XLID-Causing Mutations and Associated Genes Challenged in Light of Data From Large-Scale Human Exome Sequencing

Abstract

Because of the unbalanced sex ratio (1.3–1.4 to 1) observed in intellectual disability (ID) and the identification of large ID-affected families showing X-linked segregation, much attention has been focused on the genetics of X-linked ID (XLID). Mutations causing monogenic XLID have now been reported in over 100 genes, most of which are commonly included in XLID diagnostic gene panels. Nonetheless, the boundary between true mutations and rare non-disease-causing variants often remains elusive. The sequencing of a large number of control X chromosomes, required for avoiding false-positive results, was not systematically possible in the past. Such information is now available thanks to large-scale sequencing projects such as the National Heart, Lung, and Blood (NHLBI) Exome Sequencing Project, which provides variation information on 10,563 X chromosomes from the general population. We used this NHLBI cohort to systematically reassess the implication of 106 genes proposed to be involved in monogenic forms of XLID. We particularly question the implication in XLID of ten of them (AGTR2, MAGT1, ZNF674, SRPX2, ATP6AP2, ARHGEF6, NXF5, ZCCHC12, ZNF41, and ZNF81), in which truncating variants or previously published mutations are observed at a relatively high frequency within this cohort. We also highlight 15 other genes (CCDC22, CLIC2, CNKSR2, FRMPD4, HCFC1, IGBP1, KIAA2022, KLF8, MAOA, NAA10, NLGN3, RPL10, SHROOM4, ZDHHC15, and ZNF261) for which replication studies are warranted. We propose that similar reassessment of reported mutations (and genes) with the use of data from large-scale human exome sequencing would be relevant for a wide range of other genetic diseases.

Main Text

Introduction

Intellectual disability (ID, formerly called mental retardation) is a developmental brain disorder commonly defined by an IQ below 70 and limitations in both intellectual functioning and adaptive behavior. ID can originate from environmental causes and/or genetic anomalies, and its incidence in children is estimated to be of 1%–2%.^1,2 As a result of an excess of males affected by ID (the male-to-female ratio is 1.3–1.4 to 1) and the identification of many families presenting with a clear X-linked segregation, much attention has been focused for the last 20 years on genes located on the X chromosome and thus responsible for X-linked ID (XLID, previously known as XLMR) when mutated.^3,4 One of the first genes identified as involved in XLID is FMR1 (MIM 309550), a target of the unstable expansion mutation responsible for fragile X syndrome (MIM 300624); accounting for about 1%–2% of all ID cases, this mutation still remains the most common cause of XLID.^5,6 Since then, the number of genes involved in XLID when mutated has grown exponentially,^3,7,8 from only 11 in 1992 to 43 in 2002 and over 100 genes now identified thank to the efforts of various teams.^4,9,10 Half of the known genes carrying mutations responsible for XLID appear to be associated with nonsyndromic or paucisyndromic forms; the other half are associated with more syndromic forms (i.e., ID associated with defined clinical or metabolic manifestations), which facilitates the identification of causative mutations in the same gene because unrelated probands with comparable phenotypes can be more easily matched. However, the presence of “milder” mutations (in RPS6KA3 [RSK2, MIM 300075] or ARX [MIM 300382], for instance) and/or incomplete penetrance of specific clinical signs in some individuals carrying mutations in genes associated with syndromic ID can blur the distinction between syndromic and nonsyndromic ID.¹¹

Various approaches have been developed for the identification of genes and associated causative mutations responsible for XLID (see Lubs et al.⁴ for a review): (1) positional cloning based on chromosomal rearrangements or copy-number variants (CNVs) affecting the X chromosome, (2) screening of genes located in candidate intervals identified via linkage analysis in large XLID-affected families, (3) direct sequencing of candidate genes with a function or expression pattern that suggests a role in cognition or that fits with metabolic or clinical observations in affected subjects, and (4) high-throughput sequencing allowing screening of mutations in all protein-coding regions of the genome or only in the X chromosome (exome versus X exome).^10,12–14

The validation of potentially damaging mutations in a gene newly associated with XLID requires functional and/or genetic analyses, especially when the identification is based on reporting mutations in very few families or simplex cases. Functional studies are uneven in pertinence and strength. They can include direct assessment of the mutational impact at any of the protein, cellular, or organism levels or functional connection to the disease (e.g., involvement in synaptic organization or plasticity). However, functional tests showing an effect of a candidate mutation at the protein or cellular level do not necessarily imply that this effect is inherently responsible for the disease. Indeed, some false positives can lie in mutations that have been “functionally” validated (see, for instance, SRPX2 [MIM 300642] below).

Genetic validation of a mutation usually includes cosegregation analysis in the proband’s family and testing for the absence of the mutation in a population of control chromosomes. Cosegregation analyses that validate mutations in simplex cases often lack statistical power and can lead to misleading conclusions, even in families with several affected individuals. Some noncausative variants will segregate by chance with the disease only because they lie in the linked candidate region, which often contains many genes. Recently, the identification of genes implicated in ID or autism spectrum disorders (ASDs) has extensively relied on the detection of de novo events, but it is clear now that such events can also be detected in unaffected siblings and hence that the de novo criterion is not sufficient to imply pathogenicity. Moreover, for X-linked genes, because de novo mutations occur more often in the male germline, probands’ mothers are often silent carriers of a transmitted pathogenic mutation that might have occurred de novo in preceding generations. Investigating the maternal grandparents is thus very useful and informative but is often not possible. Lastly, because of the cost-efficiency limitations of Sanger sequencing, the number of control individuals sequenced is often limited to a few hundred individuals, which is therefore insufficient to allow the detection of rare innocuous variants. Some false-positive conclusions about the involvement of some genes in XLID might have arisen notably by the overinterpretation of the pathogenicity of missense variants, and this might concern genes proposed for mutation screening in XLID diagnostic panels.

