Magazine section: Focus

DNA fingerprinting back in the dock: The case of a man cleared of armed robbery has badly dented the credibility of genetic fingerprinting in British courts DNA fingerprinting back in the dock: The case of a man cleared of armed robbery has badly dented the credibility of genetic fingerprinting in British courts

New Scientist vol 137 issue 1863 - 06 March 93, page 14

On 15 March 1991, William and Ellen Gardner were driving to Romford, Essex, to go shopping with their daughter and two grandchildren. As they drove up Shepherds Hill, a winding road cutting through farmland, they saw three men stumbling from an upturned BMW. Gardner stopped the car to help. Minutes later he was looking down the barrel of a pump-action shotgun and his daughter was lying face down on the ground.

The Gardners had stumbled upon an armed robbery gone wrong. It took the police nearly two years to bring a suspect, Terence Hammond, to court. Hammond denied any involvement in the robbery or the assault on the Gardners. And at his trial, defence lawyers argued that the DNA fingerprinting techniques used by forensic scientists to connect Hammond to the robbery were unreliable. The judge, Recorder Alan Rawley, agreed. He cleared Hammond, and cast doubt on whether the Crown Prosecution Service (CPS) would be able to use DNA fingerprinting - or DNA profiling - in future. The ruling also opened the way for a flood of appeals from people convicted on DNA evidence in the past.

Unlike real fingerprints, DNA profiles are not unique to individuals. As part of their case, prosecution scientists used techniques that produced odds of millions to one against another person chosen at random having the same profile as Hammond. New Scientist has found that these methods are widely used in Britain, Europe, Australia and New Zealand, despite the belief of many independent and police scientists that the evidence they produce is inconclusive.

The robbery had taken place at an industrial estate about a mile from Shepherds Hill. The crooks snatched 100 000 from a security van and made their getaway in the BMW. The driver went too fast along Shepherds Hill and turned the car over. Two men walked away apparently uninjured, but the face of the third was covered in blood. As he climbed from the car, he left blood on his sunglasses, a door handle and a leaf. The crooks forced the Gardners from their car, jumped in and drove off.

The spots of blood gave forensic scientists a chance to track down at least one of the robbers. In Britain, for the past seven years, DNA profiling has been used widely to link suspects to blood or tissue found at crime scenes. Scientists use special enzymes to cut DNA from the samples into fragments, the lengths of which vary from person to person, and place a few millilitres of the resulting broth on an agarose gel. An electric field applied across the gel sorts the fragments by length - the shortest 'migrating' farthest up the gel.

Forensic specialists use radioactive probes that attach to specific DNA sequences to locate the precise position of particular fragments on the gel. These appear as black bands on a photographic plate. By comparing the positions of the same band on two such profiles, a scientist can judge whether the two come from the same person. Because people inherit two copies of every chromosome, a probe will generally create two bands. And to increase the discriminatory power of the test, scientists use four different probes, giving them eight bands on which to base their case.

The blood samples from the BMW were sent to Geoffrey Roe at the Metropolitan Police Forensic Science Laboratory (MPFSL), who made profiles from them. He concluded they came from the same person.

On 8 June 1992, Detective Constable Michael Geaghan of the flying squad picked up Hammond and sent Roe a hair sample for profiling. Roe compared the two sets of bands and told Geaghan that he had a match. Armed with this evidence and an alleged identification of Hammond by the Gardners, Geaghan charged Hammond with armed robbery.

But the match between Hammond's DNA profile and that from blood in the BMW was not perfect. The MPFSL uses a system in which two bands from separate samples are said to match if they are positioned within one millimetre of each other. Six of Hammond's bands matched, one band was just inside the limit and another was clearly outside. In the US, the FBI deems such mismatches inconclusive. But in Britain, forensic scientists are allowed to use their judgment to call the profiles a match.

Roe's next step was to calculate the odds of somebody else having the same profile as Hammond. Usually, this is done by a process called match-binning. The MPFSL has a database of around 2000 DNA profiles from which it estimates the likelihood of finding a band that matches one of the suspect's bands. Typically, this provides odds of 30 to 1 for a single band. But by multiplying together the odds produced by eight bands - a technique known as the product rule - scientists can generate minuscule odds of hundreds of millions to 1 that someone else could have the same profile.

However, in cases where not all eight bands match, MPFSL policy is to use an alternative method to calculate the odds: a relatively new statistical technique devised by Ian Evett, a Home Office forensic scientist. Although it still relies on the DNA database, Evett's method does not make a sharp distinction between bands from separate samples that match and those that do not. Instead, the odds vary continuously with bands that are close together generating the lowest odds, and bands that are far apart the highest. With this method, Roe calculated odds of 65 million to 1 that somebody would share the same profile as Hammond. But the MPSFL only allows scientists to proffer odds of 10 million to 1: this is what Roe wrote in his report.

Before the trial, defence barristers Robin Simpson and Mark Graffius received a copy of Roe's report. 'We knew that DNA evidence had not been tested in the UK courts in a thorough way,' says Graffius. 'A lot of people were reassured by the media hype, so many articles saying that it was infallible.' The lawyers hired Thomas Fedor to check Roe's work.

