How to Use an Article About Genetic Association: Title and subTitle BreakA: Background Concepts

A 55-year-old man consults you, worried about his risk of developing Alzheimer disease. His grandfather had dementia in his 70s, and his own father was diagnosed with dementia at 65 years. He has been a smoker since he was 20 years of age, works as an electrician, and has been taking antihypertensive medication (thiazide and β-blocker) for the last 5 years. He has never had his cholesterol level checked. He has recently read a news story about genetic tests and asks you whether he should have any genetic test for dementia risk, in particular for a gene called APOE.

Consulting your electronic medical texts, you find a section addressing genetic risk factors for dementia. You come across words such as allele, genotype, and genetic association and encounter many abbreviations, including APOE, e3/e4, and e4/e4 (Box). A colleague with an interest in genetics steers you toward a helpful Web site, the HuGE Navigator,¹ and you find that there are apparently more than 1000 publications on genetic associations of Alzheimer disease that discuss several hundred genes; the whole endeavor seems daunting. A quick Internet search reveals that many companies offer genetic tests for Alzheimer disease, often as part of testing hundreds of thousands of genetic variations. You realize that you need to review your basic genetics knowledge and discover more about how to read genetic association studies.

Additive

Allele

Candidate gene study

Chromosome

Dominant

Genome

Genome-wide association study

Genotype

Haplotype

Heterozygous

Homozygous

Isoform

Linkage

Linkage disequilibrium

Locus/loci

Messenger RNA

Mutation

Pedigree

Phenotype

Polymorphism

Recessive

Ribosome

SNP

Synonymous SNP

Variant allele

Wild-type allele

The Human Genome Project has greatly stimulated interest in genetic determinants of disease. The determinants of common mendelian diseases that involve a single gene (eg, cystic fibrosis, Huntington disease) are well established. Current research addresses the role of genetics in the chronic diseases composing the major causes of human morbidity and mortality, diseases that result from the concomitant effect of environmental, behavioral, and genetic factors. In the last year alone, major studies have tested hundreds of thousands of genetic variations (“polymorphisms”), trying to establish the genetic determinants of coronary artery disease,² type 2 diabetes,³ stroke,⁴ multiple sclerosis,⁵ breast cancer,⁶ bipolar disorder,⁷ rheumatoid arthritis,⁸ Crohn disease,⁷ and Alzheimer disease.^{9 - 11}

There are many reasons why a clinician may wish to understand the results of genetic association studies. This genetic information may shed light on pathways involved in disease and identify new targets for therapeutic interventions, advances that are of interest in basic and applied research. Genetics might be able to improve diagnosis or improve the ability to use therapeutic agents more efficiently and with less risk, an area termed pharmacogenomics. Although few genomic applications have reached this level of utility, despite the claims of companies marketing genetic tests, this might change in the future. The most likely short-term clinical effect of this genetic information is to enhance risk stratification and provide patients with information about their prognosis.

This series of 3 articles aims to provide clinicians with an introduction to reading and understanding genetic studies in medicine. This first article summarizes the key genetic concepts clinicians must understand to read articles offering genetic associations for possible clinical use; those familiar with basic genetics may wish to move straight to the second article. That article, along with the third, will focus on issues of validity and applicability fundamental to critical appraisal of such studies. Throughout the series, our presentation will be simplified but sufficiently detailed to allow readers to judge the relevant issues.

In 1953, James Watson and Francis Crick proposed a winding staircase (double helix) structure of DNA (Figure 1). The sides of the staircase or ladder, called strands, are formed by alternating sugar (deoxyribose) and phosphate molecules; the rungs of the ladder are formed by 4 nitrogen-containing ring compounds called bases: adenine (A), thymine (T), guanine (G), and cytosine (C). A pair of these bases forms each rung of the ladder; adenine always binds to thymine and cytosine always binds to guanine to form the full rung. Thus, each rung of the helix ladder is called a base pair. A single base plus its associated sugar and phosphate groups is called a nucleotide.

One long stretch of double-stranded DNA, forming the spiral staircase, constitutes 1 chromosome, on which there are many genes, a gene being a stretch of DNA that typically codes for 1 protein. Twenty-three chromosomes in the sperm and the corresponding 23 chromosomes in the egg come together at fertilization to form the entire “DNA set” of a human, called the genome. Each person therefore has 23 pairs of chromosomes, of which 1 pair is the sex chromosomes that determine sex. The other 22 pairs are numbered 1 to 22, providing each person with 2 versions of each gene, one on the maternally inherited and one on the paternally inherited chromosome (Figure 2).

DNA is the blueprint for making proteins that build cells and tissues and enzymes that catalyze biochemical reactions within a cell. The information on the DNA guides the production of proteins through a 2-step process of transcription and translation.

