fied yeast [genes] will require that thou­sands of different deletion strains be tested under a large variety of selection condi­tions," they write. 'Molecular bar-coding is ideal for this task because it allows large numbers of tagged deletion strains to be analyzed simultaneously in a highly quanti­tative fashion."

Chee says the Stanford paper illustrates "what this technology is good at—being able to analyze complex mixtures of nu­cleic acids in a parallel win. If you've got a mixture of 1,000 different sequences, you can analyze them all at the same time, whereas current methods generally re­quire you to purify them out one at a time and analyze them separately. Generally, its not feasible to contemplate the kind of analysis that's done in Dan Shoemakers paper/'

In another stud}', senior scientist David J. Lockhart and coworkers at Affymetrix, in collaboration with researchers at Genet­ics Institute, Cambridge, Mass.. have mea­sured the expression levels of thousands of genes in parallel [Nat. BiotechnoL 14, 16^5 (1996)]. The group has used a set of four chips to look at the expression of more than 6,500 human genes, all from the same sample, without any purification.

"There are various ways to do this by conventional methods/' explains Chee, a coauthor on the paper, "but they usually involve counting. You might make a cDNA [complementary DNA] library and sequence lots of individual clones and then count the frequency of each one/' By using DNA chips, he says, "you're hybrid­izing everything at once and looking at the different signal levels. It's much more sen­sitive, it's much faster, and it's probably more accurate.'" The technique has made it possible to observe changes in gene ex­pression caused by cell stimulation and to analyze alterations in gene expression as­sociated with malignancy.

Chee, Fodor, and coworkers also re­cently reported the use of DNA chips to perform comparative sequence analysis on the entire human mitochondrial genome [Science, 274, 610 (1996)]. They focused on variations in the genomes of different individuals—differences that can have pro found effects on biological function. The researchers found that 50 genomes could be read by DNA chips in the time it takes to read two genomes by conventional gel-based sequencing, and that the chip tech­nology also allows significant reductions in sample preparation time.

Right now, Affymetrix has only one commercially available chip, for analysis of drug resistance mutations in the re­

verse transcriptase and protease genes of human immunodeficiency virus (HIV). The purpose of the chip is to facilitate studies of the way in which specific HIV mutations affect the development of viral resistance to AIDS drugs.

"We do make and sell other chips/' says Chee, "but so far they're mostly in contracts with specific partners." Af­fymetrix has arrangements with Genetics Institute, Merck & Co. (Whitehouse Sta­tion, N.J.), Incyte Pharmaceuticals (Palo Alto, Calif.), Roche Molecular Systems (Alameda, Calif.), and other companies to use chips for expression monitoring and diagnostics.

In addition, says Chee, "we have a col­laboration agreement with OncorMed (Gaithersburg, Md.) to develop a p53 chip for cancer diagnosis. There are various stud­ies in progress to see how the mutations and expression levels of genes correlate with outcome and prognosis in cancer."

Other companies besides Affymetrix are developing miniaturized array-based DNA chip technologies. For example, Nanogen, in San Diego, is working on a technology called APEX, for automated programmable electronic matrix. An APEX device uses an electric field to concentrate pieces of DNA at sites containing DNA "capture probes." Hybridization occurs between the probes and complementary sample DNA. Unhybridized sample DNA is then expelled from the chip surface by re­versing the field polarity.

According to Nanogen, the chips can discriminate single base pair mismatches in DNA sequences and offer advantages over conventional DNA hybridization technologies and even other DNA chips. For example, the company believes its

Dogma among biochemists has held that proteins—no matter how com­plicated their structures—never

contain knots. But two years ago, chem­ists at Princeton University pointed out that disulfide bonds and covalently bound metal atoms tie a number of pro­teins into classic trefoil knots. And this fall, researchers at the University of Kan­sas, Lawrence, reported the structure of a protein whose peptide backbone is

chips are the only ones that actually in­crease the rate of hybridization and dena-turation. Initially, diagnostic applications are being pursued for the technology.

Hyseq, Sunnwale, Calif., has developed DNA "SuperChips" that can sequence 64,000 bases in one reaction. The set of probes on each chip can be changed via software control, avoiding the need to de­sign and manufacture custom chips for each application. Tests for genetic diseases, can­cer, and infectious diseases can be am on the same chip, from point mutation detec­tion through complete sequencing.

In the area of genomics, the company is currently analyzing about 200,000 genomic samples per month. Our analy­sis includes the identity and relative ex­pression level of every gene expressed in even cell," says Hyseq President Lewis S. Gruber. "This approach analyzes down to the level of one message per cell for any of the 150,000 human genes expressed in that cell. As far as we know, there is no comparable technology."

Synteni, Palo Alto, Calif., is working on microarrays printed on glass with high-speed robotics for genome analysis and other applications. Incyte Pharma­ceuticals, besides collaborating with Af-fymetrix, is using ink-jet printing technol­ogy for localized attachment or synthesis of DNA to form microarrays, primarily for monitoring of gene expression.

