Little is known about the relative importance of monocyte and tissue-resident macrophages in the development of lung fibrosis. We show that specific genetic deletion of monocyte-derived alveolar macrophages after their recruitment to the lung ameliorated lung fibrosis, whereas tissue-resident alveolar macrophages did not contribute to fibrosis. Using transcriptomic profiling of flow-sorted cells, we found that monocyte to alveolar macrophage differentiation unfolds continuously over the course of fibrosis and its resolution. During the fibrotic phase, monocyte-derived alveolar macrophages differ significantly from tissue-resident alveolar macrophages in their expression of profibrotic genes. A population of monocyte-derived alveolar macrophages persisted in the lung for one year after the resolution of fibrosis, where they became increasingly similar to tissue-resident alveolar macrophages. Human homologues of profibrotic genes expressed by mouse monocyte-derived alveolar macrophages during fibrosis were up-regulated in human alveolar macrophages from fibrotic compared with normal lungs. Our findings suggest that selectively targeting alveolar macrophage differentiation within the lung may ameliorate fibrosis without the adverse consequences associated with global monocyte or tissue-resident alveolar macrophage depletion.
Van Furth and colleagues suggested an adult monocytic origin of all macrophages, but it is now clear that many tissue-resident macrophages, including alveolar macrophages in the lung, are self-renewing populations that arise from fetal progenitors and require minimal input from circuiting adult monocytes in a healthy environment (Guilliams et al., 2013; Yona et al., 2013; Scott et al., 2014; Kopf et al., 2015; van de Laar et al., 2016). In response to macrophage depletion, monocytes are recruited to the lung, where the microenvironment shapes them into cells that closely resemble tissue-resident alveolar macrophages (Landsman and Jung, 2007; Hashimoto et al., 2013; Lavin et al., 2014; Gibbings et al., 2015). Accordingly, during homeostasis, the lung harbors at least one and perhaps two ontologically distinct populations of alveolar macrophages, referred to as tissue-resident alveolar macrophages (TR-AMs) and monocyte-derived alveolar macrophages (Mo-AMs). Our current understanding of the role of alveolar macrophages is almost exclusively based on studies conducted in healthy animals during homeostasis, but the differentiation of monocytes into alveolar macrophages may differ in the local microenvironment resulting from tissue injury or fibrosis.
It has long been recognized that depletion of circulating monocytes (e.g., using Ccr2−/− mice or via the systemic administration of liposomal clodronate) ameliorates fibrosis severity in the lung and other organs (Moore et al., 2001; Gibbons et al., 2011; Wynn and Vannella, 2016). To explain these findings, investigators have suggested that during injury, monocytes rapidly differentiate into Mo-AMs and that both TR-AMs and Mo-AMs are polarized toward a profibrotic or “M2” phenotype in the fibrotic lung (Zhou et al., 2014). This “macrophage polarization” model is based on limited evidence and is conceptually incomplete (Xue et al., 2014; Nahrendorf and Swirski, 2016). For example, even in the healthy lung, alveolar macrophage differentiation is a slow process that unfolds over a time course longer than most fibrosis models (weeks), raising the possibility that incompletely differentiated alveolar macrophages rather than “polarized” alveolar macrophages contribute to fibrosis (Lavin et al., 2014; Gibbings et al., 2015). Furthermore, the relative contributions of Mo-AMs compared with TR-AMs to the development of fibrosis is not known. Finally, it is not clear whether alveolar macrophages are reconstituted after injury through the differentiation of recruited Mo-AMs or by the proliferation of TR-AMs and, if the former, whether these ontologically distinct populations are functionally different. Answers to these questions are important for the design of monocyte/macrophage–targeted therapies. In particular, monocyte-depletion strategies are limited by the requirement for a continuous supply of monocyte-derived cells for homeostasis in the gut and other tissues; therefore, it is critical to know whether selectively targeting Mo-AMs after they have lost canonical monocyte markers can ameliorate fibrosis. Similarly, strategies that nonselectively deplete both TR-AMs and Mo-AMs might paradoxically worsen fibrosis by inducing further recruitment of monocytes and threaten tissue homeostasis through the loss of TR-AM function (Janssen et al., 2011; Gibbings et al., 2015).
To address these questions, we developed a novel lineage tracing system in mice to unambiguously identify Mo-AMs and TR-AMs during the development of fibrosis and over the subsequent life span of the animal. We used this system to show that lung fibrosis was ameliorated when Mo-AMs were driven to necroptosis during their differentiation without affecting monocytes, thereby establishing a causal link between Mo-AMs and the pathobiology of fibrosis. In contrast, depletion of TR-AMs before the induction of lung fibrosis did not alter fibrosis severity. In 14-mo-old untreated mice, >95% of AMs were TR-AMs, but in mice treated with bleomycin at 4 mo of age, the alveolar macrophage pool was comprised of both TR-AMs and Mo-AMs, suggesting that a subpopulation of Mo-AMs persists after the resolution of fibrosis. Using transcriptomic profiling of monocyte-derived populations collected over the course of fibrosis and its resolution, we show that monocyte to macrophage differentiation unfolds slowly over weeks. Genes identified by others as causally linked to the development of fibrosis were most highly expressed in Mo-AMs early during their differentiation and were progressively down-regulated as Mo-AMs differentiated into mature alveolar macrophages. TR-AMs and Mo-AMs showed significant differences in profibrotic gene expression during fibrosis; however, 10 mo after the injury, these differences were no longer evident. Human homologues of many of the profibrotic genes expressed in Mo-AMs during bleomycin-induced fibrosis were differentially expressed in flow-sorted alveolar macrophages obtained from the lungs of patients with fibrotic lung disease. These results reveal remarkable heterogeneity in alveolar macrophage function during lung fibrosis with important implications for the design of targeted therapy.
In the absence of tissue injury, monocyte to macrophage differentiation is thought to be regulated by epigenetic changes in response to factors present in the lung microenvironment (Lavin et al., 2014). Depletion of circulating monocytes using Ccr2−/− mice or the administration of liposomal clodronate reduces fibrosis severity, implicating monocyte-derived cells in the development of fibrosis (Moore et al., 2001; Gautier et al., 2012; Lavin et al., 2014; Gibbings et al., 2015). However, we know little about the importance of monocytes or monocyte-derived cells after they leave the circulation. We have previously shown that 5 d after administration of bleomycin, the number of interstitial macrophages (identified as CD64+Silgec F−) was increased, whereas the number of alveolar macrophages (CD64+Siglec F+) was reduced. In contrast, during the fibrotic phase (day 21 post–bleomycin administration), the number of interstitial macrophages decreased and the number of alveolar macrophages increased because of the appearance of alveolar macrophages characterized by lower expression of Siglec F (Siglec Flow; Fig. 1, A and B; Misharin et al., 2013). To determine whether this new population of cells represented an expansion of TR-AMs or the recruitment of Mo-AMs, we developed a lineage tracing system that allows us to distinguish alveolar macrophage ontogeny during injury and over the subsequent life span of the animal (Fig. 1 C). Using this system, we showed that the increase in alveolar macrophages during fibrosis was completely attributable to monocyte-derived cells (i.e., Mo-AMs). Furthermore we found that differential expression of Siglec F reliably distinguished Mo-AMs and TR-AMs over the course of bleomycin-induced fibrosis (Fig. 1, B, E, and F; Misharin et al., 2013), and we used this gating strategy (Fig. S1) to identify TR-AMs and Mo-AMs in subsequent studies.
