Over the last few decades, scientists have recognized that to further our understanding of human biology – and in turn, human pathology – a complete picture of the human proteome is required. After all, proteins are the “workhorses” of cells and the targets of almost all pharmaceutical drugs. However, working towards a complete picture has been no easy feat.

Advances in analytical technologies have certainly increased capabilities for proteome analysis, but there have been bottlenecks in the process that have limited the speed at which proteomics can impact human healthcare and drug discovery. This has led to a perception that proteomics approaches “failed” to deliver what was initially hoped and expected.

In a recent interview with Technology Networks, Roy Smythe, chief executive officer at SomaLogic, explained that this failure has been understandable, but “is now a thing of the past”. SomaLogic is a protein biomarker discovery and clinical diagnostics company known for its proprietary SomaScan® Assay, a technology that is gaining traction as a “breakthrough” proteomics technology. The company recently announced an agreement with the biopharmaceutical company Amgen that aims to advance Amgen’s drug discovery and development programs through comprehensive proteomic profiling. SomaLogic’s technology will be applied to 40,000 samples, including samples from Amgen’s clinical trials.

We spoke with Smythe to discover what barriers have been holding proteome analysis back, and how the collaboration with Amgen aims to change the landscape of drug discovery.

Molly Campbell (MC): Why is it important to leverage proteomic data to gain a deeper understanding of human health?

Roy Smythe (RS): It has been thought for some time that proteins would perhaps be the best source of data to collect from the human body to understand its past conditions, current state and predict future states (like the development or worsening of disease). Why  is this? Proteins are the structural and functional molecules of life, and the targets of basically all drugs. In addition to the fact that they drive almost all human biology, they are also dynamic. In other words, proteins change with age, with environmental exposures, based on whether or not we are ill and if we are taking medications, etc.

Genes cannot do this, as they are static. Genetic abnormalities can predict some things, but unless there is a known “dominant mutation” in a gene or genes, they are not all that predictable of future states. Approximately 3–5% of human disease is driven directly by “germline” genetic predisposition, and the rest is some combination of genetic predisposition, plus a big dose of life exposures, or exposures alone. Proteins, if you can measure enough of them, give us a window into this remaining 95–97%. The problem has been measurement. There are 20,000  basic protein structures, but until we developed and refined our technology and approach, you could only measure hundreds at a time in a way that was clinically informative. We can measure 5000 now, and will increase that number to more than 7000 later this year.

Our pharma partners are interested in the ability to measure this many proteins for a number of important reasons. In the context of clinical research, this allows them to answer a number of questions including:

• Is the drug hitting the target of interest?
• Is the drug hitting targets we are concerned about?
• Can I take the data from clinical trial patients back to the lab to develop better drugs?
• Are there changes in protein expression that are important for me to understand re: the nature of the disease being treated?

We have measured this many proteins in humans hundreds of thousands of times and developed machine learning models to correlate these “protein expression patterns” to current and future body states and disease presence or risks via our first-in-class protein pattern recognition SomaSignal tests. In addition to these potentially transformational applications, there is also now the ability to measure a large percentage of the proteome and overlay that onto population-wide genomic studies.

The location of genetic abnormalities could be drug targets. If you can only measure a small percentage of the canonical 20,000, the likelihood of finding the protein that the gene makes, or the downstream gene that is being affected (“in trans”) and its protein is unlikely. We are enabling this work. Finally, from the standpoint of human health in general, our ability to manipulate that process to our benefit, requires that we understand the body’s biology more succinctly. If you can only measure several hundred of the structural and functional molecules of life – when there  are 20,000 basic and potentially innumerable altered forms – then how can you understand human biology? Our academic and pharma partners are interested in all of the above, and we all should be as well.

Laura Lansdowne (LL): What limitations exist that have prevented proteomics data from impacting healthcare? How does the SomaScan platform overcome such limitations?

RS: Most of the limitation has been around two sticking points. One, measuring enough proteins to get a “full body biologic signal”, and two, having the computing power and approaches to take the data from those measurements and make new inferences.

LL: How can the assay help to improve the assessment and management of therapeutic responses?

RS: It’s pretty simple – almost all drugs target proteins. The ability to see if positive or negative protein targets are being affected is one very important consideration, as well as understanding the body’s dynamic biologic response to a drug via protein expression.  In addition to providing this data from our assay, we are now offering our SomaSignal tests – where we are able to diagnose or predict things of interest – back into the Pharma and academic market. An example is our primary cardiovascular risk test – it measures the risk of a heart attack or stroke in the next four years based on protein pattern expression and machine learning.  If the trial you are running is one where you are evaluating the ability of a new drug to lower cardiovascular risk, we give the potential for an entirely new and more effective way to measure impact dynamically. Another example is nonalcoholic steatohepatitis (NASH). We have a test that can predict with high sensitivity whether or not someone has NASH without a liver biopsy and whether or not a treatment is having an effect on the liver. This allows you to enter patients into NASH trials and evaluate response without the need for a liver biopsy.

LL: What challenges exist when utilizing proteomic data in the context of drug discovery?

RS: While we are demonstrating enormous benefit from correlation of large scale protein expression patterns, rather than measuring single proteins to predict and measure things for the first time, there is not a full understanding of what protein expression patterns mean from a causation standpoint – how the changes actually drive biology. However, our partners in Pharma and academic settings are making a lot of progress.

MC: There is an apparent “failure” of clinical proteomics to deliver. What is your opinion on this matter, and what advances do you envision will take place in this space over the next few years?

RS: The failure has been understandable, but is now a thing of the past. It has been based on what I described earlier – an inability to measure sufficient numbers of proteins in a single sample at one point in time, an inability to understand complex protein expression data patterns if you do, and the inability to have an assay that meets all the requirements of a commercial clinical platform, such as cost, accuracy, speed, depth/breadth and reproducibility. We have solved for a lot of this at SomaLogic, and there is more to come.

Roy Smythe was speaking to Molly Campbell and Laura Elizabeth Lansdowne, Science Writers for Technology Networks.