Patient characteristics

A total of 186 patients with DLC were screened from March 2022 to March 2024. The study comprised two stages (Phases Ia and Ib), as shown in Fig. 1. Phase Ia enrolled 15 patients who received MSC infusions following a “3 + 3” dose-escalation protocol. Three patients were enrolled in Cohorts I, II, and IV, while Cohort III included six patients. As per the study protocol, the enrollment of additional patients in Cohort III was necessitated by an adverse event (rash, level 1, possibly related) observed in one patient on the second day post-infusion. After the 28-day follow-up, MSC dosages and the interval of Phase Ib were determined based on Phase Ia results. Phase Ib enrolled nine patients. Cohort A (1.0 × 10⁸ cells per dose, three doses) included three patients, while Cohort B (2.0 × 10⁸ cells per dose, three doses) included six patients to comprehensively evaluate safety.

Fig. 1

Study design. a Phase Ia; b Phase Ib. scRNA-seq (single-cell RNA sequencing) profiles transcriptomes across thousands of individual cells, whereas CyTOF (cytometry by time of flight) uses metal-tagged antibodies to simultaneously quantify >40 protein markers at single-cell resolution. Integrated analysis enables cross-validation between immune cell surface protein expression and transcriptional states

The median age of the participants was 51.0 ± 10.1 years; 15 (62.5%) were males, and 9 (37.5%) were females. The median MELD score was 12.38 ± 3.42, and the median Child‒Pugh score was 8.12 ± 1.36. The leading etiologies were HBV infection (9/24, 37.5%) and alcohol use (9/24, 37.5%), with fewer cases attributed to primary biliary cirrhosis (5/24, 20.8%) or HCV infection (1/24, 4.2%). Patients exhibited liver dysfunction (median MELD score: 12.38 ± 3.42; Child‒Pugh score: 8.12 ± 1.36) and typical complications: ascites (Phase Ia: 13/15, 86.7%; Phase Ib: 6/9, 66.7%), esophageal/gastric varices (Phase Ia: 11/15, 73.3%; Phase Ib: 8/9, 88.9%), and hepatic encephalopathy (Phase Ia: 3/15, 20%; Phase Ib: 4/9, 44.4%). The comorbidities included gastrointestinal diseases (15/24, 62.5%), diabetes (4/24, 16.7%), and hypertension (1/24, 4.2%). The baseline characteristics of all patients are presented in Table 1. All patients completed the 28-day follow-up without death, withdrawal, or loss to follow-up.

Table 1 Clinical and demographic featuresSafety

The MSC infusion was well tolerated, with no adverse hemodynamic or respiratory changes observed during or up to 30-min post-infusion. Adverse events (AEs) up to Day 28 are summarized in Table 2. In total, ten AEs were reported in phases Ia and Ib. Four AEs occurred in Phase Ia: one rash (level 2, unlikely to be related) in Cohort II; one case each of rash (level 1, possibly related) and renal stone (level 2, unlikely to be related) in Cohort III; and one case of influenza-like symptoms (level 1, unlikely to be related) in Cohort IV. The rash in Patient 2 of Cohort III was deemed possibly related to MSC infusion, prompting the inclusion of three additional patients in this cohort; none developed a new rash. Six AEs occurred in Phase Ib: one case of diarrhea (level 2, unlikely to be related), one case of bleeding gums (level 2, unlikely to be related), one upper respiratory tract infection (level 3, possibly related) in Cohort A, and two cases of fever (level 1, possibly related) and one case of hemorrhoids (level 2, unlikely to be related) in Cohort B. The three above-reported AEs were judged as possibly related to the MSC infusion due to the close time relationship. Nevertheless, they may also be attributable to DLC or other concomitant drugs and may be unrelated to MSC treatment.

Table 2 Adverse events up to Day 28

Throughout the 28-day observation period, there were no dose-limiting toxicities (DLTs), AEs leading to drug discontinuation, withdrawals due to AEs, serious adverse events (SAEs), or suspected unexpected serious adverse reactions (SUSARs).

