Developmental dyslexia is a common specific childhood learning disorder with a strong heritable component. Previous studies using different genetic approaches have identified several genetic loci and candidate genes for dyslexia. In this article, we have integrated the current knowledge on 14 dyslexia candidate genes suggested by cytogenetic findings, linkage and association studies. We found that 10 of the 14 dyslexia candidate genes (ROBO1, KIAA0319, KIAA0319L, S100B, DOCK4, FMR1, DIP2A, GTF2I, DYX1C1 and DCDC2) fit into a theoretical molecular network involved in neuronal migration and neurite outgrowth. Based on this, we also propose three novel dyslexia candidate genes (SLIT2, HMGB1 and VAPA) from known linkage regions, and we discuss the possible involvement of genes emerging from the two reported genome-wide association studies for reading impairment-related phenotypes in the identified network.
Developmental dyslexia (or specific reading disability) is defined as a specific and significant impairment in reading ability that cannot be explained by deficits in either intelligence, learning opportunity, motivation or sensory acuity. Dyslexia is the most common childhood learning disorder.1, 2, 3
The prevalence of dyslexia ranges from 3.6% in primary school children in The Netherlands4 to as high as 10% in US school-aged children,5 which is likely due to the fact that many different psychometric tests and diagnostic criteria for dyslexia exist (Box 1 ).2, 3, 6 Twice as many boys as girls are affected with dyslexia.4, 7, 8
Results from functional magnetic resonance imaging studies and studies using other functional brain imaging modalities (such as event-related potentials and magnetoencephalography), indicate that three regions in the left brain hemisphere are important for fluent reading (for recent reviews, see Shaywitz and Shaywitz9 and Hoskyn10). These regions are an ‘anterior’ system, that is the inferior frontal gyrus (Broca's area), and two ‘posterior’ systems, the dorsal parietotemporal and ventral occipitotemporal systems. Functional brain imaging studies have repeatedly demonstrated that dyslexic readers are characterized by a so-called ‘neural signature,’ in that the anterior system of their left brain hemisphere is slightly overactivated during reading tasks whereas, in contrast, the two posterior systems are underactivated.9, 10
Evidence from family and twin studies shows that dyslexia is a highly heritable disorder, and up to 75% of the phenotypic variance can be explained by genetic factors.2, 3, 4 Dyslexia behaves as a multifactorial (or complex) disorder, in which combinations of genetic and environmental factors contribute to disease risk. This article will concentrate on the genetic factors involved in dyslexia etiology. The genetic model underlying most cases of dyslexia is likely one in which several to multiple genetic factors, all of small or moderate individual effect size, contribute to disease risk.3, 5 In the current study, we reviewed the available evidence for dyslexia loci and genes, including linkage studies (candidate gene based) association studies and reports of chromosomal aberrations cosegregating with dyslexia. We found that 10 of the 14 dyslexia candidate genes that were suggested by the literature had been shown to directly or indirectly interact and could be integrated into a molecular signaling network contributing to dyslexia etiology, responsible for regulating neuronal migration and neurite outgrowth. Our work builds on findings of other researchers in the field of dyslexia genetics,5, 11, 12, 13 as well as existing knowledge about genes involved in neuronal migration disorders and the KEGG pathway for axon guidance (http://www.genome.jp/dbget-bin/show_pathway?map043600). Moreover, based on the signaling network, we propose three novel dyslexia candidate genes from linkage regions.
Several chromosomal aberrations have been reported to cosegregate with dyslexia. As reported above, the DYX1C1 and ROBO1 genes in the DYX1 and DYX5 loci, respectively, were identified through their position in the breakpoints of translocations in dyslexic individuals.
Our own group recently reported PCNT, DIP2A, S100B and PRMT2 to be (partially) deleted in a small deletion of chromosome band 21q22.3 that cosegregated with dyslexia in a father and his three affected sons.56 Two of these genes seem valid candidate genes for dyslexia.
