NEW YORK (GenomeWeb) – Researchers at Rice University have devised a new tool to screen infectious samples for the presence of bacteria and identify them.
The universal microbial diagnostics (UMD) approach, developed by Rice's Richard Baraniuk and his colleagues, relies on random DNA probes and a mathematical signal recovery technique to tease out the identity of bacteria from within a sample based on the probes' binding patterns. As the researchers reported in Science Advances today, their approach could identify 11 known strains of pathogenic bacteria using only five probes. The method could be useful for the quick identification of pathogens in both healthcare and defense settings, they wrote.
"The problem that we're addressing is the problem of identifying from some kind of sample ... that there is bacteria present and, second, classification of which bacterium it is," Baraniuk told GenomeWeb.
Currently, bacteria are commonly identified by using probes tailored to each bacterium of interest. Baraniuk argued that this can be cumbersome: for example, when a lab wants to test for the presence of a hundred or so different bacteria, it needs to use a hundred or so different probes. And while sequencing might be another approach for identifying bacteria in a sample, it is expensive and time consuming.
Their UMD method instead aims to be a rapid identification approach. It draws on a relatively recent mathematical theory called compressive sensing that has been used in digital signal processing, such as for digital cameras and phones. This theory suggests, Baraniuk said, that researchers don't have to use one probe per bacterium to identify what's there.
"What the math tells you is that if you are in a situation where, in a given sample, there's only one or a small number of bacteria present — which is pretty typical in medical situations — then you can actually use far, far fewer probes than you would think," he said.
The UMD approach instead examines the pattern of probes that bind to a sample. To do this, the researchers generated random hairpin DNA probes 38 nucleotides in length with a conserved four-nucleotide stem. These probes hybridize to bacterial genomes at different spots and to different extents that the researchers then gauge. By comparing the hybridization patterns of an unknown sample against the pattern of known bacteria from a database using a computational algorithm, they can determine what type of bacteria that unknown sample contains.
This approach is also nimble, Baraniuk said. If there's a new bacterium that they want to detect, they don't have to change the probes; they only have to update the software.
As part of their in vitro proof-of-concept study, Baraniuk and his colleagues combined five UMD molecular beacons — the probes — with genomic DNA from nine infectious bacterial strains that included Escherichia coli, Staphylococcus aureus, and Campylobacter jejuni. The sequence of some of these strains was known, while that of others was not known, the researchers noted. They determined the probe-bacterial DNA-binding affinity using FRET.
Based on receiver-operator characteristic curves and normalized root mean square analyses, they concluded that their screening performance was acceptable and thermodynamic modeling of the hybridization was accurate. In addition, they noted that when they expanded their reference genome database to comprise 40 genera — including common human pathogens — the detection performance remained high.
They also tested the identification of two other bacteria — Bacteroides fragilis and Enterobacter aerogenes — with four random probes.
Through simulations, the researchers suggested that their five-probe approach might be able to distinguish among 40 different strains and, with various numbers of probes, between 24 different Staphylococcus species and between 23 Vibrio species. Baraniuk noted that there are tradeoffs between the number of probes used in this approach and how specific it can be in its identification.
Such a tool, Baraniuk said, could help clinicians home in on what bacteria are present in a patient sample. "It takes so long to figure out what someone is infected with that doctors just immediately use these really, really broad spectrum antibiotics, which are not necessarily guaranteed to work, and second, are bad for antibiotic resistance reasons," he said. He added that he and his colleagues are interested in exploring whether they can get to strain-level and antibiotic resistance information using UMD.
The tool could also be used in defense or security scenarios in which experts want to determine whether a pathogen is being spread and what that pathogen is so that they can quickly understand what threat there might be.
Baraniuk noted that now that he and his colleagues have shown their approach works, they are interested in commercializing it. He added that though they used molecular beacons in their study, the method could use other hardware, such as microarrays.
That's something he said paper co-author Rebekah Drezek's lab, which has experience in commercialization and designing and building systems, is focused on.
He added that they are also interested in expanding UMD so it can detect and classify viruses as well as tweak the algorithm to boost the number of bacteria that can be detected and classified within a sample.