Reduced BMPR2 expression induces GM-CSF translation and macrophage recruitment in humans and mice to exacerbate pulmonary hypertension

In human PAECs derived from central PAs, we established, by ELISA, that TNF induced a dose-dependent increase in GM-CSF production, beginning as early as 6 h after stimulation. Pretreatment with a high dose (100 ng/ml) of BMP-2 had a modest but significant inhibitory effect on TNF-induced GM-CSF production in both PAECs and PASMCs. However, when BMPR2 mRNA and protein were reduced by RNAi to <50% of the control siRNA value in PAECs (Fig. 1, A and B), TNF mediated a similar increase in GM-CSF mRNA (Fig. 1 C), but GM-CSF protein levels were enhanced by up to twofold (Fig. 1 D). This observation was confirmed with two different control (scrambled) and two different GM-CSF siRNAs (see Materials and methods). This suggested that reduced BMPR2 might be enhancing TNF-mediated GM-CSF mRNA translation. Similar findings were evident in PASMCs (not depicted). To determine whether this could be a global effect on TNF-mediated cytokine production, we also investigated changes induced in IL-6, IL-8, and MCP-1 (Fig. 1, E–J). Loss of BMPR2 also resulted in an increase in TNF-mediated p–IL-6 and IL-8 protein secretion without a comparable increase in mRNA levels. However, the TNF-mediated increase in MCP-1 mRNA and protein was unaffected by loss of BMPR2.

To assess translation of GM-CSF mRNA, we performed polysome analysis by sucrose gradient centrifugation. Effective separation of polysomes was confirmed by using Western immunoblotting to detect ribosomal subunit proteins S6 and L13a in the fractions (Fig. 2, A and B). Based on polysome profile and Western immunoblot, we divided the polysome fractions into three groups: untranslated (unbound or monosomes, fractions 1–7), moderately translated (light polysomes, fractions 8–11), and actively translated (heavy polysomes, fractions 12–15). In cells transfected with control siRNA and stimulated with TNF, 60% of the GM-CSF mRNA transcript was found in the moderately translated pool, whereas in cells with reduced BMPR2, 60% of the GM-CSF mRNA was distributed in the actively translated fraction (Fig. 2 C). Reduced BMPR2 did not affect the distribution of a housekeeping gene, GAPDH (Fig. 2 D). These data provided further evidence that impaired BMPR2 function enhances TNF-induced GM-CSF production by increasing mRNA translation. Loss of BMPR2 did not affect the stability of GM-CSF mRNA, as assessed by measuring GM-CSF mRNA levels after transcriptional inhibition by 5,6-dichloro-1-β-d-ribofuranosyl-benzimidazole (DRB; Fig. 2 E).

To dissect the mechanism by which reduced BMPR2 enhances GM-CSF mRNA translation, we next assessed p38 and MK2 activity, members of a signaling pathway previously shown to be enhanced with loss of BMPR2 (Hagen et al., 2007) and to influence the transcriptional and posttranscriptional regulation of cytokine production (Hitti et al., 2006; Anderson and Kedersha, 2008). In PAECs transfected with control siRNA, TNF induced p38 phosphorylation (p-p38) at 10 min, but the response was transient, with return to baseline levels at 30 and 60 min. However, in PAECs transfected with BMPR2 siRNA, p-p38 was sustained 30 and 60 min after TNF treatment (Fig. 3, A and B). Consistent with these results, phosphorylation of MK2 was significantly prolonged in cells transfected with BMPR2 siRNA versus control siRNA at 30 and 60 min after TNF treatment (Fig. 3, A and C). TNF-mediated activation of the transcription factor NF-κB, assessed by I-κB phosphorylation, was similar in cells transfected with BMPR2 siRNA and control siRNA (Fig. 3, A and D). When p38 and MK2 activation were inhibited using SB202190, the increase in TNF-mediated GM-CSF production in PAECs transfected with BMPR2 siRNA was reduced (Fig. 3, E and F). These results are consistent with the requirement of sustained activation of p38 mitogen-activated protein kinase (MAPK) and MK2 to enhance GM-CSF mRNA translation when there is reduced expression of BMPR2.

We next tested the hypothesis that the enhanced TNF-mediated GM-CSF mRNA translation in response to reduced BMPR2 is related to impaired assembly of stress granules. We first assessed the phosphorylation of eukaryotic translation initiation factor (eIF2α), a requisite modification for stress granule assembly, and observed that TNF-mediated eIF2α phosphorylation was reversed when BMPR2 levels were reduced (Fig. 4 A). The decrease in p-eIF2α associated with reduced levels of BMPR2 was p-p38 dependent, as it was lost when p-p38 was inhibited by SB202190 (Fig. 4 B). We then investigated whether p38 activation enhanced phosphatase activity, necessary to dephosphorylate eIF2α and promote mRNA translation. The GADD34–PP1 complex dephosphorylates eIF2α, but we found no change in the levels of GADD34 or PP1 (Fig. 4 C) or in the amount of complex formed (Fig. 4 D) in response to reduced BMPR2 and prolonged p-p38. However, salubrinal, a chemical inhibitor of GADD34–PP1 activity, reversed both the dephosphorylation of eIF2α (Fig. 4 E) and the enhanced GM-CSF production (Fig. 4 F) associated with reduced BMPR2. Decreasing PP1 by siRNA (Fig. 4 G) also reversed the enhanced GM-CSF production observed in TNF-stimulated PAECs with reduced levels of BMPR2 (Fig. 4 H). Collectively, our experiments support a model in which loss of BMPR2 prolongs p-p38 activity required for GADD34–PP1 activity, and this dephosphorylates eIF2α and impairs assembly of stress granules facilitating GM-CSF mRNA translation.

