Genetic bases of language control in bilinguals: Evidence from an EEG study

Abstract

Previous studies have debated whether the ability for bilinguals to mentally control their languages is a consequence of their experiences switching between languages or whether it is a specific, yet highly‐adaptive, cognitive ability. The current study investigates how variations in the language‐related gene FOXP2 and executive function‐related genes COMT, BDNF, and Kibra/WWC1 affect bilingual language control during two phases of speech production, namely the language schema phase (i.e., the selection of one language or another) and lexical response phase (i.e., utterance of the target). Chinese–English bilinguals (N = 119) participated in a picture‐naming task involving cued language switches. Statistical analyses showed that both genes significantly influenced language control on neural coding and behavioral performance. Specifically, FOXP2 rs1456031 showed a wide‐ranging effect on language control, including RTs, F(2, 113) = 4.00, FDR p = .036, and neural coding across three‐time phases (N2a: F(2, 113) = 4.96, FDR p = .014; N2b: F(2, 113) = 4.30, FDR p = .028, LPC: F(2, 113) = 2.82, FDR p = .060), while the COMT rs4818 (ts >2.69, FDR ps < .05), BDNF rs6265 (Fs >5.31, FDR ps < .05), and Kibra/WWC1 rs17070145 (ts > −3.29, FDR ps < .05) polymorphisms influenced two‐time phases (N2a and N2b). Time‐resolved correlation analyses revealed that the relationship between neural coding and cognitive performance is modulated by genetic variations in all four genes. In all, these findings suggest that bilingual language control is shaped by an individual's experience switching between languages and their inherent genome.

This study addresses the genetic bases underlying bilinguals’ language control ability to simultaneously juggle two or more languages, critically targeting that language control is shaped by inherent language ability or is the product of the language ability interweaving with executive function. The results showed that genetic variations extensively influenced language control processing reflected by modulating the predictability of neural coding on cognitive performance. Therefore, our findings complement the crucial assumption of the Adaptive Control hypothesis, that language control is an adaptive part of domain‐general control, and demonstrate that language control is the consequence of switching experience interweaving with inherent genome.

1. INTRODUCTION

Language is part of our genetic makeup that allows us to interact in sophisticated ways and in a variety of contexts. For individuals who speak more than one language, we are often intrigued by the fact that these individuals appear to effortlessly juggle between their languages as necessitated by their listeners and environmental cues (Abutalebi & Green, 2007; Liu et al., 2021). It seems that bilinguals have the unique ability to temporarily “ignore” one language while using another to the extent that they rarely suffer unwanted intrusions from the irrelevant language. This ability is generally thought to be supported by the mental process known as language control. Based on the adaptive control hypothesis (ACH; Green & Abutalebi, 2013), language control among bilinguals is an adaptative mechanism that meets situational and communicative needs by recruiting executive functions, such as inhibition, attentional control, updating, conflict monitoring, and working memory (Abutalebi & Green, 2007; Bialystok et al., 2005; Calvo & Bialystok, 2014; Coderre et al., 2016; Declerck et al., 2017; Kovács & Mehler, 2009; Kwon et al., 2021; Liu, Schwieter, Liu, et al., 2022; Prior & Gollan, 2013; Verreyt et al., 2016). However, accumulating evidence challenges the belief that language control and executive functions share underlying mechanisms and instead, argues for the specificity of language control (Antón et al., 2014; Calabria et al., 2012; Calabria et al., 2015; Cattaneo et al., 2020; Declerck et al., 2015; Duñabeitia et al., 2014; Liu, Dunlap, et al., 2016; Paap et al., 2014). These inconsistencies may be partially due to specific genes, identified in some bilinguals, that influence the relationship between language control and executive function (Liu, Schwieter, Liu, et al., 2022). Accordingly, the present study aims to explore whether language control among bilinguals depends on language‐related genes or whether it inseparably relates to the particular genes that are essential for executive function.

1.1. Language control during two phases of language switching

Switching between languages causes a processing delay, typically called a switch cost. These switch costs are measurable indicators of language control (Blanco‐Elorrieta et al., 2018; Costa & Santesteban, 2004; Declerck & Koch, 2022; Declerck & Philipp, 2015; Liu et al., 2021; Schwieter & Sunderman, 2008; Zhu et al., 2022) which represent transient, trial‐to‐trial control processes and engagement of additional cognitive resources (Christoffels et al., 2007; Jackson et al., 2001; Linck et al., 2012; Liu et al., 2021; Liu, Liang, et al., 2016; Martin et al., 2013; Misra et al., 2012; Verhoef et al., 2009; Zhu et al., 2022). The picture‐naming task with cued language switches is one of the most common measures of language control in bilinguals. The cues serve to indicate in which language (i.e., the language schema phase involved in a switching task) the participant is to name the picture (i.e., the lexical response phase). In the first phase, switching between language schemas (e.g., a first language, L1, or a second language, L2) requires multiple components of executive functions, such as inhibiting the nontarget language and updating the new language schema. This increased demand for cognitive resources on switch trials has a significantly larger effect than on non‐switch trials (Christoffels et al., 2007; Martin et al., 2013; Verhoef et al., 2009; Zheng et al., 2020). These findings have been evidenced by an event‐related potential (ERP) component, the N2a, which is believed to be an indicator of attentional control during language schema selection (Liu et al., 2014; Liu et al., 2018; Misra et al., 2012; Verhoef et al., 2010).

The second phase involves lexical processing, which can be reflected by the late positive component (LPC) (Jackson et al., 2001; Martin et al., 2013; Roelofs, 2003). This ERP component is associated with the retrieval of the correct lexical item in the intended language and reflects the reconfiguration of stimulus–response mappings (Jackson et al., 2001; Liotti et al., 2000; Martin et al., 2013). Some research has revealed that language control may be implicated during the lexical response phase in which an N2 effect occurred after picture onset (N2b), suggesting that selective inhibition was engaged to reduce competition during lexical selection and/or phonological encoding (Cheng et al., 2010; Piai et al., 2014; Roelofs, 2003; Shao et al., 2014). Other research, however, has suggested that the lexical response phase involves cognitive processes related to lexical processing (Green, 1998; Linck et al., 2012; Liu et al., 2014, 2018; Misra et al., 2012), such as conceptual identification, lemma retrieval, and word‐form encoding. Whether executive functions also play an important role in both the language schema and lexical response phases is an ongoing question (Cheng et al., 2010; Jackson et al., 2001; Martin et al., 2013; Roelofs, 2003; Verhoef et al., 2010). If language control is cultivated by mental switching exercises, this control would solely emerge in the language schema phase; otherwise, it means that language control benefits from integration with executive function. In the present study, we examine how language control is affected by specific genes related to executive functions and language‐related genes during the two aforementioned phases of picture naming.

1.2. Genetic contributions to language control and executive control

Bilinguals' language control ability may be inherently constrained by language‐related genes. Several studies have found that human language abilities are influenced by genetic variations, particularly in the gene forkhead box protein P2 (FOXP2) (Chabout et al., 2016; Crespi et al., 2017; Fisher & Scharff, 2009; Mozzi et al., 2017). The FOXP2 gene is critically involved in the development of the neural systems that mediate human speech and language acquisition (Fisher & Scharff, 2009; Liégeois et al., 2003). Many studies on language impairment have suggested that FOXP2 single nucleotide polymorphisms (SNPs) are linked to dyspraxia (i.e., the reduced ability to accurately sequence speech sounds), impaired expressive and receptive linguistic abilities, reduced lateralization in brain speech‐related areas (Chabout et al., 2016; Liégeois et al., 2003; Reuter et al., 2017; Vernes et al., 2006), as well as abnormalities in speech and language processing (Badcock, 2010; Ford et al., 2014; Španiel et al., 2011). In a study by Pinel et al. (2012), the researchers examined the brain activation of 94 healthy individuals during a sentence reading task and found that FOXP2 rs6980093 polymorphism selectively modulated brain activity in the left frontal cortex, showing that AA homozygotes elicited stronger brain activity than GG homozygotes. Furthermore, GG homozygotes could exhibit an advantage in a non‐native speech learning task, showing shifting faster to procedural learning strategies (Chandrasekaran et al., 2015). In addition, Padovani et al. (2010) reported that rs1456031 polymorphism had an effect on verbal fluency and phonological fluency in patients with frontotemporal lobar degeneration, specifically among the patients with homozygous dominant (TT) genotype (Padovani et al., 2010). Inconsistently, Crespi et al. (2017) found a significant effect of FOXP2 rs1456031 variation on inner speech rather than speech fluency.

