Structural basis for interdomain communication in SHIP2 providing high phosphatase activity

SH2-containing-inositol-5-phosphatases (SHIPs) dephosphorylate the 5-phosphate of phosphatidylinositol-3,4,5-trisphosphate (PI(3,4,5)P3) and play important roles in regulating the PI3K/Akt pathway in physiology and disease. Aiming to uncover interdomain regulatory mechanisms in SHIP2, we determined crystal structures containing the 5-phosphatase and a proximal region adopting a C2 fold. This reveals an extensive interface between the two domains, which results in significant structural changes in the phosphatase domain. Both the phosphatase and C2 domains bind phosphatidylserine lipids, which likely helps to position the active site towards its substrate. Although located distant to the active site, the C2 domain greatly enhances catalytic turnover. Employing molecular dynamics, mutagenesis and cell biology, we identify two distinct allosteric signaling pathways, emanating from hydrophobic or polar interdomain interactions, differentially affecting lipid chain or headgroup moieties of PI(3,4,5)P3. Together, this study reveals details of multilayered C2-mediated effects important for SHIP2 activity and points towards interesting new possibilities for therapeutic interventions. DOI: http://dx.doi.org/10.7554/eLife.26640.001

The SHIP 5-phosphatase (5-Ptase) family consists of two members, SHIP1 and SHIP2. SHIP1 expression is limited to hematopoietic cells and spermatocytes, whereas SHIP2 is ubiquitously expressed (Liu et al., 1998;Pesesse et al., 1997). SHIP1 plays an important role in myeloid homeostasis and exhibits reduced levels or mutational inactivation in various leukemias and lymphomas, suggesting a role as tumor suppressor by negatively regulating the PI3K/Akt pathway (Brauer et al., 2012;Lo et al., 2009;Luo et al., 2003). SHIP2 polymorphisms associate with susceptibility to type 2 diabetes mellitus and hypertension (Hao et al., 2015;Kagawa et al., 2005;Kaisaki et al., 2004) as well as opsismodysplasia (Below et al., 2013;Huber et al., 2013;Iida et al., 2013), a rare but severe type of skeletal dysplasia. SHIP2 knockout mice were reported to be resistant to dietary obesity (Sleeman et al., 2005) and mice expressing catalytically inactive SHIP2 displayed several developmental defects and lower insulin secretion (Dubois et al., 2012). Several studies associate SHIP2 with signaling triggered by growth factors, including EGF and FGF (Jurynec and Grunwald, 2010;Olsen et al., 2006). Like SHIP1, SHIP2 also plays important roles in cancer, although its behavior is more complex and tissue dependent. Similar to SHIP1, SHIP2 functions as a tumor suppressor in glioblastoma, erythroleukemia and squamous cell carcinoma (Giuriato et al., 2002;Taylor et al., 2000;Yu et al., 2008), but is oncogenic in breast carcinomas and colorectal cancer (Hoekstra et al., 2016;Prasad et al., 2008). On the one hand, this ambivalent oncogenic behavior of SHIP2 in different cell types likely depends on how efficient the SHIP2 product, PI(3,4)P 2 , is further metabolized to PI(3)P, a reaction catalyzed by inositol polyphosphate 4-phophatases (INPP4). Accumulation of PI (3,4)P 2 in a case where the INPP4 reaction is rate limiting can be oncogenic, since PI(3,4)P 2 can still partially activate the Akt pathway (Li and Marshall, 2015), but is no longer accessible for degradation by PTEN. On the other hand, the SHIP product can be directly tumor suppressing by shutting down mTORC1 signaling (Marat et al., 2017).
SHIP1 and SHIP2 are large~140 kDa multidomain proteins that share a very similar domain organization ( Figure 1A). The SHIP Ptase domain belongs to a family of Mg 2+ -dependent inositol phosphatases with specificity for 5 P dephosphorylation of inositol rings. These 5-Ptases have different substrate recognition specificities (Jefferson and Majerus, 1996) with the SHIP 5-Ptase exhibiting a strong preference for 3-phosphorylated substrates (Chi et al., 2004;Pesesse et al., 1998;Trésaugues et al., 2014). Further, the inositol 5-Ptase family displays distant homology to Mg 2+dependent DNase I and Apurinic/Apyrimidinic (AP) endonucleases (Dlakic´, 2000;Whisstock et al., 2000) and the conservation of catalytic residues suggests a common mechanism of catalysis (Whisstock et al., 2002). The existence of a C2 domain following the Ptase domain was previously predicted for SHIP1 (Ong et al., 2007), however, the sequence identity to other C2 domains is low and a C2 fold has not been structurally confirmed. The SHIP1 C2 domain was reported to allosterically upregulate Ptase activity by binding to the PI(3,4)P 2 product, a feature not found to be conserved in SHIP2.
Here, we report the crystal structure of SHIP2 containing the 5-Ptase and a C2 domain (Ptase-C2), revealing that the two domains tightly associate via an extensive interface. We show that, both the Ptase and Ptase-C2 regions of SHIP2 bind phosphatidylserine (PS), which likely localizes the rigid Ptase-C2 region close to its substrate. Importantly, even in the absence of a localization effect, the C2 domain provides activating signals to the Ptase active site, resulting in a significant increase in catalytic turnover. Combining enzyme kinetics, molecular dynamics (MD) simulations and mutagenesis we identify distinct regions at the domain interface with different effects on SHIP2 activity. Hydrophobic Ptase-C2 interactions increase SHIP2 turnover with the lipid chain containing PI(3,4,5)P 3 substrate, whereas polar interactions communicate from the domain interface to a substrate binding loop mainly affecting catalysis of the soluble inositol 1,3,4,5-tetrakisphosphate (IP 4 ) substrate. Together, our data provide novel structural and mechanistic insights on a C2 assisted functioning of SHIP2, and by homology likely of SHIP1.

