RNAi screening identifies a mechanosensitive ROCK-JAK2-STAT3 network central to myofibroblast activation

Myofibroblasts are key effector cells in wound healing, but also play a leading role in pathological fibrosis and contribute to tumor progression (Hinz et al., 2012). The myofibroblast cytoskeleton is characterized by the incorporation of α-smooth muscle actin (αSMA) into densely bundled actin stress fibers (Hinz and Gabbiani, 2003). These stress fibers also contain myosin, allowing myofibroblasts to generate the large contractile forces that are necessary for matrix remodeling and tissue scarring, as well as tension-mediated activation of latent transforming growth factor-β (TGF-β) (Wipff et al., 2007). Multiple biochemical and biomechanical pathways appear to intersect in regulating the fibroblast and myofibroblast cytoskeleton (Prager-Khoutorsky et al., 2011; Sandbo and Dulin, 2011), but a systematic loss-of-function screen of this cellular phenotype has not been previously reported. Although TGF-β is widely recognized as a key regulator of myofibroblast activation, therapeutic targeting of TGF-β itself has remained challenging because of its pleiotropic roles in inflammation and tissue homeostasis. Thus, discovery of novel gene and pathway regulators of myofibroblast activation may have substantial translational impact for the development of interventions that target the formation and function of myofibroblasts.

To this end, we developed a novel high-content RNAi screening approach to measure the length and intensity of αSMA-positive stress fibers in lung fibroblasts. We screened a ∼7000 siRNA library focusing on kinases, phosphatases, G-protein coupled receptors (GPCRs) and other druggable targets, and identified numerous known pro-fibrotic genes, as well as a panel of novel putative myofibroblast regulatory genes. Importantly, screen hits were validated on hydrogels of physiologically relevant stiffness, mimicking the transition zone between normal and fibrotic tissue stiffness, and the efficacy of selected hits confirmed in primary lung fibroblasts from patients with idiopathic pulmonary fibrosis (IPF), demonstrating relevance in disease-derived primary cells. One target emerging from this screen, signal transducer and activator of transcription 3 (STAT3), was further investigated by siRNA and small-molecule inhibition. Knockdown or inhibition of STAT3 suppressed fibroblast traction force generation and extracellular matrix gene expression and deposition in lung fibroblasts, as well as primary hepatic stellate cells and cardiac fibroblasts, consistent with a broad role of STAT3 in myofibroblast activation across multiple organs. Further investigation implicated matrix stiffness, acting in a ROCK- and JAK2-dependent but ligand/receptor-independent fashion, in STAT3 activation. While STAT3 has been previously implicated in fibrotic disease models (Milara et al., 2018; O'Donoghue et al., 2012; Pang et al., 2010; Pedroza et al., 2015), it also plays important roles in tissue inflammation, injury, repair and regeneration (Jacoby et al., 2003; Mair et al., 2010; Milner et al., 2015; Moh et al., 2007; Pickert et al., 2009), and thus its direct role in fibrosis per se has remained ambiguous. Therefore, we inhibited STAT3 using a therapeutic regimen (days 14-28) after bleomycin-induced lung injury, when inflammation has largely subsided and fibroblast activation and pulmonary fibrosis are ongoing (Moeller et al., 2008). Our results document an important direct role for STAT3 in the fibrotic phase of lung remodeling post injury. Together, these findings validate our RNAi screening approach, identify a panel of high-confidence myofibroblast regulatory genes, and point towards STAT3 as an important molecular target for regulation of myofibroblast activation in pulmonary fibrosis and other fibrotic diseases.

To identify novel regulators of the transition of fibroblast to myofibroblasts, we developed a robust and quantitative αSMA stress fiber assay in IMR-90 fibroblasts derived from fetal human lung. The motivation for our quantitative image analysis approach was the observation that αSMA, when not localized in stress fibers, was still diffusely observed in cells (Fig. 1A). Assay conditions therefore included low serum and exogenous TGF-β stimulation (Fig. S1) for 72 h to maximize formation of dual-positive F-actin and αSMA stress fibers, a hallmark of myofibroblasts (Sandbo and Dulin, 2011). To visualize such stress fibers, fixed and permeabilized cells were stained with phalloidin to positively identify filamentous F-actin, and the colocalization of αSMA immunostain with ‘bundled’ phalloidin-stained structures was used to define αSMA stress fibers (Fig. S2). A custom image analysis algorithm was created to quantify the length and intensity of these fiber structures (detailed in the Materials and Methods). The utility of the assay was confirmed with positive control siRNAs known to ablate fibroblast stress fibers, including Rho-associated coiled-coil-containing protein kinase 1 (ROCK1), TGF-β receptor 1 (TGFBR1), and megakaryoblastic leukemia translocation 1 (MKL1; Fig. 1A). Using TGFBR1 siRNA pools as positive control on all screening plates, the robustness of the αSMA stress fiber assay was acceptable for high-throughput screening, returning, on average, a Z-prime value of 0.2.

Using this approach, we screened 7317 siRNA pools (Dharmacon), focusing on kinases, phosphatases, GPCRs and other druggable targets (Fig. 1B; a complete gene list can be found in Table S1). Each pool contained four siRNAs targeting a single gene and pools were assayed in triplicate plates. Toxic siRNA pools, defined as reducing cell count below 30% of non-targeting controls, were removed. In general, αSMA fiber signal was unaffected by cell count (Fig. S3). We classified hits into weak, moderate and strong inhibitors and enhancers, based on robust z-score (Fig. 1C). The ‘hit’ cut-off was an absolute z-score of 2 (∼15% change in αSMA fiber signal compared with non-targeting control siRNA) and ‘hitting’ in at least two of three replicates. A weighted score for each siRNA pool replicate was used to finalize hit strength (detailed in Materials and Methods). This approach identified 526 hits: 51 strong, 135 moderate and 129 weak inhibitor siRNA pools, and 17 strong, 87 moderate and 107 weak enhancers (Fig. 1, Table S2). The hit rate in this primary screen was ∼7% overall, with ∼1% of siRNAs registering as strong hits (absolute z-score >3). Analysis of phalloidin-stained F-actin stress fibers was generally less robust than the αSMA fiber assay, returning Z-prime values ∼0. However using this read-out, we did identify several inhibitor and enhancer siRNAs that the αSMA fiber analysis was unable to identify and carried several of these hits forward for further validation (Table S3). Analysis of screen hit siRNA seed sequences was carried out to identify potential off-target effects. Genome-wide enrichment of seed sequence matches (GESS) (Sigoillot et al., 2012) did not identify any significant off-target gene transcripts (Fig. S4). A different seed clustering analysis [Dharmacon (Sudbery et al., 2010)], identified 3 seeds enriched in the primary screen hits, contained in 26 siRNA pools (Table S4).

