Comparative analysis on the anti-inflammatory/immune effect of mesenchymal stem cell therapy for the treatment of pulmonary arterial hypertension

Despite the advancement of targeted therapy for pulmonary arterial hypertension (PAH), poor prognosis remains a reality. Mesenchymal stem cells (MSCs) are one of the most clinically feasible alternative treatment options. We compared the treatment effects of adipose tissue (AD)-, bone marrow (BD)-, and umbilical cord blood (UCB)-derived MSCs in the rat monocrotaline-induced pulmonary hypertension (PH) model. The greatest improvement in the right ventricular function was observed in the UCB-MSCs treated group. The UCB-MSCs treated group also exhibited the greatest improvement in terms of the largest decrease in the medial wall thickness, perivascular fibrosis, and vascular cell proliferation, as well as the lowest levels of recruitment of innate and adaptive immune cells and associated inflammatory cytokines. Gene expression profiling of lung tissue confirmed that the UCB-MSCs treated group had the most notably attenuated immune and inflammatory profiles. Network analysis further revealed that the UCB-MSCs group had the greatest therapeutic effect in terms of the normalization of all three classical PAH pathways. The intravenous injection of the UCB-MSCs, compared with those of other MSCs, showed superior therapeutic effects in the PH model for the (1) right ventricular function, (2) vascular remodeling, (3) immune/inflammatory profiles, and (4) classical PAH pathways.

Protective effects of UCB-MSCs on PAH-induced vascular remodeling. Next, the pulmonary arterial medial wall thickness and perivascular fibrosis in rat lungs were compared between the MCT + Saline and the MCT + MSC-treated groups to assess the therapeutic effects of the MSCs. The MCT treatment significantly increased the medial wall thickness and perivascular fibrosis compared with CON group (Fig. 2a,d). Overall, the treatment of all three MSCs significantly reduced the MCT-induced increase in the media wall thickness for MCT + AD, BM, and UCB groups, respectively (Fig. 2a,d). Intriguingly, the MCT + UCB group showed the most significant the reduction compared with the other two MSC groups. Similarly, the increased perivascular fibrosis in the MCT + Saline group was significantly reduced in the MCT + AD, MCT + BM, and MCT + UCB groups, and the MCT + UCB group had more significant attenuation than the MCT + AD and MCT + BM groups (Fig. 2b,e). Moreover, we next compared the efficacy of the three MSCs in attenuating vascular cell proliferation thorough PCNA, proliferation marker (Fig. 2c,f). The increased percentage of PCNA-positive cells in the MCT + Saline group was significantly reduced in the MCT + AD, MCT + BM, and MCT + UCB groups, and the MCT + UCB group showed the greatest effect (Fig. 2f). These data collectively demonstrated that all three MSC treatments showed protective effects on the medial wall thickening, perivascular fibrosis, and vascular cell proliferation in the MCT + Saline group, and that the MCT + UCB had significantly greater effects than the other two MSC types.

Effective engraftment of the injected MSCs in lungs.
To evaluate the effectiveness of the engraftment of the injected MSCs, we next measured the levels of the human stem cell markers (CD44, CD90, CD29, human nuclear antigen (HNA), and human Arthrobacter luteus (Alu)) within the MCT-treated lungs harvested at days 1, 3, 5, 7, and 14 post-MSC injection. Their mRNA levels can reflect the amounts of the engrafted MSCs at days post-MSC injection. The mRNA levels of the three markers gradually decreased over time (sFig. 2). Intriguingly, significant levels of the three markers were stably detected up to day 7, suggesting that significant portions of all three MSCs were stably engrafted in the lungs. Among the three MSCs, the MCT + UCB showed the highest mRNA levels of the three markers, especially at days 3 and/or 5 post-MSC injection, suggesting the most effective engraftment of the UCB-MSCs.