Preliminary exploratory endpoints

On day 28, the Child‒Pugh score decreased in 53.3% (n = 8) of the participants in Phase Ia, while increases were observed in 13.3% (n = 2), limited to Cohorts 1 and 2, with no increases recorded in Cohorts 3 and 4. In Phase Ib, the Child‒Pugh score decreased in 88.9% (n = 8) of the participants, with 3 patients in Cohort A and 5 in Cohort B, and no increases were reported in either cohort. Overall, the higher-dose groups were more likely to have lower Child‒Pugh scores across both phases. (Fig. 2)

Fig. 2

Changes in individual Child‒Pugh scores from baseline to Day 28. A child class A; B child class B; C child class C

From baseline to day 28 in Phase Ia, 33.3% (n = 5) of the participants had decreased MELD scores, with reductions in Cohorts I, II, III, and IV by 1, 0, 1, and 3 cases, respectively. In contrast, the MELD scores increased in 2, 2, and 3 cases in Cohorts I, II, and III, while Cohort IV had no increases. In Phase Ib, 44.4% (n = 4) of participants had decreased MELD scores, with decreases of 1 case in Cohort A and 3 cases in Cohort B; no increases were observed in Cohort B (Table 3). Data at additional time points are provided in Supplementary Table 1,2.

Table 3 Clinical outcome indicators at Day 28

To assess the impact of MSC therapy on patients’ nutritional status and liver synthetic function, we analyzed changes in albumin, prealbumin, and cholinesterase levels and in prothrombin activity. Albumin levels improved in 53.3% (n = 8) of Phase Ia participants and 88.9% (n = 8) of Phase Ib participants. Prealbumin levels increased in 53.3% (n = 8) of the Phase Ia participants and 66.7% (n = 6) of the Phase Ib participants. Cholinesterase levels increased in 73.3% (n = 11) of Phase Ia and 77.8% (n = 7) of Phase Ib participants, whereas prothrombin activity improved in 53.3% (n = 8) of Phase Ia and 66.7% (n = 6) of Phase Ib participants.

For quality-of-life assessments, chronic liver disease questionnaire (CLDQ) scores improved in 55.6% (n = 5) of Phase Ia participants and 88.9% (n = 8) of Phase Ib participants, with all Cohort B participants showing improvements. The EuroQol 5-dimension self-report questionnaire (EQ-5D) scores improved in 20% (n = 3) of Phase Ia participants and 55.6% (n = 5) of Phase Ib participants, with Cohort B accounting for 66.7% (n = 4) of these improvements. These results suggest that higher or multiple doses may improve quality of life. (Table 3)

Regarding cirrhosis complications within 28 days after MSC treatment, no cases of liver failure, hepatocellular carcinoma, or death occurred in any cohort.

Dynamic single-cell landscape of PBMCs following MSC treatment

We performed single-cell RNA sequencing (scRNA-seq) on peripheral blood mononuclear cell (PBMC) samples from three patients in Cohorts I, II, and IV at five time points (baseline, Day 3 [D3], Day 7 [D7], Day 14 [D14], and Day 28 [D28], Fig. 3a). Moreover, we included 28 PBMC samples from healthy donors in the public dataset to identify alterations in CAID patients at baseline. After rigorous quality control and batch effect removal, we obtained 467,831 single cells (Fig. 3b). Following dimensionality reduction and annotation on the basis of canonical markers, we identified four main immune cell lineages: T cells, B cells, NK cells, monocytes, and other myeloid immune cells (Fig. 3b, Supplementary Table 3). These lineages were further subdivided at higher resolution. We subsequently calculated the proportion of cell lineage alterations at baseline and their dynamics following MSC treatment. At baseline, the proportions of monocytes and B cells were elevated. In contrast, the proportions of T cells and NK cells were reduced (Fig. 3c). After MSC treatment, the proportion of these cell lineages gradually shifted toward the healthy state compared with the baseline state, with the most notable recovery observed on D7. To validate the cellular dynamics of immune cell lineages under MSC treatment, we also performed cytometry by time-of-flight (CyTOF) staining on PBMC samples from 10 patients (one patient in Cohort I, three patients in Cohort II, and six patients in Cohort III) at three time points (baseline, D3, and D7). The cells were categorized into seven clusters corresponding to the scRNA-seq data (Supplementary Fig. 1a). T cells and NK cells also tended to increase with MSC treatment, while B cells tended to decrease (Supplementary Fig. 1b). Interestingly, monocytes tended to increase on D3 in most patients but decreased on D7. Considering that monocytes are diverse among PBMCs, we hypothesized that various monocyte subpopulations underwent distinct changes during MSC treatment. These results indicate that monocytes had the most significant changes in DLC patients compared with healthy individuals among all the immune cell subsets. Moreover, MSC treatment had a relatively minor effect on the proportions of immune cells other than monocytes. This, to some extent, indicates that changes in monocytes may be the primary mechanism through which MSCs exert their immunomodulatory functions. Furthermore, the impaired proportion and function of monocytes are closely associated with cirrhosis progression.4,26 The important role of monocytes and the varying trends observed in different cohorts prompted us to analyze monocyte subpopulations.

