Molecular and Comparative Genetics of Mental Retardation
Jennifer K. Inlow, Linda L. Restifo


Affecting 1-3% of the population, mental retardation (MR) poses significant challenges for clinicians and scientists. Understanding the biology of MR is complicated by the extraordinary heterogeneity of genetic MR disorders. Detailed analyses of >1000 Online Mendelian Inheritance in Man (OMIM) database entries and literature searches through September 2003 revealed 282 molecularly identified MR genes. We estimate that hundreds more MR genes remain to be identified. A novel test, in which we distributed unmapped MR disorders proportionately across the autosomes, failed to eliminate the well-known X-chromosome overrepresentation of MR genes and candidate genes. This evidence argues against ascertainment bias as the main cause of the skewed distribution. On the basis of a synthesis of clinical and laboratory data, we developed a biological functions classification scheme for MR genes. Metabolic pathways, signaling pathways, and transcription are the most common functions, but numerous other aspects of neuronal and glial biology are controlled by MR genes as well. Using protein sequence and domain-organization comparisons, we found a striking conservation of MR genes and genetic pathways across the ∼700 million years that separate Homo sapiens and Drosophila melanogaster. Eighty-seven percent have one or more fruit fly homologs and 76% have at least one candidate functional ortholog. We propose that D. melanogaster can be used in a systematic manner to study MR and possibly to develop bioassays for therapeutic drug discovery. We selected 42 Drosophila orthologs as most likely to reveal molecular and cellular mechanisms of nervous system development or plasticity relevant to MR.

MENTAL RETARDATION (MR) is a common form of cognitive impairment affecting between 1 and 3% of the population of industrialized countries (Roeleveldet al. 1997; Aicardi 1998). Although there is debate over the definition and classification of MR (Leonard and Wen 2002), it is often defined by an IQ of <70, with deficits in adaptive skills included as diagnostic criteria (Luckassonet al. 1992; Dailyet al. 2000). Behavioral and cognitive therapies can help mentally retarded patients reach their maximum potential (Bathaee 2001; Butleret al. 2001), but they are not curative and often focus on treating habit disorders, aggression, or self-injurious behavior that can accompany MR (Long and Miltenberger 1998; Dosen and Day 2001). MR due to congenital hypothyroidism is now largely preventable through screening and hormone replacement (Gruterset al. 2002). Aside from this, the only molecular-based therapeutic approaches are dietary restrictions and supplements for inborn errors of metabolism such as phenylketonuria (Dashman and Sansaricq 1993; Levy 1999; Kabra and Gulati 2003). Few, if any, clinical conditions affect such large numbers of children and young adults and yet have no effective pharmacological therapy. One reason for the lack of drug treatments is the limited understanding of the molecular and cellular bases for MR.

Many environmental and genetic factors can cause MR, including premature birth, prenatal infections, chromosomal abnormalities, and single-gene mutations (Kinsbourne and Graf 2000). An etiology can be established in 60-75% of cases of severe MR, but only in 38-55% of mild cases. Estimates of genetic causes of severe MR range from 25 to 50% (McLaren and Bryson 1987). There are two categories of hereditary MR. Isolated MR with no other consistent defining features is known as nonspecific or nonsyndromal MR. To date, all but one of these (Molinariet al. 2002) are X-linked, but other autosomal genes may have eluded identification because of the considerably greater difficulty of mapping disorders to autosomal loci. MR also occurs, with variable penetrance and expressivity, as a phenotypic feature of numerous hereditary syndromes. The challenge of understanding the biological bases of hereditary MR is heightened by its enormous genetic heterogeneity and the limited knowledge of cellular phenotypes in the brains of mentally retarded individuals. Recent rapid progress in human genetics, however, has provided us with an opportunity for a comprehensive analysis of the biochemical and cellular processes underlying the MR phenotype. A search for “mental retardation” in the Online Mendelian Inheritance in Man (OMIM) database (Hamoshet al. 2002) yields >1000 entries, suggesting that hundreds of human genes can mutate to a MR phenotype. We conducted a detailed analysis to determine how many MR genes have been molecularly identified and what molecular and biological functions they encode.

Controversies over the definition of MR are based on both sociopolitical and biological considerations (Leonard and Wen 2002). Narrow definitions of MR restrict it to cases of nonprogressive cognitive impairment present from birth and categorize as “dementia” cases of progressive cognitive deterioration beginning some time after a period of normal development. Nonetheless, hereditary neurodegenerative disorders are often said to cause MR (see Stevensonet al. 2000), even when the onset is in late childhood or adolescence (e.g., progressive epilepsy with mental retardation, one of the neuronal ceroid lipofuscinoses; CLN8). Moreover, the distinction between MR and dementia blurs in disorders such as Rett syndrome (MECP2), where phenotypes span a wide spectrum of severity and clinical course (Hammeret al. 2002). For the purpose of our analysis of hereditary MR, we chose a broader, albeit less precise, definition that includes progressive disorders with onset of cognitive impairment in childhood and, occasionally, as late as adolescence.

In parallel with human genetics research, progress in Drosophila melanogaster genetics and genome sequencing (Adamset al. 2000) allows a comparative approach to the biological study of MR. Not only do homologous mammalian and fruit fly genes share biological functions (Padgettet al. 1993; Boniniet al. 1997; Johnstonet al. 1997; Leuzingeret al. 1998; Nagaoet al. 1998; Dearbornet al. 2002), but also Drosophila provides useful models of human disease, including spinocerebellar ataxia (Warricket al. 1998), Parkinson’s disease (Feany and Bender 2000), Huntington’s disease (Jacksonet al. 1998), and type 1 diabetes (Rulifsonet al. 2002). Moreover, neurodegeneration in the Drosophila model of Huntington’s disease can be suppressed by treatment with a specific peptide (Kazantsevet al. 2002). Hence, we propose that this neurogenetic model system can reveal cellular phenotypes responsible for hereditary MR and will provide bioassays for potential drug therapies. By searching the Drosophila genome, we found candidate functional orthologs for the majority of molecularly identified human MR genes. Several dozen of these genes are most likely to have mutant phenotypes due to primary developmental defects of neurons or glia and thereby provide clues to the causes and treatment of MR due to single-gene mutations. Treatment strategies based on the understanding of hereditary MR may be useful for acquired MR as well.


