National policies for mitigating the COVID-19 pandemic include stricter measures for people considered to be at high risk of severe and potentially fatal cases of the disease. Although older age and pre-existing health conditions are strong risk factors, it is poorly understood why susceptibility varies so widely in the population.

People with cardiometabolic diseases, such as diabetes and liver diseases, or chronic inflammation are at higher risk of severe COVID-19 and other infections including pneumonia. These conditions alter the molecules circulating in the blood, providing potential ‘biomarkers’ to determine whether a person is more likely to develop a fatal infection. Uncovering these blood biomarkers could help to identify people who are prone to life-threatening infections despite not having ever been diagnosed with a cardiometabolic disease.

To find these biomarkers, Julkunen et al. studied blood samples that had been collected from 105,000 healthy individuals in the United Kingdom over ten years ago. The data showed that individuals with biomarkers linked to low-grade inflammation and cardiometabolic disease were more likely to have died or been hospitalised with pneumonia.

A score based on 25 of these biomarkers provided the best predictor of severe pneumonia. This biomarker score performed up to four times better within the first few years after blood sampling compared to predicting cases of pneumonia a decade later. The same blood biomarker changes were also linked with developing severe COVID-19 over ten years after the blood samples had been collected. The predictive value of the biomarker score was similar for both severe COVID-19 and the long-term risk of severe pneumonia.

Julkunen et al. propose that the metabolic biomarkers reflect inhibited immunity that impairs response to infections. The results from over 100,000 individuals suggest that these blood biomarkers may help to identify people at high risk of severe COVID-19 or other infectious diseases.

The coronavirus disease 2019 (COVID-19) pandemic affects societies and healthcare systems worldwide. Protection of those individuals who are most susceptible to a severe and potentially fatal COVID-19 disease course is a prime component of national policies, with stricter social distancing and other preventative means recommended mainly for elderly people and individuals with pre-existing disease conditions. The prominent susceptibility to severe COVID-19 for people at high age has been linked with impaired immune response due to chronic inflammation caused by ageing processes (Akbar and Gilroy, 2020). However, large numbers of seemingly healthy middle-aged individuals also suffer from severe COVID-19 (Zhou et al., 2020; Atkins et al., 2020; Williamson et al., 2020); this could partly be due to similar molecular processes related to impaired immunity. A better understanding of the molecular factors predisposing to severe COVID-19 outcomes may help to explain the risk elevation ascribed to pre-existing disease conditions. From a translational point of view, this might also complement the identification of highly susceptible individuals in general population settings beyond current risk factor assessment.

Pneumonia is a life-threatening complication of COVID-19 and the most common diagnosis in severe COVID-19 patients. As for COVID-19, the main factors that increase the susceptibility for severe community-acquired pneumonia are high age and pre-existing respiratory and cardiometabolic diseases, which can weaken the lungs and the immune system (Almirall et al., 2017). Based on analyses of large blood sample collections of healthy individuals, biomarkers associated with the risk for severe COVID-19 are largely shared with the biomarkers associated with the risk for severe pneumonia, including elevated markers of impaired kidney function and inflammation and lower HDL cholesterol (Ho et al., 2020). This may indicate that these molecular markers may reflect an overall susceptibility to severe complications after contracting an infectious disease.

Comprehensive profiling of metabolic biomarkers, also known as metabolomics, in prospective population studies have suggested a range of blood biomarkers for cardiovascular disease and diabetes to also be reflective of the susceptibility for severe infectious diseases (Ritchie et al., 2015; Deelen et al., 2019). Metabolic profiling could therefore potentially identify biomarkers that reflect the susceptibility to severe COVID-19 among initially healthy individuals. However, such studies require measurement of vast numbers of blood samples collected prior to the COVID-19 pre-pandemic. Conveniently, a broad panel of metabolic biomarkers have recently been measured using nuclear magnetic resonance (NMR) spectroscopy in over 100,000 plasma samples from the UK Biobank.

Here, we examined if NMR-based metabolic biomarkers from blood samples collected a decade before the COVID-19 pandemic associate with the risk of severe infectious disease in UK general population settings. Exploiting the shared risk factor relation between susceptibility to severe COVID-19 and pneumonia (Ho et al., 2020), we used well-powered statistical analyses of biomarkers with severe pneumonia events to develop a multi-biomarker score that condenses the information from the metabolic measures into a single multi-biomarker score. Taking advantage of the time-resolved information on the occurrence of severe pneumonia events in the UK Biobank, we mimicked the influence of the decade lag from blood sampling to the COVID-19 pandemic on the biomarker associations, and used analyses with short-term follow-up to interpolate to a scenario of identifying individuals susceptible to severe COVID-19 in a preventative screening setting. Our primary aim was to improve the molecular understanding on how metabolic risk markers may contribute to increased predisposition to severe COVID-19 and other infections.

A flow diagram of eligible study participants and case numbers is shown in Figure 1. Clinical characteristics of the study population are listed in Table 1. Among the 105,146 UK Biobank study participants with complete data on metabolic biomarkers and severe pneumonia outcomes, and no prior history of diagnosed pneumonia, there were 2507 severe pneumonia events recorded in hospital or death registries after the baseline blood sampling (median follow-up time 8.1 years).

Figure 1 with 1 supplement see all

Download asset Open asset

For the severe COVID-19 analyses, there were 652 PCR-confirmed positive cases diagnosed in hospital (inferred as severe cases in this study) among the 92,725 individuals with COVID-19 data linkage available per 3rd of February 2021. The number of severe COVID-19 cases in the UK Biobank closely followed the trends in hospitalised individuals for COVID-19 in England (Figure 1—figure supplement 1). In February 2021, the age range of study participants was 49–84 years. The median duration from blood sampling to the COVID-19 pandemic was 11.2 years (interquartile range 10.0–12.6). The prevalence of chronic respiratory and cardiometabolic diseases was similar for study participants who developed severe pneumonia and those who contracted COVID-19 and required hospitalisation, with the exception of COPD. There were 33 overlapping cases between severe pneumonia and COVID-19.

Metabolic biomarkers and severe pneumonia risk

Figure 2A shows the associations of 37 biomarkers with severe pneumonia events occurring during the follow-up in the entire study population (n = 105 146). The biomarkers highlighted here are those with a regulatory approval for diagnostics use in the Nightingale Health NMR platform. These biomarkers span most of the different metabolic pathways captured with the NMR platform; results for all 249 metabolic measures quantified are shown in Figure 2—figure supplements 1–3. Strong associations were observed across several metabolic pathways: increased plasma concentrations of cholesterol measures, omega-3 and omega-6 fatty acid levels, histidine, branched-chain amino acids and albumin were associated with lower susceptibility to contracting severe pneumonia. Increased concentrations of monounsaturated and saturated fatty acids, as well glycoprotein acetyls (GlycA, a marker of low-grade inflammation) were associated with elevated susceptibility to contracting severe pneumonia.

Figure 2 with 3 supplements see all

Download asset Open asset

Figure 2—source data 1

Numerical tabulation of odds ratios, betas, standard errors, and p-values for results shown in Figure 2.

https://cdn.elifesciences.org/articles/63033/elife-63033-fig2-data1-v2.csv

Download elife-63033-fig2-data1-v2.csv

Since all the biomarkers are quantified in the same single measurement, we examined if even stronger associations with severe pneumonia could be obtained using a combination of multiple biomarkers. We derived this multi-biomarker combination, denoted ‘infectious disease score’, using logistic regression with LASSO for variable selection, considering the 37 clinically validated biomarkers in a half of the study population as the training set. This resulted in an infectious disease score comprised of the weighted sum of 25 biomarkers, with the weights selected by the machine learning algorithm (Supplementary file 1). Broadly similar results were obtained using all 249 metabolic measures quantified in the Nightingale Health NMR platform to derive the multi-biomarker score.

