Introduction

There are basically two genetic approaches to finding disease genes that are inherited to the offspring. The first is the genome-wide search (also referred to as ‘positional cloning’ since the gene is identified using its position in the genome). The second is the ‘functional candidate gene approach’ where genes are selected for investigation based on a known function.

Linkage analysis is used to approximately point out a chromosomal region, which is more often inherited by the affected individuals within families. Association analysis on the other hand depends upon the same allelic variant being present in seemingly unrelated individuals with a common ancestor or ‘founder’. Association analysis has been a method for fine-mapping and for candidate gene studies since the resolution is much higher in comparison to linkage. The number of meiotic events available to ‘cut’ up the chromosomes during recombination are few in linkage (one for each generation and child), while association is the result of a common ancestor many generations ago, and therefore, includes many meiotic events. Unlike linkage analysis, association studies depend upon the analysis of unrelated individuals, where a common sample includes a set of cases and control individuals but the so-called ‘trio’ families (one child and two parents) can also be used. Genetic linkage and association methods are based on DNA analysis and the use of DNA polymorphisms. When it comes to genome-wide studies, microsatellites have been the markers (genetic variants) traditionally used. However, due to the advances in single-nucleotide polymorphism (SNPs) technology, SNPs are now the variants of choice. SNPs can be used for linkage studies but more importantly also for genome-wide association (GWA) studies. Recently, the knowledge about copy number variants (CNVs) has exploded. These variants have been difficult to detect but new technology makes these a target for disease association studies as well. In addition to genetic linkage and association, analysing differential expression of genes and proteins can help by identifying pathways and interesting candidate genes. There is a large amount of mRNA and protein-expression data involving CD including whole-genome mRNA expression profiling studies.1, 2 To limit the scope, we will not discuss this approach further in the present review.

Coeliac disease (CD) is a disease with complex inheritance. After the remarkable success with positional cloning of genes responsible for monogenic traits, there has been comparably few success stories with complex polygenic traits.3 The main problem with linkage and association analysis on phenotypes with complex genetic inheritance compared to monogenic inheritance is that the relationship between a variant/mutation influencing the disease and the disease phenotype is weak and consequently difficult to detect. This is mainly due to the small contribution from each locus to the overall disease susceptibility. A number of problems encountered when dealing with complex inheritance and positional cloning strategies are listed in Box 1.

Box 1 Difficulties in complex inheritance
Full size table

In light of these problems (described in Box 1), it is important to carefully think through and plan the study design before starting to collect samples on diseases with complex inheritance patterns. For example, very large study populations could be necessary. Whether the whole data set is analysed together or if a stratification approach is adopted, sample size is clearly one of the determinants of success. Thanks to the HapMap project, which was completed and published in October 2005,4 it is now possible to select a limited number of SNPs or CNVs to capture most of the variation in a given segment of the genome. Recent advances in technology have also made it possible to analyse thousands or even millions of genotypes in a few days at a cost that is reasonable. Not only thousands of samples but also thousands of different variants can be analysed.

Information gathered from each subject is of great importance; diagnosis, age of disease onset, disease symptoms, other diseases in the family, family history, ethnic background and birthplace of grandparents. All of these are examples of information, which should be carefully considered and could be very useful when conducting a genetic study. Every individual study and disease is likely to require an optimised study design that takes into account the unique characteristics of both the sample population and the phenotypes studied.5, 6, 7 In some cases, it might be useful to define the phenotype differently compared to the traditional phenotypic characteristics used when making the disease diagnosis. The phenotype–genotype correlation is, in most studies, the weakest link.6, 7

In this paper, we will review what is known so far about the genetics of CD. Several reviews have been written about CD and the human leukocyte antigen (HLA) component8, 9 as well as the epidemiology and diagnosis of CD.10, 11, 12 However, the so-called non-HLA genetic component of CD is currently under intense investigation and this will be the main focus of the present review.

