The New Genetics and Chronic Obstructive Pulmonary Disease

Abstract

Epidemiological and family studies provide evidence for genetic factors contributing to chronic obstructive airways disease (COPD) susceptibility. Studies to date have focused on candidate genes implicated in the pathogenesis of the disease. In general, many of these studies have been underpowered or have not been extensive enough in investigating the full extent of genetic variation in these genes. This has resulted in conflicting data with potential false positives or findings that have not been replicated. More recently, larger studies and extensive coverage of candidate genes have implicated genetic variants that may contribute to the disease. The use of unbiased genome-wide association studies offer the prospect of identifying new genes involved in COPD susceptibility and genetic modifiers of disease phenotypes. There is cause for optimism as a number of major complex diseases have been successfully tackled in this way. The review will highlight what has been achieved by genetic studies to date, some of the related problems and the future impact of high throughput technologies such as genome-wide association studies on our understanding of the genetic basis of COPD.

Key words: :

• Chronic Obstructive Pulmonary Disease
• Candidate genes
• Genome-wide association

The abbreviations used are
AATD	=	alpha1-antitrypsin deficiency
COPD	=	chronic obstructive pulmonary disease
GWAS	=	genome-wide association studies
LD	=	linkage disequilibrium
MAF	=	minor allele frequency
SERPIN	=	serine proteinase inhibitor

INTRODUCTION

Chronic obstructive pulmonary disease (COPD) is characterised by progressive largely irreversible airflow obstruction accompanied by chronic inflammation, resulting in fibrosis and narrowing of the small airways and emphysema arising from loss of alveolar architecture. COPD represents a substantial economic and social burden worldwide and the prevalence of the disease will increase with an aging population ([1]). It is frequently under-diagnosed and under-treated, affecting over 150 million people worldwide with nearly 3 million deaths per annum. Once lung function (FEV₁) declines to below 50% predicted, there are generally a limited number of strategies to influence the outcomes. These patients often develop acute exacerbations, which have a significant effect on the quality of life for many months ([2]).

Phenotypically, COPD is a heterogeneous disease. The term includes patients with chronic obstructive bronchiolitis and small airways obstruction and emphysema with destruction of lung parenchyma and enlargement of air spaces. A number of criteria are used in the diagnosis of COPD including clinical presentation, decline in lung function, imaging techniques including CT scans, diffusing capacity for carbon monoxide and biomarkers such as exhaled nitric oxide. Factors that influence increased risk of death include body mass index, degree of airflow obstruction as measured by FEV₁, dyspnea and exercise tolerance, referred to as the BODE index with a higher BODE score reflecting a higher risk of death.

Pathogenesis and genetics

There are many aspects of COPD which are poorly understood, though it is recognised that some key pathways are involved in its pathophysiology. Chronic inflammation is a major feature of the disease. This primarily affects the small airways and lung parenchyma with the process of tissue destruction and remodelling leading to fibrosis and narrowing of the small airways and loss of lung parenchyma. There is a direct correlation between disease severity and the degree of inflammation. Some of the key inflammatory cells involved include neutrophils, macrophages and CD8+ cytotoxic T lymphocytes. Other cells, including epithelial cells and fibroblasts are also thought to contribute to the disease ([3]). The net effect of the interactions of these various cell types is to increase the number of inflammatory mediators. These include low molecular weight lipid mediators, such as leukotriene B4, pro-inflammatory cytokines and chemokines such as IL-6 and IL-8 and proteases involved in tissue remodelling (Figure 1). A consequence of inflammation is the release of reactive oxygen species from activated inflammatory cells thus promoting tissue damage ([3]).

Figure 1 A simplified diagram showing key inflammatory pathways and cells implicated in the pathogenesis of COPD (reproduced with permission from Resp. Research – Wood et al., 2006).

Known genetic and environmental factors in COPD

Cigarette smoking is by far the most important environmental factor in the aetiology of the disease, but family-based studies suggest that there are also likely to be important genetic factors. There are several lines of evidence to support this. It has been suggested that only about 10–20% of chronic heavy cigarette smokers develop symptomatic COPD, though this figure may be an underestimate. There is a spectrum of COPD that may include many cigarette smokers who develop mild forms of the disease ([4], [5]). The variability of expression of the disease for similar exposure to cigarette smoking also implicates genetic factors that either contribute to susceptibility, or protection to explain the variability of onset of disease.

