|Home | About | Journals | Submit | Contact Us | Français|
Elena Grigorenko: Yale University School of Medicine, New Haven, CT 0620, USA
Karen M. Smith: Yale University School of Medicine, New Haven, CT 0620, USA
Hanna Stevens: Yale University School of Medicine, New Haven, CT 0620, USA
Increased brain size is common in children with autism spectrum disorders. Here we propose that an increased number of cortical excitatory neurons may underlie the increased brain volume, minicolumn pathology and excessive network excitability, leading to sensory hyper-reactivity and seizures, which are often found in autism. We suggest that Fibroblast Growth Factors (FGF), a family of genes that regulate cortical size and connectivity, may be responsible for these developmental alterations. Studies in animal models suggest that mutations in FGF genes lead to altered cortical volume, excitatory cortical neuron number, minicolum pathology, hyperactivity and social deficits. Thus, many risk factors may converge upon FGF-regulated pathogenetic pathways, which alter excitatory/inhibitory balance and cortical modular architecture, and predispose to autism spectrum disorders.
Autism spectrum disorders (ASD) including autism, Asperger’s Disorder, Pervasive Developmental Disorder (PDD), and Pervasive Developmental Disorder Not Otherwise Specified (PDD-NOS), encompass a spectrum of debilitating symptoms beginning early in childhood that can be classified into three broad categories: impairments in social communication and reciprocity; speech and language delay; stereotypic movements and circumscribed interests (American Psychiatric Association, 1994; Volkmar et al., 2004). It is often felt that social disability is at the core of the pathophysiology of these disorders. Despite this, individuals with autism spectrum disorders can vary greatly in the extent of their symptoms.
Although facial dysmorphology or other somatic features are present in rare forms of ASD, the more common non-syndromic form of these disorders does not appear to have grossly altered phenotypic traits. However, in his seminal description of autism in 1943, Kanner noted that a large head circumference was a common finding in children with autism (Kanner, 1943). Subsequent studies have identified macrocephaly, which is a head circumference greater than 2 standard deviations above the population mean, to be present in approximately 20% of individuals with ASD (Davidovitch, Patterson & Gartside, 1996; Woodhouse et al., 1996; Fombonne et al., 1999; Miles, Hadden, Takahashi & Hillman, 2000). More recently, systematic studies have found that an abnormally elevated increase in head growth (assessed by cranial circumference) is present in a majority of children with ASD. Intriguingly, the larger head circumference is not present at birth, but develops within the first two years of life (Lainhart et al., 1997; Courchesne, Carper & Akshoomoff, 2003; Dementieva et al., 2005; Redcay & Courchesne, 2005), more specifically, at around the age of 4 months (Courchesne & Pierce, 2005, Redcay & Courchesne, 2005;Gillberg, 2002 #234}). In a meta-analytic study, Redcay and Courchesne (2005) have established a nonlinear dynamic of brain size fluctuation, with its relative proportion in individuals with autism changing from 13% smaller at birth, to 10% larger at 1 year of age, to 2% larger in adolescence when compared to control individuals. When the rates of change in head circumferences were compared among children with ASD and children from a normative sample, it was found that the rates were accelerated in the autism sample from birth to 12 months, but were no different from the normative sample from 12 to 36 months (Dawson et al., 2007). Similarly, a study carried out on a large sample of high-risk infants for the manifestation of ASD (the Baby Sibs Research Consortium), revealed increased growth velocity and decreased acceleration of head circumferences associated with ASD (Hodapp & Urbano, 2007). Specifically, though having started at approximately the same head circumferences, growth curves for children with ASD and for non-diagnosed children diverged early in the first year of life and remained divergent until about 28 months. Also, of note is the observation that children with autism have been reported to have increased body length and body weight as well. However, several studies, both retrospective and prospective, comparing individual growth curves from birth to 2–3 years, concluded that head circumference increases in autism were greater than expected for corresponding body height (Lainhart et al., 2006; Szatmari et al., 2007; Fukumoto et al., 2008). Hence, macrocephaly in autism cannot be attributed exclusively to an increase in body growth.
