LEM domain–containing protein 3 antagonizes TGFβ–SMAD2/3 signaling in a stiffness-dependent manner in both the nucleus and cytosol

Transforming growth factor-β (TGFβ) signaling through SMAD2/3 is an important driver of pathological fibrosis in multiple organ systems. TGFβ signaling and extracellular matrix (ECM) stiffness form an unvirtuous pathological circuit in which matrix stiffness drives activation of latent TGFβ, and TGFβ signaling then drives cellular stress and ECM synthesis. Moreover, ECM stiffness also appears to sensitize cells to exogenously activated TGFβ through unknown mechanisms. Here, using human fibroblasts, we explored the effect of ECM stiffness on a putative inner nuclear membrane protein, LEM domain–containing protein 3 (LEMD3), which is physically connected to the cell's actin cytoskeleton and inhibits TGFβ signaling. We showed that LEMD3–SMAD2/3 interactions are inversely correlated with ECM stiffness and TGFβ-driven luciferase activity and that LEMD3 expression is correlated with the mechanical response of the TGFβ-driven luciferase reporter. We found that actin polymerization but not cellular stress or LEMD3–nuclear-cytoplasmic couplings were necessary for LEMD3–SMAD2/3 interactions. Intriguingly, LEMD3 and SMAD2/3 frequently interacted in the cytosol, and we discovered LEMD3 was proteolytically cleaved into protein fragments. We confirmed that a consensus C-terminal LEMD3 fragment binds SMAD2/3 in a stiffness-dependent manner throughout the cell and is sufficient for antagonizing SMAD2/3 signaling. Using human lung biopsies, we observed that these nuclear and cytosolic interactions are also present in tissue and found that fibrotic tissues exhibit locally diminished and cytoplasmically shifted LEMD3–SMAD2/3 interactions, as noted in vitro. Our work reveals novel LEMD3 biology and stiffness-dependent regulation of TGFβ by LEMD3, providing a novel target to antagonize pathological TGFβ signaling.

Transforming growth factor-␤ (TGF␤) signaling through SMAD2/3 is an important driver of pathological fibrosis in multiple organ systems. TGF␤ signaling and extracellular matrix (ECM) stiffness form an unvirtuous pathological circuit in which matrix stiffness drives activation of latent TGF␤, and TGF␤ signaling then drives cellular stress and ECM synthesis. Moreover, ECM stiffness also appears to sensitize cells to exogenously activated TGF␤ through unknown mechanisms. Here, using human fibroblasts, we explored the effect of ECM stiffness on a putative inner nuclear membrane protein, LEM domaincontaining protein 3 (LEMD3), which is physically connected to the cell's actin cytoskeleton and inhibits TGF␤ signaling. We showed that LEMD3-SMAD2/3 interactions are inversely correlated with ECM stiffness and TGF␤-driven luciferase activity and that LEMD3 expression is correlated with the mechanical response of the TGF␤-driven luciferase reporter. We found that actin polymerization but not cellular stress or LEMD3-nuclearcytoplasmic couplings were necessary for LEMD3-SMAD2/3 interactions. Intriguingly, LEMD3 and SMAD2/3 frequently interacted in the cytosol, and we discovered LEMD3 was proteolytically cleaved into protein fragments. We confirmed that a consensus C-terminal LEMD3 fragment binds SMAD2/3 in a stiffness-dependent manner throughout the cell and is sufficient for antagonizing SMAD2/3 signaling. Using human lung biopsies, we observed that these nuclear and cytosolic interactions are also present in tissue and found that fibrotic tissues exhibit locally diminished and cytoplasmically shifted LEMD3-SMAD2/3 interactions, as noted in vitro. Our work reveals novel LEMD3 biology and stiffness-dependent regulation of TGF␤ by LEMD3, providing a novel target to antagonize pathological TGF␤ signaling.
Transforming growth factor ␤ (TGF␤) 2 plays important roles in human development, wound repair, and pathology. In particular, overactivation of TGF␤ signaling has been implicated as a pathological driver of fibrosis in several organ systems, including the lung, kidneys, skin, and liver (1)(2)(3). Recently, the Food and Drug Administration approved the use of pirfenidone in pulmonary fibrosis, which acts in part to antagonize TGF␤ signaling. Pirfenidone has led to a slowing of disease progression, indicating that this is a promising yet not fully realized therapeutic axis for addressing fibrotic pathologies (4 -7).
Previous work has shown a clear connection between extracellular matrix (ECM) stiffness and TGF␤ activation. TGF␤ is deposited in the ECM in an inactive state in latent TGF␤associated protein complexes. To activate TGF␤, cells can mechanically disengage TGF␤ from the latent complex through ␣v integrin-mediated mechanical stress, after which TGF␤ acts in a paracrine fashion. Stiffer ECMs allow for more force transmission to the latent complex and potentiate the liberation of TGF␤. This allows for pathological feedback whereby TGF␤ signaling drives matrix synthesis and cellular stress, and a stiffer matrix increases the free amount of TGF␤ available to cells (8,9). In addition to activating TGF␤, several in vitro studies have indicated that a stiff ECM potentiates cellular responses to exogenously activated (pre-activated) TGF␤ (10 -13). Here, our work focused on understanding this cellular potentiation to TGF␤ to identify additional targets of possible therapeutic intervention along the TGF␤-signaling pathway (i.e. targets that specifically address matrix-driven sensitization to TGF␤ and not matrix-driven activation of TGF␤).
We focused this study of rigidity-driven cellular sensitization to TGF␤ on LEM domain-containing protein 3 (LEMD3). LEMD3 is an integral inner nuclear membrane protein and a compelling target for two reasons: 1) it is a known inhibitor of TGF␤ signaling; 2) it is biophysically connected to the cell's cytoskeleton, which is remodeled in an ECM stiffness-dependent fashion. LEMD3 binds and inactivates the downstream transcription factor partners of both TGF␤ and bone morphogenic protein (BMP), the receptor proteins' mothers against decapentaplegic (Smads) 2 and 3 and 1, 5, and 8, respectively (14 -21). LEMD3's RNA recognition motif (RRM) domain, on its C-terminal end, competes with other transcription factors for binding to Smad2/3 and promotes Smad2/3 dephosphorylation and nuclear export by acting as a coordinating scaffold for protein phosphatase, Mg 2ϩ /Mn 2ϩ -dependent 1␣ (PPM1␣) (16,22).
In humans, heterozygous loss of LEMD3 is frequently, but not always (23), associated with the development of Buschke-Ollendorf syndrome (BOS), in which patients develop some constellation of cutaneous collagenomas and elastomas (from aberrant TGF␤ signaling) and osteopoikilosis (from aberrant BMP signaling), hyperostic lesions in long bones, which can mimic osteoblastic metastatic lesions radiographically (20, 24 -32). Depending on the penetrance and specific genotype, heterozygous loss of LEMD3 can also be associated with more severe ECM pathologies, such as melorheostosis, a progressive sclerosis frequently in long bones that can lead to disfigurement and joint destruction (20,24).
In addition to its clear role in antagonizing TGF␤ signaling both in vitro and in vivo, LEMD3 is retained in the inner nuclear membrane through its connection to the nuclear lamin superstructure by the activity of its N-terminal end, which contains the molecule's Lapb2/Emerin/MAN1 (LEM) domain (14,(33)(34)(35)(36)(37). This connection creates a direct biophysical link to the cell's actin microfilament network through the nesprin-sun linker of nucleus and cytoskeleton (LINC) complex (38 -40).
Given LEMD3's role in antagonizing Smad2/3 and its biophysical connection to the cell's cytoskeleton, we hypothesized that LEMD3's inhibition of Smad2/3 would be reduced by cytoskeletal stress thereby potentiating TGF␤ signaling in an ECM stiffness-dependent fashion. Here, dermal fibroblasts displayed ECM stiffness-dependent TGF␤ responsiveness, and LEMD3 knockdown or overexpression modulated this stiffness responsiveness. We also showed that LEMD3's interactions with Smad2/3 were negatively regulated by ECM stiffness and were independent of TGF␤ dosing. LEMD3-Smad2/3 interactions were increased by disruption of the actin cytoskeleton, but these complexes were not potentiated by disruption of the LINC complex or overexpression of the LEM domain. Unexpectedly, we discovered cytosolic interactions between LEMD3 and Smad2/3, which called into question basic assumptions about the current understanding of LEMD3's biology and its regulation of TGF␤. We identified and genetically localized Nand C-terminal LEMD3 fragments, which separated LEMD3's Smad-binding RRM and lamin-binding LEM domains. We established that these fragments were created through posttranslational proteolysis by a serine protease and were differen-tially regulated by lamin integrity. We verified that a consensus C-terminal fragment of LEMD3 bound Smad2/3 in a stiffnessdependent fashion, antagonized Smad2/3's ability to complex with Smad4, and showed that both endogenous LEMD3 and this LEMD3 fragment bound in the cytoplasm the PPM1␣ protein phosphatase previously shown to be coordinated by LEMD3 to antagonize TGF␤ signaling. Finally, we found correlates between our in vitro findings and human lung tissue from idiopathic pulmonary fibrosis (IPF) and non-IPF patients. Interactions between LEMD3 and Smad2/3 demonstrated greater intra-patient/total variability in frequency and were more frequent in the cytoplasm in human lung core biopsies from IPF patients relative to non-IPF tissues. Additionally, IPF tissues had unique low-frequency LEMD3-Smad2/3 interaction regions, possibly indicative of local fibrotic disease. Our work demonstrates novel aspects of LEMD3 biology as well as identifies a potential target for TGF␤ antagonism relevant to fibrosis that complements current medical management.

