November 7, 2023 | Within the next five years, general practitioners may have at their disposal a simple blood test for predicting the risk of Alzheimer’s disease (AD) two decades before any symptoms appear. The feat is being made possible by physicists at The Australian National University (ANU) who have coupled nanotechnology with machine learning to search for known biomarkers of disease onset, according to Professor Patrick Kluth.
He and ANU Ph.D. researcher Shankar Dutt have been leading the charge to identify causal proteins via solid-state nanopore sensing, culminating in a paper that published recently in Small Methods (DOI: 10.1002/smtd.202300676) describing the methodology. The multidisciplinary effort involved not only experts in physics who understand the properties of matter and energy, but materials scientists to design porous membranes enabling better signals from the protein biomarkers, computer scientists and engineers to help design and build the technology, and medical school colleagues who best appreciate the clinical implications of an AD diagnostic test.
While AD currently has no early intervention nor treatment, a logarithmic increase in discovery is underway that could soon be providing hope unavailable to previous generations of patients, says Kluth. An early diagnosis also gives individuals time to adjust their lifestyle to help preserve their cognitive function as well as the chance to benefit from an investigative treatment by participating in a clinical trial.
Detection of AD is already possible, although the only reliable means is an expensive and invasive spinal tap to measure amyloid and tau levels, he says. Numerous blood tests are under development, but the technology being created by the ANU team is taking on the challenge of identifying similar-sized proteins at the single-molecule level. It requires only a small blood sample and produces results in near-real-time.
The diagnostic technique is straightforward, says Kluth. A nanoscale silicon chip spans a thin membrane separating two electrolyte reservoirs that features a single pore roughly the size of the molecules of interest. An analyte is added to one side and a voltage is applied across the membrane, driving the molecules through the nanopore one at a time to create a small electric current change that gets analyzed.
The pore is temporarily blocked when a molecule is passing through, he explains, leading to a decrease in the current that can be measured with a timed resolution of 25 nanoseconds and is indicative of the protein. Many current drop signals are generated by the molecules and each one gets measured as it passes through the membrane. The machine learning algorithm gets trained on sets of these current signatures, so it knows how to pick out the same molecules in the future.
It is easy enough to control the size of the nanopore, says Kluth, adding that it only needs to be large enough for the targeted protein biomarkers to go through. The algorithm can be trained to pick out the exact diameter of the pores to accommodate a bit of variation between the sizes of the proteins.
Envisioned use of the test in the clinic would begin with customary centrifugation and filtration techniques on a blood sample taken from patients. A small amount of blood plasma would then be put on a device about the size of an iPhone that is connected to a laptop, Dutt says. The trained algorithm would report on the concentration level of the biomarkers indicating whether an individual is or is not at risk for AD. The biomarkers have all been well studied so the cutoff values are known, he adds.
Machine learning is necessary only for proteins of a similar molecular weight and/or size that are otherwise hard to tell apart, Kluth says, such as bovine serum albumin (BSA) and human serum albumin (HSA) which share over 70% of the same amino acid sequence. For the latest study, the algorithm succeeded in distinguishing between these proteins and two others, hemoglobin and concanavalin A.
“You can in principle measure any kind of protein biomarker [with the testing methodology] if you train your algorithm right,” says Kluth. The focus has been on neurological diseases—also including Parkinson’s disease, multiple sclerosis, and amyotrophic lateral sclerosis—because the low concentration of biomarkers in the blood has made them difficult to measure.
For the blood test to move closer to the clinic will take more work, including mapping out well-defined mixtures of proteins in real-world samples, he says. The machine learning platform still needs to be automated to produce results, so no special know-how is required to analyze the data. The algorithms may also need to be optimized to align with known standards and account for biological and technological variability that can occur when analyzing complex clinical samples, and the membrane technology will be further enhanced so the proteins produce better signals.
Nanopore sensing is a burgeoning field of study, and a lot of work is underway combing it with machine learning, Kluth says. In 2021, researchers in Japan reported on an “artificially intelligent nanopore” capable of identifying four types of coronaviruses similar in size, including SARS-CoV-2 (Nature Communications, DOI: 10.1038/s41467-021-24001-2).
Proteins are harder to measure than viruses, he adds, because they are far smaller as well as much softer and thus more easily deformed. A great deal of research is also being done around the detection of neurodegenerative diseases in blood that rely on methods other than solid-state nanopore sensing.
The technology underpinning the device could just as easily enable health monitoring on a personal level as it could help clinicians detect biomarkers of debilitating disorders of the nervous system, says Kluth. Because the test is “very cheap,” individuals might opt to test themselves on a continuous basis to get a longitudinal view of their risk.
The ANU team is believed to be the first to combine nanopore sensing and artificial intelligence for detecting similarly sized proteins, notes Dutt. One of the enablers is an Our Health in Our Hands funding initiative from ANU that pools expertise at the university to improve personalized healthcare.