In a nondescript office park in the northeast side of Menlo Park, CA, the next genomics revolution may be taking place. There, 12 prototypes of a new sequencing machine developed by startup Pacific Biosciences are churning out reams of DNA sequence as fast as built-in cameras can record it. The deep freezer-size boxes, covered for the time being in red plastic sheeting, are performing a novel feat: reading single strands of DNA in real time. The machine’s creators hope that innovation will result in a process that operates fast and cheaply enough to make sequencing a routine component of medical care.

The company, founded in 2004, made a splashy debut almost a year ago, showcasing its technology to the scientific community for the first time at a sequencing conference in Florida. Steve Turner, the company’s founder and chief technology officer, wowed the audience with a video of single molecule sequencing in progress, the product of a proof-of-principle experiment reading a 150-base-pair length of DNA. Since then, the startup has garnered $100 million more in funding–for a total of $178 million to date.

Pacific Biosciences’ central innovation is a small chip composed of a 100-nanometer-thick metal film deposited on a silicon-dioxide substrate and dotted with thousands of tiny wells, each only tens of nanometers in diameter. Before sequencing begins, an enzyme called DNA polymerase is immobilized at the bottom of the well, along with the strand of DNA to be read. Fluorescently labeled bases–each of the four DNA letters labeled with a different marker–are then flooded onto the plate, randomly diffusing into each well. When the correct base diffuses into the bottom of a well, the enzyme attaches it to the growing strand of DNA.

The wells are so small that fluorescent light shined through the bottom of the plate penetrates only the lower 20 to 30 nanometers of each well, meaning that only the bases being actively attached to the DNA molecule light up. That allows a camera to capture the sequencing reaction as it happens, leaving any irrelevant chemical activity in the dark. “The waveguide is the first technology to allow observation of the polymerase in action at physiologically relevant concentrations,” says Turner.

In a video showing the sequencing reactions in action, a series of lights scattered across the screen burst and fade in quick succession. (A computer detects which base is added with each burst by its position within a well.) The machines can currently sequence 12 million bases of DNA per hour, about one-third of a percent of a human genome. The video must actually be slowed to be viewed: the reactions occur too fast to be visible to the human eye.

Despite its early success, it’s not yet clear whether the company’s innovative approach will surpass “next generation” sequencing technologies already in use. Pacific Biosciences plans to release a commercial product in 2010 and will announce the target sequencing capacity for that machine early next year. New types of sequencing machines, such as those developed by Roche Applied Sciences and Illumina, have already revolutionized genomics, allowing scientists to sequence huge volumes of DNA and reportedly dropping the cost of a human genome to less than $100,000.

But Turner is confident that his method has advantages, especially in the clinical diagnostics market. The most advanced sequencing approaches currently on the market stop the sequencing reaction after the addition of each base, wash away extra bases, snap a picture, and then repeat. Real-time sequencing is faster and uses fewer chemicals, making it much cheaper. “In the long run, the reagent cost dominates in sequencing,” says Chad Nusbaum, codirector of the Genome Sequencing and Analysis program at the Broad Institute, in Cambridge, MA.

Both Roche and Illumina have ramped up speed by running a massive number of sequencing reactions in parallel. However, these methods generate relatively short lengths of sequence, about 50 to 200 bases pairs, which must then be computationally assembled into a complete sequence. The shorter the piece, the more computationally difficult it is to sew them together. “With a 35 base read, you can’t assemble 25 percent of the genome,” says Turner.

Pacific Biosciences has already generated what Turner believes to be “the longest sequencing trace ever made.” In a proof-of-principle experiment, scientists continuously sequenced the same 135 base circle of DNA 12 times–it was like repeatedly driving around a rotary. While the company hasn’t yet repeated the feat with natural DNA, the ability to generate long reads could be especially important when sequencing unknown DNA or stretches with a highly repetitive series of bases. “Many of the dynamic regions of the genome that are associated with disease consist of duplicated and complex repetitive sequences that cannot be accurately assessed at present,” says Evan Eichler, a geneticist at the University of Washington, in Seattle. “Long-sequence reads are necessary to comprehensively understand human genetic variation.”

Pacific Biosciences’ machine may also enable scientists to generate more accurate sequences. Existing methods generate a consensus sequence by reading the same section of multiple copies of DNA, which may have some copying errors, and pooling the results. “All short-read technologies are less accurate than traditional methods,” says Nusbaum. With Pacific Biosciences technology, scientists would theoretically be able to sequence the same piece of DNA multiple times. “That means we’ll be able to detect rare mutations with unprecedented accuracy, orders of magnitude better than others,” says Turner.

He adds that the long reads, high accuracy, and quick turnaround time make the technology ideal for diagnostics, such as those for cancer, which involves long stretches of repetitive sequences, and infectious disease, in which small sequence changes may drastically change the infectiousness and virulence of a pathogen.

Perhaps the biggest excitement comes from Turner’s predictions for the future. He says that it will be easy to further improve the company’s technology with higher-resolution cameras, faster DNA polymerases, and more densely packed chips. Currently, each chip has thousands of wells, only about one-third of which are used. (The remaining two-thirds house either no enzyme molecules or more than one, and thus fail to generate useful information.) But the chips, which are created using semiconductor fabrication methods, have the capacity to hold 10 million wells.

Turner calculates that with a camera that can track one million wells, a polymerase that operates at about 50 bases per second (the current rate is 10), and full use of all the wells on the plate, Pacific Biosciences technology could read 100 gigabases an hour. That translates to full coverage of a human genome–the same genome sequenced 15 times–in just 15 minutes.

“I’m pretty excited about the possibilities of this technology, but I remain to be convinced,” says Nusbaum. “They still have some significant technical hurdles ahead of them.”