Coeliac disease: the phenotype

CD is characterised by an immunological response to dietary gliadins in wheat and the corresponding prolamins in rye and barley. Individuals with CD characteristically develop a small bowel enteropathy with a villous atrophy, crypt hyperplasia, increased number of intra-epithelial lymphocytes and inflammation of the lamina propria as well as circulating disease-specific IgA-antibodies (anti-endomysium antibodies (EMA) and tissue transglutaminase 2 (TG2) antibodies).

From having been regarded as a gastrointestinal disease, CD is today rather considered a multisystemic complex inflammatory disease.13 Clinically CD is highly variable. Life-threatening states with severe diarrhoea, malnutrition, lethargy, oedema and/or anaemia are seen (even if they are rare) as well as clinically silent, asymptomatic persons without obvious symptoms at all. In dermatitis herpetiformis, a skin manifestation of gluten intolerance, severely itching blisters appear on elbows, knees and buttocks, most often in addition to symptomatic or non-symptomatic enteropathy.

Several diseases are associated with CD. Malignancy, especially in the gastrointestinal tract and particularly intestinal lymphoma, was already in the 1930s found to be more common in CD. A large Swedish registry follow-up study on 12 000 subjects verified this risk in adults.14 The standard incidence ratio (SIR) for all types of cancers was 1.3 (1.2–1.5; 95% confidence interval (CI)), for malignant lymphoma 5.9 (4.3–7.9) and for small intestinal cancer 10 (4.4–20). The risk of cancer declined with increasing length of follow-up after CD diagnosis and was not significantly increased after 10 years. Interestingly, in individuals first hospitalised with the diagnosis of CD before the age of 10 years, no significantly increased risk for malignant lymphoma was found. In contrast, individuals first hospitalised with CD older than 20 years had a considerably higher risk (SIR 7.0, 95% CI: 5.0–9.5). Further, the risk decreased over time for diagnosis of malignant lymphoma during the study period from SIR 12 (95% CI: 3.8–28) in 1970–1979 to 3.4 (95% CI: 1.9–5.7) in 1990–1995.

Osteopenia/osteoporosis is present at diagnosis of CD in all ages and in patients who fail to adhere to the gluten-free diet.15, 16 Treatment with gluten-free diet reduces or normalises the osteopenia.17 A special form of epilepsy with occipital calcifications is also linked to CD.18

Association between CD and autoimmune disorders has drawn special attention. Type 1 diabetes mellitus (T1D), thyroid diseases, Sjögren's syndrome and other autoimmune diseases are overrepresented in CD and CD is overrepresented when patients with these disorders are serologically screened for CD.19, 20, 21, 22, 23, 24 An Italian multicentre study demonstrated a strong correlation between increasing age at CD diagnosis and occurrence of other autoimmune disease.25 Failure to adhere to the gluten-free diet increases the risk for teenagers of having circulating organ-specific autoantibodies and autoimmune disease26 and first-degree relatives of CD patients have an increased prevalence of autoimmune diseases as well as silent CD.27 These findings point to the possibility of common genetic pathways in CD and autoimmune diseases, and raise the question if the upregulated immunological activity in untreated CD may contribute to the development of autoimmune disease in genetically susceptible individuals.

A higher prevalence of CD is found also among girls with Turner's syndrome28 and children with Down's syndrome.29

Epidemiology, population genetics

Within a few years in the early 1980s, a three to four times increase of childhood CD incidence was described in Sweden, but not in the neighbouring countries.30, 31, 32, 33 After 1996, a sudden decrease was noticed which was followed by a slow increase and a shift to higher ages and more silent phenotypes.33 The changes were parallel to changes in the infant-feeding pattern. Studies showed a higher gluten consumption in Swedish infants compared to countries with a lower incidence and to earlier Swedish studies.34 Also, a higher frequency of terminated breast feeding before the gluten introduction in children with diagnosed CD versus controls was found.35 The question whether these environmental factors are affecting the phenotype (ie, the intensity of the symptoms), and thus, the chance of being diagnosed rather than triggering the onset of the disease has been under debate.36

