Ultra-High-Throughput Proteomics To Discover New Biomarkers To Predict COVID-19 Severity

One of the puzzles of the COVID-19 pandemic is why infection with the SARS-CoV-2 virus can cause severe illness and death in some patients, while others experience only mild symptoms, or are asymptomatic and unaware they are infected. Being able to identify which people are more likely to experience severe illness could enable earlier action and administration of treatments to help improve patient outcomes. In a study recently published in Cell Systems, a new ultra-high-throughput proteomics method was used to identify 27 blood biomarkers that distinguished COVID-19 patients with mild or severe responses.

To learn more about the study and how the method could potentially be used in the clinic to diagnose COVID-19 severity, we spoke to Dr Markus Ralser, Einstein Professor of Biochemistry and Head of Biochemistry, Charitè University Medicine, and Group Leader, Molecular Biology of Metabolism Laboratory, The Francis Crick Institute.

Anna MacDonald (AM): Can you tell us about some of the limitations of methods currently used in the diagnosis of COVID-19?

Markus Ralser (MR): The COVID-19 pandemic has led to unprecedented times and economic impacts across the globe, with about 3 million people under lockdown. While some nations have instituted widespread testing to detect current infections, others have restricted these tests to those apparently at higher risk, such as patients admitted to hospital with suspected SARS-CoV-2 infection – at least during the early phase of the outbreaks. As public health restrictions become lifted, and fears of a second wave of infection abound, the race is on to develop diagnostic tests that do more than detect an ongoing infection. We now need tests to assess previous exposure and immunity to SARS-CoV-2. The two main methods being used for tests of SARS-CoV-2 infection are PCR-based assays and antibody-based assays.

Real-time RT-PCR is widely applied for diagnosing viral infections and there are now several such tests available for diagnosing COVID-19. These are proving essential for contact tracing and testing. Once the viral genome was sequenced, PCR primers were designed – which was relatively easy. The PCR itself can be performed using standard reagents, oligonucleotides, positive controls, and equipment in clinical labs, so PCR is really a robust technology that can be relied on when establishing new diagnostic tests for routine clinical lab services before pre-formulated assays are available. This makes it a good bet for rapid implementation and scaling up in response to infectious disease outbreaks. When all you want is a simple and effective test that gives you a clear yes or no answer for diagnosis, PCR is the best. But PCR-based assays cannot provide all the answers we need. If we want to get a fuller picture of the clinical situation, especially to guide treatment, other methods are available that can complement PCRs.

Serological immunoassays, such as ELISAs, instead tests not for the presence of the viral genome, but if an individuum has developed an immunity against the virus. Immunoassays can effectively look back in time to detect whether someone who has recovered was infected in the first place. This makes this kind of assay a vital tool for contract tracing, especially in the long term. Moreover, this test can also be used to verify whether a vaccine is working or not. Again, there are a few technical challenges. Developing immunoassays takes considerable time and requires prior knowledge of the antigen epitopes and the disease mechanisms. It can be quite tricky deciding on which epitope to target, as it has to be just right in order for the assay to be sensitive and specific enough. Also, if the developers really get it right and the epitope is biologically meaningful for the disease process, an antibody-based assay could inform clinicians about a patient’s immunity, for example. That’s useful because if we know that the epitope is highly immunogenic, thereby probably raising neutralizing antibodies in the patient, we know that the patient is immune and so he/she can go back to work knowing that they won’t pass the infection on to anyone else.

With MS-based proteomics, one can inform clinical decision making.

In contrast to PCR- and serological antibody based assays, mass spectrometry (MS)-based proteomics don’t rely on affinity reagents like antibodies and, unlike PCRs, they can be set up in an untargeted fashion without the need for prior knowledge of the disease. These assays are fast, precise and robust too, and they provide data that is clinically and biologically meaningful. MS analysis of blood or serum provides a readout that identifies and quantifies a large number of proteins or peptides present. This gives us a snapshot of what’s going on in the body, integrating its various processes and mechanisms. It’s a much bigger and more detailed picture, as it were, than what we get with PCR- or antibody-based assays, which only tells us if the virus is present or not. By comparing the proteomic profiles of patients against those of healthy controls, we can find new biomarker proteins and biomarker proteomic signatures. It also tells us a lot more about what proteins and associated biochemical pathways are involved in a disease mechanism. That’s why MS-based proteomics has so much potential to become an ideal technology for situations when rapid response is needed to address public health emergencies.

While it’s true that MS-based proteomics workflows are currently more established in research labs – where they’re used largely for biomarker discovery and profiling, this technology is becoming more applicable for clinical settings. For example, OVA1 is an in vitro diagnostic multivariate index assay (IVDMIA) that detects 5 protein biomarkers, which is used to inform clinicians about a patient’s risk for ovarian cancer, when deciding on whether to refer the patient for a surgical biopsy. It’s the first such IVDMIA approved by the FDA (for the management of women with pelvic masses that are suspected of being ovarian cancer). So, you may imagine, MS-based proteomics is being used to find biomarkers of COVID-19 and SARS-CoV-2 infection, as well as potentially for categorizing patients for personalized intervention based on their prognosis. Indeed, those prognoses could potentially be predicted by complex proteomic biomarker signatures, once big data sets have been analyzed using machine learning and other artificial intelligence (AI).

AM: Why hasn’t the potential of MS-based proteomics in clinical practice been realized so far?

MR: In a way, with examples like OVA1, you could say it has been realized – just not to a widely adopted degree yet.