Databases and bioinformatics tools: The OMIM database [McKusick-Nathans Institute for Genetic Medicine, Johns Hopkins University and National Center for Biotechnology Information (NCBI), National Library of Medicine; Hamoshet al. 2002] was accessed online ( to search for genes and mental retardation disorders. BLASTP (Altschulet al. 1997) at the NCBI ( and the Homophila Human-Disease-to-Drosophila-Gene database (Reiteret al. 2001; were used to search for D. melanogaster homologs of the human MR genes. Pairwise sequence alignments were performed with LALIGN (; Huang and Miller 1991). DotPlot and TransMem of the Accelrys GCG Wisconsin Package were accessed through the Arizona Research Laboratories Biotechnology Computing Facility and were used to compare homologous protein sequences by dot matrix analysis (Maizel and Lenk 1981) and prediction of transmembrane regions, respectively. The InterPro resource for protein families, domains, and sites (Apweileret al. 2001; was used to determine and compare the locations of functional domains in homologous proteins. The Gene Ontology (GO) database (Gene Ontology Consortium 2001) was accessed online ( to determine the molecular-function classification of MR gene products. FlyBase (FlyBase Consortium 2002) was accessed online ( to obtain information on Drosophila genes. Newly isolated P-element insertions were found through the P-Screen Database (

Identifying human mental retardation genes through OMIM: We searched all OMIM fields on February 21, 2002, using the phrase “mental retardation” and reviewed each of the resulting 1010 entries. To include very mild MR, we also searched for “cognitive impairment” and “learning disability,” obtaining 38 additional entries for evaluation. In retrospect, “developmental delay” and “psychomotor retardation” would have been useful search phrases as well. Other MR genes were identified by periodic literature searches through September 30, 2003, using NCBI’s PubMed.

Careful evaluation of individual OMIM search results and cross-referencing with literature-search results revealed both false positives and false negatives. OMIM contains many partially redundant entries, which makes it impossible to equate numbers of entries obtained from a search for a specific phenotype with the number of genes that can mutate to that phenotype. OMIM entries for a genetic disorder or gene are organized into some or all of the following fields: title, MIM number, gene map, clinical synopsis, text (literature summary), allelic variants, references, and contributors. When different mutations of a single gene cause distinct disorders, there are separate OMIM entries for each disease, but only one contains a list of disease-associated alleles (“allelic variants” field). For example, mutations in the L1CAM gene result in one of at least three MR disorders (Weller and Gartner 2001): MASA syndrome (mental retardation, aphasia, shuffling gait, and adducted thumbs), HSAS (hydrocephalus due to congenital stenosis of the aqueduct of Sylvius), or SPG1 (spastic paraplegia 1). There is a separate OMIM entry for each of these disorders and a fourth entry for the L1CAM gene. There is some text redundancy among the four entries, but only the L1CAM entry includes the allelic variants field. On the basis of this organizational scheme, OMIM searches restricted to entries containing the allelic variants field should eliminate redundant results. However, this strategy would cause false negatives because entries that list allelic variants do not necessarily contain complete phenotype descriptions. For example, entry 600514, which lists the allelic variants of reelin (RELN), does not contain the phrase “mental retardation,” whereas entry 257320 for Norman-Roberts type lissencephaly syndrome due to RELN mutations contains the search phrase but does not list allelic variants. In principle, the “clinical synopsis” field could offer a useful search strategy for disease phenotypes, but some are incomplete (e.g., the clinical synopsis for Norman-Roberts lissencephaly does not include MR although it is a consistent phenotype of this disorder) and many entries have no clinical synopsis at all.

Errors in the clinical synopsis fields also contributed to the many (∼15%) false-positive entries (see Table 1). For example, entries 167200 and 167210 for pachyonychia congenita types 1 and 2 include MR in their clinical synopses, but the only evidence for MR is in the much rarer type 4 (Feinsteinet al. 1988). Other false positives result from statements such as “neither [patient] had evidence of mental retardation” (entry 243605). In other entries MR is not a feature of the disorder being described, but some atypical patients are mentally retarded due to deletion of adjacent genes (e.g., entry 312865). Finally, MR may be mentioned because related disorders have a MR phenotype. For instance, MR is a phenotype of a subset of hereditary spastic paraplegias, so it is mentioned in the text of the entries for most forms. Boyadjiev and Jabs (2000) noted similar difficulties in extracting information from OMIM. To obtain complete information from OMIM, one must search in a manner that yields redundant and irrelevant entries. This minimizes false negatives, but, to interpret the search results accurately, one must be willing to review individual entries carefully. Even using a broad OMIM search strategy, we missed 45 MR genes that were revealed through various literature search strategies.

Functional classification of human mental retardation genes: We searched for the 282 MR gene products in the molecular-function category of the GO database and used information from the literature to classify those not yet in the database. The GO database is composed of three parallel schemes for classifying gene function: biological process, cellular component, and molecular function (Gene Ontology Consortium 2001). Each ontology is a hierarchical classification scheme (directed acyclic graph) of structured vocabulary terms that differs from a simple hierarchical tree, such as a pedigree, in that each term may be a “child” of multiple independent “parents.” There are 24 occupied top-level terms in the molecular-function ontology, i.e., terms that do not have parents themselves. When GO assigned gene products to multiple molecular functions, we chose the most specific term for each. For example, we classified the α-subunit of Gs, the adenylate cyclase-stimulating guanine nucleotide-binding protein (GNAS), as a “nucleotide-binding protein” rather than as a “hydrolase,” the other GO assignment. For genes considered by GO to have “unknown function,” we found that most could be provisionally classified on the basis of data in the literature.

The “biological function(s)” assignments were based on literature reviews for each gene, including neuroimaging, gene expression, and neuropathological data from human patients, as well as studies of wild-type and mutant mice. We first designated the basic cellular process in which the gene is primarily involved, e.g., cytoskeleton or chromosome structure. We then identified the site of primary organ system function, relative to MR: endocrine system, central nervous system, or neither. For those genes that directly impact central nervous system (CNS) development and/or function, we ascertained the tissue type (neuron, glia, or blood vessel) and the specific cellular process affected (e.g., cell identity or differentiation). We also considered whether MR caused by mutation of the gene is secondary to toxicity or secondary to energy or fuel deficiency.

Identifying Drosophila orthologs of human mental retardation genes: We used bioinformatics tools to determine if the human MR genes have likely functional orthologs in D. melanogaster. For MR genes encoding tRNAs, we aligned the human and fly tRNA homologs using LALIGN and calculated the percentage identity. For each protein-coding MR gene, we searched the D. melanogaster sequences of the NCBI nonredundant database with NCBI’s BLASTP. We used an E-value cutoff of 1 × 10-10 (1e-10), a threshold commonly used for humanfly gene comparisons (Fortiniet al. 2000; Lloydet al. 2000; Reiteret al. 2001). The Homophila database (Reiteret al. 2001) is designed for such comparisons but, due to its organizational features and infrequent updates, we found it easier and more reliable to do our own BLAST searches. For one MR gene, we concluded that Drosophila does not have a biologically meaningful homolog despite a published claim of one. Grunge (FBgn0010825) is the most similar fly gene to human DRPLA (Zhanget al. 2002), but has a BLASTP E-value of 5E-2, which does not meet our threshold. Moreover, sequence similarity is limited to the extreme C terminus and the Grunge protein does not possess the same domain organization as DRPLA.