The first step involves transcribing DNA into messenger RNA (mRNA) (Figure 3). The mRNA molecule migrates out of the nucleus to the cytoplasm to reach the protein-building machinery of the cell, known as the ribosome. Here, during the second step, the mRNA molecule is translated into protein, using the code to link amino acids one at a time. These processes of transcription and translation convert the genetic information at one gene, called the genotype, into the protein that, in concert with other genes, their proteins, and environmental exposures, determines the final attribute, the phenotype (eg, hair color, height, thrombophilia).

Sequencing the human genome—identifying the entire sequence of base pairs in the 25 000 genes that constitute human DNA—has revealed that the sequence is more than 99% identical across different people.¹² However, the human genome includes 3.3 billion base pairs; thus, even with this high level of similarity, there are still more than 12 million potential variations between 2 people's genomes.^{13 - 15} Differences in gene structure that occur infrequently in populations, ie, less than 1%, are called mutations, whereas differences that occur more frequently, ie, greater than or equal to 1%, are called polymorphisms (Figure 4). These polymorphisms may take a number of different forms:

The different forms or variants that a particular polymorphism may take are called alleles. For example, the APOE gene in the clinical scenario has 3 alleles, named e2, e3, and e4 (e for epsilon; for historical reasons, there is no e1). The location along the DNA strand at which a particular allele is present is called a locus. In a feral moment, geneticists decided to call the form of the gene that is most common in the population the wild type; we will use the less colorful term common allele; and the less common allele(s), variant allele(s). The different alleles—in the case of SNPs, only the nonsynonymous SNPs—result in production of different forms of the protein for which the gene is responsible. These different proteins are called isoforms.

Of the APOE alleles, allele e3 is most common in white populations (78%) and thus represents the common allele; e2 and e4 are variant alleles (with a frequency of 6% and 16%, respectively, among white individuals). Physiologically, the apoE protein carries a form of cholesterol and binds to the apoE receptor on the surface of cells for the cholesterol to be metabolized. Of the 3 protein isoforms that result from the 3 corresponding APOE alleles, the e2 isoform has decreased strength of binding, or affinity, to the apoE receptor. The proteins resulting from the e3 and e4 alleles, the e3 and e4 isoforms, have higher affinity compared with e2. (For the sake of simplicity, we have presented the 3 alleles of the APOE gene, leading to the 3 isoforms, as if they are changes at one point in the DNA. In fact, these isoforms are defined by changes at 2 points in the DNA, leading to changes in 2 amino acids (positions 112 and 158) that make up the apoE protein: alleles e2, e3, and e4 at the DNA level lead to amino acids cysteine/cysteine, cysteine/arginine, and arginine/arginine at these 2 sites, respectively).

Because each person has 2 versions of each chromosome, one from the mother and one from the father, individuals have 2 APOE genes. The 2 relevant chromosomes (in the case of apoE, the 19th chromosome) may carry the e3 allele (denoted e3/e3) or 1 each of e3 and e2 alleles (denoted e2/e3), or any other combination, eg, e2/e4, e3/e3. This shorthand denotes the “genotype” of the individual. Individuals with 2 copies of the same allele are said to be homozygous, or a homozygote; ie, an individual with e3/e3 is homozygous for e3, or an e3 homozygote. Conversely, an individual with 2 different alleles, eg, e2/e3, is a heterozygote, or heterozygous.

Individuals with e2/e3 produce some e2 protein and some e3 protein. The question then follows: which protein function wins? If the variant allele is dominant, it need be present in only 1 of the 2 genes to carry all the biological activity. In such cases, the allele on the other chromosome will remain biologically “silent.” Conversely, an allele that is recessive will need to be present on both genes, ie, homozygous, to affect function; it will otherwise remain silent. If 2 differing protein isoforms resulting from 2 different alleles (eg, e2, e3) share function, the model of protein function is known as an additive, or per-allele, model. These dominant, recessive, and additive/per-allele models are called models of inheritance, or genetic models.

As it turns out, the APOE alleles act pretty much in an additive manner; thus, e2/e3 individuals have overall apoE function that is partway between the reduced affinity of an e2 homozygote and the higher affinity of an e3 homozygote. With most genetic association studies for complex diseases, the inheritance model remains unknown.

