We have found evidence for a novel QTL locus for ocular refraction on chromosome 1p36 in an Ashkenazi Jewish population. The linkage peak on chromosome 1 has not previously been reported for either qualitative (i.e., myopia) or quantitative ocular refraction phenotypes.
The statistical procedure implemented in MERLIN-REGRESS estimates the test statistic of individual linkage signals based on asymptotic properties and large-sample central limit theorem. Sham et al. found their test statistic to be unbiased and that it provided correct type I error rates for simulated sibships ranging in size from 2 to 6 (Sham et al. 2002
). The authors maintain that the method is applicable to general pedigrees of all types although they limited their simulation studies of non-nuclear families to cousin pedigrees only. Our gene-dropping simulations suggest that the magnitude of the linkage statistics were elevated in our study, yielding inflated type-1 error rates if P
values had not been determined empirically. This may have been the result of extreme between-relative discordance in refraction having a large influence on the regression slope. In addition, this upward bias in LOD scores under the null hypothesis of no linkage was greater for REF than for LTR even though the sample’s marginal trait distributions were similar for the untransformed (skewness=− 0.48, kurtosis=3.47) and transformed (skewness= 0, kurtosis=3.37) data, with the transformed trait being slightly closer to normality. Sham et al. (2002)
suggest that liberal results can occur when a small number of families contribute inordinately to the linkage statistic. MERLIN-REGRESS provides a measure of family informativeness, conditional on the trait values of the family members, to assess each family’s expected contribution to the test statistic. For the untransformed trait, a single family accounted for 19.3% of the total informativeness and the five most informative families contributed 45.4% to the informativeness index. After transformation, the distribution of family informativeness was much less skewed. For LTR, the most informative and five most informative families accounted for 7.5% and 30.8% of the total informativeness, respectively. This demonstrates the importance of appropriate data transformation and of calculation of empirical significance levels for this statistic. The empirical genomewide P
value (a family-wise error rate or FWER) associated with our maximum LOD score of 9.5 for REF was 0.065 whereas our maximum LOD of 8.7 for LTR yielded a genomewide significance smaller than 0.005. Lander and Kruglyak (1995)
suggested that allele-sharing LOD scores between 3.3 and 3.8 were needed to establish genomewide significance for non-parametric linkage scans at a genomewide P
value (FWER) of 0.05. However, these guidelines were established for allele sharing methods and relative pair designs and are unlikely to be applicable in general pedigree situations. In our sample of extended pedigrees, the simulations of LTR indicate that a MERLIN-REGRESS LOD score of 4.69 or larger is required to exhibit a genomewide significant p-value of 0.05 or less.
The discrepancy between empirical LOD scores obtained through simulation of unlinked markers and theoretical expectations emphasizes the importance of performing computer simulation studies to determine study significance, especially when the applicability of asymptotic theory is in question, such as with small sample sizes and extended, and unbalanced, pedigree structures. In addition, careful consideration must be given to choosing an appropriate transformation of the data since departures from distributional assumptions can have a significant effect on study error rates, as illustrated by the differences in empirical type-1 error rates between REF and LTR.
We compared the results from the regression-based approach to standard variance components linkage analysis, which yielded LOD scores of 1.8 (P=0.002) for refraction and 2.23 (P=0.0007) for LTR at D1S552. These results support the evidence for linkage to this locus and suggest that the regression-based approach may be more powerful than the variance components method in highly ascertained samples.
We carried out linkage analyses while varying the statistical model parameters in order to determine the sensitivity of the MERLIN-REGRESS linkage statistic to possible parameter misspecifications. Our sensitivity results indicate that alternate specifications of the statistical model parameters within a plausible range have little effect on the magnitude of the maximum linkage signal in our data (). Specifically, the linkage peak on chromosome 1 decreased substantially only for extreme population mean and heritability specifications. Empirical P values associated with the maximum linkage peak on chromosome 1 were below a genomewide significance level of 0.05 for all models except those that specified either very low variances or high heritabilities (). Even using our sample mean (which was most likely skewed towards myopia due to our ascertainment scheme) as the underlying population mean would have yielded a LOD score profile similar to that of our putative model. Therefore, assuming that our linkage signal is true, it appears that the regression method implemented in MERLIN-REGRESS is robust to misspecifications of the statistical model parameters.
Using the same population, we have previously reported linkage of myopia, characterized as a binary trait, to a locus on chromosome 22 (Stambolian et al. 2004
). The current QTL analysis shows only suggestive evidence for linkage of ocular refraction to this region (peak LOD score on chromosome 22=0.93 at 9.1 cM for LTR, empiric locus-specific P
=0.03). In addition, the prior parametric analysis did not reveal linkage of the dichotomous trait, myopia, to chromosome 1. While these results may appear incongruous, several differences may account for this discrepancy. First, QTL linkage analyses take account of the entire sample distribution of the trait. In particular, the QTL regression method implemented in MERLIN-REGRESS capitalizes on between-relative trait differences and sums to infer statistical linkage. Hence, individuals who are highly discordant for a quantitative trait (for example, a −1 D myope and a −15 D myope) may be in the same outcome group in binary trait analyses. In our sample, ocular refractions ranged from −15.875 to +6.75 D and 87% of individuals had myopia of 1 D or more. Therefore, quantitative analyses provided additional resolution of the phenotype within broad categorical classes. Second, ocular refraction is determined by various anatomical components that modulate the eye’s refractive power as well as its size. Myopia results primarily from a relative axial elongation of the posterior chamber, a component of the axial length, which has been shown to be a highly heritable trait in a variety of human populations (Alsbirk 1979
; Hammond et al. 2001
; Lyhne et al. 2001
; Biino et al. 2005
). A “myopia” gene may affect the sign of optical defocus whereas QTLs affecting ocular component growth may be responsible for regulating differences in the degree of refractive error. Hence, binary trait analyses may allow for the detection of “myopia” genes but be unable to discriminate between varying degrees of myopia whereas QTL methods would be more useful in identifying genes that regulate refraction across the entire phenotypic spectrum.
