|Home | About | Journals | Submit | Contact Us | Français|
Rationale: Extracellular matrix (ECM) is a dynamic tissue that contributes to organ integrity and function, and its regulation of cell phenotype is a major aspect of cell biology. However, standard in vitro culture approaches are of unclear physiologic relevance because they do not mimic the compositional, architectural, or distensible nature of a living organ. In the lung, fibroblasts exist in ECM-rich interstitial spaces and are key effectors of lung fibrogenesis.
Objectives: To better address how ECM influences fibroblast phenotype in a disease-specific manner, we developed a culture system using acellular human normal and fibrotic lungs.
Methods: Decellularization was achieved using treatment with detergents, salts, and DNase. The resultant matrices can be sectioned as uniform slices within which cells were cultured.
Measurements and Main Results: We report that the decellularization process effectively removes cellular and nuclear material while retaining native dimensionality and stiffness of lung tissue. We demonstrate that lung fibroblasts reseeded into acellular lung matrices can be subsequently assayed using conventional protocols; in this manner we show that fibrotic matrices clearly promote transforming growth factor-β–independent myofibroblast differentiation compared with normal matrices. Furthermore, comprehensive analysis of acellular matrix ECM details significant compositional differences between normal and fibrotic lungs, paving the way for further study of novel hypotheses.
Conclusions: This methodology is expected to allow investigation of important ECM-based hypotheses in human tissues and permits future scientific exploration in an organ- and disease-specific manner.
Extracellular matrix is important in driving cellular phenotypic responses in fibrotic and other lung disorders. Typical investigation using planar, rigid dishes coated with isolated matrix proteins may not accurately reflect the impact of matrix on cells in vivo.
We define a new method to prepare acellular lung matrix that can be used to study the impact of extracellular matrix on lung cell phenotype. This process results in a more physiologic substrate that retains normal stiffness, architecture, and matrix composition compared with native lungs. By using human lungs and cells, this new system provides a novel way to culture cells in vitro to investigate matrix contributions to lung pathology in a disease-specific manner.
The extracellular matrix (ECM), long believed to simply serve as a structural support for cells, is now recognized as an important mediator in health and disease; ECM is a highly dynamic complex that varies in composition according to its tissue localization and physiological circumstances. In addition to providing structural integrity, ECM delivers important spatial and contextual cues to drive varied behaviors such as proliferation, migration, epithelial-to-mesenchymal transition, apoptosis, and production of various secreted mediators (as reviewed in References 1–3, among others). ECM also acts as a reservoir for growth factors and cytokines that, during tissue injury and repair, help to localize and direct the wound healing response. The ECM is critical during embryogenesis, dictating key developmental events such as branching morphogenesis in the lung (4, 5) as well as cardiovascular and neural development (6, 7). Thus, ECM is an integral contributor to tissue homeostasis and development.
In the human lung, ECM consists primarily of collagen I (which provides tensile strength), elastins (which provide elasticity), and laminins and collagen IV (which constitutes the bulk of the alveolar and airway basement membranes). Other ECM proteins (e.g., fibronectin and glycosaminoglycans) serve to maintain lung cell polarity and survival. Although the groundwork for our understanding of individual ECM component functions has been established using standard culture techniques, mechanistic studies of human lung ECM have been hampered by lack of a biologically relevant model. Although reductionist studies using cells grown on planar ECM-coated plastic dishes have informed our understanding of cell signaling and behavior, this approach does not account for the complexity of a three-dimensional (3D), deformable microenvironment in the lung where cells encounter a variety of different proteins simultaneously in a unique spatial geometry. To address these issues, some investigators have used human fibroblast-derived 3D matrices to determine the effects of ECM on cell phenotype, often with strikingly disparate results compared with planar matrices (8). This approach, although more appropriate than culturing cells on planar ECM-coated dishes, does not recapitulate an organ-specific ECM structure and composition. More recently, newer technologies involving artificial polymeric or hydrogel scaffolds (9) or bioengineered cantilever posts that deform under stress (10) have been used to evaluate cell–ECM interactions, but these studies are often limited to single ECM proteins rather than the complex mixture found in situ and lack the topographic variability of the organs they are designed to mimic. Notably, none of these models allows for ECM analysis in an organ- or disease-specific manner.