We next sought to determine whether Mo-AMs, TR-AMs, or both contributed to the development of fibrosis. First, to target Mo-AMs, we used mice with targeted depletion of capsase-8 in macrophages (Cuda et al., 2014, 2015). Caspase-8 is a cysteine protease that serves the dual function of activating the extrinsic apoptotic pathway in response to death ligands and suppressing receptor-interacting protein kinase (RIPK)–mediated necroptosis during development and inflammation (Lu et al., 2014; Salvesen and Walsh, 2014). We deleted Casp8 from different monocyte/macrophage populations using two independent Cre drivers (CD11cCreCasp8flox/flox and LysMCreCasp8flox/flox) that both efficiently target alveolar macrophages. We then administered intratracheal bleomycin to Casp8flox/flox, CD11cCreCasp8flox/flox, and LysMCreCasp8flox/flox mice and compared the severity of fibrosis after 21 d. We found that although a substantial fraction of control Casp8flox/flox mice died in response to bleomycin, survival was improved in both CD11cCreCasp8flox/flox and LysMCreCasp8flox/flox animals (Fig. 2 A). Furthermore, the severity of fibrosis was significantly worse in Casp8flox/flox animals than in CD11cCreCasp8flox/flox and LysMCreCasp8flox/flox animals (Fig. 2, B–D). We confirmed these results in another model of lung fibrosis induced by the intratracheal administration of an adenoviral vector encoding an active form of TGF-β (Lam et al., 2014; Morales-Nebreda et al., 2015). Mice with macrophage-targeted deletion of Casp8 showed attenuated Ad-TGF-β–induced fibrosis (Fig. S2, A–C). One model for alveolar macrophage development involves the sequential differentiation of monocytes into interstitial macrophages (IMs), which in turn differentiate into alveolar macrophages, although it is also possible that monocytes differentiate directly into alveolar macrophages. We observed reduced numbers of Mo-AMs and increased numbers of IMs in CD11cCreCasp8flox/flox and LysMCreCasp8flox/flox animals when compared with Casp8flox/flox animals (Fig. 2, E and F; and Fig. S2, D and E), suggesting that Mo-AMs are lost during the differentiation of IMs or monocytes.
Our findings could be explained by the induction of necroptosis in Mo-AMs as they differentiate in the lungs of Casp8-deficient mice during fibrosis. If so, we reasoned that deletion of RIPK3 in CD11cCreCasp8flox/flox and LysMCreCasp8flox/flox animals should restore the Mo-AM population and “rescue” fibrosis. To test this hypothesis, we generated Casp8flox/flox, CD11cCreCasp8flox/floxRIPK3−/−, and LysMCreCasp8flox/floxRIPK3−/− mice and treated them with bleomycin. The loss of RIPK3 restored the population of Mo-AMs to control levels, and the severity of fibrosis in Casp8flox/flox, CD11cCreCasp8flox/floxRIPK3−/−, and LysMCreCasp8flox/floxRIPK3−/− mice was similar to that seen in Casp8flox/flox animals (Fig. 2, A–F). These data suggest that RIPK3 is activated during monocyte to macrophage differentiation. To exclude a functional role for RIPK3 in alveolar macrophage differentiation, as has been suggested in bone marrow–derived cells (Dannappel et al., 2014; Moriwaki et al., 2014; Vlantis et al., 2016), we flow-sorted Mo-AMs from Casp8flox/flox, CD11cCreCasp8flox/floxRIPK3−/−, and LysMCreCasp8flox/floxRIPK3−/− mice and measured their transcriptomes using RNA sequencing (RNA-seq). We found that only a small number of genes were differentially expressed in all investigated cell types, suggesting RIPK activation is largely dispensable for alveolar macrophage differentiation and fibrosis in this model (Fig. 2, G and H).
Both CreLysM and CreCD11c should effectively target both Mo-AMs and TR-AMs; therefore, we wondered whether the protection we observed in the CD11cCreCasp8flox/flox and LysMCreCasp8flox/flox mice be partially attributed to deletion of Casp8 in TR-AMs. Both CD11cCreCasp8flox/flox and LysMCreCasp8flox/flox mice had normal numbers of TR-AMs in comparison to Casp8flox/flox mice (Fig. 3 A). Surprisingly, the levels of Casp8 mRNA in TR-AMs from naive CD11cCreCasp8flox/flox mice were similar to those in the Casp8flox/flox mice (Fig. 3 B). This suggested to us that RIPK-mediated necroptosis might occur as fetal monocytes differentiate into TR-AMs during the early postnatal period, thereby providing a selection pressure favoring TR-AMs that escape recombination. Consistent with this hypothesis, Casp8 was efficiently deleted in TR-AMs from the RIPK3−/− animals (Fig. 3 B). To confirm that the loss of Casp8 prevented the differentiation of monocytes into Mo-AMs, we compared the ability of bone marrow from Casp8-deficient donors to reconstitute the alveolar macrophage pool in the presence or absence of RIPK3. Compared with wild-type bone marrow, bone marrow from the Casp8-deficient donors on a wild-type background inefficiently repopulated the alveolar macrophage pool, as shown by ongoing recruitment of alveolar macrophages with low levels of Siglec F expression in CD11cCreCasp8flox/flox relative to Casp8flox/flox mice (Fig. 3, C and D). Deletion of RIPK3 (CD11cCreCasp8flox/floxRIPK3−/−) restored the monocyte to alveolar macrophage differentiation, as indicated by normal levels of Siglec F. We confirmed these findings by demonstrating a reduced ability of bone marrow from CD11cCreCasp8flox/flox relative to Casp8flox/flox mice to reconstitute the alveolar macrophage, but not circulating monocyte, pool in 1:1 completive chimera experiments (Fig. 3 E).