The multi-biomarker infectious disease score was then tested for association with severe pneumonia in the other half of the study population. The magnitude of association for the infectious disease score was approximately twice as strong with severe pneumonia compared to any of the individual biomarkers (Figure 2B). The odds for contracting severe pneumonia was increased 67% per 1-SD increment in the infectious disease score. This corresponds to close to fourfold higher risk for contracting severe pneumonia among people in the highest quintile of the infectious disease score, compared to those with a score in the lowest quintile.

To assess the robustness of the multi-biomarker score association with severe pneumonia, we adjusted the analyses for prevalent diseases and performed analyses stratified by age and sex (Figure 3). The association was attenuated by ~10% in magnitude when adjusting for, or omitting, individuals with a diagnosis of prevalent diseases at time of blood sampling (cardiovascular diseases, diabetes, lung cancer, COPD, liver diseases, renal failure, and dementia; panels 3A and 3B). The association was similar across age groups, and also for men and women analysed separately (panels 3C and 3D).

Figure 3

Download asset Open asset

Figure 3—source data 1

Numerical tabulation of odds ratios, betas, standard errors, and p-values for results shown in Figure 3.

https://cdn.elifesciences.org/articles/63033/elife-63033-fig3-data1-v2.csv

Download elife-63033-fig3-data1-v2.csv

To mimic the influence of the decade-long lag from blood sample collection to the COVID-19 pandemic, we tested the association of the multi-biomarker infectious disease score with severe pneumonia events occurring during 7–11 years after the blood sampling (Figure 4A). Since there were only few severe pneumonia events recorded with more than 9 years of follow-up, we could not fully mimic the decade long time lag to the COVID-19 pandemic. The risk elevation observed in this time-lag accounting scenario was only approximately half of that observed for severe pneumonia events occurring within the first 7 years (odds ratio 1.43 vs 1.75 per 1-SD, respectively; and 2.59 vs 4.27 for individuals in the highest vs lowest quintile of the infectious disease score).

Figure 4 with 1 supplement see all

Download asset Open asset

Figure 4—source data 1

Numerical tabulation of odds ratios, betas, standard errors, and p-values for results shown in Figure 4.

https://cdn.elifesciences.org/articles/63033/elife-63033-fig4-data1-v2.csv

Download elife-63033-fig4-data1-v2.csv

To interpolate to a screening scenario conducted today, we also tested the association with short-term risk of severe pneumonia by analysing events occurring within the first 2 years after the blood sampling (Figure 4B). The association magnitude in this analysis of short-term risk scenario was approximately twice as strong as for severe pneumonia events occurring more than 2 years after blood sampling (odds ratio 2.21 vs 1.59 per 1-SD; 7.95 vs 3.35 for individuals in the highest vs lowest quintile of the infectious disease score). The elevated susceptibility to severe pneumonia associated with the multi-biomarker score was therefore three to four times stronger when examining short-term risk as compared to risk of severe pneumonia events occurring almost a decade after the blood sampling.

The elevation in the short-term risk for severe pneumonia for high levels of the infectious disease multi-biomarker score remained strong when adjusting for BMI, smoking and prevalent diseases (odds ratio 6.10 for individuals in the highest vs lowest quintile; Figure 4—figure supplement 1).

We further explored the risk gradient for a future onset of severe pneumonia along increasing levels of the infectious disease score, since non-linear effects could potentially facilitate the identification of thresholds for individuals at high susceptibility. Figure 5A shows the increase in the proportion of individuals who contracted severe pneumonia according to percentiles of the score. The risk increased prominently in the highest quintile, and particularly for the highest few percentiles. The time-resolved plot of the cumulative probability of severe pneumonia during follow-up is shown in Figure 5B. The susceptibility to severe pneumonia was particularly elevated among individuals with the very highest levels of the multi-biomarker infectious disease score. This was observed already during the first few years of follow-up, corroborating the results for long-term and short-term risk shown in Figure 3. The prominent and immediate elevation in susceptibility to severe pneumonia was also observed when limiting analyses to individuals without chronic respiratory and cardiometabolic diseases at the time of blood sampling (Figure 5—figure supplement 1).

Figure 5 with 1 supplement see all

Download asset Open asset

Figure 5—source data 1

Numerical tabulation of event rates for each percentile in Figure 5A.

https://cdn.elifesciences.org/articles/63033/elife-63033-fig5-data1-v2.csv

Download elife-63033-fig5-data1-v2.csv

Metabolic biomarkers and severe COVID-19

Figure 6 shows the associations of the 37 clinically validated biomarkers and the infectious disease score with the future onset of severe COVID-19 (defined as PCR-confirmed positive inpatient diagnosis). Many of the individual biomarkers had significant associations (p-value<0.001) with increased risk for severe COVID-19. These biomarkers for susceptibility to severe COVID-19 include lower levels of omega-3 omega-6 fatty acids as well as albumin, and higher levels of GlycA. We observed a high concordance in the overall pattern of COVID-19 biomarker associations with severe pneumonia (Figure 2A), with a Spearman correlation of 0.89 between the overall biomarker association signatures for severe pneumonia and severe COVID-19 (Figure 7).

Figure 6

Download asset Open asset

Figure 6—source data 1

Numerical tabulation of odds ratios, betas, standard errors, and p-values for results shown in Figure 6.

https://cdn.elifesciences.org/articles/63033/elife-63033-fig6-data1-v2.csv

Download elife-63033-fig6-data1-v2.csv

Figure 7

Download asset Open asset

Figure 7—source data 1

Numerical tabulation of odds ratios, and 95% confidence intervals for results shown in Figure 7.

https://cdn.elifesciences.org/articles/63033/elife-63033-fig7-data1-v2.csv

Download elife-63033-fig7-data1-v2.csv

The multi-biomarker infectious disease score derived for the future onset of severe pneumonia was also robustly associated with the future onset of severe COVID-19. The odds ratio was 1.40 per 1-SD increment and 2.90 for comparing individuals in the highest quintile of the multi-biomarker infectious disease score to those in the lowest quintile. This magnitude of association with susceptibility to severe COVID-19 was similar to that observed with severe pneumonia events occurring during the interval of 7–11 years after the blood sampling.

We further examined the association of the multi-biomarker infectious disease score with severe COVID-19 after adjustment or exclusion for prevalent diseases, and conducted stratified analyses for age and sex (Figure 8). The association with severe COVID-19 was attenuated, but remained significant when adjusted for BMI, smoking and prevalent diseases (panel 7A). The association magnitudes were approximately 20% weaker when limiting the COVID-19 analyses to individuals without prevalent diseases at time of blood sampling (panel 7B). There was no robust evidence of differences in association magnitude according to age (panel 7C) and odd ratios were broadly similar for men and women (panel 7D).

Figure 8

Download asset Open asset

Figure 8—source data 1

Numerical tabulation of odds ratios, betas, standard errors, and p-values for results shown in Figure 8.

https://cdn.elifesciences.org/articles/63033/elife-63033-fig8-data1-v2.csv

Download elife-63033-fig8-data1-v2.csv

Finally, we examined the technical repeatability and biological stability of measuring the multi-biomarker infectious disease score. The measurement repeatability was high (Pearson correlation 0.94 in blind duplicate samples; Figure 9A). Even though the blood samples were primarily non-fasting, the levels of the infectious disease score remained broadly stable during 4 years based on blood samples from repeat visits (Pearson correlation 0.61 between baseline and repeat visit measurements; Figure 9B).

Figure 9

Download asset Open asset

Most biomarker studies on COVID–19 have focused on characterising already infected patients and their disease prognosis (Kermali et al., 2020; Shen et al., 2020; Messner et al., 2020; Dierckx et al., 2020). In contrast, in the largest blood metabolic profiling study to date, we explored biomarker associations for susceptibility to severe pneumonia and COVID-19 in general population settings. We developed a multi-biomarker score for increased susceptibility to a severe infectious disease course, and demonstrated that this biomarker score captures an increased risk for COVID-19 hospitalisation a decade after the blood sampling.