A new powerful resource, the National Heart, Lung, and Blood Institute (NHLBI) data set of over 6,500 sequenced exomes (available on the Exome Sequencing Project Exome Variant Server [EVS]) can now be used for ascertaining the frequency of potential mutations in large cohorts of adults initially selected for cardiac, lung, or metabolic phenotypes but a priori not enriched with neurological or cognitive defects. This very large project originally aimed to identify genes underlying complex heart, lung, and metabolic disorders and provides detailed exome-variation information on unrelated African American and European American individuals (2,443 males and 4,060 females), amounting to a total of 10,563 X chromosomes.¹⁵ Because participants had to provide written informed consent,¹⁶ one can expect that these cohorts do not contain individuals with moderate or severe ID and that even the mild form is presumably underrepresented as compared to its incidence in the general population. This cohort can hence be considered as a “general population” for the analysis of rare variations in genes with a proposed implication in ID when mutated.

List of 106 Genes with Reported Mutations Involved in XLID

We established a list of 106 genes proposed to be associated with XLID by compiling data from reviews, lists of genes proposed for XLID diagnosis, and recent primary publications (Figure 1). Two of these 106 genes (MTM1 [MIM 300415] and GJB1 [MIM 304040]) appear to be misassociated with ID, given that very few cases were reported with both ID and centronuclear myopathy (MIM 310400) or Charcot-Marie Tooth neuropathy (MIM 302800) and that they carry deletions encompassing other genes in addition to MTM1 or GJB1.^17,18 Both genes should thus be removed from XLID lists. For the remaining 104 genes, we systematically screened the NHLBI EVS data. We assumed that out of the ∼2,500 males of the NHLBI population, there would be fewer than 50 males (2%) potentially affected by ID. Consequently, a previously published XLID mutation reported in two NHLBI males would imply that this mutation is responsible for 4% (2/50) of all ID cases and is therefore more common than the fragile X expansion mutation, which is very doubtful. Such a variant should therefore be considered a nonpathogenic (or low penetrance) rare single-nucleotide variant (SNV) with respect to monogenic forms of ID. The same logic can be applied if truncating variants (nonsense and splice variants) are identified in more than one male from the NHLBI population in a gene associated with ID; this would confer it a major role in ID, which is suspicious. This led us to further assess the association between some genes and ID when (1) truncating variants are observed in males within the NHLBI cohort (Table S1, available online), (2) previously described ID mutations are detected in this population at a frequency inconsistent with a causative effect on cognitive impairment (Tables S2 and S3), and (3) the implication of this gene in XLID relies on a single piece of evidence. For all such genes, we discuss the current evidence of their association with monogenic forms of ID on the basis of both the literature and our EVS data analysis (Supplemental Data and Table 1). All other genes were considered to be convincingly implicated in ID.

Figure 1.

Representation along the X Chromosome of the 106 Genes in which Mutations Have Been Reported in XLID and Classification According to Both the Type and Number of Mutations Reported in OMIM

Genes for which involvement in XLID is convincing are listed on the left side of the chromosome, and genes for which implication in ID either is questionable or needs to be replicated are listed on the right side of the chromosome. GJB1 and MTM1 are linked to diseases with no real ID association and are thus not included in the figure. The color code was designed according to the total number of mutations reported in OMIM. If at least one truncating mutation (splice mutations, nonsense mutations, frameshifts, and large deletions that might encompass several exons) was described, genes are in red (more than five mutations in total), orange (two to five mutations in total), or green (only one mutation). If only missense mutations or mutations involved in expression decrease were identified, genes are in violet (more than five mutations in total), blue (two to five mutations in total), or purple (only one mutation). Asterisks denote genes discussed in this paper but whose implication in ID remains very likely. In parentheses next to each gene are lists in which that gene is included (abbreviations are as follows: L, Lubs⁴; R, Ropers⁸; G, Greenwood Genetic Center; E, Emory Genetics Laboratory; and A, Ambry Genetics). References for the genes not reported in any list are as follows: [1] Honda et al., [2] Houge et al., [3] Voineagu et al., [4] Huang et al., and [5] Witham et al.

Table 1.

List of All 28 Genes Discussed and Associated Evidence Supporting or Questioning Their Involvement in ID