Fibroblast TGF␤ responsiveness is potentiated by ECM stiffness and inhibited by LEMD3 expression
We sought to validate that fibroblasts in our system displayed stiffness-dependent TGF␤ responsiveness using primary human dermal fibroblasts (HFFs) stably transfected with a Smad-driven luciferase. We characterized stiffness and morphology phenotypes associated with mechanotransduction using optical microscopy and atomic force microscopy (AFM). With increasing stiffness, fibroblasts underwent cellular compliance-matching (p ϭ 0.0022) and increased their cell-spreading surface area (p ϭ 0.1801) and polarization (p ϭ 0.0263) seen in Fig. 1A and Fig. S1, respectively. Luciferase-transfected HFFs showed a sigmoidal responsiveness to recombinant human TGF␤ as a function of substrate stiffness (Fig. 1B). This stiffness potentiation was dose-dependent with increasing picogram/ml doses of TGF␤ appearing to shift the transition point of the sigmoid toward softer substrates ("soft-shift", p ϭ 0.07, Fig. 1B). Transition points in the picogram/ml dose-range occurred at stiffness over a range from Ϸ1-4 kPa. These cells also showed the expected TGF␤ dose-response (increasing luminescence with increasing TGF␤ dose) on stiff substrates (p Ͻ 0.0001, Fig.  1B). Previous studies have also shown stiffness-dependent modulation of TGF␤ signaling, and our work improves the resolution of these results by more finely probing the stiffness space across a range from 0.5 to 50 kPa (10 -13).
To test which components of the cell's microfilament cytoskeleton control the stiffness-response of HFFs to TGF␤, we used chemical inhibition of the structural (cytochalasin D, a globular actin (g-actin) stabilizer) and contractility machinery (blebbistatin, a myosin II inhibitor). Stiffness-dependent TGF␤ responses required actin polymerization but did not require cellular contractility (Fig. 1D). Treatment with 10 M blebbistatin (a myosin II inhibitor) did not significantly shift the transition point nor increase the maximal luminescence of cells relative to HFFs treated with TGF␤ alone. Cells treated with 2 M cytochalasin D showed a low, flat luminescent profile over increasing ECM stiffness, and the data did not converge to a sigmoidal model. Interestingly, siRNA against LEMD3 did not rescue the cytochalasin D phenotype, although the degree of LEMD3 knockdown may not have been sufficient (Fig. S2B). These data indicate that the actin cytoskeleton, but not cellular stress per se, is needed for stiffness-modulated TGF␤ signaling. A, fibroblasts demonstrated mechano-sensitivity to ECM stiffness through cytoskeletal compliance matching as measured by AFM on increasingly stiff matrices (p ϭ 0.001). B, fibroblasts stably transfected with Smadresponsive luciferase demonstrated a dose-dependent stiffness modulation of their TGF␤ responsiveness. Increasing doses of TGF␤ were associated with a softer sigmoidal inflection point (p ϭ 0.07), and cells demonstrated a dose-response to TGF␤ on stiff surfaces (p Ͻ 0.0001). C, LEMD3 expression was correlated with a stiffer transition point in TGF␤ stiffness responsiveness (p Ͻ 0.0001) and with decreased luminescence (p ϭ 0.04). D, cytoskeletal depolymerization but not myosin II inhibition was associated with a decreased and flattened luminescent response of fibroblasts to TGF␤. LEMD3 KD does not rescue actin-depolymerization phenotypes. Treatments with cytochalasin D did not converge to a sigmoidal model, and the data are represented as mean with standard error of the mean to convey the heterogeneity of the results. All cell stiffness/morphology phenotypes and stiffness model parameters were statistically analyzed using an ANOVA test for trends analysis. All data are represented by the mean with S.E. unless otherwise noted. Bleb, blebbistatin.

LEMD3-SMAD2/3 complexes are inversely correlated to substrate stiffness and occur throughout the cell
Given that LEMD3 modulated the stiffness response of fibroblasts to TGF␤, we used proximity ligation assays (PLA, methodologically reviewed in Ref. 42) to examine the stiffness dependence of LEMD3-Smad2/3 interactions on hydrogels whose stiffness is representative of physiological (1 kPa) and fibrotic (25 kPa) lung tissue in humans (43). LEMD3-Smad2/3 interac-tions were negatively correlated with substrate stiffness in both the cytoplasm and the nucleus (Fig. 2, A and B). This negative correlation was independent of TGF␤ dosing (50 pg/ml, inside the range of stiffness-responsive doses identified in Fig. 1B), consistent with previous findings that LEMD3-Smad2/3 interactions are phosphorylation-independent (21). Although LEMD3-Smad2/3 interactions were unperturbed by TGF␤ dosing, we explicitly tested the ability of LEMD3 to bind phos-LEMD3, a stiffness-dependent inhibitor of TGF␤ signaling pho-Smad2/3 (Fig. S4). We found that HFFs treated with 100 pg/ml TGF␤ on glass had significantly higher LEMD3-phospho-Smad2/3 interactions than untreated fibroblasts (p ϭ 0.0333, Fig. S4) and that there was a significant negative correlation between substrate stiffness and LEMD3-phospho-Smad2/3 interactions (p ϭ 0.0305, Fig. S4). These data reveal the stiffness dependence of LEMD3-Smad2/3 interactions. The data also confirm previous observations that LEMD3 can bind both phosphorylated and unphosphorylated Smad2/3 (21).
We extended our findings by performing PLA across a finer range of stiffness with both HFFs and CCL210s, an adult pulmonary-derived fibroblast line (Fig. 2C). Both CCL210s and HFFs showed a biphasic response to stiffness with peak values at 1 kPa. CCL210s demonstrated a greater dynamic range in response to stiffness as well as a more gradual loss of PLA interactions on progressively stiffer substrates relative to HFFs. Additionally, the abrupt loss of LEMD3-Smad2/3 interactions between 1 and 4 kPa in HFFs was inversely correlated with the stiffness dependence of the sigmoidal transition point of the luciferase signal measured in Fig. 1B.
Interestingly, we found that approximately half of the PLA interactions on both substrates occurred in the cytoplasm of the cells. Across a broader range of stiffness, we found that the nuclear proportion of LEMD3-Smad2/3 interactions was inversely correlated to substrate stiffness in both HFFs (p Ͻ 0.0001) and CCL210s (p ϭ 0.0199), as seen in Fig. 2D. Given that LEMD3 is thought to be an integral protein of the inner nuclear membrane, we validated our findings through V5-Smad2/3 PLA in cells transfected with pFLAG-LEMD3-V5 (Fig. 2, E and F). Near identical trends were observed with this independent PLA reaction: a reduction in PLA frequency overall (p Ͻ 0.0001, Fig. 2F), and reductions in both the cytoplasm (p Ͻ 0.0001, Fig. 2F) and nuclear (p ϭ 0.0018, Fig. 2F) compartments with increasing substrate stiffness. To control for the degree of recombinant LEMD3 expressed across stiffness conditions, we found a significant difference in the linear regression between PLA puncta per cell and the per cell FLAG intensity in cells on both soft and stiff surfaces (p Ͻ 0.0001, Fig. 2G). The steeper slope observed for cells on soft substrates confirmed a higher rate of LEMD3-Smad2/3 complex formation per arbitrary unit of LEMD3 expressed (as measured by the fused FLAG epitope) relative to stiff substrates.

