Abstract
Emerging research has provided valuable insights into the structural characteristics of the bilingual brain from studies of bilingual adults; however, there is a dearth of evidence examining brain structural alterations in childhood associated with the bilingual experience. This study examined the associations between bilingualism and white matter organization in bilingual children compared to monolingual peers leveraging the large-scale data from the Adolescent Brain Cognitive Development (ABCD) Study. Then, 446 bilingual children (ages 9–10) were identified from the participants in the ABCD data and rigorously matched to a group of 446 monolingual peers. Multiple regression models for selected language and cognitive control white matter pathways were used to compare white matter fractional anisotropy (FA) values between bilinguals and monolinguals, controlling for demographic and environmental factors as covariates in the models. Results revealed significantly lower FA values in bilinguals compared to monolinguals across established dorsal and ventral language network pathways bilaterally (i.e., the superior longitudinal fasciculus and inferior frontal-occipital fasciculus) and right-hemispheric pathways in areas related to cognitive control and short-term memory (i.e., cingulum and parahippocampal cingulum). In contrast to the enhanced FA values observed in adult bilinguals relative to monolinguals, our findings of lower FA in bilingual children relative to monolinguals may suggest a protracted development of white matter pathways associated with language and cognitive control resulting from dual language learning in childhood. Further, these findings underscore the need for large-scale longitudinal investigation of white matter development in bilingual children to understand neuroplasticity associated with the bilingual experience during this period of heightened language learning.
1 | INTRODUCTION
A growing body of literature has started to investigate to what extent brain structure is associated with learning and use of two languages (Green & Abutalebi, 2013; Hernandez et al., 2018; Li et al., 2014; Pliatsikas et al., 2020; Pliatsikas & Luk, 2016). Historically, research examining how the brain is shaped by language acquisition has predominantly focused on monolingual speakers (Friederici, 2011; Lee et al., 2007), but approximately half of the world's population is bilingual or multilingual (Grosjean, 2021). Importantly, bilinguals are not two monolinguals in one person (Grosjean, 1985), rather there is a broad range of heterogeneity in the fluency and balance bilinguals have in their two languages, associated with different exposures and contexts in which they use their languages (Grosjean, 2021). An increasing number of studies suggest that bilingual adults compared to monolingual peers show characteristic structural brain alterations in cortical and subcortical structures and corresponding white matter pathways (Hayakawa & Marian, 2019). These findings suggest that bilingual experience may interact natural maturational processes in similar ways as other types of long-term experiences to significantly shape the structure of the brain (Bermudez et al., 2009; Maguire et al., 2000; Mills & Tamnes, 2014; Tamnes et al., 2010).
White matter pathways are one brain structural characteristic which provide a vehicle to examine experience-induced neuroplasticity (Mechelli et al., 2004). The process of white matter maturation is characterized by reorganization and plasticity and is known to be shaped by environmental input and experience. Recent studies using diffusion tensor imaging have revealed that throughout development, there are continuous increases of fractional anisotropy (FA) in white matter as a result of maturation (Giedd, 2004; Lebel et al., 2019; Lebel & Deoni, 2018) and in conjunction with the acquisition of new skills (Scholz et al., 2009). White matter, comprised of myelinated axon bundles, facilitates efficient neuronal communication across the brain and enables the acquisition of corresponding skills. White matter differences in language-related pathways have been established in lifelong bilingual older adults compared to their monolingual peers (Gold et al., 2013; Luk et al., 2011; Schweizer et al., 2012). Highly immersed bilingual young adults also show structural differences in white matter tracts compared to their monolingual peers suggesting that learning and actively using a second language can induce white matter alterations regardless of when the second language is acquired (Pliatsikas et al., 2015). However, there is a limited body of research examining the effects of bilingualism-induced structural plasticity from a developmental perspective as bilingual children learn their languages. To further understand these effects during development, the present study examines white matter organization in school-age bilingual children compared to a group of rigorously matched monolingual peers leveraging the large-scale data set from the Adolescent Brain Cognitive Development (ABCD; Casey et al., 2018).
1.1 | Pathways underlying language networks
Previous research studies have identified neural networks related to the core functions of language. These language networks are generally organized around dorsal and ventral white matter pathways that connect language-relevant areas in the temporal and parietal lobes with areas in the frontal lobe (Friederici & Gierhan, 2013). One prominent bilateral dorsal pathway is the superior longitudinal fasciculus (SLF), which includes the arcuate fasciculus, and connects language and reading areas of the temporal cortex and the parietal cortex to the frontal cortex ipsilaterally (hereafter referred to as SLF; Catani et al., 2002). The dorsal pathway is thought to support several aspects of language processing, such as mapping phonemic representations to motor representations (Glauche & Abel, 2008; Middlebrooks et al., 2017), processing complex syntactic structures (Friederici, 2012; Wilson et al., 2011), and supporting phonological planning and processing for language and reading processing (Janelle et al., 2022; Middlebrooks et al., 2017; Wandell & Yeatman, 2013). Ventrally, there are also two major bilateral pathways that have been associated with the language network connecting areas of the frontal cortex to the temporal, parietal, and occipital cortices. The inferior frontal-occipital fasciculus (IFOF) forms a direct ventral pathway extending from the inferolateral and dorsolateral frontal cortex with posterior areas of the temporal and occipital lobes (Catani et al., 2002; Friederici & Gierhan, 2013; Vigneau et al., 2006). The indirect ventral pathway is composed of the inferior longitudinal fasciculus (ILF) and the uncinate fasciculus (UF), connecting the temporal pole to the occipital cortex and the inferior frontal gyrus to the superior temporal gyrus. This pathway has been proposed to play a role in reading, visualorthographic, semantic processing, and simple syntactic processing (Friederici & Gierhan, 2013; Griffiths et al., 2013).
