Abstract

Background

   The metabolic patterns of human placental-derived mesenchymal stem cell
   (hP-MSC) treatment for primary sclerosing cholangitis (PSC) remain
   unclear, and therapeutic effects significantly vary due to individual
   differences. Therefore, it is crucial to investigate the serological
   response to hP-MSC transplantation through small molecular metabolites
   and identify easily detectable markers for efficacy evaluation.

Methods

   Using Mdr2^−/− mice as a PSC model and Mdr2^+/+ mice as controls, the
   efficacy of hP-MSC treatment was assessed based on liver pathology,
   liver enzymes, and inflammatory factors. Serum samples were collected
   for ^12C-/^13C-dansylation and DmPA labeling LC–MS analysis to
   investigate changes in metabolic pathways after hP-MSC treatment. Key
   metabolites and regulatory enzymes were validated by qRT-PCR and
   Western blotting. Potential biomarkers of hP-MSC efficacy were
   identified through correlation analysis and machine learning.

Results

   Collectively, the results of the liver histology, serum liver enzyme
   levels, and inflammatory factors supported the therapeutic efficacy of
   hP-MSC treatment. Based on significant differences, 41 differentially
   expressed metabolites were initially identified; these were enriched in
   bile acid, lipid, and hydroxyproline metabolism. After treatment, bile
   acid transport was accelerated, whereas bile acid production was
   reduced; unsaturated fatty acid synthesis was upregulated overall, with
   increased FADS2 and elongase expression and enhanced fatty acid
   β-oxidation; hepatic proline 4-hydroxylase expression was decreased,
   leading to reduced hydroxyproline production. Correlation analysis of
   liver enzymes and metabolites, combined with time trends, identified
   eight potential biomarkers: 2-aminomuconate semialdehyde,
   l-1-pyrroline-3-hydroxy-5-carboxylic acid, l-isoglutamine, and maleamic
   acid were more abundant in model mice but decreased after hP-MSC
   treatment. Conversely, 15-methylpalmitic, eicosenoic, nonadecanoic, and
   octadecanoic acids were less abundant in model mice but increased after
   hP-MSC treatment.

Conclusions

   This study revealed metabolic regulatory changes in PSC model mice
   after hP-MSC treatment and identified eight promising biomarkers,
   providing preclinical evidence to support therapeutic applications of
   hP-MSC.

Supplementary Information

   The online version contains supplementary material available at
   10.1186/s13287-024-03967-y.

   Keywords: Mesenchymal stem cell, Metabolomics, Mdr2^−/− mouse, Primary
   sclerosing cholangitis, Chemical isotope-labeled liquid
   chromatography-mass spectrometry, Efficacy biomarker

Introduction

   Primary sclerosing cholangitis (PSC) is a liver condition characterized
   by damage to bile ducts within or outside the liver, or both. The
   disease process involves an interaction of inflammation, fibrosis, and
   cholestasis [[40]1]. No genetic or environmental factors are associated
   with or known to trigger PSC [[41]2, [42]3]. Several drugs targeting
   biliary components, immunomodulation, fibrosis, or the gut microbiome
   have been tested in PSC patients [[43]4, [44]5]. Ursodeoxycholic acid
   (UDCA) has limited efficacy in altering the long-term course of the
   disease and improving survival [[45]6]. Immunosuppressive medications,
   including steroids, azathioprine, and cyclosporine, are effective in
   specific cases according to some reports; however, their overall
   efficacy remains uncertain and they have substantial adverse effects
   [[46]7]. Currently, liver transplantation remains the sole definitive
   treatment. The 10-year recurrence rate after liver transplantation is
   approximately 20% [[47]8]. Consequently, it is imperative to develop
   novel therapeutic interventions. Recently, cellular therapy,
   particularly mesenchymal stem cell (MSC) therapy [[48]9, [49]10], has
   received considerable attention as a promising therapeutic modality.

   Friedenstein et al. first identified MSCs in mouse bone marrow, and
   subsequent evidence has demonstrated that they originate from pericytes
   and adventitial progenitor cells in almost all tissues [[50]11,
   [51]12]. Human placenta-derived mesenchymal stem cells (hP-MSCs), found
   in the fetal membranes of the full-term placenta, are easily obtained
   using non-invasive techniques. Studies have investigated the potential
   therapeutic applications of MSCs in mouse models of PSC and organoid
   models [[52]13–[53]15]. MSCs ameliorate bile duct hyperplasia,
   peribiliary fibrosis, and inflammation in PSC; they are neither toxic
   or carcinogenic [[54]15], partially confirming the safety and efficacy
   of this therapeutic approach. However, the metabolic patterns after MSC
   therapy remain unclear, and the response to MSC therapy is variable.
   MSCs improve the clinical phenotype in some individuals, whereas others
   do not respond or even deteriorate after receiving cell therapy. Thus,
   it is imperative to discover new biomarkers suitable for evaluating the
   preclinical safety and effectiveness of MSC transplantation in PSC.
   Furthermore, the precise pattern of changes in serologic substances
   after MSC treatment remains unclear.

   Metabolomics is the extensive examination of biomolecules with
   molecular weights below 1000 Da in biological fluids, cells, and
   tissues [[55]16, [56]17]. This approach is invaluable for identifying
   potential biomarkers [[57]17]. Chemical isotope-labeled liquid
   chromatography-mass spectrometry (CIL LC–MS) is widely used in
   metabolomics. This method classifies metabolites based on their
   chemical groups, uses specific reagents to label each type of
   metabolite, and subsequently performs LC–MS analysis [[58]18].
   Differential isotope labeling enables relative quantification of
   labeled metabolites in samples [[59]19]. This approach offers enhanced
   detectability, comprehensive coverage of organismal metabolites, and
   precise quantification [[60]20]. Specifically, ^12C-/^13C-danylation
   labeled LC–MS can be applied to amine/phenol metabolite analysis,
   whereas 2,2-dihydroxy-propanethial (DmPA) labeling is suitable for
   carboxylate compounds [[61]18].

   This study used Mdr2^−/− mice as a PSC disease model and transplanted
   hP-MSCs as a therapeutic intervention. The efficacy of hP-MSC therapy
   was validated through comprehensive pathological and biochemical
   assessments. Mouse serum was collected at multiple time points, and
   semi-quantitative metabolomics analysis was conducted using Danyl and
   DmPA labeling LC–MS. Through the identification of metabolic pathways
   enriched in differentially abundant metabolites and validation of key
   metabolites and enzymes, we elucidated changes in the metabolic
   regulatory network after hP-MSC treatment. Furthermore, bioinformatics
   analysis enabled us to identify potential biomarkers for assessing
   hP-MSC therapeutic efficacy.

