A basic protein, N25, from a mollusk modifies calcium carbonate morphology and shell biomineralization

Biomineralization is a widespread biological process in the formation of shells, teeth, or bones. Matrix proteins in biominerals have been widely investigated for their roles in directing biomineralization processes such as crystal morphologies, polymorphs, and orientations. Here, we characterized a basic matrix protein, named mantle protein N25 (N25), identified previously in the Akoya pearl oyster (Pinctada fucata). Unlike some known acidic matrix proteins containing Asp or Glu as possible Ca2+-binding residues, we found that N25 is rich in Pro (12.4%), Ser (12.8%), and Lys (8.8%), suggesting it may perform a different function. We used the recombinant protein purified by refolding from inclusion bodies in a Ca(HCO3)2 supersaturation system and found that it specifically affects calcite morphologies. An X-ray powder diffraction (XRD) assay revealed that N25 could help delay the transformation of vaterites (a metastable calcium carbonate polymorph) to calcite. We also used fluorescence super-resolution imaging to map the distribution of N25 in CaCO3 crystals and transfected a recombinant N25-EGFP vector into HEK-293T cells to mimic the native process in which N25 is secreted by mantle epithelial cells and integrated into mineral structures. Our observations suggest N25 specifically affects crystal morphologies and provide evidence that basic proteins lacking acidic groups can also direct biomineralization. We propose that the attachment of N25 to specific sites on CaCO3 crystals may inhibit some crystal polymorphs or morphological transformation.

Biomineralization is a crucial and widespread process among various species from prokaryotic magnetic bacteria to advanced eukaryotic organisms. Typical structures like magnetosome, spicule, shell, teeth, and bone are the products of these processes. These biominerals, which mainly consist of inorganic materials, offer organisms the functions of hunting, navigating, and defending (1,2). Calcium carbonate is one of the most abundant biomineralization materials utilized by numerous creatures because Ca 2ϩ and CO 3 2Ϫ are common metabolic sub-stances both inside and outside cells and CaCO 3 crystals possess excellent mechanical properties and plasticity (3,4). CaCO 3 is mainly deposited in three anhydrous crystalline polymorphs, namely calcite, aragonite, and vaterite. Amorphous calcium carbonate (ACC) 3 is considered the precursor of different polymorphs of CaCO 3 and indeed exists in several mineral structures, although it is unstable and likely to turn to vaterites before ultimately transforming to other polymorphs in ambient conditions (1,5).
One of the areas of most concern in biomineralization is the molecular mechanism that regulates the ordered deposition of inorganic ions into solid phases. It has been shown that organic macromolecules, mainly containing proteins, chitin, and other components, may play substantial roles during these processes, although the content of macromolecules is less than 5% of the dry weights of the biominerals (6,7). The bivalve shell of Pinctada fucata has been well investigated in previous studies as a model for pearl formation as well as shell biogenesis (8,9). The typical structure of the shell is constituted by three primary parts: the periostracum, prismatic layer mainly composed of calcite, and nacreous layer mainly made of aragonite, also identical to the pearl both in components and structure (10). In common with other biominerals, shell formation is considered to be under the control of organic materials. A typical model was proposed in which the insoluble chitin and silk fibroin built the frame inside, and calcites or aragonites are regulated by matrix proteins and deposited into organized structures (8,11).
Over 50 different kinds of matrix proteins from P. fucata have been reported, and some of them are well investigated (9,10). Proteins Pif97 and Pif80 are believed to regulate the lamellar sheets of the nacreous layer, and during this process, Pif97 functions as a chitin-binding protein and probably recruits Pif80, possibly together with N16, to induce the deposition of aragonite on chitin membrane in specific c-axis orientation (12)(13)(14). The protein Nacrein, isolated from the nacre layer, is a carbonic anhydrase with potential calcium-binding sites that may induce local Ca 2ϩ supersaturation followed by nucleation with HCO 3 Ϫ generated by its anhydrase activity (15,16). ACCBP is a specific ACC-binding protein that inhibited undesired growth of aragonite and could assemble to a decamer as a regulator for ACC's maintenance (17,18). Other proteins such as Aspein (19,20), MSI60 (21), N19 (22), PfN23 (23), PfN44 (24), Prismalin14 (25), and PfY2 (26) as well as the N16 family (12,27), lysine-rich matrix protein family (28), and Shematrin family (29) have also been thoroughly characterized and reported; however, functional information about the rest of the components is still lacking.
In previous research (30), several candidate matrix protein genes with tandem repeat sequences (10) were discovered by microarray analysis between different developmental stages of P. fucata from fertilized egg to juvenile stage (5,31,32). These candidate genes were up-regulated at least 20 times at the juvenile stage and were thought to be closely related to biomineralization because the typical shell structures begin to form during this period. One of the candidates, the protein PfY2, was identified as a novel matrix protein and has been reported as an inhibitory regulator of CaCO 3 precipitation (26), whereas the other candidates are still poorly understood. A deeper investi-gation was conducted in this study, and a new matrix proteincoding gene, N25, was cloned, and this protein was further identified as a matrix protein exhibiting calcite binding and morphological modifying abilities.

