Urinary phenotyping of SARS-CoV-2 infection connects clinical diagnostics with metabolomics and uncovers impaired NAD⁺ pathway and SIRT1 activation

Study design

The present research investigated the metabolite profile of urine samples collected from Acute COVID-19 patients (AcuteCOV) (Heidelberg University Hospital). An independent pre-pandemic cohort of SARS-CoV-2-negative subjects was used as control (CTR) (data provided by Bruker BioSpin GmbH).

The overall study can be subdivided into 4 major aims: (1) Aim 1: Identification of the metabolite changes in AcuteCOV urine samples compared with CTR using ¹H NMR. This section analyzed the metabolite data obtained with ¹H NMR spectroscopy and Bruker IVDr method in urine samples to investigate differences between AcuteCOV (Werner Siemens Imaging Center, University Hospital Tübingen) patients and healthy CTR subjects (Data provided by Bruker BioSpin GmbH); (2) Aim 2: Characterization of the AcuteCOV urinary metabolic signature using ¹H NMR and LC–MS data. To enhance the characterization of AcuteCOV urinary metabolic phenotype, additional LC–MS experiments were carried out (Australian National Phenome Centre and Computational and Systems Medicine, Murdoch University). The information provided by this analysis integrated with the ¹H NMR data to derive more detailed metabolite profiles of AcuteCOV samples, which were then analyzed to identify potential disease severity, age, sex, BMI-dependent metabolic signatures; (3) Aim 3: Analysis of antiviral ddhNs and their associations with metabolite profiles and clinical variables. Here, we investigated whether the concentration of 4 antiviral ddhNs, namely 3′-deoxy-3′,4′-didehydro-cytidine (ddhC), 3′-deoxy-3′,4′-didehydrocytidine-5′-carboxylic acid (ddhC-5′CA), 3′-deoxy-3′,4′-didehydrouridine (ddhU), and 3′,5′-dideoxy-3′,4′-didehydrocytidine-5′-homocysteine (ddhC-5′Hcy), measured in urine samples by LC–MS, was associated with either variations in urinary metabolite concentration or changes in clinical parameters; (4) Aim 4: Investigation of the potential association between SARS-CoV-2-induced kynurenine pathway dysregulation, NAD⁺ biosynthesis, and sirtuins. In this section, we sought to provide a biological interpretation underlying the SARS-CoV-2-induced perturbations in the tryptophan metabolites, by investigating whether a dysregulation in this pathway could affect the activation of the NAD⁺-dependent deacetylases sirtuins. To this aim, in addition to the results obtained by ¹H NMR and LC–MS in urine samples, we included the concentration of inflammatory mediators measured in matched serum samples of the same cohort of patients [11].

Study cohort information

Sample processing and quantitative ¹H NMR spectroscopy analysis

¹H NMR spectra were generated using the standardized Bruker Avance IVDr platform (Avance III HD spectrometer) and the quantification was performed using the different Bruker IVDr packages (https://www.bruker.com/de/products-and-solutions/mr/nmr-clinical-research-solutions/b-i-methods.html), as previously described [48]. Urine sample preparation was performed according to B.I.-embedded standard operation procedure (SOP) [49]. For quality control (QC), the B.I. BioBank QC™ method; and for quantification B.I. Quant-UR eTM (e – extended, up to 150 metabolites quantified) modules were utilized. Briefly, for each sample a 600 μL aliquot was transferred into an autoclaved 2 mL microcentrifuge tube (MCT) and centrifuged at 2000 RCF for 10 min at 4 °C (Heraeus Megafuge 8R, Thermo Electron LED GmbH, Osterode am Harz, Germany). Then, 585 μL of the supernatant were added with 65 μL of the standard Bruker urine buffer of pH 7.4 (order number AH0621-10, provided by Bruker BioSpin GmbH, Ettlingen, Germany; contains 1.5 M KH₂PO₄, 2 mM NaN₃, 0.1 % TSP-d₄) and the mixture was vortexed for 30 s (VORTEX-GENIE 2, Scientific Industries, Inc., Bohemia, NY, USA). A volume of 600 μL of the processed sample was then transferred to a 5 mm Bruker NMR tube.