Structure of SHIP2 Ptase-C2
A construct of human SHIP2 containing residues 420-878 corresponding to the Ptase and a C2 domain (Ptase-C2) was crystallized and diffraction data collected to 1.96 Å . The protein crystallized in the P2 1 2 1 2 1 space group with eight molecules per asymmetric unit (for crystallographic and refinement statistics, see Table 1). Each of the eight molecules (A-H) contains a 5-Ptase and a C2 domain ( Figure 1B). The overall structure of the Ptase domain in Ptase-C2 is similar to the one previously reported (Mills et al., 2012;Trésaugues et al., 2014), displaying a central b-sheet sandwich decorated by seven helices (a1-a7) and several loops (L1-L4) ( Figure 1B). However, there are significant  differences. Loop residues 587-594 (L3) are located at the interface with the C2 domain and are due to extensive interactions with the C2 domain well ordered ( Figure 1C). Importantly, a loop proximal to the active site, spanning residues 674-684 (L4), was in previous Ptase structures disordered or closed over the active site (Mills et al., 2012;Trésaugues et al., 2014) (Figure 1D; hereafter referred to as L4-in conformation), whereas in our structure 6 of the Ptase-C2 monomers (A, B, C, E, F, G) exhibit a conformation where L4 points away from the active site (L4-out). These L4-out conformations are stabilized by interactions of R682 in L4 with one or two aspartic acids (D613/D615) adjacent to helix a5 ( Figure 1E-F, hereafter referred to as singly or doubly bound L4-out conformations). In molecule D, R682 is half way between the L4-in and L4-out conformation and in monomer H R682 is disordered. Together, these conformations suggest extensive mobility of L4 with a displacement of the guanidinium group of the R682 side chain by~20 Å between L4-in and L4-out conformations. L4-out conformations appear to be favored by the presence of the C2 domain, since they are only observed for Ptase-C2, although crystal packing analysis suggests ample space in Ptase crystals in this region. In L4-out structures where R682 is doubly bound to D613 and D615 (molecules B and F), a network of hydrogen bonds is established that includes H674, T683 and N684 on L4, rigidifying the loop ( Figure 1F). The C2 domain interacts with the Ptase on a face that is opposite to the active site ( Figure 1B). The Ptase and C2 domains share an extensive interface of 940 Å 2 ( Figure 1C), resulting in stabilization of the Ptase (Supplementary file 1). The interface is constituted largely of hydrophobic interactions, with A592, F593, L597, Y697 and P698 in the Ptase and Y770, H820, L822, E836, V838 and A840 in the C2 domain forming a hydrophobic core ( Figure 1C, Figure 1-figure supplement 1). A significant peripheral polar contact is formed by R649 in the Ptase, forming hydrogen bonds with E862 and a cation-p interaction with F855 in the C2 domain. Notably, R649 is located in a loop that connects to a stretch of three helices (a5-7, colored grey in Figure 1), which at the other end join to D613 and D615, the docking site for R682 in the L4-out conformation.

The C2 domain of SHIP2
Our structure confirms the presence of a b-sheet sandwich structure typical of C2 domains within residues 746-874 of SHIP2. Using the DaliLite server (Holm and Rosenströ m, 2010), we identified the closest structural homologues as the C2 domains of the itchy E3 ubiquitin protein ligase (pdb code 2NQ3, Z score = 12.7, RMSD over 106 residues = 2.1 Å ) and dysferlin (pdb code 4IQH, Z score = 12.4, RMSD over 110 residues = 2.4 Å ), which both belong to the PKC type C2 subfamily (Zhang and Aravind, 2010). A sequence-based similarity search using HH-pred (Sö ding et al., 2005) also identified the C2 domain of dysferlin as the closest related C2 domain (E-value = 0.049), although with 16% sequence identity the homology is low. The topology adopted by the SHIP2 C2 domain is as for itchy and dysferlin that of type II (or P-variant) (Corbalan-Garcia and Gó mez-Fernández, 2014). PKC type C2 domains usually bind phospholipids, most frequently phosphatidylserine (PS), often in a Ca 2+ -dependent manner via acidic residues present in three Ca 2+ -binding loops (CBLs) at the tip of the C2 domain ( Figure 2A-B) (Cho and Stahelin, 2006). Structure-based sequence alignment indicates that only 2 of 5 acidic residues on CBL1 and CBL3 are conserved between PKCa and SHIP2 (D829 and E832) and two are replaced by serines (S761 and S827). To test for lipid binding and calcium dependency, we performed surface plasmon resonance (SPR) experiments with immobilized PS vesicles and protein lipid overlay (PLO) assays. We found that both the Ptase and Ptase-C2 bind PS, with the C2 domain enhancing the interaction ( Figure 2C Figure 2D), an effect that was not detected in PLO experiments ( Figure 2-figure supplement 1). In summary, we conclude that both, the Ptase and Ptase-C2 bind PS and the small effect of Ca 2+ suggests that the interaction is dominated by direct PS binding to basic residues, rather than Ca 2+ -mediated binding to acidic residues and serines. Nevertheless, the low conservation of Ca 2+ -binding residues in CBL1 and CBL3 of the SHIP2 C2 domain might weakly contribute to Ca 2+ -mediated lipid binding ( Figure 2D).