To confirm primary screen results and eliminate potential false positives due to off-target effects, we carried out deconvolution screening on a selected subset of 355 siRNA pools (315 inhibitors and 40 enhancers; Fig. 2A). We prioritized strong and moderate inhibitors, and consistent but weak inhibitors identified in the primary screen were carried forward for this step. Some strong and moderate enhancers were also picked for deconvolution, as well as several strong or moderate inhibitors from the F-actin only analysis. For each selected gene, four individual siRNAs were analyzed in the αSMA stress fiber assay. To be considered validated, at least two out of four siRNAs were required to meet an absolute z-score cut-off (>2 or ∼15% change) in the αSMA stress fiber assay, and ‘hit’ in at least two of three replicates. In total, 200 gene targets were validated (56% confirmation), 61 with three or four individual siRNAs validating (Fig. 2A,B, Table S5). A weighted score for each siRNA replicate was used to finalize hit strength (see Materials and Methods). A single score of weak, moderate or strong for each inhibitor or enhancer was given to validated targets, based on the weighted score for the single best performing siRNA. Overall, deconvolution screening confirmed 176 inhibitors (96 strong, 52 moderate, 28 weak) and 24 enhancers (4 strong, 10 moderate, 10 weak) of αSMA stress fibers (Fig. 2B). Of the confirmed hits, 28 were recovered from the ‘F-actin only’ category of primary screen hits. Hits that contained duplexes with enriched seed sequences that may act on off-targets (as identified in seed clustering analysis, Table S4) had either 1 or 2 additional active duplexes, increasing confidence that these targets are valid.

To evaluate the relevance of siRNA hits under more physiologically relevant stiffness conditions, we re-tested selected hits on polyacrylamide hydrogel-coated plates with a Young's modulus of 13 kPa to mimic the transition zone between normal and fibrotic tissue stiffness (Booth et al., 2012; Brown et al., 2013; Liu et al., 2015; Liu et al., 2010). We prioritized hits with multiple (3 or 4) validated individual siRNAs and consistent activity across plate replicates. Robustness of the αSMA fiber assay was acceptable on hydrogel 96-well plates, returning Z-prime values between 0.2 and 0.6 throughout screening. For 167 selected targets (146 inhibitors, 21 enhancers), 4 individual siRNAs were tested on triplicate plates, and the same hit criteria and scoring as in deconvolution screening was applied. In total, 118 gene targets were validated (71% confirmation), 53 with 3 or 4 individual validating siRNAs (Fig. 2C, Table S6). This compliant hydrogel validation screening identified 97 inhibitors (47 strong, 42 moderate, 8 weak) and 21 enhancers (16 strong, 5 moderate). As expected, genes with 4 of 4 siRNA duplexes hitting in the previous deconvolution screening validated at the highest rate (100%, 13 of 13), while those with 2 of 4 or 3 of 4 validated at a lower rate (69%, 105 of 153). The αSMA fiber score for these gel plate hits generally correlated well (R²=0.66) with their scores on plastic plates (Fig. S5). Interestingly, the 49 inhibitors that failed to meet hit criteria on hydrogels included both weak and strong inhibitors from the previous plastic deconvolution screen, indicating that performance of the assay on plastic plate conditions was not uniformly predictive of performance on more compliant hydrogels. And while only 66% of inhibitors tested (97 of 146) passed this second analysis, all enhancer siRNAs tested (21 of 21) were validated. Although we cannot rule out selection bias, one interpretation of these results is that screening on hydrogels of physiologically relevant intermediate fibrotic stiffness is a more selective approach at identifying inhibitors of myofibroblast activation, and conversely, is a more sensitive system to identify enhancers of myofibroblast activation compared with screening on stiff plastic tissue culture plates. Thus, future efforts at identifying and testing inhibitors and activators of myofibroblasts might be best conducted on hydrogels at either end of the stiffness transition range between normal and fibrotic tissues.

High ranking hits from the deconvolution step, with two or more positive siRNA duplexes, and with an effect equal to or greater than 35% change in αSMA are shown in Fig. 2D. As expected, siRNAs targeting ACTA2 (encoding αSMA), TGFBR1 and TGFBR2 were strong inhibitors (example images in Fig. S6). Several other known pro-fibrotic molecules were identified, including serum response factor (SRF) (Small, 2012; Yang et al., 2003), megakaryoblastic leukemia translocation 1 (MKL1) (Crider et al., 2011; Small, 2012; Small et al., 2010), and bromodomain containing 4 (BRD4) (Ding et al., 2015; Tang et al., 2013a,b), which strongly validates this approach to identify bona fide myofibroblast regulatory genes. While a rigorous unbiased network analysis of our candidate genes is not possible due to the selection of candidates at each step of screening, we nevertheless employed pathway analyses (Ingenuity) to visualize connections between individual candidates validated in the hydrogel deconvolution step. Interestingly, the highest ranking interconnected gene network that emerged included not only expected interactions between α-smooth muscle actin (ACTA2), SRF and MKL1, but also suggested a central role for STAT3 as an essential gene/protein hub in myofibroblasts (Fig. 2E).

To evaluate the relevance of the siRNA hits in primary cells isolated from patients with fibrosis, we re-tested 53 selected hits (47 inhibitors and 6 enhancers) in three fibroblast cultures derived from patients with idiopathic pulmonary fibrosis (IPF). siRNA-transfected cells were cultured on compliant hydrogel plates (13 kPa Young's modulus) in the presence of TGF-β for 72 h, and αSMA stress fibers were quantified as in prior screening steps. Additionally, procollagen I protein expression was assessed by parallel quantification of immunostaining (Fig. S7). Assay performance in the three cell lines was monitored throughout this medium-throughput screening step using TGFBR1 siRNA as positive control (Fig. S7). The two most potent siRNAs per gene from previous screening steps (or in some cases four siRNAs where the original duplexes were no longer available) were tested in two separate trials in each IPF fibroblast culture. To score positively, hits had to confirm the αSMA fiber phenotype in at least one of the previously validated siRNAs, two cell lines and both trials (any combination of siRNAs tested). Thirty-seven genes (70% of 53 tested) were validated by this analysis (Fig. 3A, Table S7). Of the target genes that did not confirm, 12 genes were validated in a single IPF fibroblast culture, and 5 genes were not validated in any IPF fibroblast culture. Twenty-one of 37 hits (57%) also scored positively in the procollagen read-out (Fig. 3B, Table S8). The conservation of hits between primary IPF fibroblasts and the IMR-90 fibroblast line, and overall correlation in αSMA score (Fig. 3C) confirmed that IMR-90 cells are a suitable, although imperfect cell choice for high throughput identification of fibrosis regulatory genes that are relevant to primary disease-derived fibroblasts.

