Meta-Analysis of 13 Genome Scans Reveals Multiple Cleft Lip/Palate Genes with Novel Loci on 9q21 and 2q32-35

Abstract

Isolated or nonsyndromic cleft lip with or without cleft palate (CL/P) is a common birth defect with a complex etiology. A 10-cM genome scan of 388 extended multiplex families with CL/P from seven diverse populations (2,551 genotyped individuals) revealed CL/P genes in six chromosomal regions, including a novel region at 9q21 (heterogeneity LOD score [HLOD]=6.6). In addition, meta-analyses with the addition of results from 186 more families (six populations; 1,033 genotyped individuals) showed genomewide significance for 10 more regions, including another novel region at 2q32-35 (P=.0004). These are the first genomewide significant linkage results ever reported for CL/P, and they represent an unprecedented demonstration of the power of linkage analysis to detect multiple genes simultaneously for a complex disorder.

Introduction

Orofacial clefts, particularly cleft lip (CL), cleft palate (CP), and cleft lip with or without cleft palate (CL/P), are very common structural birth defects with a complex etiology (Murray 2002). Birth prevalence ranges from 1/500 to 1/2,000, depending on the population, with Native American and Asian populations having the highest prevalence and African populations having the lowest. Individuals with these disorders often require multidisciplinary treatment into adulthood. In addition to the financial costs of such treatment, individuals with orofacial clefts and their families experience health burdens, such as increased morbidity and mortality, and a variety of psychosocial implications (Berk and Marazita 2002). Despite >200 years of research, the complex etiology of orofacial clefts in humans remains unclear, although the pace of discovery has quickened in recent years.

In 1757, the first description of a family with several affected members was published (Trew 1757). Many other studies followed that evaluated the family patterns of orofacial clefts (Marazita 2002a). The consensus from these studies was that there is a familial component to orofacial clefts but not a straightforward single gene-dominant or -recessive pattern. Indeed, statistical modeling of recurrence risk data in families with orofacial clefts suggests that 2–14 loci are likely to be involved (Schliekelman and Slatkin 2002).

Recently, investigators have been attempting to locate the genes predisposing to orofacial clefting, using linkage and/or association methods. The groundwork for these studies was laid by application of complex statistical analyses that allowed direct comparison of multifactorial and single-gene models in CL/P (reviewed by Marazita [2002b]), which demonstrated that single-gene models provided an equal or better fit to the family data than did the “large number of equal and additive loci” specified by standard multifactorial models (Marazita 2002b). The possibility that one or a few loci might explain orofacial clefts made linkage and association studies feasible.

After several negative linkage and association studies, the first positive linkage finding with orofacial clefting was at the F13A locus on chromosome 6p (Eiberg et al. 1987), and the first positive allelic association was between CL/P and a TaqI RFLP in the transforming growth factor α (TGFA) locus (Ardinger et al. 1989). Since those early linkage and association studies, there have been many others (Marazita 2002a, 2002b). Positive linkage and/or association results in CL/P, CP, or both have been found, primarily on seven chromosomes (1, 2, 4, 6, 14, 17, and 19) (Marazita 2002a).

The first CL/P genome scan reported results in English affected sib pairs (Prescott et al. 2000); since then, 12 additional genome scans have been performed (see table 1), each of which identified a number of positive signals, although no single study produced results that reached the standard levels of genomewide significance. Clearly, for complex traits such as orofacial clefts, large numbers of families and/or combined results across studies are essential. Therefore, a consortium of four research groups (M.L.M./L.L.F., J.C.M., A.C.L., and M.A.-B.) was formed to undertake genome scans of a large number of extended multiplex families with nonsyndromic CL/P (NS CL/P) from seven populations (see table 1).

Table 1.

