Transforming growth factor β1 alters the 3′-UTR of mRNA to promote lung fibrosis

Idiopathic pulmonary fibrosis (IPF) is a chronic disease characterized by the pathological remodeling of air sacs as a result of excessive accumulation of extracellular matrix (ECM) proteins, but the mechanism governing the robust protein expression is poorly understood. Our recent findings demonstrate that alternative polyadenylation (APA) caused by NUDT21 reduction is important for the increased expression of fibrotic mediators and ECM proteins in lung fibroblasts by shortening the 3′-untranslated regions (3′-UTRs) of mRNAs and stabilizing their transcripts, therefore activating pathological signaling pathways. Despite the importance of NUDT21 reduction in the regulation of fibrosis, the underlying mechanisms for the depletion are unknown. We demonstrate here that NUDT21 is depleted by TGFβ1. We found that miR203, which is increased in IPF, was induced by TGFβ1 to target the NUDT21 3′-UTR, thus depleting NUDT21 in human and mouse lung fibroblasts. TGFβ1-mediated NUDT21 reduction was attenuated by the miR203 inhibitor antagomiR203 in fibroblasts. TGFβ1 transgenic mice revealed that TGFβ1 down-regulates NUDT21 in fibroblasts in vivo. Furthermore, TGFβ1 promoted differential APA of fibrotic genes, including FGF14, RICTOR, TMOD2, and UCP5, in association with increased protein expression. This unique differential APA signature was also observed in IPF fibroblasts. Altogether, our results identified TGFβ1 as an APA regulator through NUDT21 depletion amplifying pulmonary fibrosis.

preferred for poly(A), leading to transcripts with long 3Ј-UTRs (20). However, in the cells that exhibit robust protein expression, proximal polyadenylation sites are preferred for poly(A), leading to 3Ј-UTR shortening (20). The shortening of the 3Ј-UTR can lead to robust protein expression due to the lack of mRNA degradation elements on the 3Ј-UTR (16 -18). Nudix hydrolase 21 (NUDT21), also known as cleavage factor Im 25 (CFIm25), which is a key subunit of the poly(A) machinery, is one of the main regulators of APA in cancer cells, neurons, and stem cells (21)(22)(23)(24). NUDT21 reduction by siRNA is sufficient to promote global mRNA shortening by APA-associated increased protein expression and cell proliferation in HeLa and stem cells (21,24). Furthermore, down-regulation of NUDT21 increases tumor growth in the glioblastoma xenograft model. Although NUDT21 reduction is implicated as a key regulator for APA and enhanced protein expression in pathological and physiological conditions, the mechanisms that promote NUDT21 depletion are not fully known.
Our recent studies demonstrated that APA contributes to pulmonary fibrosis, which is characterized by uncontrolled production of profibrotic mediators and ECM (25). NUDT21 is down-regulated in IPF fibroblasts, and knockdown of NUDT21 in human lung fibroblasts promotes global mRNA shortening through APA. Furthermore, it was demonstrated that global mRNA shortening through APA and NUDT21 reduction enriched profibrotic pathways such as TGF␤, Wnt, and vascular endothelial growth factor, and NUDT21 deletion in lung fibroblasts worsened lung fibrosis induced by bleomycin in vivo, suggesting that NUDT21 down-regulation in IPF exacerbates the severity of the disease. Moreover, knockdown of NUDT21 was sufficient to promote collagen and fibronectin expression in human lung fibroblasts through APA (25). These data suggest that down-regulation of NUDT21 worsens lung fibrosis by up-regulating ECM production, presumably via global mRNA shortening of profibrotic genes. Despite the importance of NUDT21 reduction in pulmonary fibrosis, the mechanisms governing the reduction are still unknown. Knowledge of the mechanisms of NUDT21 depletion can potentially lead to innovative therapies. In this study, we sought to investigate the mechanisms by which lung fibroblasts down-regulate NUDT21 and therefore promote APA for profibrotic pathways.

Transforming growth factor ␤1 (TGF␤1) down-regulates NUDT21 in primary human lung fibroblasts
To identify a profibrotic cytokine that down-regulates NUDT21 in human lung fibroblasts, primary fibroblasts (CCD8Lu) were cultured with seven cytokines that are associated with pulmonary fibrosis: TGF␤1, insulin-like growth factor 1, connective tissue growth factor, platelet-derived growth factor BB, interleukin (IL) 4, IL6, and IL13 for 7 days (26 -28). TGF␤1 down-regulated NUDT21 to the greatest extent in fibroblasts (Fig. 1A). Next, to longitudinally determine the capacity of TGF␤1 to deplete NUDT21, fibroblasts were cultured with TGF␤1 and harvested each day for Western blot analysis (Fig. 1B). NUDT21 was down-regulated by TGF␤1 after day 3. To validate NUDT21 reduction by TGF␤1, primary human lung fibroblasts isolated from three donor lungs were cultured for 4 days. NUDT21 was found to be consistently down-regulated following TGF␤1 treatment, suggesting that NUDT21 reduction by TGF␤1 may be a general cellular feature in human lung fibroblasts (Fig. 1C). We previously reported that fibroblasts from IPF patients displayed reduced levels of NUDT21 (25). To determine whether TGF␤1 can further deplete NUDT21 in IPF fibroblasts, we treated the IPF fibroblasts with TGF␤1 and found that TGF␤1 additionally reduced NUDT21 in the fibroblasts that had high basal NUDT21 levels Primary normal human lung fibroblasts (CCD8Lu) were cultured with human recombinant TGF␤1, insulin-like growth factor 1 (IGF1), connective tissue growth factor (CTGF), platelet-derived growth factor BB (PDGF BB), IL4, IL6, and IL13 for 5 days. The protein lysates were harvested for Western blot analysis to probe NUDT21 and ␤-actin protein levels. ␤-Actin was used as a loading control (A). The lung fibroblasts were cultured with TGF␤1 for up to 6 days. ␣SMA was used as a positive control for TGF␤1 treatment (B). Primary human lung fibroblasts from three independent human donors were cultured with TGF␤1 for 4 days (C). The primary human lung fibroblasts were challenged with TGF␤1 for 2 days and then harvested for real-time qPCR analysis. 18S was used as a normalization control for the assay (D). ****, p Ͻ 0.0001. Veh, vehicle; D, day; HD, human donor. Error bars represent S.D. (Fig. S1). In addition to protein reduction, NUDT21 mRNAs were also reduced following 24 h of treatment with TGF␤1 in CCD8Lu cells and in the three primary human fibroblasts, suggesting that NUDT21 protein reduction correlates with mRNA reduction (Fig. 1D).

