Long-term cognitively challenging experiences modulate metabolite concentrations in the healthy brain: the case of bilingualism

Cognitively demanding experiences, including complex skills acquisition and processing, have been shown to induce brain adaptations, at least at the macroscopic level, e.g. on brain volume and/or functional connectivity. However, the neurobiological bases of these adaptations, including at the microstructural cellular level, remain poorly understood. Here we use bilingualism as a case study to investigate the microscopic correlates of experience-based brain adaptations. We employ Magnetic Resonance Spectroscopy to measure concentrations of metabolites in the ventral striatum, a region critical to language control which is reshaped by bilingualism. Our results revealed increased concentration of myo-Inositol in bilinguals compared to monolinguals. This metabolite is linked to synaptic pruning, a process underlying experience-based brain restructuring. Crucially, concentration was predicted by relative amount of bilingual engagement. Our results suggest that (degree of) long-term cognitive experiences have measurable effects at the microcellular level, which might accompany, if not drive, the observed macroscopic brain adaptations.

microscopic level, in this case metabolic ones. Importantly, changes in the concentrations of INS, NAA and GLX have also been reported in the occipital cortex of sighted participants after visual training 27 . This suggests that the glial cells might have a central role in brain restructuring as a result of cognitive and sensory experiences, beyond ageing or pathology.
In line with the above, it is judicious to hypothesise and empirically test that similar effects in metabolite concentrations can result from long-term, cognitively challenging experiences known to affect brain structure and function. One of these experiences is bilingualism. It is widely accepted that the mental juggling of more than one language in a single mind/brain is cognitively demanding. Indeed, the need for use between the two languages is unpredictable. As a result, both languages are continuously active at all times irrespective of apparent need or intent. This reality requires an efficient system of control for appropriate selection of one language for comprehension and production alongside simultaneous suppression of the irrelevant language to a low level of idle activation for whenever the other may become needed 28,29 . This constant competition taxes domain general executive control abilities and their underlying brain structures, leading to long-term adaptations in domain general cognition 30 , and in brain function 31 , structure 6 and metabolism 32 . Notably, such effects can be both observed after extended long-term exposure in a bilingual environment 33,34 and after shortterm intensive language training 35,36 . Moreover, recent literature has shown that the nature and location of these effects is modulated by quantitative measures of the depth and intensity of experiences bilinguals have with using their languages and opportunities to switch between them [37][38][39][40] .
Over all, it seems reasonable to predict that the continuously challenging task of handling two (or even more) languages would result in brain changes at the neurochemical level; bilingualism requires greater and more sustained efficiency in brain regions subserving language and cognitive control, such as the anterior cingulate gyrus (ACC) and parts of the basal ganglia such as the caudate nucleus and putamen. In fact, all these regions have been shown to change in shape and/or volume as a response to bilingualism 28,41-44 . Therefore, it is possible that these structural changes might have their correlates in changes in metabolite concentrations. For example, and drawing parallels from the findings from healthy ageing and neuroplasticity in the blind 23,24 , the observed restructuring of these regions might be characterised, at least partly, by expansion of glial cells, which is in itself marked by increases in regional CHO, CRE and INS.
Crucially, this hypothesis could provide the biological basis of bilingualism-induced regional neuroplasticity, which is currently not well understood 6 .
To the best of our knowledge, only one study has looked at correlates in metabolic concentrations of the effects of bilingualism on cognition and the brain. Weekes and colleagues 45 compared young bilingual adults and age-and education-matched monolingual controls on metabolite concentrations (NAA, CHO, CRE, INS and GLX) in the anterior cingulate cortex (ACC), a region critical for domain general cognitive control 28 . Contrary to their predictions, Weekes and colleagues reported lower levels of NAA, CHO, CRE and GLX in bilinguals than monolinguals 1 , and interpreted this finding as an indication of more efficient control monitoring of the bilingual brain as a result of prolonged bilingual experience. However, they reported no significant correlations between these concentrations and executive control abilities as were measured by a Flanker Task, nor a significant difference in task performance between the two groups. This study suggests that there might be effects of bilingualism on brain metabolism that are not detectable behaviourally, echoing some evidence in the neuroimaging literature suggesting that behavioural measures might not always capture latent effects of bilingualism on brain function 38 . More to the present point, this set of results provided the first evidence that bilingualism-induced neuroplasticity might have its roots in changes in metabolite concentrations.
The present study expands on Weekes et al. 45 by looking at the effects of bilingualism on metabolite concentrations in the ventral striatum, a subcortical grey matter nucleus comprising the caudate nucleus and the putamen, key structures for language selection and cognitive control 28 , that have both been shown to be affected structurally by bilingualism 41,43 . In order to study such effects across the adult lifespan, the present study comprises a relatively large sample that spans an age range representative of the adult lifespan, additionally accounting for the known effects of age on metabolite concentrations 19 . We looked at relative concentrations of four key metabolites (NAA, INS, CHO, GLX) as proportions of a fifth one (CRE). Based on previous literature, we predicted that increased age will lead to increased INS and decreased NAA concentrations in the ventral striatum, potentially accompanied by increases in GLX but not CHO [19][20][21] . In terms of the effects of bilingualism, we consider the data in two ways: following the most traditional practice in the field 45 , we split our participant groups into Bilinguals and Monolinguals according to experiences in using more than one language (for details, see Methods). For this comparison, we predicted overall higher concentrations of INS and CHO in bilinguals, compared to monolinguals, an effect that would signify glial expansion, which could contribute to the observed restructuring of the ventral striatum in bilinguals 23,24 .
Moreover, if the effects of age and bilingualism are based on similar mechanisms, then the combined effect of these two factors should lead to steeper increases with age of the INS and CHO concentrations in bilinguals. Following from more recent suggestions on how bilingualism can differentially affect brain structure and function 38,39 , we also looked at our entire sample to investigate whether the concentrations the metabolites of interest might be modulated by intensity of engagement with (bilingual) language experiences. We predicted that the more intense and sustained the bilingual experience, the greater the concentration increases would manifest.

i) Effects of age and bilingualism
The results from the first set of analyses are illustrated in  Figure 1 illustrates the relative concentrations per group across age for each metabolite. from the second set of analyses are presented in Table 2. Figure 2 illustrates the effects on the INS concentrations.

