Dyslexia and age related effects in the neurometabolites concentration in the visual and temporo-parietal cortex

Several etiological theories, in particular neuronal noise and impaired auditory sampling, predicted neurotransmission deficits in dyslexia. Neurometabolites also affect white matter microstructure, where abnormalities were previously reported in dyslexia. However findings from only few magnetic resonance spectroscopy studies using diverse age groups, different brain regions, data processing and reference scaling are inconsistent. We used MEGA-PRESS single-voxel spectroscopy in two ROIs: left temporo-parietal and occipital cortex in 36 adults and 52 children, where half in each group had dyslexia. Dyslexics, on average, had significantly lower total N-acetylaspartate (tNAA) than controls in the occipital cortex. Adults compared to children were characterized by higher choline and creatine in both areas, higher tNAA in left temporo-parietal and lower glutamate in the visual cortex, reflecting maturational changes in cortical microstructure and metabolism. Although the current findings do not support the proposed etiological theories of dyslexia, they show, for the first time, that tNAA, considered to be a neurochemical correlate of white matter integrity, is deficient in the visual cortex in both children and adults with dyslexia. They also point that several neurotransmitters, including ones previously used as reference, change with age.

Dyslexia is a brain-based difficulty in acquiring fluent reading skills that affects around 10-15% of the population 1 . Since this disorder is heterogeneous, many different etiological theories have been proposed to explain its symptoms 2 . Two were specifically linked to neurotransmission deficits, even though still little is known about neurochemical underpinnings of reading disorder. The first theory suggests that dyslexia might be a consequence of neuronal hyperexcitability, which contributes to learning deficits by heightened noise and instability in information processing 3 . This theory emphasizes the importance of balanced levels of excitation and inhibition within cortical pyramidal-interneuron networks, which support tuning to sensory input, neural timing, and information processing. Excitation-inhibition balance and the precise timing of neural activity can be disrupted by cortical hyperexcitability. On the level of neurotransmitters, this theory predicts heightened level of glutamatergic (Glu or Glx) signalling in dyslexic individuals 3 , since glutamate concentrations were found to be positively correlated with cortical excitability 4 . The loss of neuronal synchronization is hypothesized to lead to deficits in low-level temporal auditory processing, the oscillatory neural processes that sample and encode sensory information and impairments in multisensory integration, all of which are key components of reading development. Other theory of impaired auditory sampling 5 suggests that the cause of reading disorder lies in a general deficit in synchronization of neuronal oscillations specifically in auditory cortex (phase-locking) in response to external input. However, studies examining cortical oscillations in dyslexia are inconclusive. Some point to synchronization deficits in the delta and theta range of 2-10 Hz 5 which corresponds to syllable stress pattern, other show impairments in the low gamma band [25][26][27][28][29][30][31][32][33][34][35] , which corresponds to phonemic rate pattern. The two mentioned theories of www.nature.com/scientificreports www.nature.com/scientificreports/ neurochemistry in dyslexia. Furthermore, so far only three studies directly compared dyslexic to typically reading subjects 20,22,25 , but all had relatively small samples of dyslexic subjects.
Here we aimed at systematic analysis of potential neurometabolite abnormalities in dyslexia by examining the concentration of glutamate (Glu and Glx), choline, NAA and GABA in both adult and children samples. Additionally, we tested whether Cr level can be utilized as a reference. We hypothesize that, if reading disorder is associated with poor phase synchronization in low frequency bands, as suggested by Goswami 5 , dyslexic groups should show abnormalities in the concentration of glutamate and/or choline. If, however, as suggested by Lehongre et al. 6 , dyslexia is related to poor synchronization in the gamma band, concentration of GABA should be altered. Heightened glutamate in dyslexics compared to control subjects is also expected by the neuronal noise theory 3 . Finally, changes in NAA and choline can inform about white matter microstructure differences previously implicated in dyslexia 21 . Since still little is known about whether dyslexia effects change with age, besides the effect of group (dyslexic vs. typical readers), we examined both age effects and interactions on neurometabolites' concentration. Thus the present study, as compared to previous ones examining similar age ranges 23,25,27 , has an additional insight of examining the commonalities and differences between school-aged children and experienced adult readers in the level of neurotransmiters in two brain regions present in the previous studies (i.e. occipital cortex, and left temporo-parietal cortex including the angular gyrus). By examining two different brain regions in the same subjects we can infer about the specificity of neurotransmiter differences in dyslexia. Finally, in the supplementary analyses we compare different reference scaling methods and suggest directions for future research. In order for the reader to compare current findings with the previous literature we report all effects surviving the nominal significance of p < 0.05, but focus more on results surviving more stringent statistical threshold corrected for multiple comparisons.