Until quite recently, MS technology has been relatively slow and costly, because to get the proteomic depth needed in research studies, scientists had to use long measurement times, which were rather expensive. However, MS vendors like SCIEX are developing a lot of innovations and advances to speed up MS runs and make them more accessible. Another challenge is that the level of expert knowledge required to robustly run MS or rather liquid chromatography (LC)-MS in a clinical environment is quite high. Particularly for routine clinical applications, MS-based proteomics methods need to combine precision, reproducibility and robustness with low cost and high-throughput.

AM: Can you give us an overview of the new platform you have developed and the associated workflow?

MR: We developed this workflow to analyze disease susceptibility and progression in COVID-19 patients, so we needed this MS-based proteomic assay to be precise, relatively comprehensive, quick, cost-effective and able to cope with a high throughput of samples. To achieve this, we redesigned all the steps of the workflow: from sample preparation and LC through to data acquisition and data processing.

First, we automated the sample preparation, using robots to do all the liquid-handling – that is, all the pipetting and mixing steps, which reduced hands-on-time and meant that we could scale up to high sample numbers. We also incorporated several novel strategies to effectively mitigate any longitudinal mis-quantification effects produced by batch variations in the sample preparation reagents – a major bottleneck for large-scale proteomic experiments until now.

This in turn allowed us to use short-gradient high-flow LC – which has been used in several FDA-approved clinical assays, and which wasn’t very applicable before for high-throughput proteomic applications. The measurements were reduced to 5-minute gradient lengths with high flow rates of 800μL/min. These novel LC modifications substantially increased sample turnover, as well as analytical precision and retention time stability, while at the same time reduced costs.

For data acquisition on the mass spec, we switched from nanoflow to microflow – something we could do with the TripleTOF 6600 instrument without sacrificing analytical sensitivity, which resulted in reduced run-to-run variability across large series of samples. The instrument also meant we could achieve a sufficiently fast mass spectrometric duty cycle to record sufficient data points per chromatographic peak, with a very fast sampling rate and with SWATH Acquisition, a data-independent acquisition (DIA) method specifically developed by SCIEX to minimize stochastic elements in data acquisition. Applying SWATH Acquisition also reduced the runtime as it works by using 25 windows with variable window size to more quickly scan the precursor mass range of interest.

With data processing when using DIA schemes that don’t sample each peak individually, conventional software can’t properly deconvolute the short-gradient data, which is full of signal interferences. So, to deconvolute the complex data recorded, we improved on our DIA-NN software by adding several new algorithms and deploying a form of AI: deep neural networks. This boosted the number of true positive precursor identifications made in the short-gradient DIA-MS runs, as well as corrected for any interferences and improved the throughput capability of the data processing.

AM: Can you tell us more about how it was recently used to analyze COVID-19 patient samples in Germany? What were your main findings?

MR: After we benchmarked the platform using samples from the Generation Scotland (GS) large-cohort study, we applied it to investigating the COVID-19 pandemic outbreak in Germany. We obtained samples from 31 of the first patients hospitalized with COVID-19 at the Charité University Hospital in Berlin, with no other inclusion criteria beyond hospitalization due to a COVID-19 infection, because we needed to act very quickly during that early phase of the pandemic. From those 31 patients, we identified 27 biomarkers that either went up or down in quantity, depending on the severity of the disease. We then validated these biomarkers using a smaller cohort of 17 independent COVID-19 patients and 15 healthy volunteers.

The biomarkers were able to distinguish between severe COVID-19 cases and milder forms of COVID-19 disease, as set out by the WHO grading, which was introduced in April 2020. We also found several proteins that hadn't previously been associated with COVID-19 severity. These included proteins that suggest there is a role played by complement factors, the coagulation system, several inflammation modulators, as well as pro-inflammatory signalling pathways. The change in the quantities of these proteins reflect a progression from mild to severe COVID-19 (ranging from Scale 3: hospitalized, no oxygen therapy, to Scale 7: most critical), and so could potentially form the basis of a new clinical test using prognostic biomarkers for COVID-19.

AM: What implications could the identification of these biomarkers have for the diagnosis and treatment of COVID-19 patients?

MR: If we could develop a clinical test using these prognostic biomarkers to predict whether a patient is likely to progress to a more severe disease grade, we could intervene earlier to prevent that progression. That would be especially important given the lung damage, scarring and other longer-term complications, which we are just learning about from COVID-19 survivors. Further, if a test based on these biomarkers could also forecast disease progression risk in infected but asymptomatic patients or those with very mild grade disease (not hospitalized), it could allow clinicians to decide on which patients to pre-emptively hospitalize and treat – and which to allow to stay at home. Indeed, as well as vaccines, there may be need for a post-infection prophylactic to prevent the excessive inflammation in the lungs that seems to indirectly cause most of the damage associated with SARS-CoV-2. Moreover, several of our 17 biomarkers correspond with biochemical pathways identified as potential therapeutic targets, such as the interleukin-6-mediated proinflammatory cytokine signalling pathway.

Our study has shown that LC-MS is a viable, robust and powerful tool, capable of coping with the high-throughput demands of clinical research and practice in a global public health crisis. Such ultra-fast and reliable MS-based proteomics technologies can play a vital role both in clinical classification, as well as in the rapid identification of new therapeutic targets against novel infectious agents.

AM: Can you tell us about any future work you have planned?

MR: So far, our assays have been classifying patients, but of much greater clinical value would be the ability to look into the future, and to generate prognostic assays. We are now analyzing much larger and longitudinal COVID-19 cohorts, and use these to generate assays that, ideally, will predict likely disease trajectories, and hence help physicians to identify the patients that have the highest risk and hence deserve most attention.

Markus Ralser was speaking to Anna MacDonald, Science Writer for Technology Networks.