The most common method of familial searching involves generating a list of possible relatives of the unknown person (whose DNA was picked up at a crime scene) by performing a profile search of a forensic DNA database that is designed to find “partial matches” between crime-scene evidence and offender profiles. In the Combined DNA Index System (CODIS) the term “search stringencies” describes one of three search modalities. A “high-stringency” search means that all alleles of the loci that are present in both DNA profiles (crime-scene DNA and database profiles) are identical. This represents the standard search for a cold hit, where the crime-scene profile matches exactly one and only one profile on the database. In a “moderate-stringency” search the alleles of a locus (among two DNA profiles) with the least number of distinct alleles must be present in the corresponding locus of the other DNA profile. For example, under “moderate stringency” one DNA profile with STR alleles 7, 10 will generate a match with another profile with STR alleles 7, 7. The heterozygote 7, 10 is deemed a “moderate-stringency” match with the homozygote 7, 7.
For a “low-stringency” match (used to find parent-child relationships), each locus that is compared between two DNA profiles must have at least one allele of that locus present in the other DNA profile. Thus a locus with STR 7, 10 would meet the standard of a “low-stringency” match with locus STR 3, 7.6
In a second familial search method police may conduct what is called a “rare-allele” search, in which they analyze the crime-scene DNA for highly unusual DNA signatures. Close relatives of those matches are then tracked down and asked to “voluntarily” provide a DNA sample.7 A third method of DNA profile searching is a variant of the second. It considers both the number of matching alleles resulting from a low-stringency search and their frequencies in a population. It is designed to reduce the number of false positives, that is, identification of individuals who are not relatives of the perpetrator.
Although this relatively new use of forensic DNA databases has led to a handful of somewhat remarkable success stories, familial searching, when practiced routinely, effectively expands DNA databases to all the close blood relatives of the individuals in the database, subjecting entire families (and perhaps even neighborhoods or even ethnic populations) to lifelong genetic surveillance. This gives rise to a number of social and ethical questions that deserve serious consideration.
Techniques and Practices
The premises behind familial searching are twofold. First, close matches in DNA profiles are more likely to indicate that the sources of the DNA are close family members rather than two unrelated individuals. Second, the closer the DNA match is, the higher the likelihood that the individuals are related, particularly when the matching alleles are rare in the general population. The most common and simplest technique used to identify potential relatives of suspects of a crime in the absence of a complete match with the database is allele counting, where forensic investigators compare the overall number of alleles shared between the crime-scene evidence and the database profiles. Generally speaking, the larger the number of shared alleles, the greater the probability of family ties. Siblings, on average, have about 18 of 26 possible alleles in common, while unrelated individuals average about 8 out of 26 in common.
BOX 4.1 Serial Killer Found from Sister’s DNA
In 2006, 49-year-old James Lloyd was convicted of four rapes and two attempted rapes that had occurred between 1983 and 1986. A father of two and described as “a wealthy businessman,” Lloyd was discovered to be the notorious “Dearne Valley Shoe Rapist,” who had tied his women victims up with tights and stolen their shoes. The police got to Lloyd through his sister, who was in the database for a drunken-driving conviction. Forensic scientists narrowed an initial partial-match list and provided the South Yorkshire Police the names of 43 people in the database who were possible relatives of the sex offender. In following up on the 43 names, police knocked on the door of Lloyd’s sister, who told the police that her brother roughly fit the age and height of the wanted individual and agreed to contact him. The sister called her brother and warned him that the police were after him. Upon hearing that he had become a suspect, Lloyd attempted suicide by trying to hang himself in his home but was saved by his 17-year-old son. While arresting him, police found more than 100 pairs of stiletto-heeled shoes hidden behind a secret trapdoor on the premises of Lloyd’s printing firm, where he worked. Lloyd pled guilty to four rapes and two attempted rapes and was sentenced to life imprisonment.
