Steroid-induced fibroblast growth factors drive an epithelial-mesenchymal inflammatory axis in severe asthma

Abstract

Asthma and inflammatory airway diseases restrict airflow in the lung, compromising gas exchange and lung function. Inhaled corticosteroids (ICSs) can reduce inflammation, control symptoms, and improve lung function; however, a growing number of patients with severe asthma do not benefit from ICS. Using bronchial airway epithelial brushings from patients with severe asthma or primary human cells, we delineated a corticosteroid-driven fibroblast growth factor (FGF)–dependent inflammatory axis, with FGF-responsive fibroblasts promoting downstream granulocyte colony-stimulating factor (G-CSF) production, hyaluronan secretion, and neutrophilic inflammation. Allergen challenge studies in mice demonstrate that the ICS, fluticasone propionate, inhibited type 2–driven eosinophilia but induced a concomitant increase in FGFs, G-CSF, hyaluronan, and neutrophil infiltration. We developed a model of steroid-induced neutrophilic inflammation mediated, in part, by induction of an FGF-dependent epithelial-mesenchymal axis, which may explain why some individuals do not benefit from ICS. In further proof-of-concept experiments, we found that combination therapy with pan-FGF receptor inhibitors and corticosteroids prevented both eosinophilic and steroid-induced neutrophilic inflammation. Together, these results establish FGFs as therapeutic targets for severe asthma patients who do not benefit from ICS.

INTRODUCTION

Inhaled corticosteroids (ICSs) are the mainstay controller therapy for patients with asthma and the standard of care in long-term asthma management (1). Although ICSs are sufficient to limit symptoms, exacerbations, and lung function decline in the majority of patients, individuals with severe disease uncontrolled by maximum doses of ICS represent a growing unmet need (2). As we have learned during the past decade, severe asthma is a heterogeneous disease, with a variety of underlying biologies and endotypes identified (3, 4). Combination therapy with long-acting β agonists (LABA) or long-acting muscarinic antagonists is an effective rescue medication (1), and new therapeutics for severe asthma targeting underlying inflammatory pathways (5, 6) are showing promise; however, understanding how and why some individuals with severe asthma do not benefit from ICS may help guide more targeted approaches. A “steroid-insensitive” asthma phenotype has been described with patients presenting with more neutrophil-associated disease (7); however, the underlying mechanisms contributing to this phenotype are poorly understood and not met with current treatment options.

ICSs, including fluticasone propionate (F.p.) and budesonide, bind to the ubiquitously expressed cytoplasmic glucocorticoid receptor (GRα), liberating it from chaperone proteins such as Hsp90 (8). GRα molecules translocate into the nucleus, form homodimers, and bind to glucocorticoid response elements (GREs) in the promoter region of a variety of genes, leading to trans-activation or trans-repression (9, 10). GRα-mediated inhibition of specific inflammatory genes is therapeutically beneficial for the majority of patients with asthma (11). GRα-mediated trans-activation of target genes has been widely studied, with many examples of GRα-mediated amplification of inflammatory cytokines (12, 13), prevention of apoptosis (14), and up-regulation of surface receptors (15), either directly or synergistically (16) with other signals. It is noteworthy that some of the GRα-mediated anti-inflammatory properties have been attributed to the transactivation of inhibitory pathways (17, 18). Of particular interest, glucocorticoids have been reported to up-regulate fibroblast growth factors (FGFs) in a variety of cell types and across a variety of species (19–24). However, whether ICSs similarly regulate FGFs in the airways of individuals with asthma is unclear.

In mammals, 22 FGFs have been classified into paracrine and endocrine subfamilies based on sequence homology and developmental characteristics. Paracrine FGFs exert pleiotropic effects on a variety of cells after the formation of ternary complexes with heparan sulfate proteoglycans (HSPGs) and one of four FGF receptors (FGFRs). FGF function can be facilitated through one of three FGF binding proteins (FGFBPs), which bind to FGFs and HSPG, increasing the bioavailability of FGFs (25). Upon binding, receptor dimerization and phosphorylation of the essential docking protein, FGFR substrate 2 α (FRS2α) (26), occur, which orchestrate a signaling cascade involving rat sarcoma virus/mitogen-activated protein kinase (RAS/MAPK), phosphoinositide 3-kinases/protein kinase B (PI3K/AKT), phosphoinositide phospholipase Cγ (PLCγ), and signal transducer and activator of transcription (STAT) pathways (25). FGFR signaling is essential for development, from preimplantation (27) throughout organogenesis. In the lung, FGFs form an essential epithelial-mesenchymal signaling axis throughout development (28), with dysregulated expression of FGFs and FGFRs observed during a variety of pulmonary diseases (29), including severe asthma (30–36). In vitro epithelial cells (37), lung fibroblasts (38, 39), endothelial cells (40), and airway smooth muscle cells (41, 42) are all responsive to FGFs; however, whether dysregulated FGFs contribute to the pathogenesis of asthma through a similar epithelial-mesenchymal axis is unclear.

Here, we identified a suite of FGFs, FGFBPs, and FGFRs that were increased in bronchial brushings from a subset of individuals with severe asthma relative to healthy controls or individuals with moderate asthma. Using primary human bronchial airway epithelial cells (BAECs) in vitro, we found that two commonly prescribed ICS treatments, F.p. and budesonide, induced FGF1, FGF2, FGF4, and FGF18 expression and that F.p.-induced FGFs were sufficient to drive FGFR-dependent responses in primary human lung fibroblasts. These observations suggest that ICS-induced epithelial FGFs can function as mediators between epithelial and mesenchymal cells, as previously reported during lung development (28). Furthermore, F.p.-inducible FGFs were sufficient to induce a variety of disease-relevant pathogenic responses in fibroblasts and primary human precision-cut lung slices, suggesting that ICS-induced FGFs may contribute to the pathogenesis of asthma through the induction of FGFs. Testing this model in vivo, we demonstrated that F.p. treatment inhibited type 2 cytokine responses and airway eosinophilia after house dust mite (HDM) exposure while concomitantly inducing FGFs and enhancing airway neutrophilia. By disrupting FGFR signaling in fibroblasts, blocking granulocyte colony-stimulating factor (G-CSF), or degrading hyaluronan, we delineated a fibroblast-specific FGFR-dependent G-CSF and hyaluronan axis that drove airway neutrophilia and lymphocyte recruitment and retention in the lung. This supported the notion that ICS-driven FGFs are pathogenic. Therapeutically, combining F.p. with an FGFR inhibitor prevented F.p.-driven neutrophilia and maintained inhibition of eosinophilia, providing a scientific rationale for the development of therapeutics targeting FGF or FGFRs to prevent untoward steroid-driven responses in some individuals with neutrophilic severe asthma.

RESULTS

FGF expression is dysregulated in epithelial brushings from individuals with severe asthma

BAECs, which line the major connecting airways, receive signals from both the host and the environment and have proved to be an invaluable sample site to characterize and understand the nature and phenotype of airway-related diseases, particularly asthma (43–45). Expression of FGFs, which form an essential epithelial-mesenchymal signaling axis, has been observed in a variety of pulmonary diseases (29), including severe asthma (30–36). Testing the hypothesis that FGFs contribute to the severity of asthma, we analyzed gene expression data in bronchial brushings from three patient cohorts (table S1), including healthy volunteers (HV), individuals with moderate asthma (Mod), and individuals with severe asthma (Severe) (43, 45–47). We identified a suite of FGFs (FGF1, FGF2, FGF3, FGF4, FGF7, FGF8, FGF9, FGF16, and FGF18) that were elevated in bronchial brushings from a subset of individuals with severe asthma (Fig. 1), relative to healthy volunteers or individuals with moderate asthma. In addition to FGFs, we also found elevated expression of the FGFBPs FGFBP1 and FGFBP3 as well as the FGFRs FGFR1 and FGFR4 in individuals with severe asthma (fig. S1, A and B). FGFBP2 expression was unchanged, and FGFR2 and FGFR3 expression was down-regulated.

Fig. 1. Expression of epithelial FGFs is dysregulated in severe asthma.

Transcriptional data are shown for bronchial airway epithelial brushings isolated from healthy volunteers (HV), patients with moderate asthma (Mod), or patients with severe asthma (Severe), as previously described (43, 45, 47). Data are presented as log₂ relative expression. Data were analyzed using a Kruskal-Wallis test with Dunn’s multiple comparisons test. *P < 0.05.

