Severe Hospitalization-Requiring Viral Infection With Influenza or COVID-19: Metabolic Pathway Analysis

Abstract

Background

Influenza and SARS-CoV-2 can cause severe respiratory failure, but the metabolic pathways that lead to clinical deterioration are not fully uncovered. Tryptophan catabolism has been linked to disease progression and adverse outcomes. We aimed to find out whether the link between tryptophan catabolism and disease progression is shared by the 2 viral infections and, in an exploratory manner, to assess other metabolic pathways.

Methods

Adults hospitalized due to influenza or SARS-CoV-2 from 3 prospective studies were pooled in a nested case-control study design. Cases were defined by disease progression: an increase in oxygen supplementation, intensive care unit admission, or death within 28 days. Cases were matched 1:2 to nonprogressors by pathogen and initial disease severity. We tested associations of plasma kynurenine, tryptophan, and the kynurenine/tryptophan ratio with disease progression. Metabolic profiles were investigated by unsupervised clustering-based, pathway-resolved methods.

Results

We included 303 patients hospitalized with influenza or SARS-CoV-2. Higher levels of kynurenine and higher kynurenine/tryptophan ratios were associated with disease progression (odds ratio per log₂ increase [95% CI], 1.81 [1.21–2.70] and 1.89 [1.26–2.84], respectively) independent of pathogen. Two metabolite modules were associated with disease progression. One module contained multiple amino acids, including kynurenine and 10 other tryptophan catabolism metabolites. The other module contained mainly lipids and xenobiotics.

Conclusions

Several groups of metabolites were associated with disease progression independent of the pathogen. This indicates that biological mechanisms related to disease severity are shared in influenza and COVID-19. These mechanisms could be used for risk stratification of patients for potential disease-modifying treatments.

Respiratory viral infections are major contributors to morbidity and mortality worldwide [1] and have the potential of leading to pandemics, most notably the COVID-19 and 2009 H1N1 swine flu pandemics. Despite virologic differences across respiratory pathogens, there are many similarities in their clinical presentation, such as the risk of respiratory failure, the need for oxygen supplementation, and variation in disease severity across patients. Although most influenza and SARS-CoV-2 infections cause only minor symptoms, some of those infected require hospitalization, and disease progression can involve organ complications, intubation, and death. The biological mechanisms that lead to disease progression and unfavorable outcomes are largely uncovered. The gain from more knowledge on these mechanisms will be twofold: it will increase the understanding of pathogenesis in severe respiratory viral infections, and it will offer a potential for novel therapies targeting such pathways to improve the prognosis.

Previous studies have investigated the indoleamine-2,3-dioxygenase (IDO1)–mediated depletion of tryptophan to kynurenine (also known as tryptophan catabolism) in influenza [2–4] and COVID-19 [5–8]. These studies have consistently found low levels of tryptophan, high levels of kynurenine, and/or a high kynurenine to tryptophan (K/T) ratio in the blood associated with disease progression and/or death. Tryptophan is 1 of 8 essential amino acids that must be supplied through diet [9]. Most tryptophan in mammals is degraded through the kynurenine pathway, where the first and rate-limiting step is mediated by the IDO1 enzyme. IDO1 is activated by interferons during inflammation [2, 10] and has been proposed as a dual function that may be pathogen specific: one being tryptophan starvation of the pathogen to limit viral replication and disease, the other being immune dampening to allow infectious tolerance and prevent damage to the body caused by immune activation [11, 12]. Kynurenine has been proposed to be contributing to this infectious tolerance by avoiding hyperinflammation responses [13]. Kynurenine has been shown to help with differentiation and function of T_reg cells through induction of FoxP3 expression and in promoting the differentiation of proinflammatory Th17 cells through inhibition of RORγ [14]. The full IDO1-mediated conversion of the tryptophan pathway is involved in the de novo biosynthesis of NAD⁺, which regulates macrophage immune function [5] and contributes to T-cell differentiation [14]. These findings indicate that an active tryptophan-to-kynurenine conversion in the blood of patients infected with influenza or SARS-CoV-2 might reflect a virus-driven dampened immune response, allowing for more severe infections and disease outcomes.