In genetics, it is usual to describe the distribution of the alleles of interest in the population. In the same way that most continuous variables in medicine observe a normal distribution, most allele distributions observe what is called Hardy-Weinberg equilibrium (HWE). The Hardy-Weinberg law states that if there are 2 alleles at a particular locus, named A and a, with frequency p and q, respectively, then after 1 generation of random mating the genotype frequencies of the AA, Aa, and aa groups in the population will be p², 2pq, and q², respectively. Given that there are only 2 alleles possible, A or a, then p + q = 1, and p² + 2pq + q² = 1. It is general practice in a genetic association study to check whether the allele frequencies observe HWE proportions. Deviations from HWE in the population may be due to the following:

Deviations from HWE may also signal methodological problems with the genetic study (eg, error in genotyping or population stratification), a possibility discussed in the next article in this series.

Investigators conducting genetic association studies may target genes for investigation according to the known or postulated biology and previous results, an approach known as candidate gene association. Alternatively, they may screen the entire genome for associations, an approach that has, in recent years, transformed the field of genetic association studies. Researchers now investigate hundreds of thousands of SNPs across the entire genome, without any previous hypotheses about potential mechanisms or candidates. This type of “agnostic” study, “genome-wide association,” or GWA, has greatly accelerated the pace of discovery of genetic associations.^{16 - 17}

As discussed in the next article in this series, testing so many potential genes simultaneously carries the risk of finding many spurious associations. For this reason, SNPs that seem to have strong or suggestive statistical signals in an initial GWA study are then tested for replication in other large data sets or studies. To ensure that the discoveries are not just statistical flukes, replication studies are often published along with the initial data.⁷ The boundaries between candidate gene studies and agnostic GWA studies can become blurred, and the 2 types of studies are not mutually exclusive: GWA studies propose new candidates for replication but may also interrogate traditional candidates.

Whether hypothesis-driven or GWA studies, genetic association studies usually represent population-based investigations in which diseased and nondiseased individuals are unrelated. Genetic studies (either candidate or genome-wide) may also be carried out among family members in large pedigrees affected by a rare disease caused by a rare mutation. In such studies, the presence of certain stretches of DNA or genes in family members with disease and their absence in those without disease is called linkage analysis. The methods and interpretation of family studies differ radically from those of population-based studies,^{18 - 19} and this series is restricted to discussion of the latter.

Because detecting SNPs is relatively easy and they are responsible for most of the genetic variation in humans, they have been the focus of most of the research exploring gene-disease association. The basic idea of a gene-disease association study is relatively simple. In the same way that variation in an exposure is linked to an outcome (eg, cholesterol and myocardial infarction) in a traditional epidemiologic study, variation in a gene is linked to an outcome (eg, a SNP in the cholesteryl ester transferase gene and myocardial infarction) in a genetic association study.

One goal of traditional epidemiology, elucidating causation, can be frustrated by noncausal associations. Traditional epidemiology tries to deal with this problem by adjusted or multivariate analysis. For example, when examining the association of hypertension with stroke, investigators will simultaneously adjust for other variables, including age, sex, smoking, and obesity.

In genetic association studies, one goal may be to establish whether a SNP is causally associated with the outcome (eg, presence or absence of disease). This requires isolating the function of a particular SNP from the other SNPs that may be nearby in the gene. In practice, because stretches of the genome tend to be inherited together as a unit (a situation known as linkage disequilibrium), this may be difficult. Thus, the association of a SNP with an outcome, no matter how strong, may be noncausal. It is possible that some other SNP in linkage disequilibrium with the one under study is the true causal variable. Although this is an important distinction if the aim is to understand the underlying biology to develop a new therapeutic agent, it is not critical if the aim is to use the SNP simply as a marker of risk.

The linkage disequilibrium phenomenon results in “haplotype blocks,” stretches of DNA defined by the presence of high linkage disequilibrium among the SNPs present. Two or more SNPs that are in linkage disequilibrium in the same haplotype block can define haplotypes, specific combinations of variants across these SNPs. The combinations of variants that can arise are dictated by the extent of linkage disequilibrium.

As an example, consider SNP A with a common allele frequency (A) of 80% and SNP B with a common allele frequency (B) of 60%. If there is no linkage, allele A at SNP A and allele B at SNP B will be found together in the same person 0.80 × 0.60 = 36% of the time, ie, consistent with chance. With perfect linkage disequilibrium, eg, where the SNPs are very close together, it may happen that allele A is always found with allele B. The extent of this linkage disequilibrium may be expressed a number of different ways, with metrics such as r² (a plain correlation coefficient) or D′. Describing the properties of each metric goes beyond the scope of this article¹⁶ ; however, in general, the greater these measures, the greater the degree of linkage between the variants, with a D′ of 1 indicating that 2 alleles are always found together.

Armed with this background knowledge of genetic concepts and terminology, you are now ready to embark on an evaluation of validity, results, and applicability in genetic association studies. The remaining 2 articles in this series will address these issues.