Interleukin-13 (IL-13) regulates a suite of responses in the bronchial airway epithelium, including expression of periostin (45). In individuals with asthma, elevated FGFs did not correlate with POSTN expression (fig. S1C), suggesting that FGFs are not regulated by IL-13. Furthermore, in a separate cohort of individuals with moderate asthma with high or low IL-13–associated responses (45), we did not observe any differences in FGFs between individuals characterized as IL-13^Hi or IL-13^Lo, indicating that FGFs are not enriched in individuals with IL-13–associated asthma (fig. S1D). Likewise, we observed only minor differences in FGFBPs or FGFRs between individuals characterized as IL-13^Hi or IL-13^Lo (fig. S1, E and F). To further explore the relationship between FGFs and IL-13, we analyzed bronchial brushings from individuals with severe asthma treated before, and 12 weeks after, lebrikizumab (anti–IL-13) monoclonal antibody (mAb) treatment (48). There was no appreciable change in epithelial expression of FGFs, FGFBPs, or FGFRs after lebrikizumab treatment (fig. S1G), further suggesting that FGFs, FGFBPs, and FGFRs are uncoupled from IL-13–associated asthma and may represent a distinct endotype in severe asthma.

Corticosteroids regulate expression of FGFs in primary human BAECs grown at air-liquid interface

It has been widely reported that glucocorticoids can regulate FGFs in a variety of cell types and across a variety of species (19–22). ICSs are first-line options for the management of asthma, with individuals with severe asthma often prescribed greater than 500 μg per day (49). We therefore asked whether the commonly prescribed ICS, F.p. or budesonide, regulated FGFs in primary human BAEC grown at air-liquid interface (BAEC-ALI). When added to the apical surface of BAEC-ALIs, F.p. induced the secretion of FGF1, FGF2, FGF4, and FGF18 (Fig. 2A), without changing cellular metabolism, survival, or density (fig. S2A). F.p. had little impact on FGF3, FGF7, FGF8, FGF9, or FGF16 (Fig. 2A).

Fig. 2. Corticosteroids regulate expression of FGFs in primary human BAECs grown at air-liquid interface and in precision-cut lung slices.

(A) Secreted FGFs and (B) IL-8 were measured by Luminex (A to I) or ELISA (J), respectively, in basal supernatants of primary human BAECs grown at air-liquid interface (BAEC-ALI) after exposure to F.p. for 24 hours. Each line represents a different donor. * denotes P < 0.05 using a one-way analysis of variance (ANOVA) with Dunnett’s multiple comparisons test; unless the control group was zero, ^# denotes P < 0.05 using Wilcoxon paired t tests. (C) Secreted FGF1, FGF2, FGF4, FGF18, (D) IL-8, and (E) G-CSF were measured in supernatants from F.p.-stimulated precision-cut lung slices. Three precision-cut lung slices were incubated per well per donor and exposed to F.p. (10⁻⁷ M) for 24 hours, with three replicates per experiment. One of three experiments is shown. *P < 0.05 with a paired t test.

A similar FGF secretion profile was observed with budesonide-exposed BAEC-ALIs (fig. S2B), suggesting that this is a common response to ICS. As expected, F.p. or budesonide significantly inhibited IL-8 secretion from BAEC-ALI (P < 0.05; Fig. 2B and fig. S2C) (50). In these BAEC-ALI assays, F.p. exposure led to an increase in FGF1, FGF2, FGF4, and FGF18 transcription, rather than steroid-induced secretion of preformed intracellular FGFs; F.p. also increased expression of the steroid-inducible gene FKBP5 (fig. S2D) (43, 51).

To determine whether steroid exposure correlated with FGF expression in vivo, we assessed the correlation between expression of FKBP5 [as a surrogate for steroid exposure (51)] and FGF1, FGF2, FGF4, and FGF18 in bronchial brushings from individuals with severe asthma. FKBP5 correlated significantly with FGF1, FGF2, FGF4, and FGF18 in individuals with severe asthma (P < 0.03; fig. S2E), but not in individuals with moderate asthma who were not being treated with ICS (fig. S2, F and G), suggesting that steroid exposure may contribute to the up-regulation of these FGFs in vivo.

We extended these studies beyond the epithelium into more complex organoid systems using primary human precision-cut lung slices that contain epithelial, mesenchymal, and immune cell compartments. Similar to BAEC-ALI, exposure of precision-cut lung slices to F.p. led to increased secretion of FGF1, FGF2, FGF4, and FGF18 (Fig. 2C) and inhibition of IL-8 (Fig. 2D). F.p. also led to a significant increase in G-CSF secretion from precision-cut lung slices (P < 0.05; Fig. 2E), in line with previous reports of steroid-induced G-CSF (13) and reports of steroid-induced neutrophilia (52–55). Together, these data indicate that steroid exposure can concomitantly inhibit inflammatory cytokine responses and induce FGF1, FGF2, FGF4, and FGF18 in BAEC in vitro. Moreover, ICS exposure correlates with FGF expression in vivo in individuals with severe asthma.

Corticosteroid-induced FGFs mediate an epithelial-mesenchymal inflammatory axis

BAECs sit on an acellular reticular basement membrane, often thickened in severe asthma (56), separating epithelial cells from mesenchymal cells. FGFs have been identified as important mediators between epithelial and mesenchymal cells during development (28), facilitated by basement membrane components (57); however, it is unclear if dysregulated FGFs contribute to epithelial-mesenchymal cross-talk during disease. We therefore asked whether the FGFs that were elevated in bronchial brushings from individuals with severe asthma and that were steroid inducible in vitro (FGF1, FGF2, FGF4, and FGF18, hereafter referred to as FGF^C) had the potential to relay signals from epithelial cells to fibroblasts. Cultured human lung fibroblasts in three-dimensional (3D) rafts were treated with FGF^C to model the effects of multiple FGFs. RNA sequencing (RNA-seq) after 24 hours of FGF^C exposure identified over 4000 genes that were differentially regulated, with an equal number of up and down-regulated transcripts (Fig. 3A). FGF^C stimulated a suite of clinically relevant inflammatory cytokines, including IL33 and TSLP, both of which are showing promise as new therapeutic targets for severe asthma (58, 59), along with IL24, IFNA1, IL1B, and LIF and the granulocyte-promoting cytokines, CSF2 (encoding granulocyte-macrophage CSF) and CSF3 (encoding G-CSF) (Fig. 3B and fig. S3A). In addition to these inflammatory mediators, a suite of chemokines (CCL2, CCL7, CXCL5, CXCL6, and CXCL8), genes encoding ion channels involved in airway inflammation (NAV3, TRPA1, and TRPV2), and genes involved in hyaluronan synthesis (HAS2) (Fig. 3B) (60) were up-regulated. Genes involved in tissue remodeling (CTGF, MMP1, MMP8, WSP1, PDGFRA, ADAM8, LOXL1, and COL1A1) were also up-regulated. Beyond transcriptional responses, FGF^C increased secretion of G-CSF and hyaluronan (fig. S3B), supporting both granulocyte recruitment and lymphocyte retention, respectively; hyaluronan is an extracellular matrix ligand for CD44 (61), which is up-regulated on a variety of activated lymphocytes (62).

Fig. 3. Corticosteroid-induced FGFs support an epithelial-mesenchymal inflammatory axis.

(A) Normal human lung fibroblasts were cultured in 3D collagen-rich rafts, as described in Methods. Cells from five different donors were stimulated with a cocktail of FGF1, FGF2, FGF4, and FGF18, all at 10 ng/ml, for 24 hours before RNA isolation for RNA-seq. Differentially expressed genes (DEGs) where P < 0.05 are displayed. nRPKM, normalized reads per kilobase of transcript per million mapped reads. (B) Expression, in RPKM, of select DEGs is shown. *P < 0.05 by a paired t test. (C) The experimental setup depicts BAEC-ALI stimulation with F.p. (10⁻⁷ M) or diluent control for 24 hours and transfer of basal supernatant, mixed 50:50 with fresh medium, to fibroblast cultures for 24 hours. (D and E) Expression of CSF3 (D) and HAS2 (E) was measured in fibroblasts after exposure to supernatant from BAEC-ALI, as described above. Media, 100% fresh medium; Cont. Sup., 50% medium from diluent-exposed BAEC-ALI with 50% fresh medium; F.p. Sup., 50% medium from F.p.-exposed (10⁻⁷ M) BAEC-ALI with 50% fresh medium. Erdafitinib (Erd; 100 nM) was added, as indicated, to the fibroblast cultures. Data are expressed as fold change (FC) from fibroblasts cultured with medium alone. (F to H) F.p.-exposed (10⁻⁷ M) precision-cut lung slices were incubated in the presence (100 nM) or absence (diluent control) of erdafitinib. Three precision-cut lung slices were incubated per well for 24 hours, with three technical replicates per experiment. One of three experiments is shown. IL-8 (F), G-CSF (G), and hyaluronan (H) concentrations were measured by Luminex or ELISA. Error bars show SEM. *P < 0.05 with an unpaired t test.