Summary of the 13 Populations Included in the GSMA

	Population Studied
	East Asians				Whites of European Descent					Other
PopulationDescription	Philippinesa	China-Ab	China-Bc	Total	Englandd	United States (Pittsburgh)e	United States (Ohio)f	Australiag	Total	Colombiaf	Indiah	Turkey-Ai	Turkey-Bc	Argentina, Mexico, andUnited States (Iowa)j	Syriak	Total	CIDR-7 Totals (N=7 Studies)	Grand Totals (N=13 Studies)
Genome scan performed by	CIDR-7	Marshfield	CIDR-7		In-housel	CIDR-7	CIDR-7	In-housel	…	CIDR-7	Marshfield/CIDR-7	Marshfield	CIDR-7	CIDR	CIDR	…	…	…
No. of families	220	36	75	331	92	24	9	1	126	48	38/11	18	1	10	2	117	388	574
Type of families	Extended multiplex	Extended multiplex	Extended multiplex	…	Affected sib pairs	Extendedmultiplex	Extendedmultiplex	Extendedmultiplex	…	Extended multiplex	Extendedmultiplex, some consanguineous	Consanguineous	Consanguineous	Extendedmultiplex	Extended multiplex, some consanguineous	…	…	…
No. of individuals	2,770	466	504	3,740	372	163	145	29	709	497	551/162	185	9	102	197	1,541	4,250	5,990
No. of individuals genotyped	1,676	310	347	2,333	338	122	61	19	540	261	272/81	55	3	59	61	711	2,551	3,584
No. of affectedindividuals genotyped	567	80	98	745	207	36	26	3	272	98	82/18	21	2	29	18	250	845	1,267
No. of unaffectedindividuals genotyped	1,109	230	249	1,588	131	86	35	16	268	163	190/63	34	1	30	43	461	1,706	2,317
Overall weightm	3.023	1.300	1.375	…	1.358	.816	.577	.322	…	1.193	1.218	.548	.128	.567	.577	…	…	…
Subgroup weightm	1.591	.684	.724	…	1.768	1.062	.751	.419	…	1.692	1.727	.777	.181	.804	.818	…	…	…

^aSchultz et al. 2003.

^bMarazita et al. 2002.

^cM.L.M. and L.L.F., unpublished data.

^dPrescott et al. 2000.

^eNeiswanger et al. 2003; Weinberg et al. 2003.

^fMoreno et al. 2003.

^gEdwards et al. 2003.

^hField et al., in press.

ⁱMarazita et al. 2004.

^jZeiger et al. 2003.

^kWyszynski et al. 2003.

^lIn-house = genotyping performed in the principal investigator’s laboratory or by in-house genotyping service.

^mWeighting factor for the GSMA, calculated as the square root of the number of individuals genotyped in a particular study, divided by the mean across all studies (Levinson et al. 2003). Weighting factors were calculated separately for each of the three subgroups as well as for the total study.

In addition, we performed a meta-analysis in which results from the seven populations were combined with six other published genome scans of CL/P (see table 1). Various methods have been proposed for meta-analysis in genetic studies, dating back to Fisher (1925), that require either tests of the same hypotheses or the same statistical tests across studies. The 13 available genome scans of NS CL/P did not test the same hypotheses (i.e., different marker sets were used) and did not all calculate the same statistics. Therefore, we applied the genome scan meta-analysis (GSMA) method (Wise et al. 1999; Levinson et al. 2003). Simulation studies that used this method detected linkage with power comparable to or greater than that obtained by performance of a combined linkage analysis of all data (Levinson et al. 2003). The major goal of the current study was to identify those genomic regions that contain genes predisposing to CL/P across populations.

Material and Methods

Study Populations

There were 13 study populations with genome scans available for families with NS CL/P (summarized in table 1). There were 574 families in the 13 populations, with 5,990 individual family members. Of the total family members, 3,584 were genotyped (1,267 affected; 2,317 unaffected). Most of the studies involved extended multiplex kindreds; that is, multigenerational families with at least two affected individuals. One study involved affected sib pairs (England), and all or part of the families in four studies were consanguineous (India, Turkey-A, Turkey-B, and Syria). Two countries each had two genome scans (China-A and China-B; Turkey-A and Turkey-B). Note that each genome scan consisted of unique families—that is, there is no overlap between the families of China-A and China-B or between those of Turkey-A and Turkey-B. The India study had 38 families genotyped by Marshfield (Center for Medical Genetics), of which 11 families were also genotyped as part of CIDR-7 (see below).

The phenotype analyzed in each study was NS CL/P; in all studies, families were ascertained through probands, and additional relatives were recruited. For families to be included, it was necessary that the proband have isolated CL with or without CP (i.e., no other anomalies) and that no other family member have an indication of a clefting syndrome (e.g., lip pits). Each study included evaluations of family members by clinical geneticists to rule out syndromic forms of CL/P. Each study had approval by the appropriate institutional review boards; to participate, all study subjects provided informed consent. Refer to each study reference for more specific details on ascertainment, genotyping, and family structures (see table 1).

