Lee et al. BMC Medical Genetics 2010, 11:68 http://www.biomedcentral.com/1471-2350/11/68 RESEARCH ARTICLE Open Access RPeesedarcHh aurtincleter 2.0 and its usage to characterize the founder structure of the Old Order Amish of Lancaster County Woei-JyhLee1, ToniIPollin2, JeffreyRO'Connell2,3, RichaAgarwala1 and AlejandroASchäffer*1 Abstract Background: Because they are a closed founder population, the Old Order Amish (OOA) of Lancaster County have been the subject of many medical genetics studies. We constructed four versions of Anabaptist Genealogy Database (AGDB) using three sources of genealogies and multiple updates. In addition, we developed PedHunter, a suite of query software that can solve pedigree-related problems automatically and systematically. Methods: We report on how we have used new features in PedHunter to quantify the number and expected genetic contribution of founders to the OOA. The queries and utility of PedHunter programs are illustrated by examples using AGDB in this paper. For example, we calculated the number of founders expected to be contributing genetic material to the present-day living OOA and estimated the mean relative founder representation for each founder. New features in PedHunter also include pedigree trimming and pedigree renumbering, which should prove useful for studying large pedigrees. Results: With PedHunter version 2.0 querying AGDB version 4.0, we identified 34,160 presumed living OOA individuals and connected them into a 14-generation pedigree descending from 554 founders (332 females and 222 males) after trimming. From the analysis of cumulative mean relative founder representation, 128 founders (78 females and 50 males) accounted for over 95% of the mean relative founder contribution among living OOA descendants. Discussion/Conclusions: The OOA are a closed founder population in which a modest number of founders account for the genetic variation present in the current OOA population. Improvements to the PedHunter software will be useful in future studies of both the OOA and other populations with large and computerized genealogies. Background dementia [5], diabetes [6], blood pressure [7,8], hip frac- The Old Order Amish (OOA) of Lancaster County in tures/osteoporosis [9], and vision phenotypes [10]. Inves- Pennsylvania are a closed founder population, with tigators at the University of Maryland School of Medicine approximately 40,000-50,000 living individuals that can (UM-SOM) [11] have recruited over 4,800 OOA from be connected into a single 14-generation pedigree. Other Lancaster County for several studies of complex adult- large OOA populations live in Ohio and Indiana. There onset diseases beginning in 1993 with the initiative of Dr. have been numerous medical genetics studies on the Alan R. Shuldiner [6,12]. These OOA studies have used OOA. For several decades, the medical genetics studies genome-wide linkage analysis [13-15], candidate gene focused on monogenic diseases such as brittle hair dis- association analysis [16], and most recently genome-wide ease [1,2]. association studies (GWAS) to discover variants influenc- More recently, research interests have broadened to ing traits such as cardiac repolarization [17], type 2 dia- include complex traits, such as Parkinson disease [3,4], betes [18], hypertension [19], and fasting and post- prandial triglyceride levels [20] or to validate locus asso- * Correspondence: [email protected] ciations first discovered in other populations, e.g., diabe- 1 National Center for Biotechnology Information, National Library of Medicine, tes [21], human height [22], fasting glucose levels [23], National Institutes of Health, 8600 Rockville Pike, Bethesda, Maryland 20894, USA Full list of author information is available at the end of the article © 2010 Lee et al; licensee BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons At- BioMed Centraltribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited. Lee et al. BMC Medical Genetics 2010, 11:68 Page 2 of 13 http://www.biomedcentral.com/1471-2350/11/68 waist circumference [24], bilirubin levels [25] and In this paper, we report on new features in PedHunter response to the anti-clotting agent clopidogrel [26]. and on how those features can be used to quantify the The OOA comprise one of several groups in the more impact of the pedigree structure of a founder population general category of Anabaptists; other Anabaptist groups on the amount of genetic variation present in the current living in North America include other Amish, Menno- population. Under a model of genetic drift, the number of nites, and Hutterites. Medical geneticists have been inter- founder alleles present in the current population will ested in Anabaptist populations because they are closed depend on factors such as sibship size and number of new populations and have written genealogies and other fea- genomes that enter each generation. From AGDB we can tures such as a homogeneous, rural lifestyle and high answer this question: How many OOA founders contrib- standard of living [27-29]. Their genealogies were mostly uted what expected percent of the genetic material to the published in paper books, which make integration and present-day living