Inhibitory effects of UCB-MSCs on innate and adaptive immunity. We next compared the level of infiltrated inflammatory cells within the MCT-treated lungs harvested on day 14 post-MSC injection. The immunostainings and mRNA levels of the representative markers CD80 and CD206 for the pro-inflammatory macrophage (MΦ) subtype (M1) and pro-fibrotic MΦ subtype (M2), respectively, were significantly increased in the MCT + Saline group. The increase was reduced in all MSC-treated groups, and such reduction was relatively stronger in the MCT + UCB group (Fig. 3a,b). Moreover, we analyzed immunostaining and mRNA levels (c-f) The tricuspid regurgitation maximal pressure gradient (TR max PG), pulmonary velocity acceleration time (PVAT), tricuspid annular plane systolic excursion (TAPSE), and RV fractional area contraction (RV FAC) of each treatment group are shown. Each group was examined in n = 7 mice. Data are shown as the mean ± SD. Statistical significance is marked with 3 symbols. * for the comparison with control; $ for the comparison with the MCT + Saline group; # for the comparison with the MCT + UCB group; *, $, # P < 0.05; **, $$, ## P < 0.01; ***, $$$, ### P < 0.001 from Mann-Whitney test; CON: control ; MCT: monocrotaline; MSC: mesenchymal stem cell; RV: right ventricule; LV: left ventricle; AD: adipose tissue-derived; BM: bone-marrow-derived; UCB: umbilical cord bloodderived; TR max PG: tricuspid regurgitation max pressure gradient; PVAT: pulmonary velocity acceleration time; TAPSE: tricuspid annular plane systolic excursion; RV FAC: right ventricular fractional area contraction. www.nature.com/scientificreports/ of the cytokines that represent the activated M1 (Tnf-α) and M2 (Tgf-β) (Fig. 3c,d). The mRNA levels of these cytokines were increased in the MCT + Saline group. Similarly, the increase was significantly reduced in all the MSC-treated groups, and the reduction was relatively stronger in the MCT + UCB group (Fig. 3c,d). Furthermore, we examined whether the attenuated innate immunity further modulates adaptive immunity by measuring the mRNA levels of Cd8 and Cd20, T and B cell markers, respectively. Immunostainings and mRNA levels of these markers were significantly increased in the MCT + Saline group, although MCT + UCB group had the largest reversing effects (Fig. 3e,f). Similar immunostaining and mRNA expression patterns of Il-10 and Il-8 that represent T and B cell activations, respectively, were observed in the MCT + Saline and MSC-treated groups (Fig. 3g,h). Collectively, these data suggest that UCB-MSCs more strongly attenuate innate and adaptive immunity associated with inflammation and fibrosis in the MCT + Saline model. . Among G1-8, we focused on the four groups (G1-4) whose expression levels were differently changed by MCT + Saline and the MSCs, which can explain the effects of the MSCs in the MCT + Saline model (Fig. 4b). The largest number of DEGs were identified from MCT + UCB/ www.nature.com/scientificreports/ MCT + Saline (Fig. 4a), and G1-4 included a larger number of DEGs from MCT + UCB/MCT + Saline compared with those from MCT + AD/MCT + Saline and MCT + BM/MCT + Saline, suggesting that UCB-MSCs most strongly affected gene expression in the MCT + Saline model (Fig. 4b). Next, we identified cellular processes associated with G1-4 by performing enrichment analysis of GOBPs using the DAVID software 21 (Fig. 4c). G1-4 were strongly associated with PAH-related cellular processes, such as (1) inflammation (inflammatory response, Mɸ cytokine production, and wound healing); (2) vascular remodeling (cell adhesion, extracellular matrix organization, cell proliferation, and blood vessel remodeling); (3) angiogenesis (response to hypoxia and blood vessel development); (4) apoptosis (apoptotic signaling pathway); and/or (5) cytoskeleton organization (cilium organization, microtubule-based movement, and regulation of actin cytoskeleton organization. Collectively, these data suggest that UCB-MSCs most greatly modulate PAH-related process including inflammation and vascular remodeling (Fig. 4c). . The enrichment analysis of the GOBPs revealed that the 108 up-regulated genes were mainly involved in cellular processes related to inflammatory response (defense, inflammatory, and immune responses, and cytokine production) (Fig. 5a). The 88 down-regulated genes were mainly involved in the processes related to immune response (defense response, and responses to cytokine and virus), cell development and migration, and extracellular matrix organization (Fig. 5b). Together with the findings in Fig. 3, these data suggest that the inflammation and immune response-related processes may play critical roles in the inhibitory effects of UCB-MSCs on PAH phenotypes. Next, we thus selected 27 genes related to inflammation (Fig. 5c) and 15 genes related to immune and defense responses (Fig. 5d) and found that they included both upregulated and downregulated genes by MSCs. To sort out cellular processes upregulated and downregulated predominantly by UCB-MSCs in the MCT-treated condition, we built network models describing interactions among these upregulated (Fig. 5e) or downregulated (sFig. 4b) genes involved in the inflammatory and immune responses. The network model for the upregulated genes ( Fig. 5e) showed that UCB-MSCs predominantly downregulated pro-inflammatory pathways activated by MCT, including damage-associated molecular patterns, such as heat shock protein A member 1b (Hspa1b), its receptor [toll-like receptor 5 (Tlr5)), and other modulators and their cofactors (Arg2, Chia, Spp1, and Gp2), as well as resolution of inflammation (Bpifa1/b). The UCB-MSCs also negatively regulated of the MCT-induced T and B cell activation (Tp73, and Foxj1). These data suggest that UCB-MSCs, compared with AD-and BM-MSCs, more strongly decrease the inflammatory responses induced by MCT + Saline in general, including early inflammation and late resolution of inflammation, as well as activation of adaptive immune responses (T and B cell activations). On the other hand, the network model for the downregulated genes (sFig. 4b) showed that UCB-MSCs predominantly upregulated the pathways induced by pathogen infections, which were downregulated by MCT, including complement cascade (Fcnb and C4bpa), NK cell activation (Klrb1a/c, Spon2, and Slit2), cytotoxic T cell-mediated cell killing pathways-MHC1 class 1 presentation (RT1-A/A1, RT1-CE2/3/4/12/14/15/16, RT1-S2/3, and Tap1/2), and killing of target cells (Gzma/b/c/k/m, and Prf1). These data suggest that the injected MSCs are killed as foreign cells by these pathways, and the predominant activation of these pathways in the MCT + UCB group reflects a prolonged survival of UCB-MSCs, compared with AD-and BM-MSCs, consistent with the findings illustrated in sFig. 2. www.nature.com/scientificreports/ Finally, we confirmed stronger upregulation and downregulation of the representative genes involved in inflammatory and immune responses (Arg2, Hspa1b, Spp1, Gp2, Bpifb1, Chia, Tp73, Foxj1, and Cfi; Fig. 5F) and killing of target cells (Fcnb, Spon2, RT1-A, RT1-CE4, and RT1-CE12; sFig. 4c) in the MCT + UCB group, compared with in the MCT + AD and BM groups. Taken together, all these data suggest that UCB-MSCs, compared with AD-and BM-MSCs, more strongly suppress the progression of inflammation induced by MCT and thereby the resolution of inflammation as well.

Inhibitory effects of UCB-MSCs on PAH progression through the PAH and inflammation-related pathways.
