Professor Luigi Luca Cavalli-Sforza is a man whose eminence is matched only by his elegance. Erect of posture, even in his late seventies, his silver hair always immaculately groomed, he is equally at home in the busy conference rooms of the academic circuit by day and the exclusive restaurants that welcome the most distinguished delegates by night. His contributions and influence in the field cannot be over-estimated. Scientists who once studied under him, either in Italy or later at Stanford University in California, today hold many of the important academic positions in the discipline of human population genetics. It was Luca who first formulated the theory which had come to dominate European prehistory over the preceding quarter-century. According to this theory, or at least the version believed by archaeologists, farmers from the Near East had overwhelmed the descendants of the Cro-Magnons, who themselves had replaced the Neanderthals. This was a large-scale replacement which meant that most Europeans traced their ancestry back not to hunter–gatherers but to farmers.

Having collected together the records of thousands of blood and other genetic tests from all over Europe, Luca had amalgamated the results into a gradient of gene frequencies that summarized this mountain of data. These gradients were organized into simple vectors, called principal components, which were projected as lines on a map. The most striking, the first principal component, led diagonally across Europe from Anatolia in Turkey to Britain and Scandinavia in the north-west. To Luca and his colleagues, this was the signature of a massive influx of people into Europe from the Near East. The fit between the south-eastnorth-west axis of this genetic slope and the routes followed by the early farmers according to the archaeology available at the time was convincing. The farmers had overrun Europe.

The influence of Cavalli-Sforza’s conclusion spread far beyond the narrow bounds of human genetics, through archaeology and related disciplines. Although there were some archaeologists who did not share his conclusion and saw in the record signs of only minimal population movement, they had a hard time making their views heard. Like any academic discipline archaeology has its fashions, and the fashion in Europe was for a large-scale settlement by incoming farmers. It had not been so when Cavalli-Sforza and his colleague, the American archaeologist Albert Ammerman, first put forward their ideas in the 1970s. At that time the contemporary taste was for entirely indigenous development; for the gradual adoption of agricultural methods and practice by the mesolithic hunter–gatherers of Europe without a large-scale movement of people. The original argument put forward by Ammerman and Cavalli-Sforza was for at least some movement, some migration from the Near East. Launched in a hostile intellectual atmosphere, this process was described in a term which sounded unchallenging. It was called ‘demic diffusion’. Demic means ‘to do with people’, and diffusion is a gentle phrase implying the gradual inching outward of the farmers from their stronghold in the Near East. However, demic diffusion was not just a descriptive idea; it had a strong mathematical basis. It took as its foundation a mathematical model developed by Arthur Mourant’s mentor, the great statistical geneticist R. A. Fisher, who produced equations to describe the spread of anything – animals, people, genes, ideas – outwards from a growing centre. This mathematical model was given the dramatic title the ‘wave of advance’.

Over the past twenty-five years, the ‘wave of advance’, the name of the mathematical model, has gradually taken over from ‘demic diffusion’ as the description of the spread of farming. I don’t entirely understand the reasons for this. It may be that as the model became more widely accepted there was no need to present it in a tone conciliatory to the intellectual atmosphere into which it was introduced, resistant to any theories which suggested large-scale movements of people; or it may just have been that archaeologists were beguiled by the power of the phrase ‘wave of advance’. In any event, somehow the dramatic had taken over from the gentle. The idea of a gradual influence of incoming agriculturalists had been replaced in the collective psyche by the image of an unstoppable tidal wave of land-grabbing farmers that swept away everyone and everything in its path. The notion that the farmers overwhelmed the original inhabitants became the prevailing lore among archaeologists.