Expression and distribution pattern of N25
After cloning the full-length cDNA of N25, it was necessary to determine the sites where N25 was synthesized and expressed and where N25 protein localized. The real-time PCR (RT-PCR) assay was conducted in different tissues, including foot, viscus, gonad, mantle edge, mantle pallium, gill, and adductor muscle, to reveal the tissue-specific expression pattern. The result showed that the N25 mRNA was narrowly synthesized in the mantle tissue, which is thought to be the most substantial organ directing shell formation, with a much higher amount than in other tissues (Fig. 1D). Correspondingly, N25 protein was also detected in both the extracts from the decalcified prismatic insoluble matrix (PISM) and nacreous insoluble matrix (NISM), but the signal from PISM was stronger than that of NISM (Fig. 1C). These results implied that N25 was A, a denaturation-renaturation method was applied to the purification of N25 followed by detection with SDS-PAGE. B, purified N25 was mixed with calcite, aragonite, and chitin, respectively. MBP was used as a control, and all materials were washed and boiled before SDS-PAGE. MBP had a calculated molecular mass of 45 kDa; N25 protein was near 35 kDa. C, N25 protein localization in four matrix extracts from the nacreous layer and prismatic layer was analyzed by Western blotting. D, gene expression pattern of N25 in various tissues was determined by RT-PCR. Error bars represent S.D. (or S.E.). NSM, nacreous EDTA-soluble matrix; PSM, prismatic EDTA-soluble matrix.

Functions of N25 on CaCO 3 morphology
mainly expressed in mantle tissue and finally integrated into shell structure as a part of the insoluble frame.

Synthesis and secretion of N25 in HEK-293T
It is believed that the mantle tissue mainly controls shell formation, during which cells facing the inner side of the shell are responsible for the secretion of matrix proteins and other components. However, it is not possible to observe the process of how N25 is synthesized and secreted by the mantle epithelial cells due to a lack of both an appropriate bivalve cell line and an efficient gene transduction system. As a compromise, the eukaryotic 293T cell was chosen as a substitute. N25-EGFP with secretion signal peptide (Fig. S1) was expressed initially and distributed uniformly in the cytoplasm, and the subsequent appearance of tiny fluorescence spots represented the local regions with highly condensed recombinant protein. We speculated that these spots were vesicles. In addition, we found that the average fluorescence intensity of N25-EGFP decreased after imaging the cells with a time-lapse method. The absolute intensity of the region of interest o (ROIo) declined significantly from 0 to 42 min (Fig. 2E), and correspondingly, the amounts of spots in ROIo deceased compared with 0 min (Fig. 2, A and B). The region of interest c (ROIc) was chosen as a control to eliminate or minimize the possibility that the decreased fluorescence intensity was induced by nonspecific photobleaching. The relative intensity of ROIo, which also dropped at 42 min (Fig. 2F), is represented by the intensity ratio of ROIo to ROIc to subtract the influence of the decline of the base intensity. The motion image is available in Movie S1. To further investigate the exact secretion process of a single vesicle, we performed TIRF imaging of the cells, and only those vesicles located near or even anchored to the bottom membrane were observed. Fig.  2, G-J, shows the secretion process, during which a spot disappeared by rapid diffusion between 130 and 130.3 s, as indicated with the white arrows. The dynamic image is available in Movie S2. It could be concluded from these observations that the predicted signal peptide of N25 worked in the eukaryotic cell and the synthetic proteins were involved in a secretion pathway that could mimic similar events in mantle cells.

Expression and purification of N25
Several methods have been applied to explore the expression output of protein in Escherichia col; however, it was fairly difficult to obtain condensed and relatively pure N25 protein because most of the protein exists in inclusion bodies except for a tiny amount in soluble form. The inclusion bodies were washed with 1% Triton X-100 and dissolved in denaturant followed by purification with affinitive chromatography. Denatured N25 was further linearized by reducing disulfide bonds with DTT and dropped into a refolding environment with no denaturant, leading to an instant dilution of N25 into a low enough concentration that prevented the unfolded N25 from aggregating; then the protein molecule would refold spontaneously to a stable native packed conformation. After affinitive

Functions of N25 on CaCO 3 morphology
chromatography with Ni-NTA, the relatively pure N25 was harvested and condensed by ultrafiltration before analyzing by SDS-PAGE (Fig. 1A). The apparent molecular mass of N25 was about 35 kDa, consistent with the bands in Fig. 1C but nevertheless different from the calculated value of 25 kDa. To further identify the purified protein, the band in SDS-PAGE was analyzed and confirmed by mass spectrometry (Fig. S5). Posttranslational modification analysis (Table S1) and practical molecular mass measurement (Fig. S6) were also conducted by LC-mass spectrometry (MS) and QTOF-MS, respectively. The measured molecular mass of N25 was 25,776 Da, which was 124 Da larger than the predicted value of 25,652 Da, which suggests that the structure or shape of the molecule probably contributed to the altered migration.

