Structural modeling defines transmembrane residues in ADAM17 that are crucial for Rhbdf2–ADAM17-dependent proteolysis

The cell surface metalloproteinase a disintegrin and metalloproteinase 17 (ADAM17) regulates signaling through the epidermal growth factor receptor (EGFR) by controlling the ectodomain shedding of several membrane-anchored EGFR ligands, such as transforming growth factor α (TGFα) and heparin-binding (HB)-EGF (Blobel, 2005; Jackson et al., 2003; Peschon et al., 1998; Sternlicht et al., 2005). In addition, ADAM17 is responsible for the release of the pro-inflammatory cytokine tumor necrosis factor α (TNFα, also known as TNF) from cells (Black et al., 1997; Horiuchi et al., 2007a; Moss et al., 1997), and is involved in the proteolytic processing of a variety of other membrane proteins (Weber and Saftig, 2012). A major function of the EGFR is to control the skin and intestinal barriers, and therefore patients lacking the EGFR or ADAM17 suffer from barrier defects, which are also the major side effects of pharmacological inhibition of EGFR signaling (Blaydon et al., 2011; Franzke et al., 2012; Lichtenberger et al., 2013; Campbell et al., 2014). ADAM17-dependent EGFR signaling can be rapidly activated through a variety of signaling pathways, raising questions about the underlying mechanism (Horiuchi et al., 2007b; Le Gall et al., 2009, 2010; Maretzky et al., 2011a, 2011b; Sahin et al., 2004; Gschwind et al., 2003; Inoue et al., 2012; Myers et al., 2009). Structure–function experiments have uncovered a role for the transmembrane domain (TMD) of ADAM17 in regulating its rapid activation (Le Gall et al., 2010), whereas the cytoplasmic domain was found to be dispensable for this purpose (Doedens et al., 2003; Le Gall et al., 2010; Schwarz et al., 2013).

Recently, the inactive rhomboid-like proteins Rhbdf1 and Rhbdf2, which have seven-membrane-spanning domains, have emerged as key regulators of ADAM17 (Adrain et al., 2012; Christova et al., 2013; Li et al., 2015; Maretzky et al., 2013; McIlwain et al., 2012; Siggs et al., 2012). Rhbdf2 is crucial for the maturation and function of ADAM17 in myeloid cells, and therefore mice lacking Rhbdf2 are protected from ADAM17- and TNFα-dependent septic shock and inflammatory arthritis (Issuree et al., 2013; McIlwain et al., 2012). Interestingly, myeloid cells isolated from mice carrying a conservative point mutation (termed sinecure), in the first TMD (TMD1) of Rhbdf2, also have a defect in the release of TNFα that is very similar to mice lacking Rhbdf2 (Rhbdf2^−/−) (Siggs et al., 2012). Mice lacking Rhbdf2, or homozygous sinecure mutant mice (Rhbdf2^sin/sin), appear normal because the related Rhbdf1 has compensatory or redundant functions in other tissues besides immune cells (Li et al., 2015; Christova et al., 2013). Therefore, inactivation of both Rhbdf1 and Rhbdf2 prevents the maturation and function of ADAM17 in all tissues, and results in pre-natal or perinatal lethality, depending on the source of the Rhbdf1-deficient mouse strain (Li et al., 2015; Christova et al., 2013).

Much remains to be learned about how Rhbdf1 and Rhbdf2 control the maturation and regulation of ADAM17 (Adrain et al., 2012; Christova et al., 2013; Li et al., 2015; Maretzky et al., 2013; McIlwain et al., 2012). Since the rapid activation of ADAM17 requires its TMD (Le Gall et al., 2010), and since the release of the ADAM17 substrate TNFα from myeloid cells was strongly reduced through the sinecure point mutation in the TMD1 of Rhbdf2 (Siggs et al., 2012), we hypothesized that an interaction between the TMD1 of Rhbdf2 and the TMD of ADAM17 is responsible, at least in part, for the regulation of ADAM17. Here, we explored the functional properties of the Rhbdf2 sinecure mutant in immune cells, and made the conclusions more generally applicable by characterizing Rhbdf2^sin/sin mice that also lacked Rhbdf1. To predict potential modes of interactions between ADAM17 and Rhbdf2 we used molecular modeling and all-atoms molecular dynamic (MD) simulations, and the resulting models provided the basis for structure–function studies in cell-based assays to determine how mutations at the predicted interaction sites in ADAM17 affect its function. Our results support the hypothesis that an interaction between the TMD1 of Rhbdf2 and the TMD of ADAM17 is crucial for the regulation of Rhbdf2–ADAM17-dependent shedding events.

