Structural Insight on Functional Regulation of Human MINERVA Protein

MINERVA (melanoma invasion by ERK), also known as FAM129B, is a member of the FAM129 protein family, which is only present in vertebrates. MINERVA is involved in key signaling pathways regulating cell survival, proliferation and apoptosis and found upregulated in many types of cancer promoting invasion. However, the exact function of the protein remains elusive. X-ray crystallographic methods were implemented to determine the crystal structure of MINERVAΔC, lacking C-terminal flexible region. Trypsin digestion was required before crystallization to obtain diffraction-quality crystals. While the N-terminal pleckstrin homology (PH) domain exhibits the typical fold of PH domains, lipid binding assay indicates specific affinity towards phosphatidic acid and inositol 3-phosphate. A helix-rich domain that constitutes the rest of the molecule demonstrates a novel L-shaped fold that encompasses the PH domain. The overall structure of MINERVAΔC with binding assays and cell-based experiments suggest plasma membrane association of MINERVA and its function seem to be tightly regulated by various motifs within the C-terminal flexible region. Elucidation of MINERVAΔC structure presents a novel fold for an α-helix bundle domain that would provide a binding platform for interacting partners.


Introduction
In human proteome, the pleckstrin homology (PH) domain is present in nearly 300 proteins involved in diverse physiological processes. A typical PH domain comprises about 100 amino acid residues structured into a seven-stranded antiparallel β-sheet and a C-terminal α-helix [1]. The most well-known physiological property of PH domains is recognition of phosphoinositides with high affinity and specificity, which regulate localization of PH domain-containing proteins to and from intracellular membranes [2]. PH domains recognizing phosphoinositide display polar surface charge distribution around its β1-β2 region for accommodation of negatively charged head groups of phosphoinositides. On the other hand, PH domains also contribute to binding interaction partner proteins, including G-protein coupled receptors (GPCRs), GTPase exchange factors (GEFs), and focal adhesion kinase (FAK) [3,4]. Phosphotyrosine-containing peptides and polyproline helixes may bind on pockets created by the β-sheet and the C-terminal α-helix of the PH domain [5]. Considering its pivotal role in regulating protein subcellular localization, mutations within the PH domains can lead to aberrant cell signaling [4].
Here, we report the first crystal structure of MINERVA lacking the C-terminal flexible region, which contains the PH domain and a helix bundle domain with a novel fold. In addition, lipid binding ability of the PH domain is assessed for structural and functional analyses of MINERVA. This study shows structural basis for membrane association of MINERVA and its importance in governing multiple signaling pathways in cancer progression.

Functional Analyses of Physiological Role of MINERVA
Previous studies have shown that high levels of MINERVA expression were related to increased tumor cell invasion, delayed onset of apoptosis, and ultimately low survival rate of patients with a variety of cancer types [11,12,16]. The overall survival rate of breast, brain, and pancreatic cancer patients plotted against MINERVA expression level through Kaplan-Meier analysis from The Cancer Genome Atlas datasets (TCGA 2012 provisional) [21] shows the negative effect of high MINERVA expression in cancer prognosis ( Figure S1). Although it is known that MINERVA is overexpressed in various cancer cell types, the expression level was verified in a limited number of cell lines (HeLa, U251, LN229, PC14, HaCaT, or patient-derived non-small cell lung cancer (NSCLC) cells) [11,[14][15][16].
To obtain more comprehensive data on expression levels of MINERVA, cancer cell lines from different tissue origins were tested. Protein expression levels of MINERVA relative to GAPDH in the cancer cells were significantly elevated compared to that of the human embryonic kidney (HEK293FT) cells used as a negative control (Figure 1a). The expression data suggest a ubiquitous role of MINERVA in cancer progression.   ) were cultured in optimal media and cell lysates were prepared individually, then immunoblotted against MINERVA. GAPDH was used as a loading control. Immunoblotted proteins from experiments performed in triplicate were quantified using Image J software (NIH, Bethesda, MD, USA). Full blot can be found in Figure S4. (b) HeLa cells reverse-transfected with either siRNA against MINERVA (siMINERVA) or scrambled siRNA (siCTL) were seeded into a 96-well plate. Images acquired from was used as a loading control. Immunoblotted proteins from experiments performed in triplicate were quantified using Image J software (NIH, Bethesda, MD, USA). Full blot can be found in Figure S4. (b) HeLa cells reverse-transfected with either siRNA against MINERVA (siMINERVA) or scrambled siRNA (siCTL) were seeded into a 96-well plate. Images acquired from the IncuCyte ZOOM™ system at the indicated times were analyzed. Cell confluence was measured in triplicate wells for each time point. Error bars indicate the mean ± SEM for three independent experiments. * p < 0.05, ** p < 0.01. Protein extracts from siRNA-transfected cells in 72 h were analyzed for the MINERVA knockdown in western blot. Full blot can be found in Figure S5. (c) HeLa cells reverse-transfected with siRNA against MINERVA were incubated in culture medium with either H 2 O 2 or vehicle for 4 h. Cell death was assessed using Annexin V/propidium iodide (PI) staining by flow cytometry. The dead cell portion was measured based on the number of annexin V and PI single-stained cells. Error bars indicate the mean ± SEM for three independent experiments. ** p < 0.01. (d) Proteins extracted from siRNA-transfected cells were analyzed by western blotting using anti-PARP1 and anti-β-actin antibodies. The proform of PARP (116 kDa) and cleaved PARP (85 kDa) are indicated. Full blot can be found in Figure S6.
Next, we verified changes in cell growth and apoptosis rate in MINERVA knockdown cells. We observed that the cell growth rate of the HeLa cells treated with MINERVA siRNA was significantly reduced compared with cells treated with scrambled siRNA (siCTL) in IncuCyte ZOOM™ system ( Figure 1b). In addition, apoptosis rate in response to hydrogen peroxide (H 2 O 2 ) was markedly increased compared to that in siCTL-treated cells as analyzed by Annexin V and propidium iodide (PI) staining ( Figure 1c). We also observed more PARP cleavage in response to TNFα treatment in siMINERVA-treated cells, compared to those treated with siCTL (Figure 1d), which is consistent with the previous observations by Chen et al. [16]. These data indicate that MINERVA plays a critical role in cancer cell proliferation and survival, especially under oxidative stress condition.

