Positive Association of Tyrosine Hydroxylase Microsatellite Marker to Essential Hypertension
Abstract—Despite advances in the understanding of monogenic hypertensive disorders, the genetic contribution to essential hypertension has yet to be elucidated. The position of tyrosine hydroxylase (TH) as the rate-limiting enzyme in catecholamine biosynthesis renders it a candidate gene for the etiology of hypertension. The TH gene contains an internal, informative microsatellite marker (TCAT)9. We undertook (1) an association study in a group of well-characterized hypertensive subjects (HT) and control subjects (NT) and (2) an affected sibling pair (ASP) study using sibships from our local family practices. Two hundred twenty-seven hypertensive patients (pretreatment systolic/diastolic blood pressure [BP] range, 139/94 to 237/133 mm Hg; age range [SD], 30 to 71 [8.5] years) were age- and gender-matched with 206 control subjects (BP range, 96/62 to 153/86 mm Hg; age range, 40 to 70 [7.6] years). One hundred thirty-six affected sibling pairs were recruited for our linkage study; 73 young borderline hypertensive subjects (YHT) (pretreatment BP range, 123/76 to 197/107 mm Hg; age range, 20 to 51 [9.4] years) were also recruited in whom recent pretreatment norepinephrine and epinephrine levels were available. All subjects were white. The TH short tandem repeat (STR) was amplified using specific polymerase chain reaction cycling conditions in all subjects, and products were run on an ABI 373A sequencer. TH alleles were assigned using Genescan and Genotyper software. Five TH alleles were present and designated A through E. Allele frequencies in the NT population (A, B, C, D, and E: 0.24, 0.17, 0.13, 0.20, and 0.26, respectively) were significantly different from the HT cohort (A, B, C, D, and E: 0.24, 0.19, 0.11, 0.11, and 0.35, respectively), P<0.0005 (Pearson’s test χ2=19.94; 4 df). The E allele appears overrepresented in the HT group, whereas the D allele appears to be overrepresented in the NT group. TH genotype frequencies were also significantly different between cases and controls (P<0.001; χ2=36.57; 14 df). Both groups were in Hardy-Weinberg proportion. There was a trend (NS) for the D allele to be associated with a lower BP when BP was analyzed as a quantitative trait. ASP linkage data was analyzed using Splink, a nonparametric program. Expected values for sharing 0, 1, and 2 alleles (Z0, Z1, and Z2, respectively) may be expected to be 25%, 50%, and 25%, respectively, by chance (assuming identity by descent). These probabilities were calculated by Splink as 34, 68, and 34, respectively, and compared with observed values of 36.8, 67.9, and 31.3, respectively; thus, there was no excess sharing of TH alleles among affected sibling pairs (P=0.59; logarithm of odds ratio score, 0.0). TH allele frequencies in our YHT group (A, B, C, D, and E: 0.24, 0.20, 0.12, 0.15, and 0.29, respectively) were similar to those of our NT cohort (P>0.05). There was a trend for lower pretreatment plasma norepinephrine levels with the D allele in this YHT cohort. A common and potentially functional variant at codon 81Val→Met within exon 2 of the TH gene (which we show to be in linkage disequilibrium with TH-STR) was also typed in our YHT but did not associate with catecholamine levels and is therefore unlikely to account for our findings with D and E TH-STR. In conclusion, the TH locus strongly associates with essential hypertension in a case-control model using well-characterized hypertensive and control groups. An ASP linkage model was negative, presumably because of lack of power. This study suggests that the TH gene, or a nearby gene, may be involved in the etiology of essential hypertension.
- hypertension, essential
- molecular biology
- tyrosine hydroxylase
- microsatellite marker
The role of catecholamines (CAs) in essential hypertension (EH) was extensively investigated during the 1970s and 1980s. At that stage, few individual studies acknowledged in their design and size the likelihood that EH is a polygenic condition in which only a minority of patients might have elevated CA secretion. Many methodological problems emerged, such as the interpretation of plasma or urine CA levels as indices of sympathetic activity, the relative roles of increased secretion versus reduced clearance, and above all the need for careful matching of patients and control subjects. Despite these difficulties a consensus emerged, especially from the meta-analysis of plasma CA studies1 and our own larger study of urine CA during the Medical Research Council Hypertension Trial,2 that CA secretion is increased in some patients during the development of hypertension. In older patients with EH, the role of increased CA secretion was less easy to establish, possibly because it was superseded by the secondary changes of remodeling. Even in the positive studies, the difficulty remained of determining whether increased CA secretion was a primary or secondary event. One of the attractions of the search for genetic abnormalities in EH is that these abnormalities cannot be secondary to hypertension.