1.2 | Theoretical models for bilingualism-induced neuroplasticity
Models of bilingualism posit that different aspects of the bilingual experience (e.g., diversity of language use contexts, language switching, degree and duration of second language use, relative proficiency) lead to experience-induced structural changes. These models suggest that the brain may enhance or repurpose existing structures and networks through continuous interaction with the environment in response to changing bilingual language exposure and use (Claussenius-Kalman et al., 2021; Hernandez et al., 2018, 2019). The adaptive control hypothesis (Abutalebi & Green, 2007; Green & Abutalebi, 2013) suggests that for successful language switching and language control based on the conversational context, bilinguals recruit several cognitive processes (i.e., for goal maintenance, interference suppression, monitoring, and opportunistic planning) during language processing. These cognitive control processes engage a wide network of cortical and subcortical areas of the brain (the bilateral inferior frontal gyrus, the anterior cingulate cortex [ACC], the inferior parietal lobe, the basal ganglia, the thalamus, and the cerebellum, to name key areas) and the white matter connections between them. These connections include the corpus callosum (CC; for interhemispheric communication anteriorly via the forceps minor, medially through the midbody of the CC, and posteriorly through the forceps major); the anterior thalamic radiation (ATR; connecting anterior and medial portions of the thalamus to the frontal cortex) and the cingulum (linking medial areas in the frontal, parietal, and temporal lobes with the parahippocampal region, the thalamus, and the ACC). The conditional routing model (Stocco et al., 2010, 2014), posits that the basal ganglia (the caudate and putamen, in particular) are trained through the continuous use of two or more languages. The brain is then able to shortcut cortical-to-cortical connections using connections through the basal ganglia during language processing for more efficient language selection and language switching.
Finally, the dynamic restructuring model (Pliatsikas, 2020) offers a framework of stages for the experience-induced plasticity observed with increasing exposure and proficiency in a second language. The first stage of initial exposure to a second language is characterized by increased cortical gray matter in structures implicated in language, cognitive control, and short-term memory. With increased experience and immersion in the second language and a shift toward increased efficiency in language processing, bilinguals reach the second stage, consolidation, exemplified by increased subcortical and cerebellar gray matter, and return to baseline levels of cortical gray matter. Consolidation and efficiency are also reflected through increased plasticity in white matter pathways connecting cortical and subcortical regions. Finally, as experience leads to automation of second language use and control, the peak efficiency stage is represented by increased cerebellar gray matter and a shift from processing in anterior to posterior regions. These models of bilingualism, supported by empirical evidence, explain brain alterations characteristic of the bilingual brain in adulthood, for lifelong bilingual adults and late learners of a second language as they increase proficiency in the language. However, there is limited evidence on whether and how these putative bilingual characteristics develop in childhood, considering the variability of language use and proficiency as children acquire and learn their languages during a heightened period of language immersion and brain plasticity.
1.3 | White matter in bilingual adults
The field has acquired valuable insights into our conceptual understanding of the structural characteristics of the bilingual brain from studies of bilingual adults. Pliatsikas et al. (2015) found that bilingual young adults (Mage = 29.8 years; SDage = 6.5 years) who were late but highly proficient learners of English had higher FA (a measure of the directionality of water diffusion in white matter tracts; Christodoulou et al., 2017; Vandermosten et al., 2012) in pathways connecting cortical and subcortical structures (the CC, bilateral SLF, right IFOF, and right UF) compared to age-matched native speakers of English. In a follow-up analysis from the same data, Rahmani et al. (2017) used diffusion magnetic resonance imaging (dMRI) connectometry to measure quantitative anisotropy (QA) and found increased connectivity in selected pathways (i.e., the CC, bilateral cingulum, bilateral SLF, and the left IFOF) for the bilinguals compared to the monolinguals. Further, correlation analysis revealed that QA for all these white matter pathways, except for the cingulum, had a direct positive correlation with the duration of the language immersion period of the bilingual group. Rossi et al. (2017) investigated white matter differences in university students (age range = 18–27) in a group of monolingual English speakers compared to late learners of Spanish. The researchers found that the bilinguals significantly presented higher FA values in a broad network of left hemisphere white matter pathways, including the ATR, the left IFOF, the left UF, and the left ILF. Moreover, the FA values in these tracts were positively correlated with the age of acquisition of the second language. These findings suggest that neural plasticity, and in particular FA values as a measure of white matter organization, may be directly impacted by the amount and length of immersion in a bilingual environment and the level of proficiency acquired in the second language.
However, in populations of older bilingual adults (aged 65 and older), research has provided contradicting results. Luk et al.'s (2011) results reflect the pattern of differences in FA from the young adult studies by Pliatsikas et al. (2015) and Rossi et al. (2017) in older lifelong bilinguals compared to their age-matched monolingual peers (Mage = 70.5 years; SDage = 3 years). However, in a similar study, Gold et al. (2013) reported significantly lower mean FA values in white matter pathways (i.e., IFOF, IFL, fornix, and in multiple areas of the CC, including the forceps major) in lifelong bilingual older adults (Mage = 64.3 years; SDage = 4.8 years) compared to monolingual controls. Cross-sectional findings from studies of older adults, lifelong bilinguals compared to monolinguals, can be difficult to interpret because of lifelong variability in language use and language proficiency. Therefore, to clarify these mixed findings, it is of utmost importance to (a) compare bilingual and monolingual white matter organization earlier in development, and (b) account for variability in language use and proficiency within bilinguals as it relates to white matter structure.
1.4 | White matter in bilingual children
There are only a handful of studies examining white matter organization in bilingual children, which have also yielded mixed findings likely due to small sample sizes and heterogeneity in the ages of participants (Pliatsikas et al., 2020). In an initial study, Mohades et al. (2012) compared mean FA in four major pathways (i.e., left SLF, left IFOF, the forceps minor, and the midbody of the CC) in forty children (age range = 8–11) subdivided into three language groups: 15 simultaneous bilinguals (exposed to both languages since birth), 15 sequential bilinguals (acquired second language after age 3), and 10 monolinguals. No significant differences were found between the groups in the left SLF or the midbody of the CC. However, the researchers found significantly greater FA values in the left IFOF for the simultaneous bilinguals compared to both the sequential bilingual and the monolingual children and significantly lower FA values in the forceps minor in the simultaneous bilinguals compared to the monolinguals. Interestingly, when the children were retested after 2 years (when children were ages 10–13 years old), the researchers found that mean FA values in the left IFOF had increased for all groups, but that the increase over time was significantly higher for the simultaneous bilinguals (Mohades et al., 2015). The researchers interpreted this finding as maturation and myelination of a key pathway for semantic processing resulting from increased exposure and experience to using more than one language.