MINERVA Crystallization and Structure Determination
The sequence analysis result of MINERVA using various web servers such as XtalPred, PSIPRED, and DISOPRED [22][23][24] predicted that the C-terminal region of MINERVA (Asp575-Phe746) is disordered, while the N-terminal PH domain and its flanking helix-rich regions are well-conserved among the members of the FAM129 family ( Figure S2). The domain composition and the known posttranslational modification (PTM) sites of MINERVA are denoted in Figure 2a. We successfully purified MINERVA FL and MINERVA ∆C (Figure 2b) and attempted crystallization of the proteins. Unfortunately, not only MINERVA FL but also MINERVA ∆C produced poorly diffracting rosette-shaped crystals, which were not fit for data collection. We suspected that internal disordered regions were hindering crystal formation and implemented limited proteolysis by trypsin. The limited trypsin treatment effectively digested MINERVA ∆C and yielded two separate fragments with molecular weights of approximately 14 kDa and 50 kDa, as confirmed by SDS-PAGE (Figure 2c, left). N-terminal sequencing identified the first five residues of the fragments as Gly 2 -Asp 3 -Val 4 -Leu 5 -Ser 6 and Ser 146 -Gly 147 -Ser 148 -Ala 149 -Pro 150 , respectively. The sequencing results corresponded with the prediction from PeptideCutter [25], that trypsin would cleave the protein after Lys145 residue. Other predicted trypsin cleavage sites were likely protected by protein folding, which prevented complete fragmentation of the protein.
The trypsin-digested MINERVA ∆C fragments co-eluted from the size-exclusion chromatography column (Figure 2c), meaning that the two fragments remained intact due to extensive interaction between them and the trypsin-digestion did not disrupt the overall folding of MINERVA ∆C . Trypsin digestion of MINERVA FL also gave fragments of the same sizes (data not shown), justifying our MINERVA ∆C construct design of the C-terminal flexible region removal. The trypsin digestion greatly aided production of large single crystals of MINERVA ∆C suitable for data collection, and hence was implemented for crystallizations of the native and SeMet-substituted MINERVA ∆C . sites, as well as the Keap1-binding motif 708 -DLGX7ETGE-721 . The sites of phosphorylation by epidermal growth factor receptor (EGFR) (Tyr593) and by ERK (Ser641, Ser646, Ser665, Ser681, Ser692, and Ser696) are marked with red and orange circles, respectively. The structured domain (residues Gly2-Arg560) is colored in rainbow spectrum from N-terminus to C-terminus. (b) Size exclusion chromatography-multiangle light scattering (SEC-MALS) results of MINERVA FL and MINERVA ΔC show monomeric states of the proteins. (c) SEC and SDS-PAGE results of trypsin-treated MINERVA ΔC show that the protein after trypsin-cleavage is divided into two fragments but retains its integrity in solution. (d) The overall MINERVA ΔC structure in cartoon representation colored in the same rainbow spectrum scheme as in (a). Secondary structure elements and the N-and C-termini are labeled. Unmodeled region, Nmyristoylation and lipid binding site on the pleckstrin homology (PH) domain are indicated. (e) Surface representations of the MINERVA ΔC structure showing surface electrostatic potentials. Front view is in the same orientation as MINERVA ΔC in (d). Top view shows a surface composed of positively charged residues for membrane association (yellow outline). Front and back views show two distinct crevices traversing the whole molecule that are approximately 13 Å wide and 50 Å or 60 Å long each. Bottom view shows an entrance to a large cavity inside the molecule (green outline).
We collected X-ray diffraction data from crystals of the trypsin-treated native and SeMetsubstituted MINERVA ΔC at 1.65 Å and 2.39 Å resolutions, respectively. Due to the absence of known similar structures, the phase was obtained from the SeMet-substituted MINERVA ΔC crystal by the single-wavelength anomalous diffraction (SAD) method. Using the SeMet-substituted structure as a molecular replacement (MR) model, we could successfully resolve the native MINERVA ΔC structure We collected X-ray diffraction data from crystals of the trypsin-treated native and SeMet-substituted MINERVA ∆C at 1.65 Å and 2.39 Å resolutions, respectively. Due to the absence of known similar structures, the phase was obtained from the SeMet-substituted MINERVA ∆C crystal by the single-wavelength anomalous diffraction (SAD) method. Using the SeMet-substituted structure as a molecular replacement (MR) model, we could successfully resolve the native MINERVA ∆C structure for residues Gly2-Lys145 and Leu152-Arg560 at higher resolution ( Table 1). The C-terminal region as well as Ser146-Ile151 could not be modeled due to high flexibility in these regions. As the last residue on the carboxy-terminal of the model is Arg560, the missing region Lys561-Ser574 was likely cleaved by trypsin. In an asymmetric unit (ASU) of P 2 1 2 1 2 1 space group, there was only one MINERVA ∆C . According to proteins, interfaces, surfaces, and assemblies (PISA) analysis [26], the largest surface area a MINERVA ∆C molecule shares with an adjacent one is 573.2 Å 2 , suggesting that MINERVA ∆C exists as a monomer. Oligomeric states of MINERVA FL and MINERVA ∆C in solution were assessed by measuring molecular mass using size exclusion chromatography-multiangle light scattering (SEC-MALS) method, supporting that both constructs of MINERVA exist as monomers in solution as seen in the crystal (Figure 2b).