Genome-Scan Genotyping

Seven of the studies (Philippines, China-B, India, Colombia, United States-Pittsburgh, United States-Ohio, and Turkey-B; 388 families) (see table 1) were part of a collaborative genome-scan project (between M.L.M./L.L.F., J.C.M., A.C.L., and M.A.-B.) and thus were all genotyped by the Center for Inherited Disease Research (CIDR) at the same time. We will refer to these studies as the “CIDR-7 studies.” The other six studies (186 additional families) were genotyped either by CIDR, by Marshfield Laboratories Mammalian Genotyping Service (Center for Medical Genetics), or in-house by study investigators (see table 1 for details). In most cases, Weber screening-set markers were used (Yuan et al. 1997); that is, high-quality STRPs with an average spacing of ∼9 cM (with a range of 1–19 cM). For details of the markers and genotyping, refer to the Web sites for CIDR and Marshfield and to the references for each genome scan (see table 1).

Statistical Genetic Analyses of Genome-Scan Markers

Each of the 13 studies analyzed the genome-scan data in a variety of ways, by use of both parametric and model-free methods of linkage analysis and, in most cases, also by application of tests of linkage in the presence of association, such as the transmission/disequilibrium test (TDT) (see individual references for details). When different studies have exactly the same markers and exactly the same analysis methods, it is straightforward to combine results. The CIDR-7 families were genotyped for exactly the same markers and were analyzed using the same methods as in M.L.M.’s research group. For the other six studies, different markers were used, and, in some cases, different statistical analyses were performed. Therefore, results across all 13 studies were combined into a meta-analysis (see the “GSMA” section).

For purposes of combining the results from the CIDR-7 studies, we utilized multipoint parametric linkage statistics, in particular the heterogeneity LOD score (HLOD). HLODs are based on the admixture heterogeneity test (Smith 1963), in which the recombination fraction (θ) and the proportion of linked families (α) are estimated simultaneously. Recent simulation studies have shown that, although the estimate of the proportion of linked families may not be precise, HLODs are a powerful method for detecting linkage in the presence of heterogeneity (Greenberg and Abreu 2001; Vieland et al. 2001; Hodge et al. 2002). The descent-graph method (Sobel et al. 1996; Sobel and Lange 1996; Lange 2002) implemented in the computer program SIMWALK2 was used for the multipoint HLOD calculations of the CIDR-7 data.

The inheritance of each marker in each of the CIDR-7 study populations was analyzed with PedCheck (O’Connell and Weeks 1998) to test for inconsistencies due to nonpaternity or other errors. Parametric linkage approaches such as the HLOD approach used here require estimates of marker-allele frequencies and parameters of the genetic model. Allele frequencies were estimated in the founders of the families, separately by study population because the seven populations were of diverse ethnicities. HLOD calculations were done under the best-fitting dominant and recessive models for each study population (estimated from segregation analysis; see individual references). Maximizing LOD scores over a range of genetic models is valid for simultaneously evaluating linkage and determining the most likely genetic model (without adjustment to significance levels and without need to correct for ascertainment), provided that there is indeed linkage (Hodge and Elston 1994). Furthermore, if an oligogenic model is suspected (as seems likely for CL/P) or if significant heterogeneity exists, some of the causal genes may act in a dominant fashion and others in a recessive one.

To combine the multipoint results across the CIDR-7 studies, we summed the multipoint HLODs. Because each study population had different allele-frequency estimates and different genetic-model parameters, it was not possible simply to perform a combined linkage analysis over all families. Further, simulation studies by Vieland et al. (2001) showed that combined HLOD analyses that pool multiple data sets resulted in a loss of power for detecting linkage, compared with summing the HLODs from individual studies.

The meta-analysis method that was used to combine the results from all 13 studies (see the “GSMA” section) can combine any statistics; for maximum consistency across studies, we utilized linkage statistics, including affected–sib pair tests (identical-by-descent–based statistics) from the England study, model-free two-point NPL statistics from the Argentina/Mexico/Iowa study, and two-point parametric LOD scores from all other studies. For the CIDR-7 families, we calculated two-point LOD scores using the Elston-Stewart algorithm (Elston and Stewart 1971), employing the LINKAGE program with recent updates to speed calculations (VITESSE and FASTLINK) (Cottingham et al. 1993; Terwilliger and Ott 1994; O’Connell and Weeks 1995).

