From CNTNAP2 to Early Expressive Language in Infancy: The Mediation Role of Rapid Auditory Processing

Abstract

Although it is clear that early language acquisition can be a target of CNTNAP2, the pathway between gene and language is still largely unknown. This research focused on the mediation role of rapid auditory processing (RAP). We tested RAP at 6 months of age by the use of event-related potentials, as a mediator between common variants of the CNTNAP2 gene (rs7794745 and rs2710102) and 20-month-old language outcome in a prospective longitudinal study of 96 Italian infants. The mediation model examines the hypothesis that language outcome is explained by a sequence of effects involving RAP and CNTNAP2. The ability to discriminate spectrotemporally complex auditory frequency changes at 6 months of age mediates the contribution of rs2710102 to expressive vocabulary at 20 months. The indirect effect revealed that rs2710102 C/C was associated with lower P3 amplitude in the right hemisphere, which, in turn, predicted poorer expressive vocabulary at 20 months of age. These findings add to a growing body of literature implicating RAP as a viable marker in genetic studies of language development. The results demonstrate a potential developmental cascade of effects, whereby CNTNAP2 drives RAP functioning that, in turn, contributes to early expressive outcome.

Introduction

Over the first months of life, the time course of early expressive language development varies greatly among children (Henrichs et al. 2011). Spontaneous production of words emerges at about 12 months and during the following months children slowly but surely increase the number of words they know and say. This slow and steady vocabulary growth is followed by an impressive burst in acquisition with a rapid increase in the number of words produced starting at about 16–20 months of age (Goldfield and Reznick 1990; Fenson et al. 1994).

Twin studies have assessed the etiology of individual differences in language acquisition, demonstrating the crucial role of genetic influences on 20- to 24-month-old expressive vocabulary with heritability estimates ranging from 0.16–0.38 (Dale et al. 1998). Molecular genetic studies clearly suggest that multiple genes are involved in the etiology of language (Rodenas-Cuadrado et al. 2014), and convergent research emphasized a compelling role of CNTNAP2 on language acquisition both in clinical (Alarcon et al. 2008; Vernes et al. 2008; Rodenas-Cuadrado et al. 2014) and in general population samples (Whalley et al. 2011; Whitehouse et al. 2011; Kos et al. 2012). CNTNAP2 is a gene located on chromosome 7q36, and it codes for a neurexin superfamily member, whose functions in the nervous system are relevant to cell–cell interactions and ion channel expression (Poliak et al. 2003).

CNTNAP2 is highly expressed in the frontal and anterior regions of the cortex that correspond to cortico-striato-thalamic circuitry associated with language-related processing (Alarcon et al. 2008). In families of probands affected by language impairment, an association has been found between the rs2710102 single nucleotide polymorphism (SNP) and both receptive and expressive outcomes (Alarcon et al. 2008; Vernes et al. 2008). Interestingly, a significant association was identified in a general population sample between a measure of parent-reported communicative behavior at 24 months of age and an intronic cluster of CNTNAP2 SNPs, including rs2710102, suggesting that genetic variation at this locus may have a role in normal individual differences in expressive language at early stages of the development (Whitehouse et al. 2011).

However, because of the complexity of language, a one-to-one relationship between CNTNAP2 and quantitative measures of language has not emerged consistently across studies, and the key pathways connecting them remain obscure (Rodenas-Cuadrado et al. 2014). To better understand the link between molecular mechanisms and language, researchers have recently used neuroimaging and event-related potentials (ERPs) to examine the association of CNTNAP2 with functional brain outcomes related to specific linguistic processes (Poliak et al. 2003; Folia et al. 2008; Snijders 2010; Whalley et al. 2011; Whitehouse et al. 2011; Kos et al. 2012). Whalley et al. (2011) performed a functional magnetic resonance study in healthy adults during a sentence completion task and showed significant associations between 2 CNTNAP2 SNPs (rs7794745 and rs2710102) and brain activity in contralateral areas of traditional language regions. Furthermore, other brain studies reported that CNTNAP2 rs7794745 played a crucial role in sentence processing and in artificial syntax acquisition (Folia et al. 2008; Snijders 2010; Kos et al. 2012).

