A universal reading network and its modulation by writing system and reading ability in French and Chinese children
Abstract
Are the brain mechanisms of reading acquisition similar across writing systems? And do similar brain anomalies underlie reading difficulties in alphabetic and ideographic reading systems? In a cross-cultural paradigm, we measured the fMRI responses to words, faces, and houses in 96 Chinese and French 10-year-old children, half of whom were struggling with reading. We observed a reading circuit which was strikingly similar across languages and consisting of the left fusiform gyrus, superior temporal gyrus/sulcus, precentral and middle frontal gyri. Activations in some of these areas were modulated either by language or by reading ability, but without interaction between those factors. In various regions previously associated with dyslexia, reading difficulty affected activation similarly in Chinese and French readers, including the middle frontal gyrus, a region previously described as specifically altered in Chinese. Our analyses reveal a large degree of cross-cultural invariance in the neural correlates of reading acquisition and reading impairment.
Introduction
A large proportion of published studies on reading concern English, even though this language may be considered as an outlier within alphabetic writing systems, compared to Finnish or Italian for example, due to its highly opaque grapheme-phoneme correspondences. Thus, international efforts should progressively extend the results obtained in English subjects, or more generally in Western languages, to other writing systems and languages, notably during childhood. Daniels and Share, 2018 listed 10 dimensions that might affect reading acquisition and dyslexia phenotypes, and that fall into three main classes: the structure of the oral language, the complexity of the visual shapes, and the translation rules between those two domains. With regard to language, the number of phonemes, the syllabic structure and the complexity and regularity of morphological markers can modulate the ease with which children construct explicit representations of speech, which need to be converted in, or deduced from, writing (Goswami, 2008). The visual shapes of letters and characters also vary in number, uniformity, and complexity (Daniels and Share, 2018). Finally, the grapheme-to-phoneme correspondences vary on several dimensions across writing systems, including granularity, complexity, transparency, and consistency (Daniels and Share, 2018). All these factors may influence the speed and effectiveness with which children learn to read, and at least one of these dimensions, orthographic transparency, has been robustly reported to affect reading acquisition in Western languages (Seymour et al., 2003), with a reported impact on brain activation in the adult reading circuit (Paulesu et al., 2000).
Beyond these surface differences, the fundamental logic of reading remains the same from one writing system to another: all of them comprise a restricted number of visual symbols whose combinations allow to access the spoken language network through vision. Although accessing linguistic information from the visual system is a natural possibility for the brain, as demonstrated by the capacities for image naming, lip reading or sign language, writing introduces an additional step that involves converting an arbitrary visual form into speech. Grapheme-phoneme correspondences are arbitrary and have been shown to depend on a precise region of the visual system, the visual word-form area (VWFA) and on the posterior superior temporal cortex, as shown by numerous brain imaging studies in adult readers (Baker et al., 2007; Dehaene et al., 2015; Stevens et al., 2017). However, most of those brain imaging studies suffer from the bias noted above: they concern mainly Western alphabetic languages and mainly adults. Despite the existence of a few previous investigations (Chee et al., 1999; Nakamura et al., 2012; Rueckl et al., 2015; Szwed et al., 2014; Xu et al., 2017), controversy still surrounds the question of whether different cognitive and neural processes are involved in reading non-alphabetic material, or whether reading is based on a similar network regardless of the target language, with only minor variations in the degree of involvement of the different nodes of the network according to the graphic complexity and consistency (Paulesu et al., 2000) or to the linguistic grain size that predominates in a given writing system (Perfetti, 2003; Ziegler and Goswami, 2005). For instance, Chinese characters, like letters, eventually map onto phonology, but they do so at the syllable level and with a considerable degree of irregularity, with (in most cases) no parts in a character corresponding to phonological segments such as phonemes.
An additional source of bias in the literature is that most publications on brain imaging have focused on adults, that is, reading experts. It is entirely conceivable that scanning children during the acquisition of reading would yield different results. The transient mechanisms of learning, in an immature and inexperienced brain, could be based on different mechanisms than those observed in the mature brain (Kersey et al., 2019). This is especially the case for a highly cultural, education-dependent activity such as reading (Dehaene and Cohen, 2007). Children may rely on different and possibly broader regions of the brain before converging onto the adult expert network for reading, for instance transiently recruiting parietal regions for effortful reading (Dehaene-Lambertz et al., 2018; Martin et al., 2015). Conversely, they may also enlarge their responses with skill acquisition, as described in the fusiform region (Olulade et al., 2013). Furthermore, given the differences between writing systems outlined above, initial reading might rest upon different brain areas before converging to a common circuit. Therefore, functional neuroimaging studies in young children, comparing reading acquisition in different writing systems, although difficult, are highly desirable.
Another approach to studying the universality and specificity of the neurocognitive bases of reading is to investigate whether reading-impaired children from alphabetic and non-alphabetic languages exhibit similar brain anomalies. Neuroimaging studies of dyslexia have revealed common neural deficits in different alphabetic languages, with consistently decreased activation (Martin et al., 2016; Richlan et al., 2009) and reduced gray-matter volumes (Linkersdörfer et al., 2012; Richlan et al., 2013) in several left-hemispheric posterior regions, including the left temporoparietal and left ventral occipitotemporal regions. According to two recent meta-analyses, the left ventral posterior occipitotemporal cortex (including the Visual Word-Form Area, VWFA) appears to be the most reproducible and consistent site exhibiting hypoactivation in dyslexic individuals across several alphabetic writing systems regardless of orthographic depth (Martin et al., 2016). Given its sensitivity to visual features (e.g. line junctions) and its efficient reciprocal projections to language areas (Bouhali et al., 2014; Hannagan et al., 2015; Saygin et al., 2016), this area has been proposed as one of the candidates for a universal effect of reading disability (Martin et al., 2016; Pugh, 2006).
By contrast, several neuroimaging studies of Chinese dyslexic children have emphasized the differences between Chinese and alphabetic languages, underscoring the role of the left middle frontal gyrus (LMFG) (Liu et al., 2013; Siok et al., 2004; Siok et al., 2009): Chinese dyslexic children showed a decreased activation in the LMFG compared with typical readers (with eight 11-year-old children in each group) during a homophone judgment task (Siok et al., 2004). This decreased activation in the LMFG was replicated by Siok et al., 2008, with twelve 11-year-old children in each group. It was associated with smaller gray-matter volume at the same location, and no other functional or structural differences was found in the regions singled out by other studies in alphabetic languages. These results were interpreted as showing a clear dissociation of the biological basis of reading disabilities between alphabetic and logographic writing systems. However, these findings contradict behavioral data that show similar profiles in Chinese and alphabetic-language dyslexics (Goswami et al., 2011; Ziegler and Goswami, 2005) and similar predictors of reading abilities in both writing systems. For instance, phonological awareness and morphological awareness (lexical compounding) in 4-year-old Chinese children predict character recognition at 11 years of age, while naming speed (RAN) and vocabulary predict reading fluency (Su et al., 2017). These results are consistent with those obtained in alphabetic languages, which link phonological awareness to reading accuracy and RAN to reading fluency (Landerl et al., 2013; Moll et al., 2014). Moreover, another fMRI study revealed remarkably few differences in brain activity between 11 English and 11 Chinese dyslexic adolescents (13 to 16 years) once all confounding variables (e.g. stimuli and task in fMRI) were controlled for (Hu et al., 2010). It is worth noting that previous fMRI studies of Chinese children with reading disability used a variety of tasks, e.g. picture and semantic matching (Hu et al., 2010), homophone judgment (Siok et al., 2004), font-size perceptual matching (Siok et al., 2009) and a morphological task (Liu et al., 2013), which may partially explain the inconsistency of the findings between studies.
Therefore, to clarify the question of the commonalities and specificities of reading acquisition across different writing systems, we used a similar experimental protocol in 96 Chinese and French 10-year-old readers (48 in each language), with different reading proficiency. Half of the children had reading difficulties, meeting the criteria for dyslexia in French children (more than 2 years of reading delay) though not as marked in Chinese children. (Table 1). As in our previously published fMRI studies of reading (Dehaene-Lambertz et al., 2018; Monzalvo and Dehaene-Lambertz, 2013), all children performed the same passive viewing task with words, faces, and houses, with the mere goal to detect an occasional target star. We studied the effect of reading proficiency in whole-brain analyses and in the specific ROIs highlighted in the literature, in both Chinese and French children. The two analyses are complementary: whole-brain analyses can reveal any regions with differences between Chinese and French children, or between children with and without reading difficulties, including at unexpected brain sites, while ROI analyses have a better sensitivity to detect small differences between groups by summing voxel activity in the cluster and decreasing the number of repeated measures and thus the severity of the correction for multiple comparisons.
Characteristics of the four groups.
Chinese | French | |||
---|---|---|---|---|
Typical readers | Poor readers | Typical | Poor readers | |
Sample size | 24 | 24 | 24 | 24 |
Age in months (SD) | 123 (11) | 123 (10) | 123 (11) | 123 (10) |
Sex | 13M/11F | 16M/8F | 13M/11F | 16M/8F |
Reading ability (CI 95%) | 0.67 (0.49 ~ 0.86) | −1.74 (−1.95 ~ −1.54) | 0.73 (0.39 ~ 1.07) | −2.16 (−2.38 ~ −1.94) |
-
Table 1—source data 1
Demographic information of the four groups.
- https://cdn.elifesciences.org/articles/54591/elife-54591-table1-data1-v2.xlsx
Because group analyses leave open the possibility that the observed group differences might be due to spatially more variable activations in poor readers relative to typical readers, we also performed individual-based analyses and compared the location and activation values of the most responding voxels in each child. Finally, because classical analyses may mask the presence of fine-grained activity patterns that are specific to a given subject or a given category, we also quantified the stability of subject-specific activation patterns within reading-related ROIs using multivariate pattern analyses. The goal of these analyses is to circumvent the blurring effect of group analyses, which may hinder the discovery of genuine but more dispersed activations in poor readers relative to typical readers.
In addition to activations to words, we were also interested in how reading acquisition, in French and Chinese, may differently reorganize the ventral visual areas. Fusiform regions have a distinct maturation profile than more medial regions (Gomez et al., 2017) and face-specific activations expand slowly with age (Golarai et al., 2007; Golarai et al., 2015). Several studies in alphabetic languages suggest that, during reading acquisition in children, words, and faces may compete for cortical territory within the left fusiform gyrus (Centanni et al., 2018; Dehaene-Lambertz et al., 2018; Hervais-Adelman et al., 2019; Li et al., 2013; Ventura et al., 2013). Given the complexity of Chinese characters and their frequently reported bilateral activation, we investigated how face activation might be differently modulated by different reading abilities in Chinese and French children. We thus performed the same analyses for faces than those described above for words and also considered more precisely the development of the anterior-posterior gradient of activations for these two categories.
Results
Behavioral Results
Within the scanner, Chinese children responded faster to the target star than French children (main effect of Language: F (1,92)=60.94, p<0.001). There was no significant effect of Reading ability (F (1,92)<1) nor Language × Reading ability interaction (F (1,92)<1) (Chinese typical readers: 534.95 ± 71.47 ms, Chinese poor readers: 536.13 ± 72.84 ms; French typical reader: 661.15 ± 102.29 ms, French poor readers: 689.32 ± 99.35 ms).
Whole brain analyses
Category-specific activations
We first examined the brain activations to each category (i.e. Words, Faces, and Houses) relative to the other two categories among all participants (see Figure 1A and Table 2). The [Words > Faces + Houses] analysis yielded the usual reading-related regions: fusiform gyrus, posterior superior temporal region, planum temporale, intra-parietal sulcus and inferior frontal regions in the left hemisphere and the posterior superior temporal gyrus in the right hemisphere. We also observed the classic mosaic of category-specific ventral visual areas, with category-specific activation to Houses occupying a medial parahippocampal location, Faces an intermediate fusiform location, and Words a lateral location in the left occipito-temporal sulcus (VWFA). Amygdala responses to Faces were also clearly seen. Those results were seen in each of the four groups of subjects, with the interesting exception that the left VWFA was not seen in both Chinese and French poor readers, contrary to the typical readers, at this classical threshold (voxel-level p<0.001, cluster-level uncorrected) (Figure 1B and Supplementary file 3).

Category-specific circuits.
(A) Category-specific circuits across all participants (voxel-level p<0.001, cluster-level FWE corrected p<0.05). On the left, the reading circuit identified by the contrast Words > [Faces, Houses] and on the right: Face-selective (Faces > [Words, Houses]) and House selective regions (Houses > [Words, Faces]). (B) Category-specific circuits in each of the four groups (voxel-level p<0.001, cluster-level uncorrected). On the left, category-specific activations in Chinese typical readers (above) and Chinese poor readers (below). On the right, category-specific activations in French typical readers (above) and French poor readers (below). Green: regions selectively activated by words (Words > [Face, House]); Red: regions selectively by faces (Face > [Word, House]); Yellow: regions selectively activated by Houses (House > [Word, Face]).
Regions of significant activations for each visual category vs. the two others across all participants.
Region | MNI coordinates | Peak p-value | Peak z-value |
---|---|---|---|
Words > others | |||
Left inferior frontal gyrus | −48 12 30 | 2.06e-19 | 8.93 |
Left precentral | −39 0 36 | 1.45e-18 | 8.72 |
−51 6 39 | 2.91e-14 | 7.51 | |
Left superior temporal gyrus/sulcus | −57 −30 3 | 2.39e-19 | 8.92 |
Left middle temporal gyrus | −51 −42 6 | 7.89e-18 | 8.52 |
Left fusiform gyrus | −48 −57 −15 | 1.69e-17 | 8.43 |
Left Inferior parietal sulcus | −45 −39 42 | 4.29e-14 | 7.46 |
Right superior temporal sulcus | 57 −27 3 | 8.94e-10 | 6.02 |
Faces > others | |||
Left fusiform gyrus | −39 −48 −21 | 3.28e-17 | 8.35 |
Right fusiform gyrus | 42 −54 −18 | 6.14e-26 | 10.47 |
Right amygdala/hippocampus | 18 −9 −18 | 6.11e-22 | 9.56 |
Left amygdala/hippocampus | −18 −9 −18 | 3.91e-15 | 7.77 |
Houses > others | |||
Left fusiform gyrus | −30 −48 −6 | 9.01e-53 | 15.24 |
Right fusiform gyrus | 30 −45 −9 | 2.90e-50 | 14.86 |
27 −63 −9 | 2.40e-22 | 9.65 | |
Left calcarine | −18 −54 9 | 8.66e-10 | 6.02 |
Reading-related differences
Reading scores across all 96 children were significantly correlated with fMRI activation in the [words > fixation] contrast, in the classical regions of the reading circuit in the left hemisphere (fusiform, superior temporal sulcus, middle frontal region and precentral) plus some of their right counterparts (Figure 2 and Table 3). When the French and Chinese groups were considered separately, similar regions (bilateral FFG, bilateral PCG, bilateral MFG, left STS,) were observed in each of the two languages (Figure 2B and C).

Neural correlates of inter-individual variability in reading scores.
(A) The figure shows the regions whose activation in the words versus fixation contrast was positively correlated with reading scores across all participants at the whole-brain level (voxel-wise p<0.001 and cluster-wise p<0.05 FWE corrected).(B, C) Regions positively correlated with reading scores in Chinese and French children (voxel-wise p<0.001, cluster-level uncorrected).
Regions significantly correlated with reading scores across all participants at the whole-brain level.
Region | MNI coordinates | Peak p-value | Peak z-value | R-value |
---|---|---|---|---|
Left fusiform gyrus | −42 −45 −18 | 6.65e-6 | 4.36 | 0.490 |
Right fusiform gyrus | 42 −66 −24 | 9.96e-6 | 4.27 | 0.512 |
Left precentral | −36 −3 57 | 2.75e-6 | 4.54 | 0.467 |
Right precentral | 54 18 33 | 1.29e-5 | 4.21 | 0.432 |
Left middle frontal gyrus | −36 12 27 | 1.92e-5 | 4.12 | 0.468 |
Right middle frontal gyrus | 45 6 54 | 7.72e-6 | 4.32 | 0.460 |
Left superior temporal sulcus | −57 −24 0 | 1.00e-5 | 4.26 | 0.510 |
Right middle occipital gyrus | 27 −69 42 | 6.34e-6 | 4.37 | 0.448 |
We also observed a significant positive correlation, across all children, between reading scores and the face-evoked activation (vs fixation) in the left fusiform gyrus (left [−33 −60 −15], 74 voxels, Z = 4.31, pFWE_corr = 0.001). This region was close to coordinates of the classic left FFA (Scherf et al., 2007). A small cluster was also observed close to the classic right FFA, but did not survive the cluster-level correction ([45 −48 −15], 12 voxels, Z = 3.70).
Group comparisons
No significant difference was found between typical and poor readers in either direction when analyzing either the [Words > fixation] or the [Words > Faces+Houses] contrast across the whole brain. However, a few voxels reached the voxel-wise statistical threshold (p=0.001) in the left fusiform gyrus, left precentral and left superior temporal sulcus with greater activation in typical readers relative to poor readers (see Figure 1—figure supplement 1A).To improve the sensitivity to differences between groups and decrease the risk of false negatives, we restricted our analyses to reading sensitive regions defined by the mask comprising all voxels showing a preference for words relative to the two other categories across all participants. Typical readers showed larger activations relative to poor readers in a left precentral cluster (79 voxels, pFWE_corr = 0.027, Z = 3.69 at [−51 15 33]) and Chinese relative to French in the left intra-parietal sulcus (55 voxels, pFWE_corr = 0.004, Z = 4.29 at [−30–60 39]) in the words vs. fixation contrast (Figure 1—figure supplement 1). No region showed more activation in French children relative to Chinese nor a significant language × reading ability interaction, even when a very lenient voxel-wise threshold of p<0.05 was considered.
To summarize the results so far, our analyses recovered, in a large group of 10-year-old children, the classical activations for words, faces, and houses described in adults. Reading proficiency modulated the response to words in the classical reading circuit and contralateral regions, but also to faces in the fusiform region, in both languages. However, a binary classification of the participants in typical and poor readers was less powerful, recovering a few voxels with significant hypo-activations at expected locations in the poor readers relative to the typical readers, which did not survive corrections for repeated measures. To circumvent this reduced power, we next focused on brain regions which have been reproducibly shown to be under-activated in dyslexics or modulated by Chinese writing, and studied whether and how reading ability and writing system affected their response to words.
Literature-driven analyses
Figure 3A presents all foci reported in four published meta-analyses of dyslexia in alphabetic languages (Linkersdörfer et al., 2012; Maisog et al., 2008; Richlan et al., 2009; Richlan et al., 2011) and in four meta-analysis of Chinese typical reading (Bolger et al., 2005; Tan et al., 2005a; Wu et al., 2012; Zhu et al., 2014). As seen in Figure 3A, dyslexia in alphabetic languages is consistently characterized by dysfunctions in the left occipito-temporal, temporoparietal and frontal regions. All ROIs (except two ROIs in the right hemisphere) fell within the reading circuit identified in our participants (Words > other categories; see Figure 4 and Figure 4—figure supplement 1). Figure 3B presents the correlation (FDR-corrected) between reading score and activation to words in these ROIS. This correlation reached FDR-corrected significance in PCG, STS, MTG, and FFG, as well as the MFG and RIOG previously reported in Chinese readers (Figure 3B).

Regions of interest (ROIs) used to analyze the data.
(A) Each sphere represents a peak reported in the literature; Labels in white background indicate foci reported in meta-analyses of dyslexia in alphabetic languages; Labels in black background indicate foci reported in meta-analyses of Chinese reading. (B) ROIs used in the current study. Coordinates of foci (see the upper graph) belonging to the same functional region were averaged to create 6-mm-radius spheres at the averaged coordinates. Dots are colored according to their correlation with reading scores across all children. Red dots represent ROIs whose activation to words versus fixation were positively correlated with individual children’s reading scores, while gray dots represent non-significant ROIs (pFDR < 0.05). MFG: Middle Frontal Gyrus, PCG: Pre-Central Gyrus, STS: Superior Temporal Sulcus, MTG, Middle Temporal Gyrus, FFG: Fusiform gyrus, IOG: Inferior Occipital Gyrus.

