1.1 Bilingual experience and the Caudate Nucleus

Several studies indicate that the caudate nucleus (CN) is one of the brain structures involved in bilingual language control. The Adaptive Control hypothesis (Green & Abutalebi, Reference Green and Abutalebi2013), arguably the most popular account of neural mechanism of bilingual language control, attributes a specific role in language control ability to the left CN. Supporting this claim, several studies have reported left CN engagement in various aspects of language control including translation (e.g., Lehtonen et al., Reference Lehtonen, Laine, Niemi, Thomsen, Vorobyev and Hugdahl2005), language selection (e.g., Branzi et al., Reference Branzi, Della Rosa, Canini, Costa and Abutalebi2016), and language switching (e.g., Abutalebi et al., Reference Abutalebi, Della Rosa, Ding, Weekes, Costa and Green2013). Although the evidence is more scarce for right CN, its activity has also been linked to language switching (Ma et al., Reference Ma, Hu, Xi, Shen, Ge, Geng, Wu, Guo and Yao2014) and simultaneous interpretation (Hervais-Adelman et al., Reference Hervais-Adelman, Moser-Mercer, Michel and Golestani2015). Moreover, bilingualism-induced neuroplastic changes have been shown for the CN bilaterally (e.g,, Burgaleta et al., Reference Burgaleta, Sanjuán, Ventura-Campos, Sebastian-Galles and Ávila2016; DeLuca et al., Reference DeLuca, Rothman and Pliatsikas2019b; Pliatsikas et al., Reference Pliatsikas, DeLuca, Moschopoulou and Saddy2017; Zou et al., Reference Zou, Ding, Abutalebi, Shu and Peng2012). These findings are in line with the notion that the CN, among other cognitive functions, plays an important role in executive functioning in general. Specifically, the CN has been shown to contribute to tasks that span from working memory (Cools & D'Esposito, Reference Cools and D'Esposito2011) to regulation of goal-directed behavior through the selection of appropriate action schemas and sub-goals following the evaluation of the action-outcome relationship (Grahn et al., Reference Grahn, Parkinson and Owen2008). Moreover, CN atrophy has been related to the so-called dysexecutive syndrome, i.e., an impairment of attentional and mnemonic executive abilities, supporting the argument of its key role in the frontostriatal executive circuit (Macfarlane et al., Reference Macfarlane, Looi, Walterfang, Spulber, Velakoulis, Crisby, Örndahl, Erkinjuntti, Garde, Waldemar, Wallin and Wahlund2013).

Coming back to bilingual language control, the CN seems to play a peculiar role among areas of the language control network, lying at the intersection of cortical and subcortical areas, as we detail below. The CN is implicated in two main theoretical accounts modeling the underlying mechanisms of bilingualism-induced neuroplastic trajectories and consequent cognitive benefits – namely, the bilingual anterior-to-posterior and subcortical shift (BAPSS; Grundy et al., Reference Grundy, Anderson and Bialystok2017) and the dynamic restructuring model (DRM; Pliatsikas, Reference Pliatsikas2020). The BAPSS model posits that increasing bilingual experience leads to improvements in the efficiency and automatization of language control and thus of executive functioning. This in turn results in a shift of the site of bilinguals’ executive-related neural activation from frontal to posterior cortical and to subcortical brain regions. In the BAPSS, the CN is argued to behave similarly to all other subcortical structures involved, which are increasingly relied on by bilinguals to perform executive tasks. In turn, the DRM models the trajectory of neurostructural changes associated with increasing bilingual experience. Interestingly, for all subcortical structures involved in bilingual language control, except the CN, the DRM posits a steady volumetric increase associated with increasing bilingual experience.

Similarly to BAPSS, this feature reflects the brain's augmented reliance on these structures for language/executive control. For the CN, however, the structural neuroplastic trajectory predicted by the DRM is different from those of other subcortical structures. The CN would increase in volume at initial stages of second language (L2) acquisition in order to cope with cognitive demands associated with the novel task of language control. This novel cognitive demand would induce the CN to undergo structural changes via the formation of new synaptic connections. Subsequently, increasing bilingual experience would lead to an increase of CN's functional efficiency. This increased functional efficiency, in turn, renders the previously accumulated “extra” structural resources no longer necessary for optimized language control. At this stage, surplus connections would thus be eliminated via synaptic pruning. In line with theories of synaptic reorganization (see e.g., Lövdén et al., Reference Lövdén, Wenger, Mårtensson, Lindenberger and Bäckman2013), only the most efficient synaptic connections are maintained during the pruning phase. This process would cause CN's structural substrate to return to pre-bilingualism levels in gross volumetric terms, while its enhanced synaptic connectivity has been reorganized towards higher efficiency. The DRM bases this claim on existing evidence of expansion-renormalization of CN volume (Burgaleta et al., Reference Burgaleta, Sanjuán, Ventura-Campos, Sebastian-Galles and Ávila2016; DeLuca et al., Reference DeLuca, Rothman and Pliatsikas2019b; Pliatsikas et al., Reference Pliatsikas, DeLuca, Moschopoulou and Saddy2017; Zou et al., Reference Zou, Ding, Abutalebi, Shu and Peng2012). The increased efficiency resulting from this structural and functional reorganization (in the CN as well as in other cortical and subcortical structures of the language control network) is argued by the DRM to underlie the neuroprotective effects observed in senior bilinguals.