To understand the effects of the MSCs on the classical PAH pathways (prostacyclin, endothelin, and NO pathways), we reconstructed a network model describing how the MSCs modulated the MCT-induced expression changes of the components in the PAH pathways. The network model showed down-regulation of prostacyclin and NO pathways (Gucy1a2/a3/b3 in NO pathway and Gnas and Adcy2/9 in prostacyclin pathway) by MCT, which promote vasodilation (Fig. 6a, upper part). In contrast, the network model showed up-regulation of endothelin pathway (Ereg and Agt) by MCT, which promotes vasoconstriction, consistent with the previous findings (Fig. 6a, middle part) 22 . Intriguingly, these alterations in the three pathways were restored by the MSCs (Gucy1a3 in NO pathway; Prkcb, Ptgir/s, Adora2b, and Adcy2/5 in prostacyclin pathway; and Agt and Pla2g2a in endothelin pathway). Moreover, the network model showed the effects of the MSCs on Ca 2+ signaling pathway that closely interacts with the above PAH pathways (Fig. 6a, right part). Ca 2+ transporters were altered (downregulation of Atp2a3/2b1/2b3 and up-regulation of Cacna1h) by MCT to increase the cytosolic Ca 2+ concentration and activate Ca 2+ signaling (Calml3/4) and motor protein complexes (Dnas and Kifs), which promote vasoconstriction. These alterations were also restored by the MSCs (Cacna1h and Calml3/4). Furthermore, the network model showed up-regulation of inflammation-related pathways, cytokine (Ccl21, Grem1, and Gdf15) and coagulation factor (Kng2 and F2/5) pathways by MCT, which can promote vasoconstriction (Fig. 6a, left). These alterations were also restored by the MSCs. These data suggest that the MSCs can contribute to the restoration of the MCT-induced changes in the PAH and inflammation pathways. However, the comparison of the numbers of the components in the network model reversed by the individual MSCs revealed that UCB-MSCs (38 components) had the largest number of the components affected by MCT compared with AD-(28 components) and BM-MSCs (23 components). Next, we validated these MSC-induced recovery patterns caused by MCT for the following representative components in the PAH or inflammation pathways by qRT-PCR analysis: (1) Ptgis and Adcy2 in the prostacyclin pathway; (2) Gucy1a3 in the NO pathway; (3) Agt in the endothelin pathway; (4) Calml4 in the Ca 2+ signaling pathway; and (5) Ccl21, Gdf15, Kng2, F2, and F5 in inflammation pathways (Fig. 6b). Overall, all these data suggest that MSCs reverse the MCT-induced changes in the PAH and inflammation pathways. The UCB-MSCs, among the MSCs, can serve as the most potent MSC with such antagonizing effects.

Discussion
Despite the recent advent of the use of stem cell paracrine factors, such as exosomes and small molecules, as a treatment option in cardiovascular medicine 23,24 , MSC cell therapy remains a potent and valuable biological treatment option. Although cell-free exosome treatment has been shown to replicate the effects of stem cell therapy to a certain extent, the heterogeneity of the exosomes, low reproducibility, and short half-life bring into question whether the treatment effects could be consistently sustained [25][26][27] . Doubts have also been raised about whether small molecules derived from stem cells, such as microRNAs, could outperform the pleotropic effects of stem cells. Embryonic stem cells can give rise to virtually all cell lineages, although they are predisposed to teratoma formation and are difficult to obtain without encountering ethical issues 28,29 . The discovery of induced pluripotent stem cells has resolved such ethical issues 29 , but the genetic instability observed under in vitro culture conditions, expensive media costs, propensity for teratoma formation, and relatively long period required for cell reprogramming have been obstacles for their immediate clinical application [30][31][32] . MSCs are free from ethical issues 29 , easy to obtain from human tissues 10,13 , immune tolerant between species or individuals 11,12,14 , stable under various types and lengths of in vitro culture, and able to be rapidly expanded in an economic fashion 10,13 , which makes them an attractive treatment option.
The most widely investigated MSCs are the AD-, BM-, and UCB-MSCs. All three MSC types have shown promising treatment effects in animal PH models, although a comparison between the three has not been made to date 16,19,33,34 . We therefore systematically performed a comparative analyses of PAH phenotypes (inflammatory profiles, histological features of vascular remodeling, and echocardiographic parameters of the RV), engraftment features, genome-wide mRNA expression profiles, and the associations with the classical PAH cellular pathways. The MSCs significantly restored the PAH phenotypes ( Fig. 1-3), modulated PAH-related cellular processes (Fig. 4), and interacted with the classical PAH pathways (NO, prostacyclin, and endothelin) and inflammationrelated pathways (Fig. 6). Interestingly, the tissue transcriptome revealed that a substantial portion of the genes that showed significantly different expression compared with the MCT group were genes involved in the immune response and inflammation (Fig. 4). The attenuation of the recruitment of total alveolar macrophages and the M1 and M2 subsets (Fig. 3a) are associated with the decrease of both pro-and anti-inflammatory responses (Fig. 3b) 35 . The subsequent decrease of adaptive immunity was substantiated by the diminished recruitment of T and B cells (Fig. 3c) and the levels of cytokines (Fig. 3d). These features support the immunomodulatory role of MSCs in PAH.