Effects of Reading ability and Language on the words versus fixation contrast in the selected ROIs.
Brain slices showed the literature-based ROIs (cyan) overlaid on the reading circuit (red-yellow) in our participants (Words > [Faces, Houses]). Plots show the mean activation for words > fixation, in each of the four groups and ROIs. The words ‘Reading ability’ and ‘Language’ indicate a significant main effect of reading ability and a main effect of language in the ANOVA (after FDR correction for a total of 13 ROIs). Note that no ROI showed a significant interaction of language × reading ability. Brackets indicate significant post-hoc analyses: **p<0.005, **<0.01, *<0.05, †=0.052.
-
Figure 4—source data 1
Activation to words in each of the four groups and ROIs.
- https://cdn.elifesciences.org/articles/54591/elife-54591-fig4-data1-v2.xlsx
We submitted the activation to words relative to fixation in each ROI to a 2 × 2 ANOVA with Language and Reading ability as between-subject factors. Below, we report only the p-values that survived an FDR correction over the 13 ROIs.
A significant main effect of Language was observed in the left middle frontal gyrus (F (1,92)=15.23, p<0.001, pFDR_corr < 0.001), superior parietal lobule (SPL, F (1,92)=8.13, p=0.005, pFDR_corr = 0.022) and posterior superior temporal gyrus (pSTG, F (1,92)=9.04, p=0.003, pFDR_corr = 0.020), always due to larger activation in Chinese readers than in French readers (see Figure 4).
Significant reduced activation was observed in poor readers relative to typical readers in the left fusiform gyrus (F(1,92) = 21.08, p<0.001, pFDR_corr < 0.001), middle frontal gyrus (F (1,92)=10.17, p=0.002, pFDR_corr = 0.009), superior temporal sulcus (F (1,92)=11.88, p=0.001, pFDR_corr = 0.006), and precentral gyrus (F (1,92)=7.78, p=0.006, pFDR_corr = 0.020). Importantly, it was the case within each language group, except for the pSTG where a significant difference between typical and poor readers was only observed in French but not Chinese children (Figure 4). No ROI showed a significant language × reading ability interaction (see Figure 4 and Figure 4—figure supplement 1), even the pSTG (F (1,92)=2.815, p=0.097 before FDR correction).
We used Bayesian ANOVAs on the Word activation to assess the likelihood of the null hypothesis H0 over H1 in the case of the interaction language × reading ability. As also explained in the methods, the Bayes factor (BF10) is the ratio of the amount of evidence for H1 above H0 (BF01 for H0 above H1). Evidence for H1 against H0 is generally considered as moderate for BF10 ≥ 3, and strong for BF10 ≥ 10. Bayesian analyses provided similar conclusions to frequentist analyses for the main effect of Language and Reading ability (Table 4). The BF10 for an effect of language was 80.21 in the MFG, 7.05 in the SPL and 9.91 in the pSTG. The likelihood of an effect of reading ability in the FFG was BF10 = 1528.97 higher than that of the null hypothesis; the same BF10 was 37.51 in the STS, 9.95 in the MFG and 5.63 in the PCG respectively. It was also the case in the post-hoc analyses within each language (Table 4). Notably, there was strong evidence of a reading ability effect in the fusiform gyrus (FFG) and moderate evidence in the STS and PCG in both languages. In the pSTG and also paradoxically in the MFG described as a dyslexic marker in Chinese, the evidence of a reading ability effect was strong in French (BF10 = 23.74 and 29.35 for each site respectively) but absent or weak in Chinese children (BF10 < 1 and 1.62). Sensitivity analysis revealed that the Bayes factor stayed about the same for a wide range of prior specifications (Cauchy prior width: 0–1.5) in the FFG, STS, PCG, and MFG for the comparisons between typical and poor readers in each language except for the pSTG in Chinese children.
Bayes factor (BF10) in Bayesian ANOVA analyses of children’s activation to words versus fixation.
ROIs | Main effect of language | Main effect of reading ability | Interaction | Post-hoc analyses | |||
---|---|---|---|---|---|---|---|
Typical readers (Chinese vs. French) | Poor readers (Chinese vs. French) | Chinese children (typical vs. poor readers) | French children (typical vs. poor readers) | ||||
FFG | 0.216 | 1528.966 | 0.313 | 0.227 | 0.337 | 13.577 | 93.163 |
MFG | 80.211 | 9.948 | 0.305 | 2.837 | 309.651 | 1.616 | 29.351 |
STS | 0.220 | 37.507 | 0.309 | 0.387 | 0.259 | 6.933 | 5.076 |
PCG | 1.532 | 5.625 | 0.275 | 1.782 | 1.075 | 2.662 | 2.661 |
SPL | 7.052 | 0.562 | 0.323 | 4.486 | 1.782 | 0.822 | 0.332 |
pSTG | 9.908 | 0.289 | 1.054 | 0.616 | 74.014 | 0.215 | 23.744 |
-
The value of Bayes factor BF10 means that data are n times more likely under alternative hypothesis (H1) than null hypothesis (H0). The alternative hypothesis in comparisons between typical and poor readers is group 1 (typical) > group2 (poor); the alternative hypothesis in comparisons between languages is group 1 (Chinese) > group2 (French).
Turning now to the effect of language in each reading group, differences between languages were mainly observed between the groups of poor readers: Chinese children showed larger activations than French in the MFG (BF10 = 309.65) and pSTG (BF10 = 74.01) although the same tendency was present in typical readers. The above Bayes factor stayed the same for a wide range of prior specifications (Cauchy prior width: 0–1.5) in the sensitivity analysis. However, there was no evidence for a significant interaction language × reading ability in all these ROIs. On the contrary, the null effect was supported by moderate evidence in the FFG, MFG, STS, PCG, and SPL (respectively BF10 = 0.31, 0.31, 0.31, 0.27, and 0.32, i.e. the likelihood of the null hypothesis BF01 = 3.23, 3.23, 3.23, 3.70, and 3.13); there was no evidence in either direction in pSTG (BF10 = 1.05) (Table 4). Sensitivity analysis revealed that the Bayes factor stayed about the same for a wide range of prior specifications (Cauchy prior width: 0–1.5).
These analyses were replicated for the activation to houses and to faces, no main effect nor interactions were found in any of these ROIs.
To summarize, when analyses were focused on specific ROIs outlined in the literature as sensitive to reading performance or to differences between writing systems, we replicated the reduced activation to words in poor readers relative to typical readers in the FFG, MFG, STS, and PCG. Crucially, this reduction was observed in both French and Chinese participants with no significant interaction language × reading ability. Activation in SPL, pSTG, and MFG was also modulated by language, with greater activation in Chinese than in French children. For the pSTG, the language effect was mainly observed in the poor readers, due to a large reduction in activation in French poor readers than Chinese poor readers.
Individual analyses
The above analyses were carried out in a standardized way at the group level. It is therefore possible that the observed group differences were due to a greater inter-individual variability in brain localization in the reading-impaired group than in the typical group. This possibility would lead to a completely different interpretation of the results: each child might have a well-organized brain activity for reading, with the only anomaly being a greater anatomical dispersion in the group of poor readers compared to the group of typical readers. To test this possibility, we performed two individual-based analyses, one based on the comparison of the location and activation values of the most responding voxels and the other examining the stability of the pattern of responses across runs through a multi-voxel pattern analysis (MVPA). We focused on the ROIs previously showing significant differences due to reading ability (i.e. left FFG, MFG, PCG, and STS) and Language (i.e. left MFG, SPL, and pSTG).
Peak analyses
Considering the locations of the individual centers of mass for word activations in the left FFG, MFG, STS, PCG, SPL, pSTG, their Euclidean distance to the group peaks did not differ between poor and typical readers (Supplementary file 6), suggesting a similar dispersion among poor readers and typical readers. These results of a null effect in peak location were further confirmed by Bayesian analyses (the BF10 for an effect of reading ability were 0.22, 0.25, 0.23, 0.36, 0.23, 0.24 for left FFG, MFG, STS, PCG, SPL, and pSTG, respectively, thus supporting the null hypothesis). Furthermore, even after having selected the best responding voxels in each child, the word activation remained weaker in poor than typical readers in the left FFG, MFG, STS, PCG, and pSTG (all pFDR_corr < 0.05). French also yielded weaker activations than Chinese children in the left MFG, pSTG and SPL (all pFDR_corr < 0.05). There was no significant Language × Reading ability interaction in any of these analyses, supported also by the following BF10 in Bayesian analyses (peak location: BF10 = 0.38, 0.30, 0.70, 0.35, 0.31, 0.35; peak activation: BF10 = 0.35, 0.38, 0.59, 0.35, 0.00, 0.53 for left FFG, MFG, STS, PCG, SPL, and pSTG). These results thus corroborated the standard analyses.
As concerns faces, we did not observe any effect of reading ability or Language on the peak locations and activations in bilateral fusiform face gyrus (FFA) in both frequentist (all p>0.09) and Bayesian statistics (all BF10 < 1).
MVPA of the activations to words
The previous analysis asked whether, at the individual level, the peak activity is reduced in poor readers. In the present section, we ask the same question about the stability of the response pattern across runs (i.e. the within-subject reliability of the activation patterns). Indeed, the reduced activation in the group analysis could be due to two distinct causes. Poor readers could have a genuinely reduced and erratic activation, but alternatively, poor readers could possess an identifiable and reproducible circuit similar to typical readers, only with a spatially more dispersed extent. In that case, the within-subject reproducibility of multivariate activation patterns may not differ between typical and poor readers.
To examine the within-subject reliability of activation across runs, we used a multi-voxel pattern analysis (MVPA) focused on the same regions than above. We computed the correlation between pattern of activations in run 1 and 2, separately for within-category patterns (words in run 1 and words in run 2) versus between-category patterns (average of words-faces, words-houses and faces-houses, each in run 1 versus run 2). A reproducible reading circuit should result in a significant effect of condition (greater within-category correlation than between-category correlation). Furthermore, if the activation pattern is less reproducible in poor readers than in typical readers, there should be a significant interaction of this Condition effect with Reading ability.
In all the ROIs, when pooling over all subjects, there was an overall replicable pattern of activation evoked by words, as indicated by a significant main effect of condition, with a greater correlation coefficient within words than between words and other categories (all pFDR_corr < 0.001) (see Figure 5 and Figure 5—figure supplement 1A). Crucially, we also observed a significant interaction of condition × reading ability (typical vs. poor readers) in the left FFG (F (1, 83)=10.14, p=0.002, pFDR_corr = 0.006) and in the left pSTG (F (1, 83)=15.75, p<0.001, pFDR_corr < 0.001). Post-hoc analysis found that typical readers, but not poor readers, exhibited a significantly similar pattern of activation from one run to the next. Those results show that the above differences between typical and poor readers were not due to an artifact of group averaging, and that children struggling with reading acquisition exhibited a genuinely more erratic activation pattern for words in these regions (Figure 5).

Multivariate pattern analysis indicating that the word-induced activation is less reproducible in poor readers.
Within the designated ROIs, we computed the correlation coefficient of the multivoxel patterns for the word versus fixation contrast in run 1 and in run 2 (within-category correlation). For the between-category coefficient, the plots show the average correlation coefficient between words and faces, words and houses and faces and houses between run 1 and run 2. In each plot, the correlation is presented for the Chinese children on the left of the plot and for French children on the right. The words ‘Condition × Reading ability’ indicate a significant interaction between condition (within vs. between) and the status of children (typical vs. poor readers) (after FDR correction for a total of 13 ROIs). Typical readers, but not poor readers, exhibited a significant similar pattern of activation to words from one run to the next in the left FFG and pSTG, suggesting that the weaker activation for words in poor readers was not due to greater anatomical variability in this group but was related to genuinely less reproducible activation patterns than in typical readers.
-
Figure 5—source data 1
Correlation coefficients for within-category patterns and between-category patterns in each of the four groups and ROIs.
- https://cdn.elifesciences.org/articles/54591/elife-54591-fig5-data1-v2.xlsx
MVPA of the activations to faces
For the MVPA analysis in the bilateral fusiform face areas (bilateral FFA), only the main effect of condition reached significance (pFDR_corr < 0.001), with a greater correlation coefficient for within-category patterns (faces-faces) than for between-category patterns (faces-words, faces-houses, words-houses). Neither the main effect of reading ability nor the interaction of condition × reading ability reached significance. These results suggest an equally replicable pattern of activation to faces in poor and typical readers bilaterally in the fusiform face area (see Figure 5—figure supplement 1B). This pattern resulted in a triple interaction of category (words versus faces)×condition × reading ability when the fusiform data from words and faces were analyzed together, indicating that the reduced stability of activations in poor readers was limited to words.
In summary, individual analyses substantiated a genuine reduction and instability of the activation to words in poor readers relative to typical readers. Importantly, this instability was not universally present in all visual categories (as might be the case if, for instance, the poor readers had greater noise or motion), because the activations to Faces in the fusiform regions did not differ between groups.
Anterior-to-posterior gradient in the visual cortex
Finally, given the massive impact of reading on the organization of the ventral temporal areas, we focused on these regions and examined the anterior-posterior gradient of responses for the different visual categories vs. fixation.
Words
Keeping constant x = ±48 and z = −16, we studied the activation to Words vs. fixation along the y-axis (ranging from −79 to −22). Firstly, we observed greater response to words in Chinese children compared to French children in the right hemisphere at several y coordinates, leading to a significant triple interaction of language × hemisphere × ROI. Secondly, we observed larger activation in the posterior relative to anterior sites as revealed by the main effect of ROIs. Crucially, the reading ability × ROIs interaction was significant. In more detail, compared to typical readers, poor readers had weaker activation to words at several consecutive sites (y-axis at −64, −55, −46, and −37). However, when we considered separately Chinese and French children, only one site (y = −46) survived correction for multiple comparison in both Chinese and French children. This site is only slightly anterior to the classic VWFA site (Cohen et al., 2002) (see Figure 6B).

Mosaic of preferences for different visual categories (Words, Faces and Houses) in the ventral visual cortex.
(A) Slices show the activation difference between a given category and the others across all participants. (B) fMRI signal change of Words relative to fixation in both Chinese and French children in successive cortical sites along the y-axis (green dots on the left cortical slices) with constant x = ±48 and z = −16. Both Chinese and French poor readers have significantly lower activations relative to their controls at a specific y site of y = −46 (pFDR_corr = 0.048 and pFDR_corr = 0.012 respectively for Chinese and French children) corresponding to the classical coordinates of the VWFA (Cohen et al., 2002). (C) fMRI signal change of Faces relative to fixation in both Chinese and French children in successive cortical sites along the y-axis (red dots on the left cortical slices) with constant x = ±39 and z = −16. (D) fMRI signal change of Houses relative to fixation in both Chinese and French children in successive cortical sites along the y-axis (yellow dots on the left cortical slices) with constant x = ±30 and z = −16. Additional files.
-
Figure 6—source data 1
Activaions to different categories relative to fixation in ROIs in the ventral visual cortex.
- https://cdn.elifesciences.org/articles/54591/elife-54591-fig6-data1-v2.xlsx
Faces
Keeping constant x = ±39 and z = −16, we also studied the activation to Faces vs. fixation along the y-axis (ranging from −79 to −22). We observed a significant Hemisphere × ROI interaction. This effect was due to greater right than left face activation at several y coordinates (y = −73, −64, −55, −46). Besides, the main effect of reading ability reached significance, due to a lower activation to faces in poor readers compared to typical readers bilaterally and in both languages (Figure 6C).
Houses
Along the medial house-specific activation at x = ±30 and z = −16, we similarly studied the activation to Houses vs. fixation. We found a significant triple interaction of language × hemisphere × ROI. French children had greater right than left activation at each of the six anterior-posterior y coordinates (all pFDR_corr < 0.005) while Chinese children had the same pattern only at four sites (y = −73, −46, 37, 28). We also observed a significant reading ability × ROI interaction, with decreased activation to Houses in poor readers in several sites (Figure 6D).
In summary, we recovered the expected anterior-posterior and medial to lateral gradient of preferences for the different visual categories. Chinese children had larger right-hemispheric activation to words than French children; and poor readers, in both languages, showed lower activation than typical readers at the VWFA site. Congruent with the whole-brain analyses, we observed an effect of reading ability on Face activations in both hemispheres. The activation to Houses was also reduced in poor readers at several sites.
Discussion
In the present study, we examined the universality and specificity of the neural bases of reading among novice readers by comparing two very different writing systems: French and Chinese. Our goal was to study whether the same neural circuit was involved in the success and failure of reading acquisition, regardless of the complexity of the written symbols and the size of the speech units that are mapped onto them. We investigated this question in 96 10-year-old children using a similar paradigm in both countries with a minimally demanding task (i.e. detecting a star) that did not directly tax reading or reading-related cognitive skills and thus was equally easy for everyone in four age-matched groups (French and Chinese × typical and poor readers).
First, in a whole-brain analysis, we recovered the classical category-specific activations for words, faces and houses in extra-striate visual areas across all participants but also in each group (Figure 1). Second, reading scores were correlated with the word activations in common key regions of the reading circuit (left VWFA, posterior superior temporal gyrus/sulcus, middle frontal gyrus and precentral gyrus) but also in the right hemisphere (middle occipital and fusiform gyri and precentral). Third, analyses based on ROIs from the literature confirmed these results, including an effect of reading ability in the left middle frontal gyrus which was true in both French and Chinese children, contrary to previous suggestions of a Chinese-specific effect at this site. The main effect of reading ability across languages was replicated even when analyzing only the most responsive voxels in each individual.
We also observed a few differences in activations depending on the children’s native language. Chinese reading tended to engage more symmetrical activations in the visual system, with stronger activations in the right hemisphere than French readers when we specifically tested the anterior-posterior organization of the fusiform region. Chinese children also had stronger activations than French children in the left parietal region, middle frontal region and posterior superior temporal gyrus.
We concluded our analyses by examining the reproducibility of the activation patterns between runs. The within-subject pattern of activity evoked by words was reproducible across runs in typical readers in all key reading regions, underscoring that the reading circuit is stable after 3 years of learning to read and can be reliably measured in a single fMRI run even in children. However, such was not the case for poor readers, whose activity was significantly less reliable in left fusiform and posterior superior temporal gyrus in both Chinese and French poor readers. We now discuss each of these results in turn.
The reading network is largely universal, and modulated by reading skill
A long-standing debate in reading research is whether the neurobiological circuitry for reading is universal across languages. Previous cross-cultural fMRI studies have compared brain activations in adult readers in different languages and suggested that the expert reading network may be universal across languages (Nakamura et al., 2012; Paulesu et al., 2000; Rueckl et al., 2015). The current study extends this finding to young children by showing common activation patterns to words in children in both alphabetic and non-alphabetic writing systems after only 3–4 years of primary school. Script invariance across English and Chinese was already reported after only one year of reading acquisition in the VWFA (Krafnick et al., 2016). This region is also the most strongly modulated by reading proficiency in both French and Chinese, with no difference (Bayes factor < 1) between French and Chinese typical readers but also between French and Chinese poor readers, thus confirming that VWFA activation is a universal marker of reading proficiency. The same observation can be made to a lesser extent for the STS and the PCG, the activation of which also depends on the reading proficiency in both writing systems in a similar way.
These cultural invariance findings are in agreement with the neuronal recycling hypothesis, according to which recent cultural acquisitions (e.g. reading) rely on the preemption of universal pre-existing circuits of the human brain, with only small culture-dependent modulations (Dehaene and Cohen, 2007). In any language, reading recruits a pre-existing circuit that connects visual areas capable of learning to recognize orthographic symbols, and spoken language processing areas (Hannagan et al., 2015; Saygin et al., 2016). Successful literacy acquisition is thus the result of the convergence of visual and speech processing systems, both of which are likely to be largely universal and partially laid down under genetic control. Indeed, the spoken language network is already present at its usual left-hemispheric location in 2-month-old babies (Dehaene-Lambertz et al., 2002; Dehaene-Lambertz et al., 2006; Dehaene-Lambertz and Spelke, 2015). The specific subpart of the ventral visual pathway which is used to recognize written characters seems to be, at least in part, determined by its pre-existing connections with this language network (Barttfeld et al., 2018; Bouhali et al., 2014; Hannagan et al., 2015; Saygin et al., 2016) and by its micro-structure (Weiner et al., 2016). Thus, according to the neuronal recycling hypothesis, in spite of variations in language and writing systems, a considerable amount of inter-cultural convergence should be expected.