Overall Structure of MINERVA ∆C
The overall structure model of a MINERVA ∆C monomer assumes a novel globular fold that spans approximately 70 Å by 70 Å by 60 Å in size. The structure of MINERVA ∆C could be divided into two domains, the PH domain (residues Ile69-Asn196) and a helix bundle domain comprising the rest of the molecule (Figure 2a), which exhibits a distorted L-shape where the PH domain is held at the corner of the helix bundle ( Figure 2d). The spatial arrangement of the two domains allows the N-terminus of the protein to be located adjacent to the PH domain in proximity, presumably for cooperative action of myristoylated Gly2 residue and the PH domain for membrane localization of MINERVA (Figure 2d). Surface representation of the protein reveals that MINERVA possesses a positively charged patch on its membrane-interacting face, which is mainly composed of positively charged residues Lys83, Lys84, Arg86, Lys104, Arg109, Arg114, Lys145, and Lys153 of the PH domain and Arg256, Lys258, and Lys260 of the helix bundle domain, presumably enabling stable interactions of MINERVA with the plasma membrane ( Figure 2e, top view). To our surprise, the surface representation revealed two distinct and characteristic crevices that are approximately 13Å wide and 50 Å or 60 Å long each on the front and the back, respectively (Figure 2e, front and back views), and a large cavity inside the molecule (Figure 2e, bottom view), which might serve as binding platforms for interaction partners.
The helix bundle domain is composed of 11 α-helices (α1-α2 and α6-α14) and six 3 10 helices (η1-η6) that are held together mostly by hydrophobic contacts. The overall architecture of the helix bundle domain resembles no other known structure, as analyzed by Dali structural similarity search algorithm [27], suggesting that it is of a novel fold. Sequence alignment of the three FAM129 family members shows that this specific fold, including the N-terminal PH domain, is well-conserved among the family members ( Figure S2). Interestingly, the helices α8-η2-α9 spanning 119 residues (Pro261-Glu379) are severely bent around the helix η2 (Gln340-Phe342) to encircle the entire protein, forming the characteristic L-shape (Figure 2d, green-to-yellow). Other helices in the bundle interact with the helices α8-α9 via hydrophobic contacts. Helices α2 and α7 are associated with the helix α8, while helices α10 through α14 interact with the helix α9 ( Figure 2d). The N-terminal PH domain is tightly held by the helix bundle domain that acts as a pincer-like scaffold, with a large contact area of 1573.4 Å 2 . Residues on helices α6, 7, 9, and 10 comprise an extensive network of hydrogen bonds and salt bridges with residues on the PH domain, such as Glu67 and His94, and Gln434, Arg192, Gln431, Asn195, and Gly231.
The N-terminal PH domain adopts a typical fold of a PH domain containing a central curved antiparallel β-sheet (β5↑-β6↓-β7↑-β1↓-β2↑-β3↓-β4↑) followed by a C-terminal α-helix, α5 ( Figure 2d). Sequence alignment of MINERVA PH domain with a representative PH domain from Pleckstrin (PDB ID: 2I5F) [28] showed moderate sequence identity and similarity of 22% and 35%, respectively. Moreover, structure-based sequence alignment of the PH domains of MINERVA and Pleckstrin revealed that the core secondary structures including the characteristic antiparallel β-sheet are well conserved ( Figure 3a) with root-mean-square deviation (RMSD) of 1.30 Å over 91 equivalent C α positions. On the other hand, there are two additional α-helices α3 and α4 inserted between β3 and β4, and β5 and β6, respectively, in the MINERVA PH domain (Figure 3a,b, left). A major difference between the PH domains of MINERVA and Pleckstrin arises from a loop region connecting the helix α4 and the strand β6 (residues Gly136-Phe158) of MINERVA, which is unusually long and disordered ( Figure 3a). Residues Ser146-Pro151 within this region could not be modeled in our structure due to lack of electron densities. The corresponding region in Pleckstrin (residues Asn303-Glu311) was shorter but also disordered. Incidentally, the trypsin cleavage site (after Lys145) is located within the α4-β6 loop, meaning that this loop is intrinsically disordered and exposed to solvation.

PH Domain of MINERVA Recognizes Phosphatidic Acid and Phosphatidylinositol 3-Phosphate
PH domains are well-known for their ability to recognize phosphatidylinositols [2]. Structure-based sequence alignment of the PH domains from MINERVA and Pleckstrin in complex with d-myo-inositol 1,2,3,5,6-pentakisphosphate (IP 5 ) (Figure 3a) suggests that putative binding site for lipids on MINERVA PH domain is composed of residues Gln77, Gln79, Lys84, Arg86, Arg88, Arg114, and Arg168, which are mainly located on strands β1 and β2, forming a positively charged face of its ligand binding pocket (Figure 3b). On Pleckstrin, residues Lys253, His256, Arg257, Arg258, Lys262, Arg264, and Tyr277 establish grossly positively charged surface that form extensive hydrogen bond network with the bound IP 5 . Among these residues, His256-Arg258 seemed to be the major contributors in stabilizing the bound IP 5 by encapsulating the ligand with their long and positively charged sidechains. MINERVA possesses negatively charged amino acid residues (Glu80-Asp81-Ser82) instead on the positions corresponding to His256-Arg258 of Pleckstrin (Figure 3a), which form a negatively charged face within the ligand binding site (Figure 3b, right). The oppositely charged faces within the ligand binding site would restrict types of ligands that can bind in the MINERVA PH domain to those with fewer phosphates at their head groups. Furthermore, efficient IP 5 binding on Pleckstrin is facilitated by an absence of side chain on Gly255 residue, while MINERVA possesses Gln79 on the corresponding position (Figure 3b, left and Figure S3a), which would cause steric hindrance to incoming ligand.
Since MINERVA is recruited to plasma membrane under certain growth-inhibited conditions [6], affinity towards the membrane lipids of the PH domains was expected. In fact, phosphatidylinositol (3)-phosphate (PI3P) was recognized as the innate ligand of the MINERVA PH domain [29]. We implemented lipid dot blot assay using PIP Strips™ to determine which lipids are recognized by MINERVA FL and MINERVA ∆C . The dot blot assay revealed that both MINERVA FL and MINERVA ∆C recognized phosphatidic acid (PA) and PI3P, but to different extents (Figure 3c). MINERVA FL showed significantly higher affinity towards PA over PI3P, while MINERVA ∆C preferred PI3P. As shown in Figure 3b, we expected the lipid binding pocket of MINERVA to possess specific affinity towards lipids with few phosphates at their head groups due to oppositely charged faces within the ligand binding site and the steric hindrance elicited by side chain of the Gln79. PA and PI3P both satisfy these criteria, though it is not clear how other similar lipids such as PI4P and PI5P are discriminated against. Nevertheless, the lipid dot blot assay results suggest that the C-terminal flexible region (residues Asp575-Phe746) could play a role in switching specificity of the N-terminal PH domain between PA and PI3P.
Next, we attempted to solve the structure of MINERVA ∆C in complex with PA or PI3P to elucidate the mechanism of ligand recognition. However, we could only obtain structures of ligand-unbound forms of MINERVA ∆C . We instead performed computational Glide-induced fit docking [30] of the head groups of PA or PI3P against the MINERVA PH domain to predict their binding modes and calculated binding free energy (∆G MM-GBSA ). As expected, both PA and PI3P were modeled well at the ligand binding site of the PH domain with binding free energies of −51.9 kcal mol −1 and −36.8 kcal mol −1 , respectively, suggesting more stable binding of PA with the MINERVA PH domain. The docking results showed that the phosphate groups on PA and PI3P both establish multiple polar interactions with the positively charged residues (Lys84, Arg86, Arg88, Arg114, and Arg168) at the ligand binding site (Figure 3d). Interestingly, the negatively charged residues Glu80-Asp81 are flipped away from the ligand binding site to better accommodate the bound ligands in this model.