LEMD3-Smad2/3 complexes are inhibited by actin polymerization and not potentiated by disrupting cytoplasmic-nuclear-lamina-LEMD3 coupling
We hypothesized that the frequency of LEMD3-Smad2/3 complexes would be decreased by transmission of ECM-driven cytoskeletal tension to LEMD3 through a two-part physical linkage as follows: 1) the nesprin-sun LINC complex joining the nuclear lamina and actin cytoskeleton; and 2) through nuclear lamina-LEMD3 coupling via the LEM domain. We tested this hypothesis by disrupting the molecular linkages at either the level of the LINC complex, by expressing a previously validated dominant-negative nesprin construct, DN-Kash-mCherry (40), or by disrupting LEMD3-lamin interactions, by expressing a novel DN-LEM-mCherry constructs and assessing LEMD3-Smad2/3 complex formation by PLA. Neither DN Kash nor DN LEM expression increased the LEMD3-Smad2/3 PLA frequency relative to mCherry-only expressing fibroblasts on glass (Fig. 3, A and B). To assess whether our findings were biased by the degree of DN Kash or DN LEM expression, we correlated PLA puncta against the mCherry expression of the fusion protein per cell. None of the Pearson's coefficients (r ϭ Ϫ0.251 and r ϭ Ϫ0.087 for DN Kash and DN LEM, respectively, Fig. 3C) varied significantly from 0 (p ϭ 0.0579 and p ϭ 0.43 for DN-Kash and DN-LEM, respectively, Fig. 3C), indicating that the degree of dominant-negative protein expression was not a likely explanation for our findings. These results also corroborated findings in Fig. 1D, where cellular contractility inhibition, through blebbistatin antagonism of myosin II, did not significantly modify the TGF␤ stiffness response of fibroblasts.
Because g-actin-stabilizing agents did modify the fibroblast stiffness response (Fig. 1D), we examined the effect of actin polymerization on LEMD3-Smad2/3 complex formation by PLA. Actin polymerization was significantly negatively correlated to the frequency of LEMD3-Smad2/3 complexes ( ). B, quantification of PLA interactions grouped by substrate stiffness and by TGF␤ dose. Total LEMD3-Smad2/3 interactions were negatively correlated to substrate stiffness (p Ͻ 0.0001 for 1 kPa versus 25 kPa for both 0 and 50 pg/ml TGF␤) but not correlated to TGF␤ dose. Cytoplasmic (p ϭ 0.0164 for 0 pg/ml TGF␤ and p ϭ 0.0087 for 50 pg/ml TGF␤) and nuclear compartment (p ϭ 0.0428 for 0 pg/ml TGF␤ and p ϭ 0.1228 for 50 pg/ml TGF␤) interactions were also negatively correlated to substrate stiffness. C, total HFFs and CCL210s LEMD3-Smad2/3 interactions by PLA normalized to 0.5 kPa on surfaces with stiffness of 0.5, 1, 4, 8, 25 kPa and glass. Each fibroblast population showed a biphasic trend, centered around a peak of LEMD3-Smad2/3 interactions at 1 kPa. CCL210 demonstrated greater dynamic range in interaction frequency and a slower loss of interactions on stiffer substrates than HFFs. D, subcellular location of LEMD3-Smad2/3 PLA interactions in HFFs and CCL210s from C. Each cell line demonstrated a cytoplasmic shift in location with increasing substrate stiffness (p Ͻ 0.0001 and p ϭ 0.0199 for HFFs and CCL210s, respectively; ANOVA test for trend). E, micrographs of V5-Smad2/3 PLA interactions with pFLAG-LEMD3-V5 on soft (top row) and stiff (bottom row) matrices (PLA in green, f-actin in blue, FLAG in red, and nucleus in white). Area above the dashed line in 25-kPa image indicates filler space. F, V5-Smad2/3 PLA interactions were also negatively correlated with substrate stiffness (for 1 kPa versus 25 kPa: total PLA, p Ͻ 0.0001; cytoplasmic PLA, p Ͻ 0.0001; nuclear PLA, p ϭ 0.0018) and also occurred in the cytoplasm. G, V5-Smad2/3 PLA interactions were significantly higher on soft substrates independent of the degree of pFLAG-LEMD3-V5 expression (difference in linear regression slopes, p Ͻ 0.0001). All PLA groups were statistically compared using a two-way ANOVA with Tukey post-test unless noted. All scale bars, 10 m. All data represented by the mean with S.E. except for G, where individual data points are plotted. *, p Յ 0.05; **, p Յ 0.01; ****, p Յ 0.0001.  Fig. 1A and Fig. S1, showing increased cell spreading, polarization, and stiffness as a function of substrate stiffness, these data suggested that actin polymerization in response to substrate stiffness helps coordinate the LEMD3-dependent stiffness response of cells to TGF␤.

LEMD3 fragments are generated by a serine protease and differentially regulated by the integrity of the nuclear lamina
We observed cytosolic LEMD3-Smad2/3 interactions with PLA antibody pairs against the native protein and against a full-length recombinant LEMD3 protein, expressing a V5 epitope tag. These data were supported by fractionated cell Western blottings of endogenous LEMD3, which revealed an Ϸ50-kDa LEMD3 fragment in the cytosol (Fig. S5B). These cytosolic LEMD3-Smad2/3 interactions seemed to be stiffness-regulated given their negative correlation with actin polymerization (Fig. 3

, D and E).
To understand how LEMD3 could interact with Smad2/3 in the cytosol, we performed MS on lysates from cells transfected with pFLAG-LEMD3-V5 constructs (Fig. 4A). Results from a FLAG-purified 60-kDa fragment showed two transitions in the peptide spectral matches to LEMD3's primary sequence, first at p.294 -325 and second at p.579 -647 (these protein coordinates are offset by 20 amino acids inserted with the N-terminal FLAG tag in the pFLAG-LEMD3-V5 construct from the native protein sequence). The primary sequence relationship between these two regions and other known domains of

LEMD3, a stiffness-dependent inhibitor of TGF␤ signaling
LEMD3 are shown in Fig. 4B. Deletion mutants (pFLAG-LEMD3p.⌬294 -325-V5 and pFLAG-LEMD3p.⌬579 -647-V5) exploring these two regions showed differential fragment presentation in Western blottings that probed either the N-terminal FLAG tag or the C-terminal V5 tag of the recombinant protein (Fig. 4C). Specifically, pFLAG-LEMD3p.⌬294-325-V5 lysates lacked the 60-kDa FLAG fragment (the FLAG fragment at Ͻ50 kDa was also missing but was not consistently found in all blots) and the 46-kDa V5 fragment. Interestingly, the 46-kDa V5 fragment was similar in size to the cytoplasmi-cally localized fragment of LEMD3 noted earlier from endogenous lysates (Fig. S5B). The pFLAG-LEMD3p. ⌬579 -647-V5 lysates did not appear to change the FLAG fragment presentation but did eliminate V5 fragments at 60 and 85 kDa. These genetic data suggested that the protein is modified in at least two regions, one in the nucleoplasm (⌬294 -325) and one in the peri-nuclear membrane space (⌬579 -647). Moreover, all the V5/C-terminal fragments identified by our genetic mutations contained the CTF used in Fig. S3, indicating that they are sufficient for binding , and RRM domains, and FLAG and V5 epitope tags. C, Western blots from full-length and each deletion mutant using N-terminal FLAG tag (top two blots) and C-terminal V5 tag (bottom two blots) at 12 and 24 h after electroporation. FLAG blots consistently produced a 60-kDa fragment, whereas V5 blots produced 85-, 60-, and 46-kDa fragments. D, representative Western blots for protease and cell cycle inhibitor experiments using N-terminal FLAG tag (top two blots) and C-terminal V5 tag (bottom two blots). E, quantification of blots from D showing that 60-kDa FLAG fragment was significantly reduced relative to the full-length protein when cells were treated with DCI (p ϭ 0.0153), MG-132 (MG) (p ϭ 0.0238), cathepsin-G inhibitor (Cat-Gi) (p ϭ 0.0075), or roscovitine (Ros) (p ϭ 0.0051). V5-tagged 46-kDa fragment was similar in that DCI (p Ͻ 0.0001) and MG-132 (p ϭ 0.0238) treatments decreased its abundance but was dissimilar in that roscovitine increased its abundance (p Ͻ 0.0001). V5-tagged 85-kDa fragment was increased with MMP inhibitor treatment (p ϭ 0.0004) and E64D treatment (p ϭ 0.0044). All treatment groups, except MG-132, were tested statistically using ANOVA test for trends with a correction for multiple hypotheses using a false discovery rate of ␣ ϭ 0.05. MG-132 was compared with DMSO treated lysates with a Mann-Whitney test and then also corrected using the false discovery rate approach above. All data in E are represented by the mean with S.E. *, p Յ 0.05; **, p Յ 0.01; ***, p Յ 0.001; ****, p Յ 0.0001.

