Turner syndrome (45, X, TS, monosomy X) is the genetic disorder resulting from the absence of all or part of one X chromosome in females. The complex TS phenotype includes short stature, ovarian failure, and a characteristic neurocognitive profile 1. Although severe developmental disorders do not predominate in TS, the risk of selective deficits in certain cognitive domains is substantially increased. Girls and women with 45, X TS commonly demonstrate deficits in visual-spatial abilities, visual-perceptual abilities, motor function, nonverbal memory, executive function and attentional abilities when compared to normal females matched for age, height, IQ, and socioeconomic status 23456789.

Both hormonal and genetic factors may influence the cognitive development of TS females. Hormonal influences are attributable to deficient ovarian production of estrogen, androgen, or both. Subtle estrogen effects on motor function and processing speed and androgen effects on working memory have been demonstrated 101112. However, many of the cognitive deficits characteristic of TS are consistent across a wide age range, including children as well as estrogen-replaced adults 41112. Furthermore, these deficits are generally not seen in females with premature ovarian failure and intact X chromosomes 1314, suggesting a predominant role of genetic factors in the etiology of TS cognitive deficits.

The genetic abnormality in TS is determined by the absence of one copy of genes on the X chromosome. Most aspects of the TS phenotype, including the cognitive phenotype, are thought to be due to half-normal gene dosage, or haploinsufficiency, of X-linked genes that escape inactivation 1516. Although one X chromosome undergoes inactivation in normal females during early embryogenesis, about 15% of all X chromosome genes, mostly situated on the short arm (Xp), remain active to some degree on both X chromosomes 17. Some of these genes have functional Y-linked homologs that are thought to balance their dosage between males and females. The pseudoautosomal region (PAR1), a 2.6 Mb interval at the tips of the X and Y short arms, is a plausible location for TS genes because PAR1 genes are all expressed at diploid dosage in both males and females 1819. One PAR1 gene, SHOX [Mendelian Inheritance in Man (MIM) 12865], has been implicated in TS short stature 20. Whether any PAR1 genes contribute to other aspects of the TS phenotype is unknown.

One way to deduce the underlying genotype-phenotype relationships in TS is to compare the phenotypes of individuals missing various portions of one X chromosome in order to assign specific features to "critical regions." A trait maps to a region if deletion of that region accounts for the variance in that trait. This approach was used to map short stature to the SHOX gene on Xp 20 and ovarian failure to regions of the long arm (Xq) 21. We have previously applied this phenotype mapping methodology to define genetic correlates of the cognitive phenotype in children with TS 22.

We used discriminant function analysis to derive a mathematically defined TS cognitive phenotype to test for association with deletions of specific regions of the X chromosome 2223. The initial results from 34 subjects with deletions of varying portions of Xp, mapped mostly by fluorescence in situ hybridization (FISH), identified a probable association between deletion of the distal ~10 Mb of Xp and the TS neurocognitive phenotype 22. However, those conclusions relied upon identification of the TS neurocognitive phenotype in a small number of individual subjects; the sample was too small to permit rigorous statistical inference. The TS cognitive phenotype was defined as a dichotomous measure, rather than a quantitative trait, using a cutoff score. In addition, the population was heterogeneous, including both children and adults, and did not include any subjects with Xq deletions, who commonly receive the TS diagnosis.

In the present study, we attempted to confirm and expand the initial findings using 47 adult females with nonmosaic deletions distributed along both Xp and Xq. Potential sources of variation in cognitive outcome have been minimized by including only adults with similar estrogen-replacement status. In this study we treated the TS cognitive phenotype as a quantitative rather than a qualitative trait and compared the results of women with deletions of varying portions of the X chromosome to that of 45, X TS and normal controls. We hypothesized that an association exists between the defined TS neurocognitive phenotype and deletion of Xp22.3, which includes the distal 8.3 Mb of Xp.

An earlier descriptive study suggested that Xp22.3 contains one or more genes that influence cognitive ability in Turner syndrome 22. The current results provide additional support for this hypothesis. The strength of the conclusion is based on a larger, all adult study population that included both Xp and Xq deletion subjects, with uniform estrogen replacement status for those with ovarian failure. Additionally, the previous study defined a critical region by determining the minimal overlapping deletion among a few individuals with the TS neurocognitive phenotype, assigned as a dichotomous trait. The present study used a quantitative measure of the TS neurocognitive phenotype, the TSCS, to test for associations with deletions, reducing error due to variability of individual results.