Traditional tissue culture approaches are inadequate to study disease states that are themselves defined by disordered ECM. Idiopathic pulmonary fibrosis (IPF) is a progressive disorder of the lungs characterized by deposition and remodeling of the lung ECM, which ultimately leads to chronic respiratory failure and death. At present, no effective therapeutic options exist to slow progression of IPF (11). This has primarily been attributed to the fact that no animal model faithfully recapitulates all features of IPF (12), and in vitro investigation typically relies on standard culture techniques.
Because ECM is a critical regulator of health and disease, some investigators have relied on acellular matrices for organ regeneration. In experimental rodent models, we and others have demonstrated an ability to regenerate functional lungs (13, 14) and heart (15) by repopulating acellular matrices with various cell types. Perhaps more importantly, this approach has been successfully used in humans to correct large airway defects (16, 17).
To better study the influences of human lung ECM on cell behavior in vitro, we developed a biologically relevant in vitro model using decellularized human lung matrices. Our method results in a lung matrix devoid of cellular and nuclear material. Structural and ultrastructural analysis confirms the normal architecture of the matrix, with normally aligned matrix fibers and intact, homogeneous basement membranes. In contrast, acellular IPF lung matrix reveals marked ECM deposition in the lung interstitium with disorganized alignment and disrupted basement membranes. Proteomic analysis of acellular lung matrices identifies significant differences in ECM composition between normal and IPF lung. Atomic force microscopy confirms that acellular normal lung matrices exhibit an elastic modulus similar to that expected in vivo, whereas IPF acellular matrices demonstrated increased but heterogeneous stiffness. Finally, when cultured in normal or in IPF acellular lung matrices, normal human fibroblasts adopted starkly different phenotypes, confirming the active role of the ECM in shaping cell behavior. Taken together, these data highlight the feasibility of using an entirely human in vitro system to study lung ECM biology in a manner that takes into account tissue stiffness, architecture, and matrix composition. We believe this new approach will enable biologically relevant, mechanistic study of ECM in a fully humanized system that can be adapted to any disease state. Moreover, our data provide novel compositional information regarding the human lung ECM, opening new potential avenues of research.
Some of these results have been previously presented in abstract form (18).
Detailed methods can be found in the online supplement.
Primary human lung fibroblasts were isolated as we have described previously (19). Written informed consent was obtained from all subjects in accordance with the University of Michigan Institutional Review Board, and cell lines were derived from patients in a blinded fashion, without regard to clinical data except diagnosis. Cells were maintained in Dulbecco’s modified Eagle medium with 10% fetal calf serum, antibiotics, glutamine, and N-2-hydroxyethylpiperazine-N′-ethane sulfonic acid. The ALK5 inhibitor A83-01 (20) was from Tocris Bioscience (Minneapolis, MN) and was used at a concentration of 20 nM in dimethyl sulfoxide.
Human control lungs deemed to be unsuitable for lung transplantation were obtained from beating-heart (or warm autopsy) donors through Gift of Life Michigan. Human deidentified IPF lung samples were obtained from explants of patients with IPF undergoing lung transplantation at the University of Michigan. The University of Michigan Institutional Review Board has deemed these approaches exempt from oversight as all subjects were considered deceased.
Decellularization was achieved based on modifications of a previously published protocol (21). A detailed protocol used for these studies is included in the online supplement.
Cylinders of acellular lung matrices were obtained with a sterile 12-mm biopsy punch (Acuderm, Inc., Fort Lauderdale, FL), embedded in 2% warm ultra-low melt agarose (SeaPrep; Lonza, Rockland, ME), and cooled at 4°C. Sections 1,000 μm thick were cut on a standard Vibratome (Leica Biosystems, Richmond, IL). Sections were gently warmed to remove the agarose surrounding and within tissue sections. Before culture, slices were sterilized in 0.18% peracetic acid and 4.8% ethanol solution and rinsed extensively in phosphate-buffered saline (PBS). Slices were stored at 4°C in PBS containing antibiotics until use in culture. Using this technique with human lungs, a minimum of 60 cylinders can be obtained per lung lobe (depending on lobe size), and each cylinder can be sliced into 6 to 10 uniform slices.
Acellular matrices were placed in individual wells of a 24-well dish without media. Fibroblasts were harvested from culture, resuspended in media at 1 × 107 cells/ml, and 10 μl of cells were seeded dropwise into slices. After 30 minutes, tissue slices were transferred to individual wells of a Teflon-coated 24-well plate and media slowly added to the side of the well. Slices were cultured for indicated time points, changing media every other day.
