Type II SH2 domain–containing inositol 5-phosphatase (INPPL1, or SHIP2) plays an important role in the control of insulin sensitivity. INPPL1 mutations affecting gene function have been found in rat models of type 2 diabetes and hypertension and in type 2 diabetic patients. We investigated the influence of nucleotide variation in INPPL1 on components of the metabolic syndrome. Following comprehensive resequencing of the gene, we genotyped 12 informative polymorphisms in 1,304 individuals from 424 British type 2 diabetes families that were characterized for several metabolic phenotypes. We have found highly significant associations of single nucleotide polymorphisms (SNPs) and haplotypes of INPPL1 with hypertension as well as with other components of the metabolic syndrome. In a cohort of 905 French type 2 diabetic patients, we found evidence of association of INPPL1 SNPs with the presence of hypertension. We conclude that INPPL1 variants may impact susceptibility to disease and/or to subphenotypes involved in the metabolic syndrome in some diabetic patients.
The metabolic syndrome, a cluster of metabolic abnormalities that can include type 2 diabetes, hypertension, central obesity, and dyslipidemia, occurs at a high rate in the U.S., Europe, and Scandinavia (1–4). Insulin resistance is the focal component of the metabolic syndrome, and both genetic and environmental factors contribute to its development (4). Much progress has been made toward understanding the regulation of insulin action and the molecular defects that give rise to insulin resistance (5).
INPPL1 is a negative regulator of insulin signaling, and Inppl1 inactivation in mice results in increased insulin sensitivity (6). INPPL1 is in human chromosome 11q13-14, which has suggested linkage to type 2 diabetes, hypertension, and insulin resistance (7–9). In the rat, Inppl1 is in an interval on chromosome 1 linked to glucose intolerance and adiposity in the spontaneously diabetic (type 2) Goto-Kakizaki rat (10,11) and to hypertension in the spontaneously hypertensive rat (12). We identified a missense variant in Inppl1 specific to both rat strains that increases inhibition of insulin signaling compared with wild-type Inppl1 (10). Furthermore, in type 2 diabetic patients, we found a 16-bp deletion of an adenylate/uridylate-rich element in the 3′ untranslated region of INPPL1, which in vitro causes changes consistent with a role in insulin resistance (10). Seven of nine patients with this mutation were hypertensive, and five were obese.
To investigate the role of genetic variation in INPPL1 in the metabolic syndrome, we first resequenced the gene (15.2 kb), including all exons and introns in a panel of 64 individuals. We describe a total of 49 variants (Fig. 1; a table of single nucleotide polymorphisms [SNPs] is available in the online appendix at http://diabetes.diabetesjournal.org and at http://www.well.ox.ac.uk/rat_mapping_resources/SHIP2). At first we genotyped all polymorphic markers, but in later stages we discarded rare markers, and for SNPs 1, 3, 5, and 6, which were in complete linkage disequilibrium (Δ2 = 1), only snp1 was used for genotyping (linkage disequilibrium results are available in the online appendix). The 12 markers chosen (Fig. 1) represent 79% of haplotype variation in the gene. These markers were genotyped in 1,304 individuals from 424 British type 2 diabetic families (Diabetes in Family [DIF] study collection) and were confirmed to be in Hardy-Weinberg equilibrium.
The computer program Transmit (13) was used to analyze the data with a family-based test of linkage and association to type 2 diabetes, hypertension, central obesity, general obesity, or the metabolic syndrome. Its score-test function minimizes problems caused by population stratification, and it provides a complementary test to the transmission disequilibrium test (14). Transmit makes full use of available parental data and, in addition, efficiently incorporates data from families with missing parental genotypes, common in late-onset diseases. Because parents were available for only 8% of our families, these features of Transmit were deciding factors in our choice of analytical method. Empirical P values were calculated using methods similar to those found in Martin et al. (15) and were corrected for multiple phenotypes testing for the first associated marker in Table 1.