Although it is clear from genetic expression, neuroimaging, and ERP studies that the language neural networks can be influenced by CNTNAP2 effects, the underlying mechanisms and/or the pathway between gene and language are still largely unknown.

A large literature has emphasized the role of rapid auditory processing (RAP), in terms of processing brief and rapidly occurring successive auditory cues, in language development (Tallal and Piercy 1973; Tallal 1980; Tallal and Gaab 2006). Previous studies have shown that RAP abilities are highly heritable (Dionne et al. 2013; Brewer et al. 2016), and a substantial genetic influence was seen on nonspeech-based auditory processing skills, with heritability estimates ranging from 0.31 to 0.74 (Brewer et al. 2016).

Interestingly, an animal study showed RAP abnormalities in Cntnap2 KO mice (Truong et al. 2015). In particular, this study compared Cntnap2 KO and control mice on their performance in a series of auditory processing tasks (i.e., silent gap detection, embedded tone, and pitch discrimination). Cntnap2 KO mice displayed enhanced processing in pure frequency tasks (embedded tone and pitch discrimination), whereas they were impaired in a spectrotemporally complex task (silent gap detection), suggesting that deficits were more related to auditory temporal-envelope processing with a high temporal demand, rather than to simple frequency discrimination. Moreover, stereological analysis critical structures of the central auditory pathway revealed fewer and smaller neurons in the medial geniculate nucleus of Cntnap2 KO mice compared with controls; more strikingly, neuronal size in the medial geniculate nucleus significantly predicted performance in the silent gap detection task, suggesting that morphological changes may underlie the spectrotemporally complex auditory processing abnormalities found in Cntnap2 KO mice.

Previous research has shown that deficits in auditory processing can be identified very early in the first year using converging behavioral and electrophysiological methods (e.g., Benasich and Tallal 1993; Benasich and Tallal 2002; Benasich et al. 2002; Guttorm et al. 2005; Leppanen et al. 2010; van Zuijen et al. 2012; Cantiani et al. 2016; Piazza et al. 2016). Studies using ERPs found that the ability to perform fine nonspeech acoustic discrimination in early infancy is critically important to and highly predictive of later language development in both typically developing infants and in infants at familial risk for language learning disorders (Benasich et al. 2002; Guttorm et al. 2005; Choudhury and Benasich 2011; Cantiani et al. 2016; Piazza et al. 2016), supporting RAP as early predictors of language development and markers of risk.

Despite the fact that RAP has been consistently found to be a strong predictor of language development (Benasich et al. 2002; Guttorm et al. 2005; Choudhury and Benasich 2011; Cantiani et al. 2016; Piazza et al. 2016) and the recent evidence for substantial genetic influences on RAP (Dionne et al. 2013; Brewer et al. 2016), and specifically of CNTNAP2 (Truong et al. 2015), no human studies have examined simultaneously the role of CNTNAP2 on RAP and language.

Based on these data, we conducted a prospective longitudinal study of 96 Italian infants and measured RAP abilities at 6 months of age as a possible mediator of the impact of rs7794745 and rs2710102 common variants on 20-month-old language outcome.

To our knowledge, no previous study has attempted to examine the association between CNTNAP2, RAP, and language using a prospective, longitudinal study design. Using structural equation modeling (SEM), the current study aimed to determine: (1) if there is an association between CNTNAP2 and RAP and (2) within a longitudinal framework, if 6-month-old RAP is a mediator of the contribution of CNTNAP2 common variants upon expressive language at 20 months of age.

Methods and Materials

Participants

The sample consisted of 2 subsamples of infants of 6 months of age. The first included 77 infants recruited by local advertisement from 2 hospitals in Northern Italy. The second subsample, recruited at the Medea Institute, consisted of 19 infants at high risk for language and/or learning impairment (LLI), who had at least 1 sibling with a formal diagnosis of LLI. The present sample partially overlaps (48%) the sample examined in a recently published study (Cantiani et al. 2016). At 6 months of age, electroencephalography (EEG) data were recorded, DNA samples were collected, and the cognitive subscale of the Bayley Scales of Infant Development (Bayley 1993) was administered. In addition, socio-demographic and pre/perinatal information was collected using a self-administered questionnaire for parents. Finally, at 20 months of age, follow-up information on expressive language development was collected by means of a parental questionnaire.