Sources: Tony Lake, chief constable, Lincolnshire Police, presentation before the FBI Symposium on Familial Searching and Genetic Privacy, Arlington, VA, March 17–18, 2008; Paul Sims, “20 Years After His Evil Reign, Shoe Rapist Is Unmasked by His Sister’s DNA,” Daily Mail (London), July 18, 2006; Andrew Norfolk, “Genetic Bar Code That Reopened the Case,” The Times (London), July 18, 2006.
Allele Counting: High, Moderate, and Low Stringency
As discussed in chapter 2, in the U.S. DNA data-banking system a complete match of two DNA samples occurs when all 26 alleles in the 13 loci (two alleles per locus) are identical, that is, each of the alleles at the locus has the same number of short tandem repeats (STRs). The FBI’s definition of a “partial match” requires that the crime-scene DNA and the offender DNA profiles share at least one allele at each genetic location tested, but the definition was not intended to address familial (or kinship) relations.8 In the hypothetical example in table 4.1, four samples, A, B, C, and D, are compared at locus number 1. The number of STRs for alleles at locus 1 is identical for samples A and B. If all 13 loci show the same concordance, we have an exact match for the samples.
In the event of a partial match, STR values will not be identical for all 26 alleles. Sample C matches the crime sample at moderate stringency, while sample D matches A at low stringency.
TABLE 4.1 High, Moderate, and Low Stringency for a Single Locus
The U.S. CODIS software system was designed to run moderate-stringency searches. This means that when a search is conducted against CODIS, the computer compares a given DNA profile with the database, looking not only for perfect matches but also cases where, for one or more loci, one sample contains only one allele (sample C, 9) and the other is heterozygote (sample A, 9, 13), with one of the alleles the same as the single one. This search criterion was designed not for purposes of conducting familial searches, but instead for capturing cases where DNA from a crime scene is partially degraded (so that the crime-scene DNA profile is a partial profile), or where the crime-scene sample may contain a mixture of two or more DNA profiles, or where “allelic dropout” (failure to detect an allele during sampling or failure to amplify an allele during the polymerase chain reaction) or a mistyping may have occurred. In other words, the intent of the moderate-stringency search was to build in a safety factor so that suspects who do have a profile that is identical to that of the crime stain but who would be overlooked by a high-stringency match search will in fact be identified. The CODIS software system was not set up to run low-stringency searches routinely, although it presumably would not be difficult to change the search parameters of the software.
For familial searches, forensic investigators must decide the criteria for a partial match. Some states use 13 alleles out of 26 to define a partial or low-stringency match. The state of Florida requires 21 out of 26. Investigators can examine the allelic similarities in partial matches and make predictions about how closely related the donors of the DNA samples are. They do this by making use of statistical information from large databases of the allelic homology of the DNA from family relations.
Henry Greely and colleagues reported on the allelic matching statistics for first-, second-, and third-degree relatives.9 A close or first-degree relative is a parent or sibling. On average they are expected to share about 50 percent of one another’s DNA variants (between 13 and 16 alleles). A father or mother and child match at no fewer than 13 alleles. Second-degree relatives include uncles, aunts, nephews, nieces, grandpar
ents, grandchildren, and half brothers and sisters. These relatives share about one-quarter of their DNA variations. Finally, third-degree relatives, who consist of great-grandparents and great-grandchildren, share about one-eighth of their DNA variations (see table 4.2).