To further assess stress granule assembly in PAECs with and without loss of function of BMPR2 by immunofluorescence microscopy, we first costained the RNA binding proteins TIA-1 and Hu-related protein R (HuR), associated with stress granules under conditions of TNF stimulation. Arsenite was used as a positive control for stress granules (Fig. 5, A–H). Additional experiments were performed using G3BP as a marker for stress granules. 3 h after the addition of TNF or exposure to hypoxia (1% O₂; Brown et al., 2011), abundant stress granules appeared (Fig. 5, I–P). In all conditions studied (TNF, arsenite, and hypoxia), loss of BMPR2 significantly reduced stress granule formation.

To establish the clinical significance of our findings in patients with IPAH, Western immunoblot for GM-CSF was performed on whole lung tissue from 10 IPAH patients and 8 control unused donor lungs (Tables 1 and 2 provide clinical and demographic information on the population studied). Elevated levels of GM-CSF were observed in the IPAH versus control samples (Fig. 6, A and B). We then localized the heightened expression of GM-CSF protein using immunohistochemistry. GM-CSF protein was not readily detected in the PAs of unused donor control lungs (Fig. 6 C), but there was moderate to intense GM-CSF immunoreactivity in the endothelium, thickened intima, and hypertrophied media of the PAs of the IPAH patients (Fig. 6, D–F). GM-CSF co-distributed with increased immunoreactivity for TNF in IPAH PAs (Fig. 6, D–F). There was also minimal if any detection of TNF protein in control PAs (Fig. 6 C). Semiquantitative analysis for GM-CSF immunoreactivity (not depicted) indicated statistically significant differences in GM-CSF and TNF expression in IPAH versus control vessels of the same size and location.

As GM-CSF can recruit inflammatory and progenitor cells to the vessel wall, we assessed the distribution and nature of GM-CSFRα–expressing cells in the PAs of IPAH patients by immunohistochemistry and by microfluidic-based single cell transcriptional analysis. In the donor control lungs, GM-CSFRα immunoreactivity was not detectable in the PAs but was seen in alveolar macrophages (Fig. 7 A). In the IPAH patient lung tissues, GM-CSFRα co-distributed with cells expressing CD31 and CD68 in the thickened intima, in the media, and in neoangiogenic channels in plexiform lesions (Fig. 7, B and C).

Consistent with the immunohistochemical analyses, flow cytometry of freshly isolated lung CD31-positive cells revealed that the fraction of GM-CSFRα–positive cells was significantly expanded in IPAH versus unused donor control lungs (77.4 ± 17 vs. 20.9 ± 12%, P < 0.05; Fig. 7, D and E). Single cell transcriptional analysis using microfluidic-based real-time quantitative RT-PCR (qRT-PCR) identified two subpopulations with distinct transcriptional signatures. 46% of the cells were assigned membership to a subpopulation with high expression of other endothelial genes, including VWF, KDR, TEK, ENG, and CD34 (Fig. 7 F, left), and the remaining 54% of the cells comprised a different subpopulation with high expression of monocyte/macrophage-associated genes, including CD14, CD68, CD163, vimentin, and CD44 (Fig. 7 F, right). The proportion of GM-CSFRα–positive cells expressing monocyte/macrophage markers was considerably higher in the IPAH lungs than in the control tissue, i.e., 35 versus 3.5% (not depicted). These experiments suggest a strong association between elevated levels of GM-CSF and the recruitment of GM-CSFRα–positive cells that could be of endothelial or hematopoietic lineage. There was no evidence that GM-CSF at doses of 0.1–10 ng/ml could appreciably induce proliferation of cultured PAECs or PASMCs (not depicted), so we propose that its major function may lie in recruiting cells that could contribute to vascular pathology.

To determine whether up-regulation of the GM-CSF–GM-CSFRα axis induces or increases the severity of PAH, we administered murine GM-CSF to mice exposed to room air or to chronic hypoxia for 3 wk, as described in the Materials and methods. GM-CSF administration negatively impacted body weight in mice exposed to chronic hypoxia for 3 wk (Table 3). The total number of circulating white blood cells tended to be higher in the GM-CSF–treated group but did not reach statistical significance; the proportion of neutrophils and monocytes was, however, greater in the GM-CSF–treated mice (Tables 1 and 2). Hematocrit was, as expected, higher in hypoxic animals but was unaffected by GM-CSF administration. Left ventricle (LV) function and cardiac output were unaffected by GM-CSF treatment (Tables 1 and 2). However, right ventricle (RV) systolic pressure (RVSP) was significantly higher in the GM-CSF– versus vehicle-treated group after exposure to hypoxia (Fig. 8 A), consistent with more severe RV hypertrophy (Fig. 8 B). Morphometric analysis using barium-injected lung tissues revealed that the percentage of muscularized small PAs induced by chronic hypoxia was also significantly higher in GM-CSF–treated animals (Fig. 8, C and D), but the reduced number of distal arteries per 100 alveoli was unchanged by GM-CSF administration (Fig. 8 E). Immunohistochemistry revealed a significantly higher number of perivascular cells expressing a macrophage marker (Mac-3) in GM-CSF–treated room air animals, similar to that seen with hypoxic exposure. The combination of GM-CSF and hypoxia was additive in recruiting perivascular Mac-3–positive cells (Fig. 8, F and G).

