Pleiotropic Effects of a Chromosome 3 Locus on Speech-Sound Disorder and Reading

Introduction

Speech-sound disorder (SSD) is a complex behavioral disorder characterized by deficits in motor control of the articulatory mechanism and/or deficits in the general processing, organization, and cognitive representation of linguistic information. Thus, children’s speech-sound productions may reflect deficits in underlying phonological processes, such as phonological memory and speech-sound coding. SSD has a high prevalence in preschool children, estimated at ∼16% at age 3 years (Shriberg 2002), with 3.8% of children continuing to present with speech delay at age 6 years (Shriberg et al. 1999). More than half of these children encounter later academic difficulties in language, reading, and spelling (Shriberg and Kwiatkowski 1988; Aram and Hall 1989; Bishop and Adams 1990; Menyuk et al. 1991; Felsenfeld et al. 1995; Shriberg and Austin 1998; Lewis et al. 2000). The residual effects of preschool SSD may be lifelong, and yet, for the majority of individuals, the etiological basis of this disorder is unknown.

There is a growing body of literature suggesting that susceptibility to SSD is genetic, including twin studies (Lewis and Thompson 1992; Bishop et al. 1996; Tomblin and Buckwalter 1998), familial aggregation studies (Lewis et al. 1989; Lewis 1992; Felsenfeld et al. 1995) and segregation and linkage analyses (Lewis et al. 1993; Schick et al. 2002). Recently, studies of specific language impairment (SLI), defined as a failure to acquire language within normal limits in the absence of a generalized developmental delay, have reported linkage to chromosome 16q for nonword repetition (SLI Consortium 2002) and to chromosome 19q for expressive language (SLI Consortium 2002). There has also been a report of linkage to chromosome 7q31 for verbal dyspraxia in a large family (Fisher et al. 1998; Vargha-Khadem et al. 1998). Multiple genetic influences could also be at work, with different genetic factors contributing to different types of domains in language and speech-sound acquisition.

Pennington and colleagues (Pennington and Lefly 2001; Tunick and Pennington 2002; Raitano et al., in press) proposed that early developmental problems in spoken language predict the later emergence of dyslexia in children from high-risk families. The cognitive linguistic deficits associated with dyslexia include impairments in phonological awareness (explicit knowledge of the sound system of a language), phonological memory (as measured by the repetition of nonsense words), and the ability to decode unfamiliar words (reading decoding), each of which underlie spoken and written language. Although SSD and dyslexia are assessed through different test batteries, possibly at different ages, the similar cognitive linguistic bases of these disorders suggest that they share some common genetic etiology (see fig. 1).

Whereas there have been few molecular genetic data reported for SSD, there is an extensive genetic literature on dyslexia. The genetic mechanisms proposed to underlie the behavioral manifestation of dyslexia are complex (see Fisher and DeFries 2002). Various components of dyslexia have been analyzed, and linkages have been found to chromosome 1 (Rabin et al. 1993; Grigorenko et al. 2001); word recognition and reading decoding linkage has been found on chromosome 2 (Fagerheim et al. 1999; Francks et al. 2002; Petryshen et al. 2002; Kaminen et al. 2003); phonological awareness, rapid naming, and verbal short term memory linkage has been found on chromosome 3 (Nopola-Hemmi et al. 2001, 2002); phonemic awareness, reading decoding, single-word reading, orthographic coding, vocabulary, rapid naming, and spelling linkage has been found on chromosome 6 (Cardon et al. 1994, 1995; Grigorenko et al. 1997; Gayan et al. 1999; Fisher et al. 1999, 2002a; Nothen et al.1999; Petryshen et al. 2001; Kaplan et al. 2002); word recognition and spelling linkage has been found on chromosome 15 (Grigorenko et al. 1997; Schulte-Körne et al. 1998; Nothen et al. 1999); and single-word reading and phonological and orthographic processing linkage has been found on chromosome 18 (Fisher et al. 2002a). However, recent evidence suggests that some of these chromosomal regions harbor genes with pleiotropic effects (Marlow et al. 2003).

A recent study by Nopola-Hemmi et al. (2001) analyzed a large Finnish pedigree segregating for developmental dyslexia in an autosomal dominant fashion and found linkage to the pericentromeric region of chromosome 3. On the basis of the hypothesis that SSD and dyslexia share genetic determinants, we genotyped markers in the dyslexia candidate region on chromosome 3 to determine whether linkage would be observed in families ascertained through a proband with SSD. This is the first study, to our knowledge, to examine the common genetic basis of SSD and dyslexia on chromosome 3.