To exclude the possibility that the partial loss of Casp8 we observed in TR-AMs from CD11cCreCasp8flox/flox and LysMCreCasp8flox/flox relative to Casp8flox/flox mice might account for the protection these mice exhibited against bleomycin-induced fibrosis, we compared the transcriptomes of sorted naive TR-AMs from Casp8flox/flox, CD11cCreCasp8flox/flox and LysMCreCasp8flox/flox mice during fibrosis (day 14) and found that <1% of transcripts were differentially expressed (Fig. 3 F), suggesting that the protection we observed in CD11cCreCasp8flox/flox and LysMCreCasp8flox/flox mice against fibrosis was attributable to the loss of Mo-AMs. To exclude the possibility that TR-AMs also contribute to the development of fibrosis, we selectively depleted TR-AMs by administering liposomal clodronate intratracheally 3 d before the administration of bleomycin. The administration of liposomal clodronate effectively depleted TR-AMs without inducing the recruitment of neutrophils (Fig. 4, A and B). Although TR-AMs were reduced in intratracheal liposomal clodronate–treated animals over the course of bleomycin-induced fibrosis, there was no difference in the recruitment of Mo-AM, interstitial macrophages or in the severity of fibrosis (Fig. 4, C–G), suggesting TR-AMs do not contribute to the development of fibrosis in this model.
Our data suggest that Mo-AMs contribute disproportionately to the development of lung fibrosis. To better understand the function of these cells during lung fibrosis, we flow-sorted monocytes, interstitial macrophages, Mo-AMs, and TR-AMs 14 and 19 d after the administration of bleomycin and analyzed their gene expression using RNA-seq (Fig. 5 A). Principal-component analysis (PCA) of the differentially expressed genes showed clustering of the replicates according to cellular populations, suggesting that flow cytometry identifies differences in cellular phenotypes that can be reproducibly detected at the level of the transcriptome (Fig. 5 B; Fig. S3, A and B; and Table S1). We then performed k-means clustering of the entire transcriptional dataset (12,968 genes), which revealed five distinct clusters of genes (Fig. 5 C, Fig. S3 C, and Tables S2 and S3). Clusters I and II contained core gene signatures that were progressively up-regulated and down-regulated during monocyte to alveolar macrophage differentiation, respectively. Genes in these clusters were similar to those others have associated with monocyte to alveolar macrophage differentiation during homeostasis (Fig. S3 G; Lavin et al., 2014; Schneider et al., 2014). Genes in clusters IV and V were up-regulated early and late during fibrosis, respectively. Early genes (cluster IV) included those involved in biosynthetic pathways, whereas later genes (cluster V) included those involved in epigenetic regulation of gene expression, consistent with the described epigenetic regulation of alveolar macrophage differentiation in the lung (Lavin et al., 2014). These data show that monocyte to alveolar macrophage differentiation unfolds slowly over the course of fibrosis.
Because Mo-AMs substantially outnumber TR-AMs during the development of fibrosis, it is possible that both populations respond to fibrosis similarly, but Mo-AMs predominate numerically. To address this question, we compared the expression of profibrotic genes in Mo-AMs and TR-AMs during the development of fibrosis. Cluster III included genes and pathways that are not known to be expressed in alveolar macrophages during homeostasis, are not associated with processes involved in cellular differentiation, and are associated with fibrosis (Fig. 5 C). These genes were differentially expressed in monocyte-derived cells (interstitial macrophages and Mo-AMs compared with TR-AMs). We performed pairwise comparisons of Mo-AMs, interstitial macrophages, and TR-AMs collected 14 or 19 d after bleomycin with naive TR-AMs (Fig. 6 A and Fig. S4 A). 14 d after the administration of bleomycin, we identified 3,232 genes differentially expressed between TR-AMs and naive TR-AMs and 5,039 genes differentially expressed between Mo-AMs and naive TR-AMs (FDR P < 0.05; Fig. 6 A). Functionally similar genes were differentially expressed 19 d after the administration of bleomycin, but the number of differentially expressed genes was smaller (Fig. S4 A). Gene ontology analysis of the differentially expressed genes revealed pathways involved in cell adhesion, cell migration, and fibrosis (Fig. 6, B and C; and Fig. S4 B). Indeed, of the 387 genes in cluster III that were differentially expressed in Mo-AMs compared with TR-AMs, 130 were linked to fibrosis in PubMed (Fig. 6 D). For 29 of the genes, other groups have reported that mice lacking the gene were protected against fibrosis in mouse models, including lung (Moore et al., 2006; Flechsig et al., 2010; Riteau et al., 2010; Aschner et al., 2014; Shibata et al., 2014; Tan et al., 2014; Vuga et al., 2014; Agassandian et al., 2015; Brodeur et al., 2015; Wang et al., 2015; Lv et al., 2016), heart (Li et al., 2009; Fan et al., 2014; Nishikido et al., 2016; Toba et al., 2016), liver (Seki et al., 2009; Kim et al., 2010; Madala et al., 2010; Fiorotto et al., 2016; Marí et al., 2016), kidney (Grgic et al., 2009; Lai et al., 2014; Yan et al., 2016), bone (Ishizuka et al., 2011), eye (Chan et al., 2013), muscle (Sinadinos et al., 2015), and skin (Wang et al., 2015). Genes exclusively expressed in TR-AMs included those involved in lipid metabolic processes and blood microparticle formation (Fig. 6 B). We performed a functional network analysis of the top 100 differentially expressed genes in cluster III using DAVID and Fgnet tools (Fontanillo et al., 2011; Aibar et al., 2015). This analysis revealed 14 highly overlapping metagroups with a core set of hub genes implicated in fibrotic signaling processes (e.g., Adam8, Arg1, Apoe, Itga6, Mfge8, Mmp12, Mmp13, Mmp14, and Pdgfa; Fig. 6 E and Table S4).
Macrophage polarization refers to distinct sets of inflammatory (M1) or fibrotic (M2) genes that are expressed by macrophages in cell culture that are first induced toward differentiation and then treated with LPS and IFN-γ or IL-4, respectively (Martinez and Gordon, 2014; Murray et al., 2014). It has been suggested that polarization of alveolar macrophages toward a profibrotic or “M2” phenotype contributes to the development of fibrosis (Murray et al., 2010; Larson-Casey et al., 2016). Our transcriptional data allowed us to directly test this hypothesis by examining the expression of 20 “M1” and 44 M2 genes during bleomycin-induced lung fibrosis (Misharin et al., 2014). Interestingly, we found that both Mo-AMs and TR-AMs up-regulated M1 and M2 genes in response to bleomycin without a discernible shift in gene expression toward a M1 or M2 phenotype in either cell population. When compared with naive TR-AMs, 25% of M1 genes and 18% of M2 genes were differentially higher in interstitial macrophages and Mo-AMs during fibrosis (Fig. 6 F and Fig. S4 C). In addition, many M2 genes were higher in naive TR-AMs than in TR-AMs isolated during fibrosis (Pparg, Tgm2, Cebpb, and Tgfb1). Collectively, these results show that Mo-AMs and TR-AMs alter their expression of both M1 and M2 genes during bleomycin-induced fibrosis and argue against an exclusive role for macrophage polarization toward a M2 phenotype to explain their contribution.