Results
Behavioral measures. In adults Adult Reading History Questionnaire(ARHQ) score (40.6 ± 17.9) was highly correlated with word reading (r = −0.57, p < 0.001) and pseudoword reading (r = −0.73, p < 0.001). In both reading tests adults with dyslexia performed significantly lower than adult control group (p < 0.001, see Table 1). Dyslexic adults were also significantly slower than controls at naming letters and numbers in Rapid automatized naming (RAN), but had similar performance in the objects and colors subtests. This pattern of results remained unchanged when several subjects were removed from the analysis due to spectral artifacts. It is important to mention that both ARHQ and pseudowords reading test were used to classify subjects to the groups in adult sample.
Dyslexic children had significantly worse performance than children from the control group in all behavioral tests -word and pseudoword reading, both subscales of RAN (objects & colors and letters & digits), and phonological task based on pseudowords. This pattern of results was not affected by removing subjects with spectral artifacts. Descriptive statistics and results of the statistical comparisons between the dyslexic and control groups are presented in the Table 1.
Spectroscopy -data quantity and quality. In adult group 4 spectra from visual cortex and 3 from left temporo-parietal cortex were rejected from the analysis due to strong artifacts. Two additional GABA estimates from visual cortex were rejected after GANNET preprocessing. In the end, 33 spectra from the left temporo-parietal lobe (17 dyslexic and 16 control adults) and 32 spectra from the visual cortex (16 dyslexic and 16 control adults) were being taken under consideration (30 GABA estimates from visual cortex − 15 dyslexic and 15 control adults).
In children sample, after data reconstruction 1 subject was rejected due to heavy, presumably motion artifacts in both spectra. Eight more spectra from left temporo-parietal cortex and 1 from visual cortex were declassified upon detailed visual inspection of LCModel printout. Additionally, 4 GABA estimates were rejected after preprocessing in GANNET (3 from left temporo-parietal cortex and 1 from visual cortex). Invalid spectra were misaligned during spectral registration and subsequently characterized with an abnormal shape of peaks and baseline. As a result, 43 spectra from left temporo-parietal lobe (20 dyslexic and 23 control children) and 50 www.nature.com/scientificreports www.nature.com/scientificreports/ spectra from visual cortex (25 dyslexic and 25 control children) were further analyzed (40 − 17 dyslexic and 23 control children and 49 − 25 dyslexic and 25 control children GABA estimates, respectively).
Gamma-aminobutyric acid (GABA). In both brain areas, neither main effects of age and group nor interaction between age and group reached significance for GABA.
Brain -behavior correlations. We performed correlations within the group of adults and children, combining both control and dyslexic individuals. This analysis showed that choline concentration correlated with RAN objects and colors differently in adults and children. The correlation was positive in adult sample (r = 0.516, p = 0.002** in the left temporo-parietal cortex and r = 0.449, p = 0.01 in the visual cortex), meaning the higher the concentration of this metabolite the slower naming speed (see Fig. 2). In children the correlation between RAN and choline was of a different direction, i.e. the higher the concentration of choline the faster naming speed, but it did not approach significance (r = −0.299, p = 0.052 in the left temporo-parietal cortex and r = −0.109, p = 0.45 in the visual cortex). Importantly, the difference in correlations coefficients between adults and children was significant (z = 3.64, p < 0.001 for the temporo-parietal and z = 2.51, p = 0.012 for visual cortex). Several weaker correlations were found, but they did not survive the correction for multiple comparisons. Specifically, in children there was a positive correlation between choline concentration in the visual cortex and words reading (r = 0.284, p = 0.045) and a negative correlation between tNAA in the temporo-parietal cortex and time needed to name objects and colors (r = −0.372, p = 0.014). Phonological awareness in children was also correlated with metabolite concentration, but no significant effects were found. In adults, glutamate in the left temporo-parietal cortex was positively correlated with RAN objects and colors (r = 0.367, p = 0.036 -Glu; r = 0.385, p = 0.027 -Glx).

