October 17, 2023 | The rise and fall of blood protein levels over time is an astonishingly good way to assess the risk of major atherosclerotic cardiovascular disease (ASCVD) events such as myocardial infarction (heart attack), as well as an ideal biomarker of the progression of disease and its regression in response to treatment, according to Kari Stefansson, M.D., Ph.D., CEO of deCODE Genetics, a subsidiary of Amgen. In fact, the biopharmaceutical company now routinely utilizes proteomic risk scores when designing clinical trials across multiple therapeutic areas.
Had the risk assessment mechanism been in place when Amgen was planning its FOURIER (Further Cardiovascular Outcomes Research with PCSK9 Inhibition in Subjects with Elevated Risk) trial, it would have been able to decrease the size of the 27,564-person study by 35%, he reports. In a recently published retrospective study in JAMA (DOI: 10.1001/jama.2023.13258), proteomic risk scores were shown to not only reflect the likelihood of having a major cardiovascular event but to do so in the absence of information on patients’ medical history and classical risk factors such as a family history of cardiovascular events or a prior ASCVD since the proteome handily captures these details.
The study was based on an analysis of plasma taken from over 13,500 Icelanders decades ago who had no history of a prior ASCVD event but later developed one, together with over 6,000 participants in the FOURIER trial who had suffered ASCVD before plasma sampling. The scientists at deCODE used artificial intelligence to pare down a list of 5,000 plasma proteins to fewer than 200 that were collectively signaling a problem, Stefansson says. Individual scores were all produced with proteomics data from a single plasma sample.
In the primary analysis on a population of people with a median age of 54 years, the proteomic risk score captured significantly more risk than a polygenic risk score. The reason, Stefansson explains, is because participants had active atherosclerosis ongoing in their arteries and the proteomic risk score is probably documenting early steps in its pathogenesis.
As autopsy studies have separately suggested, there are earlier steps in the pathogenesis of the disease that would likely have made a polygenic risk score more useful had the participants instead all been about 34 years old. Intriguingly, he points out, “the polygenic risk score and the proteomic risk score are uncorrelated, meaning that after the pathogenesis has been going on for a certain period of time, forces take over but independent of the pathogenesis of onset.”
Classical risk factors cannot be reversed but proteomics can capture the risk conferred to individuals by prior events and their history, Stefansson points out. “You can reverse the proteomic risk if a drug diminishes the probability of a cardiovascular event,” as was evidenced in the FOURIER trial by a lowered proteomic risk score among participants taking the PCSK9 inhibitor evolocumab (Repatha) for 24 weeks.
The algorithm being used here is agnostic as to which proteins matter, he adds. “We are not diving into the biochemical mechanism of pathogenesis... we are just looking for the signal and this is interesting because you can hardly find a protein in blood that does not correlate with many other proteins.” Unlike genomics data, proteomics data are highly structured into coregulated groups of proteins.
A proteomic risk score is likewise fundamentally different from an immutable polygenic risk score, which looks at the effects of multiple sequence variants in germline DNA and is set at birth—and is therefore most useful when people are relatively young to assess their lifetime probability of developing disease. Signals from proteins provide “an indication as to whether the event is nearby because the protein levels rise and fall as a function of time to [the disease] event,” says Stefansson.
By giving scientists a target engagement biomarker for assessing a drug’s clinical efficacy at the trial design stage, proteomic risk scoring will have a “huge impact on all clinical development” potentially industry-wide. Everything discovered by deCODE has been openly shared via its published papers, Stefansson says, so companies other than Amgen could theoretically recreate the ASCVD risk prediction instrument.
Since its launch in 1996, deCODE has been on a mission to understand human diversity, says Stefansson. This includes its study on the use of polygenic risk scores to assess thyroid cancer risk (PNAS, DOI: 10.1073/pnas.1919976117).
As he broadly defines it, human diversity includes the risk of different diseases conferred upon individuals by the sequencing in their genomes as well as environmental exposures. The environmental drivers of human diversity include age, which deCODE has previously shown to influence genetic recombination.
The internal environment is a part of that picture, Stefansson says, noting that atherosclerosis affects circulating protein levels and other happenings in the body. “All of the common diseases of man, including atherosclerosis and other cardiovascular diseases are diseases of delayed onset. They are mostly onset after 50 years of age, and that focuses our attention on age... [which] is just a funnel that collects the environmental influences.”
The proteome can be viewed as a large net for capturing those influences, he continues, which are consequential. Genomic impacts happen via proteins as do genetic diseases. “Short of stepping in front of a truck, most of the effects of the environment in our biology go through proteins.”
Most recently, deCODE has demonstrated that variants in the genome interact with each other and with the environment to affect cholesterol levels and thus the risk of cardiovascular disease (Cell, DOI: 10.1016/j.cell.2023.08.012). The environmental influences include alcohol and oily fish consumption, both of which can be tempered by genetic factors.
Measuring levels of proteins in the blood enables risk assessment for all kinds of diseases and the probability of all types of biological events, says Stefansson. That it allows the capture of time is especially important when it comes to preventive healthcare, because it will give payers what they have long said they need—evidence that a drug is dealing with a disease, not just diminishing the risk of developing it in the first place.
Using the proteomic risk score for predicting the probability of a heart attack, he offers as an example, an argument can be made that it is documenting atherosclerosis “as it is taking place... [and] in the end will lead to a myocardial infarction.”