Other studies have investigated the whole metabolome (all measurable small molecules) in blood and identified unique metabolic signatures of survivors of influenza H1N1 as compared with nonsurvivors [15], metabolic signatures of acute respiratory distress syndrome in H1N1 influenza pneumonia [16], and metabolic signatures of COVID-19 disease severity [17, 18]. This strengthens the belief that metabolites, groups of metabolites, or metabolic pathways could be associated with similar disease outcomes for both pathogens and may reveal shared metabolic disruptions rendering certain people more susceptible to disease progression than others. However, previous results are based on single-pathogen cohorts (influenza or SARS-CoV-2) and most with relatively small sample sizes, single-country populations, or animal or cell models. Given the importance of understanding the underlying common biological mechanisms, we aimed to investigate shared metabolic patterns of disease progression in a clinically well-defined international cohort consisting of people hospitalized with influenza or SARS-CoV-2 infection. Our objectives were to uncover biological pathways suspected to lead to unfavorable outcomes: specifically, (1) to assess the kynurenine pathway through levels of kynurenine, tryptophan, and the K/T ratio as well as (2) to explore the whole metabolome and lipidome to find potential new associations of metabolites and pathways with disease progression.

METHODS

Study Population

The study population included adults hospitalized due to influenza or SARS-CoV-2. Individuals were recruited from 3 independent cohorts:

• The international observational study INSIGHT FLU003 (ClinicalTrials.gov NCT01056185) [19] consisting of patients with influenza and updated to include COVID-19 cases following the onset of the pandemic in early 2020

• An observational study in patients with COVID-19 performed at Copenhagen University Hospital, Hvidovre, Denmark [20]

• The randomized clinical trial ProPAC (EudraCT 2020-001198-55) with patients infected with COVID-19 from several hospitals in Denmark [21]

Of note, the INSIGHT FLU003 was used for a prior study exploring tryptophan catabolism and clinical outcomes [3], although no individuals in the present study were analyzed previously.

Participants were divided into categories based on their requirement for oxygen supplementation at time of hospital admission: 1, no supplemental oxygen; 2, face mask or nasal cannula; 3, high-flow nasal cannula, noninvasive ventilation, continuous positive airway pressure, or bilevel positive airway pressure; 4, invasive mechanical ventilation or extracorporeal membrane oxygenation. Categories were based on prior findings [20, 22]. A low initial risk of disease progression was defined as supplemental oxygen categories 1 and 2 (low-risk group) and a high initial risk of disease progression as oxygen categories 3 and 4 (high-risk group). Cases (hereinafter, progressors) were those who progressed to a worse disease state within 28 days of enrollment, defined as a higher level of oxygen supplementation vs that at enrollment (eg, progressing from oxygen category 1 to 2 or from 2 to 4), admission to the intensive care unit (ICU), or death. If a person was enrolled from the ICU, only increased oxygen supplementation or death defined that person as a case. Controls (hereinafter, nonprogressors) were those who did not progress during follow-up. A nested case-control design was used, including all identified cases that were matched 1:2 to controls by viral pathogen and initial risk of disease progression (high or low risk). Six additionally available nonprogressors were randomly matched to existing sample pairings causing six 1:3 matches.

Blood samples were collected from all participants at study enrollment, which for most participants was close to hospital admission (median time after admission, 2 days; IQR, 1–4). Clinical and demographic data were sourced from each inclusion study and were determined at hospital admission by a relevant physician.

Metabolomics and Lipidomics Analysis

Relative metabolite abundances were measured in plasma by untargeted ultrahigh-performance liquid chromatography–tandem mass spectrometry by Metabolon. Absolute lipid concentrations were measured using Metabolon's Complex Lipid Platform. The metabolite abundances were batch normalized, and the metabolite and lipid measurements were log₂ transformed resulting in a median of 0 per molecule. Details are provided in the supplementary material (Metabolomics and Lipidomics Data section, Supplementary Figure 1).

Statistical Analysis

Potential differences in metabolite abundances between progressors and nonprogressors were assessed by conditional logistic regression with a binary outcome for disease progression and accounting for matching by initial risk of disease progression and pathogen. Power calculations are provided in the supplementary material (Power Calculations section).

