Combining proteomics techniques reveals importance of protein structure to health
In a study undertaken in more than 10,000 individuals, and published today in Nature Communications, scientists have shown that integrating information derived from different technologies to measure proteins can identify otherwise hidden links between proteins and human health and disease. The international research team led by scientists from the Berlin Institute of Health at Charité (BIH) and from the Medical Research Council (MRC) Epidemiology Unit at the University of Cambridge provides a detailed view on how hundreds of plasma proteins are linked to hundreds of diseases and measures of health, showing that in several cases it might rather be the structure of a protein rather than its blood concentration that may mediate the effect.
Proteins are the essential functional units of the human body that are composed of amino acids and coded for by our genes. Naturally occurring sequence variation in the human genome can influence both the amount of a protein present and the function of the protein. Pharmaceutical companies and researchers now use this information to identify novel drug targets and treatment options. However, how we measure proteins shapes the insights we can gain, and up to now it has been unclear how this translates across thousands of proteins potentially linked to the whole spectrum of human health and disease.
More than 800 proteins measured in 10,000 participants
The researchers compared two commercially available proteomics technologies that each measure hundreds to thousands of proteins in blood samples simultaneously. The SomaScan v4 assay uses short, single-stranded DNA fragments, called aptamers, that fit to the structure of the protein like a key to a lock, whereas the Olink proximity extension assay uses sophisticated antibodies similar to those used in clinical labs. Using these complementary technologies, they measured the abundance of 871 circulating proteins overlapping between both technologies in more than 10,000 Fenland Study participants. They combined this with genetic data to chart a map showing how proteins contribute to diseases and health more generally.
One third of the more than 500 associations between individual proteins and disease or measures of health were only seen with one technology, highlighting a huge synergistic value of integrating technologies that goes beyond simple verification of results.
Novel Biology identified
Dr Maik Pietzner lead author of the study and researcher at the Berlin Institute of Health at Charité (BIH) and at the MRC Epidemiology Unit said: “To understand how proteins contribute to diseases we have to go beyond measuring the amount in blood. It was remarkable to see how taking differences in the results between the two technologies and combining these with genetics enabled us to identify novel biology. For example, we show that the role of the protein GDF-15 in severe nausea in pregnancy and childhood obesity, possibly through altering taste perception, does not depend on whether there is more or less of it in the blood, but rather on a minor change in its shape introduced by a genetic variant of which one in four persons carries at least one copy.”
However, the researchers also identified examples in which the results of the two proteomic assays for the genetic relationship of a protein with a particular phenotype differed. For example, a protein called PILRA seems to be causally involved in Alzheimer’s pathology, but whereas the Olink assay suggests higher levels of PILRA would be beneficial in preventing Alzheimer’s disease, the SomaScan assay correctly identifies that the opposite holds true. The researchers attributed such effects to the differences in the affinity of the two technologies for what they called ‘proteoforms’ of PILRA. The term proteoform describes proteins that differ in their shape or function due to common genetic variation that either changes single amino acids in the protein sequence or leads to multiple changes in the protein sequence by altering the genetic template that is used to encode the protein.
Synergistic insights into disease mechanisms
On a more technical note, the authors observed that both assays agreed moderately to well on roughly half of the proteins tested. They found that technical factors explained poor agreement, but also that common genetic variations that change the shape of proteins has a strong influence, basically disturbing the lock-and-key model. Despite unexpectedly high variation between both assays, they show that two in three genetic variants related to proteins were consistent between both platforms.
Dr Claudia Langenberg, division head at the Berlin institute of Health at Charité (BIH) and at the MRC Epidemiology Unit, is senior author on the paper. She concluded: “We are just about to start to make full use of the enormous potential of combining protein measurements in blood with genetic variation to identify novel biology and diseases. Our study - for the first time - provides systematic evidence that is not only important for how we measure proteins, but also shows that complimentary technologies can provide synergistic insights into disease mechanisms and the development of treatment strategies. We generated a benchmark for future studies that aim to combine different technologies to maximize sample size for future discoveries.”
Pietzner M., Wheeler E., et al. Synergistic insights into human health from aptamer- and antibody-based 1 proteomic profiling. Nature Communications 24 November 2021. DOI:10.1038/s41467-021-27164-0.
Interview: Maik Pietzner and Claudia Langenberg
Interview: Maik Pietzner and Claudia Langenberg
How proteins cause diseases - new publication provides information on the importance of the measurement method
Which questions did you want to answer?
