ABSTRACT

   The heterogeneity in severity and outcome of COVID-19 cases points out
   the urgent need for early molecular characterization of patients
   followed by risk-stratified care. The main objective of this study was
   to evaluate the fluctuations of serum metabolomic profiles of COVID-19
   patients with severe illness during the different disease stages in a
   longitudinal manner. We demonstrate a distinct metabolomic signature in
   serum samples of 32 hospitalized patients at the acute phase compared
   to the recovery period, suggesting the tryptophan (tryptophan,
   kynurenine, and 3-hydroxy-DL-kynurenine) and arginine (citrulline and
   ornithine) metabolism as contributing pathways in the immune response
   to SARS-CoV-2 with a potential link to the clinical severity of the
   disease. In addition, we suggest that glutamine deprivation may further
   result in inhibited M2 macrophage polarization as a complementary
   process, and highlight the contribution of phenylalanine and tyrosine
   in the molecular mechanisms underlying the severe course of the
   infection. In conclusion, our results provide several functional
   metabolic markers for disease progression and severe outcome with
   potential clinical application.

   IMPORTANCE Although the host defense mechanisms against SARS-CoV-2
   infection are still poorly described, they are of central importance in
   shaping the course of the disease and the possible outcome. Metabolomic
   profiling may complement the lacking knowledge of the molecular
   mechanisms underlying clinical manifestations and pathogenesis of
   COVID-19. Moreover, early identification of metabolomics-based
   biomarker signatures is proved to serve as an effective approach for
   the prediction of disease outcome. Here we provide the list of
   metabolites describing the severe, acute phase of the infection and
   bring the evidence of crucial metabolic pathways linked to aggressive
   immune responses. Finally, we suggest metabolomic phenotyping as a
   promising method for developing personalized care strategies in
   COVID-19 patients.

   KEYWORDS: COVID-19, SARS-CoV-2, longitudinal study, metabolomics,
   virus-host interactions

OBSERVATION

   More than a year has passed since the World Health Organization (WHO)
   announced the COVID-19 outbreak as a pandemic in March 2020, following
   the rapid spread of the SARS-CoV-2 virus ([46]1). The clinical course
   of COVID-19 is versatile; the infection of the SARS-CoV-2 virus not
   only varies in its severity from asymptomatic or mild and moderate
   respiratory disease (80%) to clinically severe or critical
   life-threatening disease (20%) but also varies in a range of organs the
   disease can affect ([47]2, [48]3). Diverse clinical trajectories seem
   to be the result of the immune response differences between individuals
   ([49]4). Multiple innate and adaptive immune system pathways that
   produce inflammatory molecules against the virus and virus-infected
   human cells are triggered after the SARS-CoV-2 entry in the cell, with
   characteristic overexpression of proinflammatory cytokines (e.g., IL-6,
   TNF-α, IFN-γ) known as cytokine storm in the most severe cases
   ([50]4[51]–[52]6). The host's immune responses typically involve
   changes in metabolic processes at the cellular level, reflecting the
   host-defense mediators and underlying mechanisms ([53]7).

   Detailed understanding of the molecular mechanisms behind COVID-19
   pathogenesis and inflammatory response is needed to predict and reduce
   individual risks, develop therapeutic strategies, and reduce the
   overall mortality rate (around 2% globally according to WHO data)
   ([54]8). The human blood sera metabolome (defined as small molecules
   <1500–2000 Da) reflects the organism's metabolic state and is widely
   used to gain a deeper understanding of the pathogenesis of diseases.
   Recent reports of metabolomics studies highlight the pivotal role of
   cellular metabolites in programming immune response to SARS-CoV-2
   infection, but nevertheless, none of the studies so far have addressed
   the metabolomic changes during the recovery of infection in a
   longitudinal manner ([55]9[56]–[57]13). Considering the extremely high
   heterogeneity of the COVID-19 disease and lack of promising predictive
   biomarkers, we believe that implications of longitudinal metabolite
   profiling may be beneficial in understanding the underlying mechanisms
   of the diverse course of the disease and promote the early
   identification of people at increased risk of severe illness from
   COVID-19 and related complications.