DIP2A encodes the ‘disco-interacting protein 2 homolog A’ (DIP2A).95 (In our original report,56 based on incorrect gene annotation information, we identified this gene as encoding the ‘DLX-interacting protein 2’, which is an alternatively spliced human variant of glutamate receptor interacting protein 1.101) The Drosophila homolog of DIP2A is involved in neuronal connectivity in the visual system.102 In humans, DIP2A is a nuclear protein that is ubiquitously expressed. In a complex with the transcription factors DMAP1,95, 103 DNMT1 and HDAC2,103 it negatively regulates neurite outgrowth104 and synaptic plasticity.105 Synaptic plasticity is crucial for cognitive processes such as learning and memory,105 which seem partially impaired in dyslexia (see above).63
The S100B gene is mainly and highly expressed in the brain (especially in the hippocampus).106 S100B encodes a calcium-binding peptide that is mainly produced by astrocytes and that exerts paracrine and autocrine effects on neurons and glial cells. S100B is known to bind to the receptor for advanced glycation end products (RAGE) protein in (hippocampal) neurons, which activates a signaling pathway resulting in neurite outgrowth.107 The protein has been reported to be involved in modulating synaptic plasticity,108 although the mechanism by which this is brought about is currently unknown.106 The concentration of the S100B peptide is increased in blood and cerebrospinal fluid in various clinical brain conditions, including psychiatric disorders such as schizophrenia and Tourette syndrome.106
In a very recent study, a microdeletion of the DOCK4 gene was found to cosegregate—though imperfectly—with dyslexia in individuals from two independent families in a study on rare structural DNA variants predisposing to autism and/or reading impairment/dyslexia.109
A deletion of chromosome 7q11.23 results in Williams–Beuren syndrome (WBS), a human neurodevelopmental syndrome that has multisystemic manifestations including mild-to-moderate mental retardation and cognitive deficits. WBS has also been associated with reading difficulties and ‘full’ dyslexia.112, 113, 114 The deletion of GTF2I as part of the WBS deletion is responsible for the rather typical neurocognitive profile associated with WBS and, more specifically, for the visuospatial construction deficits that are very often seen in individuals with WBS.115, 116 Visuospatial construction deficits are also found in people with dyslexia.117 Another indication that GTF2I is involved in the etiology of language-related disorders is the fact that in a recent study of people with a 7q11.23 microduplication syndrome involving GTF2I, all 13 subjects investigated in the study also had some degree of specific language impairment,118 a disorder that is comorbid with dyslexia.8 In these individuals, the expression of only two genes in the duplicated region was upregulated, that is GTF2I and CYLN2.118 Interestingly, mice with haploinsufficiency for Cyln2, the other gene that was upregulated in the duplicated region, have features of WBS, including mild growth deficiency and deficits in motor coordination.119 GTF2I is highly expressed in the brain, where it has a role in normal brain development. It encodes the transcription factor TFII-I.44 As indicated above, TFII-I binds to the promoter of the DYX1C1 gene at a binding site containing rs3743205,44 the SNP associated with dyslexia.43, 50, 51, 53, 54 TFII-I acts as an enhancer or suppressor of the transcription of DYX1C1 depending on the context of other transcription factors.44
From single candidate genes to an integrated molecular network
As discussed earlier, 14 genes from the literature had (at least some) evidence for involvement in the etiology of dyslexia, that is DYX1C1, DCDC2, KIAA0319, C2ORF3, MRPL19, ROBO1, KIAA0319L, FMR1, PCNT, DIP2A, S100B, PRMT2, DOCK4 and GTF2I. An analysis with the GOstat bioinformatics tool (http://gostat.wehi.edu.au/)120 showed that after correction for multiple testing, 6 of the 10 most significantly enriched gene ontology terms in these 14 candidate genes are related to nervous system development in general and/or neurite outgrowth in particular, that is generation of neurons, neurogenesis, axonogenesis, neuron projection morphogenesis, neuron projection development and neuron development (Table 4 ). Whereas three genes from the set carried gene ontology terms implicating them in neurite outgrowth (that is DCDC2, ROBO1 and S100B), it was clear from the above review that more genes could be linked to this process and the related neurodevelopmental process of neuronal migration.
Therefore, we systematically searched PubMed (http://www.ncbi.nlm.nih.gov/sites/entrez) and the Uniprot Protein Knowledgebase (http://www.uniprot.org/uniprot)95 for the (proposed) function of the proteins that are encoded by the 14 dyslexia candidate genes that were described above.
As shown and described in (the legends of) Figures 1 and 2, 10 of the 14 proposed dyslexia candidate genes (that is ROBO1, KIAA0319, KIAA0319L, S100B, DOCK4, FMR1, DIP2A, GTF2I, DYX1C1 and DCDC2) fit into a neurodevelopmental and theoretical molecular network. All the corresponding proteins could be directly or indirectly linked to processes regulating and modulating cytoskeletal microtubules and actin filaments that are involved in neuronal migration and/or the directed outgrowth of neurites/axons. The proposed network therefore fits and extends the currently dominating hypothesis of dyslexia as a neuronal migration disorder.5, 11, 12, 13, 57, 58, 59, 65, 75, 81, 82, 83
In addition to the 14 candidate genes described above, the delineation of the molecular network enabled us to propose three additional dyslexia candidate genes from linkage regions in which no candidates had been described thus far.