NMR spectroscopy was performed using a 5 mm triple resonance (TXI) RT probe that was controlled via TopSpin software, version 3.6.2. The sample jet was set to a cooling temperature of 5 °C, then sample spectral data were acquired at 300 K. The standard one-dimensional (1D) NMR experiment with solvent suppression was acquired with 32 scans (65,536 data points, 20.0186 ppm spectral width). total experimental time of 4 min per sample. The obtained spectra were uploaded to the Bruker data analysis server for automated quantification. The sample and shim quality of the experiment was tested by the full width at half maximum (fwhm) of the TSP signal (δ 0.00) being less than 1.30 Hz, sharp singlet symmetrical peak. The residual water signal after solvent suppression is less than 30.0 mmol/L.

A statistical total correlation spectroscopy (STOCSY) was performed on each metabolite to find the signal patterns and their chemical shift [50]. Metabolites that could not be determined by its pattern and chemical shift by STOCSY were further investigated by 2D experiments (¹H–¹H COSY, ¹H–¹H TOCSY, ¹H–¹³C HSQC and ¹H–¹³C HMBC). Each spectral region corresponding to the metabolites was integrated and correlated to the IVDr measurements to validate the quantification. Those that did not meet the criteria but the correct concentration is salvageable were done manually in R, otherwise removed from the analysis. Only metabolites passing the correlation of lineshape metabolite signals with calculated fit over 80 % were included in the study. Additionally, maleate quantification was excluded due to the recent report of urine NMR spectra in the same spectral region [32].

NMR metabolites were quantified via an external reference based on an ERETIC calibration [51]. The calibration transfer to each individual sample is facilitated by the PULCON principle [52]. An artificial reference peak representing a concentration equivalent of 10 mM can be added to the solvent suppressed 1D NMR spectrum in a region without metabolite signals. Quantitative calibrations need to be obtained and verified using a dedicated reference sample with known compound concentrations according to B.I. Methods QC procedures. Human body fluid sample preparation and NMR protocols were derived from an article of Dona et al. [49].

Sample processing and targeted quantitative LC–MS

Urine samples were thawed at 4 °C and prepared for analyses following previously reported metabolic phenotyping methods [53], with minor modifications. For the quantification of biogenic amines and amino acid metabolites, separation was performed by ultra-high-performance liquid chromatography (UHPLC) using an Acquity UPLC (Waters Corp, Milford, MA) coupled to a Bruker Impact II QToF mass spectrometer (Bruker, Daltonics, Billerica, MA). Resulting data files were processed for integrations and quantification using Target Analysis for Screening Quantification (TASQ) software v2.2 (Bruker, Daltonics, Bremen, Germany) where calibration curves were linearly fitted with a weighting factor 1/x. For the measurement of tryptophan and associated catabolites, separation was performed using Acquity UHPLC coupled to a Waters Xevo TQ-XS mass spectrometer (Waters Corp, Wilmslow, U.K.). Obtained raw files were processed for peak integrations and metabolite quantifications using the TargetLynx package within MassLynx v4.2, where calibration curves were linearly fitted for each metabolites using a weighting factor of 1/x. Resulting data matrices were quality controlled and combined prior to statistical analysis.

Enzyme-linked immunoassay (ELISA) based analysis to assess serum SIRT1

SIRT1 serum concentration was measured with a commercially available Human SIRT1 ELISA Kit (Thermo Fisher Scientific Inc., Germany. Inter assay coefficient of variation <10 %, Intra assay coefficient of variation <12 %, detection limit: 1.23 ng/mL), according to the manufacturer’s instructions. Serum samples of 171 patients were used in this experiment. Briefly, each well was filled with 50 µL of undiluted serum and 50 µL of proprietary diluent. Seven different concentrations of a lyophilized human SIRT1 standard were prepared to build a calibration curve ranging from 300 ng/mL to 1.229 ng/mL. Absorbance was measured at 450 nm using a plate Reader (Tecan Trading AG, Switzerland). SIRT1 concentration was then calculated based on the standard curve and by adjusting for sample dilution.