Materials and methods

Animal models of primary sclerosing cholangitis and hP-MSC treatment

   Male multidrug resistance gene 2 knockout (Mdr2^−/− or Abcb4^−/−) mice
   (FVB.129P2-Abcb4tm1Bor/J) and Mdr2^+/+ mice (wild-type, WT) were
   procured from The Jackson Laboratory (Bar Harbor, ME, USA) and
   allocated to three groups: Mdr2^+/+ mice as a control group, Mdr2^−/−
   mice treated or not with hP-MSCs as treatment groups, and the PSC model
   group. At week 8, the control and model groups received a caudal vein
   injection of 100 μL phosphate-buffered saline (PBS), whereas the
   treatment group received an injection of 5 × 10^5 hP-MSCs/100 μL PBS.
   Each group used to assess treatment effectiveness contained six mice;
   for serum metabolomics research, the control group comprised 12 mice,
   and the model and treatment groups each contained 15 mice. All mice
   were housed in a specific pathogen-free (SPF) facility at the
   Laboratory Animal Center of The First Affiliated Hospital of Zhejiang
   University School of Medicine, under a 12 h light/dark cycle with
   standard mouse chow and water. Mice were anesthetized by isoflurane
   (95% oxygen and 5% isoflurane, RWD, Shenzhen, China) inhalation and
   euthanized via cervical dislocation when necessary. All operations were
   performed with humane care. The experimental animal protocol was
   approved by the hospital’s Ethics Committee (Approval No: 2020-1088).

Serum sample pretreatment

   Blood samples were collected from mice at weeks 0, 2, 4, and 8 after
   treatment (mouse ages 8, 10, 12, and 16 weeks). After incubation at
   room temperature for 1 h, samples were centrifuged at 4 °C and 3500 rpm
   for 15 min to separate the serum layer. Isolated serum was stored at
   − 80 °C for future use. To precipitate metabolites, serum samples were
   mixed with three volumes of methanol (Fisher Chemical, Waltham, MA,
   USA) chilled to − 20 °C and incubated at − 80 °C for 5 min. Then,
   samples were centrifuged at 15,000 rpm for 30 min at 4 °C and
   freeze-dried with a Labconco machine (Kansas City, MO, USA) to produce
   powdered serum metabolites. An 80-μL aliquot of serum was aspirated
   from each sample; all aliquots were combined to create a pooled sample
   that was subjected to the same procedure to obtain powdered
   metabolites.

Dansylation and DmPA-labeling

   Individual samples were labeled with ^12C-dansyl chloride (Dns-Cl)
   (Nova Medical Testing, AB, Canada); pooled samples were labeled with
   ^13C-dansyl chloride (Nova Medical Testing). The metabolite extract (25
   μL) was mixed with sodium carbonate/sodium bicarbonate buffer (0.5 M,
   pH 9.5, Sigma–Aldrich; 12.5 μL) and acetonitrile (ACN; Sigma–Aldrich;
   12.5 μL). After thorough mixing, a freshly prepared solution of either
   ^12C-dandyl chloride (18 mg/mL in ACN, Sigma–Aldrich) or ^13C-dandyl
   chloride (18 mg/mL in ACN) was added (25 μL). The dansylation reaction
   was carried out at 40 °C for 60 min using a water bath (Thermo Fisher
   Scientific, Waltham, MA, USA). Subsequently, NaOH solution (250 mM, 5
   μL; Sigma–Aldrich) was added and incubated at 40 °C for 10 min to
   remove excess dansyl chloride. Finally, formic acid prepared with
   ACN/H[2]O (425 mM, 25 μL) was added to neutralize and acidify the
   solution. DmPA bromide labeling was performed according to the method
   of Peng et al. [[62]21] with modifications. Single samples were labeled
   with ^12C-DmPA bromide (Nova Medical Testing), whereas pooled samples
   were labeled with ^13C-DmPA bromide (Nova Medical Testing). Pooled
   samples were prepared by combining equimolar amounts of individual
   samples. Both labeled quality control (QC) samples consisted of a
   mixture containing equimolar masses of ^12C-labeled and ^13C-labeled
   pooled samples.

LC–MS analysis

   Labeled samples were analyzed using the UltiMate 3000 UHPLC system
   (Thermo Scientific) coupled with an Impact II Quadrupole Time-of-flight
   (QTOF) mass spectrometer (Bruker, Billerica, MA, USA). Mobile phase A
   (MPA) consisted of 0.1% (v/v) formic acid in water; mobile phase B
   (MPB) contained 0.1% (v/v) formic acid in ACN. The gradient profile for
   metabolite separation was as follows: at t = 0, the composition was 25%
   B; at t = 10 min, it reached 99% B and was held constant until
   t = 13 min; at t = 13.1 min, it returned to the initial condition of
   25% B and was held constant until t = 16 min. The flow rate was set to
   400 μL/min. To assess instrument repeatability and stability before
   destination sample analysis, five consecutive QC samples were tested.
   All samples were randomly arranged, with a blank sample and a labeled
   standard sample included after every 10 completed runs of destination
   samples. MS spectra were acquired in positive ion mode with a spectral
   acquisition rate of 1 Hz.

Mouse serum and liver biochemical tests

   Serum levels of alanine aminotransferase (ALT), aspartate
   aminotransferase (AST), alkaline phosphatase (ALP), triglyceride (TG),
   and high-density lipoprotein cholesterol (HDL) were quantified using a
   dry biochemical analyzer (DRI-CHEM 4000ie; Fujifilm, Tokyo, Japan).
   Serum free fatty acid (FFA), total cholesterol (T-CHO), low-density
   lipoprotein cholesterol (LDL), liver malondialdehyde (MDA), and
   superoxide dismutase (SOD) levels were assessed with commercial kits
   (Jian Cheng Bio-engineering Institution, Nanjing, China). Serum
   interleukin-6 (IL-6), IL-10, IL-1β, and tumor necrosis factor-α (TNF-α)
   levels were determined using the LEGENDplex™ multifactor detection kit
   (BioLegend, San Diego, CA, USA), then analyzed using CytoFLEX LX
   (Beckman Coulter, Brea, CA, USA).

