A Genome-Wide Scan in Families With Maturity-Onset Diabetes of the Young: Evidence for Further Genetic Heterogeneity

We studied 23 families fitting strict MODY criteria of having at least one affected subject diagnosed with type 2 diabetes ≤25 years and at least two generations of type 2 diabetes. An additional three families with characteristics falling just outside these criteria were included (families FR624, FR743, and DK19). Table 1 gives the clinical criteria for all subjects. All families were of European ancestry except DUK149 (East-Asian Jamaican). These families underwent three separate genome-wide linkage studies; 5 families were genotyped in Malmö, Sweden, and have been reported as part of a previous study of young-onset families from Sweden and Finland (26), 7 were genotyped in Oxford and Exeter, U.K., and 14 families were genotyped in Lille, France. These families consisted of 236 individuals of which 125 were considered affected (105 with diabetes, 11 with impaired glucose tolerance [IGT], and 9 with impaired fasting glucose [IFG]). Figure 1 shows the pedigrees, and Table 2 summarizes the family structures. Of the affected subjects in the five MODY families from the Swedish scan, 65% had HLA risk alleles (DQB02 and/or 0302). The same criteria for a subject to be considered affected were used among all participating groups as follows:

1) Diabetes was defined according the World Health Organization 1999 criteria: fasting plasma glucose >7.0 mmol/l, 2-h plasma glucose ≥11.1 mmol/l, or a previous diagnosis of diabetes with ongoing treatment with oral agents and/or insulin (27).

2) IFG was defined as fasting plasma glucose >6.1 and 2- h blood glucose <7.8 mmol/l, and IGT was defined as fasting plasma glucose <7.0 mmol/l and 2-h blood glucose ≥7.8 mmol/l. Patients with IFG or IGT were considered affected only if they had the same status on two or more consecutive occasions of an oral glucose tolerance test. All Swedish/Finnish subjects classified as IGT had a 2-h blood glucose ≥8.5 mmol/l. If subjects only reached these values on a single test, they were classified as status unknown. There were 20 patients with IGT or IFG among the three sets; all have first-degree relatives with diabetes and are of high risk to develop diabetes later in life (28).

Exclusion of mutations in known MODY genes was performed using one or more of the following methods: 1) linkage exclusion (logarithm of the odds [LOD] scores less than −2.0) using microsatellite markers flanking known MODY genes (microsatellites included D12S366, D12S321, D12S807, D12S820, D12S342, D12S1349, and D12S395 for HNF-1α; D20S170, D20S96, D20S119, ADA microsatellite, D20S17, and D20S197 for HNF-4α; GCK1, GCK2, D7S667, D7S519, and D7S2506 for glucokinase; D17S1788, D17S927, and D17S800 for HNF-1β; and D13S221, D13S1254, and D13S289 for IPF1); 2) forward and reverse direct sequencing of all exons and intron-exon boundaries of known MODY genes according to previously described methods (11,29,30); and 3) automated fluorescent single-strand conformation polymporphism or denaturing high-performance liquid chromatography technique, followed by sequencing of variant profiles (31). The presence of a A3243G mitochondrial DNA mutation was either excluded by previously described genetic methods (11) or by the absence of a history of maternal diabetes and deafness. In addition, the recently described β-cell–specific alternative promoter and associated exon1D of HNF-4α (resulting in the splice variant HNF-4α7) was excluded by forward and reverse direct sequencing using a proband from each family and previously described methods (32).

Genotyping data were pooled from three different genome scan studies, consisting of 14 (from France), 7 (from U.K.), and 5 (from Sweden) families.

Genotyping was performed using 445 microsatellite markers with an average spacing of 10 cM. After electrophoresis on an ABI 377 DNA sequencer, semiautomated fragment sizing was performed by use of GENESCAN and GENOTYPER software (ABI). Each genotype was reviewed independently by two members of the research team to confirm the accuracy of allele calling. Incompatibilities were searched for with the PED-CHECK 1.1 program (33) and inconsistencies resolved. For nonresolvable errors, genotypes for families or subfamilies involved were deleted.