Several other academic and commer­cial groups also are doing research on miniaturized DNA chip devices. But the recent series of reports on the Affymetrix DNA chips puts the current focus on this technology as it moves toward broader adoption and applicability.

Stu Borman

twisted into what most people would consider a knot.

"These findings are delightful," com­ments David C. Richardson, professor of biochemistry at Duke University. "We used to boast about proteins not having knots. If your protein structure came out knotted, you'd say: 'Whoops—I'd better look again.' "

Ironically7, looking again is exactly what led Fusao Takusagawa, associate professor

Much Ado About Knotting Chemists discover knots in proteins, challenging conventional ideas about folding

DECEMBER 9, 1996 C&EN 43

science/technology Despite its spaghetti-like appearance, the substructure of the multicopper enzyme ascorbate oxidase is a single closed curve that forms a trefoil knot. The simplified schematic at right traces only those loops, linked by disulfide and metal coordination bonds, that define the knot.

Cys = cysteine His = histidine

of biochemistry at Kansas, to the knot in CS>adenos\ lmethionine synthetase (MAT). Takusagawa and coworkers published the first crystal structure of that enzyme earli­er this year [J. Biol. Chem.. 271, 136 (1996)]. But only recently did he and post­doctoral research associate Shigehiro Ka-mitori notice that one end of the polypep­tide chain was slipped through a loop in the middle of the protein [/. Am. Chen ι Soc. 118,8945(1996)].

"if you somehow grabbed the ends of the protein and pulled, you would

A subunit of (S)-adenosylmethionine synthetase (MAT) shows the unusual knot structure of the polypeptide chain, in which the orange N-terminus is threaded through a loop formed by the green, blue, yellow, and white sections.

end up with a knot," Richardson notes. That's not the case with virtu­ally any other protein of known structure: Tugging on the ends of their polypeptide chains would result in a straight chain.

Takusagawa and Ka-mitori describe finding a

knot in the polypeptide chain of MAT as "dramatic." entitling their paper A Real

Knot in Protein" and writing that "there has been no report so far of knots in native pro teins or polypeptides." They discount as "pseudoknots" knotted structures in which

peptide chains are tied together by disul­fide and metal coordination bonds. But Princeton chemistry professor Kurt Mis-low —who with postdoctoral research asso­ciate Chengzhi Iiang had earlier described several such knots—objects strongly to the Kansas researchers' statement.

"We had already report-i ed knots in native proteins," jf Mislow says [/. Am. Chcm. I Soc, 116, Π 189 (1994)and ο 117, 4201 (1995)]. "If you I trace the chain, including "J disulfide bridges and metal 1 atoms, you see a genuine 8 trefoil knot. Protein' is not | synonymous with 'polypep­

tide' because cofactors. cys­tine cross-links, and so forth are integral parts of proteins. Their claim is simply false because what we had pre­viously found were legiti­mate knots, and the authors' characterization of them as pseudoknots—whatever that means—is absurd."

Mislow points out that mathematically a knot is de­fined as a closed curve in

space that doesn't intersect itself. The knots the Princeton researchers uncovered by searching the Brookhaven Protein Data Bank are true closed knots, whereas the knot in MAT described by Takusagawa is what's known as an open knot.

A closed knot can't be untied no matter how the strands are pushed and pulled, Mislow explains. An open knot, however, can be untied by pulling the ends out of the encircling loops. "From the point of view of topology, where everything is considered rubbery, the so-called open knot is not a knot at all." he says. Never­theless. Mislow concedes that most peo­ple would classify the sort of knot the Kan­sas researchers found in MAT as a real knot, even though it does not fit the strict topological definition.

It's reasonable for both Takusagawa and Mislow to call their protein struc­tures knots, says Marc L. Mansfield, re­search scientist and associate research professor at Michigan Molecular Insti­tute. Midland, Mich. In the strict mathe­matical sense, Mislow is correct when he says Takusagawa does not have a knot. But in that sense, neither are the knots that sailors and Boy Scouts tie true knots, although we call them knots." Mansfield is a polymer chemist interested in the general problem of knot recognition.

The really interesting question, Mans­field says, is how proteins fold into knot­ted structures. Mislow points out that the cross-links responsible for the trefoil knots he has described must form after the polypeptide chain is generated. That is, the cross-links are not explicitly en­coded by the structural gene.

Takusagawa has also been thinking about what knots can teach about pro­tein folding. He notes that the N-termi­nus of a protein is synthesized first. Find­ing the N-terminus of MAT within a loop of the protein that must have been as­sembled much later indicates that "the outcome of folding in the cell is not di­rectly controlled by the vectorial nature" of protein synthesis, he says.

Richardson comments that scientists have a long way to go to completely un­derstand protein folding. "No knots has been an amusing régulant}' of proteins that we have thought may be something fundamental." he says. "But nobody has a tight enough concept of how proteins fold for this to shake the foundations.

"Knots add to the fascination of how marvelously complex and subtle proteins are," Richardson says. "The} continue to surprise us."

Pamela Zurer

4 4 DECEMBER 9. 1996 C&EN