GSMA

To combine the genome-scan results across all 13 studies, a meta-analysis procedure was necessary. There are many proposed methods for combining results across diverse studies. Fisher (1925) proposed a simple and elegant method for combining probabilities from tests of significance across multiple studies. Province (2001) supplied an update to incorporate certain modern linkage and association statistics that are truncated at 0.0. In Fisher’s method, it is not necessary that the same statistic be used for calculating P values across studies; however, it is necessary that the P values arise from tests of the same hypothesis—that is, tests of linkage or association to the same marker. To combine P values across “i” studies, sum the quantity: (-2lnP_i), where P_i is the P value in the ith study (Fisher 1925); for parameters that are truncated at 0.0, use P=1/(2ln2)=.72 (Province 2001). Various other methods have also been proposed (reviewed by Wise et al. [1999], Province [2001], and Levinson et al. [2003]) that require either tests of the same hypotheses or that the same statistical tests be applied across studies.

The 13 available genome scans of NS CL/P did not test the same hypotheses (i.e., different marker sets were used) and did not all calculate the same statistics. Therefore, we applied the GSMA method (Wise et al. 1999; Levinson et al. 2003) to combine data across studies. The GSMA is a nonparametric rank ordering method that can combine genome-scan methods across studies with different markers and different statistical tests. In simulation studies, the GSMA detected linkage with power comparable to or greater than that obtained by performing a combined linkage analysis of all data (Levinson et al. 2003).

For the GSMA procedure, the genome was divided into bins, with bin width selected such that there were at least two bins on the smallest chromosome and that at least one marker was genotyped within each bin. Therefore, for combining the current 10-cM genome scans, a bin width of 30 cM was selected (130 bins across the genome). For each of the 13 studies, bins were assigned a rank (R, with values of 1–130) according to the maximum-linkage statistic of markers in each bin. Any tied bins were assigned equal Rs on the basis of the mean of the sequential ranks for those bins.

Because the 13 study populations covered a wide range of sample sizes (see table 1), we weighted the R statistics on the basis of sample size. Optimal weighting strategies are not determined for the GSMA (Wise et al. 1999; Levinson et al. 2003), but simulation studies showed that weighting increased the power of the GSMA to detect linked loci (Levinson et al. 2003). We used a weighting factor based on the total number of genotyped individuals—the ranks within each study were multiplied by

Weighting factors for each study are listed in table 1.

For combining the rank results across studies, the average rank (R_avg) for each bin was calculated across all 13 studies. The probability distribution for unweighted R is derived (Wise et al. 1999); however, for weighted Rs, the probability distribution must be determined empirically. To determine statistical significance, the resulting R_avg distribution was then compared with its empiric probability distribution, derived from 1,000 simulations under the assumption that ranks were assigned randomly (GSMA P values [denoted “P_avg” by Levinson et al. (2003)]). The P_ord (Levinson et al. 2003) probabilities also were calculated for each bin to provide additional information about interpreting the GSMA results with respect to linkage to each bin (see the “Statistical Significance” section below). The P_ord probability for a rank is the P_{avg Rank/order}—in other words, that bin’s place in the order of average ranks—and quantifies the consistency of the R_avg across simulations.

The GSMA procedure was done across the entire set of 13 studies, then repeated for three subgroups: “East Asian,” “white,” and “other.” Table 1 lists the composition of the subgroups, including the weighting factors used within each subgroup. Note that, unlike the East Asian and white subgroups, the “other” subgroup is a catch-all category (with no particular biological meaning). Within the “other” subgroup, there are studies of South Americans, Indians, Turks, and Syrians, but there are not enough families in any of those groups to have sufficient power to analyze them separately.

The GSMA identified 30-cM bins that are best supported across the studies. To narrow the regions of positive findings, we propose an extension of the GSMA that involves repeating the GSMA with different bin starting points and then determining the minimum region of maximum significance (MRMS). We repeated the GSMA twice, shifting the starting point for the binning procedure first to 10 cM and then to 20 cM. This determined the 10-cM MRMS for each positive finding. Given that all genome scans averaged 10 cM between markers, 10 cM is the limit of resolution for this meta-analysis.

Candidate-Gene–Analysis Methods (Markers on 9q21)

To pursue the highly significant results on 9q21, five candidate genes that lie within or immediately adjacent to the one- and two-LOD intervals of the peak HLOD score were selected for additional studies. SNPs for four of these genes (PTCH, ROR2, TGFBR1, and ZNF189) were genotyped in 219 multiplex Filipino families; microsatellite markers for two of the genes (FOXE1 and TGFBR1) were genotyped in 99 Chinese, 50 Indian, and 18 Turkish families.