Previous studies had shown that the Rhbdf2 sinecure mutation (Rhbdf2^sin/sin) affects the release of TNFα from myeloid cells (Siggs et al., 2012), raising the question of whether this intramembrane point mutation impacts the maturation and function of the principal TNFα convertase ADAM17. We found strongly reduced levels of mature ADAM17 in western blots of bone marrow (BM) and BM-derived macrophages (BMDMs) from Rhbdf2^sin/sin mice, similar to BMDM lacking Rhbdf2, where mature ADAM17 was not detectable (Fig. 1A, quantification in Fig. S1A). When Rhbdf2^sin/sin BMDMs were stimulated with 1 ng/ml lipopolysaccharide (LPS) to promote the production and release of TNFα, significantly less soluble TNFα was detected in the culture supernatant compared to wild-type (WT) controls (Fig. 1B), similar to a previous report for Rhbdf2^sin/sin BMDMs (Siggs et al., 2012). Stimulation with 10 ng/ml LPS elicited substantially more TNFα shedding from Rhbdf2^sin/sin BMDMs, approaching the levels seen in WT BMDMs, whereas TNFα shedding from Rhbdf2^−/− BMDMs remained almost completely abolished (Fig. 1B).

As an independent indicator for the function of ADAM17 in Rhbdf2^sin/sin BMDMs, we monitored levels of the ADAM17 substrate macrophage colony stimulating factor receptor (MCSFR, also referred to as colony stimulating factor 1 receptor, CSF1R) on the cell surface (Becker et al., 2015; Qing et al., 2016). We found that MCSFR levels on unstimulated Rhbdf2^sin/sin BMDMs were intermediate between Rhbdf2^−/− and WT BMDMs (Fig. 1C). Stimulation of BMDMs with 25 ng/ml of the phorbol ester 12-O-tetradecanoylphorbol-13-acetate (PMA), which is known to strongly activate ADAM17 (Horiuchi et al., 2007b; Sahin et al., 2004), triggered an almost complete downregulation of the MCSFR on WT BMDMs to the background levels of the isotype control (Fig. 1D), but only elicited a partial downregulation in Rhbdf2^sin/sin BMDMs (Fig. 1E), and very little MCSFR shedding in Rhbdf2^−/− BMDMs (Fig. 1F). Finally, a western blot analysis demonstrated that LPS stimulation of BMDMs from WT, Rhbdf2^sin/sin and Rhbdf2^−/− mice increased the levels of the pro-form of ADAM17, although there was substantially less mature ADAM17 in LPS-stimulated Rhbdf2^sin/sin cells, and mature ADAM17 was not detectable in Rhbdf2^−/− BMDMs (Fig. 1G, top panel, see also Adrain et al., 2012; McIlwain et al., 2012). The mature form of ADAM17 could be biotinylated on the cell surface of WT BMDMs, whereas the pro-form was not (Fig. 1G, middle panel, quantification shown in Fig. S1B), consistent with previous studies (Schlöndorff et al., 2000; Adrain et al., 2012; McIlwain et al., 2012). The levels of mature cell-surface biotinylated ADAM17 were substantially lower in unstimulated or LPS-stimulated Rhbdf^sin/sin BMDMs compared to WT controls, whereas no cell-surface biotinylated ADAM17 could be detected in Rhbdf2^−/− BMDMs (Fig. 1G, middle panel) (Adrain et al., 2012; McIlwain et al., 2012).

In order to learn more about the function of Rhbdf2^sin during mouse development, we generated Rhbdf1^−/− Rhbdf2^sin/sin double mutant mice through crosses with previously described Rhbdf1^−/− mice that have no evident spontaneous pathological phenotypes (Li et al., 2015) (see Materials and Methods for details). These animals were born at the expected Mendelian ratio with defective eyelids at birth (Fig. 2), and 66% died within the first day after birth (Fig. S2A). The Rhbdf1^−/− Rhbdf2^sin/sin animals thus strongly resembled Adam17^−/− (Horiuchi et al., 2007a; Peschon et al., 1998) or Rhbdf1^−/− Rhbdf2^−/− (Rhbdf1/2^−/−) double-knockout mice previously produced in our laboratory, which have open eyes at birth and suffer from perinatal lethality (Li et al., 2015). The remaining 34% of Rhbdf1^−/− Rhbdf2^sin/sin mice survived with varying degrees of pathological phenotypes, ranging from ruffled fur (7/11) to normal appearance (4/11). A histopathological analysis demonstrated that newborn Rhbdf1^−/− Rhbdf2^sin/sin mice resembled Rhbdf1/2^−/− or Adam17^−/− mice in that they also had enlarged aortic and pulmonic heart valves and an enlarged zone of hypertrophic chondrocytes in long bone growth plates (Fig. 2A) (Li et al., 2015; Horiuchi et al., 2007a; Peschon et al., 1998; Jackson et al., 2003; Hall et al., 2013).

Previous studies showed that no mature ADAM17 could be detected in tissues from Rhbdf1/2^−/− mice by Western blot (Christova et al., 2013; Li et al., 2015). However, Western blot analysis of different tissues from Rhbdf1^−/− Rhbdf2^sin/sin mice showed reduced, but clearly detectable levels of mature ADAM17 in skin, heart, liver and kidney, and almost normal levels in the lung (Fig. 2B, quantification in Fig. S2B). In addition, we noted a relative increase in the level of pro-ADAM17 in the Rhbdf1^−/− Rhbdf2^sin/sin skin, heart and liver compared to Rhbdf1^+/+ Rhbdf2^sin/sin and WT mice, and an increase in pro-ADAM17 in both Rhbdf1^+/+ Rhbdf2^sin/sin and Rhbdf1^−/− Rhbdf2^sin/sin kidneys and lungs compared to controls.

