¹H-NMR urine metabolomic fingerprint for severity discrimination in pulmonary sarcoidosis

Abstract

Background

Sarcoidosis is a systemic inflammatory immune disease characterised by noncaseating granulomas. Its course can vary from benign to very severe, requiring appropriate treatment. Few biomarkers are available to monitor the management of these patients and to identify those at risk of poor prognosis. Given the systemic nature of sarcoidosis, we hypothesised that the analysis of urine metabolites could provide valuable biomarkers for the management of severe disease.

Methods

We conducted a comparative analysis of urine metabolomics in a consecutive cohort of 37 well-phenotyped patients, using ¹H nuclear magnetic resonance (NMR) with multivariate statistical analysis, followed by metabolite identification. After NMR spectra acquisition, we used principal component analysis and partial least squares (PLS) discriminant analysis to generate predictive models.

Results

A urinary metabolomic signature predictive of severe pulmonary sarcoidosis was identified using a PLS model. We selected five metabolites with significant changes in samples from severe sarcoidosis defined by a composite physiologic index >40 compared to those from nonsevere sarcoidosis. These changes correspond to the decreased levels of taurine, hippurate, serine and creatinine and increased levels of 3-hydroxyisovalerate. This metabolite profile suggests an association with activated inflammatory pathways.

Conclusions

This study shows that urinary NMR metabolites can discriminate samples between nonsevere and severe sarcoidosis. It suggests that urinary metabolomic studies in sarcoidosis may be particularly useful to identify potentially relevant biomarkers. However, further validation in a larger cohort of patients at different disease stages is warranted to confirm the relevance of these NMR biomarkers for follow-up.

Shareable abstract

Urinary metabolites analysed by ¹H-NMR spectroscopy show a predictive signature associated with severe pulmonary sarcoidosis, suggesting an association with inflammation. It is important to test this as a potential prognostic biomarker in sarcoidosis. https://bit.ly/432YcgP

Introduction

Sarcoidosis is recognised as a systemic inflammatory immune disorder of multiple presumed aetiologies, characterised by the presence of characteristic noncaseating granulomas composed of epithelioid cells in tissue observations [1]. It most commonly affects the respiratory and lymphatic systems, but virtually all organs may be involved, reflecting the extrapulmonary extent of the disease [1, 2]. Sarcoidosis can have different clinical consequences, ranging from benign to very severe and particularly aggressive sarcoidosis requiring sarcoidosis-specific treatment [3]. Some patients do not respond to treatment or relapse, progressing to severe sarcoidosis with marked lung function impairment, such as pulmonary fibrosis and pulmonary hypertension. Demographic studies have suggested an increased risk of mortality associated with sarcoidosis, but this increased mortality rate is predominantly seen in subgroups of sarcoidosis patients classified as “severe or at risk”. Despite poor interobserver agreement, the Scadding radiographic classification [4, 5] remains useful for predicting disease progression, while other findings are associated with an increased risk of mortality, such as pulmonary hypertension, pulmonary fibrosis and markedly altered lung function [5–7]. Recently, a reliable prognostic algorithm based on the composite physiologic index (CPI), taking into account the ratio of main pulmonary artery diameter/ascending aorta diameter and the extent of fibrosis on chest computed tomography (CT), has been shown to accurately predict mortality [5–7]. While various laboratory test abnormalities have been documented during the course of sarcoidosis, there is no consensus on which tests are best suited to assess disease activity, monitor treatment or evaluate prognosis [8].

Metabolomics has been applied to a wide range of diseases, offering the potential to improve diagnostic accuracy and potentially stratify prognosis. A recent review has highlighted the contribution of metabolomics in sarcoidosis, highlighting the links with the gut or respiratory microbiome [9]. This method has been used to study various physiological fluids under pathological conditions. In a previous study using ¹H nuclear magnetic resonance (NMR) spectroscopy, we investigated the metabolomics of saliva in patients with sarcoidosis [10].

Urine metabolomics has already been shown to be highly informative in several studies, especially in inflammatory diseases and cancer [11, 12]. For example, urine autotaxin levels may reflect disease activity in sarcoidosis [13]. While several ¹H-NMR studies of sarcoidosis patients have been reported in saliva or serum [10, 14–18], to the best of our knowledge, urine has not yet been investigated in sarcoidosis.

