We performed a nonparametric genome scan on a subset of 133 families from the AGRE sample with extended phenotypic information to test the approach used in previous studies to reduce heterogeneity and strengthen linkage signals by stratification based on proband language delay endophenotypes (Buxbaum et al. 2001
; Shao et al. 2002a
). We also incorporated information from parents of probands with reported language difficulties to determine whether this would similarly strengthen linkage signals as shown by Bradford et al. (Bradford et al. 2001
As shown previously, stratification of families based on proband phrase or word delay did appear to strengthen linkage signals in a number of regions. However, none met suggested criteria for genome-wide significance or for replication (Lander and Kruglyak, 1995
). A region on chromosome 2q was of particular interest as an autism susceptibility region for two reasons: 1) it had been previously reported by two groups (Buxbaum et al. 2001
; Shao et al. 2002a
) to have increased evidence for linkage in language-delayed families; and 2) there is an association between autism and a mitochondrial gene in this region (Ramoz et al. 2004
). However, when the AGRE families included in the Buxbaum et al. analysis were removed from the current study, the peak in that region was diminished. This suggested that the same group of families contributed to the linkage signal on 2q in both studies. Thus, the present study does not provide additional support for linkage in the 2q region despite having a larger number of families than the Buxbaum report.
We did not observe any strengthening of the signal in the region of chromosome 7q identified by Bradford et al. (Bradford et al. 2001
). Our group has demonstrated suggestive linkage to this region on chromosome 7q in AGRE using a quantitative trait locus (QTL) approach based on age at first word (Alarcón et al. 2002
). Further analysis of this region has demonstrated that the linkage peak may not be related to the magnitude of language delay per se, but rather to a more general language-related susceptibility trait (Alarcón et al. 2005
). Thus, not finding an effect of stratification based on delay in the AGRE sample would not be surprising. However, we did find a region at the telomere on chromosome 7 in the phrase delayed families as well as very modest evidence (p<.05) for linkage on chromosomes 1, 2, 4, 6, 8, 9, 10, 12, and 19 in areas that have not previously been reported as linked to autism and will need to be further studied in an independent sample.
We also examined the utility of incorporating historical information regarding parental language difficulties into the linkage analysis. Bradford et al. (Bradford et al. 2001
) hypothesized that extending the specific endophenotype of language delay to other family members should increase the signal at any locus related to that phenotype. Expecting that the majority of the increased signal would come from families who also had probands with speech delay, they also stratified on phrase speech delay in a manner similar to our analysis. Their results on chromosome 7 and 13 showed a modest effect of inclusion of parent information, made stronger by combining with stratification based on the presence of proband phrase speech delay. Results of the present study did not support those described in Bradford’s report.
The lack of support observed in this study of previous linkage peaks strengthened by stratification and/or inclusion of parental information could be explained in a number of ways. First, none of the linkage results in the previous stratification studies reached genome-wide significance and therefore may have represented spurious results based on sub-setting of the cohort into smaller groups. Similarly, although the results for the Bradford et al. study (2001)
that explored the utility of including parental information in the linkage analysis was based on 50 families, they still did not reach significance for genome-wide scans (Lander and Kruglyak 1995
) and may represent spurious findings. Second, the present results may reflect a lack of power due to the small sample size that was a consequence of including only the subset of the complete AGRE sample with available parental language information. Third, in this cohort, it is also possible that the language deficits in parents are not related to the specific delay endophenotype in their children, and, thus, inclusion of parental information in the present linkage analysis increased heterogeneity rather than reducing it. This is supported by the lack of parent-child clustering of language delay within families. Finally, although the questions and methods used to obtain these data were similar to those used in other studies, it is possible that the self-report language history information we obtained from parents is not reliable.
This study provides an example of the difficulty phenotypic heterogeneity poses for the identification of autism susceptibility genes. Furthermore, the fact that different studies all using similar language-related endophenotypes yield linkage to different loci implies that even the language delay component of ASD could be genetically heterogeneous. For instance, a child could present with speech delay secondary to a true expressive language deficit (which may also be accompanied by a receptive language deficit), or due to a more specific motor speech disorder (e.g. speech apraxia). These different underlying pathologies could have distinct underlying genetic etiologies leading to the identification of a variety of loci in linkage analyses. Alternatively, this trait could co-segregate with another trait (so far undetected) that truly underlies linkage to the chromosomal regions identified. Better definition of the language endophenotype and perhaps further subgrouping of much larger samples into more homogeneous groups based on related aspects of the phenotype (e.g. receptive and expressive speech delay or speech apraxia) will be necessary to further explore this concept.