Evaluation of the Potential of Urinary Volatilomic Patterns of COVID-19 Patients Infected With SARS-CoV-2 for Disease Diagnosis. An Exploratory Study

1. Introduction

In late 2019, COVID-19, a highly infectious disease caused by the severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2), triggered a tremendous and severe pandemic outbreak. COVID-19 has been identified as a hyperinflammatory syndrome characterised by aberrant immune activation and excessive cytokine release (cytokine storm), ultimately leading to failure in multiple organs [1,2]. As the respiratory virus spread to all countries and/or regions in the world, intermittent lockdowns were imposed as a desperate and unprecedented measure to contain the propagation of the disease [3,4]. The effectiveness of this procedure is still an open debate, although it is widely acknowledged that the delay of virus propagation gave time to healthcare systems to adapt and mitigate the mortality caused by COVID-19 [4,5]. According to the World Health Organization (WHO), 768 million people were infected and over 6.94 million deaths have been estimated (Figure 1) [6]. This has caused a strong impact on society at economic, political, social, educational, environmental, and cultural levels that will last for decades [7,8,9,10].

COVID-19 had a huge impact on healthcare systems leading to an immense pressure on healthcare systems, expressed into shortages of beds, medical supplies, and healthcare workers. The businesses closing and layoffs led to millions of job losses, and the adoption remote work policies. On the education system the schools were closed, being the promotion of the online learning the found solution to overcome this. The inequalities in access to technology, and internet connectivity worsened the social differences and put huge challenges for teachers, students, and parents. The social isolation, the fear of infection, and the financial stress contributed to an increased anxiety, depression, and emotional stress, increasing the problems related to mental health, among others [4,5].

However, on the other hand, the tireless search for fast and efficient means of diagnosis and effective therapies, has boosted several areas of science and research. The importance of science in the context of COVID-19 was immense, playing a crucial role in various aspects of understanding, managing, and fighting the pandemic. In understanding the virus, in the vaccine development, in the implementation of public health measures, in epidemiological modelling and in the monitoring the emergence of new variants (Figure 2).

Different kind of tests of COVID-19, namely molecular, antigen and serological, have been a key factor for diagnosis and for managing the pandemic. Real-time reverse transcription-polymerase chain reaction (RT-PCR) and rapid antigen detection tests are the most effective and fast strategies developed to diagnose patients with COVID-19 and control infections in hospitals and communities [11,12]. However, the collection of nasopharyngeal exudates is an invasive procedure that puts healthcare workers at risk of disease transmission due to patients’ sneezing, coughing, or gag reflex [13]. Other limitations of the RT-PCR test are related to the long turnaround time and the requirement for expensive laboratory equipment and highly trained laboratory staff [12,14]. In turn, the results obtained by the antigen detection test may be unreliable because there is no RNA amplification, and in the case of low viral loads, the virus may not be detected, leading to false-negative results [15].

Additionally, the associated costs are relatively high. Besides the enormous capabilities of RT-PCR and other diagnosis platforms, more sensitive and accurate detection assays of SARS-CoV-2 are needed for an early diagnosis. In this context the development of improved analytical tests, simultaneously highly accurate, sensitive, and ultrafast, are extremely important, to diagnose early-stage and even asymptomatic individuals and, therefore, increase and improve the prevention and treatment efficiency.

Studies of secondary volatile organic metabolites (VOMs) have shown promising results in the identification of potential biomarkers for oncological, inflammatory, and respiratory infectious diseases [16,17,18,19,20,21]. VOMs are a rich source of information regarding the health status of an individual, as changes in the levels of these metabolites may be characteristic of specific disease processes [16,22]. VOMs can be found in various biological matrices such as blood, urine, saliva, exhaled air, faeces, and skin exudates [18,20,23,24,25,26,27]. Urinary VOMs are considered intermediate or final products of metabolic pathways [16,28]. Additionally, urine sampling is non-invasive and causes no discomfort to the patients. VOMs analyses require sensitive procedures to avoid contamination and sample loss. Solid-phase microextraction (SPME) is a simple, highly efficient, and easy-to-perform extraction technique that does not require a pre-concentration step prior to analysis [29,30]. The combination of SPME in headspace mode (HS) with gas chromatography coupled with mass spectrometry (GC-MS) analysis allows reliable and reproducible results and has been widely used for the analysis of urinary VOMs [16,17,20,27].

This study aimed to establish the urinary volatilomic profile of COVID-19 patients using HS-SPME/GC-MS as a strategy to identify potential biomarkers for the detection of this infection. The chromatographic data obtained were subjected to multivariable statistical analysis to identify volatile signatures that could discriminate between the presence of COVID-19 and its progression.