The TDT was used to assess association in the presence of linkage disequilibrium between the markers and CL/P. The family-based association-test (FBAT) extension of the TDT (Laird et al. 2000; Rabinowitz and Laird 2000; Horvath et al. 2001) was used to assess association between alleles at each marker and CL/P, plus association between CL/P and haplotypes of SNPs within candidate genes. Linkage was assessed using the methods described in sections above. Since TGFBR1 was assessed in all the data sets, the Fisher (1925) method was used to combine the TGFBR1 results across studies.

Statistical Significance

For determining genomewide significance for the multipoint-linkage calculations of the CIDR-7 families, standard guidelines (Lander and Kruglyak 1995) were followed. The desired α-level of .05 was divided by 400 (the approximate number of genome-scan markers) to yield .000125; therefore, the corresponding LOD-score threshold for genomewide significance was 3.2.

For the GSMA, result-interpretation guidelines presented by Levinson et al. (2003) were followed. By use of a Bonferoni correction, the criterion for genomewide significance for any particular bin would be .05/130=.0004. In the situation where more than one bin is likely to contain a linked locus, Levinson et al. (2003) determined (from simulation studies) that, if the GSMA P value (“P_avg” in the notation of Levinson et al. [2003]) and the P_ord were both ⩽.05 for a particular bin, then that bin was highly likely to contain a linked gene. Therefore, we considered the GSMA results to have genomewide significance either if the GSMA/MRMS P value was ⩽.0004 or if both the P value and P_ord were ⩽.05.

Results

Figure 1 summarizes the results of the combined multipoint linkage analyses of NS CL/P in the CIDR-7 populations. Six regions on five chromosomes had summed HLODs ⩾3.2 (chromosomes 1p12-13, 6p23, 6q23-25, 9q21, 14q21-24, and 15q15). There were nine additional regions with summed HLODs between 2.0 and 3.2 (see figs. 1A and A1 [online only]).

Figure 1.

Summary of the 10-cM genome scan of CL/P in the CIDR-7 studies. A, Summary of the maximum summed multipoint HLOD for each chromosome, under both dominant (DOM) and recessive (REC) genetic models for CL/P. B–F, Summed multipoint HLOD plots for each chromosome that had a maximum summed HLOD ⩾3.2 (under the best genetic model for each chromosome).

Figure 2 summarizes the results of the initial GSMA for all 13 studies. Depicted are the R_avg for each of the 130 bins, across all chromosomes, with the thresholds for P values of .05 and .01 indicated. Figure 3 provides a graphical representation of the bin-shifting procedure and the MRMS results for the six chromosomes with the most significant GSMA/MRMS results (see fig. A2 [online only] for all chromosomes). Under the MRMS procedure, the significance increased for most of the regions that were positive in the initial GSMA, whereas some regions that were not significant in the initial GSMA were significant after rebinning. Supplemental figure A3 (online only) shows the GSMA results, by population subgroup, for each chromosome. Table 2 summarizes the genomewide significant results from the combined analyses of the CIDR-7 studies as well as the GSMA/MRMS of the 13 studies and of the population subgroups.

Figure 2.

GSMA of 13 CL/P-study populations. Depicted are the R_avg for each of the 130 bins, across all chromosomes. The horizontal lines indicate the R_avg values corresponding to empirical P values of .05 and .01, as determined by simulation.

Figure 3.

Graphs of the MRMSs for the six chromosomes with the most significant GSMA/MRMS results for the 13 CL/P-study populations. These graphs summarize the MRMS process, repeating the GSMA with bin shifting to narrow the statistically significant region. The dashed vertical lines indicate the 10-cM MRMS for each chromosome. A–F, Graph of chromosomes 1, 2, 6, 14, 17, and 18, respectively.

Table 2.