OOA? The answer impacts the genetic search challenging and time-consuming tasks. architecture, and hence phenotypic distribution of In 1996 we set out to construct a computer-searchable important traits such as low density lipoprotein (LDL) genealogy database of the OOA of Lancaster County in and triglyceride (TG) levels. A recent study characterized Pennsylvania for use by geneticists [30]. To support the the patterns of linkage disequilibrium in the OOA [51]. In database, we developed a suite of query software that this paper, we quantify the representation of founder would solve pedigree-related problems automatically and genes in the current population that contribute to those systematically. The digital genealogy database expanded patterns of linkage disequilibrium. to include other Anabaptist populations and was renamed Anabaptist Genealogy Database (AGDB) [31- Methods 33], and the query software package was named Ped- PedHunter Versions Hunter [30,33]. The initial source to construct AGDB was PedHunter version 1.0 [30] that provided 23 queries was a 1996 computer file updating the Fisher Family History first released in 1998. In versions 1.1, 1.2, 1.3, and the cur- (FFH) book [34] edited by Ms. Katie Beiler. Later versions rent 2.0 we increased the number of queries to 50. The integrated the Amish and Amish Mennonite Genealogies full list of 50 query programs and seven utility programs (AAMG) book [35], a large computerized genealogy file is in the Additional file 1. PedHunter 2.0 has been tested from Mr. James Hostetler and two smaller updates from on platforms running the Linux, SunOS, Windows, and Ms. Beiler of recent births, deaths, and marriages in the Mac OS X operating systems. The queries are imple- Lancaster area. The queries and utility programs of Ped- mented using Transact-SQL and C version of Open Cli- Hunter are illustrated by examples using AGDB in this ent DB-Library. All PedHunter queries and utility paper. programs are available as executable files that can be used Several groups have used AGDB in their research. from the command line prompt and in UNIX shell Some worked on rare diseases, such as nemaline myopa- scripts. The current version of PedHunter is freely avail- thy [36], congenital microcephaly [37], and dystonia able and can be downloaded from http:// [38,39]. Although AGDB contains no explicit phenotype www.ncbi.nlm.nih.gov/CBBresearch/Schaffer/ped- data, lifespan can be inferred when both birth and death hunter.html. dates are available. Several studies on lifespan using PedHunter queries a genealogy database stored either AGDB have established that lifespan is heritable [40], that as a SYBASE relational database or in structured ASCII lifespan may be associated with other implicit traits plaintext files. Before version 2.0, the source codes for [41,42], and that lifespan can be associated with experi- these two variants were separate with much duplication. mentally measured traits [43]. The single set of code files in version 2.0 can be adapted A frequent problem in medical genetics is to recon- to either type of database representation with small struct the pedigree relationships among distant relatives. changes to one code header file. PedHunter was first designed to solve optimal pedigree Queries in PedHunter 2.0 connection problems and other problems related to pedi- PedHunter 2.0 supports four categories of queries as gree construction, verification, and analysis. PedHunter is basic operations: 1) testing a relationship; 2) finding all not formally tied to AGDB; in particular, access to AGDB individuals satisfying a certain relationship; 3) printing requires ethics approval from an Institutional Review information; and 4) complex queries. We use italics to Board (IRB), while PedHunter is freely available. Ped- indicate the names of specific queries. Queries that find Hunter has been used to construct genealogies for link- pedigrees print the pedigrees in LINKAGE format [52]. age and haplotype analysis on Hutterite families [44-47] In the second category, PedHunter 2.0 adds two queries and analyses of an Icelandic population [48], a Southern pertinent to our study of OOA founder structure: Italy population [49], and a Northern Italy population founder_descendant to find founders of a given individ- [50]. Lee et al. BMC Medical Genetics 2010, 11:68 Page 3 of 13 http://www.biomedcentral.com/1471-2350/11/68 ual, and a new query count_descendant to count number Each individual in AGDB is assigned a unique integer, of descendants per ancestor from an input file in the for- called the program id. Two main tables in the PedHunter mat of ancestor-descendant pairs. New complex queries pedigree storage system are the person table and the par- pertinent to our study include: calculate_r to calculate ent-child relationship table. relative founder representation (RFR) for a founder- There have been several versions of AGDB based on the descendant pair, and average_r to calculate mean RFR per sources mentioned in Background and with sizes summa- founder in AGDB, as explained below. rized at the beginning of Results. The current AGDB ver- sion 4.0 (AGDB4) was created in 2004, after the most Examples of Queries Useful for Analysis