Background
Rheumatoid arthritis (RA), psoriasis (Ps), psoriatic arthritis (PsA), systemic lupus erythematosus (SLE), Crohn’s disease (CD), and ulcerative colitis (UC) are prevalent immune-mediated inflammatory diseases (IMIDs) [
1‐
4]. This group of diseases is characterized by the aberrant and chronic activation of the immune system, affecting one or more tissues. IMIDs have a high socioeconomic impact [
1,
4,
5] and are among the main causes of morbidity, disability, and mortality in developed countries [
6‐
8]. Although each IMID targets different tissues and organs, they all share common molecular mechanisms like the activation of the Tumor Necrosis Factor cytokine pathway [
9]. Recently, genome-wide association studies have demonstrated that IMIDs also share many genetic risk loci [
10]. Consequently, the combined analysis of multiple IMIDs has the ability to leverage the identification of more relevant molecular features.
Improvements in the diagnosis of IMIDs would be of great benefit to the patient and would significantly reduce the socioeconomic burden of these diseases. There is increasing evidence that the administration of therapies, particularly biological treatments, at earlier stages of the disease results in a more effective control of the inflammatory process [
11,
12]. In RA, for example, early diagnosis and treatment have been shown to increase the probability of entering disease remission [
13‐
16], an accomplishment that was unthinkable only a decade ago. Similarly, the diagnosis of inflammatory bowel diseases CD and UC is often established too late, when severe complications have already occurred [
17]. The identification of more accurate diagnostic biomarkers would therefore have a high impact on the improvement of disease outcomes in IMIDs.
Measuring disease activity is also a challenging problem in IMIDs. The lack of objective and highly informative markers of disease activity has a negative impact in key aspects of patient management, like the decision to initiate or terminate a specific therapy. Currently, different scores are available to measure disease activity in each IMID. These scores are based on clinical, laboratory, and/or imaging measures, and although they are frequently used in clinical practice, they have important limitations [
18]. Disease activity scores are often based on unspecific and sometimes subjective variables that significantly increase their inter- and intra-observer variability, clearly reducing their accuracy and, consequently, affecting disease monitoring [
19]. The dynamic nature and highly informative properties of biological molecules (i.e., biomarkers) could provide the level of objectivity and accuracy necessary for a better management of disease activity in IMIDs.
High-throughput analysis technologies are able to generate comprehensive profiles of different molecular species from multiple biological samples. Recent developments in these technologies could provide the level of precision that is required to improve disease management [
20‐
22]. However, one limitation in the use of these approaches to study IMIDs is that the target tissue or organ cannot be easily sampled, resulting in a highly invasive procedure. Instead, the use of more accessible surrogate tissues or biofluids like blood, saliva and urine could help to circumvent this limitation. Urine, in particular, is a highly interesting sample source since its collection is very simple and is clearly non-invasive for the patient. The direct relationship with blood composition strongly supports the hypothesis that different molecular species that are present in both biological fluids like metabolites, nucleic acids, or proteins and whose variation is associated with pathological features could be highly informative biomarkers in IMIDs [
23,
24].
The profiling of the metabolite composition of biological samples, metabolomics, is one of the most rapidly evolving high-throughput analysis approaches [
25]. Metabolites could potentially serve as biomarkers in many diseases since they represent the biochemical end products of the genetic pathways, providing an accurate representation of the physiological state of an individual [
26]. Nuclear magnetic resonance (NMR), together with mass spectrometry, is one of the most widely used metabolomic technologies [
27]. NMR has been used in the determination of the metabolite profiles of tissue and biofluid samples of multiple diseases [
28,
29]. To date, however, very few studies have analyzed the metabolomic profiles of IMIDs and most lack independent validation cohorts. Further, there is a lack of studies comparing the metabolomes of this group of inflammatory diseases in parallel.