For genetic issues it is often of interest to compare epidemiological data between different regions, countries and populations. Different Caucasian populations show a prevalence of around 1% with surprisingly low variations when statistical margins of error are taken into consideration.37, 38 A female predominance of around 2:1 is generally found.39 Family studies have shown a sibling relative risk,40, 41 of approximately 10 (λs≈10) using a population prevalence of 1% and a sibling prevalence of 10%.42 In dizygotic twins, the sibling prevalence has been reported to be around 20%.43

Few epidemiological studies have been done in non-Caucasian populations. One exception is a study of Saharawi refugees, who show a surprisingly high prevalence of 5.6%, the highest population prevalence of CD ever found.44 Another study comparing populations from European and Asian origin demonstrated that the Punjabis had a four times higher incidence than Europeans.45

The positional cloning approach in CD

The first evidence of linkage in CD: chromosome 6 and HLA (‘CELIAC 1’)

The most common candidate gene region in inflammatory diseases is the major histocompatibility (MHC) complex on chromosome 6, which encodes for the human leukocyte antigens (HLA). The linkage and the association signals to HLA are typically strong in autoimmune and inflammatory diseases. Linkage and association to HLA were detected several decades ago in many diseases such as type 1 diabetes, rheumatic diseases and CD.46, 47, 48, 49, 50 However, finding the functional genes in HLA influencing these diseases has not been an easy task, in part due to the very strong linkage disequilibrium (LD) in the HLA region. Strong LD can complicate the fine mapping of disease associated genes, but in spite of this and uniquely for CD, the primary disease locus for CD in HLA has been identified.51 The HLA-DQ2 molecule, encoded by specific alleles of HLA-DQA1 and HLA-DQB1, is one of the strongest HLA associations among the HLA-associated diseases. In most populations, over 90% of CD patients carry the DQ2 heterodimer encoded by alleles DQA1*05 and DQB1*02.8 The final genetic evidence that HLA-DQA1 and -DQB1 encoded the functional variant causing CD was the discovery that the DQ2-risk molecule can be encoded from alleles DQA1*05 and DQB1*02 in cis- or in trans-51(Figure 1).

Figure 1
figure 1

The HLA-DQ2 αβ heterodimer encoded in cis and in trans.8

Full size image

Nearly all of the CD patients negative for DQ2 carry the DQ8 haplotype containing the risk allelic combination of DQA1*0301-DQB1*0302. A very small proportion of patients are both DQ2 and DQ8 negative, and they have been found to carry one of the two specific alleles coding for DQ2.52, 53 That the risks associated with the different HLA-susceptibility alleles differ between populations have been shown54 as well as a dose dependence of the DQB1*02 allele.55, 56, 57

Non-HLA-linked genes in CD

The alleles HLA-DQ2 or -DQ8 are necessary for the onset of CD but they cannot explain the whole genetic susceptibility to the disease. These alleles are also common in healthy individuals and together the population frequency is about 20–30%.51

The existence of additional susceptibility genes located outside of the HLA region is supported by the difference in concordance between monozygotic twins and dizygotic twins, who are identical in the HLA region and both carrying the susceptibility alleles. The pair wise concordance among monozygotic twins is over 70%, which is the highest reported in a disease with complex inheritance.43 In siblings sharing identical HLA haplotypes, the CD pair-wise concordance is around 20-30%, suggesting importance of genes located outside HLA in disease susceptibility.43, 58, 59 With current estimates of population prevalence (0.5–1%) the relative risk has decreased, and λs is approximately 10–20, λs (HLA) is around 5 and thus with a multiplicative model,60 the non-HLA component λs (non-HLA) is 2–4 and would confer a minor part of the total genetic component. However, these measurements should be taken with caution also considering that HLA is a necessary factor. The estimate λs is a measure of familiarity (a combination of the genetics and other factors which are shared among family members) and there could be an unknown environmental component diluting the effect of the non-HLA genes, if this component also tends to cluster within families.

