Celiac disease (CD) is an autoimmune enteropathy triggered by the ingestion of gluten from wheat, rye, barley, oats, spelt, and their cross-related varieties (Schuppan et al., 2009). Based on serologic tests for transglutaminase or anti-endomysial antibodies, the CD prevalence was estimated to account for approximately 1.4%. However, this value reduced to 0.7% based on biopsy-proven screening (Singh et al., 2018). Until the 1970s, three biopsies from the small intestine were required to diagnose CD. Nowadays, a positive serologic test together with a small intestinal biopsy is sufficient to confirm the diagnosis (Singh et al., 2018). Noteworthy, some limitations deserve a deeper clarification. Approximately 3–5% of CD patients are seronegative (Schiepatti et al., 2018). Moreover, villous atrophy was not exclusively detected in CD, therefore, a biopsy can lead to an erroneous diagnosis (Schiepatti et al., 2019). Considering the invasiveness of these methods, novel approaches need to be considered. Flow cytometry is an example of alternative methods applied for CD biomarker assessment (Leon, 2011), as well as the emerging techniques based on the detection of gluten-specific CD4+ cells in peripheral blood mononuclear cells (PBMCs) (Kurki et al., 2021).

While showing symptoms related to diarrhea, lethargy, tiredness with or without anemia, and weight loss (Garrote et al., 2008), CD individuals suffer from various degrees of intestinal inflammation due to the interaction between gluten peptides and intestinal lamina propria (Elli et al., 2015). This leads to the secretion of several proinflammatory cytokines (e.g., IL-1b, IL-8, IL-15, IL-21, TNF-α, and MCP-1) (Schuppan et al., 2009; Iacomino et al., 2016). Specifically, literature reports a chronic upregulation of the IL-15 cytokine, whose increase in the epithelium and intestinal lamina propria is considered a hallmark of mucosal damage (Abadie and Jabri, 2014).

However, other compounds over cytokines are assessed during the pathological progression of many diseases and for this reason, they have been proposed as potential biomarkers of health. Some of them are metabolites related to gut microbiota activity (e.g., free fatty acids – FFAs), and some others might derive from the host. In line with this, evidence suggests that some variations in specific metabolic pathways markedly characterize various human pathological states, including CD (e.g., glycolysis, protein metabolisms, and lipids synthesis) (Caio et al., 2019). In individuals with overt CD, the gluten-free diet (GFD) is the only therapeutic approach currently available to reduce symptoms, villus atrophy, and restore intestinal functionality (Aljada et al., 2021). Although ameliorating CD individual well-being, GFD is not conclusive. Strict adherence to GFD is not easy to be followed also considering that gluten traces have been detected in many processed foods (Olshan et al., 2020). Furthermore, due to persisting symptomatology, malabsorption, and villus atrophy, clinical data showed that a percentage of CD individuals is composed of GFD non-responders. This condition is also known as refractory CD (RCD) and up to date no targeted therapies have been developed to treat them. In line with this, evidence supports that the degree of gluten sensitivity cannot be standardized worldwide due to a large interindividual variability, reflecting the combination of multiple co-factors such as genetic, immune, eating habits, and microbiota involvements (Kivelä et al., 2021). Additionally, GFD requires the replacement of wheat flour with other types of cereals and pseudocereals that usually are characterized by a higher glycemic index, such as corn and rice flours. Considering that diet is one of the major drivers of human gut microbiota composition (Marasco et al., 2016), it is reasonable to think that GFD can directly affect the host’s physiology and metabolism (Carroccio et al., 2021), as well as the gut bacterial activity (Bonder et al., 2016). For this reason, the interactions between the host’s genetics, GFD, and microbiota need to be deepened.

To investigate the metabolites associated with diagnosis and progression of CD (treated or not with a GFD), the present narrative review, based on a systematic search, argues on specific studies published in the 2009–2020 time window. Our primary aim was to provide a critical overview of the main metabolomic variations linked to the interaction between host, gut microbiota, and diet. This inspection offers an additional point of view about the chance of exploiting some specific metabolites as pathology biomarkers.