It is not known whether the alveolar macrophage pool is restored after the resolution of injury through the persistence of Mo-AMs recruited to the lung or whether TR-AMs proliferate to repopulate the alveolar macrophage pool. Bleomycin-induced fibrosis in young mice slowly resolves over 60 d (Hecker et al., 2014). Accordingly, we treated shielded bone marrow chimeric mice with bleomycin and measured the ratio between Mo-AMs and TR-AMs 10 mo later. Extending the findings from several groups (Guilliams et al., 2013; Hashimoto et al., 2013; van de Laar et al., 2016), we found that TR-AMs were remarkably stable, constituting >95% of alveolar macrophages at 14 mo of age (Fig. 7, A and B). In contrast, in mice that received bleomycin at 8 wk post–bone marrow transfer (4 mo of age), 10 mo later, ∼50% of the alveolar macrophages were monocyte derived. To determine whether this finding was unique to bleomycin, we treated a separate cohort of shielded bone marrow chimeric mice with influenza A (A/WSN/33) virus, at a dose that was associated with severe injury but was not lethal, and obtained similar results (Fig. 7 B). This finding suggests that a severe injury early in life can permanently reshape the alveolar macrophage landscape with respect to its developmental origins.
Next, we sought to determine whether the differences in gene expression between TR-AMs and Mo-AMs during fibrosis persisted over the life span. 10 mo after bleomycin treatment or influenza A infection, TR-AMs and Mo-AMs expressed similar levels of Siglec F and were no longer distinguishable by flow cytometry (Fig. 7 C). A comparison of the transcriptomes of TR-AMs and Mo-AMs 10 mo after bleomycin administration revealed only 330 differentially expressed genes (Fig. 7 D and Table S5).
The role of immune cell infiltration in the development of human lung fibrosis remains controversial. Therefore, we used flow cytometry to assess the number of alveolar macrophages in distal lung tissue obtained from the explanted lung, and a biopsy of the donor lung, in patients with lung fibrosis undergoing lung transplantation. We observed an expansion of alveolar macrophages in lung tissue samples from patients with fibrosis compared with the donor controls (Fig. 8, A and B; Bharat et al., 2016; Desch et al., 2016; Yu et al., 2016). Expansion of the alveolar macrophage pool was confirmed by immunostaining lung sections with the alveolar macrophage markers CD206 and CD169 (Fig. 8 C). As we found that Mo-AMs are causally linked to the development of lung fibrosis in response to bleomycin and Ad-TGF-β, we sought to determine whether the human homologues of the mouse profibrotic genes we identified in Mo-AMs during bleomycin-induced fibrosis (Fig. 5 C, cluster III) were up-regulated in alveolar macrophages isolated from the lungs of patients with lung fibrosis. We flow-sorted alveolar macrophages from patients with end-stage lung fibrosis attributed to one of several etiologies (idiopathic pulmonary fibrosis, scleroderma-associated interstitial lung disease, polymyositis-associated interstitial lung disease, and mixed connective tissue disease) and alveolar macrophages from a small biopsy specimen of normal lung obtained from the donor, and we analyzed their transcriptome using RNA-seq. Of the genes differentially expressed in cluster III (Table S2), homologues of 61 genes were also differentially expressed between alveolar macrophages isolated from patients with lung fibrosis compared with donors: 51 were up-regulated in patients with lung fibrosis, including APOE, ITGA6, FGFR1, MMP12, MMP14, PDPN, and SPARC (Fig. 8 D; normalized counts in Table S6), which suggests the existence of common profibrotic pathways in human and mouse macrophages as well as highly conserved profibrotic alveolar macrophage signature across heterogeneous clinical entities. Several genes that were up-regulated in mouse Mo-AMs were down-regulated in patients with lung fibrosis, and some of these genes have been implicated in lung repair after injury (e.g., CD163, MFGE8, LYVE1, and C1QB). A pathogenic role for alveolar macrophages and other inflammatory cells has been dismissed because a human clinical trial showed the administration of corticosteroids was shown to have no effect or slightly worsened clinical outcomes in patients with idiopathic pulmonary fibrosis (Raghu et al., 2012). To determine whether corticosteroids targeted alveolar macrophages during the development of fibrosis, mice were treated with dexamethasone (equivalent prednisone dose of 1 mg/kg/day, intraperitoneal) beginning simultaneous with the administration of bleomycin. The severity of bleomycin-induced lung fibrosis in dexamethasone- and sham-treated mice was similar (Fig. S5, A–C). Dexamethasone treatment did not significantly alter the depletion of TR-AMs or the recruitment of Mo-AM during fibrosis (Fig. S5, D and E), consistent with the suggestion that corticosteroid treatment has little or no effect on the population of macrophages.
We used a genetic lineage tracing system to show that Mo-AMs and TR-AMs play distinct roles during the development of lung fibrosis. In common mouse models of lung fibrosis, the deletion of Mo-AM after their recruitment to the lung markedly attenuated the severity of fibrosis, whereas the deletion of TR-AM had no effect on fibrosis severity. These findings are bolstered by transcriptomic profiles of monocyte and macrophage populations over the course of bleomycin-induced fibrosis, which reveal substantial differences in gene expression between Mo-AMs and TR-AMs over the course of lung fibrosis. Our findings suggest revision of our consideration of alveolar macrophages as a single population of cells in the design of macrophage-targeted therapies in lung fibrosis.
