An Interview with DNA Forensics Authority Dr. Bruce Weir
William Neal Reynolds
Professor of Statistics and Genetics at North Carolina State University
The term "DNA
fingerprinting" was coined by British geneticist Alec Jeffreys
only ten years ago. Since that time, DNA forensics has become an
important tool in law enforcement. In some cases, the DNA tests have
helped convict suspects, while in others the tests have exonerated
suspects or overturned previous convictions. Recent high profile
court cases have put the spotlight on DNA forensics and created the
impression that there is a lack of agreement among the experts on the
reliability of this evidence. I spoke with Dr. Bruce S. Weir, an
expert witness for the prosecution in the O.J. Simpson case, about the
methods and controversies surrounding DNA evidence.
Q: First, let's begin with a basic question, what do
we mean when we say DNA
fingerprinting, or DNA
profiling as I prefer to call it, characterizes a small portion of
our DNA. It is a way of identifying the DNA content of an individual.
We think of fingerprints as being unique to the individual. We know
that even genetically identical twins have different fingerprints. In
contrast, DNA typing, because it uses a very small fraction of the
DNA, is certainly not unique.
Q: Let's imagine an hypothetical crime scene. When
the technical people arrive at the crime scene, what do they do, what
are they looking for?
A: The people at the crime scene look for biological
tissues, such as blood stains on the ground and semen stains on
clothing. They may take swabs from the victim's body if rape is
involved. It is also possible to extract DNA from exotic things like
the saliva on the back of a postage stamp. If it is a very old crime,
they may obtain DNA samples from a skeleton.
Q: Wouldn't a drop of blood or other specimen at a
crime scene be commingled with all kinds of other DNA from bacteria,
flora, etc.? How is this extraneous DNA excluded?
A: The methods used in forensic DNA are not so much
concerned with excluding extraneous DNA as with the identification of
human DNA. The probes used are very specific for human DNA.
Q: Tell us a bit about the techniques used in
forensic DNA testing. Are the laboratories using PCR to look at
allele combinations or RFLP (restriction fragment length
polymorphisms) or both? What are the advantages of the different
A: The laboratories use both techniques. The RFLP
tests were developed first and are better in the sense that they are
much more variable. There are many more variants in the population.
A typical RFLP will have 30 distinguishable types. The disadvantage
of the RFLP method is that it requires a relatively large amount of
DNA along with radioactive labeling visualized on a gel. To expose an
X-ray film with a radioactive-labeled probe takes about a week. You
have to do that for each system, so the entire process can take many
techniques offer the advantage of requiring only trace amounts of DNA,
and they can be done overnight. Unfortunately, PCR
systems currently in use are not very variable, allowing typically
only three of four variants.
Q: When laboratories use RFLP they are only looking
for five or six polymorphisms. Why not look for ten or 100 or 1,000?
A: The state of North Carolina uses eight
polymorphisms. I believe by the time you have eight systems even with
as few as 20 alleles, you are not going to gain anything by looking
Q: A number of statistical techniques are now used to
confirm the reliability of these methods. Can you explain how the
most fundamental technique, the allele multiplication rule, is used to
rate the odds of a match between specimens?
A: If we want to come up with a figure for the
frequency of the pattern, we rely on a statistical model. A DNA profile contains
information from, let's say, five RFLPS or five PCRs.
This provides at least ten pieces of information. The entire profile
has almost never been seen. Out of the 30,000 people who have been
typed and their profiles put into a database, no two people have had
the same profile with five loci. While the entire profile has not
been seen, each component has been seen quite often. Let's say each
variant at each locus occurs about ten percent of the time. So say we
have a ten band profile based on five loci, with each band occurring
ten percent of the time. We then multiply the ten numbers together to
gain a probability for that combination.
Q: It has been suggested that some factors could
affect the accuracy of that kind of probability calculation. For
example, some alleles are probably more likely to be seen in a given
racial or ethnic group. This led to a concept called the ceiling
principle. Can you explain this concept?
A: The ceiling principle was a method designed to be
conservative in estimating probabilities, providing a frequency which
would not overstate the strength of the evidence. The DNA databases
used are designed to be representative of the entire population. If a
crime was committed in a particular area, and the suspect in a case
belonged to a specific population group, the frequency for a certain
allele might be higher than for the population as a whole. So it
would be prejudicial to the defendant to quote the population-wide
frequencies instead of the specific ones.