Univariable analyses were used to identify potential confounders associated with disease progression in the total population (likelihood ratio test P ≤ .1; details are included in the supplementary material, Statistical Analysis section). Multivariable analyses were conducted for each tryptophan pathway metabolite (tryptophan, kynurenine, and K/T ratio) in separate models, adjusted for age (<50, 50–59, 60–69, 70–79, ≥80 years), sex (male/female), and factors that were significant in univariable analyses (see all results in Supplementary Table 1 and sensitivity analyses of nonincluded adjustment variables in Supplementary Table 2). To explore the association between metabolites and disease progression, the same unadjusted and adjusted multivariable models were assessed in patients with low and high initial risk of progression and in those with influenza and SARS-CoV-2 infection separately; however, these subgroup analyses may be underpowered and should be interpreted with caution.

To examine associations of metabolite abundances with death alone, the same univariable and multivariable analyses as previously described were performed in progressors who died and their matched nonprogressors. However, the subgroups of low initial risk and SARS-CoV-2 infection had too few progressors to perform meaningful analyses (23 and 21 progressors, respectively); therefore, these subgroups were not analyzed separately.

Pathway-Resolved Correlation Network Analysis

Exploratory pathway-resolved correlation network analysis was performed for metabolomics and lipidomics datasets individually via the Weighted Gene Co-expression Network Analysis (WGCNA) R package [23] as described previously [24]. Full description is available in the supplementary material (WGCNA section). Resulting modules were tested for correlations with clinical variables by Spearman rank correlation and presented with Benjamini-Hochberg false discovery rate (FDR)–adjusted P values.

Software

All statistical and bioinformatic analyses were conducted in R version 3.6.3 [25]. Google AI (Gemini 3) was used to improve figures.

Ethics

All contributing cohorts had obtained ethical approval and participant consent prior to the study. More details are available in the supplementary material (Ethics section).

RESULTS

Influenza and SARS-CoV-2 Patient Cohort

The study cohort included 303 participants (Table 1): 180 hospitalized with influenza infection between 2013 and 2022 (60 progressors and 120 nonprogressors) and 123 hospitalized with SARS-CoV-2 infection between 2020 and 2021 (39 progressors and 84 nonprogressors). One low-risk influenza progressor was excluded from lipidomics analyses due to insufficient sample material.

Table 1.

Patient Demographics and Clinical Data

		SARS-CoV-2		Influenza
	Total Cohort	Progressors	Nonprogressors	Progressors	Nonprogressors
Participants	303 (100)	39 (100)	84 (100)	60 (100)	120 (100)
Dates of sample collection, range	2013 Jun–2022 Mar	2020 Mar–2021 Mar	2020 Apr–2021 Mar	2013 Jul–2022 Mar	2013 Jun–2020 Mar
Male	173 (57)	23 (59)	53 (63)	34 (57)	63 (53)
Age, y	65 (20–97)	73 (36–90)	60 (27–97)	71 (39–97)	62 (20–94)
High initial risk of progression	149 (49)	14 (36)	30 (36)	34 (57)	63 (53)
ICU
At enrollment	86 (28)	1 (3)	1 (1)	28 (47)	56 (47)
Admission during follow-up	26 (9)	9 (23)	0 (0)	17 (28)	0 (0)
Death during follow-up	67 (22)	20 (51)	0 (0)	47 (78)	0 (0)
BMI, kg/m2	27 (14–56)	27 (14–48)	29 (17–51)	26 (16–44)	26 (17–56)
Comorbidity
Asthma	41 (14)	3 (8)	18 (21)	3 (5)	17 (14)
Diabetes mellitus	72 (24)	11 (28)	20 (24)	16 (27)	25 (21)
Chronic kidney disease	32 (11)	3 (8)	6 (7)	11 (18)	12 (10)
Chronic liver disease	8 (3)	0 (0)	0 (0)	3 (5)	5 (4)
Vaccination
SARS-CoV-2	2 (1)	1 (3)	0 (0)	1 (2)	0 (0)
Influenza	96 (32)	5 (13)	1 (1)	29 (48)	61 (51)
Smoking
Current smoker	44 (15)	1 (3)	3 (4)	2 (3)	38 (32)
Former smoker	113 (37)	22 (56)	33 (39)	27 (45)	31 (26)
Never smoked	124 (41)	12 (31)	43 (51)	24 (40)	45 (38)
Missing	22 (7)	4 (10)	5 (6)	7 (12)	6 (5)
Acute steroid use
Before sample collection	175 (58)	28 (72)	64 (76)	32 (53)	51 (43)
After sample collection	4 (1)	2 (5)	1 (1)	0 (0)	1 (1)
No steroids	124 (41)	9 (23)	19 (23)	28 (47)	68 (57)

Data are presented as No. (%) or median (range). Progressors were those who progressed to a worse disease state within 28 days after sample collection, and nonprogressors were those who did not progress. Progressors were matched to nonprogressors by pathogen and initial risk of disease progression. Comorbidities and vaccinations were determined at study enrollment.