Discussion
The current study is the first to examine brain neurochemistry in typically developing and dyslexic children and adults in the two brain regions: occipital cortex, and left temporo-parietal cortex including the angular gyrus. We tested the predictions of different etiological theories of dyslexia related to neurotransmission deficits and aimed to disentangle inconsistencies between previous findings by exploring how participants' age or different preprocessing steps (see supplementary materials) affect the results.
We found that dyslexic individuals, irrespective of age, had significantly lower tNAA than controls in the occipital cortex. This was the only group effect related to dyslexia, which survived the correction for multiple comparisons. NAA is the most prominent neurotransmitter detected in the normal human brain, which reflects neuronal density, function or viability 15 , but is also considered to be a neurochemical correlate of neuron-oligodendrocyte (axon-myelin) integrity 32 . Reduced tNAA in dyslexic subjects suggests white matter microstructure abnormalities since the concentration of tNAA is positively correlated with fractional anisotropy (FA) 33,34 . FA measures the amount of coherence of water diffusion which putatively reflects the amount of myelination in axonal bundles or the coherence of fiber tracts. Diffusion weighted studies identified lower FA mainly in the left temporo-parietal areas in dyslexia in both children and adults 21 , whereas occipital areas seem to be less frequently found 35 37 and other stimuli such as symbol strings 38 . NAA differences in reading skill were identified only in one recent MRS study, where higher NAA/Cr concentration in the occipital cortex predicted faster reaction times in a cross-modal matching task with language stimuli  www.nature.com/scientificreports www.nature.com/scientificreports/ (letters, words and pseudowords) in children 26 . This result would be in line with ours, where controls had higher concentration of tNAA in the occipital cortex than dyslexics. Del Tufo and colleagues suggested that higher NAA in the occipital cortex corresponds to more intact white matter reading network and specifically to left arcuate fasciculus in the temporo-parietal cortex, which connectivity was related to individual differences in cross-modal brain activity in developing readers. However, we found no effects of dyslexia in the concentration of tNAA in the left temporo-parietal cortex. Moreover, lower tNAA in the occipital cortex characteristic for dyslexic subjects in the current study was also not predicted by the two etiological theories of dyslexia. Neuronal noise hypothesis predicted heightened level of glutamate 3 in dyslexia. However no such effect was present in the current study. So far only one study reported elevated levels of Glu/creatine in dyslexic children in the occipital cortex 25 . Even though the stated effect size for this group difference was large (Cohen's d = 0.79), given the low sample of dyslexic children (n = 10) and no correction for multiple comparisons it is not unlikely that this estimate might be inflated 39 . Previous correlational MRS studies showed mixed results for glutamate, with positive correlation with phonological awareness in anterior cingulate in pre-reading children 27 and negative correlation for Glu/creatine with both reading performance and phonological awareness in the occipital cortex in school-age children 25 . In our study, glutamate concentration was not related to reading, phonological awareness or naming speed in children, whereas in adults higher glutamate concentration was related to slower naming speed, but this result did not survive correction for multiple comparisons.
Altered levels of glutamate together with choline would be also expected by impaired auditory sampling theory for theta oscillations, which correspond to speech syllables 12,14 . With respect to choline, our results suggest that its role in reading and dyslexia can change with age. Specifically, in adults, but not in children, higher choline in both occipital and temporo-parietal cortex was related with slower naming speed in the RAN test. This result is in line with Bruno et al. (2013), who found a negative correlation between choline in left angular gyrus and phonological skills in young adults 23 . RAN is considered to be a phonological skill reflecting the speed of access to phonological information in the long-term memory 40 . Similarly to PA, RAN is a reliable predictor of reading skills, especially in transparent orthographies such as Polish. In contrast to Pugh et al. (2014), where a negative correlation between Cho/creatine and reading skills was found in occipital cortex in children, we did not observe any significant correlations between choline and reading or reading related skills in children. Here, we found that the concentration of creatine might change with age, as adults had higher absolute concentration of creatine than children. Given this, pairwise correlations between Cho/creatine and reading skills not adjusted for age might be difficult to interpret 25 . Additionally, we found anecdotal evidence for an interaction between age and group in choline concentration both in temporo-parietal and occipital cortex with dyslexic children compared to control children having lower choline concentration (nominally significant), whereas dyslexic adults having higher choline concentration than typically reading adults. Importantly, such interaction was observed irrespective of scaling or correction for tissue composition (see Supplementary Materials). Future studies should test if this result holds in larger samples. Our results are not consistent with Rae's et al. 22 study, where Cho/NAA ratio was reported to be lower in temporo-parietal regions in adult dyslexic men compared to controls (but see Laycock et al. 2008 for findings in the opposite direction in the cerebellum). The discrepancy might be related to reference scaling to NAA, when they used scaling to creatine differences between groups became non-significant 22 . In fact, in the current study the absolute concentration of tNAA differed between dyslexic and control subjects in the visual cortex and between children and adults in the temporo-parietal cortex, yielding scaling to tNAA illegitimate. Our results stand also in contrast to previous findings of elevated levels of Cho/creatine in dyslexic children in the occipital cortex 25 . Since the results of both their and our study survived only nominal significance, future research in larger samples is needed to resolve these discrepant findings.
