The DYX2 locus and neurochemical signaling genes contribute to speech sound disorder and related neurocognitive domains.

A major milestone of child development is the acquisition and use of speech and language. Communication disorders, including speech sound disorder (SSD), can impair a child's academic, social and behavioral development. Speech sound disorder is a complex, polygenic trait with a substantial genetic component. However, specific genes that contribute to SSD remain largely unknown. To identify associated genes, we assessed the association of the DYX2 dyslexia risk locus and markers in neurochemical signaling genes (e.g., nicotinic and dopaminergic) with SSD and related endophenotypes. We first performed separate primary associations in two independent samples - Cleveland SSD (210 affected and 257 unaffected individuals in 127 families) and Denver SSD (113 affected individuals and 106 unaffected individuals in 85 families) - and then combined results by meta-analysis. DYX2 markers, specifically those in the 3' untranslated region of DCDC2 (P = 1.43 × 10(-4) ), showed the strongest associations with phonological awareness. We also observed suggestive associations of dopaminergic-related genes ANKK1 (P = 1.02 × 10(-2) ) and DRD2 (P = 9.22 × 10(-3) ) and nicotinic-related genes CHRNA3 (P = 2.51 × 10(-3) ) and BDNF (P = 8.14 × 10(-3) ) with case-control status and articulation. Our results further implicate variation in putative regulatory regions in the DYX2 locus, particularly in DCDC2, influencing language and cognitive traits. The results also support previous studies implicating variation in dopaminergic and nicotinic neural signaling influencing human communication and cognitive development. Our findings expand the literature showing genetic factors (e.g., DYX2) contributing to multiple related, yet distinct neurocognitive domains (e.g., dyslexia, language impairment, and SSD). How these factors interactively yield different neurocognitive and language-related outcomes remains to be elucidated.

Proficiency of communication and language skills is an important aspect of child development. However, children with numerous disorders, including speech sound disorder (SSD), have impairments in these skills. Specifically, SSD is characterized by the omission or substitution of sounds and other speech production errors that prevent understanding of speech (Lewis et al. 2006; Newbury & Monaco 2010; Pennington & Bishop 2009). The speech deficits in children with SSD are of unknown origin and focus on sound production and articulation miscues of verbal utterances. These difficulties in articulation and sound production are common in children, with prevalence estimates ranging from 4% to 16% (Lewis et al. 2006; Pennington & Bishop 2009). In addition to articulation difficulties, children with SSD exhibit deficits in manipulating and understanding phonemes – the speech sound units of words (ASHA Website 2015; Bernthal et al. 2013; Pennington & Bishop 2009). Thus, SSD includes both errors of articulation or phonetic structure (errors due to poor motor abilities associated with the production of speech‐sounds) and phonological errors (errors in applying linguistic rules to combine sounds to form words). Phonological awareness (PA) is the awareness of the sound structure of a language and includes the ability to manipulate and understand phonemes. Impairments in various facets of phonology have also been implicated in related neurodevelopmental disorders, including dyslexia and language impairment (LI) (Newbury & Monaco 2010, Pennington & Bishop 2009).

Family and twin studies have shown that there is a substantial genetic component to SSD and communication traits (Lewis et al. 2006; Stein et al. 2011). Linkage and association studies, as well as studies of rare, severe forms of SSD and/or related speech disorders, have identified several genes that influence these traits. Specifically, linkage analyses have implicated the DYX2 locus on chromosome 6p21.3 (Peter et al. 2012; Smith et al. 2005). The DYX2 locus was first shown through linkage analysis to contribute to dyslexia, although more recent studies have also implicated the locus in LI and general cognitive abilities (Eicher et al. 2014; Lewis et al. 2006; Mascheretti et al. 2014; Newbury et al. 2011; Powers et al. 2013). The DYX2 locus houses two risk genes, DCDC2 and KIAA0319 (Becker et al. 2014; Cope et al. 2005; Konig et al. 2011; Ludwig et al. 2008; Meng et al. 2005; Powers et al. 2013; Scerri et al. 2011; Schumacher et al. 2006; Zou et al. 2012), although there is evidence that other DYX2 genes contribute to these traits (Eicher et al. 2014; Pinel et al. 2012; Plomin et al. 2004). However, no fine‐mapping association studies have examined the contribution of the DYX2 locus to SSD. Therefore, it is unknown which genes are responsible for the linkage to DYX2, and more specifically, whether the risk genes DCDC2 and KIAA0319 contribute to SSD as well.