The C2 domain affects SHIP2 activity
Next, we tested the effect of the C2 domain on SHIP2 activity. For this, we employed a Malachite Green assay to compare activity of SHIP2 Ptase-C2 to that of the isolated Ptase. Activity was measured with two soluble substrates, IP 4 and PI(3,4,5)P 3 -diC8 and in all cases we observe a behavior close to conventional Michaelis-Menten enzyme kinetics ( Figure 3). We find that the C2 domain increases turnover rates for both substrates ( Figure 3A-B, Table 2). Interestingly, the effect is significantly larger with PI(3,4,5)P 3 -diC8, where the C2 domain increases the k cat~1 0 fold, while for IP 4 the increase is~1.5 fold. As a result, the C2 domain switches the substrate preference with the Ptase alone turning over IP 4 twice as fast as PI(3,4,5)P 3 -diC8, whereas Ptase-C2 displays~4 fold faster kinetics with PI(3,4,5)P 3 -diC8 compared to IP 4 ( Table 2). The K M is less affected by the C2 domain with a significant change only with PI(3,4,5)P 3 -diC8, where the absence of the C2 domain decreases the K M~2 fold ( Table 2). The fact that the distant C2 domain has specific and differential effects on catalysis of the two substrates, suggests the presence of an allosteric communication between the domain interface and the active site ( Figure 1B).
The effect of the C2 domain was further investigated by mutagenesis. In order to disrupt the hydrophobic core of the domain interface we mutated F593 and L597 to aspartates (FLDD mutant). Further, we mutated R649 in the Ptase to alanine (R649A) to prevent polar interactions with E862 and F855 in the C2 domain ( Figure 1C). Crystal structures of Ptase-C2 FLDD confirm that few hydrophobic interactions remain intact between the Ptase and C2 domains (Figure 1-figure supplement 2). The FLDD mutations also disrupt interdomain hydrogen bonds between R649 and E862. When comparing the catalytic turnover of the FLDD mutants we find that the C2 domain no longer has a    Table 2. Error bars represent SEM from at least three measurements. DOI: 10.7554/eLife.26640.010 The following source data is available for figure 3: Source data 1. Source data for plots in Figure Table 2), indicating that the FLDD mutations largely mimic the absence of the C2 domain. We note that the FLDD mutations increase intrinsic Ptase activity with PI(3,4,5)P 3 -diC8, but importantly the presence of the C2 does not cause any further change. Interestingly, the polar R649A mutation specifically eliminates the effect of the C2 domain on IP 4 kinetics, while retaining most of its effect on PI(3,4,5)P 3 ( Figure 3E-F). Together, these results suggest that the polar R649-E682/F855 interactions between Ptase and C2 domains mainly affect SHIP2 activity with IP 4 , whereas hydrophobic interactions increase activity towards PI (3,4,5)P 3 . Next, we tested whether PS binding alters the C2-mediated effects. We find that soluble PS-diC8 has little effect on IP 4 kinetics and increases activity~40% with PI(3,4,5)P 3 -diC8 ( Figure 4A-B), both for Ptase and Ptase-C2. If instead a lipid vesicle embedded PI(3,4,5)P 3 -diC16 substrate is used, the presence of 30% (mol/mol) PS-diC16 in the vesicles increases SHIP2 activity~2-3 fold for both, Ptase and Ptase-C2 ( Figure 4C). The PS-mediated increase in activity on vesicles is maintained for the FLDD mutant, whereas as with soluble substrates, the increase in activity due to the C2 domain is abrogated by the interface mutations ( Figure 4D). In summary, these data show that the effect of PS is amplified for SHIP activity on vesicles, possibly due to a localization or positioning effect and this effect is independent and additive to the enhancing allosteric signaling of the C2 domain.