Not only had this tsunami of people brought agriculture to Europe, it was also responsible, according to the distinguished Cambridge archaeologist Colin Renfrew, for the introduction and dissemination of the language family to which most European languages belong. Although it is not readily apparent to any but professional linguists, there is no doubt that, with only a few exceptions, the languages spoken in Europe today all stem from a common root. They belong to a family of languages called Indo-European. The way in which sentences are constructed and many of the words they share betray a relationship among them that may not be obvious to most of us as we struggle with our phrase books. It takes a linguist to connect English and Portuguese, Greek and Gaelic. The exceptions are the Basques’ Euskara, Finnish, Estonian, Lapp and Hungarian. While Euskara is unique among living European languages and cannot be reliably linked to any other (though some linguists see a connection with languages of the Caucasus mountains), the other four are members of the Uralic language family which has its origins further east.

The Indo-element in Indo-European is there because there is a strong connection, again visible only to linguists, between the European languages and Sanskrit. This link was discovered by William Jones in 1786 while he was working as a judge in India for the British Raj. It was an amazing piece of amateur scholarship; indeed, Jones invented the concept of language families that is still a feature of comparative linguistics today. The essential idea of a language family is that all the different languages within it have evolved from a common root, almost certainly a language that is by now extinct. This raises the question of where the original Indo-European language was spoken and, importantly, how it spread out from there. Renfrew deduced that the original Indo-European language was spoken in Anatolia in central Turkey, and was then spread to Europe by the first farmers. A massive replacement of the hunter–gatherers by the agricultural ‘wave of advance’, as demic diffusion had surreptitiously become, was just what was needed to spread the language From its base in Anatolia.

There was now a powerful confederation of genetics, archaeology and linguistics in support of the argument that the mesolithic hunter–gatherers of Europe had been overwhelmed by the neolithic farmers. So, by the time we produced our startling results, the received wisdom was that most native Europeans today were descended not from the people who had endured the rigours of the last Ice Age but from the farmers who had walked in only ten thousand years ago with a bag of seeds and a few animals. But it just didn’t fit with the ages of our DNA clusters. We were sure that the strongest signals from the mitochondrial DNA in today’s Europeans were from much further back in the past than ten thousand years. We saw these signals as the genetic echo of the hunter–gatherers. These were not the faint whispers of a defeated and sidelined people but a resonant and loud declaration from our hunter–gatherer ancestors: ‘We are still here.’

I decided to present our work at the Second Euroconference on Population History, held in Barcelona in November 1995. I knew very well that the main proponents of the ‘wave of advance’ theory would be there, so at least what I had to say would be noticed. I was given a twenty-minute slot. The conference room was vast, with four hundred delegates and room for many more. I was introduced by the convenor, Sir Walter Bodmer, Fellow of the Royal Society, a long-time associate of Luca Cavalli-Sforza and co-author with him of two influential textbooks on genetics. Walter is not widely known for his conciliatory remarks, but I did think ‘And the next speaker is Bryan Sykes who is talking about mitochondria. I don’t believe in mitochondria’ was a less than gracious introduction. I began to lay out the basis for our revision of European prehistory.

Walter and Luca were both sitting below the podium, side by side in the front row. It is surprising how much you can take in when addressing even a large audience such as this. As I went from one point to another I could see that Walter was getting agitated. He began to mutter to himself, then to Luca; at first inaudibly, then louder and louder. ‘Rubbish,’ ‘Nonsense,’ I thought I heard him say. He began to fidget, to half raise himself in his seat then sit back down, as one slide followed another in my presentation. As I came to the concluding slide, I could almost see the steam coming out of his ears.

No sooner had I finished talking than Walter and Luca were on their feet, throwing questions at me. I have known Walter for ages and seen him in action many times. I have watched him crush young researchers by his aggressive questioning, and I was determined the same would not happen to me. There is only one effective remedy with Walter, and that is to argue back. I had been expecting fireworks, and as I stood there under this barrage, I began to see it all as a piece of theatre – like a cross-examination in the High Court or a fierce exchange at the Despatch Box in the House of Commons. I began to enjoy myself.