Gene (RefSeq Accession Number)	Listsa	Reason for Highlightb	Identification Methodc	ID Structural Variations	ID Point Mutations	Occurrence of Mutations in EVSd	Truncated Variants in EVS	dN/dS Ratio (#N/#S)	Protein Size	Implication in ID
AGTR2 (NM_000686.4)	L, R, G, E, A	T	1	(X; 7) translocation in a female19	c.402del (p.Phe134Leufs∗5)19	(see Table S4)	yes (1 F)	0.731 (19/8)	363	highly questionable (already controversial)20–22
					c.62G>T (p.Gly21Val)19,23	10 M, 23 F (htz)
					c.971G>A (p.Arg324Gln)19	4 M, 24 F (htz)
					c.1009A>G (p.Ile337Val)19	NF
					c.157A>T (p.Ile53Phe)24	NF
					c.572G>A (p.Gly191Glu)23	NF
MAGT1 (IAP) (NM_032121.5)	L, R, G, E	T	3	-	c.1028T>G (p.Val311_343Gly)25	5 M, 8 F (htz)	yes (1 M)	0.453 (15/10)	367	highly questionable
NXF5 (NM_032946.2)	L, R, G, A	T	1	inv(X)(p21.1;q22) in two males;26,27 deletion disrupting the NXF cluster;28,29 duplication at Xq22.1 disrupting30 and duplicating31NXF5 in one female and one male, respectively	-	-	yes (8 M)	0.783 (30/11)	365	highly questionable
ZNF674 (NM_001039891.2)	L, R, E, A	T	1	deletion encompassing ZNF673, ZNF674, RP2, SLC9Z7, and CHST7 in one male;32 duplication encompassing ZNF673, ZNF674, and CHST733	c.352G>T (p.Glu118∗)32	NF	yes (19 M)	0.750 (28/10)	581	highly questionable (already controversial)34
					c.1235C>A (p.Pro412Leu)32	NF
					c.1028C>T (p.Met343Thr)32	49 M, 16 F (hmz), 223 F (htz)
ZNF41 (NM_007130.2)	L, R, G, E, A	M	1	(X; 7) translocation disrupting ZNF4135	c.332C>T (p.Pro111Leu)35 c.73−42A>C (p.?)35	2 M, 8 F (htz)	yes (1 M, 1 F)	0.696 (45/17)	779	highly questionable
						5 M, 9 F (htz)
ZNF81 (NM_007137.3)	L, R, G, E, A	M	1	(X; 9) translocation in one female;36 1.3 Mb duplication encompassing >30 genes at Xp11.23–p11.3 in one male;37 335 kb microduplication bearing two other genes at Xp11.2–p11.3 in one male38	c.536G>A (p.Ser179Asn)36	NF	yes (1 M)	0.595 (25/11)	661	questionable
ARHGEF6 (NM_004840.2)	L, R, G, E, A	M	1	(X; 21) translocation in one male39	c.166−11T>C (p.?)39	5 M, 13 F (htz)	no	0.456 (28/18)	776	questionable
ATP6AP2 (NM_005765.2)	L, R, G, E, A	T	2	-	c.321C>T (p.Asp107Asp) (affects splicing)40	NF	yes (1 M)	0.444 (14/10)	350	questionable
SRPX2 (NM_014467.2)	L, R, G, E, A	M	2	-	c.980A>G (p.Asn327Ser)41	3 M, 8 F (htz)	no	0.728 (31/14)	465	questionable
					c.215A>C (p.Tyr72Ser)41	1 M
ZCCHC12 (SIZN1) (NM_173798.2)	E	M	3	-	c.19C>T (p.Arg7Cys)42	5 M, 28 F (htz)	no	0.406 (19/14)	402	questionable
					c.1031C>T (p.Thr344Ile)42	NF
IGBP1 (NM_001551.2)	L, G, E	I	2	-	5′ UTR 2 bp substitution affecting IGBP1 expression43	not covered	no	0.487 (19/11)	339	never replicated
KIAA2022 (NM_001008537.2)	L, R, G, E, A	I	1	inv(X)(q13;p22) in two related males44	-	NA	no	0.441 (54/34)	1,516	never replicated
KLF8 (NM_007250.4)	L, R, G, E	I	1	(X; 21) translocation leading to a loss of KLF8 expression in a woman45	-	NA	no	0.831 (16/6)	359	never replicated
NLGN3 (NM_181303.1)	L, R, G, E, A	I	3	-	c.1411C>T (p.Arg451_471Cys) in two ASD brothers46	NF	no	0.209 (17/27)	848	never replicated
ZDHHC15 (NM_144969.2)	L, G, E	I	1	(X; 15) translocation leading to a loss of ZDHHC15 expression in a female47	-	NA	no	0.628 (17/8)	337	never replicated
ZNF261 (ZMYM3) (NM_201599.2)	R	I	1	(X; 13) translocation disrupting ZMYM3 in one female48	-	NA	no	0.232 (31/42)	1,370	never replicated
MAOA (NM_000240.3)	L, R, G, E, A	I	1	MAOA-MAOB-NDP deletion in subjects with severe ID49	c.886C>T (p.Gln296∗)50	NF	no	0,359 (14/12)	527	never replicated
CCDC22 (NM_014008.3)	-	I	3	-	c.49A>G (p.Thr17Ala)51	NF	no	0.809 (40/17)	627	awaiting replication
CLIC2 (NM_001289.4)	-	I	4	-	c.303C>G (p.His101Gln)52	NF	no	0.346 (6/5)	247	awaiting replication
CNKSR2 (NM_014927.3)	-	I	1	partial deletion of CNKSR253	-	NA	no	0.340 (22/19)	1,034	awaiting replication
FRMPD4 (NM_014728.3)	-	I	1	partial duplication of FRMPD454	-	NA	no	0.383 (57/46)	1,322	awaiting replication
HCFC1 (NM_005334.2)	-	I	2	-	chrX: 152890455A>G55	NA	no	0.227 (50/80)	2,035	awaiting replication
					c.674G>A (p.Ser225Asn)55	NF
NAA10 (NM_003491.2)	L	I	4	-	c.109T>C (p.Ser37Pro)56	NF	no	0.074 (2/8)	235	likely in Ogden syndrome but needs further replication in nonsyndromic ID
					c.346C>T (p.Arg116Trp)13	NF
RPL10 (NM_006013.3)	L, R, G, E, A	I	3	-	c.616C>A (p.Leu206Met) (ASD)57	NF	no	0.154 (2/4)	214	awaiting replication
					c.639C>G (p.His213Gln) (ASD)57,58	NF
SHROOM4 (NM_020717.3)	L, R, G, E, A	I	1	X-autosome translocations in two females;59 Xp11.22 deletion bearing SHROOM4 in one family54	c.3266C>T (p.Ser1089Leu)59	NF	yes (1 F)	0.918 (86/28)	1,493	awaiting replication
					c.1422A>G (p.Glu474Glu)59	NF
HUWE1 (NM_031407.5)	L, R, G, E, A	I	1	microduplications encompassing HSD17B10 and HUWE160,61 or HUWE1 alone62 in a total of 16 unrelated families	c.12037C>T (p.Arg4013Trp)61	NF	no	0.201 (67/105)	4,374	likely
					c.8942G>A (p.Arg2981His)61	NF
					c.12559C>T (p.Arg4187Cys)61	NF
					c.2849T>A (p.Val950Asp) (ASD)63	NF
PTCHD1 (NM_173495.2)	L	M	1	160 kb deletion leading to a null PTCHD1 in dizygotic ASD twin brothers;64 50–390 kb inherited deletions disrupting PTCHD1 in six ASD male probands;65 90 kb deletion disrupting PTCHD1 in three related ID males;60,66 200 kb deletion in two ID brothers67	c.517A>G (p.Ile173Val)66	2 M, 2 F (htz)	no	0.366 (29/25)	888	likely
					c.583G>A (p.Val195Ile)66	NF
					c.1008_1009delinsTA (p.MetLeu336_337IleIle)66	NF
					c.217C>T (p.Leu73Phe)66	1 F (htz)
					c.1436A>G (p.Glu479Gly)66	NF
					c.1409C>A (p.Ala470Asp)66	NF
					c.1076A>G (p.His359Arg)66	NF
SYN 1 (NM_006950.3)	L, R, G, E, A	M	2	-	c.1067G>A (p.Trp356∗)68	NF	no	0.347 (12/12)	705	likely
					c.1663C>T (p.Gln555∗)69	not covered
					c.152C>G (p.Ala51Gly)69	11 M, 30 F (htz)
					c.1648G>A (p.Ala550Thr)69	not covered
					c.1699A>G (p.Thr567Ala)69	not covered