For protein-coding MR genes, we also conducted a “reverse” BLASTP search using the top-scoring Drosophila BLASTP result as a query against the human sequences of the NCBI nonredundant database. A Drosophila gene was considered an ortholog of a human MR gene only if this reverse analysis (sometimes supplemented with dot-matrix plot and protein-domain comparison; see below) revealed that it was more similar to the human MR gene (or a paralog) than to another gene. For example, the Drosophila proteins most similar to human glial fibrillary acidic protein (GFAP) are the products of Lamin and Lamin C. A reverse BLASTP search revealed that, although these two proteins share a single common domain with GFAP, they are more similar over their full lengths to members of the human lamin family. In addition, both human and Drosophila lamins are localized to the nucleus (Goldmanet al. 2002), whereas GFAP is cytoplasmic (Enget al. 2000). Hence, GFAP does not have an ortholog in Drosophila.

When compared with mammals, Drosophila has relatively few duplicated genes (Durand 2003), so in some cases a Drosophila gene is the single ortholog of a paralogous set of human genes. For example, FMR1, which causes fragile X syndrome, is a member of a gene family that also includes FXR1 and FXR2, the autosomal fragile X-related genes. Drosophila dfmr1 is the only homologous fly gene, sharing significant sequence similarity and domain structure with all three human genes, suggesting that it is the sole ortholog.

To determine if orthologous genes are likely to share the same molecular and biological functions in humans and flies, we used dot matrix plots (GCG DotPlot) to assess the extent of protein sequence similarity and searched the InterPro database for known functional domains in each protein. GCG TransMem was used to predict transmembrane regions in the human and fly proteins. If the proteins share sequence similarity over most of their lengths and have similar organization of known functional domains, we considered them to be candidate functional orthologs. In some cases we also considered expression patterns, mutant phenotypes, and subcellular localization. In cases of “computed genes” predicted from the Drosophila genome sequence, the absence of experimental data made the evaluation of ortholog status more difficult.


The 282 mental retardation genes have been molecularly identified: Analysis of OMIM and literature search results allows us to present a status report on the genetics of MR. From the 1010 OMIM “mental retardation” entries obtained on February 21, 2002, we found 204 human genes that cause MR either in isolation or as part of a syndrome. Through literature searches we found 45 additional MR genes whose OMIM entries did not contain the search phrase “mental retardation.” About a quarter of these “false-negative” entries contained the phrases “psychomotor retardation” and/or “developmental delay.” To include disorders causing very mild MR, we also searched OMIM for entries containing “cognitive impairment” or “learning disability” but not “mental retardation.” Most of these 38 entries describe adult-onset, progressive cognitive impairment disorders, but literature review identified 4 of them as MR genes. Finally, literature searches between March 2002 and September 30, 2003 revealed 29 recently identified MR genes for a total of 282 human genes known to cause MR (Figure 1). On the basis of these and subsequent publications, we estimate that new MR genes are being identified at a rate of 1-2 per month. The appendix lists the 282 MR genes in alphabetical order by their gene symbols, along with their associated MR disorders, chromosomal locations, OMIM numbers, and other information explained below. As will be discussed in later sections, the MR genes control an extraordinary range of molecular and cellular functions.

Figure 1.

—Diagram of the identification of human mental retardation genes and their comparison to D. melanogaster genes. The OMIM searches were performed on February 21, 2002. The literature search was completed on September 30, 2003.

We classified the 1010 OMIM “mental retardation” entries, based on data available in spring 2002, according to the following scheme (Table 1):

  • Category 1: The disorder has been mapped to a specific gene and allelic variants have been identified (this category includes OMIM entries for the MR disorders as well as separate entries describing the genes themselves).

  • Category 2: The disorder has been mapped to one or more candidate genes in a chromosomal region (contiguous gene deletion syndromes, e.g., Prader-Willi, are in this category).

  • Category 3: The disorder has been mapped to a chromosomal region.

  • Category 4: The disorder has been mapped to a candidate chromosome.

  • Category 5: The disorder has not yet been mapped to a chromosome.

  • Category 6: The disorder is caused by a gross chromosomal abnormality and no single gene determines the MR phenotype (Down syndrome is one example).

  • Category 7: MR is not a phenotype of the disorder.

  • Category 8: The disorder does not exist.

View this table:

OMIM mental retardation entries

The number of OMIM entries in category 1 (“known gene”), 254, is greater than the number of genes, 204, because of OMIM database redundancy (see materials and methods). The nearly 600 OMIM entries in categories 2-5 represent MR disorders in which the causative genes were unknown (see below). Of the 29 recently discovered MR genes, half had “advanced” from “candidate gene” (1 gene), “chromosomal region” (9 genes), or “unmapped” (5 genes) categories. Thirteen represent new loci that can cause a known disorder. One (FKRP) causes a form of muscular dystrophy, not previously associated with MR, that had been in category 7.

Entries in category 6 (“chromosomal abnormality”) describe bona fide MR disorders, but we have not considered them further in this analysis because they appear to involve many genes (e.g., Shapiro 1999). It remains to be determined whether individual genes that contribute to MR in cases of aneuploidy or other chromosomal defects can mutate to an MR phenotype individually. The 149 OMIM entries in category 7 (“no MR phenotype”) represent false positives in which MR is not a phenotype (see materials and methods). Most of these false-positive errors could be eliminated by the adoption of a controlled vocabulary for OMIM clinical synopses, with the previously mentioned caveat that MR definitions vary. The three entries in category 8 (“nonexistent disorders”) do not represent distinct clinical entities, and one was subsequently removed from the OMIM database.

With ∼600 OMIM MR entries in categories 2-5 (Table 1), it is obvious that many more MR genes remain to be identified—but how many? Some of these disorders, particularly those in categories 4 (“candidate chromosome”) and 5 (“not mapped”), are likely to represent MR genes that are already known. This is because of both practical difficulties in mapping human phenotypes and the phenomenon of phenotypic divergence; i.e., different mutant alleles of the same gene cause distinct MR disorders (e.g., different DKC1 mutations result in dyskeratosis congenita or Hoyeraal-Hreidarsson syndrome). Similarly, novel MR genes that remain to be identified may each explain more than one disorder, especially within the large unmapped group. Hence, this set of OMIM entries is likely to represent <595 genes.