Tyrosine hydroxylase (TH) catalyzes the hydroxylation of l-tyrosine to L-DOPA and is found in abundance in certain neurons of the locus ceruleus, ventral tegmental area, and substantia nigra, as well as the adrenal medulla and sympathetic ganglia.3 A single copy gene encoding human TH encompasses 13 primary exons spanning 8 kb and is located on chromosome 11p15, adjacent to Harvey ras 1 oncogene and insulin, respectively.4 The TH gene consists of at least 4 splice variants, suggesting novel means of regulating CA levels (although transgene experiments show conflicting results for its role in the cardiovascular system5 6 ). Crucially, this gene contains an informative (heterozygosity, 78%) tetranucleotide (TCAT)9 microsatellite repeat marker located within intron 1,7 enabling investigators to study the TH locus. Previous studies have used this short tandem repeat (STR) marker to elucidate the role of TH in the etiology of psychoses.8 We therefore have used this TH polymorphism to determine whether this gene is one of the many now considered to contribute, additively or epistatically, to the pathogenesis of EH.
Using this marker polymorphism we were able to assess blood pressure (BP) as (1) a quantitative trait in an association study using a group of well-characterized hypertensive and control subjects, and (2) a qualitative trait in an affected sibling pair (ASP) linkage study using sibships from our local family practices. Furthermore, a recently described common amino acid variant within exon 2 of the TH gene causing a substitution of valine for methionine at codon 819 was typed in a recently recruited group of young, mainly borderline, hypertensive patients (YHT) to determine whether this potential functional variant was in linkage disequilibrium (LD) with TH-STR. In any study of a candidate gene in EH, it is also desirable to investigate an “intermediate phenotype” that is likely to be more directly influenced by allelic variation in the gene than BP itself. The most accessible intermediate phenotype was plasma norepinephrine (NE), although it is well recognized that there are many factors other than TH that contribute to variation in plasma NE. We undertook a separate evaluation of this intermediate phenotype in the YHT cohort in which NE and epinephrine (EPI) had been estimated before any treatment over the previous 2 years, and we investigated whether any relationship existed between plasma CA levels and TH-STR or codon 81Val→Met, which could be a functional variant.
Ethical approval for this study was obtained from the local ethics committee. All subjects were recruited from Cambridgeshire and the East Anglia region in the UK.
Hypertensive Group: HT
Two hundred twenty-seven hypertensive patients (pretreatment systolic/diastolic BP range, 139/94 to 237/133 mm Hg; age range [SD], 30 to 71 [8.5] years) were selected from a population of 635 hypertensive subjects who form part of an ongoing, long-term prospective study on the effects of antihypertensive agents. In this long-term study, patients are randomized to 1 of the 4 main groups of antihypertensive treatment. On recruitment, patients are previously untreated and have either a pretreatment diastolic BP reading >90 mm Hg or a pretreatment systolic BP >160 mm Hg, each averaged over 3 consecutive readings measured using a Datascope Accutorr 2A by the same observer and sustained for 3 months. Heart rate (HR) and BP were measured in triplicate at 5-minute intervals after subjects were supine for 15 minutes. Demographic data such as age, gender, cigarette smoking, alcohol consumption, and body mass index (BMI) were recorded along with serum cholesterol levels (Table 1⇓). Patients have been followed up for up to 10 years.
Control Group: NT
Two hundred six control subjects (BP range, 96/62 to 153/86 mm Hg; age range [SD], 40 to 70 [7.6] years) were randomly selected from our locally screened population of 30 000 healthy subjects recruited from 43 general practices since 1990. An average of 3 BP readings were recorded using the same Datascope Accutorr 2A machine in these general practices. Demographic details similar to those of our HT group were documented (Table 1⇑). Subjects were excluded if they had a previous diagnosis of cardiovascular disease, hypertension, diabetes mellitus, hypercholesterolemia, stroke, angina, or myocardial infarction.
