Introduction

Developmental dyslexia (DD) is a complex heritable condition typically diagnosed in the first school years characterized by an impairment in reading abilities in spite of normal intelligence and adequate educational opportunities.

Converging evidence from high-risk and longitudinal studies indicates that early developmental problems in oral language1, 2, 3, 4, 5, 6, 7 and mathematical skills8, 9, 10 can constitute associated features of DD. Strong genetic correlations between language, mathematical and reading traits have been consistently reported,8, 10, 11, 12, 13, 14 indicating that the degree of overlap in genetic influences is comparable across different regions of the liability distribution.

Even if addressing the issue of pleiotropy is among the major aims of contemporary genetic research on cognitive sciences,15, 16, 17 the identification of elements supporting shared genetic influence is a less straightforward question. Few molecular genetic studies have been conducted for language and mathematics18, 19 compared with the extensive molecular genetic literature on DD. As substantial heritability has been indeed reported for DD and DD-related traits,15, 20 at least nine DD loci have been mapped and specific identified genes have been consistently reported to be DD candidate susceptibility genes.21

A DD susceptibility locus on the short arm of chromosome 6, that is, DYX2, has been reported by at least five independent studies.21 From these studies of DYX2, there are two genes that stand out: KIAA0319 and DCDC2. Since the first significant association findings,22 several studies tested KIAA0319 for association with DD and DD-related traits with both positive23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33 and negative results34, 35 in clinical samples, as well as in the general population.36, 37 Furthermore, a three-single-nucleotide polymorphism (SNP) risk haplotype has been shown to associate with neuropsychological skills in reading impaired families,21 likewise in general population samples.26, 36 Finally, recent linkage studies reported significant findings for specific language impairment38, 39 and speech sound disorder40 on this genomic region.

Linkage to the peri-centromeric region of chromosome 3 has been observed for DD susceptibility in three independent genome-wide screens.41 Hannula-Jouppi et al.42 provided the primary evidence supporting ROBO1 as a DD candidate gene, reporting the observation of a translocation t(3;8)(p12;q11) in an individual who also had DD, and of a 35-Mb haplotype cosegregating with DD in the large Finnish family and including ROBO1. Recently, Bates et al.43 found associations between several SNPs of ROBO1 and reading-related measures in a general population sample, indicating that ROBO1 can be involved in language acquisition. A negative association between markers spanning within the ROBO1 and the diagnosis of DD have been newly found in a case–control study in an Indian sample.44 Furthermore, significant linkage results have been reported for specific language impairment38 and speech sound disorder45 to the DD locus on chromosome 3.

Finally, and perhaps of more interest, for both the risk variants mentioned hitherto, there is initial evidence of a proper functional role. In the presence of the risk haplotype on chromosome 6p21, the expression of the KIAA0319 gene is specifically reduced.46 This haplotype harbors the rs9461045 minor allele, which confers reduced luciferase expression in both neuronal and non-neuronal cell lines, and is implicated in a nuclear protein-binding site.28 By knocking down the expression of this specific nuclear protein-binding site, the risk haplotype KIAA0319 expression can be almost entirely rescued.28 Functional studies have also revealed that KIAA0319 likely involved in neuronal migration and axon growth,47, 48, 49, 50, 51, 52, 53, 54, 55 which is in turn consistent with the original neuroanatomical finding of focal architectonic dysplasias and neuronal ectopias in the brains of people with DD.56, 57 Some further support is offered by a recent animal study showing that embryonic RNA interference of KIAA0319 expression resulted in rapid auditory processing and spatial learning deficits.51, 52 Regarding the ROBO1 gene, Hannula-Jouppi et al.42 demonstrated that its expression is significantly reduced by the 35-Mb risk haplotype, which is associated with DD in the large Finnish family.