Abbreviations are as follows: L, Lubs et al.;⁴ R, Ropers et al.;⁸ G, Greenwood Genetic Center (see Web Resources); E, Emory Genetics Laboratory (see Web Resources); A, Ambry Genetics (see Web Resources); T, truncating variants reported in EVS; M, ID mutations reported in EVS; I, one piece of evidence or study implicating this gene in ID; htz, heterozygous; hmz, homozygous; M, number of males carrying the mutation; F, number of females carrying the mutation; NF, not found; NA, not applicable; N, number of nonsynonymous variants; and S, number of silent variants.

^aLists in which genes are included.

^bReason why we discuss this gene in the paper.

^cApproach used for the initial identification, consistent with the information from the text: 1, positional candidate found through rearrangement or CNV; 2, positional candidate located in a linked region; 3, functional candidate; and 4, through exome or X-exome sequencing.

^dOccurrence of ID mutations in EVS.

Presence in EVS of Truncating Variants Affecting Genes Involved in XLID

We first analyzed the NHLBI cohort data set to evaluate the presence, notably in males, of truncating SNVs in genes with a proposed implication in ID (Table S1). We did not include in our analysis the small indels that have been recently added to EVS (v.0.0.15, October 31, 2012) because the frequency of some of them indicates that they are presumably false-positive observations, either due to artifacts inherent from the sequencing technology or to difficulties in reliably annotating small indels (Table S4).⁷⁰ For instance, the frameshift c.1777_1780del (p.Leu593Phefs^∗7) has been reported in NLGN4X (MIM 300427) in 20 homozygous females but only in a single heterozygous one, which departs from Hardy-Weinberg equilibrium. A total of 21 truncating SNVs are present in EVS and are distributed among 15 of the 104 genes in the XLID list (Table S1). Truncating variants found at the heterozygous state in females are not inconsistent with an implication of these genes in XLID. Similarly, hemizygous truncating variants that affect only some specific splicing isoforms do not question the implication of such genes either. When a unique truncating variant affecting the main isoform of a gene is identified in a single male (such as one in MAGT1 [MIM 300715], ZNF41 [MIM 314995], ZNF81 [MIM 314998], or ATP6AP2 [MIM 300556]), the interpretation is more ambiguous, and this alone cannot rule out the gene’s role in ID. However, two different truncating variants have been reported in NXF5 (MIM 300319) in eight males and in ZNF674 (MIM 300573) in 19 males, and this seriously challenges the implication of these genes in monogenic XLID (see below and Supplemental Data).

Presence in EVS of Known XLID Mutations

We then investigated the presence among the NHLBI cohort of all XLID mutations identified in the 104 genes, which are listed in OMIM or were retrieved by a PubMed search (Table S2), and we found 22 such mutations (Table S3). All but two (c.1186C>T [p.Pro396Ser] in CASK and c.217C>T [p.Leu73Phe] in PTCHD1, identified in single heterozygous females) were detected in both males and females and are probably noncausative variants instead of real disease-causing mutations. The majority of these missense mutations (16/20) that now appear to be innocuous variants are predicted to have benign or unknown consequences on protein function. The pathogenicity of some of these missense mutations (c.515C>A [p.Pro172His] in TSPAN7, c.3872C>T [p.Pro1291Leu] in FLNA, and three missense mutations in DMD) was already questioned by the authors.^71–73 Concerning the FGD1 (MIM 300546; involved in Aarskog-Scott syndrome [AAS (MIM 305400)]) c.935C>T (p.Pro312Leu) mutation, which was detected twice in NHLBI males, the phenotype of the reported family did not fit with classical AAS, already suggesting that it might not be a truly pathogenic mutation.⁷⁴ Even more surprising was the identification of 18 males carrying NDUFA1 mutation c.94G>C (p.Gly32Arg), which was demonstrated to have functional consequences on protein activity.^75,76 Such observations question these specific mutations, but they do not raise doubt about the involvement of the cognate genes in ID given that several other convincing mutations have been reported in such genes (TSPAN7 [MIM 300096], FLNA [MIM 300017], FGD1, NDUFA1 [MIM 300078], and PDHA1 [MIM 300502]). However, for other genes (SRPX2, ZNF41, ARHGEF6 [MIM 300267], etc.), the presence of mutations at a relatively high frequency in the general population challenges their involvement in ID (see the following sections and Table 1).

