LAURA LEAY: Welcome to MRS Bulletin’s Materials News Podcast, providing breakthrough news & interviews with researchers on the hot topics in materials research. My name is Laura Leay. How do you find a needle in a haystack? What if that haystack is a sample of liquid containing multiple biological species and you only want to find certain types? This is a challenge being tackled by Jennifer Dionne from Stanford University and collaborators using a technique more commonly applied to traditional materials characterization.
JENNIFER DIONNE: Our team was excited to develop a way to identify bacteria in complex samples without necessarily having to culture the bacteria. And whether you’re talking about bacteria in blood or bacteria in sputum or bacteria in wastewater there usually aren’t very many bacteria in that sample. Generally the number of colony-forming units is quite small, of order maybe one bacteria up to maybe in some cases a few hundred bacteria per milliliter. And then also the types of bacteria you have are very numerous. For a patient who’s experiencing symptoms in the hospital I think every hour that passes without the appropriate diagnosis and treatment with appropriate antibiotics could decrease the patient’s survival chances by about seven percent.
LAURA LEAY: Culturing bacteria takes hours and then the bacteria must be identified using specific methods such as the PCR, or polymerase chain reaction test, that identifies known genetic markers on a pathogen. Instead, Raman spectroscopy could be used to rapidly detect the presence of pathogens without the need to wait so long. The Raman spectra are complex and are convoluted with spectra from other biological material in the sample such as blood cells. Jennifer’s team collected spectra of two bacteria and used these to train machine learning algorithms to detect their presence in samples that contained red blood cells. The machine learning identified key elements of the spectra that would serve as identifying features.
FAREEHA SAFIR: We trained the model on known, pure samples and then we gave it some unknowns and sort of recorded the accuracy it got. But we sort of wanted to take it further the black box of machine learning that people reference was actually picking up on relevant biological, molecular differences actually in our bacteria versus just maybe sample noise or any overfitting of the model. And so we went further and did this further machine learning technique we called wavenumber or wavelength importance determination. We wanted to see which wavenumbers is the model picking up on, or which ones are the most important for making this classification. And then we actually went back to the literature and looked at: are these wavenumbers actually corresponding to relevant biological differences.
LAURA LEAY: That was Fareeha Safir who led on this work during her PhD project at Stanford University and is now part of a start-up called PumpkinSeed, co-founded by Jennifer. The team used a few tricks to improve the signal from the bacteria. They added gold nanorods to the samples. Fareeha explains how these act as nano-antennae.
FAREEHA SAFIR: The gold nanorods are sort of, what we can consider as these nano-antennas. They basically allow us to enhance the signal from hot-spots or regions that are in contact with the nanorods. And so, the reason that we incorporated them into our sample: we can get a Raman signal from our bacteria without the rods – they’re not required for Raman itself – however without the rods you have to actually interrogate or sort of take measurements of the samples a lot longer to get, sort of, signals at the level we can differentiate. With the rods we can actually drop, let’s say, a minute measurement down to the order of, like, seconds.
LAURA LEAY: The surface charge on the nanorods meant that they tended to have bacteria stick to them which in turn meant that the signal from the bacteria was enhanced by a much greater degree than from other material in the sample. Using tiny drops of the sample, just 2 pico-liters in volume, also helped to improve the signal intensity.
FAREEHA SAFIR: Our basic goal was to choose a droplet size where we can minimize the total number of cells present in each droplet so we could get very clear Raman signal from the entirety of the droplet. We’re really looking for needle in a Haystack; we’re trying to find, you know, one bacterium amongst, let’s say, fifty billion red blood cells in that one milliliter sample. And so to get that we wanted to give the best chance to find that bacterial signal without having to culture – without having to grow and basically produce more bacteria for the signal. And so the two ways that we went about that: one was getting those nanorods to enhance the bacterial signal and then the other way was really tuning our printer so that we could get these droplets with very few cells so we’d get very clear signal from anything in the entirety of that droplet to find that signal amongst all the other cells.
LAURA LEAY: To produce these tiny droplets a bespoke acoustic printer was developed. Unlike traditional printing techniques that involve a nozzle which can block or become contaminated, the specialist printer uses focused soundwaves to break the surface tension of a larger droplet. Tuning parameters such as sound wave frequency meant that the droplets could be reliably printed. Amr. Saleh from Cairo University in Egypt explains the benefit of this method.
AMR. SALEH: Because we use the acoustic waves to do all the ejection you actually maintain the cells intact and they maintain the cell viability. So most of other, like, approached to bio-printing use the nozzles or high temperature or high pressure, you lose a percentage, like two percent of the cells.
LAURA LEAY: With further work this novel application of Raman spectroscopy combined with machine learning and advanced printing techniques can be used to rapidly identify many more pathogens, with data acquisition taking perhaps just seconds. This could lead to fast diagnosis of a huge variety of ailments. Ultimately, the technique could be used on samples where the constituents are unknown; Raman spectroscopy coupled with machine learning can identify whether any pathogen is present. This offers significant advantages over current methods. Not only is it faster but current diagnostic techniques rely on additional knowledge to narrow down what pathogens may be present so that they can be tested for. According to Jennifer this sort of interdisciplinary work could be a game-changer:
JENNIFER DIONNE: We’re in a really exciting and interesting era when it comes to understanding biology. Biology is so data-rich and so complex. I think there are estimates that if you took just a milligram of DNA, the amount of data stored in that snowflake-sized sample would be equivalent to stacking today’s top-of-the-line hard-drives to like three times the height of Mount Everest. So there’s an enormous amount of data with which to uncover and yet the techniques we have to understand biology, as incredible as they are, they’re still so limited. So I think we’re entering a really cool era where new tools can help us uncover the mysteries of biology and the mysteries of life.
LAURA LEAY: This work was published in a recent issue of Nano Letters. My name is Laura Leay from the Materials Research Society. For more news, log onto the MRS Bulletin website at mrsbulletin.org and follow us on twitter, @MRSBulletin. Don’t miss the next episode of MRS Bulletin Materials News – subscribe now. Thank you for listening.