According to the fact that gluten constitutes an essential dietary component in Western societies, a strict lifelong GFD leads to the diversity and abundance of gut microbiota. Noteworthy, nutritional regimens, including GFD, involve the production of specific microbial and human metabolites (Bascuñán et al., 2020). In the selected studies, investigators applied different techniques to describe metabolic variations, such as gas- or liquid- chromatography (GC and LC, respectively) coupled with mass spectrometry (-MS), or proton nuclear magnetic resonance (H-NMR) spectroscopy. Besides, carbon source evaluation systems and spectrometer-based assays have been also adopted in a few selected studies.

Advancements in the field of microbiota (and microbiome) characterization by means of different meta-omics approaches are continuously increasing. The literature widely supports the hypothesis that any change in dietary habits shapes the gut microbial community (Moschen et al., 2012; Senghor et al., 2018; De Angelis et al., 2020). In turn, the composition and function of gut microbiota play a pivotal role in chronic inflammatory diseases (de Vos and de Vos, 2012). Although based on the host genetic background, CD needs to be included among inflammatory diseases due to its peculiar dietary-based influence.

According to previous evidence (Ryan et al., 2015), the metabolomic signature of CD encompassed three components: impaired energy metabolism, malabsorption, and gut microbiota alterations (Figure 2). Serum analyses from U-CD showed an increased concentration of glucose (Bertini et al., 2009; Bernini et al., 2011; Upadhyay et al., 2020) but reduced levels of pyruvate and lactate (Bertini et al., 2009; Fathi et al., 2013). This energetic impairment directly impacts the weakness and the chronic fatigue reported by U-CD at diagnosis (Jelsness-Jørgensen et al., 2018) and opens the way to another signature commonly observed in selected studies, that is, the high synthesis/utilization of lipids and ketone bodies as the energy sources instead of glucose. Some of the selected studies assessed higher levels of ketogenic metabolites in U-CD than in HC (Bertini et al., 2009; Di Cagno et al., 2009; Bernini et al., 2011; Upadhyay et al., 2020). Similarly, Bertini et al. (2009) analyzed the same subjects pre- and post-GFD and observed an inverse tendency in serum levels of β-hydroxybutyric acid, which is a ketone body usually increased during the fasting phase as an alternative energy source for brain activity (Ota et al., 2021). Hence, the significant decrease of β-hydroxybutyric acid (Bertini et al., 2009) seems to indicate the return to a normal energy metabolism via carbohydrates instead of ketone bodies driven by GFD.

Among selected articles, the authors also evaluated the energy balancing through SCFA profiling. Di Cagno et al. (2009) found a critical lower concentration of total SCFAs in U-CD when compared to HC, while subjects who undertook GFD (T-CD) did not exhibit the same behavior. Contrarily, other studies reported the complete absence of differences in the levels of total SCFAs (Primec et al., 2016; Niccolai et al., 2019; Zafeiropoulou et al., 2020) or their increase (Tjellström et al., 2010; Nistal et al., 2012; Caminero et al., 2015; Baldi et al., 2021). The adoption of different methods (GC- or LC-MS), sample type (serum or feces), or data elaboration (absolute or relative concentration) could have led to these controversial results. In addition, the SCFA metabolism is dependent on the availability of SCFA precursors (e.g., glucose, pyruvate, and lactate) but is also influenced by the dietary consumption of prebiotics (Figure 3). In turn, prebiotic consumption drives the abundance of SCFA-producing microbes in the gut (Golfetto et al., 2014; Makki et al., 2018). However, as recently reviewed by Verdu and Schuppan (2021), due to controversial results, the microbiota composition in CD did not show a unique fingerprint. Although some taxa shared common tendencies (i.e., increased abundance of Proteobacteria, Neisseria, and Escherichia coli as well as a decreased abundance of Bifidobacterium and Bacteroides ovatus), not a unique microbiota fingerprint is consistent with CD. Similarly, a lack of dietary fibers was claimed to characterize GFD (Thompson et al., 2005; Hopman et al., 2006; Martin et al., 2013), although this evidence is highly dependent on dietary habits that, in turn, are mainly influenced by culture and geography (De Angelis et al., 2020).