Genotyping was performed using 400 dinucleotide repeat markers with an average spacing of 10 cM. Electrophoresis was preformed on a MegaBACE 1000 capillary array sequencer (Molecular Dynamics; Amersham Pharmacia Biotech, Sunnyvale, CA). We reviewed and manually edited only those genotypes receiving a quality score <2.0 (of a possible 10), indicating a lower confidence in the automated genotype provided by the software. Mendel checking (testing for co-dominant inheritance of marker alleles) was performed with PedCheck version 3.0 (33). Before linkage analysis, Mendel conflicts due to misinterpretation of traces were resolved through rescoring in Genetic Profiler. The Swedish Finnish Genome-Wide Scan has been described in detail elsewhere (26).

We analyzed our data by generating combined marker maps and combining raw genotyping data from the three different genome-wide scans as recommended (34).

Marker maps were created using the Marshfield Marker map and subsequently the obtained marker order was tested in MAP-O-MAT, which is a web-based program that runs CRI-MAP on Centre d’Etude du Polymorphisme Humain (CEPH) families; for details, see the appendix of this article. A total of 757 unique markers were put into a map with an average distance of 4.7 cM between unique markers (note that for any given family average spacing between markers remained at ∼10 cM, as no further genotyping was performed). Kosambi map distances were used. Markers used in more than one genome screen were coded as two separate markers with 0.01 cM distance between them.

All genotype data were rechecked for Mendelian segregation using the PEDMANAGER software (M.P. Reeve and M.J. Daly, personal communication). As the extent of genetic homogeneity was largely unknown, we decided ad hoc that four analyses should be performed in our pooled data sets. Two groups of families were used: all 26 families and a subset of 14 families fitting more strict criteria for autosomal dominant inheritance. These more strict minimum criteria were as follows: two members with age at diagnosis at <25 years and two consecutive generations of diabetes or one member with an age of diagnosis at 25 years and three consecutive generations of diabetes. Families fitting this criteria but including bilineal inheritance were also excluded. The subset of 14 families were FR30, FR538, FR694, FR725, IT4884, IT4885, ESP7, SW1168, SW20631, SW20655, SW20663, DUK149, DUK89, and DUK156. These two sets of 14 and 26 families were analyzed using two disease models as follows:

1) A parametric multipoint analysis using the GeneHunter (GH) software version 2.01 (35). The parametric LOD score and heterogeneity LOD (HLOD) score calculations were performed using the inheritance model proven successful in previous studies of early-onset diabetes (MODY3) (36). This model uses a disease allele frequency of 0.001 and four age-dependent liability classes: <10, 10–25, 25–40, and >40 years with age-dependent prevalences for carriers of one or two copies of the disease allele of 0.0003, 0.0006, 0.0012, and 0.0025, respectively. The frequency of phenocopies was assumed to be 0.5, 1, 10, and 30% for the four age-dependent liability classes, respectively.

2) NPL analysis. In the event that some of the MODYX families did not concur to the same strict parametric model of inheritance previously used to identify the MODY3 locus, we also performed NPL analysis. Evidence of linkage was assessed without specifying a mode of inheritance using GH. Due to computational limitations, GH cannot perform multipoint analyses in families of >18 bits (where number of bits is 2n − f, where n = nonfounders and f = founders). For families DUK32, DUK149, FR30, FR743, FR725, and DK19, 2, 3, 13, 7, 3, and 1 family members were excluded from the analysis, respectively. However, these subjects were predominantly young unaffected subjects who would not add great power to the analyses.

Allele frequencies used in all of the analyses were those observed in the founders of the families in the study using 18 bits. Exclusion mapping was also performed with the GH 2.01 package; exclusion was declared at an LOD score of −2.0.

We considered an HLOD score of 3.3 as the threshold for declaring significant linkage with the parametric model. For the nonparametric analysis we allowed for incomplete informativeness in the dataset and estimated the thresholds for suggestive and significant genome-wide linkage in our family collection by simulations using the GENSIM computer program designed by Mark J. Daly (unpublished data). This program created a data set matching ours with respect to marker informativeness, number of genotyped individuals, affection status, and fraction of missing data (3%). Specifically, 100 genome screens (2,200 chromosomes) were generated at the genome-scan marker density. In this fashion, the overall genome-wide threshold for suggestive linkage was determined to be at an NPL score of 2.4, while significant linkage required NPL scores of 3.6 (P ≤ 0.05), respectively. When repeating the above simulation, using only the 14 families falling under the strict criteria, the overall genome wide threshold for suggestive linkage was determined to be at an NPL score of 2.8, while significant linkage required NPL scores of 4.1 (P ≤ 0.05), respectively. In addition, we performed simulations using a 100-cM chromosome with markers with eight alleles spaced at 10 cM and the gene placed in between the two central markers. We used three heterogeneity parameters, α at 0.30 (i.e., 30% of the families share the same genetic background), 0.50 (i.e., 50% of the families share the same genetic background), and 0.80 (80% of the families with the same affection gene), assuming 54, 54, and 88% penetrance, respectively, for carriers of one or two copies of the affected gene. Simulations indicated that we had 75% power to detect linkage with an NPL of 0.84 under the 30% heterogeneity model, 75% power to detect linkage with an NPL of 2.51 under the 50% heterogeneity model, and 75% power to detect linkage with an NPL of 4.91 under the 80% heterogeneity model.