   We performed quantitative targeted metabolome analysis with liquid
   chromatography-mass spectrometry (LC-MS) in blood sera of 32
   hospitalized COVID-19 patients at the acute phase (time of admission at
   the hospital) and the recovery phase (40 ± 14.92 days) of the disease
   (see Text S1 in the supplemental material for a detailed description of
   methods). We also included a group of 39 subjects without any acute
   infection or state from the general population as controls. Written
   informed consent was obtained from every participant before their
   inclusion in the study, and the study protocol was approved by the
   Central Medical Ethics Committee of Latvia (No. 01-29.1.2/928).

   As expected, the clinical blood tests revealed abnormal hematological
   parameters for the majority of study participants at the time of
   hospitalization, with a high variation in platelet (202.94 ± 65.26 μL)
   levels and low lymphocyte measurements (0.64 ± 0.56 μL), which
   coincides with previously reported lymphopenia as the hallmark of
   severe COVID-19 cases. We also observed a high variation of several
   markers (e.g., alanine aminotransferase, bilirubin, lactate
   dehydrogenase, C-reactive protein) indicating renal and hepatic
   dysfunction, myocarditis, inflammation, and coagulation, which confirms
   the systemic response to the infection in our study cohort ([58]2,
   [59]14) (see Table S1 in the supplemental material).

   Out of 51 metabolites analyzed by LC-MS, 22 metabolites showed
   significantly altered levels (paired t test, FDR < 0.05) in the serum
   samples during the acute phase in comparison to the recovery phase
   ([60]Table 1), where concentrations for 16 compounds were significantly
   elevated, whereas 8 metabolites were decreased. The hierarchical
   clustering and principal-component analysis of the obtained metabolomic
   profiles showed clear metabolomics-based discrimination of samples
   collected in different phases of the disease and independent controls
   ([61]Fig. 1A and [62]B), indicating an altered metabolic activity
   during infection. Pathway analysis of longitudinally obtained COVID-19
   patients’ metabolite profiles revealed 13 significantly enriched
   pathways (FDR < 0.05), including phenylalanine, tyrosine, and
   tryptophan biosynthesis, d-glutamine and d-glutamate metabolism, and
   arginine biosynthesis ([63]Fig. 1C, Table S3). Statistical analysis was
   done with Metaboanalyst version 5.0 ([64]15).

TABLE 1.

   Serum metabolites showing significantly altered levels between the
   acute phase and recovery phase of the disease[65]^a
   Compound Fold change False discovery rate Avg level in acute COVID-19
   μM (±SD) Avg level in COVID-19 recovery phase μM (±SD) Avg level in
   non-COVID controls μM (±SD)
   3-Hydroxy-DL-Kynurenine 10.51 6.79E-07 4.77 (± 2.63)* 1.76 (± 2.59)
   <LOD
   4-Hydroxyproline 0.42 1.99E-06 6.4 (± 3.51)* 15.08 (± 9.34) 11.24 (±
   5.29)
   Carnitine 1.25 8.90E-05 52.74 (± 19.33) 42.52 (± 11.56) 74.52 (± 24.65)
   Citrulline 0.48 9.16E-08 16.81 (± 8.48)* 34.54 (± 17.31) 32.43 (± 9.66)
   Isovalerylcarnitine 2.01 1.78E-06 0.13 (± 0.08) 0.07 (± 0.03) 0.18 (±
   0.09)
   Kynurenine 1.71 2.67E-06 3.57 (± 1.40)* 2.22 (± 1.32) 2.84 (± 0.69)
   L-Acetylcarnitine 1.21 2.34E-02 4.53 (± 2.40)* 3.61 (± 1.26) 4.19 (±
   2.75)
   L-Asparagine 1.51 3.88E-05 42.56 (± 14.92) 30.19 (± 11.12) 44.63 (±
   14.64)
   L-Glutamic acid 1.65 4.61E-05 174.32 (± 81.10)* 122.38 (± 83.16) 144.62
   (± 75.42)
   L-Glutamine 0.72 5.71E-08 608.72 (±223.93)* 820.83 (±268.53) 732.84
   (±131.12)
   L-Isoleucine 1.21 1.25E-02 75.72 (± 18.49) 71.38 (± 33.53) 185.44 (±
   71.27)
   L-Lysine 1.08 1.15E-02 174.3 (± 54.09) 165.07 (± 32.74) 225.46 (±
   63.36)
   L-Methionine 1.47 1.62E-06 25.99 (± 10.49) 17.42 (± 5.79) 29.69 (±
   12.20)
   L-Octanoylcarnitine 0.75 3.85E-02 0.32 (± 0.19) 0.42 (± 0.17) 0.06 (±
   0.06)
   L-Phenylalanine 1.33 6.79E-07 113.06 (± 38.07)* 88.98 (± 28.46) 95.81
   (± 20.50)
   L-Proline 0.86 3.23E-03 185.87 (± 81.55)* 219.66 (± 66.13) 236.84 (±
   70.70)
   L-Threonine 1.45 2.15E-03 119.92 (± 45.29) 85.33 (± 37.11) 149.53 (±
   40.05)
   L-Tryptophan 0.83 2.87E-04 63.23 (± 20.67) 77.02 (± 16.74) 93.21 (±
   21.87)
   L-Tyrosine 1.14 2.87E-04 76.46 (± 21.50) 70.2 (± 17.15) 82.52 (± 23.17)
   L-Valine 1.26 7.73E-05 254.27 (± 73.92) 210.84 (± 56.92) 285.98 (±
   83.58)
   Ornithine 1.37 9.45E-03 93.43 (± 30.49) 74.36 (± 36.16) 131.28 (±
   72.24)
   Taurine 1.36 6.01E-03 108.82 (± 51.77)* 85.87 (± 54.56) 66.89 (± 21.32)
   [66]Open in a new tab
   ^a