Meanwhile, no significant differences in GABA concentration, which drive oscillations in high frequency bands, were found in neither of the groups. GABA is still difficult to measure compared to glutamate or choline concentration and even with an improved data processing pipeline we did not come to conclusive results. The calculated Bayesian factor (alternative/null) suggested that the data were 0.296:1 for visual and 0.427:1 for temporo-parietal in favor of the null hypothesis (no differences between controls and dyslexic), or rather, 3.38 or 2.34 times more likely to occur under a model without including an effect of dyslexia, rather than a model with it. Alternatively, it would take more than 272 subjects to prove 2% difference in GABA concentration in the left temporo-parietal cortex based on power analysis. Thus GABA (if relevant) needs to be further examined in a much larger sample with a more robust protocol.
Apart from the effect of dyslexia on the concentration of neurometabolites, we observed several maturational changes. In particular adults compared to children were characterized by lower glutamate, especially in the visual cortex, in line with previous studies 30,31 . Age-related reduction in density of glutamatergic NMDA receptors 41 and glutamate-glutamine cycle flux 42 could lead to reduced glutamate driven activation and cognitive deficits. However, since glutamate reduction was related to age but not dyslexia in our sample, the decrease in metabolites is rather the result of maturation of neuronal system and its metabolism. Such an effect, reviewed by Segovia in 2001, is linked to elevated metabolic activity in children and changes in glutamatergic neurotransmission 28 . What is new in the field, higher choline levels in adults than in children were found in both visual and temporo-parietal cortex. As stated in introduction, choline is considered a marker of cell membranes in the voxel i.e. the amount of membrane turnover from breakdown or synthesis. It has been shown that choline pool measured in 1H-MRS and i.e. glycerophosphorylethanolamine and glycerophosphorylcholine measured by 31P-MRS increases with age 18 . Our results are consistent with age-related dynamics of tissue growth measured by MR relaxometry and diffusion 43 . The relation between visible choline and maturation was not well studied in humans but other authors reported increases in Cho related to aging in rats 44 . According to the their hypothesis the uptake of acetylocholine might be impaired by aging, leading to higher signal of free choline measured by MRS. The result presented here is especially interesting in the context of previously reported discrepancies in choline findings 20,22,23,25 depending presumably on age of participants (anecdotal interaction was detected here). We also found that children had (2019) 9:5096 | https://doi.org/10.1038/s41598-019-41473-x www.nature.com/scientificreports www.nature.com/scientificreports/ lower concentration than adults of tNAA in the left temporo-parietal cortex and creatine in both brain areas in line with increase measured in 5-12 y/o children 29 . To sum up, we observed increase in tNAA, Cr, Cho considered as markers of changes in cortical microstructure and Glu, a measure of increased metabolism. Thus, these results question the use of tNAA or Cr (generally thought to be stable 45 ) as a reference when comparing different age groups i.e. in dyslexia studies. It seems that the normalization of metabolites to creatine or tNAA, limits possible interpretations and future studies measuring absolute metabolite concentrations relative to water signal would be more valuable.
A potential limitation of our study is that we used slightly different acquisition parameters in the group of children -shorter acquisition to reduce number of spectra affected by motion and smaller voxel in temporo-parietal lobe to better correspond with anatomical features. However, the applied corrections for partial volume effect and locally measured water signal make the spectra comparable between the groups with a use of applied statistical tests (assuming unequal variance where relevant). Nevertheless, measuring metabolites of low signal such as GABA is still a challenge in a reasonable amount of acquisition time. We look forward to novel methods of online motion correction for longer spectroscopic acquisition.
