Does the visual word form area split in bilingual readers? A millimeter-scale 7-T fMRI study

In expert readers, a brain region known as the visual word form area (VWFA) is highly sensitive to written words, exhibiting a posterior-to-anterior gradient of increasing sensitivity to orthographic stimuli whose statistics match those of real words. Using high-resolution 7-tesla functional magnetic resonance imaging (fMRI), we ask whether, in bilingual readers, distinct cortical patches specialize for different languages. In 21 English-French bilinguals, unsmoothed 1.2-millimeters fMRI revealed that the VWFA is actually composed of several small cortical patches highly selective for reading, with a posterior-to-anterior word-similarity gradient, but with near-complete overlap between the two languages. In 10 English-Chinese bilinguals, however, while most word-specific patches exhibited similar reading specificity and word-similarity gradients for reading in Chinese and English, additional patches responded specifically to Chinese writing and, unexpectedly, to faces. Our results show that the acquisition of multiple writing systems can indeed tune the visual cortex differently in bilinguals, sometimes leading to the emergence of cortical patches specialized for a single language.


Supplementary Text
In the main text for the English-French participants, we analyzed the activity in word-specific clusters irrespective of language (hereafter referred to as "bilingual clusters"), defined by greater activity for the average of two languages (English and French) relative to other categories (faces, bodies, houses, tools) in the localizer runs (p<0.001 uncorrected, cluster size > 4). This was motivated by the finding that there were no consistent language-specific clusters when directly contrasting English and French words, neither in the localizer nor in the main fMRI runs, and that most of the language-specific voxels were included by the bilingual voxels (whole brain, English: 91.90%, French: 86.14%), especially in the VOTC (English: 95.75%, French: 94.82%).
However, it may seem strange to pull together the two languages, and only then check whether they prefer one language over the other, and may miss language-specific clusters that didn't survive thresholding under direct comparison.
In this supplementary analysis, we show that this choice of examining word-specific clusters without language specificity (bilingual clusters) in the English-French participants was inconsequential, because it did not lead to missing voxels that would be putatively strongly selective to one language. Here we performed two separate contrasts with each single language (English words > faces, bodies, houses, tools; French words > faces, bodies, houses, tools, p<0.001 uncorrected, cluster size > 4), separated all voxels into four categories, according to whether they overlapped with the bilingual clusters, and examined the averaged activity within each voxel category. If a language-specificity was present, it would be consistently observed in the language-specific conditions for word components (LE vs. LF, BE vs. BF, QE vs. QF) and for the real words (WE vs. WF). We performed this analysis in a VOTC mask ( Figure S3A) and also in the remaining of the voxels outside the VOTC mask ( Figure S3B). To minimize the possibility that potential language-dominance differences between participants may obscure the language-specificity activity differences, we performed the analysis separately for the three subgroups. As a comparison, we also performed the same analysis for the English-Chinese participants.
For the English-French participants, both the group-averaged and the single-subject results showed that there was no consistent language-specificity for any of the three sub-groups in the single-language-specific voxels ( Figure S3 third and fourth columns). That is, the direction of language specificity for the word-components (LE vs. LF, BE vs. BF, QE vs. QF) did not match that of the real-word conditions, and the real-word conditions in the main fMRI runs were not even consistent with those of the localizer. Again, this supports our finding in the main text that the English-French participants use the same set of brain areas to process the written scripts of both English and French.
On the other hand, the language specificity of the English-Chinese participants was very consistent between the localizer and the main fMRI runs, regardless of whether they overlapped with the bilingual clusters or not. The single-language-specific non-overlapping voxels showed language specificity (higher activity amplitude), consistent with the finding that the bilingual word-specific voxels did not include all language-specific voxels. The slopes in the VOTC mask across conditions were also consistent with what we reported in the main text: In the bilingual clusters, a slope was present for both English and Chinese; however, in the Chinese-selective clusters though, a slope was present only for Chinese but not for English. VOTC word-specific activation clusters were observed in these participants. A minority of datasets (3/31) were affected by signal dropout (SB01, SB02, SE01). Panel B shows the latter two participants. Note that even in this case, the clusters in the posterior VOTC (locations around the classic "VWFA") were always present (TAL Y < -50, see Figure S2). The signal dropout was caused either by the inclusion of the ear canal into the data acquisition slab (SB01, SB02), or by not being able to use the dielectric pads in the tight space of the head coil (SE01). The resulting decrease in SNR in lateral temporal cortices can be observed in both the anatomical and functional images. In both A and B, the red lines overlaid on the anatomical images indicate the bottom edge of the data acquisition slabs, and were parallel to the in-plane phase encoding direction.     Chinese words > faces, bodies, houses, tools; and faces > bodies, houses, tools, English and Chinese words. Two were from the main fMRI runs: Chinese words > English words and viceversa. For better visibility, in the inferior view of the brains, Chinese-specific clusters localized close to face-specific clusters are marked with yellow arrows. 8/10 participants had such clusters (participant SC04 and SC08 lack them). For the English > Chinese contrast, participant SC03 was an outlier due to too many (more than 200) activated clusters at p<0.001, and the contrast was plotted at p<0.0001 instead.   B. We labeled units in the dense layer as word-selective if their mean response to words from either language was more than 3 standard deviations above their mean response to natural image categories (i.e. bodies, faces, houses, and tools). Next, each of these word-selective units was labeled as language-selective if the difference between its mean response to the two languages was greater than 1 standard deviation. In the English-Chinese bilingual network, there were on E. Classification accuracy between the word stimuli from two languages across different bilingual and illiterate network layers. We used linear discriminant analysis to identify a linear combination of features that best classified the two languages. To avoid overfitting, we used 5fold cross-validation. As expected, classification accuracy was higher for literate networks (solid lines) than for the illiterate networks (dashed lines), and between Chinese and English stimuli (red lines) than between French and English stimuli (green lines), probably due to differences in low-level visual features. Shaded error bars represent 1 standard deviation. Table S1. Individual language profiles. The age at which participants began acquiring the two languages was self-reported in the language-history questionnaire. Language dominance scores were derived from the number of words read in the one-minute word-reading test.