Functional Analyses of N-and C-Terminal Regions of MINERVA FL
In the above sections, we showed by solving protein structure that MINERVA possesses an N-terminal PH domain, and that it recognizes PA and PI3P. To assess the physiological significance of the PH domain, we generated an additional N-terminal truncation construct (MINERVA ∆N ; residues Glue307-Phe746) and compared its subcellular localization in HeLa cell along with MINERVA FL and MINERVA ∆C . The MINERVA constructs were FLAG (DYKDDDDK)-tagged for detection and their localizations were monitored by immunofluorescence analysis. In accordance with the previous observations by Chen et al. [16], MINERVA was localized at plasma membrane. Interestingly, both MINERVA constructs retaining the N-terminal PH domain (MINERVA FL and MINERVA ∆C ) were mostly localized at plasma membrane, whereas MINERVA ∆N was localized in cytoplasm as either diffused or puncta forms (Figure 4). The immunofluorescence data emphasizes the importance of the PH domain on MINERVA for membrane localization.

Functional Analyses of N-and C-terminal Regions of MINERVA FL
In the above sections, we showed by solving protein structure that MINERVA possesses an Nterminal PH domain, and that it recognizes PA and PI3P. To assess the physiological significance of the PH domain, we generated an additional N-terminal truncation construct (MINERVA ΔN ; residues Glue307-Phe746) and compared its subcellular localization in HeLa cell along with MINERVA FL and MINERVA ΔC . The MINERVA constructs were FLAG (DYKDDDDK)-tagged for detection and their localizations were monitored by immunofluorescence analysis. In accordance with the previous observations by Chen et al. [16], MINERVA was localized at plasma membrane. Interestingly, both MINERVA constructs retaining the N-terminal PH domain (MINERVA FL and MINERVA ΔC ) were mostly localized at plasma membrane, whereas MINERVA ΔN was localized in cytoplasm as either diffused or puncta forms (Figure 4). The immunofluorescence data emphasizes the importance of the PH domain on MINERVA for membrane localization. Although the C-terminal flexible region was excluded from structural study due to its high flexibility, this region has been a focus of attention for functional regulation of MINERVA by multiple reports [6,12,15]. We examined interaction between MINERVA and Keap1, by co-transfecting each of FLAG-tagged MINERVA constructs (MINERVA FL , MINERVA ΔN , and MINERVA ΔC ) with HA-tagged Keap1 and subsequently analyzing co-immunoprecipitation (co-IP). MINERVA FL and MINERVA ΔN were co-immunoprecipitated with Keap1, whereas MINERVA ΔC , which is missing the Keap1interacting motif "DLGXnETGE" located in the C-terminal flexible region [12], failed to interact with Keap1 (Figure 5a). Following the co-IP experiment, we performed surface plasmone resonance (SPR) analysis between the two proteins, with MINERVA FL as ligand and Keap1 as analyte. Keap1 showed interaction with MINERVA FL in a concentration dependent manner, and a micromolar KD was obtained from kinetics analysis (Figure 5b). Although the C-terminal flexible region was excluded from structural study due to its high flexibility, this region has been a focus of attention for functional regulation of MINERVA by multiple reports [6,12,15]. We examined interaction between MINERVA and Keap1, by co-transfecting each of FLAG-tagged MINERVA constructs (MINERVA FL , MINERVA ∆N , and MINERVA ∆C ) with HA-tagged Keap1 and subsequently analyzing co-immunoprecipitation (co-IP). MINERVA FL and MINERVA ∆N were co-immunoprecipitated with Keap1, whereas MINERVA ∆C , which is missing the Keap1-interacting motif "DLGX n ETGE" located in the C-terminal flexible region [12], failed to interact with Keap1 ( Figure 5a). Following the co-IP experiment, we performed surface plasmone resonance (SPR) analysis between the two proteins, with MINERVA FL as ligand and Keap1 as analyte. Keap1 showed interaction with MINERVA FL in a concentration dependent manner, and a micromolar K D was obtained from kinetics analysis (Figure 5b).