The overall signature of biomarker associations was similar for the susceptibility to severe COVID-19 and to severe pneumonia (Figure 7). The proportions of individuals with existing cardiometabolic diseases were also consistent for both of these infectious diseases (Table 1). We used these observations of a shared risk factor basis to draw an analogy between susceptibility to severe pneumonia and severe COVID-19, and hereby infer potential implications for preventative screening. We therefore exploited the strong statistical power and time-resolved information on severe pneumonia events for more detailed analyses than was feasible with COVID-19. This led to three important observations. First, the infectious disease multi-biomarker score was largely independent of prevalent chronic respiratory and cardiometabolic diseases (Figure 3). Second, the susceptibility to severe pneumonia was drastically elevated in the extreme tail of the multi-biomarker infectious disease score, with 5–10 times higher risk compared to individuals with normal levels of the multi-biomarker score (Figure 5). Such features might aid in establishing thresholds for identifying individuals most susceptible to a severe disease course. Third, the odds ratio of the multi-biomarker score for severe pneumonia events occurring after 7–11 years closely matched that of severe COVID-19, for which all events occurred over decade after blood sampling (Figure 4A). Yet, screening for the susceptibility to severe COVID-19 would require a strong association with the short-term risk. When confining the analyses of severe pneumonia to events occurring within the first 2 years after blood sampling, the short-term risk elevation was over four times stronger than that observed for long-term risk — individuals with high levels of the multi-biomarker score were almost 7-times more susceptible than people with low levels (Figure 4B). If similar enhancement in short-term risk extend to COVID-19, our results could potentially indicate applications for identification of individuals at high susceptibility to a severe COVID-19 disease course. However, the unavailability of metabolic biomarker data from blood samples drawn shortly prior to the pandemic prevents us from examining biomarker associations with short-term COVID-19 susceptibility, and our results should therefore be considered of hypothesis generating nature.

We observed multiple blood biomarkers commonly linked with the risk for cardiovascular disease and diabetes (Soininen et al., 2015; Würtz et al., 2017; Holmes et al., 2018; Ahola-Olli et al., 2019) to also be associated with increased susceptibility to both severe pneumonia and severe COVID-19. The biomarkers span multiple metabolic pathways, including low concentrations of lipoprotein lipids, impaired fatty acid balance, decreased amino acid levels and high chronic inflammation. This is the first study to show that many of these blood biomarkers associate with susceptibility to severe infections, potentially indicating that fatty acids and amino acids should not be considered only as biomarkers for cardiometabolic risk. The associations of omega-3 and other fatty acids with the risk for severe COVID-19 may be particularly important, as these measures are more directly modifiable by lifestyle means than common markers of inflammation. The overall pattern of biomarker associations followed a characteristic metabolic signature reflective of an increased susceptibility to a severe infectious disease. This pattern of biomarker associations is broadly similar to what has previously been reported with the risk for all-cause mortality in smaller prospective cohort studies (Deelen et al., 2019). It is therefore unlikely that the identified biomarker signature is specific to the risk for severe pneumonia and COVID-19, or even specific to infectious diseases in general. We propose that the overall metabolic biomarker perturbations observed here reflect molecular signals of low-grade inflammation that exacerbate disease severity, in case of both infectious and chronic diseases (Akbar and Gilroy, 2020; Bonafè et al., 2020). In line with this, prior studies have demonstrated that elevated levels of GlycA, the biomarker with the strongest weight in the infectious disease score, is associated with increased neutrophil activity and the long-term risk for fatal infections (Ritchie et al., 2015). Such over-activity of immune response from pneumonia or COVID-19 infection is known to cause tissue damage and organ dysfunction through cytokine storm, a common complication of severe COVID-19 (Mangalmurti and Hunter, 2020). While the specific biological mechanisms underpinning the blood metabolic biomarker associations with chronic and infectious diseases remain poorly understood, we emphasize that the observational character of our study does not allow us to conclude whether the biomarkers are contributing causally to increase the risk or are merely indirect risk markers.

Replication of novel biomarker associations is a key aspect in observational studies. We are not aware of other prospective studies with sufficient COVID-19 hospitalisation events and NMR-based metabolic biomarker data to address this. However, a preprint of the present study featured analysis of 195 severe COVID-19 cases, based on data available in UK Biobank back in June 2020 (Julkunen et al., 2020). In the present updated analyses, with over three times the number of cases, all biomarker associations with susceptibility to severe COVID-19 were similar or stronger, and hereby provide a within-cohort replication of our initial findings. In addition, a recent study used the same metabolic biomarker panel in three cohorts of hospitalised patients and observed similar overall biomarker perturbations to be predictive of COVID-19 severity (Dierckx et al., 2020). The study also reported the multi-biomarker infectious disease score to be among the strongest biomarkers for discriminating COVID-19 severity among already hospitalised patients.

Our study has both strengths and limitations. Strengths include the large sample size, which enabled the analysis of biomarkers for susceptibility to severe COVID-19 based on pre-pandemic blood samples from general population settings. We used a validated metabolic profiling platform that enables simultaneous quantification of numerous metabolic biomarkers in a scalable low-cost setup. Although the number of hospitalised COVID-19 cases was in line with the prevalence in England, we acknowledge that the statistical power was limited for prediction analyses even with close to 100,000 samples linked with COVID-19 outcome data. Furthermore, the UK Biobank study participants are not fully representative of the UK population by demographic characteristics; the individuals were enrolled on a volunteer basis and are therefore more representative of healthier individuals than average (Sudlow et al., 2015; Fry et al., 2017). Even though this is generally not a concern for investigating risk associations (Keyes and Westreich, 2019), it does limit the statistical power to explore effects of ethnicity and old age. Other limitations include the decade long duration from blood sampling to the COVID-19 pandemic. While this limits inference on how well the biomarkers predict short-term risk for severe COVID-19, our analogy with long-term risk for severe pneumonia indicates that the time lag likely attenuates the biomarker association magnitudes substantially. Conversely, the remarkably strong associations for short-term risk of severe pneumonia led us to speculate that similar enhancements in association magnitudes could also hold for severe COVID-19. However, this inference should be further tested, in particular in the light of the bacterial origin of many severe pneumonia cases and the viral origin of COVID-19. Weaker biomarker associations for severe COVID-19 compared to severe pneumonia may also arise from the UK Biobank COVID-19 data being influenced by ascertainment bias in terms of differential healthcare seeking and differential testing (Griffith et al., 2020), whereas pneumonia is anticipated to have nearly complete case ascertainment (Ho et al., 2020).

In conclusion, a metabolic signature of perturbed blood biomarkers is associated with an increased susceptibility to both severe pneumonia and COVID-19 in blood samples collected a decade before the pandemic. The multi-biomarker score captures an elevated susceptibility to severe pneumonia within few years after blood sampling that is several times stronger than the risk elevation associated with many pre-existing health conditions, such as obesity and diabetes (Ho et al., 2020). If the three- to fourfold elevation in short-term risk compared to long-term risk of severe pneumonia also applies to severe COVID-19, then the metabolic biomarker profiling could potentially complement existing tools for identifying individuals most susceptible to a severe COVID-19 disease course. Regardless of the translational prospects, these results provide novel understanding on how metabolic biomarkers may reflect the susceptibility of severe COVID-19 and other infections.

The data are available for approved researchers from UK Biobank. The metabolic biomarker data has been released to the UK Biobank resource in March 2021.