At one point Walter insisted that they (he and Luca) had never said that the farmers had overwhelmed Europe and replaced the hunter–gatherers. I had brought along a copy of their jointly written textbook Genetics, Evolution and Man against just such an assertion. In response, I opened it at a page I had already marked with a yellow sticker and read out: ‘If the population of Europe is largely composed of farmers who gradually immigrated from the Near East, the genes of the original Near Easterners were probably diluted out progressively with local genes as the farmers advanced westward. However, the density of hunter– gatherers was probably small and the dilution [of Near Eastern genes, that is] would thus be relatively modest.’ There it was in black and white, in their own words. This was massive replacement in all but name. Walter puffed one last time and sat down. The chairman closed the session. I had survived the first charge: but the fuse had been lit on a fierce debate that would not be resolved for another five years.

In science these days, international conferences like the one in Barcelona are useful for announcing new findings and getting an initial reaction. But work presented at a conference has no real validity until it is published in a scientific journal. Publication involves close scrutiny of the data, the methods and the conclusions by expert reviewers working unpaid and under an obligation to declare any conflict of interests. Though a conference presentation has to be truthful, it is only during the review process prior to publication that the assumptions, results and interpretations are thoroughly checked. Considering the fierce reaction that our radical revision of European prehistory had provoked in Barcelona, it came as no surprise to us when we submitted our manuscript to the American Journal of Human Genetics, the leading international journal in the field, that the reviewers were even more demanding than usual. They insisted that the evolutionary network method, which we had published in 1995 as an intensely mathematical and opaque article, be explained once more in an appendix. They asked for additional tables of, to my mind, old-fashioned population comparisons. But finally, they published it. ‘Palaeolithic and Neolithic lineages in the European mitochondrial gene pool’ appeared in the July 1996 issue. It was now in print. We had set out our stall; now we waited for the reaction.

For a while, nothing happened. Then we started hearing from friends that the work was being discussed as at best irrelevant or at worst just plain wrong. Surprisingly, the main target of the whispering campaign was not us but mitochondrial DNA itself, which had distinguished itself so well in solving the puzzle of the Polynesians. Suddenly it was portrayed as being unreliable, too unstable, with too many parallel mutations in the section that we had chosen to use. The mutation rate estimates were attacked as being wildly out. This meant that the dates for the clusters were much younger than we thought and thus perfectly compatible with the ‘wave of advance’ model of an essentially farming- derived gene pool. Lastly, mitochondrial DNA was accused of being just one marker, just a single witness to events whose account of prehistory could not be substantiated.

When a controversial paper is published it is not unusual for the scientific journal in which it appears to receive and publish a criticism from others in the field. This takes the form of a ‘Letter to the Editor’. The authors of the original paper are given the opportunity to respond, and if they do, both letters appear next to each other in the same issue of the journal. It was no surprise to learn that Cavalli-Sforza had composed such a criticism of our paper and that it had been accepted by the American Journal of Human Genetics. The editor sent us a copy of Luca’s letter with an invitation to reply to it.

The letter was a withering attack on mitochondria in general and on our interpretation of the control region sequence data in particular. It did, however, contain one very interesting statement that we had been waiting to hear. Although the overwhelming influence of the neolithic farmers on the make-up of the European gene pool was the main feature of Luca’s ‘demie diffusionwave of advance’ model, no figures had ever been put on their overall genetic contribution. While we had estimated that roughly 20 per cent of modern Europeans traced their mitochondrial ancestors back to these early agriculturalists, there was no comparable figure from Luca’s work that we could use as a contrast. The assumption which most people had made was that the farmers had ‘overwhelmed’ the hunters. That was certainly how a generation of archaeologists had interpreted the ‘wave of advance’ model. But the scale of the immigration had never been quantified. There was probably no need. The model had gained its own momentum and everybody knew what it meant, or thought they did. But now, for the first time, Luca put a figure on the proportion of modern European genes contributed by farmers from the Near East. It was, according to the letter, roughly equal to the proportion of the genetic variation that contributed to the first principal component which tracked the cline of genes across Europe from the south-east to the north-west. And this was 26 per cent. No mathematical proof whatsoever accompanied the statement, but we weren’t going to complain about that. It was close enough to our estimate of about 20 per cent, derived from mitochondrial DNA analysis, that it looked as if there was little left to argue about.