In the present work, we have performed a large-scale high-throughput analysis of the urine metabolome of six of the most prevalent IMIDs (RA, PsA, Ps, SLE, CD, and UC) and a cohort of healthy control individuals in order to identify new biomarkers associated with disease diagnosis and disease activity. For this objective, we have used a two-stage study design consisting of a discovery stage where the urine metabolomes of 1210 IMID patients and 100 healthy controls were analyzed, and a validation stage where the most significant candidate metabolite biomarkers from the discovery stage were confirmed using an independent cohort of 1200 IMID patients and 200 healthy controls. To our knowledge, this study provides the first comprehensive characterization of urine metabolites associated with IMIDs.
Discussion
The metabolome represents the collection of small molecules produced by cells and, therefore, its analysis is providing a unique opportunity to identify biological perturbations associated with diseases [
29,
45‐
47]. New technological advances are allowing the characterization of such biochemical variations, revealing unexpected metabolic changes associated with different human pathologies. From a translational perspective, the analysis of the metabolome is beginning to provide new and powerful biomarkers that are highly informative of specific disease processes and, therefore, could lead to more precise and efficient patient management. Despite their prevalence, there remain few studies analyzing the metabolome of IMIDs. In the present study, we report, for the first time, the results of a parallel analysis of the urine metabolome of six of the most prevalent IMIDs – RA, PsA, Ps, SLE, CD, and UC – for the search of clinically relevant biomarkers. Using a two-stage approach we have identified and validated multiple urine metabolites associated with disease diagnosis as well as disease activity. These results provide the most comprehensive analysis of the urine metabolome in IMIDs performed to date, leading to the identification of new biomarker metabolites, as well as providing strong evidence of shared metabolic pathways in this group of diseases.
The present large-scale profiling of the urine metabolome study has found unexpected strong similarities between IMIDs. Some of these metabolite variations were common across all or almost all diseases and, therefore, were considered as hub metabolites. To our knowledge, it is the first time that hub metabolites have been described in IMIDs. Among these metabolites, citrate, a central metabolite of the Krebs oxidative phosphorylation cycle, showed the strongest association to all IMIDs. Despite its essential role in cell energy production, citrate has been recently shown to have important immunologic properties [
48], modulating, for example, the production of proinflammatory factors in macrophages or being a critical factor for dendritic cell antigen presentation. Previous studies have found that citrate is present at lower concentrations in the urine of inflammatory bowel disease (IBD) patients compared to controls [
49,
50]. In RA and SLE, citrate has also been found to be in lower levels in the serum of patients compared to controls [
51,
52]. Here, we show that the previously observed citrate variation in RA and SLE is also detected in urine, a much less invasive sample source than whole blood. Finally, we also demonstrate, for the first time, that Ps and PsA patients also have low concentrations of urine citrate compared to healthy controls. Together, the results of this study provide strong evidence of the presence of hub metabolites that could become “pan-IMID” biomarkers that could be easily measured in routine clinical settings.
The parallel analysis of this group of diseases has led to unique findings. The unsupervised analysis of the urine metabolite associations showed three strong and reproducible clusters of clinically similar IMIDs: (1) IMIDs involving skin affection (i.e., Ps and PsA), (2) inflammatory bowel diseases (i.e., CD and UC), and (3) RA and SLE, two diseases characterized by having a higher prevalence in women. These results correlate with the observed shared genetic risk components observed between different IMIDs using genome-wide association studies [
53‐
56]. For example, CD and UC have shown to share more than 163 disease risk loci [
57], Ps an PsA share up to 30 risk loci [
58,
59], and SLE and RA have more than 80 common risk variants [
60]. To our knowledge, it is the first time that metabolite patterns in urine have shown to etiologically group more similar IMIDs. This result confirms the validity of the urine metabolome in the characterization of biochemical pathways that are specifically associated with this group of diseases.
When assessing the metabolic context of the disease-associated metabolites by integrating the metabolic reactions that link them, the resulting network showed a high degree of overlap of three main metabolic pathways (Fig.