Data analysis

   Using tools provided by OECloud ([63]https://cloud.oebiotech.com), data
   underwent clustering heatmap, correlation, and metabolic pathway
   quantitative enrichment analyses. SIMCA-P 14.1 software was used for
   principal component analysis (PCA) and orthogonal partial least squares
   discrimination analysis (OPLS-DA). Machine learning analysis was
   conducted using Orange software (ver. 3.35.0). All experimental data
   are reported as means ± standard errors of the mean. Statistical
   significance was assessed through one- or two-way analysis of variance
   (ANOVA) or a two-tailed Student’s t-test. All statistical analyses were
   performed using GraphPad Prism software (GraphPad Software, San Diego,
   CA, USA). In all analyses, P-values < 0.05 were deemed statistically
   significant.

Additional methods

   Details regarding hP-MSC culture and identification, flow cytometry,
   immunohistochemical and immunofluorescence staining, oil red O
   staining, Western blotting, RNA extraction, and real-time quantitative
   reverse transcription polymerase chain reaction (qRT-PCR) are provided
   in the Supplementary Materials (Supplementary Methods, Table S1).

Results

Liver injury was substantially reduced in the hP-MSC treatment group

   Osteogenic, chondrogenic, and adipogenic multipotential differentiation
   (Additional file [64]1: Fig. [65]S1A–D) and flow cytometry
   immunophenotype profiles (Additional file [66]1: Fig. [67]S1E) were
   used to confirm the MSC lineage. We monitored body weights of Mdr2^+/+
   and Mdr2^−/− mice (Additional file [68]1: Fig. [69]S2A) at four time
   points and assessed the liver-weight-to-body-weight ratio at the final
   time point. There was no significant difference in body weight among
   the three groups (Fig. [70]1A, Additional file [71]1: Fig. [72]S2B),
   but the liver/body weight ratio decreased (hP-MSC treatment group vs.
   model group, Fig. [73]1B). Hematoxylin and eosin (H&E) staining of
   liver sections showed reduced immune cell infiltration in the confluent
   area and a significant decline in hepatocyte necrosis (Fig. [74]1C).
   Furthermore, progressive worsening of bile duct and liver lesions was
   observed in the model group over time. The levels of inflammatory
   factors IL-6, IL-10, IL-1β, and TNF-α were significantly decreased in
   16-week-old mice (treatment group vs. model group, Fig. [75]1D). Serum
   ALT, AST, and ALP levels increased over time, whereas hP-MSC treatment
   partially alleviated hepatocyte injury (Fig. [76]1E–G). These findings
   provide evidence that hP-MSCs are effective for treating PSC.

Fig. 1.

   [77]Fig. 1
   [78]Open in a new tab

   Effectiveness of hP-MSC therapy for PSC. A Mdr2^+/+ mice, Mdr2^−/− mice
   treated with PBS, and Mdr2^−/− mice treated with hP-MSC, all via tail
   vein injection, served as the control, PSC model, and treatment groups,
   respectively. B Liver-weight-to-body-weight ratio in mice at 16 weeks.
   Compared to the control group, the treatment group showed an increased
   ratio, but it remained lower than that of the model group; one-way
   ANOVA. C Hematoxylin and eosin (H&E) staining of mouse livers at 8, 10,
   12, and 16 weeks; scale bar = 100 μm. D Serum IL-6, TNF-α, IL-1β, and
   IL-10 levels in 16-week-old mice; one-way ANOVA. E Changes in ALT, F
   AST, and G ALP levels; two-way ANOVA; ns, not significant, *P < 0.05,
   **P < 0.01, ***P < 0.001. (D)–(G), n = 6

Serum metabolite profiling indicated that hP-MSC treatment affected
endogenous metabolism

   Serum samples were isotopically labeled and subjected to CIL LC–MS
   analysis (Additional file [79]1: Fig. [80]S3A). Instrument performance
   remained stable throughout testing, and the signal responses of samples
   within each group were consistent (Additional file [81]1: Fig.
   [82]S3B–D). In dansyl-labeled samples, 1698 peak pairs were identified,
   compared with 1854 peak pairs in DmPA-labeled samples. This analysis
   produced 1338 (78.9%) accurately identified or inferred peak pairs with
   respect to their carbon isotope composition (^12C-/^13C). Additional
   file [83]1: Table S2 and S3 lists the accurately identified substances.

   The PCA score plot revealed that QC samples clustered together,
   indicating excellent LC–MS stability and repeatability, ensuring
   reliable data. The metabolic profile of the treatment group was more
   similar to the control than to the model group (Fig. [84]2A). The
   robustness of the OPLS-DA supervised model was confirmed by
   200-replacement validation (Additional file [85]1: Fig. [86]S3E and F).
   The OPLS-DA score plot demonstrated discernible intergroup separation
   and intragroup aggregation among the three groups (Fig. [87]2B),
   suggesting that Mdr2^−/− mice undergoing MSC treatment experienced
   substantial changes in amino/phenol hydroxyl and carboxyl metabolites
   compared with untreated mice.

Fig. 2.

   [88]Fig. 2
   [89]Open in a new tab

   Metabolomics analysis of serum. A Score plots of PCA and B OPLS-DA in
   the control, model, and treatment groups and QC samples. C Volcano
   plots of the control vs. model groups and model vs. treatment groups at
   10 weeks. |FC|> 1.2, P < 0.05. D Venn diagram of metabolites that
   differed at consecutive time points. E Heatmap of mean relative
   abundances of 41 metabolites that differed at consecutive time points.
   F Quantitative enrichment analysis of metabolic pathways based on the
   41 differentially expressed metabolites

   To identify significant metabolites responsible for metabolomic
   changes, paired t-tests and fold change (FC) values were used to
   compare changes in peak pairs. The volcano plot revealed 1896, 1678,
   and 1539 differential peak pairs between the control and model groups
   at 10, 12, and 16 weeks, respectively. Similarly, there were 888, 491,
   and 762 differential peak pairs between the model and treatment groups
   at the same time points (Fig. [90]2C, Additional file [91]1: Fig.
   [92]S4A and B). Venn diagram analysis of these differentially expressed
   metabolites from intergroup comparisons clarified common changes across
   all groups. At week 10, 636 differential peak pairs were shared between
   the control and model groups and model and treatment groups; there were
   319 peaks at week 12 and 525 peaks at week 16 (Additional file [93]1:
   Fig. [94]S4C). From these shared peaks, we identified corresponding
   substances at each time point: 149, 64, and 138 substances at weeks 10,
   12, and 16, respectively. Furthermore, the intersection of
   differentially expressed metabolites among these three groups yielded
   metabolites that consistently changed across all three time points (17
   species), as well as those that changed over two consecutive time
   points (18 + 6 species) (Fig. [95]2D). Ultimately, this approach
   identified 41 differentially expressed metabolites (Additional file
   [96]1: Table S4). Temporal changes in these metabolites were visualized
   using heat maps (Fig. [97]2E), revealing that MSC treatment induced
   significant differences in the metabolic profiles of the model and
   treatment groups at subsequent time points. These findings suggested a
   shift in the metabolic profile of the mice after hP-MSC treatment.