Infants were included in the study if: (1) both parents were native-Italian speakers, (2) gestational age was ≥36 weeks, (3) birth weight was ≥ 2000 g, and (4) Bayley cognitive score (Bayley 1993) was ≥7.

After the application of these criteria, 96 six-month-old infants (mean = 6.48 months; SD = 0.49, female-to-male ratio = 0.86) who had completed the experimental procedure and DNA collection were included in the study. Given that family risk was overrepresented in the total sample, the mediation model was first tested on the total sample, and then excluding infants at high risk to control for potential sampling biases. Descriptive statistics of demographic and clinical characteristics are reported in Table 1. The study was approved by the Ethical and Scientific Committees of the Institutes involved.

Experimental Task: RAP

RAP was assessed by means of an electrophysiological task tapping the ability to process rapidly changing and spectrotemporally complex auditory stimuli. The paradigm and stimuli were identical to those used in previous research (Cantiani et al. 2016; Piazza et al. 2016). A nonspeech auditory task paradigm was used in which pairs of brief complex tones were presented with an interstimulus interval of 70 ms. Standard tone pairs (STD stimuli) were composed of 2 identical tones (F0 = 100 Hz, duration = 70 ms). Two complex deviant tone pairs differing with respect to the second tone were presented: in “deviant for frequency” stimuli (DEVF), the second tone had a fundamental frequency of 300 Hz; in “deviant for duration” stimuli (DEVD), the second tone duration was 200 ms. The stimuli were presented in a passive oddball paradigm in which 1200 stimuli (80% STD, 10% DEVF, and 10% DEVF) were delivered in a pseudorandom order, with an intertrial interval (offset-to-onset) varying randomly from 700 to 900 ms.

EEG Data Acquisition and Preprocessing

During EEG recording, children were seated on their caregiver’s lap in a sound-attenuated and electrically shielded room. Auditory ERPs were recorded from 60 scalp sites using a dense-array EGI recording system (Electric Geodesic, Inc.). Vertex was used as an online reference. EEG was sampled at 250 Hz and bandpass filtered (0.1–100 Hz) online. After recording, data were processed using EEGLAB (Delorme and Makeig 2004) and ERPLAB (Lopez-Calderon and Luck 2014). EEG data processing procedures were identical to those used in Cantiani and colleagues (2016). Continuous EEG data were bandpass filtered offline at 0.5–30 Hz. Channels with high impedance (>50 KΩ) or visually evident noise were interpolated with a spherical spline (never more than 12 of the 60 channels). The signals were then rereferenced to an average reference. For the STD, only the responses to the immediate predeviant STD were included in the average. The continuous EEG was segmented according to stimulus type (predeviant STD, DEVF and DEVD) with 900 ms epoch lengths. The 100 ms prestimulus segment was used for baseline correction. Bad EEG epochs were identified and rejected using 2 automatic criteria, followed by visual inspection. First, a moving window (200 ms width, 50 ms step) was used to identify segments containing signals with voltage differences >150 μV. Then, trials whose spectrum (in one or more channels) deviates from baseline by +25/−100 dB in the frequencies >20 Hz were removed (Delorme et al. 2007).

ERP Extraction

Time windows and electrode sites to be submitted to statistical analyses were selected based on mass univariate analyses applied to a subset of ERP data (Cantiani et al. 2016). For each participant and each component of interest, ERPs were extracted from a subset of 18 electrodes localized in the left and right fronto-central areas. Data were then averaged in 2 clusters corresponding to left and right fronto-central areas, each including 9 channels.

To examine the role of RAP, we focused on the large positive waveform corresponding to the mismatch response (P3 component), reflecting a neural change detection process. Based on the abovementioned mass univariate analyses, mean amplitudes were calculated for different time windows: 350–550 ms time window for the STD and DEVF and a 420–620 ms for DEVD.