Therefore, in this pilot study, we conducted a ¹H-NMR spectroscopy analysis in a consecutive cohort of well-phenotyped sarcoidosis patients to identify specific changes in urinary metabolite content associated with the severity in pulmonary sarcoidosis, as previously demonstrated for other systemic diseases [19–21].

We aimed to identify discriminative urinary metabolites in a sarcoidosis population to assess their accuracy in identifying poor prognosis in relation to the CPI score. Therefore, urine metabolites were compared according to the status of severe pulmonary sarcoidosis. We found that ¹H-NMR spectroscopy analysis of urine samples could identify a significant metabolomic signature associated with the severity of pulmonary sarcoidosis most likely related to inflammatory pathways. These results need to be confirmed by larger and multicentre cohort studies.

Study population and methods

Study population

Urine samples were prospectively collected from consecutive chronic and recently diagnosed sarcoidosis patients between January 2018 and October 2019. All study patients were cared for at the Respiratory Medicine Department of the APHP Avicenne University Hospital (Bobigny, France), which is an expert centre for sarcoidosis. All included patients fulfilled the American Thoracic Society/European Respiratory Society/World Association of Sarcoidosis and Other Granulomatous Disorders diagnostic criteria for sarcoidosis [1, 22]. None of them had a history of systemic corticosteroid, hydroxychloroquine sulphate or immunosuppressive therapy in the previous 6 months. Other study exclusion criteria included age under 18 years, diabetes, use of a standard treatment regimen (including recent treatments such as antibiotics), urinary tract infection or urinary lithiasis at the time of sampling, and renal insufficiency. In accordance with French legislation, written informed consent was required from patients prior to enrolment. Approval was obtained from the French National Committee for the Protection of Human Subjects (2017/33;2017-A01555-48).

Data collected at the time of urine collection are shown in table 1. Respiratory function tests, forced vital capacity (FVC), forced expiratory volume in 1 s (FEV₁) and diffusing capacity of the lung for carbon monoxide (D_LCO) were expressed as % of predicted values. The CPI was calculated according to the following formula: CPI=91−(0.65×D_LCO)−(0.53×FVC)+(0.34×FEV₁) for each patient [7]. As previously reported, a CPI greater than 40 is considered to identify patients with poor vital prognosis, representing the so-called “severe sarcoidosis” phenotype.

TABLE 1.

Clinical data concerning the investigated patients with pulmonary sarcoidosis

Population	n=37
Female/male	11 (29.7)/26 (70.3)
Age (years)	46.4±11.7
BMI (kg·m−2)	24.8 (22.1–28.6)
Geographical origin
Europe	21 (56.8)
Sub-Saharan Africa or French Antilles	10 (27.0)
Maghreb region	5 (13.5)
Asia	1 (2.7)
Current smoker	7 (18.9)
Scadding radiographic stage
0	4 (10.8)
1/2/3	8 (21.6)/18 (48.6)/3 (8.1)
4	4 (10.8)
Fever	12 (32.4)
Impaired general condition	17 (45.9)
FAS score#	22.1±9.0
Extrapulmonary involvement	20 (54.1)
Severe extrapulmonary localisation (cardiac, neurological, renal)	4 (10.8)
Number of extrapulmonary localisation per patient	1 (1–2)
ACE (U·L−1)	75.5 (53.2–121.5)
ACE >ULN	21 (56.8)
Blood lymphocytes (per mm3)	1093 (775–1530)
FVC (% pred)	89.4±19.0
FEV1 (% pred)	83.2±18.2
DLCO (% pred)	71.4±16.9
CPI	23.0 (16.9–34.0)
CPI >40	6 (16.2)
Treatment initiation (after urine sampling)	23 (62.2)

Results are expressed as n (%), median (interquartile range) or mean±sd. Forced vital capacity (FVC), forced expiratory volume in 1 s (FEV₁) and diffusing capacity of the lung for carbon monoxide (D_LCO) are expressed as percentages of predicted values. ACE: angiotensin converting enzyme; BMI: body mass index; CPI: composite physiologic index; FAS: fatigue assessment scale; ULN: upper limit of normal. ^#: FAS score was calculated for 36 patients according to [44].