Based on the network analysis of STAT3 as a putative hub in a myofibroblast regulatory network (Fig. 2E), and confirmation of the consistent effects of its knockdown in IPF-derived fibroblasts (Fig. 3D), we further explored the functional relevance of STAT3 in TGF-β-stimulated IMR-90 cells. We analyzed expression of a panel of pro-fibrotic genes after siRNA knockdown of STAT3 by using western blotting and qPCR (Fig. 4A) at 72 h, and demonstrated significant attenuation of expression of ACTA2, which encodes αSMA, along with COL1A2, encoding the collagen type I α2 component of fibrillar type I collagen, and CTGF, which encodes connective tissue growth factor (Fig. 4B). These results are consistent with the recent demonstration of an essential role for STAT3 as an enhancer of COL1A2 expression (Papaioannou et al., 2018). We then used traction microscopy to assess the functional effect of STAT3 siRNA knockdown on cell-matrix force generating capacity, and documented a significant ablation of root mean square (RMS) traction forces with knockdown, consistent with an important role for STAT3 in myofibroblast contractility (Fig. 4C). Finally, to relate pro-fibrotic gene expression to de novo matrix synthesis and deposition, we adapted an antibody-based detection method (Vogel et al., 2014) to quantify fibroblast-deposited fibronectin, an essential early step in fibrotic matrix deposition (Schwarzbauer and DeSimone, 2011), and collagen I. STAT3 siRNA knockdown significantly reduced deposition of both fibronectin and collagen I (Fig. 4D). Together, these results indicated an essential role for STAT3 expression in TGF-β-stimulated myofibroblast functions, including extracellular matrix gene expression, deposition and contractile function. To test whether the effects of STAT3 siRNA were exclusive to TGF-β-stimulated fibroblasts, we repeated the analysis of ACTA2, COL1A2 and CTGF in the absence of TGF-β. STAT3 siRNA knockdown demonstrated a significant effect on ACTA2 and CTGF expression levels in the absence of TGF-β (Fig. 4B), further validating its relevance as a regulator of gene expression even in the absence of exogenous myofibroblast stimulation.

To extend our siRNA observations with a pharmacological approach, we employed LLL12, a small-molecule inhibitor of STAT3. LLL12 is a cell-permeant non-peptide molecule that binds to tyrosine 705 on STAT3, preventing its phosphorylation and subsequent STAT3 dimerization, nuclear translocation and downstream signaling (Lin et al., 2010). LLL12 is specific for STAT3 over STAT1, STAT2, STAT4 and STAT6, as well as a panel of 21 additional kinases (Lin et al., 2012). We first confirmed that LLL12 phenocopies the attenuation of αSMA-positive IMR-90 fibroblasts in a dose-dependent fashion (Fig. 5A), and leads to dose-dependent reversal of TGF-β-induced deposition of both fibronectin and collagen I by these cells (Fig. 5B). We then examined transcript expression of genes including ACTA2, COL1A2, FN1 (fibronectin) and CTGF. As was the case with STAT3 knockdown, LLL12 significantly reduced transcript levels for these pro-fibrotic genes, and did so in a dose-dependent manner (Fig. 5C). LLL12 also dose-dependently reduced traction forces in TGF-β-stimulated IMR-90 cells (Fig. 5D). Strikingly, the traction ablation in response to LLL12 was similar in magnitude to that observed with the Rho kinase inhibitor Y27632, demonstrating the potency of the LLL12 effect on myofibroblast contractility. Extending these observations to other primary cell types implicated in fibrotic tissue remodeling, we observed similar effects of LLL12 on pro-fibrotic gene expression (Fig. 5E) and traction forces (Fig. 5F) in hepatic stellate cells and cardiac fibroblasts stimulated with TGF-β to promote a pro-fibrotic state of activation. Together, these results indicate a broad relevance of STAT3 signaling to TGF-β-stimulated myofibroblast contractility, ECM gene expression and matrix production.

Prior work has demonstrated stable phenotypic alterations in fibroblasts isolated from patients with IPF, and implicated aberrant STAT3 signaling in IPF fibroblast activation (Pechkovsky et al., 2012; Prêle et al., 2012). To build on our screen observations of STAT3 siRNA effectiveness in IPF fibroblasts, we tested whether STAT3 pharmacological inhibition retains effectiveness in such patient-derived samples by applying LLL12 to primary, low passage fibroblasts derived from the lungs of patients with IPF. In such cells, we observed potent inhibitory effects of LLL12 on pro-fibrotic gene expression (Fig. 5E) and traction forces (Fig. 5F), confirming the preserved efficacy of STAT3 inhibition in these disease-derived cells. Notably, these experiments were conducted in the absence of exogenous TGF-β, and combined with our results in Fig. 4B suggest that the role of STAT3 inhibition in ablating myofibroblast activation is not exclusive to TGF-β signaling, but appears to extend more broadly to activated fibroblasts cultured on stiff matrices. We therefore turned our attention to the possibility that STAT3 is a mechanoresponsive signal essential to the mechanoactivation of myofibroblasts

To investigate the responsiveness of STAT3 to matrix stiffness, we cultured both IMR-90 and primary normal human lung fibroblasts on soft and stiff matrices in the absence of exogenous TGF-β. Strikingly, we observed that cells cultured on soft matrices approximating normal physiological lung stiffness (1 kPa) significantly reduced STAT3 phosphorylation observed on more rigid hydrogel or plastic matrices (Fig. 6A,B), for the first time implicating STAT3 as a mechanoresponsive transcriptional effector in fibroblasts. This result was confirmed using both traditional western blotting with antibodies to p-STAT3 (p-Y705) and total STAT3, as well as an ELISA for pSTAT3 (p-Y705). Upstream activation of STAT3 is best known to occur via JAK signaling from IL-6 family ligand-receptor interactions, raising the question of whether mechano-activation of STAT3 requires extracellular autocrine signaling. We confirmed that IL-6 stimulation was sufficient to increase p-STAT3 levels on both soft and stiff matrices (Fig. 6C), demonstrating the presence and functional capacity of IL-6 signaling machinery in both states. However, conditioned medium harvested from fibroblasts growing on stiff matrices at the time of STAT3 activation was unable to enhance p-STAT3 levels when transferred to cells on soft matrices (Fig. 6D), arguing against an important contribution from autocrine regulation. To further test this concept, fibroblasts cultured on stiff matrices were incubated with neutralizing antibodies to the IL-6 receptor and gp130, the co-receptor for multiple IL-6 family cytokine receptors upstream of STAT3 phosphorylation. While both antibodies effectively blunted IL-6-driven increases in STAT3 phosphorylation (Fig. 6E), they did not reduce constitutive levels of STAT3 phosphorylation seen on stiff matrices (Fig. 6F), ruling out the important contribution of an extracellular IL-6 family autocrine loop in sustaining p-STAT3 on stiff matrices. Thus, while IL-6 (Fig. 6C) and TGF-β (Chakraborty et al., 2017) are capable of transiently activating STAT3, and inducing autocrine signals that activate STAT3 (Albrengues et al., 2015; Albrengues et al., 2014), our results demonstrate that matrix stiffness itself is also capable of stimulating ligand-independent STAT3 activation.

We therefore turned our attention to intracellular pathways that might account for stiffness-dependent STAT3 activation. Notably, recent work has identified ROCK as a potential upstream mediator of JAK family protein phosphorylation and activation of downstream STAT3 (Huang et al., 2012a). ROCK itself is well known to be responsive to alterations in matrix stiffness (Huang et al., 2012b), and was identified in the original RNAi screen as essential for myofibroblast activation (Table S2 and Fig. 1A). Small-molecule inhibitors of both ROCK (Fig. 7A) and JAK2 (Fig. 7B, AZD1480) inhibited constitutive STAT3 phosphorylation in a dose-dependent fashion on stiff matrices, confirming their essential roles. However, less-selective inhibitors of JAK1/2 and JAK3 (Fig. 7B, Ruxolitinib and Tofacitinib) demonstrated no effect on STAT3 phosphorylation, nor did an inhibitor of the mechanosensitive focal adhesion kinase FAK (Fig. 7C) (Bae et al., 2014). Taken together, these results directly implicate ROCK and JAK2 as specific upstream ligand-independent, matrix stiffness-dependent mediators of fibroblast STAT3 activation. To further confirm the dependence of STAT3 phosphorylation on mechanosensitive signaling, we inhibited myosin II with blebbistatin, and similarly confirmed dose-dependent reductions in p-STAT3 (Fig. 7A). In contrast, inhibition of the TGF-β receptor with SB-431542 was ineffective, further emphasizing that matrix stiffness-dependent STAT3 activation is independent of pro-fibrotic extracellular ligands, including TGF-β (Fig. 7C).