Linguistic Outcome at 20 Months

Infants’ expressive language at 20 months of age (M = 20.66 months; SD = 0.43) was rated by the parent-administered Language Development Survey (LDS) that has recently been standardized on the Italian population (Rescorla et al. 2014). The number of total words that the child produces corresponds to the total vocabulary score. Norms are available from 18 to 35 months of age (Rescorla and Alley 2001). Percentile scores were entered in the analyses.

CNTNAP2 Genotyping

Saliva samples were collected to obtain DNA. Two intronic CNTNAP2 SNPs (rs7794745 and rs2710102), selected based on previous literature (Vernes et al. 2008; Whalley et al. 2011; Whitehouse et al. 2011; Kos et al. 2012; Udden et al. 2016), were analyzed by quantitative PCR and typed using TaqMan SNP Genotyping assays on a 7900 HT Sequence Detection System (Life Technology).

Statistical Analysis

Table 2 presents allele and genotype frequencies and Hardy–Weinberg equilibrium tests for both SNPs. We tested the additive genetic model and the genotypes were classified into three-level variables: A/A, A/T, T/T for rs7794745 and C/C, C/T, T/T for rs2710102.

CNTNAP2 and RAP

Given the correlations among P3 measures (average r = 0.37, Supplementary Table S1), main CNTNAP2 effects were analyzed by MANOVA for rs7794745 and rs2710102, which were entered as independent variables upon P3 measures, entered as dependent variables. Bonferroni corrected P-value threshold for significance was set at 0.025 (2 SNPs).

RAP and Language Outcome

Pearson’s product moment correlations were calculated between mean amplitude of P3 components (separate by type of stimuli and by hemisphere) and expressive vocabulary percentile score.

CNTNAP2 and Language Outcome

The association of the 2 CNTNAP2 SNPs with the language score was assessed by 2 ANOVAs. The expressive language score was entered as dependent variable, and rs7794745 and rs2710102, respectively, were set as factors. For each ANOVA, Bonferroni corrected P-value threshold for significance was set at 0.025 (2 SNPs).

CNTNAP2 and Language: The Mediation Role of RAP

As we were interested in the indirect effects of RAP on the association between CNTNAP2 and language outcome, we used SEM as implemented in the MPLUS software package (Muthén and Muthén 2014). Indirect effects were examined using the bias-corrected 5000 bootstrap technique to assess non-normality in the product coefficient (Fritz and Mackinnon 2007). Confidence intervals (95% CIs) that do not contain zero indicated significant indirect effects (Mackinnon et al. 2004; Taylor et al. 2008; Tofighi and MacKinnon 2011). This method offers the best power, confidence interval placement, and overall control for Type I error (Williams and Mackinnon 2008). SEM simultaneously models all paths, giving more powerful, accurate and robust estimation of mediation effects than more traditional tests based on sequential regressions, especially when more than 1 mediator is implemented in the model.

Although there is no golden rule for the assessment of model fit, reporting a variety of indexes is advisable (Crowley and Fan 1997) because different indexes reflect different aspects of model fit. To evaluate the goodness-of-fit of the model, model fit was evaluated by use of the Chi-square statistic, the standardized root mean square residual (SRMR, with values ≤0.08 indicating adequate fit), the root mean square error of approximation (RMSEA, with values ≤0.08 indicating adequate fit), and the comparative fit index (CFI, with values ≥0.95 indicating adequate fit).

To control for the oversampling of infants at high risk for LLI, we repeated the mediation model excluding infants at high risk.

