The medical staff knew the seriously ill woman carried a drug-resistant bacterium when she entered the National Institutes of Health Clinical Center, a 243-bed research hospital.

A New York City hospital had transferred her in June 2011 to be part of a clinical study at NIH. At age 43, she had already been through a lot, including a lung transplant for which her immune system had to be suppressed. She probably contracted the Klebsiella pneumonia during an earlier hospital stay; the bacterium, after all, causes about 6 percent of infections spread in hospitals nationwide. Worse, her strain of K. pneumonia carried genes that made it impervious to nearly every antibiotic in the medicine chest.

The NIH Clinical Center had never suffered an outbreak of K. pneumonia -- and it did not want one now.

To manage the threat, the hospital put her in isolation, limited staff contact, imposed strict rules about visitors, and required the use of gloves and gowns. During her stay, the patient spent two 24-hour periods in the intensive care unit, and the doctors treated her infection with gentamicin, tigecycline and colistin, an older and potentially toxic medicine from the 1940s considered a drug of last resort. The combination worked, however, and she recovered, leaving the hospital in July 2011. The danger of an outbreak, the NIH staff believed, had passed.

They were wrong.

Two months later, in early August 2011, a 34-year-old man with cancer became sick with a K. pneumoniae infection. A sputum culture identified the bacteria but could not confirm whether he had the same strain as the New Yorker. The two patients, after all, were never in the same ward at the same time, so his infection could have come from a separate source before he arrived at NIH. Still, the Clinical Center team responsible for controlling hospital-acquired infections began to investigate the two cases.

Like all good detectives, the team first looked for fingerprints. "For decades, we used pulsed-field gel electrophoresis to differentiate between strains," said Tara N. Palmore, M.D., the Clinical Center's deputy hospital epidemiologist. This test produces a barcode-like pattern from bacterial DNA that shows whether strains are genetically similar; if the tests showed the same bar patterns, then the two cases might have the same source. In K. pneumoniae, however, 70 percent of the strains in the United States belong to one strain type with one pulsed-field pattern. "This test is not very helpful for that organism," she said.

Moreover, even as the investigation began, new cases started to appear in the Clinical Center at the alarming rate of one a week, sending shudders through the hospital's leadership and an increasing sense of urgency in the infection-control team. For a hospital filled with the sickest of sick patients, an outbreak of multi-drug resistant K. pneumoniae could be disastrous. The microbe commonly spreads in intensive care units or ICUs -- especially in patients with suppressed immune systems -- producing a 40 percent death rate. The team rapidly applied a set of epidemiology tools to control transmission, but many questions remained unanswerable with existing techniques -- and the outbreak continued to spread.

Despite heroic efforts to isolate and separate affected patients, enforce hand hygiene and infection-control precautions and disinfect large sections of the Clinical Center -- including use of hydrogen peroxide vapor to decontaminate rooms and ripping out contaminated sink drainpipes -- the bacteria ultimately colonized 17 patients over the next five months. Six would sicken and die from the infection. Five others died from their underlying disease while actively infected. The remaining eight survived because K. pneumoniae can enter the body, colonizing the gastrointestinal tract but never causing illness. Colonized patients, however, can unwittingly serve as bacterial reservoirs capable of transmitting infections to vulnerable patients, or even infect themselves if the bacterium moves to a vulnerable part of the body.

"It is very difficult to contain an epidemic once these kinds of organism are introduced into the hospital environment," said David Henderson, M.D., NIH Clinical Center's deputy director for clinical care and associate director for quality assurance and hospital epidemiology. "They become endemic and part of the hospital flora."

A depressing prospect given K. pneumoniae's virulence.

Early in the outbreak, the Clinical Center's unique location on the 400-acre campus of the National Institutes of Health paid off. Behind the hospital, in a basic research building, a team led by Julie Segre, Ph.D., a senior investigator at the National Human Genome Research Institute (NHGRI), offered to help. Head of the epithelial biology section of the intramural Genetics and Molecular Biology Branch, Dr. Segre had been working with the Clinical Center's Clinical Microbiology Department to study the evolution of bacterial antibiotic resistance when she heard about the outbreak.