One implication of our in vitro results is that STAT3 may be involved in ligand-independent sustained activation of myofibroblasts driven by the stiffened matrix present during fibrosis progression. To test whether STAT3 inhibition can attenuate the progression of fibrosis in such a setting, we employed a relevant pre-clinical model of tissue fibrosis, one-time intratracheal bleomycin exposure, to provoke fibrotic tissue remodeling in the lungs of mice. Daily administration of LLL12 was started at day 14, with the lungs harvested at day 28, thereby testing the therapeutic relevance of STAT3 inhibition without confounding effects during the acute injury and inflammatory phase following bleomycin administration (Chaudhary et al., 2006; Foskett et al., 2014; Moeller et al., 2008). Histological examination demonstrated profound fibrotic remodeling in response to bleomycin, with relatively protected tissue architecture in LLL12-treated mice (Fig. 8A), quantified using Ashcroft score (Ashcroft et al., 1988) (Fig. 8B). We confirmed a trend toward increased STAT3 phosphorylation in bleomycin-treated lungs and significant reductions in STAT3 phosphorylation in LLL12-treated mice (Fig. 8C). Consistent with these observations, bleomycin-treated lungs exhibited a significant increase in hydroxyproline content, indicative of collagen deposition, which was significantly attenuated by LLL12 treatment (Fig. 8D). Pro-fibrotic transcripts Acta2, Col1a2 and Fn1 were also strongly induced in bleomycin-treated lungs, and their expression significantly attenuated in LLL12-treated mice (Fig. 8E). These results concur with a recent report demonstrating efficacy of another STAT3 inhibitor in a repetitive bleomycin-induced lung fibrosis model (Pedroza et al., 2015), with the key distinction that our findings using delayed intervention after a single acute bleomycin exposure emphasize a key role for STAT3 in driving the fibrotic phase of lung remodeling in vivo.

Our results offer the first comprehensive loss-of-function survey of genes essential for formation of the characteristic αSMA cytoskeleton of myofibroblasts, and are complementary to recent efforts aimed at discovering genes that confer myofibroblast activation through gain of function cDNA library expression (Davis et al., 2012, 2015). Identification of siRNA targeting ACTA2 (α-SMA), and the TGF-β receptors TGFBR1 and TGFBR2, as high-ranking hits confirmed the overall utility of our approach. Moreover, additional known regulators of myofibroblasts, including MKL1 (Crider et al., 2011; Huang et al., 2012b; Sandbo et al., 2011; Small et al., 2010) and SRF (Small, 2012; Yang et al., 2003) were similarly highlighted in our results, underscoring the fidelity of our high-content screening approach in identifying known myofibroblast regulators. Although the screen did not identify some known transcriptional regulators of myofibroblasts, including SMAD2 and SMAD3 downstream of TGF-β receptors, and YAP and TAZ (Bertero et al., 2015; Liu et al., 2015; Mannaerts et al., 2015), caution is necessary in interpreting such ‘negative’ results. For example, functional redundancy of closely related gene products may limit the effects of single-gene knockdown, while specific siRNA sequences and pools may lack efficacy for certain genes. Therefore, we focused exclusively on positive hits identified and validated through our comprehensive screening process. Although we note generally good agreement between our screen hits identified on tissue culture plastic and validation experiments performed on hydrogel matrices of more physiologically relevant stiffness, mimicking the transition between normal and fibrotic tissue stiffness, we note that primary screening on such hydrogels in the future, and across a wider range of stiffness, may identify additional regulators of fibroblast activation.

A number of the newly validated hits identified in our study are intriguing as biological effectors, and as potential targets for additional study, but have not previously been linked extensively to myofibroblast activation, wound healing or fibrosis. For example, neurotrophins such as brain-derived neurotrophic factor (BDNF) have been identified in human IPF tissue and implicated in fibroblast proliferation (Ricci et al., 2007), but have not been well studied as functional effectors in pulmonary fibrosis (Prakash et al., 2010). PCDH7 and its protein product protocadherin-7 have recently been identified as a signature gene and potential surface marker of activated hepatic stellate cells (Zhang et al., 2015). Although very little is known about its biological effects in fibroblasts beyond morphological and adhesion changes (Yoshida, 2003), our results suggest that PCDH7 plays an important functional role in myofibroblast activation that is worthy of further study. Leupaxin (LPXN) is a relatively little studied paxillin family member that functions within cell-matrix adhesions (Chen and Kroog, 2010). Intriguingly, leupaxin can shuttle to the nucleus and interact with transcriptional regulators SRF (Sundberg-Smith et al., 2008) and β-catenin (Shi et al., 2015), potentially linking cell-matrix adhesions to nuclear transcriptional effects. Given the key roles that matrix and mechano-signaling play in myofibroblast activation (Thannickal et al., 2014; Tschumperlin, 2015, 2013), the dual roles of leupaxin in cell-matrix adhesions and in the nucleus would seem to leave it well positioned to contribute to myofibroblast activation. We anticipate that further biological investigation will mechanistically link these and other hits identified from our high-content screen to myofibroblast activation, revealing novel mechanisms and networks that underlie this important phenotypic activation process.