Given that recruitment of macrophages in hypoxia was evident both in our experiments and in those recently reported (Vergadi et al., 2011), we determined whether blocking GM-CSF might prevent hypoxia-induced pulmonary hypertension. Injection of a GM-CSF–neutralizing antibody prevented hypoxia-induced pulmonary hypertension, RV hypertrophy, and associated vascular changes described above, including recruitment of perivascular macrophages (Fig. 8).

To determine whether GM-CSF might also be elevated in an inflammatory model of PAH, we monitored its expression using qRT-PCR in rats injected with the toxin monocrotaline as described in the Materials and methods. We observed an almost twofold increase in mRNA levels for GM-CSF in the lungs of rats at 14 d after injection of monocrotaline (Fig. 8 H), a time point we and others have shown is consistent with the development of significant pulmonary hypertension in this model (Rosenberg and Rabinovitch, 1988).

Clinical and experimental studies have linked reduced expression of BMPR2 with inflammation and PAH (Song et al., 2005; Hagen et al., 2007; Hong et al., 2008). Our study has uncovered a novel mechanism accounting for this link. We have shown that in response to a proinflammatory cytokine such as TNF, prolonged activation of p38–MK2 occurs if there is reduced expression of BMPR2. This activates GADD34–PP1 and dephosphorylates eIF2α, thereby impairing assembly of stress granules and enhancing mRNA translation of GM-CSF along with a subset of TNF-mediated cytokines that likely includes IL-6 and IL-8 but not MCP-1. In addition to mRNA translation, the level of secreted cytokine depends on factors that stabilize or destabilize the protein.

We show, for the first time to our knowledge, the involvement of GM-CSF in the pathobiology of pulmonary vascular lesions in IPAH patients. Heightened GM-CSF co-distributes with increased immunoreactivity for TNF throughout the vessel wall in patients with IPAH and is associated with an expanded subpopulation of intimal cells positive for GM-CSFRα. We determined by microfluidic-based single cell transcriptional analysis that these cells were divided equally into two subgroups; one subgroup coexpressed markers of ECs, and the other subgroup coexpressed markers of monocytes/macrophages. Moreover, we provide evidence that administration of GM-CSF can worsen pulmonary hypertension in mice exposed to hypoxia, in association with enhanced recruitment of perivascular macrophages, and that blockade of GM-CSF prevents this pathology. In addition, elevated GM-CSF coincides with the development of significant pulmonary hypertension in rats in response to the proinflammatory agent monocrotaline. Collectively, our experiments provide a novel mechanistic link between PAH, reduced expression of BMPR2, and an amplified response to a proinflammatory signal that promotes GM-CSF–mediated recruitment of cells that can contribute to pulmonary vascular pathology by their production of cytokines, chemokines, leukotrienes, proteolytic enzymes, and growth factors.

Several previous studies support the antiinflammatory properties of BMPR2 signaling (Song et al., 2005; Hagen et al., 2007; Hong et al., 2008). To further address the proinflammatory effects of loss of BMPR2 signaling, we stimulated cells with TNF because high circulating levels of this cytokine are evident in patients with IPAH (Itoh et al., 2006; Soon et al., 2010). Moreover, TNF can induce PAH in experimental animals, albeit in association with severe emphysema (Fujita et al., 2002) or related to heightened vasoreactivity (Stevens et al., 1992). The response to TNF is the induction of a cytokine/chemokine cascade involving GM-CSF (Crane et al., 1999). Loss of BMPR2, however, greatly amplified TNF-mediated GM-CSF production by a mechanism that we related to p-p38–dependent impaired stress granule assembly and mRNA translation.

It has been reported that MK2, a substrate of p-p38, plays a critical role in the posttranscriptional as well as the transcriptional regulation of cytokine production (Kotlyarov et al., 1999). The duration and magnitude of activity of MAPKs like MK2 are determinants of its biological effect (Keyse, 2000). As an increase in p-p38 resulting from loss of BMPR2 caused heightened IL-6 production in PASMCs (Hagen et al., 2007), we investigated whether there was also enhanced mRNA translation of IL-6 in response to TNF. Indeed, a similar mechanism of prolonged p-p38 activation with reduced BMPR2 could explain heightened mRNA translation of both IL-6 and IL-8 in PAECs in response to TNF. We can only speculate as to why heightened translation of MCP-1 escapes modulation by prolonged p-p38 activation. Future studies could be aimed at identifying the subset of TNF-regulated transcripts that are differentially translated in response to loss of BMPR2, using microarray (Sampath et al., 2008) or RNA-Seq gene expression profiling in heavy versus light polysomes, correlated with proteomic analyses of cell lysates and conditioned media.