Alongside the growing body of work on genetic factors of humans' capacity for language, several studies have identified the dopaminergic system as a possible contributing factor to L2 learning and language control (Hernandez et al., 2015; Mamiya et al., 2016; Sugiura et al., 2017; Vaughn et al., 2016; Vaughn & Hernandez, 2018). Specifically, dopamine‐related genes have been shown to play a critical role in executive function (Barnes et al., 2011; Barnett et al., 2008; Chen et al., 2004; Sannino et al., 2015; Witte & Flöel, 2012), suggesting that language control may be inseparable from the support of dopamine‐related genes. Previous studies have suggested that dopamine (DA) levels have a significant influence on working memory, set‐shifting, updating, and cognitive flexibility and stability (Barnes et al., 2011; Klanker et al., 2013; Logue & Gould, 2014; Robbins & Arnsten, 2009; Zhang et al., 2015). For instance, the enzyme catechol‐O‐methyltransferase (COMT) is widely represented in the human brain and accounts for 60% of dopamine degradation in the prefrontal cortex, making its involvement in prefrontal‐guided executive functions and second language learning of great interest to researchers. Several studies have found that a functional SNP (rs4680) for COMT influences executive functioning, working memory, fluid intelligence, and attentional control (Barnett et al., 2008; Barnett, Heron, et al., 2007; Barnett, Jones, et al., 2007; Egan et al., 2001; Flint & Munafò, 2007). Moreover, a recent study using functional near‐infrared spectroscopy showed that COMT rs4680 polymorphism exerted a significant effect on language performance and processing (Sugiura et al., 2017). Other studies have demonstrated that A‐allele (Met) carriers with higher DA availability performed better in tasks requiring stable performance, while G‐allele (Val) carriers performed better in tasks requiring flexible performance (Mier et al., 2010; Nolan et al., 2004). Another functional polymorphism, the COMT rs4818, has been reported to have differential effects on executive functions: GG homozygotes with higher DA availability had better decision‐making performance but worse planning ability than the C allele variant (Roussos et al., 2008).

There are also genetic underpinnings of human memory, a cognitive system crucial to language processing (Archibald, 2017; Baddeley, 2003, 2022; Daelemans & Van den Bosch, 2005; Declerck et al., 2013; Desmond & Fiez, 1998; Linck et al., 2014). Brain‐derived neurotrophic factor (BDNF) has been shown to regulate the structure and function of neurons involved in memory formation, and extensively implicated the long‐term potentiation (i.e., a form of synaptic plasticity that underlies long‐term memory storage) in the hippocampus (Bekinschtein et al., 2007, 2008). Studies have demonstrated that BDNF rs6265, an SNP, plays a crucial role in short‐term plasticity and learning, in which A allele is associated with diminished memory function (Egan et al., 2003; Hariri et al., 2003). In an EEG study using a Go‐Nogo task by Beste et al. (2010), the researchers reported that the BDNF polymorphism (rs6265) can selectively modulate response inhibition, with a larger Nogo‐N2 effect in A allele carriers. These findings indicate that BDNF may affect human executive functions too. Moreover, a new gene—KIBRA (also referred to as WWC1 for WW‐and‐C2‐domain containing‐protein‐1) has been shown to influence memory performance and synaptic plasticity (Almeida et al., 2008; Zhang et al., 2014). Genetic variations in KIBRA rs17070145 have been associated with episodic memory and the activation of the hippocampus during memory retrieval (Almeida et al., 2008; Papassotiropoulos et al., 2006). In these studies, the CC genotype revealed significantly worse immediate and delayed recalls scores than T‐allele carriers.

1.3. Present study

By examining the effect of genes related to executive functions (BDNF, COMT, and Kibra/WWC1) and a language‐related gene (FOXP2) on language control during the two processing phases involving language schema and lexical responses, the present study sheds new light on the nature of language control. In our experiment, we will test the following possibilities:

• If language control is a language‐specific ability, then the language‐related gene (FOXP2) is the only gene that has an impact on language control.

• If language control is the consequence of ongoing experience with language switching in daily life, the executive function genes (BDNF, COMT, and Kibra/WWC1) should significantly interact with switch costs of language schema phase (i.e., N2a), but not the language processing involved in lexical response phase (N2b and LPC).

• If language control is a function of its integration with executive function, then both the executive function genes and language‐related gene (FOXP2) should have an impact on two phases of language control (N2a, N2b, and LPC).

According to the ACH (Green & Abutalebi, 2013) and previous empirical evidence (Abutalebi & Green, 2007; Branzi et al., 2019; Guo et al., 2011; Wu et al., 2019), we hypothesize that language control is actually developed by cognitive control and that corollary, executive function genes will show an effect on processing at both the language schema level and lexical selection level.

2. METHOD

2.1. Participants

One hundred and nineteen Chinese (L1)–English (L2) bilinguals (91 females, 28 males, mean age: 22.47, range: 19–30) were recruited in the study. These individuals reported that they had begun learning English at an average age of 9.09 years (SD = 1.87 years, range = 6–13 years). All participants were right‐handed with normal or corrected‐to‐normal vision and reported no psychological, cognitive, or motor impairments. The study was approved by the Ethics Committee of Research Center of Brain and Cognitive Neuroscience at Liaoning Normal University and all participants provided their informed consent before beginning the experiment. When genotyping for the two FOXP2 polymorphisms and two COMT polymorphisms (discussed in Section 2.2), two of the participants failed to be genotyped, thus leaving 117 participants who were successfully genotyped for these four polymorphisms. Moreover, there was one participant who was not genotyped for Kibra/WWC1 polymorphism, resulting in 118 participants.

To assess L1 and L2 proficiency levels of the participants, we adopted the Oxford quick placement test (OPT; Geranpayeh, 2003; Liu, Schwieter, Liu, et al., 2022; Liu, Schwieter, Wang, et al., 2022), and asked individuals to rate their language abilities on a six‐point scale in which “1” indicated no knowledge and “6” indicated perfect knowledge (Liu et al., 2021). A one‐way ANOVA was next conducted for each polymorphism and revealed no significant differences between genotypes in age, age of L2 acquisition, self‐ratings of language abilities, and OPT scores (see Table 1). Paired‐sample t‐tests revealed that the participants' L1 was significantly stronger than their L2 in listening (t = 16.11, p < .001), speaking (t = 19.24, p < .001), reading (t = 17.37, p < .001), and writing (t = 11.97, p < .001). These self‐ratings and OPT scores are similar to intermediate Chinese–English bilinguals tested in prior research (Liu et al., 2021; Liu, Liang, et al., 2016; Liu, Schwieter, Wang, et al., 2022), implying that the individuals have unequal proficiency between their two languages.

TABLE 1.

One‐way ANOVA results of L1 and L2 proficiency.