Abbreviations: BMI, body mass index; ICU, intensive care unit.

Out of all progressors, 67 died during follow-up (20 SARS-CoV-2 infected and 47 influenza infected), and with their matched nonprogressors, they were included in the secondary outcome subpopulation (n = 194).

Metabolomics and Lipidomics Data

The analyses based on ultrahigh-performance liquid chromatography–tandem mass spectrometry detected 1307 metabolites measured in relative abundances, and the complex lipid panel measured 995 lipids quantified by micromolar concentration . A total of 139 metabolites and 102 lipids were removed from the datasets due to low variance. Final separate datasets included 1167 metabolites and 893 lipids.

Tryptophan Pathway Associations With Disease Progression

We investigated the associations of tryptophan, kynurenine, and the K/T ratio with disease progression before and after adjustment for sex, age group, and the following factors, identified as potential confounders: smoking status, chronic kidney disease, and asthma.

Higher abundances of kynurenine were associated with higher odds of disease progression in the unadjusted models. This persisted in the adjusted model for the total cohort with an estimated 81% increase in the odds per doubling of kynurenine abundance (95% CI, 1.21–2.70; P = .0036; Table 2). Although subgroup analyses were likely underpowered, this association was also observed in the subgroup of high initial risk of disease progression (odds ratio [OR], 2.22; 95% CI, 1.19–4.16; P = .0127) and influenza subgroup (OR, 1.95; 95% CI, 1.15–3.31; P = .0135), and the direction of the associations in the subgroups of low initial risk and SARS-CoV-2 were consistent with the total cohort. In the subset cohort assessing progression to death, all models rendered strong statistical evidence for higher odds of dying with increases in kynurenine abundances (Supplementary Table 3).

Table 2.

Odds Ratios for Disease Progression per Log₂ Increase (Doubling) in Metabolite Abundance

	Unadjusted OR	95% CI	P Value	Adjusted OR	95% CI	P Value
Kynurenine
Overall	1.86	1.34–2.57	.0002	1.81	1.21–2.70	.0036
Low risk	1.61	1.02–2.55	.0429	1.58	.85–2.94	.1481
High risk	2.14	1.32–3.47	.0022	2.22	1.19–4.16	.0127
SARS-CoV-2	2.15	1.12–4.13	.0223	1.31	.54–3.15	.5500
Influenza	1.77	1.22–2.56	.0024	1.95	1.15–3.31	.0135
Tryptophan
Overall	0.79	.53–1.16	.2285	0.94	.55–1.60	.8246
Low risk	0.99	.53–1.87	.9861	1.73	.69–4.31	.2409
High risk	0.67	.40–1.13	.1329	0.67	.32–1.43	.3003
SARS-CoV-2	0.79	.36–1.76	.5667	0.91	.32–2.60	.8606
Influenza	0.78	.50–1.23	.2894	0.89	.47–1.66	.7054
K/T ratio
Overall	2.04	1.48–2.81	<.0001	1.89	1.26–2.84	.0022
Low risk	1.87	1.10–3.18	.0215	1.32	.63–2.74	.4587
High risk	2.14	1.42–3.24	.0003	2.26	1.29–3.96	.0046
SARS-CoV-2	2.46	1.26–4.79	.0080	1.34	.59–3.06	.4855
Influenza	1.92	1.34–2.75	.0004	2.20	1.24–3.89	.0069

Disease progression in these analyses were defined by the primary case definition: progression to a more severe requirement of oxygen supplementation vs that at sample collection, admission to the intensive care unit following sample collection, or death—all within a 28-day follow-up period. Adjusted models were adjusted for age group, sex, smoking status, chronic renal disease, and asthma. Low and high risk refers to the patients' initial risk of disease progression as determined by their required oxygen supplementation. P values calculated by conditional logistic regression.

Abbreviations: K/T, kynurenine/tryptophan; OR, odds ratio.

We found no statistical evidence of association between tryptophan abundance and disease progression, overall or in subgroups (P = .8246 in total cohort and P = .8127 in the subset cohort with death as the outcome; Table 2, Supplementary Table 3).