Statistical analysis and data interpretation

Statistical analysis was performed using the online comprehensive tool Metaboanalyst (Version 5.0), the JMP Pro 15 software (©SAS Institute Inc., Cary, NC, USA), and SigmaPlot 11.0 (Systat Software Inc., San Jose, CA, USA).

Missing values were handled by the replacement by limit of detection (LoDs) method (1/5th of the positive value of each variable). All spectra were baseline corrected and normalized using probabilistic quotient normalization (PQN) [54]. Both univariate and multivariate analysis approaches were applied. Unsupervised agglomerative hierarchical cluster analysis was performed to identify any potential specific signature of metabolite urine concentration (NA-chip analyzer program, https://sites.google.com/site/dchipsoft/). Unsupervised principal component analysis (PCA), supervised partial least squares discriminant analysis (PLS-DA) and orthogonal partial least squares - discriminant analysis (OPLS-DA) were also carried out as multivariate analysis. Differences across experimental groups were investigated by Mann–Whitney U test, Pearson’s chi-squared test, One-way analysis of variance (ANOVA) or Kruskal–Wallis test on ranks. Two-way ANOVA was applied to investigate influences of two independent variables and their potential interaction. Tukey’s test was used for post hoc multicomparison procedure. The false discovery rate (FDR) was controlled using the Benjamini-Hochberg correction to maximize statistical power [55]. Linear correlation between variables was assessed based on Pearson’s correlation coefficient. A p-value <0.05 and a fold change (FC) >1.2 were considered significant.

The graphical abstract was generated via the biorender.com service with the help of standard COVID-19 related template (World Health Organization. https://www.who.int/emergencies/diseases/novel-coronavirus-2019).

Baseline characteristics of the study cohorts and metabolic data included in the analysis

A schematic workflow diagram of the investigation is shown in Figure 1.

Figure 1:

Consort diagram. Urine metabolomics was performed by ¹H NMR spectroscopy and LC–MS in a large cohort of SARS-CoV-2 infected individuals (n=243). The study is structured in 4 different consecutive aims. In addition to urinary metabolite profiles, Aim 4 involved the investigation of SIRT1 concentration in matched serum samples. Thirteen inflammatory mediators, whose concentration was measured within our previous study [11], were likewise included in the present analysis. AcuteCOV, Acute COVID-19 cohort; CRP, C-reactive protein; CTR, controls; ddhNs, deoxy-didehydronucleosides; LC–MS, liquid chromatography–mass spectrometry; NAD⁺, nicotinamide adenine dinucleotide; NMR, nuclear magnetic resonance; QC, quality control; SIRT1, sirtuin 1.

Aim 1: Identification of the metabolite changes in AcuteCOV urine samples compared with CTR

PCA was carried out to investigate general group separation based on urinary metabolite concentrations assessed via ¹H NMR spectroscopy-based metabolomics approach (Figure 2A). Unsupervised hierarchical clustering revealed a good discrimination between the urinary metabolite profile of AcuteCOV and CTR groups (Supplementary Material 1.2.1). Associations between metabolite concentration profiles were explored by multivariate analysis that disclosed 13 main clusters of variables (Supplementary Material 1.2.1.2, Supplementary Table S2). Univariate analysis identified a total of 22 differentially expressed metabolites, of which 11 were increased in the AcuteCOV group compared with CTR, while 11 metabolites were decreased (Figure 2B).