2. Materials and Methods

2.1. Chemicals and Reagents

Sodium chloride (NaCl, 99.5%) was obtained from Panreac AppliChem ITW Reagents (Barcelona, Spain). Ultrapure water, produced by a Milli-Q water purification system (Millipore, Bedford, PA, USA), was used to prepare the solutions of hydrochloric acid (HCl, 37%) 5 M and 3-octanol (internal standard, 99%) 2.5 ppm, both acquired to Sigma-Aldrich (St. Louis, MO, USA). Helium of purity 99.9% (He, N60, Air Liquide, Algés, Portugal) was used as the GC mobile phase. Glass vials, SPME holder, and fused silica fibre coating partially cross-linked with 50/30 µm divinylbenzene/carboxen/polydimethylsiloxane (DVB/CAR/PDMS) were purchased from Supelco (Merck KGaA, Darmstadt, Germany). DVB/CAR/PDMS fibres were conditioned according to the manufacturer’s guidelines. Prior to the first daily analysis, the fibres were conditioned for at least 10 min at the operating temperature of the GC injector port.

2.2. Study Design and Samples

This study included a set of 133 subjects: 42 individuals with no known infection (Control Group, CTRL), 61 individuals infected with SARS-CoV-2 (COVID Group, COVID) admitted to the Dr Nelio Mendonça Hospital between 15 May and 20 June 2020 (59 % male and 41% female, age average=56.8±18.6Y), and 30 recovered COVID-19 patients (Recovered Group, RECOV) (58 % male and 42% female, age average=60.7±16.3Y), with a recovery period of more than 3 months. Urine samples were collected at Dr. Nélio Mendonça Hospital (Funchal, Portugal). Samples from healthy subjects were obtained from blood donors, samples from the COVID-19 group were collected from patients diagnosed with COVID-19, and samples from the RECOV group were obtained from patients in their recovery period one month after COVID-19 infection. Upon collection, the samples were frozen at -20 °C until further analysis. This study was approved by the Ethics Committee of Hospital Dr. Nélio Mendonça. All the work described was carried out in accordance with The Code of Ethics of the World Medical Association (Declaration of Helsinki) for experiments involving humans. Informed consent was obtained from all the subjects recruited for this study, their privacy was strictly preserved, and any data beyond the SARS-CoV-2 infection were included in the study. Therefore, dimensions such as age, diet, previous diseases, sex, and sex, were not considered in this study.

2.3. Urinary Volatilome Analysis and Data Processing

HS-SPME extraction was performed according to previously optimised conditions for urine sampling [9]. Briefly, 4 mL of urine sample, previously adjusted to pH 1–2 with 500 µL of HCl (5 M), 0.8 g NaCl and 5 µL 3-octanol 2.5 ppm were placed in an 8 mL glass vial. The vial was placed in a water bath set at 50.0 ± 0.1 C with stirring at 800 rpm, and the SPME fibre was exposed to the headspace for 60 min. After extraction, the SPME fibre was collected and inserted into the injector port of the GC-MS instrument for 6 min at 250 °C, where the analytes were desorbed and transferred directly to the column. Each sample was analysed in triplicate.

The GC-MS analysis was performed in a gas chromatograph Agilent Technologies 6890N Network GC System (Palo Alto, CA, USA) equipped with a BP-20 fused silica column (30 m × 0·25 mm ID × 0·25 µm (SGE, Dortmund, Germany)), and connected to an Agilent 5975 quadrupole inert mass selective detector. The separation of the VOMs was carried out with a temperature gradient of 35 °C for 2 min, followed by an increase to 220 °C (2.5 °C min^-1), remaining at this temperature for 5 min, for a total GC run time of 81 min. The column flow rate was maintained at 1 mL min^-1. The injector port was operated in the splitless mode and maintained at 250 °C. For the 5975MS system, the temperatures of the transfer line, quadrupole, and ionisation source were 270 °C, 150 °C, and 230 °C, respectively. Data acquisition was performed in scan mode, 30-300 m/z, and the electron multiplier was set to the auto-tune procedure, with the electron impact mass spectra at 70 eV and the ionisation current at 10 µA. The VOMs were identified by comparing the mass spectra obtained with those available in Agilent MS Chemstation software (Palo Alto, CA, USA), which was equipped with a NIST05 mass spectral library with a similarity threshold of 80%.

2.4. Statistical Analyses

The data analyses were performed using Microsoft 365^® and MetaboAnalyst 5.0. The data matrix was normalised using the median, log transformation, and mean centring. The normalised data were processed through univariate analysis, specifically a t-test (p-values < 0·05) to identify statistically significant VOMs. Subsequently, multivariate analysis was performed using partial least squares discriminant analysis (PLS-DA). Important variables of the generated PLS-DA model were identified based on the variable importance in projection (VIP) score. The model was further evaluated using a 10-fold cross-validation (CV) and permutation tests. Finally, potential biomarkers were validated through receiver operating characteristic (ROC) curves created using Monte Carlo CV (MCCV) methodology to evaluate the accuracy and precision of the biomarkers.