1. 1. Cichońska A
   2. Julkunen H
   3. Würtz P

   (2021) Biobank

   ID 220. UK Biobank Nightingale biomarker data.

   https://biobank.ndph.ox.ac.uk/showcase/label.cgi?id=220

1. 1. Ahola-Olli AV
   2. Mustelin L
   3. Kalimeri M
   4. Kettunen J
   5. Jokelainen J
   6. Auvinen J
   7. Puukka K
   8. Havulinna AS
   9. Lehtimäki T
   10. Kähönen M
   11. Juonala M
   12. Keinänen-Kiukaanniemi S
   13. Salomaa V
   14. Perola M
   15. Järvelin MR
   16. Ala-Korpela M
   17. Raitakari O
   18. Würtz P

   (2019) Circulating metabolites and the risk of type 2 diabetes: a prospective study of 11,896 young adults from four finnish cohorts

   Diabetologia 62:2298–2309.

   https://doi.org/10.1007/s00125-019-05001-w

   • PubMed
   • Google Scholar
2. 1. Akbar AN
   2. Gilroy DW

   (2020) Aging immunity may exacerbate COVID-19

   Science 369:256–257.

   https://doi.org/10.1126/science.abb0762

   • PubMed
   • Google Scholar
3. 1. Almirall J
   2. Serra-Prat M
   3. Bolíbar I
   4. Balasso V

   (2017) Risk factors for Community-Acquired pneumonia in adults: a systematic review of observational studies

   Respiration 94:299–311.

   https://doi.org/10.1159/000479089

   • PubMed
   • Google Scholar
4. 1. Atkins JL
   2. Masoli JAH
   3. Delgado J
   4. Pilling LC
   5. Kuo C-L
   6. Kuchel GA
   7. Melzer D

   (2020) Preexisting comorbidities predicting COVID-19 and mortality in the UK biobank community cohort

   The Journals of Gerontology: Series A 75:2224–2230.

   https://doi.org/10.1093/gerona/glaa183

   • Google Scholar
5. 1. Bonafè M
   2. Prattichizzo F
   3. Giuliani A
   4. Storci G
   5. Sabbatinelli J
   6. Olivieri F

   (2020) Inflamm-aging: why older men are the most susceptible to SARS-CoV-2 complicated outcomes

   Cytokine & Growth Factor Reviews 53:33–37.

   https://doi.org/10.1016/j.cytogfr.2020.04.005

   • PubMed
   • Google Scholar
6. 1. Deelen J
   2. Kettunen J
   3. Fischer K
   4. van der Spek A
   5. Trompet S
   6. Kastenmüller G
   7. Boyd A
   8. Zierer J
   9. van den Akker EB
   10. Ala-Korpela M
   11. Amin N
   12. Demirkan A
   13. Ghanbari M
   14. van Heemst D
   15. Ikram MA
   16. van Klinken JB
   17. Mooijaart SP
   18. Peters A
   19. Salomaa V
   20. Sattar N
   21. Spector TD
   22. Tiemeier H
   23. Verhoeven A
   24. Waldenberger M
   25. Würtz P
   26. Davey Smith G
   27. Metspalu A
   28. Perola M
   29. Menni C
   30. Geleijnse JM
   31. Drenos F
   32. Beekman M
   33. Jukema JW
   34. van Duijn CM
   35. Slagboom PE

   (2019) A metabolic profile of all-cause mortality risk identified in an observational study of 44,168 individuals

   Nature Communications 10:3346.

   https://doi.org/10.1038/s41467-019-11311-9

   • PubMed
   • Google Scholar
7. Preprint

   1. Dierckx T
   2. van Elslande J
   3. Salmela H

   (2020) The metabolic fingerprint of COVID-19 severity

   medRxiv.

   https://doi.org/10.1101/2020.11.09.20228221

   • Google Scholar
8. 1. Fry A
   2. Littlejohns TJ
   3. Sudlow C
   4. Doherty N
   5. Adamska L
   6. Sprosen T
   7. Collins R
   8. Allen NE

   (2017) Comparison of Sociodemographic and Health-Related characteristics of UK biobank participants with those of the general population

   American Journal of Epidemiology 186:1026–1034.

   https://doi.org/10.1093/aje/kwx246

   • PubMed
   • Google Scholar
9. 1. Griffith GJ
   2. Morris TT
   3. Tudball MJ
   4. Herbert A
   5. Mancano G
   6. Pike L
   7. Sharp GC
   8. Sterne J
   9. Palmer TM
   10. Davey Smith G
   11. Tilling K
   12. Zuccolo L
   13. Davies NM
   14. Hemani G

   (2020) Collider Bias undermines our understanding of COVID-19 disease risk and severity

   Nature Communications 11:5749.

   https://doi.org/10.1038/s41467-020-19478-2

   • PubMed
   • Google Scholar
10. 1. Ho FK
    2. Celis-Morales CA
    3. Gray SR
    4. Katikireddi SV
    5. Niedzwiedz CL
    6. Hastie C
    7. Ferguson LD
    8. Berry C
    9. Mackay DF
    10. Gill JM
    11. Pell JP
    12. Sattar N
    13. Welsh P

    (2020) Modifiable and non-modifiable risk factors for COVID-19, and comparison to risk factors for influenza and pneumonia: results from a UK biobank prospective cohort study

    BMJ Open 10:e040402.

    https://doi.org/10.1136/bmjopen-2020-040402

    • PubMed
    • Google Scholar
11. 1. Holmes MV
    2. Millwood IY
    3. Kartsonaki C
    4. Hill MR
    5. Bennett DA
    6. Boxall R
    7. Guo Y
    8. Xu X
    9. Bian Z
    10. Hu R
    11. Walters RG
    12. Chen J
    13. Ala-Korpela M
    14. Parish S
    15. Clarke RJ
    16. Peto R
    17. Collins R
    18. Li L
    19. Chen Z
    20. China Kadoorie Biobank Collaborative Group

    (2018) Lipids, lipoproteins, and metabolites and Risk of Myocardial Infarction and Stroke

    Journal of the American College of Cardiology 71:620–632.

    https://doi.org/10.1016/j.jacc.2017.12.006

    • PubMed
    • Google Scholar
12. Preprint

    1. Julkunen H
    2. Cichońska A
    3. Nightingale Health UK Biobank Initiative

    (2020) Blood biomarker score identifies individuals at high risk for severe COVID-19 a decade prior to diagnosis: metabolic profiling of 105,000 adults in the UK biobank

    medRxiv.

    https://doi.org/10.1101/2020.07.02.20143685

    • Google Scholar
13. 1. Kermali M
    2. Khalsa RK
    3. Pillai K
    4. Ismail Z
    5. Harky A

    (2020) The role of biomarkers in diagnosis of COVID-19 - A systematic review

    Life Sciences 254:117788.

    https://doi.org/10.1016/j.lfs.2020.117788

    • PubMed
    • Google Scholar
14. 1. Keyes KM
    2. Westreich D

    (2019) UK Biobank, big data, and the consequences of non-representativeness

    The Lancet 393:1297.

    https://doi.org/10.1016/S0140-6736(18)33067-8

    • Google Scholar
15. 1. Khera AV
    2. Chaffin M
    3. Aragam KG
    4. Haas ME
    5. Roselli C
    6. Choi SH
    7. Natarajan P
    8. Lander ES
    9. Lubitz SA
    10. Ellinor PT
    11. Kathiresan S

    (2018) Genome-wide polygenic scores for common diseases identify individuals with risk equivalent to monogenic mutations

    Nature Genetics 50:1219–1224.

    https://doi.org/10.1038/s41588-018-0183-z

    • PubMed
    • Google Scholar
16. 1. Mangalmurti N
    2. Hunter CA

    (2020) Cytokine storms: understanding COVID-19

    Immunity 53:19–25.

    https://doi.org/10.1016/j.immuni.2020.06.017

    • PubMed
    • Google Scholar
17. 1. Messner CB
    2. Demichev V
    3. Wendisch D
    4. Michalick L
    5. White M
    6. Freiwald A
    7. Textoris-Taube K
    8. Vernardis SI
    9. Egger AS
    10. Kreidl M
    11. Ludwig D
    12. Kilian C
    13. Agostini F
    14. Zelezniak A
    15. Thibeault C
    16. Pfeiffer M
    17. Hippenstiel S
    18. Hocke A
    19. von Kalle C
    20. Campbell A
    21. Hayward C
    22. Porteous DJ
    23. Marioni RE
    24. Langenberg C
    25. Lilley KS
    26. Kuebler WM
    27. Mülleder M
    28. Drosten C
    29. Suttorp N
    30. Witzenrath M
    31. Kurth F
    32. Sander LE
    33. Ralser M