To connect prolonged activation of p-p38–MK2 with enhanced mRNA translation in association with TNF stimulation, we investigated whether assembly of stress granules was impaired under this condition. Stress granules are induced by inflammatory cytokines such as interferon-γ and TNF (Hu et al., 2010) to silence or dampen mRNA translation by mechanisms that include phosphorylation of eIF2α (Anderson and Kedersha, 2009). TNF treatment induced the phosphorylation of eIF2α, and this was inhibited by BMPR2 siRNA treatment in a p-p38–dependent manner. These data support impaired assembly of stress granules or stress granule disassembly as the mechanism involved in increasing GM-CSF mRNA translation. We speculated that p-p38 activated the GADD34–PP1 complex that promotes eIF2α dephosphorylation (Brush et al., 2003) and initiates mRNA translation. We did not find an increase in GADD34 and PP1 proteins, a mechanism previously related to heightened activity of the phosphatase complex (Oh-Hashi et al., 2001), nor did we find an increase in the formation of the complex between these proteins. We were able to show, however, that salubrinal, a selective inhibitor of GADD34–PP1 (Boyce et al., 2005), or lowering levels of PP1 by siRNA subverted the decrease in phosphorylated eIF2α, as well as the enhanced GM-CSF production attributed to reduced BMPR2. Thus, loss of BMPR2 likely activates GADD34–PP1 by a novel p-p38–dependent mechanism.

We confirmed, using immunofluorescence staining, that loss of BMPR2 could subvert stress granule formation induced by TNF or by other cellular stresses such as hypoxia. This observation implies that dysfunction of BMPR2 could interfere with stress granule formation in response to a wide variety of environmental stimuli implicated in the pathobiology of pulmonary hypertension (Speich et al., 1991; Cool et al., 2003; Stenmark et al., 2005; Marecki et al., 2006; Spiekerkoetter et al., 2008). It is possible, although unlikely, that despite the use of two different BMPR2 siRNAs and two different scrambled siRNAs, there were off-target effects independent of reducing BMPR2, which subverted stress granule assembly.

Once we confirmed heightened expression of GM-CSF in the PAs of patients with IPAH, we investigated whether this chemokine might be promoting the vascular pathology. GM-CSF was originally defined by its ability to generate granulocyte and macrophage colonies from bone marrow precursor cells after the proliferation and differentiation of these cells (Burgess and Metcalf, 1980). Later studies pursued the role of GM-CSF in regulating the function of more mature myeloid cells of granulocyte and macrophage lineages, during host defense and inflammatory reactions (Hamilton, 2008). For example, GM-CSF mediates macrophage polarization and induces production of the pro-inflammatory cytokines TNF, IL-6, and IL-12 (Fleetwood et al., 2007), as well as leukotriene B4 (Serezani et al., 2012), recently implicated in experimental and clinical PAH (Tian et al., 2013). GM-CSF responsiveness is not limited to hematopoietic lineages (Valdembri et al., 2009). ECs respond to GM-CSF and express both α and β subunits of the GM-CSF receptor (Soldi et al., 1997). GM-CSF–mediated EC activation induces both proliferation and expression of inflammatory cell adhesion molecules (Bussolino et al., 1989; Detmar et al., 1992). GM-CSF did not mediate proliferation of PAECs, and there was only a modest proliferation of PASMCs, in accordance with a previous report (Rolfe et al., 1995; unpublished data). Thus, other factors, i.e., GM-CSF–mediated recruitment of inflammatory or progenitor cells, may be relevant to the mechanism whereby GM-CSF promotes PAH or atherosclerosis (Haghighat et al., 2007) by inducing intimal cell proliferation in nascent lesions (Zhu et al., 2009).

We found heightened immunoreactivity for GM-CSF and GM-CSFRα in ECs and macrophages in IPAH PAs, and transcriptional analysis of single cells that express both CD31 and GM-CSFRα identified two subpopulations, one with endothelial and the other with monocyte/macrophage lineage markers. Collectively, it is conceivable that in the absence of normal BMPR2 signaling and in the presence of an inflammatory stimulus, an increase in GM-CSF induces GM-CSFRα–expressing PAECs and the recruitment of GM-CSFRα–positive inflammatory cells that induce adverse pulmonary vascular remodeling. It is also possible that an increase in GM-CSF converts a PASMC to a macrophage-like cell.

To address whether GM-CSF causes or exacerbates PAH, we investigated the effects of administering this chemokine to mice in room air or while exposed to chronic hypoxia. GM-CSF produced no evidence of PAH during treatment of the mice in room air but worsened hypoxia-induced PAH. Lung histology indicated that GM-CSF enhanced muscularization of distal arteries, which is consistent with its modest effect on proliferation of cultured SMCs that was observed by our group and others (Rolfe et al., 1995). It has also been shown that locally applied GM-CSF induces the accumulation of α-SM Actin–containing myofibroblasts (Rubbia-Brandt et al., 1991), and this, as well, could also explain the increased muscularization of distal vessels (Jones, 1992). More striking, however, was the observation that recruitment of perivascular macrophages or cells bearing a macrophage marker was amplified by GM-CSF. It has been suggested that macrophages can contribute to hypoxia-induced PAH (Stenmark et al., 2005; Vergadi et al., 2011) by enhancing cytokine production (Brissette et al., 1995; Li et al., 2011) and matrix metalloproteinase activity (Wu et al., 2001). That loss of BMPR2 per se does not aggravate hypoxia-induced PAH and associated vascular changes (Long et al., 2006), but does cause PAH in response to an inflammatory stimulus, suggests that greater loss of BMPR2 or a more severe hypoxia exposure may be necessary to bring out the subverted stress granule response that we see in cultured cells with both TNF as well as hypoxia. The reduction in stress granule formation is also more modest in response to hypoxia compared with TNF. Consistent with this, we showed that heightened production of GM-CSF is associated with progressive inflammatory forms of PAH, such as that induced by the toxin monocrotaline (Wilson et al., 1989). Heightened GM-CSF may also underlie the increased leukotriene B4 production by perivascular macrophages (Serezani et al., 2012), which is implicated in the inflammatory PAH produced in the athymic rat given VEGFR blockade (Tian et al., 2013), and in clinical PAH, which is also highly inflammatory (Humbert et al., 1995; Itoh et al., 2006; Soon et al., 2010).