Our transcriptomic data of flow-sorted monocyte and macrophage populations from lung homogenates suggest that monocyte to alveolar macrophage differentiation represents a continuous down-regulation of genes typically expressed in monocytes and up-regulation of genes expressed in alveolar macrophages. These findings suggest that monocytes give rise to interstitial macrophages, which in turn give rise to Mo-AMs, and are consistent with published data suggesting that monocyte to macrophage differentiation results from epigenetic reprogramming driven by factors present in the local microenvironment (Landsman and Jung, 2007; Lavin et al., 2014). We observed that the protection against bleomycin-induced lung fibrosis in both CreCD11cCasp8floxflox and CreLysMCasp8floxflox mice relative to Casp8floxflox controls. This protection was associated with preserved numbers of interstitial macrophages but a marked reduction in Mo-AMs. Caspase-8 initiates apoptosis through caspase-3/7 activation and blocks necroptosis via suppression of RIPK1–RIPK3 signaling (Hutcheson and Perlman, 2008; Oberst et al., 2011). The genetic loss of RIPK3 restored the population of Mo-AMs, and sensitivity to bleomycin induced fibrosis in caspase-8–deficient mice, suggesting these cells undergo necroptosis as they differentiate from interstitial macrophages to Mo-AMs. These results are consistent with previous studies that have demonstrated the importance of monocytes in the development of bleomycin-induced fibrosis (Gibbons et al., 2011; Larson-Casey et al., 2016) but represent an important advance with therapeutic implications. Systemic depletion of monocytes or inhibition of pathways required for monocyte migration to sites of tissue injury is necessarily limited by toxicity, as monocyte migration is required to maintain some tissue resident macrophage populations (e.g., in the gut) and for the systemic maintenance of dendritic cells (Lavin et al., 2015). Our data suggest that targeting pathways required for alveolar macrophage differentiation after monocytes have begun the process of differentiation in the lung can prevent fibrosis. Because many of these pathways are exclusively important for alveolar macrophages (Nakamura et al., 2013; Lavin et al., 2014; Schneider et al., 2014; Sennello et al., 2017), targeting them is predicted to slow the development of Mo-AMs without affecting monocyte or macrophage populations in other tissues.
Although TR-AMs increased the expression of some profibrotic genes during fibrosis, this response was substantially muted relative to Mo-AMs. We found that depletion of TR-AMs using intratracheal liposomal clodronate before the administration of bleomycin had no effect on fibrosis, suggesting they are dispensable for the development of fibrosis in this model. Furthermore, the significant protection we observed in CreCD11cCasp8floxflox and CreLysMCasp8floxflox mice cannot be attributed to deletion of caspase-8 in TR-AMs, as we found the TR-AMs in these mice had preserved expression of Casp8, suggesting they escaped genetic recombination. Indeed, there were few detectable differences in the transcriptomes of TR-AMs from CreCD11cCasp8floxflox and CreLysMCasp8floxflox mice compared with Casp8floxflox controls. Competitive chimera experiments and bone marrow reconstitution with CreCD11cCasp8floxfloxRIPK3−/− animals suggest that in the absence of caspase-8, monocyte-derived cells (likely interstitial macrophages) undergo necroptosis as they differentiate into Mo-AMs. Although this suggests that RIPK is activated during alveolar macrophage differentiation, we were unable to identify a role for this pathway in fibrosis, as the transcriptomes of CreCD11cCasp8floxfloxRIPK3−/− Mo-AMs were similar to controls. Interestingly, although interstitial macrophages demonstrated even higher levels of profibrotic gene expression than Mo-AMs and their numbers were similar in CreCD11cCasp8floxflox and CreLysMCasp8floxflox mice compared with Casp8floxflox controls, protection against fibrosis was observed in the Casp8-deficient mice. This might reflect the relatively low abundance of interstitial macrophages in comparison to Mo-AMs, a presumed difference in their life span, or their anatomical localization in the interstitium versus alveolar space, respectively.
It has long been known that complete depletion of alveolar macrophages with ionizing radiation results in their replacement by Mo-AMs that closely resemble TR-AMs (Hashimoto et al., 2013; Lavin et al., 2014; van de Laar et al., 2016). Although others have previously shown that monocytes recruited during the tissue injury can give rise to long-living self-renewing tissue-resident-like cells in the skin (Merad et al., 2002), it is not known whether Mo-AMs recruited to the lung during injury persist in the lung or whether the alveolar macrophage population is restored via proliferation of TR-AMs. We found that in the absence of injury, TR-AMs were a remarkably stable population, with <5% of cells originating from the bone marrow in 14-mo-old mice. In contrast, 1 yr after the administration of bleomycin, when fibrosis has completely resolved, ∼50% of alveolar macrophages were monocyte derived. Although Mo-AMs and TR-AMs showed marked differences in the expression of profibrotic genes during the development of fibrosis, 1 yr after fibrosis, they were indistinguishable by flow cytometry and showed only small differences in their transcriptome. Our design does not allow us to determine whether these retained Mo-AMs harbor an epigenetic memory of the inflammatory and fibrotic environment in which they were generated that is distinct from TR-AMs and respond differently to subsequent environmental challenges (Ostuni et al., 2013). Nevertheless, these results show that environmental exposures early in life can induce heterogeneity in the ontologic origins of alveolar macrophages, offering a possible mechanism to explain the enhanced and persistent fibrosis reported with repeated doses of bleomycin and during aging (Degryse et al., 2010; Hecker et al., 2014).
In lung transplant recipients examined up to 5 yr after transplantation, investigators reported that the majority of alveolar macrophages originated from the donor, suggesting the presence of a stable population of TR-AMs in humans analogous to those found in mice (Nayak et al., 2016). The decline in lung function in patients with fibrosis is not continuous, and step declines are often preceded by the development of “exacerbations,” or an acute or subacute clinical worsening of shortness of breath and hypoxemia accompanied by inflammatory infiltrates on computed tomography imaging of the chest that suggest a role for immune cells in the pathogenesis of pulmonary fibrosis (Collard et al., 2007). However, a multicenter clinical trial of corticosteroids for the treatment of pulmonary fibrosis failed to show benefit, with a nonsignificant trend toward harm (Raghu et al., 2012). As a result of this trial and studies reporting that lung inflammation is sometimes worse in mice with genetic mutations that prevent the development of fibrosis, many concluded that inflammation is dispensable for the development of fibrosis (Munger et al., 1999; Budinger et al., 2006; Rock et al., 2011). Our data challenge these findings. We detected a substantial expansion in macrophages using newly described flow cytometry markers in distal lung tissue from a small number of patients with pulmonary fibrosis compared with normal lungs from the donor. Transcriptional profiling of these cells revealed substantial overlap in the expression of human homologues of the profibrotic genes up-regulated in Mo-AMs in mice exposed to bleomycin. This led us to examine the effect of corticosteroids on macrophages in lung fibrosis. Similar to the human trial, corticosteroids had no effect on the severity of bleomycin-induced lung fibrosis; however, we found that neither the TR-AM nor the Mo-AM population was affected by steroid treatment.
