Introduction

Chronic pruritus of unknown origin (CPUO) is defined as itch lasting greater than 6 weeks, where a causative etiology cannot be determined.1,2,3 CPUO can have several systemic, neurologic, psychological, or dermatologic etiologies.4 It is associated with significant impairment in quality of life, similar to that experienced by patients suffering from congestive heart failure or stroke.5 CPUO is a prevalent disease, with generalized pruritus ranging from 3.6% to up to 44.5%.6; however, it remains underrecognized due to not having a unique ICD-10 code. Despite its prevalence, CPUO is challenging to diagnose and manage as there are no Food and Drug Administration-approved medications, and little guidance is available for clinicians in the management of CPUO.1,7

The understanding of the pathogenesis of CPUO is limited, and metabolomic studies may facilitate understanding of CPUO pathogenesis and identification of novel treatment targets. Some studies propose that CPUO may result from dysregulation of itch-specific nerve fibers, neurotransmitters such as histamine and calcitonin gene-related peptide, and neuronal receptors. 7There is evidence for metabolic dysregulation in similar neuroimmune disorders, such as multiple sclerosis, in which serum metabolomics analysis has identified alterations in phospholipid and glucose metabolism.8 Thus, we employed mass spectrometry to perform an untargeted comparative plasma metabolomics analysis of CPUO and non-itch controls. We utilized metabolomics to elucidate the metabolic changes associated with biological processes within CPUO, providing information on processes downstream of the genome and proteome. In this study, plasma samples were used to provide a more systemic view of circulating metabolites that may not be fully captured by local tissue analysis.

Results

Clinical characteristics

Plasma samples from a total of 11 CPUO patients and 11 age-, sex-, and race-matched controls were collected and analyzed (Supplemental Fig. 1, Supplemental Fig. 2). Baseline clinical characteristics and comorbidities of CPUO patients and matched controls are included in Table 1.

Table 1 Matched demographic and clinical characteristics of chronic pruritus of unknown origin (CPUO) patients and Non-itch controls.
Full size table

Potential biomarker screening

Samples from CPUO patients and controls yielded 39 metabolites, which were normalized relative to the protein concentration of each sample. The features were first correlated with each other to establish relationships; the amino acids and amino acid derivatives were significantly associated with each other, as were the components of the purine and pyrimidine synthesis pathways and components of glycolysis and TCA cycle, respectively (Fig. 1a). In the unsupervised principal component analysis (PCA) overview, principal components 1 and 2 captured 87.3% and 7.3% of the variation in the data, respectively (Fig. 1b). The principal components score plot showed a trend of separation between the CPUO patients and controls (Fig. 1b). However, the supervised partial least squares-discriminant analysis (PLS-DA) achieved better separation, and components 1 and 2 captured 87.3% and 2.2% of the variation in the data, respectively (Fig. 1c). A 3D PCA scores plot and 3D PLS-DA scores plot assisted in visualizing these separations along with the first and second components (Fig. 1d–e). A K-means cluster analysis with two specified clusters also yielded distinct separation between CPUO and HC patients (Fig. 1f). Finally, metabolites that contributed most significantly to clustering were identified based on variable importance in projection (VIP) values, with VIP values of the top 15 metabolites visualized in Fig. 1g.

Figure 1
figure 1

Initial metabolic analysis of CPUO vs. HC patients. (a) Heatmap of metabolite correlations, (b) Unsupervised principal component analysis (PCA) overview and pairwise score plot of top four principal components, (c) Supervised partial least squares-discriminant analysis (PLS- DA) overview and pairwise score plot of top four components, (d) 3D unsupervised PCA scores plot, (e) 3D supervised PLS-DA scores plot, (f K- means clusters, (g) PLS-DA variable importance in projection (VIP) score.

Full size image

Biomarker identification

Biomarkers contributing to the difference between CPUO and HC were then identified using t-test analysis. 16 metabolites differed significantly between CPUO and HC (Supplemental Table 1). A heatmap was constructed of the significantly decreased metabolites, which were grouped into three different categories comprised of: nine amino acids (isoleucine, L-tyrosine, threonine, DL-tryptophan, L-valine, methionine, glycine, lysine, and L-phenylalanine), four amino acid derivatives (creatinine, DL-carnitine, acetyl-L-carnitine, and indole-3-acrylic acid), and two aromatic and fatty acid derivatives (2-hydroxycinnamic acid and oleamide) (Fig. 2a). A volcano plot and venn diagram of analyzed metabolites were also constructed to demonstrate the features that were significantly different between the two groups (Fig. 2b, c). The metabolites significantly different between CPUO and controls, particularly the amino acids and amino acid derivatives, were utilized to create a visual representation of the general amino acid pathways containing these metabolites, with box plots of significantly different features (Fig. 3).