Only one other study to date has examined white matter organization in bilingual children. Pliatsikas et al. (2020) conducted a crosssectional study of brain development from childhood through young adulthood using a sample of 127 bilinguals (age distribution: 11 3– 6 years old, 37 6–10 years old, 20 10–14 years old, 31 14–18 years old, and 28 18–21 years old) compared to 510 monolinguals (age distribution: 84 3–6 years old, 154 6–10 years old, 125 10–14 years old, 89 14–18 years old, and 58 18–21 years old) from the Pediatric Imaging, Neurocognition, and Genetics dataset (Jernigan et al., 2016). The researchers found that within both language groups (i.e., bilinguals and monolinguals), there was evidence of reduced cortical gray matter (measured by cortical thickness) and enhanced white matter organization (measured by higher mean FA values) in older participants, reflecting the expected general developmental patterns of brain plasticity. Compared to age-matched monolingual participants, bilinguals tended to have thicker cortical gray matter during late childhood and adolescence (particularly in frontal and parietal areas of the brain) and higher FA values starting in mid-adolescence. Only one white matter tract, the striatal inferior frontal cortex (striatal IFC; which connects the striatum of the basal ganglia to areas in the inferior frontal gyrus), revealed significantly lower FA values for younger bilinguals compared to age-matched monolinguals with the groups converging until age 16 when bilinguals started showing significantly greater mean FA values than monolingual peers. These findings from Pliatsikas et al. (2020) suggest that although bilinguals, similar to monolinguals, follow certain expected developmental patterns of structural brain maturation, there may be key differences in trajectories. These differences suggest slower developmental pruning of frontal and parietal cortical areas and faster developmental maturation, and a protracted increase in white matter organization in bilinguals compared to monolinguals. Thus, while these studies provide initial evidence of white matter organization in bilinguals earlier in development, the existing empirical evidence is based on relatively small sample sizes. The present work will build on this evidence by accounting for language use and proficiency in a large-scale investigation of white matter organization in bilingual children compared to a group of rigorously matched monolingual peers.
1.5 | Goals of the present study
To address inconsistencies due to small sample sizes and insufficient characterization of language use and proficiency in bilinguals, the present study will leverage the large-scale dataset from the ABCD Study (Casey et al., 2018; Yang & Jernigan, 2021) to conduct an investigation on the relationship between bilingualism and white matter organization in bilingual children and their monolingual peers through the following research questions:
How does white matter organization (as measured by FA values) differ between bilingual and monolingual children in select language and cognitive control pathways? We hypothesized that at this age (9–10 years old), bilingual children compared to their monolingual peers would show protracted maturation across pathways related to language and cognitive control, reflected by lower FA values in these tracts when controlling for developmental and demographic factors.
What are the differential effects of demographic and developmental factors on white matter organization within each language group (bilingual, monolingual)? We expected that there would be widespread effects of age on white matter organization in both groups, in line with documented reports of higher FA values among bilingual adults (Luk et al., 2011; Pliatsikas et al., 2015; Rossi et al., 2017). We also hypothesized that (i) more use and exposure to a language other than English and (ii) proficiency in English would impact white matter organization in bilingual children, with more use and exposure to the other language and higher vocabulary scores associated with higher FA values.
2 | MATERIALS AND METHODS
The secondary analyses for this cross-sectional study were reviewed and approved by the Institutional Review Boards at the University of Houston, the University of Texas Health Science Center at Houston, and Boston University. Data were de-identified and obtained from the National Institutes of Mental Health (NIMH) Data Archive for the baseline timepoint from the ABCD Study, release 4.0 (Yang & Jernigan, 2021). For all participants in the study, parental informed consent and youth signed assent were obtained by the ABCD Study team (Garavan et al., 2018).
2.1 | Participants
The ABCD Study is a national longitudinal research effort to examine psychological and neurocognitive developmental data from over 11,500 children recruited at ages 9–10 through young adolescence to young adulthood. Data has been collected at 21 sites across the United States with care to, as best as possible, reflect the sociodemographic makeup of the US population using probability sampling in schools within defined areas for each site (Garavan et al., 2018). The full study battery involves data collection of neurocognitive measures and brain imaging from participating children (MRI session) and interviews and questionnaires for participating children and their parents. Data used for the current study is part of the baseline timepoint in data release 4.0 (Yang & Jernigan, 2021) collected between 2018 and 2020.
Our objective in selecting participants for the current study was to select a group of bilingual children and a group of comparable monolingual children from the ABCD Study data. Only participants with complete data for all our measures of interest and covariates were included in the eligible sample. From the participants with data for the baseline time point (n = 11,876), participants with no abnormal findings were selected (resulting in n = 11,235). Only participants who passed all inclusion criteria for dMRI accounting for protocol compliance and imaging quality control procedures were included, resulting in n = 8759. Finally, participants were excluded from the final samples for the following reasons: participants with no Parent Longitudinal Demographic Questionnaire information from the 1-year follow-up, necessary for language group determination (n = 481); those with no complete data for the NIH Toolbox measures (n = 1203); and those with no data for the other covariates (handedness inventory, nonverbal IQ, and pubertal status). This resulted in an overall sample of n = 5127 for eligible participants.
The language group classification (i.e., into bilingual and monolingual groups) was done using the surveys completed by parents and children following procedures from by Dick et al. (2019), where bilinguals were selected as those whose parents reported that either (a) the child's native language was not English, or (b) if the child's native language was English, the parents reported that the nonEnglish language was used frequently (i.e., equally or more often than English). As a result, 549 bilingual children met our criteria, reporting using their respective non-English language consistently or at least as frequently as English with their friends and family at the onset of the study. The eligible monolingual group was selected using the following criteria from parental report: (a) English was the child's native language, (b) English was used more than any other language in the home, and (c) the child had never been enrolled in a dual-language program. This produced an eligible group of monolinguals of 2945 children from which to match our group of bilinguals.
