Abstract

   Hair, as a non-invasive biospecimen, retains metabolic deposits from
   sebaceous glands and capillaries, reflecting substances from the
   peripheral circulation, and provides valuable biochemical information
   linked to phenotypes, yet its application in animal disease research
   remains limited. This work applied ultra-high-performance liquid
   chromatography–tandem mass spectrometry (UHPLC-MS/MS) to compare the
   hair metabolomic characteristics of healthy forest musk deer (FMD,
   Moschus berezovskii) and those diagnosed with hemorrhagic pneumonia
   (HP), phytobezoar disease (PD), and abscess disease (AD). A total of
   2119 metabolites were identified in the FMD hair samples, comprising
   1084 metabolites in positive ion mode and 1035 metabolites in negative
   ion mode. Differential compounds analysis was conducted utilizing the
   orthogonal partial least squares–discriminant analysis (OPLS-DA) model.
   In comparison to the healthy control group, the HP group displayed 85
   upregulated and 92 downregulated metabolites, the PD group presented
   124 upregulated and 106 downregulated metabolites, and the AD group
   exhibited 63 upregulated and 62 downregulated metabolites. Functional
   annotation using the Kyoto Encyclopedia of Genes and Genomes (KEGG)
   indicated that the differential metabolites exhibited significant
   enrichment in pathways associated with cancer, parasitism, energy
   metabolism, and stress. Receiver operating characteristic (ROC)
   analysis revealed that both the individual and combined panels of
   differential metabolites exhibited area under the curve (AUC) values
   exceeding 0.7, demonstrating good sample discrimination capability.
   This research indicates that hair metabolomics can yield diverse
   biochemical insights and facilitate the development of non-invasive
   early diagnostic techniques for diseases in captive FMD.

1. Introduction

   The forest musk deer (FMD, Moschus berezovskii) is a small,
   forest-dwelling ungulate endemic to East Asia. The musk produced by
   males is a crucial raw material for traditional Asian medicine and
   global fragrance manufacturing [[34]1]. Due to illicit poaching and
   habitat fragmentation, the wild FMD population has drastically
   decreased since the 1950s. In order to restore the FMD population, the
   Chinese government designated the FMD as a Class I national protected
   species in 2002 [[35]2]. The IUCN Red List categorized the FMD as
   “endangered” in 2008 [[36]3]. China initiated the artificial breeding
   of FMD in 1958 and is the only country globally to have successfully
   domesticated this species. The FMD is a species characterized by high
   sensitivity [[37]4]. Factors such as the type of enclosure, farming
   methods, and feeding operations, etc., can trigger high-stress
   conditions [[38]5,[39]6,[40]7]. This may lead to a decline in the
   immune system of FMD, resulting in an increased incidence of diseases
   such as hemorrhagic pneumonia (HP), phytobezoar disease (PD), and
   abscess disease (AD) [[41]8,[42]9,[43]10], which pose significant
   challenges to the growth of FMD breeding populations.

   HP is an acute infectious disease that frequently causes mass mortality
   in FMDs, with infected individuals typically experiencing sudden death
   without obvious symptoms while exhibiting blood-tinged secretions from
   the nasal and oral cavities [[44]11]. Pathogens isolated from pulmonary
   lesions of FMD include Pseudomonas aeruginosa [[45]8], Klebsiella
   pneumoniae [[46]12], and Escherichia coli [[47]13]. PD originates from
   indigestion due to densely aggregated plant-fiber phytobezoars, leading
   to anorexia and emaciation in FMD, ultimately resulting in nutritional
   exhaustion and fatality due to gastrointestinal obstruction by
   phytobezoars [[48]10]. AD—the most prevalent disease in FMD—exhibits up
   to 50% mortality, presenting suppurative lesions on the head, limbs,
   oral cavity, or digestive tract [[49]14]. The causative agents isolated
   from AD lesions include Trueperella pyogenes [[50]14] and Bacillus
   cereus [[51]15]. Primary pathogens such as P. aeruginosa and T.
   pyogenes have established virulence factor profiles and antibiotic
   susceptibility data [[52]14,[53]16,[54]17,[55]18]. However, development
   of prevention options for these diseases in captive FMD is hindered by
   restrictive regulations on invasive experiments due to CITES and
   Chinese protective laws. Moreover, the shy and evasive nature of FMD
   hampers the detection of diseases at an early stage. The intense stress
   response of FMD complicates the operation of health checkups, causing
   delays in disease diagnosis as well [[56]19]. Hence, there is an urgent
   need to establish a non-invasive methodology to monitor the long-term
   biochemical abnormalities associated with disease in FMD.

   The temporal scales of the physiological alterations detectable in
   biological specimens exhibit considerable variability. Blood or urine
   demonstrates changes occurring within hours, feces indicates
   alterations over days, and hair offers a record of metabolic products
   deposited over several months from sebaceous glands and capillaries in
   hair follicles [[57]20]. In addition to the prolonged biomonitoring
   window, mammalian hair samples offer significant benefits regarding
   accessibility and effectiveness. Hair can be obtained through routine
   care activities, such as plucking during vaccination or collecting
   naturally shed hair [[58]21]. This practice mitigates the stress
   responses in nervous animal species [[59]22], lowers the infection
   risks, and enables long-term, repetitive sampling. Hair is resilient to
   degradation due to its keratinized structure, improving its transit,
   storage, and processing efficiency [[60]23]. Therefore, mammalian hair
   has gained increasing recognition as a practical analytical specimen
   for current isotope, proteomic, and metabolomic research
   [[61]24,[62]25,[63]26,[64]27,[65]28].