The TSCS score is a composite, quantitative summary that partially describes the TS cognitive phenotype. For each individual, the score represents a weighted pattern of performance on a number of cognitive tests that assess multiple clinical domains. Therefore, the mean TSCS score for a group is an average of performance on this composite measure. As with all computed means, the mean TSCS score is not meant to reflect a typical performance but rather the mean of performances by individuals in that population that may contain both typical subjects and atypical subjects. There is overlap in TSCS scores among the TS and control populations. However, the difference in the means is statistically significant (P < 0.001). This overlap in TSCS scores indicates that many TS subjects function within the normal range (mean ± 2 SD for controls). If there were distinct and complete separation of both populations in any cognitive trait, then discriminant function analysis would not be necessary. This cognitive result is similar to the findings for the height deficit in TS. Although there are some TS subjects whose height is in the normal range, the mean height is significantly reduced in TS 1.

The mean TSCS score from 28 women with overlapping Xp deletions of varying size was similar to that of 45, X TS women and differed significantly from that of controls and women with Xq deletions (Fig. 3A). Their overlapping Xp deletions all included Xp22.3, the region previously hypothesized to affect TS cognition. We therefore investigated the mean TSCS score of the 15 subjects whose deletions were restricted to this interval (<Xp22.3). Their mean TSCS score was also similar to that of 45, X TS subjects and differed significantly from that of controls (Fig. 3B). Larger Xp deletions (>Xp22.3) were not associated with further reduction in mean TSCS score or increased individual variability in performance (Fig. 3B, Table 3). This supports the conclusion that deletion of Xp22.3 may be sufficient for producing this aspect of the neurocognitive profile in TS.

Previous studies have shown that the likelihood of ovarian failure in patients with terminal Xp deletions is directly related to the size of the deletion 2633. Our data also show this relationship. The mean TSCS score of Xp deletion subjects without ovarian failure was similar to that of 45, X TS subjects, suggesting that distinct Xp loci contribute to ovarian failure and neurocognitive deficits in TS.

Using an expanded panel of FISH and microsatellite markers, and in two cases CGH, we mapped the associated interval to an 8.3 Mb region of Xp22.3 containing approximately 30 genes. It is interesting that a deletion limited to the pseudoautosomal region in subject 430 was associated with mild developmental delay in her son. This is surprising, since haploinsufficiency of the entire pseudoautosomal region does not usually cause global developmental delay in females. Possible explanations include unmasking of a recessive Y pseudoautosomal allele, gender-specific effects, or coincidence. The boy did not undergo detailed cognitive testing.

While it appears that deletions limited to pseudoautosomal sequences may be sufficient to reduce the TSCS score, as previously hypothesized, this conclusion is tempered by the fact that two of these deletions were due to unbalanced Xp;Yq translocations, and Yq heterochromatin could exert position effects on nearby genes such as NLGN4X [MIM 300427]. The lack of effect of Xq deletions on the mean TSCS score suggests that the phenotype is specific for deletion of Xp22.3 rather than a nonspecific manifestation of aneuploidy or ovarian failure, which showed no correlation with TSCS score.

Several aspects of the Xp22.3 region are noteworthy. Deletions limited to Xp22.3 usually manifest short stature rather than the full TS phenotype and are associated with variable patterns of X inactivation, whereas larger deletions are associated with nonrandom inactivation of the deleted X chromosome 34. Patients with Goltz, Aicardi, or MLS syndrome, who may show deletions limited to Xp22.3 with highly skewed inactivation 34, were excluded from our study population. Most of our subjects with deletions limited to Xp22.3 did not show highly skewed inactivation. Subjects with larger Xp deletions or Xq deletions generally showed nonrandom inactivation, presumably of the deleted X chromosome. The TSCS score distributions were similar for subjects with deletions limited to Xp22.3 and non-skewed inactivation and subjects with larger Xp deletions and skewed inactivation. Furthermore, the TSCS scores of subjects with Xq deletions, all of whom showed highly skewed inactivation, differed from those of subjects with Xp deletions. Thus, TSCS score was not related to pattern of X inactivation.