Figure 2:

Identification of the metabolite changes in AcuteCOV urine samples compared with CTR. (A) Principal component analysis (PCA) scores plot illustrating general group separation based on urinary metabolite concentrations assessed via ¹H NMR spectroscopy-based metabolomics approach. The analysis excluded 4 patients based on extremely different glucose and creatinine urinary concentrations. Out of the 4 outliers, 3 were controls, while 1 subject was from the AcuteCOV group. (B) List of metabolites increased or decreased in AcuteCOV compared with CTR. Only molecules showing log2 fold change (FC) >1.2 are included. A two-color scale is used to illustrate FC: deep red denotes +1.6, deep blue denotes −1. Wilcoxon–Mann–Whitney test. Unsupervised hierarchical clustering reveals an overall SARS-CoV-2 infection-driven clusterization irrespective of (C) age, (D) sex, and (E) BMI groups. (F) Simplified diagram illustrating the dysregulated metabolites and their interactions. Increased metabolites in AcuteCOV vs. CTR are typed with red letters, while down-regulated with blue letters. Metabolites typed with black letters show similar concentration across study groups. Wilcoxon–Mann–Whitney test, p-values: ^*****<10⁻¹⁰; ^****<10⁻⁸; ^***<10⁻⁴; ^**<0.01; ^*<0.05. BMI, body mass index; Cys, cysteine; FC, fold change; met, metabolism; Phe, phenylalanine; TCA, tricarboxylic acid.

Next, we investigated whether sex, age, and BMI factors can influence the metabolite changes associated with SARS-CoV-2 infection. Unsupervised hierarchical clustering analysis of metabolite concentrations disclosed a disease-driven clusterization, indicating that the distinctive infection-dependent metabolic signature previously identified was sustained irrespective of age (Figure 2C, Supplementary Figure S1), sex (Figure 2D, Supplementary Figure S2), and BMI (Figure 2E). Consistently, for most of the differentially-expressed metabolites, the magnitude and pattern of modulation was similar to those observed within the whole sample analysis (Supplementary Table S3 and S4). However, some metabolites showed a different degree of modulation across demographic factors (sex, age, and BMI). Among these, taurine (p-value of interaction=0.042) and creatine (p=0.041) were differentially modulated between sex groups (Supplementary Table S3), whilst orotic acid (p=0.048), citric acid, (p=0.001), trigonelline (p=0.048), and glycine (p=0.009) showed different concentrations across age groups (Supplementary Table S3). Of note, lactic acid and succinic acid appeared to be modulated by both sex and age factors. Moreover, two additional classes of modulation were identified: (1) different response to SARS-CoV-2 infection across sex groups, including syringic acid (p-value of interaction <0.001), erythritol (p=0.048), and pantothenic acid (p=0.033); (2) different response to SARS-CoV-2 infection across age groups, including formic acid (p=0.001) and inosine (p=0.045). Concerning BMI, the changes induced by SARS-CoV-2 were similar across BMI groups for most of the evaluated metabolites, except for succinic acid (p-value of interaction=0.036), oxypurinol (p=0.009), dimethylamine (p=0.031), glycolic acid (p=0.019), pantothenic acid (p=0.026), and inosine (p=0.043) (Supplementary Table S4).

The metabolites differentially modulated between AcuteCOV and CTR groups were manually assigned to the following major metabolic pathways: (1) drug/food metabolite (n=5): caffeine, theobromine, pantothenic acid, oxypurinol, tartaric acid; (2) TCA cycle (n=4): succinic acid, fumaric acid, citric acid, glycolic acid; (3) one-carbon metabolism (n=3): dimethylamine, N,N-dimethylglycine, glycine; (4) fasting/ketogenesis (n=2): acetone, acetoacetic acid; (5) glucose homeostasis (n=2): glucose, lactic acid; (6) cysteine oxidation/creatine metabolism (n=2): taurine, creatine; (7) pyrimidine de novo synthesis/urea cycle (n=2): orotic acid, allantoin; (8) phenylalanine catabolism (n=1): hippuric acid; (9) nicotinic acid metabolism (n=1): trigonelline. Of interest, “one-carbon metabolism” includes only down-regulated metabolites, while “fasting/ketogenesis” and “glucose homeostasis” involve only increased molecules. Figure 2F illustrates the metabolic pathways enriched of modulated metabolites. Enrichment pathway analysis demonstrated a significant similarity to the diabetes mellitus metabolic signature (8 hits over 19 total metabolites, FDR-adjusted p-value: 7.78 × 10⁻⁸).

