Fingering Statistics On O.J.'s DNA

As lawyers in the O.J. Simpson murder trial prepare to debate the admissibility of DNA evidence, two Harvard population geneticists continue to challenge the validity of DNA fingerprinting.

Agassiz Professor of Zoology Richard C. Lewontin '50-'51 and Professor of Biology Daniel L. Hartl object to the forensic use of DNA finger-printing, or profiling. They contend that the accuracy claimed by current statistical techniques is overstated.

The O.J. Simpson case represents one of the most notable uses of DNA profiling, a technique used all over the country to convict criminals and clear innocent suspects. The statistical methods now being challenged by Harvard researchers are employed in thousands of cases each year.

If done correctly, DNA profiling has the potential to help unravel cases that would otherwise go unsolved.

Last Friday, Edward W. Honaker, a welder imprisoned for 10 years on a rape conviction, was freed after DNA profiling proved he could not have raped a 19 year old girl in 1984.

Honaker was represented by two lawyers from New York: Barry Scheck and Peter J. Neufeld. The Innocence Project, directed by Neufeld, has helped to free between 15 and 20 inmates around the country by using DNA evidence, according to the New York Times.

Simpson's defense has enlisted Scheck and Neufeld to help deal with the DNA evidence; however, the Los Angeles District Attorney's Office has Lisa Kahn on their staff. Kahn is a leading prosecution DNA expert.

Despite threats that he would bar DNA evidence because of the prosecution's delay in providing material to the defense, Judge Lance A. Ito has said that he will admit such evidence.

But "the battle is not over with" said Dr. Mark V. Bloom, assistant director of the DNA Learning Center at Cold Spring Harbor, New York. The defense will continue to fight against the DNA fingerprinting, because it could be some of the most incriminating evidence brought against Simpson.

The Science

DNA profiling is "a tool for exclusion," says George F. Sensabaugh, Jr., a professor of forensic science at the University of California at Berkeley. "If I am innocent and DNA evidence can prove it, I want testing done. [But], if I am guilty and DNA evidence will implicate me, I don't want it done."

"About 30 percent of suspects are excluded [from guilt] by DNA testing," says David H. Kaye, Regents professor at Arizona State University. "DNA profiling has resulted in many cases being resolved that would have gone unsolved."

The technique, also known as DNA typing or fingerprinting, is based on the fact that although the majority of DNA is the same in all humans, certain unique patterns exist in every person, except identical twins. DNA profiling distinguishes among samples by using these differences like a regular fingerprint. Experts have called the technique "the most significant breakthrough in forensic science since the development of fingerprinting."

The two methods of DNA profiling are restriction fragment length polymorphisms analysis (RFLP) and polymerase chain reaction analysis (PCR). RFLP, the more powerful test, analyzes specific locations of the DNA that differ widely across people (see graphics).

"RFLP methodology was first introduced and developed in England," says Paul B. Ferrara, director of the Virginia division of Forensic Sciences and chair of the American Society of Crime Laboratory Directors Accreditation Board. The test "gives rise to a virtual identification with a match at 5, 6, or 7 locations giving statistics showing us something like [only] one in 100 million" people could match.

"That method suffers, however, because it takes 8-10 weeks on average, requires a relatively large sample, and the DNA has to be in good shape," Ferrara says. "A relatively large sample would be a blood stain the size of a half-dollar or a seminal fluid spot about the size of a dime."

There is another limitation to the RFLP method. "Each additional probing adds four or five days. With each additional probing, you have to wash the gel," Ferrara says. "After the third or fourth probing, you start to lose too much DNA to maintain accuracy."

The catch-22 of the testing is that no one knows how many markers need to be tested. "No one has defined what that threshold might be," said Sensabaugh. "An ad hoc number like 8 to 10 provides the presumption of rebuttable proof of identification."

But Mark Stolorow, director of operations for Cellmark Diagnostics of Germantown, Md., says his firm generally uses five radioactive probes. "We can generally get at least 99.99 percent exclusion based on five probes," Stolorow says.

PCR analysis, using a technique whose inventors won the 1993 Nobel Prize, is a sort of molecular copying process that zeros in on and amplifies a particular region of DNA. The differences at that site can then be distinguished through specialized probes that tiger a color change or through gel electrophoresis followed by a silver stain.

"PCR is a quick process that requires very small samples but is not as specific, and contamination is more of a problem," Ferrara says. But, even "the saliva on the back of a postage stamp or on the butt of a cigarette has enough DNA to analyze."

