Patterns in Populations and Pedigrees

With all of the attention focused on silicon arrays, laser scanners, and the other “glamorous” gadgets of advanced genomic technologies, it is important to consider how much needs to be learned about the genetic basis of a disease before stepping into a DNA laboratory. In fact, a strong foundation of knowledge about a disease’s frequency in different populations and its occurrence among closely and distantly related family members (i.e., pedigrees) is absolutely essential. Without this foundation, endless gigabytes of DNA sequence data will not enable scientist–clinicians to develop a solid understanding of the causes of a disease. The human genome has not evolved in a test tube nor of inside a supercomputer; rather, it has existed for millennia in natural populations and been transmitted from generation to generation, from parents to children. Only by carefully studying the genetics of a disease in populations and pedigrees can we hope to begin to unearth the complex interactions between genes and the environment that underlie individual differences in disease susceptibility. This field of research is known as genetic epidemiology.

Genetic epidemiologists often first look at whether a disease occurs more often in some human populations than others. These comparisons can include both populations in different geographic areas as well as racial or ethnic groups living in the same region. Does the disease have more severe symptoms, more rapid rates of progression, or an earlier age of onset in some populations? Such findings suggest (but do not prove) that genetic differences that are important for the disease may exist among the populations.

Before the massive human migrations in recent centuries, most human populations existed in semi-isolation from other populations around the globe. As a consequence of natural selection (i.e., differential survival and reproduction), populations sometimes adapted genetically to their local environments. The most famous example of adaptation is the sickle cell hemoglobin variant that protects an individual against the infectious disease malaria. This variant is common among populations who live in areas where this mosquito-borne parasite has long been endemic, because it provides strong protection against the severe symptoms of this disease. The variant persists at high frequency in these populations, although persons who inherit two copies of the mutation (i.e., one from mother and one from father) are severely affected by the disease sickle cell anemia. A balance between the benefit of malaria resistance in persons who inherit only one copy of the variant versus the disadvantage of sickle cell disease keeps the variant at relatively high frequencies among populations where malaria is present. Another example of population differentiation by natural selection is the ability to digest the milk sugar lactose as an adult that evolved in Europeans in conjunction with the domestication of dairy cattle more than 8000 years ago. In addition to differentiation driven by natural selection, as is seen in these examples, a random process called genetic drift also causes populations with little or no migration between them to differentiate genetically over time. Thus, one cannot assume that every population difference observed has a functional biologic basis.

Regrettably, the comparison of periodontitis in different populations across the globe is extremely challenging because of the lack of calibrated examiners and standardized disease definitions.⁶ One of the most dramatic population differences in which data quality is not an issue is the observation that both localized and generalized forms of early-onset aggressive periodontitis occur about 10 times more frequently among African Americans as compared with Caucasians.³³ Human racial and ethnic groups often differ dramatically with regard to the frequency of mutations at genes that have major effects on disease risk. For example, cystic fibrosis is caused exclusively by recessive mutations in the CFTR gene, and it varies in frequency from 1 in 3000 Caucasians to 1 in 15,000 African Americans in the United States, whereas only 1 in 350,000 Japanese individuals are affected.⁴⁹ It is possible that the tenfold higher prevalence of early-onset aggressive periodontitis in African Americans is caused by the elevated frequency of high-risk gene variants in this population. However, additional evidence is needed before such a conclusion can be drawn. Although comparative studies of different populations may provide clues as to possible genetic mechanisms underlying a disease, the environments of the populations may also be dissimilar in important ways. It is possible that variations in diet or in exposure to pathogenic oral bacteria or to some unknown and unmeasured environmental factors could entirely explain the observed differences in the frequency of aggressive periodontitis between population groups. Until solid data confirm a genetic basis for population differences, we need to wait before drawing firm conclusions.