Recombinant human BMP-2, anti–α-tubulin antibody, anti–β-actin antibody, collagenase IA-S, protease inhibitor cocktail, and phosphatase inhibitor cocktails I and II were purchased from Sigma-Aldrich. Recombinant human TNF, GM-CSF, VEGF165, PDGF-BB, and mouse IgG were purchased from R&D Systems. BMPR2 and Mac-3 antibodies were purchased from BD, and S-6, L13a, p38, p-p38, MK2, p-MK2, I-κB, p–I-κB, eIF2α, and p-eIF2α antibodies were purchased from Cell Signaling Technology. GADD34, protein phosphatase-1, GM-CSFRα, HuR, and TIA-1 antibodies (4H1) and salubrinal were obtained from Santa Cruz Biotechnology, Inc., and G3BP antibody was obtained from BD. SB202190 was purchased from EMD Millipore. GM-CSF and TNF antibodies for immunohistochemistry were obtained from Abcam, and CD31, CD34, CD68, and α-SM actin antibodies and normal rabbit Ig were obtained from Dako. HRP-conjugated rabbit and mouse secondary antibodies and ECL and ECL Plus kits were ordered from GE Healthcare. Allophycocyanin (APC)-human CD31 and PE–GM-CSFRα antibodies were purchased from eBioscience. Recombinant murine GM-CSF was obtained from PeproTech.

Primary human PAECs from large vessels and human pulmonary microvascular ECs (PMVECs; ScienCell) were grown in commercial EC media (ScienCell), subcultured at a 1:6 ratio in gelatin-coated dishes and flasks (BD Falcon and Corning), and used at passages 4–6. Cells were starved in basal media (ScienCell) with 0.2% FBS and gentamycin/amphotericin overnight before adding the ligands or the vehicle. Primary human PASMCs from large vessels (Lonza) were grown in smooth muscle growth media-2 (Lonza), subcultured at a 1:3 ratio in uncoated flasks, and used at passages 4–6.

2.5 × 10⁴ cells were plated on 0.2% gelatin-coated 24-well plates and allowed to adhere for 24 h. Cells were washed in PBS and then starved in 0.2% FBS containing EC or SMC media for 24 h. After treatment with TNF, cell supernatants were collected and stored at −80°C. The concentration of GM-CSF, IL-6, IL-8, and MCP-1 were determined using an ELISA kit (Quantikine; R&D Systems) according to the manufacturer’s instructions.

siRNA specific for BMPR2 (Silencer Select; Ambion) or nontargeting siRNA as control (Silencer Select negative control; Ambion) were transfected into PAECs after withdrawal of antibiotics 24 h before and during treatment with Lipofectamine RNAi max (Invitrogen). Knockdown efficiency for BMPR2 was determined by Western immunoblot. We tested an additional control siRNA (Thermo Fisher Scientific) and two BMPR2 siRNAs (B2:s2045 and B3:s2046; Ambion) with similar results. Data shown were obtained with the siRNA (B1:s2044; Ambion). Knockdown efficiency, TNF-induced GM-CSF protein production, and TNF-induced phosphorylation of p38 and I-κB were analyzed and similar using either control or BMPR2 siRNA. We also performed experiments using siRNA specific for protein phosphatase 1c (s10930; Ambion).

For protein expression analysis, PAECs or PMVECs were washed with ice-cold PBS, and lysates were prepared by adding boiling lysis buffer (10 mM Tris HCl, 1% SDS, and 0.2 mM PMSF) containing protease and phosphatase inhibitors. Lysates were scraped into a 1.5-ml microcentrifuge tube and boiled for 10 min before centrifugation. Supernatants were transferred to fresh microcentrifuge tubes and stored at −80°C. Protein concentration was determined by the Lowry assay (Bio-Rad Laboratories). Equal amounts of protein were loaded onto each lane of a 4–12% Bis-Tris gel and subjected to electrophoresis under reducing conditions. After blotting, polyvinylidene difluoride membranes were blocked for 1 h (5% milk powder in 0.1% PBS/Tween) and incubated with primary antibodies overnight at 4°C. The binding of secondary HRP-conjugated antibodies was visualized by ECL or ECL Plus. Normalizing for total protein was performed by reprobing the membrane with a mouse monoclonal antibody against α-tubulin or β-actin. The antibody dilutions used were 1:200 GM-CSF, 1:5,000 β-actin, 1:250 BMPR2, 1:1,000 p38, 1:1,000 p-p38, 1:1,000 MK2, 1:1,000 p-MK2, 1:1,000 I-κB, 1:1,000 p-I-κB, 1:1,000 eIF2α, 1:1,000 p-eIF2α, 1:200 GADD34 and PPI, and 1:5,000 α-tubulin. Secondary antibodies were used with dilutions of 1:2,000 to 1:10,000.