In our study, although the main areas for reading were common to both groups, we also observed modulations of the amplitude of brain activity within culturally universal brain circuits. Chinese children had larger activations than French children in (1) left intraparietal sulcus; (2) left middle frontal gyrus (BA 9); (3) right hemisphere occipitotemporal regions; and (4) left posterior superior temporal sulcus.
Brain imaging research has well documented the role of the left intraparietal sulcus in visuo-spatial encoding and attention (Davranche et al., 2011; Offen et al., 2010), including the serial analysis of letters in a word or pseudoword (Cohen et al., 2008). We therefore speculate that the greater involvement of this region in Chinese reading is presumably due to the greater amount of spatial analysis required by Chinese character recognition. Indeed, activations in the inferior parietal region are observed in French children (Dehaene-Lambertz et al., 2018), but only transiently during initial reading instruction, when children must effortfully pay attention to the left-right succession of letters and disambiguate mirror-letters. Congruently, improvement in reading skills over the first year of school correlates with microstructural changes in the connectivity between the VWFA and the inferior parietal region (Moulton et al., 2019). Activation in the intraparietal sulcus is also needed when adults read in an unusual format (vertical French words for example) (Cohen et al., 2008) or determine the position of a letter in a word (Ossmy et al., 2014), underscoring the role of this region in orienting attention and computing the spatial relationships of word elements. Our results indicate that this process is involved in alphabetic languages as well as in Chinese, but its activation may remain weak in alphabetic languages due to the simplicity of the spatial relationships.
The greater involvement of the left middle frontal gyrus (BA9) in Chinese than in French children is in line with existing research (Bolger et al., 2005; Tan et al., 2005a). This region has been proposed to support language-specific processing required by Chinese reading, such as whole-syllable retrieval (Tan et al., 2005a), the tonal nature of Chinese phonology (Gandour et al., 2002), memory-based lexical integration of orthography, phonology and semantics (Perfetti et al., 2005), visuospatial working memory (Wu et al., 2012) or writing gesture information (Cao and Perfetti, 2016; Nakamura et al., 2012). In the reading of English, its proposed functional role is also diverse: lexical semantics (Bolger et al., 2005), phonological processing (Pugh et al., 1996), lexical selection (Bolger et al., 2008), or grapheme–phonology conversion (Jobard et al., 2003).
Perhaps the most likely explanation is that the LMFG, as well as the more medial and dorsal Exner area, are systematically involved in writing (Planton et al., 2013; Purcell et al., 2011). FMRI studies in typical adult readers, comparing French and Chinese reading, have shown that the left MFG is not specific to Chinese writing, but includes a representation of handwriting gestures that is engaged in both alphabetical and non-alphabetical languages (Nakamura et al., 2012). As proposed by Nakamura et al., 2012, the activation in the MFG and PCG, as seen here, might correspond to the ‘reading by hand’ circuit (gesture recognition system) as opposed to the ‘reading by eye’ circuit (shape recognition system in the VWFA). Furthermore, writing is not only learned in parallel with reading, but facilitates reading acquisition in children (Bara et al., 2004). Thus, the involvement of the left MFG in French children, although less important than in Chinese children, might be due to a heavier dependence on handwriting during reading acquisition than during adulthood. This interpretation is supported by a recent study showing that, in second-language learners of Chinese, viewing characters that were learned through character writing induced greater activation in LMFG compared with those learned without character writing (Cao and Perfetti, 2016).
The greater activation in the right ventral visual system in Chinese reading has been reported in previous literature and this region has been found to show a children-to-adult developmental increase for Chinese, suggesting that this region is especially important in Chinese reading acquisition, presumably because of its involvement of holistic visuo-orthographic processes (Bolger et al., 2005; Cao et al., 2014; Tan et al., 2005a).
We also found that Chinese children showed slightly larger activation than French children in the posterior superior temporal gyrus/sulcus. This region is associated with phonological processing and grapheme-phoneme conversion (Turkeltaub et al., 2003). Previous findings in adults suggested that it is more engaged in alphabetic than in logographic languages (Bolger et al., 2005; Tan et al., 2005a), but our study showed an opposite pattern. In Chinese reading, a large number of written characters correspond to the same syllable, thus phonological information is insufficient to access semantics of a printed character. As a result, Chinese readers must rely more heavily on the direct route from orthography to the lexicon (Cao et al., 2010; Shu et al., 2003). However, phoneme-level representation still plays an essential role in learning to read Chinese, all the more that syllable decoding is complex. This may be especially true in beginners, who rely on Pinyin, an alphabetic notation which allows children to manipulate different phonological units such as decomposing a syllable into onset, rime, tone and phonemes. Greater dependence on the grapho-phonological route in children than in adults persists until at least grade four in Portuguese (Ventura et al., 2007). It is also possible that the need to suppress the activation of the syllable associated with the character that competes with character identification, transitorily increases activation in that region. Furthermore, when carefully comparing the activation of the posterior temporal gyrus with other regions, we found that the greater activation in Chinese was mainly due to a weak activation in French poor readers (see Figure 4). These children were defined as dyslexics in France and might have been more impaired than the Chinese poor readers. Interpretation must be very cautious, given that the language × reading ability interaction was not significant, and that the Bayesian analysis provided no evidence favoring H0 or H1. Further research will be needed to probe whether a reduction in pSTG activation, presumably due to a phonological impairment, is a more frequent cause of reading disability in alphabetical scripts than in Chinese, as suggested by numerous prior studies of dyslexia (Hulme et al., 2002; Ramus and Szenkovits, 2008; Wydell and Butterworth, 1999).
Overall, our findings indicate that learning to read largely involves the same key regions across cultures and ages, but with quantitative modulations depending on the specific demands of the task, the learning stage, and culture-dependent characteristics. Our novice readers, who probably needed to deploy all their resources in order to succeed in reading, may be more informative than expert adults in objectifying the entire reading circuit, unless the latter are pushed to their limits by unusual, thus less automatized, format of words presentation, as was done for instance by Cohen et al., 2008.
A universal neural phenotype for reading disability
Some authors have also proposed that the cortical regions mediating reading disability are different in Chinese and alphabetic languages (left middle frontal gyrus in Chinese versus left temporo-parietal regions in alphabetic languages) (Siok et al., 2008; Siok et al., 2004). On the contrary, our results were strikingly similar in Chinese and French poor readers tested with the same paradigm, thus suggesting a universal neural phenotype for reading disability. Those children were impaired in all classical reading-related regions, most notably the VWFA in the left FFG and the left posterior STG, regions that have been consistently reported to show lower activations in dyslexics relative to controls in alphabetic languages (Blau et al., 2009; Martin et al., 2016; van der Mark et al., 2009). The present study confirms that these reductions in activation can be observed since childhood in all writing systems, whether alphabetic systems with deep or shallow orthographies, or in Chinese characters. Hu et al., 2010 reached a similar conclusion at a later age by comparing a smaller group of 11 Chinese and 11 English dyslexics.
Thanks to individual peak location and intensity analyses, as well as multivariate pattern analyses, we could reject an alternative interpretation which, to the best of our knowledge, was not explicitly tested in previous studies: the possibility that the reduced activations are an artifact of group averaging, solely due to greater inter-individual variability in the localization of reading-related circuits in the dyslexic brain. Using individual peak, we observed that the brain localization to words were not more dispersed among poor readers than among typical readers. Using MVPA, we showed that, within individual subjects, the activation patterns in the VWFA in response to written words were less reproducible across runs in poor readers than in typical readers. This was solely the case for words, not for the other visual categories. We did observe a slightly reduced activation to faces and houses in poor readers relative to typical readers, as previously reported in illiterate subjects (Dehaene et al., 2010), as well as seen in response to non-word stimuli (numbers, abstract strings) (Boros et al., 2016) and faces in dyslexics (Gabay et al., 2017; Monzalvo et al., 2012). However, the pattern of activity for faces was stable from one run to the next, contrary to what was found for words in the MVPA analyses. This observation suggests that the reduced activation to written words may not reflect a general disorganization of the extra-striate visual areas, but rather a specific difficulty with written words.
Given that the site of the VWFA does not activate strongly to written words in illiterates (Dehaene et al., 2010; Hervais-Adelman et al., 2019) and in typical children before they learn to read (Dehaene-Lambertz et al., 2018), its under-activation in poor readers may simply reflect the lack of reading practice – in other words, it might be a consequence rather than a cause of the reading deficit. Dyslexia is likely to be a heterogeneous deficit with a variety of different causes, including a phonological deficit in many children, but also visual attentional deficits, plausibly anywhere along the complex processing chain that leads from print to sound and meaning (Friedmann et al., 2010; Paulesu et al., 2014; Valdois et al., 2012). The present study cannot distinguish cause from consequence, but it does demonstrate that a reduced activation in the VWFA is one of the dominant fMRI signatures of reading poor performances.
The other major site of atypical activation reported in the dyslexia literature is the left posterior STG, a region involved in grapheme-phoneme conversion (Frost et al., 2009; Turkeltaub et al., 2003) although this location is less consistently found in children than in adults (Richlan et al., 2011). In our data, only French poor readers had a significantly lower activation relative to their controls at this location, but in both languages, poor readers exhibited an unstable activation pattern across fMRI runs, suggesting fragile phonetic representations evoked by written stimuli. In agreement with this interpretation, Vandermosten et al., 2020 also observed disrupted phonological representation in a multivariate pattern analysis but not in the univariate activation analysis in beginning readers with a family risk for dyslexia. Again, because this site is also under-activated during spoken language processing in illiterates and in preliterate children compared to adult and children readers (Dehaene et al., 2010) we cannot ascertain whether its anomalous activation is a cause or a consequence of the reading disability. Indeed, it does not have to be one or the other, but could be both, as phonological awareness is known to be both a predictor of future reading (Hulme et al., 2012; Melby-Lervåg et al., 2012), and a consequence of learning to read in an alphabetical language (Morais et al., 1986).
Finally, we also observed decreased activation in the left middle frontal gyrus, as first reported in Chinese dyslexics by Siok and collaborators (Siok et al., 2008; Siok et al., 2004). Activation at this location was modulated by both reading score and language: overall, Chinese children had larger activation than French children at this site, but in both languages, poor readers also exhibited weaker activation than their controls. Thus, the effects of reading ability and language on that region seemed to be additive. The first reports of dysfunction in this area were obtained in Chinese dyslexics and, because of a lack of evidence for a comparable deficit in alphabetic languages, were interpreted as supporting a unique contribution of this region to Chinese reading (Siok et al., 2008; Siok et al., 2004). Here, however, we observed reduced activations in poor readers relative to typical readers in Chinese, but also in French. The language × reading ability interaction was not significant, and Bayesian analyses confirmed the moderate evidence for a null interaction. Those Bayesian analyses, together with the fact that we easily detected strong differences between groups (mainly related to reading ability, but also, in a few regions, to the children’s native language) support the hypothesis that our study, with 96 subjects, was sufficiently powered and that cross-linguistic differences in neural deficits in poor readers were indeed minimal or absent, in this region as well as elsewhere in the brain. While a null effect can never be firmly asserted, the present results positively indicate that atypical activation in this region is not unique to Chinese poor readers, but can be found in alphabetic readers as well and is therefore part of a universal phenotype of reading disability. Note that these results are congruent with our previous study of French children, where LMFG exhibited significant activation to words relative to other visual categories only in typical readers, not in dyslexics (Monzalvo et al., 2012).
Unlike readers of alphabetic languages who only have to acquire a small set of letters and of grapheme-phoneme correspondences, readers of Chinese must learn a few thousands of characters. To overcome this difficulty, Chinese readers may rely more on a motor memory of writing gestures as a means for memorizing the large number of characters, and as noted earlier, writing skills predict reading ability in Chinese children (Tan et al., 2005b). The greater activation of the left MFG may therefore reflect the greater reliance of Chinese children on writing, and the dysfunction of this region in Chinese poor readers may reflect an underlying deficit in memorizing writing gestures (Ziegler, 2006). This strategy is not unique to Chinese, however: novice readers in alphabetic languages also benefit from a motor memory for hand gestures when recognizing written words (Bara et al., 2004), and a recent study found that a motor representation of handwriting gestures is automatically accessed for subliminal written words in both Chinese and French adult readers (Nakamura et al., 2012). Our findings on children further suggest that the left MFG is likely to play a pivotal role in successful reading acquisition independently of the writing system.
Limitations and future direction
Several limitations of the present study need to be considered. First, our task was orthogonal to reading, and involved the mere detection of a picture of a star. This choice was made to avoid any effect due to performance itself. A task that is more difficult for some children than from others could have induced greater activation, but also greater movement in the scanner. The disadvantage is that participants were not explicitly instructed to attend to words, which could have increased the differences between poor and typical readers, and between French and Chinese readers. Note, however, that our results indicate a strong sensitivity of our procedure to reading proficiency, since reading performance was correlated with activation in all key areas of the reading circuit (VWFA, posterior STG, MFG, and pre-central (Figure 2 and Table 3)).
Although our groups were larger than in any previous study comparing Chinese and alphabetic writing systems, and were also more homogeneous in age, whole-brain analyses at corrected-level of significance failed to show significant differences between poor and typical readers. Our conclusions are therefore based primarily on analyses of ROIs that were previously established in the literature as being critical to reading proficiency. Our small effect size may be due to the definition of our groups, which included readers with poor reading performances but not all full-blown dyslexics. Indeed, the reading delay in Chinese children was less than in French children (Table 1), but even in French children who were typical French dyslexics (more than 2 years of delay in the LUM test), whole-brain analyses were poorly sensitive, which underlines the difficulties of pediatric research. Constraints on experiment duration are severe with young children, and make it difficult to collect the same extensive datasets as in adults. A lack of sensitivity could also have arisen from the greater variance between children in terms of maturation, learning experience, attention, and concentration on the task compared to adults who can remain attentive, perform more consistently, and are generally at the peak of the skill being tested (i.e., no longer learning). Furthermore, the difference between impaired readers and typical readers may be smaller in children than in adults because all children are in the process of learning and are therefore not fully competent, whereas the adult typical readers (often recruited at university) are at ceiling. Finally, as has been shown in alphabetic languages, the slope of normal reading acquisition in children depends on the transparency of the writing system. In other words, a typical 10-year-old reader would not perform equally well in English, French, German or Italian (even if we could use exactly the same test) because he or she has not yet attained the same level of reading fluency (see Ziegler et al., 2003 for a comparison between German and English). Thus, comparing languages and assessing equivalent reading delay in different writing systems, while still matching for age and education, raises complex issues and increases the difficulties of intercultural comparisons in children compared to adults.
However, since we had a particular interest in understanding the role of previously identified brain regions, an analysis based on ROIs was a reasonable choice, having the great advantage of mitigating the problem of multiple comparisons and reducing the risk of false negatives, but also allowing us to examine activation patterns at an individual level, as in the MPVA analyses. These targeted analyses were able to show the role of MFG in alphabetic languages. As usual in science, the results will need to be confirmed in future studies using larger samples to obtain more robust inferences.
In conclusion, with several convergent analyses, we revealed that the neural bases of reading in typical children and in those struggling with reading acquisition are largely similar, but partly language-specific, in French and Chinese readers. Across these very different writing systems, the cultural invention of reading relies on similar brain resources. As previously noted in an adult fMRI study (Nakamura et al., 2012), cultural variability is merely reflected in the variable emphasis that different writing systems put on phonemes, syllables and whole words, which in turn may modulate the severity of dyslexia and the degree of anomaly that can be detected at different locations along the brain’s reading circuitry.
Materials and methods
Participants
Ninety-six children participated in the current study, including 24 Chinese poor readers (mean age = 123 months, standard deviation (SD) = 10), 24 Chinese typical readers (mean age = 123 months, SD = 11), 24 French poor readers (mean age = 123 months, SD = 10) and 24 French typical readers (mean age = 123 months, SD = 11). All children reported normal hearing and corrected-to-normal vision and no history of neurological or psychiatric disorder. Nonverbal IQ, assessed by Raven’s Standard Progressive Matrices in Chinese children and Wechsler Intelligence Scale in French children, was in the normal range for all participants. The study was approved by local institutional review boards in Beijing (China) and Kremlin-Bicêtre (France), respectively. Written consent was obtained from all children and their parents.
Chinese participants
Request a detailed protocolBecause standardized tests of dyslexia are not available in Chinese, we tested a large population of 2554 primary school children in Beijing (3rd grade- 5th grade, 10- 13 years of age) to calculate the standard norms in the following tests. The first round of tests involved: (1) Chinese Character Recognition Test (CCRT) (Wang and Tao, 1993), (2) Reading Comprehension Test (RCT) (You et al., 2011), (3) Raven Progressive Matrices Test (Raven et al., 1994), and (4) Digit Cancellation Test (Mirsky et al., 1991). Children with a CCRT Z-score below −1.25 SD were identified as children with potential reading difficulties. We then invited these children and their parents to take part in a second-round of tests, which involved MRI scanning and several individually-administered tests: (5) Chinese Phonological Awareness Test (CPAT), (6) Character Reading Test (CRT) (Li et al., 2012), and (7) Rapid Automatized Naming Test. A total of 103 children with different age and reading abilities were scanned and more information about these children can be found in a previous paper (Li et al., 2018). We considered poor readers to be those whose CCRT Z-score, or CPAT Z-score, was consistently low in both rounds of tests (specifically < −1.25 SD in the first and < −1.5 in the second) and obtained 24 children. We then selected an equal number of typically developing children (>−0.5 SD in all reading-related tests) whose age, sex and non-verbal IQ matched those of the reading-struggling group (Table 1 and Supplementary file 1). By definition of the groups, the two groups were significantly different (CCRT: t = 18.03, p<0.001).
French participants
Request a detailed protocolTo match the Chinese children, we selected 24 dyslexic and 24 typical readers from two previously published French studies: 21 dyslexics and 18 typical readers from the Monzalvo et al’s study (Monzalvo et al., 2012), and 3 dyslexics and six typical readers from the population of Altarelli et al., 2013 study. Dyslexia was diagnosed in a dedicated learning disability center based on extensive behavioral testing with nationally established criteria following INSERM recommendations (clinical examination, full-scale IQ, standardized tests for working memory, meta-phonology, spelling, rapid automatic naming, etc.). At the time of fMRI scanning, we checked the children current reading level with ‘L’alouette’, a standardized reading test classically used to detect dyslexia in French speaking children (Lefavrais, 1967). It consists in reading as fast and accurately as possible a meaningless text of 265 words within 3 min (Lefavrais, 1967). All French poor readers had a delay of 20 months, or more, in this test. French and Chinese children were matched in age and sex in each group (see also Table 1 and Supplementary file 2). By definition of the groups, the two groups were significantly different (‘Alouette’: t = 14.66, p<0.001).
Due to differences in writing systems, the reading skills expected at a given age are different in alphabetic languages and Chinese. Here, French children were assessed by reading a text and Chinese by character recognition. Therefore, the direct comparison of children's performance across languages provides an indication of the distance between typical readers and the children with reading difficulties, rather than an absolute assessment of dyslexia. While reading scores were similar in French and Chinese typical readers (t < 1), French poor readers tended to have worse scores than Chinese poor readers (t = 2.45, p=0.07), leading to an interaction between language and group (F(1,92)=3.91, p=0.051).