   The metabolome, formed by interactions among the genome, transcriptome,
   enzyme proteome, and environmental factors, yields sensitive,
   low-molecular-weight biomarkers (<1500 Da) most relevant to the
   phenotype [[66]29]. Therefore, the metabolomic characterization of
   biospecimens has emerged as an important method for investigating the
   pathophysiological changes induced by diseases [[67]30,[68]31].
   Previous metabolomic research has identified novel disease biomarkers
   in human and animal hair samples, demonstrating significant diagnostic
   potential. For instance, in the hair of pregnant women with
   intrahepatic cholestasis, a remarkable impairment of the glutathione
   metabolic pathway was noted [[69]32]. In rats genetically engineered
   for Alzheimer’s disease with amyloid-β-induced cognitive deficits, the
   biosynthesis of phenylalanine and tryptophan was found to be
   significantly upregulated in their hair [[70]33]. Despite the growing
   interest in hair metabolomics, this topic remains underexplored within
   the context of animal disease research.

   In light of this, we employed ultra-high-performance liquid
   chromatography–tandem mass spectrometry (UHPLC-MS/MS) to explore the
   hair metabolomic features of diseased FMD in cases of hemorrhagic
   pneumonia (HP), phytobezoar disease (PD), and abscess disease (AD). We
   further identified disease biomarkers through a comparative analysis
   between diseased FMD and healthy controls. This study offers a
   scientific foundation and technical methodology for non-invasive
   disease diagnosis via hair metabolomics in captive FMD.

2. Materials and Methods

2.1. Hair Collection

   [71]Figure 1 illustrates the comprehensive experimental framework
   followed in the present study. The collection of FMD hair samples was
   performed from early February to mid-March 2024 in Feng County, Baoji
   City, Shaanxi Province, China (E 106°18′ to 108°03′, N 33°35′ to
   35°06′). This region holds the largest captive FMD population in China.
   The subject animals were sourced from FMD farms in the towns of Hekou,
   Pingmu, and Liufengguan within Feng County.

Figure 1.

   [72]Figure 1
   [73]Open in a new tab

   Experimental framework for the untargeted hair metabolomics of captive
   forest musk deer (FMD) in the present study. The hair samples
   (hemorrhagic pneumonia, (n = 6); phytobezoar disease, (n = 6); abscess
   disease, (n = 6); healthy controls, (n = 6)) were plucked, washed,
   dried, weighed, and milled, followed by being subjected to ultrasonic
   extraction at low temperature using methanol, and subsequently
   underwent UHPLC-MS/MS analysis. The raw mass spectrometry data were
   normalized via Progenesis QI software (v3.0). The markedly modified
   hair metabolites were screened out by a metabolomic workflow to
   identify potential biomarkers of FMD diseases.

   Hair samples from healthy controls were obtained according to
   verification from farm personnel that no health issues had been
   documented in the preceding year. To eliminate disparities between
   healthy and diseased FMD, samples of healthy controls were obtained on
   the same day as the diseased FMD samples, from a healthy FMD of
   comparable age and gender. During the musk collection or vaccination
   process, when the animals were immobilized, hair samples of healthy
   individuals were plucked (n = 6). As shown in [74]Figure 2, the FMD’s
   shallow, spindle-shaped hair root makes it prone to shedding. The hair
   sampling caused no harm or suffering to the FMD. Hair samples of
   diseased FMD (HP, n = 6; PD, n = 6; AD, n = 6) ([75]Table 1) were
   obtained from individuals diagnosed by a veterinarian as either
   surviving or deceased due to the disease (a sample collection interval
   of ≤ 1 day post-mortem; the impact is negligible because hair growth
   and metabolism cease after death [[76]34]). [77]Figure 3 illustrates
   the necropsy findings of the lesions in the sampled diseased FMD. Hair
   samples from the dorsolateral thigh of FMD were collected (guard hair,
   ~200 mg, length ≥ 5 cm). Given that an FMD undergoes an annual molt
   from late May to November, and that poor physiological conditions may
   impair this process, resulting in partial molting for some individuals
   [[78]1], we ensured that only recently grown hair, reflecting
   metabolite deposition over the preceding year, was used for the
   experiment. We removed the aged hair based on the noticeable color
   difference—aged hair appears faded and yellowed from extended sun
   exposure, whereas new hair is darker and shinier—and collected only the
   newly grown hair for the analysis. The hair samples were stored in
   sterile ziplock bags and promptly transported to the laboratory for
   subsequent analysis. The sampled FMD were under the management of
   Beijing Tong Ren Tang Shaanxi Musk Industry Co., Ltd. (Baoji, China),
   which approved the sampling procedures in the present study.

Figure 2.

   [79]Figure 2
   [80]Open in a new tab

   Hair samples from forest musk deer featuring fine, short,
   spindle-shaped hair roots: (A) captured by camera, with red boxes
   indicating hair follicles; and (B) captured by scanning electron
   microscope.

Table 1.

   Individual information and descriptions of the forest musk deer sampled
   in this study.
   ID Age (Years) Gender Description of Conditions
   HP1 0.7 ♀ Hemorrhagic pneumonia
   HP2 3 ♂ Hemorrhagic pneumonia
   HP3 0.7 ♂ Hemorrhagic pneumonia, infant, infected from mother in the
   same enclosure
   HP4 2 ♀ Hemorrhagic pneumonia
   HP5 3 ♂ Hemorrhagic pneumonia
   HP6 2 ♀ Hemorrhagic pneumonia, pregnant female
   PD1 2 ♀ Anorexia, rumen indigestion, presence of phytobezoars
   PD2 4 ♀ Indigestion, chronic fecal impaction with undigested fibers,
   phytobezoars
   PD3 2 ♀ Death due to gastrointestinal blockage induced by phytobezoars
   PD4 6 ♂ Anorexia, vomiting, gastrointestinal obstruction caused by
   phytobezoars
   PD5 2 ♂ Gastrointestinal obstruction caused by phytobezoars
   PD6 4 ♂ Phytobezoar-induced gastrointestinal disease, anorexia, sudden
   death during treatment
   AD1 1.5 ♀ Abscess disease, continuous weight loss
   AD2 0.7 ♂ Post-mortem examination revealed abscesses in the lungs,
   sudden death in infant
   AD3 1.8 ♀ Greenish pus in the leg, continuous weight loss
   AD4 7 ♂ Post-mortem examination revealed abscesses in the thigh muscles
   and lungs
   AD5 4 ♂ Visible abscess in the leg
   AD6 7 ♂ Visible abscess in the leg
   C1 0.7 ♀ Healthy, no record of past illness
   C2 2 ♀ Healthy, no record of past illness
   C3 4 ♂ Healthy, no record of past illness
   C4 3 ♂ Healthy, no record of past illness
   C5 2 ♂ Healthy, no record of past illness
   C6 6 ♂ Healthy, no record of past illness
   [81]Open in a new tab

Figure 3.