From this group of all 3494 children, an additional 348 children were excluded because they contained outliers (±3 SD from the full sample mean) in NIH Toolbox measures (n = 75) or parental education (n = 34) and or who were missing demographics data to be used for matching (n = 239). This resulted in a group of 446 bilinguals and 2700 monolinguals with complete data to be used as inputs for propensity score matching. Propensity score matching using the MatchIt package in R (Ho et al., 2011; Randolph et al., 2014) was utilized to select the monolingual comparison group (n = 446) from the group of all 2700 eligible monolingual participants using the nearest-neighbor matching method with age, sex, parental education, household income, handedness, pubertal status, nonverbal IQ, and the gender of the reporting parent as matching characteristics.
2.2 | Measures
2.2.1 | Demographics
Demographic information for participants and their families was collected through a questionnaire based on items from the PhenX toolkit (Hamilton et al., 2011; Stover et al., 2010) and the general social survey (Smith et al., 2012). We used selected questions to determine the child's native languages, the home language environment (i.e., frequency of languages spoken in the home), the type of school program the child has attended (e.g., dual-language programs) in addition to other demographics questions including parent race and ethnicity, parental education, and family income. Basic demographic data from baseline timepoint was used, but since information about the child's native language and home language was only added to the questionnaire at the 1-year follow-up session, this was the time point used for the language variables.
2.2.2 | Language use
Additional measures of language use (proficiency and preferences) were obtained through the Youth Acculturation Questionnaire, which is a subset of questions from the PhenX Acculturation protocol (Hamilton et al., 2011; Zucker et al., 2018) with a version completed by parents and similar questions completed by children. These items asked parents and children about proficiency in English (i.e., poor, fair, good, excellent), if they speak or understand a language other than English, and, if so, which language is spoken most with friends and which language is spoken most with family (both of these were answered on a 1–5 point scale from 1 equal to “other language all of the time” to 5 equal to “English all of the time”).
2.2.3 | Pubertal status
The ABCD dataset includes measures of the Pubertal Development Scale (Petersen et al., 1988) completed by the children to self-assess the stage of pubertal development. This questionnaire asks about development for boys and girls on growth (e.g., height), body hair, skin change; facial hair, and voice change for boys only; and breast development and menarche for girls only. Responses for each of these questions are on a 4-point scale as follows: 1 = no development, 2 = development has begun, 3 = development is underway, and 4 = development is complete (menarche for girls was coded dichotomously: 1 = pre-menarcheal and 4 = post-menarcheal). These were all summed and divided by the number of questions to maintain the range from 1 to 4. In previous research, measures of youth self-assessment of pubertal status scores obtained from the Pubertal Developmental Scale have been reported to have moderate to good correlation with pediatrician physical exam (Barch et al., 2018; Shirtcliff et al., 2009). The youth self-assessment score of pubertal status was matched between groups and used as a covariate in all analyses.
2.2.4 | Handedness
The Edinburgh Handedness Inventory—Short Form (Veale, 2014) was administered as a measure of handedness. This version contains four items that ask children the hand used for writing, throwing, using a spoon, and using a toothbrush. Each item is rated on a five-point scale (from always right to always left hand). A laterality quotient is derived from these measures to identify participants as right-handed, lefthanded, or ambidextrous.
2.3 | Neurocognitive behavioral measures
2.3.1 | English vocabulary
The NIH Toolbox Picture Vocabulary Test (Gershon, Bleck, & Nowinski, 2013; Gershon, Slotkin, et al., 2013) is a variant of the Peabody Picture Vocabulary Test (PPVT; Dunn & Dunn, 2007), which was used as a measure of receptive vocabulary in English. During this assessment, administered on an iPad, children are presented with four pictures on the screen and are asked to touch the picture most closely representing the word they hear from the audio recording. The task uses computerized adaptive testing to ensure appropriate difficulty. This task has shown good test–retest reliability (ICC = 0.81) in validation with children and adolescents, and good convergent validity with the PPVT (Mungas et al., 2014). For this study, we used the agecorrected standard scores provided by the ABCD Study team with a mean of 100 and a standard deviation of 15.
2.3.2 | Nonverbal IQ
The Q-Interactive version (Daniel et al., 2014) of the Matrix Reasoning subtest of the Wechsler Intelligence Scale for Children, 5th Edition (Wechsler, 2014) was used as a measure of nonverbal intelligence. In this task, participants are presented with an incomplete array of visuospatial stimuli on an iPad. The participants must select one of the four options to complete the array. The task consists of 32 possible trials, but testing is discontinued after the participants select incorrect options for three consecutive items. Raw scores are calculated based on the number of correct responses and transformed into normative standard scores with a mean of 10 and standard deviation of 3, which were used for this study.
2.4 | Neuroimaging measures
2.4.1 | MRI acquisition
Imaging for the ABCD Study was obtained with the primary aim to examine brain development from childhood through adolescence (Casey et al., 2018). Data were collected across 21 sites using an optimized MRI protocol harmonized to be compatible across three 3 T scanner platforms (Siemens Prisma, General Electric 750, and Phillips). The MRI protocol includes fixed order of scan types starting with a localizer task, 3D T1-weighted images, two runs of resting-state fMRI, diffusion-weighted images, 3D T2-weighted images, another one to two runs of resting-state fMRI, and a taskbased fMRI session. During the structural MRI scans, the children were asked to watch a movie.
2.4.2 | MRI preprocessing
All MRI data collected for the ABCD Study were centrally preprocessed and analyzed by the ABCD Data Analysis and Informatics Center using the multimodal processing stream toolbox at the University of California, San Diego (Hagler et al., 2019), which in turn relies on other publicly available neuroimaging processing packages (FreeSurfer [Fischl, 2012] and FMRIB Software Library [Jenkinson et al., 2012]). T1-weighted images were processed in FreeSurfer version 5.3 (Fischl, 2012) to provide cortical surface reconstruction and subcortical segmentation using the Desikan atlas (Desikan et al., 2006). After preprocessing correction, dMRI data were coregistered to the T1-weighted images. Fiber tract segmentation was performed using AtlasTrack (Hagler et al., 2009) to derive the average FA values for 17 bilateral white matter and 3 commissural regions of interest. Preprocessing steps, including corrections for distortions and head motion and cross-modality registrations, were performed by the ABCD Study team (Hagler et al., 2019). We followed the criteria for dMRI data inclusion to ensure that all participants had reliable and complete dMRI data which had passed all automated and manual quality control checks (Yang & Jernigan, 2021).