   Next, we performed quantitative enrichment analysis based on chemical
   structures to obtain an overview of the main classes of differential
   metabolites. The most enriched classes between the treatment and model
   groups were amino acids and peptides, followed by fatty acids and their
   derivatives (Additional file [98]1: Fig. [99]S4D). To investigate
   changes in metabolic pathways, we performed quantitative enrichment
   analysis of 41 different metabolites using the KEGG database
   ([100]https://www.kegg.jp/) and identified 27 significantly perturbed
   pathways between the treatment and model groups (Fig. [101]2F and
   Additional file [102]1: Table S5). The major metabolic pathways
   encompassed biosynthesis of unsaturated fatty acids, tryptophan
   metabolism, alanine, aspartate, and glutamate metabolism, lysine
   degradation, nicotinate and nicotinamide metabolism, and arginine and
   proline metabolism, consistent with the prominent amino acids and fatty
   acids identified in the differential metabolite categories. Because the
   41 differential metabolites included bile acids, alcohols and
   derivatives, fatty acids, and fatty acids with proline that play
   crucial roles in pathway analysis, we hypothesized that MSC treatment
   influences specific metabolic processes, such as dysregulation of bile
   acid, lipid, and proline metabolism.

Disrupted bile acid metabolism was restored after hP-MSC treatment

   Of the 41 metabolites identified in the initial screening, changes in
   the glycine derivatives glycodeoxycholic acid and N-heptanoylglycine
   suggested alterations in bile acid content after hP-MSC treatment. We
   investigated the regulatory effects of hP-MSC on bile metabolism and
   its efficacy in alleviating bile stasis. Immunohistochemical analysis
   of cytokeratin 19 (CK19) from 16-week-old mice (Fig. [103]3A and B)
   showed significant bile duct cell proliferation and bile duct reaction
   in the hepatic parenchyma of the model group. MSC treatment reduced
   epithelial cell proliferation. Furthermore, serum total bile acid
   levels of 16-week-old mice revealed a steady decrease after MSC
   treatment (Fig. [104]3C). To elucidate specific changes in bile acids,
   we comprehensively searched the identified compounds for additional
   bile acids. This search revealed 17 bile acids (Additional file [105]1:
   Table S6). The thermograms from all time points in three groups
   (Additional file [106]1: Fig. [107]S5A and B) showed that the model
   group had higher levels of bile acid sludge and toxic bile acids
   compared with the control group, but these levels improved after
   treatment. Most bile acids exhibited the greatest improvement at
   16 weeks (Fig. [108]3D). Various bile acids, including α-phocaecholic
   acid, significantly decreased in the treatment group (Fig. [109]3E).
   Notably, reduced gene expression levels of cytochrome P450 family 7
   subfamily A member 1 (Cyp7a1) and cytochrome P450 family 27 subfamily B
   member 1 (Cyp27b1) (Fig. [110]3F) in the treatment group suggested a
   decrease in bile acid biosynthesis. The gene expression levels of
   basolateral transporter multidrug resistance-associated protein 2
   (Mrp2), Na ( +)/bile acid cotransporter (Ntcp), and organic anion
   transporting polypeptides (Oatp4) were reduced (Fig. [111]3G),
   indicating increased bile acid excretion. Collectively, MSC treatment
   improved bile acid metabolism and alleviated cholestasis in Mdr2^−/−
   mice.

Fig. 3.

   [112]Fig. 3
   [113]Open in a new tab

   hP-MSC treatment remedied bile acid metabolism disorders. A
   Immunohistochemical staining of CK19 in mouse livers; scale
   bar = 100 μm. B The quantification of CK19 positive area in mouse
   livers (n = 5). C Serum total bile acid levels in 16-week-old mice
   (n = 6). D Heat map of the 17 bile acids identified at 16 weeks. E
   Typical bile acids altered at 16 weeks (n = 12–15). F Hepatic mRNA
   levels of Cyp7a1 and Cyp27b1 and G Mrp2, Ntcp, and OATP4 in three
   groups of mice (n = 6). *P < 0.05, **P < 0.01, ***P < 0.001, one-way
   ANOVA

hP-MSC treatment strongly influenced lipid metabolism

   Metabolic pathway enrichment analysis indicated that unsaturated fatty
   acid synthesis underwent significant changes after MSC treatment. A
   comprehensive database search identified the main unsaturated fatty
   acids, particularly ω-3 and ω-6 polyunsaturated fatty acids (PUFAs).
   The majority of PUFAs increased post-treatment, approaching levels
   observed in the control group (Additional file [114]1: Fig. [115]S6).
   To identify the mechanisms regulating differential fatty acid
   expression, we assessed the gene expression level of key regulatory
   enzymes associated with unsaturated fatty acids based on KEGG
   regulatory networks (Fig. [116]4A). Delta-5 desaturase (Fads1) and
   acyl-CoA 6-desaturase (Fads2) belong to the family of fatty acid
   desaturases, which introduce double bonds at specific carbon atoms in
   fatty acid chains to regulate their degree of unsaturation. After MSC
   treatment, the expression of Fads1 increased. Conversely, no
   significant differences relative to the control group were noted in the
   expression levels of stearoyl-CoA desaturase 1 (Scd1) and acyl-CoA
   thioesterase 2 (Acot2), which regulate saturated fatty acid production
   to form monounsaturated fatty acids and catalyze acyl-CoA hydrolysis
   into free fatty acids and CoA, respectively. Furthermore, upregulated
   expression levels of ELOVL2 (elongation of very long chain fatty
   acids-like 2) and ELOVL5, involved in regulating omega-3 and omega-6
   PUFA elongation, were observed after MSC treatment. Figure [117]4B
   shows the integrated network depicting the unsaturated fatty acid
   synthesis process and its regulation by key enzymes. In summary,
   overall upregulation was observed in the unsaturated fatty acid
   synthesis pathway after treatment. Because unsaturated fatty acids had
   a substantial impact on overall lipid metabolism, we next evaluated a
   series of lipid-related parameters in mice.