   [82]Figure 3
   [83]Open in a new tab

   Anatomical examination of lesions in the sampled diseased forest musk
   deer (FMD). (A,B) Red arrows indicate pulmonary hemorrhages and pleural
   effusion in FMD diagnosed with hemorrhagic pneumonia; (C,D) yellow
   arrows indicate phytobezoars removed by surgery from the digestive
   tracts of FMD with phytobezoar disease; and (E,F) blue arrows indicate
   oral and visceral abscesses found in FMD with abscess disease.

2.2. Metabolome Extraction

   The hair samples underwent washing with isopropanol (analytical grade ≥
   99.5%, Macklin, Shanghai, China) followed by ultrapure water (PURELAB
   Prima 30, ELGA, High Wycombe, UK), each for 3 min in an ultrasonic
   bath, to remove surface contaminants [[84]35]. The samples were then
   placed in a UV-sterilized fume hood for 24 h to ensure complete drying.
   Approximately 50 mg of dried hair was sectioned into 1 cm snippets
   using sterilized scissors. Each hair sample was mixed with 400 µL of
   extracting solvent (methanol: ultrapure water = 4:1, v:v). To monitor
   the accuracy and reliability of the sample analysis,
   L-2-chlorophenylalanine (0.02 mg/mL) was used as the internal standard.
   After grinding for 6 min at −10 °C under a frequency of 50 Hz,
   ultrasound extraction was performed at 5 °C under 40 kHz for 30 min.
   The samples were placed in incubators at −20 °C for 30 min, followed by
   centrifugation of 13,000× g at 4 °C for 15 min, after which the
   supernatants were collected as analytical solutions.

2.3. UHPLC-MS/MS Profiling

   A Vanquish Horizon UHPLC system (Thermo Fisher, Waltham, MA, USA)
   equipped with an ACQUITY HSS T3 column (100 mm × 2.1 mm, 1.8 µm, Waters
   Corporation, Milford, MA, USA) was utilized for the chromatographic
   separation. The quality control (QC) samples were prepared by blending
   aliquots of all the FMD hair extractants and following the same
   experimental procedures as for the test samples. The QC samples were
   introduced at regular intervals (1 insertion per 5 samples) to assess
   the stability of the instrument. The column temperature was 40 °C, and
   the injection volume was 3 μL. The mobile phase consisted of Phase A:
   acetonitrile and water with 0.1% formic acid (5:95, v:v), and Phase B:
   acetonitrile, isopropanol, and 5% water with 0.1% formic acid
   (47.5:47.5:5, v:v:v). The separation gradient (positive ion mode) was
   as follows: mobile phase B-0%~20% (0~3 min); B-20%~35% (3~4.5 min);
   B-35%~100% (4.5~5 min); B-100% (5~6.3 min); B-100%~0% (6.3~6.4 min);
   B-0% (6.4~8 min). The separation gradient (negative ion mode) was as
   follows: mobile phase B-0%~5% (0~1.5 min); B-5%~10% (1.5~2 min);
   B-10%~30% (2~4.5 min); B-30%~100% (4.5~5 min); B-100% (5~6.3 min);
   B-100%~0% (6.4~8 min). The flow rate was 0.4 mL/min. Tandem mass
   spectrometry was conducted using a Q Exactive HF-X high-resolution mass
   spectrometer (Thermo Fisher). Electrospray ionization (ESI) was
   employed to ionize the metabolites eluted from the UHPLC system,
   functioning at a voltage of ±3500 V in both positive (ESI+) and
   negative (ESI−) ion modes. The settings for the tandem mass
   spectrometer were as follows: a base temperature of 425 °C, a sheath
   gas circulation speed of 50 arb, and an auxiliary gas circulation speed
   of 13 arb. The data collection relied on data-dependent acquisition
   (DDA) mode. A mass range of 70~1050 m/z was scanned at a resolution of
   60,000 in the full MS, whereas the MS/MS resolution was established at
   7500. The MS/MS mode operated with stepwise collision energy at 20, 40,
   and 60 au, respectively.

2.4. Hair Metabolomic Data Mining

   After completing the MS analysis of the sample, the raw metabolic data
   were imported into the Progenesis QI metabolomic processing software
   (v3.0, Waters Corporation, Milford, MA, USA), which generated a
   three-dimensional data matrix in CSV format, containing the sample IDs,
   metabolite tags, and mass spectral intensities. The internal standard
   peaks and identified invalid positives (e.g., noise, column bleed,
   derivatized reagent peaks) were eliminated. Following data filtration,
   the cleaned metabolic data were cross-referenced against the Majorbio
   in-house library for standard reference material (Majorbio Bio-Pharm
   Technology Co., Ltd., Shanghai, China) and the HMDB
   ([85]http://www.hmdb.ca/, accessed on 11 November 2024), LipidBlast
   ([86]https://fiehnlab.ucdavis.edu/projects/LipidBlast, accessed on 11
   November 2024) and Metlin ([87]http://metlin.scripps.edu, accessed on
   11 November 2024) databases. The evaluation criteria for the database
   search include the mass deviation of primary fragments, isotope
   similarity, and secondary fragment similarity. The primary mass
   deviation threshold was set at 10 ppm, and the isotope similarity was
   set to exceed 30 [[88]36]. For valid results, the fragmentation score
   (from the in-house standard libraries and spectrometry databases like
   Metlin and LipidBlast) must exceed 35, or the theoretical fragmentation
   score (from theoretical databases such as HMDB) must exceed 40
   [[89]37,[90]38]. The highest-scoring match is selected as the
   metabolite identification result. To ensure cross-sample consistency,
   the intensity of all the detected metabolites was normalized by the
   total sum value (each value/(sum of each column/coefficient obtained
   from the largest sum across columns). The coefficient = the total
   signal intensity of all the peaks in each sample/the maximum sum of
   signal intensities across all the samples) [[91]36]. Metabolites with a
   relative standard deviation (RSD) exceeding 30% of the QC samples were
   excluded. A Log10 transformation was applied to the data matrix.
   Metabolite features identified in at least 80% of samples within either
   group were retained. Null values were imputed using the minimum value
   of the corresponding metabolite.