Furthermore, siblings of patients with severe COPD have a significant risk of airflow obstruction with a mean odds ratio of 4.7 when compared with the prevalence of the disease in the general population matched for age, sex and smoking history ([6]) consistent with previous observations of a relative risk of 3 in first degree relatives ([7]). The data for pulmonary function in families also provide good evidence for a genetic contribution to the disease process. There is also a closer correlation of spirometry measurements between monozygotic twins compared to dizygotic twins. Twin studies have also shown that there is a high risk in monozygotic twins that both will develop airflow obstruction ([8], [9]).

The genetics of COPD is likely to be complex, involving a number of genes and environmental factors and interactions between genes and environment and considerable challenges lie ahead. To date, studies have focused mainly on candidate genes and genome-wide linkage studies to initially identify regions in the genome that confer susceptibility. The best recognised genetic contributory factor to COPD in smokers is alpha₁-antitrypsin deficiency (AATD), which accounts for about 2% of all patients with COPD in European populations though this figure may be an overestimate due to ascertainment bias ([10], [11]). This association has resulted in a major hypothesis for COPD, namely the protease–antiprotease imbalance. This is further supported by mouse models of the disease in which loss of protease activity makes mice less susceptible to disease and there is a mechanistic role for activation of matrix metalloproteinase by transforming growth factor–β in the disease process ([12], [13]). While the protease-antiprotease imbalance and oxidative stress hypotheses are the most developed, other genetic factors are likely to be important.

Alpha1-antitrypsin or SERPINA1 deficiency (AATD)

AATD, arising from homozygosity of the Protease inhibitor (Pi) Z variant, is the best known and most studied genetic risk factor for COPD in smokers. It has been shown, from family studies of patients with severe AATD that for similar levels of cigarette smoke exposure, there are variable outcomes implicating other modifiers of disease ([14]). Thus, even for what is apparently a single gene disorder, there may be other factors yet to be identified that influence susceptibility. Also the risk associated with partial deficiency only seems to hold for MZ COPD cases compared with controls rather than on a general population basis. Meta-analysis suggests an increased risk for Pi MZ in COPD cases compared to control subjects with an odds ratio of about 2 ([15]).

Large population-based studies of pulmonary function have often reported similar FEV₁ values in Pi MZ carriers and controls. There may be a subset of Pi MZ subjects who are at increased risk of COPD and this may be influenced by other genetic modifiers. For most complex diseases to date, the odds ratios have been less than 2 in replicated studies, though it is being increasingly recognised that within heterogeneous disease groups there may be subgroups with particular phenotypes where the risk of developing disease can be substantially increased on particular genetic backgrounds as has been shown for thyroid associated auto-immune disease, for example ([16]).

Genome-wide linkage studies in COPD

Genome-wide linkage analysis in the Boston early- onset COPD study in which 585 members of 72 extended pedigrees were studied used microsatellite markers, highly variable regions of the genome that can be used as tags, to identify regions of the genome associated with COPD. In this study 377 markers were used and by recent criteria would represent relatively low coverage of the genome. However, correlations were observed with post -bronchodilator FEV₁ responses in multiple regions, most strongly to chromosome 8p and 1p and further correlations were found with markers on chromosome 2q ([18], [19]). These studies need to be replicated but also highlight the potentially more sensitive methods of analysing quantitative traits in disease.

Genetic variation and applications in genetic studies

Principles

There is considerable variation at the single nucleotide level in the human genome. These are referred to as single nucleotide polymorphisms (SNPs). SNPs occur with a predicted frequency in different populations. Furthermore, combinations of SNPs that define a chromosomal background are referred to as haplotypes and they provide bar-codes for genetic variation in individuals. In reality, for each gene, relatively few haplotypes occur in the population circumventing the need to analyse all SNP combinations. The reason for few haplotypes occurring in any population is due, in part, to linkage disequilibrium (LD); two SNPs in LD are likely to be inherited together and LD reduces the number of SNPs that need to be screened.

So, if two SNPs are in complete LD, genotyping one SNP allows the other SNP to be accurately predicted. An example of how LD can be exploited is shown (Figure 2). Therefore, an individual gene may have say about 30 common SNPs, yet five or so haplotypes made up of four or five SNPs may be representative of most of the population being studied, thus making it unnecessary to analyse all of the SNPs in the gene in the population. These SNP-tags consequently provide barcodes for looking at genetic factors in disease in case-control studies.

Figure 2 LD and Tag SNPS. In (a), all 10 SNPs are in complete LD; genotyping one SNP will, therefore, provide information about all 10. In (b), the unshaded SNPs are in complete LD with each other as are the shaded SNPs. In this case, genotyping one shaded and one unshaded will provide information about all 10 sites.