Affected Sibling Group: ASP
Hypertensive patients already receiving antihypertensive therapy were identified by local general practitioners. These probands were asked for details of siblings who were also receiving treatment. These affected siblings were then contacted, and the general practitioners for all members of each pedigree were asked to confirm details of treatment and of pretreatment BP readings. In total, 99 pedigrees (84 men; pretreatment BP range, 180/60 to 260/160 mm Hg; mean age [range/SD], 64.7 [37 to 83/10.3] years; BMI [SD], 25.6 [6.0] kg/m2) were identified, one third of which were larger than sibling pairs.
Young Hypertensive Group: YHT
To determine whether Val81Met was in LD with the TH-STR and whether there was any quantitative relationship between plasma CA and this amino acid variant, 73 patients (40 men; pretreatment BP range, 123/76 to 197/107; mean age [range/SD], 37.7 [20 to 51/9.4] years), for whom pretreatment free plasma CA levels were available, were recruited from a separate BP treatment rotation study. Patients were excluded if they were already receiving antihypertensive treatment; were taking vasoactive drugs; or were pregnant, lactating, or females of childbearing potential but not receiving adequate contraception.
All subjects from our 4 groups (HT, NT, ASP, and YHT) were white. No subject was in more than 1 group.
TH Microsatellite Genotype Analysis
The TH-tetranucleotide microsatellite marker has an expected allele length of 313 to 329 bp and a heterozygosity of 78%. A 5′TET-fluorescently labeled forward oligoprimer, which appears green when excited by a laser from our ABI 373A sequencer, was designed. The TET-labeled sense oligoprimer sequence was 5′-TCC AAA AAA TCC AAG ATG GC-3′; the unlabeled antisense oligoprimer sequence was 5′-ACA GGG AAC ACA GAC TCC ATG-3′. DNA was extracted using a standard phenol/chloroform extraction technique.10 Polymerase chain reactions (PCR) contained 100 ng genomic DNA, 26.6 pm/μL labeled forward and unlabeled reverse primer, 50 mmol/L KCl, 10 mmol/L Tris HCl, pH 9.0, 0.1% Triton X-100, 0.2 mmol/L dNTPs, 1.0 mmol/L MgCl2, and 0.2 U Taq polymerase (Promega Ltd). Reactions were scaled to 12.5 μL PCR volumes. PCR products were denatured at 94°C for 2 minutes, followed by 28 cycles at 94°C for 1 minute and 24 seconds, 55°C for 1 minute, and 72°C for 45 seconds with a final extension step of 72°C for 10 minutes, using a Biometra TRIO-Thermoblock. TH microsatellite marker for our ASP linkage study was amplified using an oil-free, 96-well Biometra UNO-Thermoblock with similar cycling conditions as above, except the annealing temperature was reduced to 52°C, the extension time was increased to 1 minute, and the PCR cycle was repeated 35 times. PCR products were loaded onto an ABI 373A sequencer (Perkin Elmer), and data were analyzed using Genescan and Genotyper software, allowing semiautomated assignment of alleles by sibships. Of all samples, 7% to 8% were reamplified and genotype was confirmed. All alleles were assigned by a single observer.
TH Codon 81 Genotype Analysis
Subjects from the YHT cohort were genotyped for the amino acid variant Val81Met. The sense oligoprimer sequence was 5′-GGC AGA GCC TCA TCG AGG AC-3′, and the antisense oligoprimer sequence was 5′-AAA CAC CTT CAC AGC TCG GGA C-3′. Oligo primers were synthesized using the ABI 391 (PCR-MATE, ABI). PCR conditions were similar to above except that DynaZyme DNA polymerase (Flowgen) was used with a standard buffer containing 1.5 mmol/L MgCl2. Reactions were scaled to 25 μL volumes. PCR products were denatured at 95°C for 2 minutes, followed by 40 cycles at 94°C for 1 minute, 63°C for 45 seconds, and 72°C for 2.5 minutes with a final extension step of 72°C for 10 minutes. The 197-bp amplified product was subsequently digested using NlaIII, which recognizes the Met but not Val sequence. The 2 resultant products (131 bp and 66 bp) were resolved using a 2% MetaPhor agarose gel (Flowgen) stained with ethidium bromide (Figure 1⇓).
Plasma samples were collected from the YHT cohort from an indwelling cannula after 15 minutes of supine rest. CAs (NE and EPI) were assayed by alumina extraction11 followed by high-performance liquid chromatography with electrochemical detection using a model 510 pump, 460 EC detector (Waters), and Spherisob ODS2, 150×4.6-mm column (Phase Separations). The mobile-phase citrate/phosphate buffer contained 30 mmol/L octanesulfonic acid, pH 6.0, and 10% vol/vol methanol, at a flow rate of 1 mL/min.