Conclusions
Building on findings that the cognitively challenging experience of bilingualism can have knockon consequences for the structure and function of brain regions related to language acquisition and control, and the (functional and structural) connectivity between them 6,31,46 , the present study used MRS to investigate metabolic correlates of these adaptations. Recall that we focused on the concentrations of several well-understood metabolites in the ventral striatum, a region crucial for language selection and control in bilinguals 28 . Because brain effects of bilingualism can vary as a function of engagement with relevant experience, our sample included a considerable group of individuals varying in degree of bilingual engagement factors. Given that age is a comorbid factor, our sample also included a considerable range, enabling us to tease apart the effects of ageing from bilingualism. Results revealed age effects that largely corroborate previous findings in this particular brain region 21  temporal volumetric grey matter increases followed by renormalisation over increased experience of using the new skill 6 . In the case of bilingualism, this process particularly affects the ventral striatum, which is key to language control for bilinguals 28 , with the caudate nucleus and the putamen undergoing dynamic structural changes 6 . Crucially, this expansionrenormalisation process has been (at least tentatively) attributed to pruning of superfluous synapses that were formed during the skill acquisition, in order for the more efficient ones to be nevertheless, we remain cautious to overinterpreting this trend.
The importance of these findings is, at least, twofold. First, the data shows that the well documented effects of bilingualism on brain structure and function have their correlates in changes in brain metabolism. Crucially, we report markers of metabolic activity which are compatible with experience-based approaches, arguing for bilingualism-induced dynamic brain adaptations 6 . Given the important implications these adaptations may have for healthy and pathological ageing of the bilingual brain 50 , future studies should pay particular attention to these indices of neuroplasticity and how they interact with brain decline in key areas related to language processing and control. Second, and more generally, we show here that sustained, longterm cognitively challenging experiences, such as controlling two languages, might also have persisting effects on metabolite concentrations in the brain. Therefore, it is possible that similar long-term findings could be reported for other types of skill learning and experiences that have shown to restructure the brain (e.g. music, driving, exercise), and such effects are not limited to short-term training. Insofar as cognitively challenging experiences have a direct impact on metabolite concentrations in the healthy brain; they are useful in furthering our theoretical understanding of the mechanisms underlying skill acquisition and use, and the accompanying neural adaptations.

Participants
In total, 99 adults were recruited. Inclusion criteria for the study included being righthanded (self-reported), no history of speech and language disorders and no contraindication to MRI scanning. The participants were divided into two groups. The Bilingual group consisted of participants who spoke English as their second language (L2) and were resident in the UK at the time of testing, i.e. they were immersed in the L2-speaking environment. Importantly, there were no inclusion criteria relating to their native language or other language factors in order to recruit the widest possible range of linguistic experiences 38,39 . The English native comparison group (henceforth called "Monolingual" group for convenience) included individuals born and raised in the UK who had minimal or no exposure to additional languages, a typical and representative demographic in the UK. Of the participants that were recruited to the study, 33 were removed from the final cohort for several reasons 2 . The final sample consisted of 70 participants (age range 19-83), including 33 monolinguals (20 female) and 37 bilinguals (28 female). See Table 3 for full descriptors of the final sample. This research was approved by the University of Reading Research Ethics Committee. Informed consent was obtained from all participants.

Materials
Both participant groups completed the Language and Social Background Questionnaire (LSBQ) 51 which documents the participants' language use from childhood to the present day and across several settings and dimensions. The LSBQ yields two scores related to the amount of (bilingual) language use within specific communicative settings, which have been shown to predict bilingualism-induced changes in brain structure and function 38,39 . Specifically, L2 social corresponds to the degree of L2 exposure and use in societal and community settings whereas L2 home corresponds to the extent of L2 proficiency and use in home settings. Moreover, LSBQ outputs a composite score accounting for the overall bilingual experience. For all three scores, a higher value indicates increased bilingual engagement, i.e. increased (balance in) use of, and switching between, the two languages.

MRI data acquisition
Neuroimaging data were acquired on a 3T Siemens MAGNETOM Prisma_fit MRI scanner, with a 32-channel Head Matrix coil and Syngo software. A high-resolution T1-weighted  values from GM and WM at 3T were used as base to compute attenuation factors for both water and metabolites. These factors were in turn used to correct the reported values for relaxation effects (TR) dependent on voxel's tissue proportion. Last, and following standard practice in the field, we used a relative quantification method on our metabolite levels in order to avoid assumptions that the metabolite levels remained constant, and also to reduce individual variability among subjects 23,27,57 . For this study, CRE was chosen as the reference metabolite, as a preliminary analysis showed that its levels were not affected by age, bilingualism or the interaction between the two (all ps>0.3). Table 3 illustrates the mean relative concentrations for the remaining four metabolites.

Data analysis
Rejection criteria included one or more of the participants metabolite concentrations being outside the Cramer-Rao lower bound (CRLB) 58 and/or exceeding two standard deviations (2SD) from the group mean metabolite concentration (calculated separately for every metabolite). Only 3.85% exceeded the CRLB threshold (< 50% = acceptable reliability) and 15.39% the 2SD.
The corrected and quantified MRS data were analysed in R 59 with generalised additive models (GAMs), by using the bam() function of the mcgv package 60