However, the PCR method doesn't always provide numbers as small as RFLP.

The two methods can be used in concert to get even better probabilities, Ferrara says. "Let's say that you do the PCR tests and get numbers of 1 in 20 frequency. The RFLP then gives you, let's say, 1 in 100 thousand. Put them together, and you get 1 in 2 million."

The Controversy

Lewontin and Hartl have fueled controversy over the statistics, not the techniques, of DNA profiling. "The foundation of the molecular biology is well laid," says James F. Crow, professor emeritus of genetics at the University of Wisconsin. "The difficulties are matters of the interpretation" of the profiles.

The original statistical calculation used strict multiplication of probabilities.

A given part of a profile is compared to a representatives database of other profiles to determine the number of times that part would reoccur in the database.

Under the multiplication rule, if one probe gave odds of one in 20 and a second gave odds of one in five, the total odds reported were one in 100.

"Some population geneticists felt that strict multiplication was not a sufficiently valid and conservative method because the size of the databases was considered too small to justify such large numbers (e.g., a database of 100 profiles leading to a calculation of a frequency of 1 in 100 million)," Ferrara says.

The National Research Council (NRC), an arm of the National Academy of Sciences, published a report in 1992 on "DNA Technology in Forensic Science." The report developed a technique of interpretation called the "ceiling principle."

"The ceiling compromise puts a limiter on how we employ multiplication so the numbers aren't so big," says Ferrara, who does not "personally ascribe" to the principle.

Developed to make the probabilities more conservative, this principle, too, has been controversial--no statisticians or population geneticists were on the committee that created it.

Hard says that the debate over the ceiling principle is "nonsensical. All statistics is based on numbers picked out of the air."

Instead, Hartl argues that the use of databases created with profiles of people representing only certain races is inadequate. "There are clear differences within races; however, there are also differences in subgroups of the races."

Hartl contends that using specific databases representing certain racial groups, as recommended by the NRC report, biases the numbers in incorrect ways. Hartl feels strongly that it is appropriate to leave out race.

Lewontin and Hartl say in their 1991 paper that the use of racially-biassed reference databases to calculate frequencies "is liable to potentially serious errors because ethnic subgroups within major racial categories exhibit genetic differences."

But Stolorow claims that the proof exists to invalidate Lewontin and Hartl's concerns. "The amount of population substructure is negligible," Stolorow says.

And Stolorow's claim is backed up by the scientific literature.

Keith L. Monson and Bruce Budowle of the FBI's Forensic Science Research and Training Center agree that there is a lack of substructures in the four main racial groups. "For forensic purposes, profile frequency estimates from different reference populations do not deviate greatly," the researchers report in the Journal of Forensic Science.

In what Lewontin and Hartl see as "an attempt to nullify" their analysis, Daniel E. Koshland, Jr., editor-in-chief of Science, wrote in a recent editorial: "Acceptance of the validity of DNA evidence is exactly what most scientists in this area have believed appropriate."

Koshland says the judicial process has been "so slow to accept DNA evidence by failing to see that a couple of outspoken individuals were less representative of the scientific community than the vast majority of careful scholars."

When considering whether to admit scientific evidence, the courts rely on the 1923 Fryetest--requiring litigants to show that scientific techniques they use are generally accepted in the scientific community.

This is the precedent that the Massachusetts courts used to deny admittance of DNA evidence in Commonwealth vs. Lonigan and Commonwealth vs. Daggett.

"Massachusetts is one of the very few states that have reacted seriously to debate over the statistics," Stolorow says.

Recently, however, the Massachusetts Supreme Court has overturned lower courts' decisions and accepted DNA evidence.

As of now, however, there has been no precedent set in U.S. Federal Courts.

The Future

The recent Crime Bill takes heed of many of the NRC recommendations in the "DNA Identification Act of 1994." The Crime Bill authorizes grants (part of a $20 million package through facial year 2000) to public laboratories that apply and meet quality standards for forensic testing.

Another effect of the Crime Bill will be the establishment of a DNA index of the records of persons convicted of crimes, analyses of samples from crime scenes and from unidentified human remains. Currently, a national DNA identification index, the Combined Identification Index System (CODIS), is being developed by the FBI with participation from 10 states.

The index will be particularly helpful in catching repeat offenders, Ferrara said.

But Sensabaugh cautions that a DNA database of all individuals might not be a wise idea. He worries that the database could be used for new forms of genetic discrimination. "I don't want one, and you probably don't either," Sensabaugh says.