Binding of N25 to CaCO 3 and chitin
CaCO 3 and chitin binding assays were performed to explore the affinity properties and interaction between N25 and different crystal polymorphs because the attachment of proteins might play critical roles in regulating crystal characteristics. N25 protein was mixed with calcite, aragonite, and chitin, respectively, and incubated for 2 h at 4°C. 1% Triton X-100 was applied to wash away the nonspecifically absorbent proteins before analyzing the samples by SDS-PAGE. Components remaining after washing were thought to have a relatively strong interaction with the solid materials. As shown in Fig. 1B, N25 protein was able to bind all three materials; however, calcite binding was a bit stronger than that of aragonite, but both calcite binding and aragonite binding were obviously higher than that of chitin. In contrast, maltose-binding protein (MBP), used as a negative control, showed no affinity for any of the materials. These data demonstrated that N25 attached specifi-cally to calcite, aragonite, and chitin, implying its natural integration into the shell components.

N25 stabilized the vaterite during ACC transformation assay
The influence of N25 on the transformation of ACC to a stable crystalline structure was estimated by the ACC transition assay. ACC formed immediately at the moment Ca 2ϩ was mixed with CO 3 2Ϫ and transformed to calcite rapidly in less than 15 min (Fig. 3, buffer group). The compositions of buffer groups all changed into calcites at 15, 35, and 55 min (Fig. 3). MBPs present in the reaction solution slightly delayed the transition to calcites from vaterites ( Fig. 3, MBP group), a kind of crystal form that was considered as a metastable polymorph of CaCO 3 and would finally change to calcites after a long enough the time. At 15 min, 29.9% of vaterites remained in the MBP group, and the proportion declined to 22.7% at 35 min and finally to 16.7% at 55 min. In contrast, vaterites in N25 groups were maintained for a longer period with percentages of 51.3% at 15 min and 51.8% at 55 min, whereas at 35 min the proportion was 39.8%, which was relatively lower compared with the other time points but still significantly higher than those of the MBP groups. These observations and results indicated that N25 protein might help to stabilize the vaterites and delay their transformation into calcites.

In vitro calcium carbonate crystallization assays
The effects of N25 on the morphology and polymorphism of CaCO 3 crystals were explored by its addition to a reaction system containing saturated Ca(HCO 3 ) 2 solution, constituted by mixing Ca 2ϩ and HCO 3 Ϫ solution to an ultimate Ca 2ϩ concentration of about 8 mM. Compared with the typical rhombohedral calcite crystals in the control group supplied with storage buffer (Fig. 4, A and A1), the crystals in the MBP group that

Functions of N25 on CaCO 3 morphology
contained 200 g/ml MBP showed no obvious morphology alteration (Fig. 4, B and B1), whereas the crystals in the N25 group showed significant changes in crystal morphologies (Fig.  4, C and C1). The degree of morphology alteration increased with elevated concentrations of N25 from 4 g/ml to 40 and 200 g/ml, respectively (Fig. 4, D-F). We noticed that the rhombohedral faces began to shrink at the edges and corners, whereas some extra growth structures also arose randomly and irregularly on the faces of crystal particles. Fig. 4G shows, at the 4th day, an extreme state of the abnormal morphology that might be caused by a complicated cooperative effect contributed by extra overgrowth on each face together with the inhibition of the corners and edges. Raman spectroscopy was used to identify the crystal forms of these particles and indicated that most of the particles in each group were calcites with specific peaks around 283, 713 and 1088 cm Ϫ1 except the spherical particles indicated with black arrows in Fig. 4, C and G, whose characteristic peaks were 301, 620, 1075, and 1091 cm Ϫ1 , illustrating their vaterite property (Fig. 4H). These data showed that N25 may play critical roles in regulating calcite morphology.