When we explored the function of Rhbdf2^sin in mouse embryonic fibroblasts (MEFs), we found that the stimulated shedding of the Rhbdf2-selective substrates Kit Ligand 2 (KitL2) and epiregulin (EREG) was strongly reduced in Rhbdf2^sin/sin MEFs compared to that in WT controls (Fig. 3A,B). The constitutive release of KitL2 was slightly higher in Rhbdf2^sin/sin and Rhbdf1^−/− Rhbdf2^sin/sin MEFs compared to that in Rhbdf2^−/− MEFs, suggesting minor residual activity of Rhbdf2^sin. The PMA-stimulated shedding of the ADAM17 substrate TGFα was comparable to that in WT controls, just like in Rhbdf2^−/− MEFs, but was strongly reduced in Rhbdf1^−/− Rhbdf2^sin/sin double mutant MEFs (Fig. 3C). Cell surface biotinylation showed mature ADAM17 on the surface of WT, Rhbdf^sin/sin and Rhbdf2^−/− MEFs, with somewhat lower levels in Rhbdf1^−/− Rhbdf2^sin/sin MEFs and no detectable biotinylated mature ADAM17 in Rhbdf1/2^−/− double mutant MEFs (Fig. 3D, top panels; see Fig. S3A for quantification). These findings correlated with the relative levels of mature ADAM17 in western blots of whole-cell lysates of the different MEFs (Fig. 3D, middle panels; quantification of pro- and mature forms in Fig. S3B, and see also Christova et al., 2013; Li et al., 2015). Taken together, genetic studies and cell-based assays support the interpretation that the sinecure mutation in the Rhbdf2 gene is a strongly hypomorphic mutation, with similar functional consequences with respect to ADAM17 as the deletion of Rhbdf2.

In order to gain a better understanding of the molecular mechanism by which the mutation in Rhbdf2^sin affects the interaction with ADAM17, we undertook molecular modeling and molecular dynamics (MD) simulations of the systems to assess possible conformational changes and relative orientations in the two TMDs. The starting structures of the single TMD of ADAM17 and TMD1 of Rhbdf2 were both constructed as ideal α-helices and embedded in a full atomistic model of the membrane bilayer. The entire system was hydrated, and ions were introduced to simulate a physiologically relevant environment (Fig. 4A,B). Unbiased all-atom MD simulations were performed on the systems as described previously (Perez-Aguilar et al., 2014).

The results from the MD simulations for the case of the ADAM17-TMD indicate that, as the system evolved, the structure deviated from an ideal α-helical conformation as an internal bending appeared in the middle of the TMD segment (compare Fig. 4A and 4B). Moreover, configurational changes were observed in a cluster of three aromatic residues – F43, F47 and F51 – located in the middle section of the TMD on the same face of the helix. In the absence of intermolecular interactions, the flexible helix break created by the proline residue at position 50 seems to aid in bringing these aromatic residues in closer proximity through the formation of π–π stacking interactions (i.e. attractive noncovalent interactions between aromatic rings). Three other residues located in the vicinity of the aromatic cluster on that same helix face (S44, P50 and S52) join the patch (Fig. 4B). In the parallel simulations of the membrane-embedded Rhbdf2-TMD1, the helix tilted relative to the membrane normal (compare Fig. 4C and 4D), so that the site of the sinecure mutation, I387, was positioned at the middle section of the lipid bilayer. Interestingly, the extent of the tilting of Rhbdf2-TMD1 relative to the normal axis of the lipid bilayer, positions this helix in a corresponding orientation that is very similar to that observed crystallographically (PDB accession code 2IC8) for transmembrane helix 1 of the rhomboid protease GlpG, a prokaryotic homolog of Rhbdf2 (Wang et al., 2006).

Viewing the ADAM17-TMD and the Rhbdf2-TMD1 that emerged from their separate MD simulations in the same membrane patch, leads to the hypothesis that the protein segment formed by the aromatic cluster and its surrounding residues – all of which are located in the middle section of the TMD of ADAM17 – is positioned very favorably for an interaction with the TMD1 in Rhbdf2 in the region that would be affected by the missense I387F mutation in Rhbdf2^sin-TMD1. Indeed, the interaction between the two helices at an interface suggested by this juxtaposition would be further strengthened when the cluster of phenylalanine residues in the TMD of ADAM17 is presented with an even more favorable partner, the Phe replacing the WT isoleucine in the Rhbdf2^sin-TMD1. To explore the properties of this interface and the interactions it would engender, we used the respective converged structures from the MD simulation of the helices to perform coarse docking calculations that evaluate possible protein-protein interfaces (see Methods). Top candidate poses for the protein–protein complexes were selected from the docking procedure based on the knowledge-based scoring function utilized in the docking protocol (Fig. 5). The docking calculations were performed for the ADAM17-TMD interaction with the WT Rhbdf2-TMD and the I387F mutant (Fig. 5A,B). While the well known limitations of docking protocols preclude a detailed determination of the interface (Park et al., 2015; Ritchie, 2008), the pattern of inter-helix interactions resulting from the docking calculations has produced the expected set of predictions of specific local interactions for the planned testing in cell-based assays with mutated proteins (see below). We note further that when the Rhbdf2-TMD in the Rhbdf2-TMD–ADAM17-TMD complex predicted from the docking, is superimposed on TMD1 of the cognate GlpG crystal structure (PDB 2IC8), the segments of the interacting helices identified as the interface are positioned for interaction as predicted, while simultaneously allowing additional interactions with other parts of the Rhbdf2 transmembrane bundle without significant steric clashes (Fig. 5C).

