METHODS

As described in detail elsewhere,^3,4 our sample consists of 96 Canadian families (877 subjects), each containing two or more sibs diagnosed with phonological coding dyslexia (PCD). This diagnosis was used since the key problem in most reading disabled subjects is a specific difficulty in the phonological coding component of reading, where written words are sounded out using grapheme-phoneme (letter-sound) rules. The PCD diagnosis (affected, unaffected, or uncertain) was determined for all subjects primarily based on psychometric test results for phonological coding. Test results for phonological awareness, which is the ability to recognise and manipulate phonemes, and for spelling, which requires phonological and orthographic (recognition of letter patterns) coding, were used to assist in diagnosis, as was reading history for adults. The PCD phenotype was used for parametric and non-parametric linkage analyses. Scores from the phonological awareness, phonological coding, and spelling tests were used in quantitative trait variance component linkage analyses, after conversion to standard scores (for phonological coding and spelling) or age adjustment (for phonological awareness). Of the 96 families, 46 were nuclear families consisting of both parents, two or more affected children, and unaffected children if available, with an average family size of five members. The remaining 50 families were extended kindreds consisting of a nuclear family and other branches with affected and unaffected relatives, ranging in size from six to 107 members (average 18 members). Seven microsatellite markers spanning the DYX3 region were selected from the report of Fagerheim et al¹ and automated genotyping was performed using a LI-COR 4200S-2 Gene ReadIR DNA Analyzer. Marker allele frequencies were calculated from the parents of one nuclear family per pedigree. The Genethon genetic map⁵ was used for intermarker order and distances: (pter) D2S1352 - 4 cM - D2S2352 - 1 cM - D2S378 - 0 cM - D2S2279 - 0 cM - D2S2183 - 1 cM - D2S1337 - 2 cM - D2S393 (cen). This marker order corresponded to the order from the human genome sequence.⁶

Two point parametric linkage analysis was performed using FASTLINK (version 4.1P) from the LINKAGE programs,^7–11 and multipoint parametric linkage analysis under genetic heterogeneity was performed using the GENEHUNTER program (version 2.0).¹² Parametric analyses were performed under a dominant model with 1% disease allele frequency and penetrances of 0.04, 0.99, and 0.99 for aa, Aa, AA genotypes for males and 0.01, 0.85, 0.85 for females. These values were chosen to match those in the models used by Fagerheim et al.¹ Multipoint non-parametric linkage analysis (NPL) was performed using GENEHUNTER by analysing all affected family members. Multipoint variance component analysis was also performed using GENEHUNTER¹³ under four models that all included QTL additive variance, polygenic additive variance, and environmental variance, and with dominance variance at neither, both, or either the QTL or polygenes. Note that GENEHUNTER can only accommodate pedigrees of a limited size, so it was necessary to subdivide 10 large pedigrees, probably reducing the power to detect linkage. Marker haplotypes were constructed using GENEHUNTER to determine whether any families possessed the Norwegian cosegregating haplotype.

RESULTS

Results of two point and multipoint parametric linkage analyses provided weak evidence for linkage between PCD and the DYX3 region (maximum two point lod score = 0.77, θ=0.3 at D2S1352; multipoint peak hlod = 0.07 at D2S1352). NPL analysis provided stronger evidence for linkage to DYX3, with a peak NPL Z_all score of 2.33 (p=0.0087) at D2S1352, and p<0.05 from D2S1352 to D2S2352, thus surpassing the recommended p=0.01 required to claim significant evidence for linkage in a replication study.¹⁴ The discrepancy between the parametric and NPL results suggested that our parametric analysis model (based on that of Fagerheim et al¹) may not have been the most appropriate for our Canadian sample. We therefore subsequently performed two point linkage analyses using recessive, intermediate, and dominant inheritance models, all with reduced penetrance. The strongest evidence for linkage (lod = 1.42, θ = 0.1) was found to D2S378 under an intermediate model (penetrances of 0, 0.4, 0.6 for aa, Aa, AA genotypes, frequency of 25% for disease allele “A”), supporting the notion that the assumptions of the model of Fagerheim et al¹ were less accurate for our sample. Multipoint variance component linkage analyses of the phonological awareness, phonological coding, and spelling measures produced the most significant results under a model that included QTL additive and dominance variance, polygenic additive variance, and environmental variance. The peak lod scores were 3.82 between D2S2352 and D2S378 for spelling, 1.13 at D2S378 for phonological coding, and 1.01 at D2S378 for phonological awareness. These lod scores were generated under two degrees of freedom, owing to inclusion of QTL additive and dominance variance. To allow for comparison with a traditional linkage threshold of lod = 3.0, the lod scores under one degree of freedom were estimated to be 3.12 for spelling, 0.69 for phonological coding, and 0.60 for phonological awareness. Investigation of haplotypes in affected subjects failed to identify the D2S378/D2S2279/D2S2183 haplotype that cosegregated with dyslexia in the Norwegian family; however, the D2S378 allele in the Norwegian haplotype was rare in our sample (observed in only two of 877 subjects).

CONCLUSION

In conclusion, linkage analyses of dyslexia and quantitative reading measures in a large Canadian family sample have provided the first independent evidence for the DYX3 dyslexia locus on chromosome 2p15-p16, originally reported in a large Norwegian family.