Aim 2: Characterization of the AcuteCOV urinary metabolic signature

To further investigate the metabolic changes elicited during SARS-CoV-2 infection, the AcuteCOV urine samples were analyzed using LC–MS at the Australian National Phenome Centre, Murdoch University. As reported in Supplementary Table S1, we measured 44 additional metabolites and 4 antiviral ddhNs: ddhC, ddhC-5′CA, ddhC-5′Hcy, and ddhU [32].

Association between urine metabolic profiles and disease severity

Differences in the urinary metabolite concentration between hospitalized and nonhospitalized patients

We explored metabolite differences between non-hospitalized (n=186) and hospitalized (n=57) patients. Univariate analysis identified 12 up-regulated metabolites in the hospitalized group, among which there were the antiviral nucleosides and 4 metabolites involved in the tryptophan catabolic pathway, namely 3-hydroxykynurenine, kynurenine, 3-hydroxyanthranilic acid, and quinolinic acid (Supplementary Table S5). On the other hand, 8 metabolites, including citric acid, trigonelline, allantoin, and glycine were down-regulated in hospitalized patients compared with subjects who did not require hospital admission.

To depict the clinical significance of the identified metabolite differences, a volcano plot was created using both metabolites and clinical data. As shown in Figure 3, C-reactive protein (CRP) and ferritin serum concentrations showed similar fold change range and degree of statistical significance compared with the antiviral ddhNs, 3-hydroxykynurenine, and kynurenine. With regard to variables with reduced concentrations in the hospitalized compared with the nonhospitalized group, iron, citric acid, lymphocyte count, Transferrin saturation, glomerular filtration rate (GFR), platelets, and glycine, showed the most significant p-values.

Figure 3:

Differences in urinary metabolite concentration and serum clinical parameters between hospitalized and nonhospitalized patients. (A) Combined volcano plot showing the metabolites and the clinical variables significantly different in the comparison of hospitalized (n=57) vs. nonhospitalized (n=186) patients, with a fold change >1.2 and p-value <0.05. Increased variables are typed with red letters, while decreased with blue letters. Metabolites are written in bold, while clinical parameters are diplayed in a regular typeface. ^*Denotes serum variables assessed on the same day of urine collection. BMI, body mass index; BT, body temperature; CRP, C-reactive protein; EASIX, endothelial activation and stress index; GFR, glomerular filtration rate; RR, respiratory rate; Tf, transferrin.

Finally, we carried out a biomarker analysis based on area under ROC curve (AUC) calculation (Supplementary Figure S3). Peak CRP showed the greater discrimination ability between hospitalized and nonhospitalized patients (AUC: 0.802(0.743–0.862), specificity: 0.785(0.731–0.842)), sensitivity: 0.737(0.622–0.834)), followed by 3-hydroxykynurenine (AUC: 0.766(0.694–0.840, specificity: 0.763(0.707–0.817), sensitivity: 0.719(0.614–0.816)), and kynurenine (AUC: 0.7654(0.691–0.829), specificity: 0.715(0.642–0.769), sensitivity: 0.737(0.613–0.851)). Of note, the best performance in this population was provided by the ratio between platelets and peak CRP that showed an AUC of 0.837 (0.767–0.898) (Supplementary Figure S3).

Recognition of metabolic signatures associated with clinical deterioration

Metabolic profiles were analyzed according to the serum concentration of biomarkers currently used in the clinical practice to evaluate the main pathophysiological derangements leading to severe disease. Notably, based on previous studies [58, 59], we elected to use the following clinical variables: CRP (inflammation), peripheral blood lymphocyte count (immune response), D-dimer (coagulation), and oxygen saturation (SpO₂, lung function). We also included the Endothelial activation and stress index (EASIX), a predictor of endothelial complications recently validated in a cohort of COVID-19 hospitalized patients [60]. As shown in the heatmap reported in Figure 4, two main clusters were identified (euclidean distance measure, ward clustering): (1) metabolites whose concentration tend to increase along with worsening of clinical parameters; (2) metabolites whose concentration tend to decrease along with worsening of clinical parameters. As highlighted in the dendogram of the hierarchical cluster analysis, each of the two main clusters included two sub-clusters, referred to as 1A (n=2 metabolites), 1B (n=11), 2A (n=7), and 2B (n=9). Of note, Cluster 1B includes the tryptophan-related metabolites neopterin, 3-hydroxykynurenine, kynurenine, and 3-hydroxyanthranilic acid, together with the 4 antiviral ddhNs.