Pathway analysis and functional annotation of differentially expressed genes, using Ingenuity Pathway Analysis, identified suites of FGF^C-induced genes involved in inflammatory responses, including leukocyte abundance, infiltration by phagocytes, and neutrophil infiltration and activation (fig. S3C and table S2). These data indicated that FGF^C have the potential to relay signals from steroid-exposed epithelial cells and drive an array of inflammatory and tissue remodeling responses through an epithelial-mesenchymal dialogue. To test this model, rather than adding exogenous FGF^C, we treated BAEC-ALI with F.p. or diluent control for 24 hours to drive FGF secretion, recovered the secreted products in the supernatant (F.p. Sup. or Cont. Sup., respectively), and exposed human lung fibroblasts to these secreted products in the presence or absence of a pan-FGFR inhibitor, erdafitinib (Fig. 3C) (63). Erdafitinib effectively inhibited FGF2-induced CCL2 (fig. S3D). F.p. Sup. induced CSF3 and HAS2 expression in fibroblasts in an FGFR-dependent manner (Fig. 3, D and E). Similarly, F.p. Sup. induced expression of IL33, TSLP, IL24, CXCL5, and CSF2 in human lung fibroblasts in an FGFR-dependent manner (fig. S3E) (64), supporting a steroid-induced and FGFR-mediated epithelial-mesenchymal cell inflammatory axis. Next, to determine whether steroid exposure led to protein production and secretion of G-CSF and hyaluronan (secreted products of CSF2 and HAS2 induced by F.p.) and to test whether this was dependent on FGFR signaling in a more complex multicellular system, we exposed precision-cut lung slices to F.p. in the presence or absence of erdafitinib (63). F.p. inhibited IL-8 secretion (Fig. 3F), an effect independent of FGFR signaling. However, F.p. increased G-CSF and hyaluronan secretion, which was largely FGFR-dependent (Fig. 3, G and H). Although corticosteroids can potently inhibit some inflammatory cytokines, including IL-8, data presented here indicate that corticosteroid exposure may also contribute to an inflammatory module through the induction of local “alarmin-like” FGFs from epithelial cells, functioning as a mediator of epithelial-mesenchymal cell communication.

Corticosteroid exposure suppresses type 2 airway inflammation and concomitantly induces FGF expression and airway neutrophilia in mice

After the in vitro observations that F.p. exposure can initiate an epithelial-mesenchymal inflammatory axis through the induction of FGFs, we next determined whether exogenous FGFs could enhance the inflammatory response in vivo using a well-described model of HDM allergen-driven airway inflammation. Briefly, mice were exposed to HDM multiple times through the intratracheal route, before being exposed to HDM with a cocktail of FGF1, FGF2, FGF4, and FGF18 (FGF^C), or control protein (Fig. 4A). HDM exposure led to a significant increase in airway infiltrates (P < 0.05), recovered by bronchoalveolar lavage (BAL), consisting predominantly of eosinophils, lymphocytes, and neutrophils. These airway infiltrates were all further augmented by FGF^C supplementation (Fig. 4B). FGF^C alone also led to an increase in airway infiltrates consisting predominantly of neutrophils and lymphocytes. Analysis of secreted products in the BAL fluid indicated that FGF^C amplified IL-4 and IL-5 secretions and either FGF^C alone or with HDM led to a significant increase in BAL G-CSF, IL-13, and CCL11 (P < 0.05; Fig. 4C and fig. S4A). It has previously been reported that allergen-driven airway inflammation in mice leads to an increase in hyaluronan in BAL samples and lung tissue (65) and that FGF2 can stimulate hyaluronan synthesis (60). In our hands, mice exposed to HDM also had elevated hyaluronan in both BAL samples and homogenized lung tissue, which was further enhanced by FGF^C (Fig. 4D). Last, in line with the amplified response observed with FGF^C supplementation, mucus-encoding genes Muc5ac and Muc5b were also increased in mice given HDM + FGF^C, compared to mice treated with phosphate-buffered saline (PBS) or HDM alone (Fig. 4E). Modeling the elevated FGFs that were observed in severe asthma patients, these preclinical data indicate that exogenous FGF1, FGF2, FGF4, and FGF18, collectively, have the capacity to amplify airway inflammation and may contribute to disease severity in individuals with asthma.

Fig. 4. Local corticosteroid exposure suppresses type 2 airway inflammation and concomitantly induces FGF expression and airway neutrophilia.

(A) C57BL/6 mice were treated with 10 μg of HDM in 25 μl by the intratracheal (i.t.) route on days 0, 2, 4, 14, 16, and 18, followed by 10 μg of HDM + 10 μg of murine FGF1, FGF2, FGF4, and FGF18 (FGF^C) by the intratracheal route on days 28 and 30. Samples from mice were analyzed for airway inflammation on day 31. (B) Total and differential cell analysis of airway infiltrates in BAL. (C) BAL cytokines were measured by Luminex. (D) Lung tissue and BAL hyaluronan were measured as described in Methods. (E) Gene expression of Muc5ac and Muc5b was measured in RNA extracted from lung tissue. Five mice were used per group, and one of three experiments is shown. *P < 0.05 with Mann-Whitney test. (F) C57BL/6 mice were treated with 10 μg of HDM in 25 μl by the intratracheal route on days 0, 2, and 4. On days 14, 16, 18, 28, and 30, mice were given 10 μg of HDM + F.p. (1 mg/kg) or diluent control, as indicated. Samples from mice were analyzed for airway inflammation on day 31. (G) Total and differential cell analysis of airway infiltrates in BAL. (H) BAL cytokines were measured by Luminex. (I) Lung tissue and BAL hyaluronan were measured as described in Methods. (J) Gene expression of Muc5ac and Muc5b was measured in RNA extracted from lung tissue. Five mice were used per group, and one of three experiments is shown. Error bars show SEM. *P < 0.05 with Mann-Whitney test.

Given that exogenous FGFs can amplify HDM-driven airway inflammation (Fig. 4, A to E) and extending our observations that ICS can drive FGF expression and secretion in BAEC-ALI in vitro (Fig. 2), we next tested whether F.p. exposure could drive FGF expression in vivo by treating HDM-exposed mice with intratracheal F.p. (Fig. 4F). HDM alone only modestly increased Fgf2 expression and inhibited Fgf18 expression. However, HDM combined with F.p. increased the steroid-responsive gene Fkbp5, confirming effective delivery of F.p., and led to a concomitant increase in Fgf1, Fgf2, Fgf4, and Fgf18 in lung tissue (fig. S4B). As expected, and previously reported (66), F.p.-treated mice had reduced total airway infiltrates, with a significant reduction in airway eosinophilia (P < 0.05); however, airway neutrophils were significantly increased after F.p. treatment (P < 0.05; Fig. 4G). Airway lymphocytes were unchanged after F.p. treatment. In line with these observations, IL-4, IL-5, and IL-6 secretions in BAL samples were inhibited by F.p., whereas G-CSF was elevated (Fig. 4H). F.p. alone had no impact on hyaluronan in BAL or lung samples; however, when combined with HDM-driven airway inflammation, F.p. significantly increased hyaluronan in the lung (P < 0.05; Fig. 4I). It has previously been reported that F.p. treatment has little impact on, or can increase, mucus secretion during microbial-driven airway inflammation (67, 68). In our studies, F.p. had no impact on Muc5ac gene expression, but significantly reduced Muc5b (P < 0.05; Fig. 4J). Collectively, these data indicate that steroid exposure not only inhibits type 2–associated airway eosinophilia, as previously reported, but also results in an increase in FGF, G-CSF, hyaluronan, and airway neutrophilia.