Genomewide Significant Linkage and Meta-Analysis Results

	CIDR-7 Resultsa				GSMA/MRMSb
ChromosomalRegion	MaximumSummedHLODd	Peake(cM)	Modelf	Average αg(%)	10-cM Intervalh(P, Pord)	Subgroup with P⩽.05i,j	Candidate Genes in Region (OMIM Designation)c
1p12–13	3.58	116	Dominant	42.9	…	…	COL11A1
1q32	…	…	…	…	260–270 (.030, .001)	EA, O	IRF6
2q32–35	…	…	…	…	210–220 (.0004, .049)	EA, O	SATB2, WNT6, DLX2
3p25	…	…	…	…	60–70 (.017, .001)	O	WNT7A, RARB, THRB, TGFBR2
6p23	3.74	68	Dominant	52.1	…	…	F13A1, TFAP2A, BMP6, EDN1, COL11A2 (OFC1 [MIM 119530])
6q23–25	3.55	123	Dominant	19.3	120–130 (.013, .033)	EA
7p12	…	…	…	…	70–80 (.027, .001)		EGFR
8p21	…	…	…	…	60–70 (.027, .001)	EA	FGFR1, EGR3, PPP3CC, FZD3, NRG1
8q23	…	…	…	…	140–150 (.014, .001)	W	FZD6
9q21	6.62	111	Dominant	42.1	…	…	PTCH, ROR2, TGFBR1 ZNF189, TTF2(FOXE1), clf2 homology
12p11	…	…	…	…	80–90 (.018, .008)	O
14q21–24	3.92	83	Recessive	15.0	40–50 (.013, .005)	EA, W	TGFB3, BMP4
15q15	3.65	28	Recessive	27.1	…	…	ZNF291, NRG4
17q21	…	…	…	…	40–50 (.004, .017)	W, O	RARA, clf1 homology, WNT9B
18q21	…	…	…	…	60–70 (.010, .015)		MADH(SMAD2)
20q13	…	…	…	…	50–60 (.021, .001)	EA	EDN3, BMP7

^aN=7 studies; 388 families.

^bN=13 studies; 574 families. Results of genomewide significance from GSMA of 13 total studies (i.e., either P⩽.0004 or both P and P_ord⩽.05 [Levinson et al. 2003]).

^cList of potential cleft candidate genes in regions. One of the regions has the designation “orofacial cleft 1” (OFC1) on Online Mendelian Inheritance in Man (OMIM).

^dMaximum multipoint HLOD summed for the seven CIDR-7 studies (see table 1 for list; see figs. 1 and A1 [online only] for full HLOD plots); presented are those regions with HLODs ⩾3.2 (threshold for genomewide significance [Lander and Kruglyak 1995]).

^eLocation (cM) of the maximum summed multipoint HLOD.

^fGenetic model for the maximum summed HLOD.

^gEstimated proportion of linked families at the peak, averaged over the CIDR-7 studies.

^hThe 10-cM interval best supported by the GSMA/MRMS, with the corresponding P value and P_ord (see figs. 3 and A2 [online only] for MRMS graphs).

ⁱPopulation subgroups: EA=East Asian; W=white; O=“other.” (See table 1 for list of populations in each subgroup; see fig. A3 for population-specific GSMA/MRMS graphs.)

^jFor chromosomes 7p12 and 18q21, none of the population-specific GSMA results were significant (none ⩽.05), although the overall results were significant.

Of the six regions of genomewide significance in the CIDR-7 analyses, two also had genomewide significance in the meta-analysis (6q23-25 and 14q21-24). One region, 2q32-35, had a genomewide-significant GSMA P value (P=.0004), whereas nine additional regions met the combined genomewide significance criteria of both P and P_ord⩽.05 (Levinson et al. 2003) (table 2).

For the markers in 9q21 genes PTCH, ROR2, TGFBR1, and ZNF189 (see fig. 4), linkage and association were assessed and results combined across studies by use of the Fisher (1925) method. Positive associations were found with four of the candidate genes (ROR2 P=.02; PTCH SNP haplotype P=.007; TGFBR1 P=.02; and FOXE1 P=.002), and suggestive linkage was found with FOXE1 (summed maximum LOD=1.54; θ=0.05).

Figure 4.

Diagram of chromosome 9q21, with one- and two-LOD support intervals and candidate genes depicted

Discussion

Combined results from 13 genome scans were remarkable, given the genetic complexity of CL/P; one region (9q21) had an HLOD of 6.6, and another (2q32-35) had a meta-analysis P value of .0004. This demonstrates the great utility of the GSMA in combining disparate data sets. Further, we summarize here a new extension of the GSMA to determine the MRMS. The combined GSMA/MRMS allowed us to narrow the regions of significance from 30 cM to 10 cM (fig. 3 and table 2). It is interesting that some statistically significant regions were discovered only during one of the repeat bin–shifted GSMAs used to determine the MRMS (table 2), highlighting the importance of utilizing bin shifting when applying the GSMA.

The chromosomal regions that had statistical genomewide significance in this study include many of the regions reported elsewhere for NS CL/P but also identify two highly significant novel regions, 9q21 and 2q32-35. Positive association results (indicating close proximity to a CL/P locus) were also found with four of the five selected candidate genes on 9q21. Overall, 16 regions on 12 chromosomes had genomewide-significant results (table 2). Six other regions on four chromosomes had suggestive results (HLOD between 2 and 3 and/or GSMA/MRMS P value <.05 but P_ord>.05). The finding of multiple chromosomal regions with statistically significant results is consistent with an oligogenic model for CL/P, as proposed elsewhere (Farrall and Holder 1992; Mitchell and Risch 1992; Schliekelman and Slatkin 2002).

