) or the passive items from the Active–Passive Test (CTRL mean, 92.8%; LHPS mean, 88.5%; t(26) = −1.33; p = 0.196). Taken together, these results indicate that the participants in the LHPS group had language skills within the normal range despite their large LH lesions. Adapted from Newport et al. (2022), Figure 3, with inclusion of new Participants C13, C14, C15, L16, and L17 and removal of Participants C9, L1, and L14.
MRI setup and parameters
Participants underwent fMRI at Georgetown University's research-dedicated neuroimaging facility on a 3 Tesla Siemens MRI scanner. This scanner was upgraded from a Trio to a Prisma model toward the end of the data collection period. Five participants in the LHPS group and two in the CTRL group were scanned after the upgrade using a 20-channel head coil; all other participants were scanned before the upgrade using a 12-channel head coil. Scanning parameters were held as constant as possible across pre- and postupgrade scans. Several neurologically healthy participants were scanned before and after the scanner upgrade to determine whether the upgrade altered the activation maps. No systematic differences were found between pre- and postupgrade results in this control study.
Functional imaging data were acquired using an echoplanar T2*-weighted sequence that covered the entire brain in 50 horizontal slices, with a 64 × 64 matrix and an effective voxel size of 3 × 3 × 3 mm3 (in-plane resolution 3 × 3; slice thickness 2.8 mm; distance factor 7%; slice acquisition in a descending order). Repetition time (TR) was 3 s, echo time (TE) was 30 ms, and the flip angle was set to 90°. Each participant contributed two functional runs of 80 volumes each.
For assessment of lesions and as a reference to coregister the functional data with, a high-resolution T1–weighted MPRAGE scan was also obtained from each participant. It covered the brain in 176 sagittal slices with an effective voxel size of 1 × 1 × 1 mm3 (256 × 256 matrix), a TR of 2,530 ms, TE of 3.5 ms, inversion time of 1,100 ms, and flip angle of 7°.
In-scanner tasks
Auditory sentence comprehension task
Lateralization of the frontotemporal language network was determined using an auditory sentence comprehension task first introduced by Berl and colleagues (Berl et al., 2014) and administered to the participants of this study as described previously (Newport et al., 2022). Briefly, this task contrasts blocks of forward speech, during which participants listen to short English sentences, with blocks of reverse speech, during which the same sentences are played in reverse and thus rendered incomprehensible while retaining their basic auditory characteristics. The resulting activation maps reliably identify the inferior frontal and superior temporal brain areas commonly referred to as the frontotemporal language network, which in our neurologically healthy control participants is in the LH and in our participants with large perinatal LH strokes is in the RH homotopic regions, including the RH regions mirroring Broca's and Wernicke's areas (Newport et al., 2022).
Word form/place/face localizer task
Lateralization of the VWFA was determined using a visual localizer paradigm that contrasts activation evoked by rapid sequential presentation of word forms (W), places (P), and faces (F). Each participant contributed two runs of the localizer task. Each run was 4 min long and consisted of three 18 s blocks of each of the three experimental conditions during which participants performed a one-back repetition detection task, interleaved with 18 s rest (R) periods during which participants rested their eyes on a black fixation cross at the screen center. Runs were repeated if motion (as estimated by the scanner in real time) exceeded 3 mm in either direction or if participants failed to detect several repetitions in a row (indicating that they were no longer paying attention to the stimuli).
During word form blocks, participants viewed pseudowords generated by MCWord (Medler and Binder, 2005). Pseudowords varied in length from five to seven letters and were generated based on constrained trigram-based strings. We chose pseudowords rather than English words to prevent evoking semantic processing and visual imagery and because pseudowords have been shown to lead to stronger VWFA activation than words (Kronbichler et al., 2004; Ludersdorfer et al., 2013). To reduce differences in low-level visual stimulation between the word form condition and the face and place conditions, the pseudowords were presented on a background of phase-scrambled versions of the place stimuli [see Sadr and Sinha (2004) for the use of phase-scrambling as a visual control] and located at the same height as the eyes in the face condition. During place blocks, participants viewed grayscale images of scenes containing one or multiple buildings seen from the outside. During face blocks, participants viewed grayscale images of young adult faces (50% male, 50% female) smiling frontally at the camera. Each image was a close-up of the face and cut into a square shape such that little was visible of hair, clothing, and ears. None of the pictures included glasses or jewelry. Face and place stimuli were subsets of those used by Marois et al. (2004) that excluded highly recognizable place stimuli (such as the Twin Towers). All stimuli were squares subtending roughly 5° of the visual angle as viewed via a slanted mirror inside the scanner bore and presented at the screen center in front of a gray (RGB 128, 128, 128) background. Example stimuli are shown in Fig. 4. Stimuli were drawn randomly without replacement from sets of 144 for each experimental condition; no stimulus was shown twice except as a one-back repetition, to which participants responded with a button press. Which stimuli repeated was determined randomly, with the constraint that repetitions occurred at least 4 and at most 12 stimuli apart, for a total of two or three repetitions in an 18 s block.
