Loss of furin in melanoma grafts impairs A110 and A70 cleavage and the associated tumor growth, yet without depleting circulating mature Activin-A

To assess the role of Activin-A processing and its regulation by furin in vivo, we analyzed syngeneic grafts of B16-F1, FurKO, or DKO melanoma cells expressing βA versus empty control vectors in C57BL/6J mice. Analysis of tumor growth curves and tumor weights at the endpoint revealed that compared to Ctrl, B16F1-βA grafts had a marked growth advantage and significantly diminished the body weights of the host (Fig. 1D–F), as described previously^15,22. By contrast, in FurKO-βA and DKO-βA grafts lacking Furin, Activin-A failed to accelerate tumor growth compared to Ctrl groups. Importantly, a comparison of body weight of the hosts before tumor grafting and at the endpoint revealed that the systemic effect of βA expression was maintained nonetheless, as clearly seen in FurKO#2 and FurKO#3 tumors (Fig. 1F). In hosts of βA-expressing FurKO#1 and DKO tumors, a similar trend did not reach statistical significance. These two CRISPR clones form tumors more slowly than parental βA-expressing cells (Fig. 1D), despite comparable cell proliferation in vitro (Fig. S1E), suggesting they did not quickly enough form tumors of sufficient size for cachexia to manifest before reaching the endpoint.

To test if half-processed Activin-A forms in vivo, tumor lysates and plasma samples of tumor-bearing mice were analyzed by Activin-A Western blots. We found that B16F1-βA tumors contained mainly A30 together with A60. By contrast, FurKO-βA tumors lacked detectable A60 and instead accumulated A70 together with A110, confirming that no other PC substituted for furin within the tumor cells. Nevertheless, A30 formation persisted in the FurKO-βA tumors (Fig. 1G). Immunoblots of plasma, and an ELISA assay detecting specifically A30 revealed that the FurKO-βA tumors also still released circulating mature Activin-A but no half-processed or uncleaved precursor (Figs. 1H and S1F, G). Collectively, these results indicate that hosts of FurKO-βA tumors rescued both A30 formation and the associated loss of body weight. By contrast, to stimulate tumor growth, Activin-A depends on the coexpression of endogenous furin in cis that cleaves A110 and A70 within the tumor cells.

In the absence of cell contacts, half-processed Activin-A depends on further extracellular cleavage by the recipient cells to robustly activate SMAD3

Whereas furin is active both in intra- and extracellular compartments, the localization of PCLP activity is unknown. Therefore, to further characterize the furin-independent Activin-A cleavage and its role in melanoma, we first asked where it takes place. To address this, βA-expressing cells were treated with CMK, or with membrane-impermeable hexa-D-arginine (D6R) (Fig. 2A). While D6R had no effect even at the maximal concentration (200µM), treatment with 13–25µM CMK abolished the PCLP-mediated A70 formation in FurKO#1-βA cells, as well as A60 and A30 formation in B16F1-βA cells. CMK treatment also stabilized A110 in Furin wild-type cells, albeit only at elevated concentrations of 50–100µM, and incompletely compared to Furin-deleted cells. These data suggest that furin and PCLP activities in B16-F1 cells cleave Activin-A intracellularly.

Open in a separate window

Fig. 2

Effects of furin inhibitors on Activin-A processing by signal-sending versus signal-receiving cells.

A Western blot of Activin-A in SNs of B16F1-βΑ (top) and FurKO#1-βΑ cells (bottom) that were treated with increasing doses of CMK or with the membrane-impermeable PC inhibitor D6R for 36h. B Luciferase reporter assay to test if loss of furin-mediated cleavage within melanoma cells blocks Activin-A signaling. HepG2 reporter cells were incubated for 12h with 10-fold diluted SNs of the B16-F1, FurKO#1, or DKO melanoma cell lines shown in Fig. 1A, B which express βA or control (ctrl) lentivirus (top), or which were transfected with wild-type or S1 site mutant (mS1) βA plasmids (bottom). Induction of the SMAD3 reporter CAGA-Luc by SNs of transiently transfected melanoma cells is shown as a percentage relative to the fold induction by SN of βA-transfected Furin wild-type control cells. Error bars, SEM (n=3–7 experiments); **p<0.01, ordinary one-way ANOVA with Šídák’s multiple comparisons test. C Analysis of Activin-A precursor processing after 12h incubation with HepG2 CAGA-Luc reporter cells with or without the furin inhibitors CMK (100µM) or D6R (50µM). SNs of the melanoma cell lines producing the indicated forms of Activin-A were each diluted 10-fold to not exceed the dynamic range of the assay. Induction of CAGA-Luc was normalized relative to (i) the fold induction by control SN (A30+A60) of Furin wild-type B16F1-βA cells (left panel) or (ii) to the activity of 10ng/ml recombinant Activin-A (right panel). Western blot analysis of SNs after their incubation on HepG2 is shown below. To treat HepG2 cells with only uncleaved A110, we used the SN of FurKO#3-βA cells as input, because A70 formation can be inhibited in this clone by culturing it at subconfluent cell density (see supplementary Fig. S2). Error bars, SEM (n=5–7 independent experiments). *p<0.05, **p<0.01, ***p<0.001, ****p<0.0001 ordinary one-way ANOVA with Šídák’s multiple comparisons test. D Analysis of Activin-A processing and signaling after 12h in cocultures of FurKO-βA cells with HepG2.α1-PDX CAGA-Luc reporter cells that express α1-PDX to inhibit Activin-A cleavage by the signal-receiving HepG2 cells. Where indicated, the cocultures were treated with 100µM CMK to also block PCLP activity in the signal-sending FurKO#1-βA melanoma cells. Treatment with 10ng/ml Activin-A was a control of CAGA-Luc induction. Lack of luciferase activation in control cocultures with FurKO#3-βA cells validated that α1-PDX blocks Activin-A cleavage in trans by HepG2 cells. Error bars, SEM (n=2 experiments). *p<0.05, Student’s t-test. E As in (D), but in transwell chambers where the FurKO#1-βA cells were co-cultured with the HepG2. α1PDX cells as indicated. Schemas were created using BioRender.com.

To test the influence of furin-independent processing on Activin-A signaling, we incubated SNs from the βA-transduced FurKO#1-βA or DKO-βA cell lines or from Furin wild-type B16F1-βA cells with HepG2 cells carrying the SMAD3 luciferase reporter CAGA-Luc, together with CMV-Renilla for signal normalization³⁶. Interestingly, SNs from all three cell lines induced CAGA-Luc 150- to 200-fold above baseline within 12h (Fig. 2B, top). This response is well within the dynamic range of 20- to 400-fold induction by 5–50ng/ml recombinant Activin-A³⁶. SNs from B16-F1 cells lacking Furin alone or both Furin and Pcsk7 also induced CAGA-Luc to a similar extent if βA was only transfected transiently (Fig. 2B, bottom). However, no CAGA-Luc was induced by SN of mS1-transfected cells, confirming that signaling required S1 site cleavage.

To estimate the contribution of cell non-autonomous Activin-A maturation, we added CMK to the signal-receiving HepG2 reporter cells (Fig. 2C). SNs of Furin wild-type B16F1-βA cells containing A30 and A60 induced CAGA-Luc regardless of the presence of a high dose of CMK (100µM). By contrast, in reporter cells receiving FurKO#1-βA SNs, CMK administration blocked CAGA-Luc induction or, if traces of A30 still formed in input samples, severely diminished it (Fig. 2C(i), top panel). Western blots of the SNs after the incubation on HepG2 cells showed that CMK treatment stabilized both A110 and A70 (Fig. 2C(i), bottom panel). Collectively, these data suggest that if the melanoma cells lack furin coexpression in cis, signal-receiving cells complete Activin-A processing in trans.

Signal-receiving cells may cleave proActivin-A extracellularly. To test this, we produced uncleaved A110 in FurKO#3-βA cells, a clone where A70 formation fortuitously can be suppressed by culturing the cells at subconfluent density (Fig. S2A, B). During control incubations at 37 °C without HepG2 reporter cells, A110 from subconfluent FurKO#3-βA cells as well as A70 and A110 from confluent FurKO#1-βA cells remained stable for at least 12h even in the absence of CMK (Fig. S2C). By contrast, when incubated on HepG2 cells, A110 was converted to A70 and A30 and induced CAGA-Luc as strongly as did treatment with 10ng/ml Activin-A (Fig. 2C, panel (ii)). Co-treatment of the HepG2 cells with CMK or with the extracellular inhibitor D6R, respectively, blocked or severely inhibited this activity, correlating with the stabilization of A110. Thus, the signal-receiving HepG2 cells can activate even proActivin-A extracellularly.