3. Results

3.1. Characterization of the Urinary Volatilome

The volatile composition of the 132 urine samples was analysed using HS-SPME/GC-MS. The supplementary Figure S1 shows the typical chromatographic profiles obtained for the control (CTRL), COVID-19 patients (COVID), and recovered subject (RECOV) groups using the HS-SPME/GC-MS methodology. Overall, a larger number of peaks with higher intensities were observed in the chromatographic profile of the COVID group than in the CTRL profile (Figure S1, Supplementary material). Furthermore, the number and intensity of peaks in the RECOV group were intermediate between those in the COVID and CTRL groups.

A total of 101 VOMs belonging to 13 chemical families were identified. Data regarding the VOMs detected in the analysed samples, frequency of occurrence, and mean relative peak areas are available in Supplementary Table S1. As shown in Figure 3, terpenes, phenolic compounds, norisoprenoids, and ketones were the main contributors to the urinary volatile profiles of the studied groups. However, there were significant variations in the levels of these and less represented chemical families among the groups studied, with decreased levels of terpenes, phenolic compounds, benzene derivatives, hydrocarbons, and aldehydes in COVID-19 patients compared to the control subjects. In contrast, increased levels of norisoprenoids, ketones, alcohols, furans, sulfur compounds, and naphthalene derivatives were observed. For most chemical families in the RECOV group, the sum of the average peak areas was similar to that of the control group, except for benzene derivatives, terpenes, and phenolic compounds, whose levels were similar to the Covid19 group. It should also be highlighted that the variations in the relative levels of the different chemical families are broadly caused by an increase in the decrease of the same VOMs, as the numbers for each chemical family in each group do not vary significantly. The only exception are ketones, with 10, 9, and 13 ketones identified in CTRL, COVID and RECOV groups, respectively.

Volatile organic metabolites can have different origins. They can be endogenous as result of bacterial activity or pH changes, can be the product of metabolic pathways or oxidative stress. They can be influenced by external factors, including health status, diet, habits, physical stress, and environmental exposure. So, the human metabolom is highly complex, it’s difficult to understand if an increase or decrease in certain metabolites are related with a specific disease or illness.

This is why it is crucial to establish a relationship between the identified VOMs and their potential endogenous origin; however, the origin of many VOMs have not been clearly defined. Figure 4 shows the metabolomic pathways responsible for the origin of some chemical groups of endogenous VOMs.

3.2. Chemometric Analysis of Urine Samples

A data matrix of the relative peak areas of the 101 VOMs identified in the three groups under study, the COVID, RECOV, and CTRL groups (Supplementary Table S1), was processed using the Metaboanalyst software package [31]. Only VOMs with a frequency of occurrence (FO) higher than 80% in the volatile composition of urine were considered. To obtain a consistent distribution without redundant values, the variables were normalised, and univariate analysis was performed using a t-test (p < 0.05). Consequently, 17 VOMs with insignificant contributions to the statistical analysis were removed from the data matrix.

In multivariate pattern recognition procedures like Partial Least Squares Discriminant Analysis (PLS-DA), the information present in the VOMs fingerprint was utilized as multiple variables to visualize group trends and clusters. The outcome of the PLS-DA analysis revealed that there were two distinct groups well-separated for the COVID-CTRL comparison analysis. To evaluate the robustness of the model, a 10-fold cross validation performance was conducted using PLS-DA (Figure 5). Figure 5c showed R2 and Q2 values close to 1, which represent respectively the goodness of fit and the predictive ability for distinguishing between different study groups. (*) represented the best value of Q2 for the PLS-DA model. A random permutation test involving 1000 permutations, was carried out to assess the statistical significance of the class discrimination obtained (Figure 5d). Additionally, the top 10 variables of important in projection (VIP > 1) score plot was reported (Figure 5b) that illustrates the relative contributions of the metabolites in explaining the variance observed between the COVID and CTRL groups. 1,1,6-Trimethyl-dihydronaphthalene (TDN) and 2-heptanone showed a more significant contribution for the COVID groups and D-carvone and 3-Methoxy-5-(trifluoromethyl)aniline (MTA) showed a more significant contribution for the CTRL group.