Experimental design. A, Example stimuli illustrating the three experimental conditions, arranged as excerpts of an experimental block. The first two stimuli of “places” illustrate a repetition. B, Time course of one of the two functional runs, with gray, blue, green, and orange indicating rest, face, place, and word form stimulation, respectively. The other run presented the experimental conditions in the remaining three possible orders: PFW, FWP, and WPF. Rest periods occurred at the same times in both runs.
Each 18 s block contained 24 stimulus presentations lasting 500 ms each, followed by a 250 ms interstimulus interval, during which the fixation cross was shown. For two LHPS participants, we reduced the rate of stimulus presentation to half, i.e., to 12 per block, showing each stimulus for 1 s followed by a 500 ms interstimulus interval, because practice outside the scanner revealed that the fast rate of stimulus presentation generated more stress than would have been conducive to a successful scanning session. These participants' activation maps were not systematically different from those of the other participants, and removing them from the analysis did not change the inferences derived from the statistical tests. We thus decided to include them.
For this investigation of VWFA lateralization, we contrasted activation during the word form condition with activation during the place condition to control for visual stimulation because unlike the visual processing of faces (Dehaene et al., 2010; Behrmann and Plaut, 2020), processing of places has not been hypothesized to be in direct competition for vOTC territory with that of word forms (Rosenke et al., 2021). (Activation in response to face stimuli will be the focus of a separate publication.)
Data analysis
Software
Imaging data were preprocessed and modeled at the whole-brain level using SPM12 (v. 7771), and statistical comparisons of data extracted from the resulting activation maps were performed in MATLAB (v. R2023b) and Excel (v. 16.16.27). Activation maps were visualized using Mango (v. 4.1).
Preprocessing
Preprocessing steps included dicom-to-nifti conversion, realignment of each functional run's individual images to the run's mean functional image to reduce motion-related displacements between volumes, and coregistration of functional to anatomical data in native space, all using SPM's default parameters. The native-space MPRAGE was then warped to SPM's built-in MNI template using the unified segmentation and normalization approach with standard parameters except for rescaling regularization by a factor of 0.1 (as recommended by John Ashburner, one of the developers of this approach). The resulting warps were inspected for alignment with the MNI template and distortions relative to the native-space brain. Where necessary due to poor warping results, the procedure was repeated using a manually drawn lesion mask and enantiomorphic lesion healing (Nachev et al., 2008), as well as additional masks to prevent SPM from mistaking expanded subarachnoid spaces for gray or white matter. Once acceptable warps were achieved for the MPRAGE, the same warp field was used to bring the coregistered functional data into the MNI space. Lastly, the functional data were smoothed with a 6 mm full-width at half-maximum Gaussian kernel.
Modeling of individual whole-brain activations
Voxel time courses were modeled at the individual subject level using a general linear model that included, for each of the two functional runs, three predictors of interest modeling the presence of the three stimulation conditions (faces, places, and word forms) convolved with SPM's default hemodynamic response function; six predictors to capture signal changes associated with translational movement along and rotational movement around the x, y, and z axis, respectively; and a constant to capture global differences between the two runs. A high-pass filter with a 400 s cutoff was applied to remove linear trends. After modeling, we contrasted the beta map for the word form predictor and the beta map for the places predictor (W > P) to identify voxels with a significant preference for word forms over pictures of places. We also looked at activation for the word form and place stimuli relative to the resting baseline condition without visual stimulation.