On the other hand, what MR disorders might be “missing” from our analysis? First, we know that some genes, or their corresponding disorders, are present in the OMIM database but fail to appear in MR-related search results because of inconsistent use of terminology in the medical literature, curatorial errors, or differing opinions on what constitutes mental retardation (see materials and methods). Second, MR mutations occurring in small families likely represent a large number of genes not yet listed in OMIM. Some families never reach the attention of medical genetics research teams. Small pedigrees represent significant challenges for gene mapping, even on the X chromosome (Roperset al. 2003). The X-Linked Mental Retardation Genes Update Site (; Chiurazziet al. 2001) lists 57 nonspecific MR families and 110 X-linked MR syndromes for which the genes remain elusive. However, only 80 OMIM entries described X-linked MR disorders (syndromes and nonspecific) for which genes have not been identified (Table 1, X-linked entries in categories 2-4).

A third “missing” or underrecognized category is composed of essential genes of which most deleterious mutations cause early prenatal lethality and only exceptional alleles with specific molecular consequences permit viability along with an MR phenotype. In genetic model systems, complementation testing can easily show that a viable “memory mutation” is allelic to mutations causing early death with profound neuroanatomical defects (e.g., Pintoet al. 1999), but comparable mapping studies are much more difficult in humans.

Fourth, mutations in genes controlling thyroid development or function rarely cause MR in industrialized societies because of neonatal screening and treatment for hypothyroidism (Gruterset al. 2002). Hence, while a dozen known genes have been associated with MR secondary to hypothyroidism (appendix), mutations in other similar genes may not have had the “opportunity” to reveal whether they would cause MR in untreated patients. Finally, syndromal MR genes for which the MR phenotype has very low penetrance present a significant ascertainment challenge. For example, eight DNA repair genes/disorders are associated with MR in a modest fraction of patients. It seems likely that more such disorders (e.g., the rarer Fanconi anemia complementation groups) have MR as a bona fide phenotype, but, presumably because the phenotype depends on chance somatic mutations during brain development (Gilmoreet al. 2000), it is difficult to confidently assign MR to their clinical descriptions.

Given all these considerations, predicting the true number of human MR genes is difficult. A complete and accurate count may be beyond the capacity of medical science to determine directly. We believe that 282 represents substantially less than half of the total. It is easy to imagine that human MR genes could number ∼1000.

X-linked mental retardation genes: To date, eight X-linked genes are known to cause exclusively nonspecific MR (MRX genes), and 31 X-linked genes cause exclusively syndromal forms of MR (Table 2). Nonspecific MR has been the focus of much attention, in part because of the idea that genes with “pure” behavioral phenotypes, unaccompanied by gross brain abnormalities or other organ system defects, may provide greater insight into the molecular basis of cognition than the syndromal MR genes (Chelly 1999; Toniolo 2000). Indeed, several MRX genes figure prominently in Rho-type G-protein pathways (ARHGEF6, GDI1, OPHN1, PAK3, FGD1; Ramakers 2002) or are regulated by neuronal activity (PAK3, IL1RAPL1, RSK2, TM4SF2; Bodaet al. 2002). However, with the discovery that mutations of five MR genes can cause either nonspecific or syndromal MR (Table 2), the distinction between the two categories may not be as meaningful as originally proposed (see discussion in Frintset al. 2002).

For RSK2 (RPS6KA3), the phenotype difference is explained by allele type and severity. The R383W mutation that causes MRX19 is a partial loss-of-function allele, encoding a protein with 20% of wild-type kinase activity (Merienneet al. 1999). In contrast, null mutations of RSK2 cause Coffin-Lowry syndrome with prominent skeletal and connective tissue involvement (Hanauer and Young 2002). For several genes, the structure-function relationships are inferred but not directly demonstrated. The T1621M mutation of ATRX (also known as XH2 or XNP) causes nonspecific MR in the mild-to-moderate range (Yntemaet al. 2002). Although residue 1621 is within the highly conserved SNF2-related domain, it is not conserved, suggesting that some alterations at that site are compatible with partial function of this nuclear protein involved in chromatin structure and transcription regulation. Missense mutations just 7 and 12 residues upstream, however, cause a more severe, syndromal phenotype with hematologic, skeletal, and genital defects (Gibbonset al. 1995), suggesting greater disruption of ATRX function. A variety of FGD1 mutations, most of which truncate the encoded putative Rho GEF, cause Aarskog-Scott syndrome, which includes highly penetrant skeletal and genital anomalies but infrequent, and only mild, MR. In contrast, one particular missense mutation in a region of unknown function, P312L, causes severe, fully penetrant nonspecific MR (Lebelet al. 2002).

View this table:

X-linked mental retardation genes

Genotype-phenotype relationships are even more complex for MECP2 and ARX. Within and among Rett syndrome families, females with MECP2 mutations show great clinical heterogeneity, with X-inactivation patterns and mutation sites believed to explain the severity differences (Cheadleet al. 2000; Hammeret al. 2002). In addition, at least seven different missense mutations in MECP2, scattered over the length of the protein, cause nonspecific MR (Orricoet al. 2000; Couvertet al. 2001); several of these are very close to sites of Rett-syndrome-causing missense mutations (Cheadleet al. 2000; Hammeret al. 2002). For ARX, identical mutations, resulting in polyalanine tract expansion of this homeodomain protein, caused nonspecific MR in one family, but distinct neurological syndromes (West or Partington or MR with hypsarrhythmia) in various other families (Strommeet al. 2002). This suggests a major effect of genetic background on ARX phenotypes. Other ARX mutations cause a unique lissencephaly syndrome with abnormal genitalia (Kitamuraet al. 2002).

Complex genotype-phenotype relationships are also a feature of some autosomal MR disorders (e.g., FGFR1, GLI3, PEX1, PTEN, PTPN11). On the basis of X-linked MR, it is possible that some alleles of the one known autosomal nonspecific MR gene (PRSS12; Molinariet al. 2002) will be found to cause a syndromal MR phenotype. Conversely, autosomal genes presently known to cause MR syndromes may be able to mutate to a nonspecific MR phenotype.

Chromosomal distribution of human mental retardation genes: Of the 282 human MR genes, 11 are encoded by the mitochondrial genome. Figure 2A shows the chromosomal distribution of the 271 nuclear MR genes compared to the chromosomal distribution of all known and predicted human genes based on the human genome sequence (Venteret al. 2001). While ∼4% of known and predicted genes are on the X chromosome, ∼16% of the MR genes reside there—a fourfold overrepresentation. In contrast, the distribution of MR genes among the autosomes roughly parallels their relative gene contents (Figure 2A). An even greater X-chromosome overrepresentation is found among the MR disorders mapped to candidate loci (6-fold), chromosomal regions (14-fold), and chromosomes (15-fold), which correspond to categories 2, 3, and 4, respectively, of Table 1.