Based on specific putative interaction sites between the ADAM17 TMD and Rhbdf2 TMD1 predicted from the all-atom structural modeling of the interface, we introduced five individual point mutations into the middle section of the TMD of ADAM17 (F43A, S44A, F47A, F51A and S52A) that were expected to interfere with the interactions at the interface with Rhbdf2, and a sixth mutation, P50A, designed to probe the role of the proline-mediated kink.

When Adam17^−/− MEFs were transfected with ADAM17 TMD mutants (F43A, S44A, F47A, F51A S52A and P50A; Fig. 6A), the PMA-stimulated shedding of the Rhbdf2-dependent substrate KitL2 (Maretzky et al., 2013) was strongly reduced or abolished by all mutants except P50A, which had somewhat increased constitutive and stimulated activity compared to that of the WT ADAM17 (Fig. 6B). Similar results were obtained with another Rhbdf2-dependent substrate, EREG (Fig. 6C). Importantly, the stimulated shedding of TGFα, which can be supported by Rhbdf1 and Rhbdf2 (Maretzky et al., 2013), was not affected by any of the ADAM17 TMD mutants tested here, serving as a positive control for the functionality of these mutants, although the P50A mutant increased constitutive TGFα shedding (Fig. 6D; a representative western blot to determine expression of ADAM17 and the TMD mutants, and a quantification are shown in Fig. S4A,B). While the experimental results are in agreement with the predicted involvement of these residues in the interface, the observation that the P50A mutation did not block stimulated shedding of Rhbdf2 substrates is intriguing and will be the subject of further evaluation in combined computational and experimental studies because it indicates either that the main set of π–π stacking interactions at the interface between the helical domains are likely to be independent of the structural perturbation, or that the magnitude of the structural change introduced by the P50A mutation is not strong enough to preclude such interactions.

Finally, co-immunoprecipitation experiments with WT Rhbdf2 did not uncover a substantial difference in the interaction with overexpressed WT or mutant forms of ADAM17 under the conditions used here (Fig. S4C). This suggests that the ADAM17 mutants can, in principle, still interact with Rhbdf2 under these conditions, at least when both are overexpressed, even though the functional consequences of the individual mutations with respect to ADAM17-dependent shedding of Rhbdf2-selective substrates is very different from the WT control.

The main goal of this study was to test the hypothesis that ADAM17-dependent shedding is regulated by the TMDs of Rhbdf2 and ADAM17. Previous studies had shown that the Rhbdf2^sin mutation in TMD1 of Rhbdf2 affects the release of TNFα from myeloid cells (Siggs et al., 2012), yet little was known about the underlying mechanism. Here, we report that levels of mature ADAM17 are strongly reduced in Rhbdf2^sin/sin myeloid cells, a cell type in which the maturation and function of ADAM17 is known to depend on Rhbdf2 (Issuree et al., 2013), providing a plausible explanation for why shedding of the ADAM17 substrate TNFα was almost completely abolished following low-dose LPS stimulation (1 ng/ml) (Siggs et al., 2012). Interestingly, the defect in ADAM17 function could be partially overcome in Rhbdf2^sin/sin BMDMs after stimulation with a higher dose of LPS (10 ng/ml), which also led to an increase in the amount of pro- and mature forms of ADAM17 in these cells, and increased levels of mature ADAM17 on the cell surface. Similar results were obtained when ADAM17 activity was assessed by determining the downregulation of the ADAM17 substrate MCSFR in BMDMs. The relatively low levels of mature ADAM17 on the cell surface and its low activity in unstimulated Rhbdf2^sin/sin BMDMs suggest that the sinecure mutation mainly affects the transport of ADAM17 to the cell surface, most likely by impeding its release from the endoplasmic reticulum in BMDMs. However, the partial recovery of TNFα shedding following LPS stimulation suggests that Rhbdf2^sin can support ADAM17-dependent shedding when its expression is upregulated. Nevertheless, after stimulation with LPS, the levels of mature ADAM17 on the cell surface remained lower in Rhbdf2^sin/sin BMDMs than in WT BMDMs, suggesting that upregulation of the expression of ADAM17 and Rhbdf2^sin by LPS is not sufficient to overcome the defect in maturation and restore normal levels of Rhbdf2^sin or ADAM17 on the cell surface.