Regions of interest
Our analysis focuses on activation in the vOTC, which we defined liberally as including Brodmann areas 37, 36, and 20 using the Wake Forest Pickatlas included with SPM (dilation, 6). In addition to inspecting this large anatomical region for clusters with W > P preference, we also performed an analysis of activation changes averaged across an independently defined functional region of interest (ROI). To achieve an ROI that would be as spatially specific as possible to the putative location of the VWFA while also accounting for the well known interindividual variability in the exact location of the VWFA and for the fact that its location can vary depending on which contrast is used to identify it, we based this ROI on a meta-analysis. We downloaded the meta-analytic “association test” map for the term “word form” from the Neurosynth website (https://neurosynth.org/analyses/terms/word%20form/), increased the threshold (to 11) until only voxels in left vOTC remained, and then binarized the map. The resulting ROI had a volume of 3,712 mm3 and almost exclusively included voxels labeled as “Fusiform Gyrus, Brodmann Area 37” on the MNI Atlas included with Mango. This ROI thus contains only voxels for which (nearby) activation is reported more often in neuroimaging studies frequently mentioning the term “word form” than in neuroimaging studies that do not mention this term (Yarkoni et al., 2011). A corresponding RH ROI was generated by inverting the sign on the x-coordinates, mirroring the LH ROI across the midline.
Statistical analyses
Statistical analyses were conducted using Student’s T tests in Excel. We used one-tailed tests for comparing the LH and RH ROI response to the W > P contrast, which we expected to be left-lateralized in the CTRL group and right-lateralized in the LHPS group. Two-tailed tests were used for all other comparisons. Results are reported in terms of uncorrected p values, along with associated t values and their degrees of freedom.
Results
Behavioral performance in the scanner
Accuracy and reaction time data from the in-scanner task were available for all participants except one participant in the LHPS group, during whose session the response buttons did not work. We scored accuracy for our one-back task as the percentage of correctly detected repetitions and computed the average reaction time for all correct button pushes. There were no significant differences between the groups in accuracy or reaction time in any of the conditions, with detection performance ∼80% and reaction times ∼480 ms (Table 1).
Accuracy and reaction time on the in-scanner one–back tasks
Inspection of individual whole-brain activation maps for word-selective clusters in vOTC
Individual whole-brain activation maps for the W > P contrast at a single-voxel threshold of p < 0.001 revealed clusters of voxels that showed a significantly stronger response to word form than to place stimuli in the left vOTC for 13 of the 14 participants in the CTRL group and in the right vOTC for 9 of the 15 participants in the LHPS group. Three of the CTRL participants also had a significant cluster in the right vOTC, and two of the LHPS participants had a significant cluster in the left vOTC in addition to a larger one in right vOTC. Activation maps for the remaining participants (one CTRL, six LHPS) did not reveal significant vOTC activation for the W > P contrast on either side at this relatively stringent threshold. With a more lenient threshold of p < 0.01, LH activation was observed in 13 of 14 participants in the CTRL group and RH activation in 11 of 15 participants in the LHPS group. While the exact location of these putative VWFAs varied across participants, the RH activation clusters of the LHPS participants roughly mirrored the LH activation clusters of the CTRL participants. Example activation maps are shown in Fig. 5.
Examples of individual activation maps. A, In the CTRL group, clusters of voxels with a significant preference for visual word forms over pictures of places (contrast W > P) were identified in the left fusiform gyrus (BA 37), at locations consistent with published coordinates for the VWFA. B, In the LHPS group, clusters with this preference were located in the right fusiform gyrus, roughly mirroring the usual location. Unthresholded activation maps for all participants, an overview figure showing slice views of all participants, and a table summarizing peak coordinates and sizes of the VWFA activation clusters can be found at osf.io/8x49a.