miR203 targets the 3-UTR of NUDT21
NUDT21 mRNA reduction by TGF␤1 led to the hypothesis that TGF␤1 facilitates NUDT21 mRNA degradation in human lung fibroblasts (Fig. 1D). To test this hypothesis, mRNA degradation assays were performed. Human lung fibroblasts were cultured with TGF␤1 for 9 h because 24-h TGF␤1 treatment is required to reduce NUDT21 mRNA as measured by real-time PCR; this 9-h time period was intentionally chosen to follow mRNA degradation. After treatment, the cells were incubated with actinomycin D, which is a transcription inhibitor, allowing only mRNA degradation to affect the remaining transcript levels, and the samples were harvested to measure NUDT21 mRNA transcript levels in TGF␤1 and control groups. The cells with TGF␤1 treatment showed a rapid NUDT21 mRNA degradation as compared with a vehicle control (Fig. S2). This result suggests that NUDT21 reduction by TGF␤1 is due to enhanced NUDT21 mRNA degradation. TGF␤1 up-regulates various miRNAs, which can potentially affect the stability and degradation of target mRNAs (29 -32).
Therefore, it was hypothesized that TGF␤1 down-regulates NUDT21 by up-regulating miRNAs that target NUDT21 transcripts. TargetScanHuman showed that miR23a, miR27b, and miR203 are predicted to target human NUDT21 3Ј-UTR with at least two binding sites ( Fig. 2A). To examine whether NUDT21 3Ј-UTR is indeed a target of these miRNAs, a Dual-Luciferase 3Ј-UTR reporter assay was performed in vitro. In the vector, the firefly luciferase gene was conjugated with human NUDT21 3Ј-UTR, allowing the gene expression of firefly luciferase to be regulated by the 3Ј-UTR. The Renilla luciferase gene was used to normalize the number of transfected vectors in the cells. HEK293T and NIH3T3 cells were transfected with miR23a, miR27b, or miR203 and with the Dual-Luciferase vectors to monitor the effect of miRNAs on firefly luciferase expression. miR203 was found to consistently decrease the firefly luciferase expression in both cell types, indicating that exogenous miR203 targets the 3Ј-UTR of NUDT21 to regulate gene expression (Fig. 2B). To assess whether these miRNAs can affect the endogenous protein expression of NUDT21, human lung fibroblasts were transfected with the miRNA mimics. Indeed, NUDT21 protein was significantly reduced by miR203 and miR27b mimics (Fig. 2, C and D). Collectively, miR203 most consistently targeted the 3Ј-UTR of NUDT21 and most robustly reduced protein levels in all experiments. Therefore, HEK293T and NIH3T3 cells were transfected with Dual-Luciferase plasmids containing NUDT21 3Ј-UTR behind the firefly coding sequence in the presence of miR203, miR23a, miR27b, or miRNA control (miR-Con). A Dual-Luciferase assay was performed to measure luciferase intensities (B). Primary normal human lung fibroblasts were transfected with miRNA control (miR-Con), miR203, miR27b, or miR23a and cultured 2 days. The cells were harvested to probe NUDT21 protein levels (C). Densitometry analysis was performed to quantify NUDT21 in the transfected human primary samples (D). Primary human lung fibroblasts isolated from healthy normal or IPF patient lungs were lysed for RNA extraction. miR203 levels in the samples were measured. RNU6 was used as a normalization control (E). *, p Ͻ 0.05; **, p Ͻ 0.01; ****, p Ͻ 0.0001. Error bars represent S.D.

TGF␤1 regulates NUDT21 and 3-UTR in lung fibrosis
miR203 was chosen for further study. To examine whether miR203 elevation is associated with IPF, fibroblasts isolated from four different IPF lungs and four healthy lungs were used for RT-PCR analysis. In the IPF fibroblasts, miR203 was markedly up-regulated in association with an elevated ␣ smooth muscle cell actin (␣SMA) expression (Fig. 2E).