A summary of results for the DIF study families is shown in Table 1. The strongest evidence for association in this collection is between hypertension and a group of three INPPL1 SNPs, rs2276047, snp8, and rs9886 (P = 9.3 × 10−6), which also are associated with central obesity (P = 3.2 × 10−4). Together, rs2276047 and snp8 show evidence for association with type 2 diabetes (P = 6.2 × 10−4) and with the metabolic syndrome as defined by the National Cholesterol Education Program Expert Panel on Detection, Evaluation, and Treatment of High Blood Cholesterol in Adults (16) (P = 2.3 × 10−3). The variants rs2276047 and rs9886 together are associated with general obesity (P = 1.5 × 10−3). The most common haplotype, I-A-G (snp8-rs2276047-rs9886), has the single best haplotype association with all five traits (Table 2). The 16-bp deletion in the 3′ untranslated region of INPPL1 is not associated with any of the phenotypes measured but occurs in 2.0% of the diabetic patients and 0.7% of the control subjects. Bootstrap analyses implemented in Transmit, which compensate for the effects of possible genotyping errors in the data, were consistent with these findings (http://www-gene.cimr.cam.ac.uk/clayton/software/transmit.txt). The DIF study data were also analyzed using the pedigree disequilibrium test (PDT) (17; results available in the online appendix). The PDT test statistic is significant for the association of hypertension with INPPL1 SNPs, but the statistics for other phenotypes are not significant. We believe that this is due to the lower power of PDT, which we have confirmed based on investigation of both programs’ performance using simulations with the DIF study family structures.
Based on the above results, INPPL1 variants rs2276047, snp8, rs9886, and the 16-bp deletion were genotyped in 905 unrelated French type 2 diabetic patients and 380 population control subjects. Allele frequencies of the INPPL1 variants are similar between the French diabetic patients and control subjects, with some trends in the opposite direction from DIF study results (Table 3, left side). Log linear analysis of genotype and haplotype frequencies for markers rs2276047 and snp8 does not show association with type 2 diabetes in the case-control study (P > 0.15). When we compare French diabetic patients with and without hypertension, however, the I allele of snp8 exhibits association with hypertension, although the evidence is not highly significant (P = 0.011) (Table 3, right side). Also, the haplotype I-A-G (snp8-rs2276047-rs9886) is significantly more frequent in French diabetic patients with hypertension compared with those without (0.623 and 0.603, respectively; P = 0.033). No significant differences are found between hypertensive French diabetic patients and French population control subjects (data not shown).
Various reasons may be responsible for differences in INPPL1 association in the French diabetic patients compared with the DIF study families. Our ability to detect association in the DIF study collection may be due in part to the extensive phenotyping performed on those subjects, which were not fully available for the French diabetic patients and not at all for the French control subjects. The French type 2 diabetic patients are a selected clinic group that is quite different from the DIF population-based sample (Table 4). The French patients have higher BMI, although BMI is increased by sulfonylurea or insulin treatment, which are also more common for these patients. On average, they have higher HbA1c and a greater proportion of them are hypertensive. Haplotype frequencies estimated for the two collections were remarkably similar and are consistent with a common U.K. and French INPPL1 gene genealogy (see the online appendix). The possibility exists that the DIF study association of INPPL1 to diabetes may be a false positive because the French data showed some trend in the opposite direction.
Another possibility for the differences in association could be insufficient power in the case-control study. A rough estimate of the strength of the INPPL1–type 2 diabetes association can be derived from the DIF family-based study (odds ratio [OR] = 1.6, 95% CI 1.3–2.0). This calculation is based on the χ2 statistic associated with the high-risk haplotype, together with an estimate of the equivalent number of informative transmissions from the score statistics’ variance. Based on an OR of 1.6, we estimate 99% power to detect an association with diabetes with 905 case and 380 control subjects; however, if the true OR is at the lower end of the range (1.3), we would only have 81% power, i.e., a 19% chance of missing the association. For the French cohort, the estimated OR for diabetes is 1.0 (0.8–1.2) and is clearly discrepant with our findings in the DIF study collection. Further studies with larger case-control populations would seem to be indicated, particularly in light of the possible association with hypertension in the two diabetic cohorts.
In summary, in a well-characterized collection of U.K. type 2 diabetic families, we have found a significant association of variants in INPPL1 with hypertension and other features of the metabolic syndrome. We also observed significant association with presence/absence of hypertension in French diabetic patients. The finding that common polymorphisms of INPPL1 are associated with subphenotypes of the metabolic syndrome and may have an important impact on susceptibility to the conditions in these populations provides strong motivation for further studies of this gene in other disease cohorts.