   To detect the hair metabolic alterations in three FMD diseases,
   principal component analysis (PCA) and orthogonal partial least
   squares–discriminant analysis (OPLS-DA) were performed using the R
   package ropls (v 1.6.2). The robustness of the OPLS-DA model was
   evaluated through 7-fold cross-validation. The volcano plot was created
   using the R package ggplot2 (v 3.3.0). Significantly differential
   metabolites were identified based on three criteria, including a
   variable importance projection score (VIP) > 1, log[2] fold change > 1,
   and p-value (unpaired Student’s t-test) < 0.01. Pathway enrichment
   analysis of the differential metabolites was conducted using the
   scipy.stats Python package (v1.0.0). Fisher’s exact test was employed
   to identify pathways with remarkable enrichment (p < 0.05). The
   enriched pathways were annotated using the Kyoto Encyclopedia of Genes
   and Genomes Database (KEGG, [92]https://www.kegg.jp/, accessed on 11
   November 2024). To evaluate the discriminative capability of the
   candidate biomarkers, the receiver operator characteristic (ROC) curves
   were drawn using the R package pROC (version 1.12.1). The sample
   classification performance of the metabolites was assessed by the area
   under the curve value (AUC) with a 95% confidence interval (CI).

3. Results

3.1. Untargeted Metabolomic Characterization of Hair Samples from Captive FMD

   After data preprocessing, a total of 5621 valid peaks were detected in
   positive ion mode (ESI+) and 8090 peaks in negative ion mode (ESI−).
   The database alignment identified the chemical structures of 1084 and
   1035 metabolites for ESI+ mode and ESI− mode, respectively. As depicted
   in the total ion chromatograms (TICs) in [93]Figure 4A,B, significant
   differences in the response intensities of various metabolites in the
   hair samples from the three FMD disease groups (HP1–6, PD1–6, AD1–6)
   were observed compared to the healthy control samples (C1–C6) in both
   ESI+ and ESI− modes. The repeatability and stability of the analytical
   measurements were assessed using quality control (QC) samples. As
   demonstrated by the TICs of the QC samples in [94]Figure 4C,D, all the
   QC samples exhibited substantial overlap in terms of the retention time
   and peak intensity for ion signals. Therefore, the stability and
   repeatability of the instrumental conditions in this experiment were
   satisfactory, ensuring that the data quality met the requirements for
   subsequent analyses.

Figure 4.

   [95]Figure 4
   [96]Open in a new tab

   Representative total ion chromatograms (TICs) from collected samples.
   (A,B) TICs of experimental samples in positive (A) and negative (B) ion
   modes; (C,D) TICs of quality control samples in positive (C) and
   negative (D) ion modes. POS, positive ion mode; NEG, negative ion
   modes. HP, hemorrhagic pneumonia (n = 6); PD, phytobezoar disease (n =
   6); AD, abscess disease (n = 6); C, healthy controls (n = 6); and QC,
   quality control samples (n = 5).

   A correlation heatmap analysis was performed to investigate the
   relationships across all the samples. As shown in [97]Figure 5A, the
   strong correlation among the QC samples indicated good experimental
   reproducibility. Compared to the healthy control samples, the samples
   from each disease group exhibited reduced correlation and a weaker
   clustering pattern. We additionally conducted unsupervised principal
   component analysis (PCA) to examine the sample distribution. As
   illustrated in [98]Figure 5B, the QC samples were closely gathered in
   the center of the PCA plot. The samples from the healthy control group
   displayed clear separation from those in the disease groups. The first
   and second principal components of the PCA plot accounted for 43.7% of
   the observed variances, highlighting significant inter-group sample
   dissimilarity.

Figure 5.

   [99]Figure 5
   [100]Open in a new tab

   Classification of hair samples from captive forest musk deer based on
   the metabolomic compositions. (A) Correlation heatmap analysis. HP,
   hemorrhagic pneumonia; PD, phytobezoar disease; AD, abscess disease; C,
   healthy control; and QC, quality control. (B) Principal component plot.
   (C) Venn diagram. (D) Metabolite classification derived from the HMDB
   superclass level.

   As shown in [101]Figure 5C, the Venn diagram indicated that 1768
   metabolites were consistently present in all the samples. Compared to
   the healthy controls, the numbers of unique metabolites identified were
   119 for HP, 115 for PD, and 170 for AD. The classification of the
   compounds in the HMDB database is illustrated in [102]Figure 5D. The
   percentages of the FMD hair metabolites at different HMDB superclass
   levels (>1%) in descending order are as follows: lipids and lipid-like
   molecules (28.77%), organic acids and derivatives (20.58%),
   organoheterocyclic compounds (14.21%), benzenoids (9.82%), organic
   oxygen compounds (9.36%), phenylpropanoids and polyketides (4.62%),
   nucleosides, nucleotides, and analogues (2.98%), organic nitrogen
   compounds (1.99%), and alkaloids and derivatives (1.23%).