In the DYX6 locus on 18p11.2, marker D18S464—which showed genetic linkage with (single word) reading in two dyslexia linkage studies2, 31 (Table 1)—is located immediately downstream from the VAPA gene, in a linkage disequilibrium block that contains part of the VAPA gene (http://www.broad.mit.edu/mpg/haploview/). The VAPA gene is highly expressed in the brain.
The VAPA protein (other names: VAP-33 or VAMP-associated protein A) is an integral membrane protein found in the (neuronal) cell membrane, membranes of cellular organelles (such as the Golgi apparatus) and associated with microtubules.121, 122 VAPA was reported to have an important role in stimulating neurite outgrowth.123 In addition, it is involved in synaptic plasticity124 and necessary for vesicular neurotransmission at the synapse.121, 122, 125 Like DYX1C1,61 VAPA binds and forms a complex with HSP90, a protein with many functions including a role in intracellular vesicle transport.122 Recently, genetic association between several SNPs in VAPA and bipolar disorder has been reported.125
Suggestive genetic linkage of ‘irregular word spelling’ near marker D4S2633 on chromosome 4p15.32 was reported in the dyslexia linkage study by Bates et al.31 (Table 1). The gene nearest to this marker is SLIT2, which maps ∼1.3 Mb proximally. SLIT2 encodes a ligand of the ROBO family of axonal guidance receptors of which the dyslexia candidate ROBO1 is a member.86, 126
In the study by Igo et al.,26 linkage with ‘single word reading efficiency’ was reported for marker ATA5A09 on 13q12.3 (Table 1). ATA5A09 is located within the HMGB1 gene (http://www.ensembl.org/homo_sapiens). Like S100B, the encoded protein (HMGB1 or Amphoterin) binds to the RAGE receptor and is involved in neurite outgrowth.107, 127
HMGB1 can also directly activate the ETS1 transcription factor,128 which upregulates another SLIT gene, that is SLIT3.129 In addition, HMGB1 enhances the ESR1-induced expression of genes containing estrogen response elements130 (see below).
In Figures 1 and 2, we have used a color code indicating the level of ‘evidence’ that implicates the above-described/proposed dyslexia candidates in the etiology of dyslexia. Proteins with relatively robust evidence for involvement in dyslexia are indicated in orange. These proteins are encoded by genes that were disrupted by translocation breakpoints in dyslexic individuals (DYX1C1 and ROBO1) and/or found associated with dyslexia, requiring repeated association findings for those genes not implicated by translocations (DYX1C1, DCDC2, KIAA0319 and ROBO1). Indicated in yellow are proteins with less robust evidence of their involvement in dyslexia. These proteins are encoded by genes that were repeatedly found to be deleted in people with dyslexia (PCNT, DIP2A, S100B, PRMT2, GTF2I and DOCK4), genes that showed (nominal) association with dyslexia in only one study (C2ORF3, MRPL19 and KIAA0319L) and FMR1 that was mutated in people with dyslexia. The proteins indicated in purple are encoded by the three novel dyslexia candidate genes we propose based on their location in reported dyslexia linkage regions (VAPA, SLIT2 and HMGB1).
Signaling through the network can be initiated at the neuronal cell membrane by the interacting SLIT and netrin axonal guidance pathways, which regulate the direction and rate of axon/neurite outgrowth (Figures 1a and 2). The molecular guidance ‘cues’ SLIT (including the dyslexia candidate SLIT2) and Netrin (mainly the Netrin-1 protein) bind to ROBO1 and DCC, respectively.86, 126 As shown and described in more detail in (the legend of) Figure 2, the rate and direction of outgrowth is critically dependent on the exact receptor composition in the plasma membrane. As they have fibronectin type III domains very similar to those of the DCC receptor protein (http://www.ensembl.org/homo_sapiens), the KIAA0319 and KIAA0319L proteins may function as receptors for netrin-like molecules. However, these interactions are hypothetical and have not yet been confirmed by experimental data, and therefore are not shown in Figures 1 and 2. Moreover, as suggested by others, KIAA0319,11, 13, 82 and possibly KIAA0319L, might alternatively/additionally function in the direct adhesion between neurons and glial fibers during neuronal migration through their polycystic kidney disease domains.