of the Anabaptist recent published description [33], to combine all the Genealogy Database sources and updates. Some small batches of corrections To find possible participants in AGDB to be recruited and updates have been incorporated based on feedback into a study, the living query is useful to find living indi- from users. viduals. The founder_descendant query can be followed As indicated by the change in the first word of AGDB to find the founders of each living individual. For each from the original "Amish" to "Anabaptist", many of the founder-descendant pair, the calculate_r tool computes individuals in AGDB are not OOA. As explained in the RFR, defined below. The average_r tool then com- Results, we extracted a subpedigree from AGDB4 that putes the mean RFR for each founder. The person infor- should include most living OOA individuals in the Lan- mation for founders can be obtained using the caster area and their ancestors. We used the method person_info query. Prior to running the asp query to find described in the next subsection to predict which individ- "all shortest paths" pedigree(s) connecting a set of sam- uals are OOA. pled individuals, the subset query is required to first find a maximal subset of individuals that shares a common Prediction of Old Order Amish Status ancestor. An adult or near-adult (older adolescent) is of OOA sta- tus if the individual belongs to the OOA church, which Pedigree Renumbering means that the individual has chosen to be baptized. For PedHunter is also useful for genetic evaluations on large the purpose of our query, we included children and young pedigrees. For example, at the United States Department adults who were not yet baptized but lived with their of Agriculture (USDA), the number of animals in the Hol- OOA parents. For individuals in the printed FFH book, stein dairy pedigree used for genomic evaluation of determining the OOA status by eye is usually easy, important economic traits has grown from 41,000 to because for most family records, the religious affiliation is 125,000 in less than 18 months as new animals are added. included. However, there are some omissions and some The ability of PedHunter to scale to millions of subjects is cases that are unclear. Due to the syntax in FFH, the word a feature. Other pedigree storage/query packages, such as "OOA" appeared only once per family record. When PEDSYS [53], are not able to process such a large pedi- AGDB was generated, the OOA status was captured only gree (though PEDSYS adds the capability to connect large into the record of the person whose name preceded the amounts of phenotype data with individuals in smaller word "OOA". If the family head was married, the OOA pedigrees, such as the subset of more than 4,800 OOA status was usually assigned to the spouse of family head individuals that have been studied at UM-SOM). To facil- in AGDB. For example, the FFH book reads "1. Christian itate handling large pedigrees, we added the Fisher b April 26, 1757, d Nov. 19, 1838, farmer, m Bar- renumber_pedigree utility to PedHunter to number and bara Yoder (Yost Yoder) OOA, ...". Christian Fisher's per- order subject identifiers, and add missing spouses/mates son record in AGDB does not include any OOA status, if necessary into the pedigree. The renumbering process but Barbara Yoder's record does. The spouse and children ensures the property that identifiers of parents are of the person assigned OOA status should also have OOA smaller than the identifiers of their children, which status in most cases. Therefore, to predict if an individual enables more efficient calculation of kinship coefficients in the genealogy requires some inferences across rela- [[54], Ch. 5]. Adding missing parents into the pedigree is tionship links. For individuals added into AGDB by Ms. necessary for some software packages, such as LINK- Beiler after the FFH book was printed, the amount of AGE, which assume that each person has either zero or information about religious affiliation rapidly declined to two parents in the pedigree. The original identifiers are zero. We believe that virtually all individuals added in included in the output, so that information can still be updates by Ms. Beiler in 1999 and 2003 are OOA, due to connected to these identifiers. biased ascertainment. Prediction of OOA status is done in three different Anabaptist Genealogy Database Version 4.0 ways: 1) for an individual born in or before 1971, we con- AGDB was created in SYBASE SQL Server release 11.0.x sider person records of self, parents and spouses; 2) for an of the SYBASE relational database management software. Lee et al. BMC Medical Genetics 2010, 11:68 Page 4 of 13 http://www.biomedcentral.com/1471-2350/11/68 individual born in 1972-1986, we check person records of estimate missing birth years, we performed a frequency self, parents, spouses and spouses' parents; 3) for an indi- test on known parent-child pairs with known birth years vidual born in 1987 or later, we examine person records of for both individuals in AGDB4. We extracted birth year self, parents, grandparents and spouses. For an individual of each founder record with known birth year, and sub- I, if any of the relatives of I considered