Subjects and Methods

Statistical Analysis

Measures

For most probands and siblings, test scores on several variables were available at multiple time points during childhood. For all such data, we analyzed the first available observation, hypothesizing that information at the earliest time point would be most sensitive to SSD and associated verbal traits and least influenced by uncontrolled developmental processes and experiences. We examined the effects of age, age squared, sex, socioeconomic status (SES) based on the Hollingshead Four Factor Index of Social Class (Hollingshead 1975), and first-order interaction terms on these quantitative scores. Covariates and interaction terms significant at the α=0.05 level were adjusted for in a linear regression model, and standardized residuals were obtained for further statistical analyses. To adjust the distributions of scores for severity differences, preschool test scores were age-adjusted to 4 years (i.e., the mean value at age 4 years was added to each residual), whereas school-age test scores for single-word decoding and for reading and listening comprehension were adjusted to age 18 years (i.e., the mean value at age 18 years was added to each residual). We compared the mean values of affected sibs (not including the proband) and unaffected sibs through use of a t test. We also estimated correlations for the quantitative traits through use of FCOR (S.A.G.E. [2003], version 4.4).

Factor analysis

To reduce the battery of speech/language measures to a smaller set of test constructs, principal axis factor analysis with varimax rotation was conducted on the adjusted values for tests that were administered to both younger and older participants (i.e., age ⩾4 years). To identify distinct factors, scores with the high cross loadings were excluded after initial analyses, and factor analysis was repeated. The final analysis yielded three factors, accounting for 64% of the variance in scores: an articulation factor (ARTF) with loadings on GFTA and PCC, a phonological short-term memory factor (PHONF) with loadings on MSW and NSW, and a vocabulary factor (VOCF) with loadings on the PPVT-III and EOWPVT-R. The primary loadings of the tests on their respective factors were uniformly high (all >.65), with low cross loadings (all <.26). Separate analyses of data from the younger and older participants yielded similar results. For purposes of analysis, the two z scores that loaded on each factor were averaged to create three factor score composites.

Linkage analysis

We conducted model-free multipoint linkage analyses of six reading and language scores (WRMT-ID, WRMT-AT, WIAT-RC, WIAT-LC, SI, and RAN-C) and three factor scores (ARTF, PHONF, and VOCF) to determine whether component traits of SSD were influenced by QTLs in the chromosome 3 region identified by Nopola-Hemmi et al. (2001). The quantitative traits comprising the three factors were examined individually when the factors demonstrated evidence for linkage in the significant range (Lander and Kruglyak 1995).

Genotypes from parents and offspring were used to calculate identity by descent (IBD) allele sharing distributions through use of the GENIBD program of S.A.G.E. version 4.4 (S.A.G.E. 2003). Evidence for linkage was evaluated by a Haseman-Elston regression (Haseman and Elston 1972; Elston et al. 2000), as implemented in SIBPAL (S.A.G.E., version 4.4), using either the original Haseman-Elston regression (Haseman and Elston 1972) or the newest adaptation of the method, which parameterizes the sib pair’s trait value as a weighted combination of the squared trait difference and squared mean corrected trait sum adjusted for the nonindependence of sib pairs (W4 option [Shete et al. 2003]). Although the latter is more powerful asymptotically, in a finite sample the weights may not be optimally estimated, with the result that the former method outperforms it. Therefore, the smaller P values are reported. In all cases, the alternate method also gave evidence for linkage, albeit with a slightly larger P value. Asymptotic P values achieving nominal statistical significance were confirmed by empirically deriving a null permutation distribution through use of a sample of 50,000 replicate permutations of the data, as implemented in SIBPAL. These empirical P values were uncorrected for multiple testing.

Identification of sib pairs with linkage

We sought to identify the sib pairs that were contributing to the linkage results for both MSW and NSW, since separate linkage analyses of these traits suggested they were influenced independently by the putative QTL. For each sibling pair, we computed a score based on the squared sib-pair difference and the estimated sib-pair marker allele sharing, on which to rank the sib pairs for linkage informativity:

where y₁ and y₂ are the trait values for sibs 1 and 2, is the average of (y₁-y₂)² over the whole sample, and is the estimated mean allele sharing for the two sibs. This score is large (positive) either when the squared sib pair difference is small and is large or when the squared sib-pair difference is large and is small—that is, when the sibs are similarly alike, in terms of deviation from the mean, for both their traits and allele-sharing; otherwise the score will tend to be small (negative). Sib pairs with values in the upper 30th percentile of the distribution for both NSW and MSW were considered to be linked to the locus. To verify that these sib pairs indeed contributed most to the linkage effect, the pedigrees containing these sib pairs were removed from the data, and linkage analysis of both NSW and MSW was conducted again. The resulting linkage results were much less significant (fig. 3B).

Figure 3.