3.2. Establishing the OPLS-DA Model

   To further elucidate the inter-group differences and attributable
   metabolites, we conducted supervised orthogonal partial least
   squares–discriminant analysis (OPLS-DA) comparing each disease group to
   the control group. The OPLS-DA score plots exhibited significant sample
   differentiation in the comparative analysis of HP vs. control
   ([103]Figure 6A), PD vs. control ([104]Figure 6C), and AD vs. control
   ([105]Figure 6E) in the mixed ion mode (ESI+ and ESI−). The cumulative
   R^2X values for principal components 1 and 2 in the OPLS-DA models of
   the three diseases were as follows: HP vs. control: R^2X = 0.53 (>0.5);
   PD vs. control: R^2X = 0.38; AD vs. control: R^2X = 0.46, indicating
   the principal components of the OPLS-DA model constructed for the HP
   group most remarkably explained the sample variances. To assess the
   explanatory capacity of the constructed OPLS-DA models, we calculated
   the models’ R^2Y value, Q^2 value, and Q^2 intercept after conducting
   200 time permutation tests ([106]Figure 6B,D,F). The findings are as
   follows: HP vs. control: R^2Y = 0.8245, Q^2 = 0.635, and Q^2 intercept
   = −0.2835; PD vs. control: R^2Y = 0.8902, Q^2 = 0.618, and Q^2
   intercept = −0.1279; AD vs. control: R^2Y = 0.7988, Q^2 = 0.508, and
   Q^2 intercept = −0.307. The results indicate that all the OPLS-DA
   models exhibit strong discriminative capacity (R^2Y > 0.5) and
   predictive capabilities (Q^2 > 0.5), with no indications of overfitting
   (Q^2 intercepts < 0.05).

Figure 6.

   [107]Figure 6
   [108]Open in a new tab

   OPLS-DA score plot and permutation test for hair metabolome of healthy
   and diseased forest musk deer. (A,B) Hemorrhagic pneumonia vs. healthy
   control; (C,D) phytobezoar disease vs. healthy control; (E,F) abscess
   disease vs. healthy control. C, healthy controls; HP, hemorrhagic
   pneumonia; PD, phytobezoar disease; AD, abscess disease.

3.3. Screening Differential Hair Metabolites Between Healthy and Diseased
Forest Musk Deer

   As illustrated in [109]Figure 7, the volcano plot reveals notable
   changes in the hair metabolomics of diseased FMD compared to healthy
   individuals (VIP value > 1, log[2]FC > 1, p < 0.01). Compared to the
   healthy control group, the HP group showed 85 metabolites with
   increased levels and 92 with decreased levels. The PD group had 124
   upregulated and 106 downregulated differential metabolites, while the
   AD group had 63 upregulated and 62 downregulated metabolites.

Figure 7.

   [110]Figure 7
   [111]Open in a new tab

   Volcano plots of markedly differential metabolites. (A) Hemorrhagic
   pneumonia vs. healthy control; (B) phytobezoar disease vs. healthy
   control; and (C) abscess disease vs. healthy control. Metabolites were
   selected based on VIP value > 1, log[2] fold change > 1, and p < 0.01.
   Red bubbles represent upregulation, blue bubbles represent
   downregulation, and gray bubbles represent no significant difference.

   [112]Figure 8 further depicts the metabolites of interest (top10 VIP
   value, top 10 p value, and biologically significant metabolites). As
   shown in [113]Figure 8A, in comparison to the healthy controls, the
   relative concentrations of lysophosphatidylcholine (Lpc 18:1) (VIP =
   1.83, p = 0.002) and lysophosphatidylethanolamine (Lpe 18:2) (VIP =
   2.08, p = 0.006) were markedly upregulated in the HP samples, while
   metabolites such as N-(3-Oxohexanoyl) homoserine lactone (VIP = 2.82, p
   = 0.02) exhibited notable downregulation. [114]Figure 8B displays the
   distinct metabolites identified in the comparison of the PD vs. healthy
   groups. The PD samples showed significant elevations in metabolites
   such as 2alpha-hydroxyalantolactone (VIP = 3.14, p = 0.001), along with
   significantly reduced levels of succinic anhydride (VIP = 1.38, p =
   0.01), L-malic acid (VIP = 1.09, p = 0.003), etc. [115]Figure 8C
   illustrates the differential metabolites detected in the examination of
   the AD vs. healthy groups, highlighting a remarkable increase in
   metabolites such as 4-hydroxybenzoic acid (VIP = 2.21, p = 0.009) and a
   notable decrease in triisopropanolamine (VIP = 2.78, p = 0.001), etc.

Figure 8.

   [116]Figure 8
   [117]Open in a new tab

   Screening differential hair metabolites between diseased and healthy
   forest musk deer. (A) Hemorrhagic pneumonia vs. healthy controls. (B)
   Phytobezoar disease vs. healthy controls. (C) Abscess disease vs.
   healthy controls. *** p < 0.001, ** p < 0.01, * p < 0.05.

3.4. Functional Analysis of Altered Hair Metabolites

   To examine the biological functions linked to the altered metabolites,
   KEGG pathway topological and enrichment analyses were conducted.
   [118]Figure 9A,B illustrate the results of the analysis comparing the
   HP samples to the healthy controls. It can be inferred that a total of
   35 KEGG level II metabolic pathways were enriched, including “Pathways
   in cancer” (p = 0.006) and “Amoebiasis” (p = 0.01). [119]Figure 9C,D
   present the results from the functional analyses of the PD group. A
   total of 34 KEGG level II metabolic pathways were identified, which
   include “Histidine metabolism” (impact value = 0.2), “Lysine
   degradation” (impact value = 0.04), “Citrate cycle” (impact value =
   0.09), and “Pentose phosphate pathway” (impact value = 0.06).
   [120]Figure 9E,F present the results of the enrichment and topological
   analysis for the AD samples. A total of 35 KEGG level II metabolic
   pathways were recognized, including “Retrograde endocannabinoid
   signaling” (p = 0.002) and “Fc gamma R-mediated phagocytosis” (p =
   0.004).

Figure 9.