Figure 2 shows the multipoint NPL Z-scores for all 26 families, the multipoint HLOD scores for the subset of 14 strictly defined families, and the peak multipoint LOD scores >1.0 for individual families. In the set of 14 families with the strictest MODY criteria, we extracted on average 66% of the inheritance information.

The analysis of all 26 families revealed NPL Z-scores exceeding the nominally significant threshold of P = 0.05 on chromosome 6 (69–82 cM from pter; GATA29C09-D6S460, max NPL score 2.18 at 80 cM) and chromosome 10 (169–175 cM from pter; D10S212-D10S169 max NPL score 1.90 at 172 cM), although neither of these quite reached the empirical level for declaring suggestive linkage. HLOD scores using the previously described parametric model for MODY reached a maximum of 0.31 at 80 cM for chromosome 6 and 0.38 at 174 cM for chromosome 10. There was a high frequency of affected subjects with high-risk HLA alleles in the five families from the Swedish genome scan, but none of these families showed evidence for linkage to the HLA region. There were no differences in phenotypic characteristics (age at onset, BMI, or treatment) between families contributing to different loci. In addition, we analyzed these 26 families by individual scan, but this did not result in any HLOD scores reaching genome-wide significance.

The analysis of the 14 families defined using more strict criteria and using the previously described parametric model for MODY revealed a maximum HLOD score of 1.81 at 117 cM on chromosome 3 (HLOD >1.0, α = 0.36 from 91–122 cM; D3S1287-D3S2459). The only other HLOD >1.0 was observed on chromosome 5 (HLOD 1.37, α = 0.50 at 172 cM; D5S400-D5S498). NPL Z-scores for these 14 families exceeding the nominally significant threshold of P = 0.05 were obtained on chromosome 3 (87–117 cM, maximum Z-score 2.84 at 102 cM), chromosome 5 (173–175 cM, maximum Z-score 2.04), chromosome 6 (70–75 cM, maximum Z-score 2.07), and chromosome 10 (169–175 cM, maximum Z-score 1.61), although only one of these, on chromosome 3, exceeded the empirical Z-score of 2.8 for declaring suggestive linkage. The addition of four extra markers (D3S3683, D3S1303, D3S1267, and D3S1269) in the chromosome 3 region reduced the HLOD to 1.27 at 124 cM and the NPL linkage score to 2.78 at 97 cM.

The maximum LOD score possible for an individual family using the parametric model described was 2.24 (family FR30), and therefore none of the families in our study reached the required LOD score of ≥3.0 for declaring significance. Table 3 gives details of the families contributing to the evidence for linkage at each of the peaks identified in the genome scan. In our study, two families were capable of exceeding LOD scores of 2.0 and a further 13 families were capable of exceeding LOD scores of 1.0. We did not observe any significant clustering of individual family LOD scores (if so this would have resulted in significant HLOD scores), but a few results from individual families are worth noting (Table 4). Two of the three largest families FR30 and DUK149 are linked to chromosome 3 with their maximum LOD scores; FR30 has a maximum LOD of 2.24 at 106 cM, and DUK149 has a maximum LOD of 1.55 at 147 cM. There was some degree of overlap between these results, but the addition of four extra markers, while narrowing the region of overlap of multipoint linkage scores, revealed that no single marker showed evidence for linkage to both families. In addition, two families, FR571 and DUK32, have LODs >1.0 at 10 cM from pter of chromosome 22.

