Å®ÉúСÊÓƵ

Editorial: Time to rethink DNA evidence

THE rape of a woman in Albuquerque, New Mexico in September 1988 has given rise to a court case which could strike at the core of the emerging science of DNA fingerprinting – the technique which, in principle, allows an individual to be identified by an analysis of their DNA.

A group of US scientists is using the case to mount a challenge to the application of the technique in legal cases, arguing that it is not as accurate as is claimed. They face the FBI, whose methods for matching semen and tissue samples with suspects using DNA fingerprinting could stand or fall on the judge’s decision. This is due at any time.

The case in question, in which charges have been brought by the state of New Mexico, is just one of thousands of rapes tried in American criminal courts. But the FBI has chosen this one, says prosecuting attorney Mark Burtner, to defend its pre-eminince at what US authorities now call DNA ‘profiling’.

Critics of the technique have already been able to convince a few judges in the US to reject the method because of doubts about how samples are prepared and read (New Å®ÉúСÊÓƵ, 31 March). However, hundreds of other courts have accepted it, many for cases involving the death penalty.

Now, scientists are mounting a second challenge, this time focusing on the population biology underlying the technique. This group is claiming that the FBI has got its sums and methodology wrong, and as a result is placing excessive confidence in the results obtained by the methods it uses.

The idea of identifying people by their DNA, whether in criminal cases or in matters of establishing paternity, was first proposed by Alec Jeffreys of the University of Leicester in Britain. Since 1987, courts in the US have accepted the technique in several hundred cases.

Two companies, Lifecodes in New York and Cellmark in Maryland, dominate the business. Over the past two years, however, the FBI has been performing profiles for crime laboratories around the country and is expected to set the national standard.

In rape cases, semen or tissue suspected to belong to the assailant is recovered from the victim or from the scene of the crime. A DNA profile can be made from even small amounts of material. If a suspect is brought in, a sample of his blood will yield another DNA profile. The two are compared; if they match, the prosecution has a very powerful case against the suspect.

A match is only significant, however, if one knows how often two people’s profiles match by chance. To make that calculation, scientists must know how often a particular profile occurs in the population. The FBI has made such calculations, and says chance matches between two different people are generally in the range of one in several hundred thousand to one in a million.

However the FBI’s technique assumes that whites choose their partners at random and that the genes used for profiles are distributed evenly through the Caucasian population. ‘That is blatantly wrong,’ argues Charles Taylor, a population geneticist at the University of California, Los Angeles.

Taylor says the Caucasian population probably contains patches, or subpopulations, of people in whom certain genes are more common than the norm. If undetected, this patchiness ‘could throw the FBI’s statistics so far off as to render them meaningless’, says Taylor. Moreover, he adds, the assumption that whites mate randomly is ‘utterly invalid’.

Furthermore, the FBI bases its population genetics on studies of fewer than 300 of its own recruits. The recruits, says Taylor, are probably too few, and not genetically typical of the American population, to provide a good database. In an affidavit for the defence in the Albuquerque case, Taylor says no method accepted by biologists exists to calculate the true frequency of a DNA profile in a population.

Population statistics is an arcane outpost of biology visited most often by the mathematically inclined. Taylor and biologist Larry Mueller of the University of California at Irvine spent hours in the Albuquerque courtroom demystifying its subtleties for Judge Ross Sanchez, who must decide whether to admit the DNA evidence.

Briefly, a person has two copies, or alleles, of each gene. Within a given population, such as Caucasians, there may be as many as 100 variants of the alleles used for profiling. If all Caucasians mate randomly, and all 100 alleles are equally frequent, the chance of any one individual having a particular allele would be .01. Because people usually have two different alleles, the chance of a particular pairing occurring is .01 X .01, or .0001.

Furthermore, since a person inherits one band from the mother and another from the father, the statistical way of determining the frequency of this pairing, or ‘product rule’, is twice this, namely .0002. If this process is done with four probes, each result is mulitiplied by the next, making the likelihood of two people sharing the same four-probe genotype is infinitesimal.