Of the 3 types of MSCs, the UCB-MSCs showed the most promising results for the RV pressure overload and function (Fig. 1), histological properties (Fig. 2), engraftment (sFig. 1), immune/inflammation modulation (Figs. 3, 5), and classical PAH pathways (Fig. 6). The UCB-MSCs had the most potent effects in attenuating the inflammatory and innate immune responses, and in subsequently decreasing the expression of T and B www.nature.com/scientificreports/ cell markers, which may be the result of a reduction in the recruitment of lymphocytes due to the decreased activation of MΦ. Moreover, according to the network model, the UCB-MSCs had the most significant effect in restoring the PAH and inflammation-related pathways (Fig. 6). These controls of the immune responses by the UCB-MSCs may have led to the better engraftment profiles in MCT + UCB group, for which the engrafted cell survival was the highest. Interestingly, the potent immunomodulatory effects of the UCB-MSCs could be translated to improved RV pressure overload and function as shown in Fig. 1c-f. The exact mechanisms behind the anti-inflammatory effects of MSCs and improved heart function are not clearly understood. The reduced pro-inflammatory factors predominantly by UCB-MSCs may act as a potential link of the inflammation to the alleviation of the RV pressure and volume overload. Gene expression profiling and network analysis showed that the modulation of pro-inflammatory factors (Arg2, Hspa1b, Tlr5, Spp1, and Gp2) genes may help delineate the attenuated immune response and anti-inflammatory phenotypes that we observed in the MCT + UCB group. In particular, Tlr5 has been shown to be expressed in the intestinal epithelia and in endothelial cells that recognize bacterial flagellin and further to mediate the innate immune responses 36 . Recently, the involvement of Tlr5 in inflammatory diseases such as rheumatoid arthritis was also discovered 37,38 . Although the role of Tlr5 in PAH has not been elucidated, our data show that the expression level of Tlr5 is increased in the MCT + Saline group, and such increase is normalized by the MSC treatment, with the most prominent effect observed in the MCT + UCB group. This may suggest that the Tlr5-mediated innate immune response may play a role in the pathogenesis of PAH and that the down-regulation of Tlr5 may contribute to an attenuation of the development of PAH. It has been also suggested that Tlr5 may be a Tnf-α responsive gene 37 , indicating that the attenuated Tlr5 levels may reflect the low Tnf-α levels in the MCT + UCB group. Future studies are warranted to validate the relationship between Tlr5 and Tnf-α.
G2 showing no change by MCT, but upregulation by MSCs in the MCT-treated condition was associated with PAH-related processes (immune and inflammatory responses, cytokine production, cell proliferation, blood vessel development, and response to hypoxia), suggesting that MSCs induce MCT-independent upregulation of these processes. This observation seems inconsistent with the results presented in Fig. 3. Unlike the MCTindependent upregulation of PAH-related processes in G2, however, the results in Fig. 3 represent restoration of the MCT-dependent upregulation of PAH-related processes by MSCs. These PAH-related processes were represented by the genes in G1 (upregulated by MCT and downregulated by MSCs), G2, and/or G4 (downregulated by MCT and upregulated by MSCs). Network analysis showed that G1 was mainly associated with inflammatory responses induced by MCT, including early inflammation, late resolution of inflammation, and activation of adaptive immune responses, whereas G4 was associated with cytotoxic immune response related to killing of the injected MSCs. Interestingly, G2 was predominantly associated with killing of the injected MSCs (e.g., macrophage activation leading to activation of NK and cytotoxic T cells, as well as neutrophil and B cells; sFig. 4), which is closely linked to the network model for G4 (sFig. 4b). Moreover, the genes involved in blood vessel development in G2 were not closely associated with a central VEGF signaling in angiogenesis, suggesting that they are the indirect angiogenic factors related to immune responses. Also, the genes involved in blood vessel development (45.1%, 23 of 51 genes) and cell proliferation (42.0%, 55 of 131 genes) in G2 overlapped with those involved in immune and inflammatory responses, suggesting that the above conclusion (i.e., killing of the injected MSCs) for the immune and inflammatory responses may apply to these processes.