	Age	L1				L2				AoA	OPT
		Listen	Speak	Read	Write	Listen	Speak	Read	Write
FOXP2 rs6980093
F	.01	.04	.98	.92	1.14	.08	.90	1.12	1.67	1.05	.24
p	.986	.964	.379	.404	.325	.922	.409	.329	.192	.355	.800
FOXP2 rs1456031
F	.91	.55	.48	.34	1.31	.69	.62	2.24	.14	.41	.32
p	.407	.579	.617	.714	.275	.504	.542	.111	.867	.663	.725
COMT rs4680
t	.69	.02	.34	1.26	.46	.001	.06	.72	.02	2.75	1.95
p	.408	.885	.560	.264	.499	.974	.814	.397	.886	.100	.166
COMT rs4818
t	2.90	1.56	.51	1.14	.53	.90	.45	.14	.34	.20	3.61
p	.091	.215	.478	.289	.468	.344	.502	.709	.562	.657	.060
BDNF rs6265
F	.47	2.05	.46	1.31	.93	.43	.41	.64	.87	1.44	.51
p	.627	.134	.631	.275	.398	.649	.666	.529	.422	.241	.603
BDNF rs2049046
F	1.15	.25	.67	1.18	.82	.47	.04	1.82	.91	.25	.50
p	.319	.777	.516	.311	.444	.627	.965	.184	.041	.776	.611
Kibra/WWC1 rs17070145
t	.71	.26	.26	.99	.66	.18	.18	.02	.02	.09	<.001
p	.402	.615	.612	.332	.418	.669	.674	.895	.878	.769	.984

2.2. DNAs extraction and genotyping

An overview of the experimental procedures and analyses is illustrated in Figure 1. In the first step, the genomic DNAs were extracted from peripheral blood leukocytes of all participants, which were collected with anticoagulant ethylene diamine tetraacetic acid tubes. All interested polymorphisms (FOXP2: rs6980093, rs1456031; COMT: rs4680, rs4818; BDNF: rs6265, rs2049046; Kibra/WWC1: rs17070145) were genotyped using MassARRAY flight mass spectrometry. SNPs were sorted according to the dbSNP database. The gene fragments containing each SNP were amplified by PCR. Following this, for each SNP, the dNTP generated was phosphorylated by alkaline phosphatase reaction to form ddNTP. The UEP was used for a single base extension reaction in the ddNTP system to form a single base extension product complementary to the SNP genotype to be detected, and the results were subjected to resin purification and chip sampling. Finally, the samples were analyzed by a MALDI‐TOF mass spectrometer. Based on the principle that the flight time of ions generated by radiation ionization of matrix molecules in a vacuum environment is directly proportional to mass, genotyping was obtained.

FIGURE 1.

Overview of the procedure and analyses. Experimental procedures contain genotyping for all polymorphisms (step 1), experimental design and time course of the language switching task (step 2), and statistical analyses conducted in the present study (step 3).

Allele frequencies and genotyping distributions of interested gene polymorphisms (FOXP2: rs6980093, rs1456031; COMT: rs4680, rs4818; BDNF: rs6265, rs2049046; Kibra‐WWC1: rs17070145) were consistent with Hard–Weinberg expectations (see Table 2). As AA homozygotes for COMT rs4680, GG homozygotes for COMT rs4818, and CC homozygotes for KIBRA/WWC1 rs17070145 had expectedly low frequencies, we combined them with their polymorphic heterozygotes into one group. The final groups of each polymorphism were as follows: FOXP2 rs6980093 contained GG homozygotes, GA heterozygotes and AA homozygotes; FOXP2 rs1456031 contained TT homozygotes, TC heterozygotes and CC homozygotes; COMT rs4680 was divided into GG homozygotes and A carriers; COMT rs4818 included CC homozygotes and G carriers; BDNF rs6265 was divided into CC homozygotes, CT heterozygotes and TT homozygotes; BDNF rs2049046 was genotyped into AA homozygotes, AT heterozygotes and TT homozygotes; Kibra/WWC1 rs17070145 was divided into TT homozygotes and C carriers.

TABLE 2.

Hardy–Weinberg equilibrium test results.

Polymorphisms	Actual counts (frequency)	Expected counts (frequency)	χ 2	p
FOXP2 rs6980093 (N = 117)			2.37	.124
GG	19 (.16)	23.11 (.20)
GA	66 (.56)	57.78 (.49)
AA	32 (.27)	36.11 (.31)
FOXP2 rs1456031 (N = 117)			.08	.782
TT	30 (.26)	29.25 (.25)
TC	57 (.49)	58.5 (.50)
CC	30 (.25)	29.25 (.25)
COMT rs4680 (N = 117)			.93	.334
GG	56 (.48)	58.17 (.50)
GA	53 (.45)	48.65 (.42)
AA	8 (.07)	1.17 (.08)
COMT rs4818 (N = 117)			2.86	.091
CC	58 (.49)	54.02 (.46)
CG	43 (.37)	5.96 (.44)
GG	16 (.14)	12.02 (.10)
BDNF rs6265 (N = 119)			.52	.469
CC	33 (.28)	34.96 (.29)
CT	63 (.53)	59.08 (.50)
TT	23 (.19)	24.96 (.21)
BDNF rs2049046 (N = 119)			1.28	.258
AA	21 (.17)	24.05 (.20)
AT	65 (.55)	58.89 (.50)
TT	33 (.28)	36.05 (.30)
Kibra/WWC1 rs17070145 (N = 118)			1.54	.215
CC	4 (.03)	6.41 (.05)
CT	47 (.40)	42.18 (.36)
TT	67 (.57)	69.41 (.59)

2.3. Language switching task

In the second step, we administered a picture naming task with cued language switches, in which participants were asked to name simple line drawings as accurately and quickly as possible in their L1 or L2 according to a color cue. We selected 24 black‐and‐white drawings (15 cm × 15 cm) from a standardized picture inventory (Snodgrass & Vanderwart, 1980; Zhang & Yang, 2003), whose Chinese names were two‐characters in length and whose English equivalents were one‐ or two‐syllable words containing 3–6 letters. To ensure that the Chinese and English names were familiar to the participants, we recruited a group of 35 age‐matched participants from the same population, but who did not participate in the formal experiment, and asked them to judge their familiarity with the Chinese and English names of the drawings. Paired‐sample t‐tests revealed no significant differences between their familiarity with words in the two languages (L1: M = 4.85 ± .08, L2: M = 4.84 ± .12, t(23) = .69, p = .76).

Each individually‐presented drawing represented conditions that were either switch trials in which the response language is different from the immediately preceding trial or non‐switch trials where the response language is the same as in the previous trial. The experiment contained a total of 5 blocks, each of which included 98 trials. Each block contained 2 warm‐up trials followed by 48 non‐switch trials (24 L1–L1 and 24 L2–L2) and 48 switch trials (24 L1–L2 and 24 L2–L1) which were randomly distributed throughout the block. The response language (L1 or L2) was also randomized throughout the task.

In the experiment, each trial started with a red or blue square as a cue for 250 ms on the center of the screen followed by a blank screen for 500 ms. A line drawing then appeared until participants named it into a microphone or until 2000 ms passed. A blank screen was presented for 1000 ms before the next trial started. Reaction times (RTs) were recorded by a PSTSR‐BOX connected to a microphone and accuracy of responses was recorded manually by a research assistant. Unfortunately, some responses were too low to be recorded by the microphone and accordingly, we excluded these trials from data analyses (0.83%). Before the formal experiment, participants completed 12 practice trials to ensure that they fully understood the task.