Questioning Monogenic Involvement of 28 Specific Genes in XLID

Our investigation focused on 28 genes proposed to play a role in XLID when mutated and for which there is a paucity of confirmatory studies or some contradictory elements arising from the analysis of the EVS data (Table 1). This allowed us to classify them into five categories. We highlight five genes that appear very unlikely to play a role in ID with high penetrance in males (AGTR2 [MIM 300034], MAGT1, NXF5, ZNF674, and ZNF41; we refer to these as “highly questionable”) and five others for which implication in ID appears suspicious (ZNF81, ARHGEF6, ATP6AP2, SRPX2, and ZCCHC12 [SIZN1, MIM 300701]; we refer to these as “questionable”). For 15 genes, additional confirmation studies are required, but no contradictory findings were found in EVS. These include some genes that have never been replicated since their first publication more than 8 years ago (IGBP1 [MIM 300139], KIAA2022 [MIM 300524], KLF8 [MIM 300286], NLGN3 [MIM 300336], ZDHHC15 [MIM 300576], ZMYM3 [MIM 300061], and MAOA [MIM 309850]; we call these “never replicated”) or others that have been recently described but have not yet been replicated (CCDC22 [MIM 300859], CLIC2 [MIM 300138], CNKSR2 [MIM 300724], FRMPD4 [MIM 300838], HCFC1 [MIM 300019], NAA10 [MIM 300013], RPL10 [MIM 312173], and SHROOM4 [MIM 300579]; we refer to these as “awaiting replication”). Finally, for three genes—HUWE1 (MIM 300697), PTCHD1, and SYN1 (MIM 313440)—our analysis and/or recent confirmation reports support a real implication in ID (we refer to these genes as “likely”). We detail here current arguments and evidence for the five highly questionable genes and also present a few others as examples of the four other categories (SRPX2 and ARHGEF6 as questionable, MAOA as never replicated, HCFC1 as awaiting replication, and HUWE1 as likely). The detailed information for the remaining 18 genes is given in the Supplemental Data.

AGTR2: Highly Questionable

After a de novo balanced (X;7)(q24;q22) translocation disrupting AGTR2 (angiotensin II receptor 2) was identified in one female with ID, it was screened in a cohort of 590 male ID probands, 38 of whom were in families affected by probable or possible XLID.¹⁹ This led to the identification of (1) a 1 bp deletion causing a frameshift (c.402del [p.Phe134Leufs^∗5]) in a family affected by probable XLID and in another simplex case and (2) three missense mutations (c.62G>T [p.Gly21Val], c.971G>A [p.Arg324Gln], and c.1009A>G [p.Ile337Val]) absent from 510 control X chromosomes. A further study of 57 males with ID detected two individuals with the p.Gly21Val or p.Ile53Phe changes.²⁴ Other extensive studies (25 XLID-affected families showing coherent linkage to the locus, 116 affected sibling pairs, and 224 simplex cases,^9,20 as well as another set of families¹⁰) failed to detect any additional mutations (besides the p.Gly21Val variant and a missense variant that did not show full cosegregation in a family that was later shown to carry a SOX3 [MIM 313430] mutation). More recently, screening of 203 Japanese males with ID detected an additional inherited missense change (c.572G>A [p.Gly191Glu]) in a boy showing severe ID, pervasive developmental disorder, and epilepsy.²³ Meanwhile, both the p.Phe134Leufs^∗5 frameshift and the p.Gly21Val variant were reported—in rarely cited publications—in males from control cohorts, suggesting that they are unlikely to be causative.^21,22 This is now supported by the NHLBI data, given that two of the three initially reported missense variants (p.Gly21Val and p.Arg324Gln) are present in ten and four males, respectively. Moreover, although we did not include the indel data from EVS, we noted that the initial frameshift mutation was observed in 7 of 1,770 European American males (minor allele frequency [MAF] = 0.4%), but not in African American individuals, thus excluding a possible sequencing artifact (Table S4). Consequently, EVS data support the serious doubts previously raised about the implication of AGTR2 in ID, despite the suggestive phenotype of the AGTR2-knockout mouse showing altered memory capacities and abnormal dendritic-spine morphology.⁷⁷

NXF5: Highly Questionable

Mutant NXF5 (nuclear RNA export factor 5) was suspected to cause ID after pericentric inversion inv(X)(p21.1;q22), leading to the loss of NXF5 expression, was identified in an ID-affected male.^26,27 The observation of two nonsense variants (c.162G>T [p.Cys54^∗] and c.958G>A [p.Arg320^∗]) affecting a total of eight males in EVS raises serious doubts about the involvement of this gene in ID, despite some indication of its function in the brain. This gene was more recently found to be deleted or duplicated in other ID individuals, but never alone.^28,29,31

MAGT1: Highly Questionable

After the identification of an ID-causing mutation in TUSC3 (encoding a putative subunit of an oligosaccharyltransferase complex [MIM 601385]), the sequencing of its X-linked paralog gene, MAGT1 (magnesium transporter 1, formerly called IAP), in XLID-affected families revealed one missense change segregating with the ID status (LOD score 1.8) and absent from more than 267 control chromosomes.²⁵ This variant (p.Val311Gly but annotated c.1028T>G [p.Val343Gly] in EVS) was detected in 13 out of 10,557 X chromosomes from the NHLBI cohort. Moreover, one male from the same population carries a nonsense variant in this gene. Although the implication of TUSC3 in recessive ID appears well supported from the identification of additional consanguineous families affected by mutations in this gene,⁷⁸ no ID-related mutation has been reported in MAGT1 since 2008, and its disruption has been recently associated with several forms of X-linked immunodeficiency without neurological manifestations.⁷⁹ In light of these observations, there is no current support for the implication of MAGT1 in ID.

ZNF674: Highly Questionable

After the identification of one ID-associated CNV encompassing ZNF674 (zinc-finger protein 674) and four other genes (RP2 [MIM 300757], SLC9A7 [MIM 300368], CHST7 [MIM 300375] and ZNF673 [MIM 300585]), its screening in additional families led to the identification of a nonsense mutation (c.352G>T [p.Glu118^∗]) segregating with the ID phenotype in a large family (LOD score 2.51).³² Further screening of 28 families with nonsyndromic XLID exhibiting coherent linkage to the region and 309 simplex cases detected one inherited missense change (c.1235C>A [p.Pro412Leu]) in two related males and a second (c.1028C>T [p.Thr343Met]) in another small family.³² Males carrying the p.Pro412Leu missense variant also displayed partial trisomy 21 and monosomy 18 as a result of an unbalanced translocation, which precluded conclusions about the causative effect of this variant. Also, the authors concluded that the p.Thr343Met missense variant was unlikely to be pathogenic (despite its being absent from 354 controls) because it did not affect a highly conserved residue and because the replacing methionine was present in the ZNF674 chimpanzee ortholog. EVS data indicate that this variant is present in 49 males, 223 carrier females, and 16 homozygous females, fully excluding its implication in ID. The implication of ZNF674 in ID was recently questioned when it was found to be deleted together with RP2 (in retinitis pigmentosa [MIM 312600]) in two families with X-linked retinal dystrophy without cognitive impairment.³⁴ Lastly, given that two nonsense mutations (c.1324C>A [p.Asp442^∗] and c.601G>A [p.Arg201^∗]) have been reported in the NHLBI cohort and that one is present in 19 hemizygous males, ZNF674 seems unlikely to be involved in monogenic ID.