Figure 4:

Recognition of urinary metabolic signatures associated with clinical deterioration. Heatmap showing the concentration of those metabolites which resulted significantly different across the following classes of clinical variables: EASIX; low risk, vs. high risk; D-dimer: normal vs. high; CRP: normal vs. minor elevation, vs. moderate elevation vs. marked elevation; lymphocyte count: normal vs. low; SpO₂: normal vs. low-minor risk vs. low-medium risk vs. low-high risk. Clustering method: average linkage; distance metric: 1–Spearman’s rank correlation. Four main metabolite signatures are identified: (1A) (orange dendrogram), (1B) (green), (2A) (violet), (2B) (brown). A two-color scale is used to illustrate the metabolite modulation pattern: deep red denotes +5, deep blue denotes −5. Specific color-coded scales denote the different clinical parameters and their classes. CRP, C-reactive protein; EASIX, endothelial activation and stress index; SpO₂, oxygen saturation.

Aim 3: Analysis of the antiviral ddhNs and their associations with metabolite profiles and clinical variables

The 4 antiviral molecules ddhC, ddh-5′CA, ddhC-5′Hcy, ddhU accounted for a significant proportion of variability in the dataset, as indicated by PCA (Figure 5A). Thus, we explored whether these molecules were related to specific metabolite signatures and clinical manifestations. To this end, a combined factor, referred to as “Total ddhNs”, was calculated for each patient by adding up the urinary concentration of the 4 antiviral nucleosides: [ddhC] + [ddh-5′CA] + [ddhC-5′Hcy] + [ddhU]. As shown in Table 3, “Total ddhNs” was positively associated with markers of systemic inflammation, including CRP, EASIX, and LDH, whereas leukocytes and lymphocyte counts decreased along with increasing antiviral nucleoside concentrations. With regard to associations with the other metabolites, a strong positive correlation was found with neopterin and a number of molecules related to tryptophan metabolism, including quinolinic acid, kynurenine, 3-hydroxykynurenine, kynurenic acid, and 3-hydroxyanthranilic acid (Table 4). On the other hand, citric acid, glycine, and threonine showed the lowest correlation coefficients, indicating an inverse association with antiviral nucleosides. Moreover, an inverse correlation was observed between the nucleoside sum and different glucogenic amino acids entering the TCA cycle upstream to succinate, such as glycine, serine, threonine, alanine, arginine, glutamine, whereas aspartate, which is converted to oxaloacetate, resulted positively correlated.

Figure 5:

Analysis of the antiviral ddhNs and their associations with metabolic urinary profiles. (A) Principal component analysis (PCA) loading plot showing a strong influence of the 4 ddhNs on PC1. (B) PLS-DA reveals no overlap between the quartile Q1 and Q4 of “Total ddhNs”, which was calculated for each patient by adding up the urinary concentrations of the 4 antiviral nucleosides: [ddhC] + [ddh-5′CA] + [ddhC-5′Hcy] + [ddhU]. (C) Box plots illustrating the metabolites with the highest fold change between “Total ddhNs” Q1 and Q4. Kruskal-Wallis One-way ANOVA on Ranks. Normalized concentrations were obtained by applying the probabilistic quotient normalization (PQN). All the metabolites displayed here had a p-value<0.001. For each experimental group, box plots show the median [25–75], while the red line denotes the mean. LDA, linear discriminant analysis; PCA, principal component analysis; Q, quartile.