It has been proposed that the human X chromosome contains a disproportionately high density of genes for cognitive ability (Lehrke 1972; Turner and Partington 1991). This proposal generated controversy as well as speculation concerning possible underlying evolutionary mechanisms, including the intriguing suggestion that female mate selection for high male intelligence helped accelerate the rapid rise of human cognitive abilities (Turner 1996; Zechneret al. 2001). The identification of numerous MRX genes and X-linked MR syndromes (Chiurazziet al. 2001) seemed to support the proposal. Opponents, however, argued that all X-linked recessive mutations are simply easier to map and identify because their phenotypes are revealed in hemizygous males (Morton 1992; Lubs 1999). Countering this view is an OMIM-based analysis (Zechneret al. 2001) showing a 7.2-fold X-chromosome bias for MR genes, whereas genes causing common morphological phenotypes (polydactyly, cleft palate, facial dysplasia, skeletal dysplasia, and growth retardation) have, on average, only a 2.4-fold X-chromosome bias. [Zechner et al. (2001) did not take OMIM errors, such as false positives and negatives, into consideration, but such errors may be comparable across phenotypes.]

To take this question one step further, we asked whether the apparent X-chromosome overrepresentation among the molecularly identified human MR genes (Figure 2A) would disappear if we accounted for the plausible possibility that numerous autosomal loci are “hiding” among the unmapped MR genes (represented by the OMIM entries in category 5, Table 1). We attempted to overcome the ascertainment bias that favors identification of X-linked genes by making simplifying assumptions that maximize the estimate of autosomal MR genes and minimize the estimate of X-linked MR genes. First, we assumed that one OMIM entry equals one gene. Second, for the unmapped MR disorders (category 5, Table 1), we assumed that each represents a different, novel autosomal gene and that these are distributed in proportion to the overall gene distribution on those chromosomes (Venteret al. 2001). Third, for those disorders whose genes map to chromosomal regions and candidate chromosomes (categories 3 and 4, Table 1), we assumed that there will be no new X-linked genes, i.e., that each potential X-linked gene is identical to an X-linked gene already known to cause MR. However, all candidate genes (category 2, Table 1), including the X-linked genes, were assumed to be new MR genes.

Figure 2.

—Chromosomal distribution of human mental retardation genes and D. melanogaster orthologs. (A) The chromosomal distribution of the 271 molecularly identified nuclear MR genes is compared to the chromosomal distribution of all nuclear, protein-coding human genes based on human genome sequence analysis (Venteret al. 2001). Note the striking overrepresentation of X-linked MR genes. (B) The predicted chromosomal distribution of known and potential MR genes based on maximizing the assignment of genes to autosomes (see results and discussion), compared to all nuclear human genes as in A. The X-chromosome overrepresentation has been reduced, but remains almost twofold. (C) The chromosomal distribution of the Drosophila MR gene orthologs is compared to the chromosomal distribution of all nuclear, protein-coding Drosophila genes on the basis of Drosophila genome sequence analysis (Adamset al. 2000; see FlyBase at for Release 3). Drosophila homologs of human MR genes that are not orthologs were not included in this analysis.

Even when these very conservative (i.e., biased toward autosomal) assumptions are used to estimate the chromosomal distribution of the unknown MR genes, a 1.9-fold overrepresentation of MR genes on the X chromosome remains (Figure 2B). This result supports the hypothesis that the X chromosome contains a disproportionately high density of genes influencing cognitive ability. One caveat is the possibility discussed above that many autosomal MR genes may be so rare or difficult to study that they never appear in the medical literature and, hence, in OMIM. We also agree with the suggestion of Lubs (1999) that resolution of this issue would be enhanced by analyzing genome-wide brain expression data and by searching for allelic variation in single genes responsible for the high end of the intelligence spectrum.

D. melanogaster homologs of human mental retardation genes: We found that 87% of known MR genes (246/282) have at least one Drosophila homolog with a BLASTP E-value of 1 × 10-10 or better (Figure 1; appendix). Similarly, Reiter et al. (2001) found that 75% of ∼1400 human disease genes, representing all major disease categories, have Drosophila homologs at this level of sequence similarity. More important, 76% (213) of the MR genes, including syndromal and nonsyndromal types, have at least one Drosophila ortholog (see materials and methods and appendix). In fact, a handful of the human genes were named for their Drosophila orthologs, in most cases prior to their identification as MR genes (ASPM: abnormal spindle-like, microcephaly-associated; EMX2: homolog 2 of empty spiracles; PTCH: homolog of patched; PTCH2: homolog 2 of patched; SHH: sonic hedgehog; SIX3: homolog 3 of sine oculis).

The appendix lists the Drosophila homologs and orthologs of the MR genes, their FlyBase accession numbers, and the BLASTP E-values (see also Figure 1 for overview). As discussed below, several dozen Drosophila orthologs (designated “¶” in the appendix) are prime candidates for cellular and molecular study of MR. Seventeen MR genes (6%; designated with asterisk) have one or more homolog(s) that may be orthologs, but it is not possible to make a determination on the basis of sequence analysis in the absence of experimental data. Another 16 MR genes (6%; in brackets) have one or more Drosophila homolog(s) that are not orthologs on the basis of reverse BLAST results or other sequence analysis (see materials and methods). There are 36 MR genes (13%) with no Drosophila homolog, although this number may decline as final gene identification for the Drosophila genome is completed.

Some of the Drosophila genes are functional orthologs of human MR genes on the basis of experimental data. For instance, mutations of dfmr1, the Drosophila ortholog of fragile X mental retardation 1 (FMR1; Wanet al. 2000), cause specific disruptions of neuronal morphology (Zhanget al. 2001; Moraleset al. 2002; Leeet al. 2003; C. Michel, R. Kraft, B. Hassan and L. Restifo, unpublished results) and behavioral defects (Dockendorffet al. 2002; Inoueet al. 2002). Genetic and biochemical data suggest that Drosophila dFMR1 is a regulator of translation (Zhanget al. 2001; Ishizukaet al. 2002), as has been shown for mammalian FMRP (Kaytor and Orr 2001; Laggerbaueret al. 2001; Mazrouiet al. 2002; Zalfaet al. 2003). Although learning phenotypes of dfmr1 mutant flies have not yet been reported, four of the fruit fly MR gene orthologs are “learning and memory genes” on the basis of behavioral data: G protein sα60A (Connollyet al. 1996), the ortholog of GNAS; Neurofibromin 1 (Guoet al. 2000), the ortholog of NF1; cheerio (see Dubnauet al. 2003, online supplement), the ortholog of FLNA; and S6kII or ignorant (G. Putz, T. Zars, and M. Heisenberg, personal communication), the ortholog of RSK2. Additional Drosophila learning and memory genes have been proposed as candidates for MR disorders that are not yet mapped (Morley and Montgomery 2001).