has OOA status, tracted from birth year of the oldest child with known then we predict that I has OOA status. birth year. There are 7,373 known founder-and-oldest- child pairs. The mean birth year difference is "25.7 years", Pedigree Trimming on Founders and the median birth year difference is "25 years". There- Working with large pedigrees, trimming redundant or fore, we used 25 years per parent-child generation in the irrelevant individuals (as defined below) from a pedigree estimation of birth years for founders in AGDB4. will reduce the computational time of some pedigree If the birth year of a founder was unavailable, we sub- analysis methods [55]. In a nuclear family that is at the tracted 25 years from the birth year of the (known or esti- top of the pedigree and has only one child, we can replace mated) oldest child. If birth year of a child is unavailable, both parent founders with their only child in the analysis we recursively estimated birth years among the children because such a child's genomic data represents all of the child. The following example estimates a founder genomic data derived from those parents. Nuclear fami- (unknown name in AGDB4)'s birth year to be "1695" by lies at the top of pedigree are recursively trimmed, until subtracting 75 years (three generations) from the esti- no more trimming is possible. Figure 1 illustrates a mar- mated oldest child (Henry Stehly)'s estimated oldest child ried founder couple Hans Beuttschi and Margaret Zum (Magdalena Stehly)'s oldest child (Catherine Sieber) who Bach with their only son Peter Beuttschi. After one round was born in 1770. of trimming, Peter Beuttschi replaces his parents as a trimmed founder. Peter Beuttschi was married to Marga- Mean Relative Founder Representation rete Oswald, who is a female founder, and gave birth to We created and used the calculate_r tool to calculate the their only son also named Peter Beuttschi. After a second relative representation of each founder in each study par- round of trimming, Peter Beuttschi replaces his parents ticipant. We define the RFR of a given founder in a given as a new trimmed founder. descendant as the expected proportion of alleles in the descendant that were inherited identical-by-descent Estimation of Birth Years for Founders (IBD) from the founder. For example, an offspring inher- Many of the older person records in AGDB did not have its half its genome from each parent, thus the founder birth dates in the data sources. To report our analysis by representation for each parent is one-half. The expected birth cohorts, it is necessary to estimate birth years for proportion can be computed as twice the kinship coeffi- individuals; the estimated birth years are used for the cient between the parent and offspring. The kinship coef- analysis in this project, but are not recorded within ficient is defined as the probability at a given locus that AGDB. For simplicity and to allow inclusion of records the allele selected from one individual will be identical by with known dates missing the month or day, we used only descent to a randomly selected allele of a second individ- birth years but not actual birth dates. To systematically ual. For example, twice the kinship coefficient between a parent and an offspring, and the RFR of that parent in that offspring, is 0.5. In a three-generation pedigree, the grandparent-grandchild kinship coefficient is 0.125; thus the relative representation of each grandparent in a grandchild is 0.25. It should be noted that founder repre- Hans Beuttschi Margaret Zum Bach sentation represents the average across the genome, whereas at any autosomal locus the two alleles an individ- ual inherits come from at most two founders. From these definitions and the pedigree structure we use average_r to calculate the mean RFR for each founder over all study Peter Beuttschi Margarete Oswald descendants. By definition the RFRs in a given individual will sum up to 1. A subtle point is that the founder representation (expected IBD) as described above is calculated assuming Peter Beuttschi no inbreeding in the founders because by definition, their ancestors are not in the genealogy. Conceptually, the Figure 1 Sample pedigree trimming in AGDB4. Two recursive probability of inheriting a founder allele is a property of rounds of pedigree trimming on three founders down to one trimmed founder in AGDB4. Mendelian sampling in the pedigree, independent of the Lee et al. BMC Medical Genetics 2010, 11:68 Page 5 of 13 http://www.biomedcentral.com/1471-2350/11/68 probability that the two alleles of a founder are inherited born in 1930-2000. Using the living tool in PedHunter 2.0, IBD from unknown ancestors. Thus, the RFR of a parent we found 51,905 presumed living individuals, with no in an offspring is 0.5 only under the assumption of no death date recorded in FISHER. By predicting the OOA inbreeding. If the parent were completely inbred (say a status among 54,411 individuals born in 1930-2000, we mouse strain) then twice the kinship coefficient would be find 36,057 to be OOA. Among 51,905 presumed living 1.0. The former agrees with what we expect, while the lat- individuals, the results showed 34,160 presumed living ter would