Fig. 3

The dynamic single-cell landscape of DLC patients under MSC treatment. a Schematic diagram of the multiomics analysis of cohorts in the clinical trial and the validation cohort. b Uniform manifold approximation and projection (UMAP) plots of the annotated cell lineages (left) and subpopulations (right) at a relatively high resolution. The cell clusters were annotated via canonical markers. c Boxplot showing the dynamics of the four major immune cell lineages. Lineage proportions were calculated as the percentage of total immune cells. d UMAP plot of the annotated monocyte subpopulations (left) and the expression of CD14 and FCGR3A (right). e Heatmap showing the scaled signature scores of the KEGG signaling pathways across monocyte subpopulations. The signature scores were calculated via the “AddModuleScore” function in Seurat. f Boxplot showing the dynamics of the five monocyte subpopulations. Cell proportions were calculated as the percentage of total monocytes. g Boxplot showing the fold change in cell proportions of c36_Mono_LGALS2 and c38_mono_MX1 at different MSC treatment time points compared with baseline across the three cohorts of DLC patients. h Boxplot showing the dynamics of the CCR+ cMono proportion in CyTOF. i Violin plot showing the signature score of the cytokine production process in c38_mono_MX1. Related genes were derived from the Gene Ontology database. j Heatmap of bulk RNA-seq data showing the expression of MX1+ monocyte-specific genes before and after coculture with MSCs. The Wilcoxon rank-sum test was used to test the difference in cell proportion between groups (p-values highlighted in red) and between posttreatment and baseline in each cohort (p-values highlighted in the corresponding cohort color). ***p < 0.001. **p < 0.01. *p < 0.05

Alterations in myxovirus resistance 1-positive (MX1+) monocytes in response to MSC treatment

To elucidate the changes in monocyte subpopulations during MSC treatment, we analyzed five monocyte subpopulations (Fig. 3d). Among them, versican+ (VCAN+, c35) monocytes, galectin 2+ (LGALS2+, c36) monocytes, and myxovirus resistance 1-positive (MX1+, c38) monocytes expressed more CD14 and fewer Fc gamma receptor IIIa (FCGR3A, CD16), corresponding to classical monocytes. Moreover, MX1+ monocytes highly expressed IFN-stimulated genes (Supplementary Table 4). MHC class II molecules DP beta+ (HLA-DPB+, c37) monocytes expressed both FCGR3A and CD14 (transitional monocytes), whereas FCGR3A+ (c39) monocytes expressed high levels of FCGR3A (CD16) and low levels of CD14, corresponding to nonclassical monocytes. KEGG pathway analysis revealed that the MX1+ monocytes were more immunologically active, with high expression of genes related to the chemokine receptor (CCR), mitogen-activated protein kinase (MAPK), and transforming growth factor-beta (TGF-β) pathways (Fig. 3e). Further analysis of cell proportions revealed that the number of MX1+ monocytes was significantly greater at baseline than it was in healthy controls, whereas the number of LGALS2+ monocytes was notably lower (Fig. 3f). After undergoing MSC treatment, both subpopulations tended toward healthy controls; however, these changes varied across cohorts: MSCs effectively reduced the number of MX1+ monocytes and increased the number of LGALS2+ monocytes, with the most pronounced effects on D7 in Cohort IV. In contrast, MSCs demonstrated moderate efficacy in Cohort II but failed to significantly influence these changes in Cohort I (Fig. 3g). In the CyTOF data, we also identified a subpopulation with high expression of CCR4 and CCR6, corresponding to MX1+ monocytes (Supplementary Fig. 1c). The proportion of this subpopulation also decreased on D3 and D7 (Fig. 3h).