LEMD3, a stiffness-dependent inhibitor of TGF␤ signaling
Smad2/3 in a stiffness-dependent fashion and binding Smad2/3 in the cytoplasm.
Following the observation of cytoplasmic interactions with an intron-less LEMD3 cDNA construct, we hypothesized that these fragments were generated proteolytically. We used Western blottings to determine the effects of several broad-spectrum protease inhibitors against cysteine (E64 and E64d), serine (3,4-dichloroisocoumarin (DCI)), and matrix metallo-proteinases (MMP inhibitor III (MMPi)) as seen in Fig. 4, D and E. From fragments controlled by the ⌬294 -325 region, the 60-kDa FLAG fragment was reduced relative to the full-length transcript in cells treated with DCI (p ϭ 0.0153, Fig. 4E), which is a known inhibitor of cathepsin G, elastase, thrombin, plasmin, factors Xa and X11a, and granzymes A, B, and H (44). This fragment was also reduced when cells were treated with MG-132 (p ϭ 0.0238, Fig. 4E), an inhibitor of the 26S proteosome complex, an important regulator of overall protein homeostasis and particularly critical for degrading ubiquitinated proteins (45). These same trends were observed for the 46-kDa V5 fragment (DCI: p Ͻ 0.0001; MG-132: p ϭ 0.0238, Fig. 4E), which is also controlled by the ⌬294 -325 region. Monash University's PROSPER (Protease Specificity Prediction Server as described in Ref. 46)) predicted only a single serine protease, cathepsin G, cleaving between p.Val-275-Leu-276 (native protein coordinates, score ϭ 1.14) in pFLAG-LEMD3p.⌬294 -325-V5's deletion region. Using a more selective cathepsin G inhibitor, we found a significant reduction in the 60-kDa FLAG fragment (p ϭ 0.0075, Fig. 4E) but no significant trend with the 46-kDa V5 fragment. From the ⌬579 -647 region, the 60-kDa V5 fragment was not significantly modified by any protease inhibitor treatments explored here. Interestingly, the 85-kDa fragment's abundance was only potentiated by MMPi (p ϭ 0.0004, Fig. 4E) and E64d (p ϭ 0.0044, Fig. 4E) treatment. These fragment data together suggested that a serine protease, possibly cathepsin G, operates in the ⌬294 -325-V5 region, with a distinct mechanism of fragment generation occurring in the ⌬579 -647 region.
Finally, Fig. 2D showed that LEMD3-Smad2/3 interactions shifted toward the cytoplasm with increasing substrate stiffness. We tested whether this shift was driven by alterations in C-terminal LEMD3 fragment abundances as a function of substrate stiffness with Western blottings assaying HFFs transfected with pFLAG-LEMD3-V5 (Fig. S6, A and B). After 24 h, cells grown on 1-and 25-kPa hydrogels each had decreased proportions of full-length LEMD3 relative to fibroblasts cultured on tissue culture plastic (TC) (p ϭ 0.0062 and p ϭ 0.0031 for TC versus 1 kPa and TC versus 25 kPa, respectively, see Fig.  S6A). Lysates from 1-and 25-kPa hydrogels had similar abundances of the 85-and 60-kDa fragments, although the 45-kDa fragment was almost twice as abundant on 25-kPa hydrogels as on 1-kPa hydrogels (p ϭ 0.5196, Fig. S6A). We extended our findings to endogenous LEMD3 and found that full-length native LEMD3's abundance was not modulated significantly by culture on 1-or 25-kPa surfaces (Fig. S7A). LEMD3's mRNA expression level was also invariant to culture on these surfaces (Fig. S7B).

C-terminal fragment of LEMD3 binds Smad2/3 in a stiffnessdependent fashion, antagonizes Smad2/3-Smad4 complexes, and binds PPM1␣
To understand the biological significance of these novel CTFs of LEMD3 in the regulation of TGF␤ signaling, we used PLA to ascertain their association with Smad2/3 and their ability to antagonize TGF␤ signaling. The consensus (i.e. cloned past the 2nd cleavage site so as to contain the conserved sequence from all C-terminal fragments) CTF described earlier for Fig. S3 has been shown to be sufficient for Smad2/3 binding (16,18,19,21,41) and lacks the necessary domains for nuclear localization (14,18,33). We repeated the recombinant PLA experiments from Fig. 2, B and F, between CTF's V5 epitope and Smad2/3 to ascertain the following: 1) whether this fragment of LEMD3 was sufficient for generating stiffness-dependent Smad2/3 interactions; 2) whether these fragments were uniquely localized to the cytoplasm; and 3) whether the CTF-Smad2/3 interactions were dependent on the phosphorylation state of Smad2/3.
Unexpectedly, a significant fraction (Ϸ40 -55%, depending on stiffness) of CTF-Smad2/3 interactions took place in the nucleus (Fig. 5B). Although the N terminus of LEMD3 up to the first transmembrane domain has been shown to be necessary for faithful localization of full-length LEMD3 to the nucleus (14,33), the CTF does have an independent ability to bind both barrier-to-autointegration factor (a nuclear chromatin protein (34)) and DNA directly (41), which may allow for enrichment of the CTF in the nucleus. Overall, these V5-Smad2/3 PLA assays LEMD3, a stiffness-dependent inhibitor of TGF␤ signaling with recombinant LEMD3 proteins confirmed the following: 1) the negative stiffness correlation and TGF␤ independence of CTF-Smad2/3 complexes, mirroring the results observed with endogenous LEMD3 assays; and 2) the CTF was sufficient for cytoplasmic LEMD3-Smad2/3 interactions.
Having demonstrated the CTF binds Smad2/3 in a stiffness, but not TGF␤ dose-dependent fashion, we assayed whether CTF overexpression could antagonize TGF␤ signaling by examining the effect of CTF's expression on the abundance of Smad2/3-Smad4 complexes. Smad2/3-Smad4 complexation is a necessary step for efficient nuclear translocation and activity as a transcription factor and is sensitive to Smad2/3's phos-phorylation state (48,49). HFFs treated with 500 pg/ml TGF␤ on 1-kPa hydrogels showed an increase in Smad2/3-Smad4 complexes relative to untreated HFFs (p ϭ 0.001, Fig. 5D). Overexpression of the CTF significantly abrogated this TGF␤driven increase in Smad2/3-Smad4 complexes (p ϭ 0.0081, Fig. 5D). These results mirrored the finding of previous studies, which have shown that the RRM domain of LEMD3 is sufficient for Smad antagonism (16,18,19,21,41). Previous studies have shown that LEMD3 antagonizes TGF␤/Smad signaling by dephosphorylating Smads 2 and 3 in a protein phosphatase, Mg 2ϩ /Mn 2ϩ -dependent 1␣ ("PPM1␣")-dependent manner (16,22). We queried whether endogenous LEMD3 and the CTF LEMD3, a stiffness-dependent inhibitor of TGF␤ signaling could bind PPM1␣ in the cytoplasm and nucleus by PLA. Both endogenous LEMD3 and the CTF bound PPM1␣, demonstrated by interaction frequencies significantly above their respective negative controls (Fig. 5F, total interactions, p Ͻ 0.0001 for both endogenous LEMD3 versus a no primary control, and electroporation only/neon only control versus CTF). Both endogenous LEMD3 and the CTF bound PPM1␣ significantly above background in the cytoplasm (Fig. 5F, cytoplasmic interactions: p Ͻ 0.0001 for endogenous LEMD3 and p ϭ 0.0081 for the CTF), although only endogenous LEMD3 was found to significantly bind PPM1␣ in the nucleus (Fig. 5F, nuclear interactions: p Ͻ 0.0001 for endogenous LEMD3 and p ϭ 0.2569 for the CTF). These data specifically demonstrated the CTF's ability to antagonize Smad2/3 signaling and its connection to PPM1␣, a known LEMD3-associated phosphatase that antagonizes TGF␤ signaling.