DNA profiling is the technique used to determine predispositions for cancer, Down's Syndrome and a variety of other diseases. But Bloom doesn't foresee this type of a database in the near future.

"The loci of medical interest are where the DNA should be the same between everyone, only changing with disease," Bloom says. "Police are looking for the ones that change like mad, the parts that are known as 'junk DNA' by most scientists. The other part of the answer is that it's not cost effective to do millions of tests."

In fact, for paternity cases, Baird says that LifeCodes, a Stamford, Conn. lab, charges $200 per sample and $535 for forensic DNA analysis including RFLP, PCR, or both.

Kaye says that there is a big impact on the number of guilty pleas with DNA profiling. Also, Bloom says, "Eyewitnesses are not accurate. Eyewitnesses have testified that a given individual was the criminal and DNA tests have proven that individual could not have been the perpetrator."

But Ferrara cautions that DNA profiling is not the only method that should be used to convict criminals.

"DNA evidence should not be looked at in isolation," Ferrara says. "It should be looked at with means, mode, motive and opportunity. DNA adds useful incriminating evidence."

"DNA testing alone is not meant to resolve innocence or guilt," Stolorow says. "It is the litigants job to prove that to a judge and jury."

Similarly, "lawyers understand that here is a remote possibility that some other gun fired a bullet," Bloom says. "But that realization doesn't stop them from using the ballistics report."

Bloom says that regardless of the accuracy of DNA profiling, lawyers will always be able to dispute it.

"Errors lead to mistakes," Bloom says. "Manipulation and the chain-of-custody can get fouled up along the way."

But Koshland has come up with effective alternatives to even that problem. "It may soon be possible to put a DNA sample in an automated machine with both defense and prosecution acting as witnesses to the procedure," he writes in his Science editorial.

"The real question is, aside from all other evidence, what is a good enough probability to convince a jury on just a DNA profile," Ferrara says. "If I thought that the chance of it being someone else was one in a million, I could live with it."

Sensabaugh does not, however, foreser DNA profiling becoming the method of choice for identification. "I don't see a point where genetic identification will become more common that fingerprints," Sensabaugh said.

Next month, many of the questions still surrounding DNA profiling will be addressed when the second NRC committee meets for three days to discuss their new report. "There will be a public meeting, for at least a day, where a large number of people will come and speak," says Crow. The second report, "DNA Technology in Forensic Science: An Update," is expected to be finished by next summer.

"There is an irony in this new acceptance of DNA fingerprinting," Koshland says in his editorial. "Ink fingerprinting went through the same type of debate, with questions about whether more than one person could have the same print, whether there could be abuse by police, whether there would be care in sample taking, and so on."Probe and Run an Autoradlograph The membrane is soaked in a solution containing microscopic radioactive probes, which bind to speclife parts of the DNA molecule. The strands are then exposed on X-ray him. When the firm is developed, the strand with the probe attached shows up on the firm, leaving a mark that can be compared with samples in existing databases.

Despite threats that he would bar DNA evidence because of the prosecution's delay in providing material to the defense, Judge Lance A. Ito has said that he will admit such evidence.

But "the battle is not over with" said Dr. Mark V. Bloom, assistant director of the DNA Learning Center at Cold Spring Harbor, New York. The defense will continue to fight against the DNA fingerprinting, because it could be some of the most incriminating evidence brought against Simpson.

The Science

DNA profiling is "a tool for exclusion," says George F. Sensabaugh, Jr., a professor of forensic science at the University of California at Berkeley. "If I am innocent and DNA evidence can prove it, I want testing done. [But], if I am guilty and DNA evidence will implicate me, I don't want it done."

"About 30 percent of suspects are excluded [from guilt] by DNA testing," says David H. Kaye, Regents professor at Arizona State University. "DNA profiling has resulted in many cases being resolved that would have gone unsolved."

The technique, also known as DNA typing or fingerprinting, is based on the fact that although the majority of DNA is the same in all humans, certain unique patterns exist in every person, except identical twins. DNA profiling distinguishes among samples by using these differences like a regular fingerprint. Experts have called the technique "the most significant breakthrough in forensic science since the development of fingerprinting."

The two methods of DNA profiling are restriction fragment length polymorphisms analysis (RFLP) and polymerase chain reaction analysis (PCR). RFLP, the more powerful test, analyzes specific locations of the DNA that differ widely across people (see graphics).