The same multivariate analysis was performed to compare the data of SARS-CoV-2 infected urine samples with the recovered from COVID-19 urine samples (Figure 6). Even in this case, the PLS-DA segregated the COVID and RECOV samples in two well-separated clusters corresponding with the infected and on the mend patients (Figure 6a). The 10-fold CV performance and permutation test showed a good robustness of the PLS-DA model (Figure 6c,d). In the Figure 6b, VIP score plot revealed that β-damascenone α-isophorone gave a higher significant contribution to discriminate COVID group and nonanoic acid and α-terpinene gave a most significant contribution to discriminate RECOV group.

Hierarchical clustering analysis of the volatilomic data was carried out for the two comparisons, COVID-CTRL and COVID-RECOV, through the heatmap and dendogram. Heatmap was created using Spearman’s distance correlation to build a visual representation of the data set, specially focusing on the 15 most relevant metabolites to discriminate the two groups of study. Heat map allows an intuitive description of the relationship between samples and detected volatile metabolites. The coloured representation of the cells corresponds to the concentration of the detected VOMs for each sample (dark blue, less concentrated, and dark red, more concentrated).

The comparison between COVIDand CTRL groups, the analysis revealed two well-defined clusters (Figure 7a). The urinary VOMs 2-methoxythiophene, toluene, α-isophorone, TDN, hemimellitene showed a higher correlation with urine profile of COVID-19 patients. Piperitone, β-ionone, D-carvone, eudalene resulted more related with urinary profile of CTRL (control) group. Dendogram was able to split completely the samples in two groups matching the real group of study (Figure 7b). Although the heatmap was perfectly able to cluster volatilomic data from COVID-19 patients and recovered individuals, the clusters accuracy was visually lower than the first analysis (COVID-CTRL), highlighting that the COVID-19 patients’ urinary profile is closer to the recovered individuals. Urinary VOMs such as hemimellitene, furan, β-damascenone, and α-isophorone showed a higher correlation with COVID group, while 2,4-dimethylbenzaldehyde, nonanoic acid, 1-methylcycloheptene, and α-terpinene were more related with recovered indiciduals’ volatile profile (Figure 7c). Dendogram divided only partially the samples of these two different groups (Figure 7d).

For the classification of true positives and false positives and the predictive ability, multivariate exploratory receiver operating characteristic (ROC) curves were created using Monte Carlo cross-validation (MCCV) methodology. The features importance, selected using 2/3 of the samples, were utilized to construct classification models, which were validated on the remaining 1/3 of the samples that were not initially used. This process was repeated several times to determine the performance of each model and calculate confidence intervals. From these samples, the top 3, 5, 10, 20, 30, and 61 important features were identified, and the built curves were reported (Figure 8a,c). Figure 8a display the ROC curve for different sets of important features for the COVID-CTRL (COVID-19 patients and control subjects). The area under the curve (AUC) values obtained, ranging from 0.988 to 1, indicated excellent discriminative accuracy between the two groups. Plot in the Figure 8c illustrated the ROC curves for the COVID-RECOV comparison (COVID-19 patients and infected subject in the recovering period). In this case, the area under the curve (AUC) values fell in a range from 0.937 to 0.987 that also show an optimal ability to discriminate between the groups of study. These values were calculated with a 95% confidence interval, demonstrating the reliability of the results. Figure 8b,d illustrated the predictive accuracy of the biomarker models as the number of features increases. As more features are included in the models, the predictive accuracy improves. This suggests that the selected features contribute to the differentiation between the controls (CTRL) and COVID-19 (COVID) groups, and between COVID and recovered (RECOV) groups. The predicted class probabilities were assessed through the performance of the classification model for COVID-CTRL groups (Figure 8e) and COVID-RECOV groups (Figure 8f). Overall, the results demonstrate promising performance of the biomarker models, with high accuracy in distinguishing between the two groups.

The boxplots of the most important variables (VOMs) for discriminating between COVID and CTRL groups, and between COVID and RECOV, were plotted in Figure 9.

Some VOMs, such as D-carvone, MTA, TDN, and α-terpinene, are associated with diet [13,14,15,16,32,33] and as a result, their interpretation as potential biomarkers for COVID-19 infection and progression is not straightforward.

4. Conclusion

A novel, fast, and sensitive analytical approach was developed and successfully applied in diagnosis of COVID-19 infection. This exploratory study using HS-SPME/GC-MS, unveiled significant differences in the volatilomic patterns of COVID-19 patients and recovered patients compared to control subjects, evidencing that SARS-CoV-2 infection triggered metabolic changes that also affect the urinary volatile composition of the infected patients. This constitutes a signature with potential for COVID-19 diagnosing and monitoring of the disease progression with potential to be successfully used in clinical applications. Overall, the results show the feasibility of using urine samples for the non-invasive COVID-19 diagnosis and further studies with larger cohorts are desirable, envisaging the developing of this complementary tool for COVID-19 diagnosis.