To further explore the function of Rhbdf2^sin during mouse development, we generated mice that lacked Rhbdf1 and were homozygous mutant Rhbdf2^sin/sin (Rhbdf1^−/− Rhbdf2^sin/sin). These Rhbdf1^−/− Rhbdf2^sin/sin double mutant mice were born at the expected Mendelian ratio with defective eyelids, a characteristic phenotype for mutations in the EGFR pathway that is also observed in mice lacking ADAM17 (Horiuchi et al., 2007a; Peschon et al., 1998). However, whereas the eyes of newborn Adam17^−/− mice are invariably open, the eyelids of newborn Rhbdf1^−/− Rhbdf2^sin/sin mice were sometimes almost closed, but were nevertheless always clearly distinguishable from normal littermate controls. Most Rhbdf1^−/− Rhbdf2^sin/sin mice died within the first day after birth, just like Adam17^−/− or Rhbdf1/2^−/− mice (Horiuchi et al., 2007a; Peschon et al., 1998; Li et al., 2015), demonstrating that the single point mutation in the TMD1 of Rhbdf2^sin renders it unable to support the function of ADAM17 during development. The survival of a small fraction of the double mutant Rhbdf1^−/− Rhbdf2^sin/sin mice could potentially be explained by a slightly increased production or functionality of ADAM17 in the surviving animals, perhaps similar to the increase in ADAM17 activity observed in Rhbdf2^sin/sin BMDMs following stimulation with LPS. Taken together, these findings demonstrate that the Rhbdf2^sin mutation is a severely hypomorphic mutation with a high degree of penetrance, allowing only about a third of the double mutant Rhbdf1^−/− Rhbdf2^sin/sin animals to escape perinatal lethality.

Interestingly, western blot analysis of different tissues from newborn Rhbdf1^−/− Rhbdf2^sin/sin mice showed that some mature ADAM17 could be detected in heart, liver, kidney and skin, and that the ADAM17 levels in lung were comparable to those in WT mice. Previous studies have shown that expression of Rhbdf2 in the lung is relatively high (Leilei et al., 2014; Li et al., 2015), providing a possible explanation for the almost normal expression level of mature ADAM17 in the lungs of Rhbdf1^−/− Rhbdf2^sin/sin mice. However, since the perinatal lethality of Adam17^−/− mice is thought to be the consequence of multi-organ failure (Xu et al., 2013), it is difficult to draw conclusions about how the Rhbdf2^sin mutation affects the function of the mature ADAM17 in any given cell type or tissue based on the Rhbdf1^−/− Rhbdf2^sin/sin mouse phenotype. In studies with Rhbdf1^−/− Rhbdf2^sin/sin MEFs, we found a strong reduction in ADAM17 activity towards the substrates tested here, even though mature ADAM17 was present on the surface of these cells, albeit at lower levels compared to WT MEFs. The experiments in MEFs provide additional support to the notion that Rhbdf2 is not only required as a chaperone that supports the exit of ADAM17 from the ER and its maturation, but that Rhbdf2 is also involved in regulating the shedding of substrates by mature ADAM17 on the cell surface.

To understand more about the potential interaction between the TMD1 of Rhbdf2 and the TMD of ADAM17, we used all-atom simulations of the helices and computational modeling of their interaction. The results suggested a putative interface and mode of interaction between these two TMDs coming together in the mid region of the lipid bilayer. To test the involvement of the predicted interface, we identified loci for point mutations in the TMD of ADAM17, which, from the computational findings, were predicted to disrupt the interactions with the TMD1 of Rhbdf2. This led to the identification of the first specific mutations in the TMD of ADAM17 that affect its function as a sheddase in a substrate-selective manner. Specifically, the mutations in the TMD of ADAM17 only reduced its ability to shed the Rhbdf2-selective substrates KitL2 and EREG, but did not affect the release of TGFα, which can be supported by both Rhbdf1 and Rhbdf2. These results provide additional independent evidence for an interaction of the TMD of ADAM17 with the TMD1 of Rhbdf2. Furthermore, they suggest that these interactions are crucial for Rhbdf2–ADAM17-selective shedding, but not for Rhbdf1–ADAM17-dependent shedding.