Distribution of N25 in calcite crystals and the computational simulation for morphology
For a deeper understanding of how N25 affected calcite crystallization, distribution mapping assays were conducted by labeling N25 with the fluorescent dye Cy5-NHS ester before adding to the calcite crystallization reaction. The super-resolution images were harvested with z-axis stacks at a z resolution of 0.1 m and reconstructed to a 3D image by a Nikon (Japan) structured illumination microscope (N-SIM) at an excitation wavelength of 645 nm. It was observed that the fluorescence signal showed inhomogenous distribution from the front view ( Fig. 5A1) and the side view (Fig. 5A2); moreover, N25 protein seemed to encircle the crystal and reside on the surfaces. To further confirm this observation, the fluorescence intensity was analyzed for each section slice, and the 2D histograms are presented for slices 23, 53, and 83, respectively (Fig. 5, B1-B3 and C1-C3), in which most of the signal pixels are distributed in a circular pattern without fluorescent signals inside the crystals. A surface simulation was performed to further clarify the N25 distribution pattern by replacing the red fluorescent signals with white granules. These white granules were shown to attach to the main faces and the shrunken borders of calcite crystal habit (Fig. 5D2), which could be correlated to the contractive edges in Fig. 5E2. A smaller signal ring, indicated by the white arrowhead in Fig. 5D1, arose inside the large fluorescent circle near the bottom of the glass plate and corresponded to the irregularly sinking cavities presented and indicated in Fig. 5E1 with a white arrowhead. We then mimicked the theoretic morphology and crystal habit in the normal condition and a specific situation with modified morphological parameters. The growth equilibrium theory was proposed to predict the normal morphology according to the structure of calcite, and a hexahedron-shaped crystal habit was generated (Fig. 5F1), which was consistent with the typical crystal habit of calcite in ambient condition. Several growth faces were listed with different growth rates; however, the {1 0 4}, {0 1 8}, {1 1 0}, {0 1 2}, and {0 0 6} were the five most possible growth faces with relatively low growth rates. The final habit of calcite showed only one face, {1 0 4}, because this face showed the lowest growth rate and constituted the main crystal habit. In a modified condition (Table  1), the distances of the other faces were decreased to 5920.8, 6720.8, 7740.6, and 6640.6, respectively, which induced these faces to emerge (Fig. 5F2). When further decreasing the distances to 4920.8, 5720.8, 6030.6, and 5040.6, respectively, the area of these faces expanded, whereas the {1 0 4} face area declined correspondingly (Fig. 5F3). Parameters that were utilized for calculation are listed in Table 1. We could infer from these observations that N25 bound to specific crystal surfaces of calcites first and then modified the morphology, potentially through increasing the attachment energy and decelerating the growth of these crystal faces.

Discussion
Matrix proteins have been extensively researched as one of the three main components of shell biominerals besides CaCO 3 and chitin frame. They are also thought to be critical for regulating the structure constitution (10). The matrix protein profile of the prismatic layer differs from that of the nacreous layer, although some types of proteins exist in both structures (33)(34)(35). The difference in protein distribution may contribute to the prior precipitation of prismatic calcite followed by growth of nacreous tablets (11,36). Proteins from both layers exist in two main states: soluble and insoluble proteins in EDTA solution. The EDTA-soluble matrix proteins are thought to regulate the precipitation rate, crystal polymorphism, and/or morphology, whereas the EDTA-insoluble proteins mainly function as a part of the frame of the shell and are very likely to function in regulating the crystallization nucleation sites and crystal orientation (11,37).
In this study, a novel protein, N25, was identified as a matrix protein, and its properties in CaCO 3 crystallization were explored from different levels. The N25 gene expression pattern in several tissues showed its specific location in the mantle tissue, whose edge region expressed more N25 proteins than the pallial regions, whereas much weaker signals were detected in other tissues (Fig. 1D). Based on previous reports, proteins in the mantle edge were mainly responsible for controlling of the prismatic layer, whereas proteins in the mantle pallium were thought to regulate nacre formation (38,39). N25 protein could be detected in both layers by Western blotting; nevertheless, the protein level in the prismatic layer was markedly higher