The ideal would be to have data bases tailored to every crime. That
is not practical, mainly because the population groups are not well
defined. So we are pretty much obliged to use the wider population
samples. There are ways to characterize the variation in allele
frequency across population subgroups using a statistical method
developed by population geneticists in the early 1950's. We really
would like to have samples from subgroups of the population to be able
to estimate how much they differ one from another. Because these
subgroups are ill-defined we can't sample them. So we fall back on
what is available, geographic sampling. The FBI has samples from
different states and has also compiled a worldwide survey derived from
data bases compiled by forensic scientists around the world. So we
can compare the frequency of alleles in different recognizable
We have found that any particular allele can have a frequency that
differs significantly from one population to another. It is the
frequency which differs, not whether that allele occurs in a given
population. However, the variants occur within all populations. This
leads us to believe that those variations occurred before the
divergence of the various human population groups. So each individual
allele frequency varies depending on the population, but when we take
a collection of say ten alleles, the ups and downs tend to cancel out.
We find that there is not really a great deal of difference in the
profiles we have seen from one group to another. When we modify the
product rule appropriately by measures of population substructure,
then the ceiling principle is inappropriate. I think the ceiling
principle is poor science and I don't think it will be used in the
Q: Doesn't it seem that human error involved in the
laboratory work would be the weak link in the chain of DNA forensics?
A: There has been a lot of discussion about the
potential for human error. I would think the weak link would be right
at the beginning, for example, does the tube labeled 'crime scene
blood stain' reflect the true source of that material? The forensic
laboratories have a lot of safeguards built in, such as dual
observation of each step, and signing for custody of the evidence.
Forensic laboratories have a lot of experience in taking care of
evidence. But I take your point, if there is going to be an error, it
would be of the gross human kind, rather than in technique.
Q: Modern science in general relies on publication of
data in peer-reviewed journals, sharing raw data with other
researchers to confirm conclusions, etc. One criticism of forensic
DNA profiling as opposed to genetic susceptibility testing, is that
the methodologies have not passed through the usual channels of peer
review and comment.
A: In the early days that was true. Some people were
very jealous about these data. However, the data are now routinely
made available to researchers. There is now an extensive bibliography
of studies available. The FBI, Lifecodes, Cellmark, and other
laboratories have all published peer-reviewed scientific papers
explaining their protocols and methods of analysis. When they do
publish, they are then obligated to make the raw data available.
Q: Several laboratories have established proprietary
techniques in DNA
profiling. Is one RFLP the same as the next? What are the
differences between the companies' methods?
A: The RFLPs do differ from one laboratory to the
next, but the PCRs do not. Because the RFLPs are so variable, it is not trivial to
distinguish one variant from another. As a result, ad hoc methods are
used to accomplish this. We are talking about variations in length of
the regions. The regions examined may vary in length from 500 base
pairs up to 20,000 base pairs. They differ in the multiples of repeat
units of about ten. So if we are going to go from 500 to 20,000 in
sets of ten, we are talking about thousands of types, far too many to
distinguish on current gels. So binning strategies have been
developed to amalgamate alleles that are close together. Different
laboratories use different binning strategies.
Q: So it sounds like the ultimate in sensitivity and
specificity in DNA
profiling would be a combination of RFLP plus a PCR?
A: Actually the ideal would be to use sequence data,
which is being used in some contexts. The Armed Forces Institute of
Pathology, for example, uses PCR mitochondrial sequence data to identify war remains. This method can
make very specific identification using only a few hundred base pairs.
Q: It seems straightforward enough from the
scientific perspective what these tests are measuring and that they
offer high levels of reliability and accuracy. So why does there
appear to be so much debate on the validity of DNA forensics and the
statistical accuracy of DNA profiling?
A: There is a perception among the public that there
is some debate within the scientific community. I believe that there
is no such debate. When we look at the scientific literature, which
is where science is discussed, the published, peer-reviewed papers are
overwhelmingly in favor of this technology and the protocols and
analytic methods used. Much of the current debate has been outside of
the scientific literature. It typically comes from court cases, where
someone is on trial for a crime. The defendant and prosecutors each
have expert witnesses, so it looks likes half of the scientists are on
one side and half on the other. This is quite misleading and it is
not the way science operates. There are very few people who have
thought about and examined the issues carefully who remain critical of
Go to next story: Use of DNA in Identification