"We were already trying to develop clinical molecular diagnostics tools," Dr. Segre said, "We thought we could use genome sequencing to tell whether the K. pneumoniae from the first patient was the same strain as the one that infected the second patient."

Genome sequencing produces an ordered list of As, Ts, Cs and Gs that represent the chemical subunits (nucleotides) of DNA, the long thread that stores genetic information in a cell. Sequencing bacterial genomes delivers far more precise data than the pulse-field gel approach. Where the older pulse-field gel produces a barcode, sequencing gives the exact order of genetic letters, allowing researchers to spot and track even single nucleotide changes.

In essence, genome sequencing measures bacterial evolution in real time. Bacteria reproduce rapidly, some species taking only 20 minutes to produce a new generation. Despite the accuracy of DNA replication, however, each succeeding generation has a small chance of carrying a new mistake -- the substitution of one nucleotide, say an A being replaced by a G -- among the 6 million nucleotides that make up the K. pneumoniae genome. Geneticists call these changes variations, or, when they cause malfunctions, mutations. Genomic variations occur regularly, and the more variations that accumulate, the more the current generation of bacteria has evolved away from the parent strain. "Genome sequencing can be used as an evolutionary time stamp for bacteria," Segre said.

Sequencing, Dr. Segre reasoned, should provide a powerful way to show the relatedness of bacteria in many affected patients. If patients carried K. pneumoniae with essentially identical genomes, then the bacteria were passing from patient to patient within the Clinical Center. Dramatically different genomes probably meant the bacteria came from an unrelated source.

The strategy only became feasible because sequencing costs and turn-around times have plummeted in recent years, reaching a point where the NIH Intramural Sequencing Center (NISC) in Rockville, Md., run by NHGRI, could actually afford to analyze a large number of samples from patients and environmental samples and deliver the results in clinically relevant turnaround time.

"This was so cool," said NISC Director James Mullikin, Ph.D. This experienced showed that sequencing "can help solve medical mysteries in the clinical setting."

The Clinical Center quickly accepted NHGRI's offer.

Meanwhile, the sense of urgency was growing. All the early patients had been in the intensive care unit (ICU). By September, the antibiotic-resistant K. pneumoniae had spread to patients in regular rooms of the Clinical Center. The worried infection-control team had been isolating the medical staff and equipment since August, so those caring for infected patients remained only with those patients, preventing any possible cross contamination. This strategy, called cohorting, is extremely disruptive and expensive. The infection-control team began collecting rectal swabs from every patient in the hospital to search for signs of silent K. pneumoniae colonization.

The task was enormous. More than 1,100 patients passed through the Clinical Center at the height of the outbreak; they all needed regular testing.

The NHGRI and NISC teams went back to the beginning of the outbreak and sequenced the genome of the K. pneumoniae that infected the first patient. Her medical records suggested that the New Yorker had been infected for many months. During her four-week stay at NIH, doctors had isolated the bacteria from samples from her throat, lungs, groin and urine. The first step was to sequence the bacteria's genome so it could be used to determine if the outbreak started with her.

To sequence a bacterial genome, the microbe must first be isolated from a patient and grown overnight in the lab to produce enough cells. The NIH Intramural Sequencing Center's staff then purified the DNA and ran it through a sequencer. This process takes a couple of days.

"We can provide the sequencing turnaround time that is required for solving cases and saving lives," Mullikin said. "You want to get the information back to the clinical center so they know how to stop the outbreak sooner."

With the sequence data in a computer, Evan Snitkin, Ph.D., an NHGRI postdoctoral fellow in the Segre lab, started to look at the variations between different bacterial samples.

Basically, Dr. Snitkin compared bacterial genome samples to see what is the same and what is different. When one DNA letter is changed in a genome, researchers call this a single nucleotide variation or SNP (pronounced "snip"). SNPs occur all the time during normal cell division. Most SNPs produce no effect; some cause disruptive mutations. Any SNP may produce an identifiable fingerprint useful for tracking epidemics.

SNPs samples from different body sites of the New Yorker showed she had been colonized long enough for bacterial evolution to occur. The pattern of SNPs from the bacteria in her urine showed consistency with the ancestral genotype of K. pneumoniae. But the samples from her lungs and groin where different: Each shared 3 SNPS compared to the ancestral form. The throat sample showed three still other SNPS in the 6 million nucleotide genome. All of these isolates are clearly derived from the same original strain, but at the level of sequencing, even these small numbers of variants made the urine, lung, groin and throat isolates distinct from each other.