The Drosophila orthologs of the human MR genes do not have a skewed chromosomal distribution (Figure 2C). Approximately 16% of all fly genes and 16% of MR gene orthologs are on the X chromosome. Of the first two dozen Drosophila “learning and memory genes” identified, almost 50% are X-linked (reviewed in Dubnau and Tully 1998; Morley and Montgomery 2001). However, the recent isolation of 60 new autosomal memory genes (Dubnauet al. 2003) indicates that the older results reflect the previous tendency to design X-chromosome screens for behavioral and neuroanatomical phenotypes.

Molecular functions of mental retardation genes: Each of the 282 MR genes was classified in a single molecular-function category, primarily on the basis of the GO database (Figure 3; appendix; see materials and methods). The MR genes are distributed over a broad range of functions, indicating that disruption of any of a wide array of molecular processes can impair brain function so as to cause MR. Several categories are prominently represented, such as enzymes (143 genes; 51%), mediators of signal transduction (32 genes; 12%) and transcription regulation (19 genes; 7%), binding proteins (23 genes; 8%), and transporters (21 genes; 8%). Enzymes, especially those expressed in accessible peripheral tissues, make gene identification easier than that for many other proteins, so their relative representation may decline as new MR genes are discovered. Other categories with smaller numbers of MR genes include cell adhesion molecule, structural molecule, motor protein, tRNAs, apoptosis regulator, chaperone, and enzyme regulator. GO classifies ∼9% of the MR genes (25) in the “unknown function” category, but published data suggest functions for all but 10 of them (see appendix).

Within the GO molecular-function ontology, top-level categories include fundamental molecular functions (e.g., binding activity, of which there are many subcategories), as well as others related to a specific cellular process (e.g., cell adhesion molecule), and in many cases, genes could be assigned to more than one. This makes classification, analysis, and comparison to other sets of genes somewhat difficult. We did not classify any MR gene products as “defense/immunity proteins,” but IKBKG encodes a subunit of a signal transducer (our category choice) that regulates NF-κB in the immune and inflammatory response pathway (Wallachet al. 2002). We also did not use the “translation regulator” category, but EIF2AK3 encodes a kinase (our category choice) that indirectly regulates translation by phosphorylating eukaryotic translation initiation factor-2 (Maet al. 2002). Similarly, we classified FMR1 as “RNA binding,” but considerable data demonstrate that it regulates translation (Jin and Warren 2003). In addition, we could have classified some genes in the “protein stabilization” (e.g., PPGB), cytoskeletal regulator (e.g., TBCE), or “protein tagging” (e.g., UBE3A) categories. However, anticoagulant, antifreeze, antioxidant, chaperone regulator, nutrient reservoir, and toxin are top-level categories in which none of the 282 MR genes could be placed.

Figure 3 indicates the Drosophila-homolog status of the MR genes in each molecular-function category. The 213 MR genes with Drosophila ortholog(s) (solid bars) are distributed among the GO categories in roughly the same pattern as that of all the MR genes, with two exceptions. More than half of the “receptor binding” genes (4 of 7) and 36% (9 of 25) of the “unknown function” MR genes have no Drosophila homolog.

Biological functions of mental retardation genes: We devised a “biological function(s)” classification scheme for the 282 MR genes that considers both cellular- and systems-level perspectives (Figure 4; appendix; see materials and methods). The basic cellular processes controlled by MR genes take place in the nucleus, in the cytoplasm (including within organelles), and at the interface among cells, cell compartments, and the extracellular milieu. In the nucleus, MR genes affect chromosome structure (e.g., DNMT3B), DNA repair (e.g., NBS1), basal and regulated transcription (e.g., ERCC2 and SIX3, respectively), as well as rRNA processing (e.g., DKC1).

Figure 3.

—Molecular function classification of mental retardation genes. Genes were classified on the basis of GO categories (see materials and methods). In cases where the top-level parent terms include large numbers of genes (signal transduction, binding, transcription regulation, enzyme), we show the distribution of genes among the children terms. For many of the genes that have not yet been classified by the GO Consortium, we used information from the literature to assign them to a GO term. For some of the genes designated “unknown function” by GO, we were able to assign provisional functions on the basis of published literature (see appendix), but these genes are included in the “unknown function” category of this figure. As indicated by the boxed legend, each bar indicates classification of the human MR genes based on the degree of similarity to Drosophila genes.

In the cytoplasm, many MR genes have metabolic functions (see also Kahler and Fahey 2003), involving a wide range of pathways [citric acid cycle (e.g., FH), gluconeogenesis (e.g., GK), glycolysis (e.g., PDHA1), oxidation (e.g., the PEX genes), oxidative phosphorylation (e.g., MTCO1), urea cycle (e.g., OTC), and general cell integrity (e.g., GSS)] and biologically critical compounds [amine (e.g., MAOA), amino acid (e.g., OAT), carbohydrate (e.g., GALE), cholesterol (e.g., SC5DL), creatine (e.g., GATM), fatty acid (e.g., ALDH3A2), heme (e.g., PPOX), lipid (e.g., DIA), methionine (e.g., MAT1A), purine (e.g., HPRT), pyrimidine (e.g., DPYD), and cofactors (e.g., TC2)]. MR genes involved in macromolecular synthesis and modification include those required for mitochondrial translation (e.g., MTTK), translation regulation (e.g., FMR1), protein folding (e.g., BBS6), protein stability (e.g., PPGB), protein glycosylation (e.g., PPM2), and lipid synthesis (FACL4). Macromolecular degradation in lysosomal (e.g., HEXA) and proteasomal (e.g., UBE3A) pathways is also commonly disrupted by mutations in MR genes. MR genes have major effects on the cytoskeleton, including its actin (e.g., FLNA), microtubule (e.g., DCX), and intermediate filament (e.g., GFAP) components.

The major signaling pathways are represented among the MR genes, including those regulated by Sonic Hedgehog (e.g., SHH), the TGF-β family of growth factors (e.g., GPC3), Notch (e.g., JAG1), and calcium (e.g., ATP2A2). MR-related signaling cascades are mediated by diverse cell surface proteins, such as integrins (e.g., ITGA7), G protein-coupled receptors (e.g., AGTR2), receptor tyrosine kinases (e.g., NTRK1), and intracellular proteins, including small G proteins (e.g., GDI1), heterotrimeric G proteins (e.g., GNAS), and phosphatidylinositol (e.g., PTEN). Moreover, genes in a common pathway can share MR as a phenotype. SHH (Minget al. 1998), through its receptors encoded by PTCH and PTCH2, regulates GLI3, some of whose targets are also regulated by GPC3.