Contact of melanoma cells with co-cultured HepG2 reporter cells enables furin-independent signaling by half-processed Activin-A but not by its precursor

Furin is also expressed in HepG2 cells, together with PACE4 and PC7³⁷. To test if the signaling activity of half-processed Activin-A or its precursor can be blocked by inhibiting furin and/or related PCs only in the signal-receiving cells, we transduced HepG2 CAGA-Luc reporter cells with the antitrypsin variant α1-PDX³⁸. To induce CAGA-Luc expression, the resulting HepG2.α1-PDX were then co-cultured in contact with FurKO#3-βA as a source of only A110, or with FurKO#1-βA cells which secrete also A70. Alternatively, as a positive control, they were treated with 10ng/ml Activin-A (Fig. 2D). While the control treatment with Activin-A upregulated CAGA-Luc at least 20-fold, the A110 secreted by co-cultured FurKO#3-βA remained uncleaved and inactive, confirming that the presence of α1-PDX efficiently blocked Activin-A processing in trans. By contrast, in cocultures with FurKO#1-βA cells that secrete A70, the α1-PDX transgene failed to prevent CAGA-Luc induction. In this setting, treatment with CMK was essential to block signaling, correlating with the suppression of A70 formation by the melanoma cells. To test if this signaling activity required cell contacts, cocultures with HepG2.α1-PDX reporter cells were analyzed in transwell chambers (Fig. 2E). FurKO#1-βA melanoma cells again induced CAGA-Luc in HepG2.α1-PDX cells that were in the same compartment. By contrast, if the cells were co-cultured in separate chambers without contact, A70 still formed but failed to signal.

The Activin-A processing intermediates A70 and A60 bind ActR-II despite covalent linkage to their prodomain

Precursor cleavage is essential for Activin-A to bind type I and II activin receptors³⁹. To test if half-processing initiates binding at least to cognate type II receptors, we first injected comparable relative amounts of A110, A70, or A30 from B16F1-βA and FurKO-βA SNs into microfluidic chips that were precoated with ActRIIB-Fc to measure the reduction of plasmonic intensity resulting from ActRIIB-Fc binding (Fig. 3A)^40,41. To estimate non-specific background, we injected B16F1-Ctrl or FurKO-Ctrl SNs, and SNs of B16F1-mS1 cells secreting uncleavable proActivin-A. Among FurKO clones, we arbitrarily chose clone #2 which was indistinguishable from clone #1 in vitro (Fig. 1C–E). As expected, SNs from Ctrl or mS1 expressing cells barely diminished the plasmonic intensity. By contrast, injection of B16F1-βA SN resulted in a robust signal intensity drop. A comparable drop in signal intensity ensued upon injection of FurKO#2-βA SN (Fig. 3B). These data suggest that ActRIIB-Fc binds both A30 and A70.

Open in a separate window

Fig. 3

Binding of half-processed Activin-A to type II activin receptors, and analysis of its βA subunits on reducing gels and by mass spectrometry.

A Western blot of representative B16F1-mS1 and FurKO#2-ctrl or -βA cell SNs, and ImageJ quantification of the proportions of mature, hemicleaved, and uncleaved Activin-A in independent sample preparations from the indicated cell lines used as inputs to test Activin-A binding to immobilized ActRIIB-Fc by plasmon resonance measurements. The microfluidic chamber used to inject cell SNs on a nanoplasmonic chip functionalized with ActRIIB-Fc on gold nanoparticles is depicted on the right. Drops in signal intensity are induced by Activin-A binding to ActRIIB-Fc. B Binding of the different forms of Activin-A was evaluated by injecting the B16F1-βA or FurKO#2-βA cell SNs analyzed in (A). B16F1-mS1, or B16F1-Ctrl cells expressing empty lentivirus served as specificity controls. Error bars, SEM (n=3 independent experiments); *p<0.05, **p<0.01, ***p<0.001, ****p<0.0001, Student’s t-test. C Schematic of the conversion of hemicleaved Activin-A (A70) to A60 by furin cleavage of the second βA subunit, and strategy to monitor the cleavage using anti-Flag and anti-Activin-A (ActA) antibodies. In this model, an allosteric cystine bond (S–S) links the prodomain (violet) of one βA subunit to the C-terminal region (yellow) in A70 and A60. Prodomain that is not covalently attached (light shading) can be displaced by cognate type II receptors. D Anti-Activin-A (top) or anti-Flag Western blots (bottom) of A70 in SN of FurKO#1 cells expressing Flag-tagged 27F-βA analyzed before (input) and after pull-down by AIIB-Fc. SN of B16-F1 cells, and of mock-transfected FurKO#1 cells served as controls. Arrow, cleaved prodomain; n.s. non-specific. E Anti-Activin-A (top) or anti-Flag (bottom) Western blot of the indicated Flag-tagged Activin-A constructs after pull-down as in (D), but on reducing gels. FurKO#1 and B16-F1 control SNs used to pull down Flag-tagged A70 and A60 encoded by 130F-βA or 146-βA were as in Fig. S3C. F LC–MS analysis and spectral counts of peptides mapping to Activin-A in samples collected in (E). Data represent two pull-down experiments, one using ActRIIA-Fc and the other using AIIB-Fc. Schemas were created using BioRender.com.

A70 and A60 may contain a prodomain that failed to be cleaved, or which remained covalently linked to the mature region via a disulfide bond (Fig. 3C). To distinguish between these scenarios, and to assess the influence of prodomain cleavage on ActR-II binding by an alternative approach, SNs of cells expressing Activin-A were incubated with ActRIIB-Fc on protein A/G beads, followed by Western blot analysis of bound protein (Fig. 3C). To monitor the prodomain, flexible regions after residues 27, 130, or 146 were tagged with a Flag epitope (Fig. S3A). All three constructs were processed as expected both in Furin WT and KO cells (Fig. S3B), and their cleaved prodomains (39kDa) were detectable by anti-Flag antibody both on reducing and on non-reducing blots (Fig. S3C). However, the binding of anti-Flag antibody to reduced precursor monomers or to non-reduced A70 or A60 was too weak to detect them without prior enrichment. Therefore, we enriched the Flag-Activin-A construct 27F-βA by pull-down using ActRIIB-Fc beads. Non-reducing Western blot analysis of the bead eluates revealed A70 and A60, and that both reacted with the anti-Flag antibody, suggesting that they harbor the prodomain (Fig. 3D). By contrast, binding of ActRIIB-Fc to cleaved prodomain was close to background. Moreover, analysis on reducing gels after ActRIIB-Fc pull-down validated that A70 from all three Flag-tagged constructs contained the cleaved mature region and prodomain together with the uncleaved βA subunit (Fig. 3E, top panel). These results confirm that A70 corresponds to the previously described hemicleaved 70kDa processing intermediate of Activin-A²⁷. By contrast, A60 from B16-F1 cell SN retained the prodomain exclusively in its cleaved form (Fig. 3E, bottom), presumably via a disulfide bond (Fig. 3C, and see below).

Since A70 and A60 differed in their electrophoretic mobility, we wondered if the A60 prodomain might be truncated. To address this, A70 from FurKO#2-βA SN, and A60 and A30 from B16F1-βA SN were individually excised from non-reducing gels after pull-down by ActRII-Fc fusion proteins and compared to gel-purified A110 from FurKO#2-βA SN using liquid chromatography-mass spectrometry (LC–MS) (Fig. S3D, E). A pull-down from FurKO#2-Ctrl SN served as negative control. To resolve the cystine knot structure of the mature region, immunoprecipitates had to be treated with DTT before tryptic digestion, precluding the detection of potential alternative disulfide bonds. Counting of spectra that match Activin-A revealed that after the first 25 residues, almost all segments of the prodomain were represented both in A70 and A60 samples, and with similar overall coverage as in A110 in two independent experiments using ActRIIA-Fc or ActRIIB-Fc, respectively (Fig. 3F, and Supplementary Table 1). By contrast, >99% of the peptides identified in A30 samples mapped to the mature region, and no Activin-A peptides were detected in the negative control. These results suggest that even though the mature region of Activin-A remains covalently attached to the prodomain of one of its βA subunits in both A70 and A60, this linkage does not prohibit binding to ActR-II.