Our results support a model in which Rhbdf2–ADAM17-dependent shedding is regulated through an interaction between the ADAM17 TMD and the TMD1 of Rhbdf2. Previous studies had identified an essential role for the TMD of ADAM17 for its activation by various stimuli (Le Gall et al., 2010), whereas the ADAM17 cytoplasmic domain was deemed dispensable (Doedens et al., 2003; Le Gall et al., 2010; Schwarz et al., 2013). The subsequent identification of the so-called inactive rhomboid proteins (iRhoms; Rhbdf proteins) as crucial regulators of ADAM17 (Adrain et al., 2012; Christova et al., 2013; Li et al., 2015; Maretzky et al., 2013; McIlwain et al., 2012; Siggs et al., 2012) and the finding that the Rhbdf2^sin point mutation disrupts the release of the ADAM17 substrate TNFα from cells (Siggs et al., 2012), led to the hypothesis that Rhbdf2 controls the maturation and function of ADAM17 through an interaction with its TMD1. This hypothesis is now supported by the finding that the sinecure point mutation in TMD1 of Rhbdf2^sin leads to an almost complete loss of ADAM17 activity during development and in MEFs in the absence of the related Rhbdf1. Moreover, the identification, through molecular modeling, of specific amino acid residues in the TMD of ADAM17 that are required for Rhbdf2-dependent shedding events, but not Rhbdf1-dependent shedding, further strongly supports the hypothesis that interactions between the TMDs of Rhbdf2 and ADAM17 are required for the regulation of ADAM17-dependent protein ectodomain shedding.

Based on these findings, we propose that there are two pools of ADAM17, one consisting of Rhbdf1–ADAM17 and the other of Rhbdf2–ADAM17, that appear to be differentially regulated by distinct interactions with ADAM17. The function of the Rhbdf2–ADAM17 pool is reduced by the Rhbdf2^sin point mutation and by mutations in the TMD of ADAM17, presumably because they affect interactions that are important for proper formation or for the proper function of the complex between Rhbdf2 and ADAM17. Interestingly, the TMD mutations in ADAM17 did not affect its co-immunoprecipitation with Rhbdf2, suggesting that the effect of the mutations is to change the positioning or mutual orientation of the Rhbdf2–ADAM17 complex in a way that prevents its activity towards Rhbdf2-selective substrates, without affecting the release of the Rhbdf1–ADAM17-dependent substrate TGFα. It appears likely that different residues in the TMD of ADAM17 will be responsible for regulating the function of Rhbdf1–ADAM17, since replacing the entire TMD of ADAM17 with that of CD62L (also known as L-selectin) or of betacellulin abolished the stimulated shedding of the Rhbdf1 or Rhbdf2 and ADAM17-dependent substrate TGFα without affecting its constitutive shedding (Le Gall et al., 2010). These findings also raise the possibility that the recently reported activation of ADAM17 through phosphatidylserine exposure could, at least in part, regulate ADAM17 through an effect on the interaction of its TMD with that of Rhbdf2 (Sommer et al., 2016). A better understanding of the differential regulation of ADAM17 by Rhbdf1 and Rhbdf2 might aid in the development of selective inhibitors of Rhbdf2–ADAM17, which would block the release of TNFα and of Rhbdf2-selective substrates such as HB-EGF, without affecting the protective function of the Rhbdf1–ADAM17-dependent processing of the substrate TGFα in the skin and intestinal barriers.

Mice carrying the Rhbdf^sin point mutation in both alleles of Rhbdf2 (Rhbdf2^sin/sin) on a C57BL/6 background (Siggs et al., 2012) were obtained from Dr. Bruce Beutler (Center for the Genetics of Host Defense, UT Southwestern Medical Center, Dallas, TX) and used to isolate BM and primary BMDMs, with Rhbdf2^−/− and WT mice of C57BL/6 background used for comparison in the experiments in Fig. 1A–F. WT, Rhbdf1^−/− mice and Rhbdf1/2^−/− mice have been described previously and were of mixed genetic background (129Sv,C57BL/6) (Li et al., 2015). To generate Rhbdf2^sin/sin mice that also lack Rhbdf1, we crossed Rhbdf1^−/− mice of mixed genetic background (129Sv,C57BL/6; Li et al., 2015) with Rhbdf2^sin/sin (C57BL/6) mice to produce mixed background Rhbdf1^−/+ Rhbdf2^sin/+ mice, which were backcrossed with Rhbdf2^sin/sin mice. The resulting Rhbdf1^+/− Rhbdf2^sin/sin mice were kept and mated with one another. The three expected genotypes from this mating (Rhbdf1^+/+ Rhbdf2^sin/sin; Rhbdf1^+/− Rhbdf2^sin/sin; Rhbdf1^−/− Rhbdf2^sin/sin) were born at a Mendelian ratio (see Fig. S2A). Histopathological analysis was performed on newborn mice from at least three separate litters for each genotype analyzed here. All animal experiments were approved by the Institutional Animal Use and Care Committee of the Hospital for Special Surgery and Weill Cornell Medicine.

