Community-acquired pneumonia (CAP) is the most common infectious cause of death worldwide, and MIT researchers have now designed a type of “activity-based nanosensor” that can distinguish between viral and bacterial pneumonia infections based on differential protease expression, and which they hope will help doctors select the most appropriate treatment for each patient. In a study in mice, the researchers showed that their sensors could accurately distinguish bacterial and viral pneumonia within two hours, using a simple urine test to read the results.

“The challenge is that there are a lot of different pathogens that can lead to different kinds of pneumonia, and even with the most extensive and advanced testing, the specific pathogen causing someone’s disease can’t be identified in about half of patients,” said Sangeeta Bhatia, PhD, the John and Dorothy Wilson professor of health sciences and technology and of electrical engineering and computer science at MIT and a member of MIT’s Koch Institute for Integrative Cancer Research and Institute for Medical Engineering and Science. “And if you treat a viral pneumonia with antibiotics, then you could be contributing to antibiotic resistance, which is a big problem, and the patient won’t get better.”

Bhatia is the senior author of the team’s study, which appears in the Proceedings of the National Academy of Sciences (PNAS), and is titled, Host protease activity classifies pneumonia etiology.” In their paper, the team concluded, “We have created, screened, tested, and validated the sensors using in silico, in vitro, in vivo, and in situ methods, setting the stage for further development that might one day translate into a clinically useful diagnostic for the rapid and noninvasive detection of CAP … This proof-of-concept diagnostic approach demonstrates a way to distinguish pneumonia etiology based solely on the host proteolytic response to infection.”

The COVID-19 pandemic has brought CAP to “the forefront of global health priorities,” the authors wrote. “However, even before COVID-19, CAP had long been responsible for significant morbidity and mortality worldwide, with millions of people affected globally and over 100,000 deaths per year in the United States alone.”

Many different types of bacteria and viruses can cause pneumonia, but there is no easy way to determine which microbe is causing a particular patient’s illness. One reason why it has been difficult to distinguish between viral and bacterial pneumonia is that there are so many microbes that can cause pneumonia, including the bacteria Streptococcus pneumoniae and Haemophilus influenzae, and viruses such as influenza and respiratory syncytial virus (RSV).

This uncertainty makes it harder for doctors to choose effective treatments because the antibiotics commonly used to treat bacterial pneumonia won’t help patients with viral pneumonia. In addition, limiting the use of antibiotics is an important step toward curbing antibiotic resistance. “… the standard of care for patients with suspected CAP is to initiate empiric antibiotics as soon as possible based on local antibiotic resistance patterns and patient characteristics (e.g., age, comorbidities), a strategy that may exacerbate antibiotic resistance and not provide clinical relief,” the team continued. “To accurately triage, treat, and track patients with CAP due to bacterial and viral causes, new noninvasive tools that can both rapidly diagnose acute pneumonia and identify etiology must be developed.”

In designing their sensor, the research team decided to focus on measuring the host’s response to infection, rather than trying to detect the pathogen itself. Viral and bacterial infections provoke distinctive types of immune responses, which include the activation of protease enzymes that act to break down proteins. The MIT team found that the pattern of activity of those enzymes can serve as a signature of bacterial or viral infection.

The human genome encodes more than 500 proteases, many of which are used by cells—including T cells, neutrophils, and natural killer (NK) cells—that respond to infection. A team led by study author Purvesh Khatri, PhD, an associate professor of medicine and biomedical data science at Stanford University, collected 33 publicly available datasets of genes that are expressed during respiratory infections. “To our knowledge, no disease-specific signatures consisting solely of host enzymes have been created to distinguish pneumonia etiology,” the team noted. “To create such signatures for bacterial versus viral pneumonia, we curated publicly available transcriptomic datasets for respiratory infections from whole blood and peripheral mononuclear blood cells (PBMCs), filtered these datasets for human peptidases … and applied a computational multicohort framework designed to integrate gene expression data (multicohort analysis using aggregated gene expression [MANATEE]) across 33 unique study cohorts.”

By analyzing those data, the researchers were able to identify 39 proteases that appear to respond differently to different types of infection. Bhatia and her students then used those data to create 20 different sensors that can interact with those proteases. The sensors consist of nanoparticles coated with peptides that can be cleaved by particular proteases. Each peptide is labeled with a reporter molecule that is freed when the peptides are cleaved by proteases that are upregulated in infection. The reporters are then eventually excreted in the urine, which can be analyzed using mass spectrometry to determine which proteases are most active in the lungs.

“ABNs contain mass-encoded peptide linkers that are designed to be cleaved by proteases dysregulated in specific disease states, the scientists wrote. “Upon peptide cleavage by a target protease, the linked barcodes are released from the ABN, after which they are small enough to diffuse into systemic circulation for subsequent renal concentration and clearance.”

The researchers tested their sensors in five different mouse models of pneumonia, caused by infections of S. pneumoniaeKlebsiella pneumoniaeH. influenzae, influenza virus, and pneumonia virus of mice. After reading out the results from the urine tests, the researchers used machine learning to analyze the data. Using this approach, they were able to train algorithms that could differentiate between pneumonia versus healthy controls, and also distinguish whether an infection was viral or bacterial, based on those 20 sensors.

“Machine learning algorithms were used to train diagnostic classifiers that could distinguish infected mice from healthy controls and differentiate those with bacterial versus viral pneumonia with high accuracy,” the team noted. The researchers also found that their sensors could even distinguish between the five pathogens they tested, albeit with lower accuracy than the test to distinguish between viruses and bacteria. One possibility the researchers may pursue is developing algorithms that can not only distinguish bacterial from viral infections, but also identify the class of microbes causing a bacterial infection, which could then help doctors choose the best antibiotic to combat that type of bacteria.

The study, in addition, identified some patterns of host response to different types of infection. In mice with bacterial infections, proteases secreted by neutrophils were more prominently seen, which was expected because neutrophils tend to respond more to bacterial infections than viral infections.

Viral infections, on the other hand, provoked protease activity from T cells and NK cells, which usually respond more to viral infections. One of the sensors that generated the strongest signal was linked to a protease called granzyme B, which triggers programmed cell death. The researchers found that this sensor was highly activated in the lungs of mice with viral infections, and that both NK and T cells were involved in the response.

“We believe that our ABN panel represents a method to diagnose pneumonia that could augment the current diagnostic paradigm,” the investigators concluded. “… the ability of our panel to detect pneumonia and determine etiology via a noninvasive readout within 2 h of sensor administration could represent a rapid test for pneumonia. Additionally, the ability to determine pneumonia etiology quickly and accurately could help curb rising antimicrobial resistance by ruling out bacterial pneumonia on a time scale that would enable antibiotic stewardship in patients with suspected CAP.”

The urine-based readout is also amenable to future detection with a paper strip, similar to a pregnancy test, which would allow for point-of-care diagnosis. To this end, the researchers identified a subset of five sensors that could put at-home testing closer within reach. However, more work is needed to determine if the reduced panel would work similarly well in humans, who have more genetic and clinical variability than mice.

To deliver the sensors in mice, the researchers injected them directly into the trachea, but they are now developing versions for human use that could be administered using either a nebulizer or an inhaler similar to an asthma inhaler. They are also working on a way to detect the results using a breathalyzer instead of a urine test, which could give results even more quickly.