Figure 2
figure 2

(a) Heat map of significantly different metabolites between CPUO and HC. Metabolites above the black line indicate metabolites that have a q-value (FDR-adjusted p-value) < 0.01. Metabolites below the black line indicate metabolites that have a q-value > 0.01, (b) Volcano plot of metabolites between CPUO and HC, (c) Venn diagram showing overlap between metabolites in CPUO and HC.

Full size image
Figure 3
figure 3

Amino acid pathways for metabolites significantly altered in CPUO vs. HC. Terms in red indicate a significant difference between CPUO and HC. Bar charts also indicate difference between CPUO and HC. ***q-value (FDR-adjusted p-value) < 0.001.

Full size image

Receiver operating curves

Receiver operating curve (ROC) analysis is utilized in metabolomics to assess biomarker performance in predicting CPUO compared to HC. The results of ROC curve analysis of the 16 significantly different metabolites indicate good predictability for all the significantly different metabolites, determined by the value of area under the curve (AUC) higher than 0.7 (Supplemental Table 1, Supplemental Fig. 3a–o). A multivariate ROC demonstrated good predictability of all 39 of the metabolites when together (Supplemental Fig. 3p).

Association with itch severity

The downregulated and upregulated metabolites were correlated with the WI-NRS scores of CPUO patients and controls. Supplemental Fig. 4 demonstrates these correlations. All significantly downregulated metabolites in CPUO were negatively correlated with worse itch except for glycine. GAR, which was significantly upregulated in CPUO, was positively correlated with itch severity (Supplemental Fig. 4).

Metabolic pathway analysis

A metabolic set enrichment analysis (MSEA) was performed based on known human metabolic pathways. Enrichment analysis yielded many enriched pathways, of which the top 25 (organized by p-value) are outlined in Supplemental Fig. 5a-b. The most enriched associations included tyrosine metabolism, catecholamine biosynthesis, thyroid hormone synthesis, branched-chain amino acid degradation, and carnitine synthesis. Based on the direction of change of the respective metabolites of the pathway that were significantly different between CPUO and controls, these enriched pathways were likely downregulated in CPUO patients.

A more extensive metabolic pathway analysis (MetPA) was then performed as an extension of the MSEA. Pathway analysis uses additional procedures to calculate the impact on the pathways based on network topology analysis.11 Table 2 outlines the pathway analysis results, with the most affected pathways being selected based on the most significant pathway impact value, and p-values and FDR-adjusted p-values (q-values) < 0.05. The most significantly impacted pathways were phenylalanine, tyrosine, and tryptophan biosynthesis; phenylalanine metabolism; and glycine, serine, and threonine metabolism (Supplemental Fig. 5c). The most impacted pathways were then mapped with each other in a network; each network node represented a metabolite set. Two metabolite sets were connected by an edge (line) if the number of shared metabolites between the two sets was greater than 20% (Supplemental Fig. 5d). Based on the direction of change of the respective metabolites of the pathway that were significantly different between CPUO and controls, these affected pathways were likely downregulated in CPUO.

Table 2 Top 15 metabolomics pathway analysis results, organized by impact. Significant values are in bold.
Full size table

The major affected pathways established by the MSEA and MetPA were then mapped to visualize the role that downregulated metabolites play within CPUO patients compared to controls. Mapping of the phenylalanine and tyrosine pathway, which was most significantly downregulated in MetPA analyses, supports MSEA findings of alterations in catecholamine and thyroid hormone synthesis (Supplemental Fig. 6a). Mapping of the tryptophan biosynthesis pathway corroborates biomarker t-test findings of a decreased peak intensity of indole-3-acrylic acid in CPUO, a derivative of tryptophan (Supplemental Fig. 6b). Finally, the glycine, serine, and threonine pathway were also mapped (Supplemental Fig. 6c).

Discussion

Our study identified the downregulation of nine amino acid metabolites in CPUO patients compared to controls including nine amino acids (isoleucine, L-tyrosine, threonine, DL-tryptophan, L-valine, methionine, glycine, lysine, and L-phenylalanine), four amino acid derivatives (creatinine, DL-carnitine, acetyl-L-carnitine, and indole-3-acrylic acid), and two aromatic and fatty acid derivatives (2-hydroxycinnamic acid and oleamide). These metabolites were also significantly correlated with itch severity. Pathway enrichment analysis suggested significant downregulation of several pathways in CPUO patients, including phenylalanine, tyrosine, tryptophan biosynthesis; catecholamine biosynthesis; and glycine, serine, and threonine metabolism.