RESEARCH DESIGN AND METHODS
Informed consent was obtained from all subjects, and the investigation was conducted according to the principles expressed in the Declaration of Helsinki.
Type 2 diabetic family collection.
The DIF study identified British families of European descent containing a type 2 diabetic patient with one or more siblings. Table 4 provides their general phenotypic information (1,304 individuals from 424 families; 41 families had at least one parent available). Type 2 diabetes status was ascertained by patient history or fasting plasma glucose ≥7.0 mmol/l and HbA1c >6.0% or fasting plasma glucose >7.8 mmol/l, irrespective of HbA1c. Hypertension criteria were systolic blood pressure ≥160 mmHg and/or diastolic blood pressure ≥90 mmHg or antihypertensive treatment. General obesity threshold was BMI ≥30 kg/m2, and, for central obesity, waist circumference ≥102 cm for men or ≥88 cm for women (16). Metabolic syndrome was defined as by the National Cholesterol Education Program Expert Panel on Detection, Evaluation, and Treatment of High Blood Cholesterol in Adults (16).
French case-control study.
The 905 type 2 diabetic patients of European descent who participated consecutively attended the hospital Sud-Francilien and had complete annual examinations (Table 4). Diabetes was defined as fasting plasma glucose >7 mmol/l, treatment by oral antidiabetic agents, or by insulin treatment following ≥2 years of oral antidiabetic treatment (Table 4). Hypertension was defined by systolic and diastolic blood pressure >160 and >80 mmHg, respectively, and/or antihypertensive treatment. In addition, 380 self-described Caucasian control subjects from the Paris area were tested. No phenotypic data were available for the control subjects.
SNP discovery and typing.
The 15.2 kb containing the entire INPPL1 gene from genome sequence NT_033927 was resequenced in pools of DNA from 64 individuals (32 Caucasians and 32 Africans, two DNA per pool). Regions containing SNPs were resequenced in a panel of 24 U.K. type 2 diabetic patients and 24 U.K. control subjects to confirm their polymorphism in our study population. The National Center for Biotechnology Information dbSNP database was also searched for INPPL1 SNPs (http://www.ncbi.nlm.nih.gov/SNP).
SNP genotyping used primer extension methods (18,19), except for the insertion/deletion variants (snp8 and snp30), which were detected by size separation on agarose gels. Primer sequences are available in the online appendix. The genotyping percentage was 81–96% successful out of attempts and 90–94% for the markers showing association, snp8, rs2276047, and rs9886.
Statistical analysis.
The genotyping data for the DIF family-based study was checked for segregation inconsistencies with PedCheck version 1.1 software (20). Standard contingency table methods were used to compare allele frequencies in case and control subjects and to assess Hardy-Weinberg equilibrium of genotype frequencies. Case-control analysis of multiple SNPs was carried out using a log-linear modeling approach (21) and the program Phase version 2 (22).
The computer program Transmit (13) was used to apply a family-based test of linkage and association. The Transmit statistic can cause increased false-positive results when only affected siblings are used for inferring the parental alleles, but the extent of the inflation is reduced significantly upon addition of unaffected/unselected siblings (15). Because most DIF study families contained one or more unaffected siblings, we expected minimal type 1 error inflation. This was confirmed by running simulation studies (23) with matching family structures to the DIF study collection and two susceptibility gene models with low penetrance and common allele frequencies believed to be characteristic in complex diseases (see online appendix). To account for the small increase in type 1 error, empirical P values have been calculated based on the excellent fit of a linear regression of the log(empirical P value) on log(size).
Transmit implements both allele and haplotype tests of transmission disequilibrium. A forward-selection strategy was implemented to build a haplotype linked/associated with susceptibility to the various phenotypes (13). A haplotype was extended when the overall significance of the gene association was increased, based on the asymptotic P value; this takes into account the increasing number of degrees of freedom with longer haplotypes. Haplotype analyses of four or more polymorphisms were not practicable with the current version of Transmit due to computational complexity. Haplotype frequencies were calculated using Merlin/Fugue for family data (24), SNPHAP for the case-control study (D. Clayton, http://cigmr.man.ac.uk/geneticanalysis/progs/manuals/text/readme_snphap.txt), and SNPtagger to estimate coverage of haplotype variation (25).
