We studied 574 adults, 294 men and 280 women, ranging between 18 and 90 years of age. They belonged to 126 two- to four-generation pedigrees. The subjects were uniformly distributed with respect to age between 20 and 70 years. The pedigrees comprised 3 to 14 persons. The studied individuals were all Chuvashians (Caucasians) from small villages in the Chuvasha and Bashkirostan autonomies of the Russian Federation. Their population is demographically stable and they have lived there for centuries. The environmental conditions, particularly dietary influences, have been relatively constant and genetic flow has been minimal 17 . Further details on this relatively isolated population, and our contacts with them, are given elsewhere 18 .

Our genetic field work in Chuvasha consisted of a complete history, physical examination, and health questionnaire conducted by a native speaker. All diagnostic measures were in compliance with the Helsinki Declaration. Written, informed consent was obtained after due approval by the local ethics committee. Plain posteroanterior radiographs of both hands were taken from each study participant, with the x-ray source located 60 cm above and using a standard radiographic technique 18 . Digital images were created from all radiographs. The extent of OA development was evaluated for each of the 14 joints of each hand separately, in accordance with the grading system of Kellgren and Lawrence 19 . The OA evaluation was based on radiographic changes, such as presence of osteophytes, joint-space narrowing, subchondral sclerosis, lateral deformity, or cortical collapse. Development of OA at each joint was graded from 0 to 4. Since the OA scores for individual joints are intercorrelated, the total individual OA score was obtained from principal component analysis of grade sums for all assessed joints. First factor scores (FS1-OAs) were then used in further analyses as a characteristic of hand OA. We recently described the method in detail elsewhere 20 . To assess the reproducibility of our basic phenotype, the evaluations of each trait was performed twice on 50 randomly chosen radiographs by the same investigator 10 days apart. The κ statistic showed high intraobserver reproducibility for the Kellgren and Lawrence (K–L) score (κ = 0.87; P < 0.01) and was in good agreement with that found in other studies on a similar subject 21 .

DNA was prepared from peripheral blood lymphocytes by standard techniques. Single-nucleotide polymorphisms (SNPs) were previously identified when sequencing the ENPP1 gene in 23 unrelated patients with generalized arterial calcification of infancy, together with their parents and a number of control individuals 9 12 . All but the missense mutation R774C can also be found in the public databases 22 and may be referred to by their rs numbers. Genotyping was performed by Pyrosequencing™ on the PSQ™ HS 96A System (Biotage AB, Uppsala, Sweden). Primer sequences for the assays are available upon request. Amplification conditions were standard as specified by the supplier. Controls were included to exclude mix-ups and other errors during genotyping. Thus, each plate contained a well with DNA-free reaction mix to detect contamination with DNA. Another well contained a dedicated DNA, which was expected to yield identical genotypes for all plates genotyped for a given genetic variant.

We identified three new short tandem repeat (STR) markers at the ENPP1 locus at 6q, one of the (tcct)_n tetranucleotide repeat type (M06NR1A) and two of the common (ca)_n dinucleotide repeat type (M06NR2A and M06NR3A), and genotyped them in all samples in addition to the SNPs. The exact position in base pairs of the three STR markers in contig NM_006208.1 is at 8776828ff base pairs for M06NR1A, 8862795ff base pairs for M06NR2A, and 8911166ff for M06NR3A. For each marker, 6 ng of genomic DNA was amplified in a 10 μl reaction volume on an MJ PTC 225 Tetrad Cycler. The PCR mix contained 0.53 μM each primer (sequences are available upon request), 0.1 μM each dNTP, 0.5 U Taq polymerase, 1 × reaction buffer with 1.5 mM MgCl₂. The forward primers were labelled at their 5' ends with FAM. Genotypes were determined on a MegaBACE 1000 automated sequencer (Amersham Biosciences, Freiburg, Germany). For allele calling, the proprietary Genetic Profiler software version 1.5 from Amersham Biosciences was used.

Genotyping data were checked for Mendelian errors with the PedCheck program 23 . In two cases, implausible Mendelian errors were detected, and the two probands were excluded from further analysis. It was also tested whether the observed recombination rates between the markers were in accordance with their distance. We did not observe any hint of genotyping errors.

Haplotypes for all individuals were determined by using the program Genehunter version 2.1 24 . SNPs were arranged according to their location on the chromosome in the following order: rs1800949, rs858342, rs1044498 (K173Q), R774C.