2.5 | Data analyses
All data analyses were conducted using RStudio version 2022.07.0-548. Rstudio was used in analyses for subject selection, for extracting demographics and language variables from the applicable sources in the ABCD dataset, and to calculate descriptive and correlation information.
For the first research question, we compared white matter organization in selected language network and cognitive control white matter pathways in bilingual children and a group of monolingual children matched on age, sex, parental education, household income, handedness, pubertal status, and nonverbal IQ. Multiple regression models were used to predict mean FA values for each tract with language group (bilingual or monolingual) as a predictor while also controlling for demographic and developmental variables (i.e., age, sex, parental education, household income, handedness, pubertal status, and nonverbal IQ) and English language proficiency (i.e., English vocabulary scores). The model for a sample FA tract (i.e., 01 L SLF) is specified in Equation 1:
FA01 L SLF= β0 þβ1 x age + β2 x sex +β3 x edu + β4 X inc + β5 X handedness + β6 X pubst + β7 X NVIQ + β8 x Eng_voc+β9 x lang_grp
This model was used to predict FA for all the white matter tracts selected for this study.
To answer the second research question, we examined the association of the demographic and developmental characteristics with white matter organization (using FA values) within each language group (i.e., bilingual or monolingual). English language proficiency (as measured by English vocabulary) was included in the regression models for both groups, and language use with family and friends was included in the models for the bilingual group (as language use information was only available for participants who reported they spoke a language other than English).
Language and cognitive control related white matter tracts of interest were selected for these analyses based on the previous findings in the literature on pathways related to language and cognitive control in both bilinguals and monolinguals. The bilateral corticospinal tract (CST) was added as a nonlanguage dominant control tract (for a summary of the selected white matter pathways for these analyses, see Table 1). Covariates controlled in the model were selected because of effects on structural brain development found in prior research related to developmental maturation (i.e., age, sex, pubertal status, handedness, and nonverbal IQ; Budisavljevic et al., 2021; Giedd et al., 1999; Herting et al., 2012) and socioeconomic/ demographic differences usually observed when comparing bilinguals and monolinguals in the United States (i.e., parental education and household income as proxies for socioeconomic status; Brito et al., 2018). Finally, we used English vocabulary as a measure of language proficiency, which could influence the structure of the language-related white matter tracts and language use in only bilinguals to account for heterogeneity of bilingual experiences which has been shown to affect white matter structure in bilingual adults.
Skewness and kurtosis for mean FA values in all white matter tracts were calculated to ensure the data met the assumption of normality. For all pathways examined, skewness values ranged between 0.98 and 0.06, and kurtosis values ranged between 0.12 and 3.58. These values fall within the ranges of 2 for skew values and 7 for kurtosis recommended by Kim (2013) to be used as reference values for the assumption of normality in sample sizes greater than 300. The variance inflation factor (VIF), tolerance statistics, and the Durbin– Watson test were used to check for multicollinearity and independence of errors in our predictors. The mean VIF for our data were 1.22, and tolerance statistics were all above 0.2, indicating no collinearity within our data (Field et al., 2012). The Durbin-Watson statistics calculated for the regression models ranged between 1.81 and 2.10, suggesting the data met the assumption of independence of errors. In a final step, to control for multiple comparisons, false detection rate (FDR) corrections were applied to all coefficients across all analyses (Benjamini & Hochberg, 1995). Results were considered significant at FDR-corrected values of α ¼ :05 or lower.
3 | RESULTS
3.1 | Descriptive statistics and correlations for demographics and language variables
Demographic and control variables for the bilingual and monolingual groups are summarized in Table 2. After using propensity matching to select the monolingual group participants, the bilingual and monolingual groups did not significantly differ in age, sex, handedness, pubertal status, nonverbal IQ, parental education, combined household income, or gender identification of the reporting parent, as also reported in Table 2. Before the regression analyses, we examined the relationship between the control variables with English vocabulary in both language groups (bilinguals and monolinguals) and language use within the bilingual group.
The bilingual and monolingual groups significantly differed in English vocabulary scores, although both groups averaged above the mean of the standardized NIH Toolbox picture vocabulary assessment of 100 (MBil = 103.3, SDBil = 15.3; MMono = 106.7, SDMono = 16.3; t (890) = 3.246, p = .001). There was no significant relationship between English vocabulary and age, sex, or handedness in the bilingual group, as indicated in Table 3. However, in the monolingual group, English vocabulary was significantly related to the age (r(444)= 0.13, p = .01) of participants. English vocabulary was negatively correlated with pubertal status in both language groups (monolinguals: r(444) = .17, p < .001; bilinguals: r(444) = .10, p = .04), but positively correlated with nonverbal IQ (monolinguals: r(444) = .37, p < .001; bilinguals: r(444) = .32, p < .001), parental education (monolinguals: r(444) = .30, p < .001; bilinguals: r(444) = .37, p < .001), and household income (monolinguals: r(444) = .39, p < .001; bilinguals: r (444) = .40, p < .001). For the bilingual group only, we examined correlations of matching covariates with the combined language use with family and friends, which was measured on a scale from 0 to 8 (where 0 indicates English was most frequently used, 4 indicates a balanced use of languages, and 8 indicates the other language was most frequently used). Bilinguals, on average, reported a balanced use of their languages with a mean of 3.1 (SD = 1.7) on this scale. Bilinguals who spoke more of their other language with family also tended to speak their other language with friends (r(444) = .27, p < .001). Further, bilinguals who used more of their other language with family and friends, tended to score lower on the English vocabulary measure (r (444) = .22, p < .001), although this was a small negative correlation suggesting that these variables are measuring different aspects of language. There was no significant relationship between language use with family and friends and age, sex, handedness, or pubertal status. In general, nonverbal IQ, parental education, and household income were negatively correlated with language use with family and friends. For an overview of these associations, see Table 3.