Fig. 4.

   [118]Fig. 4
   [119]Open in a new tab

   hP-MSC treatment moderated lipid metabolism disorders. A Expression of
   hepatic genes involved in unsaturated fatty acid synthesis. B Network
   of ω-3 and ω-6 series unsaturated fatty acid biosynthesis. In the ω-3
   series, α-linolenic acid (ALA, 18:3) is primarily converted into
   stearidonic (18:4), eicosapentaenoic (EPA, 20:5), docosapentaenoic
   (DPA, 22:5), and docosahexaenoic (DHA, 22:6) acids. In the ω-6 series,
   linoleic acid (LA, 18:2) is mainly converted into γ-linolenic (GLA,
   18:3), dihomo-γ-linolenic (DGLA, 20:3), arachidonic (ARA, 20:4),
   adrenic (ADA, 22:4), and docosapentaenoic (DPA, 22:5) acids. Red
   indicates an increase after treatment, blue indicates no difference
   after treatment, and gray indicates undetermined substances or untested
   genes. Δ represents desaturase. (C) Serum FFA, T-CHO, LDL, HDL, and TG
   levels (n = 6). D Oil red O staining showed that the TG level declined
   after hP-MSC treatment; scale bar = 100 μm. E The quantification of
   lipid droplet area in mouse livers (n = 5). F Expression of hepatic
   genes involved in cholesterol synthesis, G lipogenesis, and H
   mitochondrial β-oxidation (n = 6). Liver MDA I and SOD J levels
   (n = 6). K Immunohistochemistry staining showed that γ-H2AX declined
   after hP-MSC treatment. L The quantification of γ-H2AX positive area in
   mouse livers (n = 5). M Nrf2, HO-1, and NQO1 gene expression levels in
   mouse livers (n = 6). ns, not significant, *P < 0.05, **P < 0.01,
   ***P < 0.001, one-way ANOVA

   Initially, we assessed serum lipid-related biochemical indices. Serum
   FFA and T-CHO levels were reduced in Mdr2^−/− mice, whereas MSC
   treatment increased these levels (Fig. [120]4C). This change was
   attributed to reduced hepatic lipid accumulation; serum T-CHO levels
   were lower in Mdr2^−/− mice than in control mice [[121]22]. LDL levels
   were reduced in both the control and treatment groups, whereas they
   were elevated in the model. HDL was decreased in Mdr2^−/− mice, but its
   level did not increase after MSC treatment (Fig. [122]4C). MSCs may
   selectively internalize LDL. There were no pronounced fluctuations in
   TG levels among the three groups. Examination of oil red O-stained
   liver samples (Fig. [123]4D and E) confirmed the absence of significant
   changes. Analysis of key genes involved in lipid and cholesterol
   metabolism—sterol regulatory element binding transcription factor 2
   (Srebp2) and 3-hydroxy-3-methylglutaryl-CoA reductase (Hmgcr)—did not
   show significant changes after MSC treatment (Fig. [124]4F). The
   lipogenic gene sterol regulatory element-binding protein 1c (Srebp1c)
   also showed no significant change compared with the model group
   (Fig. [125]4G). Previous studies have consistently revealed reduced
   mitochondrial function during cholestasis [[126]23, [127]24],
   characterized by impaired fatty acid oxidation [[128]23]. Therefore, we
   investigated pivotal genes in fatty acid oxidation. The mRNA levels of
   peroxisome proliferator activated receptor α (Pparα) and its target
   genes PPARγ coactivator 1α (Pgc1α), carnitine palmitoyltransferase 1A
   (Cpt1α), and acyl-coenzyme alternative oxidase (Aox) (Fig. [129]4H)
   were decreased in the model group; these levels increased after MSC
   treatment. This finding indicated decreased fatty acid breakdown in
   Mdr2^−/− mice during cholestasis, with potential MSC-induced
   improvement through the promotion of β-oxidation. Furthermore,
   β-oxidation of fatty acids enhanced unsaturated fatty acid synthesis.
   Excessive lipid deposition led to toxic injury; changes in hepatic MDA
   levels (F[130]ig. [131]4I) and hepatic SOD activity (Fig. [132]4J)
   indicated an antioxidant response. Immunohistochemical analysis of
   histone γ-H2AX staining (Fig. 4K, L) also indicated an attenuated
   oxidative damage response in MSC-treated mice compared with model mice.
   Increased levels of heme oxygenase 1 (Hox1), nuclear factor erythroid
   2-related factor 2 (Nrf2), and NAD(P)H quinone dehydrogenase 1 (NQO1)
   in Fig. [133]4M indicated initiation of the antioxidant defense
   mechanism.