Phenylalanine and tyrosine levels have been implicated in altered neurotransmitter synthesis and neuronal signaling.12 These biomarkers are mediators in neurotransmitter synthesis, as phenylalanine is converted into tyrosine, which in turn forms dopamine, epinephrine, and norepinephrine.13 While phenylalanine has been found to be upregulated in dermatoses such as psoriasis and allergic contact dermatitis, this metabolite likely plays a differential role in CPUO, through decreased norepinephrine production.14,15 Previous studies have found the depletion of intrathecal catecholamines such as norepinephrine to enhance the itch response in mice.16 Also, Gotoh et al. found that the addition of catecholamine neurotoxins such as 6-hydroxydopamine and α-adrenoceptor antagonists such as phentolamine into the intrathecal space produced a decreased concentration and function of norepinephrine respectively, and a subsequent enhancement of the itch response.16 Studies have demonstrated the ability of norepinephrine-regulating medications to reduce itch, supporting the mitigating role of norepinephrine in itch. Miyahara et al. found that in mice with chronic itch, pretreatment with milnacipran, a serotonin and norepinephrine reuptake inhibitor (SNRI), and mirtazapine, a noradrenergic and specific serotonergic antidepressant, decreased spontaneous scratching behavior.17 These findings are hypothesized to occur through activation of inhibitory interneurons that typically modulate itch behavior. Thus, downregulation of these neural-mediated pathways in CPUO patients suggest the contribution of the neuroimmune mediators of itch in CPUO pathogenesis.18.

Tryptophan, a precursor for neurotransmitters such as serotonin and tryptamine, was also downregulated in CPUO patients. Miyahara et al. found that administration of selective serotonin reuptake inhibitors (SSRIs), such as fluvoxamine and paroxetine, mitigated spontaneous scratching behavior in mice with chronic itch.17 Additionally, the neuronal function of serotonin may be affected by other mediators in these metabolite pathways. For example, low T4 levels resulting from downregulation of tyrosine in CPUO may impair serotonin neurotransmission and worsen itch, implicating both the tyrosine and tryptophan neuronal pathways in CPUO pathogenesis.19.

CPUO patients also presented with downregulated glycine, serine, and threonine metabolism. Glycine is an inhibitory neurotransmitter released by interneurons, and higher levels of glycine may inhibit itch responses.20 Foster et al. found that activating glycinergic inhibitory interneurons caused a reduction in itch behavior; the administration of strychnine, a glycine antagonist, also blocked glycine-mediated scratch inhibition in mice.20,21 In patients with psoriasis, glycine levels were negatively correlated with itch severity, suggesting that glycine affects cell proliferation and itch modulation in psoriasis.22 Sitter et al. found that when treating psoriasis patients with corticosteroids, tissue levels of glycine were increased, suggesting a role of glycine in immune downregulation. Future studies are warranted to examine neurotransmitters such as norepinephrine, serotonin, and glycine as potential targets of transmitter-modulating drugs, such as SSRIs and SNRIs, in itch treatment.

Along with neuromodulation, many of the biomarkers altered in CPUO also contribute to immunomodulation, pointing to immune dysregulation in CPUO pathogenesis. An additional described mechanism of tryptophan and serotonin in itch pathway regulation is in decreasing lymphocyte proliferation, as serotonin receptors are found in many immune cells, including mast cells, T cells, natural killer cells, and Langerhans cells.23 Therefore, the function of medications affecting these metabolites is two-fold, affecting both neurotransmission and immune regulation.

Limitations of this study include the cross-sectional nature and a small sample size with limited distribution across ethnicity, sex, comorbidities, which may be underpowered to identify all significant changes in metabolites in CPUO. Given strict selection criteria for metabolites, the study lacks a comprehensive list of metabolites detected in CPUO patients. Furthermore, confounders other than age, sex, and race may affect outcomes. Despite these limitations, this study provides a baseline for further examination of metabolomics in CPUO. Further studies using larger external cohorts will be necessary to demonstrate generalizability of our findings. Furthermore, metabolomic profiles from patients with other types of chronic pruritus as well as healthy controls with no underlying conditions from known origins should be examined in comparison to CPUO in future studies. Finally, catecholamine biomarkers should be directly examined in future metabolomic studies to verify the downregulation of catecholamine biosynthesis in CPUO identified in our study.