Table 2 summarizes all results of genomewide significance and indicates potential cleft candidate genes in each positive region. A brief discussion of the most notable results by chromosome follows. Although not summarized in detail, several regions also have chromosomal rearrangements reported in CL/P (Brewer et al. 1998, 1999).

Chromosome 1

The 1q32 region is the location for interferon regulatory factor–6 (IRF6) that was identified recently as the locus involved in van der Woude syndrome (VDWS [MIM 119300]) (Kondo et al. 2002). Our group recently has found highly significant evidence of association between IRF6 and CL/P (Zucchero et al., in press), with TDT P<10^-9 in a sample of 3,350 nuclear families (some families from CIDR-7). The current results that demonstrate linkage to the IRF6 region provide powerful support for the IRF6-association findings. It is interesting that there was only a weak linkage signal to the IRF6 locus itself (LOD<1.0) (Zucchero et al., in press); thus, the current study demonstrates the utility of combining genome scans across studies to detect subtle effects.

Chromosome 2

The 2q32-35 region had the most highly significant GSMA results in the current study (P=.0004) and contains the gene for the DNA-binding protein SATB2 (a.k.a. “KIAA1034”) that has been identified elsewhere, through translocation-breakpoint analysis, as a gene involved in cleft palate (FitzPatrick et al. 2003) and that also shows site- and stage-specific expression in murine palate development.

The 2p13 region contains TGFA, the gene with the first reported association with CL/P (Ardinger et al. 1989) and numerous confirmatory reports (Mitchell 1997; Marazita 2002a). TGFA has ∼40% sequence homology with epidermal growth factor (EGF) and competes with EGF for binding to the EGF receptor. The results for the 2p13 region did not reach genomewide significance but were suggestive (HLOD=2.67; GSMA P=.001; P_ord=.18).

Chromosome 4

Chromosome 4 is a notable omission from the list of chromosomes with positive results in this study. Two regions on chromosome 4 have had positive linkage or association results in the literature: homeobox 7 (MSX1 4p16.1 [MIM 106600] and 4q21-31) (Marazita 2002a). Regarding 4p16.1, positive associations between CL/P and MSX1 have been reported (Marazita 2002a), hypodontia with oral clefts is due to MSX1, mouse Msx1 knockouts have cleft palate, and complete sequencing of the MSX1 gene showed mutations in ∼2% of NS CL/P cases (Jezewski et al. 2003). If an MSX1 variant is a low-frequency gene predisposing to CL/P or if MSX1 is a modifier of expression, it would be difficult to detect by use of linkage methods. The fact that relationships between CL/P and both MSX1 and IRF6 were detected using nonlinkage approaches highlights the necessity of applying diverse, complementary approaches for adequately delineating complex traits.

MSX1-related CL/P may be limited to certain population groups; in the subgroup-specific GSMA/MRMS analyses, there was a borderline significant result (P=.05) near the MSX1 region in the “other” subgroup but not in the white or East Asian subgroups.

Chromosome 9

Region 9q21 had the most significant linkage result and also was suggestive in the GSMA/MRMS (P=.01; P_ord=.058). Mutations in the gene for the human homolog of patched (PTCH [9q22.3]) cause basal cell nevus syndrome (Gorlin syndrome [MIM 109400]), in which ∼5% of cases have CL/P. Furthermore, PTCH maps to the region homologous to the clf2 locus in the mouse (Juriloff 2002). Receptor tyrosine kinase–like orphan receptor 2 (ROR2 [9q22]) is selectively expressed in—and particularly important for—the chondrocyte lineage. Transforming growth factor β receptor type 1 (TGFBR1 [9q33-q34]) mediates the induction of several genes involved in cell-matrix interactions, including the TGFB superfamily (thought to be developmental regulators of cell proliferation and differentiation). The gene for zinc finger protein 189 (ZNF189 [9q22-q31]) is one of the estimated 300–700 ZNF genes in the human genome. ZNFs bind nucleic acids and regulate transcription. Mutations in the forkhead domain–containing transcription factor FOXE1 (a.k.a. “TTF2” or “TITF2” [9q22]) are associated with congenital hypothyroidism, thyroid agenesis, and cleft palate in humans and mice (Castanet et al. 2002; Dathan et al. 2002). The forkhead gene family (Fox), originally identified in Drosophila, encodes transcription factors with a conserved 100–amino acid DNA-binding motif called the “forkhead domain” (Lehman et al. 2003) and regulates many diverse developmental processes in eukaryotes. We assessed linkage and association with each of these five potential candidates, and we found positive associations with PTCH, ROR2, TGFBR1, and FOXE1 and suggestive linkage with FOXE1. A major current focus of our research group is fine mapping of this region.