hP-MSC treatment improved hydroxyproline-dependent collagen metabolic
disorders

   Analysis of key components in the arginine and proline metabolic
   pathways revealed significant downregulation of proline
   l-1-pyrroline-3-hydroxy-5-carboxylate, trans-4-hydroxyl-proline, and
   1-pyrroline-5-carboxylic acid (P5C) after MSC treatment. P5C is an
   intermediary product derived from glutamate and ornithine, bridging the
   tricarboxylic acid and urea cycles. However, no significant changes
   were observed in glutamate or ornithine levels after treatment
   (Fig. [134]5A and B). The intermediate step in these pathways involves
   enzymatic conversion of proline to hydroxyproline, catalyzed by prolyl
   4-hydroxylase (Fig. [135]5A). Considering the close association of
   proline and hydroxyproline metabolism with collagen formation and
   degradation [[136]25], Mdr2^−/− mice might exhibit disrupted collagen
   fiber metabolism. To assess fibrosis, we performed Sirius red and
   Masson staining (Fig. [137]5C and D), which revealed a significant
   increase in hyperplastic collagen fibers surrounding bile ducts in
   model mice; this phenomenon was considerably improved in the treated
   mice. Alpha-smooth muscle actin (α-SMA) staining showed myofibroblast
   proliferation and aggregation (Fig. [138]5C and D). Hepatic mRNA levels
   of α-SMA, collagen alpha 1 chain type I (Col1a1), TIMP metallopeptidase
   inhibitor 1 (Timp1), and transforming growth factor beta 1 (Tgf-β1)—all
   markers of fibrosis—also indicated that MSC treatment ameliorated
   collagen fiber metabolic disorders in mice (Fig. [139]5E). Paha2
   encodes a critical subunit of prolyl 4-hydroxylase, an essential enzyme
   involved in collagen synthesis, which facilitates the 4-hydroxyproline
   production necessary for correct folding of recently formed procollagen
   molecules. Studies have revealed elevated P4HA2 mRNA levels in
   individuals with PSC, highlighting its contributions to biliary tract
   reactivity and fibrosis development [[140]26]. Using qRT-PCR, we
   observed upregulated gene expression in the livers of model mice;
   however, after MSC treatment, the Mdr2^−/− mice showed decreased
   expression (Fig. [141]5F). These changes were validated at the protein
   level (Fig. [142]5G). Immunohistochemical analysis revealed predominant
   expression of P4HA2 in the fibrotic tissue areas, with decreased
   expression after treatment (Fig. [143]5 and I). According to reports,
   P4HA2-positive cells display the morphological characteristics of
   biliary reactive cells [[144]26]. To further validate this observation,
   we performed immunofluorescence staining for P4HA2 and CK7
   (Fig. [145]5J). CK7 immunostaining delineated clusters of biliary cells
   in the peribiliary region that responded to biliary liver injury.
   Consistent with the immunohistochemical findings, there was extensive
   overlap in the distribution of P4HA2^+ cells and CK7^+ biliary cells.
   Furthermore, P4HA2 expression was rarely detected in normal mice. In
   control group mice, P4HA2 expression increased, while it decreased in
   the treatment group. Consequently, our findings suggest that hP-MSC
   therapy affected hydroxyproline metabolism in mice.

Fig. 5.

   [146]Fig. 5
   [147]Open in a new tab

   hP-MSC treatment alleviated proline‑dependent collagen metabolism
   disorders. A Schematic diagram of proline metabolic pathways. Blue,
   decreased after treatment; yellow, no change after treatment; gray, not
   identified; red, key enzyme. Solid lines indicate direct
   transformation, whereas dotted lines represent omitted intermediates. B
   Relative composition of key metabolites involved in proline metabolism
   (n = 12–15). C Sirius red, Masson, and immunohistochemical staining of
   α-SMA in mouse livers; scale bar = 100 μm. D Assessment of fibrosis
   based on Sirius red-, Masson-stained and α-SMA positive regions
   (n = 5). E Hepatic expression of α-SMA, COL1A1, TIMP1, and TGF-β1
   (n = 6). F Hepatic expression of P4HA2 (n = 6). G Western blot analysis
   of the liver showed greater P4HA2 expression in the model group, which
   was downregulated by hP-MSC treatment. H Immunohistochemical staining
   of P4HA2 in mouse livers. Arrows indicate positive areas. ns, not
   significant. I The quantification of P4HA2 positive area in mouse
   livers (n = 5). J Representative confocal images of CK7, P4HA2 and DAPI
   in mouse livers. ns, not significant, *P < 0.05, **P < 0.01,
   ***P < 0.001, one-way ANOVA

Differential metabolites associated with clinical phenotypes

   The above experiments revealed the effect of MSC treatment for PSC from
   a metabolomic perspective and further illustrated the credibility of
   serum sample metabolomics in this study. We obtained 41 differentially
   expressed metabolites as potential biomarkers of treatment
   effectiveness. To filter these potential biomarkers and determine
   whether any are linked to disease severity, we conducted Spearman
   correlation analysis of their relative abundances with clinical
   phenotypes; specifically, we monitored the ALT, AST, and ALP levels of
   the three groups at each time point (Additional file [148]1: Fig.
   [149]S7A–C). This analysis showed that the effect of MSC treatment was
   more pronounced at 16 weeks. Therefore, we regarded liver enzyme data
   at 16 weeks as the clinical phenotype to further analysis. The
   calculated correlation coefficients and significance levels are
   recorded in Additional file [150]1: Table S7. As shown in Fig. [151]6A
   and Additional file [152]1: Fig. [153]S7D, most metabolite factors were
   associated with disease severity. Among these,
   l-1-pyrroline-3-hydroxy-5-carboxylic acid, a breakdown product of
   hydroxyproline, was positively correlated with liver enzymes
   (Fig. [154]6B), and monounsaturated eicosenoic acid was negatively
   correlated with liver enzymes (Fig. [155]6C). Next, we used the
   criterion |ρ|> 0.6 and p < 0.05 to screen the 41 differential
   metabolites, identifying eight differential metabolites:
   l-1-pyrroline-3-hydroxy-5-carboxylic acid, maleamic acid,
   2-aminomuconate semialdehyde, l-isoglutamine, 15-methylpalmitic acid,
   octadecanoic acid, eicosenoic acid, and nonadecanoic acid. Generally,
   metabolite levels became normalized after treatment. A significant
   difference at two or three consecutive time points might indicate a
   treatment-induced change in metabolites. Additional file [156]1: Fig.
   [157]S8 shows that all eight metabolites followed the expected trend,
   confirming their suitability as potential biomarkers of MSC treatment
   effectiveness.

Fig. 6.

   [158]Fig. 6
   [159]Open in a new tab

   Associations between metabolites and clinical phenotypes. A Correlation
   heatmap of metabolites and liver enzymes (Spearman analysis, n = 42).
   Red signifies a positive correlation, whereas blue denotes a negative
   correlation. Gray represents correlation coefficients |ρ|< 0.6 or
   p < 0.05. B–C Examples of individual liver enzyme–metabolite
   associations. Logistic regression classifiers comprising liver enzymes,
   metabolites, and their combination could discriminate effective and
   ineffective individuals in the D training and E test sets. *P < 0.05,
   **P < 0.01, ***P < 0.001

Eight metabolites as potential biomarkers for evaluating hP-MSC therapy
efficacy

   To investigate whether these eight serum metabolites could serve as
   biomarkers to determine MSC treatment effectiveness, we performed
   machine learning modeling. Serum samples from Mdr2^−/− mice
   administered either PBS or hP-MSCs were randomly assigned in a 2:1
   ratio to the training and test sets, respectively. We studied seven
   fivefold cross-validated machine learning models: logistic regression
   (LR), support vector machine (SVM), random forest (RF), neural network
   (NN), naive Bayes (NB), adaBoost (AB), and k-nearest neighbor (kNN)
   (Additional file [160]1: Tables S8–S10). Machine learning model
   classification effectiveness was assessed using metrics such as the
   area under the receiver operating characteristic curve (AUC), accuracy,
   F1-score, precision, and recall. For comprehensive comparison, we
   selected the logistic regression model. Figure [161]6D shows that
   metabolites were more effective than liver enzymes in distinguishing
   the effectiveness of MSC treatment (AUC 0.933 and 0.843). The
   combination of liver enzymes and metabolites yielded better results
   (AUC 0.947). Our conclusions were verified by data from the testing set
   (Fig. [162]6E). The AUCs for liver enzymes, metabolites, and their
   combination were 0.757, 0.932, and 0.978, respectively. The confusion
   matrix also confirmed the validity of the model and the reliability of
   the selected biomarkers (Additional file [163]1: Fig. [164]S9). Our
   study indicated that these biomarkers (Table [165]1 and Fig. [166]7)
   can effectively evaluate MSC efficacy in PSC treatment. This finding
   provides a reliable scientific basis for individualized treatment
   decisions and will inform future clinical practice.