Data were analyzed using SPSS for Windows, version 6.1.4, and SPlus, version 3.3. In our association study, BP was used as the dependent variable in a multiple regression model in which age, gender, BMI, cigarette smoking, alcohol, cholesterol level, and TH genotype were predictors of outcome. Qualitative variables (genotype and allele frequencies) were analyzed using Pearson’s χ2 test. Comparison of allele frequencies has greater power than comparison of genotype frequencies because of fewer degrees of freedom, and it is therefore more sensitive to gene-disease associations, particularly when multiallelic microsatellite markers rather than biallelic polymorphisms are used. In addition, TH genotypes were analyzed using Lathrop’s test,12 which assumes Hardy-Weinberg information from our control group. In all analyses, P<0.05 was taken as statistically significant. Linkage data from our ASP study was analyzed using the nonparametric iterative program Splink,13 designed locally at Cambridge. This program negates the need to know control population allele frequencies of markers by self-generating haplotype probabilities.
Our NT and HT populations were well matched for a variety of cardiovascular risk factors (Table 1⇑). All subjects were genotyped using the ABI 373A sequencer and Genescan and Genotyper software. Five TH alleles were identified (Figure 2⇓), which were designated A, B, C, D, and E (relative sizes: 311, 315, 319, 323, and 327 bp, respectively). There was a significant difference in TH genotype frequency between the 2 groups using classic χ2 (P<0.001; χ2=36.57; 14 df) (Table 2⇓) as well as Lathrop’s test (P<0.005; χ2=31.10; 14 df). This difference was more pronounced using the more powerful chromosome analysis χ2 test (P<0.0005; χ2=19.94; 4 df) (Table 3⇓). Allele D was overrepresented in the NT cohort, while the E allele was overrepresented in the HT group. Both groups were in Hardy-Weinberg equilibrium.
BP was assessed quantitatively using a multiple regression model in both cohorts. TH genotype showed no significant relationship to systolic or diastolic BP in either group (quantitative trait analysis of BP using genotype 14 df, P>0.05; quantitative trait analysis of BP for additive allele chromosomal effects 4 df, P>0.05). However, when the 5 TH genotypes were pooled into 3 groups (A, B, C, and DE recoded into group 1; D recoded into group 2; and E recoded into group 3), there was a trend (NS) for both systolic and diastolic BP to be lower in those with the D allele in both NT and HT populations, when BP was assessed quantitatively (exemplified in Figure 3⇓ by systolic BP against pooled TH genotype in the HT cohort). There was no relationship between TH genotype and HR in our hypertensive cohort (data not shown).
We recruited 99 pedigrees (one third of which were larger than sibling pairs), totaling 136 equivalent ASPs based on a conservative estimate of K-1, where K is the number of affecteds (Table 4⇓). Allele frequencies generated by Splink (A, B, C, D, and E: 0.16, 0.22, 0.08, 0.15, and 0.39, respectively) compared very favorably to those observed in our NT population (P>0.05; χ2=6.7; 4 df). Under Mendelian inheritance of each allele, expected values for sharing 0, 1, and 2 alleles (Z0, Z1, and Z2, respectively) are expected to be 25%, 50%, and 25%, respectively, by chance (assuming identity by descent). This expected rate of allele sharing did not differ significantly from the observed scores of 27%, 50%, and 23%, respectively (Table 5⇓). Thus, there was no excess sharing of TH alleles among affected sibling pairs (P<0.6).
Seventy-three YHT were genotyped, and TH allele frequencies (A, B, C, D, and E: 0.24, 0.20, 0.12, 0.15, and 0.29, respectively) were similar to those in our NT population (P>0.05; χ2=2.2; 4 df). The allele frequencies of the Val/Met molecular variant were 0.67/0.33, respectively. Conditional probabilities that the haplotype carries the Val allele given the TH allele were A, B, C, D, and E: 0.89, 0.17, 0.83, 0.27, and 0.97, respectively, suggesting that codon Val81Met was in LD with the TH-STR (likelihood χ2 test for LD 49.4; 8 df; P<0.0001). Sixty-four subjects in our YHT cohort had pretreatment plasma NE measured (mean±SD [range], 278.8±105.0 [73 to 556] ng/L), and 49 subjects had pretreatment EPI measured (mean±SD [range], 43.5±29.9 [10 to 129] ng/L). There was a trend for pretreatment plasma NE levels to be lower in those with the D allele (Figure 4⇓), although this did not reach statistical significance. No relationship existed between pretreatment plasma CA levels and Val81Met or plasma EPI and TH pooled genotype (data not shown).