ZNF41: Highly Questionable

ZNF41 (zinc-finger protein 41) was found to be disrupted by a de novo balanced (X;7) translocation in a girl with severe developmental delay.³⁵ Subsequent screening of 210 probands from families affected by probable or possible XLID revealed one missense change (c.332C>T [p.Pro111Leu]) and one intronic variant (c.479−42A>C) causing a splicing alteration; both are absent from over 400 control X chromosomes, and both show cosegregation in the families. Although the authors concluded carefully that ZNF41 could be an ID candidate, it is now included in most lists of ID-associated genes. However, two truncating variants have been reported in one male and one female in EVS, and the two putative point mutations initially described by Shoichet et al. have been detected in a total of seven males at a frequency too high to be fully penetrant ID mutations (see Table 1).³⁵ No other ZNF41 mutation has been reported since the initial 2003 report.

SRPX2: Questionable

In 2006, Roll et al. reported a family affected by probable dominant XLID, rolandic epilepsy, and speech dyspraxia.⁴¹ Sequencing the genes of the linkage region (LOD score 3.01) revealed only a missense variant (c.980A>G [p.Asn327Ser]) in SRPX2 (sushi-repeat-containing protein, X-linked 2). Another missense variant (c.215A>C [p.Tyr72Ser]) was detected in one epileptic male and his female relatives, all of whom showed additional bilateral perisylvian polymicrogyria. This latter variant was reported to increase the affinity of the encoded SRPX2 for the urokinase-type plasminogen activator receptor.⁸⁰ The initial p.Asn327Ser alteration was shown to affect protein glycosylation and to cause partial retention of the altered protein within the endoplasmic reticulum in a transfected-cell model.⁴¹ Despite their observed functional consequences, both variants are present in EVS in three males and eight females (all of European American origin; p.Asn327Ser) and in one male (p.Tyr72Ser). Accordingly, 0.28% of the EVS European American population carries one of these two variants, which considerably weakens the evidence supporting a role for SRPX2 in epilepsy, speech, and cognition. It is most unlikely that mutations in SRPX2 are a major cause of epilepsy given that no mutation has been reported in this gene since 2006, despite its sequencing in 100 individuals with a similar phenotype of epilepsy and speech disorder (D. Sanlaville, G. Lesca, and P. Szepetowski, personal communication). Because SRXP2 seems to play a role during brain development and was recently shown to influence neuronal migration in the developing cerebral cortex in rats (P. Szepetowski, personal communication), we cannot, however, exclude that other drastic alterations of SRXP2 function might cause cognitive impairment.

ARHGEF6: Questionable

ARHGEF6 (α-PIX or Cool-2; rho guanine nucleotide exchange factor 6) was identified as disrupted via an inherited balanced (X;21) translocation in a male with ID.³⁹ In the same study, subsequent screening of 119 unrelated males with nonsyndromic ID allowed the identification of a cosegregating intronic mutation (c.166−11T>C) affecting splicing in a large XLID-affected family; it was absent from 170 control chromosomes. However, this intronic variant has been reported in EVS in five males, all of European origin (MAF = 0.26%). An inherited duplication of ARHGEF6 was also reported in two brothers with moderate ID, but later analysis showed that the duplication spanned 1.8 Mb and encompassed 24 genes.^81,82 In the Tarpey et al. study, two missense changes were detected in single families: c.362G>A (p.Arg121His), which is found in 14 NHLBI males, and c.992A>G (p.Tyr331Cys), which is not observed in EVS. However, the latter missense variant is not in the list of likely pathogenic mutations and was not further commented on by the authors. Engineered αPix (Arhgef6)-deficient mice present altered hippocampal neuronal connectivity, impaired synaptic function, and additional cognitive deficits.⁸³ However, because no other pathogenic mutation has been described since 2003 and given the unlikely pathogenicity of the initially reported splice variant, the implication of ARHGEF6 in ID still remains unclear.

MAOA: Never Replicated

The combined loss of MAOA (monoamine oxidase A) and MAOB (MIM 309860), along with the deletion of the gene associated with Norrie disease, NDP (MIM 300658), has been described in some probands with a continuous syndrome and presenting with severe ID.⁴⁹ In 1993, Brunner et al. described a large Dutch family affected by X-linked mild ID and prominent behavioral abnormalities associated with disturbances in monoamine metabolism and segregating with a truncating mutation (c.886C>T [p.Gln296^∗]) in MAOA.^30,50 More recently, deletions of MAOA with MAOB alone were described in two males with severe developmental delay or ID.^60,84 Mouse models with total Maoa inactivation or with a spontaneous mutation present indeed enhanced aggressive behavior and additional “autistic-like” features.⁸⁵ To our knowledge, no other clearly pathogenic mutation has been reported in MAOA in other individuals in the past 20 years,⁸⁶ even though one missense variant (c.812A>T [p.Asn271Ile]) was reported in a proband with ASD.⁸⁷ However, we recently identified a pathogenic missense mutation (c.797_798GC>TT [p.Cys266Phe]) associated with deficient MAOA activity in one family affected by autism, ID, and abnormal behavior, confirming the findings from 1993 (A.P., unpublished data). A third family harboring a MAOA mutation was recently identified (M. Field, J. Gecz, and V.M. Kalscheuer, personal communication).