Stimuli and task
Request a detailed protocolFor French children, the experimental procedure was identical to Monzalvo et al., 2012. The procedure was adapted to Chinese children by replacing French words and Caucasian faces by Chinese words and Asian faces. While being scanned, Chinese and French children viewed short blocks of words, faces, and houses and of a revolving checkerboard (30 frequent regular words known by young readers and 30 black and white pictures in each category) followed by a fixation cross during 10.5 s (total bloc duration 28.5 s). In each block, 10 pairs of different images belonging to the same category (200 ms presentation for the first picture/word, 200 ms inter-stimulus, 500 ms presentation for the second picture/word) were presented, separated by a 600 ms fixation period. Besides, two stars were randomly inserted in each block, 1500 ms for each star. Children were instructed to press a button with their right index finger whenever a target star appeared. This task was designed to keep their attention focused on the visual stimuli, but without any explicit reading requirement. For Chinese children and the older French children, a supplementary category (tools) was added but not included in the present analyses as this category was not presented in the original study (Monzalvo et al., 2012) and thus missing in most of the French children reanalyzed here.
In each run, there were two blocks of each visual category and only one block of checkerboard. All the blocks were presented in a random order. Chinese children performed two runs and French children performed four runs in Monzalvo et al., 2012, and only one run in Altarelli et al., 2013.
fMRI Acquisition Parameters fMRI data were acquired on two Siemens 3T scanners using a 12-radiofrequency-channel head coil and the same gradient-echo planar imaging sequence in France and China with the following parameters: 40 contiguous 3 mm isotropic axial slices, TE/TR = 30/2400 ms, flip angle = 81°, voxel size = 3 × 3 × 3 mm, matrix = 64 × 64. A high-resolution T1 weighted volume was also acquired with the following parameters: 176 axial slices, TE/TR = 4.18/2300 ms, flip angle = 9°, matrix = 256 × 256.
Prior to the scanning session, all children underwent a training session in a mock scanner. This training session aimed to help children become familiar with the MRI environment and task instructions, and to teach them to keep their head motionless during the scan.
Data pre-processing and statistical analyses
Request a detailed protocolPreprocessing and analyses of the data were done using SPM12. The French and Chinese data were processed together. The functional images were first corrected for differences in slice-acquisition time and realigned to the first volume in the scanning session. ArtRepair toolbox was used to detect and repair bad volumes (Mazaika et al., 2007). Two criteria were used to detect bad volumes: (1) 1.5% variation in the global average BOLD signal from scan to scan and (2) 0.5 frame-wise displacement, reflecting the sum of all head movements from scan to scan (calculated from realignment parameters). The damaged volumes that exceeded these criteria were replaced by linear interpolation of previous and subsequent images or by nearest-neighbor interpolation when several consecutive images were affected.
For the anatomical image, we first checked for scanner artefacts and gross anatomical abnormalities, then we manually set the origin of T1 image to the anterior commissure for each subject. We normalized each child’s anatomy to the Montreal Neurological Institute (MNI) template using the DARTEL approach to improve segmentation accuracy and local registration among participants. Functional images were co-registered to their corresponding anatomy. Then the parameters obtained during the DARTEL wrapping process were applied to the functional images which were finally smoothed using a 6 mm Gaussian kernel.
The pre-processed functional images were then submitted to a first level statistical analysis: in each subject, a general linear model was built in which a hemodynamic response function and its time derivative were convolved with block onsets for each category and the six motion parameters entered as regressors of non-interest.
Data-driven Analyses
Whole-brain analyses
Request a detailed protocolWe implemented a mixed-model analysis of variance (ANOVA) with language (French vs. Chinese) and reading skill (typical vs. poor readers) as between-subject factors, and Category (Words vs. Faces vs. Houses) as a within-subject factor. We recovered category-specific activations through the contrasts [category X > mean of the other two categories] across the whole group (N = 96). We report effects at a threshold of p<0.001 at the voxel-level and p<0.05 family wise error (FWE) corrected for multiple comparisons at the cluster-level (denoted pFWE_corr).
To deepen our analyses of the effect of reading performances, we studied the correlation between reading performance (standard scores in reading test) and the word and face activation in the 96 children. We report effects at a threshold of p<0.001 at the voxel-level and p<0.05 FWE corrected for multiple comparisons at the cluster-level.
For these analyses, we displayed in the corresponding figures the results in each of the four groups (N = 24) or in each language (N = 48) at a threshold of p<0.001 at the voxel-level, non-corrected at the cluster-level to provide the reader with the full information on the activation patterns in each group.
Mask-restricted analyses
Request a detailed protocolTo maximize the sensitivity to differences between groups, we focused our analyses on a restricted mask of voxels corresponding to the word-specific activation across all children determined by the [Words > Faces + Houses] contrast (p<0.001 voxel-level and pFWE < 0.05 cluster-level, size ~ 7497 mm3). We performed an ANOVA with language (French vs. Chinese) and reading skill (typical vs. poor readers) as between-subject factors. To provide readers with full information, we report these results at the threshold of voxel-level p<0.001, non-corrected at the cluster-level.
Literature-driven Analyses
Request a detailed protocolAs the number of comparisons at the voxel-level might decrease the sensitivity to small differences between typical and poor readers, we focused on Regions of Interest (ROI), which have been repeatedly shown in the literature to show a reduced activation in dyslexics. We first searched published meta-analyses of imaging studies reporting brain regions showing functional dysfunction in dyslexics in alphabetic languages (Linkersdörfer et al., 2012; Maisog et al., 2008; Richlan et al., 2009; Richlan et al., 2011). To create representative ROIs for these dyslexia-related regions, we collected all of the foci reported in each meta-analysis corresponding to the anatomical location under consideration (see Supplementary file 4), and averaged the reported coordinates (x, y, z respectively) to create a 6-mm-radius sphere of the averaged locus as a ROI (see Figure 3).
Due to the limited number of published neuroimaging studies of Chinese dyslexia, no meta-analysis was available to summarize the available evidence into a pooled estimate. However, atypical activation in a lateral prefrontal region within BA nine has been reported in Chinese children with reading disability (Siok et al., 2004; Siok et al., 2009) and this region was repeatedly found to be more involved in reading Chinese than alphabetic languages (Bolger et al., 2005; Nakamura et al., 2012; Tan et al., 2005a). Besides, previous studies also often reported that Chinese reading networks are more symmetrical in the ventral visual system. We thus included the foci in both left middle frontal gyrus and right occipital cortex that were reported in several meta-analyses on Chinese typical reading (Bolger et al., 2005; Tan et al., 2005a; Wu et al., 2012; Zhu et al., 2014) and created representative ROIs as above (see Supplementary file 5). In total, we obtained 10 ROIs related to dyslexia in alphabetic languages and three additional ROIs potentially related to Chinese typical reading and dyslexia (Figure 3).
We extracted the mean contrast-weighted beta values for the words vs. fixation contrast in each ROI in each child and first considered whether these values were correlated with the reading score across all participants (Figure 3), second entered these values in an ANOVA with language (Chinese vs. French) and reading skills (typical vs. poor readers) as between-subject factors. The false discovery rate (FDR) multiple-comparison method was implemented to take into account the multiple ROIs. We did the same analyses in the same ROIs for the contrasts of faces vs. fixation and houses vs. fixation. The FDR-corrected p value is denoted as pFDR_corr.
Because some of our analyses evaluated the null hypothesis of no difference between two factors, we also performed Bayesian analyses on the same data, with the same factors as the ANOVA above, using the JASP software (https://jasp-stats.org). These analyses yield a Bayes Factor (denoted BF10), which estimates the likelihood ratio of the positive (1) over the null (0) hypothesis. A BF10 of 0.20, for instance, indicates a five-times (1/0.20) greater likelihood in favor of the null hypothesis and is equivalent to the inverse notation BF01 = 5.
Individual analyses
Peak analyses
Request a detailed protocolWe further investigated whether children poor with reading might have a greater inter-individual variability in brain localization, which could putatively explain a lower activation at the group level in each voxel. We focused on those regions showing significant main effects of reading ability or language in the group activation analysis (i.e. left FFG, MFG, precentral, STS, pSTG, and SPL) and searched for active voxels (Words > fixation) in each participant in a sphere (radius = 12 mm) centered on the whole group peak coordinates in the [Words > Faces + Houses] contrast. We eliminated voxels with t-value inferior to one, and then selected the 10 strongest activated voxels within the search area. We first derived the individual center of mass of these voxels by averaging their x, y, z coordinates and calculated the distance between this center of mass and the group peak coordinates in each child. Second, we averaged the beta values measured in these voxels to obtain the maximal activation in each child. We then separately entered distance and activation into language × reading ability frequentist and Bayesian ANOVAs to investigate whether poor readers differed in peak location and activation intensity compared to typically developing children. We performed a similar analysis on the face responses in the bilateral fusiform face areas (FFA) to investigate whether poor readers had greater inter-individual variability in the location and intensity of face activations.
Multivariate Pattern Analysis (MVPA)
Request a detailed protocolOur final investigation of putative differences related to language or to reading ability was based on MVPA analysis. We focused on the regions showing significant main effects of reading ability or language in the univariate activation analysis and drew a 9 mm radius sphere centered on the averaged coordinates of foci reported in meta-analyses, and then intersected each sphere with the mask [Words > Faces + Houses] described above to obtain a fair representation of the group activations in the mask. All the voxels within the mask were included for MVPA analysis.
Secondly, within each ROI, we quantified the within-subject reproducibility of the patterns of activation by calculating in each subject the correlation coefficients between the pattern of response evoked by words relative to fixation during the first run and the pattern of response evoked by each category (words, faces, and houses) relative to fixation during the second run. The correlation coefficients were converted into Z-scores and entered into a separate ANOVA for each ROI with language (Chinese vs. French), reading ability (typical vs. poor readers), and condition (within-category correlation (i.e. words with words), vs. between-category correlation (i.e. words with faces, words with houses, faces with houses)) as factors.
We performed a similar analysis in the bilateral face fusiform areas to investigate whether poor readers showed reproducible activation patterns to faces. Bilateral face ROIs were spheres with a 9 mm radius centered on the reported peak coordinates in the face-selective activation in previous studies (left [−39, −45, −18], right [39, −45, −18]) (Downing et al., 2006). We intersected each sphere with the whole group activations [Face > Words + Houses] to obtain a mask (~1318 mm3). All the voxels within the mask were included for MVPA analysis. We then calculated the correlation coefficients between the pattern of response evoked by faces relative to fixation during the first run and the pattern of response evoked by each category (faces, houses, and words) relative to fixation during the second run in each subject.
Note that six French typical readers and three French poor readers from Altarelli et al., 2013 completed only one run of the visual task, so that they were not included in this MVPA analysis. For those children who had four runs, we used their first two runs to calculate the correlation coefficients between runs. The FDR multiple-comparison method was again used as a correction for the multiple ROIs tested.
Anterior-to-Posterior organization of the ventral temporal cortex
Request a detailed protocolIn the following analyses, we focused on the ventral occipito-temporal region, because it is the site of one of the major changes related to reading: the emergence of a specific response to words in literates and also the site of the most reproducible hypoactivation in dyslexics compared to typical readers in alphabetic writing (Richlan et al., 2011). Because activations to words, faces, and houses display a gradient both along the anterior-to-posterior (i.e. ‘y’) axis and lateral-to-medial (i.e. ‘x’) axis (Baker et al., 2007; Scherf et al., 2007), we averaged the activity in successive spheres along the anterior-posterior axis at the ‘x’ privileged position for each visual category and compared typical and poor readers in Chinese and French participants. The spheres had a 6-mm-radius and were regularly spaced with the center positioned at y = −73, −64, −55, −46, −37, −28 respectively, x and z positions being kept constant (z = −16). The x position was based on the peak of the category-specific activation (category > others) in all participants, i.e. for Words: x = ±48; Faces: x = ±39; and Houses: x = ±30 (see Table 2).
We performed a separate ANOVA for each visual category with language and reading ability as between-subject factors, ROI (6 y-axis positions) and Hemisphere (left and right) as within-subject factors. In each ANOVA, we corrected for multiple comparisons using the FDR method.
Data availability
The processed data to generate figures in this manuscript will be shared in the Open Science Framework (Identifier: https://doi.org/10.17605/OSF.IO/C4AXG).
-
Open Science FrameworkUniversal brain mechanisms of reading acquisition: an fMRI study in normal and dyslexic French and Chinese children.https://doi.org/10.17605/OSF.IO/C4AXG
References
-
A lateral-to-mesial organization of human ventral visual cortex at birthBrain Structure and Function 223:3107–3119.https://doi.org/10.1007/s00429-018-1676-3
-
Neural correlates of orthographic and phonological consistency effects in childrenHuman Brain Mapping 29:1416–1429.https://doi.org/10.1002/hbm.20476
-
Anatomical connections of the visual word form areaJournal of Neuroscience 34:15402–15414.https://doi.org/10.1523/JNEUROSCI.4918-13.2014
-
The brain adapts to orthography with experience: evidence from english and chineseDevelopmental Science 18:12245.https://doi.org/10.1111/desc.12245
-
Writing system variation and its consequences for reading and dyslexiaScientific Studies of Reading 22:101–116.https://doi.org/10.1080/10888438.2017.1379082
-
Orienting attention in time activates left intraparietal sulcus for both perceptual and motor task goalsJournal of Cognitive Neuroscience 23:3318–3330.https://doi.org/10.1162/jocn_a_00030
-
Illiterate to literate: behavioural and cerebral changes induced by reading acquisitionNature Reviews Neuroscience 16:234–244.https://doi.org/10.1038/nrn3924
-
Domain specificity in visual cortexCerebral Cortex 16:1453–1461.https://doi.org/10.1093/cercor/bhj086
-
Atypical perceptual processing of faces in developmental dyslexiaBrain and Language 173:41–51.https://doi.org/10.1016/j.bandl.2017.06.004
-
A Cross-Linguistic fMRI study of spectral and temporal cues underlying phonological processingJournal of Cognitive Neuroscience 14:1076–1087.https://doi.org/10.1162/089892902320474526
-
Language-universal sensory deficits in developmental dyslexia: english, Spanish, and chineseJournal of Cognitive Neuroscience 23:325–337.https://doi.org/10.1162/jocn.2010.21453
-
Origins of the specialization for letters and numbers in ventral occipitotemporal cortexTrends in Cognitive Sciences 19:374–382.https://doi.org/10.1016/j.tics.2015.05.006
-
Learning to read recycles visual cortical networks without destructionScience Advances 5:eaax0262.https://doi.org/10.1126/sciadv.aax0262
-
Phoneme awareness is a better predictor of early reading skill than onset-rime awarenessJournal of Experimental Child Psychology 82:2–28.https://doi.org/10.1006/jecp.2002.2670
-
Developing, mature, and unique functions of the child's brain in reading and mathematicsDevelopmental Cognitive Neuroscience 39:100684.https://doi.org/10.1016/j.dcn.2019.100684
-
Predictors of developmental dyslexia in european orthographies with varying complexityJournal of Child Psychology and Psychiatry 54:686–694.https://doi.org/10.1111/jcpp.12029
-
ReportTest De L’Alouette, Manuel Et TestsÉditions du Centre de Psychologie Appliquée (ECPA).
-
Chinese children's character recognition: Visuo-orthographic, phonological processing and morphological skillsJournal of Research in Reading 35:287–307.https://doi.org/10.1111/j.1467-9817.2010.01460.x
-
Chinese dyslexics show neural differences in morphological processingDevelopmental Cognitive Neuroscience 6:40–50.https://doi.org/10.1016/j.dcn.2013.06.004
-
A meta-analysis of functional neuroimaging studies of dyslexiaAnnals of the New York Academy of Sciences 1145:237–259.https://doi.org/10.1196/annals.1416.024
-
Phonological skills and their role in learning to read: a meta-analytic reviewPsychological Bulletin 138:322–352.https://doi.org/10.1037/a0026744
-
Analysis of the elements of attention: a neuropsychological approachNeuropsychology Review 2:109–145.https://doi.org/10.1007/BF01109051
-
How reading acquisition changes children's spoken language networkBrain and Language 127:356–365.https://doi.org/10.1016/j.bandl.2013.10.009
-
Developmental differences for word processing in the ventral streamBrain and Language 125:134–145.https://doi.org/10.1016/j.bandl.2012.04.003
-
The universal grammar of readingScientific Studies of Reading 7:3–24.https://doi.org/10.1207/S1532799XSSR0701_02
-
Cerebral organization of component processes in readingBrain 119 ( Pt 4:1221–1238.https://doi.org/10.1093/brain/119.4.1221
-
What phonological deficit?Quarterly Journal of Experimental Psychology 61:129–141.https://doi.org/10.1080/17470210701508822
-
BookManual for Raven's Progressive Matrices and Vocabulary ScalesOxford Psychologists Press.
-
Connectivity precedes function in the development of the visual word form areaNature Neuroscience 19:1250–1255.https://doi.org/10.1038/nn.4354
-
Foundation literacy acquisition in european orthographiesBritish Journal of Psychology 94:143–174.https://doi.org/10.1348/000712603321661859
-
Properties of school chinese: implications for learning to readChild Development 74:27–47.https://doi.org/10.1111/1467-8624.00519
-
Privileged functional connectivity between the visual word form area and the language systemThe Journal of Neuroscience 37:5288–5297.https://doi.org/10.1523/JNEUROSCI.0138-17.2017
-
Effects of literacy in early visual and occipitotemporal Areas of chinese and french readersJournal of Cognitive Neuroscience 26:459–475.https://doi.org/10.1162/jocn_a_00499
-
Development of neural mechanisms for readingNature Neuroscience 6:767–773.https://doi.org/10.1038/nn1065
-
BookChinese Character Recognition Test Battery and Assessment Scale for Primary School ChildrenShanghai Education Press.
-
The face-processing network is resilient to focal resection of human visual cortexThe Journal of Neuroscience 36:8425–8440.https://doi.org/10.1523/JNEUROSCI.4509-15.2016
-
Developmental dyslexia in different languages: language-specific or universal?Journal of Experimental Child Psychology 86:169–193.https://doi.org/10.1016/S0022-0965(03)00139-5
Decision letter
-
Andrea E MartinReviewing Editor; Max Planck Institute for Psycholinguistics, Netherlands
-
Floris P de LangeSenior Editor; Radboud University, Netherlands
-
Charles PerfettiReviewer; University of Pittsburgh, United States
In the interests of transparency, eLife publishes the most substantive revision requests and the accompanying author responses.
Acceptance summary:
Your contribution is important because it shows common neural correlates for word recognition in typical developing and dyslexic children in French and Chinese. While the conclusion is based on "null effects" (i.e., Chinese and French dyslexics are similar), a set of exhaustive contemporary approaches (whole-brain MRI, ROIs, MVPA, Bayesian sensitivity analysis) gather consistent support the findings. The results provide additional insight into the ongoing debate about the universality of the neural correlates of dyslexia across cultures and writing systems. The study fills an important gap, going beyond the existing meta-analytic literature which claimed that the neural basis of dyslexia is similar across alphabetic writing systems despite difference in orthographic depths in the visual word form area (VWFA). The current work extends this claim by showing experimentally that even in completely different writing systems such as Chinese and French, the neural correlates are shared. The authors took incredible care to match across the French and Chinese languages in terms of the various parameters (scanner characteristics and demographics).
Decision letter after peer review:
Thank you for submitting your article "Universal brain mechanisms of reading acquisition: an fMRI study in normal and dyslexic French and Chinese children" for consideration by eLife. Your article has been reviewed by three peer reviewers, and the evaluation has been overseen by a Reviewing Editor and Floris de Lange as the Senior Editor. The following individual involved in review of your submission has agreed to reveal their identity: Charles Perfetti (Reviewer #3).
The reviewers have discussed the reviews with one another and the Reviewing Editor has drafted this decision to help you prepare a revised submission.
Here I have summarised the reviews to help your revision. I intend that you will use my letter as a guide as to what you should address in your revision, and refer from my points in the editorial letter to the 3 reviews as reference, such that you can read about the points I mention in the letter in more depth in the reviews. I will mainly draw your attention to the concerns from reviewers 1, 2, and 3 regarding the interpretations of your data, and the strength of the support that your statistical analyses offer to those conclusions.