Prior to urine collection, in order to limit the various factors that could influence metabolite analysis, patients were advised to reduce their fluid intake (e.g. water) and to avoid meat, coffee, tea, alcohol consumption and exercise the day before. They were also asked to avoid grilled meat and smoked fish for 48 h. Daily diet and habits that could interfere with the metabolomics results were collected (table 2).

TABLE 2.

Data on main daily habits

Population	n=37
Daily treatment	4 (10.8)
Analgesic treatment	20 (54.1)
Antibiotics in the last month	2 (5.4)
Hormone replacement therapy	2 (5.4)
Meat consumption >once a day	9 (24.3)
Grilled meat consumption	4 (10.8)
Alcohol >1 glass per day	3 (3.1)
Tea or herbal tea >3 cups per day	4 (10.8)
Coffee >3 cups per day	9 (24.3)
Spicy food >once a day	9 (24.3)

Results are displayed as n (%).

¹H-NMR spectroscopy method and multivariate analyses

Urine collection

The first morning urine sample was collected (100–200 mL) into sterile urine containers and immediately centrifuged at 4 °C at the hospital. Filters (0.22 µm) were rinsed with distilled water and used for the supernatant. pH was measured and adjusted to 6.8 and 5 mL samples were stored at −80 °C prior to NMR analysis. At the time of analysis, 100 μL of deuterium oxide (Eurisotop) was added to 600 μL of thawed urine.

Sample preparation and NMR analysis

NMR profiling was performed using a 500 MHz Advance III Bruker NMR spectrometer (BioSpin Bruker, Wissembourg, France) on an automatic 24-position sample changer with cooling capability using a quadruple resonance inverse probe. The NMR spectra were acquired using the noesy1dgppr sequence at 298 K. The spectral width was set at 10 kHz. A Fourier transform was performed on the NMR data with the line broadening set at 0.3 Hz. All spectra were phased and baseline-corrected, followed by alignment. Finally, each one-dimensional (1D) NMR spectrum was sliced into 0.001 ppm buckets, containing NMR signals. These steps were performed using NMRPipe software. Prior to statistical analysis, the NMR spectra bins were normalised using the probabilistic quotient normalisation algorithm [23] and unit variance scaled. The assignment of the ¹H-NMR spectra was performed using two-dimensional experiments, Chenomx software and the Human Metabolome Database (version 5.0).

A principal component analysis (PCA) model was generated using an in-house MATLAB® script. The aim was to exclude possible outliers. A partial least squares (PLS) discriminant model was used to identify fingerprints related to patients’ age and sarcoidosis severity using CPI values as the Y-variable. Statistical analysis and visualisation were performed on MATLAB® R2016b for macOS (Mathworks, Natick, MA, USA) with scripts based on the method described by Trygg and Wold [24]. Models were evaluated using R² (Y). Q² (Y) was calculated using seven-fold cross-validation. Q² (Y) estimates the predictability of the model. The score plot, showing the discrimination between groups, and the loading plot, showing the spectral region responsible for this discrimination, have been calculated.

Results

Patients

42 consecutive patients were initially enrolled in the study. After acquisition of the ¹H-NMR spectroscopy data, it was determined that five urine samples had poor water suppression and a low signal-to-noise ratio that did not allow analysis. These samples were excluded from the final study, which was therefore performed on the samples of 37 patients. Table 1 shows the characteristics of the patients. Most were men (70%) and 19% were current smokers.

Patients with a “recent diagnosis” of sarcoidosis (i.e. within 2 years following the urine sample) represented 62.2% of the population (n=23). Patients with a chronic sarcoidosis (i.e. diagnosed 3 years or more before the urine sample) represented 37.8% (n=14).

Patients had an extrapulmonary involvement in 54%. The mean CPI was 23.9±15.0 and six patients (16%) had a CPI above 40, indicating severe sarcoidosis [7]. It should be emphasised that 62% of the patients required a steroid and/or immunosuppressive treatment initiation (started after urine collection). The CPI of patients with a “recent diagnosis” of sarcoidosis (25.2±14.1) was not statistically different from that of patients with “chronic” sarcoidosis (26.0±13.8). Similarly, the number of patients receiving treatment after urine collection did not differ between the two groups. All 37 patients followed the recommendations described above for the days leading up to urine collection. Table 2 presents data on the most important daily habits. The interference of each of these parameters with the urine metabolite results was tested.