Results

Association Between CNTNAP2 and RAP

The MANOVA revealed a main effect of rs2710102 (Roy’s Largest Root_(6,89) = 0.175; P = 0.023) on RAP. The subsequent ANOVAs showed significant associations between rs2710102 and P3 for DEVF on the right hemisphere (F_(2,95) = 5.18; P = 0.007) and between rs2710102 and P3 for DEVD on the left hemisphere (F_(2,95) = 4.60; P = 0.012). Post hoc analyses showed significantly smaller P3 amplitude for DEVF (right) and for DEVD (left) in rs2710102 C/C genotype group (mean DEVF = 3.50; SD = 2.93; mean DEVD = 2.97; SD = 2.12) compared with the other groups (C/T group: mean DEVF = 4.83; SD = 2.36; mean DEVD = 3.90; SD = 3.04 and T/T group: mean DEVF = 6.01, SD = 3.08; mean DEVD = 5.34; SD = 2.20; post hoc LSD range P_s = 0.002–0.038). Descriptive statistics of the P3 amplitude for each genotype (C/C, C/T, and T/T), hemisphere (left and right), and condition (STD, DEVF, and DEVD) are reported in Table 3. In Figure 1, the grand average waveforms are shown for rs2710102 genotypes (C/C, C/T, and T/T) for 2 channels, 13 (corresponding to F5 in the 10/10 international system) and 59 (corresponding to F6), each representative of the 2 channel clusters, located, respectively, in the left and right fronto-central areas. Finally, the MANOVA revealed no significant main effects for rs7794745 on RAP.

Associations Between RAP and Language Outcome

The associations between ERPs components and language were assessed using Pearson’s correlations. Correlations revealed a significant positive association between mean amplitude of the P3 component for DEVF on the right hemisphere (r = 0.25; P = 0.028): infants with smaller P3 amplitude at 6 months of age produced fewer words at 20 months (see Fig. 2).

Association Between CNTNAP2 and Language Outcome

A one-way ANOVA showed no significant associations between both CNTNAP2 markers and expressive language (F_(1,79) = 0.43; P = 0.652 for rs7794745; F_(1,79) = 1.13; P = 0.328 for rs2710102).

RAP as Intermediate Phenotype: A Mediation Analysis

Having shown that individual differences in expressive language are predicted by the P3 amplitude for DEVF on the right hemisphere, we next explored a mediation model, which assumes that CNTNAP2 contributes to mean amplitude P3 DEVF component that in turn affects expressive language outcome.

The mediation model provided a good fit to the data [χ²(3) = 16.06, P = 0.001; RMSEA = 0.000, CFI = 1.00; SRMR = 0.000] and explained 11.0% of the variance in the expressive language at 20 months. Standardized estimates of path coefficients are depicted in Figure 3.

Using 5000 bootstrapping analyses and bias-corrected 95% CI (38), the significant indirect effect was the path from rs2710102 to the expressive language via the P3 DEVF component (β = 0.089; SE = 0.044; 95% CI = 0.001–0.023; P = 0.043). Examination of the indirect effect revealed that the rs2710102 C/C genotype group was associated with lower P3 amplitude on right hemisphere, which, in turn, predicted poorer expressive vocabulary at 20 months.

The mediation model excluding the oversamples of infants at high risk for LLI also provided good fit to the data (χ²(3) = 11.38, P = 0.010; RMSEA = 0.000, CFI = 1.00; SRMR = 0.000) and the indirect effect remained significant (indirect effect: β = 0.020; SE = 0.011; 95% CI = 0.002–0.040; P = 0.046).

Discussion

The aim of this study was to investigate the role of 2 common variants in CNTNAP2 on RAP skills in an Italian sample of 6-month-old infants. Moreover, for the first time, we attempted to disentangle the complex relations between CNTNAP2, the ability to process rapid auditory stimuli and expressive language, using a longitudinal perspective.

Association Between CNTNAP2 and RAP

Our results show that rs2710102 in the CNTNAP2 gene accounts for the ERP-P3 component amplitude in a nonspeech auditory oddball paradigm, in which 2 types of complex tone pairs were presented, that is, second tone being deviant either for frequency (DEVF) or for duration (DEVD). Six-month-old infant carriers of the rs2710102 C/C genotype showed reduced P3 amplitude in both DEVF and DEVD conditions compared with the other genotypic groups, suggesting an involvement of CNTNAP2 in the processing of spectrotemporally complex auditory stimuli, and more generally, supporting a role of CNTNAP2 in language-related phenotypes (Poliak et al. 2003; Folia et al. 2008; Snijders 2010; Whalley et al. 2011; Whitehouse et al. 2011; Kos et al. 2012).