Furin converts A70 to A60 independently of an N-glycosylation site and of alternative KR or RR motifs in the Activin-A prodomain

To further investigate how furin converts A70 to A60, we tested the role of potential secondary furin sites. In the absence of stronger candidates, we considered only KKR₇₁ and RR₁₁₀ (Fig. S3A). Alanine substitutions of these motifs did not impair A60 formation by furin, or signaling activity (Fig. S3F, G). As an alternative possibility, we assessed whether A60 and A70 differ in their electrophoretic mobility due to differential N-glycosylation. The only asparagine that can be N-glycosylated in proActivin-A is in the prodomain (N165). Alanine substitution of N165 in proActivin-A (mN) increased the electrophoretic mobility of A60, confirming by yet another approach that A60 contains the prodomain (Fig. S3H). Despite this size shift, the signaling activity of this mN mutant was comparable to WT Activin-A. In addition, analysis on reducing gels confirmed that only cleavage site 1 mutant (mS1) proActivin-A, but neither WT nor mN mutant A60 contained uncleaved βA chains (Fig. S3I). These data show that furin converts A70 to A60 independently of the N-glycosylation site. However, how furin accelerates the electrophoretic mobility of A60 compared to A70 remains to be determined.

Combined mutation of cysteines C35 and C38 inhibits the covalent linkage of proActivin-A to a high molecular weight complex in the ER

The amount of stably accumulating A60 relative to A30 varied considerably among independent experiments (Fig. S4A). We hypothesized that unpredictable fluctuations in the cell culture milieu could stabilize alternative allosteric disulfide bonds⁴². Activin-A prepro-protein contains nine conserved cysteines in the mature region, four in the prodomain, and one in the signal sequence (Fig. S4B, C). Covalent dimerization of the mature region by C390 is dispensable for the secretion and proteolytic maturation²⁹. By contrast, three intrachain disulfides that form the cystine knot are essential for proActivin-A folding in the ER, including C321-C391, C350-C354, and C423-C425²⁹. An additional intrachain disulfide between C314 and C322 in the mature region adjacent to the S1 cleavage site is predicted to localize within less than 20Å from cysteines C35 and C38 in a flexible loop of the prodomain (Figs. 4A and S4B). These four cysteines and the arrangement of C35 and C38 in a thioredoxin-like CXXC motif are conserved across species (Fig. S4C) and in all inhibin B isoforms, GDF8, and GDF11 (Fig. S4D)⁴³. To test the influence on Activin-A processing, we substituted each of these four cysteines with alanines. Transient transfection in FurKO#1 or DKO cells confirmed that none of them were essential for the secretion or furin-independent processing of proActivin-A (Fig. 4B, lanes 9-24). However, while mutating C38 exposed A70 to inhomogeneous PCLP cleavage (lanes 11, 19), mutating C35 or C322 markedly diminished A70 (lanes 10, 15, 18, 23). By contrast, mutation of C35 together with C38 restored the proportion of A70 (lanes 12, 20), suggesting that C35 stimulates A70 accumulation indirectly by regulating a function of C38. In particular, we noticed that the C35A and C38A mutants each formed a high molecular weight (HMW) form that was barely visible in SN of cells expressing WT βA (lanes 1, 9, 17 or, after prolonged exposure, 13, 21), and which was absent altogether if C35 and C38 were mutated jointly (lanes 12, 20). Analysis in lysates of cells treated with or without brefeldin A to inhibit ER export confirmed that C35 limits the amount of an Activin-A HMW complex that can be stabilized by C38, and vice versa (Fig. 4C). These results suggest that C35 regulates how much of Activin-A is (or remains) covalently linked to a hitherto unknown HMW complex via C38 or, if C38 is mutated, via C35 itself.

Open in a separate window

Fig. 4

The accumulation of mature Activin-A and A60 is regulated by allosteric disulfide bonds of C314 and C322 adjacent to site S1.

A Quaternary structure of proActivin-A (PDB: 5HLZ), with the mature domains colored ocher and prodomains in violet. Arginine R310 forming the scissile bond is highlighted in red, and thiols of cysteines C35 and C38 in yellow. B Anti-Activin-A Western blots of SNs and extracts (CL) from the B16-F1, FurKO#1, or DKO cell lines that were cultured for 48h after transfection with wild-type βA (WT) or with the indicated cysteine mutants. Proteins were denaturated with SDS and separated on a non-reducing gel to preserve disulfide bonds. Analysis of γ-tubulin in cell lysates served as a loading control. Data are representative of 2 experiments. Quantifications by Image J are shown below. C Western blot of SN and lysates of transfected B16-F1 cells after treatment with 5µg/ml brefeldin A for 12h. HMW, a high molecular weight form of Activin-A in the ER migrating with an apparent size of 118kDa. D Schematic illustration of allosteric disulfide bonds in Activin-A, and their proposed roles in HMW complex formation and S1 site cleavage during transport from the ER (top row) to the more oxidative trans-Golgi network (middle row) and extracellular space (bottom row). The prodomain (violet) and an interacting factor that is predicted to be covalently linked in a HMW complex are shaded in lighter color after they are cleaved off. Blue arrows indicate thiol exchanges predicted to be required to detach the mature region from its disulfide-linked prodomain (see text for details).

Covalent linkage of mature Activin-A to its cleaved prodomain in A60 primarily requires cysteines C35 and C322

A disulfide bond of C38 may form at least transiently to stabilize a HMW complex of proActivin-A with a rate-limiting interacting factor (Fig. 4D). A HMW complex of proTGF-β that is stabilized by the homologous cysteine is asymmetric³⁰. In one possible model, an analogous asymmetry in Activin-A precursor dimers enables the free C35 of the HMW-bound βA subunit to link to C322 in the mature region instead of pairing with C314 or C38 (Fig. 4D, top row). In this model, A60 accumulation reflects the perdurance of allosteric disulfide bonds of C35 with C314 or, if C314 is mutated, with C322. In keeping with this model, transient transfection in control B16-F1 cells revealed that the furin-cleaved A60 was absent upon mutating C35 alone or together with C38, but not after mutating only C38 (Fig. 4B, lanes 1–4). Importantly, the furin-cleaved A60 was also depleted or diminished, respectively, if we mutated C322 or C314 (Fig. 4B, lanes 6-8). These results suggest that covalent attachment of a cleaved prodomain to the mature region of A60 involves a disulfide bond of C35 with C322 on one βA subunit, or with C314 on the other.

Allosteric disulfide bonds in proActivin-A prevent its precocious furin-induced degradation

To rule out possible transfection artifacts, cysteine mutant βA constructs were transduced in B16-F1 cells as lentiviral transgenes. Screening by RT-qPCR identified stable cell lines expressing these constructs at comparable levels (Fig. S5A). Western blot analysis confirmed that both C35 and C38 had to be mutated to block the intracellular assembly of an Activin-A HMW complex. In addition, C38 partially compensated for the absence of C35 in stabilizing A60 in cell SN (Fig. S5B). Importantly, in line with our observation in transient transfections described above, the C₃₅C₃₈>AA mutation depleted A60 also in stable cell lines without inducing a corresponding increase of any other cleaved or uncleaved forms of Activin-A (Fig. S5B, lane 5). In sharp contrast, disrupting A60 by the alternative approach of mutating C314 and/or C322 led to a net increase in A30 (Fig. S5B, lanes 6-8). These data support our model that disulfide bond formation between C314 and C322 is inhibited or delayed by C35 and C38 to prevent precocious furin-induced Activin-A clearance (Fig. 4D).