Fedor was an outsider, an American living in Britain, who had never worked for a prosecution laboratory, leaving him outside what Graffius calls Britain's 'forensic science mafia'. Fedor was also aware of controversies over DNA profiling that have sprung up in the US. Last year, two American researchers argued that the calculated odds against two people sharing the same profile could be far too low. Ethnic groups such as blacks and Caucasians, they said, are made up of multiple subpopulations in which certain genes may be a lot more common than theory predicts (New Scientist, Science, 4 January 1992).

Simpson and Graffius drew on two arguments that derive from this work. First, the MPFSL had not demonstrated that its DNA database was representative of the general population. In addition, the product rule could be criticised because it does not allow for the possibility that some genes - and hence bands - may be inherited together.

Fedor argued that the product rule should never have been applied because Hammond's profile was not a perfect match with that taken from blood in the BMW. 'You could not tell whether the mismatch in the bands was a rare experimental error or the result of a genuinely different profile,' says Fedor. He also pointed out that Evett's method had not been published in a scientific journal.

On 30 November last year, at a hearing without the jury, Simpson used these arguments in an attempt to convince Rawley that the DNA evidence should be ruled inadmissible. As the hearing progressed, Roe and the CPS barrister, Simon Wild, made a series of major concessions. For example, Wild decided to withdraw the odds calculated with Evett's technique. That decision forced Roe to recalculate the odds using conventional match-binning.

After a week in court, Rawley made his decision. He accepted criticisms of the MPFSL database and product rule, not because they were flawed but because there were 'two respectable bodies of opinion at loggerheads in the scientific world'. For these reasons, along with statements from the Gardners that the third crook's face had been 'obscured' by blood, Rawley ruled that neither the DNA evidence nor the identifications should be put before a jury. The CPS was forced to drop the charges against Hammond who was exonerated.

The ramifications of Rawley's ruling are not yet clear. It is not binding on other courts. And in three recent contested cases, one judge sided with Rawley and two disagreed with him, allowing DNA evidence to be presented to the jury. But, even in the cases where judges have disagreed with Rawley, they have been more circumspect about DNA evidence than in the past. In one of the cases, Lawrence Verney, the Recorder of London, told the jury that blood left at the scene of a crime 'could have come from the defendant. That is all that can be said'. It seems unlikely that there will be a definitive ruling on the validity of the product rule or the database, until a test case reaches the Court of Appeal.

However these issues are resolved, it will not clear up the confusion for people who are tied to crimes by profiles in which only six or seven bands match. Evett's technique, although now published, has only been used in Britain and has yet to be universally approved. At least one statistician, David Balding of Queen Mary and Westfield College in London, believes the method is not reliable enough for court.

Match-binning in these cases has many more critics. Roe's difficulties in the Hammond case were complicated by Fedor who pointed out that the Home Office forensic laboratory and the MPFSL use different criteria to decide if two bands from separate profiles match. The Home Office deems that two bands match if the DNA fragments they represent are within 1.4 per cent of their average length. Roe agreed that if Home Office criteria were applied to the Hammond case, only six rather than seven bands matched.

At the MPFSL, there are no rules over when and how to calculate odds in cases where one or two bands do not match. Roe preferred to multiply together the odds from the six clearly matched bands to produce odds of 500 000 to 1. In general, Roger Cook, Roe's boss at the laboratory, would multiply together the odds from all eight bands. In these cases some forensic specialists say it is reasonable to call two profiles a match even if one or two bands do not align, because they can tell if the mismatch is caused by experimental error.

Another MPFSL scientist, Matthew Greenhaugh, would not calculate any odds. Instead, he believes the profiles in such a case merely 'fail to exclude' the suspect. This view is shared by Victor McKusick, who last year chaired a US National Research Council investigation into DNA profiling.

Prosecution scientists often say that disputes over DNA profiling are minor and will never turn strong evidence into weak. But the Hammond case shows that, in cases where not all eight bands match, this is not true. And such cases are not rare. Cook estimates that they now probably account for up to 5 per cent of cases. It is likely that, in the past, when laboratory procedures were less exact, the proportion was higher.

Borderline cases give prosecution scientists a problem. If they exclude all of them, guilty people may walk free from court.

In the US, the FBI has reduced the problem of imperfect matches by laying down strict rules: if there is any mismatch between bands from separate samples, the evidence is deemed inconclusive. The FBI is able to have strict rules because its criterion for calling a match between bands from separate profiles is more generous than those used in Britain. It allows bands to differ by as much as 4 per cent of the length of the DNA fragments they represent.

Cook says the MPFSL is considering changing its definition of a match to one that is closer to the FBI's criterion. But that will do little to placate defendants who, over the past seven years, have heard that the odds against somebody else sharing their DNA profile are 10 million to 1.

Because prosecution scientists often do not specify in their submissions to court that some bands within a profile do not match, laywers representing people who have been convicted on six or seven-band matches may not be aware that the prosecution's case was contestable.