imbalance the founder representation between OOA individuals born in 1930-2000 and in FISHER. The an inbred father and non-inbred mother. difference between the 34,160 individuals used here and the 40,000-50,000 estimate in the Background is due to Results children born in 2001-2009 and individuals born before The size of AGDB has grown from AGDB1, completed in 1930 who are still alive. We focused attention on the indi- 1998 containing 55,636 individuals and 12,896 marriages, viduals born in 1930-2000 because death records for to the current AGDB4, containing 417,789 individuals those born before 1930 are rather incomplete, and birth and 102,341 marriages [30-32]. In 2003, we built a special records after 2000 are currently incomplete, although we version called FISHER that includes 66,131 individuals plan to address these deficiencies in AGDB. and 15,057 marriages to accommodate three updates Table 1 reports the birth year and gender distribution since FFH was published in 1988. The results here are among above 34,160 presumed living OOA individuals in based on both FISHER and AGDB4. The subjects of FISHER. The results show increasing numbers of births interest in the ongoing genetic studies at UM-SOM are in over time. The year with the most newborns is 1995, in FISHER, but some of their ancestors who are not in which there were 1,073 birth records (488 females, 572 FISHER can be found in AGDB4. males, and 2 of unknown gender). The data consistently show more males than females born. Distribution of Old Order Amish We then located these 34,160 individuals in AGDB4 by The procedure to analyze mean RFR of founders in pre- mapping their program ids from FISHER to AGDB4. sumed living OOA individuals is illustrated in Figure 2. Using the ancestors tool in PedHunter, we constructed a By executing the age tool in PedHunter 2.0 on the 66,131 pedigree containing a total of 43,162 individuals, includ- individuals in FISHER, we collected 54,411 individuals ing the above 34,160 presumed living individuals, in FISHER AGDB4 database database For each FISHER id i: CCoonnvveerrtt FFIISSHHEERR iidd ii to AGDB4 id d. FFoorrmm ppaaiirrss ooff ttrriimmmmeed founder and pprreessuummeedd lliivviinngg OOOOA descendant. Identify all ancestor founders of d. For each founder- Was i born in 1930-2000 descendant pair (f,d): No without a known death date (as presumed living)? FFoorrmm ppaaiirrss oofff mmmaaarried founder ccooouuupppppllleees. Calculate RFR as r of founder f in Discard i. Yes presumed living OOA descendant d. FFoorr eeaacchh ffoouunnddeerr couple (p,m): Yes No Is i an OOA? Calculate mean RFR r for each founder fover all presumed living (cid:2)1 NNuummbbeerr ooff ooffffsspprriing for (p,m)? OOA descendants. ===1 Procedure to analyze Sort all trimmed founders by the mean RFR in AGDB4 RReeppllaaccee pp aanndd mm bbyy the offspring c. descending order of mean RFR r ’s. Figure 2 Procedure to analyze mean RFR in AGDB4. Flowchart to analyze mean RFR of founders in presumed living OOA descendants. Lee et al. BMC Medical Genetics 2010, 11:68 Page 6 of 13 http://www.biomedcentral.com/1471-2350/11/68 Table 1: Distribution of birth years and genders for presumed living OOA in FISHER. Birth year Female Male Unknown gender Sum 1930-1940 651 650 0 1,301 1941-1950 1,014 1,128 0 2,142 1951-1960 1,360 1,452 0 2,812 1961-1970 1,894 2,042 0 3,936 1971-1980 3,131 3,320 0 6,451 1981-1990 4,031 4,347 4 8,382 1991-2000 4,414 4,698 24 9,136 Total 16,495 17,637 28 34,160 Birth year cohorts of ten years are reported in rows, and genders in three categories are reported in columns. The total number of presumed living OOA individuals in FISHER is 34,160. AGDB4. The 43,162 individuals form a 14-generation may be inaccurate. Table 3 reports on the birth years of pedigree, in which there are 606 founders. There are 22 554 founders and numbers of their living OOA descen- living individuals recorded as founders themselves. The dants. The largest coverage number is 34,100 living OOA study at UM-SOM confirms that many of these are descendants, who are descended from the oldest founder spouses of individuals who left the Amish church, couple. Of 554 founders, 361 were born before 1800. although they may have been OOA at some time and may However, many of these founders had few descendants. have OOA children. We were interested in calculating the number of founders Among the 606 founders, 355 are female and 251 are contributing to the majority of the extant population to male. There are 175 marriages between 175 female evaluate qualitatively and quantitatively our assumptions founders and 170 male founders; five male founders had about the utility of the OOA for gene mapping. two marriages to female founders, with the second mar- We used the calculate_r and average_r tools in Ped- riage following the death of the first spouse in all cases. Hunter 2.0 to calculate the relative representation of each Among 175 marriages, 102 founder couples have exactly founder in an average living OOA individual. By ranking one child. By applying the trimming technique described founders in descending order of their founder representa- in Methods, 606 founders are reduced down to 555 tion, we were able to count the number of founders repre- trimmed founders after one round and 554 trimmed sented in the majority of expected alleles in an average founders after two rounds. All subsequent analysis of individual among living OOA individuals. founders uses this trimmed set of 554 founders. In the We computed the average expected genetic contribu- trimmed set, there are 332 female founders and 222 male tion of each of the 554 founders (332 females and 222 founders, and 136 married founder couples including 136 males) to each of the 34,160 presumed living OOA females and 131 males. descendants as summarized in Figure 3. We found that Table 2 reports the birth years for the 554 founders. 