Most of the 30 or so genes in the Xp22.3 critical region (Table 4) escape X inactivation, and many have Y-linked homologs, suggesting that they are dosage sensitive 17. Functional studies of genes in this region using animal models are hindered by the apparent absence of almost the entire region in mouse and rat, except for two genes, PRKX [MIM 300083] and STS [MIM 308100] 19.

STS encodes steroid sulfatase, deficiency of which causes X-linked ichthyosis 35. Mouse Sts has been implicated in modulating neurosteroid levels and thus GABA_A receptor function 36. Mouse Sts is pseudoautosomal and therefore dosage sensitive, and 39, X mice lacking one copy of Sts show altered GABA_A receptor expression 37. These mice also show reduced visuospatial attention 38. Restoration of Sts diploid dosage by a truncated sex chromosome carrying the pseudoautosomal region and a small number of X-linked genes normalized GABA_A receptor expression and rescued the visuospatial attention deficit 3738. The investigators concluded that haploinsufficiency of a pseudoautosomal gene, possibly Sts, is responsible for visuospatial attention deficits in 39, X mice and perhaps analogous neurocognitive deficits in humans with TS.

Other genes in the Xp22.3 region that have been linked to cognitive phenotypes include VCX3A and NLGN4X. VCX3A [variable charged X-linked gene 3A, MIM 300533] belongs to a gene family with multiple closely related members on both the X and Y chromosome. VCX3A was proposed to be involved in cognition on the basis of overlapping Xp22.3 deletions associated with mental retardation in a few males 39. However, a subsequent study found that only one out of seven males in a family with icthyosis due to a microdeletion that included VCX3A and VCX [MIM 300229] had mental retardation 40. Furthermore, expression of human VCX genes is restricted to male germ cells 41.

NLGN4X, also in Xp22.3, encodes a neuroligin, or neural cell adhesion molecule, widely expressed in brain that escapes inactivation and has a functional Y homolog 1742. Overexpression of the protein in cultured hippocampal neurons has been shown to stimulate formation of presynaptic terminals 43. Frameshift and nonsense mutations in NLGN4X have been linked to Asperger syndrome/autism [MIM 300495, 300497] 44 and X-linked mental retardation [MIM300495] 45. Although cognitive effects of reduced NLGN4X dosage have not been described in female carriers of these mutations, autistic features have been described as part of the 45, X TS neurocognitive phenotype 46, and three females with autism and deletions of distal Xp encompassing NLGN4X have been reported 47. However, none of our subjects carried a diagnosis of autism spectrum disorder, and the relationship of TSCS results and autistic features is not clear. The TSCS score is based on performance on cognitive tests and therefore has no direct relationship to autistic behavior. However cognitive impairment on visual-perceptual and spatial processing tasks may be related to increased risk of social problems and altered interpersonal relationships.

Our study has several limitations. First, nonmosaic partial X chromosome deletions are rare. In order to obtain sufficient sample size for statistical comparisons, we included eleven subjects with unbalanced translocations (eight with Xp deletions and three with Xq deletions). While none of these subjects had any obvious features of autosomal trisomy, presumably due to spread of X inactivation to the autosomal segment, we cannot exclude the possibility of cognitive effects. However, the mean TSCS score of the unbalanced translocation subjects with Xp deletions was very similar to that of all Xp deletion subjects. Similarly, because the number of Xq deletion subjects was small, we included subjects whose deletions did not all overlap. This deletion heterogeneity could mask an effect of a locus in Xq. However, the mean TSCS score of just the subjects with terminal Xq deletions was similar to controls. Our results strongly support a TS neurocognitive locus in Xp but do not absolutely exclude a neurocognitive effect of deletion of Xq.

There was ascertainment bias toward ovarian failure, particularly for subjects with Xq deletions, although this aspect of the TS phenotype did not show any correlation with TSCS score. Although the study included only adults, the age range was relatively broad (17–55 years). However, age did not correlate with TSCS score either. Because parental DNAs were not available for most of the subjects, we could not address any possible imprinting effect, as has been claimed for social-behavioral aspects of the TS cognitive phenotype 48.