Discussion
Here, we report the first crystal structure of human MINERVA ΔC , a protein found upregulated in many types of cancer with an unclear function. Sequence analysis predicted intrinsically disordered C-terminal region containing multiple phosphorylation sites and a Keap1-interaction motif, which was excluded in our construct for structure determination. The structure of MINERVA ΔC comprises the N-terminal PH domain and an α-helix bundle domain (residues Gly2-Glu49 and Glu205-Arg560; Figure 2a,d) of a novel fold. Flexible α3-β6 loop within the PH domain was cleaved by trypsin for generation of diffraction-quality crystals. To our surprise, intermolecular interaction analysis by the PISA program suggested that the cleavage allowed better stacking of the MINERVA ΔC molecules in crystal lattice. Helix α8′ (residues Ser292′-Arg306′) from an adjacent MINERVA ΔC molecule approaches the helix α4 region exposed by the trypsin cleavage, forming several intermolecular interactions with residues Asp129-Asn137 and Lys260-Gln265 ( Figure S3b, gray dashed ribbon). These residues forming the interactions would have been concealed by the α3-β6 loop before trypsin treatment, explaining poor diffraction quality from crystals of untreated MINERVA ΔC we had obtained.
Our structure has revealed that the N-terminal myristoylated residue Gly2 is positioned adjacent to the PH domain to constitute a positively charged membrane-interacting face for cooperative action of anchoring MINERVA to lipid membranes (Figure 2d,e). Previous reports, as well as our observation on the localization of MINERVA, showed that the protein is dispersed in cytoplasm as puncta form in exponentially growing HeLa cells, whereas it was recruited to cell-cell contacts whenever the cells became confluent [16]. Contact inhibition is the major factor in switching cellular functions from growth and proliferation to migration [31], and we investigated lipid binding ability of the PH domain on MINERVA to gain insight on the function of the protein. Lipid dot-blot assay indicated that MINERVA FL predominantly binds to PA, whereas MINERVA ΔC showed higher affinity towards PI3P (Figure 3c. We speculate that negatively charged residues (Glu80-Ser82) on the β1-β2 loop at the ligand binding site seem to restrict binding of lipids with multiple phosphate groups, while the side chain of Gln79 only allows binding of lipids with smaller head groups. Although PH domains are best known for their abilities to bind phosphatidylinositols [2], the lipid dot-blot assay results suggest that PA can also be recognized by the PH domain. The difference in binding affinity between MINERVA FL and MINERVA ΔC implies that the C-terminal flexible region contributes to the ligand selectivity of the N-terminal PH domain. Despite the importance of PA in cell signaling [32], domains or motifs responsible for recognizing PA have not been clearly defined. Studies on PAbinding proteins commonly noticed a short stretch of positively charged peptides responsible for the