Even though this was an important new announcement from Luca, we certainly needed to reply to his letter and the criticism of mitochondrial DNA that it contained. He had every right to be critical. It is perfectly reasonable to demand absolute clarification from anybody who is challenging a long-held view. Extraordinary claims, such as ours, demand extraordinary proof. Even so, we all felt under a lot of pressure. We were the new boys on the block up against the might of the Establishment. Nevertheless, I never doubted for a second that we were right. There was nothing for it but to answer the criticisms one by one.

We were confident that the first objection – that our chosen section of mitochondrial DNA, the control region, was so riddled with parallel mutations as to be completely unreliable – could be rebuffed. There are plenty of other base changes that can be used as molecular markers around the mitochondrial DNA circle. If we drew a new evolutionary tree using these other markers instead of the control region sequences, then one of two things would happen: either the clusters would match our own groupings or they would not. If they did match, then the control region must be reliable. If they didn’t match, then it wasn’t, and we might as well give up.

For this test we teamed up with Antonio Torroni, an Italian geneticist from Rome who had spent many years developing an intricate technical system for these other markers. He supplied us with samples he had already tested for us to sequence through the control region, and we in turn took our own sequenced samples to Rome to run through his system. The results couldn’t have been more encouraging. There was an almost exact fit between the clusters identified by Antonio’s markers and our own. The one or two minor incompatibilities were quickly resolved; those apart, the match was perfect – so much so, in fact, that we abandoned our own numerical classification for the clusters and adopted Antonio’s, based on letters of the alphabet. Now we had proof that the control region was not after all a fickle piece of DNA that could mislead and deceive but, once you got to know it, a faithful and reliable companion.

The mutation rate criticism was harder to address. It was certainly true that if we were using a gross under-estimate of the mutation rate then our cluster dates would be seriously adrift. If our estimates were out by a factor of ten, as some people suggested, then the ages of our clusters would fall from the Palaeolithic into the Neolithic and we could kiss our theory goodbye.

There are basically two ways of estimating a mutation rate. Either you can try to measure it by direct observations from one generation to the next, or you can see how many mutations have accumulated in two different groups – which could be tribes, or populations, or species – that have been separated for a known length of time. The very first estimate of the mutation rate, the speed of the molecular clock, was made by comparing the differences between humans and their closest relative, chimpanzees, and combining this with the time since they last shared a common ancestor, estimated at between four and six million years ago. Of course, precisely when that separation between the ancestors of humans and chimpanzees took place is not known, especially since there are no chimp fossils to help out. The other route that has been used is to estimate the mutation rate changes which have accumulated in native Americans, who first arrived on the continent about twelve thousand years ago. The remarkable thing is that both methods agree so well with each other and come out with a figure of around one mutation in twenty thousand years down a single maternal lineage. When tracking back to a common ancestor between two modern people, as I did when estimating the date of the common ancestor between myself and the Tsar, there are two lineages, each with a chance to mutate, going forward from our common ancestor to each of us. Only one mutation separates my control region sequence from the Tsar’s, but that mutation could have happened anywhere along the two maternal lineages leading from our common ancestor. At a rate of one mutation every twenty thousand years along a single lineage, that fixes the combined length of these two lineages to twenty thousand years. Since the Tsar and I are more or less contemporaries, the length of each lineage back to the common ancestor is therefore halved to ten thousand years. Our work in Polynesia had also shown an excellent agreement between the genetic and archaeological dates for settlement using this mutation rate. If the rate was wrong by a factor of ten in Europe, then it had to be wrong everywhere else. It would mean that chimps and humans diverged only 400,000–600,000 years ago, America was first settled only 1,200 years ago and Polynesia only 300 years ago – in fact after the Europeans go! there. This was so obviously crazy that the rates we were using couldn’t be that far out.