Immunohistochemistry on paraffin-embedded sections was performed as previously described (22), with standard hematoxylin and eosin, Masson trichrome, Alcian blue, or Verhoeff elastic stains. Anti–α-smooth muscle actin (SMA) (clone 1A4; Sigma, St. Louis, MO) was used at 1:200. A Nikon Eclipse 50i microscope fitted with an Evolution MP Color CCD camera (Media Cybernetics, Bethesda, MD) was used to visualize and photograph sections.
Scanning and transmission electron microscopy were performed by the University of Michigan Microscopy and Image Analysis Core.
A detailed protocol describing atomic force microscopy is available in the online supplement.
A detailed protocol describing the method used for mass spectrometry is available in the online supplement.
Western blot analysis of cellular lysates was performed as previously described (22). Antibodies against the EDA domain of fibronectin (clone 3E2) and α-SMA (clone 1A4) was from Sigma. Antibodies to β-tubulin, GAPDH (glyceraldehyde-3-phosphate dehydrogenase), PTEN (phosphatase and tensin homolog deleted on chromosome 10), and E cadherin were from Cell Signaling Technologies (Danvers, MA). Antibodies to vimentin were from Santa Cruz Biotechnology (Santa Cruz, CA); antibodies to cytokeratin-19 were from Novus Biologicals (Littleton, CO).
RNA extraction from cells cultured in acellular matrices was performed using a previously published method (23). Quantitative reverse transcriptase–polymerase chain reaction was performed on an Applied Biosystems (Foster City, CA) 7300 thermocycler as previously described (22), using SYBR Green for detection. All housekeeping genes were assayed using a gene array (catalog # PAHS-000; SA Biosciences/Qiagen, Valencia, CA) according to the manufacturer’s instructions.
Culturing cells in acellular human lungs to study effects of ECM on target cell behavior requires they first be devoid of endogenous cellular material that might influence experimental results. Thus, we first confirmed that the decellularization process resulted in extraction of all cellular and nuclear material. Native or decellularized normal human donor lung samples (n = 3) were fixed in 10% buffered formalin and paraffin embedded. Sections were prepared and stained with hematoxylin and eosin (H&E), Verhoeff elastic (to identify elastic fibers), Alcian blue (to identify glycosaminoglycans), and Masson trichrome stains (to identify collagens). Figure 1A shows the typical appearance of normal human lung before decellularization, with cells lining alveolar, airway, and vascular structures (H&E stain). Acellular normal human lung maintained a similar architecture, albeit without evidence of cells (Figure 1B, H&E stain). Additional stains indicated the presence of elastin fibers, glycosaminoglycans, and collagens (Figures 1C–1E). Given the dense nature of fibrotic tissue, we also wished to confirm that the decellularization procedure was capable of removing cellular material from IPF lungs as well. Figure 2 reveals the absence of cellular material in densely fibrotic regions, suggesting that the decellularization protocol is capable of removing cells from denser tissues and is consistent with the literature (16).
To confirm that cellular products were absent in the matrices, we next performed Western blot on radioimmunoprecipitation assay (RIPA) extracts from native or acellular human lungs (n = 2 each) for cellular proteins; as shown in Figure 3A, native lungs expressed all cellular proteins investigated to various degrees. In contrast, none were detected in extracts of acellular lungs, with the exception of a minimal amount of collagen I(αI) chain, which was likely leached from the matrix. Finally, to ensure RNA was effectively removed from the acellular lung, we also extracted RNA from native and acellular lung matrices. Quantitative real-time polymerase chain reaction for a panel of human housekeeping genes on extracted samples showed that mRNA expression in acellular matrices was reduced greater than 99% compared with native matrices (Figure 3B). These data suggest effective clearance of cells and nuclear material using this method and are consistent with prior approaches (16, 21).
Grossly, acellular lung maintains a 3D morphology similar in appearance to normal lungs (Figure 4), comparable to results achieved in the rodent (13). However, it was critical to ensure that the normal microscopic architecture was preserved after decellularization. Thus, sections of lung were subjected to scanning electron microscopy. Figure 5 shows alveoli (Figures 5A and 5D), airway (Figures 5B and 5E), and vessel (Figures 5C and 5F) before and after decellularization, respectively. Notably, the remaining ECM basement membranes appeared intact, with maintenance of the normal 3D morphology of lung.