2.4. Data recording and analyses

Electrophysiological data were recorded by 64‐channel caps with Ag/AgCl impedance‐optimized active electrodes (ANT Neuro). Standard electrodes sites were placed according to the extended 10–20 positioning system and impedances were kept below 5 kΩ. The continuous EEG signal was recorded with a 1000 Hz sampling rate, a low cut‐off filter of .01 Hz, and a high cut‐off filter of 100 Hz online. All electrode sites were referenced online to the CPz and re‐referenced offline to the average of the left and right mastoids. Using EEGLAB (Brunner et al., 2013; Delorme & Makeig, 2004) for preprocessing, the electroencephalographic activity was down‐sampled to 500 Hz and refiltered with a high‐pass filter of .1 Hz and a lowpass filter of 30 Hz. Ocular artifact reduction was performed through independent component analysis rejection (Makeig et al., 1996). The continuous recordings were cut into epochs ranging from −200 to 1500 ms relative to the cue onset. A 200 ms prestimulus (i.e., language cue) period was used as a baseline. Signals exceeding ±80 μV in any given epoch were automatically discarded.

Statistical analyses on the behavioral data (i.e., RTs and accuracy) were performed in R (Version 4.0.3; R Core Team, 2019), using the “lme4” package (Bates et al., 2014). The following trials were excluded from the analyses: warm‐up trials (2.04%), null trials (0.83%), and trials with a response time faster than 100 ms (1.03%). Data of the naming RT beyond M ± 3SD per participant were also excluded (0.69%). Accuracy was analyzed through a generalized linear mixed model. A linear mixed‐effect model was conducted on RTs of correct responses. To satisfy the assumption of normally‐distributed residuals for linear models, the RT values were power‐transformed using the Box–Cox method (Box & Cox, 1964; Osborne, 2010; Zhu et al., 2022), which determined the optimal transformation power to be λ = −.14. Fixed effects in the models included “language” (L1/L2), “trial type” (switch/non‐switch), and “genotypes” (for all polymorphisms). Participants and items were added as a random effect and sex was added as a covariate. Because genotypes for BDNF rs6265, BDNF rs2049046, FOXP2 rs6980093, and FOXP2 rs1456031 involved three levels, we performed a three‐way ANOVA to examine any main effects and interactions by comparing the estimated marginal means using the “emmeans” package (Lenth et al., 2020). Moreover, we conducted pairwise comparisons within each gene to assess whether language switch costs (switch vs. non‐switch) were sensitive to genetic variants. All effects were considered statistically significant at p < .05.

The electrophysiological data analyses focused on the N2a (200–350 ms) effect of the language schema phase, N2b (1000–1200 ms), and the LPC (1200–1500 ms) during the lexical response phase. Previous studies have demonstrated that the N2 effect typically reflects conflict detection and monitoring, inhibition, task updating, and heightened demands on cognitive control to resolve the competition between language schemas (Cavanagh et al., 2009; Huster et al., 2013; Kirmizi‐Alsan et al., 2006; Liu, Liang, et al., 2016). However, many studies have reported an N2 effect 200–400 ms after picture onset, suggesting that selective inhibition is engaged to reduce competition during lexical selection and/or phonological encoding (Cheng et al., 2010; Piai et al., 2014; Roelofs, 2003; Shao et al., 2014). The LPC has been linked to language control during lexical access, the disinhibition between languages, and the reactivation of previously suppressed lexical items (Liu et al., 2014; Liu, Schwieter, Wang, et al., 2022; Rodriguez‐Fornells et al., 2006). We spatially pre‐defined anterior (sensors: F5, F3, F1, Fz, F2, F4, F6, FC5, FC3, FC1, FCz, FC2, FC4, FC6) and posterior (sensors: CP5, CP3, CP1, CPz, CP2, CP4, CP6, P5, P3, P1, Pz, P2, P4, P6) ROIs, then extracted the single‐trial amplitude of these sites for the linear mixed‐effect models. Fixed effects in the models included “language” (L1/L2), “trial type” (switch/non‐switch), and “genotypes” (for all polymorphisms). Participants were added as a random effect, while sex was added as a covariate. We again performed a three‐way ANOVA followed by pairwise comparisons. The false‐positive rate was controlled using false discovery rate (FDR) correction.

To test relationships between neural coding and cognitive performance, we conducted a time‐resolved correlation analysis. To do this, we calculated L1 and L2 switch costs using participants' neural activities at each time point of the trials and tested whether they correlated with their RT switch costs. Specifically, a vector of 119 neural switch costs at a single time point was correlated with a vector of 119 behavioral switch costs. The same correlation analysis was then performed on the 200–1500 ms window. The false‐positive rate for this approach was controlled using FDR correction.

3. RESULTS

3.1. The effect of genetic variations in the language‐related gene FOXP2 on language control

Analyses on RT data revealed a significant interaction between trial type and FOXP2 rs1456031, F(2, 113) = 4.00, FDR p = .036. Follow‐up tests for rs1456031 revealed trial type effects for all genotypes (Zs >7.15, FDR ps < .001), showing that switch trials (CC homozygotes: M = 835 ± 248 ms; TT homozygotes: M = 891 ± 257 ms; TC heterozygotes: M = 832 ± 233 ms) elicited slower responses than non‐switch trials (CC homozygotes: M = 810 ± 239 ms; TT homozygotes: M = 852 ± 244 ms; TC heterozygotes: M = 799 ± 219 ms), whereas the differences between the three genotypes did not reach significance for either switch trials or repeat trials. In the accuracy data, we also found a marginally significant interaction of language × trial type × rs1456031, F(2, 113) = 3.35, FDR p = .07. Follow‐up tests revealed that only TT homozygotes elicited a significant interaction of language × trial type, b = −.58, SE = .18, z = −3.13, FDR p = .005, showing a switch cost in both L1, b = −.91, SE = .13, z = −6.91, FDR p = .002, and L2, b = −.33, SE = .13, Z = −2.53, FDR p = .011. The behavioral findings for RTs and accuracy can be seen in Figure 2(a), (b), respectively.

FIGURE 2.

Electrophysiological results for FOXP2 rs1456031 and rs6980093 polymorphisms. (a) Behavioral switch costs in RT data (left) and accuracy results for FOXP2 1456031. (b) Waveforms and mean neural switch costs of ERP components for FOXP2 rs1456031. (c) Waveforms and mean neural switch cost of ERP components for FOXP2 rs6980093. Gray shading represents early (200–300 ms), middle (1000–1200 ms), and late (1200–1500 ms) time windows. Numbers in red represent mean differences (switch trials–non‐switch trials) for that condition. *p < .05.

At the neural level, to investigate whether switch costs significantly interact with FOXP2 polymorphisms, we conducted similar analyses on the electrophysiological data and found that FOXP2 rs1456031 played a predominant role in both the language schema phase (i.e., N2a) and lexical response phase (i.e., N2b and LPC) (see Figure 2(b)), while rs6980093 limitedly showed a significant effect during lexical response phase (i.e., LPC) (see Figure 2(c)). Specifically, analyses on the N2a effect revealed that the interaction of language × trial type varied markedly across genotypes for FOXP2 rs1456031, F(2, 113) = 4.96, FDR p = .014. Follow‐up tests detected an interaction of language × trial type in TT homozygotes, b = .50, SE = .22, t = 2.29, FDR p = .033, and CC homozygotes, b = .65, SE = .22, t = 2.96, FDR p = .009, but not in TC heterozygotes, b = .20, SE = .10, t = 1.96, FDR p = .053. L2 switch costs reached significance for both TT homozygotes, b = −.45, SE = .15, t = −2.92, FDR p = .012, and CC genotypes, b = −.48, SE = .16, t = −3.10, FDR p = .004, whereas no difference was found in the L1 for neither genotype, TT: b = .05, SE = .15, t = .33, FDR p = .745; CC: b = .20, SE = .10, t = 1.96, FDR p = .363.