TGF␤1 targets NUDT21 mRNA through miR203
To investigate whether TGF␤1 can promote miR203 expression, human lung fibroblasts were cultured with TGF␤1 for up to 2 days in vitro. These experiments demonstrated that starting at 2-4 h after TGF␤1 treatment, increased miR203 levels were apparent (Fig. S3), and multiple primary human lung fibroblasts showed TGF␤1-associated miR203 expression (Fig.  3A). Studies have demonstrated that miRNA up-regulation by TGF␤1 is mediated through Smad pathways in vitro for various miRNAs (30 -32), and Smad3 acts as a master regulator for fibrogenesis and fibroblast activation through phosphorylation on serine 423/425 (33). Therefore, it was hypothesized that TGF␤1 up-regulates miR203 through phosphorylation of Smad3 on Ser-423/425. To test this hypothesis, a commercially available inhibitor of Smad3, SIS3, was used. This inhibitor inhibits phosphorylation of Smad3 specifically on Ser-423/425 residues (34). SIS3 successfully inhibited phosphorylation on Ser-423/425 on Smad3 in human lung fibroblasts (Fig. S4). Inhibition of Smad3 phosphorylation in the human lung fibroblasts was sufficient to reduce TGF␤1-mediated miR203 overexpression, demonstrating that miR203 induction is Smad3 phosphorylation-dependent (Fig. 3B). These results were consistent with SIS3 inhibiting TGF␤1-mediated NUDT21 reduction in primary healthy fibroblasts (Fig. 3, C and D). Moreover, the inhibition of Smad3 alone by SIS3 increased NUDT21 mRNA and decreased miR203 in IPF fibroblasts that had intrinsically low NUDT21 (Fig. S5). These results collectively suggest that TGF␤1 elevates miR203 via Smad3 phosphorylation.
Because Smad3 acts as a transcription factor following phosphorylation, it was hypothesized that activated Smad3 transcriptionally targets a miR203 promoter to up-regulate its expression. Based on the public chromatin immunoprecipitation (ChIP)-Seq data from the Cistrome Project and the Smad3 DNA-binding motif, two independent promoter regions of miR203 were chosen: 12 and 1.6 kb upstream from the miR203 transcription start site. Two sets of ChIP primers were designed to target these regions. In this experiment, Serpine1 was used as a positive control because this gene is a direct target of Smad3 (34). These two promoter regions were found to be bound by phosphorylated Smad3 following 3 h of TGF␤1 treatment in human lung fibroblasts, but SIS3 decreased the Smad3 binding, suggesting that phosphorylated Smad3 acts as a transcription factor for miR203 expression (Fig. 3E). To further investigate whether an endogenous elevation of miR203 following TGF␤1 treatment can regulate gene expression, GFP NUDT21 3Ј-UTR reporters were generated. GFP 3Ј-UTR WT contained the WT 3Ј-UTR of human NUDT21 following the coding sequence of GFP. However, the GFP 3Ј-UTR mutant had a human NUDT21 3Ј-UTR mutant where all three key binding sites of miR203 were mutated to random sequences, preventing the gene expression regulation by miR203 (Fig. 3F). These vectors were transfected in human lung fibroblasts followed by TGF␤1 treatment. The cells transfected with the GFP 3Ј-UTR mutant were found to be resistant to TGF␤1-mediated GFP reduction, although the GFP 3Ј-UTR WT cells showed a reduction in GFP expression as shown in Western blot analysis (Fig. 3

, G and H).
Fluorescence images consistently revealed that the GFP 3Ј-UTR WT displayed a reduction in GFP following TGF␤1 treatment, but the mutation on the miR203-binding sites prevented TGF␤1-induced GFP reduction (Fig. 3I). Furthermore, quantitative analysis of the images showed that TGF␤1 did not reduce GFP intensity or the number of GFP-positive cells in the GFP 3Ј-UTR mutant-transfected fibroblasts (Fig. 3J). These data collectively suggest that endogenous miR203 up-regulation following TGF␤1 treatment is sufficient to affect NUDT21 gene expression. To validate these data in a different experimental setting, a Dual-Luciferase assay was performed. Three Dual-Luciferase vectors were utilized: vectors with a stable SV40 3Ј-UTR for control, human NUDT21 WT 3Ј-UTR, and NUDT21 mutant 3Ј-UTR with mutations in miR203-binding sites. These vectors were transfected into human lung fibroblasts followed by TGF␤1 treatment, similar to the previous GFP 3Ј-UTR reporter experiment. Consistent with the previous data, TGF␤1 significantly reduced the firefly luciferase signal in the NUDT21 WT 3Ј-UTR group but not in the mutant 3Ј-UTR group, indicating that the reduction in gene expression is mediated through miR203 (Fig. S6). To examine whether inhibition of miR203 prevents NUDT21 reduction, antagomiR, a synthetic miRNA inhibitor, was used (Fig. S7). Human primary lung fibroblasts transfected with antagomiR203 showed an increased endogenous level of NUDT21 and were more resistant to TGF␤1-mediated NUDT21 reduction (Fig. 3, K and L). In addition, fibronectin was reduced by antagomiR203 transfection ( Fig. S7), suggesting that NUDT21 restoration through antagomiR203 prevents the activation of some profibrotic pathways by APA (25). Overall, these experiments suggest that TGF␤1 targets NUDT21 mRNA through miR203.