TABLE 4.2 DNA Allele Matches Among Relatives
Greely and colleagues have estimated the probability of unrelated people matching a certain number of alleles on the basis of different scenarios. These calculations are useful benchmarks for understanding what the chances are that a low-stringency match will yield a relative of the perpetrator of a crime in the offender’s database. They conclude:
On average, the chance that an unrelated person’s genotype will match the genotype from crime scene DNA of 13 or more of the 26 alleles, allowing for all possible ways of distributing the matches across the markers, is around three percent. However, the chance that two unrelated people match at thirteen or more sites with every marker having at least one match (as will occur for parent-child pairs) is about one in two thousand.10
Matches from these criteria, then, are likely to produce close family members. However, they are also likely to miss most family members. In a memorandum sent to Attorney General Jerry Brown of California, Michael Chamberlain, head of the state’s DNA legal unit, wrote: “Under the FBI’s definition of partial match, it would preclude detection of 99.9% of brothers, many of whom have no alleles in common at a genetic location.”11
BOX 4.2 Familial Search Reveals a Posthumous Suspect
On September 16, 1973, two 16-year-old girls were raped and strangled in southern Wales, United Kingdom. Some months later a third 16-year-old girl met the same fate. The similarity of these heinous murders suggested that they could have been committed by the same man. During the initial investigation the police focused on interviewing 30,000 individuals. Subsequently, using techniques such as psychological profiling and an intelligence-led screen, police targeted 500 potential suspects, including a man named Joseph Kappen, but there was insufficient evidence to charge any of these individuals. In 2000, 27 years after the crimes were committed, investigators in the Forensic Science Service obtained DNA profiles from clothing stains of two of the girls and submitted the profiles to the United Kingdom’s National DNA Database (NDNAD). Although no exact match was found, a low-stringency analysis indicated that the DNA partially matched the DNA profile of a man named Paul Kappen. Police surmised that someone in Kappen’s family was the murderer, and this led them back to Paul Kappen’s father, Joseph, who had since died in 1990 at age 49. Meanwhile, in 2002 a comparison of the crime-scene DNA of the third girl’s murder showed that the three crimes were linked. British law-enforcement authorities obtained DNA samples from the Kappen family, including Paul Kappen’s mother and his siblings. The close matches between the crime-scene DNA and family DNA profiles were sufficiently credible for the police to obtain a warrant to exhume the body of Joseph Kappen. After his body was exhumed on May 17, 2002, it was learned that his DNA was an exact match with the crime-scene DNA from the three murders. Forensic investigators found the murderer by familial searching, albeit posthumously.
Source: Robin Williams and Paul Johnson, “Inclusiveness, Effectiveness and Intrusiveness: Issues in the Developing Uses of DNA Profiling in Support of Criminal Investigations,” Journal of Law, Medicine and Ethics 33, no. 3 (2005): 545–558.
Limitations of Allele Counting
Partial-match searches give rise to a number of questions that are technical on the surface but quickly become ethical. First, there is the question whether partial-match searches will miss potential relatives (false negatives). The second question is whether these searches will falsely identify unrelated individuals (false positives) and send investigators off on wild-goose chases. As was seen in the Craig Harman case, partial-match searches can yield results in the thousands. Ultimately, the question becomes, “How close is close enough?” In other words, what is the appropriate threshold for determining whether a partial match is likely to have revealed a relative of the actual perpetrator, warranting further investigation?
The false-negative and false-positive problems are compounded by the search criteria employed by the CODIS software system, which is not only used in the United States but also has been exported to at least one-fifth of the European countries using forensic DNA databases.12 Canada also employs CODIS software. The FBI gave the system to the Royal Canadian Mounted Police for free.13 As noted earlier, the system was designed to perform moderate-stringency searches for the 13 CODIS markers. The original purpose behind its use was to account for instances where crime-scene samples are compromised—for example, where they are partially degraded or in cases where there is a mixture, so that it is difficult to tell which alleles come from the alleged perpetrator. As applied to looking for family members, moderate- and low-stringency searches will pick up most parent-offspring relationships, since a child necessarily inherits one allele from each of his or her parents. However, as pointed out by researchers at the University of Washington, a moderate-stringency search will miss the overwhelming majority of full as well as half sibling relationships.14 This is because for each of the 13 pairs of alleles in a DNA profile, there is a 25 percent chance that siblings will not match at either allele. But if this occurs at any one of the 13 pairs of alleles, the familial relationship will be overlooked at the level of a moderate-stringency match. It turns out that the chance is only about 1 in 1,000 that a sibling will be identified at the level of a moderate-stringency search.15
The problem becomes more complicated, however, if we move to a low-stringency search. By relaxing the search criteria, we will increase the chances of picking up real siblings, or other related individuals, thereby improving the false-negative problem. However, we will greatly exacerbate the false-positive problem; the overwhelming number of individuals identified will not in fact be related at all. The reason this occurs is simply because there is a significant sharing of alleles within the general population. What we are looking for in a familial search is not this general sharing but, instead, sharing that results from two individuals with a recent common ancestor. The problem with allele counting, whether of low or moderate stringency, is that there is no way to distinguish whether these matching alleles are an indication of relatedness—echoes of common ancestry—or are simply coincidental.