    (2020) Ultra-High-Throughput clinical proteomics reveals classifiers of COVID-19 infection

    Cell Systems 11:11–24.

    https://doi.org/10.1016/j.cels.2020.05.012

    • PubMed
    • Google Scholar
18. Website

    1. Resource UKBiobankD

    (2020) COVID-19 test results data

    Accessed February 3, 2021.

    http://biobank.ctsu.ox.ac.uk/crystal/exinfo.cgi?src=COVID19_tests;
19. 1. Ritchie SC
    2. Würtz P
    3. Nath AP
    4. Abraham G
    5. Havulinna AS
    6. Fearnley LG
    7. Sarin AP
    8. Kangas AJ
    9. Soininen P
    10. Aalto K
    11. Seppälä I
    12. Raitoharju E
    13. Salmi M
    14. Maksimow M
    15. Männistö S
    16. Kähönen M
    17. Juonala M
    18. Ripatti S
    19. Lehtimäki T
    20. Jalkanen S
    21. Perola M
    22. Raitakari O
    23. Salomaa V
    24. Ala-Korpela M
    25. Kettunen J
    26. Inouye M

    (2015) The biomarker GlycA is associated with chronic inflammation and predicts Long-Term risk of severe infection

    Cell Systems 1:293–301.

    https://doi.org/10.1016/j.cels.2015.09.007

    • PubMed
    • Google Scholar
20. 1. Shen B
    2. Yi X
    3. Sun Y
    4. Bi X
    5. Du J
    6. Zhang C
    7. Quan S
    8. Zhang F
    9. Sun R
    10. Qian L
    11. Ge W
    12. Liu W
    13. Liang S
    14. Chen H
    15. Zhang Y
    16. Li J
    17. Xu J
    18. He Z
    19. Chen B
    20. Wang J
    21. Yan H
    22. Zheng Y
    23. Wang D
    24. Zhu J
    25. Kong Z
    26. Kang Z
    27. Liang X
    28. Ding X
    29. Ruan G
    30. Xiang N
    31. Cai X
    32. Gao H
    33. Li L
    34. Li S
    35. Xiao Q
    36. Lu T
    37. Zhu Y
    38. Liu H
    39. Chen H
    40. Guo T

    (2020) Proteomic and metabolomic characterization of COVID-19 patient sera

    Cell 182:59–72.

    https://doi.org/10.1016/j.cell.2020.05.032

    • Google Scholar
21. 1. Soininen P
    2. Kangas AJ
    3. Würtz P
    4. Suna T
    5. Ala-Korpela M

    (2015) Quantitative serum nuclear magnetic resonance metabolomics in cardiovascular epidemiology and genetics

    Circulation: Cardiovascular Genetics 8:192–206.

    https://doi.org/10.1161/CIRCGENETICS.114.000216

    • PubMed
    • Google Scholar
22. 1. Sudlow C
    2. Gallacher J
    3. Allen N
    4. Beral V
    5. Burton P
    6. Danesh J
    7. Downey P
    8. Elliott P
    9. Green J
    10. Landray M
    11. Liu B
    12. Matthews P
    13. Ong G
    14. Pell J
    15. Silman A
    16. Young A
    17. Sprosen T
    18. Peakman T
    19. Collins R

    (2015) UK biobank: an open access resource for identifying the causes of a wide range of complex diseases of middle and old age

    PLOS Medicine 12:e1001779.

    https://doi.org/10.1371/journal.pmed.1001779

    • PubMed
    • Google Scholar
23. 1. Williamson EJ
    2. Walker AJ
    3. Bhaskaran K

    (2020) OpenSAFELY: factors associated with COVID-19 death in 17 million patients

    Nature 584:430–436.

    https://doi.org/10.1038/s41586-020-2521-4

    • Google Scholar
24. 1. Würtz P
    2. Kangas AJ
    3. Soininen P
    4. Lawlor DA
    5. Davey Smith G
    6. Ala-Korpela M

    (2017) Quantitative Serum Nuclear Magnetic Resonance Metabolomics in Large-Scale Epidemiology: A Primer on -Omic Technologies

    American Journal of Epidemiology 186:1084–1096.

    https://doi.org/10.1093/aje/kwx016

    • PubMed
    • Google Scholar
25. 1. Zhou F
    2. Yu T
    3. Du R
    4. Fan G
    5. Liu Y
    6. Liu Z
    7. Xiang J
    8. Wang Y
    9. Song B
    10. Gu X
    11. Guan L
    12. Wei Y
    13. Li H
    14. Wu X
    15. Xu J
    16. Tu S
    17. Zhang Y
    18. Chen H
    19. Cao B

    (2020) Clinical course and risk factors for mortality of adult inpatients with COVID-19 in Wuhan, China: a retrospective cohort study

    The Lancet 395:1054–1062.

    https://doi.org/10.1016/S0140-6736(20)30566-3

    • PubMed
    • Google Scholar

Acceptance summary:

The authors show that in over 100,000 healthy individuals a baseline multi-biomarker score comprised of 25 proteins, fatty acids, amino acids and lipids obtained simultaneously from a single sample strongly predicts the risk of severe pneumonia and COVID 19 hospitalization 7-11 years later. These findings now been validated in Belgian patients provide novel insights for both pneumonia and severe COVID -19 risk.

Decision letter after peer review:

Thank you for submitting your article "Metabolic biomarker profiling for identification of high risk for severe pneumonia and COVID-19 hospitalisation" for consideration by eLife. Your article has been reviewed by 3 peer reviewers, including Edward D Janus as the Reviewing Editor and Reviewer #3, and the evaluation has been overseen by a Senior Editor. The following individual involved in review of your submission has agreed to reveal their identity: Harin Karunajeewa (Reviewer #2).

The reviewers have discussed the reviews with one another and the Reviewing Editor has drafted this decision to help you prepare a revised submission.

We would like to draw your attention to changes in our revision policy that we have made in response to COVID-19 (https://elifesciences.org/articles/57162). Specifically, when editors judge that a submitted work as a whole belongs in eLife but that some conclusions require a modest amount of additional new data, as they do with your paper, we are asking that the manuscript be revised to either limit claims to those supported by data in hand, or to explicitly state that the relevant conclusions require additional supporting data.

Our expectation is that the authors will eventually carry out the additional experiments and report on how they affect the relevant conclusions either in a preprint on bioRxiv or medRxiv, or if appropriate, as a Research Advance in eLife, either of which would be linked to the original paper.

Summary:

The authors report that in over 100,000 healthy individuals a baseline multi-biomarker score comprised of 25 proteins, fatty acids, amino acids and lipids strongly predicts the risk of severe pneumonia and COVID 19 hospitalization 7-11 years later. The issue is whether other parameters related to inflammatory or immunological processes have been excluded. Also there are many simpler scores for determining severity and outcomes in clinical practice.

Thus there are two aspects the population susceptibility and the prediction of severe disease. There is a need to change the comparison group if looking for a useful predictor of severe disease.

The paper needs to be revised to highlight its hypothesis generation aspects which is its major point of interest, but also to cover or at least discuss the broader context of other inflammatory, immunological markers and other patient susceptibility characteristics and limitations including whether there is actual clinical utility.

Essential revisions:

This is a dense and complex analysis of a large, rich and powerful dataset that examines a clinically relevant issue that is obviously extremely timely. As an observational study it is limited to "hypothesis-generating conclusions". Its findings are intriguing and potentially of high medical and scientific importance.