Communication between the C2 domain and the active site
In order to obtain insight on the path of communication between the C2 domain and the active site we performed unbiased molecular dynamics (MD) simulations of the SHIP2 Ptase in presence or absence of the C2 domain. Simulations were started in the doubly bound L4-out conformation, where R682 interacts with D613 and D615 (based on molecule B in the crystal structure). The simulations indicate that the C2 domain induces several dynamic changes in the Ptase domain, as can be seen from root mean square fluctuations (RMSF) throughout the simulation ( Figure 5A). Not surprisingly, L3 is strongly stabilized due to its direct interaction with the C2 domain. On the other hand, L2 (residues 531-539) and L4 fluctuate stronger in presence of the C2 domain. RMSF analysis separated into different L4 states (R682 unbound, singly or doubly bound) indicate that higher L4 fluctuations occur when L4 is detached from D613/D615 ( Figure 5-figure supplement 1A). Non-loop regions fluctuate little overall but fluctuations are increased in a stretch containing helices a5 and a6 and the proximal D613/D615 docking site of R682 (shaded grey in Figure 5A-B). Principal component analysis shows as one of the principal motions a5 moving away from a6 ( Figure 5A insert, Videos 1-2). In order to analyze which regions deviate during the simulations most from the doubly bound L4-out starting conformation, we calculated root mean square deviations (RMSD) with respect to the L4-out starting structure, analyzed per residue and averaged over time. We find a significant difference in the region of a5, a6 and the proximal D613/D615, which in the presence of the C2 domain remain closer to the R682 doubly bound starting structure, whereas in absence of the C2 domain this region deviates significantly from the starting conformation ( Figure 5B, also observed in   cluster analysis of snapshots from the simulation reveals that for the Ptase alone all of the main clusters remain in a L4-out conformation, whereas for Ptase-C2 the third most populated cluster has R682 detached from D613/D615 ( Figure 5-figure supplement 1G-H). Together, these data support a communication between the C2 domain and the active site in the Ptase by modulating the dynamics of helices a5-7 and L4 to facilitate transitions between L4-in and L4-out conformations. Structurally, this suggests a communication via R649, since this residue connects helices a5-7 to the C2 domain.
To understand how L4 dynamics could affect catalysis, we next analyzed the mode of substrate binding in SHIP2. Co-crystallization experiments using the catalytically inactive SHIP Ptase-C2 D607A mutant with Mg 2+ and substrate failed to reveal clear electron density for the substrate. Therefore we generated a model based on the homologue INPP5B crystal structure containing the PI(4)P product (Trésaugues et al., 2014) and insights from AP endonucleases, which are thought to share a common phosphate hydrolysis mechanism (Whisstock et al., 2002). We modeled the substrate according to a Michaelis-Menten complex, where all catalytic residues are in place for catalysis ( Figure 6A-B). In this complex the catalytic base D607 stabilizes a water molecule for 5 P attack. According to our model a key 4 P interacting residue is N684 on L4. N684 can adopt two conformations ( Figures 1D-F and 5C) and needs to point towards the active site to interact with the substrate 4 P, which is seen in unbound or singly bound, but not doubly bound L4 conformations. In our model L4 adopts the L4-in conformation seen in 2 SHIP2 Ptase structures (PDB: 3NR8 and 4A9C; Mills et al., 2012;Trésaugues et al., 2014). In this conformation R682 is ideally positioned to interact with the 3 P of PI(3,4,5)P 3 . We performed unbiased MD simulations of SHIP Ptase or Ptase-C2 starting with the PI(3,4,5)P 3 -diC8 or IP 4 substrates bound according to our model shown in Figure 6A-B. Interestingly, the C2 domain significantly restricts the displacement of the headgroup of both substrates away from the Michaelis-Menten complex ( Figure 6C-D, Videos 3-4).
Although our co-crystallization experiments did not reveal bound substrate they did exhibit electron density corresponding to a Mg 2+ ion and a phosphate group ( Figure 6-figure supplement  1A). The Mg 2+ ion is on average displaced by 1.3 Å and the phosphate by 1.7 Å compared to the Mg 2+ and 5 P positions in our model ( Figure 6-figure supplement 1B). The position of the phosphate does not leave space for an attacking water bound to D607, hence this could represent the post-catalysis positions of the Mg 2+ and 5 P.

Probing the communication path by mutagenesis
The simulations and crystal structures indicate that a potential allosteric path linking the C2 domain to the active site could lead via R649 to helices a5-7 and at the other end of the helices through D613 and D615 to R682 and L4. To test this hypothesis, we generated the following mutants along the path: R649A (already generated as an interface mutant), R691A, D613A/D615A, R682A, N684A as well as active site mutants D607A and R665A ( Figure 7A). All mutants were generated for Ptase and Ptase-C2 with the exception of R691A and D613A/D615A, which appear to destabilize the isolated Ptase and only yielded soluble protein for Ptase-C2. Circular dichroism (CD) analysis indicates that most of the mutations do not alter protein structure, apart from R691A in Ptase-C2 and N684A in the Ptase (Figure 7-figure supplement 1). The mutants were tested for changes in activity with IP 4 and PI(3,4,5)P 3 -diC8 ( Figure 7, Table 3). Interestingly, D613A/D615A mutated Ptase-C2 behaves similar to the R649A mutant, in that both only affect turnover with IP 4 , but have little effect on PI(3,4,5)P 3 kinetics ( Figures 3E-F and 7B-C). This similitude in behavior supports an allosteric communication via R649 and D613/D615. Further, we find that the R682A and N684A mutants have lost most of their activity (Table 3; note that for N684A this is also the case for the CD-stable Ptase-C2 form), supporting a key role for these residues in substrate binding via 3 P and 4 P, respectively ( Figure 6A). R691 links the Ptase core b- sandwich with helices a5-7 ( Figure 7A) and its mutation inactivates Ptase-C2, which together with the changed CD spectra indicates a structural role for this residue. Unexpectedly, R665A mainly affects the Ptase domain, but has little effect on Ptase-C2 activity ( Figure 7D-E, Table 3). According to our model, R665 stabilizes a water molecule in the coordination sphere of the Mg 2+ ion ( Figure 6B). This fits with MD simulations showing C2-mediated stabilization of the substrate headgroup ( Figure 6C-D), apparently making the R665-mediated stabilization dispensable in presence of the C2 domain. Together, these results support an allosteric communication from the C2 domain via R649 and helices a5-7 to L4, which by changing their dynamic behavior affects enzyme kinetics with the IP 4 headgroup. In addition, they support an overall stabilization of the Ptase active site by the C2 domain allowing both substrates to bind in a catalytically productive mode. Lastly, to probe how C2 interactions affect activity of full-length SHIP2 in a cellular setting we introduced the FLDD mutations or a C2 deletion (DC2) into the SHIP2 sequence and transiently expressed the full-length proteins in HEK293 cells (Figure 8-figure supplement 1). Expressing wild-type SHIP2 causes a significant reduction in downstream Akt phosphorylation levels (Figure 8). Removing the C2 domain abolishes observed SHIP2 activity, whereas the FLDD mutations partially reduce activity of the wild-type protein. This confirms that C2 interactions observed in the crystal structure are important for SHIP2 activity in cells.