HCFC1: Awaiting Replication

The mutation responsible for ID in the first family reported as linked to the MRX3 locus remained unsolved for a long time after the initial linkage analysis of 1991. A targeted massive resequencing of the genomic linkage interval recently revealed a regulatory point mutation in a functional binding site for the YY1 transcription factor in the HCFC1 (host-cell factor C1) promoter; this mutation leads to an upregulation of the gene’s expression in lymphoblastoid cells. The screening of extra unsolved XLID-affected families identified an additional missense mutation (c.674G>A [p.Ser225Asn]) segregating with the disease.⁵⁵ Both mutations are absent from EVS, and other unique missense variants were also described in a female with schizophrenia and a boy with ASD.⁸⁷ However, because HCFC1 encodes a large protein of 2,035 amino acids, the probability of identifying a missense change is high (22 missense variants predicted to be possibly or probably damaging are present in EVS). Additional evidence would therefore be useful for definitely confirming the implication of HCFC1 in ID.

HUWE1: Likely

HUWE1 (Hect, UBA, and WWE domains-containing protein 1) encodes a very large protein (4,375 amino acids) with an E3 ubiquitin ligase function. Array comparative genomic hybridization analysis initially detected duplications of HUWE1 and HSD17B10, both of which showed overexpression in affected probands in six families afflicted with predominantly nonsyndromic XLID; two of these families were large XLID-affected families.⁶¹ Four similar cases were reported in an independent replication study.⁶⁰ A further CNV analysis showed that de novo duplications covering HUWE1 alone were also associated with simplex cases of ID, indicating that increased expression of this gene is deleterious.⁶² In the initial study, three different missense mutations affecting highly conserved residues and cosegregating with the ID status were identified in three unrelated XLID-affected families (c.12037C>T [p.Arg4013Trp] in a large Australian family [LOD score = 3.31]⁸⁸ and c.8942G>A [p.Arg2981His] and c.12559C>T [p.Arg4187Cys] in two smaller families; all three mutations were absent from 750 controls). More recently, a de novo missense variant (c.2849T>A [p.Val950Asp]) was observed in one ASD male but was absent in his less severely affected brother, suggesting either a different etiology or complex inheritance.⁶³ None of these variants is present in EVS. Functional studies, notably in mouse models, indicate that HUWE1 plays a role in the control of neurogenesis in the developing brain via the N-Myc pathway.^89–91 In mice, targeted inactivation of Huwe1 in the CNS or only in cerebellar granule neuron precursors and radial glia resulted in high neonatal lethality, which could explain the absence of truncating mutations reported in individuals.^89,90 All these observations further support the role of mutations in HUWE1 in XLID. However, one should be very cautious in ascribing a pathogenic role to a HUWE1 missense change identified in an individual with ID because 21 different such variants were observed in males in the NHLBI cohort (ten of these variants are qualified as possibly or probably damaging); this number is consistent with the large target size of the protein-coding sequence.

dN/dS Ratio Distribution among the Genes Involved in XLID

The latest publication on EVS data reported a total of 285,960 nonsynonymous and 188,975 silent variants on 2,440 exomes.¹⁶ Interestingly, compared to the full genome, genes implicated in XLID tend to accumulate fewer nonsynonymous variants (nonsynonymous = 2,623; silent = 2,195; p = 0.04). In order to estimate the evolutionary pressure exerted on those different genes, we used EVS to compute an intrahuman dN/dS ratio, which we calculated by using DnaSP.⁹² For this ratio, the number of nonsynonymous variants observed in EVS over the number of total putative nonsynonymous positions in the studied transcript is divided by the analogous ratio for synonymous variants (Table S5). Accordingly, a strong selection pressure is expected to engender a low dN/dS ratio, and a truly associated gene can be expected to have such a low ratio in a general population. In contrast, genes that show a high dN/dS ratio are less likely to be disease related, especially if the disease affects fitness. We generated box plots of the dN/dS ratios for the 104 selected genes with supposed involvement in XLID according to the category we classed them into (Figure S1). Interestingly, the mean dN/dS ratios for the ten genes in which mutations associated with ID are questioned was significantly higher than that for the validated genes (μ = 0.604 ± 0.150 versus μ = 0.359 ± 0.255; Student’s t test p = 0.0004). Conversely, no significant difference was observed between the validated and the “awaiting replication” pools (μ = 0.427 ± 0.259, Student’s t test p = 0.35) that might contain genes with either real or no implication in ID. Among the 11 genes with a high dN/dS ratio above the 90^th percentile (0.723; Table S5), four belong to the “questionable” class (AGTR2, ZNF674, SRPX2, and NXF5) and three are considered to be “awaiting replication” (CCDC22, KLF8, and SHROOM4).

Discussion

This work demonstrates that among the 100 genes described in the past as involved in XLID (on the basis of available data sets and sequencing technologies), at least 10 (10%) appear doubtful and 15 others should be considered with caution until validation by replication studies. The involvement in ID of two of these genes, AGTR2 and ZNF674, was already questioned in specific publications that were seldom cited.^22,34 Curiously, some fully validated genes are missing from several XLID diagnostic panels (NSDHL [MIM 300275], MBTPS2 [MIM 300294], SLC6A8 [MIM 300036], etc., Figure 1), whereas others that have been described in only a few studies and whose implication in ID is not fully convincing are systematically included (AGTR2, MAGT1, ZNF41, and SRPX2). This work also highlights that even in a well-known ID-associated gene, the pathogenicity of a novel nonsynonymous variant should be inferred with caution given that some reported nonsynonymous “mutations” in NDUFA1, TSPAN7, or PTCHD1 are likely to be nonpathogenic (or at least not fully penetrant), and this is of particular relevance for diagnostic applications (see also HUWE1 section).⁹³