The comparison of disease occurrence or severity in identical (monozygotic) versus nonidentical (dizygotic) twins is a very powerful method for distinguishing between effects caused by variation in genes versus factors in the environment. This requires us to make what is usually a reasonable assumption that the environments of pairs of identical twins are no more or less similar than the environments shared by pairs of nonidentical twins. If variation among individuals in disease susceptibility or severity is caused entirely by factors in the environment, then we expect pairs of identical twins to be no more similar to each other than pairs of nonidentical twins. All twins (whether identical or nonidentical) are expected to be more similar to their co-twins, on average, than to unrelated members of their local population because they were raised in the same family environment, with similar diets, microbial exposures, and so on. However, if genetic variation plays an important role in determining a certain trait, then genetically identical twin pairs will be more similar to each other than nonidentical twin pairs. This is because identical twins share 100% of the same genes, whereas nonidentical twin pairs share only 50% of their parents’ genes on average. Genetic epidemiologists calculate a measure called heritability that is based on these correlations and that estimates the portion of all variation in the trait that is attributable to inherited genetic variation. Traits with variation that is determined entirely by differences in environmental exposures have heritabilities of 0.0, whereas traits with variation attributable solely to inherited genetic differences without any environmental influence have heritabilities of 1.0. Heritabilities are sometimes reported as a percentage that ranges from 0% to 100%.

Most human diseases and nondisease traits fall in the middle of this range, with heritability ranging between 0.25 and 0.75. For example, in one study, type II diabetes was estimated to have a heritability of 0.26, and abnormal glucose tolerance had a heritability of 0.61.⁵⁰ Note that, for it to be feasible to use the twin method with adequate statistical power, the disease has to be fairly common so that the researcher can recruit enough twin pairs in which at least one of the twins is affected by the disease. Not surprisingly, with regard to periodontal disease, only chronic periodontitis occurs frequently enough to have been studied using the twin design. Two twin studies of modest size (i.e., 110 and 117 pairs) have been reported, and these estimate the heritability of measures of chronic periodontitis to range between 40% and 80%, thereby clearly implicating genetic variation in disease risk.^43,44 Interestingly, a study of bacteria associated with periodontitis found no difference between identical versus nonidentical twins.⁴⁵ This suggests (at least for these twins, most of whom did not have severe periodontitis) that inherited variation in risk is not mediated by genes that influence the presence of specific bacteria in subgingival plaque.

Another method used by genetic epidemiologists to understand and distinguish different mechanisms of transmission of diseases through families is called segregation analysis. This is relatively straightforward for traits in which mutation in a single gene causes the disease to develop with nearly 100% certainty in carriers, whereas persons who do not inherit the mutation are at little or no risk. For example, carriers of a single copy of the Huntington disease gene mutation or carriers of two copies of a cystic fibrosis gene mutation always develop these diseases if they reach the ages at which symptoms of these conditions normally emerge. By tracking the transmission of these diseases in families, it is obvious, for example, that Huntington disease is a dominant single-gene disorder: it is transmitted with 50% probability to offspring of affected individuals and thus it is often found occurring across many generations of large pedigrees. By contrast, parents of children with cystic fibrosis are very rarely affected themselves, and 25% of siblings are affected by cystic fibrosis when the disease is present in a nuclear family. This pattern of transmission is expected if a disease is recessive (i.e., it requires the inheritance of a mutated gene copy from both parents, who themselves have one normal and one mutated copy and so are not affected). For most common “complex” diseases, however, having a high-risk gene does not automatically lead to development of the disease; this phenomenon is called reduced penetrance. Furthermore, several genes or even dozens of different genes may influence disease susceptibility; this is known as oligogenic inheritance and genetic heterogeneity. Environmental exposures are also important modifiers of disease risk. Such highly complex combinations of multiple genetic and environmental risk factors make the challenge of deciphering genetic mechanisms by merely observing transmission patterns in families using the segregation analysis approach unfeasible. The limitations of this approach were humorously illustrated a number of years ago in an analysis that facetiously presented evidence of a recessive gene controlling the trait of attending medical school.³⁹ “Risk” for this outcome among first-degree relatives of a doctor was elevated 61 times above that of the general population. More recently, a robust quantitative analysis of the family histories of characters in the Harry Potter series suggested that a dominant gene controls the inheritance of magic abilities.⁵⁴ Because the etiology of periodontitis is likely to be highly complex, segregation analyses of this disease that have been reported in the literature should be viewed with considerable skepticism. Unfortunately, the simplifying assumptions required for this method make the results unreliable and potentially misleading. For highly complex diseases, such as most cases of periodontitis, assays at the DNA level need to be combined with careful evaluations of clinical measures among related individuals to derive robust conclusions about a disease’s genetic architecture. Some of the key features of the different techniques for studying the genetics of periodontal disease are explained in Table 6-2.