In unadjusted models, higher K/T ratios were associated with disease progression. The adjusted K/T model for the total cohort showed strong statistical evidence for association between a K/T ratio and odds of disease progression with an estimated 89% increase in the odds of progression per doubling of the K/T ratio (95% CI, 1.26–2.84; P = .0022). As with kynurenine, this association was observed in the subgroup of high initial risk of disease progression (OR, 2.26; 95% CI, 1.29–3.96; P = .0046) and influenza subgroup (OR, 2.20; 95% CI, 1.24–3.89; P = .0069), with a consistent direction of the association in the subgroups of low initial risk and SARS-CoV-2. Models for the subset cohort with death as the case definition showed strong statistical evidence for association in the full cohort and the high-risk group, with higher K/T ratios associated with higher risk of dying (Supplementary Table 3).

When testing the interactions between the pathogens (influenza and SARS-CoV-2) and metabolite abundances (tryptophan, kynurenine, and the K/T ratio), we saw no statistical evidence of pathogen-specific differences in their associations with disease progression (P > .4).

Metabolic Patterns of Disease Progression

Our clustering analyses rendered 14 metabolite modules and 12 lipid modules, all represented by a randomly generated color (see Supplementary Table 4 for number of molecules in each module). Of these, 3 metabolite modules (the red, turquoise, and pink highlighted in the black boxes of Figure 1) showed statistical evidence of a positive association with disease progression in the total cohort after FDR adjustment of P values. All 3 metabolite modules associated with disease progression were also positively associated with higher age and chronic kidney disease. Figure 1 and the metabolite content of all modules can be explored in detail in the supplementary HTML file or at http://Rasmussen-fluCOVID.chip.dk. The overall superpathway content of all modules is shown in Supplementary Figure 2.

Figure 1.

A, Heat map of Spearman rank correlations of metabolite modules derived from the WGCNA (Weighted Gene Co-expression Network Analysis) with clinical variables. Modules are represented by colors, and the number of metabolites in each module is indicated to the left. In the correlation calculations, modules were represented by the first eigengene (average abundance of all metabolites in the module). Negative correlations are shown in blue and positive ones in red. Black boxes highlight modules associated with disease progression before adjustment for confounders. High risk refers to high initial risk of disease progression. False discovery rate–adjusted P values are represented as follows: *P = .1. **P = .05. ***P = .01. B and C, Pie figures show the metabolic pathway content of the red module (B; n = 51 metabolites) and turquoise module (C; n = 345 metabolites) that were associated with disease progression after adjustment for confounders. The inner parts of the pies are the superpathways indicated by colors. The outer part represents the subpathway content (descriptions not included here). The sizes of each part represent the number of metabolites with the specific annotation. The legend describes which superpathways are paired with which colors in the figure, and the numbers indicate how many of the 1167 metabolites measured were annotated to that superpathway. An interactive version of the figures where subpathway annotations can be viewed are included in the supplementary HTML file.

The red module contained 51 metabolites (Figure 1B) of which 16 (31%) were lipids, 14 (28%) were xenobiotics (molecules not naturally produced by the human body) mostly from the xanthine metabolism, 5 (10%) were glutamines or partially characterized glutamine conjugates, and 15 (30%) were unknown molecules. Beside age and chronic renal disease, this module was associated with increases in oxygen supplementation during follow-up, death during follow-up, diabetes mellitus, and SARS-CoV-2 infection.

The turquoise metabolite module contained 345 metabolites (Figure 1C). Of these, 121 (35%) were amino acids, including kynurenine, kynurenate, 3-hydroxykynurenine, picolinate, quinolinate, indole acetate, and 6 other downstream tryptophan pathway metabolites, as well as 7 polyamine metabolites that showed the strongest association with disease progression within the module. There were 89 unknown metabolites (26%) in the module. This module was also positively associated with high initial risk of disease progression and high initial oxygen levels, as well as increases in oxygen supplementation, death, and enrollment from the ICU.

The pink metabolite module contained 47 metabolites and consisted of 35 (75%) lipids, including 13 carnitines and 13 dicarboxylate molecules. This module was also positively associated with SARS-CoV-2 infection and increases in oxygen supplementation during follow-up and diabetes mellitus, and it was inversely associated with high initial risk of disease progression, high initial levels of oxygen supplementation, and enrollment from the ICU.