Locations of polymorphisms in INPPL1. The 12 markers used in genotyping the DIF diabetic families are shown above the diagram of the exons and introns of the gene, and the position of other SNPs are shown below. The three variants showing association to components of the metabolic syndrome are enclosed in boxes.
Locations of polymorphisms in INPPL1. The 12 markers used in genotyping the DIF diabetic families are shown above the diagram of the exons and introns of the gene, and the position of other SNPs are shown below. The three variants showing association to components of the metabolic syndrome are enclosed in boxes.
Results of Transmit analysis of INPPL1 polymorphisms in the DIF family collection
| Phenotype | First marker | Second marker | Third marker | Entry of first marker | Entry of second marker | Entry of third marker | ||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| χ2 | P (corrected) | χ2 | P | χ2 | P | |||||||
| Hypertension | rs2276047 (A) | snp8 (I) | rs9886 (G) | 17.96 | 3.9E-04 | 27 | 2.4E-05 | 39.04 | 9.3E-06 | |||
| Central obesity | rs2276047 (A) | rs9886 (G) | snp8 (I) | 11.59 | 6.9E-03 | 20.48 | 3.4E-04 | 29.50 | 3.2E-04 | |||
| Type 2 diabetes | rs2276047 (A) | snp8 (I) | snp16 | 10.51 | 1.1E-02 | 19.09 | 6.2E-04 | 22.86 | 1.7E-03 | |||
| General obesity | rs2276047 (A) | rs9886 (G) | rs2276048 | 9.41 | 1.9E-02 | 16.95 | 1.5E-03 | 22.80 | 1.7E-03 | |||
| Metabolic syndrome | rs2276047 (A) | snp8 (I) | rs2276048 | 11.20 | 8.2E-03 | 15.93 | 2.3E-03 | 20.24 | 7.8E-03 | |||
| Phenotype | First marker | Second marker | Third marker | Entry of first marker | Entry of second marker | Entry of third marker | ||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| χ2 | P (corrected) | χ2 | P | χ2 | P | |||||||
| Hypertension | rs2276047 (A) | snp8 (I) | rs9886 (G) | 17.96 | 3.9E-04 | 27 | 2.4E-05 | 39.04 | 9.3E-06 | |||
| Central obesity | rs2276047 (A) | rs9886 (G) | snp8 (I) | 11.59 | 6.9E-03 | 20.48 | 3.4E-04 | 29.50 | 3.2E-04 | |||
| Type 2 diabetes | rs2276047 (A) | snp8 (I) | snp16 | 10.51 | 1.1E-02 | 19.09 | 6.2E-04 | 22.86 | 1.7E-03 | |||
| General obesity | rs2276047 (A) | rs9886 (G) | rs2276048 | 9.41 | 1.9E-02 | 16.95 | 1.5E-03 | 22.80 | 1.7E-03 | |||
| Metabolic syndrome | rs2276047 (A) | snp8 (I) | rs2276048 | 11.20 | 8.2E-03 | 15.93 | 2.3E-03 | 20.24 | 7.8E-03 | |||
In boldface type are the associated markers in order of significance, the associated allele (shown in parentheses), and the most significant P value. Empirical P values have been calculated using methods similar to those in reference 16, and P values for the first marker have been corrected for multiple phenotypes tested.
Estimated frequencies for the three most common haplotypes of the associated markers in chromosomal order
| snp8 | rs2276047 | rs9886 | Frequency (%) | Proportion transmitted to affected individuals | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Hypertension | Type 2 diabetes | General obesity | Central obesity | Metabolic syndrome | ||||||||
| I | A | G | 63.5 | 0.64 ± 0.03 | 0.61 ± 0.03 | 0.60 ± 0.03 | 0.59 ± 0.03 | 0.59 ± 0.03 | ||||
| I | G | G | 17.6 | — | — | — | — | — | ||||
| D | A | G | 13.8 | — | — | — | — | — | ||||
| snp8 | rs2276047 | rs9886 | Frequency (%) | Proportion transmitted to affected individuals | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Hypertension | Type 2 diabetes | General obesity | Central obesity | Metabolic syndrome | ||||||||
| I | A | G | 63.5 | 0.64 ± 0.03 | 0.61 ± 0.03 | 0.60 ± 0.03 | 0.59 ± 0.03 | 0.59 ± 0.03 | ||||
| I | G | G | 17.6 | — | — | — | — | — | ||||
| D | A | G | 13.8 | — | — | — | — | — | ||||
Data are means ± SE, unless noted otherwise. The remaining haplotype combinations occur at a frequency of <3%.