The manuscript identifies a group of biomarkers that is highly predictive of severe pneumonia and COVID-19 hospitalization in prospective samples that were collected many years earlier. There are several strengths of the manuscript, including the large samples size using the UK biobank. The inquiry of a consistent set of markers that predict severe response to multiple important outcomes is also important. However, there are also not notable issues with the manuscript that need to be addressed and we believe inaccurate conclusions have been drawn from the analyses.

1. Intuitively one would expect markers of inflammation and immunological susceptibility to be the strongest predictors in this disease situation. Have these been explored previously? Is it that you are testing the already available marker set heavily slanted to metabolic markers because that is what is available so its wider use is being explored?

2. The primary concern with the manuscript as written is that the controls being used in both analyses do not seem appropriate. It is unclear why controls are being used in the analyses vs. those with pneumonia without a severe response or those with COVID-19 without a severe reaction. Aren't those the real comparison groups? By just using random controls, the analysis is not taking into account that disease itself. This is a fundamental flaw in the way the analyses are presented and the conclusions drawn. Furthermore, it completely inflates that actual statistical power in the analyses, given that the control group is literally several magnitudes bigger in the case of COVID-19.

3. Having said that, at an absolute minimum, these analyses should be rerun using COVID-19+ with severe and non-severe response. In this case, the real question is being answered: using prospective samples or individuals who get pneumonia/COVID, can we predict this that have a severe response vs. those who do not? In this case we are actually deciphering the difference between the biology of individuals who we know will have a severe response and those who we know will not. In the analyses, that the authors do, we have no idea who in the control group would have a severe response and who would not. Therefore the control group is completely inappropriate for addressing that issue.

4. In the case where one is using severe/non-severe, it should be acknowledged that the actual statistical power of the tru comparison is substantially lower that what is being used with the larger UK biobank samples.

5. Some approach to assess the robustness of these findings in an additional population would be ideal; however, it is acknowledged that this may not be feasible currently.

6. Please advise specifically on this point – that the metabolomic data has been transparently documented prior to the linked UK biobank data being unlocked, so that it can be matched up with that present in the final linked dataset – ie verifies that authors were blinded and hasn't changed since unblinded. The UK biobank policies and procedures hopefully address these issues of research integrity.

7. Please expand on clinical utility. The tests and score would need to be affordable and readily available and compliment existing predictors – not only cardiometabolic diseases but many others and specifically age and be cost effective.

A key line in the introduction outlines the study's rationale as its potential value for risk stratification as "facilitating the identification of high-risk individuals missed by current clinical tools". I think the authors should therefore address this point more explicitly when framing their results in the conclusion. To what extent do their results really support this statement? This could take the form of providing important context as to the predictive ability of existing or proposed clinical risk stratification tools. There has been an explosion of these in relation to COVID, many of which show that easily collected clinical and demographic information can predict very large differences in risk of hospitalization or death – eg Halalau et al. Annals of Internal Medicine quote an odds ratio of 19 for risk of hospitalization in their highest risk strata. So how much would metabolomic profiling really add to information much more readily and easily accessible without a blood test and expensive laboratory analysis?

8. The authors need to acknowledge the limitations of the clinical data that has been used to adjust comparisons. For instance they have adjusted for cardiac and respiratory disease but there is no mention of dementia, general "frailty, residence in a nursing home and measures of independent function which are all very highly associated with the risk of pneumonia. Could the high risk metabolic profiles lie on the same causal pathway and therefore just be surrogates for these easily-measured biological risk factors? Again, goes to the question of what metabolomic profiling could add to much more readily available clinical and demographic information.

Summary:

The authors report that in over 100,000 healthy individuals a baseline multi-biomarker score comprised of 25 proteins, fatty acids, amino acids and lipids strongly predicts the risk of severe pneumonia and COVID 19 hospitalization 7-11 years later. The issue is whether other parameters related to inflammatory or immunological processes have been excluded. Also there are many simpler scores for determining severity and outcomes in clinical practice.

Thus there are two aspects the population susceptibility and the prediction of severe disease. There is a need to change the comparison group if looking for a useful predictor of severe disease.

The paper needs to be revised to highlight its hypothesis generation aspects which is its major point of interest, but also to cover or at least discuss the broader context of other inflammatory, immunological markers and other patient susceptibility characteristics and limitations including whether there is actual clinical utility.

We thank the editors and reviewers for their valuable comments. We have now revised the paper according to the comments. We have also shifted the wording to emphasise the population susceptibility rather than the prediction aspect, and toned down some of the initially proposed public health implications to be more in line with current COVID-19 preventative policies and vaccine roll outs in many countries.

We would like to emphasize two points that substantially strengthen the revised manuscript:

1. We have re-analysed biomarker associations with severe COVID-19 based on updated COVID-19 data from UK Biobank. This update more than triples the number of severe COVID-19 cases (from 195 to 652). All association magnitudes are similar or stronger in these re-analyses, and the statistical significance is therefore substantially improved: all analyses of the multi-biomarker infectious disease score vs severe COVID-19 risk are now significant and many individual biomarkers are also now significant (Figure 6). The updated analysis thus provides a within-cohort replication of our initial results. The updated analysis also overcomes a prime limitation of the initial submission, since the multi-biomarker score associations with severe COVID-19 are now robust to adjustments and omission of individuals with prevalent diseases (Figure 8).

2. After our initial submission, a medRxiv preprint (Dierckx et al. (2020), https://doi.org/10.1101/2020.11.09.20228221) has provided independent validation of our biomarker findings in hospital settings: this preprint examines the same metabolic biomarker panel in relation to COVID-19 severity in three cohorts of COVID‑19 patients in Belgian hospitals. Their results confirm that all the key changes in the metabolic biomarkers and the “infectious disease biomarker score” that we propose to be predictive of severe COVID-19 among initially disease-free individuals, are also discriminating COVID-19 severity among already hospitalised patients.

Essential revisions:

This is a dense and complex analysis of a large, rich and powerful dataset that examines a clinically relevant issue that is obviously extremely timely. As an observational study it is limited to "hypothesis-generating conclusions". Its findings are intriguing and potentially of high medical and scientific importance.

The manuscript identifies a group of biomarkers that is highly predictive of severe pneumonia and COVID-19 hospitalization in prospective samples that were collected many years earlier. There are several strengths of the manuscript, including the large samples size using the UK biobank. The inquiry of a consistent set of markers that predict severe response to multiple important outcomes is also important. However, there are also not notable issues with the manuscript that need to be addressed and we believe inaccurate conclusions have been drawn from the analyses.

1. Intuitively one would expect markers of inflammation and immunological susceptibility to be the strongest predictors in this disease situation. Have these been explored previously?

Many inflammatory biomarkers have been shown to predict COVID-19 severity among infected individuals, but only few studies have examined molecular markers of inflammation in relation to future COVID-19 risk among healthy individuals. This is likely due to the challenge of attaining the kind of study design we present here, as it requires that the biomarkers are measured from blood samples drawn prior to the pandemic. Studies with similar design based on UK Biobank have shown that hsCRP and blood cell counts are only modest predictors of COVID-19 hospitalisation and pneumonia (Ho et al., BMJ Open 2020); and those results are weaker than many of the metabolic biomarkers analysed in the present study (p. 7).

Glycoprotein acetyls (GlycA), the primary marker of low-grade inflammation in the metabolic biomarker panel analysed in our study, has indeed previously been associated with long-term risk for severe infections (Ritchie et al., Cell Systems 2015). Also PUFA% has been associated with risk of death from non-localised infections (Deelen et al., Nature Communications 2019), but no prior studies have examined the wider panel of metabolic biomarkers quantified by the NMR platform in relation to infectious diseases. Although GlycA was the strongest individual biomarker for the risk of severe pneumonia, this was not the case for the risk of severe COVID‑19 (for which blood levels of DHA and monounsaturated fatty acids were the strongest biomarkers). In addition to the expected results on the GlycA inflammatory biomarker, our study demonstrates that many cardiometabolic biomarkers, such as fatty acids and amino acids, can also be predictive of severe infectious disease risk among generally healthy individuals (p. 6). We believe that the novel evidence on how the susceptibility to severe infectious diseases is reflected in the systemic metabolic profile could broaden the understanding of impaired immunity beyond the focus on established inflammatory biomarkers. In addition, we show that the combination of many metabolic biomarkers measured simultaneously from a single assay provides stronger associations for infectious disease susceptibility than individual biomarkers.