LEMD3-SMAD2/3 complexes are more cytosolic, and their frequencies are more varied in IPF biopsies than non-IPF biopsies
We sought to validate our in vitro findings in human lung core biopsies from six patients (three with and three without IPF). IPF patients had significantly more cytosolic LEMD3-Smad2/3 interactions than non-IPF patients (p ϭ 0.0307, Fig. 6, A and C). These data are compelling because they connect the cytoplasmic PLA localization trend seen with increasing stiffness in vitro in Fig. 2D and the pattern observed in fibrosed relative to nonfibrotic human tissue ex vivo. Additionally, these data directly demonstrate that extra-nuclear LEMD3-Smad2/3 interactions occur in human lung tissue, which we also discovered in Fig. 2, indicating that extra-nuclear LEMD3 is not an artifact of in vitro culture.
Patients with IPF also had a slightly higher median frequency of LEMD3-Smad2/3 interactions relative to patients without IPF (Fig. 6D, p ϭ 0.5783). However, although each non-IPF patient had a fairly consistent LEMD3-Smad2/3 interaction "set-point," IPF patients displayed a higher degree of intra-patient heterogeneity across their sampled tissue regions (67% versus 30% average coefficient of variance for IPF and non-IPF patients, respectively). Overall, the whole IPF data set had a higher degree of kurtosis (4.765 versus Ϫ0.1391 for IPF and non-IPF patients, respectively), indicating that more of the LEMD3-Smad2/3 variance in IPF tissue comes from extreme deviations in interaction frequency. In particular, Ϸ22% of tissue regions in IPF patients had uniquely low LEMD3-Smad2/3 interaction rates (Ͻ25% the mean interaction frequency) not observed in non-IPF tissue. Previous micro-mechanical investigations of IPF tissue revealed a high degree of spatial heterogeneity to tissue stiffness in IPF lungs (43). Our findings were consistent with the interpretation that spatially heterogeneous regions of fibrosis in the IPF tissue create both increased variability in the LEMD3-Smad2/3 interaction rate and regions of locally lower LEMD3-Smad2/3 interactions. These data in aggregate showed that LEMD3 regulation of TGF␤ through Smad2/3 is locally diminished in IPF tissue relative to non-IPF tissue and that LEMD3's role is not confined to the nucleus in human tissue.

Discussion
The medical management of fibrotic diseases, such as pulmonary fibrosis, has enjoyed recent success with the approval of pirfenidone, an "anti-fibrotic agent," and nintedanib, an inhibitor of multiple tyrosine kinase receptors, after years of failed clinical trials with a variety of other agents, including anti-inflammatory and anti-oxidant drugs (4, 6, 50, 51). Subsequent study of the mechanisms of pirfenidone has shown that pirfenidone antagonizes fibrotic progression in part by inhibiting the synthesis of TGF␤ (5, 7), which is preferentially activated in fibrotic matrices (8,9). Although pirfenidone may successfully decrease the supply of TGF␤ available in fibrotic pathology, ours and others' findings indicate that the fibrotic matrix also potentiates cellular responses (e.g. synthesis of extracellular matrix and increased cellular contractility) to remaining TGF␤ (10 -13). We focused on the role of LEMD3 in this matrix-driven sensitization, and a graphical summary of our working model for LEMD3-Smad2/3 stiffness-driven interactions is shown in Fig. 7. We showed in vitro that LEMD3 overexpression antagonized TGF␤-driven transcription and stiff-shifts the responsiveness of these fibroblasts to TGF␤, whereas LEMD3 knockdown by siRNA soft-shifts the mechan-  ). B, PLA interactions from A between the V5 tag of a C-terminal fragment of LEMD3 (pFLAG-LEMD3p.⌬21-669 -V5) and Smad2/3, normalized to the total interactions observed on 1-kPa hydrogels. HFFs on 1-kPa hydrogels have significantly more interactions overall (p Ͻ 0.0001), in the cytosol (p Ͻ 0.0001), and in the nucleus (p ϭ 0.0082) relative to fibroblasts on 25-kPa hydrogels. All transfected cells had more PLA interactions than electroporation-only (neon) populations (p Ͻ 0.0001 and p Ͻ 0.0227 for 1-and 25-kPa hydrogels, respectively). Scale bars, 11 m. All statistical testing done with two-way ANOVA with Tukey's post-test. C, representative images from HFFs on 1-kPa hydrogels assayed for Smad2/3-Smad4 interactions. Cells in the 2nd and 3rd rows were treated with 500 pg/ml TGF␤ for 1 h before fixation, whereas cells in the 1st row were untreated. Cells in the 3rd row were electroporated with pFLAG-LEMD3p.⌬21-669 -V5, and cells in the top two rows were electroporation control cells (PLA in green, f-actin in blue, and nucleus in white). All scale bars, 17 m. D, normalized quantification of data from C. Fibroblasts treated with 500 pg/ml TGF␤ and overexpressing the CTF of LEMD3 had significantly fewer Smad2/3-Smad4 complexes than untransfected cells treated with 500 pg/ml TGF␤ (p ϭ 0.0081). Untransfected TGF␤ fibroblasts treated with 500 pg/ml also had significantly more Smad2/3-Smad4 complexes than untransfected fibroblasts (p ϭ 0.001). All statistical testing done with a two-way ANOVA with Tukey's post-test. E, interactions between PPM1␣ and endogenous LEMD3 or its CTF assayed by PLA on glass surfaces. The 1st and 2nd rows used antibody pairs between PPM1␣ and endogenous LEMD3. The 3rd and 4th rows used antibody pairs between PPM1␣ and V5 tag on cells transfected with pFLAG-LEMD3p.⌬21-669 -V5 (PLA in green, f-actin in blue, and nucleus in white). All scale bars are 17 m. F, normalized quantification of the interaction rates between LEMD3 or its CTF by subcellular compartment. Both endogenous LEMD3 and its CTF interacted with PPM1␣ as demonstrated by PLA frequencies above their respective negative controls (for total interactions: p Ͻ 0.0001 for both LEMD3 versus no primary and for CTF versus neon only, respectively). Both LEMD3 and CTF transfected cells also had significantly higher PPM1␣ interaction rates than their respective controls in the cytoplasm (for cytoplasmic rates: p Ͻ 0.0001 and p ϭ 0.0081 for LEMD3 and CTF pairs, respectively), but only endogenous LEMD3 had statistically significant higher interaction rate in the nucleus (for nuclear rates: p Ͻ 0.0001 and p ϭ 0.2569 for endogenous LEMD3 and CTF-expressing fibroblasts, respectively). All statistical testing was performed by a two-way ANOVA with Sidak's post-test. All data represented by the mean with S.E. *, p Յ 0.05; **, p Յ 0.01; ***, p Յ 0.001; ****, p Յ 0.0001.

LEMD3, a stiffness-dependent inhibitor of TGF␤ signaling
ical response to TGF␤ and potentiates TGF␤ signaling. We also showed in vitro that LEMD3-Smad2/3 interactions are inhibited and cytoplasmically shifted with increasing substrate stiffness and actin polymerization. Ex vivo, we identified these same phenotypic correlates in IPF biopsies relative to non-IPF tissue; IPF tissue had a pronounced cytoplasmic shift in PLA subcellular localization; and Ϸ22% of IPF tissue regions had a unique reduction in LEMD3-Smad2/3 interactions relative to non-IPF tissue and a higher degree of variation in the LEMD3-Smad2/3 interaction rate across various tissue regions, both in individual patients and overall. These low interaction regions in IPF are not likely attributable to changes in overall LEMD3 abundance in pathology. Ours and others' data showed that full-length LEMD3 was not regulated by ECM stiffness at the mRNA or protein level in vitro for endogenous LEMD3 (Fig. S7) or for recombinant LEMD3 (Fig. S6) (52, 53). Moreover, IPF patients  Fig. 2D. D, quantification of total LEMD3-Smad2/3 PLA frequency from B showed a similar frequency of interactions between IPF and non-IPF patients (p ϭ 0.5783, Mann-Whitney test). IPF tissue had a higher intra-patient variability (coefficient of variance ϭ 67 and 30% for IPF and non-IPF patients, respectively) and more extreme dispersion overall (kurtosis ϭ 4.765 and Ϫ0.1391 for IPF and non-IPF patients, respectively). 22% of IPF tissue areas sampled formed a unique low-interaction "tail" (Ͻ25% of the mean IPF interaction frequency, denoted by data under the dotted line), which was absent in non-IPF tissues. All scale bars are 10 m. Bars in C and D represent grand medians.