Previous studies have shown that MSCs derived from human embryonic stem cells (hESCs) 39 or inducedpluripotent stem cells (iPSCs) 40,41 show potent therapeutic effect. The MSCs, however, have advantages over its other two counterparts for several reasons, as the clinical availability of hESCs and iPSCs can be a major setback. The hESCs can give rise to virtually all cell lineages, although they predispose to teratoma formation and are difficult to obtain without encountering ethical issues 28,29 . The discovery of the iPSCs has resolved such ethical issues 29 , but genetic instability under in vitro culture conditions, expensive media costs, propensity for teratoma formation, and a relatively long period required for cell reprogramming have been obstacles for immediate clinical application [30][31][32] . MSCs are free from ethical issues 29 , easy to obtain from human tissues 10,13 , immune tolerant over species or individuals 11,12,14 , are stable under various and prolonged in vitro cultures conditions, and can be rapidly expanded economically 10,13 , which makes it an attractive treatment option. GOBPs represented by the upregulated (a) and downregulated (b) genes predominantly by UCB-MSCs. Enrichment significance (p) for each GOBP is displayed in − log 10 (p). (c,d) Heat maps showing the differential expression of the genes involved in the inflammatory (c) and immune responses (d). The color bar represents the gradient of the log 2 -fold-changes in the comparisons. The heat map was generated using MATLAB (imagesc.m in R2019a; www.mathw orks.com/). (e) A network model describing interactions among the predominantly upregulated genes by UCB-MSCs. Node center and boundary colors represent upregulation (red) and downregulation (green) by MCT and UCB-MSCs, respectively. Solid and dotted arrows (or inhibition symbols) denote direct and indirect activations (or inhibition), respectively. The color bar represents the gradient of log 2 -fold-changes in the corresponding comparison (MCT + Saline/CON or MCT + UCB/ MCT + Saline). The network model was generated using Cytoscape software (v.3.3.0; www.cytos cape.org/). (f) Confirmation of the predominant upregulation of the indicated three representative genes involved in the inflammatory and immune responses by qRT-PCR. The expression levels were normalized with respect to those in the control group. The normalized data are expressed as the mean ± SD. Statistical significance is marked with 3 symbols. * for the comparison with the control; $ for the comparison with the MCT + Saline group; # for the comparison with the MCT + UCB group; *, $, #P < 0.05; **, $$, ##P < 0.01; ***, $$$, ###P < 0.001 by Mann-Whitney test; GOBP: gene ontology biological processes; MCT: monocrotaline; CON: control; AD: adipose tissue-derived; BM: bone-marrow-derived; UCB: umbilical cord blood-derived. www.nature.com/scientificreports/ The data from our study were obtained from the MCT-+ Saline PH model, which may not entirely reflect the true patholobiology of human PAH. The lack of validation of our findings in other robust models, such as the hypoxic model or the Sugen-hypoxia model, may undermine the validity of our study. Further studies extending our findings in the hypoxic and sugen-hypoxia model are warranted. Additionally, we were not able to confirm our echocardiographic estimates of RV systolic pressure (RVsP) through direct right heart catheterization (RHC). Despite its conveniences in estimating RVsP or mean pulmonary artery pressure, the diagnostic accuracy of echocardiography has been a concerning issue 42 , leading to guidelines recommending right heart catheterization (RHC) as a gold standard for confirming the diagnosis of pulmonary hypertension 43 . Experimental PH models have also confirmed their findings using RHC 44 . Accordingly, we acknowledge that the lack of validation by direct pressure measurement is a major limitation of our study. Our data in this sense should be reviewed with caution.
In this study, we focused on the comparison of the effects of three different representative MSCs on PAH phenotypes and the restoration of the MSC-perturbed gene expression profiles, and found that UCB-MSCs, among the MSCs tested, had the most potent protective effects on PAH phenotypes, as well as the association of the UCB-MSC-induced gene expression changes (PAH and/or inflammation-related pathways) with the improved PAH phenotypes. Although it can be considered valuable as an initial comparative study, further mechanistic studies are warranted to elucidate the functional link of the attenuated inflammation to the improved PAH phenotypes.