Searching the entire genome: linkage studies and association studies

Several genome scans in CD have been performed in search for susceptibility genes outside the HLA region. These studies differ in respect to number of independent families and what types of families that are analysed. Some studies have only a few (multi-affected pedigrees), or even one single large pedigree, while others have adopted an affected sib-pair approach. The differences in outcome can partly be explained by these differences in family structure and partly because sample size is simply not large enough to be able to detect a gene with relatively small effect. Also, if related sib-pairs are analysed and not only independent ones, the linkage statistic can be inflated if it is not corrected for. Another reason for discrepancies is the use of different statistical methods. Some genome-wide scans have used non-parametric linkage (NPL), others have analysed the data using several different parameters and LOD scores or heterogeneity LOD scores (HLOD). Testing of different models and using different types of families (multi-affected pedigrees versus sib-pairs), makes the comparison between studies difficult. The genome-wide linkage scans are summarised in Table 1.

Table 1 Genome-wide searches and their follow-up linkage studies
Full size table

In a European collaboration, raw data from four genome scans including follow-up data were pooled and reanalysed.64 These were the second, third, fourth and fifth studies published.62, 66, 68, 69 The raw data consisted of genotypes from 442 families; 2025 individuals of whom 1056 are affected. The results pointed to chromosome 5q31–33 as being the only significant locus in these families apart from HLA (Figure 2).

Figure 2
figure 2

The results from a meta-analysis of European CD families.64 Non-parametric linkage (NPL) results displayed as Zlr130 on the y axis. The x axis indicates chromosome distance in cM.

Full size image

Interestingly, the families from the United Kingdom showed negative linkage scores in this region. This could be the result of random variation due to a small number of families and locus heterogeneity. These families also have a somewhat different makeup, being large multigeneration pedigrees with many affected individuals. The three remaining data sets consisted of small nuclear families with typically 2–3 affected individuals. It is possible that many unrelated nuclear families favour the detection of a more common susceptibility gene shared by a large portion of CD patients, while large pedigrees with many affected individuals could be the result of a gene variant, which is relatively rare but with rather strong penetrance. This would illustrate how study design could influence the outcome of the results.

The first GWA analysis in CD has been reported and the results show convincing evidence that the IL2–IL21 region could be involved in CD.78 Several variants were highly significant when analysed in three populations; UK, Dutch and Irish case–control samples (shown in bold in Table 2). Strong LD makes it difficult to isolate one variant as disease causing and both IL2 and IL21 are expressed by T cells and have been implicated in other autoimmune conditions.79, 80 These genes are definitely interesting to follow up on in additional populations. Other SNPs from this GWA analysis, which need to be confirmed, are shown in Table 2. It is noteworthy that one of these variants is located in the 13th exon of LCT, the enzyme necessary for digesting lactose. Could there perhaps be a connection where malfunctioning lactose digestion during the first years of life, increases the risk for CD?

Table 2 Genome-wide association results from van Heel et al. 200778
Full size table

Fine-mapping and positional candidates on chromosomes 5 and 19

After linkage has been established, the next step is to fine map the linked region using association analysis. A simplifying condition for finding a disease gene by association is that most of the individuals with a certain disease are descendants of the same ancestor, therefore carrying a common haplotype with a so-called founder mutation. Association or LD between two alleles is the complex result of several factors, including recombination rate, allele frequencies and mutation rate. Reviews on the subject are written by Lander and Schork81 and Ardlie et al.82 If a disease mutation occurs next to a common allele, a large proportion of the normal population as well as affected individuals will carry this allele in coming generations. A marker that lies a bit further away could by chance have a rare allele on the founder chromosome and the correlation between that marker and the disease variant will be stronger. This illustrates the strong impact allele frequencies have on association.

When it comes to chromosome 5, a few published studies have focused on specific candidate genes in this region but not been able to find association.83, 84, 85, 86, 87 In addition, a large fine-mapping study failed to single out a strong susceptibility candidate.88 There were several genes showing nominal association in this Norwegian/Swedish family sample but no single association could explain the large linkage peak found in the same sample.88 Unless, there are regions inadequately covered due to, for example, copy number variation, these finding suggests that there is too much allelic heterogeneity or too many founder haplotypes to pin a gene down. Fine-mapping of the chromosome 19 region has been more successful than the one of the 5q region. The MYO9B gene was shown to associate in two different CD cohorts.73 It is a gene likely to be involved in Rho-dependent signalling pathways and remodelling of the cytoskeleton and tight junction assembly. This seems to fit very well with the possibility of CD being caused by a more permeable intestinal barrier. Attempts have been made but so far only one group has been able to replicate this finding89 while others have failed.90, 91, 92, 93, 94 Replications of genes with a small effect, or possibly, a population-specific effect can be difficult. However, control allele frequencies, calculated from untransmitted alleles in an Italian population, were significantly different from those of the Dutch control population; while patient allele frequencies were the same in the different populations.93 This questions the Dutch control population and could indicate a false positive finding of the MYO9B gene in the first place.