The editorial concerns below must be addressed in any revision. I include the full reviews because they are highly constructive and contain feedback that can help you improve the manuscript; I encourage you to incorporate as much of the reviewer's feedback as possible. Below I will summarise the changes necessary for the revision:
1) Bayesian Sensitivity Analysis – Like Reviewer 1, I am particularly worried about how weak the Bayes factor evidence is; at the moment it is weak and the prior that was chosen is not very informed or reflective of either direction of assumptions. Thus, a sensitivity analysis where a variety of priors, representing different assumptions, should be carried out, including an informed prior based on previous literature and which stems from the contrasts suggested by Reviewers 2 and 3. For more about sensitivity analysis, please see Oakley JE, O'Hagan A. Probabilistic sensitivity analysis of complex models: a Bayesian approach. Journal of the Royal Statistical Society: Series B (Statistical Methodology). 2004 Aug;66(3):751-69.
2) Whole Brain Analysis – From Reviewer 2, I found it important to address their point that "the "universality of dyslexia" argument (assumes) there is a neural signature of dyslexia. Yet, at the whole brain level, there is no significant difference between dyslexia and controls. The rest of the analyses heavily relies on the regions identified in the ROI based approach. This should be thoroughly discussed and addressed as an important limitation."
3) By Language Analysis – Both Reviewer 2 and 3 would like to see analyses broken down by writing system. From Reviewer 2 "Since authors do show significant correlation with reading abilities in several areas, there appears to be room to explore differences in writing systems. Can the authors do further analyses and potentially build on these results (also), and if not, justify why?"
4) Discussion of Power – All reviewers were concerned about the experiment and contrasts being properly powered. Though N=96, there were only 24 per group for each of the four groups. As Reviewer 2 asks, on what basis is this sufficiently powered? On a related point, given the specificity of the claims wrt a lack of language interaction, Reviewer 3 points out that planned comparisons focused on language (if power is sufficient), in both ANOVAs and Bayes factors, are needed to support this line of argumentation.
5) Discussion of whether and how the population was identified, and reading skill instead of dyslexia – Reviewer 3 had a very helpful and constructive point: "the value of the paper is adding incremental evidence for a shared pattern of activation areas that distinguish low levels of reading skill from more typical, higher levels." But thinks that framing this result in terms of *reading skill* is more warranted than in terms of dyslexia, given the way Chinese dyslexics were identified from the -1.25SD of normal classrooms.
6) Interpretation of the results in relation to existing literature and LMFG – Lastly, the paper's main conclusion is that these results counter existing research showing brain activation patterns implicating the involvement of LMFG in logographic more than alphabetic reading. In fact, as Reviewer 3 points out, the results reported here actually align with existing studies reporting differences between logographic and alphabetic systems, and this should be highlighted. Or to put it another way, now that the LMFG is found to be involved in alphabetic reading, is it now elevated to part of the universal reading network that has previously excluded it?
Please find the full reviews attached for your reference:
Reviewer #1:
1) Reasoning stems from a fallacy – namely that the lack of evidence for a difference is evidence for a lack of difference. It is not.
2) It is problematic that the authors interpret BOLD signal/ brain activity as somehow equivalent to a mechanism or computation, without any attempt to constrain or define that computation.
3) The study of dyslexia is important, but here it clouds the inference waters about the neural substrate of reading – it cannot separate what the neural "causes" vs the neural "effects" of dyslexia are.
4) The experiment is designed to rely on interactions for scientific inference, yet appears only to be powered to detect main effects – though no power analysis is presented. The Bayes factors reported are weak evidence for the null, and worse are highly dependent on the prior. Using a more informed prior (for example, from the existing literature cited here, such as 9,10,32 and many others) will decrease evidence for the null dramatically (and killing this Bayes factor further, I suspect; moreover, it is only Bayes factors > 10 that are regarded as strong evidence for the null). That has not been done here. Ideally one needs to show a comparable range of Bayes factors with multiple, varied, or highly justified priors. That has not been done here. Mixed models should be used here over anova in order to take into account item variance simultaneously with subject variance.
5) It is not clear how this work is an advance over existing findings, see in particular citations 9, 10, and 32.
Reviewer #2:
Summary
In the current manuscript, authors show convergent results of common neural correlates of word recognition in typical developing and dyslexic children in French and Chinese. While their conclusion is based on "null effects" (i.e., Chinese and French dyslexics are similar), they use a set of exhaustive approaches (whole-brain MRI, ROIs, MVPA) and Bayesian inferences to support their findings. The results provide additional insight into the ongoing debate about the universality of the neural correlates of dyslexia across cultures and writing systems. Their results also shed light on the neural correlates of pediatric dyslexia in children showing similar characteristics to adults (though they do not compare with adults directly in their study). The study fills an important gap that goes beyond prior meta-analytic literature that showed that the neural basis of dyslexia is similar across alphabetic writing systems that are different across orthographic depths in the visual word form area (VWFA); they show using empirical data (and not meta-analyses) that even completely different writing systems such as Chinese and French are similar. The authors make pain-staking and impressive efforts to match across French and Chinese in terms of the various parameters (scanner characteristics and demographic). The results are generally convincing and manuscript well written. Detailed major comments are listed below.
1) Premise of the study: The major issue for this reviewer was that the "universality of dyslexia" argument lies on the fact that there is some neural signature of dyslexia. Yet, at the whole brain level, there is no significant difference between dyslexia and controls. The rest of the analyses heavily relies on the regions identified in the ROI based approach. This should be thoroughly discussed and addressed as an important limitation.
2) Correlation analyses with reading: Authors then move into correlation with reading ability but do not pursue further than showing correlations in the entire sample. They do not break the sample up into writing systems. Since authors do show significant correlation with reading abilities in several areas, there appears to be room to explore differences in writing systems. Can the authors do further analyses and potentially build on these results (also), and if not, justify why?
3) Power & sample size: Discussion paragraph two: Authors mention that they were sufficiently powered because they had N=96. This is still 24 per group for each of the four groups. On what basis was it sufficiently powered? Especially in reporting these "null" results, it's important to provide evidence and perhaps reiterate this evidence when mentioning their "sufficient power" in the Discussion.
4) MVPA: In general MVPA was confusing in many ways and the reviewer recommends a major rewrite.
a) Motivation: Introduction paragraph seven: The phrase about MVPA where they motivate this analysis should probably be a separate sentence as the question being asked is very different than the first part of the question. Further, this justification comes out of nowhere and should be elaborated. The authors claim that they wish to test dispersion and intensity using classical analyses, and also using MVPA. They also do not mention in the Introduction, but in the Materials and methods, mention that they are examining stability (reproducibility from Run 1 to Run 2). Please clarify in the Introduction.
b) Materials and methods subsection “Multivariate Pattern Analysis (MVPA”: This was very confusing all around. For example… (1) What values are fed into as features? Time-series correlation? Or betas for each trial? They mention "pattern of response evoked by words relative to fixation" but this is unclear. (2) They claim they are doing multi-"voxel" pattern analyses (MVPA), but they also seem to refer to each ROI having one correlation coefficient. (3) It was not clear why this is called MVPA. What was the analytical technique used, ANOVA? Correlation-based MVPA (what is this exactly?)? There are probably many more, but it's hard to judge the quality of this analyses based on their lack of details and hard to ask further questions at this point.
5) Population: The test scores in Chinese was normed based on their own data of 2554 individuals. It is not clear whether this is appropriate (though size is large). How were these children recruited; is this a nationally representative population? In addition, how French children with dyslexia was identified (and criteria) was not mentioned in this paper and the authors simply refer to another paper. Since this is quite important, it will be nice to have this detailed here in this manuscript as well.
6) Reading ability: (1) How did they combine reading skills across writing systems to combine all in a correlation analysis, especially when reading measures that fed into the dyslexia criteria were different, there is no standardized measures in Chinese and just that the measures are so different across the board? (2) This also means that although unreported, the authors can compare reading scores (as well as others such as IQ) across writing systems, and report in Table 1? Please elaborate.
Reviewer #3:
General Assessment. An interesting paper with rich results, an fMRI study comparing previously reported results of French children with new data on Chinse children. The study is characterized by careful sample matching, comparable instrumentation across sites, and multiple sophisticated analyses. The paper provides a context of previous brain imaging studies that compare dyslexia across Chinese and alphabetic writing and places its results in this context. Its message is not novel-that the brain shows a reading network that is shared across languages and writing systems-but because the research has suggested different conclusions on whether reading Chinese and reading alphabetic writing involve different brain areas, these results could add weight to two aspects of the issue. Do alphabetic and Chinese reading engage the same brain areas to the same extent? Do Chinese and alphabetic dyslexics show similar activation patterns across brain areas? The results answer the first question by "yes" and no", but the authors emphasize more the "yes" part. They answer the second part "Yes" Some revision and perhaps reframing of the contribution of the paper to the two issues engaged-language/writing system factors and dyslexia-would improve the paper.
Major comments on context and Interpretation.
1) One suggestion is to broaden the context for the question and to refine its form. Arguments that the procedures of reading are universal have been made on the basis of behavioral studies and the logic of writing systems (e.g. Perfetti, 2003; other observations by Mattingly, 1972) If graphs-the written symbols of language-encode linguistic forms-then the brain, to decode writing, must connect visual areas to language areas. Beyond that fundamental universal, the factors of mapping of phonology and morphology, and perhaps graphic complexity, could influence the details. Whether there is a shared reading circuit across languages has not been the main question. Rather, the question has been, given a shared or universal circuit, what differences might one expect based on the features of the writing system and the language?
2) Although the authors' argument is strong on the universalist side, the data show convergence with the results of Siok, Tan et al. in the greater LFMG and SPL activation for Chinese. Puzzling, as the authors note, is the difference in pSTG, also higher for Chinese, contrary to other results. Given these language differences, what does the failure to find a language x dyslexia interaction mean? Both normal and dyslexics in Chinese show the Chinese pattern (more activation in MFG and SPL) and both controls and dyslexics in French show the French pattern (Less MFG, Less SPL, and, surprisingly, pSTG), as shown in Figure 4.
3) With these language differences, an appropriate interpretative framing would be to emphasize writing system/language differences while finding no dyslexia differences across languages. The paper reports and comments on the differences findings but does not highlight them. The Abstract and conclusion should be written to reflect these findings.
4) The LFMG difference might reflect the role of writing experience in Chinese, which may make frontal brain areas near the motor cortex, especially Exner's area, more active in reading. The authors cite the comparison of French and Chinese by Nakamura et al., 2012), but the evidence from Cao & Perfetti, 2017, is also relevant because it reports that the overlap of writing and reading brain areas is greater in Chinese than English and also in learners of Chinese as L2 who had learned through character writing compared with learners who had learned through pinyin.
5) More generally on the issue of context and interpretation, the authors should consider the relevance of Perfetti, Cao, and Booth (2013), which provides a more nuanced framing of the universal/specialization issue and summarizes results from children's studies carried out by Booth et al. to identify brain regions that show developmental changes in English and Chinese, highlighting commonalities along with divergence in STG and LMFG.
6) Any comments on why MFG shows a much stronger dyslexia effect for French (strong) than Chinese children (marginal or null), according to the Bayes factor data shown in Table 4? The language difference shows more MFG involvement among Chinese than French. The reporting of the results in subsection “ROI-Level analysis” refers only to the Bayesian evidence against a language x dyslexia interaction, ignoring the post-hoc analyses (which given the goal of the study, could have been planned comparisons), which also show strong F10 support for an MFG difference between French and Chinese dyslexia but marginally "moderate" support for such a difference between French and Chinese controls.
7) Concerning the MFG, a fair summary of the results is that MFG activation was observed more for Chinese than French and more for controls than dyslexics. It would be appropriate to point out that this confirms conclusions from other research. Thus, the conclusions from Siok, Tan, et al. that the LMFG is more involved in reading Chinese than in alphabetic reading is also found here, as is their conclusion of its under-activation in Chinese dyslexia. What may be new in the present paper is evidence that under-activation of the MFG may also characterize reading French. This last conclusion is not based on the present results, however, but on results from the previously reported results for the French children. The authors should point out whether the original publication of the French results reached this conclusion or whether it emerged only in the comparisons of the present paper. The Monzalvo et al. abstract emphasizes VWFA factors in dyslexia, but not MFG.
8) The Interpretation of the VWFA as a universal site should take into account that both faces and houses showed lower activation for dyslexics along the anterior-posterior axis. This affects how to interpret the lower activation for dyslexics to words in the VWFA.
9) The conclusion of the paper needs to be revised to more accurately reflect the results of the study. This should include the differences that were found, not just those that were not found and a more contextualized grounding for the conclusion that "cultural variability is merely reflected in the variable emphasis that different writing system put on (linguistic units)". This is not new (which is OK; it's worth emphasizing) and has been pointed in other papers in which the results have shown language differences.
10) The authors acknowledge a possible task issue but an additional consideration seems relevant. The task involved reading only incidentally. We know that for a skilled reader, implicit reading engages the network non-attentively (automatically), so the task itself is ok, as are naming and semantic judgement used in other studies. The problem is that, because it is the only task used, we get a limited picture of the reading network, perhaps especially for the dyslexics who cannot be assumed to engage the reading network automatically. It is worth noting that the research reporting Chinese-alphabetic differences did not use passive reading tasks but naming, rhyming, and meaning decision tasks. I don't know whether this is relevant for the differences, but It would strengthen the contribution of the paper if it could suggest a reason for different conclusions arising in the literature or address the task issue in that context.
11) I also wonder about the degree of reading difficulties shown in the sample. The Chinese children were classified as dyslexic by showing a -1.25 SD score on the CCRT character recognition test. We need to know what the task was for the CCRT. Performance on a recognition task reflects learning through experience. A reading problem may reflect a poor learning outcome from experience but should not reflect mere experience. Also -1.25 SD seems to identify a rather large percentage of children as dyslexic, compared with other studies, which may make it more likely that reading experience is involved to some extent. This is not a problem for an approach that treats reading as a graded skill, which the authors do in some of their analyses. Making reading skill rather than dyslexia the framing of the paper could help with this.
12) The ANOVA reported in subsection “Data-driven Analyses” does not provide a main effect of category. Instead it tests one category against the mean of the other two. And this leads to multiple tests that are appropriate with a significant main effect. A standard ANOVA with 3 levels of the category variable is needed.
13) Another statistical issue is the use of separate planned comparisons within each language when no interaction involving language is reported. In subsection “Anterior-to-Posterior gradient in the visual cortex”, this comparison, without a report of a significant language interaction, is shown for the anterior-posterior gradient analysis allowing the suggestion that the only dyslexia difference for both languages was near the VFWA. In other analyses, the authors report a lack of a language interaction and no further within-language comparisons. Given the focus of the research on language differences, planned comparisons between languages is quite reasonable to do in all analysis but the authors seem to have done this only selectively. Some explanation is needed.
14) The sentences "Dyslexics might have more dispersed activations without focal peaks that would also create weaker responses at the group-level, but a reproducible pattern of activations. In that case, the within-subject reproducibility of multivariate activation patterns should not differ between normal-readers and dyslexics" from subsection “MVPA of the activations to words”. The logic here escapes me. If Ds have more dispersed activation without focal peaks their reproducibility of multivariate patterns could still differ from controls. More voxels but not the same voxels, reflecting intrinsic variability. I think the authors are right that the individual peaks analysis means the between-group variability is not due to averaging individuals. But the differences in these individual MVPA analyses mean that variability is higher with dyslexics. The reason for this should be considered.
15) Why on the MVPA are the data analyzed transformations of correlations of first run with second run? The idea is reproducibility, but the second run is a second run that provides a familiarization component that may affect controls and dyslexics differently.
16) Another point to be explained is why some analyses use fixation as the comparison (MVPA) whereas most use the contrast of one condition vs the other two.
17) Figure 2 shows areas with correlations with reading scores. Either the caption or the images should make clear the directions of the correlations and make clear this is for all subjects. Given the goal of the study, showing these separately for French and Chinese is better, even though it reduces the power.
18) Figure 3A is a bit puzzling. It shows peaks from literature on Chinese and alphabetic studies, but the only peak in LH is at MFG for Chinese. Certainly, studies of Chinese find peaks in other LH regions. Do the peaks have the same definition across studies and meta-analyses? One peak for each region of interest or region of a hypothesized reading network probably would show a different picture. As it stands the visual here contradicts universality and may be misleading. As for 3B, again it would be useful to indicate the direction of the correlations or state they are all positive if that is the case.
19) The authors say that the LFMG dysfunction "was previously claimed to be specific to Chinese dyslexia". This is more accurately described as a discovery of an LMFG deficit in Chinese dyslexics in the absence of any existing evidence of a comparable deficit having been reported in in studies of alphabetic languages. The previously published study of French children should have added the discovery that LFMG dysfunction can characterize alphabetic reading as well.
[Editors' note: further revisions were suggested prior to acceptance, as described below.]
Thank you for resubmitting your article "A universal reading network and its modulation by script type and reading abilities in French and Chinese children" for consideration by eLife. Your revised article has been reviewed by one peer reviewer, and the evaluation has been overseen by Andrea Martin as Reviewing Editor and Floris de Lange as the Senior Editor. The following individual involved in review of your submission has agreed to reveal their identity: Charles Perfetti (Reviewer #3).
The reviewers have discussed the reviews with one another and the Reviewing Editor has drafted this decision to help you prepare a revised submission.
Summary:
The reviewer and myself were encouraged by the Bayesian Sensitivity Analysis performed. However, the reviewer has 4 concise but excellent suggestions that I am confident you can address.
Revisions for this paper:
Please address points 1-4 by the reviewer; each query is concrete, concise, addressable, and clearly laid out. Please make these revisions and clearly indicated in the revised manuscript text as well as describe in a response letter how you have handled each point.
Reviewer #3:
The authors made several significant improvements in the paper and were responsive to reviewers' comments. I think the strength of the paper lies in the developmental results for a Chinese sample to compare with previously published results of a French sample, along with a better portrayal of its contribution in the context of other research. The French-Chinese comparison itself remains a question for me, but the Abstract, Discussion, and conclusion are more in line with the results.
I suggest the authors consider a few additional revisions.
1) The authors' reframing from dyslexia to reading skill is important, given the likely differences in reading skill between the Chinese and French samples. But the use of the word "impaired" to refer to lower skill readers throughout the ms does not follow through on this reframing. Dyslexia is an impairment. It might be that the French sample includes mainly or exclusively impaired readers and the authors may prefer that term to be consistent with their original publication of the French data. If so, it might be enough to have a sentence or two that explicitly address the terminology shifts or point out the possible French-Chinese differences in the low skill group. If effective reading experience is what is involved especially in Chinese-i.e. the rate at which a child has benefitted from learning opportunities, including reading practice-this might be important to mention. This would be consistent with the demands of incremental character learning and practice. It may be interesting that the distinction between impairment and delay may not matter given the authors' interpretation of the results of their comparisons.
2) If I understand correctly, the important conclusion that LMFG is part of the universal network is indirect, dependent on adding a Chinese sample to the original French sample for comparison. The French result did not show significant under-activation for impaired readers. Instead, only when the Chinese sample was included did this occur. I think this point should be made explicit. Maybe I misunderstand something, but it seems that only by adding a Chinese sample that we see LMFG problem in French children with reading problems.
3) The greater activation for Chinese than French in the posterior superior temporal gyrus/sulcus continues to be a bit of puzzle. The explanation on 19-20 does not seem very plausible. Children's use of PinYin might help explain why the PSTG in Chinese is not less activated than in French, contrary to expectations based on decoding accounts of the PSTG. However, I don't see how it can explain the reversal that was observed for children who have been reading characters for 3-5 years. Rather than this extended speculation, it might be better to just mention it as one of several possibilities, including the possibility of more complex syllable decoding because of Chinese tones and connections to spoken language functions, or a need to suppress the activation of the syllable associated with the character (because it leads to competition for character identification), or others that also are speculative. If the evidence for a letter-phoneme function of this area comes mainly from explicit reading tasks, maybe this area is doing something else related to the target task during incidental reading.