To further explore mechanistic and functional insights enabled by our screen, we focused the current investigation on STAT3, which was identified as a central node interacting with multiple screen hits (Fig. 2E). This decision was influenced by the ambiguous nature of the existing data linking STAT3 to fibroblast activation and fibrosis. Although a limited number of studies have implicated STAT3 signaling in fibroblast activation and/or fibrosis in models of kidney, liver, dermal and cardiac scarring and fibrosis (Chakraborty et al., 2017; Dai et al., 2013; Khan et al., 2012; Pang et al., 2010; Ray et al., 2013; Shi et al., 2017; Su et al., 2015; Xu et al., 2014; Zepp et al., 2017), others have suggested protective (Mair et al., 2010; Yu et al., 2015), wound healing (Pickert et al., 2009) or regenerative roles for STAT3 (Jacoby et al., 2003; Moh et al., 2007), likely reflecting context- and cell type-specific roles for STAT3. Moreover, because STAT3 is a major downstream mediator of interleukin (IL-6) and associated family members and plays a prominent role in immune cell signaling and autoimmunity (Milner et al., 2015; Radojcic et al., 2010), it has been challenging to dissociate the effects of STAT3 in mouse models and human disease from alterations in inflammatory state and immune cell function. In our studies using cultured fibroblasts, STAT3 inhibition either by siRNA-mediated knockdown (Fig. 4), or small-molecule (LLL12)-mediated STAT3 inactivation (Fig. 5), dramatically reduced myofibroblast contractile function, collagen I and fibronectin deposition, and expression of genes implicated in myofibroblast activation and fibrogenesis. These results definitively demonstrate an important cell-autonomous role for STAT3 signaling in myofibroblast function, and are supported by the recent demonstration that fibroblast-specific deletion of STAT3 exhibits protective effects in experimental dermal fibrosis (Chakraborty et al., 2017). Moreover, the consistent nature of these effects across lung and cardiac fibroblasts as well as hepatic stellate cells (Fig. 5), in parallel with similar observations in dermal (Chakraborty et al., 2017; Khan et al., 2012) and kidney (Pang et al., 2010) fibroblasts, suggests a highly conserved role for STAT3 in fibroblast activation across multiple tissues. Importantly, our results provide a direct connection between a prominent inflammatory response pathway (STAT3) and myofibroblast activation, potentially identifying a common link between inflammation and myofibroblast activation across diverse contexts, including wound healing, autoimmunity, fibrosis and tumor-stroma interactions (Barnes et al., 2011; Dauer et al., 2005; Yu et al., 2014). Our observation that STAT3 phosphorylation is maintained in a ligand-independent and matrix stiffness-dependent fashion (Figs 6 and 7) has important implications for our understanding of how myofibroblast activation can transition from a transient wound-healing state to a state of persistent mechano-activation that contributes to feedback amplification of progressive fibrosis. Our data demonstrating ROCK- and JAK2-dependent, but ligand-independent, STAT3 activation echo recent findings of a critical role for this pathway in energy homeostasis (Huang et al., 2012a), and for the first time directly implicates STAT3 as a critical mechanoresponsive factor driving fibroblast fate. Recent findings that STAT3 activation can also promote matrix remodeling and stiffening in pancreatic cancer models (Laklai et al., 2016), and our findings that STAT3 is crucial for myofibroblast activation, support a central position for STAT3 as both responder and coordinator of fibrotic tissue remodeling. Future work aimed at identifying the stiffness threshold for STAT3 activation in vivo, and the upstream molecular mechanisms linking matrix stiffness to STAT3 activation should be prioritized.

The dramatic effects of STAT3 inhibition on myofibroblast function and gene expression (Fig. 5) motivated our efforts to evaluate therapeutic targeting of this pathway in a mouse model of pulmonary fibrosis. Prior work with STAT3 inhibitors suggested anti-fibrotic efficacy in a murine unilateral ureteral obstruction model of kidney fibrosis (Pang et al., 2010), and in a repetitive intraperitoneal bleomycin model of pulmonary fibrosis (Pedroza et al., 2015). In both cases, STAT3 inhibitors were delivered while injury and inflammation were ongoing, with demonstrated effects on inflammatory cell infiltrates, complicating interpretation of STAT3 direct effects on fibroblast activation and fibrogenesis. Therefore, we applied the STAT3 inhibitor LLL12 from days 14 to 28 after one time intratracheal bleomycin-induced lung injury; this approach tests the role of STAT3 in an active fibrotic phase with minimal ongoing injury and inflammation (Moeller et al., 2008) where IL-6 levels have returned to baseline (Chaudhary et al., 2006; Foskett et al., 2014). Our results demonstrate significant attenuation of lung fibrosis with LLL12 treatment, as measured histologically and by hydroxyproline quantification, and demonstrate potent inhibitory effects on pro-fibrotic gene expression (Fig. 8). To extend the therapeutic relevance of STAT3 inhibition to human disease-derived cells, we also evaluated the efficacy of targeting STAT3 in human cells derived from patients with IPF. Increased STAT3 activation has previously been documented in samples taken from fibrotic human lung tissue (Milara et al., 2018; O'Donoghue et al., 2012; Pedroza et al., 2015). While such activation has previously been ascribed to cytokine-based cellular stimulation, our in vitro observation of stiffness-dependent but ligand independent STAT3 phosphorylation, combined with the known matrix stiffening present in fibrotic lung pathologies (Liu et al., 2016, 2015), should prompt reconsideration of the potential role of STAT3 in persistent myofiboblast mechanoactivation. In cultured primary IPF-derived fibroblasts, we found robust effects of STAT3 inhibition in reducing myofibroblast traction forces and pro-fibrotic gene expression (Fig. 5), consistent with our original image-based siRNA screen results obtained with an independent cohort of IPF fibroblasts (Fig. 3). Taken together, these results establish a strong rationale for directly targeting STAT3 as an anti-fibrotic therapeutic approach, an approach that has already shown promise in experimental dermal fibrosis (Chakraborty et al., 2017).

In summary, our screen results identify a panel of novel candidate genes that are essential for expression of a myofibroblast phenotype, and demonstrate a pivotal role for STAT3 in myofibroblast mechano-activation. Although STAT3 has historically proven challenging to target in the clinic (Yue and Turkson, 2009), recently developed strategies have shown promise (Hong et al., 2015), raising the possibility that therapeutic intervention in fibrotic pathologies via STAT3 inhibition may be tractable. Additional promise, and selectivity, may come from understanding the network of regulators that interact with STAT3, such as JAK2, and cell-specific factors that account for STAT3 activation in myofibroblasts. Beyond STAT3, the proven preclinical efficacy of interventions targeting other siRNA screen hits, including MKL1 (Haak et al., 2014; Sisson et al., 2015) and BRD4 (Ding et al., 2015; Tang et al., 2013a), raises hopes that additional candidate genes identified here will be amenable to future mechanistic dissection and potential therapeutic intervention. Having demonstrated the general utility of our screening approach, the door is now open to such pursuits.

IMR-90 cells (ATCC), from a frozen stock (<passage 10), were grown for 6 days in 10% FBS DMEM with 50 units/ml penicillin G, 50 µg/ml streptomycin sulfate. At day 5, cells were confluent. One day before transfection, 384 Cell Carrier plates (Perkin Elmer) were coated with 30 µl collagen solution (50 µg/ml in PBS, PurCol, AdvancedBioMatrix) and incubated at 4°C overnight. On transfection day, collagen was removed and plates were washed once with PBS. Plates were air dried for several hours before transfection.

On the day of transfection, cells were trypsinized (TrypLE, Invitrogen), collected in full medium, washed in serum-free DMEM, resuspended in serum-free DMEM, and counted and diluted in 0.1% FBS DMEM. Following this, Opti-MEM (15.9 µl/well) and Lipofectamine 2000 (Invitrogen; 0.1 µl/well) were mixed together and dispensed into 384 well plates using a Matrix WellMate. siRNA (1 µl of 1 µM) was added and mixed using an automated liquid handling system (Agilent Bravo). Complexes were allowed to form for 20 min at room temperature. TGF-β was added to cells (2 ng/ml final concentration) and cells were immediately dispensed (33 µl, 1500 cells/well). Total volume per well was 50 µl, with a final siRNA concentration of 20 nM. Plates were rocked gently 5 times and cells were allowed to settle and adhere for 1 h at room temperature, after which time plates were placed in the humidified 37°C, 5% CO₂ incubator.