Our study has several limitations. Although the approach we used is a powerful tool to identify the sequence of transcriptional modifications that develop during monocyte to macrophage differentiation during disease, pulsed lineage tracing techniques will be required to follow an individual population of monocytes during their differentiation into Mo-AMs (Paul et al., 2015). Second, our flow cytometry approach cannot provide anatomical detail, except when antibodies for immunofluorescence or immunohistochemistry are available. For example, it is possible that Mo-AMs that persist in the lung after injury are differentially localized to areas where the microenvironment is subtly abnormal. Third, although our data suggest that the concept of M2 polarization overly simplifies the complex changes in gene expression that occur in both Mo-AMs and TR-AMs over the course of lung fibrosis (Nahrendorf and Swirski, 2016), it is possible that a subpopulation of Mo-AMs with one of these signatures or at different developmental stages contributed disproportionately to the fibrotic gene expression signature. Single-cell RNA-seq analysis will be required to exclude this possibility. Fourthly, the CreLysM and CreCD11c promoters both target dendritic cells. Therefore, although we did not see changes in the numbers of dendritic cells in the lungs of the caspase-8–deficient mice generated using these promoters, we cannot exclude the possibility that some of the effects we observed are attributable to dendritic cells. Finally, although we are unaware of previous studies showing an expansion of the alveolar macrophage pool in patients with lung fibrosis or previous studies of transcriptional profiling of these cells during fibrosis, we cannot determine whether they play a causal role in disease pathogenesis.
In summary, our results suggest that an appreciation of macrophage heterogeneity has important implications for disease pathogenesis and therapeutic development. Mo-AMs undergo dramatic transcriptional changes as they differentiate in the injured lung through a process that unfolds slowly over time. Mo-AMs express consistently higher levels of proinflammatory and profibrotic genes than TR-AMs, and selective depletion of Mo-AMs, but not TR-AMs, ameliorates the severity of lung fibrosis. Profibrotic gene expression in Mo-AMs harvested from mice during fibrosis is homologous to genes differentially regulated in alveolar macrophages from patients with lung fibrosis. These results challenge the consideration alveolar macrophages during lung fibrosis as a single population of M2 polarized macrophages and suggest that selectively targeting Mo-AMs after their commitment to an alveolar macrophage fate may ameliorate fibrosis without the off-target effects of therapies that deplete circulating monocytes. Our finding that Mo-AMs persist in the lung after the resolution of injury shows that environmental exposures can permanently change the origins of alveolar macrophages. Determining whether retained Mo-AMs and TR-AMs respond differently to a subsequent injury or differentially change their gene expression signature with age may offer a mechanism to understand the increased risk of lung fibrosis with advancing age.
Materials and methods
All human studies were approved by Northwestern IRB and Department of Defense Human Research Protection Office. All human subjects provided written informed consent before enrolment into the study. A biopsy-size piece of the donor lung tissue and a lobe of the explanted recipient lung were obtained at the time of the lung transplant. Tissues were processed for flow cytometry and histology as previously described (Bharat et al., 2016). 10 patients with fibrotic lung disease (2 patients with idiopathic pulmonary fibrosis, 5 patients with systemic sclerosis-associated interstitial lung disease, 1 patient with mixed connective tissue disease, 1 patient with interstitial pneumonitis, and 1 patient with pneumoconiosis) were included in flow cytometric analysis.
All mouse procedures were approved by the Institutional Animal Care and Use Committee at Northwestern University. The mouse strains C57BL/6, B6.SJL-Ptprca Pepcb/BoyJ (CD45.1) were used. All strains including wild-type mice are bred and housed at a barrier- and specific pathogen–free facility at the Center for Comparative Medicine at Northwestern University (Chicago, IL). Colonies are refreshed at least yearly with mice purchased from The Jackson Laboratory. 8–10-wk-old mice were used for all experiments, although some were aged up to 14 mo in our facility. Casp8flox/flox, CD11cCreCasp8flox/flox, LysMCreCasp8flox/flox, CD11cCreCasp8flox/floxRIPK3−/−, and LysMCreCasp8flox/floxRIPK3−/− mice were described previously (Cuda et al., 2014, 2015); these mice were backcrossed to C57BL/6 background for >10 generations. Littermate controls were used in all experiments, except for generation of competitive bone marrow chimeras. All procedures were approved by the Institutional Animal Care and Use committee at Northwestern University. Bleomycin-induced lung fibrosis, TGF-β–induced lung fibrosis, and influenza A virus infection (A/WSN/33) were performed and assessed as previously described (Budinger et al., 2006; Misharin et al., 2013; Morales-Nebreda et al., 2014, 2015; Sennello et al., 2017).
Tissue preparation and flow cytometry
Tissue preparation for flow cytometry analysis and cell sorting was performed as previously described (Misharin et al., 2013, 2014; Bharat et al., 2016). Blood was collected into EDTA-containing tubes via facial vein bleed (from live animals) or cardiac puncture (from euthanized animals). Whole blood was stained with fluorochrome-conjugated antibodies, and erythrocytes were then lysed using BD FACS lysing solution. For single-cell suspension obtained from tissues erythrocytes were lysed using BD Pharm Lyse, and cells were counted using Countess automated cell counter (Invitrogen); dead cells were discriminated using trypan blue. Cells were stained with eFluor 506 (eBioscience) viability dyes, incubated with FcBlock (BD), and stained with fluorochrome-conjugated antibodies (antibodies, clones, fluorochromes, and manufacturers were described in detail in our previous publications; Misharin et al., 2013, 2014; Bharat et al., 2016). Data were acquired on BD LSR II flow cytometer (for information regarding instrument configuration and antibody panels, see Misharin et al., 2013, 2014; Bharat et al., 2016). Compensation, analysis and visualization of the flow cytometry data were performed using FlowJo software (Tree Star). “Fluorescence minus one” controls were used when necessary to set up gates. Cell sorting was performed at Northwestern University RLHCCC Flow Cytometry core facility on SORP FACSAria III instrument (BD) with the same optical configuration as LSR II, using a 100-µm nozzle and 40 psi pressure.