3.2 | White matter organization differences between bilinguals and monolinguals
The first research question aimed to examine differences in white matter organization between bilingual and monolingual children in the selected pathways of interest implicated in language and cognitive control. To do this, we used multiple regression analyses and controlled for English vocabulary and the matching covariates (i.e., age, sex, handedness, pubertal status, nonverbal IQ, parental education, and household ncome) to examine the estimated impact of language group (i.e., bilingual or monolingual) on mean FA values in each white matter pathway of interest. The only three predictors that were significant in predicting adjusted mean FA values across all models for the selected pathways were age, sex, and language group. Age was a significant predictor for mean FA values for most of the models we analyzed with β-values ranging between .0002 and .0009 (FDR-corrected p0 s < :05). This result suggests that age has a significant yet small positive association with adjusted mean FA for these pathways (see Table S1 in Supplementary Materials) over and above all the other predictors. Sex (reference group: male) was a significant predictor for the right SLF (including the tSLF and pSLF portions) and the cingulum bilaterally with β-values ranging between .0101 and .0048 (FDR-corrected p0 s < :05), indicating that in this sample lower FA was observed within females for these pathways (see Table S1 in Supplementary Materials) when controlling for the other predictors. Finally, we found that language group (i.e., being bilingual or monolingual) was a significant predictor for adjusted mean FA in bilateral language network pathways, the SLF (including tSLF and pSLF portions) and the IFOF with β-values ranging between .0108 and .0044 (FDR-corrected p0 s < :05; see Table 4). There were also differences between bilingual and monolinguals adjusted mean FA values in the forceps major, and in the right hemisphere cingulum and parahippocampal cingulum pathways (see Figures 1 and 2). Notably, across all white matter pathways with significant differences between bilinguals and monolinguals, bilingual children had lower adjusted mean FA values than monolingual children (the model-predicted adjusted mean FA values for each white matter tract by language group are detailed in Table 4).
3.3 | Demographic and developmental factors impact on white matter by language group
The second research question explored the differential relationships between the demographic and developmental characteristics used as control variables and white matter organization within each language group. We also included English language proficiency for both language groups and language use with family and friends only for the bilingual group to understand the impact of these factors on white matter FA values within each language group.
3.3.1 | Effects within the monolingual children
Multiple regressions were used to predict adjusted mean FA values in the selected language and cognitive related white matter pathways using data for only the 446 monolingual participants. These models included age, sex, handedness, pubertal status, nonverbal IQ, parental education, household income, and English vocabulary as predictors to examine the estimated impact of these demographic and developmental variables on mean FA values in each white matter tract for the monolingual participants and used FDR corrections to control for multiple comparisons. The only predictor that was a significant predictor of mean FA values for monolinguals when all covariates were in the model was age across most white matter tracts with β-values ranging between .0005 and .0012 (FDR-corrected p0 s < :05; see Table S2 in Supplementary Materials). These results suggest that within the monolingual group, even a 1-month increase in age was related to a small positive increase in adjusted mean FA values in these white matter tracts when controlling for other demographic and developmental factors. No other predictors were significant in the models in predicting adjusted mean FA values in the selected white matter pathways for the monolingual participants.
3.3.2 | Effects within bilingual children
Multiple regression analyses were conducted to predict adjusted mean FA values in the selected language and cognitive control related white matter pathways using age, sex, handedness, pubertal status, nonverbal IQ, parental education, household income, English vocabulary, and language use with family and friends as predictors with the 446 bilingual group participants. Across all models, only sex was a significant predictor for adjusted mean FA values for the bilingual group, in the cingulum bilaterally (left cingulum: β ¼ :0151, FDR-corrected p ¼ :013; right cingulum: β ¼ :0123, FDR-corrected p ¼ :039; see Table S3 in Supplementary Materials). Notably, the model predicting FA values in the left cingulum in bilinguals accounted for approximately 5% of the variability in mean FA values, and this was significantly predicted by sex (left cingulum model: F(9,436) = 3.42, Adj:R2 ¼ :05, FDR-corrected p ¼ :008). Within the bilingual group, lower mean FA values in the cingulum pathways bilaterally were observed in females. No other predictors across all models significantly predicted mean FA values in bilingual participants.
4 | DISCUSSION
The primary goal of this study was to examine differences in white matter organization in selected language and cognitive control pathways between bilingual children and a rigorously matched monolingual group while controlling for age, sex, handedness, pubertal status, nonverbal IQ, parental education, and household income. One of the limitations of previous neuroimaging studies with bilinguals is that it is difficult to disentangle the variability in the demographic and developmental factors that may also be associated with alterations in brain structure, such as socioeconomic status and age. In the present study, we leveraged the large-scale ABCD Study dataset in order to have the power to rigorously match the groups using these demographic and developmental factors and add them as control variables to the analyses. We found significant differences between bilinguals and monolinguals in some key language network pathways and connections to cognitive control regions of the brain.
Overall, our findings indicate lower FA values in the language and cognitive control related white matter pathways in bilingual school age children compared to their monolingual peers. The findings in the present study are comparable to trends observed in the cross sectional study by Pliatsikas et al. (2020), where mean FA values tended to be lower in bilingual children compared to monolinguals, but for participants over 16 years old, bilinguals tended to show higher mean FA values compared to monolinguals. Although Pliatsikas' results were only statistically significant for the striatal IFC pathways, findings in the present study support the overall trends Pliatsikas and colleagues observed. These results may be partially explained by increased recruitment of subcortical areas associated with more mature language switching and language control skills in bilinguals (Abutalebi & Green, 2016) and with changes in procedural memory circuits underlying learning and use of both first and second language (Ullman, 2020) as posited by the adaptive control hypothesis and the conditional routing model (Abutalebi & Green, 2007; Green & Abutalebi, 2013; Stocco et al., 2010, 2014).