Macrocephaly in typically-developing children is most often attributable to an increased volume of cerebral spinal fluid with no associated brain volume increase. In contrast, macrocephaly in children with autism is due to an accelerated brain growth in the first months of life. Postmortem studies have found a larger total brain weight in individuals with autism (Palmen, van Engeland, Hof & Schmitz, 2004). Structural magnetic resonance imaging (MRI) studies have compared a large number of individuals with autism, of many different ages, to both typically-developing controls or non-autistic individuals with mental retardation (Piven et al., 1995; Piven, Arndt, Bailey & Andreasen, 1996; Courchesne et al., 2001; Aylward et al., 2002; Carper, Moses, Tigue & Courchesne, 2002; Boddaert et al., 2004; Courchesne, 2004; Herbert et al., 2004; Palmen & van Engeland, 2004; Carper & Courchesne, 2005; Hazlett et al., 2005; Palmen et al., 2005; Schultz et al., 2005; Hazlett et al., 2006); brain volume was found to be consistently enlarged. Furthermore, when the cerebrum was segmented into gray and white matter compartments, both were found to be enlarged, and the white matter compartment somewhat more enlarged (Courchesne et al., 2001; Herbert et al., 2003).
Significantly, at younger ages (2–3 years), the differences between individuals with ASD and controls seem most dramatic (Hazlett et al., 2005). After age 5, the volume differences between individuals with ASD and the comparison groups decline, concurrent with a deceleration of head growth in individuals with autism (Courchesne et al., 2001; Carper et al., 2002; Courchesne et al., 2003; Carper & Courchesne, 2005; Hazlett et al., 2005; Schultz et al., 2005; Hazlett et al., 2006). These findings suggest a substantially altered developmental trajectory of the cerebrum of individuals with autism, and therefore warrant further investigation toward its causes and potential clinical significance. For example, could this increased rate of brain growth reflect an underlying altered neural structure responsible for altered information processing, a compensatory process, or rather an irrelevant epiphenomenon?
Increases in cortical gray matter could be due to increases in neurogenesis resulting in larger numbers of neurons, or to increased gliogenesis, resulting in larger numbers of glia (McCaffery & Deutsch, 2005). The cerebral cortex is assembled through the contribution of two germinal layers, the neuroepithelium abutting the ventricular lumen, the ventricular zone (VZ), and the neuroepithelium just superficial to this, the subventricular zone (SVZ) (Noctor, Martinez-Cerdeno & Kriegstein, 2007). During embryogenesis, radial glial cells divide in the VZ, generating either neurons that directly migrate to the cortex, or intermediate progenitors that migrate to the SVZ (Figure 1a). The progenitors in the SVZ undergo a limited number of divisions producing cortical neurons. This number is likely to be genetically encoded and species-specific, as the larger the number of these neurogenic divisions, the greater the number of neurons generated and the greater the volume expansion of developing cortex (Noctor, Martinez-Cerdeno, Ivic & Kriegstein, 2004; Kriegstein, Noctor & Martinez-Cerdeno, 2006) (Figure 1b). Radial glia transform into astrocytes at the end of neurogenesis, by retracting their apical process and translocating to the cortex, and disappear shortly after birth (Levitt, Cooper & Rakic, 1981; Rakic, 2003). The processes responsible for the involution of radial glia and therefore the ending of neurogenesis are not precisely known, although a combination of increased differentiation into SVZ progenitors, increased transformation into astrocytes, and possibly cell death, are all likely to be contributing factors (Cameron & Rakic, 1991). In contrast, the SVZ persists postnatally, producing astrocytes and oligodendrocytes (Marshall & Goldman, 2002; Marshall, Suzuki & Goldman, 2003). This region remains a prominent source of neurons for the olfactory bulb throughout life, whereas its ability to generate cortical neurons becomes progressively limited, such as few cortical neurons are generated in the postnatal period (Ganat et al., 2006) and none in the adult mammalian cortex (Kornack & Rakic, 2001). The genetic and molecular events drive a neuron-to-glia switch in fate in the SVZ have been the source of recent studies (Vaccarino et al., 2007).