A significant proportion of the human genome exhibits high LD (87% in Caucasians) with SNP markers falling into regions called haplotype blocks with little recombination occurring within the blocks. These blocks are flanked by recombination hotspots where LD breaks down. The average haplotype block is 16.3 kb (< 1–173 kb) in Caucasians. Typically, there are several tens of SNPs per block but because of LD, the average number of haplotype tagging SNPs per block is 4.7 in Caucasians. Nigerians, on the other hand exhibit lower LD and consequently a larger number of haplotype tagged SNPs (an average of 5.6), need to be typed per block. The haplotype blocks have been successfully exploited in genetic studies as it reduces the number of SNPs required for study.

High-throughput genetic screening and information about SNPs spread throughout the genome are available through the International HapMap Project (http://www.hapmap.org) and have made it feasible to undertake SNP and haplotype mapping of many candidate genes in large population based case-control studies ([12]). The density of markers currently available ensures that there is, on average, at least one common SNP every 1–2 kilobases (kb). The definition of a common SNP is given as a minor allele frequency (MAF) of > 5%. This equates to roughly 5 million SNPs in the database and provides over 80% coverage of the genome for genetic association studies. It is also possible to obtain an even higher resolution of SNPs, occurring at a MAF of 1–5% by DNA sequencing, thus providing even better coverage of the human genome though the value of this needs to be considered in the context of the sample size.

Study design and genetic approaches to identify genetic factors

Family-based and case-control studies

A number of approaches have been used in an attempt to identify the genetic factors that contribute to diseases such as COPD. They include family based studies such as affected sib-pair studies where affected sibs would more likely share susceptibility alleles compared with non-affected sibs and non-family based population case-control association studies. Due to the difficulties in collecting family data in a chronic disease of relatively old age, most studies of COPD have used a case-control approach. If families are available, this minimises the likelihood of population stratification which can arise from genetic admixture and give rise to spurious false positive results. However, results from recent studies of European populations at least, stratification has probably been overestimated ([20]). Sample size is an important consideration as most studies to date have been underpowered to detect genetic associations. This also applies to affected sib-pair design.

Population based case-controls are commonly used in genetic association studies to identify genetic factors that contribute to complex diseases due to the relative ease of sample collection. The success of this approach has been limited to date. This is partly due to the small effects of individual genes and the inherent problem of relatively small sample sizes that have been used in most studies. Other important factors include a robust definition of the phenotype.

Many genetic studies of COPD have often only involved looking at single or a few SNPs in studies that have been underpowered ([8], [9]). Replication is essential for establishing the authenticity of association studies, yet there are inconsistencies in deciding what findings are deserving of replication, what constitutes an adequate study and the confidence one attaches to a negative report. It is accepted that a single study is not generally conclusive and standardisation of both phenotypic criteria and the use of identical SNPs across studies are important components to replication.

The general approach, previously, has been to compare the distribution of genetic variants in candidate genes, identified through pathophysiological studies, in patients with COPD and controls. In these studies, matching for age, sex and smoking history should be considered in the design as these are potentially important contributors to the disease. Although several studies have reported associations, only some of these are reported here as they have been performed independently on more than one occasion. Even then, there are differences in the studies. Associations have been described for genes coding for microsomal epoxide hydrolase, surfactant protein B, vitamin D, glutathione S-transferase and SERPIN E2: these associations have been replicated in some studies but not others (Table 1) ([10], [11]). COPD is the end-result of a variety of pathological processes and the severity of the disease may vary and these are additional factors potentially giving conflicting results.

Consequently, replication findings have been an issue and both false positive and false negative findings are possible. Definitive studies are now feasible by judicious use of high density single nucleotide polymorphisms (SNPs) in large samples or doing genome -wide scans (see below). It is very important to ensure that phenotypes are rigorously defined and there is consistency between studies. Many of these issues are likely to be resolved by undertaking very large studies i.e., several thousand subjects as part of international collaborations. The availability of a database with haplotype variation (combination of SNPs) in human populations has proved invaluable in studies of complex disorders. Recent studies have shown for the first time significant associations of haplotypes in the alpha1-antitrypsin gene with disease ([21]) and new associations with SERPINE2 which have been replicated in some studies but not others ([22], [23], [24]).