Searching for Answers in the DNA

In theory, a genetic marker can be any type of biomolecule or assay that allows us to “read” inherited differences among individuals in their DNA sequences. Blood groups, protein isozymes, and human leukocyte antigens (HLAs) were among the first developed markers, but even simple traits that are controlled by single genes (e.g., eye color) can also serve this purpose. Genomic methods have made these methods obsolete, because researchers can now determine a person’s inherited variation directly at the DNA level for a much lower cost and with greater speed and accuracy. So-called “next-generation” DNA sequencing methods are projected to enable researchers and clinicians to obtain nearly the entire 3 billion DNA base human genetic blueprint for less than $1000 a few years from now.18,67 At present, most genetic studies use a combination of whole genome arrays that can evaluate up to 1 million variable DNA sites in a single assay in combination with lower-throughput methods that are used for the fine mapping of chromosome regions of special interest.⁵³ These regions are said to contain candidate genes of high priority for further investigation either because of the genes’ known biologic functions or because results of previous genome-wide surveys indicate strong statistical chances that disease susceptibility genes are located in certain regions of one or more chromosomes. Types of variation include single nucleotide polymorphisms (SNPs; pronounced “snips”) in which one DNA base is substituted for another; small insertions and deletions (“in/dels”) of one or more DNA bases; and larger structural changes in the DNA, such as inversions (in which a piece of DNA of hundreds or thousands of bases in size is cut out and sewn back into the chromosome in the opposite orientation) and copy number changes (in which a given segment of DNA is either missing or occurs in more than the usual two copies inherited as one copy from each parent).

Equipped with these powerful tools for rapidly measuring DNA variation, the next question to address is what kinds of study designs are most powerful for identifying the dozen or more variants that influence the risk of disease from among the millions of DNA differences that exist between any two individuals in a typical population. One method that has been highly successful for finding molecular defects related to simple genetic diseases caused by the mutation of a single gene, such as cystic fibrosis and Huntington disease, is linkage analysis.1,7 This gene-mapping strategy requires families with one or more members affected by the disease to be recruited and clinically and molecularly evaluated for a relatively small number of genetic markers. Depending on the type of marker used, as few as 500 markers or up to 10,000 markers distributed evenly across the genome are needed. Investigators usually try to recruit families with two or more close relatives, such as sibling pairs, who are affected by the disease as well as parents and other siblings who may be unaffected. With a null hypothesis that suggests that a region of the chromosome does not contain genetic variation that influences disease risk, siblings are expected to share identical genetic material inherited from their parents an average of 50% of the time. However, if the region of the chromosome being evaluated contains a gene that has a substantial effect on disease risk (e.g., increases the risk by tenfold or more), then pairs of siblings who are both affected by the disease will share the chromosome region that contains the disease gene substantially more often than 50% of the time, and the null hypothesis of 50% sharing will be statistically rejected if the study has an adequately large enough sample of such families. This simple example illustrates how linkage analysis is performed. In practice, both small and large extended families are studied; this will include the simultaneous evaluation of the sharing of genetic material among both affected and unaffected relatives. Sophisticated mathematical algorithms and computer programs are used to carry out the huge number of calculations required for data analyses.