Table 1.

   Eight identified serum biomarkers
   No RT (s) Normalized RT (s) m/z_light m/z_heavy Neutral mass (Da)
   Intensity nCharge nTag External Identifier Compound
   S144 138.6 128.7 291.1338 293.1405 129.0424 1,002,302.47 1 1
   [167]C04281 L-1-Pyrroline-3-hydroxy-5-carboxylic acid
   S192 159.8 151.9 277.1183 279.1251 115.0269 65,500 1 1 [168]C01596
   Maleamic acid
   S372 221.9 213.7 303.1341 305.141 141.0426 28,216 1 1 [169]C03824
   2-Aminomuconate Semialdehyde
   S376 222.1 214 469.2444 473.2578 146.069 1,942,488.886 1 2 [170]C16673
   L-Isoglutamine
   S1832 746.6 777.9 432.3462 434.3533 270.2559 710,872 1 1 HMDB0061709
   15-Methylpalmitic Acid
   S1846 769 814.3 446.3621 448.3688 284.2707 2,706,476.994 1 1
   [171]C01530 Octadecanoic acid
   S1847 769.9 815.2 472.3777 474.3844 310.2864 1,362,167.2 1 1
   [172]C16526 Icosenoic acid
   S1854 792.9 838.2 460.3778 462.3846 298.2865 41,572.52632 1 1
   [173]C16535 Nonadecanoic acid
   [174]Open in a new tab

   No.: individual peak pair number coded from the LC–MS experiment; RT
   (s): the retention time of the peak pair detected in sec; Normalized RT
   (s): the corrected retention time of the peak pair with Universal RT
   Calibrant data; m/z light: the mass of light-chain (^12C) labeled
   metabolite in the peak pair; m/z heavy: the mass of heavy-chain (^13C)
   labeled metabolite in the peak pair; Neutral Mass (Da): the neutral
   monoisotope mass of the metabolite (i.e., labeled mass—the mass of the
   labeling group); Intensity: the average intensity of the heavy labeled
   metabolite in the peak pair in all samples; nCharge: the number of
   charge in the peak pair detected; nTag: the number of labeling groups
   or tags in the labeled metabolite; External Identifier and Compound:
   the external identifier and name of metabolites positively identified
   using accurate mass and retention time matches against a labeled
   standard compound library, respectively

Fig. 7.

   [175]Fig. 7
   [176]Open in a new tab

   Relative levels of eight potential biomarkers at weeks 8, 10, 12, and
   16. n = 12–15. ns, not significant; *P < 0.05, **P < 0.01,
   ***P < 0.001; unpaired t-test with Welch correction

Discussion

   This study demonstrated that hP-MSC treatment affects metabolic
   disorders in Mdr2^−/− mice, a PSC model, with particularly strong
   influences on bile acid, lipid, and hydroxyproline metabolism. Although
   the beneficial effects of hP-MSCs on PSC have been confirmed through
   pathological and biochemical analysis [[177]27–[178]29], the precise
   underlying mechanism has been unclear. Therefore, a multi-timepoint
   metabolomics analysis was conducted to elucidate changes in small
   molecule metabolites during treatment. Through a series of
   bioinformatics analyses, 41 differential metabolites were initially
   identified; quantitative metabolic pathway enrichment analysis was then
   performed. Differences between MSC treatment and model groups were
   primarily observed in pathways such as unsaturated fatty acid
   synthesis, proline metabolism, and tryptophan metabolism. Differential
   metabolites included one bile acid and three glycine derivatives. Thus,
   it is reasonable to hypothesize that MSC therapy influences the
   metabolic profiles of bile acids, lipids, and proline.

   The balance of bile acid levels is disrupted in various liver
   disorders, especially those involving cholestasis and bile duct damage.
   Bile acids play a crucial role in PSC progression [[179]30]. The
   primary route for bile acid synthesis, the classical or neutral
   pathway, is initiated by CYP7A1-mediated cholesterol 7α-hydroxylation.
   This enzyme regulates the overall rate of bile acid production and
   participates in approximately 90% of all bile acid synthesis. Bile
   acids are amphiphilic molecules synthesized via hepatic cholesterol
   metabolism, comprising primary bile acids directly produced by
   hepatocytes and secondary bile acids formed upon intestinal secretion.
   In humans, most bile acids are conjugated with glycine, and the
   remainder are conjugated with taurine [[180]31]. In hepatocytes, bile
   acid transport between blood and bile is facilitated by hepatic uptake
   transporters, such as NTCP and OATPs. After attachment to intracellular
   bile acid-binding proteins, bile acids are released into tubules via
   efflux transporters, such as bile salt export pump (BSEP) and MRP.
   Subsequently, bile acids are transported via bile ducts to the
   intestinal lumen, where they aid lipid breakdown and cholesterol
   emulsification for absorption [[181]32]. Yang et al. investigated
   metabolic changes in three cholestatic mouse models, revealing elevated
   bile acid levels and reduced arginine concentrations [[182]33]. In this
   study, we identified 17 bile acids and their intermediates within the
   metabolic group; most bile acids exhibied significant elevation in the
   model group and subsequent reduction after hP-MSC treatment. MSCs
   modulate various bile acid transporters and key enzymes involved in
   synthesis, thereby regulating the entire bile acid metabolism network
   and mitigating the detrimental systemic effects of cholestasis.
   Additionally, comprehensive bile acid data throughout the study period
   demonstrated that MSCs had the greatest therapeutic efficacy at
   16 weeks.