   Only serum metabolites exhibiting significantly altered levels
   comparing measures obtained during the acute phase and recovery phase
   of the disease are shown with the corresponding fold change and false
   discovery rate values obtained from t test analysis. Average levels in
   the acute or recovery phase that differ significantly from the control
   group are displayed in bold. Asterisk indicates significantly different
   levels comparing samples from the acute COVID-19 group versus the
   control group, for which the direction of change is the same as the
   acute vs recovery group. LOD, level of detection; SD, standard
   deviation.

FIG 1.

   [67]FIG 1
   [68]Open in a new tab

   Targeted metabolomic analysis of longitudinal serum samples of
   hospitalized COVID-19 patients. (A) Heatmap and hierarchical clustering
   of top 22 significantly altered metabolites. Each column represents one
   sample: red, samples collected in the acute phase; green, samples
   collected during the recovery phase; blue, samples collected from the
   population controls. Each row conforms to a specific metabolite
   expressed in normalized, log-transformed concentration value. (B)
   Principal-component analysis showing clear discrimination of samples
   collected from COVID-19 patients in the acute (red) and recovery
   (green) phases of infection as well as control subjects (blue) based on
   the obtained metabolite profiles. (C) Scatterplot representing the most
   relevant metabolic pathways from KEGG library arranged by adjusted P
   values (obtained by Global Test pathway enrichment analysis) on the
   y-axis, and pathway impact values (from pathway topology analysis) on
   the x-axis. The node color is based on its P value and the node radius
   is determined based on their pathway impact values. (D) Boxplots
   showing the normalized levels of the most functionally relevant
   metabolites altered during the recovery of COVID-19 (red, acute phase;
   green, recovery phase; blue, controls), described as the minimum value,
   the first quartile, the median, the third quartile, and the maximum
   value, with the black dots representing outliers.

   We found l-glutamine ([69]Fig. 1D) as the most significantly changed
   amino acid between the paired patients’ serum samples with reduced
   acute phase concentrations. l-glutamine levels were also significantly
   lower in population controls. It is known that glutamine deprivation
   and decreased glutaminolysis inhibits M2 macrophage polarization, which
   may partly explain the hyperinflammatory state in severe COVID-19 cases
   ([70]16, [71]17). Moreover, the beneficial effect of glutamine has been
   proposed in multiple studies, where adding enteral l-glutamine to the
   regular nutrition shortened the duration of hospitalization and
   improved the outcome in moderate to severe COVID-19 cases ([72]18,
   [73]19).

   Two other amino acids involved in arginine catabolism through the Urea
   cycle, citrulline ([74]Fig. 1D) and ornithine ([75]Table 1), were found
   to significantly change between the acute and recovery phases. However,
   alterations in the levels of l-arginine itself were not detected. Low
   blood plasma citrulline levels have already been reported in COVID-19
   patients, whereas in patients with severe sepsis, the decreased
   citrulline levels are associated with acute respiratory distress
   syndrome ([76]11, [77]13, [78]20). Since arginine can be metabolized to
   creatine and then to creatinine, both varying highly in our study
   cohort according to regular blood tests performed during
   hospitalization, arginine catabolism may be implicated in disturbed
   kidney function of COVID-19 patients ([79]21).