The bulk of results reviewed so far shows that KIAA0319 and ROBO1 genes appear as more than plausible candidates for DD, DD-related neuropsychological phenotypes and DD-correlated traits. We already yielded evidence about the effects of the other two well-replicated DD candidate genes (that is, DYX1C1 and DCDC2) on DD and DD-related and DD-correlated phenotypes in an Italian clinical sample.58, 59, 60 In this study, we assessed the possible association with specific aspects of the phenotypic profile of DD and eight SNPs within KIAA0319—including those forming the three-SNP haplotype tested by Francks et al.23—and ROBO1, in order to further investigate the contribution of these genes to DD and DD-related phenotypes in an Italian sample. Moreover, we investigated the pleiotropic effects of the selected KIAA0319 and ROBO1 markers while using concurrent measurements of language and mathematics abilities.

Materials and methods

The protocol was approved by the Scientific Institute, IRCCS Eugenio Medea Bioethics and Scientific Board.

Sample

This study is part of an ongoing project on the genetics of reading disabilities at the Child Psychopathology Unit, Scientific Institute, IRCCS Eugenio Medea, Bosisio Parini, Lecco, Italy, and at the Centro Regionale di Riferimento per i Disturbi dell’Apprendimento—CRRDA (‘Regional Reference Center for the Specific Learning Disability’), ULSS 20, Verona, Italy.58, 59, 60, 61, 62, 63, 64

The ascertainment scheme has been reported in detail elsewhere.61 Briefly, nuclear families were recruited if probands met the criteria for DD according to the DSM-IV.65 In particular, the inclusion criteria for the present study were (1) either accuracy or speed <−2.00 s.d. on timed text reading, single words reading or single non-word reading tests, (2) a full-scale IQ above 84 and (3) absence of neurological or sensorial disorders.

After parental informed consent, the offspring underwent an extensive medical assessment and a battery of tests, which evaluate text, word and non-word reading,66, 67, 68 writing under dictation of word, non-word and sentences-containing homophones,68 forward/backward digit span,69 phonemic elision and blending,70 language abilities,71 mathematics abilities72, 73 and the Wechsler Intelligence Scale for Children, Revised74 or the Wechsler Intelligence Scale for Children, third edition.75 For all tests, standardized scores on the Italian population are provided (for an extensive description of tests, see Supplementary Information 1). Siblings were administered only two subtests of the intelligence scale, that is, vocabulary and block design, that show a high correlation (r) with, respectively, verbal IQ (r=0.82)74, 75 and performance IQ (r=0.73),74, 75 and were included if they were fully biological, older than 6 years and younger than 18 years, if they had no history of neurological and sensorial disorders and if the mean score of vocabulary and block design subtests was >7, regardless of their reading performance. Blood or mouthwash samples were obtained from all offspring and their biological parents.

Phenotypes

Eighteen phenotypes were used in association analyses (for a detailed description, see Supplementary Information 1): DD as a discrete trait, READING-Accuracy, READING-Speed, SPELLING, Auditory STM, ELISION, BLENDING, five language abilities (that is, semantic comprehension, Token test, syntactic comprehension, rapid automatized naming and semantic fluency) and six mathematics abilities (that is, mental calculation, both accuracy and speed; written calculation, both accuracy and speed; number dictation and numerical facts).

Genotyping

Sample collection and DNA extraction have been described in detail elsewhere.64 In particular, we explored the contribution of five among the most replicated markers within KIAA0319 and TTRAP (that is, rs2038137, rs114138463, rs4504469, rs9461045 and rs2143340) and of four markers within ROBO1 (that is, rs333491, rs6803202, rs9853895 and rs7644521), which have been found associated with reading-related measures in a previous study.43 Markers rs2038136 and rs2038135 within KIAA0319 were typed as a byproduct of rs2038137 sequencing.