TGF␤1 decreases NUDT21 in mouse lung fibroblasts
Primary lung fibroblasts were isolated from C57Bl/6 mice to examine whether TGF␤1 down-regulates NUDT21 in vitro. Western blot analysis revealed that NUDT21 protein was reduced following 3 days of TGF␤1 treatment in mouse lung fibroblasts (Fig. 4A). Moreover, consistent with human data, NUDT21 mRNA was decreased, suggesting that the NUDT21 protein reduction may be due to its mRNA reduction (Fig. 4B). We previously reported that fibroblasts isolated from mice with bleomycin-induced fibrosis had reduced NUDT21. We examined whether TGF␤1 can further deplete NUDT21 in these fibroblasts and found that NUDT21 was depleted only in the fibroblasts that had relatively high NUDT21 levels ( Fig. S8) (25).
To investigate the role of Smad3 phosphorylation in NUDT21 down-regulation, mouse lung fibroblasts were pretreated with SIS3 for 1 h followed by TGF␤1 in the presence or absence of SIS3 for 3 days. Smad3 inhibition prevented TGF␤1-mediated NUDT21 reduction, indicating that TGF␤1 down-regulates NUDT21 through Smad3 phosphorylation in mouse lung fibroblasts (Fig. 4, C and D). To investigate the role of miR203 in TGF␤1 regulates NUDT21 and 3-UTR in lung fibrosis TGF␤1-mediated NUDT21 reduction, primary mouse lung fibroblasts isolated from C57Bl/6 were treated with TGF␤1 for 2 days and harvested to determine miR203 levels. TGF␤1 treatment induced miR203 expression, suggesting that miR203 is associated with NUDT21 reduction in mouse lung fibroblasts (Fig. 4E). To further demonstrate that NUDT21 is the target of miR203 in mouse, an miR203 mimic was transfected into mouse lung fibroblasts. NUDT21 was significantly down-regu-

TGF␤1 regulates NUDT21 and 3-UTR in lung fibrosis
lated by the miR203 mimic and associated with the increase in matrix proteins such as collagen 1 and fibronectin (Fig. 4, F and  G). Similar to the antagomiR203 experiment in human primary lung fibroblasts, mouse cells were transfected with antago-miR203 followed by TGF␤1 treatment. TGF␤1-mediated NUDT21 depletion was attenuated by antagomiR203 treatment, indicating that miR203 is important for NUDT21 reduction in mouse lung fibroblasts (Fig. 4, H and I). To verify whether TGF␤1 could deplete NUDT21 in vivo, conditional TGF␤1 transgenic mice were utilized. As reported previously Figure 3. TGF␤1 targets NUDT21 mRNA through miR203. Primary human lung fibroblasts were challenged with TGF␤1 for 24 h. miR203 was measured in RT-qPCR (A and B). SIS3, a Smad3 phosphorylation inhibitor, was used to inhibit Smad3 signaling (B). Primary lung fibroblasts were treated with TGF␤1 in the presence or absence of SIS3 for 4 days (C). Primary lung fibroblasts that were treated with TGF␤1 and SIS3 for 2 days were harvested for RT-qPCR analysis (D). ChIP was performed for phospho-Smad3 (P-Smad3) in the primary human lung fibroblasts following 2 h of TGF␤1 treatment. IgG and Serpine1 were used as an antibody and a TGF␤1 signaling control, respectively (E). A schematic view of 3Ј-UTR GFP reporter is shown (F). Western blotting data for GFP in the GFP 3Ј-UTR-transfected cells and densitometry analysis for three independent experiments are shown (G and H). Shown are fluorescence images for human lung fibroblasts treated with TGF␤1 for 4 days following transfection with GFP 3Ј-UTR reporters. F-actin displays stress fibers in the cells (I). Quantitative data for relative GFP intensities and the number of GFP-positive cells in 10 randomly taken pictures of each group are shown (J). Primary human lung fibroblasts were pretreated with antagomiR203, a miR203 inhibitor, followed by TGF␤1 treatment and densitometry analysis for four independent experiments (K and L). *, p Ͻ 0.05; ***, p Ͻ 0.001; ****, p Ͻ 0.0001; N.S., not significant. Veh, vehicle; Con, control; IP, immunoprecipitation. Error bars represent S.D.  (35), this mouse model has an rtTA-tTS-controlled CC10driven TGF␤1 system on a C57Bl/6 background where TGF␤1, whose expression is controlled by doxycycline water administration, is secreted from club cells into the alveolar space to induce lung fibrosis. Immunohistochemistry (IHC) for NUDT21 in the lung showed reduced nuclear NUDT21 in fibrotic areas from TGF␤1 transgenic mice compared with controls (Fig. 5, A and B). Next, fibroblasts from TGF␤1 transgenic mice were isolated and harvested for quantitative analysis. The fibroblasts isolated from TGF␤1 transgenic mice showed a

TGF␤1 regulates NUDT21 and 3-UTR in lung fibrosis
reduction in NUDT21 protein expression correlated with an increased expression of collagen and fibronectin (Fig. 5C). Densitometry analysis revealed that the reduction was statistically significant (Fig. 5D). Interestingly, NUDT21 mRNA was not reduced significantly in TGF␤1 fibroblasts, although collagen 1a1 and fibronectin mRNAs and miR203 were significantly increased (Fig. 5, E and F). These findings demonstrate that TGF␤1 decreases NUDT21 in mouse lung fibroblasts.