Total RNA was extracted and purified from cells using the RNeasy Mini kits (QIAGEN). The quantity and quality of RNA was determined by using a spectrophotometer, and 2 µg was used as the template for RT-PCR with random hexamers as primers. RT-PCR was performed on a 7900HT Sequence Detection System with TaqMan Gene Expression Assays on Demand probes (Applied Biosystems). Primers used are as follows: GM-CSF, Hs00171266_m1; β2 microglobulin, Hs99999907_m1; GAPDH, Hs02758991_m1; IL-6, Hs00985639_m1; IL-8, Hs00174103_m1; and MCP-1, Hs00234140_m1 (Applied Biosystems). In the monocrotaline experiments, RT-PCR for rat samples was performed by using 4 ng cDNA of whole lung lysate with 1 µl of 10 mM mixed primers, 5 µl SYBR green master mix (QIAGEN), and 4 µl sample. Primer sequences were designed using NCBI’s Primer-BLAST function and are as follows: β-Actin (F, 5′-ACCCGCCACCAGTTCGCCAT-3′; R, 5′-CTTGCACATGCCGGAGCCGT-3′) and GM-CSF (F, 5′-GCTGCCCAACCCTGTCACC-3′; R, 5′-GGTTGCCCCGTAGACCCTGC-3′).

After treating PAECs with 10 ng/ml TNF for 4 h, 20 µg/ml DRB was added to the media. GM-CSF mRNA levels were measured by qRT-PCR before and 20, 40, 60, 80, and 100 min after transcriptional inhibition by DRB.

Polysome analysis was performed as previously described (Ying et al., 2009; Wehner et al., 2010). PAECs transfected with siRNA for BMPR2 or control siRNA were stimulated with TNF for 3 h. Translation was inhibited with 0.1 mg/ml cycloheximide, and the cells were scraped and lysed in 500 µl lysis buffer (0.3 M NaCl, 15 mM MgCl₂, 15 mM Tris-HCl, pH 7.5, 1% Triton X-100, and 0.1 mg/ml cycloheximide) on ice for 10 min. Nuclei were pelleted at 10,000 g for 10 min, and 500 µl of the resulting supernatant was fractionated through a 10–60% linear sucrose gradient by centrifugation using a SW41 rotor for 2 h and 45 min at 35,000 rpm at 4°C. 15 fractions (750 µl each) were collected using a fraction collector (ISCO with Brandel pump). RNA in each fraction (650 µl) was extracted with phenol/chloroform/isoamyl alcohol (Invitrogen) and reverse-transcribed using Superscript III (Invitrogen). Real-time qPCR was performed on a 7900HT Sequence Detection System with TaqMan Gene Expression Assays on Demand probes. GAPDH was used as a housekeeping gene. Primers used are as follows: GM-CSF, Hs00171266_m1; and GAPDH, Hs99999907_m1 (Applied Biosystems). The amount of GM-CSF and GAPDH mRNA in each fraction was calculated. Protein was extracted from 100 µl of each polysome fraction by methanol for Western immunoblot analysis to detect the 40S ribosomal subunit protein S6 (1:1,000; 54D2; Cell Signaling Technology) and the 60S ribosomal subunit protein L13a (1:1,000; Cell Signaling Technology).

Stress granule formation was assessed as previously described (Kedersha and Anderson, 2007). 5 × 10⁴ PAECs were plated on a 4-well chamber slide. After serum starvation for 12 h, cells were treated with vehicle or 10 ng/ml TNF for 3 h. To assess stress granule formation after hypoxia, cells were incubated in a hypoxic chamber (1% O₂) and studied under the same conditions. In all experiments, the cells were fixed with 4% paraformaldehyde for 15 min and then treated with methanol for 10 min. After blocking with 5% horse serum, slides were incubated with an antibody against G3BP, the protein which is known to assemble stress granules (Tourrière et al., 2003), followed by Alexa Fluor 594–conjugated secondary antibody. Sections were stained with DAPI (Invitrogen) to visualize nuclei and mounted with Prolong Gold Antifade Reagent. Stress granule formation was assessed using fluorescence and confocal microscopy (FluoView 1000; Olympus). Stress granule–positive cells were counted relative to total cells in three randomly chosen high-power fields in four different experiments.

Lung tissues from IPAH patients who underwent lung transplantation and from unused donor lungs as controls were obtained through the Pulmonary Hypertension Breakthrough Initiative (PHBI) Network, which is funded by the Cardiovascular Medical Research and Education Fund (CMREF). The tissues were procured at the Transplant Procurement Centers at Allegheny General Hospital, Baylor University, Cleveland Clinic, Duke University, Stanford University, the University of California, San Diego, University of Alabama at Birmingham, and Vanderbilt University, and de-identified patient data were obtained via the Data Coordinating Center at the University of Michigan. Lung tissues were kept in RPMI 1640 media supplemented with antibiotics for up to 24 h during transportation from the transplant procurement centers. Small pieces of the lung tissue were snap-frozen in liquid nitrogen and stored at −80°C for biochemical experiments or fixed in 10% formaldehyde overnight for immunohistochemistry. Procurement of the tissues from human subjects is approved by the Administrative Panel on Human Subjects in Medical Research at Stanford University (IRB #350, Panel 6).