MR genes also control communication and transport across cell and organelle membranes. These include cation-chloride cotransporters (SLC12A1, SLC12A6) that may be critical for inhibitory neurotransmission (Payneet al. 2003). The transmembrane linkage (ITGA7, TM4SF2) between the extracellular matrix (LAMA2) and the cytoskeleton is strongly implicated in MR, as is cell adhesion (L1CAM).

The overlap between MR and muscle disease is striking and appears to arise from at least three distinct mechanisms: reduced membrane/cytoskeletal stability (DMD, ITGA7, LAMA2); glycosylation defects associated with abnormal neuronal migration (FCMD, FKRP, LARGE, POMGNT1, POMT1); and mitochondrial dysfunction (MTCO3 and many others). The biological basis of myotonic dystrophy (DM1) is unknown.

An integrative view of MR biology: The hereditary MR disorders can be approached from two somewhat independent perspectives: (i) where the genes are expressed and function and (ii) the relationship between the mutation and pathogenesis of the MR phenotype. Genes may act selectively within the brain (“intrinsic or selective function”) or primarily outside the CNS (“extrinsic or generalized function”). MR may result from fundamental cellular defects that impair many tissues (“generic effect”), with the brain sometimes having a higher sensitivity, or MR can result from selective impairment of unique features of brain development or physiology (“selective effect”). With the caveat that MR pathogenesis is incompletely understood and that spatial expression data are limited, we consider examples of MR genes in these major categories.

Figure 4.

—Biological functions that underlie mental retardation. Diagram of a mammalian cortical neuron and associated structures in the central nervous system. The physiological connection to the endocrine system via the bloodstream is indicated in the bottom left. Sizes are not to scale. Solid triangles represent hormone molecules. Each of the unboxed terms, in roman type, is a biological function regulated by one or more MR genes or results from mutation of an MR gene (see appendix).

Extrinsic or generalized function/generic effect: ABCC8 (SUR1) and KCNJ11 gene products work together in the pancreas to regulate ATP-dependent, exocytotic insulin secretion. Mutations in either gene cause excess insulin release and hypoglycemia which, if inadequately treated, disrupts brain development and function due to systemic fuel deficiency (Vannucci and Vannucci 2001; Huopioet al. 2002). Similarly, the brain’s energy requirements make it very sensitive to genetic disruptions of mitochondrial function (Chow and Thorburn 2000). Mutations in mitochondrial genes (MTATP6, MTCO1, MTCO2, MTCO3, MTCYB, MTTE, MTTK, MTTL1, MTTS1) or in nuclear genes encoding mitochondrial proteins (BCS1L, SCO2, SURF1, TIMM8A) cause MR due to local energy (ATP) deficiency in neurons and glia (Servidei 2001).

Extrinsic or generalized function/selective effect: In the endocrine system, locally synthesized hormones enter the circulation and affect distant organs. MR genes include several tissue-specific regulators of thyroid gland development (TTF2, PAX8) or thyroid hormone synthesis (DUOX2, TG, TPO; Kopp 2002). Mutations in these cause congenital hypothyroidism, and mutations in a receptor (THRB) cause thyroid hormone resistance. In either case, brain cells cannot initiate the transcriptional cascade that controls neuronal size, migration, and dendritic morphology, as well as oligodendrocyte differentiation (Thompson and Potter 2000). Hence, neuronal circuitry and myelination are disrupted.

Many metabolic MR genes fall into this category as well. AASS is expressed in most tissues and encodes a key enzyme in lysine metabolism (Sackstederet al. 2000). In patients lacking AASS function, lysine accumulates and inhibits arginase, causing excess circulating ammonia, which interferes with neuronal and glial functions (Felipo and Butterworth 2002). Similarly, PAH is expressed mainly in nonneural tissues (Lichter-Koneckiet al. 1999), with mutations causing elevated circulating phenylalanine. This systemic toxin impairs myelination, synaptogenesis (Bauman and Kemper 1982; Huttenlocher 2000), and possibly aminergic neurotransmission (Surtees and Blau 2000). The lysosomal storage disorders, which cause macromolecules to accumulate in many tissues, may also belong to this category. Most represent degradative enzyme deficiencies, but some of the genes encode transport, stabilizer, or activator proteins (Wisniewskiet al. 2001). They are classified by the compounds that accumulate in lysosomes, such as sphingolipidoses (e.g., ARSA), neuronal ceroid lipofuscinoses (e.g., CLN1), glyoproteinoses (e.g., PPGB), and mucolipidoses (e.g., NEU1). The traditional view that the progressive brain phenotypes result “simply” from local toxicity is countered by reports of specific neurodevelopmental defects (Walkley 1998; Altarescuet al. 2002).

Intrinsic function/selective effect: For genes with selective expression or function within the CNS, the consequences of mutations are also primarily CNS selective, with variation in cell-type involvement and severity (Pomeroy and Kim 2000). The coexistence of neuropathology and cognitive deficits supports the view of MR as a disorder of brain development or plasticity. At one end of the spectrum are MR disorders with gross brain malformations. Holoprosencephaly, a failure of the right and left brain halves to form distinct hemispheres, results from mutations in genes controlling cellular identity of forebrain neuronal precursors (PTCH, SHH, SIX3, TDGF1, TGIF, ZIC2; Wallis and Muenke 2000). Schizencephaly (“cleft brain”) is due to dominant missense mutations in EMX2, which encodes a homeodomain-containing transcription factor (Faiellaet al. 1997). Abnormal neuronal migration in the rostral forebrain (the region of EMX2 expression) causes gross morphogenetic as well as more subtle lamination defects. Neuronal migration defects also cause lissencephaly (“smooth brain”) due to mutations in LIS1, DCX, and RELN, as well as ARX (some alleles) and FLNA (Olson and Walsh 2002). Agenesis (partial or complete) and dysgenesis of the interhemispheric corpus callosum (Davila-Gutierrez 2002) are relatively common MR-associated phenotypes (e.g., CXORF5, GLI3, OCRL, SLC12A6, TSC1, TSC2) and may be isolated or accompany holoprosencephaly and other abnormalities.