Allele counts and frequencies of INPPL1 variants in a case-control study of French type 2 diabetic patients (n = 905) and population control subjects (n = 380)
| Variant (allele) | Type 2 diabetes | Control | Type 2 diabetes with HTN | Type 2 diabetes without HTN |
|---|---|---|---|---|
| snp8 | ||||
| I | 1,552 (0.85) | 656 (0.87) | 1,201 (0.86) | 351 (0.81) |
| D | 280 (0.15) | 94 (0.13) | 197 (0.14) | 83 (0.19) |
| χ2 | 3.250 | — | 6.479 | — |
| P | 0.071 | — | 0.011 | — |
| rs2276047 | ||||
| G | 382 (0.22) | 144 (0.21) | 306 (0.23) | 86 (0.20) |
| A | 1,368 (0.78) | 536 (0.79) | 1,052 (0.77) | 340 (0.80) |
| χ2 | 0.12 | — | 1.040 | — |
| P | 0.729 | — | 0.308 | — |
| snp31 (16-bp deletion) | ||||
| I | 1,731 (0.99) | 739 (0.99) | 1,347 (0.99) | 416 (1.00) |
| D | 13 (0.01) | 5 (0.01) | 11 (0.01) | 2 (0.00) |
| χ2 | 0.04 | 0.484 | ||
| P | 0.841 | 0.487 | ||
| rs9886 | ||||
| C | 69 (0.05) | 46 (0.06) | 56 (0.05) | 11 (0.03) |
| G | 1,377 (0.95) | 668 (0.94) | 1,004 (0.95) | 311 (0.97) |
| χ2 | 2.65 | 1.866 | ||
| P | 0.104 | 0.172 |
| Variant (allele) | Type 2 diabetes | Control | Type 2 diabetes with HTN | Type 2 diabetes without HTN |
|---|---|---|---|---|
| snp8 | ||||
| I | 1,552 (0.85) | 656 (0.87) | 1,201 (0.86) | 351 (0.81) |
| D | 280 (0.15) | 94 (0.13) | 197 (0.14) | 83 (0.19) |
| χ2 | 3.250 | — | 6.479 | — |
| P | 0.071 | — | 0.011 | — |
| rs2276047 | ||||
| G | 382 (0.22) | 144 (0.21) | 306 (0.23) | 86 (0.20) |
| A | 1,368 (0.78) | 536 (0.79) | 1,052 (0.77) | 340 (0.80) |
| χ2 | 0.12 | — | 1.040 | — |
| P | 0.729 | — | 0.308 | — |
| snp31 (16-bp deletion) | ||||
| I | 1,731 (0.99) | 739 (0.99) | 1,347 (0.99) | 416 (1.00) |
| D | 13 (0.01) | 5 (0.01) | 11 (0.01) | 2 (0.00) |
| χ2 | 0.04 | 0.484 | ||
| P | 0.841 | 0.487 | ||
| rs9886 | ||||
| C | 69 (0.05) | 46 (0.06) | 56 (0.05) | 11 (0.03) |
| G | 1,377 (0.95) | 668 (0.94) | 1,004 (0.95) | 311 (0.97) |
| χ2 | 2.65 | 1.866 | ||
| P | 0.104 | 0.172 |
HTN, hypertension. Frequencies are in parentheses.