After adjusting all 3 metabolite module associations for sex, age, smoking, chronic kidney disease, and asthma, the red and turquoise modules remained statistically associated with disease progression. The association of the pink module was mainly confounded by older age. We also tested for potential interactions with the different pathogens but found no evidence of a difference in the associations according to type of pathogen.

None of the modules derived from the lipidomics dataset were associated with disease progression. We did, however, see multiple modules associated with each pathogen, the initial risk of disease progression, oxygen levels, being enrolled from the ICU, and receipt of steroid therapy (Supplementary Figure 3).

DISCUSSION

In this study, we observe associations of higher kynurenine and higher K/T ratios with disease progression in a clinically well-defined cohort of individuals infected with SARS-CoV-2 and influenza. The strongest associations were found in the populations with high initial risk of disease progression (those who already required high levels of oxygen supplementation) and influenza. Exploratory analyses revealed positive associations of kynurenine and 10 other downstream metabolites of the tryptophan pathway with disease progression. Other metabolic classes associated with disease progression included other amino acids, specifically polyamine metabolism, and groups of lipids and xanthine metabolism molecules. These associations were not confounded by other clinically relevant variables tested in our analysis and may reflect the wider biological pathways through which the K/T ratio influences disease progression.

The statistical analyses of tryptophan pathway metabolites indicate that the K/T ratio is associated with disease progression and that this is shared between influenza and SARS-CoV-2 regardless of a patient's initial disease severity. However, tryptophan itself was not statistically associated with disease progression in any of our analyses, indicating that it is not the level of tryptophan but the conversion of tryptophan to kynurenine and other downstream metabolites that is related to the variations in disease progression in these patients. The exact causality cannot be evaluated in this study due to the cross-sectional design. Yet, increased viral load results in higher levels of interferons, which in turn triggers IDO1 activation as a host response mechanism to starve the virus of the essential amino acid tryptophan [11]. Therefore, what we are observing may be a proxy of increased viral load, which in other studies has been associated with poorer outcomes [26]. Furthermore, it has been suggested that some viruses may take advantage of the IDO activation, which in turn dampens the immune response, to increase replication [12]. It would be valuable to explore the relationship between K/T ratios and viral load in patients with COVID-19 to see if this ratio could be a reliable proxy for viral burden in infections where reliable measurements of viral load do not exist (including influenza). Regardless of the specific mechanism, the results from this study indicate that K/T ratios may be a useful risk stratifying biomarker across viral diseases and a potential target for future interventional strategies focusing on IDO1 in patients with severe respiratory infections. Future studies should include larger cohorts with wider representation from different ethnic backgrounds.

We identified modules of metabolites and lipids that showed similar patterns of variation in abundance or concentration across the patient cohort. In correlation analyses, the modules were represented by the first principal component: a weighted average of the abundances of all metabolites in the module. When these associations are being interpreted, it is important to note that individual metabolites may show different associations with certain clinical variables as compared with the overall module.

All identified modules were associated with at least 1 clinical variable in the study, but only 2 metabolite modules (the turquoise and red modules) were independently associated with disease progression after controlling for confounding and FDR adjustment of P values. These 2 modules contained different classes of molecules: the larger turquoise module (345 metabolites) contained more amino acids while the smaller red module (51 metabolites) was dominated by fatty acids. Median log₂-transformed abundances of all metabolites in the red and turquoise modules are presented for progressors and nonprogressors in the supplementary HTML file.

Within the smaller red module, we found lipids from many subpathways, including carnitines. Eleven downstream xanthine metabolites were found, mostly from the caffeine metabolism. However, they were not the drivers of the association with disease progression, as the difference in abundance between those who progressed and those who did not was minor. Previous studies have identified caffeine to have protective effects against SARS-CoV-2 [27–29], but this could not be validated in our cohort. The strongest single-metabolite associations with disease progression in the red module were for partially characterized glutamine conjugates. The meaning of these associations remains elusive.