This is the first study assessing the role of CNTNAP2 on RAP in humans, and findings are strikingly consistent with a recent animal study (Truong et al. 2015) that tested a series of auditory processing tasks in the Cntnap2 KO mouse, which has been proposed as an animal model for autism spectrum disorder (ASD), a neurodevelopmental disorder that can be associated with language impairment. Truong et al. (2015) found that Cntnap2 KO mice showed a significant advantage in processing pure frequency tasks (e.g., pitch detection), and were impaired in processing spectrotemporally complex auditory stimuli (silent gap detection), which is in line with ASD-related complex sensory atypicalities. Our study is consistent with the Truong et al. (2015) finding in showing that CNTNAP2 is related to the impairment of auditory processing when a high temporal demand is in place, as it is the case of our stimuli characterized by rapid changes. Additionally, it crucially demonstrates that this link is relevant for language development in humans, and evident as early as the first 2 years of life. In humans, CNTNAP2 expression is enriched in frontal and perisylvian brain regions that are implicated in higher cognitive functions including language but it has also been found expressed in subcortical areas, such as the thalamus and the ventral and dorsal cochlear nuclei (Penagarikano and Geschwind 2012; Truong et al. 2015). Neuroimaging studies consistently reported auditory discrimination skills to be localized in the same cortico-striato-thalamic circuitry (Milner et al. 2014). More specifically, the source activity of P3 in both adults and infants has been localized in primary and secondary auditory cortices and in the cingulate cortex, areas that receive input from the thalamus (Jemel et al. 2002; Hamalainen et al. 2011; Piazza et al. 2016). Based on this correspondence, it is plausible that CNTNAP2 plays a role in the neural circuitry related to auditory processing and, specifically, to auditory discrimination skills.

On the other hand, we did not find significant support for an association between rs7794745 and RAP. To our knowledge, this marker has been found to be associated mostly with electrophysiological components related to morpho-syntactic aspects of language processing in older subjects (Whalley et al. 2011; Kos et al. 2012). Thus, the absence of a significant association in our study could be due to differences in the language domain and ERP components considered (RAP and P3 vs. morpho-syntactic processing and anterior negativity/P600), and/or to the different developmental levels of the 2 populations (infants vs. children/adults). It is important to note that, because language is a heterogeneous and complex phenotype, these results need replications in different study populations.

Association Between RAP and Language

We showed a significant association between amplitude of P3 for the frequency deviant and language outcome at 20 months. These results are well replicated and are in line with previous cross-linguistic research that found early brain responses to nonspeech/speech stimuli to be predictive of linguistic skills at preschool ages (Guttorm et al. 2005; Benasich et al. 2006; Leppanen et al. 2010; Choudhury and Benasich 2011; van Zuijen et al. 2012; Cantiani et al. 2016).

Association Between CNTNAP2 and Language

Surprisingly, both CNTNAP2 variants tested were not associated with expressive language at 20 months. Although several studies found significant CNTNAP2 associations with language skills across the normal range (Rodenas-Cuadrado et al. 2014), to our knowledge only 1 previous study found this association with very early language skills (Whitehouse et al. 2011). However, the discrepancy with our study could be explained by phenotypic characterization. Whitehouse et al. (2011) found direct association between CNTNAP2 and a measure of early communicative behavior, a domain more related to social aspects of language, whereas in our study no direct effect was found on early expressive vocabulary. Although replication is crucial, it could well be the case that very early expressive vocabulary is not directly associated to CNTNAP2. In line with the hypothesis driving the present study, it is conceivable that variation at CNTNAP2 better accounts for some intermediate phenotypes, such as RAP, which appears to be closely linked to genetic susceptibility and that, in turn, is related to vocabulary acquisition.

RAP Mediates the Relationship Between CNTNAP2 and Language

Using SEM and mediation analysis, we simultaneously investigated the direct effect between CNTNAP2 and language as well as the indirect effect involving RAP as a mediator in CNTNAP2-language pathway, within a longitudinal framework. We showed that the ability to discriminate auditory frequency changes at 6 months of age fully mediates the contribution of CNTNAP2 to expressive vocabulary at 20 months of age. In particular, we obtained the first evidence that RAP mediates the association between rs2710102 and early expressive vocabulary. These results support the role of RAP as a potential intermediate phenotype for later expressive language acquisition, further elucidating the relationship between genetic polymorphisms and language development. Because CNTNAP2 is highly expressed in cortico-striato-thalamic circuitry and is involved in brain development and maintenance of connectivity in these language-relevant areas (Abrahams et al. 2007; Penagarikano and Geschwind 2012), our data support the hypothesis that CNTNAP2 variants may influence changes in brain systems underlying neurocognitive intermediate phenotype of language outcome, such as RAP. In other words, it is plausible that a CNTNAP2 common variant plays a crucial role in the neural circuitry related to establishing efficient RAP abilities in early infancy, and these deficits may ultimately predispose infants to later language disorders.