LEMD3, a stiffness-dependent inhibitor of TGF␤ signaling
do not show evidence of altered LEMD3 transcription in either microarray or next generation sequencing of IPF and control patients (54 -56).
Supporting our discovery of LEMD3-Smad2/3 cytoplasmic interactions, we have discovered here that LEMD3 was posttranslationally cleaved at two sites, and we have implicated serine proteases (possibly cathepsin G) in the generation of C-terminal fragments that lack the necessary LEM domain for nuclear targeting (33,34,57). We have also shown that a C-terminal fragment of LEMD3 bound Smad2/3 throughout the cell, associated with PPM1␣, and antagonized Smad2/3-Smad4 complex abundance. These results were consistent with previous studies showing that the C-terminal RRM domain is sufficient for LEMD3's antagonism of Smad2/3 (16,18,19,21,41).
However, the underlying mechanism for a cytoplasmic shift in LEMD3-Smad2/3 interactions with increasing substrate stiffness in vitro and in IPF patient ex vivo is not clear. In vivo, there is strong evidence for an altered and generally more active serine (58), cysteine (59,60), and matrix metallo-proteinases (61)(62)(63)(64)(65) landscape in IPF tissue, although much of the attention has focused on extracellular proteolysis. This more pronounced proteolytic environment could contribute to the cytoplasmic Figure 7. Summary cartoon. Mechanical cues from the extracellular matrix potentiate TGF␤ activation, which ultimately drives the phosphorylation-dependent formation of Smad2/3/4 complexes. These complexes translocate to Smad response elements in the nucleus and facilitate the transcription of TGF␤-dependent genes. LEMD3 antagonizes this TGF␤/Smad2/3 signaling in a stiffness-dependent fashion, altering the stiffness response of fibroblasts to activated TGF␤. LEMD3 achieves this by binding both Smad2/3 and PPM1␣, a protein phosphatase previously shown to be coordinated by LEMD3, both in the nucleus and in the cytosol. Cytosolic LEMD3 fragments are post-translationally generated at two sites, which separate the nuclear-localizing LEM domain and the Smad2/3 interacting RRM domain. Processing at the nucleoplasmic site (bottom red star) generates LEM-and RRM-containing fragments, which are reduced in abundance by serine protease inhibitors but have differential responses to lamin phosphorylation inhibitors. Both nuclear and cytosolic LEMD3-Smad2/3 complexes are inhibited by actin polymerization, which is driven by mechanical cues from the matrix, thereby connecting ECM mechanics to the inhibition of an inhibitor of Smad2/3. Inhibitory interactions are shown with red block-end arrows, and activating signals are denoted with green arrows. Black arrows indicate translocation. Abbreviations used are as follows: Cat G, cathepsin G; RRM, RNA Recognition motif; LEM, Lap2b-Emerin-MAN1 domain; TGF␤-R, TGF␤ receptor; PPM1␣, protein phosphatase, Mg 2ϩ /Mn 2ϩ -dependent 1␣.

LEMD3, a stiffness-dependent inhibitor of TGF␤ signaling
shift seen in IPF patients in our study. In vitro, we found that stiffness representative of scarred tissue in IPF (25 kPa) and physiological lung tissue (1 kPa) produced statistically similar abundances of all the LEMD3 fragments we identified (Fig. S6). However, the 46-kDa fragment was nearly twice as abundant on 25-kPa matrices, prompting a need for further investigation. Overall, we have demonstrated that LEMD3 sits at an interesting intersection of mechanical and biochemical cues in pulmonary fibrosis, and LEMD3 expression may be a promising adjunctive avenue for addressing fibrosis-driven cellular sensitivity to TGF␤ in combination with pirfenidone.
Our original hypothesis for the stiffness regulation of LEMD3-Smad2/3 complexes was that LEMD3's affinity for Smad2/3 would be negatively regulated by biophysical stress from the actin network transmitted to LEMD3 through the LINC-lamina-LEM set of complexes. However, this hypothesis requires critical re-evaluation in light of our discovery of cytosolic LEMD3 fragments that bind Smad2/3 and the apparent lack of efficacy of cytosol-nuclear and lamin-LEM-disrupting constructs (DN-Kash and DN-LEM) in decreasing LEMD3-Smad2/3 complex formation in the nucleus. We have also observed that LEMD3-Smad2/3 complexes are negatively correlated to actin polymerization. However, it is not obvious how actin polymerization itself would directly modulate LEMD3-Smad2/3 complex formation given that neither LEMD3 nor Smad2/3 has a known direct association with actin. One parsimonious explanation would be that LEMD3 competes for Smad2/3 binding with other actin-regulated proteins, namely Yes-associated protein (YAP)/transcriptional coactivator with PDZ-binding motif (TAZ) and/or Rho-associated protein kinase (ROCK). YAP/TAZ and ROCK bind Smad2/3 and have been shown to be necessary for nuclear accumulation of Smad2/3 in response to TGF␤ stimulus (13, 66 -69). ROCK inhibition has been shown to modulate phosphorylation in Smad3's linker region (Ser-203, Ser-207, and Ser-212) (66), whereas TAZ has been shown to bind to the MH1 domain of Smad2 (67). Interestingly, yeast two-hybrid assays indicate that LEMD3 binds to the MH2 domain of both Smad2 and Smad3 (14). Although YAP/TAZ and LEMD3 seem to bind different locations in Smads, YAP/TAZ have been shown to negatively regulate the Smad binding of forkhead box protein H1 (FOXH1), which competes for binding to Smad2/3's MH2 domain with LEMD3 in embryonic stem cell CHIP assays (16,70). These data indicate that YAP/TAZ, in particular, are promising candidates for modulating LEMD3's ability to bind Smad2/3 in an actin polymerization-dependent fashion.
LEMD3 is also directly implicated in the development of BOS with and without melorheostosis. Our findings that LEMD3 was post-translationally processed proteolytically and has cytosolic forms raises new questions regarding the pathogenesis of BOS. Many of the identified BOS patients possess nonsense or splice-site mutations that eliminate the Smad-binding C-terminal end of LEMD3 (20, 24 -31). Interestingly, there are a five missense mutations of unknown clinical significance associated with BOS in NCBI's ClinVar database, which are associated with one of the two deletion mutants (p.⌬294 -325 and p.⌬579 -647) identified in this study 3 (71). These missense mutations could suggest that alterations to the fragment controlling regions of LEMD3 might play a role in the development of BOS. Additionally, our finding that LEMD3 was processed by a serine protease and the 26S proteosome might suggest additional targets/mechanisms in individuals with BOS who lack mutations in LEMD3 (32).
Finally, we have characterized several N-and C-terminal LEMD3 fragments and demonstrated their varying sensitivities to the following: 1) genetic perturbation through deletion mutants; 2) serine (DCI), cathepsin G, and 26S proteosome (MG-132) inhibitors; 3) lamin integrity/disassembly through Cdk1 inhibition; and 4) substrate stiffness. One particular pair of fragments, the 60-kDa N-terminal FLAG and 46-kDa C-terminal V5 fragments, is particularly interesting because both are sensitive to the same mutation region (p.⌬294 -325), the abundance of both fragments is reduced with respect to DCI and MG-132 treatment, and they plausibly sum to the mass of fulllength LEMD3. However, these two fragments are differentially regulated by integrity of the lamin network (following inhibition of Cdk1 by roscovitine) and by cathepsin G inhibition, making it less likely that these fragments are generated in a concerted proteolytic reaction. Additionally, we do not have a straight-forward explanation relating the position of our deletion mutations and the fragments' sizes they seemingly control. It is not possible to directly align our Western blotting fragment data and mass spectrometry data because of the unknown distributions of post-translational modifications in LEMD3; however, the fragments' relative masses controlled by either deletion mutant do not logically correspond with the expected size of products from cleavage in those areas. This is, to our knowledge, the first report of an integral inner nuclear membrane protein processed proteolytically into cytosolic forms. However, our observations fit a broader pattern of a more locationally dispersed role for nuclear envelope integral proteins. Several integral members of the nuclear membrane have alternative splice forms that both include and omit their transmembrane domain(s), including nesprin, lamina-associated polypeptide 2 (LAP2), torsin-1a-interacting protein 1 (LAP1), and nurim (72)(73)(74)(75). Moreover, some "localized" elements of the nuclear envelope have been found dispersed throughout the cytosol based on the cellular context. For example, nesprins, structural elements of the outer nuclear membrane that link the cytoskeleton and nucleus, are restricted to the nuclear envelope in myoblasts but not in differentiated myotubes (76,77), and certain nesprin isoforms are also decompartmentalized during muscle regeneration in Duchenne muscular dystrophy (77). Additionally, alternatively spliced "KASH-less" nesprins, which lack the nuclear envelope anchoring domain, KASH, have recently been linked to a variety of cytosolic locations, including focal adhesions and actin stress fibers (72), Golgi bodies (78), and RNA-processing bodies (79,80). Our findings expand on the observed mechanisms that regulate the localization of nuclear envelope proteins in the cell.