Discussion
Of the enzymes controlling PI(3,4,5)P 3 levels (PI3K, PTEN and SHIP1/2), the SHIP enzymes lag far behind in terms of a detailed understanding of their functioning and regulation. Surprisingly, even though the Ptase domains of PTEN and SHIP2 bear no homology at the sequence or structural level, they conceptually share an important feature in that they are rigidly linked to a C2 domain (Lee et al., 1999). Despite the low homology between the PTEN and SHIP2 C2 domains (4% identity in a structural alignment) they use a common face for Ptase interactions. Indeed, a corresponding surface is also used by the C2 domain of PLCd1 to form extensive interactions with its catalytic domain (Essen et al., 1996). It appears therefore, that the C2 domain presents an ideal scaffold to dock lipid modifying enzymes onto the membrane and via rigid interactions orient the catalytic domain for productive substrate attack. For SHIP2, we show that the C2 domain is essential for cellular function and the rigid interface enhances its efficiency (Figure 8). Assuming that the SHIP2 C2 domain engages the membrane via motifs known to bind lipids in other C2 domains (Cho and Stahelin, 2006) and taking into account several positively charged residues on the Ptase and C2 domains facing the membrane, we provide a model of SHIP2 Ptase-C2 docked to a lipid bilayer ( Figure 9A). Although, we cannot rule out domain rearrangements between the Ptase and C2 domains upon lipid binding, the extensive domain interface between the two domains argues against large changes and suggests that the Ptase-C2 portion of SHIP2 represent the minimal fully active catalytic unit. Moreover, our model indicates that the rigid Ptase-C2 fragment can bind the membrane and engage substrate without the need of extensive extraction of the PI(3,4,5)P 3 substrate from the lipid bilayer. In SHIP2, PS binding occurs through both, the C2 and the Ptase domains and, as shown previously, interactions with the Ptase enhance SHIP2 activity (Vandeput et al., 2006). A similar scenario has indeed been observed for PTEN, where regions outside the C2 domain also contribute to membrane interactions and binding of acidic phospholipids to the N-terminus enhances activity (Campbell et al., 2003;Walker et al., 2004). The fact that the PS effect is larger when embedded in vesicles compared to soluble PS-diC8 ( Figure 4) suggests that PS mediated SHIP2 localization and/or positioning on vesicles does contribute to activity on lipid membranes, however, binding via the Ptase appears sufficient for this effect. Independent to a role in membrane binding, our data suggest that the C2 domain provides important allosteric interdomain effects to increase Ptase activity. In contrast to a traditional concept of allosteric signaling, effects by the C2 domain appear to be transmitted via changes in protein dynamics and active site stabilization rather than large conformational changes. Our data suggest that hydrophobic and polar regions at the Ptase-C2 domain interface differentially affect the lipid chains or head group of the PI (3,4,5)P 3 substrate. The fact that the R649A mutant, which presumably retains hydrophobic Ptase/C2 interactions, only removes C2 effects on IP 4 but not PI(3,4,5)P 3 kinetics, indicates that R649 communicates with the PI(3,4,5)P 3 headgroup and hydrophobic interactions with PI(3,4,5)P 3 lipid chains. Since the PI(3,4,5)P 3 lipid chains are predicted to interact with L2 ( Figure 6A), this suggests a communication between hydrophobic C2 interactions and L2.
Further, our study has identified a specific path of polar connections, linking the C2 domain to the Ptase active site. This 'polar' path leads via C2 interactions with R649 in the Ptase to helices a5-7 and at the other end of the helices to D613/D615 and R682 in L4. Mutation of R691 at the interface between helices a5-7 and the Ptase core inactivates SHIP likely due to structural destabilization (Table 3, Figure 7-figure supplement 1B), and similar inactivating mutations at this interface have been associated with opsismodysplasia (P659S and W688C) and protection from diabetes (L632I) (Huber et al., 2013;Kagawa et al., 2005). More specific disruption of this path, by either R649A or D613A/D615A mutations specifically removes C2 effects on IP4 ( Figures 3E-F and 7B-C). In a cellular setting this path could be relevant, firstly for SHIP2 activity on intracellular IP 4 , whose cellular roles are only emerging for example in T-cell signaling (Huang and Sauer, 2010). Secondly, it is likely also important when the SHIP2 enzyme approaches the PI(3,4,5)P 3 substrate in membranes with the lipid chains buried in the lipid bilayer.