Sixty-eight to seventy hours post-transfection, medium was aspirated by wand, and cells were fixed with paraformaldehyde (4% PFA in PBS, Santa Cruz Biotechnology, 20 min), permeabilized (0.2% Triton-X 100 in PBS, 5-10 min), blocked (3% BSA in PBS, 1 h) and stained with anti-smooth muscle actin antibody (clone 1A4, Sigma-Aldrich, 1:600 dilution, PBS, 1%BSA, 0.02% Triton X-100, overnight, 4°C). Cells were washed twice (0.02% Triton X-100 in PBS) and then stained with goat anti-mouse IgG2a Alexa-Fluor-488-conjugated secondary antibody (LifeTechnologies, 1:500 dilution in PBS, 1% BSA, 0.02% Triton X-100, 1 h), stained with Alexa-Fluor-647-conjugated phalloidin (Life Technologies, 1:150 dilution in 0.02% Triton X-100, PBS, 20 min), stained with Hoechst 33342 (Life Technologies, 1:2000 dilution in 0.02% Triton X-100, PBS, 5-10 min), and then washed with 0.02% Triton X-100 in PBS, and finally covered with PBS, and sealed. Nine images per well were acquired within 3 days on the Opera confocal high-content microscope using a 20× objective (Perkin Elmer).

Library plates were screened in triplicate, and all three assay plates in a given set were prepared on the same day. Plate controls include negative control non-targeting control siRNA #3 (Dharmacon), and positive controls MKL1 (ONTarget Pool) and TGFBR1 (ONTarget Pool). Throughout the screen, Z-prime scores using these plate controls were monitored to evaluate assay quality. The following Dharmacon SMARTpool siRNA libraries were screened: 779 kinases/phosphatases, 516 GPCRs, 6022 druggable genome. The complete library gene list is provided in Table S1.

Microscopy images from the screen were analyzed using a customized image analysis algorithm developed using MATLAB. To segment single cells, the algorithm first performs nuclear segmentation using Hoechst channel images. Threshold is determined for the DAPI channel using the Ridler-Calvard method, then clusters of nuclei are divided based on the distance transform of the thresholded DAPI image. To identify the boundary of whole cells, the algorithm takes advantage of basal level non-specific signal in both the F-actin and αSMA channels. Thresholds of basal level signal in both channels are determined based on the non-cell background region intensity level in each image. The threshold binary masks from both channels are combined into one raw whole cell mask, where the masks of some neighboring cells are fused into each other. To separate them, an intensity-based watershed approach is applied using segmented nuclei from step 1 as seeds.

In the third step, the same approach is used to detect and reconstruct individual fibers from both F-actin and αSMA channels. The raw fiber image is first processed using steerable filtering that results in a strong response from ridge-like signals in the image, and also generates the corresponding angular map associated with the ridge response. The ridge response image is converted to a binary image using a fixed threshold value. In addition, we observed that fibers in a single cell are normally straight and largely oriented in a similar direction. The algorithm uses this feature to refine the fiber detections by estimating the preferred local fiber orientation, and then filters out any detected fiber-like objects oriented in different directions. To further refine the detection, the algorithm attempts to link broken segments of the same fibers back together by searching for fiber ends that are in close proximity with the line linking two end points in the same direction as the two segments. In order to increase the accuracy of quantification, we decided to focus only on individual fibers by removing the ones that are bundled closely with others from the detection. In the end, fibers detected from both F-actin and αSMA channels are overlaid to classify the fiber composition, and quantification was done at the single-cell level. Cell overlap was at times significant, resulting in inaccurate cell segmentation; to increase accuracy in quantification, the per area average score of detected fibers was tabulated for the entire image.

Average well values (from 9 images) were tabulated for two fiber metrics: normalized total F-actin intensity area and normalized αSMA fiber intensity area. The robust z-score [z=(x–m)/MAD, where x is the well data in question, m is the plate median, and MAD is median absolute deviation] was calculated for each SMARTpool. A SMARTpool was considered a positive if the well absolute z-score was ≥2 in at least 2 of 3 plate replicates for αSMA fiber intensity. Positives were further defined as weak (z-score ≥2 and ≤2.5, ∼15% change from non-targeting control), medium (z-score >2.5 and <3, ∼20-30% change compared with control) and strong (z-score ≥3, ∼30% change or greater). Arbitrary scores were then assigned for weak (score±1), medium (±2) and strong (±3). A combined weighted score for three replicates was tabulated, creating a possible range of −9 (all three replicates strong inhibitors) to +9 (all three replicates strong enhancers). A score of 7-9 (absolute value) was classified as a ‘strong’ hit (inhibitor or enhancer); a score of 4-6 was a ‘moderate’ hit; a score of 2-3 was a ‘weak’ hit. A separate category of positives that hit for only total-F-actin intensity was similarly tabulated.

To identify potential off-target gene transcripts, siRNA seed sequence analysis was performed comparing the siRNA seed sequence enrichment in primary screen hit siRNAs versus non-hit siRNAs. Genome-wide enrichment of seed sequence matches (GESS) (Sigoillot et al., 2012) looks for matches to 3′ untranslated regions (UTRs) of database genes and Dharmacon seed sequence analysis (Sudbery et al., 2010) scans for miRNA-like sequences.

To eliminate potential screen false positives due to off-target effects, individual siRNA duplexes from selected positive SMARTpools (4 per pool) were screened in deconvolution screening on collagen coated 384-well plates. Assay conditions, image acquisition settings and analysis were identical to primary screening. Hits were identified by plate based z-score of αSMA fiber intensity [z=(x–n)/s.d., where x is the well data in question, n is the average of 12 negative control wells, and s.d. is standard deviation of negative control wells]. A siRNA duplex was considered a positive if the well absolute z-score was ≥2 in at least 2 of 3 plate replicates. Positives were further defined as weak (z-score ≥2 and ≤3.0), medium (z-score >3.0 and <5.0) and strong (z-score ≥5). Arbitrary scores were then assigned for weak (score±1), medium (±2) and strong (±3). A combined weighted score for three replicates was tabulated, as described above. A score of 7-9 (absolute value) was a classified as a ‘strong’ hit (inhibitor or enhancer); a score of 4-6 was a ‘moderate’ hit; a score of 2-3 was a ‘weak’ hit. For a set of 4 duplexes to be validated, the ‘hit rate’ required at least 2 out of 4 siRNAs to meet the ‘weak’ criteria. A single score of weak, moderate or strong Inhibitor or Enhancer was given to each validated set, based on the single best performing siRNA. For example, if one pool has two siRNAs, one scoring SMA strong inhibitor S(I) and one scoring weak inhibitor W(I), this pool will be labeled S(I).