Portions of the region upstream of the KIAA0319 gene were amplified from genomic DNA (primer sequences and amplification protocols are available from the authors on request) and analyzed by direct sequencing. A 0.5-μl aliquot of each amplified DNA sample was labeled with a BigDye Terminator 3.1 cycle sequencing kit (Life Technologies, Monza, Italy) and sequenced on an ABI3130xl Genetic Analyzer (Life Technologies). Sequences were aligned with Autoassembler (Applied Biosystems, Monza, Italy) and scored for polymorphisms rs2038137, rs2038136, rs2038135 and rs114138463. Polymorphisms rs333491, rs6803202, rs9853895 and rs7644521 in ROBO1; rs4504469 and rs9461045 in KIAA0319, and rs2143340 in TTRAP were analyzed with quantitative PCR and were typed using TaqMan SNP Genotyping assays (Life Technologies) on a 7900HT Sequence Detection System (Life Technologies).

Genotype error checking was completed in PEDSTATS,76 and inconsistent genotypes were zeroed out and were not considered for further analysis. Allelic frequencies and Hardy–Weinberg equilibrium for the markers under consideration are shown in Table 1. As SNPs required a minor allele frequency of at least 5% in order to be considered as a potential SNP for further analyses, rs114138463G/A was excluded (Table 1). For all the remaining genotyped SNPs, P-values for deviation from the Hardy–Weinberg equilibrium were not <0.05, and thus no SNPs were excluded. The linkage disequilibrium structure of KIAA0319 and ROBO1 was analyzed using the parental genotypes only; linkage disequilibrium was extracted and plotted in Haploview 4.0 (Figures 1a and b). Genotyped markers showing r2-values 0.80 were then dropped; on this basis, four markers for KIAA0319 gene (that is, rs4504469C/T, rs2038137G/T, rs9661045C/T and TTRAP-rs2143340A/G) and four markers for ROBO1 gene (that is, rs6803202C/T, rs9853895C/T, rs333491A/G and rs7644521C/T) were then chosen for further analyses.

Table 1 Allele frequencies and HWE’s P-values for the selected KIAA0319 end ROBO1 markers
Figure 1
figure 1

(a) Haploview plot showing pairwise linkage disequilibrium (r2-values) for seven SNPs within KIAA0319 based on parents’ genotypes. Numbers in squares are r2-values. (b) Haploview plot showing pairwise linkage disequilibrium (r2-values) for four SNPs within ROBO1 based on parents’ genotypes. Numbers in squares are r2-values.

Statistical analysis

Genetic association was investigated by the quantitative transmission disequilibrium test (version 2.5.1) as modeled by Abecasis et al.77

Discrete trait analysis

Single-marker transmission disequilibrium test analyses were carried out for DD as a discrete trait using the ‘-ad’ option, which allows to test using only affected individuals. As the association between the KIAA0319-selected markers and DD as a discrete trait was already investigated in a substantially overlapping sample (nuclear families’ overlap 61.7%)78 the present study total sample, discrete trait analysis were conducted only for the ROBO1-selected SNPs.

Quantitative trait analysis

Quantitative measures were analyzed using the ‘-wega’ option, which allows to adopt commonly used variance components, that is, the environmental variance (e), polygenic variance (g) and additive major locus (a), and the ‘-ao’ option, which allows for families of any size (with or without parental genotypes), and should be used in a variance component framework, and represents a variant of the ‘-ap’ option, which evaluates the evidence for population stratification. Performance IQ was included as covariate26, 43 with the ‘-c’ option. We calculated the genetic association for the total sample for both the KIAA0319 and ROBO1 genes. Moreover, as following up association signals over different severity groups may confirm a possible true contribution of the selected SNPs to DD and DD-related performance,79 we tested the genetic association for a subsample selected by severity, that is, by selecting only the nuclear families in which at least one offspring scored 2.50 s.d. below the general population mean on either accuracy or speed in either text-, or word-, or non-word reading tasks (n=311). Only empirical P-values are reported and were computed from 1.000 Monte-Carlo permutations.

The haplotype association analysis on the three-SNP haplotype23 was performed in PBAT (http://www.hsph.harvard.edu/clange/default.htm).80, 81 To adjust for the covariate, we used Performance IQ as interaction variable,26, 43 and nominal P-values are reported.