We have performed the first large genome-wide scan of MODY families not linked to any of the six known genes. Our results are consistent with further heterogeneity existing among MODY families in which a causal mutation has not yet been identified. We have identified several putative loci, although none reached statistical significance, either by conventional criteria of HLODs of ≥3.3 (37) or NPL Z-scores of ≥4.1 (38), or by empirically computed levels of significance. This suggests the existence of multiple additional genes. Two methods of linkage analysis were used, a parametric model of inheritance previously used to identify the MODY3 locus and a model-free analysis in the event that this model was not suitable for these families. It is therefore unlikely that the methods of analysis limited the power of the study. More likely is the possibility that by increasing our sample size by pooling data we may have also inadvertently increased the amount of heterogeneity. Power calculations indicated that we had 75% power to detect an NPL score of 4.91 under the 80% heterogeneity model (80% of the families share the same gene) but <50% power to detect a similar NPL score under the 50% heterogeneity model, and 75% power to detect linkage with an NPL of 0.84 under the 30% heterogeneity model. The lack of any large families with individual LODs of >3.0 would have compounded this problem; our analysis was dependent on at least some families being linked to the same gene. If the families described in our study have a monogenic form of diabetes, as we believe, then each family should contribute to only one region of true linkage. Table 3 describes the families contributing to the loci identified in the genome scan. Some families contribute to more than one locus, but this is not unexpected—some are small enough that linkage would be expected to more than one region due to stochastic variation. In addition, it is possible that some of the families in our study have a polygenic mode of inheritance of diabetes that, through chance, has resulted in apparently dominant inheritance patterns. Finally, it is possible that the use of different marker sets across studies may have blunted linkage signals.

The regions we have identified as most likely to contain novel MODY genes include those on chromosome 3 (∼87–117 cM), chromosome 5 (∼173–175 cM), chromosome 6 (∼70–75 cM), and chromosome 10 (∼169–175 cM). Family FR30 was part of a previous genome-wide study using different markers that led to the mapping of the MODY3/HNF1α gene on chromosome 12q22 (38); family FR30 had a positive individual LOD score (Z >2.0) at D14S72 (the current genome-wide scan used a new optimized set of markers that by chance did not include this marker). A haplotype analysis with additional markers in the more centromeric region on chromosome 14q11 pointed out four markers cosegregating with diabetes in this family (39). In the current genome scan the closest marker (D14S261) to this region did not reveal any noteworthy evidence for linkage because of lack of complete informativeness within the FR30 pedigree. This indicates that with a 10-cM genome scan and uninformative markers in individual families, linkage may be missed.

To test the hypothesis of genetic heterogeneity within all families, we have also considered taking out the two large families showing linkage to chromosome 3p with their maximum LOD scores and repeated both parametric and NPL analyses in order to search for any other chromosomal location with a linkage peak rise. Only one HLOD >1.0 near the start of chromosome 12p came up in the 24 families set (after excluding FR30 and DUK149).

A search of the locations of known β-cell transcription factors did not reveal any close correlation between locations and linkage results in individual families. The exception to this was family FR253, which was linked to the region of chromosome 1 containing the LMX1 (LIM-homeobox transcription factor 1) gene (LOD 1.48). LMX1 (OMIM: 600298) binds to the insulin promoter to stimulate insulin gene transcription (40). Family FR571 was linked to the region of chromosome 20 harboring the HNF-3β and Nkx2.2 genes, but the coding regions of these genes have already been excluded in this family (15). Family SW1168 was linked to the HNF-1α/NIDDM2 locus (4,36,41) with its maximum LOD score (1.27). It is possible that this family has a mutation in an unknown regulatory element of HNF-1α or another mutation mechanism that has not been identified using conventional mutation screening techniques.

No other genome-wide scans of MODY families have been published (with the exception of the five Swedish families in this study that were part of a young-onset type 2 diabetes genome-wide scan) (26), and so we cannot look for regions of linkage overlapping with ours using the same phenotype. Additional early-onset families with type 2 diabetes have been identified (42,43), and our regions will provide starting places for any future genome-wide screens of MODYX or early-onset type 2 diabetic families. One set of families of relevance to our study is the 32 multigenerational families with diabetes segregating in an autosomal dominant pattern identified by Doria et al. (43). An analysis of 90 cM of chromosome 12 using these families identified linkage (NPL Z-score 2.9) to chromosome 12q15 that was explained predominantly by four multigenerational families (44). We did not observe any notable evidence for linkage to this region; in the analysis of all 26 families the NPL Z-scores in this region were all negative and the HLOD 0.0. In the analysis of the subset of 14 families, the NPL Z-score was 0.48 at 88 cM (P = 0.30), and HLOD scores were 0.0.