¹H-NMR spectroscopy results

¹H-NMR spectra were recorded at 298 K for urine samples. The variability observed between the different samples was first investigated using unsupervised statistical analysis. We tested variability related to sex and age, as previously reported in urine samples from a cohort of healthy men and women [25]. The PCA results of our samples show that a part of the variability could be related to sex (figure 1a). Subsequently, supervised analysis was performed to investigate the potential variability according to age, as previously investigated in other pathologies [26]. A PLS model with three components was constructed (the score and the loading plots are shown in figure 1b). With the 37 urine spectra, we obtained a Q² of 0.55, indicating the model's ability to predict the age of the patients.

FIGURE 1.

a) Principal component analysis score plot and corresponding loading plot of the 37 ¹H nuclear magnetic resonance (NMR) urine spectra from the urine of patients with severe pulmonary sarcoidosis, illustrating the differences in gender fingerprints (females: yellow dots; males: blue dots). b) Partial least squares (PLS) score plot and corresponding loading plot of the 37 ¹H-NMR urine spectra. The first predictive x-axis discriminates urine profiles according to patient age. Three components were calculated, Q²=0.55. T1: scores of the first principal component (PC1); T2: scores of the second principal component (PC2); Ppca1: loadings of PC1 (principal component analysis); Wpls1: weights of the first PLS component.

As a third step, we searched for biomarkers that correlate with the severity of pulmonary sarcoidosis. It has been shown that the CPI could be considered a significant parameter for predicting mortality in fibrosing sarcoidosis [7]. For that purpose, the data were randomly divided into two sets of results. First, a set of 29 spectra was used as a cross-validation training set and a second set of eight urine spectra was used as a test set. A PLS model with four components was identified using the CPI as the Y-variable. The R²X was 0.275, the R²Y was 0.997 and the Q² was 0.68. We tested this model on the test set and an R² of prediction was calculated. For each cross-validation submodel, the prediction R² of the test set was calculated. These submodel R² results were used to build the confidence interval for the global prediction R² of the test set. We obtained a value of 0.74 (95% CI 0.62–0.84) for this model, showing the significant quality of the prediction. The resulting score plot is shown in figure 2a and the corresponding loading plot is shown in figure 2b. A permutation test (500 permutations) was performed to evaluate the robustness of the model (figure 2c). Figure 2d presents the predicted CPI (circle) versus the real CPI (cross). Confidence intervals in the CPI prediction for the test set were obtained using the cross-validation submodels. A prediction of the CPI for the test set was calculated by each submodel, allowing the estimation of a confidence interval. Figure 2e shows the projection of the sample showing the predictive power of the model. The first five discriminant metabolites with the highest variable importance in projection score in the PLS model (table 3) were selected to build a logistic model. These metabolites were used as biomarkers to predict whether the CPI was higher than 40, since this value of the CPI is considered critical [7]. Boxplots of corresponding intensities are presented in figure 3a. After unit variance scaling of the five selected signals, a logistic regression was performed (the coefficients used in the logistic regression are shown in table 3). In patients with a CPI higher than 40, a decrease in the concentration of taurine, hippurate, serine and creatinine and an increase in 3-hydroxyisovalerate is observed. A receiver operating characteristic curve of 0.89 (0.62–1.0) (figure 3b) was obtained on the training set. The confidence interval was obtained by bootstrap resampling. All eight urine samples in the test set were correctly predicted, demonstrating the strong performance and potential utility of this biomarker. The relationship between age and CPI score was tested by calculating the p-value of the Spearman rank correlation coefficient (0.19). This value suggests that there is no significant relationship between these two variables. The difference in CPI scores between men and women was assessed using the Mann–Whitney U test. The p-value (0.28) suggests that there is no significant difference between these two distributions.