Genome-wide association studies (GWAS)

Recent success in identifying genetic risk factors by using unbiased genome-wide association studies (GWAS) for seven common diseases demonstrates the power of identifying novel factors in disease in an unbiased way ([20]). These approaches are described in further detail in two excellent reviews which outline the general principles and power of such studies and the limitations of traditional genome-wide linkage studies ([25], [26]). For GWAS it is generally recommended that 2000 cases should be used but this figure will vary, dependent on the strength of association ([20]). GWAS provide a tour de force in unravelling the genetics of complex disease yet this is only the first step in identifying the genes involved in a particular disease without necessarily understanding the biology of the gene in the disease process. Given that cigarette smoking is a key environmental trigger for the disease, understanding the factors that contribute to cigarette smoking behaviour is a focus of research.

Candidate genes versus GWAS

The candidate gene approach with high density SNPs is supported by recent success in identifying genetic modifiers for cystic fibrosis ([27]), a candidate gene in the susceptibility of lupus erythematosus ([28]) and a new candidate gene for Type 1 diabetes ([29]). For the latter a total of 6500 SNPs were screened, the findings were validated in a case-control collection and replicated in an independent family collection. Using a combination of linkage analysis and subsequent high density SNP screening a striking observation has been observed in Type 2 diabetes where common variants in the transcription factor 7-like gene confer susceptibility to Type 2 diabetes. The mechanism is thought to be through impaired insulin secretion ([30], [31]), where carriers of two risk alleles confers an odds ratio for disease of 2.41.

There is probably still merit in considering high density SNP maps of candidate genes involved in biological pathways that have been implicated in disease pathogenesis as targets for studies, particularly where this has been informed by microarray-based expression profiling, by model systems or by previous linkage studies that have identified regions of the genome likely to be involved. This approach is often referred to as the genocentric approach. In this context it may also be important to look for epistatic or gene-gene and gene-environmental interactions in closely related genes. Based on modelling and yeast studies, epistatic interactions are thought to contribute significantly to disease susceptibility in that individual mutations may not confer a phenotype but combinations of two hits may result in an extreme phenotype ([32]).

The recent recognition that about 12% of the human genome demonstrates copy number variants (CNVs) is another factor that potentially needs to be considered ([33]). Fortunately, SNPs that are in linkage disequilibrium with CNVs are available and could be used as a surrogate for CNVs and are more easily detected. High density chips consisting of at least 300,000 SNPs, including an additional panel of 70,000 SNPs to cover CNVs are available that will simultaneously look at both individual SNP variants and CNVs. GWAS are more likely to be used in future to identify genetic susceptibility factors in complex disease, though this approach is still expensive and requires large numbers of patients and controls as highlighted here.

Statistical approaches

Multiple testing in case-control association studies is an issue when considering many genes as this can give rise to spurious false positives due to the large numbers of tests that are carried out. A Bonferroni correction is often used to correct for multiple testing treating each test as an independent event but this is too strict as in genetic studies it does not take into account LD and is therefore too rigorous. Newer methods using a Bayesian approach combined with hierarchical analysis and computer simulations provide a better strategy for analysing multiple SNPs.

Theoretical and practical considerations

The availability of the Hap Map resource, high density SNP arrays and large numbers of patient for studies have greatly facilitated unbiased GWAS. The availability of high density arrays has also made it feasible to consider high density candidate gene screening by custom designed arrays and fewer samples can be used in these studies to study genes that have been implicated in the disease process. Consideration of sample size and power are well described in the literature ([25], [26], [34], [36]).

The theoretical possibility that gene-gene and gene-environmental interactions contribute to complex genetic disorders has been widely recognised, but the statistical penalties incurred by the required multiple comparisons have been regarded as prohibitive. Marchini has argued that significant increases in genetic risk due to locus interaction are detectable even with a conservative correction for multiple testing ([34]). As an example, power calculations are presented for a two-stage sequential design. Half the sample is genotyped in stage 1, the other half in stage 2 (with one-quarter of the most strongly associated markers detected in the first stage). Data from both stages are pooled for analysis; this approach is more powerful than replication-based analysis ([34]).

The recent experience of GWAS of major diseases suggest that there is sufficient power to detect common alleles with a minor allele frequency of 5% or greater. These studies do not have the power to detect rare alleles that may be important in subsets of patients or particular population groups. For a sample size of 1,000, consisting of 500 cases and 500 controls for example, the power of the study and how this varies depending on allele frequencies and genotype risk ratios are shown (Table 2). The data for this was generated using CaTS-Power calculator ([34]) with the following assumptions: multiplicative genetic model, 3% prevalence of moderate to severe disease, complete linkage disequilibrium between disease and marker alleles, alpha 0.001.