Selected positives from deconvolution screening were tested on 96-well collagen-coated polyacrylamide hydrogel plates mimicking the transition zone between normal and fibrotic tissue stiffness (Liu et al., 2015, 2010) (13 kPa Young's modulus, Matrigen). As in deconvolution screening, 4 siRNAs per gene were tested. Assay conditions were scaled for 96-well format. Opti-MEM (36.7 µl/well) and Lipofectamine 2000 (0.25 µl/well) were mixed and dispensed in 96-well plates using a Matrix WellMate. siRNA (3 µl of 1 µM stock) was added and mixed using an automated liquid handling system (Agilent Bravo). Complexes were allowed to form for 20 min at room temperature. TGF-β was added to cells (2 ng/ml final concentration) and cells were immediately dispensed (110 µl, 4000 cells/well). Total volume per well was 150 µl, with final siRNA concentration of 20 nM. Cell incubation, fixing and staining conditions were as previously described. Images were acquired (4 sites/well) with the Image Xpress Micro plate reader using a 10× objective (Molecular Devices) and analyzed for fiber metrics by MATLAB code, similar to the primary screen. Hits were identified by plate-based z-score of αSMA fiber intensity [z=(x–n)/s.d., where x is the well data in question, n is the average of 4 negative control wells, and s.d. is standard deviation of negative control wells]. A siRNA duplex was considered a positive if the well absolute z-score was ≥2 in at least 2 of 3 plate replicates. Positives were further defined as weak (z-score ≥2 and ≤4), medium (z-score >4.0 and <7.0) and strong (z-score ≥7). Arbitrary scores were then assigned for weak (score±1), medium (±2) and strong (±3). A combined weighted score for three replicates was tabulated, as previously described. For a set of 4 duplexes to be validated, the ‘hit rate’ required at least 2 out of 4 siRNA meeting the ‘weak’ criteria. A single score of weak, moderate or strong inhibitor or enhancer was given to each validated set, based on the single best performing siRNA.

Selected positives from gel plate screening were tested on 96-well collagen-coated polyacrylamide hydrogel plates (13 kPa Young's modulus, Matrigen) in three independent IPF cell cultures (IPF-13, IPF-16, IPF-33) and in two independent experiments. The strongest two siRNAs per gene, as determined from prior gel and deconvolution screening steps, were tested. Assay conditions were identical to gel plate screening. In addition to αSMA fiber measurement, intracellular procollagen was measured. Fixed and permeabilized cells were stained with anti-procollagen type I antibody (MAB1913, EMD Millipore) along with anti-αSMA antibody, overnight. Cells were washed with PBS and stained with anti-IgG₁-Alexa Fluor 546 and anti-IgG_2a secondary antibodies. Cells were washed with PBS and stained with phalloidin and Hoechst as described above. Image analysis by MATLAB algorithm generated a measurement for average procollagen intensity per cell in addition to the previously established αSMA fiber measurement. Hits were identified by plate based z-score for αSMA fiber intensity as described for gel plate screening. Hit strength was determined by combined weighted score of three replicate plates, as described above. The identical z-score and hit classification was applied to the per cell procollagen measurement.

Primary human lung fibroblasts used in siRNA screening and follow-up experiments were isolated by explant culture from the lungs of subjects diagnosed with IPF who underwent lung transplantation, or donors whose organs were rejected for transplantation (non-IPF). Cells were isolated at the University of Pittsburgh Medical Center, under a protocol approved by the University of Pittsburgh Institutional Review Board, or were generously provided by Peter Bitterman and Craig Henke at the University of Minnesota, after isolation under a protocol approved by the University of Minnesota Institutional Review Board. Primary fibroblasts were maintained in EMEM (ATCC) containing 10% FBS, unless otherwise noted. Primary human adult cardiac fibroblasts (ScienCell) were maintained in Fibroblast medium-2 (ScienCell) and primary human hepatic stellate cells (ScienCell) were maintained in stellate cell medium (ScienCell). All primary cell culture experiments were performed with cells at passage eight or less, and IMR-90 and primary cells were subject to routine mycoplasma testing.

For in vitro experiments, cells were treated as indicated prior to RNA isolation using RNeasy Plus Mini Kit (Qiagen) according to the manufacturer's instructions. Isolated RNA (250 ng) was then used to synthesize cDNA using SuperScript VILO (Invitrogen). Quantitative PCR was performed using FastStart Essential DNA Green Master (Roche) and analyzed using a LightCycler 96 (Roche). Data are expressed as a fold change by ΔΔCt relative to glyceraldehyde-3-phosphate dehydrogenase (GAPDH). For in vivo experiments, tissue was immediately frozen, and stored at −80°C. RNA isolation, cDNA synthesis and qPCR analysis were performed as above. Primers used for qPCR are shown in Table S9.

Polyacrylamide substrates with Young's modulus of 13 kPa were prepared as previously described (Marinkovic et al., 2012) and fluorescent sulfate-modified latex microspheres (0.2 μm, 505/515 excitation/emission, FluoSpheres, Life Technologies) were conjugated to the gel surfaces after treatment with 1 mg/ml of dopamine hydrochloride (Sigma-Aldrich) in 50 mM HEPES solution (pH 8.5) for 20 min. Cells were plated on the gels overnight and treated as indicated before traction force measurements. Images of gel surface-conjugated fluorescent beads were acquired for each cell before and after trypsinization using a Nikon ECLIPSE Ti microscope at 10× magnification. Two-dimensional tractions were estimated by measuring bead displacement fields and computing corresponding traction fields using TractionsForAll (http://www.mayo.edu/research/labs/tissue-repair-mechanobiology/software).

IMR-90 cells were plated onto 8-well chamber slides (Thermo) for 24 h and then the medium was exchanged for EMEM (ATCC) containing 0.1% FBS ±2 ng/ml TGF-β (eBioscience) and incubated for 72 h. In the final 24 h, the indicated concentration of LLL12 (BioVision) or DMSO was added. Cells were fixed in 3.7% formalin (Sigma), treated with 0.25% Triton X-100 (Sigma), and then blocked with 5% BSA for 1 h. Cells were incubated overnight with a FITC-conjugated mouse monoclonal antibody against αSMA (Sigma, F3777) diluted 1:200 in PBS with 1% BSA. Cells were then washed and mounted with ProLong Antifade with DAPI (Thermo). Slides were imaged using a Nikon ECLIPSE Ti microscope at 20× magnification. For scoring, αSMA-positive cells were binned using an optical threshold and the observer was blinded to the treatment.

Adapting previously published methods (Vogel et al., 2014), IMR-90 cells were plated to confluence in clear-bottom 96-well plates. After cells attached, the medium was swapped for EMEM containing 0.1% FBS ±2 ng/ml TGF-β plus the indicated concentration of LLL12 or DMSO and incubated for 72 h. Cells were then lysed and lifted from the plate with 3.2 mM NH₄OH. Wells were washed with Tris-buffered saline (TBS) and blocked with Li-Cor Odyssey Blocking Buffer for 60 min before overnight incubation in a polyclonal rabbit antibody for fibronectin (Sigma sc-9068) or Collagen I (Novus NB600-408) diluted 1:100 in blocking buffer. Wells were washed in TBS with 0.25% Tween 20 before incubation for 45 min with IR-dye-conjugated secondary antibody (Li-Cor #926-32211) diluted 1:400. Plates were imaged via a Li-Cor OdysseyXL system with quantification performed via densitometry.