TGF␤1 differentially regulates APA in human lung fibroblasts, contributing to increased protein expression
NUDT21 is a key regulator of APA in human lung fibroblasts (25). To demonstrate that TGF␤1-mediated NUDT21 reduction also causes APA in primary human lung fibroblasts, cells were challenged with TGF␤1, and RNA-Seq was conducted. RNA-Seq data were subjected to PDUI analysis (Table 1) to investigate the APA pattern in the cells (36). The APA analysis revealed that TGF␤1 treatment differentially regulated APA in human lung fibroblasts, showing 311 shortened and 297 length-ened transcripts (Fig. 6A). KEGG pathway analysis revealed that the APA target genes in the data set were enriched for pathways associated with fibrosis such as phosphatidylinositol 3-kinase/Akt, chemokine signaling, ECM-receptor interaction, and focal adhesion (Fig. 6B). Among the APA target genes, FGF14, RICTOR, TMOD2, and UCP5 were chosen for further validation because these genes were the common APA targets reported in our previous study (Fig. 6C) (25), and these genes are important for fibrogenesis and the biology of fibroblasts (37)(38)(39)(40)(41)(42)(43). To further investigate APA usage in the transcripts, two sets of primers were designed for the RT-qPCR-based APA analysis: one set targeting the coding sequence and the other targeting the distal 3Ј-UTR of transcripts (25). The APA analysis demonstrated that FGF14, RICTOR, TMOD2, and UCP5 showed 3Ј-UTR shortening both in the NUDT21 knockdown fibroblasts and the fibroblasts challenged with TGF␤1 (Fig. 6, D and E). To further demonstrate that the 3Ј-UTR shortening is associated with increased protein expression, the protein levels of the four genes were measured by Western blot analysis in human lung fibroblasts treated with TGF␤1. Western blotting revealed that their protein expression was increased following TGF␤1 treatment (Fig. 6F). These data collectively suggest that TGF␤1-mediated NUDT21 reduction promotes differential APA, and the shortened 3Ј-UTR by APA is associated with increased protein in human lung fibroblasts. This differential APA pattern by TGF␤1 was a unique pattern that has not been reported previously. To further demonstrate the pathological relevance of this result, fibroblasts were isolated from explanted lungs from patients with IPF and utilized for APA analysis. As reported previously, IPF fibroblasts had significantly reduced NUDT21 (Fig. S9) (25). APA analysis by RNA-Seq on these fibroblasts revealed that IPF fibroblasts have differentially regulated APA patterns similar to the TGF␤1 data set (Fig. S10). These findings demonstrate that TGF␤1 differentially regulates APA in human lung fibroblasts, contributing to increased protein expression.

Discussion
IPF is a chronic lung disease with an unclear etiology. Among interstitial lung diseases, IPF is the most severe type, with the 5-year survival rate in the range of 20 -30% (37). This poor patient outcome is largely due to the absence of effective medical therapies for IPF. One of the key features of IPF is abnormal fibroblast activation and therefore uncontrolled deposition of ECM, which impairs gas exchange across the alveolocapillary membrane. Although fibroblasts are one of the major cells that overexpress pathological ECM and profibrotic mediators, the mechanisms governing the protein expression are still elusive. Our study investigated a novel mechanism for the fibrotic gene expression in lung fibroblasts.
APA has been recognized as a novel post-transcriptional regulation that affects 3Ј-UTR lengths of transcriptomes and, therefore, protein expression of APA target genes. NUDT21, a key protein in the polyadenylation machinery, has been associated with various cancer cell lines and tumor growth (23,24,(44)(45)(46), with previous studies suggesting that APA via NUDT21 reduction causes elevation of oncogenes contributing to cancer progression (24). However, the roles and implications of APA in pathophysiology have been elusive, especially in noncancer dis-