   [121]Figure 9
   [122]Open in a new tab

   KEGG pathway topological and enrichment analyses of altered
   metabolites. (A,B) Hemorrhagic pneumonia vs. healthy control. (C,D)
   Phytobezoar disease vs. healthy control. (E,F) Abscess disease vs.
   healthy control. KEGG pathway level I abbreviations: OS, Organismal
   Systems; EIP, Environmental Information Processing; HD, Human Diseases;
   CP, Cellular Processes; M, Metabolism. *** p < 0.001, ** p < 0.01, * p
   < 0.05.

3.5. ROC Analysis of Biomarker Candidates

   To assess the classification power of the differential hair
   metabolites, receiver operating characteristic (ROC) curves were
   generated. The ROC analysis results of the HP group are illustrated in
   [123]Figure 10A. The area under the curve (AUC) values for the
   individual and composite metabolite panels are as follows: top 10 VIP
   value metabolite set > top 10 p value metabolite set > top 1 VIP value
   metabolite, with an elevated AUC value signifying the biomarker’s
   enhanced predictive ability for sample discrimination accuracy. The
   ranking of the AUC’s 95% confidence interval (CI) width is as follows:
   top 1 VIP value metabolite > top 10 VIP value metabolite set > top 10 p
   value metabolite set. The 95% CI evaluates the model’s stability, with
   a narrower CI width indicating greater model reliability. The ROC
   analysis results for the PD group are shown in [124]Figure 10B. The
   ranking of the AUC values for distinct metabolite sets is as follows:
   top 10 p value metabolite set = top 1 VIP value metabolite > top 10 VIP
   value metabolite set. The ranking of the 95% CI width of the AUC is as
   follows: top 10 VIP value metabolite set > top 10 p value metabolite
   set = top 1 VIP value metabolite. Likewise, the ROC analysis of the AD
   group shown in [125]Figure 10C revealed the following pattern of AUC
   value ranking: top 10 p value metabolite set > top 10 VIP value
   metabolite set > top 1 VIP value metabolite, and the 95% CI width of
   the AUC was ordered as follows: top 1 VIP value metabolite > top 10 VIP
   value metabolite set > top 10 p value metabolite set. In summary, while
   the multivariate ROC curves of the combined metabolite sets improved
   the AUC values and discrimination compared to the univariate curves of
   the single metabolites in the comparisons of HP vs. healthy controls
   and AD vs. healthy controls, this advantage was not observed in the PD
   group, where both univariate and multivariate panels performed
   similarly. The ROC analyses showed that all the AUC values for the
   tested metabolite panels exceeded 0.7, and the lowest lower CI limit
   was 0.7081, suggesting that both the single and composite metabolite
   panels based on the VIP or p values could distinguish hair samples
   between healthy and diseased groups, demonstrating diagnostic
   potential.

Figure 10.

   [126]Figure 10
   [127]Open in a new tab

   Receiver operating characteristic (ROC) curves for differential hair
   metabolites between healthy and diseased forest musk deer. (A)
   Hemorrhagic pneumonia vs. healthy control. (B) Phytobezoar disease vs.
   healthy control. (C) Abscess disease vs. healthy control. The
   specificity and sensitivity corresponding to the optimal cutoff point
   of the ROC curve are labeled on the curves. The area under the curve
   value (AUC) and 95% confidence interval (CI) are shown beneath all the
   curves. “Combine” refers to the set of all the tested metabolites.

4. Discussion

   This study employed the UHPLC-MS/MS method to conduct a comparative
   metabolomic analysis of hair samples from healthy and diseased captive
   FMD. The results revealed that the significantly altered metabolites
   and functional pathways in the hair of diseased individuals were
   potentially associated with pathological mechanisms. These findings
   underscore the research and clinical values of non-invasive hair
   sampling for long-term biomonitoring in captive-bred protected animal
   populations.

   The metabolomic analysis of the HP hair samples revealed significant
   alterations in the levels of N-(3-oxohexanoyl) homoserine lactone, a
   signaling molecule known as N-acyl homoserine lactone (AHL), which has
   been extensively studied and plays a vital role in the bacterial quorum
   sensing (QS) system [[128]39,[129]40]. AHL is a self-inducing molecule
   that triggers QS in Gram-negative bacteria [[130]41]. Previous research
   showed that the intercellular communication of pathogens through QS
   systems is essential for the development of infectious diseases
   [[131]42]. The QS system of Pseudomonas aeruginosa enhances its
   pathogenicity by modulating the expression of virulence factors, such
   as exoenzymes and toxins that may induce inflammation [[132]43].
   Meanwhile, multi-antibiotic-resistant P. aeruginosa has frequently been
   isolated from the lung tissues of FMD individuals affected with HP
   [[133]8,[134]17,[135]44]. The findings of the current investigation
   indicated that the AHL levels in the hair of the HP group were markedly
   lower than those in the healthy control group, underscoring that in
   vivo modifications to the AHL levels may be intricately associated with
   the course of HP. The downregulation of AHL in the hair of HP
   individuals may be caused by downstream effects or feedback regulation.
   Experimental validation utilizing model animals will help to clarify
   this phenomenon. Among other metabolites, lysophosphatidylethanolamine
   LPE (18:2) and lysophosphatidylcholine LPC (18:1) emerge as compounds
   of particular interest [[136]45,[137]46]. Ma et al. reported that the
   serum LPC concentrations and LPE-related ratios effectively predict the
   disease severity and 30-day mortality in community-acquired pneumonia
   (CAP) patients [[138]45]. Furthermore, Li et al. demonstrated that
   incorporating LPE acyltransferase (LPEAT) measurements into clinical
   scoring systems significantly improves the prognostic accuracy
   [[139]47]. Building on our observed alterations in the LPC and LPE
   levels in the HP samples, these lysophospholipids may also be closely
   linked to pneumonia progression in FMD. Additional studies are needed
   to confirm this association. The functional analysis demonstrated that
   the pathways of “Amoebiasis” and “Pathways in cancer” were considerably
   enriched in the HP samples relative to the healthy controls.
   Furthermore, this discovery aligns with a previous proteomics study on
   lung tissue from three FMD who died of pneumonia [[140]48], which also
   demonstrated enrichment in the “Amoebiasis” and “Pathways in cancer”
   pathways. The authors of the proteomic analysis speculated that
   analogous or shared immune response pathways may be present between FMD
   pneumonia and amoebiasis or cancer. The findings from the current hair
   metabolome analysis of FMD with hemorrhagic pneumonia have validated
   their proteomic results. The enriched metabolites in these pathways are
   closely involved in FMD’s immune regulation. For instance, in the
   “Amoebiasis” pathway, phosphatidylserine (PS) translocation across the
   plasma membrane serves as an early apoptotic marker, which can be
   recognized by extracellular receptors to initiate phagocytosis
   [[141]49] ([142]Figure 11A). The diacylglycerol (DAG) and phosphatidic
   acid (PA) within this pathway can regulate protein kinase C (PKC)
   activity, thereby influencing cellular proliferation and survival
   [[143]50] ([144]Figure 11A). In “Pathways in cancer”, the accumulation
   of 4-hydroxynonenal (4HNE) reflects intracellular oxidative stress,
   which may disrupt PKC signaling by altering the DAG/PA levels [[145]51]
   ([146]Figure 11B). Alterations in these metabolites can constitute
   pathogenic or immunological mediators of HP, suggesting their potential
   utility as diagnostic indicators or therapeutic targets.