The number of genes involved in ID when mutated is rapidly growing as a result of the exome-sequencing analysis of cohorts of affected individuals.^12–14 Although the discovery of mutations in additional genes represents an important step for further understanding the underlying genetics of ID, there is a clear need for replication studies to validate these findings (such as the recent work by O’Roak et al.).⁹⁴ Recently, RAB40AL (MIM 300405), not included in our list, was found to be associated with Martin-Probst syndrome (MPS [MIM 300519]), a disorder characterized by deafness, cognitive impairment, short stature, and distinct craniofacial dysmorphisms, with the identification (by parallel whole-genome, whole-exome, and X-exome sequencing) of a 2 nt substitution (c.176_177AC>GA) leading to a missense change (p.Asp59Gly) shown to disrupt protein localization.⁹⁵ However, this variation is present in two males in the EVS population. Also, a few months after this initial publication, Iqbal et al. reported it in four other ID-affected families; however, it was not related to the disease in at least three of the families, questioning the involvement of RAB40AL in the MPS phenotype.⁹⁶ Some other recent publications of exome sequencing in ID and ASD simplex cases identified de novo truncating variants suggested by the authors to be most likely involved in autosomal-dominant ID or ASD in genes not previously associated with these diseases. However, for ADAM33, which was initially considered a candidate gene for dominant forms of ASD after a de novo truncating mutation was identified in an ASD proband, three nonsense variants and two splice variants were reported in a total of five carrier individuals from the NHLBI cohort.⁹⁷ Similarly, six nonsense and three splice variants were detected in a total of 11 carriers in ACACB, which was also proposed as a likely candidate for dominant ASD.⁹⁸ This illustrates that an analysis similar to ours should be performed on all proposed autosomal genes for both dominant and recessive forms of ID.

We believe that this approach should be extended to other Mendelian diseases with high genetic heterogeneity, but the way to assimilate the NHLBI data has to be adapted carefully and specifically for each disease by the integration of both its frequency and proposed transmission mode (see Table 2). While this work was being completed, a thorough similar review of mutations previously associated with cardiomyopathy revealed the presence of about 15% of them in EVS, showing a cumulative frequency inconsistent with an implication in monogenic forms of the disease.⁹⁹ However, the NHLBI cohort cannot really be considered a control cohort with respect to cardiomyopathies or other cardiac diseases given that many individuals were recruited through the NHLBI because of cardiac phenotypes.

Table 2.

Proposed Guidelines for the Assessment of a Variant or a Gene Potentially Involved in ID or in Another Rare or Genetically Heterogeneous Disease^a

	X-Linked or Dominant Form	Recessive Formb
Involvement of a Variant in a Disease
Supports	not found in any malec or individual of the GP	found at an allele frequency ≤ $\sqrt{i}$
	segregates with the disease in the familyc
	appears de novo in a simplex case
Inconclusive	found in one malec or individual of the GP
Does not support	found in more than one malec or individual of the GP	found in GP at an allele frequency > $\sqrt{i}$
Association between a Gene and a Disease
Supports	additional candidate variant(s) in unrelated individuals with similar specific phenotype
	If candidate mutation(s) = protein truncation(s): - no truncating variant is reported in malesc or individuals of the GP	-
	If candidate mutation(s) = missense(s): - a low level of missense mutations predicted to be damaging is reported in malesa or individuals of the GP	-
Inconclusive	If candidate mutation(s) = protein truncation(s): - one truncating variant in one malec or individual of the GP - truncating variants in the GP affect only the last amino acids or one of several functional isoform(s)	-
Does not support	If candidate mutation(s) = protein truncation(s): - several truncating variants in malesc or individuals of the GP	homozygous truncating or damaging variants are reported in individuals of the GP
	If candidate mutation(s) = missense(s): - high level of missenses in the GP (high dN/dS ratio, e.g., >0.7)	truncating or damaging variants are reported in the GP at allele frequency > $\sqrt{i}$

The ID example (incidence = 2%): we do not expect one mutation responsible for more than 0.1% of ID cases, so i = 0.02 × 0.001 = 1 / 50,000 in the GP. Therefore, the variation should not be found at a frequency > $\sqrt{1/50,000}$ = 0.0045, meaning no more than 38 times in 8,600 European American chromosomes. Most true mutations should be even rarer given mutation heterogeneity at a given locus; for instance, the most frequent mutation in phenylketonuria (c.1222C>T [p.Arg408Trp]) is found 15 times in 8,600 European American chromosomes. The following abbreviations are used: GP, general population (e.g., EVS); and i, estimated disease incidence that can be caused by mutations in a single gene after postulated nonallelic genetic heterogeneity is taken into account.

^aFor a phenotype not enriched in the cohort used as a GP.

^bCaution: geographic origin of the GP should match that of individuals with the candidate mutation.

^cFor an X-linked disease.

Because data from projects such as the NHLBI Exome Sequencing Project are available, one should check for the absence of each candidate mutation, especially if it is a missense change. In the case of a novel candidate mutation in a gene never associated with a disease, one should also analyze other information provided by EVS, such as the presence of truncating variants and the dN/dS ratio. Indeed, this ratio is a good indicator of the selection pressure exerted on a particular gene, and if the candidate mutation results in a missense change, one should remember that there is a higher likelihood of identifying a neutral nonsynonymous change in a gene with a high dN/dS ratio.

However, one important limitation of the EVS is the impossibility of going back to the cognitive or neurologic phenotype of individuals in whom rare potentially pathogenic variants were detected. Similarly to existing CNV databases, such as Decipher,¹⁰⁰ a public database of rare exome variants linked to some phenotypic descriptions would be warranted, especially given the overwhelming amount of such sequencing information that is currently being generated in research and that will be soon engendered for diagnostic purposes of ID and autism.

Web Resources

The URLs for data presented herein are as follows:

• 1000 Genomes, http://www.1000genomes.org/

• Ambry Genetics, http://www.ambrygen.com

• dbSNP, http://www.ncbi.nlm.nih.gov/SNP/

• Emory Genetics Laboratory, http://genetics.emory.edu/egl/tests

• Euro-MRX Consortium, http://www.euromrx.com

• Greenwood Genetic Center, http://www.ggc.org

• Human Genome Variation Society, http://www.hgvs.org/mutnomen/

• NHLBI Exome Sequencing Project (ESP) Exome Variant Server, http://evs.gs.washington.edu/EVS/

• Online Mendelian Inheritance in Man (OMIM), http://www.omim.org

• PubMed, http://www.ncbi.nlm.nih.gov/pubmed/

• RefSeq, http://www.ncbi.nlm.nih.gov/RefSeq