Corticosteroid-induced G-CSF and hyaluronan support airway neutrophilia and tissue lymphocyte recruitment, respectively

After the observation that F.p. induced G-CSF and airway neutrophilia in BAL samples, a phenomenon previously observed with dexamethasone and lipopolysaccharide (69), we tested whether steroid-driven G-CSF was an important downstream mechanism contributing to airway neutrophilia by blocking G-CSF with anti–G-CSF–neutralizing mAbs (aG-CSF; Fig. 5A). In addition, F.p. increased BAL and lung hyaluronan, similar to a previous observation (70). To test whether steroid-induced hyaluronan was an additional downstream mechanism contributing to airway inflammation, we treated mice locally with hyaluronidase (Fig. 5A). First, neutralization of G-CSF led to a reduction in total BAL cells, driven predominantly by a significant reduction in airway neutrophils (P < 0.05; Fig. 5B), similar to previous reports (71). Hyaluronidase treatment led to a significant reduction in BAL cells (P < 0.05), with a broad reduction in airway lymphocytes, eosinophils, and neutrophils (Fig. 5B), similar to observations made in the lung and intestine after hyaluronidase treatment (72, 73). BAL fluid analysis confirmed reduced G-CSF after anti–G-CSF–neutralizing mAb treatment and reduced hyaluronan after hyaluronidase treatment (Fig. 5C). In addition, hyaluronidase treatment reduced BAL G-CSF, IL-5, IL-13, and CCL11, potentially explaining the reduced airway neutrophilia and eosinophilia (Fig. 5C and fig. S4C). Lung tissue granulocytes were also significantly reduced after anti–G-CSF–neutralizing mAb treatment (P < 0.05), whereas hyaluronidase treatment reduced tissue-resident CD4⁺CD44⁺ CD69⁺ and CD8⁺CD44⁺CD69⁺ lympho cytes (Fig. 5, D and E). Last, the addition of hyaluronidase to F.p. significantly reduced Muc5ac (P < 0.05; Fig. 5F), suggesting that a hyaluronan-dependent mechanism was supporting Muc5ac expression. Together, these data indicate that F.p.-induced G-CSF and hyaluronan support granulocyte (neutrophil and eosinophil) and lymphocyte recruitment and retention in the lung and provide a mechanistic explanation for how ICS may orchestrate a distinct neutrophilic phenotype.

Fig. 5. Corticosteroid-induced G-CSF and hyaluronan support airway neutrophilia and tissue leukocyte recruitment, respectively.

(A) C57BL/6 mice were treated with 10 μg of HDM in 25 μl by the intratracheal route on days 0, 2, and 4. On days 14, 16, 18, 28, and 30, mice were given 10 μg of HDM + F.p. (1 mg/kg) with anti-mouse G-CSF mAb (+, clone 67604, 400 μg/kg) or isotype control mAb (i, immunoglobulin G1, 400 μg/kg), hyaluronidase (+, 4000 U/kg), or diluent control (c), as indicated. Samples from mice were analyzed for airway inflammation on day 31. (B) Total and differential cell analysis of airway infiltrates in BAL. (C) BAL cytokines were measured by Luminex. (D) Absolute numbers of tissue neutrophils and eosinophils from whole digested lung tissue. (E) Absolute numbers of lung-resident CD4⁺CD44⁺CD69⁺ and CD8⁺CD44⁺CD69⁺ cells from whole digested lung tissue. (F) Gene expression of Muc5ac was measured in RNA extracted from lung tissue. Five mice were used per group, and one of two experiments is shown. Error bars show SEM. *P < 0.05 with Mann-Whitney test.

FGFR signaling in fibroblasts propagates steroid-induced neutrophilic airway inflammation

F.p. treatment increased FGFs in vitro (Fig. 2) and in vivo (Fig. 4), and exogenous FGFs could drive an array of responses in fibroblasts in vitro (Fig. 3) and augment HDM-driven G-CSF, hyaluronan, and airway neutrophilia in vivo (Fig. 4). However, it remained unclear whether F.p.-induced endogenous FGFs, rather than exogenous FGFs, were a critical conduit translating F.p. treatment into the observed inflammatory responses in vivo. To test this, we took advantage of the FGFR substrate 2a, Frs2a, a membrane-anchored signal-transducing adaptor required for FGFR signaling (74, 75), by deleting Frs2a in HDM-exposed animals using tamoxifen-treated R26^CRE/ERT2Frs2a^fl/fl mice. To confirm deletion of Frs2a and disrupted FGFR signaling, we treated R26^CRE/ERT2Frs2a^fl/fl or R26^CRE/ERT2Frs2a^+/+ mice with tamoxifen, isolated lung fibroblasts, and tested their responsiveness to FGF^C (fig. S5A). FGF^C induced G-CSF secretion from fibroblasts isolated from tamoxifen-treated Frs2a^+/+ mice, but not from tamoxifen-treated Frs2a^fl/fl mice (fig. S5B). As a further control, we used the FGFR small-molecule inhibitor, erdafitinib (63), which also completely prevented FGF^C-induced G-CSF from fibroblasts. After validation of the model, we treated Frs2a^+/+ or Frs2a^fl/fl mice with HDM to establish airway inflammation, followed by tamoxifen treatment to delete Frs2a, and subsequently exposed mice to HDM and F.p. (Fig. 6A). HDM- and F.p.-driven airway inflammation, including airway neutrophilia, was significantly reduced in tamoxifen-treated Frs2a^fl/fl mice, compared to tamoxifen-treated Frs2a^+/+ mice (P < 0.05; Fig. 6B). BAL fluid analysis identified that HDM- and F.p.-induced G-CSF, IL-5, and hyaluronan were largely dependent upon Frs2a (Fig. 6C). Further analysis indicated that BAL concentrations of CXCL1, CXCL10, IL-13, and vascular endothelial growth factor (VEGF) were also dependent upon Frs2a (fig. S5C), with Muc5ac, but not Muc5b, expression also reduced in the lung tissue of tamoxifen-treated Frs2a^fl/fl mice (fig. S5D). The steroid-responsive gene, Fkbp5, was similarly induced between tamoxifen-treated Frs2a^+/+ and Frs2a^fl/fl mice, indicating that mice had similar exposure to F.p. (fig. S5D).

Fig. 6. FGFR signaling in Col1a2-expressing fibroblasts propagates steroid-induced airway inflammation.

(A) R26^CRE/ERT2Frs2a^fl/fl or R26^CRE/ERT2Frs2a^+/+ mice were treated with 10 μg of HDM in 25 μl by the intratracheal route on days 0, 2, and 4. On days 14 to 18 and 21 to 25, mice were given tamoxifen (80 mg/kg per dose, intraperitoneally) dissolved in sunflower oil. On days 28, 30, 32, 35, and 38, mice were given 10 μg of HDM + F.p. (1 mg/kg). Samples from mice were analyzed for airway inflammation on day 39. (B) Total and differential cell analysis of airway infiltrates in BAL. (C) BAL cytokines were measured by Luminex. (D) Col1a2^CRE/ERT2Frs2a^fl/fl or Col1a2^CRE/ERT2Frs2a^+/+ mice were treated with 10 μg of HDM in 25 μl by the intratracheal route on days 0, 2, and 4. On days 14 to 18 and 21 to 25, mice were given tamoxifen (80 mg/kg per dose, intraperitoneally) dissolved in sunflower oil. On days 28, 30, 32, 35, and 38, mice were given 10 μg of HDM + F.p. (1 mg/kg). Samples from mice were analyzed for airway inflammation on day 39. (E) Total and differential cell analysis of airway infiltrates in BAL. (F) BAL cytokines were measured by Luminex. A minimum of five mice was used per group, and one of two experiments is shown. Error bars show SEM. *P < 0.05 with Mann-Whitney test.

In vitro studies indicated that steroid-inducible FGFs from F.p.-exposed epithelial cells could activate fibroblasts and induce CSF3 and HAS2A expression (Fig. 3, C to E), suggesting a steroid-driven epithelial-mesenchymal axis. To test whether FGFR signaling in fibroblasts was required for F.p.-mediated airway inflammation in vivo, we specifically deleted Frs2a in fibroblasts using Col1a2^CRE/ERT2 Frs2a^fl/fl mice or Col1a2^CRE/ERT2Frs2a^+/+ as controls (hereafter FIB^Frs2aΔ or FIB^Frs2a+), in a similar experimental setup as above (Fig. 6D). After confirming that tamoxifen-treated FIB^Frs2aΔ mice rendered lung fibroblasts unresponsive to FGF^C treatment (fig. S5, E and F), we exposed FIB^Frs2a+ or FIB^Frs2aΔ mice to HDM to establish airway inflammation, followed by tamoxifen treatment to delete Frs2a in Col1a2-expressing cells. We then reexposed mice to HDM and evaluated F.p.-induced airway inflammation. BAL neutrophils, lymphocytes, and eosinophils were significantly reduced in FIB^Frs2aΔ mice as compared to FIB^Frs2a+ mice (P < 0.05; Fig. 6E), along with the FGF-driven G-CSF and hyaluronan (Fig. 6F). BAL CXCL1, CXCL10, VEGF, IL-13, and CCL11 were also significantly reduced in FIB^Frs2aΔ mice, compared to FIB^Frs2a+ mice (P < 0.05; fig. S5G). Collectively, these data indicate that FGFR signaling is an integral component of F.p.-driven airway neutrophilia and, although FGFs can activate a variety of cells (25), FGFR signaling in fibroblasts is specifically required to translate steroid exposure into downstream inflammatory responses. Together, we have identified that FGFs can drive G-CSF and hyaluronan production from fibroblasts in vitro (Fig. 3), and that G-CSF and hyaluronan support granulocytic inflammation and lymphocyte lung recruitment in vivo (Fig. 5). Data presented here with FIB^Frs2aΔ mice suggest an FGF-FGFR axis in fibroblasts as a critical node in vivo, translating steroid-induced FGFs into downstream G-CSF and hyaluronan production, airway neutrophilia, and tissue inflammation.