The DRM predicts a similar structural renormalization trajectory for all the cortical language control areas, but not for other subcortical structures except the CN. The reason behind these different predictions would be that other subcortical structures (i.e., putamen, globus pallidus, thalamus) are involved in more motor-related aspects of language control, i.e., articulation and learning of motor programs related to foreign language production. They would thus follow a somewhat different learning (and thus neuroplastic) trajectory, with these latter abilities becoming more relevant at a later stage of L2 mastery. Thus, the CN's role in bilingual language control seems to lie in-between that of cortical (DRM) and subcortical (BAPSS) regions. This view of the CN as a sort of interface between cortical and subcortical language control areas well reflects its role in domain-general executive control, which is exerted via regulating dopamine release towards prefrontal areas (e.g., Badgaiyan & Wack, Reference Badgaiyan and Wack2011). As such, the CN can in principle be viewed as a relay structure between the subcortical and the cortical architecture of the executive network.

To our knowledge, only two studies have previously investigated both neuroplastic and cognitive reserve effects of bilingualism on the CN at once. The first was by Del Maschio et al. (Reference Del Maschio, Sulpizio, Gallo, Fedeli, Weekes and Abutalebi2018), who found no differences in bilateral gray matter volumes (GMVs) of the CN between bilinguals and monolinguals across young (aged 18-35 years) and senior (aged 60 years and above) age groups while, at the same time, registering a bilingualism-related cognitive reserve effect in the left CN. Specifically, senior bilinguals’ executive performance on a Flanker task was optimized irrespectively of volumetric variations in the left CN, while the performance of senior monolinguals was related to GMV variations in the region. According to the cognitive reserve theory (Stern et al., Reference Stern, Arenaza-Urquijo, Bartrés-Faz, Belleville, Cantilon, Chetelat, Ewers, Franzmeier, Kempermann, Kremen, Okonkwo, Scarmeas, Soldan, Udeh-Momoh, Valenzuela, Vemuri and Vuoksimaa2020), this effect is supported by an increase in functional efficiency enabling the individual to cope with neural atrophy in a given brain network or area, in this case the left CN. This results in an observed mitigation of the relationship between brain structure and the corresponding cognitive performance. Similarly, the second such study found no relationship between bilingual experience and neuroplastic changes in the bilateral CN, while an effect of bilingualism on cognitive reserve development in the left CN emerged, this time in a sample of young adults aged 18-35 years (Gallo et al., Reference Gallo, Novitskiy, Myachykov and Shtyrov2021b). Thus, the available evidence from studies involving both young and senior bilinguals appears to suggest that the structural and functional modifications posited by the DRM may constitute the underlying mechanism for bilingualism-related cognitive reserve accrual.

1.2 The present study

The present study is motivated by the theoretically-driven need to test the DRM and BAPSS predictions for the CN by combining neuroimaging investigations with behavioral data analyses. By doing this, we also aim to further explore the mechanisms behind the emergence of cognitive reserve effects in bilinguals’ CN in the apparent absence of bilingualism-induced volumetric changes, as reported in Del Maschio et al. (Reference Del Maschio, Sulpizio, Gallo, Fedeli, Weekes and Abutalebi2018) and Gallo et al. (Reference Gallo, Novitskiy, Myachykov and Shtyrov2021b) studies mentioned above. Following the DRM logic, for the former study we hypothesize that renormalization to pre-bilingualism volumes of the CN may have already taken place in the sample of lifelong bilinguals, thus explaining the absence of differences between linguistic groups. As per the latter study, we believe the reason for not observing a relationship between bilingual experience and volumetric variations in the CN might be that the authors did not directly test a polynomial relationship, namely the inverted U-shaped trajectory posited by the DRM. Rather, they focused on linear interactions between different sub-aspects of bilingual experience.

To tackle these issues and answer our research questions, here we used structural equation modeling to compute a comprehensive index incorporating several dimensions of bilingual experience. Deriving a single, individual and continuous measure enabled us to test curvilinear relationships between bilingual experience and volumetric changes in the CN. Showing a neuroplastic trajectory that increases, peaks, and renormalizes with increasing bilingual experience would provide direct support to the DRM's assumptions, although in a cross-sectional design (see the Discussion section for more details). We also attempted to replicate findings suggesting that (i) increasing bilingual experience benefits executive performance as measured by the Flanker task (Fan et al., Reference Fan, McCandliss, Sommer, Raz and Posner2002) and (ii) bilingualism supports the development of cognitive reserve in young adults, namely by mitigating the relationship between variations in CN volume and executive performance.