Ultrastructurally, transmission electron microscopy highlights the abnormal appearance of IPF lung compared with normal lung (Figure 6). Normal lung revealed collagen fibers and elastin arranged in discrete organized bundles traversing interstitial spaces and what appear to be intact, homogeneous alveolar epithelial and endothelial basement membranes lining alveolar spaces and vessels, respectively, whereas IPF lungs showed a markedly expanded interstitial space with heterogeneous, partially disrupted basement membrane and disorganized ECM organization. Importantly, these findings (coupled with those in Figures 1 and and2)2) support our conclusion that cells were completely removed from the tissues.
To ensure that decellularization did not result in significant ECM protein loss, we performed proteomic analysis on the effluent of the various wash steps during decellularization. More than 500 protein peptides were identified in the wash effluent; manual review of these revealed that the large majority of expressed peptides were non–matrix-associated (see Table E1 in the online supplement). Moreover, paired proteomic analyses of lungs before and after decellularization (n = 3 each) showed very few proteins that were significantly reduced after decellularization (Table E2). The observations suggest that the decellularization procedure results in the loss of few structural matrix proteins.
Tissue stiffness (measured as Young’s elastic modulus) varies depending on the organ (24), ranging from distensible (as in lung, around 1 kPa ) to very rigid (as in bone, 6.9–31.6 GPa; reviewed in ). However, standard cell culture techniques often use plastic dishes, which do not account for this variability in stiffness to which cells are exposed in vivo. Indeed, tissue culture plastic possesses a Young’s modulus ranging from 2 to 4 GPa (24), suggesting that lung cells will experience a much stiffer surface, and bone cells a much more deformable surface, when cultured on plastic dishes in vitro.
To ensure the acellular lung matrix retains its physiologic stiffness, we subjected native and acellular normal (n = 2 separate donors) and IPF (n = 2 separate donors) lung matrices to atomic force microscopy as outlined in Methods. Consistent with prior estimates, native normal human lung possessed a mean (± SEM) Young’s modulus of 1.96 ± 0.13 kPa (24) arranged in a relatively homogeneous pattern. In contrast, native IPF lung displayed a bimodal distribution of stiffness, with each sample exhibiting stiffness comparable to normal tissue as well as much higher stiffness representing fibrotic tissue (Figure 7A). Combined, mean (± SEM) stiffness of IPF tissue was 16.52 ± 2.25 kPa, significantly higher than normal tissue (P < 0.0001 by Mann-Whitney test) but still markedly less stiff than standard tissue culture plastic.
Interestingly, we found that after decellularization, normal lung matrices exhibited a mean (± SEM) Young’s modulus of 1.606 ± 0.08 kPa, which is no different than native lung (Figure 7B). In contrast, the mean (± SEM) Young’s modulus of acellular IPF lung matrices was 7.34 ± 0.6 kPa (P < 0.0001 compared with normal acellular lung by Mann-Whitney test) and somewhat more homogeneous in distribution (Figure 7B). We conclude that decellularization of IPF lungs may result in a relaxation of matrix proteins resulting in tissue that is slightly less stiff but still significantly stiffer than normal lung.