Therapeutic inhibition of FGFR signaling with ICS can prevent steroid-driven inflammation and prevent airway eosinophilia and neutrophilia

The above studies indicated that ICS treatment can effectively inhibit type 2–associated airway eosinophilia, but that a distinct and concomitant FGF-driven axis was induced by F.p. exposure. F.p. supported airway neutrophilia and lymphocyte recruitment and retention in a G-CSF and hyaluronan-dependent manner. To determine whether targeting FGFs or FGFR signaling in combination with ICS could prevent both type 2–associated airway eosinophilia and steroid-driven airway neutrophilia, we designed a proof-of-concept study by treating mice with HDM and a combination of F.p. and the FGFR inhibitor erdafitinib (63). Airway inflammation was significantly curtailed in HDM-treated mice given erdafitinib with F.p. (P < 0.05), with an almost complete prevention of airway neutrophils, eosinophils, and lymphocytes, compared to mice given HDM and F.p. without erdafitinib (Fig. 7A). BAL concentrations of IL-5, IL-13, CCL11, G-CSF, and hyaluronan were also reduced after erdafitinib treatment (Fig. 7B), with HDM-induced peribronchial and vascular lung inflammation and goblet cell hyperplasia almost completely ameliorated (Fig. 7, C and D). Despite local delivery of erdafitinib, systemic G-CSF and IL-5 were also reduced (Fig. 7E), suggesting that inhibition of FGFR signaling may prevent a systemic circuit of granulocyte mobilization. Together, these data indicate that locally administered steroids activate an FGF-dependent inflammatory axis, amplifying G-CSF and airway neutrophilia and increasing deposition of hyaluronan, which supports leukocyte recruitment and retention in the lung. Inhibition of FGFR signaling, either genetically in fibroblasts or pharmacologically with a pan-FGFR inhibitor, prevented steroid-driven neutrophilia while maintaining inhibition of type 2–driven eosinophilia.

Fig. 7. Therapeutic inhibition of FGFR signaling can prevent steroid-driven airway inflammation and block both airway eosinophilia and neutrophilia.

C57BL/6 mice were treated with 10 μg of HDM in 25 μl by the intratracheal route on days 0, 2, and 4. On days 14, 16, 18, 28, and 30, mice were given 10 μg of HDM + F.p. (1 mg/kg) with erdafitinib (30 mg/kg) or diluent control (−, HP-β-CD), as indicated. (A) Total and differential cell analysis of airway infiltrates in BAL. (B) BAL cytokines and hyaluronan were measured by Luminex and ELISA, respectively. Error bars show SEM. *P < 0.05 with Mann-Whitney test. (C and D) Lung inflammation severity (C) and airway epithelial goblet cell hyperplasia (GCH) severity (D) scores were determined by an anatomic pathologist, in a blinded manner, from hematoxylin and eosin– and Alcian blue/periodic acid–Schiff–stained tissue sections, respectively. Scale bars, 100 μm. Severity scores were compared by Kruskal-Wallis with Dunn’s multiple comparisons test. (E) Serum cytokines were measured by Luminex. A minimum of four mice was used per group, with one of two experiments shown. Error bars show SEM. *P < 0.05 with Mann-Whitney test.

DISCUSSION

Currently, a stepwise escalating dose of ICS is currently recommended for the management of asthma (1), which in the majority of cases is an effective approach to reduce inflammation, control symptoms, prevent exacerbations, and preserve lung function. Similarly, in preclinical airway inflammation models, as demonstrated here with HDM and by many others, local F.p. can reduce type 2 cytokines as well as airway and tissue eosinophilia. However, many individuals with severe asthma do not benefit from ICS (76), representing a critical unmet need. As we have learned over the past decade, severe asthma is a heterogeneous disease, with a variety of underlying biologies and endotypes identified (3, 4). Combination therapy of F.p. with salmeterol, a LABA, can provide additional benefit (77), indicating that in some individuals nonsteroid-sensitive pathways contribute to disease or that some individuals are truly steroid resistant. In this study, we put forwar an additidonal model and propose that, in some individuals, steroids may be driving responses and redirecting the disease phenotype.

The side effects and pharmacodynamic properties of corticosteroids are well described (78, 79), and attempts to better understand individuals who do not benefit from ICS have disproportionately focused on “steroid-resistant” asthma [reviewed and debated here (7, 80)]. One commonly described phenotype of steroid-resistant asthma has been loosely defined as type 2^low (45), by virtue of the absence of type 2 cytokines or evidence of type 2 cytokine–induced responses. A reciprocal type 1 phenotype, with elevated interferon-γ (IFN-γ) and CXCL10 (81, 82), and type 17 phenotype, with elevated expression of IL-17A (83–85), have also been described in steroid-resistant severe asthma. One potential explanation proposed for the dominance of T helper 1 (T_H1) lymphocytes in severe asthma is the disabling of natural killer cell–mediated killing of T_H1 cells, a process enhanced by corticosteroids (86). Beyond type 2^hi and type 2^low asthma phenotypes in severe asthma, Camiolo and colleagues (87) applied mass cytometry and RNA-seq to identify “lymphocyte-poor” and “lymphocyte-rich” severe asthma patients, representing two further molecular mechanisms that may contribute to disease severity. One common observation in many individuals with severe asthma is the presence of airway neutrophils [reviewed here (88)], although the pathways regulating neutrophilia in steroid-resistant asthma and an understanding of why symptoms in these individuals are not controlled by ICS are lacking.

To this end, we hypothesized that FGFs, which were elevated in bronchial brushings from patients with severe asthma who were taking high doses of ICS, may contribute to disease pathogenesis. Many disease-relevant factors can regulate FGF expression, including allergen exposure (35, 89–91), respiratory viral infection (92, 93), or bacterial endotoxin exposure (94). In addition to these, using a reductionist in vitro system with BAEC-ALI, we identified that the commonly prescribed ICS, F.p. and budesonide, could transcriptionally activate and increase secretion of FGFs, including FGF1, FGF2, FGF4, and FGF18. FGF transcripts correlated with the steroid-inducible transcript, FKBP5, supporting a causal relationship between ICS exposure and FGF induction. How other up-regulated FGFs (FGF3, FGF7, FGF8, FGF9, and FGF16) are induced in bronchial tissue in individuals with severe asthma is currently unclear. Dysregulated expression of FGFs has been observed in lung tissue or biofluids from patients with asthma (30, 32, 36, 95–99), and it has been observed in many different contexts that corticosteroids can up-regulate FGFs (19–22); however, the relationship between corticosteroids and FGF induction in the context of asthma was not understood. There is a large body of literature describing the role of a variety of FGFs (particularly FGF7, FGF9, FGF10, and FGF18) in lung epithelialization during development, injury, and fibrosis (28, 100–102), suggesting that FGFs could serve as local alarmin-like factors, facilitating local tissue repair and regeneration after injury and inflammation. These reparative processes are largely dependent on an epithelial-mesenchymal cross-talk (103, 104), with a recent study combining single-cell sequencing analyses with predicted cell-cell mapping, suggesting that an FGF-mediated epithelial-mesenchymal axis is present in the lungs of those with asthma (105). These observations encouraged us to ask whether steroid-inducible epithelial-derived FGFs influenced neighboring mesenchymal cells.