Within the turquoise module that contained 35% amino acids, we found 11 tryptophan pathway metabolites, including kynurenine but not tryptophan, of which 8 were generally higher in those who progressed to a worse disease state, corresponding with the results in the targeted statistical analyses. This module was also associated with chronic kidney disease, and we identified multiple metabolites associated with kidney failure, such as polyamine metabolites and creatinine [30, 31]. Since acute kidney injury has been linked to severe disease in patients with COVID-19 [32], we cannot ascertain if the association is related to disease progression or a consequence of poor metabolite clearance due to kidney injury. In this module, we found a larger subset of fatty acids and various carnitines with higher abundance in progressors: metabolites previously determined to influence the outcome of viral infections [33, 34]. Particularly for this module, it is important to note its large size. It is possible that the settings for the WGCNA did not allow multiple modules to be formed and that the turquoise module contains multiple subsets of metabolites with different relevance to the underlying biological mechanisms in the patient cohort.

Another metabolite module (pink, 47 metabolites) was associated with disease progression, but in multivariable analyses, this association was confounded by age. This module contains multiple acylcarnitines, which have been linked to healthy aging [35]. Acylcarnitines have been linked to higher risk of disease progression in an influenza cohort [4], but this association is likely confounded by age. The metabolites and metabolic pathways within the red and turquoise modules might be targets for further investigation in other cohorts or animal models to gain broader insights into how infection affects the tryptophan pathway. However, given the explorative nature of these clustering analyses, results should be interpreted with caution and used for hypothesis generation for future validating studies. Due to the mostly unvaccinated state of the population, it is likely that the associations that we found with disease progression are driven by an innate immune response.

There are limitations to this study. First, the study population is sampled from 3 studies that had slightly different inclusion criteria and were collected over various periods. Most of the individuals who were SARS-CoV-2 infected were recruited in Denmark, limiting generalizability. Second, disease severity varied among the recruitment cohorts, and clinical practices varied among countries. For instance, some oxygen supplementation machines are kept in the ICU in certain countries, meaning that ICU admission reflects variable disease severity. Third, blood samples were drawn at 1 time point, limiting an exploration of the temporal change in metabolites through disease progression, and the timing of samples relative to initiation of oxygen supplementation varied among recruitment cohorts. However, for most participants, the sample was collected within 2 days of hospital admission, and the study was designed as a cross-sectional analysis of metabolite abundances reflecting the state at or near the time of progression. Fourth, treatment information besides acute steroid usage was not available for all patients, although it was a defining feature of the patients recruited from the ProPAC cohort and could be relevant for the inflammatory response. Fifth, the metabolomics data are semiquantitative, meaning that the abundances are not directly comparable with other studies. It also means that we cannot say anything about the absolute concentrations of metabolites and that metabolites naturally present in small amounts may not be fully captured by the analysis. Sixth, the annotation of metabolites allows each metabolite to be linked to only 1 pathway. If the same metabolite is involved in multiple biological processes, which is common, this is not captured fully. Furthermore, 25% (n = 292) of the metabolites in the analyses are unknown (have no annotation), which reflects the relatively new field of metabolomics analyses. Yet, many were associated with disease progression in our analyses and could constitute currently unknown immune pathways or drug metabolites related to treatment. Including them in these types of analyses highlights the importance of continued exploration and annotation of the metabolome. Seventh, WGCNA is an exploratory analysis method, and the results should be interpreted as such and used for hypothesis generation for future studies in similar cohorts. Finally, although the analysis was adjusted for available potential confounders, residual confounding cannot be ruled out, as is inherent to observational designs such as case-control studies. Despite these limitations, we have a strong, well-described international multipathogen cohort, which allows us to link metabolite abundances and lipid concentrations to clinical phenotypes and outcomes.

CONCLUSION

Using a clinically well-defined cohort of hospitalized individuals infected with influenza or SARS-CoV-2, we found that those with higher serum levels of kynurenine and higher K/T ratios have higher odds of disease progression, defined as an increasing need of oxygen supplementation, admission to the ICU, or death, which is consistent with results from previous studies. Using unsupervised cluster-based explorative analyses, we confirmed the association of kynurenine with disease progression and found evidence of associations of other amino acid and lipid classes similar to influenza and SARS-CoV-2. On the basis of these findings, we conclude that some biological mechanisms related to variations in disease severity in these patient groups are shared among certain respiratory viruses. Furthermore, that they are potential biomarkers for risk stratification or treatment targets that warrant further exploration in similar cohorts.

Supplementary Data

Supplementary materials are available at Open Forum Infectious Diseases online. Consisting of data provided by the authors to benefit the reader, the posted materials are not copyedited and are the sole responsibility of the authors, so questions or comments should be addressed to the corresponding author.