128 founders (78 females and 50 males) accounted for The two most ancient founders are a married couple, over 95% of the average founder contribution, with a sin- estimated to have been born in 1670. Prior to 1776, birth gle founder accounting for nearly 7% of the average con- years of more than 50% of founders are not recorded in tribution. Half of the total founder representation came AGDB4. All but one founder born after 1866 have known from only 16 founders (11 females and 5 males). These birth years. data quantify the extent to which the OOA are a closed founder population ideal for elucidating the role of Old Order Amish Founder Structure genetic variation in complex diseases. Using the pedigree structure from AGDB4, one can esti- To put the coverage analysis in some chronological per- mate the contribution of the founders to the current spective, it is useful to consider Isaac Huyard who was OOA gene pool. This is an estimate because, as described born on February 7, 1865. He married into the OOA on in Methods, the calculations are based on probabilistic December 8, 1891 after working as a servant for an analysis of the pedigree structures rather than observed Amish family, and is anecdotally the last individual who DNA segregation. Moreover, there may be additional was born and reached adulthood non-OOA, and then relationships (not in the genealogy) among the desig- became OOA [56]. 125 of the highest contributing 128 nated founders and some relationships in the genealogy founders (accounting for 95% of gene pool) were born in Lee et al. BMC Medical Genetics 2010, 11:68 Page 7 of 13 http://www.biomedcentral.com/1471-2350/11/68 Table 2: Distribution of known and estimated birth years of 554 founders in AGDB4. Birth year Female Male Sum 1666-1675 1 (1) 1 (1) 2 (2) 1676-1685 8 (7) 8 (7) 16 (14) 1686-1695 5 (5) 4 (4) 9 (9) 1696-1705 13 (12) 14 (8) 27 (20) 1706-1715 19 (18) 10 (8) 29 (26) 1716-1725 15 (11) 11 (6) 26 (17) 1726-1735 21 (17) 12 (7) 33 (24) 1736-1745 23 (18) 15 (8) 38 (26) 1746-1755 27 (22) 14 (8) 41 (30) 1756-1765 17 (14) 14 (6) 31 (20) 1766-1775 18 (11) 9 (4) 27 (15) 1776-1785 22 (10) 13 (5) 35 (15) 1786-1795 19 (6) 15 (5) 34 (11) 1796-1805 19 (7) 15 (2) 34 (9) 1806-1815 14 (2) 11 (5) 25 (7) 1816-1825 10 (2) 8 (4) 18 (6) 1826-1835 13 (3) 13 (3) 26 (6) 1836-1845 8 (4) 4 (2) 12 (6) 1846-1855 6 (1) 4 10 (1) 1856-1865 4 (2) 7 (3) 11 (5) 1866-1875 4 3 (1) 7 (1) 1876-1885 4 1 5 1886-1895 1 1 2 1896-1905 6 0 6 1906-1915 5 1 6 1916-1925 2 1 3 1926-1935 2 2 4 1936-1945 1 0 1 1946-1955 6 1 7 1956-1965 9 2 11 1966-1975 5 8 (1) 13 (1) 1976-1985 5 0 5 Total 332 (173) 222 (98) 554 (271) Birth year cohorts of ten years are reported in rows, and genders are reported in columns. The numbers represent both known and estimated birth years in the age group; the numbers in parentheses are the number of founders among that cohort whose birth years were estimated, as described in Methods. or before 1865. The other three high-contributing found- genetic material into the OOA population since the mid- ers were born in 1867, 1901 and 1904. On the other hand, 19th century. 70 founders born after 1865 account for only 0.8% of the Among 70 founders born after Isaac Huyard, we found founder representation. Thus, a detailed study of the ped- that 12 founders are adopted, and another two founders igree structure of the OOA as catalogued in AGDB sup- are likely adopted. One of the improvements in going ports the notion that there has been very little influx of from AGDB3 to AGDB4 was the identification of adopted Lee et al. BMC Medical Genetics 2010, 11:68 Page 8 of 13 http://www.biomedcentral.com/1471-2350/11/68 Table 3: Distribution of birth years and numbers of descendants among 554 founders in AGDB4. Birth year 1≤#(D)<10 10≤#(D)<100 100≤#(D)<1,000 1,000≤#(D)<10,000 10,000≤#(D) Sum 1666-1715 0 0 3 30 50 83 1716-1765 8 23 58 54 26 169 1766-1815 17 51 66 19 2 155 1816-1865 24 29 19 4 1 77 1866-1915 9 10 7 0 0 26 1916-1965 25 1 0 0 0 26 1966-1985 18 0 0 0 0 18 Total 101 114 153 107 79 554 Birth year cohorts of fifty years are reported in rows, and numbers of descendants #(D) are reported in columns. Founders born earlier have larger numbers of presumed living OOA descendants. individuals, so that those "non-biological parent to child" (numbers of founders per individual ranged between 27 links are not in the pedigrees that we construct, if the and 82) of the 201 founders born before 1800. That is, the adoptive relationships are known. entire gene pool of those individuals was derived from founders born before 1800. 