After achieving many successes with the use of linkage analysis for “simple” diseases caused by the mutation of a single gene, this method was extended to complex diseases caused by combinations of multiple susceptibility genes and environmental risk factors. Unfortunately, these conditions proved to be beyond the reach of linkage analysis in most instances. Numerous studies conducted during the 1990s either failed to find any genes, or initially positive findings failed to replicate. Linkage studies of very large numbers of carefully diagnosed families for complex diseases (e.g., orofacial clefting, in which twin studies had firmly established heritability of 70%) identified at most a tiny fraction of this genetic variation. Mathematical analyses have subsequently shown that the linkage analysis gene-mapping strategy has extremely low statistical power for complex diseases in which each individual susceptibility gene has a relatively small effect on risk (e.g., twofold or less) and in which there is extensive heterogeneity among different families that have different combinations of susceptibility genes and environmental exposures.⁵⁶ Consequently, it is not surprising that linkage analysis has only been successfully applied to syndromic forms of periodontitis; this information is summarized later in this book.

Disappointment over this setback in human disease mapping caused by the initially unrecognized limitations of linkage analysis was short-lived. An alternative approach called association analysis was also available, although this had been relegated to studies of HLA and a few other markers of special interest during the prime of linkage approaches.1,7 Mathematical analyses indicated that, if we make some reasonable assumptions about the nature of genetic factors in human disease, this method could provide adequate statistical power for finding genes of small to modest effect on risk while requiring only moderately large sample sizes that would be feasible to recruit.⁵⁶ There was, however, one major catch: to search the entire genome using the GWAS method, the number of genetic markers needed was several orders of magnitude greater (i.e., 500,000 to 1 million assays needed per subject). Fortunately, advances in molecular assay technologies converged at this time period, and several methods of array-based genotyping provided such a capability at an acceptable cost.⁵³

How association studies are used to find disease susceptibility genes is illustrated for the case–control design in Figure 6-1. Association analyses are sometimes referred to as case–control studies, although this is only one of several sampling methods that can be used (including studies of families). Genotype frequencies of an inherited DNA variant for a group of periodontitis cases are statistically compared with the frequencies of the variant in a matched group of periodontally healthy control subjects. If the genotype frequencies differ so greatly that the results are very unlikely to occur by chance, then we conclude that the genotype that is more common in the cases as compared with the controls is “associated with” increased disease risk. In Figure 6-1, 59% of the cases have the CC genotype (having inherited a C allele from both of their parents), whereas only 17% of the healthy controls inherited the CC genotype. Therefore, this DNA variant could be used to predict periodontitis risk (but not until after the finding was validated in additional independent studies). Conversely, we can also say that the AA genotype is “protective” against disease because it occurs much more often in the healthy controls (50%) as compared with periodontitis cases (8%). Ideally, the cases and healthy controls are matched as closely as possible for race/ethnicity, smoking behavior, age, gender, and so on so that differences in genotype frequency are likely to be caused by real biologic effects on disease development or progression rather than as artifacts of some kind. For example, it is well known that races and ethnic groups sometimes differ dramatically with regard to genotype frequencies as a result of their historic isolation in different geographic regions. Consider, for example, a study that had mostly Swedish cases and mostly Italian controls. We know that there are thousands of DNA variants that differ substantially among these populations because of their geographic isolation throughout human history. Few if any of these variants have anything to do with differences in disease risk, but they could falsely appear to be associated because of the failure to carefully match ethnicity in cases and controls. In practice, this is usually not a problem, provided that investigators take reasonable precautions with regard to how cases and controls are selected. It is also now routine practice to use several statistical methods to check for mismatching and then adjust for this during data analysis if it occurs.

The good news is now clearly in: association studies have been a boon for the discovery of inherited genetic variation important for a wide range of complex diseases, including diabetes, cardiovascular disease, metabolic disorders, obesity, and mental illnesses. A recent review cites 24 genes identified with unquestionable statistical confidence for type 2 diabetes alone, and the list continues to grow.⁶⁴ Most of these genetic polymorphisms with elevated risk are very common in the population (i.e., from 5% up to >50%). Although each variant only increases risk slightly (i.e., twofold or less), because the risk alleles are so common, they can account for a nontrivial proportion of the occurrence of disease in the population; this is a measure that epidemiologists call attributable risk.