Other potential candidate genes on chromosome 19 include the intercellular adhesion molecule-1 precursor (ICAM-1) and CD209 both located in chromosomal band 19p13.2. A study of 489 CD patients and 257 families recently suggested that a functional variant of the CD209 promoter is associated with DQ2-negative CD patients.95 The ICAM-1 gene was found to associate to CD when analysing 180 unrelated French CD cases and 212 controls.96 When the authors stratified for age of disease onset, the same variant conferred even stronger predisposition to the adult-onset disease.96 Also from 19p13, cytochrome P450 F3 and F2 genes (CYP4F3 and CYP4F2) showed an effect on familial clustering in a Dutch CD cohort. CYP4F3 and CYP4F2 catalyse the inactivation of leukotriene B4 (LTB4), a potent mediator of inflammation responsible for recruitment and activation of neutrophils.97 Additionally on chromosome 19, a certain combination of expressed killer cell immunoglobulin-like receptor two (KIR2DL5) genotypes has shown association in a Spanish population of 413 cases and 231 controls.98

The candidate gene approach searching for non-HLA genes in CD

Cytotoxic T-lymphocyte-associated protein: CTLA4 (CELIAC 3)

CTLA4 plays an important role in maintaining tolerance to self-antigens, both as a negative regulator of T-cell proliferation through the co-stimulatory signal and as an inducer of clonal anergy. This gene has been under investigation in a number of autoimmune conditions. It is therefore an interesting candidate gene for CD. So far no evidence has come forward showing the involvement of a certain functional genetic variant of this gene in CD. However, there are many reports indicating the involvement of CTLA4 in CD.

The relative risk of CTLA4 in diabetes is estimated to be as low as 1.299 and if the same is true for CD, almost all studies in CD have too little power to be able to detect association. Still, some linkage to the region has been reported in a few studies68, 100, 74 and most published studies in CD do show signs of association as well.100, 74, 101, 102, 103, 104, 105, 106, 107, 108, 109, 110, 111 However, they are not completely congruent. Different studies show association to different haplotypes107, 112 and some studies fail to show association.113, 114, 115 Other studies with negative results could remain unpublished. Nevertheless, a negative result is not surprising if the relative risk is low and study samples are small. Different haplotypes showing association could be a result of allelic heterogeneity between and within populations but also of that CTLA4 can have different impacts on disease sub-phenotypes such as disease onset and severity. If one study includes many silent cases and another includes classical cases, this could be a reason for the different associations shown. Genotypes at HLA have also been suggested to influence the outcome of susceptibility to alleles in CTLA4.68, 110

Other non-HLA candidate genes potentially involved in CD

We searched PubMed (www.pubmed.gov) for ‘genetic AND (celiac OR coeliac) AND (associated OR association)’ and the database returned 405 hits. The genes presented in Table 3 are implicated from candidate gene studies, where a reported significance was below a P-value of 0.05. Furthermore, only those studies based on more than 100 unrelated CD patients are included below. This is indeed a very small sample size in a complex disease like CD, where odds ratios in the order of two or even less is expected. Some studies include many more individuals and thus these results could be more reliable. After excluding the HLA, chromosomes 5, 19 and CTLA4 regions, only seven studies in total fulfilled these criteria. In some cases, there have already been attempts made to replicate these findings. However, in most cases no convincing evidence can be presented (Table 3).

Table 3 Candidate gene studies showing nominal association in CD
Full size table

It is important that other research groups try and replicate suggestive findings since it is difficult to interpret the significance of genetic associations from a single report. Without a doubt, many of these studies will prove to be false-positive findings.