FIGURE 2.

a) Partial least squares (PLS) score plot and b) corresponding loading plot of the 29 ¹H nuclear magnetic resonance (NMR) spectra of the training set. The first predictive x-axis discriminates the urine profiles according to the composite physiologic index (CPI) score. Four components were calculated (R²X=0.28, Q²=0.67). Wpls1: weights of the first PLS component. c) Values of R² and Q² after 500 permutation tests. d) Application of the calculated model to the eight one-dimensional (1D) NMR urine spectra and projection. The y-axis corresponds to the prediction axis (CPI) and the x-axis corresponds to the sample number. The projection of the 1D spectrum corresponds to the circle and is compared to the CPI value (cross). e) Projection of the eight 1D NMR urine spectra of the test set onto the model.

TABLE 3.

The five metabolites with the highest variable importance in projection (VIP) scores, showing their chemical shift, Spearman correlation and logistic regression results

	Chemical shift, ppm	VIP score	Spearman correlation		Logistic regression
			Coefficient	p-value	Coefficient	p-value
Taurine	3.29	6.14	−0.34	0.03	−3.95	0.0387
Hippurate	7.87	5.14	−0.55	0.0004	−1.48	0.0179
3-Hydroxyisovalerate	1.29	4.69	0.42	0.01	1.22	0.2467
Serine	3.98	4.58	−0.40	0.01	0.64	0.2012
Creatinine	3.06	4.48	−0.10	0.52	−1.51	0.1615

FIGURE 3.

a) Boxplots of the signal intensities of the selected metabolites, from the left to right: 3-hydroxyisovalerate, creatinine, taurine, serine and hippurate. b) Receiver operating characteristic (ROC) curve obtained on the training test set using the logistic regression. ROC curve area under the curve: 0.89 (95% CI 0.62–1.0).

Discussion

Metabolomics studies in sarcoidosis are rather scarce. A review by Kim et al. [9] very clearly presented the potential benefits of metabolomics to decipher the links between sarcoidosis and the metabolome of the microbiome, whether oral, intestinal or pulmonary. For diagnosis purposes, serum metabolic disturbances have demonstrated very high sensitivity and specificity in differentiating sarcoidosis from tuberculosis [16]. Therefore, we designed the present study to identify preliminary urinary metabolites that could be considered as severity indexes and prognostic biomarkers. We chose to investigate urinary metabolites using NMR spectroscopy, which has not been yet reported in sarcoidosis.

NMR spectroscopy is one of the main analytical techniques for metabolomics, with the advantages of minimal sample preparation and high reproducibility [27]. In addition, urine sampling is very easy, noninvasive and can be repeated. We chose to collect the first morning void urine at the time of the hospital visit, following all recommendations required for standard practice [28]. We have maximally controlled the experimental conditions to exclude bacterial contamination by 0.22 µm filtration and to ensure metabolite stability by storage at −80 °C prior to analysis, with no significant delay between urination and sample processing [28]. Patients were also asked to refrain from meat, coffee, tea and alcohol consumption the day before, as well as sporting activities. Patients were also asked to avoid eating grilled meat and smoked fish for 48 h before the test. Daily diet and habits that could interfere with the metabolomic results were recorded and checked to ensure that they did not interfere with the metabolites identified in this study.

The CPI was originally developed to assess idiopathic pulmonary fibrosis, another fibrotic interstitial lung disease [29]. Regression analysis was used to generate the weighted combination of lung function variables that best matched the extent of pulmonary fibrosis on CT [29]. In sarcoidosis, a CPI greater than 40 was found to be a robust predictor of mortality in two different cohorts, with hazard ratios ranging from 3.8 to 4.2 [6, 7]. It should be noted that no patient in our cohort had pulmonary hypertension or fibrosis greater than 20% on high-resolution CT, which are two other prognostic factors associated with reduced survival. This reinforces the accuracy of CPI selection in this study [6, 7]. In our patients, a high CPI is more likely to be associated with extended granulomatous lung inflammation than with the fibrotic process. This is consistent with the observed set of discriminant metabolites reflecting inflammation.

Five discriminant urinary metabolites, i.e. a decrease in taurine, hippurate, serine and creatinine and an increase of 3-hydroxyisovalerate, were identified by PLS and selected for their association with a poor prognostic CPI score, i.e. a CPI above 40 compared to patients with a CPI below 40.