Recent studies have also examined the effects of prominent neurochemical signaling genes, including those in the nicotinic‐ and dopaminergic‐related pathways, on SSD and related language processes. Our groups and others have recently reported associations of DRD2, which encodes a dopamine receptor, with verbal language traits and SSD (Beaver et al. 2010; Berman & Noble 1995; Eicher et al. 2013; Stein et al. 2014; Wong et al. 2013). The association of genetic markers rs4938013, rs2734849 and rs1800497 in the DRD2/ANKK1 locus in the Cleveland SSD cohort used in this study (see below), was completed independently and before the genotyping and analyses discussed in this manuscript, and has been presented elsewhere (Stein et al. 2014). Dopaminergic, nicotinic and other neurochemical signaling pathways are vital to the development and execution of higher‐order cognitive processes. A recent magnetic resonance spectroscopy study found that choline and glutamate levels were correlated with reading performance (Pugh et al. 2014). However, more studies are needed in order to further link genetic variation in genes involved in neurochemical signaling and language phenotypes.

The goals of this study were to complete an association scan of the DYX2 locus to characterize its relationship with SSD, and to test for association between specific variants in genes involved in neurochemical signaling with SSD. To accomplish these goals, we first performed primary genetic association testing using language and genetic data collected in two family‐based SSD samples: the Cleveland SSD and Denver SSD cohorts. Then, we combined these genetic association results in a meta‐analysis. Our hypotheses were that the known DYX2 risk genes DCDC2 and KIAA0319 also contribute to SSD, and that neurochemical‐related genes are involved in SSD.

Methods

We used two family‐based studies in our genetic analyses of SSD: the Cleveland SSD and Denver SSD cohorts. The details regarding subjects and recruitment methods for both cohorts are described elsewhere but are summarized below (Lewis et al. 2000, 2011; Smith et al. 2005; Stein et al. 2004). The Cleveland SSD cohort was collected at Case Western Reserve University as part of an on‐going longitudinal study in the Greater Cleveland Metro area (Lewis et al. 2000, 2011; Stein et al. 2004). These analyses were restricted to individuals of self‐reported Caucasian ethnicity. Probands between the ages of 4 and 6 years, who were enrolled in therapy for moderate‐severe SSD, were referred by clinical speech‐language pathologists. Siblings and parents of probands were also enrolled in the study with siblings diagnosed with SSD based on performance on the Goldman‐Fristoe Test of Articulation (GFTA, Goldman & Fristoe 1986) and parents diagnosed based on history of speech therapy. Probands and siblings were required to meet the following criteria: (1) Standard American English as their first language; (2) normal hearing, as defined by passing a pure tone hearing screening; (3) a normal oral mechanism; (4) a Performance IQ score of 80 or above; (5) no history of neurological or developmental disorders according to parental report. All children and parents provided informed assent/consent with study protocols approved by the institutional review board of the Case Medical Center and University Hospitals of Cleveland, Ohio.

Similarly, the Denver SSD cohort is part of a longitudinal study examining SSD (Smith et al. 2005). Probands between the ages of 5 and 7 years were recruited through public and private schools in the Greater Denver Metro area, as well as through radio and newspaper advertisements. All probands were required to currently have or to have previously had SSD, as defined by prior testing by a speech‐language pathologist and/or significant intelligibility problems at 3–4 years of age according to parent report. Additionally, all Denver SSD probands were currently receiving or previously received speech‐language therapy and/or scored below the 30th percentile on the GFTA (Goldman & Fristoe 1986). Full siblings of the probands were recruited for inclusion if they were between 5 and 9 years old, independent of speech performance. However, if both the proband and siblings had received speech‐language therapy, the child with the more extreme articulation deficit on the GFTA was designated as the proband. Participating families and children were required to be monolingual English speakers. Subjects were also excluded if they had: (1) known genetic disorders or syndromes, (2) intellectual disability (e.g., nonverbal IQ < 70), (3) pervasive developmental disorder, (4) significant birth complications, (5) acquired brain injury, (6) peripheral hearing loss, or (7) structural or functional speech mechanism impairments (e.g., cleft palate).

Subjects in both cohorts completed a wide array of neurocognitive and language measures that extensively characterized his or her neurocognitive profile that are described in detail elsewhere (Lewis et al. 2000, 2011; Smith et al. 2005; Stein et al. 2004). As the aim of this set of analyses was to leverage both cohorts and use them in a meta‐analysis, we chose to focus only on phenotypes that were identical or measured the same neurocognitive domain. Thus, we examined three phenotypes described in further detail below: (1) case–control status, (2) performance on the GFTA, and (3) a composite quantitative trait measuring PA.

