Characterization of protein extracts from different types of human teeth and insight in biomineralization

The present study describes an efficient method for isolation and purification of protein extracts from four types of human teeth i.e. molar, premolar, canine, and incisor. Detailed structural characterization of these protein extracts was done by Fourier transform infrared spectroscopy (FTIR) and circular dichroism (CD) which showed that a major fraction of the proteins present are unstructured in nature including primarily random coils in addition to the other structures like extended beta (β) structure, poly-l-proline-type II (PPII) helix, turns, with only a small fraction constituting of ordered structures like alpha (α) helix and β sheets. These resultant labile structures give the proteins the necessary flexibility that they require to interact with a variety of substrates including different ions like calcium and phosphates and for other protein-protein interactions. We also did initial studies on the mineralization of calcium phosphate with the protein extracts. Nanoparticle tracking analysis (NTA) show an increase in the size of calcium phosphate accumulation in the presence of protein extracts. We propose that protein extracts elevate the crystallization process of calcium phosphate. Our current biophysical study provides novel insights into the structural characterization of proteins from human teeth and their implications in understanding the tooth biomineralization. As per our knowledge, this is the first report which focuses on the whole protein extraction from different types of human teeth as these extracts imitate the in vivo tooth mineralization.

www.nature.com/scientificreports www.nature.com/scientificreports/ bound to their interacting partners. In the vertebrate bio-mineralization, most of the proteins associated with hydroxyapatite formation and growth are IDPs 12,13 . These IDPs may stabilize the ions or ion clusters, the critical nucleus, provide epitaxial sites for the initial mineral deposition, and/or stabilize the crystal. The flexible nature of these structures enables them to perform these and many other functions.
A tooth consist of three layer of mineralized tissues. The enamel forms the outermost layer, in which 90% organic material constitutes amelogenin, dentin forms the middle layer and cementum is the innermost layer. Tooth calcification is tightly regulated by proteins that are secreted into the extracellular matrix. The observation that unstructured proteins were involved in mineralization was first testified in 1977 in a circular dichroism (CD) study of phosphophoryn, in the presence of Ca 2+ . The addition of this ion, changed the structure from a random linear chain to a β-pleated sheet 14 . Phosphophoryn, also known as dentin phosphoprotein (DPP), is the product of the cleavage of dentine sialophosphoprotein (DSPP) into DPP and dentin sialoprotein (DSP). DSPP is the most abundant protein in dentin 15 and is also expressed in enamel and in bone.
To date, however, most studies on in vitro crystallization of calcium phosphate are performed with individual proteins from enamel 16 , dentine 17 or truncated versions of these proteins 18 . The cooperative effects of enamelin and amelogenin on crystal growth morphology was also analyzed by Janet Oldak's group 19 . Contrary to these approaches we are reporting, for the first time, the effect of whole protein extracts from different types of teeth on in vitro crystallization of calcium phosphate to mimic the in vivo condition of tooth biomineralization.