Statistical evaluation of average W > P effects in an independently defined VWFA ROI
To quantitatively test whether lateralization of visual word form processing was indeed reversed in the LHPS group and to overcome the problem that conventionally thresholded activation maps did not reveal significant vOTC activation in every participant, we averaged, for each participant, the t values for the W > P contrast across a left fusiform VWFA ROI and a RH mirror version of that ROI (Fig. 6A). These ROIs were defined independently of the current dataset using the meta-analytic results of other studies as aggregated using Neurosynth (see Materials and Methods, Regions of interest). At the individual level, 13 of 14 CTRL participants showed stronger activation in the left than in the right ROI (Fig. 6A, negative sloping blue lines), and 11 of 14 LHPS participants showed stronger activation in the right than in the left ROI (Fig. 6A, positive sloping orange lines). There were no reversals of this pattern in either group; the participants (one CTRL, three LHPS) whose lateralization did not follow this pattern had similar activation in both hemispheres, likely because the ROIs were much larger than any participant’s individual VWFA and thus contained a large number of voxels without a preference for visual word forms in either hemisphere. At the group level, the CTRL group showed the expected left-lateralization (CTRL LH > RH; t(13) = 6.03; p < 0.0001 one-tailed) and the LHPS group showed right-lateralization (LHPS RH > LH; t(14) = 3.35; p = 0.002 one-tailed). This reversal of VWFA lateralization in the LHPS group was not solely driven by a decrease in LH activation, as one might expect due to the proximity of a lesion (CTRL_LH vs LHPS_LH; t(27) = 2.54; p = 0.017 two-tailed), but also by an increase in RH activation in the LHPS group relative to the CTRL group (LHPS_RH vs CTRL_RH; t(27) = 2.45; p = 0.021 two-tailed). There were no significant group differences between activations in the language-dominant hemispheres (CTRL_LH vs LHPS_RH; t(27) = 0.72; p = 0.480 two-tailed) or in the nondominant hemispheres (LHPS_LH vs CTRL_RH; t(27) = 0.81; p = 0.423 two-tailed.) Taken together, these findings confirm a reversal of the typical lateralization pattern in the LHPS group.
ROI analysis. Each data point represents a participant’s activation averaged across all voxels inside independently defined VWFA ROIs in the left and right hemisphere (shown in red on the brain inset); short horizontal lines represent the group mean; vertical lines represent the standard error of the mean. A, Contrasting the response to word form stimuli with that to place stimuli (contrast W > P). The slope of the lines connecting each participant’s LH and RH activation shows the lateralization of ROI activation. As expected, most CTRL participants showed a stronger W > P effect in the left compared with the right fusiform ROI (blue, left panel). The opposite was true for the LHPS participants (orange, right panel.) The LHPS group's activation in the RH ROI was significantly stronger than that of the CTRL group and not significantly different from the CTRL group’s activation in the LH ROI. B, Relative to the baseline (fixation only), activation of the left ROI in response to place stimuli was no weaker in the LHPS group than in the CTRL group, showing that right-lateralization of word form processing in the LHPS group was not driven by overall depressed activation in the lesioned hemisphere. C, Activation in the RH ROI in response to word form stimuli was stronger in the LHPS group than in the CTRL group even relative to the baseline (i.e., without controlling for the effect of visual stimulation).
Statistical evaluation of average ROI effects for word forms and places relative to rest
To further reduce concerns about whether the lesioned LH was not able to activate normally in the LHPS group, we also looked at LH ROI activation relative to the resting baseline, reasoning that if activation evoked by visual stimulation in general were suppressed due to the ipsilateral lesion, we should see weaker place activation in the LHPS group relative to the CTRL group. As can be seen in Figure 6B, this was not the case. At the individual level, 12 of the 14 CTRL participants and 13 of the 15 LHPS participants showed a positive response to place stimuli. Across participants, average LH ROI activation to place stimuli was comparable in both groups (CTRL_LH vs LHPS_LH for P > R; t(27) = −0.42; p = 0.679 two-tailed.)
We did not necessarily expect a significant group difference in the RH ROI response to word form stimuli relative to the baseline (fixation) because both groups should show strong activation for any visual stimulation, and this large general visual effect might overpower a small group difference driven by right-lateralization of visual word form processing in the LHPS group. Nonetheless, a group difference in RH word form activation was clearly evident and nearly reached significance even relative to the baseline (LHPS_RH vs CTRL_RH for W > R; t(27) = 2.00; p = 0.056 two-tailed; Fig. 6C).