2.1 Participants

Forty bilingual participants (L1: Russian, L2: English; mean age = 21.93, SD ± 2.75; 10 males) were recruited from the population of the HSE University students. All participants were right-handed, as confirmed by the Edinburgh Handedness Inventory scale (Oldfield, Reference Oldfield1971), and had no psychiatric or neurological impairments. Individual socio-demographic profiles for age, educational attainment, and socio-economic status (SES) were assessed with the MacArthur Scale of Subjective Social Status (MacArthur Foundation, 2007). Annual household income bands used as a proxy of SES were adapted to local standards based on the European Social Survey 2020 (ESS Round 10: European Social Survey Round 10 Data, 2020). We also assessed participants’ general intelligence via a subset of the Raven's Standard Progressive Matrices for adults (Court & Raven, Reference Court and Raven1992). Demographics and language background characteristics of the study sample are presented in Table 1. The study was approved by the local research ethics committee, and all participants gave their written informed consent.

Table 1. Demographic and language background characteristics of the study sample

2.2 Individual profiles of bilingual experience

We used the Russian version of the Language Experience and Proficiency Questionnaire (LEAP-Q; Marian et al., Reference Marian, Blumenfeld and Kaushanskaya2007) to assess several dimensions of individual bilingual experience including self-rated L2 proficiency and L2 exposure in various modalities (i.e., writing, reading, speaking, listening) and L2 age of acquisition (AoA). Moreover, we collected objective measures of L2 proficiency via the Cambridge test for adult learners (http://www.cambridgeenglish.org/test-your-english/general-english/) and a custom-made translation task including 42 Russian-to-English entries and 42 English-to-Russian entries. In the translation task, participants were presented with words of three frequency levels (low, medium, high), calculated using the Russian National Corpus (https://ruscorpora.ru/en/) and the Subtlex-UK corpus (van Heuven et al., Reference van Heuven, Mandera, Keuleers and Brysbaert2014), respectively, and had to provide what they considered to be the best translation for each entry.

2.3 Executive performance assessment

To measure their executive performance, participants underwent the Flanker task within the standard ANT set-up (Fan et al., Reference Fan, McCandliss, Sommer, Raz and Posner2002). In this task, a fixation cross appeared for 400 ms at the center of the screen, followed by an array of five arrows pointing to the left or to the right for a maximum duration of 1700 ms. Participants had to detect the direction of the central target arrow as accurately and as fast as possible by pressing the designated arrow key on the PC keyboard. The task included three conditions: congruent, where the target arrow is flanked by arrows pointing to the same direction (→→→→→); incongruent, where target and flankers point to opposite directions (←←→←←); and neutral, where the target is flanked by uninformative dashes rather than arrows (– – → – –). Incongruent trials require interference suppression to override prepotent incorrect responses, which typically entails lower accuracy and longer RT in responses. Prior to the task, participants underwent a practice run of 24 pseudo-randomized trials. Subsequently, two pseudo-randomized runs of 96 trials each (32 per each condition) were presented.

The Flanker Task is a benchmark in executive functioning assessment. The reason for selecting this task is two-fold. First, as mentioned above, one of our aims was to further expand previous findings from studies with a similar design (Del Maschio et al., Reference Del Maschio, Sulpizio, Gallo, Fedeli, Weekes and Abutalebi2018; Gallo et al., Reference Gallo, Novitskiy, Myachykov and Shtyrov2021b). Since issues with cross-task correlations have been previously raised in the field (Paap & Sawi, Reference Paap and Sawi2014), we kept the executive task constant with the previous investigations in question, for replicability purposes. Second, and more important, the Flanker task is one of the few tasks whereby the structural brain-behavior (SBB) relationship has proven to be valid, especially in young populations (Boekel et al., Reference Boekel, Wagenmakers, Belay, Verhagen, Brown and Forstmann2015). Given that our cognitive reserve analysis (detailed below) consisted in testing the potential modulation of the expected SBB relationship between CN and executive performance by bilingualism it was of crucial importance to choose a task that is known to follow this relationship in the first place.