For the N2b effect, there was a significant interaction between FOXP2 rs1456031, language, and trial type, F(2, 113) = 4.30, FDR p = .028. Separate tests revealed that only TC heterozygotes showed a significant interaction between language and trial type, b = −1.01, SE = .21, t = −4.80, FDR p = .0003. No significant interactions were observed in TT, b = −.05, SE = .30, t = −.18, FDR p = .854, nor CC homozygotes, b = −.34, SE = .30, t = −1.14, FDR p = .379. Follow‐up tests for TC genotype confirmed that L1 switch costs were significant, b = −.87, SE = .20, t = −4.39, FDR p = .002, but not L2 switch costs, b = .15, SE = .20, t = .75, FDR p = .351.

The analyses on the LPC effect found a marginally significant interaction for rs1456031, F(2, 113) = 2.82, FDR p = .060, and a significant three‐way interaction for rs6980093, F(2, 113) = 3.62, FDR p = .036. Separate t‐tests for FOXP2 rs1456031 found that only the TC genotype revealed a reversed L1 switch cost, b = −.64, SE = .16, t = −4.05, FDR p = .002, while there was no significant contrast in L2 switch costs, b = −.01, SE = .16, t = −.07, FDR p = .942. As for FOXP2 rs6980093, follow‐up tests showed a significant interaction between language and trial type in GA heterozygotes, b = −.55, SE = .20, t = −2.70, FDR p = .018, and GG homozygotes, b = −.99, SE = .43, t = −2.31, FDR p = .032, but not in AA homozygotes, b = .23, SE = .30, t = .77, FDR p = .441. Planned pairwise comparisons revealed a reversed L1 switch cost only for GA heterozygotes, b = −.66, SE = .14, t = −4.59, FDR p = .002, with a stronger LPC effect in non‐switch trials than in switch trials, but no significant difference between switch and non‐switch in the L2, b = −.11, SE = .14, t = −.78, FDR p = .437. No significant switch costs were present in GG homozygotes (L1: b = −.56, SE = .30, t = −1.85, FDR p = .128; L2: b = .43, SE = .30, t = 1.42, FDR p = .157).

3.2. The effect of genetic variations in executive‐function‐related genes on language control

3.2.1. COMT polymorphisms

Analyses on RTs and accuracy in the behavioral data revealed no significant effects. A similar linear mixed‐effect model conducted on the electrophysiological data revealed that the COMT rs4680 polymorphism exhibited a significant effect on the language schema phase (N2a) (Figure 3(a)), while the COMT rs4818 polymorphism had a wider influence on language processing as evidenced by significant effects on the N2a and N2b components and the LPC (Figure 3(b)). Moreover, during the language schema phase, we found that the interaction of language × trial type varied markedly across genotypes for COMT rs4680, b = −.61, SE = .22, t = −2.84, FDR p = .007, and COMT rs4818, b = .58, SE = .22, t = 2.69, FDR p = .007. Follow‐up tests for COMT rs4680 showed that GG homozygotes elicited robust L2 switch costs, b = −.38, SE = .11, t = −3.52, FDR p = .002, but not L1 switch costs, b = .17, SE = .11, t = 1.54, FDR p = .125. No significant three‐way interaction was present in A allele carriers, b = −.07, SE = .16, t = −.44, FDR p = .662. Follow‐up tests for the COMT rs4818 polymorphism showed a significant language × trial type effect in G allele carriers, b = .56, SE = .16, t = 3.60, FDR p = .002, but not in CC homozygotes, b = −.02, SE = .16, t = −.16, FDR p = .872. Planned pairwise comparisons for G allele carriers revealed a significant N2a switch cost in the L2, b = −.38, SE = .11, t = −3.45, FDR p = .004, but not in the L1, b = .18, SE = .11, t = 1.64, FDR p = .135.

FIGURE 3.

Waveforms and mean switch costs of ERP components for COMT rs4680 (a) and rs4818 (b). Gray shading represents early (200–300 ms), middle (1000–1200 ms), and the late (1200–1500 ms) time windows. Numbers in red represent mean differences (switch–non‐switch trials) for that condition. * p < .05.

The influence of COMT rs4818 on the N2b time phase showed a two‐way interaction between language and trial type in CC homozygotes, b = −.99, SE = .21, t = −4.66, FDR p = .002, but not in G allele carriers, b = −.30, SE = .21, t = −1.46, FDR p = .144. Planned pairwise comparisons revealed that switch trials had a stronger N2b effect than non‐switch trials in the L1, b = −.74, SE = .15, t = −4.87, FDR p = .002, but not in the L2, b = .26, SE = .15, t = 1.71, FDR p = .116.

Analyses on the LPC for rs4818 revealed showed a significant interaction between language and trial type for CC homozygotes, b = −.81, SE = .22, t = −3.64, FDR p = .002, but not for G allele carriers, b = −.02, SE = .22, t = −.08, FDR p = .938. Reversed switch costs were detected in the L1, b = −.73, SE = .16, t = −4.64, FDR p = .002, but not in the L2, b = .08, SE = .16, t = .50, FDR p = .616.

3.2.2. BDNF polymorphisms

Analyses on behavioral data revealed no significant effects. The electrophysiological analyses found that for BDNF rs6265, there was a robust effect on N2a, F(2, 115) = 7.32, FDR p = .003, and N2b, F(2, 115) = 5.31, FDR p = .015 (Figure 4(a)), whereas for BDNF rs2049046, there was only a significant effect on N2a, F(2, 115) = 3.56, FDR p = .028 (Figure 4(b)). In the language schema phase, follow‐up tests for BDNF rs6265 showed an interaction between language and trial in the TT genotype, b = −.58, SE = .26, t = −2.20, FDR p = .042, and in TC heterozygotes, b = .51, SE = .15, t = 3.54, FDR p = .001, but not in CC homozygotes, b = .38, SE = .22, t = 1.78, FDR p = .08. Specifically, for TT homozygotes, L1 switch costs were significant, b = −.68, SE = .19, t = −3.65, FDR p = .001, but L2 switch costs were not, b = −.11, SE = .18, t = −.58, FDR p = .561. For TC heterozygotes, planned pairwise comparisons showed reversed switch costs for the L1, b = .25, SE = .10, t = 2.39, FDR p = .022. This reversed switch cost effect was not detected in the L2, b = −.27, SE = .10, t = −2.61, FDR p = .018. For BDNF rs2049046, there was a significant language × trial type interaction in AA homozygotes, b = .89, SE = .27, t = 3.35, FDR p = .002. The planned pairwise comparisons showed significant switch costs in the L2, b = −.88, SE = .19, t = −4.71, FDR p = .004, but not in the L1 (switch: M = −.60 ± 8.37 μV; non‐switch: M = −.60 ± 8.36 μV), b = .01, SE = .18, t = .05, FDR p = .964. No such effects were significant in TT homozygotes, b = .12, SE = .22, t = .53, FDR p = .597, nor in AT genotypes, b = .15, SE = .14, t = 1.01, FDR p = .467.

FIGURE 4.

Waveforms and mean switch costs of ERP components for BDNF rs6265 (a), BDNF rs2049046 (b), and Kibra/WWC1 rs17070145 (c). Gray shading represents early (200–300 ms), middle (1000–1200 ms), and the late (1200–1500 ms) time windows. Numbers in red represent mean differences (switch trials–non‐switch trials) for that condition. .* p < .05.