For both single-marker TDT analyses and haplotype association analysis, Bonferroni correction for multiple testing was not implemented, as it would be too conservative for multiple correlated traits (Supplementary Table 1) and non-independent genetic markers.23, 24, 34, 60, 82, 83, 84 Thus far, we decided to adjust the significance levels by the false discovery rate method85 applied to the tests performed for each marker and haplotype (that is, 17 phenotypes), separately for each marker and haplotype.86, 87, 88, 89

Results

The total sample consisted of 493 unrelated nuclear families of probands with DD (804 offspring); except for 63 families that had only one parent available, all parents were represented, yielding a total sample of 1727 individuals, all of Italian ancestry. Of this total sample: (i) 352 families (599 offspring) had complete measures of READING-Accuracy, READING-Speed, SPELLING and auditory STM (group 1), and (ii) 179 families (229 offspring) had complete measures of ELISION, BLENDING, language and mathematics abilities (group 2; Supplementary Tables 2 and 3).

No significant associations were found between DD as a discrete trait and any of the ROBO1 markers in both the total sample and the selected-by-severity subsample.

After implementing the false discovery rate correction, market-trait associations for DD-related neuropsychological phenotypes (that is, READING-Accuracy, READING-Speed, SPELLING and Auditory STM) were not-significant in both group 1 and in the selected-by-severity subsample (Tables 2a and b). Regarding the DD-correlated neuropsychological traits, significant association was found between the ‘A’ allele of the ROBO1-rs333491A/G marker and mental calculation-accuracy (χ2=10.24; df=102; nominal P-value=0.001; empirical P-value=0.003; q-value=0.030; 62 informative families; genetic effect=−0.746; Table 3b) in group 2. The same association was found in the selected-by-severity subsample; the ‘A’ allele of the ROBO1-rs333491A/G marker was significantly associated with mental calculation-accuracy (χ2=11.90; df=89; nominal P-value 0.001; empirical P-value 0.001; q-value=0.010; 56 informative families; genetic effect=−0.458; Table 3b).

Table 2a Market-trait association empirical P-valuesa in both the group 1 of the total sample (n=352) and the selected-by-severity subsample (n=311) for KIAA0319 markers
Table 2b Market-trait association empirical P-valuesa in both the group 1 of the total sample (n=352) and the selected-by-severity subsample (n=311) for ROBO1 markers

No significant associations were found between KIAA0319 markers and any of the neuropsychological traits in both the total sample and the selected-by-severity subsample (Tables 2a and b and Tables 3a and b).

Table 3a Market-trait association empirical P-valuesa in both the group 2 of the total sample (n=179) and the selected-by-severity subsample (n=154) for KIAA0319 markers
Table 3b Market-trait association empirical P-valuesa in both the group 2 of the total sample (n=179) and the selected-by-severity subsample (n=154) for ROBO1 markers

Finally, we investigated the KIAA0319 three SNP haplotype (rs4504469, rs2038137 and rs2143340) originally identified by Francks et al.23 The 1-1-2 haplotype—where 1 is the major allele and 2 is the minor allele of each SNP23, 47—showed a frequency of 14.3% in our sample. After implementing the false discovery rate correction on the nominal P-values, significant association between this haplotype and READING-Accuracy was found in the selected-by-severity subsample (informative families=108, nominal P-value=0.003, q-value=0.025).