First, probands and siblings in both the Cleveland SSD and Denver SSD cohorts were clinically ascertained for SSD. There were 210 affected and 257 unaffected in 127 families in the Cleveland SSD cohort, and 113 affected and 106 unaffected individuals in 85 families in the Denver SSD cohort. Second, 187 subjects (affected and unaffected) from the Cleveland SSD cohort, and 175 subjects (affected and unaffected) from the Denver SSD cohort, were assessed by the GFTA for quantitative performance of articulation. Third, subjects completed multiple tasks that measure PA. Therefore, we constructed a composite PA measure in each cohort. In the Cleveland SSD cohort, the PA composite score was comprised of the Blending Words and Elision tasks of the CTOPP (Wagner et al. 1999); the raw scores were converted to a z‐score, and a weighted composite score was derived using principal component analysis. In the Denver SSD cohort, the PA composite score was comprised of the Bird and Bishop (1992) rhyme judgment task, and the Elision, Blending Words and Sound Matching subtests of the CTOPP (Wagner et al. 1999). The raw scores from these four measures were age‐corrected, then averaged and converted to a z‐score to yield a PA composite score. There were 96 and 128 individuals, including both cases and controls, with PA composite scores in the Cleveland SSD and Denver SSD cohorts, respectively.

We developed a single nucleotide polymorphism (SNP) marker panel to tag common variation in the DYX2 locus and to specifically interrogate candidate markers within neurochemical signaling genes. The development and composition of this marker panel is explained in full elsewhere and described briefly below (Eicher et al. 2013, 2014; Powers et al. 2013). First, we developed a SNP marker panel to capture the common variation in the DYX2 locus. TagSNPs in the DYX2 locus were selected using the association study design server of Han et al. (2008). In this study, the panel contained 204 SNPs within the DYX2 locus spanning approximately 1.4 Mb. Additionally, the SNP marker panel contained 32 SNPs within 11 neurochemical signaling genes (Table 1). The candidate markers were selected due to their past implications in neurochemical function and/or association with nicotine dependence. All markers were genotyped together on the Sequenom MassARRAY platform (San Diego, CA, USA) following manufacturer's guidelines at the Yale Center for Genome Analysis (West Haven, CT, USA). Briefly, markers were genotyped in nine multiplex reactions of 30–36 markers each, totaling 300 markers. A subset of the markers in this panel was neither in the DYX2 locus nor within neurochemical‐related genes, and therefore, not included in the subsequent analyses presented in this study. In addition to downstream quality control during the genetic analyses (see below), the histogram plot showing clustering of genotype groups generated by the assay for each marker was manually evaluated. Mendelian errors were detected and removed in PLINK (Purcell et al. 2007). Markers were also removed if not in Hardy‐Weinberg equilibrium (P < 0.0001), had call rates <0.85, or had minor allele frequency <0.01.

Table 1

Genetic markers within neurochemical signaling genes genotyped in the Cleveland SSD and Denver SSD cohorts

Variant	Gene	Location	Variant	Gene	Location	Variant	Gene	Location
rs2072660	CHRNB2 1	1q21.3	rs6278	DRD2 6, 7	11q23.2	rs1426153	PKNOX2 8	11q24.2
rs2072661	CHRNB2 1	1q21.3	rs11604671	ANKK1 6, 7	11q23.2	rs750338	PKNOX2 8	11q24.2
rs12466358	CHRND 2	2q37.1	rs1800497	ANKK1 6, 7	11q23.2	rs1051730	CHRNA3 9, 10	15q25.1
rs13277254	CHRNB3 3	8p11.21	rs2734849	ANKK1 6, 7	11q23.2	rs1317266	CHRNA3 9, 10	15q25.1
rs4950	CHRNB3 3	8p11.21	rs4938013	ANKK1 6, 7	11q23.2	rs578776	CHRNA3 9, 10	15q25.1
rs6474413	CHRNB3 3	8p11.21	rs7118900	ANKK1 6, 7	11q23.2	rs6495308	CHRNA3 9, 10	15q25.1
rs4075274	NTRK2 4	9q21.33	rs10893365	PKNOX2 8	11q24.2	rs8034191	AGPHD1 11	15q25.1
rs2030324	BDNF 5	11p14.1	rs10893366	PKNOX2 8	11q24.2	rs2229959	CHRNA4 12	20q13.33
rs4274224	DRD2 6, 7	11q23.2	rs11220015	PKNOX2 8	11q24.2	rs2236196	CHRNA4 12	20q13.33
rs4648318	DRD2 6, 7	11q23.2	rs11602925	PKNOX2 8	11q24.2	rs2273504	CHRNA4 12	20q13.33
rs7131056	DRD2 6, 7	11q23.2	rs12284594	PKNOX2 8	11q24.2