In conclusion, the present comparative, systemic analysis of AD-, BM-and UCB-MSC cell therapy in the rat MCT model demonstrated that although all 3 MSC types had therapeutic benefits, the UCB-MSC treatment showed the most promising features in terms of the (1) RV function, (2) histological features, (3) cell engraftment, (4) classical PAH pathways, and (5) immune/inflammatory responses. Due to the lack of direct hemodynamic pressure measurement, however, the current data needs to be interpreted with caution.

Methods
This study was identified the institutional and/or licensing committee approving the experiments, including any relevant details and confirmed that all experiments were performed in accordance with relevant guidelines (ARRIVE guidelines) and regulations.
Animal model of PH. Eight-week-old Sprague Dawley male rats (250-300 g) used in this study were in a pathogen-free and climate-controlled facility with a 12-h light/dark cycle and were fed ad libitum throughout the experiment. The study was approved by the Center of Animal Care and Use of Gachon University (AAALAC; approval number LCDI-2013-0015). The PAH rat model was induced by a single subcutaneous injection (60 mg/ kg) of MCT (cat. C2401; Sigma-Aldrich, USA) 39,40 . The control (CON) and MCT + Saline group (n = 7) was injected with saline. AD and UCB-derived MSCs were purchased from Medipost (Seongnam, Korea) and BMderived MSCs were purchased from Cefobio (Seoul, Korea). At two weeks post-MCT injection, 1.0 × 10 6 AD-(n = 7), BM-(n = 7), and UCB-(n = 7) MSCs were administered by an IV tail vein injection. Echocardiography, tissue harvesting for immunohistochemistry, histology and microarray were performed at the fourth week post-MCT injection, as shown in Fig. 1a. The survival of engrafted MSCs were assessed by measuring human MSC markers, such as Cd44, Cd90, Cd29, HNA, and Alu at day 1, 3, 5, 7, and 14 after MSC injection (sFig. 2).
Echocardiography. After rats were anesthetized with isoflurane, transthoracic echocardiography was performed using a 12 MHz probe (GE Healthcare) and a Vivid Q (GE Healthcare, Israel). The assessed transthoracic echocardiographic parameters included the maximum TR max PG, TAPSE, PVAT, and RV FAC 45 . TR max PG and TAPSE was measured in the apical 4 chamber view, whereas the PVAT and RV FAC was evaluated in the short-axis view 46 . Echocardiographic parameters were measured following the recommendations of the American Society of Echocardiography. The pulmonary artery systolic pressure was calculated as 4V 2 + right atrial pressure, where V was the maximal velocity of the tricuspid regurgitated jet 39 . The TR max PG and PVAT were measured to estimate RV pressure overload, whereas TAPSE and RV FAC were were examined as indicators of RV functional. The sample sizes of each treatment arm were n = 7 for CON, n = 7 for the MCT + Saline, n = 7 for MCT + AD, n = 7 for MCT + BM, and n = 7 for MCT + UCB.
Paraffin-embedded tissue section processing. The collected lung tissues are washed with cold PBS and stored in 4% paraformaldehyde at 4 °C for 12 h. The fixed tissue is washed for 12 h for embedding and a paraffin block was processed using a tissue processor. The paraffin blocks were cut to 7 μm using a Microtome and dried at 40 °C for 24 h. The paraffin blocks were subjected to xylene and four concentrations of ethanol (100%, 95%, 80%, and 70%) for dyeing preparation.
Hematoxylin eosin staining. Paraffin-embedded lung tissues were sectioned with a thickness of 7 μm and used for histological analyses. To measure the pulmonary medial wall thickness, sections were stained with Mayer's Hematoxylin & Eosin. Stained sections were imaged using an Axio Imager Z1 upright microscopy system (Carl Zeiss, Oberkochen, Germany). Of the pulmonary arteries (< 120 µm in diameter), 20 were randomly selected in each animal (n = 3 per group), and the medial wall thickness was calculated as previously reported 47 : (medial thickness × 2)/external diameter × 100.