Functions of N25 on CaCO 3 morphology
than that of the nacreous layer (Fig. 1C), which corresponded to the results of RT-PCR.
Knowledge about the secretion of matrix proteins is insufficient due to the inability to conveniently monitor the protein behaviors inside mantle cells because no bivalve cell line is available for transfection (40). In this study, we utilized mammalian 293T cells instead and successfully observed this process (Fig. 2). The analysis of the primary amino sequence of N25 demonstrated that there is a typical signal peptide in the N terminus (Fig. S1), implying the possibility of secretory simulation by 293T cells (41,42). During long-time imaging with DeltaVision, we observed a decrease of the average fluorescence intensity, probably caused by the release of the hypothetical secretion vesicles from the membrane, which was a rapid diffusion process accomplished within 300 ms. In the actual mantle cells of P. fucata, there might be a similar process from the synthesis of N25 or other matrix proteins to their secretion, after which these proteins could bind to particular sites of the chitin frame or crystals, as indicated by the binding assays in vitro (Fig. 1B). It could be considered that N25-EGFP underwent secretory post-translation processing that mimicked the natural event occurring in mantle epithelia, and the signal peptide prediction principle was likely to work for matrix proteins of bivalve.
Based on the primary sequence, the molecular mass of N25 was theoretically calculated to be 25 kDa, and the predicted pI was 9.21, indicating that N25 is a basic protein and unlike the common, known acidic matrix proteins (43). The acidic carboxyls of glutamic or aspartic acid residues were considered as a substantial and potential calcium-binding groups that could recruit and induce a local calcium-saturated site, leading to the initial nucleation of CaCO 3 (20,44). However, N25 was rich in Ser (13.9%) and Pro (13.5%) but insufficient in acidic residues (6.9%). However, there are considerable amounts of Asn (7.8%), Gln (5.7%), and Lys (9.1%) residues, all of which have a free amino group at the terminus of the side chain. Consequently, we inferred that both the carboxyl groups and the hydrogenbonding amino groups of N25 were able to interact with certain crystal surfaces of CaCO 3 because the carboxyl was able to interact with Ca 2ϩ and the amino group probably contributed to the interaction with HCO 3 Ϫ (45, 46). The N25 protein secondary (Fig. S2) and tertiary structure (Fig. S3) prediction results demonstrated that N25 is likely to have a disordered and highly flexible configuration that might assist N25 with attaching to the crystal surfaces (47). We assumed N25 might possess binding affinity for free Ca 2ϩ or CO 3 2Ϫ as an extension of CaCO 3 -binding capability; however, the isothermal titration calorimetry assays showed a negative result, namely that the titrations with Ca 2ϩ , CO 3 2Ϫ , or HCO 3 Ϫ caused no obvious exothermic or endothermic effects (Fig. S4). It seems contradictory, but it could be interpreted that N25 may interact with specific ion arrays on certain surfaces (48) through its flexible configuration as even the molecule itself showed little ability to combine free ions in solution.
N25 also exhibited the characteristic of modifying the calcite crystal morphology. When N25 was introduced to the precipitation reaction, the borders around the main crystal surfaces of the typical rhombohedral calcite became depauperate and shrunk (Fig. 4, C1 and E1), and some additional structures also formed on the main surfaces (Fig. 4, F1 and G1). Based on these observations, it could be postulated that both the border shrinkage and extra structure growth contributed to the morphology alteration. In the N25 groups with a concentration of Functions of N25 on CaCO 3 morphology erite forming and distributed around other calcite particles randomly; however, the diameter of these vaterites declined when deposition time extended to 4 or 5 days, as shown in Fig. 4, C and G. Consistent with our in vitro crystallization results, the ACC transformation assay showed a similar effect, that during the early precipitation period there was a relatively long vaterite phase in both MBP and N25 groups before transition to the calcite phase, but the vaterite in N25 group remained more stable than that of the MBP group (Fig. 3). Most of the proteins show a trend to attach on the solid phase because of the weak interaction contributed by the charge on the molecule surface. The mild detergent Triton X-100 could disrupt these weak and nonspecific interactions between protein and the crystal without obstructing the specific interactions; thus, it was the specific attachment of N25 to CaCO 3 that made it unable to be washed away in the binding assay (Fig. 1B). We supposed that the molecules specifically binding to the surface of vaterite might suppress the subsequent transformation to other stable types of polymorphs, which led to the longer existence of vaterite in the N25 group.
It has been reported that matrix proteins might be distributed around the surfaces of CaCO 3 or be integrated inside the crystals in vivo during the formation of biomineral (49). In the present study, the localization of N25 was consistent with the former. The Cy5-conjugated N25 used in the crystallization assay exhibited a distribution pattern in which the signal was only detected on the crystal surfaces, including the main faces of calcite and the edges of each face, whereas almost no N25-Cy5 fluorescence was present inside the crystal (Fig. 5, C1-C3), suggesting that the effect of N25's adhesion to CaCO 3 changed the crystal morphologies. Considering the periodic bond chain and the attachment energy (E-att) theories (50,51), crystal morphology is controlled by a series of bonds between the crystal lattice, and the attachment energy is the energy released when a newly formed layer slice is added to a certain crystal face. Theoretically, a crystal consists of countless faces, and each face has a specific E-att because the bonds and interactions within the lattice of the face differ from each other. The relative growth rate of a certain face is proportionate to the E-att so that faces with larger E-att absolute values grow faster than those with lower E-att (52). As a result, the most slowly growing faces encircle a volume and construct the crystal habit or morphology because fast growth would lead to the disappearance of a face. Taking these principles into account, we performed a calculated computer simulation to predict the morphology and elucidate the habit alteration. The calcite structure cell was constructed, and the possible growth face list was generated fol-lowed by calculating the E-att and relative distance, which represented the distance from the face to the growth center. The distance was proportion to the growth rate; therefore, it could be defined as equal to the absolute value of E-att. The common rhombohedron-shaped calcite habit is presented according to the setting parameters (Fig. 5F1) with only one kind of face, {1 0 4}, whose distance was 4960.4. Faces listed in Table 1 were the most probable faces because of their relative lower E-att values; however, faces {1 0 4} were the lowest and the most important faces for normal calcite morphology, whereas other faces were absent in most situations. When the E-att and the distances of other faces were modified to 5920.8, 6720.8, 7740.6, and 6640.6, respectively (Table 1), the habit changed with the edges of each face shrinking, caused by the altered distances, which were shortened enough for these hidden faces to emerge. When further shifting the E-att and distances to 4920, 5720, 6030, and 5040, these face areas expanded, leading to the apparently increased intensity of the shrinking border (Fig. 5F3). These predicted habits could be correlated to the actual calcite crystal in Fig. 5E2; the exposed faces indicated with a yellow arrowhead in Fig. 5E3 might represent the face {0 1 8} or the face {1 1 0}. Based on these results, we deduced that N25 bound to certain CaCO 3 growth faces first and then induced the attachment energies to decline followed by slowing down the growth rates of some crystal faces and ultimately altering the morphology of CaCO 3 .
A schematic diagram is shown in Fig. 6 to explain the process from protein secretion to calcite binding and morphology alteration. Because of its narrow expression pattern, N25 gene should be up-regulated by a specific pathway (unknown at present) and then locally expressed and secreted by the mantle cells facing the inner side of the shell via vesicle-membrane fusion. The released N25 protein, probably working synergistically with other matrix proteins, then binds to vaterite and prolongs its existence. Vaterite may transform to calcite in two ways simultaneously, that is by direct inner transformation of the lattice of vaterite and by the Ostwald ripening method (51) in which the Ca 2ϩ and CO 3 2Ϫ in vaterite are redissolve into free ions and redeposited to a larger and more stable particle because the larger crystal had a lower surface area to volume ratio, more stable bonds, and thus lowered total energy. Both of these mechanisms might be delayed as a result of the attachment of N25, leading to a crystal that inhibits the lattice rearrangement and the release of ions into solution. Alternatively, calcite crystals could also grow directly from an initial nucleation core followed by free ion deposition onto their surfaces. According to the predicted attachment energy, the face {1 0 4} has the lowest attachment energy, making it the most impor- Table 1 Initial and modified parameters calculated for calcite morphology The parameters used in simulating the morphology of calcite in this work are listed. hkl indicates the Miller indices of the crystal growth face. Multiplicity indicates the number of faces with equivalent indices. E-att is the attachment energy, usually a negative value. The distance is proportional to the ͉E-att͉, and in this case the coefficient was defined as 1, so that the distance was identical to ͉E-att͉.