To better understand the cellular and molecular mechanisms underlying increased brain size in ASD, we must consider that the cerebral cortex is by far the largest structure in the primate brain, and that two types of neurons populate the cerebral cortex, the excitatory pyramidal cells, which secrete glutamate and send myelinated axons outside the cortical mantle, and the inhibitory interneurons, which secrete GABA and are intrinsic to the cortex (Jones & Peters, 1984). These neurons are organized into columns each comprising a core of excitatory neurons and a surrounding area of inhibition (Mountcastle, 1997). Neurons within each column are functionally related, that is they have similar response properties and represent a feature of sensory representation or analysis (i.e., left or right eye; orientation; color; position in space). An anatomical minicolumn has been defined as a core of pyramidal neurons ascending vertically through layers VI and II, surrounded by a peripheral neural space rich in dendrite arborizations and synapses and lined by GABAergic interneurons, thought to prevent the inappropriate spread of excitation (Peters, Cifuentes & Sethares, 1997).
The anatomically identifiable minicolumns may represent a mechanism to integrate or segregate inputs and outputs of a group of cells. These pyramidal cell modules are conserved in the mammalian kingdom and likely correspond to the functional minicolumns, identified by Mountclastle as the smallest units of cortical information processing (Mountcastle, 1997). Intriguingly, post mortem analyses of the brains of autistic individuals revealed an increased number of anatomically identified cortical minicolumns with a higher packing density and no alteration in intra-column cell density (Casanova, Buxhoeveden, Switala & Roy, 2002; Buxhoeveden et al., 2006; Casanova et al., 2006). This increase in column number and density suggests that the number of excitatory neurons forming minicolumns may be increased, the inhibitory neurons that surround the minicolums may be decreased, or both processes may be occurring at the same time.
Both VZ and SVZ produce the excitatory pyramidal neurons that form the basic minicolumn scaffolding of the cortex. An increase in the pool of progenitors in the VZ or SVZ could result in larger numbers of excitatory neurons, and, arising from these, minicolumns. In lower mammals, the inhibitory neurons surrounding the minicolums arise almost exclusively from basal regions of the embryonic telencephalon, the medial and lateral ganglionic eminences. These neurons reach the cerebral cortex by a separate process of tangential migration (de Carlos, Lopez-Mascaraque & Valverde, 1996; Lavdas, Grigoriou, Pachnis & Parnavelas, 1999). In primates, however, these inhibitory neurons arise both from the ganglionic eminences and from the dorsal SVZ, from which they reach the cortex by radial migration (Letinic, Zoncu & Rakic, 2002). An absolute or relative decrease in inhibitory neuron production, migration, or survival may also be involved in the pathogenesis of autism, as it would produce a decreased distance between minicolumns and also explain the increased incidence of seizures in these children.
Although it is tempting to speculate that the increase in gray matter volume observed in neuroimaging studies in ASD may reflect an underlying increase in excitatory (pyramidal) neuron number, implicating the prenatal developmental mechanisms described above, it is important to note that we cannot exclude a postnatal trajectory in ASD pathogenesis. In particular, reduced pruning of connections between neurons, or decreased programmed cell death (McCaffery & Deutsch, 2005) may also lead to increased gray matter volume, although these mechanims would be unlikely to generate an altered columnar structure.
Increases in white matter such as those detected by MRI in the subcortical regions of the neocortex and cerebellum of adolescent patients with ASD (Courchesne et al., 2001) could reflect increases in oligodendrocytes, cells that ensheath axonal projections with myelin. However, MRI studies examining T2 relaxation have not detected increases in myelin density that would accompany such an excess of oligodendrocytes (Hendry et al., 2006). Alternatively, increases in white matter could reflect an increased neuron pyramidal neuron number, since these neurons are those that send myelinated axons outside the cortex. Thus, an increased number of cortical pyramidal neurons could explain both the increased gray and white matter volumes detected in neuroimaging studies in ASD, as well as the increased density of minicolumns.