To estimate the influence of disulfide bonds on Activin-A signaling in reporter cells, equal volumes of SNs from the melanoma cell lines shown in Fig. S5B were administered to HepG2 CAGA-Luc reporter cells with or without the furin inhibitor CMK (Fig. S5C, upper panel). Interestingly, Activin-A increased CAGA-Luc expression in CMK-treated reporter cells 30- to 40-fold regardless of any of the cysteine mutations. Purified mutant proteins are not available to compare their specific activity more accurately. However, normalization to the total signaling activity in reporter cells without CMK revealed that mutation of C314, C322, or both significantly diminished the fraction of Activin-A that remains to be processed by signal-receiving cells relative to how much it was activated by the signal-sending melanoma cells (Fig. S5C, bottom panel). These data suggest that covalent linkage of at least one prodomain to the mature region via an allosteric disulfide bond with C314 or C322 increases the reservoir of Activin-A that can be activated after secretion.

To investigate how the C₃₅C₃₈>AA mutation depleted Activin-A specifically in SN of furin-expressing cells, we considered whether it triggers ER-associated degradation. To test this, B16-F1 cells stably expressing WT or cysteine mutant forms of βA were treated for 12h with the proteasome inhibitor MG-132 with or without brefeldin A (BFA) that blocks ER to Golgi transport. As expected, cells treated with BFA alone accumulated uncleaved precursor in cell lysates instead of secreting it (Fig. S5D). In addition, BFA-treated cells accumulated traces of a 61kDa band of WT Activin-A (Fig. S5D, asterisk). Since this band migrated more slowly than A60 (58kDa) and independently of the cysteines that form A60, it may correspond to incompletely glycosylated hemicleaved Activin-A or a degradation product (Fig. 4D). Importantly, treatment with MG-132 alone or together with BFA neither stabilized A110 nor its HMW complex, regardless of the cysteine mutations examined. These results show that even the combined absence of C35 and C38 does not target Activin-A for ER-associated degradation of unfolded protein.

Hemicleaved Activin-A is partially resistant to follistatin inhibition

To test the impact of uncleaved βA subunits on physiological activin antagonists, Activin-A from B16F1-βΑ or FurKO-βΑ cells, respectively, was treated with increasing concentrations of FST, followed by incubation on HepG2 CAGA-Luc reporter cells. To avoid signal saturation, input SNs containing comparable amounts of mature or immature Activin-A, respectively, were each diluted 20-fold to match the activity of 10ng/ml of recombinant Activin-A (Fig. 5A, B). As expected, CAGA-Luc induction by the SN containing mature Activin-A was inhibited by FST no less than the induction by recombinant Activin-A. By contrast, FST only partially inhibited the signal induced by immature Activin-A from FurKO-βA cells even at the highest dosage examined (Fig. 5C, D). Western blot analysis of the SNs after their incubation on reporter cells validated that FST did not inhibit cleavage of A70 or A110 in the medium (Fig. 5E). These data suggest that the uncleaved prodomain confers partial resistance to inhibition by FST, and that furin-mediated cleavage during secretion sensitizes Activin-A more potently to this inhibitor than cleavage in trans by the signal-receiving cells.

Open in a separate window

Fig. 5

Hemicleaved Activin-A is partially refractory to inhibition by Follistatin.

A Anti-Activin-A Western blot of SNs from the indicated Furin wild-type (WT) and knockout (KO) #1 B16F1 cell lines used in (B–D). B CAGA-Luc induction in HepG2 reporter cells after 12h treatment with Activin-A (10ng/ml) (control), or with 1:10 diluted SNs of the indicated Furin WT or KO B16F1 cell lines expressing INHBA (βA). Error bars, SEM (n=5–6 independent experiments). *p<0.05, **p<0.01, ***p<0.001, ****p<0.0001, Student’s t-test. C CAGA-Luc induction in HepG2 reporter cells by 10ng/ml rActA (up), B16F1-βΑ SN (middle) or FurKO#1-βΑ SN (below) in the presence of increasing doses of FST (up to 500ng/ml). Error bars, SEM (n=3 independent experiments). IC50 (μM) values were calculated using nonlinear fit. D Estimation of the half maximal inhibitory concentration (IC50) of FST required to block Activin-A in the samples analyzed in (C). E Representative Western blot analysis of the βA-conditioned SNs shown in (A) after they were incubated for 12h on reporter cells with or without 500ng/ml FST.

Both βA subunits must be cleaved for Activin-A to bind and inhibit BMPR-II

Activin-A can also bind BMPR-II receptors, leading to SMAD1/5 signal attenuation^5,44 (Fig. 6A). To test whether hemicleaved Activin-A also binds to BMPR-II, cell SNs enriched for mature or hemicleaved Activin-A were incubated with Fc fusions of the extracellular domain of BMPR-II or, as a control, TGFBRII. We found that BMPRII-Fc pulled down both A30 and A60, albeit in low amounts (Fig. 6B, lanes 2, 5), in line with the fact that BMPR-II binds Activin-A with lower affinity than ActR-II^4,45. By contrast, binding of BMPRII-Fc to A70 was not detected (Fig. 6B, lane 3). To test the impact on BMP signaling, we supertransfected HepG2 CAGA-Luc cells with BMP-inducible BRE-Luc reporter and then treated them with the ALK4/5/7 inhibitor SB-431542 to block CAGA-Luc induction, followed by quantification of BRE-Luc induction by BMP receptor-activated SMADs (Fig. 6A). Addition of A30 or A70 dose-dependently inhibited the BRE-Luc induction by recombinant BMP-4 (Fig. 6C, top panel). Next, to prevent conversion of A70 to A60 during the assay, BRE-Luc was transfected into HepG2 CAGA-Luc cells expressing α1-PDX. Interestingly, in α1-PDX expressing reporter cells, A70 failed to inhibit BMP-4 signaling (Fig. 6C, bottom panel). These data corroborate the conclusion of our receptor pull-down assay that cleavage of both βA chains is necessary for Activin-A to interfere with BMP receptor signaling.

Open in a separate window

Fig. 6

Activin-A binding to type II receptors and its potentiation by furin, or by furin-independent inhibition of all cystine bonds of C314 and C322.

A Experimental strategy to assess which forms of Activin-A can inhibit BMP receptor signaling. HepG2 reporter cells with or without α1-PDX were transfected with the SMAD1/5 reporter BRE-Luc and treated with BMP4 alone or together with A110 in FurKO#3-βA cell SNs, or with A70 (FurKO#1-βA SN), or A30+A60 (B16F1-βA SN). Co-treatment with the activin type I receptor inhibitor SB-431542 (SB) served to prevent induction of CAGA-Luc. Created using BioRender.com. B Anti-Fc Western blot of Fc fusions of the type II receptors (left) that were used to pull down Activin-A from the concentrated cell SNs indicated on the right (input). C Induction of transiently transfected SMAD1,5 reporter BRE-Luc in HepG2 CAGA-Luc reporter cell lines with or without the α1-PDX transgene within 12h after stimulation with 10ng/ml BMP4, and its inhibition by 10ng/ml Activin-A (rActA) or by serially diluted SNs from B16F1-βΑ cells (A30) or FurKO#2-βΑ cells (A70), or FurKO-βΑ #3 (A110), or from B16F1-Ctrl cells (mock) cell lines. Cells were co-treated with the type I activin and TGF-β receptor inhibitor SB-431542 (SB, 10µM) to inhibit CAGA-Luc induction, or with empty vehicle. The sizes of colored circles indicate the 10- to 80-fold serial dilutions (from right to left). Induction was normalized to baseline luciferase values in untreated cells and then to treatment with BMP4 and SB-431542 (n=1–3 independent experiments). Error bars, SEM (ī=2–4 independent experiments); *p<0.05, **p<0.01, ***p<0.001, ****p<0.0001, ordinary one-way ANOVA with Šídák’s multiple comparisons test.