An especially attractive aspect of the GWAS approach is that, because the entire human genome is searched, we no longer have to depend on prior hypotheses about the disease’s molecular pathology. In most GWAS studies, about half of the statistically definitive findings point to genes that experts in the field had no suspicion whatsoever were involved in the disease’s etiology. This allows researchers to open up entirely new pathways for investigation that may lead to insights about the disease’s biologic mechanism and suggest novel molecular strategies for pharmaceutical or other therapeutic interventions. In more than a few cases, robust GWAS findings implicate regions of the human genome in which no genes appear to be present, thereby highlighting the limitations of our current knowledge of basic genome functions.

Although great progress has been made toward understanding the etiology of many complex human diseases by using GWAS methods, the approach has nevertheless usually failed to account for most of the heritability known to exist for these conditions.12,38 A recent study found that well-established nongenetic diabetes risk factors (e.g., gender, smoking, family history, body mass index, blood lipid and glucose levels) were better predictors of risk than a combination of the top 20 genetic markers for this disease.⁶⁵ To improve gene-based risk estimates, the missing heritability needs to be found. High hopes are now being placed on the emergence of next-generation DNA-sequencing tools. In theory, these may enable researchers to identify the less common (i.e., 1% to 5%) genetic variants that are predicted to have individual gene effects of greater magnitude on disease risk (i.e., greater than twofold but less than tenfold) that cannot readily be found with the use of either GWAS or linkage analysis methods.

Periodontitis in Genetic Syndromes and Other Diseases

A number of extremely rare conditions consistently include periodontitis among the array of clinical manifestations that define a syndrome. Many genetic syndromes involve mutations of single genes or larger chromosomal regions. However, a number of syndromes, such as fetal alcohol syndrome, are purely environmental in origin. Some of the syndromes that include periodontitis are caused by mutations in specific genes. For example, mutations in the cathepsin C gene have been shown to cause both Papillon–Lèfevre syndrome (Figure 6-2) and Haim–Munk syndrome as well as some forms of nonsyndromic prepubertal periodontitis, and they may also be associated with a risk of aggressive periodontitis.⁴⁸ Periodontitis frequently occurs with some subtypes of Ehlers–Danlos syndrome, Kindler syndrome, Down syndrome (trisomy 21), leukocyte adhesion deficiencies (Figure 6-3), hypophosphatasia, two types of neutropenia, and aplasia of the lacrimal and salivary glands. A large triracial extended family demonstrated evidence of a single gene that caused both early-onset aggressive periodontitis and dentinogenesis imperfecta. The gene has been mapped with the use of linkage to a chromosomal region that contains a dentin matrix protein gene.³⁷ Many of these conditions are so rare that few periodontists see even a single case during a lifetime of practice. However, dentists should be aware that these single-gene conditions exist; they need to be prepared to extend clinical evaluations to close relatives and to seek the assistance of or to refer to appropriately trained genetic counselors or specialists if a patient’s medical history or the presentation of multiple symptoms raises the possibility that he or she may be affected. Clinicians can obtain updated information about these conditions by accessing the publicly available Online Mendelian Inheritance in Man database and typing in “periodontitis OR periodontal disease” as the query term.² Further research is necessary to determine whether inherited variation in the genes that cause these rare syndromes may also influence the risk of nonsyndromic forms of aggressive or chronic periodontitis.

Nonsyndromic Aggressive and Chronic Periodontitis

In this section, evidence for the association of inherited genetic variation with aggressive and chronic periodontitis will be considered for cases that present without the co-occurrence of anomalies or disorders of other parts of the body or of the affected individual’s behavior. Such cases are appropriately classified as nonsyndromic periodontitis. This terminology is similar to the way that other human diseases, such as orofacial clefting, have long been recognized as occurring in both syndromic and nonsyndromic forms. The elevated risk of periodontitis that is associated with metabolic conditions (e.g., diabetes, which is addressed elsewhere in this book) is more appropriately considered a comorbidity rather than being a cause for the designation of a syndrome.