Taurine is a semi-essential amino acid that plays a role in the regulation of oxidative stress and proinflammatory cytokines [30]. In serum, taurine levels were not significantly different between sarcoidosis patients and healthy volunteers [15], and an increase was even observed in veterans with sarcoidosis [14]. However, in these studies, sarcoidosis patients were considered globally and disease severity was not a specific objective. The decrease in taurine we observed in urine collected from patients with severe sarcoidosis is consistent with reports from studies in patients with inflammatory bowel disease [21].

Hippurate is one of the most abundant microbial host cometabolites [31]. A large study of urinary metabolites in immune-mediated inflammatory diseases found a highly significant association between a decreased hippurate level and two inflammatory bowel diseases, i.e. ulcerative colitis and Crohn disease [19, 21]. Furthermore, in a study of patients with severe fever with thrombocytopaenia syndrome, a decrease in urinary hippuric acid was observed, while an increase was observed in cases with fatal progression [32]. In addition, low urinary hippurate has been reported in patients with acute pancreatitis [33] or in obesity, which has been suggested to be associated with the gut microbiome dysbiosis [34].

Serine, a nonessential amino acid, is implicated in numerous different metabolic processes, such as antioxidant defence and regulation of metabolic homeostasis [35]. Serine alterations have been reported in inflammatory bowel disease, both in patients and in experimental models in mice [36]. A low level of urinary serine has been observed by mass spectrometry in children with Crohn disease compared to those with ulcerative colitis [37] and recently, in a mouse model of colitis, inhibition of serine synthesis or depletion played a role in modulating acute or chronic colonic inflammation [36]. Recent studies have shown that serine metabolism is related to macrophage functions; for example, exogenous serine supports M1 macrophage polarisation through mammalian target of rapamycin signalling [38], which is a key pathway in sarcoidosis [39]. However, in contrast with the study by Mirsaeidi et al. [17], an increase in serum N-stearoyl serine was found in patients with a fibrosing lung pattern.

In the present study, low urinary creatinine excretion was found in patients with severe sarcoidosis. At steady state, urinary creatinine excretion is one of the most specific indices of total body muscle mass [40]. These results should be considered in of the light of a previous study of 423 Dutch patients that showed that muscle wasting was present in 25% of sarcoidosis patients and was associated with more severe pulmonary disease [41].

In contrast, an increase in urinary 3-hydroxyisovalerate was found in patients with a CPI greater than 40, i.e. severe sarcoidosis. 3-Hydroxyisovaleric acid, is a byproduct of the leucine catabolic pathway, which has also been implicated in muscle protein catabolism in response to energy expenditure and inflammation [42]. Such an increase of 3-hydroxyisovalerate has been found in the urine of patients with Crohn disease [19]. In addition, 3-hydroxyisovaleric acid has been found to increase under hypoxic conditions [43]. Consistent with our findings, increased serum levels of 3-hydroxyisovaleric acid have been described in patients with severe COVID-19 and impaired lung function [43]. This increase has been suggested to be secondary to an altered mitochondrial branched-chain amino acid metabolism related to hypoxia [43] and activation of the gluconeogenic pathway. Indeed, COVID-19 severity and increased muscle catabolism have been associated with increased gluconeogenesis [43], which further supports our findings.

Considering the aggregate of these five main discriminating metabolites, it can be hypothesised that they are associated with the inflammatory status observed in severe sarcoidosis and point to a link between the gut microbiome and pulmonary sarcoidosis, the gut–lung axis highlighted by Kim et al. [9].

In conclusion, the present pilot study using urinary metabolite identification by high-resolution ¹H-NMR spectroscopy coupled with multivariate statistical analysis identified a metabolic signature significant for severe pulmonary sarcoidosis consisting of five main discriminating metabolites (decrease of taurine, hippurate, serine, creatinine and increase of 3-hydroxyisovalerate). The five identified metabolites formed an aggregated variable that achieved the best positive predictive value for the diagnosis of severe sarcoidosis according to the CPI, an index to identify severe pulmonary sarcoidosis. The results of this pilot study demonstrating the potential of urinary metabolomics need to be confirmed in validation cohorts and through longitudinal sampling. Therefore, the simple and robust NMR spectroscopy of urine samples may be of significant assistance in the follow-up of sarcoidosis patients and in monitoring treatment decisions.