As for SCFAs, also findings concerning the protein metabolism showed a broad heterogenicity among selected studies. In the studies by Bertini et al. (2009); Bernini et al. (2011), Fathi et al. (2013); Martín-Masot et al. (2020), and Lamas et al. (2020), specific amino acids could be usually detected at low levels in U-CD serum. Oppositely, high concentrations of sulfur compounds and nitrogen derivatives have been found in urine (Bertini et al., 2009; Bernini et al., 2011; Di Cagno et al., 2011; Upadhyay et al., 2020), feces (Di Cagno et al., 2009), and saliva (Francavilla et al., 2014). On one hand, the low levels of amino acids in serum could suggest an impaired host metabolism as well as an unbalanced dietary intake (Martín-Masot et al., 2020). Instead, the collection of high concentrations of IS, p-cresol sulfate (pCS), and trimethylamine-N-ox (TMAO) require the presence of specific gut microbes able to metabolize the relative precursors (tryptophan or indole for both IS and pCS, while trimethylamine for TMAO). The occurrence of these metabolites allows us in speculating about two additional CD manifestations, which are the altered (reduced) intestinal permeability and the confirmation of high abundance of some specific taxa (e.g., Proteobacteria, Enterobacteriaceae, Escherichia, and Klebsiella) as shown by previous works (D’argenio et al., 2016; Jameson et al., 2016). Additional evidence about the impaired microbiota metabolism in CD comes to light from Bernini and co-workers (2011), who observed that using blind H-NMR-profiled urinary spectra, metabolites derived from microbiota [mainly IS, m-propionic (hydroxyphenyl) acid, and amino-acetic (phenylacetyl) acid] discerned U-CD from T-CD and HC.

About fats, the lipidomics carried out by Solakivi et al. (2009) explained those abnormalities in FA levels characterizing U-CD, as well as the benefits given by the GFD adoption. When evaluating the same individuals after 12 months of GFD, long-chain (C20–22) PUFAs increased although this variation was not sufficient to reach the levels found in HC. Long-chain (C20–22) PUFAs are important for optimal metabolism while deficiencies might contribute to the disease onset, including psychiatric diseases and depression that, indeed, are manifestations usually reported by U-CD (De Palma et al., 2014; Jelsness-Jørgensen et al., 2018). Impairments in FAs determined by CD were also argued by Jamnik et al. (2018). This research group showed that U-CD had a lower level of circulating HDL-cholesterol and apolipoprotein-A1. However, concerning HDL-cholesterol, two studies notified that GFD was able to reverse its levels after 6–8 months (De Marchi et al., 2013) and after 12 months (Lewis et al., 2009).

However, the first goal of GFD undoubtedly remains the improvement of the epithelial integrity, amelioration of the intestinal functionality, and the reduction of the inflammation status. In line with the results collected by Högberg et al. (2011) in urine profiles, also Uspenskaya et al. (2014) found that higher values of NO₂, NO₃, and total NOx discriminate serum from U-CD and HC. The assessment of high levels of these compounds suggests an ongoing inflammatory status in U-CD. Compared to HC, T-CD patients following a strict GFD showed no more significant differences in levels of all the NOx subtypes. Instead, in those T-CD that were on a semi-strict GFD (compared both with HC and T-CD on a strict GFD), higher median values of NO₃ and NOx persisted.