Gain of function mutations in Liddle’s14 15 and glucocorticoid remediable aldosteronism16 and loss of function in apparent mineralocorticoid excess17 have shown how monogenic disorders can cause hypertension. In most patients, however, the functional consequences of the genetic mutations are likely to be much more subtle, with some interacting to cause a higher or lower BP. Although we and others are now engaged in whole genome screens in a systematic attempt to find BP-responsible loci, it is unlikely that such studies will have the resolving power to detect many of the mutations expected in a common polygenic disease such as hypertension. There is therefore a recognized need for association studies with initial candidate genes chosen from known physiology, until more systematic genome-wide association studies become feasible.
Our specialized hypertensive practice coupled with our local stable and homogeneous population enabled us to undertake an allelic-association and linkage study. These 2 methods allow BP to be assessed both quantitatively and qualitatively using an internal, informative microsatellite tetranucleotide repeat marker within intron 1 of TH, our candidate gene. This is not the first attempt to use microsatellite sequences rather than biallelic polymorphisms in association studies.18 Our population-based case-control study strongly suggested that the TH locus is involved in the pathogenesis of EH (Tables 2⇑ and 3⇑). Of the 5 alleles identified, the D allele was overrepresented in the NT cohort, while the E allele was overrepresented in the HT group. Interestingly, the effect of the 2 alleles is counterbalanced when they occur together (Table 2⇑). The availability of pretreatment BP measurements allowed us, unusually, to investigate any quantitative effect with BP against TH genotypes. There was a trend for lower BP in those with the D allele in both NT and HT cohorts in a dominant model (Figure 3⇑), although this did not reach statistical significance. Conversely, there was a dominant effect of the E allele causing a higher BP (Figure 3⇑). The fact that the E allele did not result in even higher BP illustrates the polygenic and complex nature of BP control. Indeed, it is possible that the causes of EH will vary with age of onset, with EH in those affected at a younger age due to fewer but more potent genes compared with those with late-age onset, presumably resulting from a greater gene-environment interaction. However, the age range of subjects recruited for this study was not designed to test this hypothesis.
For a case-control study to demonstrate a 12% difference in allelic frequency of a common polymorphism would require 250 affected subjects with an equal number of controls to achieve a significant difference at the 5% level, with 80% power.19 Thus, although this work was well powered, the accuracy in choosing the unaffected control group in an association model is clearly crucial to these types of study. From our large database of 30 000 locally screened normotensive individuals, we randomly selected 206 NT subjects who were well matched to our HT cohort. These individuals have been followed up for 5 years. Commentators have rightly highlighted the main failings of association studies, particularly regarding recruitment of unaffected groups.20 All our subjects were recruited from the same geographical area with a relatively healthy, stable, and homogeneous population. This area has low cardiovascular environmental risk factors, amplifying any genetic contribution to hypertension.
Access to a group of young borderline hypertensives enabled us to show from conditional haplotype probabilities that a recently identified Val81Met common amino acid variant of potential functional importance9 was in LD with TH-STR. This particular YHT cohort was selected because pretreatment BP and CA levels were available; because 60% to 70% of plasma CA variance is known to be under genetic influence,21 22 we investigated whether Val81Met or TH-STR could account for this variance. Moreover, we hypothesized that a younger cohort with hypertension, albeit borderline, would increase our chances of detecting a genetic contribution to BP regulation. We were unable to demonstrate any obvious relationship between either BP or plasma CA levels and Val81Met, but a larger cohort of subjects is clearly needed to exclude this amino acid variant in the control of either of these 2 variables.
Other investigators have shown that STR sequences could by themselves be functionally active.23 Indeed, the D allele was associated with lower BP (Figure 3⇑), and this allele, as would be predicted, also correlated with lower plasma NE levels (Figure 4⇑), although neither result reached statistical significance. Plasma NE levels are difficult to measure reliably and reproducibly, but their availability in a pretreated young hypertensive cohort provided us with the ability to perform this useful genotype-phenotype analysis. Study of a larger group of hypertensives for whom pretreatment CA levels are available may confirm or refute the suggestion that the NE levels are influenced by TH-STR.