To conclude, although the current findings do not lend support to neither of the proposed etiological theories of dyslexia, they show, for the first time, that tNAA, considered to be a neurochemical correlate of white matter integrity, is deficient in the visual cortex in both children and adults with dyslexia. They also point that several neurotransmitters, including ones previously used as reference, change with age. However, current results should be interpreted with caution given our cross-sectional data. Future studies should employ a longitudinal approach to examine developmental changes in neurometabolites ideally from the pre-reading period.

Methods
The general inclusion criteria for both adult and children samples were: good overall health, right-handedness, normal (or corrected to normal) vision, normal hearing, no history of neurological illness or brain damage and no symptoms of ADHD. All participants speak Polish as their first language. The study was approved by the Warsaw University Ethical Committee and carried out in accordance to the provisions of the World Medical Association Declaration of Helsinki. All adult participants and parents gave written informed consent to the study, and the children agreed orally.
Participants -adults. 36 adult volunteers (12 females/24 males, mean age 27.3 ± 3.8 y/o, 21.1-39.8 y/o) were included in the study. All participants completed higher education.To select subjects for dyslexic and control group, we used three different criteria and the subject had to meet at least two of them to be included in the dyslexic group. First criterion was the formal dyslexia diagnosis. However, even though 14 subjects obtained dyslexia diagnosis in childhood, dyslexia was not as commonly known and diagnosed in Poland in the 1980s. We thus decided to introduce additional criteria. The second one based on the ARHQ score, commonly used to screen for the reading problem in families in the dyslexia research. A criterion was met if the ARHQ score was higher than 40 46 . Third criterion was the pseudoword reading score. The criterion was met if the subjects obtained a score lower than group median (Med = 69, Mean = 76). 4 subjects that were not previously diagnosed with dyslexia met both criteria: for ARHQ and pseudoword reading. These subjects were also classified as dyslexics. All 14 individuals with dyslexia diagnosis in childhood also scored above 40 point cut off in the ARHQ. Then from the larger group of volunteers we selected subjects without problems with reading (subjects that met at most one of the criteria described above) matching for age and sex to the dyslexic group (see Table 1). In result, we selected 18 dyslexics and 18 controls for the analysis reported in the paper.
Participants -children. 52 children (22 females/30 males, mean age 11.06 ± 0.97 y/o, range 9-12.9 y/o) from third to fifth grade of elementary school were recruited from the larger sample taking part in a different study at the Nencki Institute. The dyslexia diagnosis 47 and the IQ assessment (Wechsler Intelligence Scale for Children 48 ) was conducted within this larger study. We selected 26 children with dyslexia diagnosis, and matched (by age and gender) 26 typically reading controls. All children had average or above-average IQ. The two groups did not differ in age or sex (see Table 1). Additionally, we made sure that the children were born at term (≥37 weeks).

Behavioral measures.
Apart from neuroimaging all subjects were tested on words and pseudowords reading speed 49 as well as rapid automatized naming -RAN 50 with two subtests: objects & colors and letters & digits. The amount of words and pseudowords read per minute was compared between the groups. For the RAN test, the summed time for naming objects and colors as well as letters and digits subtests was compared. All children were tested with a battery of phonological awareness tasks based on pseudowords in which both syllable and phoneme manipulation level was included 47 . Results of behavioral measures are presented in the Table 1. MR Spectroscopy and Imaging. MRI was performed on Siemens Magnetom TRIO 3 T using 32CH receiving coil. MEGA-PRESS spectroscopy 51 (Siemens research agreement WIP-529) was acquired from two locations -occipital (visual) cortex and left temporo-parietal cortex. Voxels were carefully placed based on 3D T1-weighted images (TR = 3000 ms, TI = 900 ms, 1 × 1 × 1 mm resolution) reconstructed in 3 planes (see Fig. 3).