1). From these, the citric acid cycle is the predominant pathway identified, with citrate showing a common association to all the IMIDs. Previous studies have already shown that alterations within this metabolic pathway are related to immunity and inflammation, although the functional implications of the alterations of this pathway are still being investigated [
61]. The second major metabolic pathway was the phenylalanine metabolism pathway. The metabolites included in this pathway have shown relevant and specific associations to IBDs in this study. This finding agrees with previous metabolomic studies that have shown the importance of this pathway in the etiology of IBDs [
62]. Finally, network analysis also showed an important role for the glycine and serine metabolism pathway in IMIDs. Metabolites within this pathway act as major connectors between the two previous pathways and have been previously related with inflammatory processes. Glycine, the most connected metabolite in the resulting network, has been previously proposed to be an anti-inflammatory and immunomodulatory agent [
63]. Although not directly detected by the NMR approach used in this study, our results strongly suggest that glycine could be a highly informative biomarker to the inflammatory processes that characterize IMIDs. Future studies using alternative analysis technologies like mass-spectrometry will help to determine the utility of this metabolite as a clinical biomarker of autoimmune diseases.
In this study, we also demonstrate that the urine metabolome has great potential for assessing disease activity. Citrate, the strongest hub metabolite for IMID diagnosis, was found to correlate with high disease activity in CD, PsA, and SLE. In IBDs, we also demonstrate that hippurate has a very strong correlation with disease activity. Therefore, this urine metabolite could be used not only for early disease diagnosis but also to monitor the level of disease activity in IBDs. This result further strengthens previously reported results that show how changes in the microbiome correlate with the level of inflammation in the gut and disease activity in IBD patients [
64‐
67]. Future studies, aimed at characterizing the interrelation between bacterial species in the gut, tissue inflammation and the urine metabolites identified herein could therefore help to develop more objective and reproducible systems to monitor disease progression in IBDs.
The disease diagnostic models built in this study using the urine metabolites were found to have good performance in all IMIDs. In IBDs in particular, the classifiers were found to predict the disease with very high accuracy. These results are in agreement with previous studies [
50,
68,
69] that suggested the use of urine metabolites for the diagnosis of IBDs. Compared to previous studies, we here provide, for the first time, a validation analysis of the diagnostic predictor using an independent and large patient and control cohort. Providing an independent confirmatory analysis is an essential step for any new molecular diagnostic tool [
70]. These findings support the analysis of the urine metabolome as a simple, cost-effective and non-invasive approach for the diagnosis of IBDs.
To our knowledge, there is no evidence that the metabolite patterns associated with IMIDs in this study have been previously associated to other diseases. While variations in single metabolites like citrate have been associated with other disease etiologies, the diagnostic ability generated by the combination of multiple metabolites clearly holds a much higher potential to be the approach finally used in the clinical setting. As shown in this study, it is the integration of variation in multiple metabolites that gives the best disease prediction accuracies. In order to further consolidate these diagnostic metabolite patterns as clinically useful tools, the next steps will include the study of the urine metabolome in individuals with pre-diagnostic symptoms as well as longitudinal studies to assess biomarker variability and correlation with specific features of disease progression. Further, future developments of the disease predictors could evaluate the inclusion of other molecular features like the presence of autoantibodies in sera or, even, the identification of additional metabolites in urine using mass-spectrometry approaches. For this latter objective, the results of this study will clearly be a highly valuable starting point.
Acknowledgements
This work was supported by the Spanish Ministry of Economy and Competitiveness grants (IPT-010000-2010-36, PSE-010000-2006-6, and PI12/01362) and by the AGAUR FI grant (2013/00974).