As indicated in Figure 2, the activation of the ROBO1 and DCC receptors leads to signaling via CDC42131 and cytoplasmic ERK1 and ERK2 proteins132 resulting in changes in the cytoskeletal organization of microtubules and actin filaments and (eventually) in neurite outgrowth.86, 126, 133, 134 Two other dyslexia candidates, S100B and HMGB1, can also result in CDC42135 and ERK1/2132 activation and neurite outgrowth by binding to RAGE106, 127 (Figure 1b). In addition, S100B or HMGB1 binding to RAGE also activates RAC1136, 137 (Figure 1b). This latter pathway is also activated by DOCK4, a peripheral membrane protein95 that regulates dendritic growth and branching of hippocampal neurons110 (Figure 1b). In the cytoplasm of hippocampal neurons, FMR1 also activates ERK1 and ERK292 (Figures 1d and e). Moreover, FMR1 is involved in the translation of the mRNA of RAC196 (Figure 1d).
Activated ERK1 and ERK2 activate/phosphorylate the NF-κB (nuclear factor kappa-B),138 ETS1,139 ELK1140 and TFII-I141 transcription factors (Figure 1c). NF-κB is also activated downstream of RAC1.136 All four transcription factors seem to have a role in the dyslexia neurodevelopmental network. Activated NF-κB upregulates the transcription of the Netrin-1142 and HSP90143 genes. It also upregulates the transcription of the ELK1 gene,144 a transcriptional activator of DYX1C1.43 ELK1 is a brain-expressed transcription factor, and its activation has been associated with neurite outgrowth and hippocampus-dependent learning in rats.43 The ETS1 transcription factor regulates the transcription and expression of the SLIT3 gene.129 Activated (nuclear) TFII-I141 upregulates or downregulates the transcription and expression of the DYX1C1 gene, depending on the context of other transcription factors.43, 44
The nuclear protein DIP2A binds DMAP1,102 which is a part of the DMAP1–DNMT1–HDAC2 repressive transcription complex103 (Figure 1c). HDAC2—through which the DMAP1–DNMT1–HDAC2 complex negatively regulates neurite outgrowth104—is involved in repressing the transcriptional activity of ELK1,95 which results in a reduced expression of DYX1C1. HDAC2 also directly inhibits ESR1, the nuclear estrogen receptor α.145
The nuclear ESR1 and ESR2 proteins—when activated—upregulate the expression of genes containing estrogen response elements by directly binding to the DNA sequences containing these elements95 (Figure 1c). This process—that is important in brain development and neuronal processes such as neuronal migration and synaptic plasticity62 (see above)—is enhanced—for ESR1—by HMGB1130 and suppressed—for both ESR1 and ESR2—by DYX1C1.62 DYX1C1 also binds directly to nuclear ESR1 and ESR2, which promotes their proteosomal degradation and hence negatively regulates the function of these proteins.62
In addition to their role in the cell nucleus, ESR1 and ESR2 are also found in the cytoplasm of hippocampal neurites, where they are involved in activating ERK1/262, 146 (Figure 1d). The ESR–DYX1C1 complexes that promote the proteosomal degradation of ESR1 and ESR2 are also found in the cytoplasm of hippocampal neurites (Figure 1d), which suggests the involvement of DYX1C1 in negatively regulating estrogen signaling cascades in these neurites.62 Moreover, as both DYX1C161 and VAPA122 bind to HSP90, this makes one speculate that a complex of VAPA, HSP90 and DYX1C1 would be involved in degrading and negatively regulating cytoplasmic ESR1 and ESR2 (Figure 1d).
After being activated by any of the above-described cascades, the ERK1/2 proteins also regulate the activity of the DCDC2 protein by phosphorylating its DCX domains80 (Figure 1e). DCDC2 and FLNA—a protein that is encoded by a gene interacting with FMR197 (see above)—directly bind and modulate neuronal microtubules and actin filaments that are involved in neurite outgrowth and neuronal migration11, 12, 13, 65, 69, 78, 79, 80, 81, 95, 98, 99, 100 (Figure 1e).