References:

Wessel et al. (2010)

Saccone et al. (2009)

Saccone et al. (2010)

Beuten et al. (2007)

Beuten et al. (2005)

Huang et al. (2009)

Gelernter et al. (2006)

Chen et al. (2011)

Saccone et al. (2007)

Liu et al. (2010)

Li et al. (2005)

Amos et al. (2008).

First, genetic associations were performed in each cohort individually using GWAF v2.0, with gender included as a covariate for the three phenotypes of interest: (1) case–control status, (2) GFTA, and (3) PA (Chen & Yang 2010). GWAF uses both linear mixed effects (LME) and generalized estimating equation (GEE) models in continuous traits, and only GEE models in binary traits to account for familial correlations in association analyses (Chen & Yang 2010). Associations were performed only under additive and dominant models, as recessive models are less reliable in samples of smaller size. Next, we completed a meta‐analysis for each phenotype using the association results for both cohorts with METAL, using sample size as the weight under default settings (Willer et al. 2010). Tests of sample heterogeneity, as evaluated by I2, were performed using METAL. Associations with P < 0.05 in our meta‐analyses are reported. To correct for multiple testing and to account for the substantial linkage disequilibrium (LD) of our markers, we used SNPSpDlite to determine the number of independent hypotheses in this study (Nyholt 2004). There were 122 DYX2 and 20 non‐DYX2 independent marker loci, totaling 142 tests, in these analyses. In addition to genetic marker correlation, we have estimated the average correlation among phenotypic traits to be 0.3, meaning that 70% of the phenotypes are independent. Therefore, taking into account phenotypic and genotypic correlations, we set a threshold of 0.05/[(0.7 × 3) × 142] = 1.68 × 10−4.

Results

The results of the primary genetic association studies in the two independent cohorts, Cleveland SSD and Denver SSD, under both additive and dominant models are shown in the Supporting Information (Tables S1–S3). One association, rs16889240 in GPLD1 with PA under a dominant model (P = 2.53 × 10−6) in the Cleveland SSD cohort, survived correction for multiple testing within DYX2 (Table S3c). There was, however, no evidence for association with this marker in the Denver SSD cohort (P > 0.05, data not shown). Additionally, there was one association that approached the corrected threshold for multiple testing among the neurochemical signaling markers, rs6474413 located upstream of CHRNB3, with PA under an additive model (P = 3.15 × 10−4) (Table S3a). Again, there was no evidence of suggestive association with the marker in the Denver SSD cohort (P > 0.05, data not shown).

In order to leverage information from both samples together as opposed to relying on associations informed by just one, we combined the above genetic associations in both cohorts in a meta‐analysis. The results of these meta‐analyses are presented phenotype by phenotype, with DYX2 and neurochemical signaling genes presented together. Regional association plots of DYX2 markers for each phenotype and model are located in Figure S1. There were suggestive associations with case–control status, though none approached the multiple testing corrected threshold. These were observed mostly with markers in neurochemical signaling genes, as opposed to those in the DYX2 locus (Tables 2 and S5). Consistent with prior associations in the Cleveland SSD cohort reported independently in Stein et al. (2014), we found suggestive evidence for association of markers in similar direction of effects in ANKK1 (e.g., rs4938013, P = 0.01; rs7118900, P = 0.0163) and DRD2 (e.g., rs6278, P = 9.22 × 10−3; rs4274224, P = 0.029). Additionally, there were nominal associations with markers in BDNF (rs2030324, P = 8.14 × 10−3), PKNOX2 (rs10893365, P = 0.021) and CHRNB2 (rs2072661, P = 0.039). Within the DYX2 locus, nominally associated markers were limited to two markers in FAM65B (rs4712862, P = 0.024; rs640380, P = 0.037) and one marker in KIAA0319 (rs9295626, P = 0.047) (Tables 2 and S5, Figure S1).