Measuring mutation rates directly is a hard business. It means picking up a change between a mother and her child. We estimated that we would need to test a thousand pairs of parents and children to pick up a single new mutation. That was out of the question. Fortunately, the mutation process in mitochondria is a gradual one and, as it turned out, not too difficult to observe by a different route. Mutations happen in individual DNA molecules in individual mitochondria. However, in most people the DNA sequence of all the mitochondria in all the body cells is exactly the same. These two truths pose a paradox. A new mutation can only take place in one DNA molecule in one mitochondrion in one cell; so how does it manage to take over the whole body?

In order to be passed on to a new generation, a mutation has to occur in a female germline cell, one of the cells that divide to become eggs. Mutations also happen in other body cells – in skin, bone, blood, and so on – but, as these do not get passed on to the next generation, they play no part in the patterns of evolution. What seems to be happening is that each time a female germline cell divides it takes only a few mitochondria with it. If the mitochondrion with the new DNA mutation is one of the few to slip through this bottleneck then it can make up a much bigger proportion of the mitochondrial DNA in the new cells. When these cells divide there is a chance that the new mutation will be further enriched, and so on.

There are only twenty-four cell divisions in the female germline between one generation and the next. These are twenty-four opportunities for enrichment of a new mutation; only rarely is this enough for a complete takeover in a single generation. The individual who grows from the fertilized egg will have a mixture of two mitochondrial sequences: the old one, which is the same as her mother’s, and the new one, which began as a new mutation somewhere in her mother’s germline cells.

We looked very hard at our sequencing results over the past few years, searching for the signs of mixed mitochondria within the same person. We found that about 1.5 per cent of people do indeed have a mixture of two different mitochondrial DNAs. We then tracked these mixtures through families and found that it took an average of six generations for a new mutation to establish itself and take over completely. Remember the unusual case of the Tsar, who had a mixture of two different mitochondria in his bone cells? It looks as if he was in the transitional state where a new mutation was struggling to get established; eventually it did, as we can see in the cells of his modern-day relatives like Count Trubetskoy. As far as we could tell from our experiments, there was no inevitability in this process; some new mutations appeared to be doing well for one or two generations, then slipped back into obscurity and disappeared. We were observing directly the appearance and spread of new mutations, and from these data we could make a separate estimate of the mutation rate, independent of the complications associated with the exact dating of past events like the evolutionary separation of humans and chimps. This independent estimate, though only approximate, matched the mutation rate we had been using. We had answered the second criticism. Mitochondrial DNA had survived with its reputation intact.

The points Luca had raised in his letter, and to which we had responded, were serious and valid questions to ask of a new technology, especially one that had rewritten the version of prehistory that had dominated thinking for so long. They needed to be addressed, and they were. What happened next threatened to discredit not only our studies in Europe but all the evolutionary work using mitochondrial DNA that had ever been done on humans. We had to deal with the spectre of recombination.

Briefly, what makes the chromosomes in the cell nucleus so difficult to use for tracing evolutionary histories is their habit of scrambling information at each generation. Until the germline cells are into their final division which produces the gametes (sperm or eggs), the chromosomes lead separate lives and don’t have a great deal to do with one another. However, in that final cell division, the pairs of chromosomes which have been inherited from each parent sidle up to one another, like mating earthworms, and start to exchange bits of DNA. After this canoodling they pull apart and go off to different gametes. But now they are no longer the same chromosomes but DNA mosaics. They have undergone what is called recombination. This is the ultimate genetic reason for sex itself, the potential for creating through recombination new and better gene arrangements that can advance evolution.

Recombination has its advantages for scientists. It has greatly helped the mapping of genes for serious inherited diseases on to specific chromosomes, and has been instrumental in unravelling the sequence of the entire human genome. But as far as tracing DNA through the generations is concerned, recombination is a very big nuisance. One of the features of mitochondrial DNA that have made it such a successful instrument for probing into the deep human past is that the information it brings us is not scrambled by recombination. The only differences between my mitochondrial sequence and that of my direct maternal ancestors are the changes that have been introduced over the millennia by mutation. With recombination, there would be the prospect of having not just one line of mitochondrial ancestors but dozens of them. Everything that had been assumed about mitochondrial genetics would be in doubt.