Allometric models of cerebral development suggest that doubling the number of neurons supports a four-fold increase in neuronal connectivity, potentially generating a four-fold increase in white matter (Ringo, 1991). Thus, it can be predicted that subtle differences in neuron number, and therefore gray matter, could result in a larger, more-measurable difference in white matter volume (Casanova, 2004). Since the majority of myelination occurs postnatally, neonates born with increased numbers of pyramidal neurons and hence of minicolumns may have a normal brain size at birth, which will undergo accelerated growth postnatally due to the increased connectivity between minicolumns and the myelination of these connections, reflected in increased white matter volume. This hypothesis also implies that the primary pathogenic event leading to an increased number of minicolumns may substantially precede the manifestation of increased brain volume.
A recent imaging study (Herbert et al., 2004) has suggested that the WM increase in autism involves the outer, radiate compartment, which is composed of intrahemispheric cortico-cortical connections; indeed, several studies have shown that long-tract fibers such as the corpus callosum and internal capsule are not altered in ASD (Piven, Bailey, Ranson & Arndt, 1997; Manes et al., 1999; Hardan, Minshew & Keshavan, 2000). This important clue further suggests that the exuberant pyramidal neurons in ASD may be those that populate the upper cortical layers, which are those neurons that mostly form intrahemispheric and cortico-cortical connections. The formation of upper-layer cortical neurons occurs at the later stages of neurogenesis (Frantz & McConnell, 1996; Desai & McConnell, 2000) (Figure 2). An increase in later-developing upper layers pyramidal neurons and in their intrahemispheric connections could account for the larger proportion of subcortical radiate white matter and the relative decrease in long-range connections found in children with ASD (Herbert et al., 2003; Courchesne & Pierce, 2005).
Family and twin studies have determined that ASD are genetically complex, with some variation in expression due to environmental influences (Folstein & Rutter, 1977; Steffenburg et al., 1989; Bailey et al., 1995; Santangelo & Tsatsanis, 2005). Family studies of macrocephaly in autism suggest that this trait is heritable and that explorations of genetic factors affecting cell growth are warranted (Fidler, Bailey & Smalley, 2000; McCaffery & Deutsch, 2005). While a small percentage of ASD are inherited as monogenetic disorders, the majority of cases are due to the concurrent inheritance of several risk-conferring genetic variants (Santangelo & Tsatsanis, 2005). Although multiple genome-wide scans have been conducted, results generally have been inconsistent. Chromosomal regions 7q and 10p are supported by meta-analytic (Trikalinos et al., 2006) and high-resolution scanning studies carried by the Autism Genome Project (AGP) Consortium, respectively (Szatmari et al., 2007), but other regions may confer susceptibility to ASD.
Multiple candidate genes, including those controlling the growth of the cerebellum (Gharani et al., 2004) and glutamatergic and GABAergic synaptogenesis (Szatmari et al., 2007) also have been proposed to play a role. For example, the development or maintenance of synaptic connections may be affected in autism, and indeed, rare mutations have been reported in ASD for both neuroligins (Jamain et al., 2003; Laumonnier et al., 2004), the associated neurexins (Feng et al., 2006) and SHANK (Durand et al., 2007), molecules which are associated with both excitatory and inhibitory synapses.
Altered regulation of genes involved in the development of the GABAergic system is also likely to cause a dysequilibrium of the glutamate/GABA neuronal systems. Interestingly, the GABA receptor subunits, GABRB3, is strongly implicated in ASD by multiple association studies and is reduced in expression in human cerebral samples of ASD, particularly Rett and Angelman syndromes (Samaco, Hogart & LaSalle, 2005). Furthermore, a recent genetic study of over 700 ASD families has implicated the MET tyrosine kinase, a gene product that regulates the migration of GABA interneurons, with the disorder (Campbell et al., 2006).