Bone marrow chimeras and shielded bone marrow chimeras
Bone marrow chimeras were established by transferring 5 × 106 bone marrow cells isolated from C57BL/6 mice (this strain expresses CD45.2 alloantigen) into 8-wk-old lethally irradiated (single dose of 1,000 cGy γ-radiation using a Cs-137–based Gammacell-40 irradiator; Nordion) recipient mice (expressing CD45.1 alloantigen). Mice were maintained on autoclaved water supplemented with antibiotics (trimethoprim/sulfamethoxazole; Hi-Tech Pharmacal) for 4 wk after bone marrow transfer and then switched to normal housing regimen. CD45.2 to CD45.1 bone marrow chimeras were used for experiments 8 wk after bone marrow transfer. 8 wk after bone marrow transfer, >95% of all leukocytes and 100% of monocytes and neutrophils in peripheral blood were of donor origin. Bone marrow chimeras with thoracic shielding were used to assess the origin of pulmonary macrophages (TR-AMs vs. Mo-AMs) and were generated in a manner similar to that previously described (Janssen et al., 2011), with additional modifications (Misharin et al., 2014). To protect TR-AMs from radiation, we applied a uniform lead shield that covered the lungs during irradiation as previously described using a custom-made “Mousotron-5” apparatus. However, we found this approach was limited, as the residual recipient bone marrow in the shielded region results in incomplete chimerism; ∼20% of peripheral blood monocytes are still of recipient origin and cannot be distinguished from tissue-resident macrophages upon recruitment to the lung. This problem was corrected by the administration the myeloablative agent busulfan (30 mg/kg body weight; Sigma-Aldrich; Chevaleyre et al., 2013) 6 h after the irradiation, followed 12 h later by bone marrow infusion. 2 mo after the procedure, 100% of all alveolar macrophages in the shielded bone marrow chimeras were of recipient origin (CD45.2), whereas 100% of all circulating monocytes were of donor origin (CD45.1). Bone marrow chimeras with thoracic shielding were maintained on antibiotics for 4 wk as described above and then switched back to the normal housing regimen.
Gene expression profiling (RNA-seq) and bioinformatic analysis
In the overall design, monocyte and macrophage subpopulations were isolated from naive, bleomycin-, or influenza A virus–challenged mice at the indicated time points via FACS-sorting on BD FACSAria III instrument and RNA was extracted for subsequent transcriptomic analysis. For each genotype/condition, two to five biological replicates were used. Libraries for RNA-seq were prepared in 96-well plate format. For the bleomycin time-course experiment, 155 samples were equally distributed between two batches of the libraries preparation (78 and 77 samples), so each group/condition would be represented on each plate. For experiments using shielded chimeras with busulfan (40 samples), libraries were prepared as a single batch.
Monocyte and macrophage subpopulations were isolated on BD FACSAria III instrument, 0.5–1.5 × 105 cells were sorted into capture media (PBS, 2% BSA and 0.05% EDTA), immediately spun down and lysed in 350 μl of RLT-plus buffer (QIAGEN) supplemented with β2-mercaptoethanol. Total RNA was extracted using RNeasy Plus Mini kit (QIAGEN) and eluted in 35 μl of nuclease free water. RNA quality and quantity were assessed using Bioanalyzer 2100 or TapeStation 4200 instruments (Agilent Technologies). RNA-seq libraries were prepared in 96-well plate format using NEBNext Poly(A) mRNA Magnetic Isolation Module, NEBNext Ultra RNA Library Prep kit for Illumina and NEBNext Multiplex Oligos for Illumina (Dual Index Primers Set 1) according to the standard protocol. 50 ng total RNA was enriched for poly(A) mRNA using NEBNext Poly(A) mRNA Magnetic Isolation Module, the mRNA was subjected to chemical fragmentation in the presence of divalent cations at 94°C for 15 min, and cDNA was generated using random primers and ProtoScript II reverse transcription in the presence of mouse RNase inhibitor using the following program: 10 min at 25°C, 15 min at 42°C, and 15 min and 70°C. Second-strand DNA was generated using NEBNext Second Strand Synthesis Enzyme module (DNA polymerase I, RNase H, Escherichia coli DNA ligase) for 60 min at 16°C. The resulting double-stranded DNA was purified using 1.8× volume of SPRI beads (AMPure XP Beads; Beckman Coulter) or HighPrep PCR Beads (MagBio Genomics), eluted in 0.1× TE (Tris/EDTA) buffer, DNA ends were repaired using NEBNext End Prep Enzyme Mix (T4 PNK and T4 DNA polymerase) for 30 min at 20°C and 30 min and 65°C, followed by NEBNext adaptor for Illumina ligation using Blunt/TA Ligase master mix (T4 DNA ligase) for 15 min at 20°C and incubation with USER enzyme for 15 min at 37°C. After another round of purification using 1× volume of SPRI Beads, adaptor-ligated DNA was amplified using NEBNext Q5 Hot Start HiFi DNA polymerase in the presence of NEBNext multiplex oligos (dual index primers) for 14 cycles. PCR-enriched libraries were purified using a 0.9× volume of SPRI Beads, and their quality was assessed using Bioanalyzer 2100 or TapeStation 4200 instruments. Equimolar amount of libraries were pooled and sequenced on Illumina NextSeq 500 instrument (single end reads) using V1 chemistry high-output 150 cycles sequencing kit (bleomycin time-course experiment) or V2 chemistry 75 cycles high-output sequencing kit (aged-shielded bone marrow chimeras).
Computation intensive analysis was performed using “Genomics Nodes” on Northwestern’s High Performance Computing Cluster, Quest (Northwestern IT and Research Computing). Reads were demultiplexed using bcl2fastc, quality was assessed using FastQC, reads were trimmed and aligned to mm10 reference genome using TopHat2, read counts were associated with genes using the GenomicRanges (Lawrence et al., 2013), and differential gene expression was assessed using edgeR (Robinson et al., 2010; McCarthy et al., 2012) R/Bioconductor packages. Genes with less than one normalized read count across at least half of the samples were filtered from all analyses. Pearson correlation matrix and clustering heat maps were built using GENE-E (J. Gould, 2013; GENE.E: Interact with GENE-E from R. R package version 1.12.2; http://www.broadinstitute.org/cancer/software/GENE-E). Gene ontologies were evaluated using GOrilla (Eden et al., 2007, 2009). The dataset is available on the GEO database under accession number GSE82158.
For all fibrosis measurements, the data were analyzed using two-way ANOVA for repeated measurements with Bonferroni post-test to compare differences between the groups. All analyses were performed using GraphPad Prism version 7.00 (GraphPad Software). Data are shown as means ± SEM.
Online supplemental material
Fig. S1 shows the gating strategy used to isolate subsets of monocytes and macrophages during the course of bleomycin-induced lung fibrosis. Fig. S2 shows data demonstrating protection from Ad-TGFb-induced lung fibrosis in Casp8-deficient mice. Fig. S3 shows global transcriptome differences between populations of macrophages and relevant GO processes. Fig. S4 (related to Fig. 6) shows differences between TR-AMs and Mo-AMs at day 19 of bleomycin-induced lung fibrosis and absence of evidence supporting clear M1/M2 pattern of macrophage polarization in vivo. Fig. S5 shows glucocorticoids have no effect on the course of bleomycin-induced lung fibrosis. Table S1 (related to Fig. 5 and Fig. S3) contains Pearson’s correlation coefficients for transcriptomes of the individual macrophage populations. Table S2 (related to Fig. 5) contains k-means cluster assignment. Table S3 (related to Fig. 5 and Fig. S3) contains GO processes and statistics for each cluster. Table S4 (related to Fig. 6) contains results of Fgnet enrichment. Table S5 (related to Fig. 7) contains list of differentially expressed genes between TR-AMs and Mo-AMs 10 mo after bleomycin-induced lung fibrosis. Table S6 (related to Fig. 8) contains normalized gene counts for human homologues of the mouse genes from cluster III. Tables S1–S6 are included as Excel files.