So, when two papers claiming evidence for mitochondrial recombination appeared in the March 1999 issue of the prestigious Proceedings of the Royal Society, they sent shock waves around the world. Editorials in the leading popular science journals, Science in Washington and Nature in London, immediately publicized this fundamental challenge to the authority of mitochondrial DNA. If recombination really was occurring, as these papers were suggesting, then it meant that all the work published over the previous decade on mitochondrial DNA in human evolution was completely undermined.

The wide publicity accorded to these articles was due not only to the claims they advanced but also to the great distinction of the author of one of them: John Maynard Smith, the undisputed doyen of British evolutionary biologists, the author of textbooks and other influential works, and still an active presence in his eighties. Condemnation by such an eminent figure, with no obvious axe to grind, spelled obliteration for us and everybody else in the field – if the claims for recombination could be substantiated. The substance of Maynard Smith’s largely theoretical argument was that there was too much variation in mitochondrial DNA to have arisen by mutation alone. It was not so much a proof of recombination as an elimination of other mechanisms that could account for what Maynard Smith saw as a higher than predicted number of mutations. The reasoning was reminiscent of Sherlock Holmes’ advice to Dr Watson in The Sign of Four: ‘When you have eliminated the impossible, whatever remains, however improbable, is the truth.’ But what made Maynard Smith’s argument so seductive was the announcement in an adjoining paper of actual evidence for recombination in mitochondria from the tiny and remote island of Nguna in the Pacific. And the leading author (one of six) of the second paper was Erika Hagelberg.

Erika, you will recall, had worked in my laboratory on the first recovery of DNA from human bone back in the late 1980s. She had since made a name for herself in the field of ancient DNA and become involved in some celebrated forensic cases, most famously when she and her colleagues had recovered DNA from the remains of Joseph Mengele, the infamous Nazi doctor who carried out unspeakable human experiments on prisoners in the Auschwitz extermination camp. With these and other cases under her belt she had built up a reputation as an imaginative scientist. However, despite occasional attempts on both our parts to heal the rift that had grown up during the difficult final days Erika spent in my laboratory, she and I had endured an uneasy relationship ever since. This tension added an extra dimension to the drama that was about to unfold.

The essence of Erika’s evidence for recombination was that a particular mitochondrial mutation, at position 76 in the control region, was cropping up in several different clusters on the small island of Nguna. Like the Maynard Smith paper that accompanied it, this wasn’t direct evidence for mitochondrial recombination. However, mutations at position 76 were exceedingly ran; elsewhere in the world, so to find it frequently and in different clusters on the same island did deserve a special explanation. It would mean either that the mutation had happened spontaneously several different times in different clusters, which was extremely unlikely, or that a new mutation at 76 in one cluster had somehow spread to the others. And the only way for that to happen was by recombination.

For mitochondrial recombination to occur, two things have to happen. First, there needs to be a way for two circular mitochondrial DNA molecules to snuggle up to each other and exchange DNA. That didn’t seem too unlikely. There are about eight DNA molecules in each mitochondrion and they enjoy free access to each other. So it would not be hard for them to exchange DNA. More difficult to accept was that there had to be two very different mitochondrial genomes in the same cell. If all the mitochondria in the cell had exactly the same sequence, they could exchange DNA between themselves as much as they liked and it would not make any difference. All the mitochondria would still have the same DNA sequence. Only if there were two different mitochondria exchanging DNA would anything be noticed. So the Nguna observation demanded that there were, or had been in the past, people who had mixtures of mitochondria. One component of the mixture would have to be the DNA belonging to one cluster, let’s call it A, and with a mutation at position 76 in the control region. The other would be mitochondrial DNA from a completely different cluster, which we can call B, without the mutation at position 76. These two mitochondria w7ould then exchange segments of DNA so that a piece from A, which included the mutation at position 76, ended up on B.