Is it that you are testing the already available marker set heavily slanted to metabolic markers because that is what is available so its wider use is being explored?

All the metabolic markers analysed in our study are indeed obtained simultaneously in a single assay using NMR spectroscopy. We have now clarified this point in the manuscript (p. 3 and 7). The measurements of these metabolic biomarkers in >100,000 samples from the UK Biobank were undertaken prior to the COVID-19 pandemic. It is thus a pre-specified set of biomarkers that has been related primarily to risk for cardiovascular disease and diabetes in earlier epidemiological studies based on smaller cohorts.

The reason for focusing on this particular set of metabolic biomarkers was two-fold: first, we wanted to investigate whether these biomarkers of systemic metabolism, and commonly described as cardiometabolic risk markers, could be also associated with susceptibility to a severe infectious disease course in general population settings. The second reason was that these biomarkers can be measured simultaneously in a single blood assay at low cost. This allows using multi-biomarker scores for enhanced risk prediction without additional costs. These points can both provide molecular insights into COVID-19 etiology, and potentially open for novel tools to support identification of individuals who are most susceptible. The panel of biomarkers shown in Figures 2 and 6 are those from the employed NMR platform that have been certified for clinical use and which would therefore facilitate potential translational applications, as now noted on p. 3 and 7.

2. The primary concern with the manuscript as written is that the controls being used in both analyses do not seem appropriate. It is unclear why controls are being used in the analyses vs. those with pneumonia without a severe response or those with COVID-19 without a severe reaction. Aren't those the real comparison groups? By just using random controls, the analysis is not taking into account that disease itself. This is a fundamental flaw in the way the analyses are presented and the conclusions drawn. Furthermore, it completely inflates that actual statistical power in the analyses, given that the control group is literally several magnitudes bigger in the case of COVID-19.

The use of the entire non-case study population as controls in our statistical analyses is the established approach for biomarker association testing with disease risk in the general population settings – regardless of whether it is for infectious diseases, cardiovascular disease or any other disease. The same selection of controls is used by many other COVID-19 related publications based on UK Biobank (e.g. Petermann-Rocha et al., BMC Medicine 2020). The choice of using all non-cases as controls, rather than only test-positive mild cases, is also taken in high profile papers on risk factors associated with COVID-19-related death, e.g. the OpenSAFELY study on determining the relative risk ascribed to the prevalence of many chronic conditions (Williamson et al., Nature 2020).

We are convinced that the chosen control group and the statistical analysis framework is the appropriate one to address study question we are after: for a person with a given value of the biomarker score, what is the risk increase for the disease outcome (severe pneumonia or COVID-19 hospitalisation) compared to the risk in a reference population (such as people in the general population with median levels of that biomarker score). This is the same principle as when risk factors for COVID-19 mortality are reported, e.g. the two-fold risk increase linked to class III obesity vs people with normal weight (Williamson et al., Nature 2020). The chosen statistical analyses can thus be used to identify individuals who are at increased risk compared to their peers according to their metabolic biomarker profile. We have now revised the wording throughout the manuscript to emphasize that we estimate the susceptibility for the disease outcome with respect to that person’s peers, rather than the relative risk of a given person for getting severe disease vs mild disease. We have also clarified the selection of control group in detail on p. 8.

The statistical power in the analyses is dictated by the number of cases when there is strong class-imbalance between the number of cases and controls. We address this point further below in answer #4.

3. Having said that, at an absolute minimum, these analyses should be rerun using COVID-19+ with severe and non-severe response. In this case, the real question is being answered: using prospective samples or individuals who get pneumonia/COVID, can we predict this that have a severe response vs. those who do not? In this case we are actually deciphering the difference between the biology of individuals who we know will have a severe response and those who we know will not. In the analyses, that the authors do, we have no idea who in the control group would have a severe response and who would not. Therefore the control group is completely inappropriate for addressing that issue.

As reasoned above, we are answering a slightly different study question than whether an individual will have a severe response or not; instead, we deliberately address the study question of the elevated risk for severe disease relative to peers from the general population as the reference group. We have now framed this point as identification of individuals at increased susceptibility in the general population, e.g. in the title and the abstract of the revised manuscript.

The chosen study design was also deemed necessary due to the limited COVID-19 testing performed in the UK at the time of the analysis. At the time of submission, the COVID-19 testing was heavily focused on healthcare workers and those with severe symptoms. Therefore, the subset of UK biobank participants who were tested for COVID-19 were far from being a random sample of the study participants. This is known to cause bias and potentially spurious results, and it prevents us from robustly addressing the question of the risk for severe vs mild response for an infected individual. For instance, being a health care worker influences the likelihood of being tested at all, and in turn more healthcare workers than average end of being tested positive – if health care workers generally have more healthy biomarker profiles than mild cases who never get tested in the general population, it can inflate the differences between mild and severe cases.

The various biases introduced by different case-control designs of COVID-19 analyses in the UK Biobank have been described in detail recently (Griffith et al., Nature Communications 2020); we have now referred to this in the revised manuscript (p. 7 and 8). In the case of the suggested mild vs severe cases design the authors note: when sampling conditional on having a positive test for COVID-19 infection and testing is unlikely to be random, then conditioning on the positive test result introduces bias by factors causing severe infections as well as those causing increased likelihood of testing. The ascertainment bias induced by these factors would have therefore caused the associations to not reflect patterns in the general population of interest. Please also see our response above explaining the study question we aim to address with the chosen study design.

Finally, we note that the fraction of the population with severe COVID-19 in UK Biobank largely follow the hospitalisation rate in the UK, see Figure 1—figure supplement 1. Only a small fraction of hospitalised cases is likely to be missed (due to lack of inpatient positive PCR test); however, many mild cases continue to be missed throughout the pandemic, which further complicates addressing the study design and related scientific questions suggested by the reviewers.

4. In the case where one is using severe/non-severe, it should be acknowledged that the actual statistical power of the tru comparison is substantially lower that what is being used with the larger UK biobank samples.

Please see our response above explaining the reasoning behind the chosen study design.

We acknowledge that the statistical power is dictated by the number of cases rather than the total of the study population. We now further note this in the revised manuscript (p. 7).

The tripling of the number of severe COVID-19 cases in our analyses of the most recent UK Biobank alleviates this limitation. Nonetheless, we would like to emphasise that a sample size of >100,000 individuals is required to obtain sufficient number of cases in the prospective population-based study design that we present – and that this a major novelty compared to studies based on samples from already infected individuals.

5. Some approach to assess the robustness of these findings in an additional population would be ideal; however, it is acknowledged that this may not be feasible currently.

We agree that replication and validation of the findings are essential for biomarker studies. No replication data existed at the time of submission, owing to difficulties in accessing pre-pandemic blood samples with subsequent COVID-19 infection. This has now changed, with two sources providing replication and validation of our initial findings as detailed in a new paragraph on p. 6.