Discussion
Here, we report the first crystal structure of human MINERVA ∆C , a protein found upregulated in many types of cancer with an unclear function. Sequence analysis predicted intrinsically disordered C-terminal region containing multiple phosphorylation sites and a Keap1-interaction motif, which was excluded in our construct for structure determination. The structure of MINERVA ∆C comprises the N-terminal PH domain and an α-helix bundle domain (residues Gly2-Glu49 and Glu205-Arg560; Figure 2a,d) of a novel fold. Flexible α3-β6 loop within the PH domain was cleaved by trypsin for generation of diffraction-quality crystals. To our surprise, intermolecular interaction analysis by the PISA program suggested that the cleavage allowed better stacking of the MINERVA ∆C molecules in crystal lattice. Helix α8 (residues Ser292 -Arg306 ) from an adjacent MINERVA ∆C molecule approaches the helix α4 region exposed by the trypsin cleavage, forming several intermolecular interactions with residues Asp129-Asn137 and Lys260-Gln265 ( Figure S3b, gray dashed ribbon). These residues forming the interactions would have been concealed by the α3-β6 loop before trypsin treatment, explaining poor diffraction quality from crystals of untreated MINERVA ∆C we had obtained.
Our structure has revealed that the N-terminal myristoylated residue Gly2 is positioned adjacent to the PH domain to constitute a positively charged membrane-interacting face for cooperative action of anchoring MINERVA to lipid membranes (Figure 2d,e). Previous reports, as well as our observation on the localization of MINERVA, showed that the protein is dispersed in cytoplasm as puncta form in exponentially growing HeLa cells, whereas it was recruited to cell-cell contacts whenever the cells became confluent [16]. Contact inhibition is the major factor in switching cellular functions from growth and proliferation to migration [31], and we investigated lipid binding ability of the PH domain on MINERVA to gain insight on the function of the protein. Lipid dot-blot assay indicated that MINERVA FL predominantly binds to PA, whereas MINERVA ∆C showed higher affinity towards PI3P (Figure 3c). We speculate that negatively charged residues (Glu80-Ser82) on the β1-β2 loop at the ligand binding site seem to restrict binding of lipids with multiple phosphate groups, while the side chain of Gln79 only allows binding of lipids with smaller head groups. Although PH domains are best known for their abilities to bind phosphatidylinositols [2], the lipid dot-blot assay results suggest that PA can also be recognized by the PH domain. The difference in binding affinity between MINERVA FL and MINERVA ∆C implies that the C-terminal flexible region contributes to the ligand selectivity of the N-terminal PH domain. Despite the importance of PA in cell signaling [32], domains or motifs responsible for recognizing PA have not been clearly defined. Studies on PA-binding proteins commonly noticed a short stretch of positively charged peptides responsible for the interaction [33]. Sequence analysis of the C-terminal flexible region of MINERVA revealed a similar motif 615 -KRRRAK-620 (Figure 6a), which might be responsible for enhancing ligand specificity of the PH domain on MINERVA FL towards PA. The facts that MINERVA ∆C showed higher affinity against PI3P over PA, and that PI3P was recognized as the innate ligand of MINERVA PH domain alone [29], hint the importance of this positive residue stretch in the C-terminal flexible region for PA recognition. In cells, PI3P is predominantly populated on endosomes, for sorting out protein targets for endocytosis [34]. Moreover, N-terminal myristoylation on Gly2 residue strongly suggests that cytoplasmic fraction of MINERVA is in fact endosome-associated, and the cooperative action of the N-terminal myristic acid and the PH domain is supported by our structure (Figure 2d). We speculate that intracellular localization of MINERVA is regulated by the PH domain, depending on the level of cell confluency [16]. Once cells become confluent, MINERVA would be recycled back to the plasma membrane for activation of cell migratory signaling pathways via, for example, FAK phosphorylation [11]. interaction [33]. Sequence analysis of the C-terminal flexible region of MINERVA revealed a similar motif 615 -KRRRAK-620 (Figure 6a), which might be responsible for enhancing ligand specificity of the PH domain on MINERVA FL towards PA. The facts that MINERVA ΔC showed higher affinity against PI3P over PA, and that PI3P was recognized as the innate ligand of MINERVA PH domain alone [29], hint the importance of this positive residue stretch in the C-terminal flexible region for PA recognition. In cells, PI3P is predominantly populated on endosomes, for sorting out protein targets for endocytosis [34]. Moreover, N-terminal myristoylation on Gly2 residue strongly suggests that cytoplasmic fraction of MINERVA is in fact endosome-associated, and the cooperative action of the N-terminal myristic acid and the PH domain is supported by our structure (Figure 2d). We speculate that intracellular localization of MINERVA is regulated by the PH domain, depending on the level of cell confluency [16]. Once cells become confluent, MINERVA would be recycled back to the plasma membrane for activation of cell migratory signaling pathways via, for example, FAK phosphorylation [11].  The "DLGXnETGE" motif within the C-terminal flexible region was pointed out to be responsible for lifting Keap1-mediated Nrf2 degradation [12]. Keap1-Nrf2 is a well-known defense system against cellular oxidative stress which is frequently observed in cancer progression. Nrf2 is ubiquitinated and degraded by interaction with Keap1, a scaffold protein for Cul3-containing E3 ubiquitin ligase, whereas under the conditions of oxidative stress, Nrf2 is disrupted from interaction with Keap1 and then is translocated into the nucleus for transcriptions of antioxidant genes [35,36]. MINERVA competes with Nrf2 for Keap1-binding through the C-terminal region, resulting in Nrf2 activation [12]. Our SPR analysis of MINERVA FL and Keap1 provides direct evidence for physical association between the two proteins with micromolar binding affinity (Figure 5b). Moreover, MINERVA ∆C failed to interact with Keap1, implying the importance of the C-terminal region of MINERVA for cancer progression through Nrf2 mediated regulation (Figure 5a). On the other hand, the effect of phosphoserines on MINERVA function was elusive since the first report of MINERVA as a target of ERK phosphorylation [6]. We noticed that the sequences around these serine residues exhibited a tandem proline-rich consensus motif "(PESPP)PASP", where underlined serine residues are the targets of phosphorylation and residues in brackets are less conserved (Figure 6a). This motif is reminiscent of a Src specific SH3-domain binding motif "(PXXPP)PψXP", where ψ and X represent hydrophobic and any amino acid residues, respectively [37,38]. Thus, it would be interesting to further investigate whether the interaction with Src presents a possible role of MINERVA in regulation of cell motility. Overall, the C-terminal flexible region seems to harbor multiple sequence motifs and PTM sites for interaction with other proteins for regulation of signal transduction (Figure 6a,b). We can hypothesize that MINERVA recognizing PA or PI3P by the N-terminal PH domain, as well as EGF-mediated ERK phosphorylation of the C-terminal "(PESPP)PASP" motifs, might allow spatial and temporal regulation of Src activation to promote cell proliferation and migration.
We present the first structure of the globular domain from MINERVA that displays distinct crevices and a cavity shown from surface representations (Figure 3e). We suggest that these characteristic crevices and the cavity provide a binding platform for MINERVA interacting partners. The widths and lengths of the crevices seem adequate for accommodating 20-30 amino acid long loops each. Although a certain region within the C-terminal flexible region may be the inherent occupant of the crevices, we excluded such possibility as crystallization attempts with trypsin-digested MINERVA FL produced mostly the same structure model with no additional electron densities at the crevices. However, it should be noted that the presented structure of MINERVA ∆C might not necessarily represent the in vivo structure since it lacks the C-terminal flexible region harboring post-translational modifications sites and motifs for interaction with other proteins. Binding of interaction partners often elicits conformational changes required for protein function [39]. Interactome analysis indicates that interacting partners of MINERVA includes not only signaling molecules discussed above (EGFR, H-Ras, ERK, and Keap1), but also proteins responsible for endosomal sorting, such as SNX2 and SNX6, and VPS26A, VPS29, and VPS35 (https://thebiogrid.org/122328) [40]. The fact that E-cadherin was suggested as another binding partner of MINERVA hints at the role of MINERVA in cell migration upon confluency [41]. These interacting partners may exploit the crevices and cavity present on the structured domain of MINERVA ∆C , since the binding regions of the other signaling molecules are focused on the C-terminal flexible region. Regarding its possible influence over multiple signaling pathways regulating cell growth, proliferation, motility, oxidative stress, and survival and apoptosis, the structural elucidation of MINERVA would present valuable information on cancer progression and therapy.

Cell Lines
All human cell lines were purchased from the American Type Culture Collection (ATCC; Manassas, VA, USA) and were maintained at 5% CO 2

DNA Construct and siRNA
For the construction of the FLAG-MINERVA, a PCR-amplified DNA fragment encoding MINERVA was inserted between the NotI and XbaI sites of the pcDNA3 or pcDNA3.1 vector. The deletion mutants of the FLAG-MINERVA (MINERVA ∆C ; residues Met1-Ser574 and MINERVA ∆N ; residues Thr307-Phe746) were constructed with NotI and XbaI sites of the pcDNA. siRNA targeting the respective genes of interest and negative control siRNA (non-targeting pool) were purchased from Genolution Inc. (Seoul, Korea). The following siRNA sequences were used for the indicated target genes; siMINERVA_#1 5 -UCACGGACAUGAACCUGAACGUCAU-3 and siMINERVA_#2 5 -CAGUAUGGCGUGGCUCUCUUCAACA-3 .

Immunoprecipitation
MINERVA and its truncated forms were tagged with a FLAG epitope and Keap1 was tagged with a human influenza HA epitope. Each combination of the proteins was co-expressed in HEK293T cells.

Cell Proliferation and Viability Assays
siRNA-transfected cells were seeded at 4 × 10 3 per well in a 96-well plate, which was placed in the IncuCyte ZOOM™ system (Essen BioScience, Ann Arbor, MI, USA). After incubation for the indicated times live-cell images were obtained using a 10× objective lens (four images per well) within the instrument, and cell density was analyzed using IncuCyte ZOOM 2016B software.
Cell viability was determined by annexin V and PI staining following standard protocols at the indicated time periods (556547, BD Biosciences, San Jose, CA, USA). Cells negative for both annexin V and PI were considered live cells. The proportion of dead cells was measured based on the number of annexin V and PI single and both-stained cells. The fluorescence of stained cells was detected using the LSRFortessa fluorescence activated cell sorting (FACS) analyzer (BD Biosciences).