A critical observation in our study is that L4, proximal to the active site, can switch between a closed (L4-in) and an open (L4-out) conformation ( Figure 1D-F) and our simulations indicate that the  Table 3. Error bars represent SEM from at least three measurements. DOI: 10.7554/eLife.26640.024 The following source data and figure supplement are available for figure 7: Source data 1. Source data for plots in Figure  where efficient opening and closing of L4 is important for high turnover ( Figure 9B, Video 5): To initiate the cycle, L4 needs to open to allow substrate entry to the active site. Initial interactions of the PI(3,4,5)P 3 head group include 4 P binding to N684, which prevents the formation of a hydrogen bond network seen in the R682 doubly bound L4-out conformation ( Figure 1F). With this hydrogen network disrupted L4 is more mobile (see e.g. Figure 1E) facilitating the switch to L4-in. We propose closure of L4 over the substrate as an important second step in catalysis, allowing R682 interactions with the substrate 3 P. With the substrate tightly bound and correctly positioned, catalysis proceeds. Once 5 P cleavage has occurred we propose that L4 needs to open again for product release. In absence of 3 P and 5 P interactions, reestablishment of the hydrogen network as seen in the R682 doubly bound L4-out conformation that orients the N684 sidechain towards H674 ( Figure 1F), might aid ejection of the product, hence closing the catalytic cycle. It has been previously proposed that the 3 P specificity of SHIP2 is conferred by R682 (Trésaugues et al., 2014). Our model shows that R682 in the L4-in conformation is well placed to interact with the 3 P of PI(3,4,5)P3 ( Figure 6A) and mutation of R682 confirms an important role for this residue in catalysis (Table 3). In other 5-Ptases the loop corresponding to L4 is significantly shorter and residues corresponding to R682 (K516 in INPP5B) and N684 (R518 in INPP5B) on L4 interact with the 4 P and belong to a set of residues defined as the P4 interacting motif (P4IM) (Trésaugues et al., 2014). Possibly, R682 in SHIP2 might also contribute to 4 P interactions, which could account for the observed PI(4,5)P 2 activity of SHIP2 (Elong Edimo et al., 2016;Nakatsu et al., 2010;Taylor et al., 2000), however, several reports demonstrate a clear preference for PI(3,4,5)P 3 (Chi et al., 2004;Pesesse et al., 1998;Trésaugues et al., 2014;Vandeput et al., 2007). Another difference in SHIP2 is the somewhat weaker Mg 2+ coordination compared to other 5-Ptases. An asparagine in INPP5B (N275) providing a Mg 2+ coordination is replaced by a glycine in SHIP2 (G434) and the coordination is likely provided by a water molecule ( Figure 6B). This water can be stabilized by R665, which appears important only in absence of active site stabilization by the C2 domain ( Figures 6C-D and 7D-E, Table 3).
Other head group interactions are conserved among 5-Ptases and the mechanism initiating 5 P attack is well described and shared among 5-Ptases as well as the AP endonucleases (Dlakic´, 2000;Whisstock et al., 2000Whisstock et al., , 2002. However, the exact mechanism of catalysis is still being debated. Different models have for AP nucleases proposed the involvement of either one or two static, or one moving metal ion (Beernink et al., 2001;Mol et al., 2000;Oezguen et al., 2007). A 2-metal hypothesis for AP nucleases is supported by a crystal structure of APE1, which has two Pb 2+ ions bound 5 Å  Table 3. All values are [PO 4 ] (in mM). Numbers (#i) above data indicate independent experiment number. Most kinetic parameters (k cat , K M ) are extracted from curves shown in Figure 6, for which Source data are available with this figure apart from each other, with one ion at the site seen in 5-Ptases (A-site) and the other at a more buried site (B-site) (Beernink et al., 2001). Similar to a recent structure of INPP5B (Mills et al., 2016), we observe in our Ptase-C2 D607A structure the Mg 2+ displaced on average by 1.3 Å from the A towards the B-site and a phosphate group shifted by 1.7 Å from the 5 P location towards the attacking water ( Figure 6-figure supplement  1B). As suggested by Mills et al. (2016) this likely represents the post-cleavage position for Mg 2+ and the 5 P, since there is no longer space for the attacking water. In contrast, for OCRL a free phosphate is observed at the pre-cleaved 5 P site and the Mg 2+ at the A-site (Trésaugues et al., 2014), hence there appear to be subtle differences in the post-cleavage positioning of the Mg 2+ and phosphate for different 5-Ptases. We note that the movement of the Mg 2+ seen in INPP5B and SHIP2 from the A towards the B-site is opposite to the B to A-site movement proposed to occur during catalysis in AP nucleases (Oezguen et al., 2007). SHIP1 shares an overall 45% sequence identity with SHIP2 (65% within the 5-Ptase and 43% in the C2 domain), suggesting that many of the mechanisms described here are conserved among the two enzymes. Residues at the Ptase-C2 domain interface are mostly conserved, suggesting that the interface as well as a possible communication to L2, which is fully conserved, are also present in SHIP1. The polar path is partially conserved with R649 replaced in SHIP1 by H632, D613/D615 are with D593 and E598 similar and R682 is substituted by K665. The equal length of L4 suggests that it will be able to sample a similar space and likely K665 provides the 3 P specificity in SHIP1.
In conclusion, our study provides important insights into how the C2 domain assists SHIP2 catalysis via a combination of membrane positioning, active site stabilization and allosteric signaling, which together is crucial for efficient cellular functioning (Figure 8). Since SHIP enzymes are highly implicated in disease, this will likely aid the design of novel strategies to target SHIP enzymes with high specificity.