Functions of N25 on CaCO 3 morphology
tant face for the calcite morphology in a normal aqueous environment. N25 affected calcite morphology through interaction between the protein molecule and immobilized ions of calcite, which might block the sites for newly formed crystal layer, thus increasing the energy cost for deposition and decreasing the growth rates of some faces. And probably due to the difference in bounding intensity, the relative growth rates of several faces such as {1 1 0} and {0 1 8} declined more rapidly than that of {1 0 4}, causing the regular edges and corners to shrink and the {1 0 4} faces to externally grow with the help of redissolved small vaterite particles. The exact physiological meaning of morphology modification is not fully understood, but we speculate that morphology alteration might be crucial for the formation of some specific structures or might be just a concomitant phenomenon of inhibitory regulation of crystal growth. Because there have been several reports (20,22,24,25,26,27,35) on the inhibitory effects of matrix proteins, we suggest that N25 might integrally and subtly work together with other matrix proteins to regulate crystallization.

Conclusion
In this work, we repot that the novel matrix protein N25 functions as a morphology regulator of CaCO 3 through attachment to the crystal surfaces. A computer simulation implies that N25 protein may induce a decrease of certain face attachment energy and finally alter calcite morphology. This study extends the understanding of biomineralization regulation by matrix proteins and provides a possible aspect of exploring the mechanism by combining a simulative method and experimental observation.

Ethics statement
All animal studies were approved by the Animal Ethics Committee of Tsinghua University, Beijing, China.

Animals
Pearl oysters, P. fucata, were from the Pearl Farm in Zhan Jiang of China and were cultivated in artificial seawater. These animals were fed with yeast or spirulina powder dissolved in seawater every 3 days.

Gene cloning and sequence analysis
Total RNA was extracted from mantle tissue using the standard protocol for TRIzol reagent (Life Technologies, Thermo) for general cloning of genes. The RNA concentration was measured with a NanoDrop2000 at 260 nm (Life Technologies, Thermo). The quality of RNA was evaluated by agarose gel electrophoresis and A 260/280 plus A 260/230 . For RACE PCR, the cDNA was reverse transcribed by a SMARTer RACE cDNA Amplification kit (Clontech). For RT-PCR, RNAs from mantle edge, mantle pallium, adductor muscle, gill, foot, gonad, and viscus were extracted the same way, and each cDNA template was synthesized from 600 ng of total RNA with PrimeScript TM RT Master Mix (Takara), respectively. Primers N25-SP1-F, N25-SP2-F, and N25-SP3-F were used in 3Ј-RACE together with the combination of primers UPMlong, UPMshort, and NUP. Primers N25-SP1-R, N25-SP2-R, and N25-SP3-R were used in 5Ј-RACE the same way, following the standard procedures. The whole sequence of N25 was confirmed with primers N25-F and N25-R. The putative protein sequence was predicted online with the ExPASy Translate tool (https://web. expasy.org/translate/) 4 , and the signal peptide was analyzed by SignalP 4.1 (http://www.cbs.dtu.dk/services/SignalP/) 4 (57).