Western Blotting
Cells were harvested in ice-cold radioimmunoprecipitation assay (RIPA) lysis buffer (50 mM Tris-HCl (pH 7.4), 150 mM NaCl, 1% NP-40, 0.5% Na-deoxycholate, 0.1% SDS, and 1 mM EDTA) containing protease inhibitor cocktail (Roche Applied Bioscience, Penzberg, Germany) and phosphatase inhibitor (Sigma-Aldrich, St. Louis, MO, USA). Soluble lysate fractions were isolated by centrifugation at 20,000× g, for 20 min at 4 • C and quantified using the Pierce bicinchoninic acid (BCA) Protein Assay kit (Thermo Fisher Scientific, Waltham, MA, USA). Samples were resolved by sodium dodecyl sulfate-polyacrylamide gel electrophoresis (SDS-PAGE) using equal concentrations of protein and transferred to PVDF membranes. The membranes were blocked with 5% skim milk and then probed with the indicated primary and secondary antibodies following standard protocols. Immunoblotted proteins were quantified using ImageJ software (NIH, Bethesda, MD, USA). Data are expressed as the mean ± standard deviation (SD) or standard error of the mean (SEM) which are from at least three independent experiments. Statistical significance was calculated using Student's t-test in GraphPad Prism. A value of p < 0.05 was considered statistically significant (* p < 0.05; ** p < 0.01).

Immunofluorescence Analysis
To determine the localization of MINERVA, immunostaining against MINERVA was performed using Hela cells. After they were plated onto glass coverslips in complete medium for 18 h, cells were incubated in control (3 × FLAG only) and MINERVA domain construct mixture treatment for 48 h. For immunostaining, cells were rinsed with ice-cold PBS, fixed with 3.7% formaldehyde for 20 min, and permeabilized with 0.3% Triton X-100 in PBS (PBS-T) for 5 min. After rinsing with PBS-T, cells were blocked with blocking buffer (5% BSA in PBS-T) for 1 h, incubated with primary antibodies against MINERVA or FLAG in blocking buffer for 16 h in 4 • C, washed three times with PBS-T and then incubated with secondary antibodies conjugated with fluorescent probe in PBS-T for 1 h at room temperature. After three PBS-T washes, cells were stained with DAPI (4 ,6-diamidino-2-phenylindole) for 10 min and mounted on microscope slides. A confocal fluorescence microscope (Zeiss LSM 780, Oberkochen, Germany) was used for imaging and analysis.
The transformed cells were grown for 16 h in Luria Broth media without induction to be used as seeds for the scale-up culture. The uninduced E. coli seeds containing MINERVA plasmids were inoculated into a scale-up culture media at a ratio of 1:50. The cells in scale-up culture were grown in Luria Broth media and induced with 0.5 mM isopropyl β-D-1-thiogalactopyranoside, followed by further incubation at 20 • C for 16 h. Harvested cells were lysed by a cell sonicator (SONICS) in a lysis buffer containing 20 mM Tris-HCl (pH 7.5), 500 mM NaCl, 35 mM imidazole, and 1 mM PMSF. Cell debris was removed by centrifugation at 35,000× g for 50 min at 4 • C, then the supernatant was filtered with 0.22 µm filter to remove cell debris and any aggregated proteins. The filtered supernatant was applied onto a HiTrap Chelating HP column (GE Healthcare, Chicago, IL, USA) equilibrated with the lysis buffer for affinity chromatography. Unbound or weakly bound proteins were removed from the HiTrap Chelating HP column with six column volume elutions of the washing buffer containing 20 mM Tris-HCl (pH 7.5), 500 mM NaCl, and 50 mM imidazole. The retained proteins were eluted with a gradually increasing addition of a buffer containing 20 mM Tris-HCl (pH 7.5), 500 mM NaCl, and 500 mM imidazole. The eluted protein samples were further purified by size-exclusion chromatography with HiLoad 16/600 Superdex200 pg column (GE Healthcare, Chicago, IL, USA) equilibrated with a buffer containing 20 mM Tris-HCl (pH 8.0) and 200 mM NaCl. The purification steps for SeMet-substituted MINERVA ∆C were as above except that the buffer for the final size exclusion chromatography additionally contained 1 mM tris(2-carboxyethyl)phosphine (TCEP).
4.9. Cloning, Expression, and Purification of Keap1 321-609 for Affinity Assay N-and C-terminally truncated Keap1 protein (residues Ala321-Thr609) containing the Kelch domain was amplified using PCR and cloned into the pET-28a(+) vector to contain an N-terminal His 6 -tag. The plasmid containing Keap1 321-609 was transformed into Rosetta 2(DE3)pLysS E. coli strain. The purification steps for Keap1 321-609 were as above except that the buffer for the final size exclusion chromatography contained 20 mM Tris-HCl (pH 7.5) and 5 mM dithiothreitol (DTT).

Limited Trypsin Digestion of MINERVA Proteins
Native and SeMet-substituted MINERVA ∆C were subjected to digestion by trypsin (Sigma-Aldrich, St. Louis, MO, USA) by incubating purified MINERVA ∆C at a 5000:1 trypsin ratio at 4 • C for 20 h prior to crystallization in order to improve diffraction quality of crystals. Digested proteins were further purified by size-exclusion chromatography with HiLoad 16/600 Superdex200 pg column equilibrated with a buffer containing 20 mM Tris-HCl (pH 8.0) and 200 mM NaCl. The purified trypsin-treated MINERVA ∆C proteins were concentrated to 4.8 mg mL −1 for crystallization.