4) The authors use "script" where they actually refer to writing system. I strongly suggest a change in the title and elsewhere to writing system. The difference between French and Chinese is one of writing system and this inevitably brings along a difference in script. The difference between French Cursive and French Gothic and Chinese Traditional and Chinese Simplified are differences of script.
https://doi.org/10.7554/eLife.54591.sa1Author response
The editorial concerns below must be addressed in any revision. I include the full reviews because they are highly constructive and contain feedback that can help you improve the manuscript; I encourage you to incorporate as much of the reviewer's feedback as possible. Below I will summarise the changes necessary for the revision:
1) Bayesian Sensitivity Analysis – Like Reviewer 1, I am particularly worried about how weak the Bayes factor evidence is; at the moment it is weak and the prior that was chosen is not very informed or reflective of either direction of assumptions. Thus, a sensitivity analysis where a variety of priors, representing different assumptions, should be carried out, including an informed prior based on previous literature and which stems from the contrasts suggested by Reviewers 2 and 3. For more about sensitivity analysis, please see Oakley JE, O'Hagan A. Probabilistic sensitivity analysis of complex models: a Bayesian approach. Journal of the Royal Statistical Society: Series B (Statistical Methodology). 2004 Aug;66(3):751-69.
Following editor and the reviewers’ suggestions, we have conducted sensitivity analyses in the post-hoc pairwise comparisons (as supplementary material, see below). We now report the results in each language as well as between languages (please see main text and figures below).
“We used Bayesian ANOVAs on the Word activation to assess the likelihood of the null hypothesis H0 over H1 in the case of the interaction Language × Reading skills. As also explained in the Materials and methods, the Bayes factor (BF10) is the ratio of the amount of evidence for H1 above H0 (BF01 for H0 above H1). Evidence for H1 against H0 is generally considered as moderate for BF10 ≥ 3, and strong for BF10 ≥ 10. Bayesian analyses provided similar conclusions to frequentist analyses for the main effect of Language and Reading skills (Table 4). The BF10 for an effect of language was 80.21 in the MFG, 7.05 in the SPL and 9.91 in the pSTG. The likelihood of an effect of Reading skill in the FFG was BF10 = 1528.97 higher than that of the null hypothesis; the same BF10 was 37.51 in the STS, 9.95 in the MFG and 5.63 in the PCG respectively. It was also the case in the post-hoc analyses within each language (Table 4). Notably, there was strong evidence of a reading skills effect in the fusiform gyrus (FFG) and moderate evidence in the STS and PCG in both languages. In the pSTG and also paradoxically in the MFG described as a dyslexic marker in Chinese, the evidence of a reading skill effect was strong in French (BF10 = 23.74 and 29.35 for each site respectively) but absent or weak in Chinese children (BF10 < 1 and 1.62). Sensitivity analysis revealed that the Bayes factor stayed about the same for a wide range of prior specifications (Cauchy prior width: 01.5) in the FFG, STS, PCG and MFG for the comparisons between controls and impaired readers in each language except for the pSTG in Chinese children (see Author response image 1)”.
Turning now to the effect of language in each reading group, differences between languages were mainly observed between the groups of impaired readers: Chinese children showed larger activations than French in the MFG (BF10 = 309.65) and pSTG (BF10 = 74.01) although the same tendency was present in controls. The above Bayes factor stayed the same for a wide range of prior specifications (Cauchy prior width: 0-1.5) in the sensitivity analysis (see Author response image 2). However, there was no evidence for a significant interaction Language × Reading skill in all these ROIs. On the contrary, the null effect was supported by moderate evidence in the FFG, MFG, STS, PCG, and SPL (respectively BF10 = 0.31, 0.31, 0.31, 0.27 and 0.32, i.e. the likelihood of the null hypothesis BF01 = 3.23, 3.23, 3.23, 3.70 and 3.13); there was no evidence in either direction in pSTG (BF10 = 1.05) (Table 4).

Sensitivity analysis with a variety of priors in the post-hoc pairwise comparisons between controls and impaired-readers in each language (Left: Chinese children; Right: French children).

Sensitivity analysis with a variety of priors in the post-hoc pairwise comparisons between Chinese and French children separately in control and impaired-reading children (Left: Controls; Right: impaired-readers).
We would also like to point out that, contrary to the reviewer’s worries, we have strong positive evidence, using both Bayesian statistics and frequency statistics, in the main ROIs described in the literature as sensitive to reading performance or proposed to be Chinese-specific. In particular, 2 sites, the FFG and the MFG which have been hypothesized to be crucial for reading acquisition respectively in alphabetic (FFG) and Chinese (MFG) languages are significantly activated and showed significant differences between impaired and typical children in each language. Activation in the VWFA has been emphasized as the main marker of reading skills in alphabetical languages but there is disagreement on its role in other writing systems, notably Chinese because of the large differences in visual shapes, phonological content and relation with handwriting between Roman letters and Chinese characters.
The likelihood of an effect of Reading skill in the VWFA region was strong in both French and Chinese children: BF10 = 93.16/13.58 in agreement with Krafnik et al’s (2016) conclusion of script invariance in that region in first graders. Likewise, the MFG (which has been described as the main marker of reading difficulties in Chinese, but not in alphabetic languages), exhibits a strong effect in French children but a weak one in Chinese: BF10 = 29.35/1.62, i.e. the converse of what was predicted by the culture-specific model, thus emphasizing a more universal role of this region.
Finally, the likelihood of a significant interaction between language and reading ability as suggested by the past literature on these regions, was not at all supported. It resulted in a BF10=0.31 in both cases (thus a moderate likelihood for the absence of interaction).
In summary, our results provide positive (and relatively strong) evidence that the VWFA and the MFG are involved in both alphabetic and logographic writing systems even if, indeed, there is modulation of the activation in some of these areas by different factors including the writing system and reading competence, a point which we now discuss in more detail.
2) Whole Brain Analysis – From Reviewer 2, I found it important to address their point that "the "universality of dyslexia" argument (assumes) there is a neural signature of dyslexia. Yet, at the whole brain level, there is no significant difference between dyslexia and controls. The rest of the analyses heavily relies on the regions identified in the ROI based approach. This should be thoroughly discussed and addressed as an important limitation."
Although our groups were larger than in any previous study comparing Chinese and alphabetic writing systems, and were also more homogeneous in age, whole-brain analyses at corrected-level of significance failed to show significant differences between reading-impaired children and typical children. However, the most important results in our study do not arise from a binary group comparison but from the correlation of reading scores with brain activity. When we analyzed a more continuous variable (the reading scores), we recovered significant correlations in classical regions of the reading circuit in Chinese and French participants at the whole brain level (As suggested by reviewer 3, we now present results in each group).
In many brain imaging studies in children, comparisons between groups are often not significant at the whole brain level (see for example Pleisch et al., (2019) testing 18 vs 20 first-grade German children with poor and typical reading skills, and at varying familial risk for developmental dyslexia). This is probably due to the higher variance between children within each of the groups in a variety of factors: brain maturation, learning experience, attention and focus in the task, etc. All of these factors are different in adults who can remain attentive, perform more consistently, and are generally at the peak of the skill being tested (i.e., no longer learning). Furthermore, the difference between typical readers and dyslexics may be smaller in children than in adults because even the normal children are still in the process of learning and are therefore not fully competent, whereas the adult typical readers (often recruited at university) are at ceiling.
The small effect size in our study may be partly due to the definition of our groups, which included reading-impaired children that were not full-blown dyslexics. Indeed, the reading delay in Chinese children was less than in French children (Table 1), but even in French children who were typical French dyslexics (more than 2 years of delay in the LUM test), whole-brain analyses were poorly sensitive, which underlines the difficulties of pediatric research. Finally, as has been shown in alphabetic languages, the slope of normal reading acquisition in children depends on the transparency of the writing system. In other words, a typical 10-year-old reader would not perform equally well in English, French, German or Italian (even if we could use exactly the same test) because he or she has not yet attained the same level of reading fluency. Thus, comparing languages and assessing equivalent reading delay in different writing systems, while still matching for age and education, raises complex issues and increases the difficulties of intercultural comparisons in children compared to adults.
However, since we had a particular interest in understanding the role of previously identified brain regions, an analysis based on ROIs was a reasonable choice, having the great advantage of mitigating the problem of multiple comparisons and reducing the risk of false negatives, but also allowing us to examine activation patterns at an individual level, as in the MPVA analyses. These targeted analyses were able to show the role of MFG in alphabetic languages. As usual in science, the results will need to be confirmed in future studies using larger samples to obtain more robust inferences.
We have added above points at the end of the discussion as limitations as suggested.
3) By Language Analysis – Both Reviewer 2 and 3 would like to see analyses broken down by writing system. From Reviewer 2 "Since authors do show significant correlation with reading abilities in several areas, there appears to be room to explore differences in writing systems. Can the authors do further analyses and potentially build on these results (also), and if not, justify why?"
We have added the requested analyses and performed the correlation analysis separately in each writing system. As predicted by Reviewer 3, the split decreased the statistical power. To be fully informative for the reader, we presented results at the threshold of voxel-level p<0.001 and cluster-level uncorrected (Figure 2B and 2C). As it can be seen, similar regions (left STS, bilateral PCG, bilateral MFG, bilateral FFG, right MOG) were sensitive to the reading performance in the two languages.
4) Discussion of Power – All reviewers were concerned about the experiment and contrasts being properly powered. Though N=96, there were only 24 per group for each of the four groups. As Reviewer 2 asks, on what basis is this sufficiently powered? On a related point, given the specificity of the claims wrt a lack of language interaction, Reviewer 3 points out that planned comparisons focused on language (if power is sufficient), in both ANOVAs and Bayes factors, are needed to support this line of argumentation.
The issue of power in fMRI studies is an extremely complicated one, very different from behavioral studies with a single dependent measure. There is no agreed-upon way to precalculate power for an fMRI study, which involves typically ~200,000 voxels, each with their own level of signal and noise.
Furthermore, power cannot be computed a-posteriori. Let us cite from Mumford’s (2012) reference article on this topic: “Due to the relationship between the Type I error rate and power, running a power calculation on a dataset to predict the power of that study is not informative (Hoenig and Heisey, 2001; Levine and Ensom, 2001). […] What can you do if you do not detect anything significant for one of your hypotheses and reviewers request a posthoc power analysis? Since it is not informative and there are plenty of references explaining why (Goodman and Berlin, 1994; Hoenig and Heisey, 2001; Levine and Ensom, 2001), you can respectfully decline to run a posthoc power analysis. One alternative, which is often suggested, is to supply a confidence interval for the estimate (Goodman and Berlin, 1994). The reason behind this is that the confidence interval provides intuition for the range of values supported by the data that we do have.”
Besides, it is generally recommended to have at least 20 subjects in fMRI activation studies to obtain stable results (Mumford, 2012). In the meta-analysis on reading presented by Martin et al., 2015, the average number of children (7-12 years) in 20 studies was 19.75 (5 to 64) compared to 14 in the adult studies (4 to 37) and for the meta-analysis on dyslexia (Richlan, et al., 2009), the average number of participants was 23 in each group (12 to 41). The fMRI studies comparing writing systems in children also had a similar number of participants: 8 children in Siok et al., 2004 and 12 adolescents in Siok et al., 2008, 17 first graders in Krafnick et al., 2016,
21 adult participants per language (Spanish, English, Hebrew, and Chinese) in Rueckl et al., 2015, and 34 simultaneous Hindi-English 9 year-old readers in Cherodath et al., 2015. Thus, even considering each subgroup, with N=24, we are in a classic range of participants in fMRI studies. Furthermore, given the differences between writing systems and some previous studies and especially when we focused on specific ROIs, we should have expected large interactions between language and reading system. Not only did we not find such large interactions, but we found moderate Bayesian evidence in favor of the null hypothesis of no interaction, and positive evidence that the effects were present in each writing system (see preceding point).
We added a discussion of those points in the main text, and well as the post-hoc comparisons in each language.
5) Discussion of whether and how the population was identified, and reading skill instead of dyslexia – Reviewer 3 had a very helpful and constructive point: "the value of the paper is adding incremental evidence for a shared pattern of activation areas that distinguish low levels of reading skill from more typical, higher levels." But thinks that framing this result in terms of *reading skill* is more warranted than in terms of dyslexia, given the way Chinese dyslexics were identified from the -1.25SD of normal classrooms.
We thank the reviewer for this suggestion, with which we fully agree. We have replaced “dyslexia” by “impaired readers” and “reading-impaired children”. In the Results section, we used “the effect of reading skill” rather than “the effect of dyslexia” to describe the corresponding results. We also revised the Discussion section.
6) Interpretation of the results in relation to existing literature and LMFG – Lastly, the paper's main conclusion is that these results counter existing research showing brain activation patterns implicating the involvement of LMFG in logographic more than alphabetic reading. In fact, as Reviewer 3 points out, the results reported here actually align with existing studies reporting differences between logographic and alphabetic systems, and this should be highlighted. Or to put it another way, now that the LMFG is found to be involved in alphabetic reading, is it now elevated to part of the universal reading network that has previously excluded it?
We agree that we have emphasized the commonalities between writing systems, and not discussed enough the language specificities that we observed. We have corrected our discussion in this respect, and notably discussed the functional significance of the LMFG in Chinese and alphabetic languages. Consistent with our previous study (Nakamura, et al., 2012), we propose that LMFG is one of the region that participate in reading in all cultures, but that its activation intensity is modulated by the learning experience, e.g. more activation in Chinese due to more writing practice in Chinese children than children in alphabetic language.
“The greater involvement of the left middle frontal gyrus (BA9) in Chinese than in French children is in line with existing research (Bolger et al., 2005; Tan et al., 2005a). This region has been proposed to support language-specific processing required by Chinese reading, such as whole-syllable retrieval (Tan et al., 2005a), the tonal nature of Chinese phonology (Gandour et al., 2002), memory-based lexical integration of orthography, phonology and semantics (Perfetti et al., 2005), visuospatial working memory (Wu et al., 2012) or writing gesture information (Cao and Perfetti, 2016; Nakamura et al., 2012). In the reading of English, its proposed functional role is also diverse: lexical semantics (Bolger et al., 2005), phonological processing (Pugh et al., 1996), lexical selection (Bolger et al., 2008), or grapheme–phonology conversion (Jobard et al., 2003).
Perhaps the most likely explanation is that the LMFG, as well as the more medial and dorsal Exner area, are systematically involved in writing (Planton et al., 2013; Purcell et al., 2011). FMRI studies in typical adult readers, comparing French and Chinese reading, have shown that the left MTG is not specific to Chinese writing, but includes a representation of handwriting gestures that is engaged in both alphabetical and non-alphabetical languages (Nakamura et al., 2012). As proposed by Nakamura et al., 2012, the activation in the MFG and PCG, as seen here, might correspond to the “reading by hand” circuit (gesture recognition system) as opposed to the “reading by eye” circuit (shape recognition system in the VWFA). Furthermore, writing is not only learned in parallel with reading, but facilitates reading acquisition in children (Bara et al., 2004). Thus, the involvement of the left MFG in French children, although less important than in Chinese children, might be due to a heavier dependence on handwriting during reading acquisition than during adulthood. This interpretation is supported by a recent study showing that, in second-language learners of Chinese, viewing characters that were learned through character writing induced greater activation in LMFG compared with those learned without character writing (Cao and Perfetti, 2016).”
Please find the full reviews attached for your reference:
Reviewer #1:
1) Reasoning stems from a fallacy – namely that the lack of evidence for a difference is evidence for a lack of difference. It is not.
We totally agree with the Reviewer but here through multiple analyses, we tried to provide the readers with all the neural evidence we obtained when these 4 groups of children were exposed to words. Indeed, we did not find strong evidence supporting large differences between Chinese and French children. We have added a paragraph on the limitations of our study. At least, we have no evidence of a major difference between Chinese and alphabetic languages.
2) It is problematic that the authors interpret BOLD signal/ brain activity as somehow equivalent to a mechanism or computation, without any attempt to constrain or define that computation.
We have changed the word “mechanisms” to the more neutral “neural correlates”. This is a study of reading circuits at the whole-brain level. Its aims are to identify the correlates of reading skills in young children, and to probe the putative differences between those networks in two different writing systems. The purpose is not to attempt to define the precise computation underlying each region. However, we have added a more thorough discussion of what this computation could be (e.g. the role of the MFG in encoding writing gestures).
3) The study of dyslexia is important, but here it clouds the inference waters about the neural substrate of reading – it cannot separate what the neural "causes" vs the neural "effects" of dyslexia are.
Indeed, this criticism applies to all fMRI studies. Whether or not the reduced activations in reading-impaired relative to typical readers reflect the causes or the consequences of dyslexia, they highlight the regions involved in reading. Causal inference is not the only means of progressing in a scientific domain. With respect to adult reading, we have both causal evidence (if a brain infarct destroys the VWFA, the patient can no longer read (pure alexia) (Déjerine, 1891)) and correlational evidence (when illiterates are presented with words, a reduced activation is observed in the VWFA relative to literates, which is a consequence of having not learn to read (Dehaene, et al., 2010)). Both arguments point to the VWFA as a key region in the reading circuit. We used the same logic here, and we do not understand how “it clouds the inference water about the neural substrate of reading”. If in two different writing systems, dyslexia (or insufficient reading skills) is associated with similarly reduced activations relative to the typical readers, it is logic to infer that the corresponding sites are jointly involved in reading acquisition in both writing systems.
This point was already stated in the previous version of the manuscript: “Given that the
VWFA is also under-activated by written words in illiterates (Dehaene, et al., 2010; Hervais-Adelman, et al., 2019) and in typical children before they learn to read (Dehaene-Lambertz, et al., 2018), its under-activation in impaired readers may simply reflect the lack of reading practice – in other words, it might be a consequence rather than a cause of the reading deficit.” and “Again, because this site (pSTG) is also under-activated during spoken language processing in illiterates and in preliterate children compared to adult and children readers (Dehaene, et al., 2010) we cannot ascertain whether its anomalous activation is a cause or a consequence of the reading impairment. Indeed, it does not have to be one or the other, but could be both, as phonological awareness is known to be both a predictor of future reading impairments (Hulme, et al., 2012; Melby-Lervåg, et al., 2012), and a consequence of learning to read in an alphabetical language (Morais, et al., 1986).”
As this citation shows, we did not over-interpret the findings we discovered. We never tried to explain the group differences we found between typical and dyslexic readers as causes of dyslexia. Thus, we disagree that our study “clouds the inference waters about the neural substrate of reading”.
Finally, although it is important to understand what causes dyslexia, we believe that, for intervention purposes, it is equally important to understand what are its consequences, and whether they vary across languages. In this way, we progress in our ability to provide effective theoretical bases and guidance for the intervention of children with reading impairment. In the domain of developmental neurological impairments, the causes of the deficits may well turn out to be impossible to cure, but it may still be possible to intervene and help those children compensate for their deficits.
4) The experiment is designed to rely on interactions for scientific inference, yet appears only to be powered to detect main effects – though no power analysis is presented. The Bayes factors reported are weak evidence for the null, and worse are highly dependent on the prior. Using a more informed prior (for example, from the existing literature cited here, such as 9,10,32 and many others) will decrease evidence for the null dramatically (and killing this Bayes factor further, I suspect; moreover, it is only Bayes factors > 10 that are regarded as strong evidence for the null). That has not been done here. Ideally one needs to show a comparable range of Bayes factors with multiple, varied, or highly justified priors. That has not been done here. Mixed models should be used here over anova in order to take into account item variance simultaneously with subject variance.
Please see the response to the previous reviewer. Also, our analysis already uses one-sided alternative hypothesis as more informed prior, such as group1 (control) > group2 (impaired readers) and group 1 (Chinese) > group 2 (French) so that the Bayes factor we reported before were not being further killed. Besides, according to previous literature, evidence for H1 against H0 is generally considered as moderate for BF10≥3, and strong for BF10≥10. In the main analysis, the Bayes factors are larger than 9.9 in the FFG, MFG and STS. Post-hoc comparisons in each writing system show effects in the same direction. Given the literature, we were expecting large differences between Chinese and French in the VWFA and MFG except paradoxically the MFG in Chinese children. Following Editor’s suggestion, we have added a sensitivity analysis which confirmed that our Bayes factors are not largely driven by the priors we choose.
5) It is not clear how this work is an advance over existing findings, see in particular citations 9, 10, and 32.