Evolutionary perspective

When searching for genes causing a complex disorder, it is interesting to try and understand the evolutionary and historical implications and how the disease is moulded by time.

One would think that individuals with variants conferring risk to a disease would be reduced in the population over time. It could be argued that if intolerance to gluten was an original state for human beings, we would never have adapted to this food source. However, in a disease like CD, it is possible that the disease state is the wild type or ‘natural state’ much like lactose intolerance is considered the wild type for humans. There are several factors that can be used to argue for how ‘intolerance’ variants could have been widespread, which in turn, would explain how humans adapted to the new food source all the same.

When looking at the HLA molecule known to cause CD – DQ2 – this molecule is frequent in the population and is considered ‘normal’ (ie, no damaging mutation hindering a protein to function properly). How is this possible, from an evolutionary perspective?

First, the complexity of the disorder makes individuals respond very different. Many affected individuals live to a high age and reproduce in spite of the disease.

Second, variants conferring susceptibility to disease could give an advantage in other situations or genetic backgrounds. HLA genes are examples of such genes; many variants of HLA are beneficial at the same time as they confer susceptibility to disease. If an individual is lacking the non-HLA-susceptibility genes it could be an advantage to have the DQ2 molecule conferring susceptibility to CD and vice versa.

Third, when it comes to CD, the percentage of gluten in today’s crops is increased by human intervention. When the first humans started to eat diploid wild-wheat the amount of gluten ingested was much lower than in today’s tetraploid breed of wheat. Although very small amounts of gluten induces relapse of the flat mucosa in CD patients, we do not know so much about the amount of gluten needed to trigger the disease. Maybe, the environmental trigger (gluten) was not strong enough to cause the disease in many individuals even if they had the genetic susceptibility variants.

Factors that contribute to keeping the CD-susceptibility genes in HLA fairly common in the population can equally keep non-HLA genes common.

Taking these factors in consideration it is possible that non-identified CD-susceptibility alleles might have a similar allele frequency as the HLA-susceptibility alleles, and could then also be considered the wild-type allele. If a CD-susceptibility locus was polymorphic 10 000 years ago, then this can mean that the genetic variant influencing CD could be common and ancient. For the search of this kind of variation, the haplotype-mapping approach and GWA analysis are very suitable.127, 128, 129

On the other hand, before humans started to eat wheat, CD-susceptibility variants could have had any frequency. If the allele influencing CD was the wild-type and non-polymorphic at the time when gluten was introduced in the population, it is possible that protective and relatively rare variants have developed over the years. These variants would have escalated in the population because of an increased selective pressure in a changing environment/gluten consumption. More recent mutations would suggest increased allelic heterogeneity. From this follows that analysing patients jointly from different parts of the world or even different parts of the same country, should be done with caution. However, if this would be the case, then we would expect a high variability of disease prevalence in different populations and the fact that CD seems equally prevalent in many populations speaks against this.

Future challenges

Current technologies have made it possible to also use association analysis for whole-genome scans. Although it is under debate if this is a good strategy to use, several such studies have already reported disease associations and more will follow. The IL2–IL21 region in CD is one of these findings and even if this is not the only non-HLA region influencing disease, there is evidence that one of these genes plays a part in CD susceptibility.

Two linked regions have reached a level of significant genome-wide linkage according to Lander and Kruglyak. These regions are 5q31—33, where linkage has been verified in several different independent populations64, 65 and 19p13.1 which show linkage in the Dutch population.73 One would think that these regions should be the first place to look for susceptibility genes. However, so far no convincing evidence has pinpointed a certain gene in any of these regions. New strategies using innovative phenotyping and using gene–gene interaction models might be necessary to locate disease genes. It is very possible that a combination of several different genes in the linked regions alone may be influencing the outcome of the disease and a strategy that takes this and other gene–gene interactions into account might be the way to move forward.

The puzzle is slowly being pieced together. Undoubtedly, several genes and genetic variants outside of the HLA complex add to the disease risk. Whether this risk increase is mainly due to rare variants, common ancestral variants or both remain to be seen.