A.V. Misharin is supported by National Institutes of Health (NIH) National Institute of Arthritis and Musculoskeletal and Skin Diseases grant AR061593, an American Thoracic Society/Scleroderma Foundation research grant, Department of Defense grant PR141319, and a BD Bioscience immunology research grant. P.A. Reyfman is supported by Northwestern University's Lung Sciences Training Program (5T32 HL076139-13). C.M. Cuda is supported by NIH grant AR064313. S. Chiu is supported by Northwestern University’s Transplant Surgery Scientist Training Program (NIH grant T32 DK077662) and the American Society for Transplant Surgery Foundation. A. Bharat is supported by NIH grant HL125940 and matching funds from Thoracic Surgery Foundation, a research grant from the Society of University Surgeons, and an American Association of Thoracic Surgery John H. Gibbon Jr. Research Scholarship. B.D. Singer is supported by NIH grant HL128867 and the Parker B. Francis Research Opportunity Award. J.I. Sznajder is supported by NIH grants AG049665, HL048129, HL071643, and HL085534. G. Multlu is supported by NIH grants ES015024 and ES025644. K. Ridge is supported by NIH grants HL079190 and HL124664. G.R.S. Budinger is supported by NIH grants ES013995, HL071643, and AG049665; Veterans Administration grant BX000201, and Department of Defense grant PR141319. H. Perlman is supported by NIH grants AR064546, AG049665, and HL134375 and funds provided by Mabel Greene Myers Chair. The Northwestern University Flow Cytometry Facility and Center for Advanced Microscopy are supported by a National Cancer Institute cancer center support grant P30 CA060553 awarded to the Robert H. Lurie Comprehensive Cancer Center. The Genomics Computing Cluster is jointly supported by the Feinberg School of Medicine, the Center for Genetic Medicine, and Feinberg’s Department of Biochemistry and Molecular Genetics, the Office of the Provost, the Office for Research, and Northwestern Information Technology and maintained and developed by Feinberg IT and Research Computing Group.
The authors declare no competing financial interests.
Author contributions: A.V. Misharin contributed to conceptualization, methodology, validation, formal analysis, investigation, resources, data curation, writing, visualization, supervision, project administration, and funding acquisition. L. Morales-Nebreda contributed to conceptualization, methodology, validation, formal analysis, investigation, data curation, writing, and visualization. P.A. Reyfman contributed to conceptualization, methodology, software, validation, formal analysis, investigation, data curation, writing, visualization, and supervision. T.J. Yacoub contributed to conceptualization, methodology, software, validation, formal analysis, data curation, writing, and visualization. C.M. Cuda contributed to conceptualization, methodology, validation, formal analysis, investigation, resources, data curation, and writing. H. Abdala-Valencia contributed to investigation, resources, data analysis, data curation, writing, visualization, and project administration. J.M. Walter contributed to formal analysis, investigation, data curation, writing, and visualization. A.C. McQuattie-Pimentel contributed to methodology, validation, formal analysis, investigation, data curation, and writing. C. Chen contributed to methodology, validation, formal analysis, investigation, and data curation. N. Joshi contributed to methodology, validation, formal analysis, investigation, and data curation. K.J.N. Williams contributed to methodology, validation, formal analysis, investigation, and data curation. M. Chi contributed to methodology, validation, formal analysis, investigation, and data curation. S. Chiu contributed to methodology, validation, formal analysis, investigation, data curation, and writing. F.J. Gonzalez-Gonzalez contributed to methodology, validation, formal analysis, investigation, and data curation. K. Gates contributed to methodology, validation, formal analysis, investigation, and data curation. A.P. Lam contributed to methodology, validation, formal analysis, investigation, and data curation. T.T. Nicholson contributed to investigation and data curation. P.J. Homan contributed to methodology, validation, formal analysis, investigation, and data curation. S. Soberanes contributed to methodology, validation, formal analysis, investigation, and data curation. S. Dominguez contributed to methodology, validation, and investigation. V.K. Morgan contributed to methodology, validation, investigation, and data curation. R. Saber contributed to validation, formal analysis, investigation, and data curation. A. Shaffer contributed to validation, formal analysis, investigation, and data curation. K.R. Anekalla contributed to methodology, software, validation, formal analysis, investigation, data curation, writing, and visualization. S.A. Marshall contributed to methodology, formal analysis, and investigation. M. Hinchcliff contributed to resources, data curation, writing, and supervision. A. Bharat contributed to resources, data curation, writing, and supervision. S.M. Bhorade contributed to resources, data curation, writing, and supervision. B.D. Singer and S. Berdnikovs contributed to methodology, validation, formal analysis, writing, and visualization. E.T. Bartom contributed to methodology, validation, software, and formal analysis. W.E. Balch contributed to conceptualization, resources, data curation, writing, supervision, and project administration. R.I. Morimoto contributed to conceptualization, resources, data curation, writing, and supervision, project administration. J.I. Sznajder contributed to conceptualization, resources, data curation, writing, supervision, and project administration. N.S. Chandel contributed to conceptualization, data curation, writing, and supervision. G.M. Mutlu contributed to conceptualization, resources, data curation, writing, and supervision. M. Jain contributed to conceptualization, data curation, writing, and supervision. C.J. Gottardi contributed to conceptualization, resources, writing, supervision, and project administration. K.M. Ridge contributed to conceptualization, resources, writing, supervision, and project administration. N. Bagheri contributed to conceptualization, methodology, formal analysis, resources, data curation, writing, and visualization. A. Shilatifard contributed to methodology, resources. GR.S. Budinger contributed to conceptualization, methodology, validation, formal analysis, investigation, resources, data curation, writing, visualization, supervision, project administration, and funding acquisition. H. Perlman contributed to conceptualization, methodology, validation, formal analysis, investigation, resources, data curation, writing, visualization, supervision, project administration, funding acquisition.
A.V. Misharin, L. Morales-Nebreda, and P.A. Reyfman contributed equally to this paper.
G.R.S. Budinger and H. Perlman contributed equally to this paper.