Plasmid construction
The gene sequence without the signal peptide was amplified by PCR with primers N25NOSIG-28NCOL and N25NOSIG-28RXHOI and then inserted into the prokaryotic expression vector pET-28a(ϩ), giving the recombinant plasmid pET28a-N25C with an His 8 tail at the C terminus of N25 protein. The gene sequence with the native signal peptide was amplified with primers N25EGFPN1-XHOL and N25EGFPN1-KPNR followed by insertion into the eukaryotic mammalian expression vector pEGFP-N1, generating the recombinant vector pEGFPN1-N25, which produced the corresponding fusion protein N25-EGFP with EGFP at the C terminus of N25.

Functions of N25 on CaCO 3 morphology N25 expression specificity analysis with RT-PCR
cDNA templates from seven main tissues were applied to detect the mRNA abundance of N25 by RT-PCR in a LightCycler 480II (Roche Applied Science) with the actin gene as an internal control. Primers RT-25F, RT-25R, RT-ACTIN-F, and RT-ACTIN-R were used with SYBR Premix Ex Taq TM (Tli RNase H Plus) kit (Takara) accordingly. All primers are listed in Table 2.

Distribution of N25 in different shell extracts
The nacreous layer was separated from the prismatic layer by manual polishing, and then the layers were decalcified by EDTA to extract the EDTA-soluble matrix and EDTA-insoluble matrix, respectively. Nacreous EDTA-soluble matrix and prismatic EDTA-soluble matrix were obtained after dialyzing and condensing the EDTA solution with ultrafiltration. NISM and PISM were harvested by boiling the insoluble pellet of the shell in the protein loading buffer for SDS-PAGE. Protein N25 was detected by Western blotting with polyclonal antibody from rabbit.

Protein expression and purification
Purified plasmid pET28a-N25C was used to transform E. coli strain Transsetta (DE3) (Transgene, China) for expression. Transformed E. coli were cultured in LB medium at 37°C at 200 rpm, induced with 0.6 mM isopropyl 1-thio-␤-D-galactopyranoside (Sigma) when OD 600 reached 0.7-0.8, and then cultured at 37°C at 200 rpm for another 12 h. The E. coli cell pellet was collected after centrifugation at 6000 ϫ g for 5 min at 4°C and then suspended in lysis buffer (50 mM Tris, 100 mM NaCl, 5% glycerol, 1 mM DTT, 1 mM EDTA, pH 8.0) after which the ultrasonic homogenizer was applied to disrupt the bacteria cells. After centrifuging at 15,000 ϫ g for 10 min, the pellet was resus-pended and washed in lysis buffer with 1% Triton X-100 (Promega) to isolate the N25 inclusion bodies. The washed inclusion bodies were dispersed and dissolved in lysis buffer with 6 M urea (binding buffer), and the filtered supernatant was applied to a 1-ml Ni-NTA resin (Sangon Biotech, China) column. The resin was washed with 30 mM imidazole in binding buffer before eluting with 500 mM imidazole, which was also prepared with binding buffer containing 6 M urea. The fractions were collected, boiled, and resolved by SDS-PAGE, and the corresponding bands were cut and utilized to immunize the rabbit.

Protein MS analysis
For identification of peptide coverage and post-translational modifications, N25 protein was subjected to SDS-PAGE, and the band was excised. The protein sample in the gel band was then treated by the method described previously (53) before LC-MS analysis with an OrbiTrap Fusion LUMOS (Thermo Fisher). Trypsin was used to digest the protein, and the missed cleavage site value was set to 2 when searching the sequence. The post-translational modification results are described in Table S1. For measuring relatively precise molecular weight, purified N25 protein was dialyzed against the sample buffer (2 mM Tris, 30 mM NaCl, pH 7.8), and then the sample was analyzed by QTOF in a SYNAPT TM G2-Si HDMS system (Waters) (results are presented in Fig. S6). All of the tests and data analyses were conducted in the Center of Biomedical Analysis, Tsinghua University.

Protein refolding
Similar to the method mentioned above, the inclusion bodies were dissolved in lysis buffer but with 6 M guanidine HCl instead of urea and 1 mM DTT for 1 h at room temperature, then centrifuged, and filtered with a 0.45-m filter. The con- CGTCGCCGTCCAGCTCGACCAG Primers for pET-28a and pEGFP-N1

Functions of N25 on CaCO 3 morphology
centration of this stock solution should be adjusted to 1-2 mg/ml. 3 ml of the stock solution was added drop by drop into 200 ml of rapidly stirred refolding buffer (20 mM Tris, 500 mM NaCl, pH 8.0) at the middle of side wall of the whirlpool within 10 min (54). After the dilution, the solution was kept still for 1 h and then filtered using a 0.45-m filter before loading to the balanced Ni-NTA resin. The resin was washed with 30 mM imidazole and ultimately eluted with 500 mM imidazole, both of which were prepared with refolding buffer. N25 was dialyzed to storage buffer (20 mM Tris, 500 mM NaCl, pH 7.8), the concentration was determined using a NanoDrop2000 (Thermo) at A 280 nm, and the protein was detected by SDS-PAGE.