Figure 11.

   [147]Figure 11

   [148]Figure 11

   [149]Figure 11
   [150]Open in a new tab

   Regulatory networks of metabolic pathways enriched in hair samples from
   forest musk deer (FMD) with three diseases. For hemorrhagic pneumonia
   (HP), the (A) pathways “Amoebiasis” and (B) “Pathways in cancer” are
   displayed; for phytobezoar disease (PD), the (C) “Citrate cycle” (TCA
   cycle) and the (D) “Pentose phosphate pathway” are shown; for abscess
   disease (AD), the pathway (E) “Retrograde endocannabinoid signaling” is
   presented. Red dots represent considerably increased metabolites in the
   hair of sick FMD, green dots denote strongly downregulated metabolites,
   and blue dots show spots where both upregulated and downregulated
   metabolites coexist.

   Metabolic analysis of the PD hair samples demonstrated significant
   downregulation of features related to energy metabolism, including
   succinic anhydride (a dehydrated form of succinic acid, and succinic
   acid acts as a substance in the mitochondrial electron transport chain
   [[151]52]) and L-malic acid (a vital TCA cycle intermediate [[152]53]).
   Functional analysis revealed that the pathways within KEGG level I,
   “Metabolism”, exhibited a substantially higher enrichment in the PD
   samples (35.4%) than in the AD samples (31.1%) and the HP samples
   (27.3%). The specific pathways enriched in the PD samples within the
   “Metabolism” category are closely linked to carbohydrate metabolism,
   such as the “Citrate cycle” (TCA cycle) and the “Pentose phosphate
   pathway”. It is widely agreed that the TCA cycle is the fundamental
   pathway of cellular oxidative phosphorylation, essential for energy
   metabolism, biosynthesis, and redox homeostasis in cells and organs
   [[153]54]. In the PD samples, the specific metabolites decreased within
   the TCA cycle include isocitrate and malate ([154]Figure 11C). The
   isocitrate drives cycle progression via oxidative decarboxylation to
   α-ketoglutarate, while malate regenerates oxaloacetate to restart the
   cycle [[155]55]. The pentose phosphate pathway functions as a
   derivative of glycolysis in the initial stage of carbohydrate
   metabolism, is essential for ribonucleotide synthesis, and serves as
   the main provider of NADPH [[156]56]. In addition, the downregulated
   metabolites in PD, including glycerate and 2-deoxy-D-ribose-5-phosphate
   derivatives (G3P/F6P), provide rapid energy substrates by feeding into
   glycolytic intermediates during periods of heightened energy demands
   [[157]57,[158]58] ([159]Figure 11D). Previous studies have shown that
   in the case of intestinal inflammation, a significant increase in
   energy consumption occurs due to the high energy demand for promoting
   cell division and repairing damaged mucosa, which leads to lower levels
   of TCA-cycle-related molecules [[160]59,[161]60]. Therefore, the
   declines of the metabolites within the energy metabolism pathways in
   the PD samples may also be a compensatory mechanism in FMD to meet the
   energy needs under conditions of digestive tract obstruction, which
   requires further experimental validation.

   By examining the differential metabolites set of the AD group, we
   identified 4-hydroxybenzoic acid as a key compound. 4-hydroxybenzoic
   acid has been reported to be positively correlated with Shigella
   infection in the gut and is involved in this pathogen’s QS mechanism
   [[162]61]. Elevated levels of 4-hydroxybenzoic acid were also detected
   alongside pathogenic bacteria in mice with sepsis [[163]62]. Therefore,
   the presence of 4-hydroxybenzoic acid in the FMD hair samples from the
   AD group may be derived from pathogenic metabolic byproducts as well.
   Furthermore, our functional pathway analysis revealed alterations
   related to stress levels in the diseased FMD hair samples, as the
   “Retrograde endocannabinoid signaling” pathway was significantly
   enriched in the HP, PD and AD hair samples. Previous studies have
   documented the regulatory role of the endocannabinoid system in
   animals’ stress response [[164]63]. The activity of cannabinoid
   receptors could significantly influence the
   hypothalamic–pituitary–adrenal (HPA) axis and directly regulate
   cortisol release [[165]64]. In “Retrograde endocannabinoid signaling”,
   the significantly differential DAG in the AD samples can be converted
   into 2-arachidonoylglycerol (2-AG) via DAGL α/β, activating presynaptic
   cannabinoid receptor 1 (CB1) to modulate synaptic plasticity and stress
   responses [[166]65] ([167]Figure 11E). 2-AG is an important
   endocannabinoid controlling the release of neurotransmitters. This
   process plays an important role in synaptic plasticity, neural
   regulation, and stress response [[168]66]. The spatiotemporal
   regulation of 2-AG signaling is tightly controlled by MAGL-mediated
   degradation and ABHD6-driven hydrolysis [[169]67], with MAGL and ABHD6
   also being observed with remarkable changes in the AD samples
   ([170]Figure 11E). The cortisol levels in hair and plasma samples were
   found to be significantly correlated with the 2-AG and anandamide (AEA)
   contents [[171]68,[172]69]. This is consistent with previous studies
   that have extensively recognized altered cortisol levels in the hair
   and feces of FMD as reliable indicators of significant physiological
   stress [[173]70,[174]71]. Such stress in captive FMD is largely due to
   a conflict between their solitary biological nature and the constraints
   of captivity, which is further exacerbated by their retained wildness
   from a brief domestication history. In line with this, our hair
   metabolome analysis supports the existing literature that links
   excessive stress responses and chronic cortisol release to
   physiological “exhaustion” and impaired immune function in FMD
   [[175]6,[176]7,[177]71]. Consequently, this results in increased
   disease susceptibility, which was notably observed in the sick captive
   FMD individuals in the present study.

