From The Economist September 25, 2004

Finding drugs in the library: Researchers are searching for new medicines lurking in old herbal texts

IT IS a miracle that the "Ambonese Herbal", a 17th-century medical text compiled by Georg Everhard Rumpf, a German botanist, ever made it to the printing press. Rumphius, as the author styled himself in Latin, was an employee of the Dutch East India Company. He was stationed on Ambon, in the Malay archipelago (now part of Indonesia). He began collecting and drawing plants in1657, and continued even after going blind in 1670. Four years later he survived an earthquake that killed his wife and daughter, but he then lost all his workin a fire in 1687. Undaunted, Rumphius dictated a new version of his book, the first volume of which was shipped to Europe in 1692, only to be sunk by the French. Fortunately there was a copy, and Rumphius went on to compile six more volumes, completing the last just before his death in 1702. His employers sat on the book for decades, however, fearing that rival nations would benefit from the medical knowledge it contained. Finally, a botanist in Amsterdam published the work between 1741 and 1755.

The "Ambonese Herbal" explains the medical uses of nearly 1,300 species native to the Malay archipelago, based on Rumphius's quizzing of the local population. Medicines shipped from Europe were either useless or unavailable in sufficient quantities, Rumphius complained in the preface, so using local remedies made much more sense. His epic work is just one of many historical texts that contain such "ethnomedical" information. The medicinal value of plants is still recognised. Roughly half of the anti-cancer drugs developed since the 1960s, and about 100 other drugs on the market, are derived from plants. In the past, figuring out which plants to screen for therapeutic potential involved ethnomedical study in which traditional healers-from village shamans to tale-telling old wives-were asked to identify valuable species. More recently, this approach has given way to high-throughput screening, in which thousands of random specimens are methodically tested by robot technicians.

But both methods have their drawbacks: the knowledge of traditional healers is being lost as they die out, and high-throughput screening has not proved to be very efficient. That is why a team led by Eric Buenz, a researcher at the Mayo Clinic College of Medicine in Rochester, Minnesota, has proposed a new, hybrid approach. Hundreds of unstudied herbal texts, dating from Ancient Greece to the modern age, are sitting in libraries around the world. By sifting through these texts and comparing the results with modern medical databases, it should be possible to identify promising candidate species for further examination and screening. The researchers explain this strategy in a paper published this month in Trends in Pharmacological Sciences.

To test their idea, Mr Buenz and his colleagues analysed the first volume of the "Ambonese Herbal". The text, originally in Dutch and Latin, is in the process of being translated into English. Two reviewers went through the English translation of the first volume and extracted all the medical references. They then drew up a table listing each species, the symptoms for which it was prescribed, and hence its probable pharmacological function. The sap of Semecarpus cassuvium, the wild cadju tree, for example, is listed as a treatment for shingles. This suggests that it has antiviral properties. The list of species was then checked against a database called the International Plant Names Index, to identify misspellings and synonyms. After that, each species was looked up in NAPRALERT, a database listing all known biochemical and ethnomedical references to plants, to see if it had been mentioned in the medical literature. It was thus possible both to determine how accurate the information in the "Ambonese Herbal" is, and to identify candidates for further investigation. Of the 42 plants described in Rumphius's first volume as having medical properties, 24 had biochemical matches in NAPRALERT, which suggests that they are indeed effective. Nine of the others had ethnomedical matches, which means their potential use as medicines is already known about, but has not been followed up by modern science. But nine plants did not appear in NAPRALERT at all, and are therefore potential sources of novel drugs.

The next step, says Mr Buenz, is to scale up and automate the process. "Our work with the Rumphius herbal was a proof of concept," he says. "The push now is to make the project high throughput with bioinformatics." Book scanners, he observes, have become cheaper and more efficient in recent years. The latest models can scan 1,000 pages an hour, yet are gentle enough to handle old and delicate tomes. And by using natural-language processing software to look for particular expressions, and cross-referencing potential matches with medical and botanical databases, the text can be analysed quickly. Manually combing through the text of the first volume of the "Ambonese Herbal " took four weeks, says Mr Buenz, but his experimental automated system did the same work in a few hours. The big challenges are dealing with foreign languages, old typefaces and variations in terminology-but translation systems and databases are improving all the time. Text mining will never replace other methods of drug discovery, but tapping the accumulated medical expertise locked up in old documents could, it seems, provide some helpful shortcuts.