Moreover, the specificity of the results involving ERP lateralization requires further investigation before strong conclusions can be drawn. Overall, the present results that show a more substantial role of the right hemisphere in predicting early expressive language have replicated our previous results in a larger sample (Cantiani et al. 2016). Earlier studies have found similar lateralization patterns with the right hemisphere (Guttorm et al. 2005; Leppanen et al. 2010), or both hemispheres contributing to language prediction at later ages (Choudhury and Benasich 2011). The presence in the literature of these contrasting results, together with the known “inverse problem” of ERPs (preventing assumptions about precise brain localizations), requires a note of caution in interpreting our findings. The use of analytic techniques that allow age-appropriate source localization will be required in the future, in order to better explore these hemispheric effects (Hamalainen et al. 2011; Piazza et al. 2016).

Finally, we would like to call attention to several limitations of this study. First, as in the mediation model, the genotype–phenotype direct association (CNTNAP2—expressive language) is not significant, the results should be interpreted cautiously. However, it has been argued that mediation is still appropriate when the direct association does not reach significance (MacKinnon and Fairchild 2009). Second, although the sample size is relatively smaller compared with classical molecular genetic studies, using a more powerful characterized phenotype for tracing genetic effects on language and an a priori hypothesis, we considered that this might have a negligible impact on the alpha threshold to infer significance. However, a replication in independent samples or using larger datasets may further refine the observed effects.

Finally, although both language (dis)abilities (Liederman et al. 2005) and RAP (Peiffer et al. 2004; Truong et al. 2013) have shown sex differences in both animal and human studies, with a clear disadvantage in males compared with females, we deemed inappropriate to test the effect of sex upon the mediation model, due to the limited power of the sample. Future research will need to explore this effect more carefully with properly powered sample sizes.

This study paves the way for further hypotheses to be tested in the future. For example, CNTNAP2 is a gene of great interest also for ASD (e.g., Rodenas-Cuadrado et al. 2014; Truong et al. 2015), and although sensory processing abnormalities in ASD have a complex etiology, RAP seems to be impaired and predictive of language impairment in ASD (for a review see O'Connor 2012). Future directions in research might include testing RAP as a mediator between CNTNAP2 and ASD traits, comparing mediation models in infants from the general population, at high risk for ASD and/or language impairment, and affected individuals.

In summary, the current findings add a growing body of literature implicating RAP as a viable alternative marker in the genetic study of early language development. Moreover, the model used here demonstrates a potential developmental cascade of effects, whereby a common variant in CNTNAP2 drives RAP functioning that, in turn, contributes to early expressive outcome. This study may open new perspectives for intervention. As early treatments focused on RAP have been shown to improve the prognosis of language development in children (Benasich et al. 2014), they may be especially warranted in risk allele carriers of CNTNAP2, hopefully with enduring consequences on later educational and psychosocial outcomes.

Supplementary Material

Supplementary material is available at Cerebral Cortex online.

Funding

This work was supported by the Italian Ministry of Health (Grant “Ricerca Corrente” 2013; “Ricerca Corrente” 2016; “5 per 1000” 2015).

Notes

The authors wish to thank the nursing and clinical staff of the Department of Gynecology & Obstetrics of the Manzoni Hospital of Lecco and of the Hospital of Curate Brianza. We are also grateful to Roberta Bettoni, Giulia Mornati and Martina Gallo for their help in data collection and analysis. Finally, special thanks go to all infants and their parents participating in this study. Conflict of Interest: The authors declare that there is no conflict of interest regarding the publication of this paper.

References