A handful of MR genes and their primary cellular phenotypes are glia specific. Dominant missense mutations in GFAP cause Alexander disease due to astrocytic accumulation of abnormal intermediate filaments and secondary demyelination (Johnson 2002). In contrast, PLP1 is expressed solely in oligodendrocytes and encodes the most abundant CNS myelin protein. Myelin integrity is very sensitive to PLP1 gene dosage, with duplications, deletions, and missense mutations all causing Pelizaeus-Merzbacher disease (Koeppen and Robitaille 2002).

At the other end of the spectrum are the many hereditary MR disorders for which routine neuropathological data are unavailable or fail to show consistent defects. Higher-resolution Golgi staining has revealed dendritic abnormalities of cortical neurons in fragile-X (FMR1; Irwinet al. 2000) and Rett syndromes (MECP2; Armstrong 2001) and possibly in Rubinstein-Taybi syndrome (CREBBP; Kaufmann and Moser 2000). All three likely result from misregulated gene expression in the brain, but which target genes are responsible for the dendritic defects remain to be determined.

For MR disorders with no known anatomical lesions, such as nonsyndromal MRX, gene function in the CNS is inferred from molecular analyses. For example, GDI1 (MRX41, MRX48; Bienvenuet al. 1998) encodes a brain-specific regulator of Rab-type G proteins. One of its targets is believed to be Rab3A, which controls activity-dependent synaptic vesicle recruitment to axon terminals (Leenderset al. 2001). Given the structure-function relationships underlying developmental synaptic plasticity (Cohen-Cory 2002), it seems likely that neuroanatomical phenotypes for this and other MRX disorders will eventually be found.

Regardless of the scheme used, many disorders defy straightforward classification. For example, the role of homocysteine in CNS development and function (Mattson and Shea 2003) belies the “metabolic” classification of the MR genes CBS, MTHFR, MTR, MTRR, and TC2. The MR genes SC5DL and DHCR7 encode enzymes in cholesterol biosynthesis, making them also primarily “metabolic.” However, because Sonic Hedgehog protein function is absolutely dependent on covalent linkage to cholesterol (Ingham and McMahon 2001), the enzymatic deficiencies may impair SHH signaling. It may be that, with sufficient research on molecular and cellular pathogenesis, few if any MR genes will be considered “just metabolic.”

View this table:

Human mental retardation genes: prime candidates for study in D. melanogaster

The role of D. melanogaster in MR research: In terms of primary amino acid sequence and protein-domain organization, the degree of MR gene conservation between humans and Drosophila is remarkable (Figure 1; appendix). Not only individual genes but also whole pathways have been retained through ∼700 million years of evolution. These include protein glycosylation (ALG3, ALG6, B4GALT1, DPM1, FUCT1, GCS1, MGAT2, MPDU1, PMI, PPM2), as well as signaling pathways, notably the Hedgehog pathway (SHH, PTCH, PTCH2, GLI3, GPC3) and those mediated by small G proteins (ARHGEF6, GDI1, OPHN1, PAK3, FGD1, GPH, RSK2, and others).

Given this remarkable conservation of MR genes, we propose that Drosophila genetics can be used in a systematic manner to study MR. We have selected 42 fly genes (the orthologs of 43 human MR genes) as “prime candidates” for such analyses (Table 3). These genes most likely act selectively within the brain during development to establish the anatomical and physiological substrates for experience-dependent plasticity. The majority of prime-candidate orthologs currently have fly mutants available (about the same fraction as have mouse mutants available) and the rest can be mutagenized through the mobilization of nearby transposable elements or studied using RNA interference methods (Adams and Sekelsky 2002). About half are already known to have neural phenotypes, behavioral or anatomical, in Drosophila (Table 3 and references therein). The anatomical defects involve neurons (e.g., cubitus interruptus), glia (e.g., Neurofibromin 1), and neural precursor cells (e.g., division abnormally delayed) and result from problems with proliferation (e.g., hedgehog), migration (e.g., breathless), and process extension or arborization (e.g., Pak, dfmr1). For a few genes, neuronal defects in the mushroom bodies, an arthropod learning and memory center (Zars 2000), have been demonstrated (e.g., Lissencephaly 1; Drosophila fragile-X mental retardation 1).

How will the Drosophila developmental neurogenetics system contribute to the understanding and treatment of these challenging human disorders? First, cellular phenotypes, including those detected in primary neuronal culture (Kraftet al. 1998; R. Kraft, J. Kurtis and L. Restifo, unpublished results), could provide bioassays for drug testing. Second, genetic interaction studies will likely identify novel MR genes, as well as reveal the interconnected structure of MR gene pathways.

The degree to which fly mutant phenotypes “match” those of human patients remains to be seen, but it may not matter nearly as much as the genetic pathways involved, as these are likely to guide targeted drug discovery. For example, the fly ortholog of the MR gene ATP2A2 was identified in a screen for enhancers of Notch (Periz and Fortini 1999). The role of the Notch pathway in MR is revealed by Alagille syndrome due to mutations in JAG1 (Krantzet al. 1998), which encodes a ligand of Notch (the Drosophila ligand is Serrate). The biological relevance of genetic interactions in human MR is well demonstrated by some of the Bardet-Biedl syndromes (BBS2 and BBS6; see appendix) in which clinical manifestations result from “triallelic inheritance,” homozygosity at one locus and heterozygosity at another (Katsaniset al. 2001). Genetic interaction tests in Drosophila could help clarify the functional relevance of the physical interaction between mammalian ZIC2 and GLI3 proteins (Koyabuet al. 2001).

The number of MR genes is very large, but they may be involved in a relatively small number of interconnected pathways. If so, a modest number of pharmacological treatment strategies might be effective for many MR patients. In fact, some types of acquired MR might benefit from the same drugs. Diagnoses of hereditary MR are typically made early in life at a time when developmental brain plasticity provides an opportunity for therapeutic intervention. The widespread functional conservation of MR genes in Drosophila indicates that this genetic model system could play a critical role in the discovery of novel treatment strategies for MR.

View this table:

Human mental retardation genes and D. melanogaster homologs


The authors thank Brian Blood for recent literature searches and BLAST analyses to update the list of human MR genes and their Drosophila homologs. The authors are grateful to colleagues David Mount for advice on bioinformatics methods, Terrill Yuhas and Nirav Merchant for computer support, Charles Hedgecock for assistance with computer graphics, and John Meaney and Robert Erickson for helpful discussions about human genetic disease. This work was funded by the National Institutes of Health (grant no. P01 NS028495).


  • Note added in proof: Evaluation of recently updated OMIM entries revealed three more MR genes whose molecular identifications were published prior to September 30, 2003. They are AAAS (OMIM 605378), COH1 (OMIM 607817), and MLC1 (OMIM 605908).

  • Communicating editor: R. S. Hawley

  • Received August 14, 2003.
  • Accepted November 14, 2003.


View Abstract