Statistical comparison of the strongest individual word form responses between groups and hemispheres
Because the Neurosynth ROI is bigger than we would expect any individual VWFA to be, averaging responses across the entire ROI has the disadvantage of including voxels that are not responsive to the stimulus of interest at the individual level. Thus, our final analysis was a top-voxel analysis focused on the eight voxels with the strongest word form response inside the Neurosynth “Word Form” ROI and its RH mirror version. The constraint to the Neurosynth ROI excludes voxels from early visual areas whose strong responses are predominantly driven by visual stimulation in general and not specific to word forms.
As can be seen in Figure 7, the results of this W > Rest top-voxel analysis support the same conclusions as the ROI-wide W > P analysis: activation was left-lateralized in the CTRL group (CTRL LH > RH; t(13) = 3.90; p < 0.001 one-tailed) and right-lateralized in the LHPS group (LHPS RH > LH; t(14) = 4.67; p < 0.001 one-tailed). Compared with the CTRL group, the LHPS group had significantly weaker activation on the left side (CTRL_LH vs LHPS_LH; t(27) = 2.30; p = 0.030 two-tailed) and significantly stronger activation on the right side (LHPS_RH vs CTRL_RH; t(27) = 3.02; p = 0.005 two-tailed). Also as in the previous analyses, there were no significant group differences regarding activation in each group’s language-dominant hemisphere (CTRL_LH vs LHPS_RH; t(27) = −0.95; p = 0.352 two-tailed) or in each group’s nondominant hemisphere (CTRL_RH vs LHPS_LH; t(27) = 0.30; p = 0.764 two-tailed).
Comparison of individual activation peaks between groups and hemispheres. Each data point represents the average t value of a participants eight voxels with the strongest activation for word forms relative to rest (contrast W > Rest) inside an independently defined VWFA ROI (shown in red on the brain inset). As in Figure 6, short horizontal lines represent the group mean of these individual top activations; vertical lines represent the standard error of the mean. The pattern of results is identical to that shown in Figure 6A for the ROI-wide analysis of the W > P contrast, with reversed lateralization in the LHPS group relative to the CTRL group.
In summary, we have converging evidence across whole-brain, independent ROI, and top-voxel analyses investigating activation to visual word forms relative to a visual control stimulus (places) and relative to the baseline that in our participants with RH language due to perinatal LH stroke, the VWFA is right-lateralized.
Discussion
This fMRI study investigated the lateralization of visual word form processing in a sample of 15 adolescents and young adults with RH dominance for spoken language processing due to a large perinatal arterial ischemic stroke to the left perisylvian cortex. Their blood oxygenation level-dependent response to rapid visual presentation of pseudowords (relative to rapid visual presentation of pictures of places and relative to a fixation-only baseline) was compared with that of neurologically healthy controls (n = 14), all of whom showed the typical LH-dominant response to spoken language. The control group showed the expected left-lateralized response for visual word form processing in a region of vOTC that has been dubbed the VWFA. In contrast, the participants with RH language dominance subsequent to LHPS showed a stronger response (both compared with their own left vOTC and compared with the control group) in the corresponding region of right vOTC. This difference was not driven by a general depression of left vOTC activation in response to visual stimuli but rather by an atypically strong visual word form response in right vOTC. The strong visual word form response in the RH was similar to the atypically strong RH response to comprehension of spoken sentences already reported for this participant group (Newport et al., 2022).
This finding makes an important contribution to the debate on whether lateralization of the VWFA is influenced by lateralization of the language system. Right-lateralization of the VWFA has already been reported for neurologically healthy adults with atypical language dominance (Cai et al., 2010; Van der Haegen et al., 2012; Gerrits et al., 2019) as well as for cases of direct early damage to the usual (left midfusiform) location of the VWFA (Cohen et al., 2004; Liu et al., 2019). However, the participants at the center of our study are unlike these other two groups in that their atypical RH language dominance was not genetically driven (and thus not potentially part of a more general reversal of cerebral lateralization) but rather was secondary to their perinatal stroke, with no direct damage to the canonical location of the VWFA in left vOTC or to the posterior occipitotemporal pathways connecting this location with lower stations along the ventral visual stream. If the VWFA were simply a station at the higher end of the visual object processing hierarchy and independent of the language system, then our participants' VWFA should have developed in left vOTC as it does in the vast majority of neurologically healthy adults. The fact that it instead developed in the RH is in line with the hypothesis that the VWFA typically colateralizes with the language system (Dehaene et al., 2015; Behrmann and Plaut, 2020).