The dynamic restructuring model (Pliatsikas, 2020) provides a framework of sequential stages for experience-induced alterations to gray matter, then white matter, and finally, subcortical structures as bilinguals obtain increasing exposure and proficiency through peak efficiency in their languages. Interpreted through the lens of the dynamic restructuring model, the results from the present study may suggest that bilinguals in our sample may still be in stage 1 of initial exposure to the second language, where structural brain changes resulting from bilingualism are primarily impacting increases to cortical gray matter structures. During stage 1, white matter connections between these areas are in a stage of slower maturation compared to monolingual peers. Rapid increases in white matter organization would be predicted in stage 2, consolidation, resulting from increased experience with the second language (i.e., English for most bilinguals in this study) and a shift toward increased efficiency in language processing. In stage 2, bilinguals are expected to have increased subcortical and cerebellar gray matter compared to monolingual peers, and return to baseline levels of cortical gray matter. As bilinguals reach peak efficiency and automaticity in processing their languages, the dynamic restructuring model predicts increased cerebellar gray matter and a shift from processing in anterior to posterior regions. The dynamic restructuring model has been used primarily to explain brain alterations characteristic of the bilingual brain in adulthood. However, the results from the current study, alongside the changes documented in cortical thickness reported by Vaughn et al. (2021) with the same group of bilingual children from the ABCD Study baseline timepoint, align with the proposed framework of sequential changes posited by the dynamic restructuring model. Future work using longitudinal developmental data are necessary to determine if this continues to be the developmental trajectory for language network pathways in bilinguals.
In the dorsal language network, we found significantly lower mean FA values in bilingual children bilaterally in the SLF (including the tSLF and pSLF portions) compared to their monolingual peers. These findings contrast findings from previous studies comparing white matter organization in bilingual and monolingual children, which did not find any differences between the groups in dorsal language pathways. Mohades et al. (2012, 2015) examined the left SLF, but they did not find any significant differences between their bilingual and monolingual children in this dorsal pathway at either the initial time point (when children were 8–11 years old) or in the second timepoint (when children were 10–13). Although there is a lack of consistency across studies in whether and how the SLF may be implicated in bilingualism, the findings from the present study suggest a bilateral effect using this large, rigorously controlled sample.
The present study also found significantly lower FA values in the bilingual group compared to the monolinguals in the bilateral IFOF pathways of the ventral language network. These results differ from the findings of Mohades et al. (2012), who found significantly greater FA values for simultaneous bilinguals compared to sequential bilinguals and monolinguals in the left IFOF. The significantly greater mean FA values for the simultaneous bilinguals compared to the other language groups in the left IFOF persisted 2 years later when Mohades et al. (2015) reassessed the children when they were ages 10– 13 years old. Findings from the present study may seem in direct conflict with findings from bilingual adult studies, which find higher FA values in bilinguals in the bilateral IFOF and UF compared to monolinguals (Luk et al., 2011; Pliatsikas et al., 2015). However, our hypothesis of different developmental trajectories as bilingual children acquire experience with their languages would support these conflicting results.
Aside from the traditional language-related pathways, the present study also found significantly lower FA values for bilinguals in certain interhemispheric connections (i.e., the forceps major) and in right hemisphere connections to cognitive control areas (i.e., the right cingulum and right parahippocampal cingulum). The cingulum (anterior portion) is correlated with attention and executive function (Kantarci et al., 2011) and the parahippocampal cingulum seems to be more closely related to learning and verbal memory (Ezzati et al., 2016). It is not surprising that the present study reveals differences between bilinguals and monolinguals in white matter pathways associated with the dorsal and ventral language networks, supporting key functions of language processing (e.g., semantics, syntax, and phonology) and pathways associated with cognitive control (particularly those connecting regions involved in bilingual language switching and control). Although we hypothesize that these pathways are adjusting to increasing experience with language switching and language control before we expect increases in FA values in these bilingual children, longitudinal comparisons of developmental trajectories are necessary to reveal these trends.
Although results from the current study differed from the results in previous research with bilingual children of comparative age ranges (Mohades et al., 2012, 2015; Pliatsikas et al., 2020) some of the dissimilarities may be driven by the differences in the sample sizes and the characteristics of the participants in the studies. The large-scale sample size used for the present study (n = 446 in each language group) allowed for the examination of multiple bilateral language and cognitive control pathways and careful control of demographic and developmental variables in contrast to the total sample size of 40 (n = 15 simultaneous bilinguals, n = 15 sequential bilinguals, and n = 10 monolinguals) in the Mohades' studies and the sample of 37 bilinguals and 154 monolinguals of comparable ages in Pliatsikas' cross-sectional study. Further, details in the participant characteristics also differed between the studies. The previous studies had a wider range of ages for their participants than the current study (Mohades et al., 2012: 8–11 years old; Pliatsikas et al., 2020: 6–10; current study: 9–10 years old). Further, the participants in Mohades et al. were limited to French- or Dutch-speakers as the first language and restricted to Romance or Germanic languages for the second language in bilinguals who reported frequent (or balanced) use of both languages. The participants for the current study, similar to those in Pliatsikas et al., are more representative of bilinguals in the United States who tend to speak English at school and in the community, and who come from a variety of home language backgrounds. Finally, bilinguals in this study were, on average, balanced bilinguals but varied widely on this measure and tended to use English primarily with their friends and their other language primarily with family.
It is also worth mentioning the pathways where we did not find differences between children in the bilingual and monolingual groups in this study. As expected, we did not find any differences in the nonlanguage dominant control tract, the bilateral CST. However, there were several language network pathways where being bilingual did not significantly predict mean FA value differences between the language groups. One of the ventral pathways in the language network is the indirect ventral pathway made up of the IFL and UF connecting areas of the IFG and with the hippocampus, parahippocampal gyrus, amygdala, temporal pole, with other areas in the temporal and occipital lobes. The exact functions of the ILF and UF pathways are not well established, although it has been posited that the ILF plays a role in reading and visual orthographic processing of words and the UF seems to facilitate the integration of emotion, episodic memory, and simple syntax into language processing (Middlebrooks et al., 2017; Von Der Heide et al., 2013). Similarly, we did not find widespread differences in pathways associated with cognitive control and procedural memory, which have been associated with language control in bilinguals and where there is evidence of differences between bilingual and monolingual adults. On the one hand, development in certain language networks or cognitive control pathways may be less influenced by the bilingual experience and could be unfolding during childhood in similar ways for bilinguals and monolinguals. Perhaps, through rigorously controlling for demographic and developmental factors in the models we can only detect dissimilarities in pathways associated with the bilingual experience. On the other hand, from the perspective of the DRM, bilinguals may be in stage 2 consolidation for certain functions where the increased exposure to English as the second language has accelerated the rate of increase in these white matter pathways in bilinguals. Instead of having lower mean FA values for these tracts, these pathways look similar in bilinguals and monolinguals at this level of immersion in the language. Further longitudinal studies examining white matter change over time for these pathways are needed to support these hypotheses.