Furin converts hemicleaved Activin-A to A60 where one cleaved prodomain remains disulfide-linked to the mature region

Using LC–MS analysis and Western blots under reducing conditions, we confirmed that one βA subunit of the furin-independent Activin-A processing intermediate A70 was uncleaved. We named it A70 because hemicleaved Activin-A was first described as a protein of approximately 70kDa²⁷. Interestingly, further processing by furin after secretion converted A70 to A60, where the mature dimer remained covalently bound to the prodomain by a disulfide bond. This implies that the responsible disulfide bond did not abolish access of furin to either of the two βA subunits, even though it protected Activin-A against precocious furin-induced degradation in the secretory apparatus (Fig. 4D, box). Before testing the existence of such a disulfide bond, we considered whether A60 could correspond to uncleaved monomers. This scenario was ruled out by our findings that A60 was generated by furin and contained no uncleaved subunits on reducing gels. In TGF-β, a disulfide bond between cysteine C33 in the prodomain and a cysteine in the mature region delays cleavage of one subunit at least until after secretion^46,47. Cysteine C33 can also be connected to latent TGF-β binding protein (LTBP) in the ECM or to other cell type-specific transporters^48,49. Given the known stoichiometry of only one LTBP per TGF-β, this complex must be asymmetric, with C33 of one TGF-β subunit forming a different disulfide bond with LTBP than the other if both subunits are covalently attached^30,50. By contrast, the Activin-A prodomain to our knowledge has never been reported to form such disulfide bonds. Here, the apparent molecular weight of a fraction of proActivin-A in extracts of BFA-treated cells was shifted from 104 to 118kDa on non-reducing gels, depending on cysteine C38 or, in its absence, C35. Since the position of C38 corresponds to that of C33 in TGF-β1, the 118kDa form likely relies on an interacting factor that can at least transiently remain disulfide-linked to the prodomain. Reversible covalent attachment to C38, possibly followed by proteolytic breakdown would explain why such a factor has not been identified yet. The spacing and the sequence between C35 and C38 are reminiscent of the CXXC motif of thioredoxins that reduce oxidized cysteines in their substrates to mediate cleavage of disulfide bonds^51–53. Thus, the disulfide linkage of proActivin-A via one of these cysteines might be cleaved by a thiol attack from the other. In one possible model which is consistent with our data, the ability of the thiol of C35 to attack C38 is controlled by the connectivity of cysteines C314 and C322, which in turn is regulated by precursor cleavage and by its expected impact on their flexibility (Fig. 4D). Even a non-covalent asymmetric transient association with an interacting factor may change the conformation of one βA subunit of proActivin-A to allow cleavage by PCLP by promoting an allosteric disulfide bond of C35 with C322. On the other subunit, C35 is predicted to instead pair by default with C314, thereby reducing its flexibility and hence the accessibility of the scissile bond of this and the furin recognition motif. Importantly, this model also predicts that allosteric disulfide bonds among these cysteines will at least transiently link the Activin-A prodomain to the mature region.

About 40 allosteric disulfide bonds have been identified in various proteins, but existing technologies do not allow to determine their potential to undergo thiol exchanges in proteins containing as many cysteines as Activin-A⁴². The default connectivity of C35 and C38 cannot be inferred from structural models because these residues had to be removed to determine the existing crystal structure of proActivin-A. Consequently, C314 and C322 paired only with one another³⁹. We propose that C35 by default pairs first with cysteines in the mature region (Fig. 4D). This would explain why A60 was absent when we mutated cysteines C314 and C322, or C35. In stably transduced melanoma cells, C38 partially compensated for C35. The availability of C38 to form A60 may be limited by the expression levels of the interacting factor in the ER. Based on these considerations, we propose that HMW complexes of proActivin-A are stabilized primarily by C38. Accordingly, PCLP cleavage became inhomogeneous in the C38A mutant (Fig. 4B, lanes 11 and 19). In our model, absence of C38 destabilizes C314-C322 disulfides by liberating C35 to attack C322. Alternative mechanisms to account for the regulation of stepwise cleavage of the two βA subunits by furin and PCLP cannot be excluded at this stage. However, they too would have to consider that mutation of one cysteine will influence the connectivity among others. For example, loss of C35 or C38 alone stabilizes traces of HMW complexes even in conditioned medium, as expected if C38 can be attacked by C35 to release a covalently bound factor. Allosteric disulfides may also modulate the fate of the prodomain, including possible alternative and degradative cleavages (Fig. 4D).

Disruption of allosteric disulfide bonds exposes Activin-A to precocious furin-induced depletion

Mutation of C33 impairs TGF-β1 activity in vivo, leading to multiorgan inflammation and digestive tract cancers⁵⁴. Here, we found that combined mutation of C35 and C38 exposed all forms of Activin-A to precocious furin-induced depletion. Conversely, mutating C314 and/or C322 increased the net accumulation of secreted A30 (Fig. S4B, lanes 6-8). Interestingly, addition of mutant forms to HepG2 reporter cells showed no proportionate increase in signal activity, but rather a decrease in the reservoir of total Activin-A activity that can be activated after secretion. These findings suggest that disulfide bonds among these cysteines serve to restrain furin-induced turnover of immature forms and the predicted associated flux of Activin-A-interacting factors. Previously, the influence of C314 or C322 on Activin-A signaling has been tested by mutating them individually. When added to rat pituitary cells, these mutants labeled fewer cell surface receptors and signaled three- to fivefold less potently than wild-type Activin-A, suggesting that C314 and C322 stabilize ligand/receptor complexes²⁹. Our structural modeling and functional analysis revealed that these cysteines inhibit receptor binding rather than stimulating it (see below), indicating that they likely attenuate receptor-mediated turnover. In keeping with this interpretation, furin cleavage also facilitates endocytosis of the related Nodal precursor^34,55. Future studies should test whether precocious cleavage diverts Activin-A to endosomes or lysosomes. In line with a possible role for endosomal furin, blocking furin in endosomes requires a 10-fold higher concentration of the membrane-permeable inhibitor CMK than in the TGN or at the cell surface³², and a similar dosage effect of CMK was observed here when comparing furin-mediated cleavage of one βA subunit compared to the other. As a possible alternative, we considered if the cysteines mediating allosteric disulfide bond formation are important to prevent aberrant Activin-A folding and degradation in the ER. Contrary to this hypothesis, proActivin-A degradation in the ER appeared to be minimal, concurring with the fact that furin normally is inhibited in the ER by its prosegment.

Control of juxtracrine versus paracrine Activin-A signaling by stepwise precursor cleavage

Hemicleavage of Activin-A, defined here as the processing of one βA subunit of the precursor dimer at the S1 site, enabled significant binding to the extracellular domains of type IIA and IIB activin receptors both in plasmonic resonance measurements and in pull-down assays. In proActivin-A, a non-covalent dimer of the prodomain occupies the predicted type I receptor binding surface and several type II receptor-binding epitopes³⁹. The same study confirmed that mature Activin-A remains bound to its prodomains also after precursor cleavage and secretion. However, cognate signaling receptors bind mature Activin-A with higher affinity and readily extract it from this non-covalent complex^39,56–58. Our in vitro pull-down experiments suggest that activin type II receptors can also compete away the prodomain if one βA subunit of Activin-A is still uncleaved. By contrast, binding to BMPR-II was only detected after cleavage of both βA subunits, in keeping with its approximately 100-fold lower affinity for Activin-A compared to ActR-II^4,45.

An available proActivin-A crystal structure did not explain how furin cleavage facilitates type II receptor binding, possibly because C35 and C38 were removed to ensure that C314 and C322 can only pair with one another³⁹. Indeed, the so-called lasso region adjacent to C35 and C38 of the Activin-A prodomain is known to fold back onto the mature region of the same βA subunit and thus has been hypothesized to help prevent precocious binding of βA monomers to ActR-II already in the ER. We propose that this protective function of the lasso in proActivin-A requires a disulfide bond that links the mature region to C35, but which later can be resolved by a thiol exchange with the nearby C38 or C314. To trigger such an attack, this model predicts that at least the correct positioning of C314 requires prior proteolytic processing at the nearby furin motif for the necessary conformational change to occur (Fig. 4D). In line with this model, we here found that binding of the cleavage mutant proActivin-A to Fc fusions of cognate type II receptors in pull-down assays was inhibited by cysteines C314 and C322. This strongly suggests that before pairing with one another, the cysteines C314 and C322 first form alternative allosteric disulfide bonds that are required to prevent precocious type II receptor binding in the secretory apparatus.