TGF␤1 regulates NUDT21 and 3-UTR in lung fibrosis
eases. Our previous work demonstrated that IPF is associated with APA and NUDT21 reduction, and NUDT21 reduction exacerbates the development of fibrosis (25). As a follow-up report, our current study identified TGF␤1 as a major inducer of APA, a posttranscriptional regulation, in lung fibroblasts via NUDT21 reduction. We demonstrate here that TGF␤1, a profibrotic cytokine, down-regulates NUDT21 in primary human and mouse lung fibroblasts at protein and mRNA levels. Furthermore, it was shown that TGF␤1 elevates miR203, which targets the 3Ј-UTR of NUDT21 via Smad3 phosphorylation in human lung fibroblasts. TGF␤1 transgenic mice revealed that TGF␤1 reduces NUDT21 in vivo. In addition, RNA-Seq data demonstrated that human lung fibroblasts treated with TGF␤1, as well as IPF patient fibroblasts, exhibited a distinctive APA pattern of global shortening and lengthening. Understanding the mechanism driving the APA in IPF is important because this novel pathway can be potentially targeted to halt pulmonary fibrosis. Several studies have demonstrated that initial injuries in alveolar epithelial cells are the trigger of pulmonary fibrosis in genetically and environmentally predisposed subjects, leading to pathological fibroblast activation (47,48). In addition to the concept of epithelial injury, it has been proposed that fibrotic environments triggered initially by epithelial injuries constitute a self-sustaining niche for aggravated fibrotic progression, which is independent of epithelial injuries (12,14). This suggests that pulmonary fibrosis may have two distinct phases of pathogenesis that should be therapeutically targeted as independent entities: initiation in epithelial cells and progression in fibroblasts (47). NUDT21 conditional knockout in fibroblasts does not induce spontaneous fibrosis in vivo as reported previously (25), suggesting that NUDT21 depletion in fibroblasts does not significantly cause the initial phase of epithelial injury or activate resident fibroblasts for pathogenesis. Interestingly, NUDT21 conditional knockout mice exhibited significantly worsened fibrosis upon bleomycin administration. This result demonstrates that NUDT21 reduction in fibroblasts may amplify the severity of fibrosis in the progression phase by overexpression of profibrotic mediators and matrix protein, therefore enhancing the establishment of a self-sustaining profibrotic niche in the microenvironment (14). TGF␤1, as a major profibrotic factor, may amplify fibrosis by down-regulating NUDT21 and inducing APA.
NUDT21 is the key regulator of APA in human lung fibroblasts, and the knockdown of NUDT21 by siRNA is sufficient to induce global APA enriched for profibrotic mediators and is associated with increased protein expression (25). We previ- Figure 6. TGF␤1 differentially regulates APA in human lung fibroblasts, contributing to the increased protein expression. Primary human lung fibroblasts were challenged with TGF␤1 for 4 days and then harvested for RNA-Seq. The PDUI algorithm was used to plot shortening and lengthening of the 3Ј-UTR for transcripts (A). The APA genes were used for KEGG pathway analysis (B). The top APA genes that are commonly found in the NUDT21 KD and TGF␤1 RNA-Seq were chosen to show PDUI values (C). RT-qPCR-based APA analysis was performed using two sets of primers to measure the PAS usage on mRNAs in the NUDT21 KD and lung fibroblasts that were challenged with TGF␤1 (D and E). Western blotting was performed to probe the overexpressed APA genes in the human lung fibroblasts (F). PI3K, phosphatidylinositol 3-kinase; MAPK, mitogen-activated protein kinase; Col1, collagen 1; FN, fibronectin; FGF, fibroblast growth factor. Error bars represent S.D.

TGF␤1 regulates NUDT21 and 3-UTR in lung fibrosis
ously observed a unique pattern of APA on the 3Ј-UTR, primarily shortening. However, in this study, the human lung fibroblasts challenged with TGF␤1 displayed not just shortening but also lengthening of the 3Ј-UTR through APA globally (Fig. 6A). This differential APA by TGF␤1 was presumably because TGF␤1, as a potent inducer of various signaling pathways, regulates multiple regulators of APA, including NUDT21. For example, TGF␤1 RNA-Seq data revealed that Fip1, which is another regulator of 3Ј-UTR lengthening by APA, is down-regulated (36) (Fig. S11). In addition to Fip1, TGF␤1 reduces other APA regulators that are associated with 3Ј-UTR shortening such as CFIm68 (49,50). This unique differential APA pattern is also observed in fibroblasts isolated from IPF patients (Fig.  S7), suggesting that APA in patients is controlled by multiple factors rather than a single factor.
Patient care for IPF has been challenging due to the lack of effective drug therapies for the disease. Drug development for IPF has been unsuccessful for decades presumably because the clinical drugs target only single molecular pathways associated with IPF and thus fail (1). Given that IPF is a heterogeneous disease in which multiple pathogenic pathways are activated, it has been suggested that drugs developed for IPF should target diverse signaling pathways (1). The only drugs that slow down the progression of IPF are pirfenidone and nintedanib, and they indeed target multiple signaling molecules (51,52). Nonetheless, a definitive cure remains to be found because patients on these medications still show a constant decline in lung function and do not have a significantly improved morbidity compared with placebo groups (53)(54)(55)(56). The APA regulation by NUDT21 is novel and clinically important because this pathway is a physiological mechanism that can impact multiple pathways. NUDT21 down-regulation, which is also observed in IPF patients, leads to global mRNA shortening, especially enriched in important profibrotic pathways such as Wnt, TGF␤1, and vascular endothelial growth factor signaling, suggesting that targeting APA through NUDT21 may be a potential therapeutic approach for pulmonary fibrosis. Previously, it was demonstrated that NUDT21 overexpression in IPF fibroblasts reduces the protein expression of profibrotic mediators and ECM (25). In this study, it was further shown that preventing NUDT21 reduction by antagomiR203, an inhibitor of miR203, decreases ECM proteins and inhibits fibroblast activation. These data collectively suggest that targeting APA by NUDT21 overexpression or antagomiR203 is a potential therapeutic approach for pulmonary fibrosis.
In conclusion, our study demonstrated that TGF␤1 downregulates NUDT21 in human and mouse lung fibroblasts. The detailed mechanism includes the transcriptional induction of miR203 by TGF␤1 through Smad3 activation targeting the 3Ј-UTR of NUDT21. In turn, lung fibroblasts challenged with TGF␤1 and fibroblasts isolated from IPF patients display differential APA patterns, and the shortening of APA genes was associated with increased protein expression. These findings unveil a novel mechanism (Fig. 7) for APA induced by NUDT21 depletion that is associated with deadly diseases such as IPF and cancer. The results of this research will help develop innovative therapies by targeting the pathways involved with post-tran-scriptional regulation of profibrotic mediators and pathological genes.