There was only one way to get two mitochondria from completely different clusters in a cell: one of them had to be coming not from the egg but from the sperm. So, if this claim of recombination turned out to be true, it would be a lethal double blow. Not only would it be impossible to trace mitochondrial lineages back in time because of the scrambling implicit in recombination, but it would also follow that mitochondrial inheritance was not, after all, exclusively maternal. No longer would it be safe to assume that our mitochondrial DNA had come from an ancestral line of mothers. It could have come from fathers as well. Something had to be done. We held an emergency meeting.

Vincent Macaulay, who trained as a physicist and was a formidable mathematician, and had joined the team two years previously, went off to check and recheck the sequence data used in the Maynard Smith paper. Incredibly, a lot of them were wrong. Either they had been incorrectly copied from the public databases, or the raw sequences themselves which had been deposited in these databases had mistakes in them (actually a common enough occurrence). The cumulative effect of both sorts of error made it look as though there were more mutations in the mitochondria than there really were. After correcting these mistakes in the data and redoing the Maynard Smith calculations, it was obvious that the force of the theoretical argument for recombination was seriously diluted. We wrote at once to Maynard Smith, who gracefully accepted the error.

The claim for recombination advanced by Erika Hagelberg was a more serious proposition. Even though it fell short of an actual proof of recombination, which would require a definition of the segments that had been exchanged between the two different mitochondria, it was still a piece of evidence that was hard to explain by any other mechanism. As far as I could see, it could only be wrong if there had been a massive systematic error in the sequencing of the Nguna samples. This seemed very unlikely, given that Erika was an experienced scientist who would be familiar with the rule that extraordinary claims needed extraordinary proofs. Conventionally, these sequences would have been repeated and checked several times before making such a radical claim that she must have realized would have such profound implications.

Nguna itself is a tiny island lying off Espirito Santo in Vanuatu, west of Fiji, and Vanuatu was one of the island groups which we had included in our earlier work on Polynesia. We had been given a few samples and, checking back, I found that four of them came from Nguna itself. In those days we did not report mutations lower than position 93, because the systems we used at the time sometimes gave unreliable readings below that. So it was no surprise that our computer records showed no mutations at the crucial site of position 76. However, we still kept the old X-ray films on which the sequence was displayed as a series of bands. By some miracle I managed to locate the Nguna plate dated 2 June 1992, and the quality was perfect. I could easily read the sequence down to 76 and beyond. There was no sign of a change at 76 in any of the samples. I went at once to my colleague in the Institute who had supplied me with the original blood samples and explained what I had found. He had some more from Nguna, and we tested those for the change at 76. Not one of them had it. It seemed incredible that we couldn’t find the 76 mutation in twenty samples from such a tiny island when Erika was reporting it in nearly half of hers from the same place.

The situation was serious enough to warrant contacting Erika, and I emailed her in Dunedin, New Zealand, where she had recently taken up a post at the University of Otago. Given our strained relationship, I was as diplomatic as possible and stuck to the point. I explained that we had found no sign of the crucial mutation at position 76 in samples from the same small island. Would she let me know the source of the relevant Nguna samples, and send me samples so that I could replicate her findings? She replied that she was sure of the sequences and would re-check the results as soon as she could, that the possibility of a sequencing mistake is always there but that she had been reassured by the sheer mass of data. Considering the gravity of the situation and the impact even the suspicion of mitochondrial recombination was having on the reputation of the field as a whole, I then made a second request for samples of the Nguna DNA. This is unusual but not unheard of. I mentioned earlier that whenever a scientific paper is published there is an implicit undertaking, where possible, to make the raw material available for verification. This principle is at the very foundation of scientific progress. Without independent verification, or at least the opportunity to do so, scientific results have no validity. In most cases an actual test is unnecessary because the findings are quickly overtaken by new results. But here we had a situation where an entire field had been threatened with extinction. The truth about the Nguna samples, whatever it was, had to come out. And quickly.