The snag is the Hardy-Weinberg equilibrium, the theoretical model of the distribution of alleles that assumes a completely mixed population. People do not pick their partners randomly. Like may attract like, or people more often marry those living in close proximity. Over time, there occurs a ‘pooling’ of alleles in certain subpopulations. The sickle-cell anaemia gene among blacks or the Tay-Sachs gene among people of the Mediterranean are examples.

If a small subpopulation possessed an allele in 50 per cent of its members, rather than one per cent, the resulting genotype frequency could be off by a factor of 50. For four probes, the potential for error is 50 to the fourth power – potentially millions off the mark, says Mueller.

Mueller’s concern is based on more than numerical gamesmanship. The FBI has assumed the Hardy-Weinberg equilibrium in its population databases, from which it sets a standard for the frequency at which certain alleles occur. Mueller and his colleagues claim to have evidence that this equilibrium is a fantasy.

Mueller examined the FBI’s data on Hispanics, which was drawn from two groups: one from Miami, Florida, 70 per cent of whom are of Cuban lineage; and another from Houston, Texas, 88 per cent of whom are of Mexican lineage.

Looking at the alleles used by the FBI in DNA profiling, Mueller found that some alleles were much more common in one group than in another. One was four times more common. In fact, the genetic ‘distance’ between them was twice that seen in protein-coding genes in different races. ‘One (allele) . . . has a genetic distance which is approaching the human-chimpanzee level,’ says Mueller.

If all four alleles used in a DNA profile were four times as common as believed – a worst case hypothesis, Mueller acknowledges – a genetic profile could actually be many thousand times more common than what the FBI’s calculations would imply.

These findings, says Mueller, suggest that the same could be true among Caucasians. One explanation is that the small, repeated pieces of DNA that profilers use mutate unusually fast, as often as five per cent in each generation. Indeed, it is the variability of these points, or loci, on the genome that makes them good for distinguishing individuals. The substructures persist because groups do not mix and mate very much. Like a mode of dress or an unusual custom, a genetic trait becomes more common over time in an isolated group while becoming rarer elsewhere.

Evidence that people make predictable choices of partners is mostly anecdotal. Two scientists at Oxford University, however, studied marriages in San Francisco in 1980. They found that people much prefer familiarity. Of 402 grooms of English descent, 86 chose British brides, and another 119 chose other European mates. Only 38, fewer than 10 per cent, chose an Asian bride even though Asians made up 22 per cent of the city’s population. Similar patterns showed up among Latin American, Middle Eastern, Asian and black populations.

When subgroups occur in a population, their genetic pooling may also lead to a condition called linkage disequilibrium, which would further undercut the FBI’s statistics. Suppose obvious traits, such as hair and eye colour, are each determined by a single gene. These two genes reside on different chromosomes, so they are not physically linked.

Many Scandinavians, for example, have two copies of the blue-eye allele and two of the blond-hair allele, and are thus blue-eyed and blond. In the melting pot of the US, the instance of both traits should be no better than chance. Yet they are not, especially among individuals of Scandinavian descent, even though the two genes are not physically linked. This statistical assocation between the traits is linkage disequilibrium and the FBI’s techniques assume it does not occur.

What does all this mean for a criminal investigation? If, for example a rape victim in Providence, Rhode Island, a city in the east of the US with a high density of Portuguese immigrants, identifies her assailant only as Portuguese. A Portuguese fisherman is brought in and his DNA profile matches semen taken from the victim.

The chance of an accidental match based on a freely mixing white population is said to be several millions to one, says the FBI. But providence may not be what it seems; the alleles used for the matching may be thousands of times more common among the tightly knit, intermarrying population of Portuguese than the FBI’s Caucasian database.