In accordance with the hypothesis that genes regulating neuronal number are important candidates for study in ASD, mutations in PTEN, a gene product that down-regulates the phosphatidylinositol 3-kinase (PI3K) pathway and is involved in cell cycle regulation, have been found in a portion of autistic subjects with macrocephaly (Goffin et al., 2001; Butler et al., 2005). The PI3K pathway increases precursor cell proliferation and survival and is thought to be regulated in part by FGF receptor signaling (Ong et al., 2001). Recently, an animal model for PTEN mutations was developed which showed macrocephaly and social phenotypes reminiscent of autism (Kwon et al., 2006).
In appreciation of this mosaic of findings, the consensus hypothesis is that the etiology of ASD is predominantly oligogenic and likely includes gene–gene and gene–environment interactions. These observations nevertheless suggest that in autism there may be convergence of genetic risk factors to pathogenetic pathways affecting developmental processes that regulate the balance of excitatory/inhibitory neurons and cortical architecture (Persico & Bourgeron, 2006).
A possible cause for the excess of glutamate-containing excitatory neurons and the minicolumn pathology in some ASD patients may be represented by variations in expression or function in the genes coding for Fibroblast Growth Factors (FGF). FGF-mediated signaling regulates the number of excitatory neurons and the growth of the cerebral cortex (Vaccarino, Schwartz, Hartigan & Leckman, 1995; Ortega et al., 1998; Vaccarino et al., 1999b; Vaccarino et al., 1999a) and may therefore be implicated in the etiology of autism. This hypothesis arises from several lines of accumulated evidence (McCaffery & Deutsch, 2005), and specifically animal models developed in several laboratories.
The receptors for FGFs are present on neural stem cells and cortical neuron progenitors in vitro and in vivo (el-Husseini, Paterson & Shiu, 1994; Wuechner et al., 1996; Wilke, Gubbels, Schwartz & Richman, 1997; Vaccarino et al., 1999b; Vaccarino et al., 1999a; Blak et al., 2005). FGF binding to these receptors maintains neuronal progenitors in a proliferative state and therefore promotes their expansion (Vaccarino et al., 1995; Vaccarino et al., 1999a; Tanaka et al., 2001; Ohkubo et al., 2004). One of the FGF ligands, Fgf2, increases the genesis of cortical pyramidal neurons (Vaccarino et al., 1995; Vaccarino et al., 1999b) and of granule neurons in cerebellum and hippocampus (Tao, Black & DiCicco-Bloom, 1996; Wagner, Black & DiCicco-Bloom, 1999; Cheng, Black & DiCicco-Bloom, 2002). A single intraventricular injection of Fgf2 in rat embryos increases the genesis of excitatory pyramidal neurons in the neocortex, leading to a permanent increase in the number of cortical neurons and astroglial cells and enlarged cortical size (Vaccarino et al., 1999b) (Figure 3). In contrast, disruption of the Fgf2 gene decreases cell proliferation in the VZ during development and leads to a loss of cortical pyramidal neurons and astrocytes, without any change in the number of inhibitory neurons (Vaccarino et al., 1999b; Raballo et al., 2000). Significantly, the excitatory neuron loss is more prominent in anterior cortical regions (Korada et al., 2002).
There are 4 FGF receptor genes, Fgfr1-Fgfr4, of which Fgfr1, Fgfr2, Fgfr3 are expressed in the developing brain. Fgfr1 is enriched in dorsomedial regions, which include the hippocampal primordium. Consistently, the conditional deletion of the Fgfr1 gene from radial glial cells results in lifelong hippocampal atrophy (Ohkubo et al., 2004) and altered midline connectivity (Smith et al., 2006) but no significant alterations in cerebral cortical development.
Functionally disrupting all Fgf receptors during brain development by overexpressing an inactive Fgf receptor does disrupt cerebral cortical development. Inactive Fgf receptor overexpression results in a smaller neocortex with loss of glutamatergic pyramidal neurons in both frontal and temporal regions, emphasizing the importance of global Fgf signaling for excitatory neuron development (Shin et al., 2004). Recent data suggest that Fgfr2 may be the crucial receptor in this respect. The selective deletion of Fgfr2 in radial glial progenitor cells results in decreased cortical volume and decreased pyramidal cell number and density in the neocortex, more pronounced in medial prefrontal (mPF) cortical regions; furthermore, the volume of subcortical white matter is dramatically reduced due to decreased number of axonal fibers (unpublished data). To visualize cortical minicolumns in the Fgfr2 mutants, we immunostained the apical dendritic bundles with an antibody to the microtubule-associated protein MAP2. The data show a decrease in density of minicolumns in both mPF and cingulate cortical regions, consistent with the decrease in density of pyramidal cells (Figure 4).