I am sad to report that my requests for samples to verify the Nguna sequences did not produce results. Nor did I know of other laboratories that had tried to contact Erika to replicate the results. In the meantime, the reputation of mitochondrial DNA as a reliable evolutionary tool was spiralling downwards. The undergraduates had heard all about it. In the 1999 biological anthropology exams at Oxford, the demise of mitochondria featured in many of the students’ answers. At a packed meeting in the zoology department at which some new work from Maynard Smith was being presented by one of his colleagues, I found myself in the distinctly uncomfortable position, during questions at the end of the lecture, of having to defend the reputation of mitochondria in front of an audience of very distinguished and influential evolutionary biologists who seemed only too eager to write it off.

I was pretty sure by now that Erika’s Nguna data were wrong. Still, it was no good my thinking that. It was not really much use publishing our own results from the same island, either. There would still be uncertainty, and the original paper would still stand. If it were wrong, then it had to be corrected in the scientific press by Erika herself. In the meantime, I had also contacted co-authors of the paper who cooperated as far as possible: but still no sign of the samples.

In September of 1999 there was to be a conference in Cambridge at which both Erika and I were down to speak. It was a conference about Europe, and I gave a paper early on about our European work. Erika had been invited to talk about the Pacific islands and, we all assumed, about mitochondrial recombination. Generally speaking, scientific conferences are intensely polite affairs. There is a brief introduction by the session chairman; the speaker comes to the front and presents the paper, usually illustrated by a few slides or overheads; there is polite applause, a few questions from the audience, perhaps a bit more applause; the chairman introduces the next speaker. On this occasion, by the time it came for Erika to speak, there was a tangible atmosphere of anticipation, the expectation of a showdown in the air. The audience was completely silent, not wanting to miss a single word.

Erika began by saying that she was not going to talk about recombination. A murmur of surprise spread round the audience. Why had she come halfway round the world to a meeting on the genetic history of Europe if not to talk about mitochondrial recombination? As she went through her text on other aspects of her work in the Pacific, I knew I had to ask her about her Nguna work during questions, even if it had not featured in the presentation itself. It was the only way to get the matter cleared up. Was she sticking to her story or not? As Erika finished speaking, I raised my hand and the chairman called me to put my question. I was very nervous indeed, and could feel my heart pounding. But the issue was so important that I pressed on, in as unemotional a tone as I could manage.

‘Erika,’ I began, ‘although you did not refer to this specifically in your talk, there has been, as you know, considerable interest in your claim of finding examples of mitochondrial recombination on the island of Nguna. As you also know, my laboratory did not find evidence for recombination in samples from the same small island. There has been a suggestion in the scientific press [which there had, and not by me] that there may be a systematic error in the DNA sequences which appeared in the article. How do you respond to this suggestion?’

She answered instantly that she had checked the sequences and stood by them.

I had to keep going. ‘In that case, Erika,’ I replied, ‘why have you refused my requests for samples of the original DNA so that the sequences could be independently verified?’

The entire conference hall froze into complete silence.

‘I did not refuse,’ she answered.

‘But you did not reply to my request, which amounts to the same thing,’ I argued.

This was turning into a Grade One row. Erika accused me of having noi scientific but personal motives for pursuing the matter. Fortunately, before I could answer this charge, someone else asked a related question about the recombination data and got what seemed to me to be an equally unconvincing reply. And yet, though by now many in the audience must have had their douhts about her original paper, at the end of the meeting it was still standing. There was no retraction. Not yet.

After that conference, Erika came under pressure from some of her co-authors on the original paper to clarify the position. Eventually, she conceded that the sequences were indeed wrong and, in August 2000, nearly eighteen months after the first paper appeared, the corrcction was published. For some unexplained reason, the sequences from the first part of the control region had been shifted by ten bases. This is something that can happen if the sequencing machine is playing up. The base that the machine had scored as a mutation at position 76 was actually the normal base for position 86. So there were no mutations at 76 after all. Getting to the truth had been an exhausting, unpleasant and distressing experience. Everyone makes mistakes. But to take so long to set the record straight on such an important issue with so many ramifications seems to me completely contrary to the spirit of scientific enquiry. But there it was. Mitochondria had survived the recombination scare.