To better understand the composition of lung ECM, samples of acellular lung matrix were next evaluated using liquid chromatography/tandem mass spectrometry (LC-MS/MS). To ensure appropriate sampling, matrix for mass spectrometry was obtained from regions located adjacent to histologically confirmed representative samples (Figure 8). We identified 94 proteins in the normal lung “matrisome” (as defined in ), comprising 61 core ECM proteins (glycoproteins, collagens, and proteoglycans) and 33 ECM-related proteins (Table 1). As expected, we identified collagens (VI, XIV, XII, and IV), laminins, fibronectin, and various proteoglycans and their associated core proteins in normal lung. Notably, elastin and its peptides (desmosine, isodesmosine) were not identified, likely due to the inability of trypsin (used in the preparation for LC-MS/MS) to cleave this hydrophobic protein (28). As evidence of the presence of elastin, however, we did identify EMILIN-1 (elastin microfibril interfacer-1) and fibrillin-1 (both of which are necessary for formation of elastin fibers) in our samples. Moreover, our immunohistochemistry (Figure 1) and electron microscopy (Figure 6) studies clearly demonstrate the presence of elastin. In contrast, IPF matrices expressed 85 matrisome proteins (64 core ECM proteins and 21 ECM-related proteins) with a different protein signature than normal lungs (for example, collagens I, V, and XV were only expressed in IPF lungs) (Table 1). Despite the relatively equal numbers of various ECM proteins between normal and IPF lungs, the data also showed marked differences in the relative quantities of important proteins (Table 2). For example, IPF lungs were enriched for matrisome-associated proteins that might be predicted, such as glycosaminoglycans and latent transforming growth factor (TGF)-β binding protein 1, as well as relatively deficient in others, such as laminin α-3, β-3, and γ-2 chains, all components of the alveolar basement membrane. Moreover, we identified other proteins enriched in IPF lung, such as matrix Gla protein and nephronectin, that have not previously been associated with human IPF, although evidence suggests that they may have a role in experimental fibrosis models (29, 30). Similarly, collagens III and VI and periostin (a matricellular protein downstream of IL-13 signaling and implicated in lung fibrosis ) were enriched in IPF matrices.
To determine whether acellular lung matrices would influence the phenotype of exogenously added cells, we next cultured primary human lung fibroblasts (1 × 105) in acellular normal and IPF ECM slices (prepared as in Methods). Cells were allowed to adhere for 30 minutes before adding media to the wells and were then cultured for 48 hours before harvesting cell lysates for Western analysis.
Fibroblasts cultured in such a manner migrate into the 3D matrix (Figure E1). Importantly, we observed a significant phenotypic difference in cells plated on the two different matrices. Whereas fibroblasts in normal lung ECM retained a phenotype characterized by low α-SMA production, cells in the IPF lung ECM demonstrated a significant increase in the production of α-SMA, consistent with myofibroblast differentiation (22) (Figure 9A). Immunohistochemistry corroborated this finding (Figure 9B).
To determine whether the profibrotic mediator TGF-β could account for these findings, we next cultured primary normal human lung fibroblast-seeded acellular matrices in transwells opposite mink lung epithelial cells stably expressing the firefly luciferase gene under the control of the PAI-1 promoter (a reporter system to identify TGF-β activity ). Surprisingly, we observed no difference in TGF-β activity between normal and IPF tissues (Figure 10A), a finding that was corroborated by both enzyme immunoassay (EIA) measurement of TGF-β in acid-activated ECM slices showing no difference in TGF-β levels (Figure 10B) and the lack of effect on fibroblast α-SMA expression in fibroblasts cultured in IPF ECM in the presence of A83-01, an ALK-5 inhibitor (Figure 10C). Together, these data indicate that IPF lung matrix is capable of inducing myofibroblast differentiation in part through TGF-β–independent mechanisms and is consistent with recent literature (33).
The ECM is dynamic tissue providing structural support and spatial and temporal contextual cues to cells during many processes, including embryonic development, wound repair, and homeostasis. A great deal has been learned about the role of ECM in health and disease using standard cell culture techniques, although it is clear that these approaches mimic neither the compositional heterogeneity of ECM in vivo nor the dimensionality of tissues. As a result, data gleaned using traditional approaches may not accurately reflect cell behavior that occurs in vivo. Along the same lines, experimental animal models using transgenic approaches have helped to elucidate the role of many ECM proteins in various disease states (34–36). Unfortunately, animal models do not always recapitulate human disease, raising questions about the ability to translate results from animal to human. For example, anatomic differences in the lung between humans and rodents, such as lobar distribution, branching patterns, presence of bronchial submucosal glands, and goblet cells (37), likely substantially influence results. Indeed, these factors may account for the lack of effective therapies for patients with idiopathic pulmonary fibrosis; numerous compounds that are efficacious in rodent models of lung fibrosis are ineffective in the human disease (11).
To overcome some of these limitations, we have developed a fully humanized novel model for culturing cells in vitro in an organ- and disease-specific manner. Our approach retains the 3D spatial orientation, relative stiffness, complex ECM composition, and architecture found in the human lung; it is likely that these features will be maintained regardless of the organ tested. In fact, prior studies using decellularized heart (15), bone (38), and murine lung (39) as matrices to support regeneration imply these features remain true to the native organ, although this has not been directly shown previously. Our model also provides physiologically relevant tissue stiffness and ECM composition, as compared with the standard planar rigidity of tissue culture plastic. Moreover, it allows investigators to study the roles of ECM and cellular processes in a disease-specific manner, which is a major advance over currently used techniques.