Table 2

Associations with P < 0.01 in meta‐analysis of SSD case status in the Cleveland SSD (210 Affected, 257 Unaffected) and Denver SSD (113 Affected, 106 Unaffected) under (a) additive and (b) dominant models

Marker	Chr	BP*	Gene	Allele 1†	Allele 2†	P‐value	Direction‡	MAF Cleveland	MAF Denver	I2	Het. P‐value
(a)
rs2030324	11	27726915	BDNF	T	c	8.14 × 10−3	++	0.448	0.491	59.6	0.116
(b)
rs6278	11	113280724	DRD2	t	G	9.22 × 10−3	− −	0.136	0.107	0	0.847

Chr, chromosome; BP, base pair; MAF, minor allele frequency; I2, Heterogeneity test‐statistic; Het. P‐value, P‐value for test of heterogeneity.

Base pair assigned by NCBI Reference Genome GRch37.p13.

Upper case signifies major allele and lower case minor allele.

Direction of effect is in regards to Allele 1. − −’ denotes allele 1 is associated with decreased risk of SSD in both cohorts. ‘++’ denotes allele 1 is associated with increased risk of SSD in both cohorts.

Similar to SSD case–control status, there were several nominal associations between neurochemical signaling markers and GFTA, which measures articulation skills (Tables 3 and S6). The strongest associations were seen with markers in CHRNA3, including rs1051730 (P = 2.51 × 10−3) and rs8034191 (P = 4.16 × 10−3) under a dominant model (Table 3b). Within the DYX2 locus, there were nominal associations throughout DCDC2 (e.g., rs12192947, P = 6.87 × 10−3) as well as with markers in KIAA0319, NRSN1, FAM65B and a gene‐free region located telomeric to DCDC2 (Tables 3 and S6, Figure S1). None of these DYX2 associations approached the corrected threshold for multiple testing, however.

Table 3

Associations with P < 0.01 in meta‐analysis of GFTA performance in the Cleveland SSD (n = 187) and Denver SSD (n = 175) under (a) additive and (b) dominant models

Marker	Chr	BP*	Gene	Allele 1†	Allele 2†	P‐value	Direction‡	MAF Cleveland	MAF Denver	I2	Het. P‐value
(a)
rs1051730	15	78894339	CHRNA3	t	C	2.82 × 10−3	− −	0.324	0.332	0	0.414
rs8034191	15	78806023	CHRNA3	T	c	6.14 × 10−3	++	0.327	0.325	0	0.945
rs9379651	6	24314900	DCDC2	a	G	7.18 × 10−3	++	0.140	0.128	59.1	0.118
rs12192947	6	24318570	DCDC2	A	g	7.48 × 10−3	++	0.150	0.146	0	0.637
rs3789221	6	24189538	DCDC2	a	G	8.99 × 10−3	++	0.125	0.123	75.1	0.0451
(b)
rs1051730	15	78894339	CHRNA3	t	C	2.51 × 10−3	− −	0.324	0.332	0	0.913
rs8034191	15	78806023	CHRNA3	T	c	4.16 × 10−3	++	0.327	0.325	0	0.377
rs9356927	6	24101847		t	C	4.63 × 10−3	++	0.093	0.125	67.4	0.0801
rs12192947	6	24318570	DCDC2	A	g	6.87 × 10−3	++	0.150	0.146	0	0.507
rs3789221	6	24189538	DCDC2	a	G	7.03 × 10−3	++	0.125	0.123	73	0.0544

BP, base pair; Chr, chromosome; Het. P‐value, P‐value for test of heterogeneity; I2, Heterogeneity test‐statistic; MAF, minor allele frequency.

Base pair assigned by NCBI Reference Genome GRch37.p13.

Upper case signifies major allele and lower case minor allele.

Direction of effect is in regards to Allele 1. ‘− −’ denotes allele 1 is associated with decreased performance in both cohorts. ‘++’ denotes allele 1 is associated with increased performance in both cohorts.