First, UK Biobank updates the information on COVID-19 hospitalisations on a monthly basis. This has more than tripled the number of cases with severe COVID-19 since the initial submission, from 195 to 652 individuals with metabolic biomarker data. When re-analysing the data using the enhanced number of severe COVID-19 cases, all association magnitudes are now similar or even stronger. This provides within-cohort replication of our results. Second, our findings have later on been supported by study on Belgian COVID-19 patient cohorts (Dierckx et al. 2020, MedrXiv preprint). The authors report strong associations with COVID-19 severity for the same set of metabolic biomarkers that we observe, e.g. inflammatory markers, amino acids and fatty acids. The multibiomarker score derived in the present study, “Infectious Disease Score”, was also found to be strongly associated with COVID-19 severity in these already infected patients (odds ratios of 4 and 4.5 with COVID-19 for having the most severe disease course, compared to mild-albeit-hospitalised disease course, in both patient cohorts with such data available). These results provide further support to the robustness of our findings.

6. Please advise specifically on this point – that the metabolomic data has been transparently documented prior to the linked UK biobank data being unlocked, so that it can be matched up with that present in the final linked dataset – ie verifies that authors were blinded and hasn't changed since unblinded. The UK biobank policies and procedures hopefully address these issues of research integrity.

The measurements of the metabolic biomarker data have been conducted blinded prior to linkage to the UK Biobank health outcomes. We have now clarified this point in the revised manuscript. We also emphasized that all the metabolic biomarker data analyzed for this paper are becoming available for the research community through UK Biobank, with data released on 23 March 2021 (p. 8 and 10).

We would like to note that the metabolic biomarker measurements have been undertaken in accordance with a prespecified protocol approved by UK Biobank’s biomarker advisory committee and UK Biobank laboratory. This includes detailed quality control assurance using blinded duplicate blood samples measured throughout the project. These aspects are detailed along with the release of the metabolic biomarker data for the research community by UK Biobank.

7. Please expand on clinical utility. The tests and score would need to be affordable and readily available and compliment existing predictors – not only cardiometabolic diseases but many others and specifically age and be cost effective.

We have toned down the wording on clinical utility throughout in the revised manuscript to account for the latest public health policy developments. Instead, we have emphasized the novel biological insights gained from the analyses of both pneumonia and severe COVID-19 risk.

The re-analyses of the COVID-19 data to the latest follow-up in UK Biobank, which more than tripled the number of cases with severe disease, reinforces the utility to identify high-risk individuals over and above existing risk factors. This includes age, which is accounted for in all the models. Nonetheless, the real-world clinical utility relies on the premise that the 3-fold increase in short-term risk observed for severe pneumonia, compared to long-term risk, also applies for risk of severe COVID-19. The inherently long follow-up time between UK Biobank blood sampling and COVID-19 prevents us from demonstrating this point with currently available data. We have acknowledged this limitation in the manuscript; further studies with shorter time between blood sampling and COVID-19 are under way, but accessing large amounts of samples with ascertained COVID-19 hospitalized and blood drawn prior to the pandemic is challenging.

With a view to translation, we would like to highlight that employed NMR platform has received various regulatory approvals, and the 37 biomarkers included in the present study have been certified for diagnostics use.

A key line in the introduction outlines the study's rationale as its potential value for risk stratification as "facilitating the identification of high-risk individuals MISSED BY CURRENT CLINICAL TOOLS". I think the authors should therefore address this point more explicitly when framing their results in the conclusion. To what extent do their results really support this statement? This could take the form of providing important context as to the predictive ability of existing or proposed clinical risk stratification tools. There has been an explosion of these in relation to COVID, many of which show that easily collected clinical and demographic information can predict very large differences in risk of hospitalization or death – eg Halalau et al. Annals of Internal Medicine quote an odds ratio of 19 for risk of hospitalization in their highest risk strata. So how much would metabolomic profiling really add to information much more readily and easily accessible without a blood test and expensive laboratory analysis?

We have now reworded the introduction and toned down the claims on potential clinical utility throughout, as described for the previous point. Despite the improved results in Figure 8 in the revised manuscript, the specific utility compared to other risk factors rely on the assumption of stronger associations for short-term risk of severe COVID-19, as demonstrated for short-term risk of severe pneumonia. We have also strongly clarified this point in the Abstract, Introduction and Discussion (p. 1, 2, 5 and 7).

We agree that many studies have suggested risk stratification tools for patients with acute COVID-19 infection, such as the cited paper by Halalau et al. That is very different from our study, which aims to identify high risk individuals prior to being infected, i.e. the susceptibility to a severe disease course in case they are infected. The risk differences are generally much smaller in this type of study setting. One of the most highly cited papers with such study setting is Williamson et al. Nature 2020, where they report hazard ratios of ~2-3 for many diseases that are commonly used to shift middle-aged people into the group of “high-risk individuals”, such as uncontrolled diabetes, class III obesity, and newly diagnosed cancer. The results we present on Figure 8 are of similar hazard ratios, and based on our hypothesis generating results for short-term risk of severe pneumonia, the odds ratios are likely to be considerably stronger magnitude for short-term risk.

8. The authors need to acknowledge the limitations of the clinical data that has been used to adjust comparisons. For instance they have adjusted for cardiac and respiratory disease but there is no mention of dementia, general "frailty, residence in a nursing home and measures of independent function which are all very highly associated with the risk of pneumonia. Could the high risk metabolic profiles lie on the same causal pathway and therefore just be surrogates for these easily-measured biological risk factors? Again, goes to the question of what metabolomic profiling could add to much more readily available clinical and demographic information.

We have now revised the analyses to also account for presence of dementia. The results were essentially unaltered because of this – in fact, none of the 652 severe COVID-19 cases had a dementia diagnosis at the time of blood sampling. This point reflects the general characteristics of the study population in the UK Biobank, with baseline age 40–70 and volunteer enrollment skewing the disease prevalence towards more healthy than average. All study participants had to be able to attend the assessment centres by their own means, and there was no enrollment at nursing homes. We have now clarified these points in the revised manuscript (p 4).

We agree that the metabolic biomarkers highlighted in our study correlate with ageing, and that they likely provide a molecular reflection of frailty, immune response and overall susceptibility to severe disease. However, the associations with future risk for pneumonia and severe COVID-19 were only weakly attenuated after accounting for prevalent cardiovascular disease, diabetes, lung cancer, COPD, liver diseases, renal failure and dementia – we are therefore convinced that they do reflect aspects of frailty and disease risk that is not easily captured by readily available clinical and demographic information.

Author details

1. Heli Julkunen

   Nightingale Health Plc, Helsinki, Finland

   Contribution

   Conceptualization, Data curation, Software, Formal analysis, Investigation, Visualization, Methodology, Writing - original draft, Writing - review and editing

   Competing interests

   HJ is employee and holds stock options with Nightingale Health Plc.

   "This ORCID iD identifies the author of this article:" 0000-0002-4282-0248

2. Anna Cichońska

   Nightingale Health Plc, Helsinki, Finland

   Contribution

   Conceptualization, Data curation, Software, Formal analysis, Supervision, Investigation, Visualization, Methodology, Writing - original draft, Project administration

   Competing interests

   AC is employee and holds stock options with Nightingale Health Plc.

3. P Eline Slagboom

   1. Molecular Epidemiology, Department of Biomedical Data Sciences, Leiden University Medical Center, Leiden, Netherlands
   2. Max Planck Institute for Biology of Ageing, Cologne, Germany

   Contribution

   Investigation, Methodology, Writing - original draft, Writing - review and editing

   Competing interests

   No competing interests declared

4. Peter Würtz

   Nightingale Health Plc, Helsinki, Finland

   Contribution

   Conceptualization, Supervision, Funding acquisition, Investigation, Visualization, Methodology, Writing - original draft, Project administration, Writing - review and editing

   For correspondence

   peter.wurtz@nightingalehealth.com

   Competing interests

   PW is employee and shareholder of Nightingale Health Plc.

   "This ORCID iD identifies the author of this article:" 0000-0002-5832-0221

5. Nightingale Health UK Biobank Initiative

   Nightingale Health Plc, Helsinki, Finland

• 4,577

  Page views

• 442

  Downloads

• 74

  Citations

Article citation count generated by polling the highest count across the following sources: Crossref, Scopus, PubMed Central.