The overall prevalence of COPD in western populations has been estimated at about 10% with a range of 4 to 15%, dependent on smoking history, so that the above power calculations are conservative. A summary of efficiency and power is provided in the literature ([36]). The results show that common disease alleles (common disease/common variant model) associated with genotype relative risks (GRR) > 1.5 can be readily detected; rarer disease alleles (q = 0.1) associated with greater susceptibility (GRR > 1.75) can also be detected. The variation of power with the dominant and recessive models with allele frequencies are highlighted graphically (Figure 3).

Figure 3 The effects of sample size on power as a function of allele frequency in a recessive and dominant model of inheritance. The bold line shows that for a dominant model, a sample size of 2,000 has 90% power to detect an association for a minor allele frequency of 5%, but only about 40% power in a recessive model with a minor allele frequency of 15% (power figures obtained using CaTS software and graph drawn from data.

The use of new validated algorithms such as PROC CASE CONTROL in SAS Genetics and HAPLOSTATS provide powerful tools for data analysis. Further developments are likely to provide even better methods for data analysis. Bespoke software packages are now available for genetic analyses including commercial packages such as Disease Miner from deCode Genetics and publicly available software such as PLINK and tools available from HapMap.

COPD remains a challenge as the phenotype is often heterogeneous. The use of large sample sizes should enable dissection of some of the associated phenotypic traits such as lung function. The emergence of some common pathways in a variety of diseases studied using GWAS is encouraging in the context of COPD as adequately powered studies should provide significant new insights into the disease.

Future prospects

Genotyping errors can result in spurious associations and this may reflect poor data quality and errors in the algorithms used for assigning genotypes. Algorithms such as CHIAMO, have been used to look at cluster plots to validate genotype calling ([20]). This algorithm uses a scatter plot to look at both the genotyping and the genotype calling for consistency. By the application of thresholds in the algorithms, the accuracy of genotyping can be improved significantly thus minimising the report of false positive results. The use of computer algorithms for imputation of missing genotypes significantly enhances the use of genotypes in genetic studies.

Other newer methods, such as algorithm information content (AIC), have been way to minimise the likelihood of false positives ([37], [38]). This method involves the computation of specific functions under different algorithms (one of the simplest being the conventional correlation function) and the subsequent determination of key signatures related to the non-randomness (or the information algorithmic content) of the dataset. This AIC method is more general than conventional principal component analysis and is adaptable through the chosen algorithm to incorporate specific biological (or otherwise) knowledge of the dataset. The method of AIC is particularly suited to searches for correlations between any number of SNPs and the disease status of an individual – i.e. epistatic interaction though it has not been used in case control genetic studies. The use of computer simulations in implementing the AIC method also provides a convenient way to minimise the likelihood of false positives. It is possible within AIC to cross validate any features/patterns found, and thereby largely eliminate the undesired effect of multiple testing.

Future replication studies

Another large independent family-based collection consisting of 1052 complete families collected from 10 centres and funded by GlaxoSmithKline (GSK) has also been completed. This study has replicated a modest but significant association of the SERPINE2 gene with COPD ([23]) and further results are likely to be published from these studies. Family based studies are particularly sensitive to genotyping errors as this can often lead to false inferences as the test distribution depends on the assumption that parental genotypes are correct. This tends to inflate the false positive rate ([34]). However, the use of sophisticated algorithms for data analysis can minimise the potential sources of genotyping error as has been exemplified by the use of CHIAMO ([20]).

What have we learned from genetic studies?

Alpha1-antitrypsin deficiency remains a paradigm for COPD and is still the most important known genetic risk factor for COPD. Another related gene, SERPINE2 has been proposed as a potential risk factor in two independent studies but not confirmed in another ([22], [23], [24]). This finding needs to be tested further and remains a plausible candidate based on the central hypothesis of protease/anti-protease imbalance. Other candidates have not been reliably replicated and if these associations are real, these may be revealed in GWAS.

The way forward

The use of GWAS on very large resources combined with replication i.e., several thousand samples are likely to detect modest genetic effects and identify genetic contributions to COPD in an unbiased way. This can only be achieved by the pooling of resources on an international scale to provide the power to detect modest risk. It is conceivable that gene-gene and gene-environmental interactions may have profound effects on phenotype and further statistical tools are required to facilitate systematic analysis of these interactions in the most efficient way. These studies will undoubtedly be done and the expectation is that they will further elucidate the factors that contribute to COPD. At least they offer the best hope. We have reached an exciting stage in the genetic studies of COPD. The combined use of GWAS in well defined patients, together with the use of animal models for the disease and other advances in proteomics and expression profiling will allow an integrated approach and will lead to better insights into the disease.

We thank the European Union for financial support – Programme Grant QLG1-CT-2001-01012.