Thomas Caskey, a geneticist at Baylor University, Texas, defends the FBI and says arguments over statistics amount to hairsplitting. ‘Even being extremely conservative, the numbers behind the match are still convincing,’ says Caskey. Some defence lawyers disagree. ‘Those sort of figures, in the millions, have a powerful effect on a judge or jury unschooled in statistics,’ argues William Thompson, a lawyer and statistician who is leading the challenge in the Albuquerque case and other cases.

‘Good statistical methods would not produce the huge numbers,’ says Seymour Geisser, director of the statistics department at the University of Minnesota. For example, the profiles determine how common an allele is based on its occurrence in a few hundred people; the FBI uses about 300 of its recruits. Because the data – the occurrence of alleles in a person’s genome – may not be statistically independent of each other, probabilities for accidental matches should not be more than about one in 300, says Geisser.

‘The technique is very good for determining nonmatches,’ says Geisser, ‘but I think they’ve come out too soon on this. I guess the commercial companies are trying to make a buck.’

The FBI argues that defence lawyers are, as their duty dictates, looking for any opening to raise doubt in the judge’s mind. ‘We have developed a system that is reliable,’ says Bruce Budowle, a geneticist from the FBI, ‘and the statistics do not (create) prejudice (against) the defendant unfairly.’

Budowle counters the critics point by point. He says data on allele frequencies among Canadian, French, and American Caucasians show no significant pooling. If the key profiling genes mutate by five per cent per generation, that has not yet shown up in differences between blacks and whites.

As for the Hispanic data, Budowle says even larger genetic distances within populations have been found before ‘and nobody batted an eye’. Besides, the FBI uses the higher allele frequency rate of the two groups. That way, the calculated commonness of an allele increases and knocks a zero or two off the chance of an accidental match, which favours the defendant.

Mueller’s arguments, such as his calculation that the frequency of a genotype could be many thousand times more common ‘are extreme’, says Budowle. Nor are the FBI’s 300 recruits a skewed database, he says. ‘It’s ludicrous. Is there a gene for law enforcement?’

The suspect in the Albuquerque case may serve time anyway; lawyers say there is plenty of other evidence besides his own DNA. Or the mountain of statistical testimony could amount to nothing more than a molehill. Budowle says: ‘If it (the chance of a mistake) is one in six million or one in 600 000, who cares?’ But the population biologists are taking their arguments on the road, and will testify in a murder case in Ohio this month. And if Judge Sanchez finds the statistics lacking, it could be a rude awakening for the technology the lawyers describe as ‘the prosecution’s dream and the defence’s nightmare’.

* * *

How genes reveal their secrets

IF TWO people have brown eyes, one can say confidently that they both have the copy of the eye-colour gene that codes for brown. But the fine structure of each gene will probably differ. Different forms of the same gene are called alleles. People doing DNA profiles use alleles of genes called VNTRs, which vary widely from person to person, making them good candidates for ‘prints’. The concept is similar to that of fingerprints, discovered by a British surgeon in a prison in India in the late 19th century.

Making a DNA fingerprint begins with the extraction of DNA from a sample. The DNA is broken into millions of fragments of different lengths; the number of pieces and the variety of lengths varies with each human, like noodles in soup. Into this soup, scientists add radioactively labelled probes that can seek out a specific gene.

The DNA noodle soup is then spread out on an electrophoretic gel. What looks like a smear is actually a ladder-like stack of fragments, each slightly smaller than the one above. The fragments cannot be seen with the naked eye, except for the one a radioactive probe attaches to. It shows as a dark blot, or band. Humans usually have two copies of each gene, so usually two bands appear. That is a DNA print for a particular gene.

Such a print from the suspect in Albuquerque was set beside a print made from semen and blood taken at the scene of rape. The bands in each were exactly the same size, appearing at the same point on the gel, and thus shared the same genotype.

To narrow the risk that the suspect and the guilty party share the same genotype by chance, people who perform DNA profiles repeat the process four times, each time with a probe for a different gene. The resulting print, with eight bands, is called a multilocus genotype. The chances of the genotype of two people being the same, say the FBI, range from hundreds of thousands to several million to one.