In the lexical response phase (N2b effect), follow‐up tests for BDNF rs6265 confirmed an effect of language × trial type in TT homozygotes, b = −1.43, SE = .35, t = −4.15, FDR p = .003, and CC homozygotes, b = −.77, SE = .29, t = −2.66, FDR p = .012, but not in TC genotypes, b = −.22, SE = .20, t = −1.09, FDR p = .274. Specifically, for TT homozygotes, there was a significant main effect of trial type in L1, b = −1.07, SE = .25, t = −4.33, FDR p = .002, which indicated that switch trials evoked a stronger N2b than non‐switch trials. No such difference was reliable in the L2, b = .37, SE = .24, t = 1.51, FDR p = .131. The L1 switch cost was also significant for CC genotypes, b = −.69, SE = .20, t = −3.37, FDR p = .003, while no such effect was found in the L2, b = .08, SE = .20, t = .40, FDR p = .691. No significant effects were found in the LPC analyses.

3.2.3. Kibra/WWC1 polymorphism

Kibra/WWC1 rs17070145 also exhibited a significant effect on N2a, b = −.72, SE = .22, t = −3.29, FDR p = .002, and a marginally significant effect on N2b, b = −.57, SE = .29, t = −1.97, FDR p = .072 (see Figure 4(c)). Follow‐up tests on N2a confirmed that only TT genotype, not C allele carriers, present a significant interaction between language and trial type (TT: b = .57, SE = .15, t = 3.89, FDR p = .002; C carriers: b = −.14, SE = .17, t = −.83, FDR p = .406). Specifically, TT homozygotes elicited both a reversed switch cost of L1, b = .20, SE = .10, t = 1.96, FDR p = .066, while a typical switch cost of L2, b = −.37, SE = .10, t = −3.55, FDR p = .002. On the other hand, follow‐up tests on N2b detected the interaction between language and trial type was significantly present in C allele carriers, b = −.92, SE = .23, t = −4.03, FDR p = .002, but marginally significant in TT homozygotes, b = −.36, SE = .20, t = −1.82, FDR p = .068. Planned pairwise comparisons showed similar patterns for two genotypes, which showed only a typical L1 switch cost in TT, b = −.41, SE = .14, t = −3.00, FDR p = .006, and C allele carriers, b = −.69, SE = .16, t = −4.23, FDR p = .002. No significant effects were found in the LPC analyses.

3.3. The differential role of genes in bilinguals' brain–behavior correlation

Having separately established the robust impact of genetic variants on language control, we next examined whether the relationship between neural coding and behavioral performance varied across all SNPs. We took participants' L1 and L2 neural switch costs (i.e., the difference between switch trials and non‐switch trials) at each time point and correlated the values with their respective RT switch costs. These analyses revealed distinct patterns among different genotypes during the time windows of interest (i.e., 200–1500 ms following cue onset).

3.3.1. FOXP2 polymorphisms

Correlating neural switch costs with behavioral differences separately on L1 or L2 confirmed that FOXP2 rs6980093 polymorphism triggered distinct correlation patterns among the three genotypes (see Figure 5(a)). The analysis on AA homozygotes detected a marginally positive brain–behavior relationship in the L1 (time window: 452–504 ms, mean rho = .36, FDR p = .057), while there was a significant negative correlation in the L2 (first‐time window: 200–250 ms, mean rho = −.46, FDR p = .03; second‐time window: 608–666 ms, mean rho = −.43, FDR p = .031; third‐time window: 716–1030 ms, mean rho = −.45, FDR p = .030; fourth‐time window: 1140–1162 ms, mean rho = −.37, FDR p = .043; fifth time window: 1188–1286 ms, mean rho = −.41, FDR p = .034). Contrarily, GA heterozygotes showed that L1 neural switch costs negatively correlated with L1 behavioral switch cost (first‐time window: 690–770 ms, mean rho = −.31, FDR p = .027; second‐time window: 1188–1226 ms, mean rho = −.27, FDR p = .038; third‐time window: 1254–1326 ms, mean rho = −.34, FDR p = .025; fourth‐time window: 1366–1454 ms, mean rho = −.29, FDR p = .031), but a positive association was detected in the L2 (first‐time window: 320–560 ms, mean rho = .31, FDR p = .033; second‐time window: 720–750 ms, mean rho = .27, FDR p = .040; third‐time window: 774–836 ms, mean rho = .29, FDR p = .036; fourth‐time window: 990–1068 ms, mean rho = .27, FDR p = .039).

FIGURE 5.

Time‐resolved correlations between neural and behavioral switch costs for FOXP2 and COMT polymorphisms. (a) FOXP2 rs6980093. (b) FOXP2 rs1156031. (c) COMT rs4680. (d) COMT rs4818. Solid lines represent the genotypes, dashed lines in corresponding colors depict 95% confidence bounds for spearman correlation coefficient, and brackets represent significant time windows surviving from FDR correction.

Similarly, the analysis on FOXP2 rs1456031 showed that a negative L1 brain–behavior correlation was significant in TT homozygotes (see Figure 5(b); first‐time window: 1250–1336 ms, mean rho = −.42, FDR p = .037; second‐time window: 1442–1468 ms, mean rho = −.40, FDR p = .038) and TC heterozygotes (Figure 5(b); first‐time window: 1158–1214 ms, mean rho = −.31, FDR p = .032; second‐time window: 1280–1330 ms, mean rho = −.32, FDR p = .032), but positively present in CC homozygotes (Figure 5(b); first‐time window: 368–440 ms, mean rho = .42, FDR p = .033; second‐time window: 476–696 ms, mean rho = .52, FDR p = .019; third‐time window: 954–1046 ms, mean rho = .50, FDR p = .017; fourth‐time window: 1096–1158 ms, mean rho = .40, FDR p = .038; fifth‐time window: 1172–1284 ms, mean rho = .46, FDR p = .022). Moreover, the significant positive association in the L2 was only present in TC heterozygotes (Figure 5(b); first‐time window: 368–426 ms, mean rho = .31, FDR p = .037; second‐time window: 460–484 ms, mean rho = .29, FDR p = .038).

3.3.2. COMT polymorphisms

Two polymorphisms of COMT showed similar patterns in L2 switch costs (see Figure 5(c), (d)). We detected a positive neural‐brain switch cost association in the GG homozygotes for rs4680 (Figure 5(c); first‐time window: 360–418 ms, mean rho = .33, FDR p = .041; second‐time window: 478–518 ms, mean rho = .28, FDR p = .047; third‐time window: 878–904 ms, mean rho = .28, FDR p = .047) and G allele carriers for rs4818 (Figure 5(d); time window: 322–490 ms, mean rho = .34, FDR p = .023). Contrarily, L1 neural switch costs were negatively correlated with cognitive performance in GG homozygotes for COMT rs4680 (Figure 5(c); first‐time window: 718–740 ms, mean rho = −.29, FDR p = .044; second‐time window: 1302–1338 ms, mean rho = −.30, FDR p = .044; third‐time window: 1362–1396 ms, mean rho = −.30, FDR p = .044).

3.3.3. BDNF polymorphisms

For BDNF rs6265, L1 neural switch costs showed a significant correlation with L1 behavioral switch costs only in TT homozygotes (Figure 6(a); first‐time window: 1112–1200 ms, mean rho = −.44, FDR p = .049; second‐time window: 1238–1322 ms, mean rho = −.49, FDR p = .048), while L2 switch costs showed a positive brain–behavior relationship only in CC homozygotes (Figure 6(a); first‐time window: 268–324 ms, mean rho = .43, FDR p = .034; second‐time window: 1124–1146 ms, mean rho = .37, FDR p = .047).

FIGURE 6.