Overall, the strongest associations observed in the meta‐analyses were between a cluster of markers in the 3′ region of DCDC2 and performance on the PA composite measure, a distinct deficit in children with SSD (e.g., rs7764902, P = 1.43 × 10−4; rs9460973, P = 1.52 × 10−4; rs6456596, P = 1.56 × 10−4; rs9467062, P = 1.66 × 10−4) (Tables 4 and S7, Figure S1). These associations met experiment‐wide correction for multiple testing of 1.68 × 10−4. There were other associations in the DYX2 locus, including markers in the genes GMNN, MRS2, FAM65B and a gene‐free region telomeric to DCDC2; however, none of these were as strong or consistent as those seen with the markers in the 3′ region of DCDC2 (Table 4). There were few neurochemical signaling markers showing association with PA, with only one marker in CHRNB3 (rs6474413, P = 0.02) and one marker in ANKK1 (rs11604671, P = 0.036) showing nominal associations (Tables 4 and S7).

Table 4

Associations with P < 0.01 in meta‐analysis of composite PA score in the Cleveland SSD (n = 96) and Denver SSD (n = 128) under (a) additive and (b) dominant models

Marker	Chr	BP†	Gene	Allele 1‡	Allele 2‡	P‐value	Direction§	MAF Cleveland	MAF Denver	I2	Het. P‐value
(a)
rs7764902	6	24174834	DCDC2	a	T	3.00 × 10−4	− −	0.022	0.049	0	0.863
rs9460973	6	24175021	DCDC2	a	T	3.05 × 10−4	− −	0.025	0.047	0	0.912
rs6456596	6	24174390	DCDC2	a	C	3.26 × 10−4	− −	0.023	0.047	0	0.897
rs9467062	6	24173382	DCDC2	T	c	3.32 × 10−4	++	0.025	0.048	0	0.922
rs554400	6	23998157		T	c	8.03 × 10−3	++	0.105	0.063	18	0.269
rs9467046	6	24120269		A	g	9.06 × 10−3	− −	0.464	0.446	0	0.387
(b)
rs7764902*	6	24174834	DCDC2	a	T	1.43 × 10−4	− −	0.022	0.049	0	0.965
rs9460973*	6	24175021	DCDC2	a	T	1.52 × 10−4	− −	0.025	0.047	0	0.927
rs6456596*	6	24174390	DCDC2	a	C	1.56 × 10−4	− −	0.023	0.047	0	0.932
rs9467062*	6	24173382	DCDC2	T	c	1.66 × 10−4	++	0.025	0.048	0	0.917
rs9467046	6	24120269		A	g	1.29 × 10−3	− −	0.464	0.446	0	0.350
rs554400	6	23998157		T	c	5.57 × 10−3	++	0.105	0.063	35	0.215

BP, base pair; Chr, chromosome; Het. P‐value, P‐value for test of heterogeneity; I2, Heterogeneity test‐statistic; MAF, minor allele frequency.

Survives correction for multiple testing (P < 1.68 0 × 10−4).

Base pair assigned by NCBI Reference Genome GRch37.p13.

Upper case signifies major allele and lower case minor allele.

Direction of effect is in regards to Allele 1. ‘− −’ denotes allele 1 is associated with decreased performance in both cohorts. ‘++’ denotes allele 1 is associated with increased performance in both cohorts.

Discussion

In this study of the genetics of SSD, we assessed the association of markers within the DYX2 locus and in genes important in neurochemical signaling with SSD and related endophenotypes. In order to increase sample size and to gain insight from more than one study population, we used data from two family‐based SSD cohorts to first perform genetic associations in each individual cohort, and then combined these results in a meta‐analysis. The strongest, most consistent associations were with markers spanning the 3′ region of DCDC2 in the DYX2 locus. These associations were consistent across both studies, as indicated by similar direction of effect and non‐significant heterogeneity. There were various other associations of markers in the DYX2 locus, notably GPLD1, and within multiple previously characterized neurochemical signaling genes, including ANKK1, DRD2, CHRNA3 and BDNF. In our associations with neurochemical signaling and DYX2 markers, there was a consistent difference in the nature of these associations, as neurochemical signaling markers tended to associate with articulation traits (i.e., case–control status and GFTA), while DYX2 markers associated with phonology (i.e., the PA composite measure). The results of these analyses suggest that the DYX2 gene DCDC2 contributes to SSD and related neurocognitive domains, as well as provides limited evidence of dopaminergic and nicotinic signaling pathways influencing SSD.