The reported impact of the steroid-inducible FGFs (FGF1, FGF2, FGF4, and FGF18) on fibroblasts is diverse, ranging from anti-fibrogenic responses with up-regulation of matrix metallopeptidase 1 (MMP1), collagenase, and DNA synthesis after FGF1 exposure (106, 107); hyperproliferation and transforming growth factor–β (TGF-β) antagonism after FGF2 exposure (108); proangiogenic responses with increased MMP9 and VEGF after FGF4 exposure (109); and inhibition of cell growth and apoptosis with enhanced migration after FGF18 exposure (110). To characterize the biological impact of the FGFs we observed in bronchial brushings from individuals with severe asthma and that were steroid inducible, we treated primary human lung fibroblasts with a cocktail of all four FGFs. FGF^C exposure increased expression of disease-relevant genes, including inflammatory cytokines and chemokines, growth factors, genes encoding ion channels involved in airway inflammation, and several tissue remodeling factors, along with a suite of pathways that can influence neutrophil biology. Of particular interest was the up-regulation and secretion of G-CSF, a well-described growth and antiapoptotic factor for neutrophils (111), providing a putative mechanistic link between steroid exposure, FGF secretion, and neutrophilia in severe steroid-resistant asthma (7, 112). A similar observation has been reported with dexamethasone-enhanced G-CSF from human mononuclear cells (13), supporting a potential role for steroid-induced G-CSF and airway neutrophilia. Furthermore, it has been appreciated for a long time that endogenous glucocorticoids elicit neutrophil mobilization from the bone marrow (113), forming the basis of the prednisone test used to determine bone marrow reserves in neutropenic patients (114). In addition, F.p. can prolong neutrophil survival by inhibiting apoptosis (115, 116), and inhaled F.p. can cause neutrophilia in otherwise healthy individuals (117). These observations suggest that ICS, similar to systemic prednisone, may contribute to neutrophilia in some individuals with severe asthma, as previous reports have suggested (118, 119). In vivo data presented here with anti–G-CSF mAbs confirmed an important role for G-CSF in F.p.-induced neutrophilia (71), and the use of Frs2a^fl/fl mice or the pan-FGFR inhibitor, erdafitinib, confirmed an important upstream role for FGFR signaling for G-CSF and subsequent neutrophilia in HDM- and F.p.-treated mice. A feed-forward FGF2-induced neutrophil circuit may perpetuate airway neutrophilia, with previous reports indicating that FGF2 can directly engage FGFRs on neutrophils contributing to airway neutrophilia in vivo (93) or neutrophil chemotaxis in vitro (120). Neutrophil-derived neutrophil elastase can also induce FGF2 (89), closing the neutrophil recruitment circuit. Data presented here provide a mechanistic link between steroid exposure and neutrophilia, mediated, in part, by nonredundant intermediary steps involving FGFs and FGF-induced G-CSF. Airway eosinophilia is also sensitive to FGFR signaling, with FGF^C-treated mice developing increased airway eosinophilia. Although F.p. inhibited airway eosinophilia, IL-5, IL-13, and CCL11, the deletion or inhibition of FGFR signaling further reduced airway eosinophilia. These data suggest that basal or F.p.-induced FGFs also contribute to airway eosinophilia.

The presence and persistence of leukocytes, including lymphocytes and neutrophils, in the asthmatic lung has been previously established (112); however, the mechanisms of cell retention in the lung are poorly understood. In line with and extending previous observations in FGF-treated fibroblasts (60, 121), we observed that FGF^C treatment increased hyaluronan synthase 2 (HAS2), an enzyme responsible for hyaluronan production (61), as well as hyaluronan, in fibroblasts in precision-cut lung slices and in BAL fluid and lung tissue of F.p.-treated mice. Hyaluronan is a glycosaminoglycan component of the extracellular matrix required for appropriate lung development and often dysregulated during disease [reviewed here (122)]. Hyaluronan is also a ligand for CD44, which is required for T cell and neutrophil extravasation and retention in inflamed tissues (123, 124). It therefore stood to reason that FGF^C-induced hyaluronan may contribute to steroid-supported inflammation and leukocyte retention in the lung. Hyaluronidase treatment, which reduced hyaluronan in the lung, led to a marked reduction of airway infiltrates and tissue lymphocyte retention in HDM- and F.p.-treated mice. Similar hyaluronan-dependent leukocyte retention has been observed after respiratory (65, 73), skin, and colon inflammation (72). Of particular relevance, FGF2 combined with corticosteroids (125), or formoterol, a LABA, combined with corticosteroids (126) can synergize to enhance hyaluronan, suggesting that, once FGF2 is induced, subsequent corticosteroid and LABA exposures may further amplify hyaluronan production. This results in another steroid-amplifying axis that may lead to increased hyaluronan in the lung, potentially retaining more immune cells. These observations suggest that steroid-driven FGF^C and the subsequent increase in G-CSF and hyaluronan contribute to recruitment and retention of leukocytes in the lung and provide a steroid-driven mechanism that may alter disease trajectory.

Beyond steroid-inducible FGF^C-driven G-CSF and hyaluronan, which we demonstrate to be critically important components of the steroid-driven response, FGF^C-treated fibroblasts also had increased IL33 and TSLP expression. Therapeutics targeting these pathways are both showing promising results in the clinic (5, 58, 59) and have been associated with elevated neutrophils in severe steroid-resistant asthma (127). Similarly, FGF^C up-regulated TRPA1 and TRPV2, both of which have important functions in neurogenic airway inflammation and bronchoconstriction in asthma (128–130). Whether steroid-driven FGFs contribute to IL33, TSLP, TRPA1, or TRPV2 expression in individuals with severe uncontrolled asthma is unclear, but it is tempting to speculate that these non–type 2 pathways are aggravated by steroid-driven FGFs, reshaping the disease phenotype. The modest, and potentially enhancing (67), impact of F.p. on mucus has previously been reported in both clinical and preclinical settings. The observation that pan-FGFR inhibitor treatment could reduce Muc5ac and goblet cell hyperplasia in F.p.-treated mice, in addition to reducing airway neutrophilia, suggests that FGFR-dependent responses may contribute to mucus production. FGF2 has been proposed as a remodeling biomarker, correlating with severe neutrophilic inflammation (30) and mucus hypersecretion (131).

There are several limitations that should be considered before translating these findings into therapies for airway diseases. First, therapeutically targeting FGFs or FGFR signaling kinases in oncology is being considered (132), but whether a suitable therapeutic index could be achieved for nononcology, obstructive airway indications is currently unclear. Specifically, the on-target hyperphosphatemia along with a variety of other adverse events observed after FGFR antagonism would need to be very carefully monitored. Second, there is significant redundancy among the FGFs and the FGFRs, and it is currently unclear whether a single FGF or a single FGFR is responsible for the benefit observed in this study. A detailed single-cell resolution map of FGF-producing and FGFR-expressing cells in individuals with obstructive airway diseases may help in delineating this. Last, these data indicate that some individuals’ severe asthma present with elevated concentrations of FGFs, which correlates with ICS use. A detailed and closely monitored clinical study is required to determine whether ICSs are a trigger of FGF production and downstream neutrophilic airway diseases in these individuals and whether FGFs contribute to disease progression or exacerbations of disease.

Nevertheless, in summary, we put forward an additional model to the prevailing steroid resistance model and suggest that some individuals with steroid-resistant asthma who present with neutrophilic inflammation may not be “resistant” to steroids, but instead may be fully responsive and present with a steroid-driven phenotype. We provide a mechanistic explanation for this model, through induction of G-CSF and hyaluronan, and identify FGFs as important steroid-inducible factors. Using combination therapy of F.p. with a pan-FGFR inhibitor, we provide a proof of principle to this model, supporting the further exploration of FGFs as therapeutic targets for individuals with severe, uncontrolled asthma.

METHODS

Study design

The objective of this study was to determine whether FGFs contribute to severe asthma. This was done by exploring the expression of FGFs in patient biopsies and testing whether any of the observed FGFs in severe asthma contributed to discernable pathways in vitro and using in vivo preclinical models. For all experiments, the number of replicates and statistical test used is reported in the figure legends. The reported replicates refer to biological replicates. All in vitro experiments in the main text were performed at least three times, and no outliers or other data points were excluded. For in vivo experiments, cages of mice were randomly assigned to treatment groups. The scientists treating the mice were not blinded to the names of treatment groups.