"RFLP methodology was first introduced and developed in England," says Paul B. Ferrara, director of the Virginia division of Forensic Sciences and chair of the American Society of Crime Laboratory Directors Accreditation Board. The test "gives rise to a virtual identification with a match at 5, 6, or 7 locations giving statistics showing us something like [only] one in 100 million" people could match.

"That method suffers, however, because it takes 8-10 weeks on average, requires a relatively large sample, and the DNA has to be in good shape," Ferrara says. "A relatively large sample would be a blood stain the size of a half-dollar or a seminal fluid spot about the size of a dime."

There is another limitation to the RFLP method. "Each additional probing adds four or five days. With each additional probing, you have to wash the gel," Ferrara says. "After the third or fourth probing, you start to lose too much DNA to maintain accuracy."

The catch-22 of the testing is that no one knows how many markers need to be tested. "No one has defined what that threshold might be," said Sensabaugh. "An ad hoc number like 8 to 10 provides the presumption of rebuttable proof of identification."

But Mark Stolorow, director of operations for Cellmark Diagnostics of Germantown, Md., says his firm generally uses five radioactive probes. "We can generally get at least 99.99 percent exclusion based on five probes," Stolorow says.

PCR analysis, using a technique whose inventors won the 1993 Nobel Prize, is a sort of molecular copying process that zeros in on and amplifies a particular region of DNA. The differences at that site can then be distinguished through specialized probes that tiger a color change or through gel electrophoresis followed by a silver stain.

"PCR is a quick process that requires very small samples but is not as specific, and contamination is more of a problem," Ferrara says. But, even "the saliva on the back of a postage stamp or on the butt of a cigarette has enough DNA to analyze."

However, the PCR method doesn't always provide numbers as small as RFLP.

The two methods can be used in concert to get even better probabilities, Ferrara says. "Let's say that you do the PCR tests and get numbers of 1 in 20 frequency. The RFLP then gives you, let's say, 1 in 100 thousand. Put them together, and you get 1 in 2 million."

The Controversy

Lewontin and Hartl have fueled controversy over the statistics, not the techniques, of DNA profiling. "The foundation of the molecular biology is well laid," says James F. Crow, professor emeritus of genetics at the University of Wisconsin. "The difficulties are matters of the interpretation" of the profiles.

The original statistical calculation used strict multiplication of probabilities.

A given part of a profile is compared to a representatives database of other profiles to determine the number of times that part would reoccur in the database.

Under the multiplication rule, if one probe gave odds of one in 20 and a second gave odds of one in five, the total odds reported were one in 100.

"Some population geneticists felt that strict multiplication was not a sufficiently valid and conservative method because the size of the databases was considered too small to justify such large numbers (e.g., a database of 100 profiles leading to a calculation of a frequency of 1 in 100 million)," Ferrara says.

The National Research Council (NRC), an arm of the National Academy of Sciences, published a report in 1992 on "DNA Technology in Forensic Science." The report developed a technique of interpretation called the "ceiling principle."

"The ceiling compromise puts a limiter on how we employ multiplication so the numbers aren't so big," says Ferrara, who does not "personally ascribe" to the principle.

Developed to make the probabilities more conservative, this principle, too, has been controversial--no statisticians or population geneticists were on the committee that created it.

Hard says that the debate over the ceiling principle is "nonsensical. All statistics is based on numbers picked out of the air."

Instead, Hartl argues that the use of databases created with profiles of people representing only certain races is inadequate. "There are clear differences within races; however, there are also differences in subgroups of the races."

Hartl contends that using specific databases representing certain racial groups, as recommended by the NRC report, biases the numbers in incorrect ways. Hartl feels strongly that it is appropriate to leave out race.

Lewontin and Hartl say in their 1991 paper that the use of racially-biassed reference databases to calculate frequencies "is liable to potentially serious errors because ethnic subgroups within major racial categories exhibit genetic differences."

But Stolorow claims that the proof exists to invalidate Lewontin and Hartl's concerns. "The amount of population substructure is negligible," Stolorow says.

And Stolorow's claim is backed up by the scientific literature.

Keith L. Monson and Bruce Budowle of the FBI's Forensic Science Research and Training Center agree that there is a lack of substructures in the four main racial groups. "For forensic purposes, profile frequency estimates from different reference populations do not deviate greatly," the researchers report in the Journal of Forensic Science.

In what Lewontin and Hartl see as "an attempt to nullify" their analysis, Daniel E. Koshland, Jr., editor-in-chief of Science, wrote in a recent editorial: "Acceptance of the validity of DNA evidence is exactly what most scientists in this area have believed appropriate."