As already stated above, two GWAS related to reading disability have been reported to date. The first was a GWAS using a quantitative trait loci approach comparing the two extremes of the reading ability spectrum of the normal population in 1502 children from the UK Twins Early Development Study TEDS.41 The authors used a pooled design and SNP arrays of only 100.000 SNPs, which greatly reduced the power of the study. Nevertheless, they nominally replicated some of their findings in 4258 additional samples from TEDS, resulting in 10 SNPs of interest for reading ability in early childhood.41 Six of these SNPs are located in introns of protein-coding genes (http://www.ensembl.org/homo_sapiens), three of which could be directly linked to our molecular network for dyslexia: CDC42BPA (rs1320490; P=0.030), TIAM1 (rs2409411; P=0.004) and DPF3 (rs2192595; P=0.003).41 CDC42BPA is a downstream effector of CDC42 involved in modulating the neuronal cytoskeleton95 and hence neurite outgrowth.12 TIAM1 is a cytoplasmic protein that is involved in neurite outgrowth by directly activating RAC1.147 DPF3 is a nuclear protein that belongs to the neuron-specific chromatin remodeling complex (also called the nBAF complex),95 which has an important role in neural development and dendritic outgrowth of neurons.148
Recently, the results of a GWAS for an electrophysiological endophenotype of dyslexia (that is the so-called ‘mismatch negativity component’ or MMN) in a discovery sample of 200 and a replication sample of 186 dyslexic children were reported.42 Genome-wide significance was reached for two SNPs in high linkage disequilibrium in CLSTN2 (rs1365152 and rs2114167) in the discovery sample, though this finding was not replicated. As the replication sample was less strongly affected with dyslexia than the discovery sample, the authors suggest that CLSTN2 should not yet be discarded as a dyslexia candidate.42 The gene encodes calsyntenin 2, a brain-specific protein localizing to postsynaptic specializations of asymmetric synapses in the adult mouse brain149 and shows association with verbal working memory,150 a cognitive process that is repeatedly found to be impaired in dyslexic subjects.42, 63
Although its function is currently unknown, in analogy with its homologue calsyntenin 1,151 CLSTN2 might have a role in axonal transport, and may therefore fit well into our proposed neuronal migration and neurite outgrowth network.
In addition to the finding in CLSTN2, genome-wide significant association was reported in the combined discovery and replication sample between the late component of the MMN and rs4234898 on 4q32.1, an SNP in a gene desert, that ‘trans-regulates’ the expression of SLC2A3 (other name: GLUT3) on 12p13.31. SLC2A3 encodes the predominant facilitative glucose transporter in the brain. Therefore, the authors suggest that the ‘trans-regulation’ effect of rs4234898 (and a haplotype of this SNP with rs11100040) on SLC2A3 might lead to decreased glucose uptake in dyslexic children that could in turn explain their attenuated MMN results.42 The expression of the SLC2A3 gene is known to be upregulated by NF-κB152 and through estradiol/ESR1 activation.153 Moreover (glutamatargic) excitation of the neuronal cell membrane causes a translocation of SLC2A3 to the cell membrane leading to increased glucose uptake,154 which implies that SLC2A3 is a dyslexia candidate that may have an indirect role in neuronal migration and/or neurite outgrowth as an energy supplier.
In summary, we have comprehensively reviewed the literature related to the molecular genetic findings for developmental dyslexia. To our knowledge, this article constitutes the first attempt to integrate all reported genetic findings for dyslexia into a molecular signaling network. Furthermore, we have postulated a number of additional candidate genes for dyslexia that act in this network. The theoretical molecular/cellular model we have presented mechanistically links genes involved in directed neurite outgrowth and neuronal migration, two distinct neurodevelopmental processes that use an overlapping molecular (that is genetic) machinery. Our model does not provide further insights into which of these neurodevelopmental processes would be most relevant to the etiology of dyslexia, or where in the brain these processes are localized to selectively impact on neural circuitry determining reading and spelling performance. Although we were able to place 10 of the 14 dyslexia candidate genes that were suggested by the literature in the theoretical network, it is unlikely that there would just be a single explanatory model that connects all dyslexia candidate genes and their corresponding proteins on the molecular level. Rather, several etiological cascades contributing to dyslexia are likely to exist.
That being said, we think that the model identified in this paper makes important predictions about genes and functional relationships between genes that need to be directly tested in future studies.
Therefore, we think that a number of additional genetic association and mutation studies as well as functional studies—for example neuroimaging studies and studies in animal models—will be needed to sustain some of the claims made here. In this respect, we also hope that in the next few years, results of additional, large genome-wide studies of dyslexia and reading-related phenotypes will become available in order to provide more clues for further unraveling the genetic etiology of dyslexia.
We gratefully acknowledge the families who have made all these studies possible. We are also indebted to the many investigators whose work drives the dyslexia genetics field forward. This work was supported by a research grant from the ‘Hersenstichting Nederland’ (Brain Foundation of The Netherlands).