Pseudotime analysis via Monocle3 and validation via Slingshot indicated that VCAN+ monocytes represent the root of differentiation, with FCGR3A+ monocytes serving as the terminal state (Supplementary Fig. 1d). As differentiation progresses, cells increasingly express MHC class II molecules, including HLA-DRA, HLA-DRB1, HLA-DQA1, HLA-DPA1, and HLA-DMA (Supplementary Fig. 1e). At baseline, the FCGR3A+ monocytes exhibited an earlier phenotype compared to the healthy state (Supplementary Fig. 1f). MSC treatment could trigger MSC differentiation toward the terminal phenotype, especially in Cohort IV.

To further explore the functional changes in monocyte subpopulations, we analyzed the potential differentiation process of monocyte subpopulations. Cytokine production decreased in MX1+ monocytes following MSC treatment, and this downward trend was more pronounced in Cohort IV, with the strongest effect observed on D7 (Fig. 3i). Next, we analyzed which cytokines were influenced by MSC treatment. The production of proinflammatory cytokines, interleukin-15 (IL15), and macrophage stimulating 1 (MST1) in MX1+ monocytes, which were elevated in DLC patients at baseline, were decreased by MSC treatment. However, MSCs further increased the expression of tumor necrosis factor (TNF) superfamily members (TNFSF10, TNFSF13, and TNFSF14), which are known to stimulate T-cell activation (Supplementary Fig. 1g).

To validate the function of MX1+ monocytes, we isolated CD14⁺ monocytes from the peripheral blood of an independent cohort of five DLC patients and cocultured them with MSCs, followed by bulk RNA-seq analysis. Using scRNA-seq as a reference, we examined the functional changes in MX1⁺ monocytes before and after coculture. The results revealed that, after coculture with MSCs, the expression of genes related to cytokine production, phagocytosis, and antigen presentation in MX1⁺ monocytes were reduced, which confirms the immunomodulatory effect of MSCs on MX1⁺ monocytes (Fig. 3j). To assess whether MX1⁺ monocytes are associated with clinical outcomes, we categorized patients from Cohorts I, II, and IV into four groups based on changes in their Child‒Pugh scores (i.e., +1/+2, 0, −1, and −2) and analyzed the changes in the proportion of MX1⁺ monocytes (Supplementary Fig. 1h). The results showed that in the “+1/+2” group, the proportion of MX1⁺ monocytes increased to above baseline by D7, while in the “0” group, the proportion increased above baseline by D14. In contrast, patients in the “−1” group showed a decrease below baseline on D3, while the proportions on D7 and D14 were comparable to those at baseline. In contrast, the “−1” group showed little difference between D7 and D14 compared to baseline. More importantly, in the “−2” group, the proportion of MX1⁺ monocytes remained consistently below baseline at all four posttreatment time points. These findings suggest that the proportion of MX1⁺ monocytes is associated with clinical outcomes.

These results suggest that MSC treatment has varying effects on different monocyte subpopulations. The modulatory effect of MSC treatment on the highly immune-active c38 subpopulation is dose-dependent, indicating that MX1+ monocytes may be a key monocyte subpopulation that responds to MSC therapy. Moreover, changes in the cytokine secretion function of MX1+ monocytes suggest that it may influence the function of other immune cells.

MX1+ monocytes mediated the immunomodulatory function of MSCs

The observed variations in cytokine production indicate the immunomodulatory function of MX1+ monocytes. To identify how MX1+ monocytes influence other immune cells, we first analyzed other immune cell subpopulations with relatively higher proportions in DLC patients, which differed from those in healthy individuals, and detailed their changes following MSC treatment. All 223,953T cells were clustered into 17 subpopulations (Fig. 4a, Supplementary Table 5). Both regulatory CD4+ T cells (c5_CD4Treg_FOXP3) and effector CD8+ T cells (c10_CD8Teff_NKG7) were elevated at baseline. Following MSC treatment, CD4Treg exhibited a gradual decrease trend on D3 and D7. For CD8Teff cells, MSC treatment did not directly inhibit the further increase in the proportions of CD8Teff cells. In CyTOF, the corresponding subpopulations of CD4Treg and CD8Teff cells were also identified (Fig. 4b). Most of the patients whose CD4Treg proportions increased on D3 were in the high-dose group. CD8Teff cells remained stable or slightly increased in most patients, which is consistent with the trend observed via scRNA-seq. Among the eight indicated NK cell subpopulations, c22_NK_GZMK was identified as a CD56bright NK cell subset (Supplementary Fig. 2a, Supplementary Table 6). Compared with healthy individuals, this subpopulation presented an increased proportion at baseline. Interestingly, the proportion of this subpopulation varied across different patients: a decreasing trend was observed in Cohort I, whereas an increasing trend was observed in some patients from Cohorts II, III, and IV (Supplementary Fig. 2b). For 9 B-cell subpopulations, two naïve subpopulations (c26_NaiveB_IL4R and c27_NaiveB_TCL1A) and germinal center B cells (c33_gcB_CD38) increased at baseline, whereas three memory subpopulations (c29_MemB_AIM2, c30_MemB_TNFRSF13B, and c31_MemB_FCRL5) decreased (Supplementary Fig. 2c, Supplementary Table 7). Following MSC treatment, the number of naïve B cells remained stable and gradually decreased on D3 or D7 (Supplementary Fig. 2c, d).