Time‐resolved correlations between neural and behavioral switch costs for BDNF and Kibra/WWC1 polymorphisms. (a) BDNF rs6265. (b) BDNF rs2049046. (c) Kibra/WWC1 rs17070145. Solid lines represent the genotypes, dashed lines in corresponding colors depict 95% confidence bounds for spearman correlation coefficient, and brackets represent significant time windows surviving from FDR correction.

BDNF rs2049046 induced opposite correlation patterns in L1 switch costs, which showed that the brain–behavior relationship was significantly positive in AA homozygotes (Figure 6b; time window: 948–1056 ms, mean rho = .55, FDR p = .022), but negatively present in TT homozygotes (Figure 6(b); first‐time window: 736–756 ms, mean rho = −.40, FDR p = .033; second‐time window: 982–1076 ms, mean rho = −.41, FDR p = .032; third‐time window: 1106–1364 ms, mean rho = −.45, FDR p = .030; fourth‐time window: 1372–1486 ms, mean rho = −.40, FDR p = .034). Moreover, the AA genotype also revealed a positive correlation between neural and behavioral switch costs in the L2 (Figure 6(b); time window: 1120–1192 ms, mean rho = .51, FDR p = .029).

3.3.4. Kibra/WWC1 polymorphism

Analyses on Kibra‐WWC1 rs17070145 revealed a negative brain–behavior relationship in the L1 among C allele carriers (Figure 6(c); time window: 738–758 ms, mean rho = .29, FDR p = .048), and an L2 positive correlation in TT homozygotes (Figure 6(c); time window: 364–386 ms, mean rho = .26, FDR p = .047).

3.4. Summary of results

Statistical analyses for FOXP2 rs1456031 revealed a significant effect of rs1456031 polymorphism on behavioral switch costs in RT and accuracy and neural switch costs in N2a, N2b, and LPC. For COMT rs4818, we found that rs4818 significantly interacted with neural switch costs in N2a, N2b, and LPC, while BDNF rs6265 and Kibra/WWC1 rs17070145 showed influences on neural switch costs in N2a and N2b. Polymorphisms for COMT rs4680 and BDNF rs2049046 showed a limited effect on processing of language schema phase (N2a). Brain–behavior relationship analyses found that FOXP2 and BDNF polymorphisms broadly modulated the predictability of neural switch costs on behavioral switch costs in two phases, whereas polymorphisms for COMT and Kibra/WWC1 narrowly affected such predictability. The results from pairwise comparisons on behavioral and electrophysiological data are shown in Tables 3 and 4, respectively.

TABLE 3.

Summary of the pairwise comparisons for the behavioral data.

Gene		RT	ACC
FOXP2 rs1456031	TT	L1: switch > non‐switch L2: switch > non‐switch	L1: switch > non‐switch L2: switch > non‐switch
	TC	L1: switch > non‐switch L2: switch > non‐switch	‐
	CC	L1: switch > non‐switch L2: switch > non‐switch	‐

TABLE 4.

Summary of the pairwise comparisons for the electrophysiological data.

Gene		Language schema phase	Lexical response phase
		N2a	N2b	LPC
FOXP2 rs6980093	GA	‐	‐	L1: non‐switch > switch
FOXP2 rs1456031	TT/CC	L2: switch > non‐switch	‐	‐
	TC		L1: switch > non‐switch	L1: non‐switch > switch
COMT rs4680	GG	L2: switch > non‐switch	‐	‐
COMT rs4818	CC	‐	L1: switch > non‐switch	L1: non‐switch > switch
	G+	L2: switch > non‐switch	‐	‐
BDNF rs6265	TT	L1: switch > non‐switch	L1: switch > non‐switch	‐
	TC	L1: non‐switch > switch L2: switch > non‐switch	‐	‐
	CC	‐	L1: switch > non‐switch	‐
BDNF rs2049046	AA	L2: switch > non‐switch	‐	‐
Kibra/WWC1 rs17070145	TT	L1: non‐switch > switch L2: switch > non‐switch	L1: switch > non‐switch	‐
	C+	‐	L1: switch > non‐switch	‐

Note: G⁺ represents G allele carriers for COMT rs4818; C+ means the C allele carriers for Kibra/WWC1 rs17070145.

4. DISCUSSION

This study examined the interaction between language control and genetic variations and whether these variations modulated the relationship between neural coding and cognitive performance. We conducted a picture‐naming task with cued language switches and tested the effects of a language‐related gene (i.e., FOXP2) and executive function‐related genes (i.e., COMT, BDNF, and Kibra/WWC1 polymorphisms), and found that: (1) FOXP2 rs1456031 showed a wide‐ranging effect on language control, including behavioral performance (RTs and accuracy) and neural coding across three‐time phases (N2a, N2b, and LPC); (2) COMT rs4818, BDNF rs6265, and Kibra/WWC1 rs17070145 polymorphisms significantly influenced two‐time phases (N2a and N2b); (3) further brain–behavior relationship analyses indicated that language‐related gene and executive function‐related genes can modulate the predictability of neural switch costs on behavioral switch costs in two phases of language control. Taken together, these findings demonstrate that language control in bilinguals is not simply an adaptive function of executive control, but rather its essence is the integration of language‐switching experience and inherent genomes.

4.1. Endogenous language ability influences language control

An individual's capacity of acquiring speech and language must derive, at least in part, from their genome. A substantial body of research has shown that disruptions or structural variants (i.e., chromosome translocation or inversion) of FOXP2 can cause complications with speech motor programming, which in turn, affects production, sequencing, timing, and stress (Crespi et al., 2017; Fisher & Scharff, 2009; Liégeois et al., 2003; Marcus & Fisher, 2003; Morgan et al., 2017). Our study further demonstrated that FOXP2 variations influenced language control from the language schema phase until the utterance of the response. For rs1456031 polymorphism, TT homozygotes and TC heterozygotes displayed significant L2 switch costs during the language schema phase, suggesting that these carriers recruit additional executive functions to inhibit a more dominant language (i.e., the L1). However, during the lexical response phase, only TC heterozygotes displayed significant L1 switch costs during the N2b time phase and reversed L1 switch costs during the LPC phase. This finding suggests that selective inhibition engaged in lexical selection facilitates subsequent lexical retrieval. We found no modulatory effect of FOXP2 rs1456031 polymorphism on RTs, as TT, TC, and CC exhibited significant switch costs in both the L1 and L2. However, with respect to accuracy, TT homozygotes displayed L1 and L2 switch costs, while TC and CC did not. Neuroimaging studies have reported that rs6980093 can modulate speech category learning and development of language networks during reading tasks (Chandrasekaran et al., 2015; Pinel et al., 2012). Specifically, the A allele, compared to GG, has been associated with greater activation of the left inferior frontal gyrus during reading (Pinel et al., 2012), while the GG genotype has displayed higher accuracy in a non‐native learning task relative to AA due to a more efficient switch to a reflective, procedural‐based learning system that involves the executive corticostriatal loop (Chandrasekaran et al., 2015). Our study, however, found that only GA heterozygotes elicited reversed L1 switch costs during the LPC, indicating a higher cognitive demand on non‐switch trials.

Furthermore, we found that FOXP2 polymorphisms widely modulated the brain–behavior relationship of neural and behavioral switch costs. The AA genotype for rs6980093 showed a positive relationship with the L1 in the language schema phase but a negative relationship with the L2 in both the language schema and lexical response phases. The GA genotype showed a negative relationship with the L1 in the later language schema phase (690–750 ms) and lexical response phase and a positive relationship with the L2 during both phases. Contrarily, the CC genotype for rs1456031 displayed a positive relationship with the L1 in the language schema and lexical response phases, and with the L2 in the language schema phase, while the TT genotype showed a negative relationship with the L1 in the lexical response phase. The TC genotype revealed a negative relationship with the L1 during the lexical response phase and a positive relationship with the L2 during the language schema phase. Together, these findings suggest that FOXP2 polymorphisms extensively modulate the predictability of neural coding and cognitive performance and that this predictive relationship is exhibited in different phases of language control processing. These findings underscore the coding preferences of language control processing for each genetic variation.