IMID Consortium: Emilia Fernández1, Raimon Sanmartí1, Jordi Gratacós2, Víctor Manuel Martínez-Taboada3, Fernando Gomollón4, 5, Esteban Daudén6, Joan Maymó7, Rubén Queiró8, Francisco Javier Lopez Longo9, Esther Garcia-Planella10, José Luís Sánchez Carazo11, Mercedes Alperi-López8, Carlos Montilla1, José Javier Pérez-Venegas12, Benjamín Fernández-Gutiérrez13, Juan L. Mendoza13, José Luís López Estebaranz14, Àlex Olivé15, Juan Carlos Torre-Alonso16, Manuel Barreiro-de Acosta17, David Moreno Ramírez18, Hèctor Corominas19, Santiago Muñoz-Fernández20, José Luis Andreu21, Fernando Muñoz22, Pablo de la Cueva23, Alba Erra24, Carlos M. González9, María Ángeles Aguirre-Zamorano25, Maribel Vera21, Francisco Vanaclocha26, Daniel Roig19, Paloma Vela27, Cristina Saro28, Enrique Herrera29, Pedro Zarco14, Joan M. Nolla30, Maria Esteve31, José Luis Marenco de la Fuente32, José María Pego-Reigosa33, Valle García-Sánchez25, Julián Panés4,1, Eduardo Fonseca34, Francisco Blanco34, Jesús Rodríguez-Moreno30, Patricia Carreira26, Julio Ramírez1, Gabriela Ávila35, Laia Codó36, Josep Lluís Gelpí36, Andrés C. García-Montero37, Núria Palau35, María López-Lasanta35, Raül Tortosa35
1Hospital Clínic de Barcelona and IDIBAPS, Barcelona, Spain. 2Hospital Parc Taulí, Sabadell, Spain. 3Hospital Universitario Marqués de Valdecilla, Santander, Spain. 4CIBERehd, Madrid, Spain. 5Hospital Clínico Universitario, Zaragoza, Spain. 6Hospital Universitario de la Princesa and IIS-IP, Madrid, Spain. 7Hospital del Mar, Barcelona, Spain. 8Hospital Universitario Central de Asturias, Asturias, Spain. 9Hospital Gregorio Marañón, Madrid, Spain. 10Hospital de la Santa Creu i Sant Pau, Barcelona, Spain. 11Hospital General Universitario, Valencia, Spain. 12Hospital de Jerez de la Frontera, Cádiz, Spain. 13Hospital Clínico San Carlos, IDISSC, Madrid, Spain. 14Hospital Universitario Fundación Alcorcón, Madrid, Spain. 15Hospital Universitari Germans Trias i Pujol, Badalona, Spain. 16Hospital Monte Naranco, Oviedo, Spain. 17Hospital Clínico Universitario, Santiago de Compostela, Spain. 18Hospital Virgen de la Macarena, Sevilla, Spain. 19Hospital Moisès Broggi, Barcelona, Spain. 20Hospital Universitario Infanta Sofía, Madrid, Spain. 21Hospital Universitario Puerta de Hierro, Madrid, Spain. 22Complejo Hospitalario de León, León, Spain. 23Hospital Universitario Infanta Leonor, Madrid, Spain. 24Hospital Sant Rafael, Barcelona, Spain. 25Hospital Universitario Reina Sofía, Instituto Maimónides de Investigación Biomédica de Córdoba (IMIBIC), Universidad de Córdoba, Córdoba, Spain. 26Hospital Universitario Doce de Octubre, Madrid, Spain. 27Hospital General de Alicante, Alicante, Spain. 28Hospital de Cabueñes, Gijón, Spain. 29Hospital Virgen de la Victoria, Málaga, Spain. 30Hospital Universitari de Bellvitge, Barcelona, Spain. 31Hospital Universitari Mútua de Terrassa, Barcelona, Spain. 32Hospital del Valme, Sevilla, Spain. 33Hospital do Meixoeiro, Vigo, Spain. 34Complejo Hospitalario Juan Canalejo, INIBIC, A Coruña, Spain. 35Rheumatology Research Group, Vall d’Hebron Hospital Research Institute, Barcelona, Spain. 36Life Sciences, Barcelona Supercomputing Centre, National Institute of Bioinformatics, Barcelona, Spain. 37Banco Nacional de ADN Carlos III, University of Salamanca, Salamanca, Spain.