Forty 8-week-old female C57/BL6 mice were purchased from Charles River Laboratories. Mouse lung fibrosis was induced in 20 mice with bleomycin (BLEO; Fresenius Kabi) delivered intratracheally (3 U/kg) to the lungs using MicroSprayer Aerosolizer (Penn-Century). The 20 control (Ctrl) mice received sterile 0.9% saline instead using identical methods. Mice were weighed every 24-48 h, and both groups were then randomized at day 14 into LLL12 and DMSO treatment groups, matching for the degree of weight change. Four mice in the bleomycin group died or were euthanized based on predetermined criteria prior to treatment, thus 8 bleomycin-treated mice each were randomized to LLL12 and DMSO treatment groups. The STAT3 inhibitor LLL12 was administered everyday intraperitoneally (i.p.) for 14 days. The control groups of mice received the equivalent vehicle dose of dimethylsulfoxide (DMSO) instead. Two mice in the Bleo/DMSO group died, at days 15 and 20. One mouse in the Bleo/LLL12 group died on the first day after treatment onset (day 14). Two mice in the Saline/LLL12 group died during the study, the first acutely shortly after the first administration of LLL12, and the second at day 22. The remaining number of mice in each group for analysis was: 10 Saline/DMSO mice, 8 Saline/LLL12 mice, 6 Bleo/DMSO and 7 Bleo/LLL12. Following the final injection, mice were euthanized by intraperitoneal injection of pentobarbital (100 mg/kg) and the right lungs were inflated with 4% paraformaldehyde (PFA) and further incubated in 4% PFA for 24 h prior processing for paraffin embedding. The left lobe of the lung was snap frozen in liquid nitrogen for RNA isolation and hydroxyproline assay. All animal experimental procedures were approved by the Mayo Clinic Institutional Animal Care and Use Committee and the animals were handled in accordance with their guidelines.

Five-µm-thick sections were cut from Paraffin-embedded lung tissues, and the sections were stained either with hematoxylin and eosin (H&E) or with Masson's Trichrome stain kit (Abcam). All H&E-stained slides and trichrome-stained slides were reviewed in a blinded fashion by a thoracic pathologist. The severity of interstitial and peribronchiolar lung immature and mature fibrosis was estimated on a numerical scale according to Ashcroft et al. (1988). For scoring purposes, all H&E-stained slides were systematically scanned at 100× magnification and successive 100× fields were scored. Scoring was based on the following scale: 0 (no fibrosis), 1 (minimal interstitial and/or peribronchiolar thickening due to fibrosis), 3 (moderate thickening without obvious architectural distortion), 5 (increased fibrosis with formation of fibrous bands and/or small masses), 7 (severe architectural distortion with large areas of fibrosis and areas of honeycomb changes) and 8 (total fibrous obliteration of the field). The predominant score for each field was recorded. The mean of all scores was calculated for each case.

Hydroxyproline content was measured using a hydroxyproline assay kit (Biovision) according to the manufacturer's instructions with slight modification. The lung tissues were weighed, homogenized in sterile water (10 mg of tissue per 100 μl H₂O) and hydrolyzed in 12 M HCl in a pressure-tight, Teflon capped vial at 120°C for 3 h followed by filtration through a 45 μm Spin-X Centrifuge Tube filter (Corning). Ten µl of the hydrolyzed samples was dried in a Speed-Vac for 2 h, followed by incubation with 100 μl of Chloramine T reagent for 5 min at room temperature and 100 μl of 4-(Dimethylamino)benzaldehyde (DMAB) for 90 min at 60°C. The absorbance of oxidized hydroxyproline was determined at 560 nm. Hydroxyproline concentrations were calculated from a standard curve generated using known concentrations of trans-4-hydroxyl-L-proline. The total amount of protein isolated from the weighed tissues was determined by using a protein assay kit (Bio-Rad, absorbance at 595 nm). The amount of collagen was expressed in μg/mg total protein.

IMR-90 or primary human lung fibroblasts derived from IPF patients were plated into 6-well plates. Total protein was isolated using RIPA buffer (pH 8.0) containing Pierce Phosphatase Inhibitor (Thermo) and Halt Protease Inhibitor Cocktail (Thermo). For mouse lung tissue analysis, formalin-fixed paraffin-embedded tissue was deparaffinized in xylene and rehydrated before protein isolation as above. Lysate total protein concentration was determined using Pierce BCA Protein Assay Kit (Thermo) and samples were run on a 10% polyacrylamide gel. Blots were incubated overnight with the following primary antibodies: pSTAT3 (Tyr 705, Santa Cruz, sc-8059), STAT3 (Santa Cruz, sc-482) and GAPDH (Cell Signaling, 14C10) diluted 1:1000 in Li-Cor Odyssey Blocking Buffer. Blots were washed with TBS-Tween before 60 min incubation with IR-dye-conjugated secondary antibodies (Li-Cor) diluted 1:10,000. Plates were imaged using a Li-Cor OdysseyXL system with quantification performed via densitometry. Each antibody produced one obvious band of the appropriate size, and STAT3 antibody was validated using STAT3 siRNA.

NHLF cells (2×10⁵ cells, Lonza) were plated onto soft hydrogel (1 kPa, Young's modulus) or tissue culture plastic in 12-well plates at 37°C for 24 h in FGM-2 medium (Lonza) containing 0.1% FBS. Positive control response to IL-6 was determined by adding 50 ng/ml of IL-6 (R&D) to the cells and incubating for 60 min followed by measuring STAT3 phosphorylation. STAT3 phosphorylation at tyrosine residue 705 was examined in cell lysates using STAT3 (pY705) ELISA kit (Invitrogen, KHO0481) according to the manufacturer's protocol. Extracellular IL-6 receptor activation was blocked using anti-IL-6r (30 mg/ml, R&D) or anti-gp130 (30 mg/ml, R&D) at 37°C for 30 min prior incubation with IL-6, or for 30 min on 1 kPa or tissue culture plastic. To test roles in STAT3 phosphorylation, inhibitors of ROCK (Y27632, 3-30 µM, STEMCELL Technologies), myosin II (Blebbistatin 3-30 µM, Selleckchem), JAK1 (Ruxolitinib, 1-10 µM, Selleckchem), JAK2 (AZD1480, 1-10 µM, Selleckchem), JAK3 (Tofacitinib, 1-10 µM, Selleckchem), TGF-β type I receptor (1-10 µM, Selleckchem) and FAK (PF562271, 1-10 µM, Selleckchem) were added to the cells and incubated for 1 h prior to lysis and measurement of STAT3 phosphorylation.

For follow-up experiments, cells were transfected using Lipofectamine RNAiMAX (Life Technologies) with siGENOME siRNA (Dharmacon) targeting STAT3 (D-003544-04) or a nontargeting SMARTpool (D-001810-10-05). Cells were plated into 6-well plates and after attachment medium was exchanged for EMEM containing 0.1% FBS ±2 ng/ml TGF-β and the manufacturer's suggested concentration of RNAiMAX with 25 nM of each targeting siRNAs or equivalent amounts of nontargeting siRNA. Cells were cultured for 72 h before collecting material for qPCR or western blot analysis. For traction force microscopy experiments, the cells were transfected in the hydrogel assay plate.

In experiments comparing two groups, groups were compared using unpaired t-test with Welch's correction. In experiments comparing three or more groups, groups were compared by one-way ANOVA with Tukey's multiple comparison test. All statistical tests were carried out using GraphPad Prism 6 with statistical significance defined as P<0.05. Results are expressed throughout as the mean±s.e.m.

The authors wish to acknowledge Andreas Schnapp at Boehringer-Ingelheim and Jennifer Smith, Stewart Rudnicki, Sean Johnston, David Wrobel and Jen Nale at the Harvard Medical School Institute of Chemistry and Cell Biology Screening Facility, and Stephanie Mohr and Ian Flockhart at the Drosophila RNAi Screening Center for helpful discussions regarding design, implementation and interpretation of siRNA screen.