The transmembrane helices for the ADAM17-TMD and Rhbdf2-TMD1 were modeled as ideal helices and oriented into the lipid membrane as predicted by the E(z) potential (Senes et al., 2007). The systems were investigated by unbiased all-atom MD simulations in a physiologically relevant environment as described previously (Perez-Aguilar et al., 2014). Briefly, the protein systems were embedded into a 1-palmitoyl-2-oleoyl-sn-glycero-3-phosphocholine (POPC) lipid membrane followed by addition of TIP3P water molecules (Jorgensen et al., 1983) and by the inclusion of ions so as to obtain a salt concentration of 0.15 M NaCl. The TIP3P water model uses three interaction points to describe the water molecule interactions. The hydrated systems were investigated with unbiased all-atom MD simulations using Nanoscale Molecule Dynamics (NAMD) (Phillips et al., 2005) and the Chemistry at Harvard Macromolecular Mechanics (CHARMM) force field parameters for proteins and lipids (Mackerell et al., 2004). NAMD is a highly scalable software that performs MD simulations, while the CHARMM force field contains the atomistic parameters that describe the different interactions for biomacromolecules. Constant temperature (310 K) and pressure (1 atm) were maintained by Langevin dynamics and the hybrid Nosé–Hoover Langevin piston, respectively (Feller et al., 1995). The protein–protein docking calculations used the converged configurations from the MD simulations for both transmembrane systems (ADAM17-TMD and Rhbdf2-TMD1). Docking was carried out with the GRAMM-X docking protocol (Tovchigrechko and Vakser, 2005), that utilizes a fast Fourier transformation approach to search for suitable protein–protein complexes where the protein conformations are treated as rigid body structures. During the global searching stage, the methodology uses a smoothed Lennard-Jones potential to estimate the intermolecular potential, while knowledge-based potentials are used to rescore candidate poses.

To generate Western blots of the pro- and mature form of ADAM17 in different tissues of newborn (P1) WT, Rhbdf2^sin/sin and Rhbdf1^−/− Rhbdf2^sin/sin mice, we harvested the skin, heart, liver, kidney and lung, and processed these tissues as previously described (Li et al., 2015). Comparable amounts of protein were loaded onto 10% SDS-PAGE gels and transferred onto nitrocellulose membranes (Pall Corp., Port Washington, NY). After blocking in 5% non-fat milk at room temperature for 1 h, the membranes were incubated in primary antibody generated against the cytoplasmic domain of ADAM17 overnight at 4°C (1:500 dilution, for validation of specificity using Adam17^−/− MEFs, please see McIlwain et al., 2012), washed three times in Tris-buffered saline (TBS) with 0.5% Tween 20 and then incubated in 1:1000 horseradish peroxidase (HRP)-labeled donkey anti-rabbit-IgG secondary antibody. Bound antibodies were detected using the ECL system (Amersham Biosciences, Waltham, MA) and a Chemdoc image analyzer (Bio-Rad, Hercules, CA), and the images were assembled using Microsoft Powerpoint software. Loading controls were generated by running the same samples on a separate gel, or in some cases, by stripping membranes for 30 min at 55°C in stripping buffer (2% SDS, 50 mM 2-mercaptoethanol in 62 mM Tris-HCl pH 6.7) followed by incubation with anti-ADAM9 antibodies at 1:500 (for validation of the specificity of these antibodies in Adam9^−/− cells and tissues, see Weskamp et al., 1996). Anti-α-tubulin rabbit antibodies were from Science Signaling (Cat. # 2144S, Lot 5) and were used at a 1:1000 dilution.

After a pre-incubation for 30 min at 4°C, BMDMs or MEFs were washed with biotinylation buffer (0.5 mM MgCl₂, 1 mM CaCl₂ in PBS) twice and incubated with NHS-LC-Biotin (1 mg/ml) for 45 min at 4°C. Cells were washed with biotinylation buffer including 50 mM glycine to quench the reaction. Cell lysates were incubated with StrepAvidin–Sepharose for 2 h at room temperature to isolate biotinylated proteins from the lysate, which were then analyzed by western blotting.

Cos7 (originally from ATCC, mycoplasma negative) were grown in six-well tissue culture plates to 80% confluency and then transfected with plasmids encoding C-terminally T7-tagged Rhbdf2 (McIlwain et al., 2012) or C-terminally HA-tagged WT ADAM17 (Horiuchi et al., 2007b) or ADAM17 mutants (F43A, S44A, F47A, F51A, S52A or P50A) or combinations of Rhbdf2 and WT or mutant forms of HA-tagged ADAM17. When the cells were confluent, they were harvested and lysed in 180 µl cell lysis buffer (Ca²⁺ and Mg⁺²-free PBS, 1% Triton X-100, 5 mM 1,10-phenantroline, 2 µg/ml leupeptin, 0.4 mM benzamidine, 10 µg/ml soybean trypsin inhibitor and 0.5 mM iodoacetamide) per well. The lysates of two identically treated wells were combined (total volume 360 µl) and spun at 16,100 g in an Eppendorf table-top centrifuge for 20 min. After harvesting the supernatants, 90 µl of each sample was removed and used to determine the transfection levels (input). Cell lysis buffer was added to the remainder of the supernatants to adjust the volume to 1 ml, and 1 µg of anti-T7 monoclonal antibody (Novagen, T7-Tag mAb, Cat. No. 69522-3, Lot D00108177) was added, and the sample incubated on a rotating incubator at 4°C for 4 h. Then 20 µl of Protein-G–Sepharose slurry (1:1, Sigma) was added to the supernatants, incubated for 30 min at room temperature and then pelleted for 30 s in a microcentrifuge. The beads were washed three times with 500 µl cell lysis buffer, then boiled in 75 µl sample loading buffer to elute the bound proteins. Comparable amounts of protein were loaded onto 10% SDS-PAGE gels and transferred onto nitrocellulose membranes (Pall Corp.). After blocking in 5% nonfat milk at room temperature for 1 h, the membranes were incubated in primary anti-HA (Roche, anti-HA High Affinity rat mAb, Cat. No. 11867431001, Clone 3F10; 1:5000), anti-T7 antibodies (1:10,000) or anti-ADAM9 cytoplasmic domain (1:1000) antibodies (Weskamp et al., 1996), as indicated, overnight at 4°C. The membranes were washed three times in TBS with 0.5% Tween 20 before adding 1:1000 HRP-labeled secondary antibody and incubation for 1 h at room temperature. Bound antibodies were detected using the ECL system (Amersham Biosciences, Waltham, MA) and a Chemdoc image analyzer (Bio-Rad, Hercules, CA), and the images were assembled using Microsoft Powerpoint software.