4.2. Executive function‐related genes influence language control

The findings confirmed that executive function‐related genes play a critical role not only in language schema phase but also in the lexical response phase. For instance, COMT rs4818 significantly interacted with three ERP components (N2a, N2b, and LPC). Specifically, G allele carriers exhibited L2 switch costs in the language schema phase, whereas the CC genotype had significant L1 switch costs in the lexical response phase. Contrastively, BDNF rs6265 and Kibra/WWC1 rs17070145 significantly interacted with the N2a and N2b components. The TT genotype for BDNF rs6265 revealed significant L1 switch costs for both N2a and N2b, while TC displayed reversed L1 switch costs and typical L2 switch costs. The CC genotype showed L1 switch costs during the time phase related to N2b. In addition, we found that the Kibra/WWC1 polymorphism exhibited different effects on the language schema and lexical response phases. The TT homozygotes displayed reversed L1 switch costs and typical L2 switch costs in the language schema phase, but only L1 switch costs in the lexical response phase (N2b). The C allele carriers had significant L1 switch costs in the lexical response phase (N2b) but note in the language schema phase. These findings demonstrate that bilinguals' language control is not exclusively a consequence of switching experiences. Among the three executive function genes of interest in the present study, COMT rs4818 was the only polymorphism that affected all three ERP components. Studies have reported that COMT polymorphism (rs4680) is related to L2 learning, as evidenced by changes in white matter tracts among G allele carriers, but not among AA genotype (Mamiya et al., 2016). Sugiura et al. (2017) found that six‐ to eight‐year‐old children carriers of A allele performed better on language abilities than those with the GG genotype for COMT rs4680. Our findings revealed that COMT rs4818 exerts a broader influence across the time course of language control than COMT rs4680, whose effects are limited to the language schema phase.

Moreover, the executive function‐related genes also modulate the relationship between neuroprocessing and cognitive performance for language control during the language schema and lexical response phases. The GG genotype for COMT rs4680 exhibited a negative L1 brain–behavior relationship but a positive relationship in two phases. The G allele carriers for COMT rs4818 exhibited a negative brain–behavior relationship with the L2 during the language schema phase. The TT genotype for BDNF rs6265 showed a negative L1 relationship in the lexical response phase, while CC homozygotes exhibited a positive L2 relationship in both language schema and lexical response phases. The brain–behavior relationship varied by genetic variations of Kibra/WWC1, in which C allele carriers showed a negative L1 relationship in the language schema phase, while the TT carriers showed a positive L2 relationship in the language schema phase. These findings suggest that polymorphisms for executive function‐related genes may modulate the relationship between neural coding and cognitive performance.

4.3. Language control is shaped by the integration of language‐switching experience and inherent genomes

The configuration of language control and executive function shares multiple subsets of cognitive skills, such as inhibitory control, conflict monitoring, updating, and shifting (Abutalebi et al., 2012; Abutalebi et al., 2013; Blanco‐Elorrieta & Pylkkänen, 2016; Branzi et al., 2016; Coderre et al., 2016; De Baene et al., 2015; De Bruin et al., 2014; Emmorey et al., 2008; Wu et al., 2019). This claim is central to the ACH (Green & Abutalebi, 2013) which argues that bilinguals' language control is adaptive to situational and communicative needs. This view is strongly supported by studies that have reported overlapping neural substrates for both language and cognitive control (Abutalebi, 2008; Blanco‐Elorrieta & Pylkkänen, 2016; Branzi et al., 2016; Coderre et al., 2016; Crinion et al., 2006; Hernandez, 2009; Hernandez et al., 2001; Rodriguez‐Fornells et al., 2002; Wang et al., 2007). Further to these findings, the present study provides compelling evidence that language control is the integration of an individual's switching experience and their inherent genome. We found that, like the language‐related gene FOXP2, the executive function‐related genes COMT, BDNF, and Kibra/WWC1 exhibited robust effects in both the language schema and lexical response phases, and modulated the brain–behavior relationships on language control. These findings point to a similar influence of the language‐related gene and executive function‐related genes on language control processing. Moreover, our analyses on the behavioral data revealed that the language‐related gene FOXP2 exhibited a broader impact on language control.

To our knowledge, our study is the first to explore the influence of genetic bases of language ability and executive function on the processes of how bilinguals control their two languages. Previous studies have reported on how variations of the FOXP2 gene affect language and language development (Chabout et al., 2016; Crespi et al., 2017; Fisher & Scharff, 2009; Liégeois et al., 2003; Mozzi et al., 2017), on how the COMT genes affect prefrontal‐guided cognition (Barnett et al., 2008; Chen et al., 2004; Klanker et al., 2013; Logue & Gould, 2014; Sannino et al., 2015; Witte & Flöel, 2012; Zhang et al., 2015), and how the BDNF and Kibra/WWC1 genes modulate memory functioning (Almeida et al., 2008; Beste et al., 2010; Egan et al., 2003; Hariri et al., 2003; Zhang et al., 2014). In our study, we have shown that genetic variations of these genes differentially influence bilinguals' language control processing. Polymorphisms of FOXP2 rs1456031, COMT rs4818, BDNF rs6265, and Kibra/WWC1 rs17070145 showed significant effects during language schema and lexical response phases, whereas FOXP2 rs6980093, COMT rs4680, and BDNF rs2049046 only affected the language schema phase. Finally, the brain–behavior correlation analyses revealed differential effects of each polymorphism on the predictability of neural coding and cognitive performance. Specifically, we found that COMT rs4818 and Kibra/WWC1 rs17070145 narrowly modulated the relationship between neural coding and cognitive performance in the language schema phase, however, the other polymorphisms showed effects during the language schema and lexical response phases. These findings further suggest that these particular genetic variations affect processing preferences and recruit more cognitive resources during different phases of language production.

4.4. Limitations

The present study demonstrated that bilinguals with different genotypes flexibly adapt to external needs (i.e., the cues) in the dual‐language context. According to Green and Abutalebi (2013), there are three language contexts that have different requirements for language control: single‐language context, dual‐language context, and dense code‐switching context. The present study did not reveal that genes affect the adaptation of language control on these different language contexts. On the other hand, this study used EEG to examine genes' effects on electrical activity patterns. It is still unclear as to whether language‐related genes and executive function‐related genes will elicit similar effects on the cortical activation patterns or brain network of bilingual language control.

5. CONCLUSION

This study uniquely sheds new light on the genetic basis of bilingual language control. In the study, we found that the language‐related gene and executive function‐related function genes exhibit robust influences on bilingual language control during language schema and lexical response phases. Our findings have also revealed the modulatory nature of brain–behavior relationships on language control. As hypothesized, our findings suggest that language control among bilinguals is neither a language‐specific ability nor merely an adaptive part of executive functions, but rather a reflection of an individual's switching experience and their inherent genome.

CONFLICT OF INTEREST STATEMENT

We have no known conflicts of interest to disclose.

Liu, D. , Xing, Z. , Huang, J. , Schwieter, J. W. , & Liu, H. (2023). Genetic bases of language control in bilinguals: Evidence from an EEG study. Human Brain Mapping, 44(9), 3624–3643. 10.1002/hbm.26301

DATA AVAILABILITY STATEMENT

The datasets generated and analyzed in this study are available in the OSF repository: Liu, H. (2022, August 16). The genetic bases of language control. Retrieved from osf.io/fm432.