Fig. 4

The variation in other immune cells and their interactions with monocytes in DLC patients under MSC treatment. a UMAP plot of the annotated T-cell subpopulations (left) and boxplots showing the dynamics of the major T-cell subpopulations (right). Cell proportions were calculated as the percentage of total T cells. Subpopulations with a maximum median proportion across groups over 5% are displayed. b t-SNE plot of the annotated T-cell subpopulations in CyTOF (left) and boxplot showing the dynamics of the three T-cell subpopulations in CyTOF (right). c Heatmaps showing the strength of ligand‒receptor pairs between c38_Mono_MX1 and four other immune cell subpopulations in four groups: healthy, baseline, D3, and D7. d Pie chart illustrating the correlations among cell subpopulations and indicating the differences between DLC patients at baseline and healthy individuals. The corresponding Spearman correlation coefficients are shown in the top-right corner. e Scheme showing the optimal efficacy timing of MSC treatment, its impact on immune cell subpopulations, and the key regulatory role of MX1+ monocytes. The Wilcoxon rank-sum test was used to test the difference in cell proportion between groups (p-values highlighted in red) and between posttreatment and baseline in each cohort (p-values highlighted in the corresponding cohort color). ***p < 0.001. **p < 0.01. *p < 0.05

We then explored cell–cell communications between altered immune cell subpopulations under MSC treatment. Both the outgoing and incoming signals of each subpopulation were elevated at baseline compared with those of healthy controls and decreased following MSC treatment (Supplementary Fig. 3a). MX1+ monocytes predominantly act as signal emitters, while CD8Teff receives the most signals. Among the signals emitted by MX1+ monocytes, two pathways changed with MSC treatment: galectin was elevated in DLC patients but decreased on D7 in Cohorts II and IV. On the other hand, resistin was reduced in patients but increased on D3 and D7 in Cohort IV (Supplementary Fig. 3b, c). We also analyzed the variation in ligand‒receptor pairs between altered subpopulations (Fig. 4c, Supplementary Fig. 3d). At baseline, MX1+ monocytes promote the transendothelial migration of CD4Treg, CD8Teff, γδ T, and LGALS2+ monocytes via intercellular cell adhesion molecule-1 (ICAM1). Additionally, MX1+ monocytes interacted with other subpopulations through the amyloid protein precursor (APP)-CD74 axis. The inhibitory function of CD4Treg cells was strengthened by MX1+ monocytes via the CD86-cytotoxic T-lymphocyte-associated protein 4 (CTLA4) interaction, while the chemotaxis and phagocytosis of LGALS2+ monocytes were enhanced by Annexin A1 (ANXA1) produced by MX1+ monocytes. Following MSC treatment at D3 and D7, the chemotactic activity of MX1+ monocytes was reduced, whereas their antigen-presentation capacity was enhanced. Furthermore, MX1+ monocytes can modulate CD4Treg function via the IL16-CD4 interaction. To further investigate the immunomodulatory role of MX1+ monocytes, we analyzed their correlations with other altered subpopulations (Fig. 4d). We found that MX1+ monocytes were positively correlated with CD4Treg, γδ T, and naïve B cells but negatively correlated with CD8Teff, LGALS2+ monocytes, and memory B cells. Based on the identified cell proportion changes and interactions among cell subpopulations, the regulatory effects of MSCs on immune cells can be illustrated with a schematic diagram (Fig. 4e).