These distinctions would provide powerful clues. Where the pulse-field gel method made all the samples look identical, genome sequencing could differentiate them with precision, down to a single genetic letter. Now the NIH team could track how the bacteria spread.

Traditionally, epidemiological detectives track the order in which patients become sick and try to connect the dots with possible overlap in specific wards on specific days. The doctors normally expect Patient 1 to give the infection to Patient 2 and Patient 2 to give it to Patient 3, and so on. That works well if patients get sick in the order in which they were colonized.

This outbreak would not be that simple. The New Yorker had been discharged three weeks before the second patient became sick. The third case did not arise for another 10 days, more than a month after the first patient left the Clinical Center. Moreover, there was no obvious thread to connect these dots. Patient 2 never overlapped with Patient 1 in the hospital.

The genomic data initially complicated things even further. The genome of Patient 2 was 2 SNPs different from Patient 1, suggesting it did not come directly from Patient 1. As the sequence data poured in, the picture became clearer and the hospital infection-control staff and Drs. Palmore and Segre's team created a transmission map. The results were surprising: Patient 1 transmitted the bacteria to other patients on two separate occasions, from infections on different parts of her body, creating two major clusters of infected patients.

The first transmission went from Patient 1 to Patient 3, the second one to shows signs of sickness; Patient 3 gave it to Patient 2, the first new patient to get sick. Epidemiological tracking showed that Patient 3 did overlap with Patient 1 in the ICU where the transmission presumably occurred. These individuals were infected with bacteria that shared SNP patterns with Patient 1's groin and lung samples.

A second, completely separate transmission event linked to Patient 1's throat isolate -- identified only by genomic data -- to Patient 4, for whom no connection with Patient 1 could be found. Patient 4 went on to infect Patient 10 who passed the K. pneumoniae isolate to the largest cluster of patients affected in this outbreak.

Genomic data also identified a final transmission from Patient 1 to Patient 8, for whom no epidemiological link could be found.

The intensive investigation led to several startling discoveries: K. pneumoniae is hardier than previously shown in the hospital environment, which may spread the bacteria. Strains from the outbreak were found in six sink drains that were then ripped out and replaced, and on a ventilator that had been thoroughly cleaned after use by Patient 6. Bacteria from the outbreak also were found in the room of Patient 8 after it had been cleaned.

"They didn't know how the bacteria were being transmitted," NISC's Mullikin said. "They assumed it was from patient to patient, but it was going through a sink. They would never have imagined that as possibility without the sequence data to prove it."

The bacteria even survived standard decontamination procedures. "If Patient A is in the hospital room, and you clean the room thoroughly, and the next patient gets sick with the same isolate, then you know your cleaning method is not sufficient," Dr. Segre said. The genomic results started changing hospital procedures.

Sequence analysis showed all infections originated with Patient 1 and not from different sources, a conclusion difficult to prove with other techniques. The infection-control study, however, never really determined the exact mechanism by which the bacteria passed from one patient to the next; there may have been some individuals or patients who were silently colonized and passed the bacteria on to susceptible patients. It also may have been passed by hospital personnel and equipment.

The infection-control interventions proved successful, and by the end of the year, no new cases arose in the Clinical Center, stemming the outbreak.

"This study makes it clear that genome sequencing, as it becomes more affordable and rapid, will become a critical tool for healthcare epidemiology in the future," said the Clinical Center's Dr. Henderson. His team is preparing a paper that will outline the use of genome sequencing within the methodology of infection control for similar outbreaks. "Now that we know what genome sequencing can do," he said, "I anticipate this methodology will be rapidly adopted by the hospital epidemiology community."

More than one million healthcare-associated infections occur across the spectrum of healthcare each year; in hospitals alone, the Centers for Disease Control and Prevention estimates that one in 20 hospitalized patients has a healthcare-associated infection. These infections can be life threatening and also add to our growing health care costs, accounting for billions of dollars in excess health expenditures each year. Multi-drug resistant K. pneumoniae is among the more dreaded infections because few effective treatments exist and it has a mortality rate of 40 percent.