Human and mouse sections from formaldehyde-fixed and paraffin-embedded lung tissues were deparaffinized and rehydrated. Epitope retrieval was performed by boiling the sections in citrate buffer, pH 6.0. Sections were reacted with hydrogen peroxide to block endogenous peroxidase, washed, and blocked with 5% goat serum. The sections were then incubated with the primary antibodies against GM-CSF (1:400; Abcam), TNF (1:400; Abcam), α-SM actin (1:1; Dako), CD31 (1:50; Dako), CD34 (1:1; Dako), CD68 (1:100; Dako), GM-CSFRα (1:50; Santa Cruz Biotechnology, Inc.), and Mac3 (1:200; BD) overnight at 4°C. After streptavidin-biotin amplification (LSAB2 kit [Dako] or Vectastain Elite ABC kit [Vector Laboratories]), the slides were incubated with 3, 3′-diaminobenzidine and counterstained with hematoxylin. A negative control was performed using mouse IgG or rabbit Ig instead of the primary antibody. The localization and intensity (0, 1+, 2+, 3+) of immunoreactivity were assessed by two independent examiners, blinded to the diagnosis of PAH or to the mouse treatment group. In the murine experiments described below, the number of Mac-3–positive cells per PA was counted for all vessels in the section and a mean value calculated per mouse.

Proliferation was assessed by cell counts. Human PAECs or PMVECs were seeded at 2.5 × 10⁴ cells per well in a gelatin-coated 24-well plate in a 500-µl volume of growth medium and allowed to adhere for 24 h. Cells were washed three times and starved in 0.2% FBS containing EC media for 12 h and then incubated for 24 h in the presence or absence of the different ligands. Cells were counted in a hemocytometer (Bright-Line; Hausser Scientific). PASMCs were seeded as described above but starved in 0.1% FBS for 48 h before stimulation.

Left lung sections (3 × 3 × 2 cm) were transported in RPMI 1640 medium, then chopped into small pieces (1 × 1 × 2 mm) using scissors, and placed in a 50-ml conical tube. The minced tissue was digested in 20 ml HBSS (Gibco) containing 0.8 mg/ml collagenase IA-S (Sigma-Aldrich) on the rotator platform for 45 min at 37°C. The cell suspension was filtered using a cell strainer (100 µm, Falcon; BD) and centrifuged at 250 g for 5 min. The pellet was resuspended in 2∼4 ml of PBS containing 0.1% BSA. Magnetic beads (30 µl, Dynabeads; Invitrogen) conjugated with anti-CD31 antibody were added and incubated with the cell suspension for 20 min at 4°C. The cells bound to the beads were recovered with a magnet rack (Dynal) and washed five times with PBS containing 0.1% BSA. The cells were counted with a hemocytometer, and up to 2 × 10⁶ cells were used for FACS.

After blocking nonspecific binding with media containing 5% FBS for 15 min on ice, cell suspensions were centrifuged at 250 g for 5 min, resuspended in 100 µl FACS buffer (0.2% fetal bovine serum, 2 mM EDTA, and PBS, pH 7.2) containing APC-conjugated CD31 and PE-conjugated GM-CSFRα antibodies, and incubated on ice for 20 min. Then cells were washed with 0.1% BSA/PBS, resuspended in FACS buffer, and analyzed and sorted using a multiparameter cell sorter (FACS Aria, BD). Single cells from subpopulations of CD31⁺GM-CSFRα⁻ or CD31⁺GM-CSFRα⁺ cells were sorted into a 96-well plate that had been preloaded with RT-PCR reagents (reverse transcription enzyme and primers) for each gene target.

As previously described (Thorsen et al., 2002; Diehn et al., 2009), single cells were sorted into each well of a 96-well plate that had been preloaded with 10 µl of a master mix containing Tris-EDTA buffer, pH 7.0, Superscript III Platinum reverse transcription/Taq DNA polymerase enzyme mix (Invitrogen), Cells Direct reaction mix (Invitrogen), target gene–specific TaqMan assay (primer/probe) sets (Applied Biosystems), and SUPERase-In RNase inhibitor (Applied Biosystems). Exon-spanning primers were used where possible to avoid amplification of genomic background. Reverse transcription was performed (20 min at 50°C, 2 min at 95°C), followed immediately by a 22-cycle preamplification in the same reaction tube (denature at 95°C for 15 min, anneal at 60°C for 4 min, each cycle). Resultant single cell cDNA was mixed with sample loading agent (Fluidigm) and Universal PCR Master Mix (Applied Biosystems) and loaded into 48.48 Dynamic Array chips (Fluidigm) along with TaqMan gene expression assays and assay loading agent (Fluidigm). Products were analyzed on the BioMark reader system (Fluidigm) using a hot start protocol to minimize primer-dimer formation. 30 RT-PCR cycles were performed. Gene expression data from all individual cells were normalized relative to the median expression for each gene in the pooled sample and converted to base 2 logarithms. Absolute bounds of ±5 cycle thresholds (corresponding to 32-fold increases/decreases in expression) were set, and zero-expressers were assigned to this floor. After normalization of datasets from multiple chip runs, a pooled analysis was used to examine the characteristics of the entire population of cells. To detect overlapping patterns within the single cell transcriptional data, we used an adaptive fuzzy c-means clustering algorithm for partitioning of the cells based on similar transcriptional profiles (Glotzbach et al., 2011).