In addition to defining a new model culture system, our data also show marked differences between the normal and (IPF) disease state, which may provide insight into disease pathogenesis. For example, we observed that IPF lung ECM markedly expresses glycosaminoglycans, matrix Gla protein, and microfibrillar-associated proteins. These observations are notable in that, although heparan sulfate– and chondroitin sulfate–containing glycosaminoglycans are known to be produced by various lung cell types (40), their presence in IPF has only been indirectly shown. Similarly, matrix Gla proteins, although identified in TGF-α–induced rodent lung fibrosis (29), have not previously been implicated in human lung fibrosis.
Studies of tissue fibrosis in general, and IPF specifically, clearly implicate the profibrotic effects of TGF-β (41–44). Thus, there has been an intense effort to delineate the role of TGF-β in IPF and other fibrosing disorders. These efforts suggest that TGF-β promotes myofibroblast differentiation (45), the fibroblast antiapoptotic phenotype (46), epithelial–mesenchymal transition (47), and ECM production (48). However, our current work suggests that myofibroblast differentiation (as measured by expression of α-SMA and cellular fibronectin) occurs at least partially independent of TGF-β in our system. Whether this finding applies to the true in vivo condition is not known, because our data show (Table E2) that latent TGF-β–binding proteins-2 and -4 are stripped away during the decellularization process. Regardless, evidence supports the notion that myofibroblast differentiation in the absence of TGF-β signaling can be induced in fibroblasts merely by altering the stiffness of the underlying substrate (49). Our current work is highly concordant with a seminal finding by Liu and colleagues (49), namely that fibroblasts seeded on polyacrylamide-crosslinked hydrogels maintained a quiescent phenotype or underwent apoptosis at low stiffness levels (0.1 to ~ 3 kPa), whereas higher stiffness levels (20–50 kPa) induced an appearance of activation characterized by accumulation of cells aligned in parallel clusters (49). This is very similar to our own findings (Figure 9), in which fibroblasts cultured in stiffer IPF matrices clearly adopted an activated myofibroblast phenotype.
The pathogenesis of IPF remains incompletely understood. In part, this is due to the lack of an animal model that faithfully recapitulates the human disease (12). However, the bleomycin model (in which a dose of intratracheal bleomycin is instilled in rodents and fibrosis is measured 21–35 d later) remains one of the most widely used to address mechanistic questions about IPF. Although this model shares some features with human fibrosis, its complete reliance on the preceding inflammatory injury makes it a less-than-ideal model to study IPF. As proof of this shortcoming, many therapeutic agents that are shown to be successful in bleomycin-induced lung fibrosis are often ineffective in patients with IPF (50, 51). Thus, new models that more closely resemble the human disease are necessary to further our understanding. The in vitro model described herein provides such a tool to begin to dissect, mechanistically, the contribution of the ECM to IPF pathogenesis. Moreover, we believe this model can be adapted to other organs and diseases as a means to enhance the translatability of biomedical research.
The authors thank the patients and families who donated lung tissue for this study.
Supported by National Institutes of Health grants R01 HL085083 (E.S.W.) and U01 HL111016 (E.S.W. and L.E.N.), the Drews Sarcoidosis Research Fund (E.S.W.), the Martin Edward Galvin Fund for Idiopathic Pulmonary Fibrosis Research (E.S.W.), and the Health Effects Institute at Georgia Institute of Technology (T.H.B.). A.J.B. and S.A.M. are supported by National Institutes of Health training grant T32 HL07749.
Author Contributions: A.J.B. performed experiments and analyzed data. R.H., A.M.C., A.A.D., S.A.M., J.L.T., and K.W. performed experiments and provided intellectual input. V.F.F. and T.H.B. performed atomic force microscopy experiments and analyzed data. B.B.M., F.J.M., J.C.H., and L.E.N. provided critical intellectual guidance of experiments and assisted with data analysis. E.S.W. conceptualized the project, performed experiments, analyzed data, and wrote the manuscript.
This article has an online supplement, which is accessible from this issue’s table of contents at www.atsjournals.org
Originally Published in Press as DOI: 10.1164/rccm.201204-0754OC on August 30, 2012