Our study identifies a distinct functional circulating plasma metabolic deficit in CPUO patients, including amino acids, amino acid derivatives, and aromatic and fatty acid derivatives. These findings must be considered in the context of itch pathogenesis, particularly the role of neuronal and immune modulators in pruritus. This is the first study to utilize plasma metabolomics for CPUO, which provides novel insights into pathogenesis and identification of therapeutic targets.

Methods

Sample collection

This research was a cross-sectional study conducted at Johns Hopkins University (Baltimore, MD, USA). Eleven patients with chronic pruritus of unknown origin (CPUO) were recruited from the dermatology outpatient clinics at Johns Hopkins Hospital from August 2018 to February 2020. CPUO patients were identified as patients with chronic pruritus at least 6 months before the screening visit, with no evidence of pruritus secondary to other conditions (e.g. dermatologic, neuropathic, systemic, or pharmacologic). Eleven controls, matched by age, sex, and race, were recruited. Clinical information, including demographic information and Worst Itch-Numeric Rating Scale (WI-NRS) scores, were collected from patients.

Sample preparation

Patient plasma samples were subjected to metabolite extraction by adding 100% HPLC-grade methanol (Fisher Scientific) to a final concentration of 80% (vol/vol) methanol. The samples were centrifuged. The supernatant with metabolites was separated from the protein fraction and cell debris. Speed vacuum was used to evaporate the methanol from the samples, and the samples were lyophilized and re-suspended in 50% (vol/vol) acetonitrile diluted with mass-spec-grade water.

Ultra-performance liquid chromatography-mass spectrometry (UPLC-MS) and raw data processing

The metabolite samples in 50% (vol/vol) acetonitrile were subjected to metabolomics data acquisition using a Thermo Scientific Q Exactive Plus Orbitrap Mass Spectrometer with a Vanquish UPLC at our own Metabolomics Facility at Johns Hopkins Medical Institutions. The Vanquish UPLC auto-sampler systems kept at 4 °C was used to inject 2 µl of each sample for analysis. Reverse phase chromatography with MS-grade water containing 0.1% formic acid as mobile aqueous phase (solvent A) and 98% acetonitrile containing 0.1% formic acid as the mobile organic phase (solvent B) was used. The gradient was initiated at 2% B to 100% B over 5 min at a flow rate of 0.3 mL/min. The total runtime was 13 min. A Discovery® HS F5 HPLC Column (3 μm particle size, L × I.D., 15 cm × 2.1 mm, Sigma) and a suitable guard column (Sigma) maintained at 35 °C were used. Data were analyzed using Thermo Scientific Compound Discoverer® and Thermo Scientific TraceFinder® software. The raw intensities were then normalized based on protein concentration of each sample to get the final normalized intensities. Intensity relative to active group was obtained by dividing the normalized intensity of each sample by the average normalized intensity of the HC group.

Statistical Analysis

Missing values were first replaced with 1/5 of the minimum positive value of each variable, and data were log-transformed to assume normal distribution before comprehensive downstream analysis using MetaboAnalyst 5.0 (https://www.metaboanalyst.ca).9,10 Unsupervised principal components analysis (PCA) and supervised partial least squares discriminate analysis (PLS- DA) were applied to distinguish CPUO patients from controls, and variable importance in projection (VIP) values were used to identify potentially contributory biomarkers. A two-tailed t-test identified significant changes in metabolite levels. To correct for multiple comparisons and thus to minimize false positives, false discovery rates (FDRs) were calculated based on the Benjamini–Hochberg procedure. FDR-corrected p-values (or q-values) < 0.05 were considered statistically significant. To increase the robustness of our analyses, we combined the t-test and PLS-DA VIP score in order to determine a reliable list of metabolites that significantly contributed to the difference between CPUO and HC. ROC curve analysis was performed via MetaboAnalyst 5.0 on biomarker candidates individually or combined to evaluate the predictive performance of the significantly different biomarkers. For metabolite enrichment analysis via MetaboAnalyst 5.0, the Homo sapiens SMPDB pathway library was used as a reference, and for metabolite pathway analysis, the Homo sapiens KEGG pathway library was used as a reference.

Study approval

Reviewed and approved by the Johns Hopkins School of Medicine Institutional Review Board (IRB); approval #00231694. The study was performed in accordance with the institution’s guidelines and regulations. All participants signed written informed consent forms.