Drug Discovery In the Library

HIGH THROUGHPUT WITH A TWIST

Although it seems radical, the technique dovetails with more traditional 21st-century solutions to drug discovery, including data mining and high-throughput screening, according to Buenz. More than half of current chemotherapy drugs and more than 100 marketed pharmaceuticals are derived from plant sources, he notes.

High-throughput screening can sometimes draw a blank, Buenz notes. Recently, a large random library screen of more than half a million small molecules failed to identify any compounds with antiretroviral activity in vivo, he says.¹ This was disappointing, admits Edward Garvey of the department of molecular screening at GlaxoSmithKline, and the study's author, though he says it's not particularly surprising, given that the aim was to find inhibitors of a specific molecular interaction. "High-throughput screening is very valuable and should be one of several approaches used to find starting points for drug discovery," concludes Garvey.

Buenz says his ultimate aim is to bring the power of high-throughput screening to bear upon the wealth of traditional knowledge that lies buried in old books. "There really has to be something to bridge these two together," he says. Once enough texts have been digitized and the useful information extracted, it should be possible to subject herbal extracts to high-throughput screenings.

NERVOUS LIBRARIANS

Tangled Taxonomy: The Name Game

Courtesy of Kirtas Technologies

G.E. Rumphius' 18th-century text Herbarium Amboinense contains a chapter on the wild dragon tree that includes this passage: "They put these leaves on the privy parts of both men and women, when they have been harmed or inflamed by fire-piss." A potentially promising drug could be missed if the language-processing software can't determine that "fire-piss" most likely means gonorrhea. Extracted terms are automatically run through online databases such as MedHist http://medhist.ac.uk and the History of Medicine Division of the National Library of Medicine http://www.nlm.nih.gov/hmd, which can help translate these outdated terms.

And it's not only medical terms. Plant names are likely to have changed over the years. Plant names are revised using online resources such as the International Plant Names Index http://www.ipni.org or Missouri Botanical Garden's TROPICOS http://mobot.mobot.org/W3T/Search/vast.html. Currently, only English texts are being scanned, but future plans call for online translation tools to mine foreign-language herbal texts for pharmacological leads.

The key to the process is an automated scanner designed by Kirtas Technologies, Victor, NY. Using the device, Buenz is able to whip through ancient texts at 1,000 pages per hour. According to the company, the cost is less than 3 cents per page, if the machine is used two shifts a day for five days a week over the lifetime of the machine. "Such acceleration and streamlining will enable the rapid uploading of literally thousands of historical texts in a matter of days or weeks, as opposed to years," Buenz and his colleagues reported in a recent issue of Trends in Pharmacological Sciences.²

This rate is now feasible, agrees Rory Mcleod of the digitization studio at the British Library in London. However, because this kind of automation could damage the texts, such technology is suitable for only certain parts of the collection, he says. "I personally would still be very reluctant to use books of 100 to 150 years old, but could see us implementing the technology for serials or monographs."

Buenz has already scanned several original texts in that age range, though he says it was not possible to use books from the Mayo Clinic's history of medicine collection. "The preservationists were very hesitant to allow anything to be sent over to New York to actually get it looked through." Several colleagues volunteered their own antique copies, the oldest text being The Dispensatory of the United States of America published in 1845. Kirtas Technologies claims that the robotic arm, which creates a vacuum to flip from page to page, is "gentler than the human hand," and Buenz is inclined to agree. A preservationist made a thorough study of the books both before and after scanning, he says. "So far, we haven't had any problems."

The alternative could be a bit more time consuming. According to archivists at the Missouri Botanical Gardens, a manual scan of some 20,000 pages would take four years to complete.

EXTRACTING USEFUL INFORMATION

Once the book comes out of the scanner, useful information must then be extracted from the digital pages. Buenz and his colleagues customized a Mayo Clinic software program normally used to pull relevant information from patients' notes. The software records anything that looks like a plant name or plant part, and symptoms and disorders are used to suggest possible pharmacological functions.

Quirky medical nomenclature and outmoded taxonomy can make things tricky, but online resources help to make sense of confusing terminology (see Box). For example, Rumphius describes how the Ambon people of Indonesia used the wild cadju tree, Semecarpus cassuvium: "... they rub the same burning sap also on margins of the shingles of Indian Curap so that it will exulcerate, die, and not progress further," he notes. Sifting through this passage, the software enables Buenz to surmise that the sap of S. cassuvium was being used to treat shingles, so it might have antiviral properties and could be worth a closer look.

Following the successful analysis of Rumphius' Herbarium Amboinense, Buenz and his colleagues plan to study the ethnobotany of the Tewa Indians from southwestern United States. Although the texts are not copyrighted and can be scanned without permission, Buenz is keen to involve the Indian tribe in the process. "Most of these tribes that we're looking at have lost all of their traditional knowledge regarding healing," he says, "and so hopefully we're going to be trying to bring some of that back."

Henry Nicholls hnicholls@the-scientist.com