Important clues about the mechanisms underlying autism come from the monogenic disorders Rett’s syndrome and Fragile × syndrome (see Essay by Kelleher and Bear, page 401 of this issue). Rett’s syndrome, caused by mutations in the human MECP2 gene, which encodes the methyl-cytosine binding protein, occurs almost exclusively in girls (mutations in this X-linked gene are lethal to males). Affected girls show normal development for 1–2 years, followed by the appearance of stereotyped repetitive hand movements and a regression of neurological and social skills. Similarly, one of the best-known disorders associated with autistic symptoms is the Fragile × syndrome. Males with Fragile × syndrome typically have developmental delays (for example, delays in walking or language) and somatic dysmorphology (for example, large ears, a large head), and so Fragile × is generally not referred to as autism per se. Although there are many syndromic forms of autism, it is also important to stress that 40% to 60% of autistic children show some degree of mental retardation (Giacometti et al., 2007), although measuring IQ in autism may pose challenges. Also, some pediatric neurology clinics report that ~30% of autistic patients suffer from seizures (Fombonne, 1999).

As clinical genetic testing becomes more frequent, more autistic children will receive specific genetic diagnoses. The same genetic defects associated with broader syndromes are now recognized to cause a spectra of abnormalities, including those found in some higher-functioning children with autism. For instance, deletion of 22q11 (Vorstman et al., 2006), Fragile × gene abnormalities, or milder MECP2 mutant alleles (Moretti and Zoghbi, 2006) are occasionally seen in children without the typical broader syndromes associated with these conditions. Why similar genetic abnormalities cause a range of phenotypes is an enduring mystery and may reflect the involvement of other genetic or nongenetic factors.

One of the areas that has seen the most rapid recent progress in autism genetics has come with the realization that up to 7%–10% of children with autism have a variety of de novo chromosomal deletions and duplications (Sebat et al., 2007;Marshall et al., 2008; Kim et al., 2008). These deletion syndromes typically cause a spectrum of phenotypes that includes autism. Some chromosome aberrations have been helpful in identifying specific genes mutated in autism, and some of these genes appear to have brain-specific functions. For instance, × chromosome deletions implicated the NLGN3 and NLGN4 genes (Jamain et al., 2003) encoding neuroligins 3 and 4, which are synaptic adhesion molecules. Subsequent studies identified an NLGN4 mutation inherited in a large family with many affected males showing mental retardation and/or autism (Laumonnier et al., 2004). These findings suggest that NLGN3 and NLGN4 gene mutations are causes, albeit rare, of autism, mental retardation, and potentially other neuropsychiatric syndromes. The SHANK3 gene, which encodes a cytoplasmic binding partner of the neuroligins, was also identified based on a chromosomal deletion (Durand et al., 2007). Chromosome deletions (Kirov et al., 2007; Moessner et al., 2007) and translocations (Kim et al., 2008) involving the NRXN1 neurexin gene, which encodes an extracellular binding ligand for neuroligins, have implicated this gene in autism as well. Rare changes in CNTNAP2, encoding contactin associated protein-like 2 (a neurexin superfamily member), are associated with autism, and common alleles of CNTNAP2 may increase the risk of developing autism. Even these mutations, which can affect the neuronal synapse, tend not to be highly specific for “pure autism” but rather cause a broader mental retardation phenotype in some patients and pure autism in others.

Why do many mutations associated with autism occur de novo, meaning they are present in the child but not in the parents or earlier ancestors? This is actually not surprising. Autistic children rarely marry and have children, so that the condition is subject to “negative evolutionary selection,” with affected patients having decreased fertility. This negative selection means that for mutations to be observed in the population they must frequently occur as de novo mutations. Such negative evolutionary selection is common to mutations associated with other severe diseases, especially those of childhood, for example mental retardation, epilepsy, or congenital heart disease (Lupski, 2007). A recent study focused on identifying autosomal recessive loci in pedigrees with autism and recent shared ancestry (Morrow et al., 2008). Autosomal recessive mutations, like sporadic mutations, also frequently reduce reproductive fitness. In this study, consistent with the finding of copy number variations (CNVs) in autism, multiple heterogeneous loci were implicated, but these loci represent potential inherited causes. Hence, by tracing shared ancestry this approach may provide an important way to identify inherited loci in heterogeneous neurodevelopmental conditions that reduce reproductive fitness.

Other genes associated with autism seem to affect neuronal excitability, or the ratio of excitation to inhibition (Rubenstein and Merzenich, 2003). Indeed, mutations in neuroligins appear to affect the balance between excitatory and inhibitory synapses (Tabuchi et al., 2007). MeCP2 mutations in mice also appear to regulate the balance of excitatory and inhibitory synapses (Chao et al., 2007; Dani et al., 2005). Furthermore, evidence is accumulating that several genes associated with autism play roles in the development of cortical inhibitory interneurons including c-Met (Campbell et al., 2006, 2007), DLX (Hamilton et al., 2005), and ARX, where rare, partially inactivating mutations cause mental retardation or autism (Kato et al., 2004; Turner et al., 2002). A common theme of these studies is that autism may relate less to generic synaptic function and more to levels of excitation or inhibition, and the balance between the two.

Another hypothesis to explain the specificity of autism for affecting language and social behavior holds that these are simply the most difficult tasks ever learned by the human brain, so that mild global defects in synaptic function may be sufficient to impair them. Language acquisition, at ~2 years of age is one of the last developmental milestones, whereas learning about social relationships continues into adulthood. The late appearance of these landmarks may reflect the complex circuitry they require, which may be susceptible to subtle defects that spare simpler tasks. Alternatively, genes implicated in autism may include those that have been under recent evolutionary selection (Baron Cohen and Belmonte, 2005). After all, the ability of humans to live in large social groupings—that is, civilization—was a recent evolutionary development, occurring only after the appearance of anatomically modern humans, and probably as recently as 50,000 years ago. There is evidence that some genes (Williamson et al., 2007) and genomic regions (16p11.2) in which mutations cause autism show signatures of recent evolutionary selection.

An even more remarkable example of the reversibility of autism-related genetic and neurobiological defects comes from the study of mouse models of Rett’s syndrome. The Bird and Jaenisch labs created mice that lacked MeCP2, and when the mice were born they developed the typical symptoms of Rett’s syndrome (Giacometti et al., 2007; Guy et al., 2007). Remarkably, reinstating the MeCP2 gene after birth caused partial (Giacometti et al., 2007) or seemingly complete (Guy et al., 2007) resolution of the abnormalities. These findings suggest that MECP2 is not essential for the earliest wiring of the nervous system but instead is required later for activity-dependent processes. This is consistent with the normal development of Rett’s syndrome in girls until 18–24 months of age, and it highlights the potential for reversing the symptoms of Rett’s syndrome in humans, if effective therapies can be found.