Human clinical cohorts

Patients with moderate asthma and healthy controls were participants in the Study of the Mechanisms of Asthma (MAST; clinicaltrials.gov: NCT00595153). Bronchoscopic Exploratory Research Study of Biomarkers in Corticosteroid-refractory Asthma (BOBCAT) (47) was a multicenter observational study conducted in the United States, Canada, and the United Kingdom and included 67 adult patients with moderate-to-severe asthma. Inclusion criteria required a diagnosis of moderate-to-severe asthma, confirmed by a forced expiratory volume in 1 s (FEV1) score between 40 and 80% of predicted value, as well as evidence within the past 5 years of greater than 12% reversibility of airway obstruction with a short-acting bronchodilator or methacholine sensitivity (20% decline in FEV1, PC20) < 8 mg/ml that was uncontrolled [as defined by at least two exacerbations in the prior year or a score of greater than 1.50 on the asthma control questionnaire 5 (ACQ5) while receiving a stable dose regimen (at least 6 weeks) of a high-dose ICS (at least 1000 mg of fluticasone or equivalent per day)] with or without a LABA. Post hoc analysis of datasets generated from bronchial brushings from healthy control and moderate and severe asthma patients was performed using Ingenuity Pathways Analysis (IPA) (133). Airway epithelial brushing transcriptomic data [GSE4302, National Center for Biotechnology Information (NCBI) Gene Expression Omnibus (GEO)] were derived from 86 individuals including 28 healthy individuals, 16 smokers, and 42 individuals with moderate asthma, in which diagnosis of asthma was based on the National Heart, Lung, and Blood Institute and National Asthma Education and Prevention Program Expert Panel Report 3 criteria. A dataset, T_H2 low and high phenotypes, was described previously (45). Transcriptomic data were based on Affymetrix Human Genome U133 Plus 2.0 Array (Thermo Fisher Scientific). CLAVIER (NCT02099656) was a randomized, placebo-controlled trial evaluating effects of lebrikizumab on airway eosinophilic inflammation and remodeling in uncontrolled asthma (48). Bronchial brushing RNA isolation and microarray analysis from both cohorts were conducted as a single experiment to mitigate potential batch effects. A randomized block design was used to balance factors of interest across RNA isolation and microarray batches; blocking factors included (i) cohort and (ii) disease severity, gender, age, medication usage, and type 2 biomarker concentrations (134).

Animals

Wild-type C57BL/6, R26^CRE/ERT2Frs2a^fl/fl, R26^CRE/ERT2Frs2a^+/+, Col1a2^CRE/ERT2Frs2a^fl/fl (FIB^Frs2aΔ), and Col1a2^CRE/ERT2Frs2a^+/+ (FIB^Frs2a+) mice were housed and bred at Genentech Inc. Wild-type C57BL/6J mice were obtained from The Jackson Laboratory. Frs2a^fl/fl mice were provided by F.W. (75). Littermate wild-type and gene-manipulated mice were used to eliminate microbiota biases. All mice were kept in a standard 12-hour light-dark cycle under specific pathogen–free (SPF) conditions and were allowed free access to sterile food and water. All mouse experimentation protocols were approved by the Laboratory Animal Resources Committee (LARC) at Genentech Inc. and adhered to the National Institutes of Health (NIH) Guide for the Care and Use of Laboratory Animals.

Airway inflammation models

For HDM-induced airway inflammation, mice were anesthetized with an intraperitoneal injection of ketamine (40 mg/kg) and dexmedetomidine (1 mg/kg). Once anesthetized, mice were treated with HDM (Greer, 10 μg in 25 μl) through the intratracheal route. After HDM treatment, mice were administered an intraperitoneal injection of atipamezole (5 mg/kg) to accelerate recovery from anesthesia. As indicated in figure legends, mice were given HDM (Greer, 10 μg in 25 μl) on days 0, 2, 4, 14, 16, 28, and 30 and analyzed on day 31. In some experiments, mice were treated with HDM supplemented with FGF^C (FGF1, FGF2, FGF4, and FGF18, 10 μg of each) or control protein in 25 μl. In some experiments, mice were treated with 10 μg of HDM + F.p. (1 mg/kg) or diluent control, as indicated in figure legends. Anti-mouse G-CSF (MAB414, clone 67604 diluted in sterile-filtered PBS) or isotype control (rat immunoglobulin G1 diluted in sterile-filtered PBS) was dosed at 400 μg/kg per intratracheal dose, as indicated in figure legends. Endotoxin-depleted hyaluronidase [Calbiochem, suspended in 20 mM sodium phosphate (pH 7) and 0.45% NaCl], as previously described (73), was dosed at 4000 U/kg or diluent [20 mM sodium phosphate (pH 7) and 0.45% NaCl], as indicated in figure legends. Endotoxin was removed from hyaluronidase using endotoxin removal spin columns (Pierce, 88276) and confirmed using an endotoxin quantitation kit (Pierce, 88282). Erdafitinib [JNJ-42756493, formulated in 20% hydroxypropyl-β-cyclodextrin (HP-β-CD)] or diluent control (HP-β-CD) was dosed at 30 mg/kg, as previously reported (63). To confirm tamoxifen-mediated Cre recombinase induction and deletion of loxP-flanked exons, we treated R26^CRE/ERT2Frs2a^fl/fl, R26^CRE/ERT2Frs2a^+/+, Col1a2^CRE/ERT2Frs2a^fl/fl, or Col1a2^CRE/ERT2Frs2a^+/+ mice with tamoxifen (Sigma-Aldrich) on days 0 to 4 and 7 to 11 (80 mg/kg per dose, intraperitoneally) dissolved in sunflower oil. On day 14, lungs were excised and primary lung fibroblasts were cultured as previously described (135). Briefly, lungs were isolated, washed in sterile-filtered PBS, cut up into 1-mm³ pieces and placed onto a scored petri dish, and cultured for 14 to 21 days in Dulbecco’s modified Eagle’s medium (DMEM) containing 10% heat-inactivated fetal bovine serum (FBS), 2 mM L-glutamine, and 0.5% penicillin-streptomycin (complete DMEM), washing away nonadherent and dead cells throughout. R26^CRE/ERT2Frs2a^fl/fl, R26^CRE/ERT2Frs2a^+/+, Col1a2^CRE/ERT2Frs2a^fl/fl, and Col1a2^CRE/ERT2Frs2a^+/+ mice were treated with 10 μg of HDM in 25 μl through the intratracheal route on days 0, 2, and 4. On days 14 to 18 and 21 to 25, mice were given tamoxifen (80 mg/kg per dose, intraperitoneally) dissolved in sunflower oil. On days 28, 30, 32, 35, and 38, mice were given 10 μg of HDM + F.p. (1 mg/kg). Samples from mice were analyzed for airway inflammation on day 39. In all experiments, mice were euthanized with pentobarbital sodium (390 mg/ml) and phenytoin sodium (50 mg/ml) (Euthasol, Virbac Animal Health) (200 mg/kg), with blood recovered during terminal exsanguination. Serum was separated using serum separator tubes (Sarstedt). BAL was collected in two sequential recoveries. First, 500 μl of ice-cold PBS was used to lavage the airspaces for cellular and analyte analysis. A second lavage of 2 × 500 μl was used for additional cellular recovery. Total BAL cells and differential cell counts were determined from all BAL washes by fluorescent flow cytometry using an XT-2000 veterinary hematology analyzer (Sysmex). Wright-Giemsa–stained cytospins were then prepared and reviewed by microscopy for morphology and agreement with automated counts. Lungs were recovered in RNAlater and stored at 4°C for 24 hours, before longer storage at −80°C followed by RNA recovery. For pathology, lungs were inflated with formalin and submerged in formalin for 24 hours before processing, then trimmed, processed, and paraffin-embedded, and 4-μm sections were stained with hematoxylin and eosin stain or Alcian blue/periodic acid–Schiff. Lung inflammation and airway goblet cell hyperplasia severity were scored blinded to treatment group using semiquantitative scales from 0 to 4 and 0 to 3, respectively. Severity score group comparisons were by Kruskal-W allis test with Dunn’s multiple comparisons test. For lung tissue cellular flow cytometry analysis, mice were euthanized and bled before all lung lobes were recovered. Lobes were minced using gentleMACS system in digestion buffer [RPMI 1640, Liberase (200 μg/ml), and bovine deoxyribonuclease I (0.67 mg/ml)] for 30 min at 37°C in an orbital shaker at 200 rpm. Tissue was further processed with gentleMACS to produce single-cell suspension with Lung_01 program and filtered through a 70-μm pore strainer. Recovered cells were resuspended in 1 ml of ice-cold magnetic-activated cell sorting (MACS) buffer (PBS containing 1% FBS) and aliquoted for antibody staining. Cells were stained with a viability dye using a Vi-CELL XR cell counter (Beckman Coulter) and treated with Fc blocking reagent (Miltenyi Biotec, catalog no. 130–092-575) before adding a flow cytometry antibody panel. The following fluorochrome-conjugated antibodies were used: LIVE/DEAD (UV-Thermo Fisher Scientific), eFluor 780, 1:1000; CD3 (17A2), Pacific Blue (PB), 1:200; CD19 (eBio1D3), phycoerythrin (PE), 1:200; CD49b (DX5), fluorescein isothiocyanate (FITC), 1:200; Siglec-H (440c), Brilliant Violet (BV510), 1:200; CD11c (N418), allophycocyanin (APC), 1:200; CD11b (M1/70), PE, 1:200; major histocompatibility complex class II (M5/114), APC or FITC, 1:200; Ly6C (HK1.4), APC, 1:200; Ly6G (1A8); Alexa Fluor (AF) 488, 1:200; Siglec-F (E50–2440), PE-CF594, 1:200; CD8α (53–6.7), APC, 1:200; CD4 (RM4–5), APC, PB or FITC, 1:200; T cell receptor β (H57), AF488, 1:200; AF647 or APC Cy7, 1:200; CD44 (IM7), APC or FITC, 1:200; and CD69 (H1.2F3), PB, 1:200.