Primary fibroblast cultures and transfection, human lung biopsies, and other reagents
Primary HFFs were procured from the ATCC (ATCC SCRC-1041, ATCC, Manassas, VA) and routinely cultured in DMEM (ThermoFisher Scientific, Waltham, MA) supplemented with 15% FBS and 1% penicillin/streptomycin (ThermoFisher Scientific, Waltham, MA) in 5% CO 2 at 37°C in a humidified incubator to passage 13. CCL210s, a primary human pulmonary fibroblast line, were procured from the ATCC (ATCC CCD-19Lu) and routinely cultured in Eagle's minimal essential media (ThermoFisher Scientific) supplemented with 10% FBS and 1% penicillin/streptomycin (ThermoFisher Scientific). For DNA transfections, 50,000 HFFs were resuspended in 10 l of Buffer R and electroporated using a 1700-V, 20-ms pulse width, 1-pulse program with 1 g of DNA in a Neon transfection system (ThermoFisher Scientific). For siRNA transfections, 8000 -10,000 cells/cm 2 were treated with 25 or 200 nM siRNA with Lipofectamine2000 (ThermoFisher Scientific) according to the manufacturer's recommendations. IPF and non-IPF human lung core biopsies were generously provided by Dr. Eric White and the Lung Tissue Research Consortium (NHLBI: HHSN2682016000021). Selected cores all came from distal sections of lung parenchyma. Human subjects' approval for this tissue was obtained through the University of Michigan Institutional Review Board. This study abides with the Declaration of Helsinki principles. For information on the oligos, antibodies, and uncommon chemicals used in this work, please refer to the Tables 1-3 for supplier and use information.

Generation of recombinant constructs
LEMD3 constructs were derived from the pSVK3-FLAG-MAN1 construct, kindly provided by Dr. Howard Worman (Addgene plasmid no. 26002) (33). SDM was performed using the Q5 SDM kit (New England Biolabs, Ispwich, MA) according to their protocol, including the use of "NEB BaseChanger" applet to design primers and select PCR conditions. Primers and associated melting points for each mutagenesis are listed in Table 1. Constructs were confirmed with Sanger sequencing (MacrogenUSA, Rockville, MD).

Western blotting
For nonphosphoprotein Western blottings, cell lysates from tissue culture plastic or from 2-or 25-kPa, 150-cm Petrisoft dishes (Matrigen, Brea, CA) were either harvested directly in 4ϫ protein loading buffer (Licor, Lincoln, NE) supplemented with protease and phosphatase inhibitors (no. A32959, Thermo-Fisher Scientific) according to the manufacturer's instructions or processed with the NE-PER compartment isolation kit (ThermoFisher Scientific) according to the manufacturer's instructions. Lysates were treated with 0.3 l of benzonuclease (ThermoFisher Scientific) for 15 min at room temperature with vigorous shaking. Samples were heated to 95°C for 5 min and run on a 4 -12% BisTris gel in 1ϫ MES buffer and run for 70 min at 150 V. The gel was then transferred to a nitrocellulose membrane using the XCell Blot Module (ThermoFisher Scientific) at 25 V for 65 min. Membranes were air-dried for an hour, blocked with 5% nonfat milk in 1ϫ PBS for an hour at room

LEMD3, a stiffness-dependent inhibitor of TGF␤ signaling
temperature, and then incubated with primary antibodies as described in Table 2 in 1ϫ PBS-T with 1% milk for 16 -24 h at 4°C. Near-IR or HRP-conjugated secondaries as described in Table 2 were then incubated with the blot and imaged subsequently on either an Odyssey CLx system for near-IR dyes (Licor, Lincoln, NE) or a ChemiDoc MP system (Bio-Rad) with SuperSignal West Femto reagents (ThermoFisher Scientific) for HRP secondaries according to the manufacturer's instructions. Band analysis was performed using the associated Licor or Bio-Rad software.
For phosphoprotein Western blots, the procedure above was followed with the following changes. Cells were treated with 100 pg/ml TGF␤ 90 min before lysis. Membranes were incubated in and washed in buffers formulated with TBS instead of PBS. Membranes were blocked in 5% BSA in 1ϫ TBS, and all antibody incubations were done in 1% BSA in 1ϫ TBS-T.
For recombinant LEMD3 Western blots, two samples of HFFs were prepared and electroporated as above and plated onto each plastic or hydrogel surface for 12 h (protease inhibitor experiments) or 24 h (LEMD3 fragment abundance and deletion mutant experiments) before harvesting.

qPCR
RNA was isolated from cultured cells using RNeasy Plus Mini kits (Qiagen, Hilden, Germany) according to the manufactu-rer's instructions. cDNA libraries were constructed using RT 2 First Strand kit (Qiagen, Hilden, Germany) according to the manufacturer's instruction. 25-l qPCRs were prepared using SYBR Green Master Mix (ThermoFisher Scientific) with 250 nM primers, listed in Table 1, and 10 ng of input cDNA. Reactions were carried out on a StepOnePlus (ThermoFisher Scientific) for 40 cycles. Melt curves were visually inspected after each run to confirm a single defined peak in the derivated intensity plot. Data from each run was imported into LinRegPCR, which performed baseline correction and measured PCR efficiency by amplicon group as described previously (81). The relative number of transcripts was calculated by dividing the C q threshold by the PCR efficiency for the reaction raised to the power of C q , the cycle number at which the threshold is reached for a given target. For relative quantification between conditions, LEMD3 transcripts were either normalized for each condition to the mean of two reference genes, 18S and ACTB, if the samples were all from a single stiffness condition or normalized to 18S alone if substrate stiffness was an experimental condition.

Smad-driven luciferase assays
HFFs were transfected with Cignal Lenti Reporter virus (Qiagen, Hilden, Germany) with a multiplicity of infection of 40 and subsequently selected using 400 ng/ml puromycin until Table 2 Primary, secondary, and other antibodies used in this work Table 3 Chemicals and recombinant proteins used in this work LEMD3, a stiffness-dependent inhibitor of TGF␤ signaling untransduced HFFs died. Stably transfected HFFs, with or without additional LEMD3-focused genetic treatments as described above, were plated at 10,000 cells/well on an HTS plate (Matrigen, Brea, CA) containing glass surfaces and polyacrylamide gels, which ranged in stiffness from 0.2 to 50 kPa, all functionalized with 10 g/ml plasma-purified human fibronectin according to the manufacturer's instruction. Cells adhered to the surfaces for 4 h in serum-free DMEM supplemented with 1% BSA and 1% penicillin/streptomycin. Media were then supplemented with the desired amounts of TGF␤ and, when applicable, with cytoskeletal agents as described under the "Results." HFFs were then incubated for 16 h at 37°C before the luciferase reaction. Wells were supplemented to 2 mM VivoGlo D-luciferin (Promega, Madison, WI) using a plate reader-controlled auto-injector system, and light production was subsequently quantified every 5 min for 30 min on a Synergy H4 plate reader (BioTek, Winooski, VT). The raw luciferase signal was the median intensity measured over this time course minus the intensity from luciferase-transfected cells not treated with TGF␤ to account for any TGF␤ elaborated and activated by the cells in situ. Cells were then fixed with 4% paraformaldehyde, permeabilized with 0.1% Triton X-100, and then stained with 1:10,000 Hoechst stain 33342 in 1ϫ PBS for 30 min at room temperature (ThermoFisher Scientific). The raw luciferase signal was then normalized to the nuclear signal, which was inside a linear standard curve from cells plated on glass and on polyacrylamide within each plate. Data for each condition was then fit to a three-parameter logistic equation using a least-squares minimization fit in Prism (GraphPad, La Jolla, CA). For curves that did not converge with this model, individual data points are shown as the mean with the standard error of the mean to illustrate data heterogeneity. Otherwise, the trends across the maximal luciferase signal ("Top" variable from the model) or the mid-transition point observed between "Bottom" and "Top" signals ("EC 50 " variable from the model, called "transition point" in text), were statistically compared using an ANOVA with "test for trends," which restricts the hypothesis testing to a particular order of the data sets (e.g. maximum luciferase signal as a function of increasing TGF␤ dose or transition point as a function of increased LEMD3 expression). Each data point represents at least three biological replicates.