Treatment of various cell types with purified proActivin-A potently stimulates SMAD2/3 signaling⁵⁸. We found that this involves cell non-autonomous proteolytic processing, and that signal-receiving cells also converted hemicleaved Activin-A to mature form. Without further maturation, hemicleaved Activin-A was inactive in the signal-receiving HepG2 cells, except if the signal-sending melanoma cells were in direct contact. Cell contacts can mechanically unmask receptor-binding sites in latent complexes of TGF-β through integrin α_vβ6 or α_vβ8 pulling forces^59–61. While proActivin-A contains no known integrin binding site, exposure of receptor-binding epitopes by proteolytic processing of the first βA subunit may lead to a receptor-induced conformational change on the uncleaved subunit in a cell contact-dependent manner. Alternatively, cell contacts may facilitate access to a cell surface-bound protease. The cell-non-autonomous precursor processing was blocked by a membrane-impermeable PC inhibitor, suggesting it happens extracellularly. During early embryogenesis, both furin and the related protease PACE4 function cell non-autonomously^62,63. Therefore, and since furin is expressed in HepG2 cells³⁷, we did not further investigate how cell non-autonomous cleavage in trans is mediated.

RT-qPCR analysis

Cells were cultured in 6-well plates and total RNA was extracted using the ReliaPrep RNA Cell Miniprep System (Promega, Madison, WI, USA) following the manufacturer’s protocol. For total RNA extractions from tissues, snap-frozen pieces were sonicated or homogenized in 1mL of QIAzol (Qiagen, Hilden, DE), followed by chloroform extraction. Total RNA was isolated using RNeasy mini kit (Qiagen) according to the manufacturer’s protocol. cDNA was synthesized from 1μg RNA using the PrimeScript RT-PCR Kit (Takara, Kusatsu, JP). The analysis was performed using the SYBR Green GoTaq Master Mix (Promega) on a QuantStudio 6 instrument (Applied Biosystems, Waltham, MA, USA). qPCR primers are listed in Supplementary Table 1.

Expression vectors and cloning

Lentiviral vector pLenti-EF1alpha-ActA-SV40-puro expressing myc-tagged proActivin-A has been described¹⁵. Briefly, nucleotides 218-1609 of the human INHBA reference sequence NM_002192.2 containing a myc epitope (EQKLISEEDL) instead of the glutamic acid residue E311 near the N-terminus of the mature region was subcloned as a Spe I fragment into the Xba I site of pLenti-EF1alpha-MCS-SV40-puro vector. To derive the cleavage site 1 mutant (mS1), the myc-INHBA coding sequence was amplified by overlap extension PCR (OE-PCR) using the forward and reverse primers 5′-GAGATCTAGACTCGAGACGCAAGGC-3′ and 5′-GAGAATCGATACCGTCGACT AGAG-3′, together with the mutagenic internal primers listed in Supplementary Table S2 to thereby replace the wild-type cDNA in pLenti-EF1alpha-ActA-SV40-puro. The KKR71>AAA and RR110>AA mutant constructs mS2′ and mS2, as well as the N-glycosylation site mutant N165>A were generated analogously using the indicated internal OE-PCR primers (Supplementary Table S2). To tag the prodomain with a Flag epitope after E27, G130, or D146, the same external OE-PCR primers were used together with the internal primers encoding the amino acid sequence DYKDDDDK. To substitute cysteines 35, 38, 314, and/or 322 in proActivin-A with alanines, the internal mutagenic OE-PCR primers instead encoded the desired mutations (Supplementary Table S2). All mutant constructs were validated by Sanger sequencing of both cDNA strands.

Lentiviral vector expressing the bicistronic α1-PDX-IRES-GFP cassette was generated by subcloning α1-PDX cDNA (gift of Dr. Gary Thomas, University of Oregon) into attP1 and attP2 sites of pDONR221 vector (Addgene), and from there into the gateway destination vector pDEST between the human EF1α promoter and an IRES-eGFP cassette (Dest-vector IRESeGFP, gift of Dr. Jörg Huelsken, EPFL).

Collection of cell SNs after lentiviral transduction or transient transfection

CRISPR-edited FurKO clones #1 to #3 and a Furin and Pcsk7 double knockout (DKO) clone³² were transduced with βA or empty control lentivirus as described previously for B16F1-Ctrl and -βΑ cell lines¹⁵. In short, HEK293T cells were co-transfected with CMVΔR8.74 (Addgene, Watertown, MA, USA 22036), pMD2.VSVg (Addgene 12259), mycINHBA, or empty transfer plasmid. Lentiviral particles were collected from filtered culture supernatant by ultracentrifugation and resuspended in 50µL sterile PBS and stored at −80°C or used freshly at a 50-500 dilution to infect cells of interest that were plated at a density of 1 × 10⁵ cells/well in 12-well plates. Non-clonal B16-F1 cell lines stably expressing wild-type or the indicated mutant βA transgenes or empty vector control (Ctrl) were selected using puromycin. FurKO-βA and DKO-βA cell lines and respective empty vector controls were not selected because the parental CRISPR-edited FurKO cell line was already resistant to available antibiotics. Instead, individual stably transduced clones were plated at limiting dilution and expanded. Activin-A secretion was assessed by incubating the SNs on CAGA-Luc reporter cells confirming activity in 21 of 22 clones analyzed. Lentiviral α1-PDX particles to transduce HepG2 CAGA-Luc reporter cells were produced by transfecting HEK293T cells with α1-PDX-IRES-GFP together with the packaging vectors CMVΔR8.74 and pMD2.VSVg. The expression of α1-PDX was validated by flow cytometric sorting of the GFP⁺ cells.

For transient transfection, B16-F1, FurKO, or DKO cells were seeded in 24-well or 6-well plates at a density of 2.5×10⁵ or 5×10⁵ cells/well, respectively. On the following day, the cells were rinsed with PBS and incubated for 4h in OptiMEM medium (Invitrogen) containing Lipofectamine 3000 (Invitrogen) together with the indicated expression vectors (1µg of plasmid per well in 24-well plates, or 2µg per well in 6-well plates). After transfection, the cells were rinsed with PBS and cultured for 48–72h in complete DMEM. Alternatively, to increase the sensitivity of Activin-A Western blots on non-reducing gels, SNs were collected from cells after 48–72h in fresh OptiMEM.

Western blot analysis

To monitor the expression and processing of Activin-A, SNs were prepared as described above either in complete DMEM for analysis on non-reducing gels, or without fetal bovine serum in OptiMEM for reducing gels. Cell lysates were obtained by treating adherent cells with trypsin/EDTA and resuspending them on ice in RIPA buffer (50mM Tris pH 8.0, 150mM NaCl, 0.1% SDS, 0.5% Na-Deoxycholate, 1% NP-40, supplemented with protease inhibitor cocktail (A32953, Roche). Conditioned media were cleared of debris by centrifugation for 5min at 5000 × g, followed by the addition of ice-cold acetone to precipitate proteins that were pelleted by centrifugation, air-dried, and resuspended in cell extraction buffer PBS containing 1mM EDTA, 0.5% Triton-X100 and protease inhibitor cocktail. For loading, samples were boiled in Laemmli buffer or, for analysis on reducing gels, in Laemmli buffer containing 50mM DTT. After size fractionation on 9% SDS-PAGE gels, proteins were transferred to nitrocellulose membranes using the Trans-Blot Turbo transfer system (BioRad). Membranes were blocked using 5% skim milk (Sigma) in PBS containing 0.1% Tween-20 before incubation overnight at 4°C or 2h at room temperature with polyclonal goat anti-Activin-A antibody (AF388, R&D Systems) or, if proteins were separated on non-reducing gels, with mouse monoclonal anti-Activin-A (1:500, Abcam ab89307), anti-myc (1:1000, Santa Cruz biotechnology sc40, clone 9e10), or anti-Flag M2 antibody (1:1000, Sigma F1804) in the same buffer. Alternatively, to detect Activin-A in immunoprecipitates or after receptor-mediated pull-down, or in tumor lysates and in plasma of tumor-bearing mice, Western blots were probed with biotinylated anti-Activin-A antibody (BAM3381, R&D Systems). After washing three times in Tris-buffered saline containing 0.1% Tween-20 to remove unbound primary antibodies, blots were incubated with the HRP-coupled secondary antibodies sheep anti-mouse (GE Healthcare), or with streptavidin-HRP labeled secondary antibody (016-030-084, Jackson Immunosearch). HRP activity was detected using the ChemiDoc imaging system (BioRad) with Super-ECL reagent to reveal Activin-A, or with ECL for all other Western blots (ThermoFisher Scientific). While all effects of furin and of conserved cysteines on Activin-A processing reported in this study were highly reproducible, the actual proportion of A60 and A30 detected in SNs of βA expressing cells fluctuated among different series of experiments conducted over the course of several years. Possible factors contributing to this variability include the cell culture medium (DMEM versus OptiMEM), duration of the medium conditioning, concentration of the sample and amount of protein loaded for SDS-PAGE, as well as the availability and use of different primary antibodies and mode of chemiluminescence detection. However, none of the conclusions of this study require that absolute amounts of stably accumulating A30 or A60 or their proportion have to be invariable across independent series of experiments.