In practice, both low- and moderate-stringency searches may very well send law enforcement off on wild-goose chases, especially in cases where they are conducted against large databases. As Greely and colleagues reported, although the chance that two unrelated people match at 13 or more sites with every marker having at least one matching allele is small (1 in 2,000), even this small percentage can yield a high number of false leads when one is dealing with a database of several million.16 More recently, George Carmody, a population biologist at Carleton University and a member of the New York State Forensic Science Commission’s DNA Subcommittee, estimated that a low-stringency search against the U.S. national database would generate on the order of 8,000 false partial matches for every real sibling match.17 This number of false leads will necessarily grow as databases continue to expand.
The number of false positives will depend in part on the relative rarity of the alleles in the crime-scene sample. As Greely and colleagues have cautioned, “The partial match is only a lead—a relatively weak one for a common genotype though possibly a very strong one for a rare genotype.”18
Similarly, David Paoletti and colleagues have pointed out that the value of a given partial match is dependent on a number of important parameters, including the relative frequencies of the matching alleles, the number of initial suspects considered, and the population size, or the number of potential alternative suspects. They conclude that it is not possible to come up with a single threshold (such as “number of shared alleles”) for use in determining when a partial match warrants further investigation. A partial match of as few as 5 alleles out of 26 might be significant and sufficient ground
s for follow-up investigation when all 5 of those alleles are rare and the alternative suspect pool is small. But when all the matching alleles are relatively common, as many as 15 alleles might need to be shared.19
Mathematical models (discussed in the next section) that predict kinship relationships from DNA can support a hypothesis that two people are related on the basis of a few shared alleles even when they are not related. In other words, aggressive police work can pursue family members of partially matched individuals on the basis of tenuous or tentative mathematical assumptions. In one study of 194 Caucasians profiled by the FBI on 13 loci, pairwise comparisons of the profiles were made. In this sample 1,654 pairs of individuals partially matched at 9 loci, and 797 pairs partially matched at over 9 loci.20 Unless the mathematical models are validated and become canonical, their use in forensics will remain controversial.
BOX 4.3 The Case of the Grim Sleeper
A serial killer, colloquially named the Grim Sleeper, who had murdered African American women in South Los Angeles since 1985, left his saliva and other DNA at several killing sites. Los Angeles police connected the perpetrator to 10 victims. A genetic sample preserved from one of the crime scenes matched samples collected from a 14-year-old killed in 2002, the body of a woman killed in 2003, and another victim found in 2007. The Los Angeles police ran the DNA found at the crime scene against millions of genetic profiles of convicted criminals in the state DNA database. Police tried to locate the unknown killer’s relatives in the hope that there would be similar DNA patterns by comparing the crime-scene DNA profile (under low-stringency conditions) with those in the state’s DNA database of more than 1 million felons. Initially no suspects were found. Eventually the state forensic lab found a partial match in the Grim Sleeper case with Christopher Franklin, recently convicted of a felony, that suggested a father-son relationship. The trial led to Christopher’s father, Lonnie D. Franklin Jr., 57, when police found a DNA match from his saliva on a discarded pizza slice. In July 2010 Lonnie Franklin was declared the Grim Sleeper by Los Angeles police and charged with 10 counts of murder and one count of attempted murder.
Genetic Justice Page 9