3-UTR reporter assays
Dual-Luciferase control plasmids (psiCheck2) and plasmids with human WT NUDT21 3Ј-UTR were given by Dr. Eric Wagner as a gift. From the WT NUDT21 Dual-Luciferase plasmid, NUDT21 3Ј-UTR mutant plasmids that have mutations on miR203-binding sites were generated through site-directed mutagenesis in Gibson assembly (New England Biolabs, Ipswich, MA). Mutation was confirmed by Sanger sequencing (Genewiz, Plainfield, NJ). Primary human lung fibroblasts were electroporated with the Dual-Luciferase vectors at 120 V and 500 microfarads in a 2-mm cuvette in a Bio-Rad Gene Pulser system. The transfected cells were cultured in antibiotic-free media for 2 days and then challenged with TGF␤1. Dual-Luciferase assay was performed following the manufacturer's protocol (GenDEPOT). For GFP 3Ј-UTR reporter, pCMV3-GFP-Spark vector was purchased from Sino Biological (Beijing, China). The 3Ј-UTR of human NUDT21 was introduced into the vector through Gibson assembly. Similar to the mutation in the Dual-Luciferase vector, for NUDT21 3Ј-UTR mutant, miR203-binding sites were mutated through site-directed mutagenesis in Gibson assembly. This vector was incorporated into primary human lung fibroblasts through electroporation as described above. The GFP-transfected cells were cultured on glass coverslips. The cells transfected with GFP with WT NUDT21 3Ј-UTR or mutant 3Ј-UTR were challenged with TGF␤1. On day 4, the cells were fixed with 3.7% formaldehyde in PBS for 10 min at room temperature. 70% acetone in PBS was

TGF␤1 regulates NUDT21 and 3-UTR in lung fibrosis
added and incubated at Ϫ30°C for 10 min. The cells were incubated with 1% BSA in PBS for 30 min to decrease background signals and then incubated with Alexa Fluor TM phalloidin to stain F-actin (stress fibers) for 30 min at room temperature (Thermo Fisher Scientific). The coverslips were mounted with anti-fading DAPI (Thermo Fisher Scientific). The ␣SMA fluorescence staining was performed as described previously (57).

ChIP assay
miRStart (http://mirstart.mbc.nctu.edu.tw/) 3 was used to predict promoters of miR203 in the human genome. The predicted promotors were further examined for Smad3-binding sites in Promo 3.0 (http://alggen.lsi.upc.edu/). 3 Moreover, previously published Smad3 ChIP data were used to confirm Smad3-binding sites in the miR203 promoters (Cistrome) (60). The Smad3-binding sites that were both predicted in Promo 3.0 and reported in Cistrome were targeted for ChIP assay. Two sets of primers were designed: one set targeting 12 kb upstream and the other set targeting 1.6 kb upstream of pri-miR203. ChIP assay was performed using an EZChIP kit from Millipore Sigma (Burlington, MA). Human primary cells were pretreated with SIS3 or vehicle control for 1 h and then challenged with TGF␤1 in the presence or absence of SIS3 for 2 h. The cells were crosslinked by 1% formaldehyde for 10 min at room temperature and quenched by glycine. The cells were then sonicated for 10 s four times at 30% power. Then the samples were incubated with IgG control or phospho-Smad3 antibodies (Ser-423/425; Cell Signaling Technology, Danvers, MA) at 4°C overnight while being rotated. Agarose beads were added to pull down antibodies. The cross-link was reversed, and real-time PCR was performed to detect the Smad3-binding sites.

Human samples
Primary human lung fibroblasts were isolated directly from human lungs (explant IPF lungs or transplant-grade healthy lungs) in the laboratory following institutional protocols. IPF lungs were deidentified and obtained from the Methodist Hospital Cardiothoracic Transplant Center or the Lung Transplant Center at Memorial Hermann-Texas Medical Center. Protocols for the use of human samples in this project were approved by the University of Texas Health Science Center at Houston (UTHealth, Houston, TX), the Methodist Hospital, and the Memorial Hermann Hospital institutional review boards. IPF 7. Working model for NUDT21 reduction by TGF␤1-miR203 axis. TGF␤1 is elevated upon lung injury, leading to the phosphorylation of Smad2 and Smad3. Smad2/3 acts as a transcription factor to target miR203. The up-regulated miR203 targets the 3Ј-UTR of NUDT21 to reduce the mRNA and protein, and the reduction of NUDT21 causes shortening of profibrotic mRNAs and ultimately promotes enhanced protein expression for the genes. TGF␤IR and TGF␤IIR, TGF␤ receptor types I and II, respectively.

TGF␤1 regulates NUDT21 and 3-UTR in lung fibrosis
lungs were collected during lung transplant surgery and were processed on site within 20 min. The transplant-grade human lung tissues were obtained from the International Institute for the Advancement of Medicine.

Mouse generation and treatment
C57BL/6 mice were purchased from Envigo and housed following the guidelines of the Animal Welfare Committee at UTHealth. All experiment designs were reviewed and approved by the institutional review committee.
CC10-tTS-rtTA-TGF-␤1 transgenic mice with C57BL/6 background were bred and housed in the animal facility at Brown University. As described previously, 6 -8-week-old mice and control littermates were randomized to either regular water or doxycycline water (0.5 mg/ml) (58). The transgenic and control mice were sacrificed and harvested for fibroblast isolation and histology as described below.