CAGA-Luciferase assay

3 × 10⁴ HepG2 CAGA-Luc reporter cells were seeded into 96-well plates in complete DMEM and incubated for 12h with fresh medium containing 20ng/mL recombinant ActA (R&D Systems), or with 3- to 10-fold diluted control melanoma cell SNs, which induces CAGA-Luc expression within its dynamic range (20- to 200-fold increase above baseline) without reaching signal saturation. Where indicated, the reporter cells were co-treated with the pan-PC inhibitor CMK (100µM) or empty vehicle (H₂O). Cells were lysed in 50μL potassium buffer (10mM) containing 0.2% Triton-X100. Five microliters of each lysate were transferred into white Nunc 96-well plates. After adding 50μLP/R A (Firefly) and P/R B (Renilla) reagents⁶⁸, Firefly and Renilla luciferase luminescence were measured using a Centro XS³ LB960 luminometer (Berthold Technologies). Firefly luciferase measurements were normalized to Renilla values in each well on two technical replicates per experiment for each sample.

To monitor the inhibition of hemicleaved Activin-A by recombinant FST, SNs from 1 × 10⁶ B16F1-βA or FurKO-βA cells were collected after 72h and cleared of debris by centrifugation. Titration on CAGA-Luc reporter cells determined that SNs of FurKO#1-βA or control B16F1-βA had to be diluted 10-fold to match the induction of luciferase expression by 10ng/ml of recombinant Activin-A. To compare the IC50 of the inhibition of mature versus hemicleaved Activin-A by specific inhibitors, 10× diluted SNs were incubated on HepG2 CAGA-Luc reporter cells in 96-well plates after addition of 1–500ng/ml of FST-300 (R&D Systems).

To assess BMP receptor signaling, HepG2 CAGA-Luc reporter cells were transfected with the BMP-inducible SMAD1/5 luciferase reporter BRE-Luc⁶⁹. Briefly, one day after plating, HepG2 CAGA-Luc cells at a density of 5x10⁵ cells/well in 6-well dishes were incubated for 4h in OptiMEM containing 2µg of BRE-Luc in Lipofectamine 3000, followed by overnight culture in complete DMEM. The following day, cells were collected by trypsin/EDTA treatment and re-seeded in 96-well plates and incubated for 12h with 10ng/ml of recombinant BMP-4 (Biotechne) or Activin-A, or with 10- to 80-fold diluted B16F1- or FurKO-βA SN from 1 × 10⁶ cells that had been cultured for 72h in complete DMEM. In addition, the reporter cells were co-treated with 100μM CMK and, where indicated, with 10μM SB-431542 (S4317, Sigma-Aldrich).

For co-culture experiments, 5 × 10⁴ B16F1 FurKO#1-βA cells were seeded in the upper or lower chamber of 24-well plates with 0.4µm transwell inserts (Millicell). The following day, 4 × 10⁴ HepG2-CAGA reporter cells were seeded in the other compartment. After cell lysis in 100µL 50 μL10 mM potassium buffer containing 0.2% Triton-X100, 5µL of each extract was used to measure Firefly and Renilla luciferase luminescence.

Analysis of ER-associated degradation

To assess ERAD, 2 × 10⁵ cells were cultured for 12h in 24-well plates in complete DMEM and treated with 5µg/mL (17.83µM) BFA (LubioScience, S7046) and/or 10µM MG-132 (Sigma-Aldrich, C2211), or empty vehicle control. Effects of BFA and MG-132 on the accumulation of Activin-A in cell lysates and SNs were analyzed by Western blot.

Receptor-mediated pull-down

Lentivirally transduced HEK293T cell lines secreting AIIB-Fc, AIIA-Fc, TGFBRII-Fc, or BMPRII-Fc³⁶, were cultured at a density of 1 × 10⁶ cells/plate in 10cm tissue culture dishes during 72h in complete DMEM. SNs were collected, cleared from cell debris by centrifugation, and concentrated 10-fold using 4kDa cut-off ultrafiltration tubes (Merk Millipore). SNs containing the indicated Fc receptor fusion proteins were loaded on Pierce magnetic protein A/G beads (88802, Thermo Ficher Scientific) by overnight incubation at 4°C on a rotating wheel. The next day, beads and bound receptors were washed twice and incubated with 10-fold concentrated SNs of Ctrl- or βA-transduced B16-F1 and FurKO cells on a rotor at 4°C for 4h. Alternatively, the protein A/G beads were incubated with concentrated SNs from transiently transfected B16-F1 and FurKO cells expressing E27 Flag-tagged βA, or GFP (control). After magnetic pull-down, beads were washed three times in buffer containing 30nM Tris pH 7.4, 150nM NaCl and 0.05% Nonidet P-40 to remove unbound proteins, resuspended in 50µL of 4× Laemmli buffer and boiled before loading on a non-reducing 9% SDS-PAGE gel for Activin-A Western blot analysis.

Liquid chromatography and tandem mass spectrometry (LC–MS/MS)

Samples for LC–MS/MS analysis were separated on preparative non-reducing 9% SDS-polyacrylamide gels that were stained by colloidal coomassie blue (1610803, Bio-Rad) overnight and extensively washed in Milly-Q H₂O for 1h. The indicated gel pieces were excised and washed twice with 50% ethanol in 50mM ammonium bicarbonate (AB, Sigma-Aldrich) for 20min and dried by vacuum centrifugation. Proteins were reduced with 10mM dithioerythritol (Merck-Millipore) for 1h at 56°C, washed and dried as described above, before alkylation with 55mM Iodoacetamide (Sigma-Aldrich) for 45min at 37°C in the dark. After another wash-and-dry cycle, proteins were digested overnight at 37°C using MS Grade Trypsin or Chymotrypsin (Pierce) at a concentration of 12.5ng/µl in 50mM ammonium bicarbonate supplemented with 10mM CaCl₂. The resulting peptides were extracted for 20min twice in 70% ethanol containing 5% formic acid (Merck-Millipore), dried by vacuum centrifugation, and finally desalted on C18 Stage Tips (Pierce) as described⁷⁰. For LC–MS/MS analysis, each sample was resuspended in 2% acetonitrile; 0.1% formic acid and nano-flow separations were performed on a Dionex Ultimate 3000 RSLC nano UPLC system connected online with an Exploris 480 Orbitrap Mass Spectrometer. A capillary precolumn (Acclaim Pepmap C18; 3 μm-100 Å; 2cm × 75 μm internal diameter) was used for sample trapping and cleaning. Analytical separations were performed at 250 nL/min over a 90min biphasic gradient on a 50cm long in-house packed capillary column (75 μm ID; ReproSil-Pur C18-AQ 1.9 μm silica beads). Acquisitions were performed through Top Speed Data-Dependent acquisition mode with 2seconds cycle time and using an inclusion list of expected masses of human INHBA chain (Uniprot entry sequence: P08476). Initial MS scans were acquired at a resolution of 120,000 (200m/z). The most intense parent ions were selected and fragmented by High energy Collision Dissociation (HCD) with a Normalized Collision Energy (NCE) of 30%, using an isolation window of 1.7m/z. Fragmented ion scans were acquired with a resolution of 30,000 (200m/z). Selected ions were then excluded for the following 20s. The resulting raw data were processed using SequestHT, Mascot and MS Fragger in Proteome Discoverer v.2.5 against the Uniprot Human reference proteome database (i.e., 79,052 sequences). Enzyme specificity was set to Trypsin or Chymotrypsin. Peptides were identified using a minimum of six amino acids. Up to two missed cleavages were allowed, and a 1% FDR cut-off was applied both at peptide and protein identification levels. For the database search, oxidation (M), phosphorylation (S, T, Y), acetylation (N-term Prot) and Gln to pyro Glu were considered as variable modifications, whereas carbamidomethylation (C) was set as invariable. Final data visualization used Scaffold Viewer (v. 5).