Proximity ligation assays
Ex vivo PLA-Lung biopsy samples were prepared by OCT embedding flash-frozen tissue (Tissue-Tek, Sakura Finetek, Torrance, CA) and preparing 10-m cryosections on a Cryo-Star NX70 (ThermoFisher Scientific). Sections were fixed for 10 min at 4°C in 2% paraformaldehyde in 1ϫ PBS. Proteinprotein PLAs were performed with antibody pairs listed in Table 2 as described previously (42) with the following modifications: all primary antibody incubations were done for 16 -24 h at 4°C; sections were blocked in 0.5% Tween 20, 0.1% Triton X-100, 0.5% gelatin, 5% donkey serum, 2% bovine serum albumin (BSA), and 5 g/ml poly(dI-dC) DNA in PBS, and sections were actin counter-stained with 1:40 phalloidin-488 (Thermo-Fisher Scientific) for 30 min at room temperature following the PLA reaction using the IF protocol recommended by Sigma's PLA Resource Center. Slides were analyzed using a Axiovert 200M microscope (Carl Zeiss Microscopy, Jena, Germany) with an UltraVIEW spinning disk (PerkinElmer Life Sciences) and Flash 4.0v2 cMOS camera (Hamamatsu Photonics, Hamamatsu City, Japan). Low resolution images (ϫ20) were collected using a 0.8NA Plan-Apochromat, and high resolution images (ϫ63) were collected using a 1.4NA Plan-Apochromat objective (Carl Zeiss Microscopy, Jena, Germany). Image acquisition and analysis were performed in Volocity (PerkinElmer Life Sciences). To analyze the incidence of LEMD3-Smad2/3 interactions between IPF and non-IPF biopsies, at least six random points were chosen by Volocity in each section and registered to a serial section of the same tissue that was prepared as a no primary antibody PLA control. Tiled images with 20% overlap were acquired at ϫ20, covering a 628 ϫ 334-m area with z-slices acquired every 0.6 m. The PLA signal was processed in Volocity using a consistent intensity threshold to identify PLA puncta, which were then spatially filtered by excluding PLA signal not associated with the actin or DAPI tissue-based signals. The remaining PLA signal was processed using the "subtract" function to remove tissue autofluorescence, measured independently in an unused spectral window (615 nm, width 70 nm). The PLA signal was then normalized to the measured nuclear volume in the image slice. Relative PLA incidence between conditions was calculated by subtracting the normalized PLA signals between experimental and no primary antibody at each co-registered pair of points. High-resolution imaging (ϫ63) of each tissue was performed over at least six randomly chosen areas, each 111 ϫ 106 m with z-slices acquired every 0.2 m, to measure the cytosolic and nuclear PLA compartments in tissue. Signal processing was similar to the low-resolution processing described above except that the PLA signal was subdivided using the "exclude non-touching" function in Volocity to measure the nuclear and non-nuclear associated signal.
In vitro PLA-Protein-protein PLA was performed with antibody pairs listed in Table 2 as described previously (42) with the following modifications: cells were plated on fibronectincoated glass or hydrogels (Matrigen, Brea, CA). All primary antibody incubations were performed for 16 -24 h at 4°C, and cells were counter-stained with 1:40 phalloidin-488 or phalloidin-546 in 1ϫ PBS for 30 min at room temperature and/or further processed for IF imaging using the procedure recommended in Sigma's PLA resource center. For all recombinant LEMD3 and C-terminal fragment experiments, HFFs were electroporated as above and allowed to culture for 24 h on 10 g/ml fibronectin-coated glass or hydrogels before fixation. For LEMD3-phospho-Smad2/3, LEMD3-Smad2/3 ϩ TGF␤, and Smad2/3-Smad4 PLA assays, cells were treated 90 min before fixation with 50 pg/ml TGF␤ for LEMD3-and CTF-Smad2/3 assays, 100 pg/ml TGF␤ for LEMD3-phospho-Smad2/3 experiments, and 500 pg/ml TGF␤ for Smad2/3-Smad4 experiments. For FLAG post-staining of PLA-processed tissues, cells were blocked with unlabeled AffiniPure donkey anti-rabbit antibodies (Jackson ImmunoResearch, West Grove, PA) at a 1:10 dilution in 1ϫ PBS-T at room temperature for 1 h before further IF processing with 1:1000 rabbit anti-FLAG (no. F7425, Sigma, Darmstadt, Germany) for 1 h in PBS-T at room temperature. Following three washes with PBS-T for 5 min, LEMD3, a stiffness-dependent inhibitor of TGF␤ signaling each sample was stained with 1:500 goat anti-rabbit 546 in PBS-T for 1 h at room temperature, washed again, and then counter-stained as above. For V5 post-staining of PLA-processed samples, cells were incubated with 1:200 647-labeled anti-V5 antibody (R&D Systems, Minneapolis, MN) diluted in PBS-T at room temperature for 1 h before washing three times in PBS-T for 5 min each and then counter-stained as above. Imaging and analysis was performed as described for high-resolution ex vivo PLA signals except that no tissue autofluorescence signal was measured or needed to be subtracted and, when appropriate, other IF signals (e.g. FLAG) were measured in parallel. For cells on Matrigen HTS plates (Fig. 2, C and D), the confluence of cells prevented identifying individual cell boundaries. Instead, the PLA signal was quantified as the PLA volume divided by the nuclear intensity for each image. Each data point represents at least three unique surfaces with at least 12 cells measured per surface. For Smad2/3-Smad4 experiments, only two independent hydrogels were analyzed for the "CTF 500 pg/ml" and "Neon 500 pg/ml" groups due to loss of hydrogels during processing. Each bar in Fig. 5D represents the mean with standard error of the mean for all the cells analyzed from that condition.

Atomic force microscopy and cell morphology
Experiments were conducted on an MFP-3D AFM (Asylum Research, Santa Barbara, CA) on an inverted optical microscope (Nikon, Melville, NY). MCLT O-10 cantilevers (Bruker Nano, Camarillo, CA) were functionalized manually with 4.47-m polystyrene beads. Probes were calibrated before each experiment using a combination of glass indention and thermal fluctuation measurements. Cellular and hydrogel measurements were acquired using a 2 nanonewton relative force trigger with a tip velocity of 2 m/s. Hydrogels were measured over a 400 m 2 space without cells, sampled as a 4 ϫ 4 array. Young's modulus was calculated using custom scripts in MATLAB (TheMathWorks, Waltham, MA) that use a linearized Hertz approach for force-indentation measurements 40 -60 nm into each surface. Cells and hydrogels were assumed to be incompressible (Poisson's ratio, ϭ 0.5). After AFM analysis, cells were fixed in 4% paraformaldehyde for 10 min and stained with 1:40 phalloidin-488 and 1:1000 Hoechst stain 33342 for 30 min. Cell spread area and morphology were analyzed from images taken at ϫ63 in Volocity (as above). Each data point represents at least three unique surfaces with at least 10 cells per surface.

Mass spectroscopy on LEMD3 fragments
LEMD3 fragments were affinity-purified from RIPA-extracted lysates using FLAG (clone M2, no. A2220, Sigma, Darmstadt, Germany) or V5 (clone V5-10, no. A7345, Sigma, Darmstadt, Germany) affinity agarose according to the instructions of the manufacturer. Purified fragments were separated by SDS-PAGE on a 4 -12% BisTris gel in 1ϫ MES buffer and visualized by SimplyBlue staining (ThermoFisher Scientific). In gel digestion, nano-LC-MS/MS, and peptide identification was performed as described previously (82) with the following modifications. Reverse-phase chromatography was performed using an in-house packed column (40 cm long ϫ 75 m inner diameter ϫ 360 outer diameter, Dr. Maisch GmbH ReproSil-Pur 120 C18-AQ 1.9-m beads) and a 120-min gradient. The Raw files were searched using the Mascot algorithm (version 2.5.1) against a protein database constructed by combining the FASTA file for LEMD3 with a contaminant database (cRAP, downloaded November 21, 2016 from https://www.thegpm. org/crap/index.html) 4 via Proteome Discoverer 2.1. Only peptide spectral matches with expectation value of less than 0.01 ("high confidence") were used.