Protein expression and purification
The experimental details regarding the cloning, expression and purification of human SHIP2 Ptase-C2 have been described in (Le Coq et al., 2016). The Ptase was cloned, expressed and purified following the same protocol, whereas the isolated C2 could not be purified due to insolubility of the protein.

Structure determination, Refinement, and Analysis
Crystallographic data for all protein crystals were measured at 100K, on beamlines ID23-2 (l = 0.873 Å ) and ID29 (l = 0.976 Å ) at European Synchrotron Radiation Facility (ESRF, Grenoble) and BL13-XALOC-BL13 (ALBA, Barcelona, for preliminary data collection) using a Pilatus 6M detector. XDS (Kabsch, 2010) was used for indexing and integration and AIMLESS (Evans and Murshudov, 2013) for scaling all diffraction data. The best dataset for SHIP2 Ptase-C2 WT diffracted to a resolution of 1.96 Å and reveals a P2 1 2 1 2 1 space group with unit cell parameters a = 136.04 Å , b = 175.84 Å , c = 176.89 Å . The Matthews coefficient is 2.54 Å 3 .Da À1 corresponding to 8 molecules in the asymmetric unit (asu) with a solvent content of 51.7% (Le Coq et al., 2016). The crystal structure of human SHIP2 Ptase model (PDB: 3NR8) was used as search probe for molecular replacement (MR) using Phaser (McCoy et al., 2007), which identified 8 clear solutions for the Ptase domain. MR using various available C2 structures as search probes failed. C2 domains where therefore built using ARP/wARP (Langer et al., 2008) and manual building with COOT (Emsley and Cowtan, 2004;Emsley et al., 2010). Translation/Libration/Screw (TLS) and Maximum likelihood of restrained refinement were performed using REFMAC5 (Murshudov et al., 1997). Final R-factors for SHIP2 Ptase-C2 WT are 18.0/20.8 (Rwork/Rfree). SHIP2 Ptase-C2 D607A crystallized in the same space group as wild type SHIP2 Ptase-C2, but Phaser (McCoy et al., 2007) was used to generate the initial model Figure 9 continued C2-PS interaction is based on PDB entry 1DSY and PI(3,4,5)P 3 interactions on the model shown in Figure 6A. (B) The catalytic cycle. Top: Docking of Ptase-C2 to the membrane orients via the rigid domain interface the Ptase active site towards its membrane substrate.
Step 2: Initial engagement of the substrate 4 P with N684 destabilizes the doubly bound L4-out conformation and initiates the switch to L4-in, which allows R682 interactions with the substrate 3 P. L2 opens and penetrates the membrane to interact with PI(3,4,5)P 3 lipid chains.
Step 3: With the substrate correctly positioned, catalysis proceeds and the cleaved 5 P and the Mg 2+ ion move towards the metal B-site (see text).
Step 4: Product interactions are weakened by L4 moving 'out', releasing R682 from the product 3 P. Further, N684 releases 4 P interactions by switching to bind H674 (see Figure 1F).
Step 5: The weakly bound product is released and a new substrate can be engaged. DOI: 10.7554/eLife.26640.033 Video 5. The SHIP2 catalytic cycle. The conformational changes between L4-out doubly bound, L4-out singly bound and L4-in conformations are shown as morphs between observed crystal structures and assembled in sequence according to the model of the catalytic cycle explained in Figure 9B. DOI: 10.7554/eLife.26640.034 since it produced better initial R values than directly using the SHIP2 Ptase-C2 WT model. Refinement was performed as described above. A Mg 2+ and a PO 4 3were placed in the active site of 6 molecules in the asu (monomers C-H). SHIP2 Ptase-C2 FLDD crystallized in two different space groups, one in P2 1 (a = 44.03, b = 81.12, c = 128.90, b = 92.85) with 2 molecules per asu and the other in I 2 (a = 43.74, b = 73.43, c = 158.0, b = 90.7) with 1 molecule per asu. The Ptase-C2 WT model was used in Phaser (McCoy et al., 2007) to provide initial phases. Initial rebuilding was performed with ARP/wARP (Langer et al., 2008) and further refinement was performed as described above. All data processing and structure refinement statistics are summarized in Table 1. All residues in the final models lay within the favored region of the Ramachandran plot with the exception of the model of SHIP2 Ptase-C2_D607A which has 0.2% Ramachandran outliers.

CD spectroscopy
Circular dichroism was used to assess and compare the overall protein fold of the different constructs. Far-UV CD spectra between 250 and 200 nm of protein at 0.25 mg/mL were run at 20˚C for each sample.

Thermal melting
Tm's were measured using a ThermoFluor assay. Protein at 1-10 mM was mixed with SYPRO Orange and subjected to a temperature gradient of 0.5˚/min from 20 to 95˚C and fluorescence recorded at 570 nm.

Lipid binding
For the PLO assay dipalmitoyl phosphatidylserine or dipalmitoyl phosphatidylcholine (Echelon Biosciences, Salt Lake City, UT) were spotted on a nitrocellulose membrane, dried and blocked with 5% skimmed milk. The membranes were then incubated for 1 hr with 2.5 mg/ml of GST fused SHIP Ptase or Ptase-C2 in blocking solution in presence or absence of 1 mM CaCl 2 , washed and developed using an HRP conjugated anti-GST antibody. For SPR measurements vesicles containing 30% PS (16:0; Echelon Biosciences) and 70% PC (from chicken egg; Avanti Polar Lipids, Alabaster, AL) or 30% PC (16:0, Avanti Polar Lipids) and 70% PC (chicken egg) were prepared at a final total lipid concentration of 1.5 mM. Organic solvent was removed by rotary evaporation for 1.5 hr at 45˚C. The lipid film was resuspended in SPR running buffer (20 mM HEPES, pH 7.5, 150 mM NaCl, 1 mM TCEP) and subjected to six cycles of freezethaw and passed 15 times through a membrane with 100 nm pore size, using a mini extruder (Avanti Polar Lipids). SPR experiments were performed on a Biacore X100 instrument (GE Healthcare, Chicago, IL). A L1 sensor chip was washed with two 1 min injections of a 2:3 (v/v) isopropanol: 50 mM NaOH solution at a flow rate of 10 ml/min and then coated by injecting PS containing vesicles in the active flow cell (Fc2) and PC vesicles in the reference flow cell (Fc1), for 15 min at a flow rate of 2 ml/min, followed by two 1 min injections of 10 mM NaOH solution at 10 ml/min to remove loosely bound vesicles and for stabilization. SHIP2 Ptase or Ptase-C2 proteins were injected for 2 min at a flow rate of 30 ml/min in running buffer. SPR responses plotted in Figure 2C-insert and Figure 2D correspond to response units (RUs) at 10 s after injection, when a relatively constant steady state binding phase is reached (dashed line in Figure 2C). Injections in presence of Ca 2+ were performed with 0.5 mM CaCl 2 . For all data shown in Figure 2C-D, buffer injections (with or without calcium, accordingly) are subtracted.