99% of participants studied Relevance of Founder Structure to Published and Ongoing are descended entirely from a subset (numbers of found- Studies ers ranged between 21 and 110) of the 255 founders born As of October 2009, the OOA studies at UM-SOM before 1900. A subset of 127 founders with more than 50 included approximately 4,800 participants. 30% of these living descendants comprises the entire gene pool of 90% participants studied are descended entirely from a subset of the study participants. A subset of 102 founders with Figure 3 Cumulative mean RFR in AGDB4. Cumulative mean RFR of 554 founders in 34,160 presumed living OOA descendants. Lee et al. BMC Medical Genetics 2010, 11:68 Page 9 of 13 http://www.biomedcentral.com/1471-2350/11/68 more than 100 descendants comprises the entire gene and kinship coefficients are both increased in temporal pool of 77% of the study participants. statistics. Analysis of cumulative mean RFR shows that 128 Excess of Female Founders founders accounted for over 95% of the average founder The number of female founders (332) is larger than the contribution among all living OOA descendants. Such number of male founders (222) among 554 founders. We results confirm that the OOA in Lancaster County are evaluated three possible causes for the excess of female truly a closed population. The combination of lack of new founders: 1) remarriages of male founders after widow- genetic material, lack of socio-economic variation, and hood; 2) marriages comprising one founder and one non- detailed genealogies make the OOA ideally suited for founder; 3) possible overestimate of female founders due identifying some rare variants that are associated with to the difficulty of tracking relationships between females complex phenotypes [20]. The examples of GWAS repli- with different married surnames. cations cited in Background prove that some trait-associ- We mentioned previously that there are 136 marriages ated SNP alleles seen in other populations can also be between 136 female founders and 131 male founders. In found in the OOA. Our characterization of the founder addition to the 136 married founder couples, there are structure provides an explanation for why other trait- 196 female founders married to non-founder males, and associated alleles did not enter the OOA population; for there are 91 male founders married to non-founder example, cystic fibrosis is one of the most common reces- females. Thus, remarriage of widowed male founders is sive diseases in individuals of European ancestry, but only a minor factor in the excess of female founders. The hardly seen among the OOA [57]. propensity for female founders to marry non-founders is That the OOA have a small number of founders, but a more substantial factor, but this may be confounded by linkage disequilibrium (LD) patterns similar to outbred difficulty in correctly determining the founder status of European populations [51] may seem paradoxical. The females. apparent paradox is explained by noting that LD is a mea- The lack of knowledge about relationships among sure of non-random association of alleles at different loci. female founders is partly due to the male bias in the FFH Linkage between the loci maintains this association and AAMG books. Among 332 female founders, only 263 under random mating and drift, while recombination will have partial names given, and only 207 have surnames, decay the association. Thus, allele and haplotype frequen- making it difficult to track relationships. There are 168 cies drift together. For example, suppose two SNPs have distinct surnames using identical spelling, and 159 after D' = 1 in the founders, so that only three of the four possi- grouping some sets such as {Moser, Mosser, Musser} that ble haplotypes are observed, with frequencies 0.15, 0.35 are likely to be different spellings of the same surname. If and 0.5, respectively, but drift to 0.20, 0.40 and 0.40 in the any individuals with the same surname share ancestry, we current population. Both allele and haplotype frequencies would be able to reduce the number of female founders. have changed, but LD has not changed. The assumption Occasionally, the books hint at a hidden relationship of common alleles is important as random drift over the among founders, via footnotes, but the footnotes were 14 generations spanned by AGDB will not eliminate com- not used systematically as they are often speculative. We mon variation. For low frequency alleles there is longer conclude that the excess of female founders is mostly due LD in the Amish due to linkage as shown in [51] and also to cryptic or unknown relationships, so that some of the in some of our association studies [20,58], where the most apparent female founders are not really founders. significant signal can be more than 500 kbp from the causal allele. Discussion/Conclusions One weakness in our characterization of the OOA With new queries and utility programs in PedHunter ver- founder structure is the apparent excess of female found- sion 2.0, we identified 34,160 presumed living OOA indi- ers. We tested three hypotheses and concluded that the viduals born in 1930-2000 and connected them into a 14- excess is mostly due to