2.4 MRI data acquisition and preprocessing

A Philips Intera 1.5T MRI scanner was used to acquire T1-weighted images with the following parameters: TR = 25 ms, TE = 4.6 ms; flip angle = 30, FOV = 240 × 240, resolution = 1 × 1×1 mm, matrix = 256, TA = 5.35 min, mode = 3DFFE, number of slices = 191. We used region-based morphometry in CAT12 (Computational Anatomy Toolbox, https://neuro-jena.github.io/cat/) within SPM12 (Statistical Parametric Mapping, https://www.fil.ion.ucl.ac.uk/spm/) to extrapolate the total amount of GMV in atlas-based regions of interest (ROIs) for the bilateral CN. First, we visually inspected images to check for gross field distortions and movement artifacts; no participants were discarded for this reason. We then manually set the origin for each image to correspond to the anterior commissure – posterior commissure (AC-PC) line. Subsequently, we followed a two-step procedure for GMV extraction:

i) Raw structural images were segmented into different tissue classes – gray matter (GM), white matter (WM), and cerebrospinal fluid (CSF) – via CAT12 segmentation routine. This routine utilizes an adaptive Maximum A Posterior (aMAP) technique that reduces the need for a priori information about tissue probabilities (Rajapakse et al., Reference Rajapakse, Giedd and Rapoport1997) and accounts for local variations and inhomogeneities of GM intensity (Dahnke et al., Reference Dahnke, Ziegler and Gaser2012). After the aMAP segmentation, CAT12 also performs a Partial Volume Estimation (PVE) of mixed tissue-classes (GM-WM and GM-CSF; Tohka et al., Reference Tohka, Zijdenbos and Evans2004) resulting in a higher segmentation accuracy via estimations of the percentage of pure tissue of each type within each voxel. Segmentation was further improved via a spatial-adaptive non-local means (SANLM) denoising filter in a pre-segmentation step (Manjón et al., Reference Manjón, Coupé, Martí-Bonmatí, Collins and Robles2010). After segmentation, each brain was co-registered to the ICBM (International Consortium for Brain Mapping) European brain space template via the affine regularization option.

ii) GMV values were extracted from bilateral ROIs in the CN via an in-built CAT12 function that allows for GMV estimation in non-normalized native space using maximum tissue probability labels derived from the Neuromorphometrics Atlas (http://www.neuromorphometrics.com/). Participants’ total intracranial volume (TIV) was extracted by summing native space global volumes of GM, WM and CSF to control for individual differences in brain size.

2.5 Procedure

Participants were tested at the experimental facilities of the Institute for Cognitive Neuroscience, HSE University. The testing procedure was as follows. Participants were first instructed on the experimental procedure, and they signed an informed consent. They then were seated in an electrically shielded and acoustically dampened chamber. All questionnaires and tasks were presented on a 75 cm-diagonal computer screen. First, participants underwent Raven's Standard Progressive Matrices and filled out the LEAP-Q, which were presented using Google Forms. Subsequently, they performed the Flanker task, which was presented via OpenSesame (v. 3.3.7). Finally, the Cambridge test for adult learners and the translation task were presented to participants also via Google Forms. All tasks were performed in the same experimental session, which lasted around 60 minutes. MRI acquisition was performed on a separate day in a specialized MRI center in Moscow using a clinical-grade Philips Intera scanner (see above).

2.6 Statistical analyses

We used structural equation modeling in order to combine information from different sub-dimensions of bilingual experience and obtain a single, continuous, and individual index of bilingualism. The model included a 2-level hierarchical structure with first-level latent variables computed for L2 exposure, subjective and objective L2 proficiency. In turn, these latent variables, together with L2 AoA, informed a higher-level latent factor, the comprehensive bilingual index. The reason behind the choice of a hierarchical SEM structure is the following: the L2 AoA variable is informed by a single score (i.e., the answer to the question “How old were you when you began understanding/speaking English?”) rather than multiple scores or answers as in the case of L2 exposure and subjective/objective L2 proficiency. Thus, had we not grouped observables informing the other bilingual experience factors and adopted a hierarchical structure, the contribution of AoA to the BI could have been obscured among the contributions of the multiple lower-level observables that inform exposure, objective and subjective proficiency. By selecting this hierarchical structure, we constrained the model to consider the weight of AoA on the latent Bilingual Index as comparable to those of the other three bilingual experience factors.

Maximum likelihood estimation was used to estimate the model. Model fit was assessed using conventional criteria and omitted paths were explored using modification indices. The final model fit to the data was acceptable (CFI = 0.946; TLI = 0.922; RMSEA = 0.09). The model is presented in Figure 1. Individual predicted values of the latent bilingual index were extracted. The resulting variable was normally distributed with a mean of 0 (range: -1.64, 1.53).

Figure 1. Standardized estimates from the best-fitting structural equation model for the bilingual index latent variable.

As for the Flanker data, neutral, incorrect and false start (RT < 100 ms) trials were removed, as well as outlier trials displaying RTs beyond 3 SDs from individual mean RT values (Baayen & Milin, Reference Baayen and Milin2010). This data pre-processing procedure resulted in discarding 1.13% of the total data.

All statistical analyses were conducted using Stata 17 (StataCorp, 2021).