Specific patterns of protein levels in our blood could be used to provide a comprehensive “liquid health check” that gives a snapshot of health and potentially an indication of the likelihood that we will develop certain diseases or health risk factors in the future, according to research by scientists in the U.S. and U.K. working with SomaLogic. The results of their proof-of-concept study involving more than 16,000 participants, and published in Nature Medicine, showed that while the accuracy of models based on specific protein expression patterns varied, they were all either better predictors than models based on traditional risk factors, or would constitute more convenient and less expensive alternatives to traditional testing.
“This proof of concept study demonstrates a new paradigm that measurement of blood proteins can accurately deliver health information that spans across numerous medical specialties and that should be actionable for patients and their healthcare providers,” said Peter Ganz, MD, co-leader of this study and the Maurice Eliaser distinguished professor of medicine at UCSF and director of the Center of Excellence in Vascular Research at Zuckerberg San Francisco General Hospital and Trauma Center. “I expect that in the future we will look back at this Nature Medicine proteomic study as a critical milestone in personalizing and thus improving the care of our patients.” The team’s published study is titled, “Plasma protein patterns as comprehensive indicators of health.”
Preventative medicine programs such as the U.K. National Health Service’s Health Check and Healthier You programs are aimed at improving individuals’ health and reducing the risk of developing diseases. While such strategies are inexpensive, cost-effective, and scalable, they could also be made more effective using personalized information about an individual’s health and disease risk, the authors suggested. The application of “big data” in healthcare, assessing and analyzing detailed, large-scale datasets, makes it increasingly feasible to make predictions about health and disease outcomes and enable stratified approaches to prevention and clinical management. “Protein scanning” represents a potential approach to bridging the gap between the need for practicality and low cost, and the potential for “personalized, systemic, and data-driven medicine.”
Proteins regulate biological processes and can integrate the effects of genes with the effects of environment, age, existing diseases, and lifestyle behaviors, the authors explained. Our genomes contain about 19,000 genes that code for some 30,000 different proteins. Up to 2,200 of these proteins, including hormones, cytokines, and growth factors, are purposefully secreted into the blood, to orchestrate biological processes in health or in disease. Other proteins enter the blood through leakage from damaged or dead cells. Both secreted and leaked proteins can inform health status and disease risk.
In a proof-of-concept study based on five observational cohorts involving 16,894 participants, the researchers scanned 5,000 proteins in single blood plasma samples taken from each participant, “to simultaneously capture the individualized imprints of current health status, the impact of modifiable behaviors, and incident risk of cardiometabolic diseases (diabetes, coronary heart disease, stroke, or heart failure).”
To analyze the proteins in each sample the researchers used a technique that harnessed fragments of DNA known as aptamers, which bind to the target protein. In general, only specific fragments will bind to particular proteins. Using existing genetic sequencing technology, the researchers could then search for the aptamers and determine which proteins are present and in what concentrations. In total, the study carried out about 85 million protein measurements in the nearly 17,000 participants.
The researchers analyzed the results using statistical methods and machine learning techniques to develop predictive models—for example, that an individual whose blood contains a certain pattern of proteins is at increased risk of developing diabetes. The models covered a number of health states, including levels of liver fat, kidney function and visceral fat, alcohol consumption, physical activity, and smoking behavior, and for risk of developing type 2 diabetes and cardiovascular disease.
The accuracy of the models varied, with some showing high predictive powers, such as for percentage body fat. Other models demonstrated only modest prognostic power, such as that for cardiovascular risk, but even this was “still modestly better than traditional risk factors and could also add value in overcoming the incomplete utilization of risk calculation in primary care,” the team wrote. Many of the proteins measured linked to a number of health states or conditions. Leptin, for example, modulates appetite and metabolism, and was informative for predictive models of percentage body fat, visceral fat, physical activity, and fitness.
The researchers pointed out that a key feature of the study is that it used information from just one source, a single blood draw, for protein-phenotype models, the authors pointed out. “This was a key objective of our health check proof of concept.” The team didn’t include demographic or known risk factors in their models—unless absolutely necessary. They also didn’t test whether they models could be improved by adding in other features, such as history, laboratory tests, or genetic information. “It is possible that these multi-source models could improve absolute models’ performance, although their inclusion has potential implications for increasing costs and loss of convenience.”
One difference between genome sequencing and proteomics approaches is that whereas the genome is fixed, the proteome changes over time, possibly as an individual becomes more obese, less physically active, or smokes, for example. These changes in proteins could be used to track changes in an individual’s health status over a lifetime.
“Proteins circulating in our blood are a manifestation of our genetic make-up as well as many other factors, such as behaviors or the presence of disease, even if not yet diagnosed,” said Claudia Langenberg, MD, from the MRC Epidemiology Unit at the University of Cambridge. “This is one of the reasons why proteins are such good indicators of our current and future health state and have the potential to improve clinical prediction across different and diverse diseases.”
While this study shows a proof-of-principle, the researchers acknowledged that there were limitations, and suggested that as technology improves and becomes more affordable, it is feasible that a comprehensive health evaluation using a battery of protein models derived from a single blood sample, could be offered as routine by health services. “It is thus conceivable that, with further validation and the potential for expansion of the number of tests, a comprehensive, holistic health evaluation using a battery of protein models derived from a single blood sample could be performed. The next step is to test the applicability of the protein models that we have derived and validated in observational cohorts under research conditions in real-world healthcare systems.”
“It’s remarkable that plasma protein patterns alone can faithfully represent such a wide variety of common and important health issues, and we think that this is just the tip of the iceberg,” said study lead Stephen Williams, MD, chief medical officer at SomaLogic, which is developing its SomaScan® Platform and SomaSignal® tests for a wide range of human diseases. “We have more than a hundred tests in our SomaSignal pipeline and believe that large-scale protein scanning has the potential to become a sole information source for individualized health assessments.”