Reagents, enzyme-linked immunosorbent assays, Luminex, and in vitro assays

F.p. (Sigma-Aldrich) and budesonide (Thermo Fisher Scientific) were depleted of any endotoxin using endotoxin removal spin columns (Pierce, 88276) and confirmed using an endotoxin quantitation kit (Pierce, 88282). F.p. and budesonide were dissolved in dimethyl sulfoxide (DMSO) at a concentration of 100 mg/ml. Erdafitinib (JNJ-42756493) was diluted in DMSO at 5 mM for in vitro use or formulated in 20% HP-β-CD for in vivo use and dosed at 30 mg/kg, as previously reported (63). Recombinant FGF proteins [mouse: FGF1, 4686-FA-025/CF (R&D Systems), FGF2, 3139-FB (R&D Systems), FGF4, 5846-F4 (R&D Systems), and FGF18, NBP2-35028 (Novus); human: FGF1, 233FB025CF (R&D Systems), FGF2, 232FA025CF (R&D Systems), FGF4, 7460F4025CF (R&D Systems), and FGF18, 8988F18050 (R&D Systems)] were resuspended in sterile-filtered PBS and used in vitro, where indicated, with heparin sulfate (1 μg/ml), as previously described (136). Human and murine chemokine, cytokine, and human FGF1 and FGF2 were assayed by Luminex technology using Bio-Plex Pro (Bio-Rad Laboratories) according to the manufacturer’s instructions. Fluorescence intensities (FIs) from the labeled beads were read using FlexMaps instrument (Luminex Corp.). FIs from diluted standards were used to construct standard curves using Bio-Plex Manager software (Bio-Rad Laboratories) using either 4- or 5-pl regression type. Data are presented as average of duplicate measurements. Enzyme-linked immunosorbent assay (ELISA) kits were used to determine concentrations of FGF3 (US Biological), FGF4 (Thomas Scientific), FGF7 (Thomas Scientific), FGF8 (Thomas Scientific), FGF9 (Abcam), FGF16 (Thomas Scientific), and FGF18 (US Biological). Cell viability was determined by alamarBlue (Invitrogen) according to the manufacturer’s guidelines. Hyaluronan in BAL, whole lung lysates, or precision-cut lung slice lysates were measured using a Hyaluronan ELISA kit, following the manufacturer’s recommendations (R&D Systems).

Cell culture

BAECs were purchased from Lonza (CC-2540S), expanded in bronchial epithelial growth medium (CC-3170), and seeded on 6.5-mm transwell plates in a humidified incubator at 37°C with 5% CO₂. After forming a confluent monolayer, the medium was lifted from the cells and the medium in the lower chamber was changed to PneumaCult-ALI medium (STEMCELL Technologies, 05001) to allow cells to differentiate. After 21 days, beating cilia could be observed on 60 to 80% of the transwell surface. BAEC-ALIs were exposed to F.p. (Sigma-Aldrich) and budesonide (Thermo Fisher Scientific) diluted in PBS and added to the apical surface. Normal human lung fibroblasts (NHLF) were purchased from Lonza (CC-2512) and cultured for a single passage in FGM-2 fibroblast medium (CC-3132) before being incorporated into Lonza collagen rafts (200,000 cells per raft) per the manufacturer’s instructions. Precision-cut lung slices were derived from healthy whole lungs. Briefly, lungs were received from the National Disease Research Interchange. The smallest lobe was cut free, exposing its main bronchiole, and inflated with 2% (w/v) low–melting point agarose solution. Once the agarose had solidified, the lobe was sectioned. Cores of 8 mm in diameter were made in which a small airway was visible. The cores were placed in the Precisionary Instruments Oscillating Vibratome (Precisionary Instruments, VF-300), and the speed was set to produce slices at about 1 per 30 s. Precision-cut lung slices (thickness, 350 μm) were transferred in sequence to wells containing Ham’s F-12 (Thermo Fisher Scientific) medium to identify contiguous airway segments. Suitable airways on slices were selected on the basis of the following criteria: presence of a full smooth muscle wall (cut perpendicular to direction of airway), presence of beating cilia and internal folding of epithelium to eliminate blood vessels, and presence of unshared muscle walls at airway branch points to eliminate possible counteracting contractile forces. Slices were then incubated at 37°C on a rotating platform in a humidified air/CO₂ (95%/5%) incubator.

RNA-seq and quantitative reverse transcription polymerase chain reaction

Tissue and cell samples were collected RNAlater (Qiagen) for storage or directly into TRIzol (Thermo Fisher Scientific) for processing. After tissue dissociation and lysis, lysates were mixed with chloroform and centrifuged at 13,000g for 15 min. The resulting aqueous phase was mixed with a 1.5× volume of absolute ethanol before purification using Qiagen RNeasy columns following the manufacturer’s recommendations. RNA integrity was assessed using a bioanalyzer before RNA-seq or quantitative reverse transcription polymerase chain reaction (qRT-PCR). One hundred nanograms of purified RNA was reverse-transcribed using iScript (Bio-Rad) following the manufacturer’s instructions, and the following TaqMan probes (Thermo Fisher Scientific) were used: human: FGF1, Hs01092738_m1; FGF2, Hs00266645_m1; FGF3, Hs00173742_m1; FGF4, Hs00173564_m1; FGF7, Hs00384281_m1; FGF8, Hs00171832_m1; FGF9, Hs00181829_m1; FGF16, Hs00175752_m1; FGF18, Hs00826077_m1; CSF3, Hs99999083_m1; HAS2, Hs00193435_m1; and FKBP5, Hs01561003_m1; mouse: Muc5ac, Mm01276739_g1; Muc5b, Mm00466391_m1; Fkbp5, Mm01300962_m1; Fgf1, Mm00438906_m1; Fgf2, Mm00433287_m1; Fgf4, Mm00438917_m1; and Fgf18, Mm00433286_m1. For RNA-seq, transcriptome profiles were generated using TruSeq RNA Access technology (Illumina). RNA-seq reads were processed using the HTSeqGenie R package (v. 4.2.2). Briefly, RNA-seq reads were first aligned to ribosomal RNA sequences to remove ribosomal reads. The remaining reads were aligned to the mouse reference genome (GRCm38) using GSNAP (1,2) version 2013–11-10, allowing a maximum of two mismatches per 75 base sequence (parameters: ‘-M 2 -n 10 -B 2 -I 1 -N 1 -w 200000 -E 1-pairmax-rna = 200000 – clip-overlap). To quantify gene expression, we counted the number of reads aligning within exons of gene models provided by GENCODE basic (v. 27). We used the DESeq2 package (137) to determine differential expression between test and control samples. We used package defaults for size factor calculation, dispersion estimation, and model fitting.

Statistical analysis

All raw, individual-level data where n < 20 are presented in data file S1. All data were analyzed and graphed using Prism 9 software (GraphPad). Comparisons between two groups were analyzed by performing an unpaired or paired, two-tailed Student’s t test, as indicated in figure legends. Comparisons between more than two groups were analyzed using a one-way analysis of variance (ANOVA) with multiple comparisons test, as indicated in the figure legends. For all experiments, individual dots or lines on each graph represent unique biological samples and not technological replicates. For marking significance, * indicates P < 0.05, as indicated in the legends. All experiments were performed at least twice. Error bars indicate SEM.

Data and materials availability:

All data associated with this study are present in the paper or the Supplementary Materials. Airway epithelial brushing transcriptomic data are available at NCBI GEO (GSE4302). Frs2a mice are available from F.W.