Enzyme kinetics
Activity measurements were performed using a Malachite Green phosphatase activity assay. Proteins were incubated for 2 min at 23˚C with the substrate in a reaction buffer containing 20 mM Hepes, 150 mM NaCl, 2 mM MgCl 2 , 1 mM TCEP, pH 7 in a total volume of 25 mL. The reaction was quenched by addition of 5 mL of 0.5 M EDTA, pH 8. Subsequently, 25 mL of the reaction was mixed with 100 mL of Malachite Green solution (Echelon Biosciences) and left to incubate for 15 min at room temperature and the optical density was measured at 620 nm. Each data point is measured at least in triplicates. For activity assays shown in Figures 4A-B 100 mM soluble PS-diC8 or PC-diC8 were included.
Cell assay HEK293 cells were maintained in Dulbecco's modified Eagle's medium (DMEM) containing 10% fetal bovine serum (FBS) and supplemented with antibiotics. The cells were transiently transfected with the empty vector (pOPINJ) or vector expressing GST fused full-length SHIP2 (WT, FLDD or DC2). The transient transfection was performed using polyethylenimine (Hsu and Uludag , 2012). After 48 hr in DMEM with 10% FBS the cells were harvested and levels of total Akt, phospho-T308 of Akt (Cell signaling, Danvers, MA, AB_2255933, AB_329827) and SHIP2 (Abcam, UK, AB_2686895) expression levels were assessed by immunoblotting. Quantifications were performed with ImageJ on eight independent experiments, each in triplicates (n = 24). Standard errors in Figure 8 are calculated using values of pAkt/Akt ratios and for Figure 8-figure supplement 1 SHIP2/Vinculin (AB_477629) ratios. The significance is calculated using a two-tailed unpaired Student t test assuming Gaussian distribution.

MD simulations
For apo simulations, the initial model was based on molecule B of the SHIP Ptase-C2 WT structure, which has a L4-out conformation. Missing residues in the linker (residues 731-746) were modelled with ModLoop (Fiser et al., 2000) and energy minimized. For substrate bound models L4 was replaced with an L4-in conformation from the SHIP2 Ptase (3NR8). The position of PI(3,4,5)P 3 , Mg 2+and catalytic waters were based on the INPP5B crystal structure bound to PI(3,4)P 2 (3MTC, adding the 5 P and 3 P) and based on AP endonucleases (Whisstock et al., 2002). Lipid parameters for IP 4 and PI(3,4,5)P 3 and a net charge of À5 for protein bound IP 4 /PI(3,4,5)P 3 were based on previous studies (Rosen et al., 2011;Slochower et al., 2013). For Ptase simulations the C2 and linker residues (733-874) were omitted. Hydrogens were added using the H++ server (Gordon et al., 2005). The systems were solvated in a cubic simulation cell of 80 Å edge length, constructed from an equilibrated box of TIP3P water molecules (Mahoney and Jorgensen, 2000). The size of the simulation cell was chosen such that a distance of at least 10 Å between the surface of the protein and the cell boundaries was maintained. The systems were neutralized with Na + /Cl -. All calculations were carried out with periodic boundary conditions. All complexes were minimized in three stages. In the first stage, the water molecules and the ions were energy minimized, while harmonic restraints of 1000 kJ/(molÁnm 2 ) were applied to the protein and the lipid; 500 steps of steepest descent and 500 steps of conjugate gradient minimization were carried out. In the second stage the lipid harmonic restrains were removed and another 500 steps of minimization were carried out. In the last stage, all restrains were removed and the entire system was minimized for 2000 steps. The resulting configurations were used for MD production: the lipid systems were simulated for a total of 300 ns each, and the non-lipid simulations for 3 Â 1.5 ms each (three independent simulations for each system). The first 50 ns of each simulation were considered equilibration time and omitted from the analysis. All minimization and MD production were performed using version 4.6 of the molecular simulations package GROMACS (Hess et al., 2008;Van Der Spoel et al., 2005) with the AMBER99SB*-ILDN force field (Lindorff-Larsen et al., 2010) with the dihedral corrections of Best Hummer (Best and Hummer, 2009). The isobaric-isothermal (NPT) ensemble was employed for all MD calculations at 300 K with a velocity-rescale thermostat (Bussi et al., 2007) and a time step of 2 fs.

RMSF and RMSD analysis
RMSF and RMSD were calculated on the C a of the 5-Ptase domain from the total 4.35 ms MD simulations, using the g_rmsf and g_rms tools from the GROMACS package. RMSDs were calculated per residue, using the L4-out structure as reference, and averaged over time. For plots shown in Figure 5-figure supplement 1A-F, RMSF and RMSD were calculated separately over snapshots, in which R682 was unbound, singly bound to either D613 or D615, or doubly bound to D613 and D615. Bonds were considered for minimal R682-D613/D615 distances below 4 Å (excluding hydrogens), hence including hydrogen bond and electrostatic bonding character.

Principal component analysis
PCA was performed on all C a atoms, applying g_covar and g_anaeig tools from GROMACS package on the full MD trajectories, to calculate, diagonalize and analyze the covariance matrix respectively. PCA identifies collective motions associated with the largest variance and the lowest frequencies shown in the first few eigenvectors, thus, we analyzed the first eight PC of each simulation.

Cluster analysis
Cluster analysis was performed on all non-hydrogen atoms using g_cluster with the gromos algorithm (Daura et al., 1999) and a RMSD cutoff of 0.15 nm. The structure with the smallest average distance to the others was written as the representative structure in each cluster.

Funder
Grant reference number Author The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.