Koshland says the judicial process has been "so slow to accept DNA evidence by failing to see that a couple of outspoken individuals were less representative of the scientific community than the vast majority of careful scholars."

When considering whether to admit scientific evidence, the courts rely on the 1923 Fryetest--requiring litigants to show that scientific techniques they use are generally accepted in the scientific community.

This is the precedent that the Massachusetts courts used to deny admittance of DNA evidence in Commonwealth vs. Lonigan and Commonwealth vs. Daggett.

"Massachusetts is one of the very few states that have reacted seriously to debate over the statistics," Stolorow says.

Recently, however, the Massachusetts Supreme Court has overturned lower courts' decisions and accepted DNA evidence.

As of now, however, there has been no precedent set in U.S. Federal Courts.

The Future

The recent Crime Bill takes heed of many of the NRC recommendations in the "DNA Identification Act of 1994." The Crime Bill authorizes grants (part of a $20 million package through facial year 2000) to public laboratories that apply and meet quality standards for forensic testing.

Another effect of the Crime Bill will be the establishment of a DNA index of the records of persons convicted of crimes, analyses of samples from crime scenes and from unidentified human remains. Currently, a national DNA identification index, the Combined Identification Index System (CODIS), is being developed by the FBI with participation from 10 states.

The index will be particularly helpful in catching repeat offenders, Ferrara said.

But Sensabaugh cautions that a DNA database of all individuals might not be a wise idea. He worries that the database could be used for new forms of genetic discrimination. "I don't want one, and you probably don't either," Sensabaugh says.

DNA profiling is the technique used to determine predispositions for cancer, Down's Syndrome and a variety of other diseases. But Bloom doesn't foresee this type of a database in the near future.

"The loci of medical interest are where the DNA should be the same between everyone, only changing with disease," Bloom says. "Police are looking for the ones that change like mad, the parts that are known as 'junk DNA' by most scientists. The other part of the answer is that it's not cost effective to do millions of tests."

In fact, for paternity cases, Baird says that LifeCodes, a Stamford, Conn. lab, charges $200 per sample and $535 for forensic DNA analysis including RFLP, PCR, or both.

Kaye says that there is a big impact on the number of guilty pleas with DNA profiling. Also, Bloom says, "Eyewitnesses are not accurate. Eyewitnesses have testified that a given individual was the criminal and DNA tests have proven that individual could not have been the perpetrator."

But Ferrara cautions that DNA profiling is not the only method that should be used to convict criminals.

"DNA evidence should not be looked at in isolation," Ferrara says. "It should be looked at with means, mode, motive and opportunity. DNA adds useful incriminating evidence."

"DNA testing alone is not meant to resolve innocence or guilt," Stolorow says. "It is the litigants job to prove that to a judge and jury."

Similarly, "lawyers understand that here is a remote possibility that some other gun fired a bullet," Bloom says. "But that realization doesn't stop them from using the ballistics report."

Bloom says that regardless of the accuracy of DNA profiling, lawyers will always be able to dispute it.

"Errors lead to mistakes," Bloom says. "Manipulation and the chain-of-custody can get fouled up along the way."

But Koshland has come up with effective alternatives to even that problem. "It may soon be possible to put a DNA sample in an automated machine with both defense and prosecution acting as witnesses to the procedure," he writes in his Science editorial.

"The real question is, aside from all other evidence, what is a good enough probability to convince a jury on just a DNA profile," Ferrara says. "If I thought that the chance of it being someone else was one in a million, I could live with it."

Sensabaugh does not, however, foreser DNA profiling becoming the method of choice for identification. "I don't see a point where genetic identification will become more common that fingerprints," Sensabaugh said.

Next month, many of the questions still surrounding DNA profiling will be addressed when the second NRC committee meets for three days to discuss their new report. "There will be a public meeting, for at least a day, where a large number of people will come and speak," says Crow. The second report, "DNA Technology in Forensic Science: An Update," is expected to be finished by next summer.

"There is an irony in this new acceptance of DNA fingerprinting," Koshland says in his editorial. "Ink fingerprinting went through the same type of debate, with questions about whether more than one person could have the same print, whether there could be abuse by police, whether there would be care in sample taking, and so on."Probe and Run an Autoradlograph The membrane is soaked in a solution containing microscopic radioactive probes, which bind to speclife parts of the DNA molecule. The strands are then exposed on X-ray him. When the firm is developed, the strand with the probe attached shows up on the firm, leaving a mark that can be compared with samples in existing databases.