Masson's trichrome staining. For the assessment of perivascular fibrosis, sections were stained with Masson's Trichrome ((MT, cat. HT15; Sigma-Aldrich, St. USA)). Perivascular fibrosis was determined by the area of the MT-stained fibrotic area divided by the short vessel diameter using ImageJ software (NIH, Bethesda, USA). Quantitative real-time PCR. Total RNA and cDNA were prepared from rat lungs as described above.
cDNA was used for quantitative real-time PCR (qRT-PCR) using SYBR green (cat. RR820A; Takara, Japan). All primers were purchased from Cosmogentech (Seoul, Korea). Relative mRNA levels of the genes were normalized to β-actin (Atcb), and the relative expression differences were obtained using the 2 -ddCt method 49 .
Microarray experiments. Gene expression profiles were generated for lung tissues obtained from CON and MCT + Saline rats, as well as MCT + Saline rats injected with AD-, BM-, and UCB-MSCs using an Agilent-028279 Rat 8 × 60 k chip, which includes 62,976 probes corresponding to 18,584 annotated genes (Agilent, Santa Clara, CA, USA), according to the manufacturer's instructions 50 . In each condition, we performed gene expression profiling of two biological replicates obtained from independent rats. Total RNA was extracted from rat lung using Trizol reagent (Invitrogen Life Technologies, Grand Island, NY). The integrity of the total RNA was analyzed using an Agilent 2100 Bioanalyzer. The RNA integrity values for all samples were larger than 8.8. Reverse transcription was then performed to generate cRNA, which was amplified and hybridized onto each array according to the manufacturer's instructions. The array was scanned using a SureScan Microarray Scanner (Agilent).

Analysis of gene expression profiles.
Probe intensities from the arrays were converted to log 2 -intensities and normalized using the quantile normalization method 51,52 . Differentially expressed genes (DEGs) were identified using an integrative statistical method as previously described 53,54 . Briefly, for each gene, we calculated a T-statistic value using a Student's t-test and a log 2 -median-ratio test for each comparison. Empirical distributions of the T-statistic values and log 2 -median-ratios for the null hypothesis that the genes were not differentially expressed were estimated by performing random permutations of the samples and by then applying the Gaussian kernel density estimation method to the T-statistic values and log 2 -median-ratios that resulted from the random permutations. Using the empirical distributions, we computed the adjusted P values for the t-test and log 2 -median-ratio test for each gene and then combined these P values with Stouffer's method 54 . False discovery rates (FDRs) of each gene for the combined P values were then computed with the Storey method 53 . Finally, the DEGs were defined as the genes with an FDR ≤ 0.05 and an absolute log 2 -median-ratio ≥ 0.58 (1.5-fold change).
Heat map generation and GOBP enrichment analysis. The DEGs were categorized into 8 groups  (G1-4), the DEGs were then sorted according to their up-regulation, no change, and down-regulation patterns in the order of the four comparisons. The heat map for the DEGs in G1-4 was finally generated using their log2-fold-changes in the four comparisons with "imagesc.m" in MATLAB (R2019a; www.mathw orks.com/). Additionally, enrichment analysis of the gene ontology biological processes (GOBPs) for a set of the DEGs was performed using the DAVID software 21 . The GOBPs represented by the DEGs were defined as those with an enrichment P value < 0.05 calculated by DAVID, where P is the significance of the GOBPs being represented by the DEGs in G1-4. The enrichment P values were converted into − log10(P), and the heat map for the enrichment significance of GOBPs was generated using their − log10(P) values with "imagesc.m" in MATLAB (R2019a; www.mathw orks.com/).

Network analysis.
To reconstruct a network model, we first selected a subset of the DEGs involved in the traditional pathways related to PAH progression (the endothelin pathway, NO pathway, and prostacyclin pathway). We then collected the protein-protein interactions (PPIs) among the selected DEGs from five protein-protein interactome databases: the biomolecular interaction network database 22