Using variance components linkage analyses, Biino et al. (2005)
recently reported suggestive linkage (LOD>2) of ocular axial length to a locus at 2p24 (marker D2S1360
at 38.3 cM) in Sardinian families. They also reported suggestive evidence for linkage of anterior chamber depth to chromosome 1 and of radius of corneal curvature to chromosomes 7, 2 and 3. This group did not report linkage results for ocular refraction although the values of these biometric components, and their relation to one-another, fully determine refraction. We found mild evidence for linkage of refraction (maximum LOD for LTR on chromosome 2=1.70, empiric locus-specific P
=0.01) to a locus near marker D2S405
at 50.4 cM, or approximately 12 cM from Biino et al.’s signal for axial length. This may implicate ocular axial length, which is the primary determinant of myopia, as the underlying cause for our linkage signal on chromosome 2.
Previous studies have mapped high (or “pathological”) myopia susceptibility loci to chromosomal regions 7q (Naiglin et al. 2002
), 12q (Young et al. 1998a
), 17q (Paluru et al. 2003
) and 18p (Young et al. 1998b
). We found no evidence for linkage of refractive error to loci on chromosomes 12 or 18. However, we found a small linkage peak (LOD=1.13 for LTR, empiric locusspecific P
=0.01) at 10.7 cM on chromosome 17. Paluru and collaborators (2003)
have previously mapped a locus for autosomal dominant high myopia to a region flanked by markers D17S5787
located at 75.0 and 82.6 cM on chromosome 17. Given the distance of more than 60 cM between our weak linkage signal on chromosome 17 from this previously reported linkage peak, it is unlikely that the two studies are giving indication of linkage to the same locus.
Using a parametric statistical approach, Naiglin et al. (2002)
reported suggestive evidence for linkage of familial high myopia (multipoint LOD score of 2.81) to a locus at 7q36 flanked by markers D7S798
(169.0 cM) and D7S2546
(173.0 cM). We found mild evidence for linkage of refraction to marker D7S3058
at 173.7 cM on chromosome 7 (LOD=1.33 for LTR, empiric locus-specific P
value=0.02). Although our peak was comparatively shallow, the signal is intriguing in view of its close proximity to Naiglin et al.’s locus and represents support for a myopia susceptibility locus on chromosome 7.
One previous genomewide QTL analysis has been carried out for ocular refraction. In a study of 221 dizygotic British twin pairs, Hammond et al. (2004)
reported significant (i.e., LOD>3.2) linkage peaks at 40 cM (LOD=6.1) on chromosome 11p13 as well as to loci at 3q26 (LOD=3.7), 8p23 (LOD=4.1), and 4q12 (LOD=3.3). Interestingly, the linkage region on chromosome 11 contains the PAX6
gene at 39–42 cM, which is fundamentally related to eye development in a number of species (Wawersik and Maas 2000
; Dominguez et al. 2004
) and in which mutations are known to cause a variety of ocular abnormalities in humans (Hanson et al. 1999
). Our analyses revealed a peak LOD score of 1.02 (empiric locus-specific P
value=0.03) at 90.6 cM on chromosome 11 for LTR. Since our peak is about 50 cM away from the previously mapped locus, our results should not be viewed as confirmatory of an ocular refraction QTL on chromosome 11. However, they are interesting in light of Hammond et al’s linkage results and should be followed-up with a denser marker map and/or the typing of additional families.
The area of most significant linkage, with a maximum LOD of 8.7 (empirical genome-wide P<0.005) observed between markers D1S552 (LOD=5.09) and D1S1622 (LOD=3.44), maps roughly to a region between 19 Mb and 30 Mb on chromosome 1. In our simulations for LTR, a genomewide significance of P<0.05 corresponded to a LOD score of 4.69, marking this as the most likely region to contain a QTL for ocular refraction. This region is predicted to contain approximately 189 genes. However, the linkage region on chromosome 1 that contains LOD scores of two or greater is considerably wider and spans a genetic distance of approximately 35 cM (or more than 20 Mb) from D1S552 to D1S3721. This large region includes several genes that are expressed in human eye tissue (). While none of these genes are obvious candidates for refractive regulation, they will be carefully examined in our follow-up studies of this region. Our future investigations will involve replicating this linkage signal with an independent sample of Ashkenazi Jewish families. Furthermore, we will attempt to refine the area of linkage by using a dense SNP map in the region, which will also permit us to perform family-based association studies.
Genes expressed in human eye tissue in chromosome 1 linkage region
Since the eye can be decomposed into individual components that contribute to refraction, additional linkage studies should begin to focus on these constituents. Given the multiplicity of loci identified to date for high myopia in various populations, it is possible that considerable locus heterogeneity also exists for ocular refraction. Our use of a relatively isolated population may have led to the identification of a population-specific locus for refraction. Nevertheless, considering the ubiquity of less severe forms of myopia worldwide it is likely that a common underlying genetic mechanism(s) is responsible for refractive development among humans.