Citation 9 (Rueckl, et al., 2015) and citation 10 (Szwed, et al., 2014) are cross-cultural fMRI studies based on adults who are proficient readers; impaired readers were not included in these two studies. Citation 32 (Goswami, et al., 2011) was a cross-cultural behavioral study that compared children with and without dyslexia in different writing systems but without fMRI data.
The reasons why we think our work is more advanced than previous studies are as follows:
a) First, the fact that the neural bases of different writing systems have already been compared in adults does not reduce the interest of performing similar (but more demanding) studies in children. Here, we study the mechanisms of learning which, in an immature and inexperienced brain, could conceptually be based on different mechanisms than those observed in the mature brain. This is especially the case for a highly cultural, education-dependent activity such as reading. In principle, brain activations in children could be much more distributed or, on the contrary, more restrained than in adults. Stability across ages, experience and culture is thus a remarkable fact. It is not such a trivial observation, if we compare this stability with the plasticity of the human brain, which allows for a considerable degree of reorganization after early brain injury (Cohen et al., 2004).
b) Second, it is still an ongoing debate about whether the neural network for reading is universal or different across languages. Although citation 9 and 10 consistently revealed the language-universal neural bases underlying proficient reading and citation 32 found that the common cognitive deficits in dyslexia across different writing systems, there are several studies that claim to provide support for the language-specific view of reading and reading disorders (Siok, et al., 2020; Siok, et al., 2004). Thus, the current literature is inconclusive about the question we asked in our work: to what extent are the brain mechanisms for reading acquisition similar across writing systems, and do the same neural abnormalities underlie developmental dyslexia in alphabetic and ideographic systems? However, up to now, there are very few comparable studies, probably owing to the difficulty of scanning large groups of children, and of organizing strictly comparable multi-site fMRI studies.
c) Last, we are in the best possible experimental situation to perform cross-linguistic comparisons: our groups are perfectly paired, very homogeneous in age and comprise a large number of children (24 in each group, for a total of 96 children; this is no small feat when considering the constraints that face such developmental studies). We used the same experimental task and brain imaging acquisition parameters in France and China, thus controlling for many confounding factors.
Reviewer #2:
Summary
In the current manuscript, authors show convergent results of common neural correlates of word recognition in typical developing and dyslexic children in French and Chinese. While their conclusion is based on "null effects" (i.e., Chinese and French dyslexics are similar), they use a set of exhaustive approaches (whole-brain MRI, ROIs, MVPA) and Bayesian inferences to support their findings. The results provide additional insight into the ongoing debate about the universality of the neural correlates of dyslexia across cultures and writing systems. Their results also shed light on the neural correlates of pediatric dyslexia in children showing similar characteristics to adults (though they do not compare with adults directly in their study). The study fills an important gap that goes beyond prior meta-analytic literature that showed that the neural basis of dyslexia is similar across alphabetic writing systems that are different across orthographic depths in the visual word form area (VWFA); they show using empirical data (and not meta-analyses) that even completely different writing systems such as Chinese and French are similar. The authors make pain-staking and impressive efforts to match across French and Chinese in terms of the various parameters (scanner characteristics and demographic). The results are generally convincing and manuscript well written. Detailed major comments are listed below.
1) Premise of the study: The major issue for this reviewer was that the "universality of dyslexia" argument lies on the fact that there is some neural signature of dyslexia. Yet, at the whole brain level, there is no significant difference between dyslexia and controls. The rest of the analyses heavily relies on the regions identified in the ROI based approach. This should be thoroughly discussed and addressed as an important limitation.
We have added a paragraph on the limitations of our study at the end of the discussion. Nevertheless, we want to point out that testing the literature predictions by looking at specific ROIs is a valid way to improve statistical sensitivity by both decreasing the impact of corrections for repeated measures and by averaging activations on a set of voxels. Furthermore, we provided the figures with non-corrected activations to provide the reader with a complete information.
2) Correlation analyses with reading: Authors then move into correlation with reading ability but do not pursue further than showing correlations in the entire sample. They do not break the sample up into writing systems. Since authors do show significant correlation with reading abilities in several areas, there appears to be room to explore differences in writing systems. Can the authors do further analyses and potentially build on these results (also), and if not, justify why?
Following the suggestion, we did the correlation analysis in each writing system and added the results in the manuscript. As we can see from the Figures 2 B and C, similar regions (left STS, bilateral PCG, bilateral MFG, bilateral FFG) were sensitive to the reading performance in both languages.
3) Power & sample size: Discussion paragraph two: Authors mention that they were sufficiently powered because they had N=96. This is still 24 per group for each of the four groups. On what basis was it sufficiently powered? Especially in reporting these "null" results, it's important to provide evidence and perhaps reiterate this evidence when mentioning their "sufficient power" in the Discussion.
Please see response to the Editors’ point 4.
4) MVPA: In general MVPA was confusing in many ways and the reviewer recommends a major rewrite.
a) Motivation: Introduction paragraph seven: The phrase about MVPA where they motivate this analysis should probably be a separate sentence as the question being asked is very different than the first part of the question. Further, this justification comes out of nowhere and should be elaborated. The authors claim that they wish to test dispersion and intensity using classical analyses, and also using MVPA. They also do not mention in the Introduction, but in the Materials and methods, mention that they are examining stability (reproducibility from Run 1 to Run 2). Please clarify in the Introduction.
We have rephrased and clarified why and how we did the MPVA analysis in the Introduction section.
“Because group analyses leave open the possibility that the observed group differences might be due to spatially more variable activations in impaired-readers than in typical readers, we also performed individual-based analyses and compared the location and activation values of the most responding voxels in each child. Finally, because classical analyses may mask the presence of fine-grained activity patterns that are specific to a given subject or a given category, we also quantified the stability of subject-specific activation patterns within reading-related ROIs using multivariate pattern analyses. The goal of these analyses is to circumvent the blurring effect of group analyses, which may hinder the discovery of genuine but more dispersed activations in reading impaired children relative to controls.”
b) Materials and methods subsection “Multivariate Pattern Analysis (MVPA”: This was very confusing all around. For example… (1) What values are fed into as features? Time-series correlation? Or betas for each trial? They mention "pattern of response evoked by words relative to fixation" but this is unclear. (2) They claim they are doing multi-"voxel" pattern analyses (MVPA), but they also seem to refer to each ROI having one correlation coefficient. (3) It was not clear why this is called MVPA. What was the analytical technique used, ANOVA? Correlation-based MVPA (what is this exactly?)? There are probably many more, but it's hard to judge the quality of this analyses based on their lack of details and hard to ask further questions at this point.
We first computed the correlation of the activations (beta values) of all voxels within a ROI of run1 with the activations of the same voxels in run 2, either within a given visual category (e.g. Word-Word) or between categories (e.g. [Word-Face, Word-House, Face-House]). Then an ANOVA was performed to compare the correlation coefficients within a category vs between categories. This was done for each ROI and separately for the Word category and the Face category. We have rephrased the Materials and methods and Results sections.
5) Population: The test scores in Chinese was normed based on their own data of 2554 individuals. It is not clear whether this is appropriate (though size is large). How were these children recruited; is this a nationally representative population? In addition, how French children with dyslexia was identified (and criteria) was not mentioned in this paper and the authors simply refer to another paper. Since this is quite important, it will be nice to have this detailed here in this manuscript as well.
The 2554 Chinese children were recruited from eight public primary schools in Beijing. Of these schools, five were located in the downtown area (in two districts whose total GDP ranked 1st and 2nd out of Beijing's 16 districts in 2017) and the last three in rural areas (in two districts whose total GDP ranked 8th and 13th out of Beijing's 16 districts in 2017). All Children in 3rd – 5th grades in each school were tested as a group in their classroom. Among the 2554 children, there were 368 3rd graders, 1158 4th graders and 1061 5th graders. Test results were standardized on these 2554 children because the previous standardized tests were based on 1993 data (e.g. Chinese Character Recognition Test) and were no more suitable for screening for dyslexia in 2013 because children can read many more characters than their counterparts of 20 years ago. Given that Beijing is one of the most developed cities in China, both economically and educationally (www.bjstats.gov.cn/tjsj/tjgb/ndgb/201511/t20151124_327764.html), it is difficult to determine whether these 2554 children could represent the national population, but we think they could at least represent Beijing.
Following the suggestion of Reviewer 3 and of the Editor, we have replaced "dyslexics" with "impaired readers" and "children with reading impairments" because the standard deviation we used (<-1.25) is not a strict criterion for dyslexia.
We have added the criteria for identifying French impaired readers in the Materials and methods section.
6) Reading ability: (1) How did they combine reading skills across writing systems to combine all in a correlation analysis, especially when reading measures that fed into the dyslexia criteria were different, there is no standardized measures in Chinese and just that the measures are so different across the board? (2) This also means that although unreported, the authors can compare reading scores (as well as others such as IQ) across writing systems, and report in Table 1? Please elaborate.
1) In the correlation analysis across all participants, we used standard scores in the main screening test of dyslexia in each language (Chinese character recognition test for Chinese children and “L’alouette” for French children). Although the tests are not equivalent, z-scoring allows to index each child reading lag relative to a normal cohort. The Chinese character recognition test (Wang and Tao, 1993) has been widely used to identify Chinese dyslexia (Feng, et al., 2017; Liu, et al., 2012; Qi, et al., 2016) and “L’alouette” is a standardized reading test classically used to detect dyslexia in French-speaking children (Altarelli, et al., 2014). These two tests are time-limited with penalties for errors.
2) Instead of comparing the reading scores, we added the CI 95%, which offers a better image of the values distribution in each group. The French reading impaired group can be considered as genuinely dyslexic (CI 95% [-2.34 -1.94] SD), whereas the Chinese group is less deviant (CI95% [ -1.54 -1.95]).
Other measures were not done systematically in all groups. For example, for French children, we considered children with normal reading scores who were in their appropriate academic level given their age with no complaint from the parents or the teacher, as typical children. A full IQ battery was performed in dyslexic French children but in a learning dedicated center at distance from the scan day. At the scan day, we only performed some subtests of Wechsler's WISC III or IV to ensure that they were in the normal range for all children (e.g. cubes for PIQ).
Chinese children were tested with the Raven's Progressive Matrices. The goal of these tests was only to dismiss any child with general cognitive difficulties. The z-score values of PIQ in these tests (with CI95%) are in Author response table 1:
Mean | CI 95% | |
---|---|---|
Chinese typical readers | 0.67 | 0.44 ~ 0.89 |
Chinese impaired readers | 0.69 | 0.47 ~ 0.91 |
French typical readers | -0.41 | -0.85 ~ 0.36 |
French impaired readers | -0.11 | -0.63 ~ 0.41 |
Reviewer #3:
General Assessment. An interesting paper with rich results, an fMRI study comparing previously reported results of French children with new data on Chinse children. The study is characterized by careful sample matching, comparable instrumentation across sites, and multiple sophisticated analyses. The paper provides a context of previous brain imaging studies that compare dyslexia across Chinese and alphabetic writing and places its results in this context. Its message is not novel-that the brain shows a reading network that is shared across languages and writing systems-but because the research has suggested different conclusions on whether reading Chinese and reading alphabetic writing involve different brain areas, these results could add weight to two aspects of the issue. Do alphabetic and Chinese reading engage the same brain areas to the same extent? Do Chinese and alphabetic dyslexics show similar activation patterns across brain areas? The results answer the first question by "yes" and no", but the authors emphasize more the "yes" part. They answer the second part "Yes" Some revision and perhaps reframing of the contribution of the paper to the two issues engaged-language/writing system factors and dyslexia-would improve the paper.
Major comments on context and Interpretation.
1) One suggestion is to broaden the context for the question and to refine its form. Arguments that the procedures of reading are universal have been made on the basis of behavioral studies and the logic of writing systems (e.g. Perfetti, 2003; other observations by Mattingly, 1972) If graphs-the written symbols of language-encode linguistic forms-then the brain, to decode writing, must connect visual areas to language areas. Beyond that fundamental universal, the factors of mapping of phonology and morphology, and perhaps graphic complexity, could influence the details. Whether there is a shared reading circuit across languages has not been the main question. Rather, the question has been, given a shared or universal circuit, what differences might one expect based on the features of the writing system and the language?
We thank the reviewer for this comment and have since implemented this suggestion. We have revised the Abstract and the Discussion section to emphasize the language differences we found in the current study.
2) Although the authors' argument is strong on the universalist side, the data show convergence with the results of Siok, Tan et al. in the greater LFMG and SPL activation for Chinese. Puzzling, as the authors note, is the difference in pSTG, also higher for Chinese, contrary to other results. Given these language differences, what does the failure to find a language x dyslexia interaction mean? Both normal and dyslexics in Chinese show the Chinese pattern (more activation in MFG and SPL) and both controls and dyslexics in French show the French pattern (Less MFG, Less SPL, and, surprisingly, pSTG), as shown in Figure 4.
We agree with the comment. However, as also pointed by the reviewer, the differences between writing systems are due to modulation, not to a radical difference in circuits.
3) With these language differences, an appropriate interpretative framing would be to emphasize writing system/language differences while finding no dyslexia differences across languages. The paper reports and comments on the differences findings but does not highlight them. The Abstract and conclusion should be written to reflect these findings.
We thank the reviewer for this comment and have since implemented this suggestion. We have revised the title, Abstract, Discussion and conclusion section to emphasize the language differences we found in the current study.
4) The LFMG difference might reflect the role of writing experience in Chinese, which may make frontal brain areas near the motor cortex, especially Exner's area, more active in reading. The authors cite the comparison of French and Chinese by Nakamura et al., 2012), but the evidence from Cao & Perfetti, 2017, is also relevant because it reports that the overlap of writing and reading brain areas is greater in Chinese than English and also in learners of Chinese as L2 who had learned through character writing compared with learners who had learned through pinyin.
We thank the reviewer for this comment and have since implemented this suggestion. We have discussed more the function of L.MFG, emphasizing that the greater involvement of this region in Chinese readers might be related to their massive writing experience which reflects culture-modulation on reading circuits. We also added the mentioned paper.
5) More generally on the issue of context and interpretation, the authors should consider the relevance of Perfetti, Cao, and Booth (2013), which provides a more nuanced framing of the universal/specialization issue and summarizes results from children's studies carried out by Booth et al. to identify brain regions that show developmental changes in English and Chinese, highlighting commonalities along with divergence in STG and LMFG.
We thank the reviewer for this comment and we have reframed the manuscript.
6) Any comments on why MFG shows a much stronger dyslexia effect for French (strong) than Chinese children (marginal or null), according to the Bayes factor data shown in Table 4? The language difference shows more MFG involvement among Chinese than French. The reporting of the results in subsection “ROI-Level analysis” refers only to the Bayesian evidence against a language x dyslexia interaction, ignoring the post-hoc analyses (which given the goal of the study, could have been planned comparisons), which also show strong F10 support for an MFG difference between French and Chinese dyslexia but marginally "moderate" support for such a difference between French and Chinese controls.
We have added the Bayes factors of post-hoc analyses between controls and impaired readers in each language. A much stronger dyslexia effect for French (strong) than Chinese children (marginal or null) in LMFG may due to the fact the French impaired readers (<-2 SD) in the current study were more severe than Chinese impaired-readers in reading performance (<-1.25 SD).
7) Concerning the MFG, a fair summary of the results is that MFG activation was observed more for Chinese than French and more for controls than dyslexics. It would be appropriate to point out that this confirms conclusions from other research. Thus, the conclusions from Siok, Tan, et al. that the LMFG is more involved in reading Chinese than in alphabetic reading is also found here, as is their conclusion of its under-activation in Chinese dyslexia. What may be new in the present paper is evidence that under-activation of the MFG may also characterize reading French. This last conclusion is not based on the present results, however, but on results from the previously reported results for the French children. The authors should point out whether the original publication of the French results reached this conclusion or whether it emerged only in the comparisons of the present paper. The Monzalvo et al. abstract emphasizes VWFA factors in dyslexia, but not MFG.
We thank the reviewer for this comment and have since implemented this suggestion. We have rephrased and emphasized that the language effect and reading skill effect in LMFG in the current study is consistent with the conclusions from previous studies. We also mentioned the original publication of the French children in which we did observe different activation pattern of LMFG in controls and dyslexic children (see Monzalvo et al., 2012) although the difference was not significant at the whole brain level.
8) The Interpretation of the VWFA as a universal site should take into account that both faces and houses showed lower activation for dyslexics along the anterior-posterior axis. This affects how to interpret the lower activation for dyslexics to words in the VWFA.
The observation of a general reduction in activation in reading impaired children relative to controls, along the anterior-posterior axis does not modify the interpretation of the VWFA as a universal site of reading. We have added a short paragraph to discuss these findings:
“Thanks to individual peak location and intensity analyses, as well as multivariate pattern analyses, we could reject an alternative interpretation which, to the best of our knowledge, was not explicitly tested in previous studies: the possibility that the reduced activations are an artifact of group averaging, solely due to greater inter-individual variability in the localization of reading-related circuits in the dyslexic brain. Using individual peak, we observed that the brain localization to words were not more dispersed among dyslexic participants than among controls. Using MVPA, we showed that, within individual subjects, the activation patterns in the VWFA in response to written words were less reproducible across runs in dyslexics than in typical readers. This was solely the case for words, not for the other visual categories. We did observe a slightly reduced activation to faces and houses in dyslexics relative to controls, as previously reported in illiterate subjects (Dehaene et al., 2010), as well as seen in response to non-word stimuli (numbers, abstract strings) (Boros et al., 2016) and faces in dyslexics (Gabay et al., 2017; Monzalvo et al., 2012). However, the pattern of activity for faces was stable from one run to the next, contrary to what was found for words in the MVPA analyses. This observation suggests that the reduced activation to written words may not reflect a general disorganization of the extra-striate visual areas, but rather a specific difficulty with written words.”
9) The conclusion of the paper needs to be revised to more accurately reflect the results of the study. This should include the differences that were found, not just those that were not found and a more contextualized grounding for the conclusion that "cultural variability is merely reflected in the variable emphasis that different writing system put on (linguistic units)". This is not new (which is OK; it's worth emphasizing) and has been pointed in other papers in which the results have shown language differences.
We thank the reviewer for this comment that help us reframe the manuscript. We agree that we may have focused too much on the neural responses common to Chinese and French children without sufficient discussion of the differences. We have revised the Abstract and the Discussion section to emphasize the language differences we found in the current study.
10) The authors acknowledge a possible task issue but an additional consideration seems relevant. The task involved reading only incidentally. We know that for a skilled reader, implicit reading engages the network non-attentively (automatically), so the task itself is ok, as are naming and semantic judgement used in other studies. The problem is that, because it is the only task used, we get a limited picture of the reading network, perhaps especially for the dyslexics who cannot be assumed to engage the reading network automatically. It is worth noting that the research reporting Chinese-alphabetic differences did not use passive reading tasks but naming, rhyming, and meaning decision tasks. I don't know whether this is relevant for the differences, but It would strengthen the contribution of the paper if it could suggest a reason for different conclusions arising in the literature or address the task issue in that context.
We thank the reviewer for this comment and have since implemented this suggestion. In the revised manuscript, we pointed out in the Introduction and Discussion section that previous literature, particularly with Chinese dyslexics, used different kinds of reading tasks, e.g. picture and semantic matching task (Hu, et al., 2010), homophone judgment task (Siok, et al., 2004), font-size perceptual matching task (Siok, et al., 2009) and morphological task (Liu, et al., 2013), whereas our findings were based on a passive reading task. This may partly explain why we found a culturally-universal neurobiological basis of impaired reading, e.g. decreased activation in LMFG in both Chinese and French impaired readers while previous studies did not. We also discuss in greater detail the importance of the choice of a task where all children can perform equally well.
11) I also wonder about the degree of reading difficulties shown in the sample. The Chinese children were classified as dyslexic by showing a -1.25 SD score on the CCRT character recognition test. We need to know what the task was for the CCRT. Performance on a recognition task reflects learning through experience. A reading problem may reflect a poor learning outcome from experience but should not reflect mere experience. Also -1.25 SD seems to identify a rather large percentage of children as dyslexic, compared with other studies, which may make it more likely that reading experience is involved to some extent. This is not a problem for an approach that treats reading as a graded skill, which the authors do in some of their analyses. Making reading skill rather than dyslexia the framing of the paper could help with this.