Embryonic fibroblasts were isolated from E13.5 Rhbdf2^sin/sin and Rhbdf1^−/− Rhbdf2^sin/sin embryos to generate primary MEFs for immortalization, as previously described (Sahin et al., 2006). Briefly, the head and viscera of E13.5 embryos were removed and the remaining tissue was minced and then trypsinized in 0.25% trypsin for 15 min at 37°C. Cells were released through mechanical trituration and grown in Dulbecco's modified Eagle medium (DMEM) supplemented with antibiotics and 10% fetal calf serum. Primary MEF cells were immortalized by transfection with a plasmid carrying the simian virus 40 (SV40) large T antigen. The cell lines used in this study were authenticated by PCR genotyping and were not contaminated.

Adam17^−/− and WT control MEFs were grown to 80% confluency and then transfected with plasmids encoding alkaline phosphatase-tagged transforming growth factor α (TGFα), Kit Ligand 2 (KitL2) or epiregulin (EREG) together with WT or mutant ADAM17 (Sahin et al., 2004) using Lipofectamine 2000 (Invitrogen, Grand Island, NY). When the cells were confluent, they were washed in Opti-MEM (Gibco, Grand Island, NY) for 1 h. Then fresh Opti-MEM with or without 25 ng/ml PMA was added for 45 min to stimulate shedding, as indicated. After incubating with the alkaline phosphatase substrate, 4-nitrophenyl phosphate, the alkaline phosphatase activity in the supernatant and lysate were measured by determining the absorbance at 405 nm. Three identical wells were prepared, and the ratio of alkaline phosphatase activity in the supernatant and that in the cell lysate plus supernatant was calculated. Each experiment was conducted at least three times.

The ADAM17 mutants were generated by using mouse ADAM17 cDNA as the template (kindly provided by Dr. Gillian Murphy, Dept. of Oncology, Cambridge University, UK), except for P50A, which was generated from ADAM17Δ-cyto (Le Gall et al. 2010). The QuikChange^® site-directed mutagenesis kit (Stratagene, CA) was used to generate the point mutations of ADAM17 (F43A, S44A, F47A, I49N, F51A, S52A and P50A). The plasmid primers used for this purpose are: F43A (TTC→GCC), forward, 5′-GTTGGGTCTGTTCTGGTTGCCTCCTTGATATTTTGG-3′ and reverse, 5′-CCAAAATATCAAGGAGGCAACCAGAACAGACCCAAC-3′; S44A (TCC→GCC), forward, 5′-GTTGGGTCTGTTCTGGTTTTCGCCTTGATATTTTGGATTC-3′ and reverse, 5′-GAATCCAAAATATCAAGGCGAAAACCAGAACAGACCCAAC-3′; F47A (TTT→GCT), forward, 5′-CTGGTTTTCTCCTTGATAGCTTGGATTCCTTTCAGC-3′ and reverse, 5′-GCTGAAAGGAATCCAAGCTATCAAGGAGAAAACCAG-3′; F51A (TTC→GCC), forward, 5′-CTTGATATTTTGGATTCCTGCCAGCATTCTTGTCCAC-3′ and reverse, 5′-GTGGACAAGAATGCTGGCAGGAATCCAAAATATCAAG-3′; S52A (AGC→GCC), forward, 5′-GATATTTTGGATTCCTTTCGCCATTCTTGTCCACTGTG-3′ and reverse, 5′-CACAGTGGACAAGAATGGCGAAAGGAATCCAAAATATC-3′; and P50A (CCT→GCT), forward, 5′-TCCTTGATATTTTGGATTGCTTTCAGCATTCTTGTC-3′ and reverse, 5′-GACAAGAATGCTGAAAGCAATCCAAAATATCAAGGA-3′.

All values are expressed as mean±s.e.m. (data from at least three independent experiments). The two-tail Student's t-test was used for all statistical analyses. Animals from at least three litters were used for the analysis, and no animals were excluded. Histopathological analyses were done in a blinded manner in that the genotype was not known during the analysis. P<0.05 was considered as statistically significant.

We thank Sarah Loh and Elin Mogollon for technical assistance.