Our DYX2 results centered on the associations between markers in the 3′ region of DCDC2 and scores on a PA composite measure in children with SSD. DCDC2 has been associated with both dyslexia and LI, disorders where deficits in phonology play a major role (Pennington & Bishop 2009). Children with SSD often exhibit impairments in phonology in addition to the primary deficits seen in articulation (ASHA Website 2015; Pennington & Bishop 2009). DCDC2 appears to influence phonological deficits in children with SSD as well as phonological impairments in other neurodevelopmental disorders. Many studies of DCDC2, including those from our group, have focused on the complex tandem repeat READ1 (GenBank Accession Number BV677278) and a naturally occurring microdeletion encompassing it, located in intron 2 of DCDC2 (Ludwig et al. 2008, Meng et al. 2005; Powers et al. 2013). There is evidence suggesting that READ1 may influence the transcriptional activity of DCDC2 (Meng et al. 2011; Powers et al. 2013). However, our associations did not center near intron 2 of DCDC2. In fact, association results of READ1 in these cohorts did not show evidence of association (P > 0.05, data not shown). Instead, we found consistent associations with the 3′ untranslated region (UTR) of DCDC2. The 3′ UTR of genes typically contains regulatory sequence with microRNA (miRNA) binding sites that influence gene and protein expression. Here, the associations we observe of the 3′ UTR in DCDC2, in addition to the literature implicating READ1 in DCDC2, suggests that pre‐ and post‐transcriptional regulation of DCDC2 expression levels may contribute to phonology and language traits. Biologically, the implications of changes in DCDC2 expression still remain to be elucidated, although knockout and knockdown experiments suggest that neuronal migration, ciliary function and N‐methyl‐d‐aspartate receptor (NMDAR) activity are strong candidates (Che et al. 2013; Ivliev et al. 2012; Massinen et al. 2011; Meng et al. 2005).

In our analyses, we found evidence suggesting that markers in dopaminergic‐related genes, ANKK1 and DRD2, as well as nicotinic‐related genes, CHRNA3 and BDNF, influence SSD and articulation skills. Our groups and others have recently shown the association of markers within ANKK1 and DRD2 with endophenotypes of communication and language skills (Beaver et al. 2010; Berman & Noble 1995; Eicher et al. 2013; Stein et al. 2014; Wong et al. 2013). ANKK1 and DRD2 have previously been implicated in many neurocognitive processes, including working memory, reinforcement learning and executive function, as well as neuropsychiatric disorders, including schizophrenia and bipolar disorder (Bertolino et al. 2010; Bolton et al. 2010; London et al. 2009; McAllister et al. 2008; Ripke et al. 2014; Yang et al. 2007). Variation in DRD2 and other dopaminergic signaling genes may alter these processes, contributing to the risk of impairment in articulation and speech skills. In addition to dopaminergic factors, the results also showed suggestive evidence for CHRNA3 and BDNF, which encode components of nicotinic pathways. CHRNA3 is a neuronal nicotinic acetylcholine receptor, most widely known for its relationship with nicotine dependence (Berrettini & Doyle 2012). Its relationship with language and communication traits has been far less studied, although markers implicated in nicotine dependence, including ones examined in this study, have been previously associated with neurocognitive performance (Winterer et al. 2010, Zhang et al. 2010). BDNF is one of the most widely studied genes in neuropsychiatric and neurobehavioral traits, including Alzheimer's disease, depression and bipolar disorder. From these studies, evidence has emerged that markers in BDNF are also associated with neurocognitive traits (Bath & Lee 2006; Goldberg & Weinberger 2004). Functionally, studies have demonstrated that BDNF can modulate nicotinic receptor activity (Fernandes et al. 2008; Massey et al. 2006; Zhou et al 2004). Our results, in addition to these studies, suggest that changes in nicotinic signaling may modulate articulation and other communicative processes, and thus contribute to SSD.

Here, we implicate genetic factors in SSD that had previously been associated with related neurodevelopmental disorders, dyslexia and LI. The associations we observe between ANKK1, DRD2, and DCDC2 in SSD and related endophenotypes add to the growing body of evidence implicating genes in multiple related traits. As opposed to specific associations between genes and individual disorders, it appears that genes are more generally associated with language and phonological processes. The complex, polygenic nature of these neurodevelopmental disorders suggests that many interacting genetic and environmental factors account for risk of impairment. Therefore, in order to discern how genetic predisposition leads to clinical presentation, future studies need to thoroughly examine the combinatorial effects of genes and environment on the development of communication, language and other neurobehavioral traits.