Spectroscopy acquisition was done with TR = 2000 ms and TE = 68 ms. Acquisition protocol was validated in the group of adults with 192 averages for both edited and non-edited spectra (NA = 192*2). A special attention was paid to quality of MEGA-PRESS edited spectrum thus longer acquisition time was used to better estimate differential spectrum of GABA. However, spectra averaged across >10 min were frequently corrupted due to bad initial calibration, field drift or subjects' motion. Further, we decided to shorten the acquisition in the whole group of children to NA = 96*2 (from 12 m 50 s to 6 m 25 s) and repeat whole measurement in case of strong artifacts (with the subject's consent). We think that such modification was legitimate, since all metabolites of interest apart from GABA still withstood conservative threshold of CRLB < 15% 52 (Cramér-Rao lower bound -an error of (2019) 9:5096 | https://doi.org/10.1038/s41598-019-41473-x www.nature.com/scientificreports www.nature.com/scientificreports/ signal modeling used for estimating maximum variance in metabolites concentration). The acquisition parameters were comparable to recent studies where standard PRESS sequence was used 23,27 . In adults the voxel size was 30 × 30 × 15 mm in both the visual cortex and left temporo-parietal cortex. In children the voxel was of the same size in visual cortex while the size was slightly reduced in temporo-parietal lobe (from 30 × 30 × 15 to 25 × 25 × 15 mm) to better correspond with angular gyrus anatomical features.
Water unsuppressed scan (NA = 16) was acquired to reference the metabolites concentrations. T1-weighted volumes were segmented in SPM8 (New Segment) and coregistered with individual voxel masks in order to correct for partial volume effects. Spectral analysis was performed in Gannet 53 -GABA and LCModel 53 -alanine, aspartate, creatine, GABA, Glu, Gln, GSH, choline compounds (Cho), lactate, myo-inositol, NAA, NAAG, scyllo-inositol, and taurine. Spectra were assessed by a spectroscopist with a regard to SNR, shape of baseline, residual noise, symmetry and width of singlets (NAA, Cr) and voxel position. GABA estimation was performed in Gannet with the raw data (Siemens TWIX) with the following steps: (1) coil channel combine and phase correction, (2) frequency drift correction, (3) calculation of differential spectrum, (4) fitting with the gaussian peak. (5) Then, the results of GABA were corrected for partial volume effects in respect to CSF, GM/WM ratio as well as tissue specific relaxation times 54 . Some spectra were misaligned during GANNET preprocessing and also disqualified from analysis.
LCModel was used to calculate absolute metabolite concentration based on MEGA-OFF spectrum (Fig. 4). Siemens RDA (spectra averaged on the scanner) were fitted with the simulated basis set consisting of 14  www.nature.com/scientificreports www.nature.com/scientificreports/ metabolites 55 . LCModel's reference water concentration was corrected based on properties of voxel specific tissue volume fractions 56 .

Metabolites of interest and confounding factors.
According to presented literature we hypothesised that Cho, Glu (and Glx), NAA or GABA concentration would differ in dyslexia. Additionally, we tested whether Cr level can be utilized as a reference.
Type of analysis, multiple comparison correction. We used a 2 × 2 ANOVA model with group (dyslexic vs. control) and age (adult vs. children) as factors. Post-hoc pairwise comparisons were performed if main effects or interaction were found significant. In order to compare current findings with the previous literature 20, 22,25 we report all effects surviving the nominal significance of p < 0.05, but focus more on results surviving more stringent statistical threshold corrected for multiple comparisons (5 metabolites) resulting in nominal threshold of p < 0.01. The results that survive correction for multiple comparisons are marked with 2 or more asterisks. Bayesian ANOVA was performed for estimating ratio of the likelihood probability of two competing hypotheses -Bayes factor 57 . Here we assume that for BF 10 > 3 there is a substantial evidence for alternative hypothesis comparing to null and symmetrically for BF 10 < 1/3 for null comparing to alternative. Bayes factor does not require multiple comparison correction 58 .
Applied corrections and confounding factors. Different scaling methods may potentially lead to bias results i.e. when considering age and voxel position. We decided to show fully corrected concentrations for relaxation, tissue composition and scaled with water concentration within a voxel as a main result. Additionally, we were obliged to test for any effect of scaling methods used previously by referenced authors such as: scaling to tNAA or Cr and less sophisticated tissue correction. Such information is presented in supplementary materials for the purpose of comparisons with published studies and thus was not taken into account in multiple comparison correction.
Brain behavior correlations. Furthermore, similarly to previous studies 23, 25 we performed Pearson's correlation between Cho, Glu, Glx, GABA, tNAA and behavioral tests (words and pseudowords reading and RAN). The correlations were performed separately for adults and children.

Data Availability
The datasets generated and analyzed during the current study are available from the corresponding authors on request.