Several genome-wide screens have been completed in common type 2 diabetes, and some of these have focused on younger-onset families or stratified results by age at onset. These include studies in North European populations and have identified, through ordered subset analysis (computation of linkage in increasing numbers of families ranked according to a potential confounding phenotype, such as age at onset or BMI) or stratification, regions linked to type 2 diabetes in patients diagnosed at an earlier age.

These include regions on chromosome 4 and 20 (45) in Ashkenazi Jews, chromosome 6 in Finnish subjects ranging in age at diagnosis from 29 to 44 years (46), chromosome 5 in Swedish subjects (47), chromosome 3 in French subjects diagnosed with type 2 diabetes before 45 years of age (48), chromosomes 5 and 22 in U.K. subjects from families where the average age at diagnosis was <45 years (49), and chromosomes 1, 5, 7, 8, 11, 15, and 21 in Swedish/Finnish subjects (26). Although it is not clear from all these studies the exact ages at diagnosis, most appear to represent subjects diagnosed under the age of 45 years. These regions may therefore be of some relevance to our study (interquartile range of diagnosis 21–49 years). Of these regions of interest, we have some degree of overlap with the U.K. study and the Lindgren et al. study. The region identified on chromosome 5 in the U.K. sib-pair study (maximum NPL 2.46 at 171 cM) coincides with our chromosome 5 region (maximum NPL 1.83 and HLOD 1.22 at 175 cM in 14 families). In addition, this region also has a maximum NPL score of 2.4 at 162 cM in 26 Swedish/Finnish young-onset families, although these 26 families include the five analyzed in our combined scan. For the chromosome 22 region (maximum NPL 2.63 at 22 pter), we observed an NPL score of 1.59 at 10 cM in all 26 families. The families in our study showing positive NPL scores in these regions are a mixture of French and U.K. Although this overlap of results should be interpreted with caution, as multiple tests have been performed they will be of interest if replicated in further studies of young-onset subjects.

In conclusion, using data from three genome-wide scans of MODYX families, we have not identified a single major novel locus but provided evidence for further heterogeneity in this important form of diabetes in childhood and in young adults.

Accession numbers and URLs for data in this article are as follows:

Online Mendelian Inheritance in Man (OMIM), http://www.ncbi.nlm.nih.gov/omim/ for MODY1 [MIM#601283], MODY2 [MIM#601407], MODY3 [MIM#600496], MODY4 [MIM*151690], MODY5 [MIM#100640], and MIDD [MIM*229300].

The GENEHUNTER software is found at Whitehead Institute for Biomedical Research/MIT Center for Genome Research, http://www.genome.wi.mit.edu/ftp/distribution/software/gh2/.

More detailed information about the GIFT collaboration can be found at http://www.gift.med.ic.ac.uk, and a description of the GIFT database can be found on http://www.biocomputing.fi/statgene/index.html.

Marker map order was obtained from http://research.marshfieldclinic.org/genetics/Map_Markers/mapmaker/MapFormFrames.html, and marker order and distances were tested in MAP-O-MAT, http://compgen.rutgers.edu/mapomat and http://www.ncbi.nlm.nih.gov/cgi-bin/Entrez/maps.cgi.

Funding is gratefully acknowledged from the EEC (QLRT-1999-00546 [GIFT]).

We are grateful to the patients and their families for their participation in our efforts to study the causes of MODY. We thank Marju Orho-Melander from the Swedish team. From the French team, we thank Philippe Gallina and Dr. Christine Bellanné-Chantelot for their contribution to family collection. From the U.K. team, we thank Amersham Parmacia Biotech for the funding of part of the study and acknowledge the MegaBACE Genotyping R&D team, especially David Shen, for their work in the early stages of this project. We also thank our colleagues at the Wellcome Trust Center for Human Genetics for their support, especially the Genotyping Core, Dr. Helen Williamson, who donated all PCR primers, and Prof. Mark McCarthy. Samantha Mason and Dr. Simon Fisher gave advice on specific PCR protocols.