One limitation of the present study is that we did not formally assess reading skill. However, we have the educational levels for each participant and know that participants were either at the expected grade level in school or had already graduated from high school. We also have parent reports of grade-level–appropriate reading performance and informal observations during consenting and testing to confirm that all participants were fluent and competent readers. Some studies have shown correlations between task performance and language lateralization (Donnelly et al., 2011; Bartha-Doering et al., 2018). Thus, if participants in the LHPS group had greater difficulty with reading the pseudowords than participants in the CTRL group, this would pose a potential confound and could be an alternative explanation for the stronger RH activation in the LHPS group. However, there were no differences between the two groups in accuracy or reaction times (even when accuracy was not at ceiling) on the in-scanner task, nor any differences regarding language skills as assessed outside the scanner. We therefore feel confident that we can rule out this alternative explanation and instead interpret the findings as a reversal in lateralization of the VWFA in association with the lateralization of the rest of the language network.
Another limitation is that our study did not include participants who remained LH dominant for language after a large perinatal stroke to the LH. Such cases are rare (or perhaps nonexistent) because large perinatal strokes almost always affect the frontotemporal language network. One future direction for our research is to recruit participants with small strokes to determine which lesion locations are compatible with developing typical LH language dominance and to ask whether the VWFA consistently develops in the left vOTC in those cases. There is evidence of left-lateralization for the VWFA in a group of patients with resections of parts of the posterior vOTC due to epilepsy (Lopes et al., 2015). These resections spared the midfusiform location of the VWFA, as well as its connections to portions of the temporal lobe that are involved in language processing, but damaged connections to the occipital lobe. Taken together, these studies suggest that lateralization of the VWFA is linked closely to lateralization of the frontotemporal language system.
This conclusion is in agreement with other studies using nonlesion approaches. For example, Moore and colleagues demonstrated in an fMRI study that an area in left vOTC whose coordinates correspond well with those published for the VWFA showed an increased response to face stimuli in a group of participants who had been trained to associate different faces with different letter sounds relative to control groups who had not learned this “face alphabet” (Moore et al., 2014). As the authors argue, this provides strong support that the VWFA serves as an interaction point for spoken language processing and processing of complex visual stimuli (even visual stimuli that, in the absence of learned association with speech sounds, do not drive this area strongly) and thus colateralizes with spoken language processing for maximum efficiency (Dehaene et al., 2015). The emergence of a preference for visual word form processing (as opposed to face processing) in the language-dominant hemisphere has also been demonstrated in computational simulations using an artificial neural network (Plaut and Behrmann, 2011). Furthermore, top–down effective connectivity from the left inferior frontal cortex was found in neurologically healthy literate adults for a functionally defined VWFA in left vOTC, but not for the corresponding right VWFA or for left vOTC voxels with a significant preference for faces rather than word forms (Canário et al., 2020), demonstrating that connectivity between the VWFA and frontal and temporal language areas is lateralized to the LH (Bouhali et al., 2014; Chen et al., 2019).
In summary, our study demonstrates that major perinatal strokes to the LH that result in RH dominance for spoken language processing also typically lead to development of the VWFA in the RH, even if the lesion does not damage the LH tissue in which the VWFA normally develops or its connections to earlier stations in the visual object processing hierarchy. This accords with the hypothesis that development and lateralization of the VWFA are determined by connectivity with the language system. It also supports the notion that the two hemispheres are equipotential early in life such that either of them can develop a functional language system, including the ability to respond to print.
Footnotes
This research was supported by National Institutes of Health Grants R01DC016902 to E.L.N and W.D.G and P50HD105328 to the DC-IDDRC and Children’s National Hospital and Georgetown University, the Solomon James Rodan Pediatric Stroke Research Fund, the Feldstein Veron Innovation Fund, and the George Bergeron Endowment to the Georgetown-MedStar Center for Brain Plasticity and Recovery. We thank Soojin Park for sharing the face and place stimuli, Ashley Carroll and Kasey Stack for their assistance with data collection and curation, and our participants and their families for their time and dedication to this research.
The authors declare no competing financial interests.
- Correspondence should be addressed to Anna Seydell-Greenwald at as2266{at}georgetown.edu.