As with any complex trait, the presence and severity of the TS neurocognitive phenotype is variable. This study used discriminant function analysis, a relatively uncommon method of defining cognitive phenotype. This method was used because it allows a quantitative approach to cognition using a group of subjects who, though similar, have a moderate degree of cognitive heterogeneity. There is clearly overlap between the distributions of TSCS scores of 45, X TS versus controls: some controls have lower TSCS scores than the 45, X average, and some 45, X TS subjects have higher TSCS scores than the control average. Because of this overlap, the ability to infer genotype/phenotype correlations based on individual subjects is limited. For this reason, we based our conclusions on mean TSCS scores of groups of subjects with similar deletions. This approach does not require that every individual deleted for the cognitive locus manifest the identical cognitive phenotype.

The TSCS reflects performance in multiple cognitive domains, and includes visual-motor tasks, visual-perceptual tasks, executive function/attention tasks and memory tasks. Given the complex TS cognitive phenotype in TS, it is not expected that the absence of a single gene leads directly to this complete phenotype or even to a specific cognitive trait associated with TS. Therefore, it is unlikely that within Xp22.3 there is a specific gene that codes for a specific cognitive dysfunction. Rather, absence of this gene most likely affects a biological activity that is intimately involved in the development of the phenotype, manifesting as a complex and nonfocal neurocognitive phenotype. The TSCS is still an incomplete representation of the TS cognitive phenotype. There are other cognitive features, including motor and social function not included in the TSCS. Therefore, it is likely that the TS cognitive phenotype is due to multiple cognitive determinants and multiple genes, each contributing to the phenotypic variance.

Molko et al. 49 described cerebral structural changes in TS compared to female controls. These regions included the right intrapatietal sulcus and bilateral superior temporal sulci. These regions have classically been associated with spatial-perceptual ability and spatial-relational memory. Dysfunction in these cognitive domains is part of the TS cognitive phenotype. Therefore, it is reasonable to speculate that the cognitive change associated with deletion of Xp22.3 may be mediated through structural changes in these anatomic regions.

These results are clinically relevant for patients with partial X chromosome deletions. TS is classically defined as monosomy X with typical physical and cognitive phenotypes. Early karyotype/phenotype correlations studies implicated loss of the short arm in growth failure. The discovery of the SHOX gene suggested a continuum between idiopathic short stature [MIM 300582], due to deletion of only the SHOX gene, and TS, which could be viewed as a contiguous gene deletion syndrome. For purposes of diagnosis and treatment, the distinction has been made between deletions smaller than Xp22.3 associated only with short stature (idiopathic short stature or dyschondrosteosis [MIM 127300]) and larger Xp deletions associated with both short stature and ovarian failure (TS) 50. However, it must now be recognized that patients with Xp22.3 deletions, regardless of diagnosis, are at risk for TS neurocognitive deficits and should be evaluated accordingly. Furthermore, in the absence of mosaicism, the risk of these cognitive deficits in patients with Xq deletions is low.

In summary, haploinsufficiency of genes located in Xp22.3 appears to cause at least part of the multifaceted cognitive phenotype of 45, X TS, as indexed by TSCS score. This critical region contains 30 or so genes. The lack of murine orthologs of most of these genes (with the exception of STS and PRKX) and the limitations of mouse models for complex human cognitive abilities precludes identifying the causative gene by knockout models. An alternative approach is to test for association between genetic variation in Xp22.3 genes and the TSCS in 45, X TS subjects. In addition, high resolution genomic technologies such as array comparative genomic hybridization may detect submicroscopic deletions associated with TS cognitive deficits that would identify the causative gene(s), as was the case for short stature due to SHOX deletion. Identification of the TS cognitive gene would facilitate early diagnosis and intervention in individuals with the associated neurocognitive deficits.

The author(s) declare that they have no competing interests.

ARZ and JLR conceived of the study, participated in its design and coordination and wrote the manuscript. DR participated in the interpretation of the results and manuscript preparation. GS participated in the design of the neurocognitive assessment and manuscript preparation. PR performed molecular genetic studies. FE performed karyotypes. HK performed statistical analyses. KK coordinated subject recruitment and supervised neurocognitive testing. All authors read and approved the final manuscript.