As described previously, twin studies have shown that chronic periodontitis has very substantial heritability, and we know that aggressive periodontitis aggregates strongly in families. Because aggressive periodontitis occurs so rarely, it is not feasible to perform a twin study to confirm the heritability of this condition. Neither segregation analyses nor gene-mapping linkage studies are capable of providing reliable information about the genetic etiology of a highly complex disease such as periodontitis. However, large numbers of susceptibility genes have been identified for complex disorders, such as diabetes and cardiovascular disease, using association analysis. It seems reasonable to expect that similar successes could be achieved for periodontitis using this approach. In fact, during the past decade, several hundred papers have reported associations of nonsyndromic aggressive and chronic periodontitis with polymorphisms in a number of candidate genes. Certain classes of genes (e.g., cytokines) that have long been a focus of attention by immunologists and cell biologists studying pathogenic mechanisms associated with periodontitis have received the most attention. Early reports of relatively weak associations with variation in interleukin-1 (IL-1) genes led to a large number of attempts to replicate and extend these findings. Unfortunately, with very few exceptions, association studies of periodontitis have been very inadequately powered to detect genetic variation with modest effects on disease risk or progression (i.e., sample sizes that are much too small). In addition, inconsistency with regard to the methods used to classify subjects as periodontal cases versus controls or to quantitatively measure disease severity and extent greatly limit our ability to draw sound conclusions by comparing results reported in different studies.

The reason why association studies of periodontitis have largely failed will be challenging to fully address. The first issue is straightforward and simply a matter of numbers. It is noteworthy that the successes achieved for many complex diseases (e.g., diabetes) with the use of the GWAS mapping approach were based on sample sizes involving thousands of cases and controls, with multiple replications by independent teams of investigators. Statistical theory shows that, to detect genes of modest effect, these large sample sizes are absolutely essential. With the use of a statistical power calculator developed for case–control studies,⁵² sample sizes required for 80% power are shown in Figure 6-4 for a study that involves only a single genetic marker as well as for studies that evaluate 5, 50, or 500 independent genetic markers in which the effects of multiple comparisons need to be accommodated. In this example, we assume that the risk gene acts in a dominant manner, with the high-risk allele occurring at a frequency of 25% in the population, and that this allele causes risk among carriers to increase by twofold (i.e., a greater effect on risk than observed for many susceptibility alleles found in GWAS studies of other complex diseases). Many periodontitis association studies reported in the literature involved multiple markers in each publication, and often the same research team reported positive findings for other genes in subsequent papers. Furthermore, because of the difficulties of publishing negative findings (i.e., when no association is found), many research teams working in this area may assay 50 or more genetic markers over the course of their work over several years. Results shown in Figure 6-4 demonstrate that, to obtain 80% power, a study of 50 markers would require more than 200 cases and 200 controls. Even if a research team assays only 5 SNPs, their study would still require 100 cases and 100 controls to achieve adequate power.

Our search of PubMed in early 2010 using the search term “(periodontal disease OR periodontitis) AND (SNP OR SNPs OR polymorphism OR polymorphisms OR linkage)” identified 311 periodontitis gene association tests with a P value of 0.05 or less for at least one statistical test reported. These results are summarized in Figure 6-5, in which the x-axis indicates the number of cases included in the study; the P value for the strongest finding is plotted on the y-axis. When more than one statistical test was reported, only the one with the smallest P value is shown in this figure. In some cases, this inflated the actual statistical significance, because investigators rarely adjust their findings for these multiple tests when the findings are reported in a publication. This analysis shows clearly that most association findings have been drastically underpowered if periodontitis is assumed to be a complex disease. The majority (66%) of these association reports for chronic and aggressive periodontitis are based on samples of 100 cases or less, and 41% are based on less than 60 cases. As shown in Figure 6-4, studies of such small sample size have little power to detect a susceptibility gene that increases risk by twofold. Given the added concern about publication bias (i.e., that positive findings are more likely to be accepted for publication), we can have little confidence that even the more statistically significant findings are valid and likely to be independently replicated if they are based on such very inadequate samples sizes. With only a few exceptions, which are noted later in this chapter, publications since 2010 have continued to involve small numbers of cases and to provide marginal statistical support for association.

Aside from important lessons about how not to carry out association studies of a complex disease, there are some tentative conclusions that can be drawn from the data available thus far. Detailed results of our review and manual curation of 298 publications reporting association findings for periodontitis are presented in Table 6-3, and key findings are condensed in Table 6-4.