   Three out of 22 metabolites showing significantly changed levels
   between acute infection and recovery phase are involved in the
   tryptophan-kynurenine pathway: l-tryptophan, kynurenine, and
   3-hydroxy-DL-kynurenine ([80]Fig. 1D, [81]Table 1).
   3-hydroxy-DL-kynurenine was the marker with the most pronounced fold
   change between the acute and recovery phases and remarkably was below
   the detection limit in any of the control samples. The reduction of
   tryptophan levels and increase of kynurenine and
   3-hydroxy-DL-kynurenine in the acute phase supports the conclusions of
   previous reports and confirms this pathway's key role in severe
   COVID-19 cases ([82]10, [83]11). In vitro experiments have shown that
   tryptophan deprivation sensitizes T cells to apoptosis, inhibits
   proliferation of T cells, and plays a role in CD8 T-cell suppression in
   cancer ([84]22, [85]23). Notably, the tryptophan-kynurenine pathway
   shows a modulatory effect on the macrophage-mediated responses by
   targeting the synthesis of the metabolic coenzyme NAD+ ([86]24,
   [87]25). According to Thomas et al., dysregulated tryptophan
   metabolism, an essential regulator of inflammation and immunity, may be
   a potential explanation for severity in older COVID-19 patients
   ([88]11).

   Finally, we observed a significant difference in l-phenylalanine
   ([89]Fig. 1D) and tyrosine levels in our cohort between the analyzed
   disease phases, both already suggested as metabolic hot spots of
   COVID-19 before ([90]26). In sepsis and HIV-1 infection, the increased
   phenylalanine serum concentrations are linked to immune activation and
   increased cardiovascular event risk ([91]27[92]–[93]29). Although the
   mechanisms behind this association are not well studied, it is in line
   with microvascular endothelial damage and higher coagulation risk
   characteristic to both: coronary heart disease and severe COVID-19
   ([94]29). Phenylalanine and tyrosine are catabolized to dopamine and
   epinephrine, and the latter has been employed in cardiac arrest as a
   result of cytokine storm, characteristic of severe COVID-19 patients
   ([95]30).

   We also performed a correlation analysis between the significant
   metabolites ([96]Table 1) and available biochemical parameters (Table
   S1); however, none of the identified correlations reached statistical
   significance either in the acute or recovery stage.

   The main limitation of our study is the relatively small sample size,
   which is mainly caused by the limited attainability of hospitalized
   COVID-19 patients during the pandemic. Nevertheless, we believe that
   this limitation is overcome by applying the longitudinal study design,
   which provides higher statistical power and minimizes the potential
   interference of individual-level confounding variables such as age and
   sex, meanwhile assuring the possibility to detect subject-specific
   effects. For technical reasons, we did not include lipidomics and
   untargeted metabolomics that could lead to a more comprehensive set of
   altered metabolites. Another aspect one may consider as the limitation
   is the consideration of serum samples obtained from independent
   subjects without any acute infection as control measures instead of
   measures of the same COVID-19 patients made before the onset of the
   infection, which would reflect the true serum metabolome modifying
   effects of the virus. During the particular study, the pre-infection
   serum samples were not collected, though they should be considered for
   similar future studies.

   In conclusion, our study shows that metabolomic profiling provides
   novel insights into the pathogenesis of host-defense mechanisms and may
   be further applied for rapid biomarker discoveries in infectious
   diseases. These discoveries could show novel therapeutic strategies as
   indirect targets to fasten the recovery process after severe COVID-19.
   In addition, we believe that our data may contribute to the development
   of metabolomics-based personalized care strategies in COVID-19 patients
   comparable to already existing clinical approaches applied in health
   services such as newborn screening of inborn errors. To the best of our
   knowledge, this is the first longitudinal study covering metabolomic
   profiling during the recovery of severe COVID-19.

Data availability.

   Metabolomics data have been deposited to the EMBL-EBI MetaboLights
   database (DOI: [97]https://doi.org/10.1093/nar/gkz1019, PMID:31691833)
   with the identifier MTBLS3852.

ACKNOWLEDGMENTS