The Animal Care Committee of Stanford University approved all protocols, in keeping with the regulations of the American Physiological Society.

8-wk-old C57BL/6 mice were exposed to 10% oxygen or kept in ambient air, and both groups were either treated with a daily i.p. injection of GM-CSF (0.4 µg/dose; PeproTech) or saline vehicle for five consecutive days per week, for 3 wk using a protocol previously described (Haghighat et al., 2007). Subgroups in normoxia and hypoxia were treated for 2 wk with continuous s.c. administration of 7 µg/d rat anti–mouse GM-CSF–IgG (eBioscience) or unspecific isotype control rat IgG2a (eBioscience) via s.c. mini-osmotic pump (model 1002; Alzet).

12 male Sprague Dawley rats (220–280 g) were randomly selected for a single injection with 60 mg/kg monocrotaline (8 animals) or 0.9% saline (4 animals). Lungs of the control animals were harvested at days 0 and 21, and lungs of monocrotaline-treated animals were harvested at days 2, 7, 14, and 21 for whole lung lysate mRNA extraction.

After 20 d of GM-CSF treatment, cardiac function and output were measured by echocardiography under isoflurane anesthesia (1%, 1 liter/min oxygen) using a Vivid 7 ultrasound machine (GE Healthcare) and a 13-MHz linear array transducer as in our previous study (Hansmann et al., 2008).

RVSP was measured under isoflurane anesthesia (1.5%, 1 liter/min oxygen) by inserting a 1.4F catheter (Millar Instruments) via the right jugular vein as previously described (Hansmann et al., 2008). Pressure measurements were repeated three times and the mean value used. Data were collected by Power Lab Data Acquisition system (AD Instruments) and analyzed using LabChart software (AD Instruments) by an investigator blinded to the treatment group.

A blood sample was taken by heart puncture after the completion of the RVSP measurement, and a complete blood cell count that included a differential count of white blood cells was performed in the diagnostic laboratory of the Stanford Animal Care Facility.

RV hypertrophy was assessed by the weight of the RV relative to LV plus septum.

After the hemodynamic measurements were completed, the isolated lungs were perfused with saline, fixed in 10% formalin overnight, and then embedded in paraffin for immunohistochemistry. Mac-3–positive cells adherent to the distal alveolar duct and wall PAs were counted in a full-length section of each lung, and a mean was calculated for each mouse. A subset of left lungs was prepared for morphometric analysis of the vasculature, using the barium injection method that has been reported in detail (Guignabert et al., 2009).

Sections parallel to the hilum in the mid-portion of the barium-injected left lung were taken and embedded in the same manner and stained by the Movat pentachrome method. All barium-filled arteries 15–50 µm in external diameter at alveolar duct and alveolar wall level were assessed for the presence of muscularity. These distal PAs and alveoli were counted in six randomly chosen lung fields per mouse, and the ratio of the number of distal PAs to 100 alveoli was calculated. All measurements were performed by two investigators blinded to the experimental group being studied.

Values from multiple experiments are shown as mean ± SEM. Statistical significance was determined using one-way ANOVA followed by Bonferroni’s multiple comparison test. When only two groups were compared, statistical differences were assessed with unpaired two-tailed Student’s t test. A p-value of <0.05 was considered significant. The number of samples or animals in each group is indicated in the figures or figure legends.

We greatly appreciate the editorial and technical assistance of Dr. Michal Bental Roof in preparing both the manuscript and figures and the administrative help of Ms. Michelle Fox. Lung tissues from IPAH and control patients were provided by the PHBI, which is funded by the CMREF. The tissues were procured at the Transplant Procurement Centers at Baylor University, Stanford University, University of California, San Diego, and Vanderbilt University, and de-identified patient data were obtained via the Data Coordinating Center at the University of Michigan.

This work was supported by grants from the National Institutes of Health National Heart, Lung, and Blood Institute (R01 HL087118 and N01-HV-00242 HHSN268201000034C to M. Rabinovitch). H. Sawada was supported by a grant from Mie University Graduate School of Medicine and from the Oak Foundation, J.P. Glotzbach by the Oak Foundation, R. Chan by a studentship and Y.-M. Kim by a fellowship from the American Heart Association, N.P. Nickel by a fellowship from the German Research Foundation (DFG), T.-P. Alastalo by the Sigrid Juselius Foundation, Instrumentarium Foundation, Finnish Foundation for Cardiovascular Research, Finnish Cultural Foundation, Finnish Foundation for Pediatric Research, and the Academy of Finland, and V. de Jesus Perez by a grant from the Amos Foundation. M. Rabinovitch was supported by the Dunlevie Chair of Pediatric Cardiology at Stanford University.

The authors have no conflicting financial interests.