Primary fibroblast isolation
To isolate primary fibroblasts from mouse lungs, fresh lungs were perfused, harvested under sterile conditions, and excised into ϳ1-mm 3 fragments using a McIlwain Tissue Chopper. To attach the cells on a culture dish, the cells were cultured without media for 2 h in an incubator. Then the cells were further cultured in high-glucose Dulbecco's modified Eagle's medium supplemented with 10% FBS, 1 mM sodium pyruvate, and a mixture of antibiotics/antimycotic for 2 weeks to allow fibroblasts outgrowth. Cultured fibroblasts were used for experiments between passages 1 and 3. Primary human lung fibroblasts were isolated and cultured following the same method except that human lungs were not perfused, and the cells were cultured in Eagle's minimum essential medium. The human cells were used for experiments until passage 6.

Western blotting
Western blot analysis was performed as described previously (59). The following antibodies were used for experiments: rabbit anti-NUDT21 from Proteintech; mouse anti-␤-actin, rabbit anti-fibronectin, and mouse anti-␣ smooth muscle cell actin from Sigma-Aldrich; mouse anti-GAPDH from Life Technologies; mouse anti-GFP from GenDEPOT; rabbit anti-collagen 1 from Abcam; and mouse anti-vinculin and rabbit anti-phospho-Smad3 (Ser-423/425) from Cell Signaling Technology. UCP5 and RICTOR antibodies were from Novus. FGF14 and TMOD2 were from Abcam and Proteintech, respectively. Membranes were incubated with the listed primary antibodies overnight at 4°C and then with the corresponding secondary antibodies conjugated to horseradish peroxidase (Cell Signaling Technology). Membranes were developed using Pico Dura ECL solutions (GenDEPOT).

RNA sequencing (RNA-Seq)
CCD8Lu cells were treated with TGF␤1 (GenDEPOT) in the presence or absence of Smad3 inhibitor SIS3 (Tocris). The cells were lysed using TRIzol (Thermo Fisher Scientific), and RNA was processed using an RNeasy Mini kit (Qiagen, Valencia, CA). RNA-Seq was performed by HiSeq 2000 (Novogene, China). This allowed us to capture both gene expression changes and the diversity of transcripts in both coding and noncoding RNA species. Paired-end RNA-Seq was carried out to yield a minimal 80 million reads per sample. RNA-Seq gene expressions were quantified by RSEM in combination with a de novo transcriptome assembler. Briefly, the significance of the difference of mean PDUIs between the expression levels of short and long 3Ј-UTRs in two samples in each condition were first computed using Fisher's exact test and further adjusted using the Benjamini-Hochberg procedure to control the false discovery rate at a 5% level. Genes with an absolute difference of mean PDUIs no less than 0.15 and an absolute log 2 ratio (-fold change) of mean PDUIs no less than 1 were identified to have significant shifted 3Ј-UTR events.

Real-time quantitative PCR
Total RNA was extracted from primary cells using TRIzol. As described previously (46), the TRIzol was mixed with chloroform for phase separation. The upper layer containing total RNA was purified through Qiagen RNA columns. The purified RNA was treated with DNase and used for cDNA synthesis using iScript TM Reverse Transcription Supermix (Bio-Rad). Real-time PCR was performed using a LightCycler 96 (Roche Applied Science). The primers used for experiments are shown in Table S4. Predesigned primers for RNU6 and miRNAs were purchased from Qiagen. Data were quantified and presented as mean ratio to 18S RNA.
As described previously (24), two sets of primers were designed to detect polyadenylation site (PAS) usage: one set targeting an exon to measure the total transcript level and the other set targeting a 3Ј-UTR region just upstream of the distal polyadenylation site to detect transcripts with long 3Ј-UTRs. Percentage of PAS usage was calculated as ⌬CT ϭ CT distal Ϫ CT total . Data were presented as log -fold changes normalized to controls by calculating ⌬⌬CT ϭ ⌬CT average target Ϫ ⌬CT average of control .

IHC and immunofluorescence
Harvested mouse and human tissues were fixed in PBS-based 10% formaldehyde and embedded in paraffin. Paraffin tissue blocks were sectioned with 4-m thickness. Sectioned slides were deparaffinized and hydrated by gradient. Background signals in sections were quenched with BLOXALL (Vector Laboratories, Burlingame, CA) and incubated in citric acid buffer (Vector Laboratories) for antigen retrieval. Sections were blocked with Avidin/Biotin Blocking System (Vector Laboratories) and then 5% normal goat serum.
For double IHC and immunofluorescence staining for NUDT21 and ␣SMA, sections were incubated with antibodies for NUDT21 (1:200) for 24 h and then with anti-rabbit antibodies conjugated with peroxidase (1:1000; Vector Laboratories) for 1 h at room temperature. Then the slides were developed with ImmPACT 3,3Ј-diaminobenzidine peroxidase substrate (Vector Laboratories). After development, the slides were further incubated with anti-␣SMA antibodies conjugated with Cy3 dye (1:1000; Sigma-Aldrich) overnight at 4°C.

Statistical analysis and densitometry
For all two-group analyses, Student's t test was performed with unpaired and two-tailed settings. For comparisons among multiple samples greater than two, one-way analysis of variance was used with Tukey's multiple comparison test. p values smaller than 0.05 were considered statistically significant. All statistical analyses were performed with GraphPad Prism (La Jolla, CA). For densitometry, ImageJ was used to measure the intensity and area of bands in Western blotting, and their values were normalized to loading controls such as GAPDH and ␤-actin.