The present genome-wide linkage scan of SRS scores was performed using 518 relatives from a unique resource of large extended and smaller multiplex ASD families from Utah. We found genome-wide significant linkage on chromosome 15q13.3, and several other suggestive findings. Peaks on chromosomes 7q, 13q and 15q were also found in our previous genome scan of affected-only subjects in these families based on clinical diagnosis [32
]. SRS quantitative analyses also replicated a finding on chromosome 11 p reported in a previous genome scan of the SRS [20
], providing some evidence for genetic consistency of this measure used as a quantitative trait across studies.
The SRS was developed as a continuous measure of autistic traits, with variation intended to extend across affected subjects and unaffected relatives. Among affected subjects, our data suggest substantial correlations between clinical measures of autism from the ADI and SRS total scores, particularly for the social behavioural domain. ADI and SRS are expected to differ to some degree because ADI scores are based on information from the developmental period between ages 4-5 and also on the most severe presentation through the individual's life, while the SRS focuses on current behaviour. In addition, most clinically affected subjects scored in the spectrum or affected ranges on the SRS, though as expected, we observed a substantial range of scores. Most of the clinically unaffected relatives scored in the normal range on the SRS (87.4%), but considerable variation also occurred in these scores.
In our previous genome-wide scan, all relatives with no clinical diagnosis were considered unknown, so results were based on phenotypic information from 192 genotyped affected cases. Using the SRS in the present study, we added information for 337 clinically unaffected relatives, though 11 clinically affected subjects from the previous scan were missing SRS scores and were not included here. A small percentage of clinically diagnosed relatives (N
= 14) scored within the normal range on the SRS and therefore also contributed different information to the current scan. Our families also replicated previous evidence for significant SRS spouse resemblance (r
= 0.33, P
= 0.001) [17
The new information provided by the SRS in these families revealed additional findings beyond our previous affected-only clinical diagnosis scan [32
]. Not surprisingly, given the overlap in classification of affected subjects based on the SRS or the ADI/ADOS, the best overlap in findings between the current study and our previous study resulted when the SRS was treated as a qualitative trait. The use of a SRS-defined spectrum category and the addition of information from other clinically unaffected relatives provide the primary difference in findings between this study and our previous clinical diagnosis scan. The investigation of the whole range of quantitative information produced additional results, which again is perhaps not surprising. We used parametric quantitative methods in an effort to increase the similarity between qualitative and quantitative analyses. We realize that multiple tests may have increased our chances of a false positive result. If we conservatively assume that models are uncorrelated, our significance thresholds would be adjusted by log10
(5) = 0.7 LOD score units [58
] for these five tests: dominant, recessive, NPL qualitative tests and parametric and NPL quantitative tests.
Our genome-wide significant result on chromosome 15q13.3 was supported by the qualitative NPL and recessive models and some lesser evidence also by the quantitative analysis (see Figure ). In our previous genome scan our best results also occurred on chromosome 15q in three locations that may represent a single peak (29,459,872 bp, 36,837,208 bp, and 55,629,733 bp) [32
]. Liu et al
] have also reported evidence for linkage in this region (LOD = 4.01 at 15q13.3-q14).
The region on chromosome 7q31.3-32.3 also occurred in our previous clinical diagnosis scan, and was supported consistently across analysis models. This peak (125,450,000 bp - 130,300,000 bp) has been implicated previously by other studies [59
]. Our findings suggest that SRS information strengthened this peak over the 7q peak in previous clinical diagnosis scan (LOD = 1.97), though results here are still only suggestive (LOD = 2.91). We note that the CNTNAP2 gene of recent interest in autism [64
] is at 145,444,386 bp, downstream of our peak. The final peak showing overlap with our clinical diagnosis scan was on 13q12.3. A second downstream peak at 13q32.1 appeared in the current SRS study that was not evident in the clinical diagnosis scan. While some studies have reported positive findings on chromosome 13 for autism, attention has focused on the neurobeachin gene (NBEAL2) [65
], which lies in between our two regions and is, therefore, not supported by our data.
Other peaks were observed on chromosomes 6 and 10. The regions on chromosome 6p25.3 and 6p22.1 have appeared in schizophrenia studies [66
] and also in bipolar disorder [69
]. The region also harbors a copy number variation (CNV) at 27,827,354-28,119,631 implicated in autism [71
]. Interestingly, chromosome 10p12.1 and 10q22.1-q22.2 also has been reported for schizophrenia and bipolar disorder in many of the same samples that showed chromosome 6 findings [69
]. Somewhat strikingly, two genome-wide associations with attention deficit and hyperactivity disorder (ADHD) have also been found on chromosome 10p12.1 at 27,672,044-27,694,523 and 10q22.1 at 72,126,813-72,120,519 [75
]; both findings are quite close to our linkage peaks. Additional evidence exists in two other samples of an association between ADHD and autism using the SRS [76
A quantitative analysis also revealed findings of interest. While results using the quantitative range of the SRS often paralleled the qualitative findings, there were interesting exceptions. We note that while we used the same genetic analysis software, the variation in the trait was treated quite differently, so direct overlap of findings would not necessarily be expected. For the quantitative analysis, we also note that adjusting for age and gender may have been a conservative adjustment, as clinical affection status is associated with age and gender. As a result of this adjustment, our quantitative results are unlikely to be due to behaviours that change with age or behaviour specific to gender. Our highest quantitative peak was on chromosome 7 in the same region as described above. Our next highest quantitative peak was on chromosome 11p15.5-p15.4, spanning a region from 7,300,000 bp to 22,400,000 bp (12.11 - 37.63 cM). This peak replicates a previously reported SRS peak from 15-45 cM found in an analysis of quantitative SRS scores [20
]. Liu et al
] also found linkage in this region, primarily using a phenotype of delayed phrase speech, but with supporting evidence also from other phenotypes. A previous genome scan of clinical diagnosis in 1181 multiplex families also identified this region [78
]. On chromosome 19q13.4, our quantitative evidence spanned a region from 57,150,000 bp to 63,780,000 bp. This region overlaps with previous autism linkage evidence [79
] and also has several CNVs associated with autism [71
Finally, we observed peaks in two other regions previously identified in our families in scans using affection status. We observed a quantitative peak on chromosome 9p24.3, from the telomere to about 4,330,000 bp. This peak has been previously reported for obsessive compulsive disorder (OCD) [80
]. The glutamate transporter genet SLC1A1 is near this region (4,480,000 bp - 4,580,000 bp), and has shown some association with OCD [82
While it is important to seek replication of these findings in independent samples, we intend first to pursue these findings within our sample families. It is possible that findings may be unique to our family sample; indeed, a rare genetic finding for autism would be our expectation. However, if we can find genetic variation responsible for these statistical findings, the results may still provide relevant evidence regarding biological mechanisms and genetic pathways implicated in autism. Our family data will allow us to investigate segregation of the genetic variants responsible for these peaks and to determine more detailed associations with phenotypes related to autism. The extended family design allows a more powerful tool to investigate rare genetic variants in relation to phenotype than the genome-wide association study design.