Fabrication and functionalization of the plasmonic biosensor

Ctrl-, βA-, or mS1-transduced B16-F1 and FurKO cell lines were cultured at a density of 1 × 10⁶ cells/58 cm² for 72h in OptiMEM. Aliquots of SNs were acetone precipitated and analyzed on non-reducing gels by anti-Activin-A Western blot as described above. For plasmon resonance measurements, gold nanohole array plasmonic chips fabricated at EPFL^40,41 were functionalized by precoating with anti-Activin-A antibody (BAM050, R&D Systems), or with AIIB-Fc (339-RBB-100). Plasmonic chips were manufactured using deep ultraviolet (DUV) lithography to achieve a wafer-scale and robust fabrication. In brief, after Radio Corporation of America (RCA) cleaning of a fused silica wafer, an e-beam evaporator (Alliance Concept EVA 760) was used to deposit Ti (10nm thickness) and Au (120nm thickness) on the wafer. Thereafter, the wafer was coated with a photoresist, and an array of nanoholes was patterned on the surface using a DUV stepper (ASML PAS5500/300). The pattern was developed and transferred to the Au and Ti layers by ion beam etching (Oxford Instruments PlasmaLab 300), followed by a final resist removal step using oxygen plasma.

For coating with AIIB-Fc, the plasmonic biosensor surface was pre-cleaned by sequential washing with acetone, isopropanol, and Milly-Q water (5min each) and treated with a UV-ozone cleaner for 20min, followed by immersion in a 1:4 mixture of COOH/OH-functional PEGylated alkanethiols (ProChimia) in absolute ethanol that coat the surface with a conventional self-assembled monolayer (SAM). After overnight incubation, the carboxylic group on the chip surface was activated by incubation for 20min at room temperature (RT) in MES buffer (pH 6) containing 200 mM N-3-dimethylaminopropyl-N′-ethylcarbodiimide-hydrochloride and 50mM sulfo-N-hydroxysuccinimide and then immersed with protein A/G (50µg/mL, Thermo Fisher Scientific) in acetate buffer (pH 4.5) for 2h at RT. Unreacted carboxyl groups in the resulting covalently linked biorecognition layer were inactivated by immersion in 1M ethanolamine solution in Milli-Q water for 5min at RT. Thereafter, the surface of the chip was loaded with recombinant AIIB-Fc receptor (50µg/mL, R&D Systems, 339-RBB-100) for 2h at RT on a thermomixer, followed by incubation at RT for 1h with 1% bovine serum albumin (BSA) as a blocking agent to reduce non-specific protein binding.

For coating with Activin-A antibody, the biosensor surface was first pre-cleaned and UV treated as described above. A SAM layer was formed by immersing the chip in biotin/OH-functional PEGylated alkanethiols (ProChimia) at a molar ratio of 1:9 (2mM final concentration) in absolute ethanol overnight at RT. Thereafter, the surface was washed for 5min with ethanol three times and once with Milly-Q water, dried using pressurized nitrogen, and then incubated with streptavidin in PBS for 2h at RT. After washing three times with PBS for 5min each to remove unbound streptavidin, the surface was incubated with biotinylated Activin-A antibody (40µg/mL) in PBS for 2h at RT on a thermomixer. After washing the surface three times with PBS (5min each), it was incubated at RT for 1h with 1% BSA solution as a blocking agent.

Monitoring of surface plasmon resonance by nanoplasmonic platform

After immobilizing the desired capturing agents on the surface, the plasmonic biosensor was mounted in a polydimethylsiloxane (PDMS) microfluidic device (microchannels of 500 μm width and 180 μm height) and connected to a syringe pump for sample injection. Samples were injected into microchannels over the surface at a flow rate of 5µL/min. For each measurement, one channel was used for the sample and one for referencing with a constant flow of PBS. Ligand binding was monitored on an imaging-based nanoplasmonic platform comprising a gold nanohole array biosensor, a narrow-band illumination source, a camera, and a microfluidic device⁴¹. In brief, after sample injections, interactions between Activin-A and the capturing agents changed the refractive index on the biosensor surface leading to redshifts in the chip transmission spectrum. The camera translated such spectral variations into intensity drops over time. The platform had a large dynamic range covering up to 12,000 arbitrary units for intensity changes without requiring prism-coupling configurations. Signals from three different regions of interest in the field of view were measured every 30s and averaged to obtain the final sensograms.

Syngeneic grafting and measurement of tumor growth

B16-F1, FurKO, or DKO cells stably expressing βA or empty control vector were injected intradermally either into the right flank of 8–12-week-old female C57BL/6J or Rag1^-/- mice (Charles River laboratories) or on both sides in contralateral grafts (10⁵ cells each). Tumors were measured with a caliper every second day. Tumor volumes were calculated using the formula V = [1.58π × (length × width)^3/2]/6⁷¹.

Quantification of Activin-A plasma levels by ELISA

To determine the influence of proteolytic maturation on the reactivity of Activin-A in ELISA assays, SNs from B16F1-βA or FurKO-βA cell were analyzed before and after treatment with recombinant furin. In brief, 3μL (6 units) of soluble furin (P8077L, Biolabs) were incubated for 5h at 37°C with 300μL aliquots of SNs from 1 × 10⁶ cells that had been cultured for 72h in complete DMEM. Thereafter, furin-treated and control SNs were analyzed using an anti-Activin-A ELISA kit (R&D Systems, DAC00B). To quantify circulating levels of mature Activin-A in tumor-bearing hosts, their blood was collected at the end of the experiment in heparin-coated tubes (Microvette 500 LH, SARSTEDT AG & Co.). Plasma was separated by centrifugation and aliquoted into new Eppendorf tubes and analyzed by ELISA.

Molecular modeling

To generate a model of proActivin-A dimer including all protein segments, the structure model of full-length proActivin-A was retrieved from the AlphaFold database⁷² (AF-A4D1W7-F1-model_v4_human Activin beta-A chain). After duplication, two single chains of the proActivin-A model were docked using the PyMOL program (Schrödinger). Briefly, the crystal structure of the mature Activin homodimer³⁹ (PDB: 5HLZ) was used as a template to align the mature domains from the proActivin-A models in a homodimeric fashion. The resulting proActivin-A dimer model was subjected to an additional step of energy minimization based on discrete molecular dynamics using the automated Chiron protocol provided on the Dokhlab webserver⁷³ (https://dokhlab.med.psu.edu/chiron/login.php). To generate a model of an artificial heterodimer GDF11:Activin-A bound to ActR-II and ALK5 receptor heterodimer, the known crystal structure of human GDF11 bound to ActR-II and ALK5 (PDB: 6MAC)⁷⁴ was aligned with one mature subunit (residues 311-426) of our proActivin-A dimer model (Fig. S6A). After alignment, the mature subunit superimposed with GDF11, as well as the two prodomains were removed (Fig. S6B). For all the figures, the final views were rendered on PyMOL.

Statistical analysis

In all experiments that were repeated at least three time, statistical tests were performed using the Prism software (GraphPad) to show the mean±SEM or SD as indicated in the figure legends. When comparing two groups, normal distributions were analyzed by the Shapiro-Wilk normality test. P-values were calculated by Student’s t-test (normal distribution) or Mann–Whitney’s test (non-parametric test). One-way ANOVA was used to compare several groups of unpaired values. Data points identified as outliers by the regression and outlier (ROUT) removal method in Prism 9 with a false discovery rate ≤1% were excluded. Power Analysis was waved by the animal experimentation authorities due to pre-existing data about the effect sizes of Activin-A on tumor growth and on the associated loss of body weight. Tumor volumes at the endpoint were compared by ANOVA or Student’s t-test, as indicated in the figure legends.

The authors are grateful to Drs. Diego Chiappe and Maria Pavlou of the EPFL Proteomics core facility for their support with the analysis and identification of Activin-A processing intermediates. We also would like to thank the members of the Center of PhenoGenomics for the animal housing. This work was supported by grants 31003A_179330 of the Swiss National Science foundation and KFS-4454-02-2018 from the Swiss Cancer League to DBC, and by services of the Center of PhenoGenomics and Flow Cytometry Research Core Facilities at the School of Life Sciences of EPFL.