   Our study is the first comparative analysis of the untargeted hair
   metabolomics in healthy and diseased FMD, revealing that hair specimens
   hold powerful biomarkers. The AUC values in the ROC analysis for all
   the tested single and composite metabolite sets surpassed 0.7,
   indicating significant diagnostic potential for differential
   metabolites exhibiting high VIP or p values in the comparative
   analysis. The FMD is classified as an endangered species. Consequently,
   biological experiments involving live FMD are strictly prohibited, and
   invasive sampling necessitates an elaborate animal ethics review. The
   symptoms of FMD diseases are often difficult to detect and progress
   rapidly. The collection of samples from FMD that died of disease often
   exhibits a notable time delay, leading to degradation of the DNA and
   RNA, thereby obstructing a full investigation of FMD pathophysiology.
   In contexts where invasive sampling is challenging, non-invasive
   samples like hair and feces have proven to be valuable biospecimens in
   FMD studies [[178]6,[179]72]. Furthermore, for FMD diseases
   characterized by prolonged courses with unnoticeable symptoms, such as
   phytobezoar disease and abscess disease, hair provides a monitoring
   window that extends over several months, making it an exceptionally
   appropriate sample type [[180]73]. The limited sample size (only six
   samples per group) may compromise the generalizability of our findings
   and lead to inflated AUC values. Future multi-center, large-sample
   collaborative research is necessary to validate the disease-associated
   biomarkers identified in this study. Furthermore, targeted metabolomics
   analysis should be employed for a deeper exploration of these
   biomarkers, and their biological functions need to be validated through
   studies in laboratory animals. RNA extraction from FMD hair follicles
   for transcriptome sequencing was attempted; however, the results were
   unsatisfactory due to the insufficient RNA quality and quantity. As a
   pilot study, this research serves as an initial step. Future research
   may focus on improving the sample processing protocol to obtain
   high-quality genetic material from hair follicles, as well as
   validating the correlation between hair metabolites and plasma/tissue
   metabolites. Incorporating multi-omics approaches, such as genomics,
   transcriptomics, and proteomics, could help to comprehensively analyze
   the non-invasive disease markers in FMD. Relative studies will serve as
   an effective research model for disease precautions in domesticated
   wild animals.

5. Conclusions

   In summary, this study represents the first comparative investigation
   of the hair metabolome between healthy and diseased captive FMD. We
   identified significantly altered metabolites in the hair of FMD
   suffering from hemorrhagic pneumonia, phytobezoar disease, and abscess
   disease compared to healthy individuals. Some of these metabolites have
   been previously reported as biomarkers for the corresponding diseases
   or pathogens. In the ROC analysis, both the individual and composite
   metabolite panels demonstrated good sample classification performance.
   Functional analysis revealed that the differential metabolites were
   significantly enriched in pathways related to cancer, parasitism,
   energy metabolism, and physiological stress. Future studies should
   employ larger sample sizes, include experimental animal models, and
   cross-reference diverse tissue samples to validate the hair biomarkers
   identified in this study. The findings highlight the potential of
   non-invasive hair sampling combined with untargeted metabolomics as a
   promising tool for biomonitoring and early disease diagnosis in captive
   FMD.

Author Contributions

   L.Y. and D.H. developed the conceptual structure for the study,
   conducted the literature investigation, and authored the initial draft
   of the manuscript. H.J., Y.L. and Z.X. contributed to the collection of
   the samples and the designs of the experiments in this study. H.Z. and
   D.H. conducted a review of the manuscript and provided suggestions for
   improvement. All authors have read and agreed to the published version
   of the manuscript.

Institutional Review Board Statement

   Ethical review and approval were waived for this study as the hair
   sampling procedure caused no harm or suffering to the FMD.

Informed Consent Statement

   Informed consent was obtained from the owner of the animals.

Data Availability Statement

   The original contributions presented in this study are included in the
   article. Further inquiries can be directed to the corresponding author.

Conflicts of Interest

   Author Zongtao Xu was employed by Beijing Tong Ren Tang Shaanxi Musk
   Industry Co., Ltd.; author Yajun Li was employed by Feng County
   Xinfenghe Animal Hospital Co., Ltd. Both authors declare that the
   research was conducted in the absence of any financial relationships
   that could be construed as potential conflicts of interest. The
   remaining authors also declare no conflicts of interest.

Funding Statement

   This work was funded by the Chinese National Natural Science Foundation
   Project, Grant No. 31872962.

Footnotes

   Disclaimer/Publisher’s Note: The statements, opinions and data
   contained in all publications are solely those of the individual
   author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI
   and/or the editor(s) disclaim responsibility for any injury to people
   or property resulting from any ideas, methods, instructions or products
   referred to in the content.

References