Despite the absence of fully concordant results, some alterations affecting U-CD can be highlighted in this study. The end-products from gluten metabolism activate the immune response in the hosts resulting in the overexpression of different pathways involved in the cascade of inflammation signaling. Already at the level of intestinal epithelial cells, an ongoing inflammation status was emphasized by the high levels of NOx found in serum and urine (Högberg et al., 2011; Uspenskaya et al., 2014). Consequently, many amino acids involved in the proliferation, repair, and healing of cells are altered being found increased or decreased based on their biological function as well as on the degree of inflammation (Upadhyay et al., 2020). For example, as deeply evaluated by Lamas et al. (2020), the Trp-derivatives activating the AhR-pathway were too low in U-CD determining an impairment in the AhR signaling of immune responses at the mucosal barrier site level. Moreover, both the bioavailability and the absorption of specific amino acids can directly affect the glycolysis that, in most U-CD patients, they were down-regulated (high glucose/low pyruvate and lactate) requiring the synthesis of alternative sources of energy (e.g., ketone bodies). The great heterogenicity of data reflects the broad combination of different affecting factors. First, although being under long-term GFD, the degree of gluten deprivation (strict/semi-strict) is widely determinant. In this line, evidence underlines how many patients can be labeled as non-responders to GFD (Kivelä et al., 2021). Furthermore, dietary habits, in terms of macro- and micronutrient intake, are pivotal factors in determining the microbiota fingerprint and the resulting gut metabolome. Therefore, the only assessment of gluten deprivation is not sufficient to stratify T-CD patients within a cohort. According to this consideration, when future metabolomic studies are developed, it may be useful to gather much more information on the type of GFD followed through the determination of macro- and micronutrient intake. Only after stratifying patients according to their dietary intake will the involvement of GFD also be further understood.

Further four studies (Sellitto et al., 2012; Kirchberg et al., 2016; Auricchio et al., 2019; Sen et al., 2019) fully verified our inclusion criteria but, considering their peculiar study designs, need to be separately discussed. Kirchberg et al. (2016) and Sellitto et al. (2012) performed two longitudinal cross-sectional studies enrolling infants. In CD-predisposed infants aged 4, the evaluation of metabolomes was not sufficient to anticipate CD-onset (Kirchberg et al., 2016). For this reason, the authors suggested that metabolomics was poorly predictive in the first years, reflecting that the serum metabolic pathways will be mainly affected only later in life. Due to the age of enrolled infants, also Sellitto et al. (2012) did not report differences in the aforementioned metabolites. The authors found that samples collected during the first 24 months after birth clustered according to age more than based on gluten introduction during weaning (Sellitto et al., 2012). This is in accordance with previous studies on infants, which reported how various prenatal and perinatal factors influence the early gut microbiota development (Koenig et al., 2011; Raspini et al., 2020). In addition, although the feeding and the time of solid food introduction were demonstrated to be among the main postnatal factors affecting gut microbiota in early phases of life (Differding et al., 2020; Raspini et al., 2021; Vacca et al., 2022), the fact that only one infant showed positive antibodies before 24 months of age in Sellitto et al. (2012), in our opinion, was not statistically sufficient to raise conclusions about effects related to an earlier or delayed gluten introduction in infants with genetically predisposed CD. Lipidomes, on the contrary, have provided more information as predictive tools to assess the CD onset later in life, suggesting that specific phosphatidylcholines might differentiate CD infants before gluten exposure (Auricchio et al., 2019). Sen et al. (2019) instead argued how several nonessential triacylglycerols were upregulated in infant CD progressors.

Celiac disease is a condition involving the host’s genetics and environment. The current literature widely supports the hypothesis that perturbations of the intestinal homeostasis exacerbate the CD symptomology through different mechanisms, and, in this context, the microbiome structure and function markedly contribute. Metabolomics gives the opportunity of studying the results coming from the interaction between host, diet, and microbiota, highlighting variations in healthy and diseased states. However, currently, it is not possible to give a unique statement about a metabolomic fingerprint related to CD because of the specific influence of diet, host genetics, disease grade, race or age, microbiome, and methodological approaches. The great heterogenicity of study designs, applied methodological approaches, and the limited number of research articles precludes the possibility to conclude. Ad hoc prepared procedures and databases collecting data and related metadata, together with standardized analysis protocols, are required. Therefore, even if there is a growing interest in exploring the factors triggering CD through meta-omics approaches, the prospective to having a CD diagnosis by applying a metabolomic non-invasive investigation is still very far away also considering the multiple expression forms of gluten-related disorders.

MV and FMC: conceptualization. AP, MV, TL, and II: investigation. MV, FMC, AP, and GC: writing—original draft preparation. All authors contributed to writing—review and editing. All authors have read and agreed to the final version of the manuscript.

DP was employed by the Giuliani SpA. The remaining authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. The handling editor declared a past co-authorship with several authors, MV, FMC, and MD.

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.