Open in a new tab

Construction of core haplotypes in affected sib pairs

To characterize the smallest region of allele sharing, we constructed haplotypes in the concordantly affected sib pairs with linkage, defined as being in the upper 30th percentile of the score distribution and both sibs having low scores for NSW and MSW. The most likely haplotypes for each individual were constructed using Merlin (Abecasis et al. 2002). The procedure used the genotype data from the 15 markers in the parents and sibs of the linked sib pairs (which were identified as described above). Merlin uses sparse gene flow trees to estimate the most likely haplotypes according to a Markov process, under the assumption that no recombination occurs between consecutive markers.

Results

We examined speech-sound phenotypes in 77 pedigrees ascertained through a proband with SSD (table 2). Previous research has shown that SSD is more prevalent in males than in females (Shriberg et al. 1999), and this was indeed the case in our sample, since there were more than twice the number of affected males than females (P=.019). Language disorder was present in 23.4% of the children, and reading disorder was present in 21.6% (table 2). Other comorbidities were reported by the parents, but each represented <15% of the sample (table 2). The majority of the sample was white, and most of the sample was from the middle- to upper-class SES strata.

Discussion

SSD is a common multifactorial disorder with characteristics that overlap with those of developmental dyslexia (see fig. 1). Consistent with the possibility that these disorders share susceptibility loci, we report, in an independent sample, linkage of intermediate phenotypes for SSD to a locus on chromosome 3 that was originally identified as predisposing to developmental dyslexia in a single Finnish family (Nopola-Hemmi et al. 2001). Scores on factors and tasks that quantify a speaker’s skills in phonological short-term memory (PHONF, MSW, and NSW) showed strong evidence of linkage. Other verbal trait variables, including those indexed by the articulation factor (ARTF), verbal short term memory (SI), and single-word decoding (WRMT-AT and WRMT-ID) also showed some degree of evidence of linkage, suggesting that this putative QTL may act in a pleiotropic fashion, influencing both SSD and developmental dyslexia. The present findings expand the scope of developmental influence of the 3p12-q13 dyslexia locus from the Nopola-Hemmi et al. (2001) study indicating monogenic linkage in an extended family to consequences including SSD. More generally, our linkage finding for speech-sound production (ARTF), typically not considered a correlate of dyslexia or SLI, adds to the genetics of cognitive linguistic processes.

These results provide the first molecular genetic data supporting Pennington and Lefly’s (2001) hypothesis that early deficits in phonological processing skills and ability to repeat unfamiliar words, as in the NSW task, are predictors of subsequent dyslexia. High correlation among numerous dyslexia spectrum phenotypes has previously been reported (Fisher et al. 2002a; Grigorenko et al. 2003). In the present analysis, we found that similarly high correlations exist for traits measured in children ascertained for SSD (table 6), and the measures that demonstrated the greatest evidence for linkage were presented auditorily (MSW and NSW) rather than visually, with many of the participant children not yet readers. These measures identify early deficits in phonological memory and speech-sound coding of unfamiliar words. Nopola-Hemmi et al. (2002) observed that the neurocognitive type of dyslexia segregating in the Finnish family with linkage to 3p12-q13 consisted of deficits in phonological awareness, verbal short-term memory, and rapid naming. Findings for our phonology variables (PHONF and SI) are consistent with their findings, but, unlike in their findings, analyses of rapid naming (RAN-C) did not provide evidence for linkage. The rapid naming task in the Nopola-Hemmi et al. (2002) study required reading, which may have invoked visual abilities associated with other phonological representational skills, such as phoneme-grapheme correspondence. The linkage findings for the two measures that require reading skills (WRMT-AT and WRMT-ID) but assess processing constructs similar to that invoked in the auditorily presented measures further strengthen the support for this hypothesis. Results from both studies together suggest that the putative locus on chromosome 3 influences phonological processing.

Another genome scan in samples ascertained for dyslexia found that phenotypes for single-word reading, phonemic awareness, reading decoding, and orthographic coding were linked to markers located between 90 cM and 136.6 cM on chromosome 3 (Fisher et al. 2002a). The location of a second dyslexia linkage signal within 10–25 cM of our maximum lends further support for the possibility that at least one gene for phonological processing is located in this region of chromosome 3. However, the investigators did not provide sufficient detail for us to be able to fully compare the results. A genome scan for SLI found potential evidence of linkage to 3p24 but not to the 3p12-q13 region described in the present article (SLI Consortium 2002). This genome scan was conducted using the receptive and expressive scales of the CELF-R and a test for nonword repetition, a measure of phonological short-term memory. Extrapolating from our current study that cognitive domains common to both SSD and dyslexia are both influenced by a locus on chromosome 3, we anticipate that other loci linked to dyslexia and or SLI may also contribute to SSD susceptibility. We have not addressed these hypotheses but envisage that future studies will elucidate these relationships.