To further understand the association between the antiviral nucleosides and the concentration of the other metabolites, we stratified the population according to quartiles (Q) of “Total ddhNs”: (1) Q1, from 0.1 AU to 214.6 AU; (2) Q2, from 214.6 AU to 756.7 AU; (3) Q3, from 756.7 AU to 2,220.9 AU; (4) Q4, from 2,220.9 AU to 6,939.4 AU. PLS-DA found an excellent discrimination between Q1 and Q4 (Figure 5B). These results were confirmed by univariate analysis performed to investigate differences in the metabolite concentrations across the quartiles (Figure 5C and Supplementary Table S6). Among the metabolites whose concentration increased along with higher “Total ddhNs”, kynurenine, neopterin, 3-hydroxykynurenine, and quinolinic acid showed the highest FC between Q1 and Q4. Conversely, the reduced metabolites were citric acid and trigonelline.

Aim 4: Investigation of the potential association between SARS-CoV-2-induced kynurenines/trigonelline dysregulation, NAD⁺ biosynthesis, and sirtuins

As schematized in Figure 6A, the kynurenine pathway mediates NAD⁺ de novo synthesis from tryptophan. Since SARS-CoV-2 can inhibit NAD⁺ generation through down-regulation of the enzymes NAD Synthetase 1 (NADSYN1) and quinolate phosphoribosyltransferase (QPRT) [61], the accumulation of kynurenine metabolites could be a consequence of the virus-induced blockade of the downstream biochemical steps. In this view, the reduced urinary concentration of 1-methylnicotinamide and trigonelline could reflect an increased activation of the NAD⁺ salvage pathways, in an effort to improve NAD⁺ recycling. Therefore, a pilot study was performed to investigate whether the derangement in NAD⁺ biosynthesis could have influences on the activation of sirtuins, NAD⁺-consuming enzymes with potent immunomodulatory and antiviral properties.

Figure 6:

Association between kynurenine/trigonelline dysregulation, NAD⁺ biosynthesis, and sirtuins. (A) Simplified diagram of the tryptophan pathway and its connection to NAD⁺ biosynthesis, which, in turn, is crucial to activate sirtuins. Metabolites typed in red fonts have increased urinary concentrations in patients with worst clinical condition, in males, in elderly, and/or in subjects with higher ddhN urinary release, whereas metabolites typed in blue showed an opposite modulation pattern, with lower urinary concentrations in at least one of the patient classes mentioned above. Enzymes involved in the biochemical reactions presented here are typed in capital letters; the expression of enzymes framed in green is modulated by SARS-CoV-2 [61]. Differences in the concentration of (B) selected metabolites, (C) clinical variables, and (D) serum inflammatory mediators between patient groups derived based on absence (n=122, white box) or presence (n=49, green box) of SIRT1 in serum samples. Box plots illustrate median [25–75], red lines denote the mean of each study group. Wilcoxon–Mann–Whitney test; p-values: ^*p<0.05; ^**p<0.01; ^***p<0.001. CRP, C-reactive protein; EASIX, endothelial activation and stress index; LDH, lactate dehydrogenase; IDO, 2,3-dioxygenase and indole 2,3-dioxygenase; NADSYN1, NAD synthetase 1; NAPRT, nicotinic acid phosphoribosyltransferase; NNMT, nicotinamide N-methyltransferase; QPRT, quinolate phosphoribosyltransferase; SIRT1, sirtuin 1.

We assessed SIRT1 expression in blood samples and found detectable concentrations of this enzyme only in 49 patients (presence of SIRT1), while 122 subjects showed no serum Sirt1 (absence of SIRT1). Next, we investigated the potential association between SIRT1 and the metabolic perturbations and clinical deterioration observed in COVID-19. Univariate analysis revealed that absence of SIRT1 was associated with significantly higher kynurenic acid, quinolinic acid, and neopterin levels in both serum and urine samples (Figure 6B). Moreover, patients with no SIRT1 expression had higher concentration of the inflammatory indexes ferritin, CRP, EASIX, and LDH compared with those patients showing detectable SIRT1 amounts in their blood (Figure 6C). Consistently, absence of SIRT1 was associated with increased serum concentration of the chemokines IL-8 and MCP-1 (Figure 6D).