Calcium carbonate and chitin binding assay
0.1 g of calcite, aragonite (Alfa Aesar), or chitin (Biodee, China), respectively, was washed and supplemented with 1 ml of storage buffer. Each material was mixed with 30 g of N25 protein, whereas the control group was supplied with equal MBP that was purified from pMAL-c5x following the manufacturer's instructions (New England Biolabs). The mixtures were rolled and incubated at 4°C for 2 h and then centrifuged to remove the supernatant. The pellets were suspended in 1 ml of storage buffer three times followed by suspending in storage buffer containing 1% Triton X-100 six times, ending with another wash with storage buffer three times. The washed materials were ultimately treated with loading buffer and analyzed by SDS-PAGE.

The synthesis and secretion of N25-EGFP in the eukaryotic cell
HEK-293T cells (China Infrastructure of Cell Line Resources, China) were thawed and cultured in DMEM (Gibco) with 10% fetal bovine serum at 37°C with 5% CO 2 . Plasmid pEGFPN1-N25 was transfected into HEK-293T cells via Vigofect (Vigorous, China) according to the manufacturer's instructions. 0.5 g of vector was transfected into cells cultured in a 35-mm confocal dish, 12 h after transfection the medium was replaced with fresh complete DMEM, and then the cells were cultivated for another 36 h. These cells were imaged directly by a DeltaVision time-lapse system (GE Healthcare) by recording every 2 min for about 3 h, and the results were processed with ImageJ software. For observing the secretion process using a Ti TIRF microscope (55) system (Nikon), cells cultivated for 36 h were digested with trypsin and diluted to separate individual cells, and then plated to a new confocal dish. Six hours later, the medium was replaced with fresh DMEM without fetal bovine serum, and cells were cultured for another 12h before imaging.

In vitro CaCO 3 crystallization
The saturated calcium bicarbonate solution was prepared by mixing 100 mM NaHCO 3 and 50 mM CaCl 2 to a final Ca 2ϩ concentration of 8 mM (18,56) in Milli-Q water immediately before mixing with proteins. 15 l of saturated solution was mixed, respectively, with 5-l protein samples whose concentrations were uniformly modified to 4, 40, and 200 g/ml, respectively. MBP was used as a control at a final concentration of 200 g/ml.
The mixtures were loaded on a silicified cover plate and incubated in an enclosed box supplied with water at room temper-ature for 48 h. Crystals were washed gently with Milli-Q water and dried in air, and then a LabRAM HR Evolution system was applied to analyze the crystal polymorphs with an extinction wavelength at 514 nm and scanning range from 100 to 1500 cm Ϫ1 . An FEI Quanta 200 scanning electron microscope was used to image the crystals.

ACC transformation assay
N25 was diluted into 7.5 ml of 50 mM CaCl 2 solution to a final concentration of 130 g/ml, and then 7.5 ml of 50 mM Na 2 CO 3 was added to this solution. The tube was immediately turned upside down several times. Three replicates were repeated, and the reactions were terminated at 15, 35, and 50 min, respectively, by centrifuging and washing the pellets with ethanol three times. The pellets were dried in air and analyzed via XRD.

Cy5 labeling and SIM imaging
N25 protein was dialyzed into labeling buffer (0.1 M phosphate, 500 mM NaCl, pH 8.3), and the concentration was estimated by A 280 . 1 mg of Cy5-NHS ester dye (AAT Bioquest) was dissolved in 100 l of DMSO and then diluted to 0.1 g/l with DMSO to form the dye solution. The amount of dye (mg) used was calculated via an empirical formula: 0.01 ϫ 855 ϫ weight of protein (mg)/25,000. Then the dye solution was added into the dialyzed N25 and mixed quickly followed by putting the solution in a dark box for 30 min. The labeled N25 was desalted into storage buffer (20 mM Tris-HCl, 500 mM NaCl, pH 7.8) by HiTrap TM desalting (GE Healthcare) in an ÄKTA system (GE Healthcare). The labeled N25 was further used in in vitro CaCO 3 crystallization assays, and after incubating for 48 h, the crystals were washed and air-dried before imaging with a superresolution Ti NSIM structured illumination microscopy system (Nikon).

Image processing and editing
All images in this study were processed in Photoshop CC2017, figures were generated from Origin2018, surface rendering was performed with and processed by iMaris, and the morphology simulation was calculated using Materials Studio following the instruction manual.