Crystllization, Data Collection, and Structure Determination
Crystals of the trypsin-treated SeMet-substituted MINERVA ∆C were grown at 22 • C by the hanging drop vapor diffusion method by mixing equal volumes of the protein at 4.8 mg mL −1 and a crystallization solution containing 0.2 M sodium tartrate and 20% (w/v) polyethylene glycol (PEG) 3350. Crystals of the trypsin treated native MINERVA ∆C were grown at 22 • C by the same methods with a crystallization solution containing 0.2 M lithium acetate and 20% (w/v) PEG 3350. Crystals of diffraction quality were cryoprotected with the reservoir solution containing additional 25% (v/v) glycerol and flash-cooled in a nitrogen gas stream at 100 K. Diffraction data from the SeMet-substituted and the native crystals were collected up to resolutions of 2.39 Å and 1.65 Å, respectively. Raw X-ray diffraction data were indexed and scaled using the HKL-2000 program suite or the XDS program package [42,43]. Single-wavelength anomalous diffraction (SAD) phases of the SeMet-substituted MINERVA ∆C were initially calculated with AUTOSOL in the PHENIX software suite [44] and further improved by the automatic model building program RESOLVE [45], resulting in an initial model. The initial model was further refined to the final model using iterative cycles of model building with Coot [46] and subsequent refinement with Refmac5 in the CCP4 suite [47] and phenix.refine [48]. The crystal structure of the native MINERVA ∆C was determined by molecular replacement with the MolRep program [49], using the refined structure of SeMet-substituted MINERVA ∆C as a phasing model. Validation of crystal structures was implemented with MolProbity [50] and the Research Collaboratory for Structural Bioinformatics (RCSB) Protein Data Bank (PDB) validation server. Statistics for the data collection and refinement are summarized in Table 1. The coordinate and structure factor of the native MINERVA ∆C have been deposited in the PDB (http://www.rcsb.org) under accession code 7CTP.

Size-Exclusion Chromatography with Multi-Angle Light Scattering (SEC-MALS)
The oligomeric states of the recombinant MINERVA FL and MINERVA ∆C proteins were assessed by SEC-MALS experiments using an ÄKTA fast protein liquid chromatography (FPLC) system (GE Healthcare, Chicago, IL, USA) connected to a Wyatt DAWN Heleos II MALS instrument and a Wyatt Optilab T-rEX differential refractometer. A Superdex 200 Increase 10/300 GL gel-filtration column (GE Healthcare, Chicago, IL, USA), which was pre-equilibrated with buffer A containing 20 mM Tris-HCl (pH 8.0) and 200 mM NaCl, was normalized using ovalbumin. The proteins (2 mg) were injected at a flow rate of 0.5 mL min −1 . Data were analyzed using the Zimm model for fitting static light-scattering data and were graphed using Easy Analytic Software Inc. (EASI) graph with a UV peak in the ASTRA 6 software (Wyatt, Santa Barbara, CA, USA).

In Silico Docking Experiment
Induced fit docking (IFD) experiment (Schrödinger Suite Induced Fit Docking protocol; Glide version 7.4, Prime version 4.7, Schrödinger, LLC, New York, NY, USA) was implemented in order to predict the binding mode of lipids to the PH domain of MINERVA. The MINERVA ∆C structure was prepared for docking calculations using the Protein Preparation Wizard implemented in Maestro (Schrödinger, LLC, New York, NY, USA). The structure of MINERVA ∆C was built in Maestro 8.5, and the possible conformations of the lipids were generated using LigPrep (version 4.1, Schrödinger, LLC, New York, NY, USA). The head groups of lipids were docked onto MINERVA ∆C using the following steps. (i) The receptor grid of MINERVA ∆C was defined as an enclosing box at the centroid of the key amino acid residues in the binding pocket. (ii) In the initial Glide docking stage, a potential docking with van der Waals radius scaling of 0.5 for the protein and ligand was performed retaining a maximum number of 20 poses per ligand. (iii) Residues within 5.0 Å of ligand poses were kept free to move in the Prime refinement step, and the side chains were further optimized. (iv) Poses within 30 kcal mol −1 of the energy cutoff in the previous step were re-docked using Glide XP. (v) The binding free energy (∆G bind ) for each output pose was computed using Prime-MMGBSA (version 3.0).

Lipid Dot Blot Assay
The PIP strips (Echelon Biosciences, Salt Lake City, UT, USA) were blocked with phosphate-buffered saline containing 0.5% (v/v) Tween-20 (Amersham Biosciences, UK) and 3% (w/v) bovine serum albumin (BSA; Bovogen, East Keilor, Australia) for 1 h and incubated with MINERVA FL or MINERVA ∆C for 1 h. MINERVA FL bound strips were treated with α-MINERVA rabbit antibody (5122; Cell Signaling Technology, Danvers, MA, USA) for 1 h then with HRP-conjugated α-rabbit IgG goat antibody (A120-101P; Bethyl Laboratories, Montgomery, TX, USA) for 1 h. MINERVA ∆C bound strips were treated with HRP-conjugated α-His 6 -tag antibody (Santa Cruz Biotechnology, Dallas, TX, USA). The strips were washed with the blocking buffer for 5 min three times between each step. The strips were visualized with an ECL kit (Amersham Biosciences, UK).

Surface Plasmone Resonance (SPR) Analysis of MINERVA
SPR experiments for the interaction of MINERVA FL with Keap1 321-609 were carried out in a Biacore T200 system (GE Healthcare, Chicago, IL, USA) using a buffer comprising 10 mM HEPES and 150 mM NaCl at pH 7.4. MINERVA FL in 10 mM sodium acetate at pH 5.0 was immobilized by the amine coupling method on a CM5 sensor chip according to the manufacturer's protocol (GE Healthcare, Chicago, IL, USA). A reference flow cell was equally treated but without protein as a control and the response by the control was subtracted from each sample. Keap1 321-609 at concentrations of 78.1, 156, 313, 625, 1250, 2500, 5000, and 10,000 nM were injected over the MINERVA FL chip at a rate of 30 µL min −1 . Biacore T200 evaluation software was used to estimate the affinity and kinetics fittings using the steady state affinity model and 1:1 binding model, respectively, to calculate the kinetics data. Average and standard deviation were calculated from two separate kinetics' data measured.  Conflicts of Interest: The authors declare no conflict of interest. The funders had no role in the design of the study; in the collection, analyses, or interpretation of data; in the writing of the manuscript, or in the decision to publish the results.