Chromosome 14

The region containing transforming growth factor β 3 (TGFB3) (14q24) was significant in both CIDR-7 and the GSMA/MRMS. The mouse Tgfb3 knockout has cleft palate, and there are some positive results from association studies of CL/P and TGFB3 (Marazita 2002a). We also recently found suggestive evidence that the CL/P phenotype within ectrodactyly-ectodermal dysplasia-clefting syndrome (EEDC1) maps to 14q, indicating that genes involved in NS CL/P may modify the phenotypic expression of the EEC1 gene (Ray et al., in press).

Chromosome 17

The 17q21 region had significant results in the GSMA/MRMS analyses and suggestive results in the CIDR-7 linkage analysis (HLOD=2.23). It is notable that this region is syntenic with the region harboring the mouse clf1 mutation (Juriloff 2002). Further, one of the early associations observed with NS CL/P was with the gene for retinoic acid receptor α that maps near this region (RARA [17q12]) (Marazita 2002a).

Chromosome 19

Linkage and association have been seen in a few studies between CL/P and APOC2, BCL3, or nearby anonymous markers (19q13) (Marazita 2002a). The current results for 19q13 are also suggestive (HLOD=2.66; GSMA P=.012; P_ord=.088). Genes in this region include TGFB1 (19q13.1) and polio-virus receptor (PVR)/polio-virus receptor–like 2 (PVRL2), both at 19q13.2-q13.3. It is interesting that mutations of PVRL1 (11q23-q24) cause the autosomal recessive Margarita Island clefting syndrome, in which it has been suggested that heterozygotes have an increased risk of clefting (Sozen et al. 2001). A family segregating a balanced translocation and clefting has its breakpoint in the CLPTM1 gene in this region (Yoshiura et al. 1998).

* * *

The studies summarized here combine results from diverse populations of families with CL/P. Therefore, the positive results necessarily will be those that are in common across populations. Note that a number of chromosomes had population subgroup–specific differences in the patterns of GSMA results (summarized in table 2). There is no formal test of heterogeneity within the GSMA, but the results summarized in table 2 (also see fig. A3 [online only]) could guide prioritization for fine-mapping and gene identification in particular population subgroups.

In summary, the remarkably significant evidence for linkage from this large study demonstrates the advantage of combining data sets to further our understanding of the genetics of a complex trait such as CL/P. This is an unprecedented demonstration of the power of linkage analysis to detect simultaneously multiple genes for a truly non-Mendelian disease in which no single major genes are evident. To identify and characterize those genes, fine mapping and candidate-gene sequencing has begun in our research groups, as have investigations of gene-gene interactions and genotype-phenotype correlations. After almost 250 years of research regarding the familial nature of CL/P, evidence is accumulating that a relatively small number of chromosomal regions contain causal genes for NS CL/P. With the molecular and statistical tools available today, and with our rich resource of extended multiplex families with CL/P, this becomes a manageable number of chromosomal regions to characterize in order to identify the causal genes for CL/P.

Appendix

Figure A1.

Summary of the 10-cM genome scan of CL/P in the CIDR-7 studies. A–V, Summed multipoint HLOD plots for the autosomes, under the best genetic model for CL/P on that chromosome (dominant or recessive). W, Plot of the maximum summed two-point LOD scores on the X chromosome (θ at the peak LOD was 0.30).

Figure A2.

Graphs of the MRMSs for all chromosomes analyzed in the 13 CL/P-study populations. These graphs summarize the process of repeating the GSMA analyses, shifting the bins to narrow the region of potential involvement in CL/P. For those chromosomes with GSMA/MRMS P values ⩽.05, the dashed vertical lines indicate the 10-cM MRMS. A–V, Autosomes. W, X chromosome.

Figure A3.

Graphs of GSMA results by population subgroup (all, East Asian, white, or “other”). The results from the GSMA procedure that had the most significant result (i.e., of the three rebinned GSMA procedures) for each subpopulation are depicted. A–V, Autosomes. W, X chromosome.