We thank the reviewer for this comment and have since implemented this suggestion. Followed the Reviewer suggestion, taken up by the editor, we have made reading skill rather than dyslexia the framing of the paper. We used “impaired readers” and “reading-impaired children” rather than dyslexia to describe the group of poor readers and we focused on the effect of “reading skill” rather than “effect of dyslexia” in the manuscript as below -1.25 SD of the norm is not a strict criterion for dyslexia screening.
The Chinese character recognition test (CCRT) is a word-compounding test and consists of 210 single Chinese characters, which were divided into ten groups based on their difficulty. Based on each of the given characters/morphemes, the participants were asked to write down a compound word or phrase on an answer sheet, which included the given character or morpheme as a part. For example, they may write down the word of “池 塘 /chi2tang2/(meaning pond)” for the character of 池/chi2/(meaning pool). The participants were asked to writing down as many as possible in 45 min. The number of correct answers was the raw score. 2554 children from grades 3–5 were tested with this test for their own grades. The mean and standard deviation of the performance of character recognition were calculated for each grade.
12) The ANOVA reported in subsection “Data-driven Analyses” does not provide a main effect of category. Instead it tests one category against the mean of the other two. And this leads to multiple tests that are appropriate with a significant main effect. A standard ANOVA with 3 levels of the category variable is needed.
We were interested by category-specific activations contrasts (category X > mean of the other two categories) not the main effect of category. We corrected the text.
13) Another statistical issue is the use of separate planned comparisons within each language when no interaction involving language is reported. In subsection “Anterior-to-Posterior gradient in the visual cortex”, this comparison, without a report of a significant language interaction, is shown for the anterior-posterior gradient analysis allowing the suggestion that the only dyslexia difference for both languages was near the VFWA. In other analyses, the authors report a lack of a language interaction and no further within-language comparisons. Given the focus of the research on language differences, planned comparisons between languages is quite reasonable to do in all analysis but the authors seem to have done this only selectively. Some explanation is needed.
We have added the post-hoc within-language comparisons between control and impaired readers, as well as the between language comparisons when language× Reading skills interaction was not significant.
14) The sentences "Dyslexics might have more dispersed activations without focal peaks that would also create weaker responses at the group-level, but a reproducible pattern of activations. In that case, the within-subject reproducibility of multivariate activation patterns should not differ between normal-readers and dyslexics" from subsection “MVPA of the activations to words”. The logic here escapes me. If Ds have more dispersed activation without focal peaks their reproducibility of multivariate patterns could still differ from controls. More voxels but not the same voxels, reflecting intrinsic variability. I think the authors are right that the individual peaks analysis means the between-group variability is not due to averaging individuals. But the differences in these individual MVPA analyses mean that variability is higher with dyslexics. The reason for this should be considered.
We have entirely rewritten this paragraph to clarify the logic in both Introduction and Results sections.
“Because group analyses leave open the possibility that the observed group differences might be due to spatially more variable activations in impaired-readers than in typical readers, we also performed individual-based analyses and compared the location and activation values of the most responding voxels in each child. Finally, because classical analyses may mask the presence of fine-grained activity patterns that are specific to a given subject or a given category, we also quantified the stability of subject-specific activation patterns within reading-related ROIs using multivariate pattern analyses. The goal of these analyses is to circumvent the blurring effect of group analyses, which may hinder the discovery of genuine but more dispersed activations in reading impaired children relative to controls.”
“The above analyses were carried out in a standardized way at the group level. It is therefore possible that the observed group differences were due to a greater inter-individual variability in brain localization in the reading-impaired group than in the control group. This possibility would lead to a completely different interpretation of the results: each child might have a well-organized brain activity for reading, with the only anomaly being a greater anatomical dispersion in the reading-impaired group compared to the control group. To test this possibility, we performed two individual-based analyses, one based on the comparison of the location and activation values of the most responding voxels and the other examining the stability of the pattern of responses across runs through a multi-voxel pattern analysis (MVPA). We focused on the ROIs previously showing significant differences due to Reading skill (i.e. left FFG, MFG, PCG and STS) and Language (i.e. left MFG, SPL and pSTG).”
15) Why on the MVPA are the data analyzed transformations of correlations of first run with second run? The idea is reproducibility, but the second run is a second run that provides a familiarization component that may affect controls and dyslexics differently.
The visual stimuli (except checkerboard) were different in each run. Furthermore, a familiarization component (i.e. due to the task itself) should similarly affect activation to words and faces.
16) Another point to be explained is why some analyses use fixation as the comparison (MVPA) whereas most use the contrast of one condition vs the other two.
When we were interested in category-specific/preferential activation, we used the contrast of (one category vs others). For example, in Figure 1A, we showed word-specific and facespecific activation maps as these regions had greater activation than the mean of other categories. When we were comparing children’s response to a visual category of interest, e.g. word response in the current study, we used the contrast of (words vs fixation). It allows to more clearly see the direction of the differences between children because for example a larger response to words in controls or a weaker response to others in dyslexics would induce a significantly larger effect in the contrast (word > others) * (controls > dyslexics).
17) Figure 2 shows areas with correlations with reading scores. Either the caption or the images should make clear the directions of the correlations and make clear this is for all subjects. Given the goal of the study, showing these separately for French and Chinese is better, even though it reduces the power.
We thank the reviewer for this comment and have since implemented this suggestion. We did the correlation analysis in each writing system and accordingly revised the figure.
As predicted by Reviewer 3, the power is reduced when data are split according to the language. To provide the readers with full information, we presented the results at the threshold of voxel-level p<0.005 and cluster-level uncorrected. Figure 2B and 2C separately displayed regions positively correlated with reading scores in Chinese and French children. As we can see from the figures, similar regions (left STS, bilateral PCG, bilateral MFG, bilateral FFG) were sensitive to the reading performance in the two languages.
18) Figure 3A is a bit puzzling. It shows peaks from literature on Chinese and alphabetic studies, but the only peak in LH is at MFG for Chinese. Certainly, studies of Chinese find peaks in other LH regions. Do the peaks have the same definition across studies and meta-analyses? One peak for each region of interest or region of a hypothesized reading network probably would show a different picture. As it stands the visual here contradicts universality and may be misleading. As for 3B, again it would be useful to indicate the direction of the correlations or state they are all positive if that is the case.
a) In ROI analysis, we aimed to test regions who had been reported sensitive to dyslexia/impaired reading in previous literature with our data. We thus first searched metaanalyses of imaging studies in dyslexics rather than in typically developing readers. However, due to the limited number of published neuroimaging studies of Chinese dyslexia, no metaanalysis was available to summarize the available evidence of into a pooled estimate. However, atypical activation in a lateral prefrontal region within BA 9 has been reported in readingimpaired Chinese children (Siok et al., 2004; Siok et al., 2009) and this region was repeatedly found to be more involved in reading Chinese than alphabetic languages (Bolger et al., 2005; Nakamura et al., 2012; Tan et al., 2005a). Besides, previous studies also often reported that Chinese reading networks are more symmetrical in the ventral visual system. We thus included the foci in both left middle frontal gyrus and right occipital cortex that were reported in several meta-analyses on Chinese typical reading (Bolger et al., 2005; Tan et al., 2005a; Wu et al., 2012; Zhu et al., 2014) and created representative ROIs. Notably, all ROIs (except two ROIs in the right-hemisphere) fell within the reading circuit identified in our participants (Words > other categories; see Figure 4 and Figure 4—figure supplement 1).
b) We have revised the Figure 3. We used red and blue dots to separately represent ROIs whose activation to words versus fixation were positively and negatively correlated with reading scores across all participants. Gray dots represent ROIs that did not reach significance in the correlation analysis.
19) The authors say that the LFMG dysfunction "was previously claimed to be specific to Chinese dyslexia". This is more accurately described as a discovery of an LMFG deficit in Chinese dyslexics in the absence of any existing evidence of a comparable deficit having been reported in in studies of alphabetic languages. The previously published study of French children should have added the discovery that LFMG dysfunction can characterize alphabetic reading as well.
We thank the reviewer for this comment and have rephrased in the revised manuscript.
Besides, we did observe different activation pattern of LMFG between controls and dyslexic children in our original publication of the French children (see Monzalvo et al., 2012) although the difference was not significant at the whole brain level. We mentioned this in the Discussion section.
[Editors' note: further revisions were suggested prior to acceptance, as described below.]
Revisions for this paper:
Please address points 1-4 by the reviewer; each query is concrete, concise, addressable, and clearly laid out. Please make these revisions and clearly indicated in the revised manuscript text as well as describe in a response letter how you have handled each point.
Reviewer #3:
The authors made several significant improvements in the paper and were responsive to reviewers' comments. I think the strength of the paper lies in the developmental results for a Chinese sample to compare with previously published results of a French sample, along with a better portrayal of its contribution in the context of other research. The French-Chinese comparison itself remains a question for me, but the Abstract, Discussion, and conclusion are more in line with the results.
I suggest the authors consider a few additional revisions.
1) The authors' reframing from dyslexia to reading skill is important, given the likely differences in reading skill between the Chinese and French samples. But the use of the word "impaired" to refer to lower skill readers throughout the ms does not follow through on this reframing. Dyslexia is an impairment. It might be that the French sample includes mainly or exclusively impaired readers and the authors may prefer that term to be consistent with their original publication of the French data. If so, it might be enough to have a sentence or two that explicitly address the terminology shifts or point out the possible French-Chinese differences in the low skill group. If effective reading experience is what is involved especially in Chinese-i.e. the rate at which a child has benefitted from learning opportunities, including reading practice-this might be important to mention. This would be consistent with the demands of incremental character learning and practice. It may be interesting that the distinction between impairment and delay may not matter given the authors' interpretation of the results of their comparisons.
We agree with the Reviewer’s suggestion and replaced the term “impaired readers” by “poor readers”. In the participants section, we re-introduced the term dyslexics to describe the French children, as these children met the diagnostic criteria for dyslexia, but we pointed out, as suggested, that the Chinese poor readers might not be proper dyslexics as French children were.
We replaced the word “controls” by “typical readers” in the figures to be consistent with the text.
2) If I understand correctly, the important conclusion that LMFG is part of the universal network is indirect, dependent on adding a Chinese sample to the original French sample for comparison. The French result did not show significant under-activation for impaired readers. Instead, only when the Chinese sample was included did this occur. I think this point should be made explicit. Maybe I misunderstand something, but it seems that only by adding a Chinese sample that we see LMFG problem in French children with reading problems.
We thank the reviewer for this comment and apologize that we were not clear but his/her statement is not correct.
First, we did observe different activation patterns of LMFG in dyslexics and controls in our original publication of the French children (please refer to Figure 1 and Table 3 in Monzalvo et al., 2012), but no difference survived correction for repeated measures at the whole brain level. Because we were interested at that time by the organization of the visual ventral regions, we restricted our analyses to a ventral mask for the comparisons between dyslexic and typical readers. Thus, we did not emphasize the putative LMFG difference in this paper. The other paper, Altarelli et al. 2013, was also only focused on the occipito-temporal region. Here in the new set of French children (The 24 children were chosen in both previous sets in function of the Chinese children age and gender), we observed an effect of reading performance in the LMFG (see Figure 2). When the French participants are split in typical and dyslexic readers, we obtained a significant cluster (33 voxels, pFWE_corr = 0.007, Z = 4.46 at [-54 12 30]) showing larger activation in typical readers than in dyslexics.
Second, in the ROI analysis in the current study, both the frequentist and Bayesian analyses revealed that the reading skill effect in LMFG was robust in French children.
Therefore, the LMFG is a genuine site also in the French participants which has been underestimated due to previous analyses strategies. It cannot be claimed that this site is specific to Chinese reading.
3) The greater activation for Chinese than French in the posterior superior temporal gyrus/sulcus continues to be a bit of puzzle. The explanation on 19-20 does not seem very plausible. Children's use of PinYin might help explain why the PSTG in Chinese is not less activated than in French, contrary to expectations based on decoding accounts of the PSTG. However, I don't see how it can explain the reversal that was observed for children who have been reading characters for 3-5 years. Rather than this extended speculation, it might be better to just mention it as one of several possibilities, including the possibility of more complex syllable decoding because of Chinese tones and connections to spoken language functions, or a need to suppress the activation of the syllable associated with the character (because it leads to competition for character identification), or others that also are speculative. If the evidence for a letter-phoneme function of this area comes mainly from explicit reading tasks, maybe this area is doing something else related to the target task during incidental reading.
We thank the reviewer for this comment and have added sentences following his/her suggestions.
4) The authors use "script" where they actually refer to writing system. I strongly suggest a change in the title and elsewhere to writing system. The difference between French and Chinese is one of writing system and this inevitably brings along a difference in script. The difference between French Cursive and French Gothic and Chinese Traditional and Chinese Simplified are differences of script.
We have reviewed the entire manuscript to have a consistent terminology. The revised title is now “A universal reading network and its modulation by writing system and reading ability in French and Chinese children.”
References
Altarelli, I., Leroy, F., Monzalvo, K., Fluss, J., Billard, C., Dehaene-Lambertz, G., Galaburda, A.M., Ramus, F. (2014) Planum temporale asymmetry in developmental dyslexia: Revisiting an old question. Hum Brain Mapp, 35:5717-35.
Cherodath, S., Singh, N.C. (2015) The influence of orthographic depth on reading networks in simultaneous biliterate children. Brain & Language, 143:42-51.
Déjerine, J. (1891) Sur un cas de cécité verbale avec agraphie suivi d’autopsie. Mémoires de la Société de Biologie, 3:197-201.
Dehaene-Lambertz, G., Monzalvo, K., Dehaene, S. (2018) The emergence of the visual word form: Longitudinal evolution of category-specific ventral visual areas during reading acquisition. PLoS biology, 16:e2004103.
Feng, X., Li, L., Zhang, M., Yang, X., Tian, M., Xie, W., Lu, Y., Liu, L., Belanger, N.N., Meng, X., Ding, G. (2017) Dyslexic Children Show Atypical Cerebellar Activation and Cerebro-Cerebellar Functional Connectivity in Orthographic and Phonological Processing. Cerebellum, 16:496-507.
Liu, L., Wang, W., You, W., Li, Y., Awati, N., Zhao, X., Booth, J.R., Peng, D. (2012) Similar alterations in brain function for phonological and semantic processing to visual characters in Chinese dyslexia. Neuropsychologia, 50:2224-32.
Mumford, J.A. (2012) A power calculation guide for fMRI studies. Social cognitive and affective neuroscience, 7:738-742.
Pinel, P., Lalanne, C., Bourgeron, T., Fauchereau, F., Poupon, C., Artiges, E., Le Bihan, D., Dehaene-Lambertz, G., Dehaene, S. (2015) Genetic and Environmental Influences on the Visual Word Form and Fusiform Face Areas. Cerebral Cortex, 25:2478-2493.
Pleisch, G., Karipidis, I.I., Brauchli, C., Röthlisberger, M., Hofstetter, C., Stämpfli, P., Walitza, S., Brem, S. (2019) Emerging neural specialization of the ventral occipitotemporal cortex to characters through phonological association learning in preschool children. NeuroImage, 189:813-831.
Qi, T., Gu, B., Ding, G.S., Gong, G.L., Lu, C.M., Peng, D.L., Malins, J.G., Liu, L. (2016) More bilateral, more anterior: Alterations of brain organization in the large-scale structural network in Chinese dyslexia. Neuroimage, 124:63-74.
Siok, W.T., Jia, F., Liu, C.Y., Perfetti, C.A., Tan, L.H. (2020) A Lifespan fMRI Study of Neurodevelopment Associated with Reading Chinese. Cerebral Cortex.
https://doi.org/10.7554/eLife.54591.sa2Article and author information
Author details
Funding
National Natural Science Foundation of China (81171016)
- Xiangzhi Meng
Agence Nationale de la Recherche (ANR-06-NEURO-019-01)
- Franck Ramus
Fondation Bettencourt Schueller
- Stanislas Dehaene
Agence Nationale de la Recherche (ANR-11-BSV4-014-01)
- Franck Ramus
Agence Nationale de la Recherche (ANR-17-EURE-0017)
- Franck Ramus
Agence Nationale de la Recherche (ANR-10-IDEX-0001-02 PSL)
- Franck Ramus
National Natural Science Foundation of China (81371206)
- Xiangzhi Meng
National Natural Science Foundation of China (31971039)
- Xiangzhi Meng
China Scholarship Council (201706040117)
- Xiaoxia Feng
The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.
Acknowledgements
We would like to thank all the children and their parents for their collaboration in this study. We are grateful to Mengyu Tian, Weiyi Xie, Feng Ai, Manli Zhang for their help in Chinese children’s behavioral and MRI data collection. We thank Le Li for providing suggestions and help on data analysis.
Ethics
Human subjects: The study was approved by Institutional Review Boards at Beijing Normal University in China and local ethics committee (CPP Ile de France VII, N° 11-008) in Kremlin-Bicêtre, France (#20130331). Written consent and consent to publish was obtained from all children and their parents.
Senior Editor
- Floris P de Lange, Radboud University, Netherlands
Reviewing Editor
- Andrea E Martin, Max Planck Institute for Psycholinguistics, Netherlands
Reviewer
- Charles Perfetti, University of Pittsburgh, United States
Version history
- Received: December 19, 2019
- Accepted: October 26, 2020
- Accepted Manuscript published: October 29, 2020 (version 1)
- Version of Record published: November 16, 2020 (version 2)
Copyright
© 2020, Feng et al.
This article is distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use and redistribution provided that the original author and source are credited.
Metrics
-
- 2,562
- Page views
-
- 460
- Downloads
-
- 33
- Citations
Article citation count generated by polling the highest count across the following sources: Scopus, Crossref, PubMed Central.
Download links
Downloads (link to download the article as PDF)
Open citations (links to open the citations from this article in various online reference manager services)
Cite this article (links to download the citations from this article in formats compatible with various reference manager tools)
Further reading
-
- Developmental Biology
- Neuroscience
The hippocampus executes crucial functions from declarative memory to adaptive behaviors associated with cognition and emotion. However, the mechanisms of how morphogenesis and functions along the hippocampal dorsoventral axis are differentiated and integrated are still largely unclear. Here, we show that Nr2f1 and Nr2f2 genes are distinctively expressed in the dorsal and ventral hippocampus, respectively. The loss of Nr2f2 results in ectopic CA1/CA3 domains in the ventral hippocampus. The deficiency of Nr2f1 leads to the failed specification of dorsal CA1, among which there are place cells. The deletion of both Nr2f genes causes almost agenesis of the hippocampus with abnormalities of trisynaptic circuit and adult neurogenesis. Moreover, Nr2f1/2 may cooperate to guarantee appropriate morphogenesis and function of the hippocampus by regulating the Lhx5-Lhx2 axis. Our findings revealed a novel mechanism that Nr2f1 and Nr2f2 converge to govern the differentiation and integration of distinct characteristics of the hippocampus in mice.
-
- Developmental Biology
- Evolutionary Biology
Gene expression has been employed for homologizing body regions across bilateria. The molecular comparison of vertebrate and fly brains has led to a number of disputed homology hypotheses. Data from the fly Drosophila melanogaster have recently been complemented by extensive data from the red flour beetle Tribolium castaneum with its more insect-typical development. In this review, we revisit the molecular mapping of the neuroectoderm of insects and vertebrates to reconsider homology hypotheses. We claim that the protocerebrum is non-segmental and homologous to the vertebrate fore- and midbrain. The boundary between antennal and ocular regions correspond to the vertebrate mid-hindbrain boundary while the deutocerebrum represents the anterior-most ganglion with serial homology to the trunk. The insect head placode is shares common embryonic origin with the vertebrate adenohypophyseal placode. Intriguingly, vertebrate eyes develop from a different region compared to the insect compound eyes calling organ homology into question. Finally, we suggest a molecular re-definition of the classic concepts of archi- and prosocerebrum.