Data analysis was carried out in two steps. We first used variance component analysis as implemented in the FISHER program 25 to assess the contribution of genetic and common environmental factors in families to OA variation as compared to the influence of potential covariates, such as sex, age, body weight, and height. We recently described the method in detail elsewhere 26 . Briefly, the program uses a maximum likelihood ratio test (LRT) as a model-fitting technique. It simultaneously assesses the contribution of each of the potential covariates (sex, age, etc.) and the contributions of the putative sources of the familial variation, namely, additive genetic effect (V_AD) and common environment shared by parents (V_SP), by siblings (V_SB), and by all members of nuclear pedigrees/household (V_HS). First factor scores (FS1-OAs) were adjusted for all significant covariates, that is, age and sex, before proceeding to the second step of the analysis.

As a second step, we then employed transmission disequilibrium tests (TDTs) to detect an association between hand OA as a quantitative trait (FS1-OA) and selected DNA marker alleles. Three TDT-like tests were carried out for each pair of dependent variable and specific genetic marker or haplotype. They included the orthogonal test (OT) proposed by Abecasis and colleagues 27 and implemented in the quantitative transmission disequilibrium test (QTDT) program, the family-based association test (FBAT) proposed by Horvath and colleagues 28 and implemented in the FBAT program, and the extreme offspring design t-test (EOT) proposed by Malkin and colleagues 29 and implemented in the MAN-6 package. The OT is the maximum likelihood test based on orthogonal decomposition of genotype scores. The significance of the additive impact of within-family genotype score on the phenotype is tested in the OT. The FBAT examines a similar hypothesis – that the phenotype is independent of a specific genotype – but by different statistical algorithms. The EOT extends the ideology of Allison's Q3 test 30 by optimal choice of the extreme offspring in each nuclear family.

The simultaneous application of different tests requires interpretation and pooling together of the various P values obtained in testing the same pedigree sample. We computed the combined P value, the probability of erroneous rejection of the general null hypothesis of no linkage disequilibrium, which unites certain null hypotheses for separate tests. The combined test can be constructed in two ways: one (A) using the asymptotic χ² distribution, and the other (B) using the simulated joint distribution of three separate test values.

Table 1 gives the demographic data and OA measures for the subjects. The data are presented according to 15-year age ranges. The agewise distribution of the subjects was fairly uniform. The men were larger than the women. Body mass index (BMI) increased until middle age and then either decreased (men) or remained constant (women). Both genders were nearly equally affected by OA, in a strongly age-dependent manner. As expected from many previous studies, a practically linear increase of the FS1-OA was seen in both the male and the female cohorts after the age of 30 years (Fig. 1).

Variance decomposition analysis was performed to estimate the contribution of genetic factors to the interindividual FS1-OA variation in comparison with the effect of the potential covariates (Table 2). The age effect was highly significant in both genders, and the corresponding regression parameter estimates were in good agreement with those obtained by the least mean square method as used in Fig. 1. The correlation with age was not sex-dependent and explained 74.3% of the total variation. However, sex differences were significant at the intercept of the regression equation. The body weight and height of the individual exerted negligible effect on variation in OA of the hand. Of the familial influences, putative genetic effects were statistically a most significant factor by the LRT (P < 0.001). Nearly 25% of the age-adjusted FS1-OA variation was attributable to genetic factors. Common environment shared by spouses also made a significant contribution (approximately 13%) to FS1-OA variation. Constraining this effect to zero was rejected by the LRT (P < 0.05).

We selected three STR and four SNP markers of the ENPP1 locus to obtain a fairly complete coverage of the gene region (Fig. 2). Analysis of linkage disequilibrium between each of the three STR markers on the one hand and the SNPs on the other revealed sufficient coverage of the entire ENPP1 locus to detect a functional SNP via linkage disequilibrium by the three STRs (data not shown). The markers were genotyped in all individuals of the population sample. To test whether particular alleles of any of the markers were significantly associated with age-adjusted FS1-OA, we used three different TDT-like tests and PDT (Table 3). For the three TDT-like tests, we also estimated the combined probabilities of the null hypothesis rejections, assuming either that the three tests (A) were or (B) were not independent. Using the different tests and combined analyses, we were able to demonstrate a number of significant (P < 0.05) or even highly significant (P < 0.001) associations between the rare-allele pool of M06NR1A ('allele' 4F), the C-allele of K173Q, and various haplotypes of two or three adjacent SNPs (Table 3). For the haplotypes, the combined three test P values ranged from 0.0082 to 0.000018, and from 0.020 to 0.0006 for the A and B types of computation, respectively. When the rare-allele pool of M06NR1A, alleles 5 + 6 + 10 + 11 (Fig. 3) was split into its components, allele 10 and the combination of the adjacent alleles 10 and 11 showed the strongest association signals with A-type combined P values, as low as 0.0001 and 0.000004, respectively (Table 4). Even after Bonferroni correction for the number of tested alleles per marker, all P values remained significant. PDT results were generally in agreement with the TDT results (Table 4). Even though the results of three tests for particular haplotypes were no longer significant, the others achieved statistical significance, with P values between 0.0278 and 0.0002.