cryptic or unknown relationships, generation pedigree descended from 554 founders, after such that some female founders are not really founders. trimming. Because of the small number of founders, the Analyses of mitochondrial genomes could provide evi- frequency of consanguineous marriages in OOA steadily dence regarding some of these cryptic relationships. The increased over time and reached approximately 85% for male lineage and Y chromosome inheritance have been individuals born in 1940-1959. Among consanguineous validated [59] a unique Y chromosome STR haplotype marriages, the median kinship coefficient stayed stable in bred true within each of the 28 male lineages represented the 19th century, but rose in the 20th century. Table 4 in the UM-SOM study sample. The 27 surnames of the reports kinship coefficients of married OOA couples who males comprising these lineages captured 98% of the had at least one offspring in AGDB4. Numbers of couples households in the 1998 Lancaster County Amish Address Book, which contains 42 distinct surnames. This number Lee et al. BMC Medical Genetics 2010, 11:68 Page 10 of 13 http://www.biomedcentral.com/1471-2350/11/68 Table 4: Distribution of birth years and kinship coefficients among married OOA couples with offspring in AGDB4. Birth year #(couples) Average 25% percentile Median 75% percentile 1866-1875 98 0.0244 0.0164 0.0258 0.0315 1876-1885 113 0.0272 0.0164 0.0258 0.0350 1886-1895 161 0.0272 0.0183 0.0267 0.0363 1896-1905 184 0.0286 0.0194 0.0292 0.0382 1906-1915 245 0.0289 0.0222 0.0309 0.0374 1916-1925 366 0.0300 0.0218 0.0297 0.0387 1926-1935 467 0.0317 0.0241 0.0318 0.0402 1936-1945 739 0.0310 0.0241 0.0326 0.0392 1946-1955 1,103 0.0315 0.0247 0.0319 0.0396 1956-1965 1,390 0.0335 0.0274 0.0344 0.0408 1966-1975 1,731 0.0352 0.0296 0.0352 0.0414 1976-1985 607 0.0363 0.0302 0.0359 0.0416 Union set 7,273 0.0326 0.0259 0.0333 0.0402 Birth year cohorts of ten years are reported in rows. A couple is assigned to the interval containing the average of the two birth years. Numbers of married OOA couples #(couples) and kinship coefficients are reported in columns. Couples born later tend to have larger kinship coefficients. is much lower than the 222 total male founders because cess on pedigrees in large genealogies. For example, of genetic drift, conversion of some earlier settlers to AGDB has been used to estimate the encatchment popu- other Anabaptist sects, and westward migration [35]. lation of specific hospitals to determine the denominator One limitation of our RFR calculations is that the vari- of the hip-fracture incidence [60]. For a study of osteo- ance of the kinship coefficient due to Mendelian sampling genesis imperfecta, AGDB was used prospectively to variance is not included in the calculation. For example, identify individuals likely to carry a genetic variant of the kinship coefficient is 0.25 for both parent-offspring phenotypic interest [61]. A pedigree constructed from and full sibs, but the variance is zero for parent-offspring AGDB made it possible to trace the origin of a rare and non-zero for full sibs. Analytical expressions of kin- APOC3 null allele conferring a favourable lipid profile ship variances are not possible for complex pedigrees, but and apparent cardioprotective phenotype to a couple empirical variances can be obtained through simulation born at the turn of the 19th century [20]. Additionally, by repeated gene-dropping of distinct founder alleles PedHunter has been used in other pedigrees, to discover through the pedigree. Computationally efficient gene- genetic drift and founder effects [62]. dropping can be achieved using the ancestor-first pedi- In sum, we quantified the founder structure of the gree renumbering described above. We may implement a OOA and implemented numerous improvements to the function r-empirical in a future version of PedHunter. software PedHunter that will be useful in future studies of The notion of "founder" we used is with respect to a both the OOA and other populations with large, comput- trimmed genealogy, not including any explicit conditions erized genealogies. on the locations of birth or death. The data sources focus on individuals who lived at least some years in the present Additional material United States, but not exclusively. To quantify this, we looked at the source information on 220 founders with Additional file 1 PedHunter 2.0 queries and utility programs. known or estimated birth years in and before 1755, and we found that 81 founders (44 females and 37 males) Abbreviations AAMG: Amish and Amish Mennonite Genealogies; AGDB: Anabaptist Geneal- probably died in Europe or on the sea. Therefore, there ogy Database; FFH: Fisher Family History; GWAS: genome-wide association may be even more female founders who never reached studies; IBD: identical-by-descent; LD: linkage disequilibrium; OOA: Old Order North America. Amish; RFR: relative founder representation; UM-SOM: University of Maryland School of Medicine. To increase the utility of AGDB and other digitized genealogies, we continue to enhance PedHunter with Competing interests queries and utility programs to assist the discovery pro- The authors declare that they have no competing interests.