We were interested in how different technologies to measure proteins from blood samples compare to each other and how this might influence conclusions about biology and diseases we draw from such data. We were in particular interested in how the technology we use affects which regions of our genome we identify to be associated with more or less of a specific protein in blood.
How did you come up with it, why did you care?
A core interest of our group is the identification of novel disease mechanisms to identify therapeutic options. Integrating common variation in the genome that predisposes us to a higher risk for a disease with proteins measured in blood provides a unique opportunity to discover possible treatment targets or harmful mechanisms. However, how we measure proteins matters, each technique has its advantages and disadvantages, and it was up to now unclear, how different techniques influence the conclusions we draw from the data. This is relevant if we want to identify the exact mechanism by which a protein may cause a certain disease. Most people still assume that too much or too little of a protein may cause a disease, but we know that the function of a protein can be impaired in various ways, including a different shape induced by variation in our genetic code.
How did you proceed? What experiments did you perform?
We measured the concentration of 871 proteins in blood from a large cohort of up to 10,000 participants of the Fenland study with two different techniques. For the entire cohort we collaborated with a company called SomaLogic, that uses small pieces of DNA that fit to the structure of the protein like a key to a lock. For a subset of about 500 from those same participants we also measured the identical set of proteins using a different technology provided by Olink, which works more like what we are currently using in clinical routine. This allowed us not only to directly compare how well both assays agree with each other but more importantly we used machine learning methods to identify technical and biological factors that explain differences.
To get to the core of our aim, we performed millions of statistical tests to identify which single letter substitutions of our genetic code led to more or less protein in the blood, or more precisely to more or less recognition by the two different techniques used. Having identified such protein quantitative trait loci (pQTLs), we tested whether they are also relevant for common complex diseases using publicly available data on thousands of diseases and measures of human health.
What results did you get?
We observed a remarkable synergy between both technologies when we systematically tested for the clinical relevance of pQTLs, identifying more than 500 examples in which a protein was robustly linked to a disease or other phenotypes via shared genetic variation. More than a third of these biologically and clinically relevant insights were only seen with one of the methods. That means, technical differences between both platforms enabled unique insights that would not have been otherwise possible to discover. For example, we show that not more or less of the protein GDF-15 in blood may account for its effect on severe nausea in pregnancy and childhood obesity, possibly via taste perception, but rather a minor change in its shape introduced by a genetic variant of which one in four persons carries at least one copy. We find similar examples on how both technologies capture different properties of proteins with relevance for Alzheimer’s or autoimmune diseases, such as rheumatoid arthritis.
On a more technical note, we observed that both assays agreed moderately to well on roughly half of the proteins we tested. We found that technical factors explained poor agreement, but also that common genetic variation that changes the shape of proteins has a strong influence, basically disturbing the lock-and-key model. Despite the rather high variation between both assays, we observed that two in three genetic variants related to proteins were consistent between both platforms.
Did that surprise you? Why? What was new?
We were staggered by the synergistic nature of both technologies to generate genuine biological and clinical insights, providing a whole new perspective to the field. We started out on a rather technical question, but our observations rather indicate that technical differences can be exploited to uncover novel mechanisms by which proteins contribute to human health and diseases. Our study provides not only a benchmark for future efforts to employ proteomic techniques to identify protein – disease links, but more importantly provides bespoke biological examples how different properties of proteins can now be studied in humans to understand disease mechanisms.
Who did you cooperate with and why?
This work has only been possible through our long-term engagement with the MRC Epidemiology Unit at the University of Cambridge, who have established the outstanding, deeply characterised Fenland cohort. We also teamed up with SomaLogic, who performed protein measurements for the entire Fenland cohort.
Does your result have any relevance for medicine? For patients?
We identified numerous examples of how malfunction of proteins is likely implicated in disease onset and impairs human health, providing tangible candidates for pharmacological modulation. Our results will further help to identify the best mode of action for medication. For example, human analogues of GDF-15 are currently in clinical trials to treat obesity, but our results indicate that the type of GDF-15 protein used to do so may have a strong impact on the success of such efforts.
What´s next?
We will use our results in multiple ways. Firstly, it informs our strategy for upcoming even larger studies including tens of thousands of participants to identify more regions in the genome that affect human health through proteins. Secondly, we are currently following up bespoke examples using experimental models to understand underlying mechanisms in detail. Finally, we apply our results in the search for novel biological insights into the current pandemic.
Original article: Maik Pietzner, … Claudia Langenberg, Nature Communications: „Synergistic insights into human health from aptamer- and antibody-based proteomic profiling“ http://www.nature.com/ncomms DOI 10.1038/s41467-021-27164-0