The present study also examined the differential relationships between demographic and developmental characteristics, including English proficiency and language use, with white matter organization in the selected white matter paths within each of the language groups. We hypothesized that we would observe small but widespread effects of age on white matter organization in both groups. The results within the monolingual group supported this hypothesis with widespread albeit small significant effects. However, contrary to our hypothesis, age did not significantly impact white matter within the bilingual group within the narrow range of children ages 9 to 10. It is not well established if patterns of white matter development in bilinguals follow similar patterns as for monolinguals. However, this pattern may support the hypothesis that white matter connections between these areas involved in language and cognitive control processes in bilinguals are in a stage of slower maturation at this very narrow age range where increases in age do not significantly affect white matter FA values. More research is needed to investigate these patterns of white matter development over time in bilingual children. We also hypothesized that more use and exposure to a language other than English and proficiency in English would impact white matter organization in bilingual children, with more use and exposure to the other language and higher vocabulary scores resulting in higher average FA values. In contrast to findings reported by Vaughn et al. (2021) where English vocabulary and language use were significantly related to cortical thickness within the group of bilingual children from the ABCD Study baseline timepoint, we did not find any significant relationship between white matter pathways and English vocabulary in either language group or between white matter and language use for bilinguals. Perhaps the language measures used for this study do not capture the full range of variability in language proficiency and language use (e.g., proficiency in the non-English language, the amount of codeswitching, etc.), which may shape white matter organization. Future research is needed to further investigate these relationships.
4.1 | Limitations
The ABCD Study provides the opportunity for researchers to examine brain structure alterations resulting from exposure and use of two languages in a large-scale cohort representative of bilingual and monolingual children growing up in the United States. This allows us to leverage this large neuroimaging data set to rigorously control factors that could also impact structural brain alterations in our participants. Since the ABCD Study was not designed to study bilingualism, we had limited information about language practices and abilities, and neurocognitive measures of language were limited to English language measures. Future studies should consider evaluation of language abilities in both of the child's languages for more insight into the general language skills of the participants.
Many previously published research examining white matter structure has used traditional tract-based spatial statistics (TBSS) analyses and coordinate for their data. Although there are some advantages to TBSS analyses, including the synthesis of our results in future meta-analyses, TBSS techniques also have limitations in ensuring that any voxel corresponds to the same tract across participants (for comparison of voxelwise and tractography methods, see GoodrichHunsaker et al., 2018). These limitations are more potentially complex when harmonizing pediatric data, which is experiencing rapid development. The ABCD Study dataset provides the advantages of a large scale data set, but since it has been collected across multiple sites with site-specific scanner variability, there are challenges to using and processing the raw dMRI data. To overcome this challenge, the ABCD team has harmonized data across sites, performed the white matter parcellations, and ran the whole-brain tractography resulting in quality-controlled neuroimaging data for over 9000 of their participants (Cetin-Karayumak et al., 2023). The present study utilizes the harmonized data provided by the ABCD Study researchers to avoid potential processing issues.
Other measures traditionally included in bilingualism studies, such as the age of acquisition or initial exposure to each language, were included in the ABCD surveys, but needed to be consistently completed by participants for inclusion in our analyses. Therefore, although we can reliably identify a bilingual group based on parental and child reports of languages spoken at home and with friends, we acknowledge that bilinguals are not a homogeneous group and that there could be within-group differences that we have not been able to detect with the current measures. For instance, Mohades et al. (2012) was able to detect differences in white matter tracts in simultaneous bilinguals compared to both sequential bilinguals and monolinguals. Theories such as the dynamic restructuring model (Pliatsikas, 2020) suggest that increased experience with the second language may account for the differences between the groups since simultaneous bilinguals will have more experience with the second language than sequential bilinguals. Experience might not be the only factor. The sensorimotor hypothesis (Hernandez & Li, 2007) and the neuroemergentism framework (Claussenius-Kalman et al., 2021; Hernandez et al., 2019) suggest that the environment and brain maturation are constantly interacting so that when both languages are experienced in infancy, they develop jointly from low-level sensory information acquired through interactions with the (bilingual) environment, may be related to earlier-developing areas of the brain, and efficient connections between these areas may already exist by mid-childhood. When languages are learned sequentially, the first language provides existing linguistic knowledge to contrast and build off knowledge of the second language, which may engage later-maturing brain regions.
Finally, comparisons of bilingual and monolingual participants are often confounded by differences between the groups that are unrelated to language experience. For example, in the United States, bilingual children often have different racial, ethnic, cultural, and socioeconomic background than monolingual children. In the current study, we attempted to minimize this variability by matching participants on parental education and household income, but cannot control for cultural differences between the groups, which might also impact brain development. Finally, it is unclear whether brain alterations found in bilingual participants result from the bilingual experience or if these alterations result from another aspect in the environment which facilitates development of proficient bilinguals. It is important to note that this study was restricted to the narrow age range between 9 and 10 years old for these children. This is an important initial step to capture developmental alterations that may result from the bilingual experience. However, longitudinal and intervention studies are necessary to better understand the relationship between bilingualism and brain structure during development.
5 | CONCLUSION
The strength of the ABCD Study dataset is the large-scale sample of bilingual and monolingual children from across the United States, representing the experiences of bilinguals growing up in the United States. By rigorously matching a monolingual group to our bilingual group using demographic and developmental factors that included age, sex, handedness, pubertal status, nonverbal IQ, parental education, and household income and including these variables as covariates in our analyses, we were able to find significant differences between the groups that we can interpret as the effect of the bilingual experience. Our results suggest that the bilingual experience is associated with alterations in some well-established dorsal and ventral language network pathways bilaterally and some right hemisphere connections to areas related to cognitive control and memory. Further studies examining structural changes longitudinally in monolingual and bilingual children are necessary to understand how bilingualism is associated with neuroanatomy alterations from a developmental perspective.