Discussion

This study was primarily designed to verify the reproducibility of previous findings of an involvement of the KIAA0319 and ROBO1 genes in DD, in a sizable sample of Italian families ascertained for DD. We already yielded partial initial evidence about the effects of the KIAA0319 and ROBO1 genes on reading-related neuropsychological phenotypes in a gene-by-environment association analysis of quantitative DD-related phenotypes.64 We now integrate the study of these genes by exploring and confirming their contribution to DD, DD-related and DD-correlated phenotypes by association analyses. In particular, we provide evidence toward a role of one SNP in ROBO1 (that is, rs333491A/G) in DD-related neuropsychological phenotypes, in a large Italian clinical sample. The associations were supported by the variant of the selected polymorphism, for which a positive association was found in previous works.43

Our data show that ROBO1 gene is associated with the mathematics cognitive phenotypes (that is, mental calculation), suggesting that this gene can also account for part of the reduced mathematical skills that are observable in these children. To our knowledge, the only study that addressed a similar question19 reported several SNPs associated with mathematical skills in chromosomal regions that do not pertain to the gene examined in this hypothesis-driven study. Our result is quite surprising; a significant, even if low, correlation has been indeed observed between ‘Mental Calculation’ and ‘Reading, accuracy’, ‘Spelling’ and ‘Elision’ (r=0.21, r=0.17 and r=0.21, respectively; see Supplementary Table 1), and, more generally, ROBO1 has been reported as a plausible candidate for DD, DD-related neuropsychological phenotypes and DD-correlated traits in previous studies.41, 42 Nevertheless, negative association between markers spanning within the ROBO1 and DD has been also reported.44 Similarly, we did not find any association between ROBO1 and DD, DD-related and DD-correlated phenotypes. This lack of association may be due to the low correlation between mathematics skills and DD-related and DD-correlated neuropsychological traits (see Supplementary Table 1). These results suggest that ROBO1 gene can account for not only part of the disabilities proper of DD, but also for part of the reduced mathematical skills that are observable in these children, thus following a pleiotropic mode of action.15, 16, 17, 90

Turning to the results with the linguistic phenotypes, our negative results contrast with previously published studies38, 39, 40, 45 that reported pleiotropic effects of DD candidate genes for linguistic traits. Similar to what we found in a previous study,60 this discrepancy, however, could be connected to the age of participants in our study and the type of tests that constitute adequate measures of linguistic skills at different ages. The studies that yielded positive results38, 39, 40 were indeed assessing phenotypes in children of a much younger age than ours and were able to test articulation skills, as it is appropriate for younger children. Our measures of language abilities may have turned out to be insensitive to association with these genes because the deficits in linguistic skills measured by the previous positive studies may undergo spontaneous catch-up with time and development, so that by assessing older children we found linguistically better functioning children for crucial phenotypes, such as ‘articulation’. Moreover, this discrepancy may also be due to the tests used; our linguistic tests were simply not measuring the same types of deficits as measured by previous positive studies.45

Finally, our findings further support the effect of the main ‘risk’ rs4504469–rs2038137–rs2143340;1-1-2 haplotype23 in the etiology of reading abilities,23, 26, 36 by showing that it had a detrimental effect on Performance IQ-adjusted reading measure within the most severely affected children.

Main conclusions

This study provide evidence about the association between ROBO1 and DD-related neuropsychological phenotypes. Moreover, these findings likely indicate shared biological and cognitive processes that underlie reading and mathematics, and represent a first evidence in favor of pleiotropic effect of the ROBO1 gene on reading-correlated neuropsychological skills in a sizable sample of Italian families ascertained for DD. Therefore, these data provide further evidence about the influence of this gene on the etiology of DD and its correlated neuropsychological traits, by exploring their effects in a non-English-speaking sample.

Nevertheless, our study should be viewed with some limitations in mind. First, even if by assuming a conservative estimate of heritability for each phenotype of 0.3, for a sample of 493 nuclear families, the conditional power is >95% for an α-level of 0.05 (PBAT Power Calculator, http://www.biostat.harvard.edu/~fbat/pbat.htm), these data need replication in independent samples in order to evaluate the influence of these genes in non-English-speaking populations. Second, as with all analyses of complex traits, the significance of the current association findings with DD-correlated phenotypes should be interpreted cautiously until they can be replicated in other large, directly selected for specific language impairment, speech sound disorder and/or developmental dyscalculia samples.