Characteristics of DIF family members and French diabetic patients (no phenotypic data are available for the control subjects)
| DIF nondiabetic subjects | DIF diabetic patients | P | French diabetic patients | P vs. DIF diabetic patients | |
|---|---|---|---|---|---|
| n | 723 | 581 | — | 905 | — |
| Sex (M/F) | 305/417 (42/58) | 260/321 (45/55) | <0.0001 | 365/540 (40/60) | 0.095 |
| Age at this study (years) | 61 (52–68) | 64 (57–71) | <0.0001 | N/A | — |
| Age at diagnosis (years) | — | 54 (46–62) | — | 48 (41–55) | <0.0001 |
| Waist (cm) | 94 (82–108) | 103 (91–117) | <0.0001 | N/A | — |
| BMI (kg/m2) | 27.8 (23.5–33.0) | 29.9 (25.1–35.6) | <0.0001 | 33.2 (27.8–39.7) | <0.0001 |
| HBA1c (%) | 5.8 (4.7–7.0) | 7.7 (6.3–9.4) | <0.0001 | 8.6 (6.6–11.1) | <0.0001 |
| Total cholesterol (mmol/l) | 5.8 (4.7–7.0) | 5.5 (4.5–6.7) | <0.0001 | 5.8 (4.5–7.0) | — |
| HDL cholesterol (mmol/l) | 1.28 (0.97–1.71) | 1.12 (0.85–1.46) | <0.0001 | 1.08 (0.80–1.47) | 0.063 |
| Triglycerides (mmol/l) | 1.4 (0.8–2.3) | 1.7 (1.0–2.9) | — | 1.95 (1.09–3.49) | — |
| Hypertension (No/Yes)* | 428/295 (59/41) | 224/397 (38/61) | <0.0001 | 208/697 (23/77) | <0.0001 |
| Diabetes treatment (diet only/oral/insulin) | Not applicable | 284/199/98 (49/34/17) | 19/545/319 (2/62/36) | <0.0001 |
| DIF nondiabetic subjects | DIF diabetic patients | P | French diabetic patients | P vs. DIF diabetic patients | |
|---|---|---|---|---|---|
| n | 723 | 581 | — | 905 | — |
| Sex (M/F) | 305/417 (42/58) | 260/321 (45/55) | <0.0001 | 365/540 (40/60) | 0.095 |
| Age at this study (years) | 61 (52–68) | 64 (57–71) | <0.0001 | N/A | — |
| Age at diagnosis (years) | — | 54 (46–62) | — | 48 (41–55) | <0.0001 |
| Waist (cm) | 94 (82–108) | 103 (91–117) | <0.0001 | N/A | — |
| BMI (kg/m2) | 27.8 (23.5–33.0) | 29.9 (25.1–35.6) | <0.0001 | 33.2 (27.8–39.7) | <0.0001 |
| HBA1c (%) | 5.8 (4.7–7.0) | 7.7 (6.3–9.4) | <0.0001 | 8.6 (6.6–11.1) | <0.0001 |
| Total cholesterol (mmol/l) | 5.8 (4.7–7.0) | 5.5 (4.5–6.7) | <0.0001 | 5.8 (4.5–7.0) | — |
| HDL cholesterol (mmol/l) | 1.28 (0.97–1.71) | 1.12 (0.85–1.46) | <0.0001 | 1.08 (0.80–1.47) | 0.063 |
| Triglycerides (mmol/l) | 1.4 (0.8–2.3) | 1.7 (1.0–2.9) | — | 1.95 (1.09–3.49) | — |
| Hypertension (No/Yes)* | 428/295 (59/41) | 224/397 (38/61) | <0.0001 | 208/697 (23/77) | <0.0001 |
| Diabetes treatment (diet only/oral/insulin) | Not applicable | 284/199/98 (49/34/17) | 19/545/319 (2/62/36) | <0.0001 |
Data are geometric means (SD range); n (%) for sex, hypertension, and diabetes treatment; and median (interquartile range) for age at study and age at diagnosis. For the French patients, BMI, HbA1c, and lipid measurements represent the highest value over time of attending the diabetes clinic. Values for DIF subjects were collected at the time of this study.
Hypertension is defined as blood pressure ≥160/90 mmHg or antihypertensive treatment. N/A, not available.
Additional information for this article can be found in an online appendix at http://diabetes.diabetesjournals.org.
Article Information
This work is supported by the Wellcome Trust (057733), the Wellcome Cardiovascular Functional Genomics Initiative (066780/Z/01/Z), and Diabetes U.K. (RD96/0001270). D.G. holds a Wellcome senior fellowship in basic biomedical science. S.P.W. is a Wellcome Prize Student in bioinformatics and statistical genetics.
The U.K. collection of diabetic families was developed by the late Professor R. Turner. We are grateful to Gbenga Kazeem for running the case-control haplotype analysis.