   Considering the crucial role of bile acids in regulating lipid and
   energy balance, changes in lipid metabolism are inevitable during
   cholestatic liver disease [[183]22, [184]34, [185]35]. Lipid levels in
   PSC typically exceed those recommended for treatment with
   lipid-lowering drugs [[186]36]. In the analysis of 41 differential
   metabolites and their pathway enrichment, we observed that unsaturated
   fatty acid synthesis plays a central role in metabolic processes. Most
   unsaturated fatty acids were found at lower levels in model group,
   whereas they showed significant increases in the control and treatment
   groups. PUFAs are widely acknowledged for their health-promoting
   properties and crucial physiological functions, particularly in terms
   of regulating cholesterol levels and inflammatory responses [[187]37,
   [188]38]. Studies have demonstrated that PPAR binds to the ω-3 fatty
   acids docosahexaenoic acid (DHA) and eicosapentaenoic acid (EPA),
   ultimately inhibiting cytokine production associated with the NF-κB
   inflammatory master transcription factor [[189]39]. To elucidate the
   impact of hp-MSCs on lipid metabolism, we conducted supplementary
   experiments. Serum biochemical data demonstrated that hP-MSCs
   effectively reduced LDL content in the model group, but no significant
   impact on HDL content was observed. We attributed this finding to MSCs’
   preferential affinity for LDL loading. Furthermore, metabolic profiles
   indicated that cholestasis led to disrupted lipid metabolism. Detection
   of key genes involved in lipid metabolism showed that MSC treatment
   primarily enhances fatty acid β-oxidation. Mitochondrial dysfunction
   plays a pivotal role in cholestatic disorders and is frequently
   associated with impaired fatty acid β-oxidation [[190]40]. Reduced PUFA
   levels promote oxidative stress, thereby exacerbating hepatic injury
   [[191]41, [192]42]. Lipid accumulation in hepatocytes and subsequent
   reactive oxygen species production can cause liver inflammation and
   injury by activating Kupffer cells and hepatic stellate cells
   [[193]43]. The primary mechanism underlying this hepatotoxicity is
   believed to be the inhibition of fatty acid β-oxidation [[194]44].
   Assessments of liver MDA and SOD levels, along with γ-H2AX pathological
   staining, revealed that hP-MSCs effectively mitigated oxidative
   stress-induced hepatocyte damage. Furthermore, we observed upregulation
   of the Nrf2 signaling pathway after treatment. Recent findings indicate
   that the transcription factor Nrf2 plays a role in maintaining lipid
   balance, along with its known antioxidant and anti-inflammatory
   properties [[195]45, [196]46]. Nrf2 depletion resulted in reduced
   β-oxidation gene expression [[197]47].We speculate that the alleviation
   of lipid accumulation-induced oxidative injury results from activation
   of the Nrf2/HO-1 pathway facilitated by hP-MSCs.

   Proline metabolism is important in the development of chronic
   cholestatic disorders [[198]48, [199]49]. We observed a substantial
   decrease in the amount of l-1-pyrroline-3-hydroxy-5-carboxylic acid, a
   breakdown product of hydroxyproline, as well as the levels of
   hydroxyproline in the control and treatment groups. Collagen synthesis
   regulation is intricately linked to proline and hydroxyproline
   metabolism [[200]50]. Based on collagen staining and the assessment of
   key fibrosis-related genes, we propose that hP-MSCs effectively
   ameliorate aberrant collagen turnover in mice. Prolyl 4-hydroxylase is
   a crucial regulatory node in the proline metabolism network that
   catalyzes the formation of 4-hydroxyproline, a pivotal enzyme in
   collagen synthesis [[201]51, [202]52]. Zhang et al. reported that the
   livers of individuals with PSC express more P4HA2, specifically in
   close proximity to cells involved in the ductal response, suggesting
   that P4HA2 targeting could be a useful therapeutic approach for biliary
   response and fibrosis [[203]26]. P4HA2 targeting may also reduce
   collagen deposition in hepatocellular carcinoma [[204]53]. Our
   investigation revealed significantly elevated P4HA2 expression in
   treatment group livers compared with those of model mice; this
   phenomenon was reversed after treatment intervention with hP-MSCs.
   Therefore, we hypothesize that hP-MSCs can improve collagen metabolic
   disorders by P4HA2 targeting, but the specific mechanism requires
   further investigation.

   We identified altered metabolic pathways and sought potential
   biomarkers from the differentially expressed metabolites. After initial
   screening, we detected 41 differential metabolites. Through correlation
   analysis and validation of machine learning models, we ultimately
   assessed the efficacy of eight potential biomarkers. Of these,
   maleamic, 15-methylpalmitic, octadecanoic, eicosenoic, and nonadecanoic
   acids are fatty acids or their derivatives; conversely, 2-aminomuconic
   acid semialdehyde, l-isoglutamine, and
   l-1-pyrroline-3-hydroxy-5-carboxylic acid are amino acids or their
   derivatives. A notable upward trend in the levels of all fatty acids or
   derivatives was observed post-treatment, except for maleamic acid,
   which may be attributed to increased unsaturated fatty acid synthesis.
   In contrast, a downward trend was evident in several amino acid
   derivatives after treatment, especially concerning
   l-1-pyrroline-3-hydroxy-5-carboxylic acid, a hydroxyproline metabolite
   for which reduction may be linked to decreased collagen hydroxyproline
   synthesis. Furthermore, 2-aminomuconic acid semialdehyde is a byproduct
   of tryptophan metabolism, suggesting that its alteration is associated
   with changes in gut microbiota activity [[205]54]. The abundances of
   serum biomarkers may be useful in establishing an effective system for
   MSC product monitoring and assist future PSC applications. Our AUC
   analysis indicated that the accuracy of these biomarkers is superior to
   that of liver enzymes. Nevertheless, these results should be verified
   in larger studies and human participants; this need for validation
   constitutes a limitation of our study. In addition, further studies are
   needed to uncover the precise mechanisms behind the metabolic changes
   during hP-MSC treatment.

Conclusion

   We clarified the therapeutic benefits of hP-MSCs in PSC by improving
   imbalances in bile acid, lipid, and collagen metabolism in a
   hydroxyproline-dependent manner. Furthermore, we identified eight
   potential biomarkers that can evaluate hP-MSC therapy effectiveness in
   PSC. These findings offer novel insights into clinical applications of
   stem cell therapy.

Supplementary Information

   [206]Additional file 1.^ (3.7MB, docx)

Acknowledgements