A linkage study was undertaken with ASPs from a separate cohort using the TH marker. With 136 equivalent sibling pairs and a marker (calculated in this study) with a heterozygosity of 74.5%, our effective sample size was 67.2 sibling pairs (ie, the number of fully informative sibpairs that would carry the same information), allowing in this study a maximum logarithm of odds ratio (LOD) likelihood score of 4.1, assuming a λs of 3.5.24 With this number of ASPs, we were unable to demonstrate linkage, but an LOD score of 0.0 (Table 5⇑) does not exclude the TH locus. Thus, this study illustrates the wide divergence in power, discussed recently by Risch and Merikangas,25 between the association and linkage approaches. To detect a small genetic contribution to EH onset by the TH locus would require many thousands of ASPs, compared with the relatively practical number required for an allelic-association model. Moreover, although case-control studies cannot exclude the possibility of unsuspected population stratification in 1 of the groups, they are at present the only pragmatic type of association study in the absence of the parental genotypes required for the family-based alternatives.25 26
Naturally, the results of this work apply best to the population studied, but this study supports the role of the TH locus (or a nearby gene) in the pathogenesis of EH.
Dr Sharma is a British Heart Foundation Clinician Scientist. We are grateful to Claire Dickerson and Chryssie Brown for help with recruitment of our YHT group and sibling pairs, respectively, and Jennie Fatibene for extracting DNA from the sibling blood samples. We are indebted to our patients and local family practitioners.
Reprint requests to Dr P. Sharma, Addenbrooke’s Hospital, Hills Rd, Cambridge CB2 2QQ, UK.
- Received December 30, 1997.
- Revision received January 22, 1998.
- Accepted June 2, 1998.
Goldstein DSK. Plasma catecholamines and essential hypertension: an analytical review. Hypertension. 1983;5:86–99.
Brown MJ, Causon RC, Barnes VF, Brennan P, Barnes G, Greenberg G. Urinary catecholamines in essential hypertension: results of 24-hour urine catecholamine analyses from patients in the Medical Research Council Trial for mild hypertension and from matched controls. Q J Med. 1985;57:637–651.
Polymeropoulos MH, Xiao H, Rath DS, Merril CR. Tetranucleotide repeat polymorphism at the human tyrosine hydroxylase gene. Nucleic Acids Res. 1991;19:3753. Abstract.
Meloni R, Leboyer M, Bellivier F, Barbe B, Samolyk D, Allilaire JF, Mallet J. Association of manic-depressive illness with tyrosine hydroxylase microsatellite marker. Lancet. 1995;345:932. Letter.
Lahiri DK, Nurnberger JI. A rapid non-enzymatic method for the preparation of HMW DNA from blood for RFLP studies. Nucleic Acids Res. 1991;19:5444.
Shimkets RA, Warnock DG, Bositis CM, Nelson-Williams C, Hansson JH, Schambelan M, Gill JR, Ulick S, Milora RV, Findling JW, Canessa CM, Rossier BC, Lifton RP. Liddle’s syndrome: heritable human hypertension caused by mutations in the beta subunit of the epithelial sodium channel. Cell. 1994;79:407–414.
Undlien DE, Bennett ST, Todd JA, Akselsen HE, Ikaheimo I, Reijonen H, Knip M, Thorsby E, Ronningen KS. Insulin gene region-encoded susceptibility to IDDM maps upstream of the insulin gene. Diabetes. 1995;44:620–625.
Caulfield M, Newell-Price J. The angiotensin converting enzyme gene in cardiovascular disease. Br Heart J. 1995;74:207–208.
Bennett ST, Lucassen AM, Gough SCL, Powell EE, Undlien DE, Pritchard LE, Merriman ME, Kawaguchi Y, Dronsfield MJ, Pociot F, Nerup J, Bouzekri N, Cambon-Thomsen A, Ronningen KS, Barnett AH, Bain SC, Todd JA. Susceptibility to human type 1 diabetes at IDDM2 is determined by tandem repeat variation at the insulin gene minisatellite locus. Nat Genet. 1995;9:284–292.
Risch N, Merikangas K. The future of genetic studies of complex human diseases. Science. 1996;273:1516–1517.
Risch N, Merikangas K. Genetic analysis of complex diseases. Science. 1997;275:1329–1330.