In analyzing NSW and MSW, we found that the putative chromosome 3 locus contributes not only to their shared variance but also uniquely to NSW (fig. 3A). Although both purportedly test phonological memory, they likely assess cognitively different processes. MSW uses words that may be in the subject’s lexicon, possibly relying on long-term memory. In contrast, NSW uses novel stimuli, which makes it a less confounded and possibly more sensitive test of phonological working memory. ARTF also shows suggestive linkage to this region (fig. 2), and its component traits, GFTA and PCC, are highly correlated with NSW and MSW (table 5). The GFTA measures the articulation of consonant sounds in familiar single words, and the PCC measures the articulation of consonant sounds in spontaneous conversational speech. Although both the GFTA and PCC assess articulation, they also invoke phonological memory, because the child is required to code information phonologically for temporary storage in short-term memory while performing each task. Finally, vocabulary measures (PPVT-III and EOWPVT-R) were included in the study to detect comorbid language impairment (LI). Vocabulary scores emerged as a significant factor in our analyses, possibly because of the high rate of comorbid LI in our sample (∼50%). Another explanation may be found in the emerging literature on the relationship between speech-sound development and vocabulary (Walley and Metsala 1990; Metsala 1997). Phonological information in long-term memory may play an active role in recall of short-term memory (Roodenrys et al. 2002). Thus, lexical familiarity may influence articulation accuracy. For example, children may repeat real multisyllabic words more accurately than nonsense words, because real words may be associated with phonological information stored in long-term memory. Thus, as with many cognitive test batteries, a single task relies on an array of neural processes.

The original article by Nopola-Hemmi et al. (2001) identified a core haplotype extending from D3S3049 to D3S1291 (spanning 16 markers over 41 cM), which was shared by 19 of 21 affected family members. Our evidence for linkage is strongest in this same region and is consistent for several of the traits analyzed. Using a simple method, we have identified sibs that are the source of our linkage signal and have examined the extent of haplotype sharing in these sibs. Our region of sharing overlaps with the conserved haplotype identified in the Nopola-Hemmi et al. (2001) article. In contrast to their results, however, both our haplotype analyses and our variance components analysis suggest a recessive or additive mode of inheritance for SSD.

To fully comprehend the biological basis of a multifactorial phenotype such as SSD, any binary classification, which is likely a function of multiple quantitative traits (Risch 2000), is probably too broad to adequately dissect the phenotypic overlap. Further, the lack of consensus on the etiological basis for SSD and its comorbidity with language and reading disorders makes the use of a standard qualitative affection phenotype less reliable. In contrast, well-chosen quantitative traits may be more closely tied to gene expression and therefore biologically more informative compared with the sensitivity of a clinical diagnosis or a threshold scale (Rice et al. 2001; Shriberg 1993). This approach also reduces misclassification of individuals into incorrect categories (Elston 1979). Moreover, as was done in the present analyses, utilizing appropriate quantitative traits for complex disorders may increase power to detect linkage (Duggirala et al. 1997) and may illustrate common quantitative QTLs that underlie these related disorders. Use of severity indices for gene mapping has been successful for dyslexia (Cardon et al. 1994, 1995; Grigorenko et al. 1997, 2003; Fisher and DeFries 2002; Fisher et al. 2002a; Kaminen et al. 2003; Marlow et al. 2003), as well as for other related traits such as specific language impairment (SLI Consortium 2002) and attention-deficit/hyperactivity disorder (Fisher et al. 2002b).

As pointed out by Fisher and DeFries (2002), interpretation of univariate QTL mapping data like that reported for the present study is challenging. In this study, we found at least nominal evidence of linkage to a region on chromosome 3 for about half of the SSD traits that we examined. Although we cannot say with certainty that this chromosomal region specifically affects deficits common to both SSD and developmental dyslexia, the reports reviewed and findings from the present results support that conclusion. Further, our linkage analysis suggests that this locus on chromosome 3 affects a core domain of SSD, articulation, which, to our knowledge, we are the first to examine. Since our variables for phonological memory (NSW/MSW) are correlated with the articulation measures (see table 5), we are unable to specify whether the linkage with the ARTF factor is due to articulation alone. We are in the process of testing a larger sample and plan to use multivariate linkage analysis (cf. Marlow et al. 2003; Stein et al. 2003) to address this issue. Our data also cannot be used to distinguish between the presence of one or two genes in this region. A similar pattern was seen in another quantitative trait analysis of dyslexia (Grigorenko et al. 2003). Because we have a moderately sized sample, we may not have the statistical power to resolve accurately a single versus multiple QTL model (Cordell 2001). Such information awaits a multivariate analysis of an independent sample.

Lastly, we acknowledge that additional loci with larger effect sizes associated with the domains depicted in figure 1 can be identified through genomewide scanning of this cohort. We are in the process of assembling a larger cohort, so that we can undertake this task. Because our focus is on speech-sound production, we are especially interested in identifying loci that would not ordinarily be identified through scans for dyslexia and SLI.