William Heberden in 1806 was probably the first to suggest "some degree of heredity" for rheumatoid arthritis (RA). Subsequent twin- and family-based studies have shown that both genetic and environmental factors influence susceptibility to RA, leading to its description as a complex or multifactorial condition. There have been many attempts to elucidate both the genetic and environmental components, but the aetiology remains largely unclear. In common with many other autoimmune chronic inflammatory conditions, associations with genes of the human leucocyte antigen (HLA) complex have been described. The original observation by Stastny in 1978 [1] of an association with the HLA DRB1 locus in 80 patients has become one of the few examples of a consistently associated gene in RA. Estimates suggest that the HLA locus probably accounts for no more than a third of the total genetic component of susceptibility [2], leaving the majority still to be determined. This review considers some of the approaches currently available for the investigation of the non-HLA genetic basis of susceptibility to RA.

Complex diseases do not lend themselves to parametric linkage analysis, as this technique depends on following the inheritance of genetic markers in extended pedigrees to look for cosegregation of marker alleles in affected individuals, under a defined model of inheritance. RA clearly does not follow Mendelian inheritance patterns, and large, extended pedigrees are difficult or impossible to ascertain. Thus, until recently, the search for RA disease genes was targeted to potential candidate genes tested in disease-association studies.

Nonparametric (model-free) linkage analysis can be carried out on small, nuclear families, such as affected sibling pairs (ASPs), provided sufficient families can be collected. Technological developments in the early 1990s making possible high-throughput genotyping of informative markers (microsatellites; simple repetitive DNA sequences, highly polymorphic between individuals in terms of the number of repeats) combined with the collection of large numbers of small nuclear families led to groups in Europe [2], Japan [3], the USA [4], and the UK (Worthington J, unpublished data) to embark upon whole-genome screens in search of RA disease genes. This exciting approach of systematically scanning the genome for regions containing disease loci provides the opportunity to identify previously undescribed genes that would never be found by the candidate-gene approach.

The results published to date all represent the first stage of genome screens, in which many potential loci have been identified, but as yet, with the exception of DRB1, none has reached the level of statistical significance (P < 2.2 × 10^–5) recommended for such approaches [5]. As a result, many of these loci will turn out to be false positives, and only replication studies in independent cohorts will determine the true regions of linkage. Further, the individual studies published so far are limited by a lack of power to exclude regions or to detect loci with modest effects – the likely scenario in RA – and this may require the use of as many as 2000 ASPs [6]. All groups are collecting more families, and plans are already in place to carry out meta-analysis of the data, which may prove to be the most effective way to achieve the necessary power, if the differences between the cohorts are taken into account. We must expect, then, to wait some time before whole-genome screens in RA accurately direct us to disease loci.

On a more encouraging note, even in the initial publications, a number of promising loci have been detected in more than one study. This is perhaps particularly surprising in view of the relative lack of concordance between whole-genome screens reported for some other conditions (e.g. multiple sclerosis, schizophrenia). This lack of agreement has, unfortunately, led to some scepticism about whole-genome screens but actually does not suggest an inherent flaw in the approach: it is more likely a reflection of heterogeneity between the cohorts studied. The possibility of both phenotypic and genetic heterogeneity in complex diseases is a potential difficulty that must be taken into account when attempting to define the genetic basis of a complex disease, whether using linkage- or nonlinkage-based approaches.

Establishing evidence of linkage using ASP methods is just the first step, and one of the greatest challenges to molecular genetics is to identify a disease gene from a region of linkage. The size of an initial linkage region may be many tens of centimorgans, and recent publications have shown that chance variation in the location estimate is substantial [7,8], suggesting that it may be necessary to follow up a large area on either side of the linkage peak. Experience so far suggests that even with fine mapping using a dense marker map, it has not been possible to narrow down regions to less than 10 cM in any complex disease. As regions of this size may contain hundreds of genes, it is vital to define a smaller region by linkage-disequilibrium mapping before moving on to target genes.

The rapid pace of developments in molecular genetics and molecular medicine make it almost impossible to accurately predict more than a couple of years into the future. It seems likely that our investigation of RA genetics will continue, in the immediate future, to be based on a combination of linkage and association studies, with refinements to improve the power and sensitivity. Linkage mapping of ASP collections will probably use a higher density of markers, and information derived from other sources such as animal models may be used to target the linkage studies. The investigation of loci homologous to regions mapped in rodent models of disease has proved fruitful in a number of diseases, including arthritis [16]. With the human genome now sequenced and the mouse sequence expected within the year, the accurate targeting of homolgous regions for linkage analysis will be greatly facilitated.

The ASP collections may also become the samples of choice for association-based studies. case–control and TDT methods have routinely used sporadic cases, but in a complex disease such as RA, any study design based on sporadic cases may be selecting more for environmental than for genetic factors. Risch [17] has advocated the use of ASPs in an association study design with unrelated controls as the most powerful approach to detect disease genes. For example, using 102 ASPs (408 individuals in total) has an 80% power to detect a genetic relative risk of 2 (for the heterozygote) for a disease allele of 20% frequency with a significance level of P = 5 × 10^–8. Intuitively, taking cases with a family history should increase the chances of detecting a genetic effect. A recent publication demonstrated an association to the tumour necrosis factor receptor II in two independent data sets in which the case had a family history of disease (defined as at least one affected first-degree relative) [18]. This association was not observed in a cohort of sporadic cases.

So far, association studies in RA have concentrated on a few, well established candidate genes. In theory, it is possible to search the whole genome by association methods. The likelihood of success using this approach depends upon the number of markers typed and the extent of linkage disequilibrium that exists between markers. It is beyond the scope of this review to discuss whole-genome linkage-disequilibrium mapping and whether it will be applied to RA. More thorough discussion of the feasibility of this approach has recently been published [17,19].

With the completion of the human genome sequence, all transcribed genes should soon be identified. Additional information about tissue expression and functional domains will allow us to make much more educated decisions about which genes to target. With initiatives such as the SNP consortium releasing >300,000 SNPs into the public domain, it will soon be possible to select SNPs in candidate genes from a list of all transcribed genes in the genome. This more focused approach may lead to greater success in detecting disease genes, because testing potentially functional SNPs within genes for association decreases the dependence upon linkage disequilibrium. However, even for a single gene, there is no clear consensus about how many SNPs one might need to analyse. A recent publication examining SNPs around the APOE locus failed to find an association with the majority of the common SNPs within a 1.5-Mb region of the gene [20]. case–control studies have most commonly been used to look at single markers; analysing several markers within a gene or small region has been more problematic, because it is difficult to determine haplotypes in the absence of family information. There is now a concentrated effort to evaluate methods of haplotyping unrelated individuals, and a recent publication successfully identified the APOE locus using haplotyping methods [21]. Drysdale et al used haplotypes in the β₂-adrenergic receptor to detect association with drug responsiveness; they suggested that haplotypes were more successful in detecting associations than in analysing individual SNPs [22].

Having reached the landmark event of sequencing the human genome, perhaps we are now in a position to really begin dissecting the aetiology of RA. Ultimately, this will be achieved only by using a combination of the techniques described in this review, together with high-quality phenotypic and epidemiological data. This will also require the development of methods of analysis based on more sophisticated models of complex disease which allow for gene–gene and gene–environment interactions.