Thus, the deletion of Fgfr2 in radial glial cells mimics in part the phenotype of Fgf2 null mice, in that the Fgfr2 mutants have a concomitant decrease in cortical volume, excitatory neurons, and astrocytes, with preservation of cortical layer architecture.
Mutations in Fgf17, a ligand for Fgfr2, also result in reduced mPF cortical areas (Cholfin & Rubenstein, 2007). Furthermore, mice lacking Fgf17 have deficits in social interactions without changes in cognitive abilities (Scearce-Levie et al., 2007). Mouse pups lacking Fgf17 have fewer ultrasonic vocalizations (an important component of mother-infant social interaction in rodents) when separated from their mothers. In adult, opposite-sex pairs of Fgf17 mutant mice spend less time interacting than opposite-sex pairs of control mice (Scearce-Levie et al., 2007). These social deficits, which mimic autistic features, may be attributed to the decreased activation of the prefrontal cortex in mutant mice. In conclusion, converging evidence suggest that FGF signaling may be controlling the modular (columnar) expansion of the cortex in mammals. Therefore, FGF ligands and receptors are an important class of candidate genes for mediating the accelerated brain growth evident in ASD.
The human FGF8 (10q24) and FGFR2 (10q26) genes are in proximity to previously identified broad autism linkage peaks on chromosome 10q, supporting the plausibility that these genes may be contributing risk for autism. As noted above FGFR2 binds FGF2, FGF17 and FGF8, which interact with the D3 immunoglobulin-like domains and the D2-D3 linker region of this receptor (Olsen et al., 2003). Like Fgf2, the Fgf8 gene regulates cortical volume (Meyers, Lewandoski & Martin, 1998) and, like Fgf17, plays a role in the patterning of neocortical areas and in the development of intracortical connections (Huffman, Garel & Rubenstein, 2004). These data are intriguing, considering that a “disconnection” between prefrontal cortex and other cortical areas is thought to underlie the social disability in ASD patients (Geschwind & Levitt, 2007).
In line with the genetic heterogeneity of ASD, there are likely to be many genetic determinants of the excitatory/inhibitory imbalance in the brain. The contributions of an exuberance of pyramidal neurons is in accordance with some of the symptoms and difficulties of patients with ASD. The excitatory overactivity may cause disruptions in cortical rhythms and seizures. Interestingly, seizures are found in approximately 30% of children with ASD, indicating that cortical development must be disrupted in ways that eventually alter the excitatory/inhibitory balance. Depending on the regional extent of this phenomenon, exuberant pyramidal neuron number and connectivity may cause noisy neuronal processing, or impaired long-range interactions among cortical areas and cortical-subcortical regions, which in turn may lead to diverse symptoms such as decreased attentional focus to socially relevant situations, inability to process global features, language deficit, perseveration, and poor impulse control. Future research will reveal whether the exuberance in neuron number is more prominent in certain regions of the cortex and how it may lead to different forms of autism.
In conclusion, in autistic children, the balance between growth and regression critically important to brain development may be biased in favor of growth, particularly in the first year of life. A dysregulation in amount or timing of FGF gene expression may conceivably be responsible in some patients for aberrantly regulated brain growth and altered ratio of different neuronal subtypes, ultimately resulting in altered cortical modularity and connectivity. A clarification of their potential role in this syndrome may lay the foundations for understanding how aberrant brain growth could be treated or reversed.
This work was supported by NIH grants MH067715, Autism Speaks and the NARSAD Foundation. We thank Shawna Ellis for technical assistance and all members of the Vaccarino lab for helpful discussions.