SAN FRANCISCO (GenomeWeb) – Ten years ago, the concept of using metagenomic sequencing to diagnose infection was first put to test retrospectively on samples of transplant patients who had all died of unknown infection.
But it wasn't until 2013 that it's true potential was demonstrated whenresearchers from the University of California, San Francisco used metagenomic next-generation sequencing (mNGS) as a last-ditch effort to identify the cause of a teenage boy's encephalitis. The effort worked and physicians changed the boy's treatment, likely saving his life.
Since that case, the UCSF team has launched a clinical mNGS test for encephalitis and meningitis out of its CLIA-certified laboratory, and a number of other academic groups are following suit, according to presentations at the Infectious Diseases Society of America IDWeek meeting held here this week.
While much progress has been made, researchers reported at the meeting that there are still a number of challenges, including with bioinformatics, signal-to-noise ratio due to a high background of human nucleic acids, and the fact that some infections cannot be detected by sequencing.
Nonetheless, researchers said that the sequencing-based tests are performing well and are hopeful that they can improve on the diagnostic rate for challenging infections.
Many acute infectious diseases remain undiagnosed, including between 60 percent and 80 percent of meningitis and encephalitis cases, Steve Miller, director of the clinical microbiology lab at UCSF, said in a presentation at IDWeek. Using traditional diagnostic techniques like culture and even PCR, "we don't do a good job of diagnosing infections or ruling out infection," he added.
To improve on this, the UCSF team has developed an mNGS test that analyzes both DNA and RNA from cerebrospinal fluid in patients with suspected central nervous system infections. It launched its clinical test last year. The test runs on an Illumina HiSeq instrument and makes use of an informatics pipeline called SURPI+ that the team developed to identify pathogens from metagenomic sequence data.
Miller said the team decided to focus on encephalitis and meningitis both because there is an unmet need for proper diagnosis and because interpreting metagenomic sequence data from cerebrospinal fluid is easier than it would be for sputum or fecal samples for respiratory or gastrointestinal infections. CSF is a "technically clean" environment, Miller said, so does not have as much background microbial DNA, making the results easier to interpret.
Bioinformatics, though, has been challenging, Miller said. He noted that microbial databases, such as GenBank, while very comprehensive, contain many errors. Researchers have "deposited sequences that are misannotated or just wrong," he said. "So we have to do a lot of filtering to scrub those out."
Another challenge is that metagenomic sequencing by nature sequences everything that's there, including a large host background, which can make pinpointing the pathogen difficult. Miller said that the group uses bioinformatics to filter out human reads, but is also looking at making those human reads useful. In particular, he said, the human RNA could be especially useful for looking at the human response to infection. "We're mining this data and actively investigating how to use this to better inform response," he said.
Currently, the lab batches samples for a weekly run on the HiSeq, starting on Monday with results reported by Friday. However, he noted that the lab aims to increase the number of runs to twice per week, which would reduce turnaround time to 72 hours.
In his presentation, Miller described a recent validation of the assay using 95 CSF samples, 73 of which were positive and 22 of which were negative for an infectious organism. The mNGS test was compared to all the previous tests that had been done on the samples and found to have a sensitivity of 73 percent and specificity of 99 percent.
However, as also described in a paper on the BioRxiv preprint server, PCR testing on discrepant samples confirmed that in 10 cases in which the mNGS test detected a pathogen but prior clinical testing had not, the mNGS test was correct. In addition, there were several cases that were negative by the mNGS test that were also not able to be confirmed via targeted PCR, but only by antibody testing.
Miller said that after doing discrepancy testing, the positive predictive agreement of the mNGS test was around 81 percent and negative predictive agreement was just over 90 percent.
The UCSF then ran the test on 20 prospective samples from pediatric hospitals and achieved a sensitivity and specificity of 92 percent and 96 percent, respectively. It had one false negative, a case of West Nile virus, which the assay also did not detect in the initial validation test.
In another IDWeek presentation, Anne Piantadosi, an infectious disease physician at Massachusetts General Hospital, discussed efforts her group is making to develop an mNGS test for CNS infections and described a potential reason why sequencing tests may not always detect some viruses.
Piantadosi said her lab compared mNGS against both PCR testing and serology-based tests on 68 patient samples. In 14 cases, both PCR and mNGS testing did not find a causative virus, but serology testing did. Those 14 cases included four West Nile virus cases, four varicella zoster virus cases, and four cases of herpes virus infection.
The biology of West Nile virus infection is well understood, Piantadosi said, with symptoms often persisting long after the infection has cleared, so it is possible that the virus could not be detected with the mNGS test because it was simply too late in the course of the infection.
That's why West Nile is routinely diagnosed with serology," she said, because the antibodies can still be detected even after the virus has cleared. "This is a fundamental, unavoidable limitation of the mNGS technique — it only works if the nucleic acid is there," she said.
Piantadosi said the group is also investigating whether other viruses that the mNGS test missed, like varicella zoster virus, are similar to West Nile in that the nucleic acids are cleared before symptoms go away. "You have to think about the biology of the infection," she said.
Piantadosi said her lab is now trying to figure out the next steps. "There is a lot of discussion about how to implement this into clinical practice," she said.
Meantime, Patricia Simner, who directs the medical bacteriology and parasitology laboratories at Johns Hopkins Medicine, said her team is looking at developing a clinical metagenomic sequencing test for CNS infections and is also evaluating various sequencing protocols to do antimicrobial resistance testing.
Simner said her group is interested in using sequencing to do susceptibility and resistance profiling because of the potential to get fast, comprehensive results. The current paradigm, she said, can involve multiple culturing steps and multiple antimicrobial susceptibility testing panels. And, "if you have an organism that is multidrug resistant, you have to set up even more AST panels."
The process can take days, and although PCR and other technology advances have helped to speed up the process, such methods are not comprehensive, she added.
Simner described one case where her lab retrospectively analyzed a patient sample to see whether whole-genome sequencing could have resulted in different management. The patient, a 64-year-old female who had developed sepsis following a liver transplant was placed on therapy but developed resistance to it. Based on the results of an AST panel, the physician added another drug to the regimen. Ultimately, she received the proper therapy about 72 hours post coming into the hospital. Using nanopore sequencing, her team determined that they could have arrived at the results within 28 hours.
Following that initial test, the group next analyzed 40 cases of Klebsiella pneumoniae infection, 31 of which were resistant to the antibiotic carbapenem and 12 of which were susceptible.
The researchers designed two different nanopore sequencing pipelines, one of which made use of the real-time analysis feature of the technology and another that was assembly-based and involved some error correction.
Under the standard-of-care procedures, of the 28 patients who had resistant K. pneumoniae, 22 initially received inadequate therapy and the median time for getting effective therapy was 61 hours.
With the nanopore sequencing approaches, the median time to getting a full profile that could have predicted the correct therapy was 41 hours with the real-time approach and 35 hours using the assembly approach.
Simner is also working on developing mNGS tests for analyzing cerebrospinal fluid for potential CNS infections and analyzing gastrointestinal infections from rectal swabs.
The lab is validating its CSF-based mNGS test for diagnostic purposes, and Simner said that it would be similar in to UCSF's test. In addition, she said, the group has been testing both Illumina and nanopore metagenomic sequencing on rectal swab samples on a research basis.