This study is subject to several limitations. First, the size of the individual Cleveland SSD and Denver SSD cohorts is relatively small. Smaller sample size can reduce statistical power and hamper our ability to detect genetic associations. However, the use of a meta‐analysis strategy allowed us to increase our statistical power and gain confidence in the fidelity of our associations. Second, subjects in both the Cleveland SSD and Denver SSD cohorts completed a wide variety of neurocognitive assessments. The number of measures shared between the two cohorts was limited; therefore, some key neurobehavioral domains such as reading and working memory could not be specifically examined in our meta‐analyses. Third, we selected specific markers in neurochemical signaling genes of interest as well as a candidate locus of interest, DYX2. However, our DYX2 marker panel was constructed to tag as much common variation in the locus as possible and to allow for an unbiased association scan of the locus. In the future, it would be of interest to similarly saturate these neurochemical signaling genes to fine map putative associations. Fourth, the minor allele frequencies of associated markers in DCDC2, ranging from 2.2% to 2.5% in Cleveland SSD and 4.7% to 4.9% in Denver SSD, were relatively low (Table 4). However, the consistency of associations with multiple markers in this region, suggestive associations of other markers in the locus (e.g., rs3789221and rs9460974 with GFTA in Table 3), and the same direction of effect in both cohorts, increases the confidence we have in these associations. In addition, some of the individual SNP‐phenotype associations in the meta‐analysis showed significant heterogeneity based on the I2 parameter. However, I2 should be interpreted with caution when using only two small studies as we have here (Thorlund et al. 2012). Furthermore, our most significant findings did not show heterogeneity, showing these results are indeed robust.

In conclusion, we associated markers in the 3′ UTR of DCDC2 within the DYX2 locus, as well as provide suggestive evidence in dopaminergic‐ (ANKK1 and DRD2) and nicotinic‐related genes (CHRNA3 and BDNF), with SSD, and related neurocognitive domains. We expand the role of the DYX2 locus, especially DCDC2, to also influence phonological skills in children with SSD. These results also further implicate genetic variation in important neurochemical signaling pathways in higher‐order cognitive and communication skills, and their respective disorders. Future studies should further examine shared genetic risk factors among communication and language disorders, as well as elucidate the common mechanisms among these disorders. Discerning the underlying genetic and neurological mechanisms, whether they reside in neurochemical signaling, neuronal migration, and/or ciliary function or a combination of these pathways, is vital for improving treatment of affected individuals.

Table S1: Associations of DYX2 and neurochemical signaling genetic markers with SSD case status in the individual Cleveland SSD (Table S1a and Table S1c) and Denver SSD (Table S1b and Table S1d) cohorts.

Click here for additional data file.

Table S2: Associations of DYX2 and neurochemical signaling genetic markers with GFTA in the individual Cleveland SSD (Table S2a and Table S2c) and Denver SSD (Table S2b and Table S2d) cohorts.

Click here for additional data file.

Table S3: Associations of DYX2 and neurochemical signaling genetic markers with PA in the individual Cleveland SSD (Table S3a and Table S3c) and Denver SSD (Table S3b and Table S3d) cohorts.

Click here for additional data file.

Table S4: (a) Genotype breakdown of markers with associations P < 0.05 in meta‐analysis of SSD case status under an additive model. (b) Genotype breakdown of markers with associations P < 0.05 in meta‐analysis of SSD case status under a dominant model.

Click here for additional data file.

Table S5: (a) Associations with P < 0.05 in meta‐analysis of SSD case status in the Cleveland SSD (210 Affected, 257 Unaffected) and Denver SSD (113 Affected, 106 Unaffected) under an additive model. (b) Associations with P < 0.05 in meta‐analysis of SSD case status in the Cleveland SSD (210 Affected, 257 Unaffected) and Denver SSD (113 Affected, 106 Unaffected) under a dominant model.

Click here for additional data file.

Table S6: (a) Associations with P < 0.05 in meta‐analysis of GFTA performance in the Cleveland SSD (n = 187) and Denver SSD (n = 175) under an additive model. (b) Associations with P < 0.05 in meta‐analysis of GFTA performance in the Cleveland SSD (n = 187) and Denver SSD (n = 175) under a dominant model.

Click here for additional data file.

Table S7: (a) Associations with P < 0.05 in meta‐analysis of composite PA score in the Cleveland SSD (n = 96) and Denver SSD (n = 128) under an additive model. (b) Associations with P < 0.05 in meta‐analysis of composite PA score in the Cleveland SSD (n = 96) and Denver SSD (n = 128) under a dominant model.

Click here for additional data file.

Figure S1: DYX2 association results on case‐control status. DYX2 association results on GFTA performance. DYX2 association results on PA performance.

Click here for additional data file.