Materials and Methods
Isolation of proteins from human teeth. This study was approved by the Institute Ethics Committee for Post Graduate Research of All India Institute of Medical Sciences (Ref no. IECPG-387/29.06.2016). All experiments were performed in accordance with relevant guidelines and regulations. Written informed consent was received from all participants. In the case of individuals below 18 years, informed consent was obtained from parents/legal guardian for study participation. Human teeth samples used in this study were from living, healthy individuals, which were extracted due to orthodontic reasons. The age group was 17-40 years. Samples were collected from the Center for Dental Education and Research (CDER), AIIMS, New Delhi. Collected teeth were kept at 10% NaCl to inhibit bacterial growth and contamination. Teeth were washed thoroughly with water (2-3 times). After washing, teeth were crushed in a pestle and mortar in the presence of liquid nitrogen until a fine powder was formed. 10 g of this powder was taken and mixed with 20 ml of demineralization solution containing 0.1 M EDTA, 100 mM NaCl and protease inhibitors (Roche) at 20 °C for 2 days, centrifuged at 10,000 rpm and collected supernatant was dialyzed against HEPES buffer, pH 7.4. The clarified supernatant was further concentrated by the membrane-based cut off filters (3.5 kDa), which helps to remove impurities and salts as well as also reduced the loss of smaller proteins from our extracts. The protein concentration was measured by BCA kit (Pierce) according to the manufacturer's instructions. The samples were run on 12% SDS-PAGE and subsequently stained using silver staining.
UV absorbance spectroscopy and circular dichroism (CD). Absorbance measurements in the range from 200-350 nm were carried out using Cary Varian Bio 100 UV-Vis spectrophotometer (Varian Inc., California, USA). Quartz cuvettes were used for the measurements. A cell of the path length of 1.0 cm was used and a baseline correction was carried out with a blank.
CD spectra were recorded on a JASCO J-1500 spectropolarimeter (Jasco Corporation, Tokyo, Japan) using cylindrical quartz cell of path length 1 mm. The instrument was constantly purged with nitrogen. Before the experiment, the calibration of dichrometer was checked by using d-10-campher sulfonic acid. The protein concentration of 0.2 mg/ml was used. Three consecutive spectral scans were averaged and corrected by subtracting corresponding blanks. Secondary structure was monitored in the far UV region between 190 and 250 nm.
Attenuated total reflectance Fourier transform infrared spectroscopy (ATR-FTIR). The protein extract was placed on the SensIR ATR reflection element. Spectra were recorded by Bruker Tensor 37 equipped with a DTGS detector at 4 cm −1 resolution. The experiments were performed using liquid samples at room temperature. Each sample spectrum of 128 scans was ratioed against a 128 scans background spectrum. We have subtracted the buffer and water content from each sample and used the buffer as a background.Collected spectra were analyzed by OPUS software. We have also performed the water vapor correction on each spectrum. Secondary structure estimation was done by curve fitting of the amide I region by Matlab (Mathworks Inc) program Kinetics written by Erik Goormaghtigh 20 . The water vapor subtraction was done and normalized to equal band amplitudes. Before the curve fitting a straight baseline passing through the ordinate at 1700 and 1600 cm −1 was subtracted. The baseline was modified again by the least-squares curve-fitting program which allows for a horizontal baseline to be adjusted as an additional parameter to obtain the best fit. To determine the initial peak position for curve fitting, a second derivative spectrum was used 21 .
Bioinformatics predictions for structural disorder. Twenty proteins were selected that were well known for their occurrence in the human tooth. Charge hydropathy (CH) plot was made. The CH plot represents a plot between mean hydrophobicity and mean net charge. The ProtParam program at the EXPASY server (https://web. expasy.org/protparam/) was used to determine the number of positively and negatively charged amino acids at pH 7 for the proteins shown on the CH-plot. The mean net charge (R) was calculated by determining the value of the absolute difference between the positively and negatively charged residues and dividing this by the total number of residues. Mean hydrophobicity was calculated by ProtScale program of EXPASY server (https://web.expasy. org/protscale/). Hydrophobicity scale of Kyte and Doolittle (1982) was chosen 22 . Plotting H versus R generated www.nature.com/scientificreports www.nature.com/scientificreports/ the CH-plot. The boundary line corresponds to the equation H = (R + 1.151)/2.785, which defines the boundary between disordered proteins (left side) and ordered proteins (right side) 23 .
Moreover, out of these 20 proteins, 10 were randomly selected to analyze their sequence-based disorder by two software programs, PONDR-VSL2 and IUPred 24 . These predictors have been used extensively in the literature and their performance has been thoroughly evaluated and compared to other predictors. The accuracy is on the order of 80-90%, with the software programs trained on the DisProt database. They return a score that ascribes the probability of structural disorder at residue level: a score above 0.5 in both means that the residue is likely to fall into local disorder. In the case of whole proteins, the overall disorder content was calculated by dividing the number of residues with a score higher than 0.5 by the length of the protein. A protein was considered as "mostly disordered" if more than 50% of its residues had a score above 0.5 and "fully disordered" if more than 90% of its residues fell into local disorder.
In vitro mineralization studies. Stock solutions of calcium (50 mM) and phosphate (5 mM) were prepared using reagent grade CaCl 2 (Sigma, >99.0% pure) and KH 2 PO 4 (Sigma, >99.0% pure). All solutions (except protein extracts) were filtered (0.22-μM filters, Millipore) before use. The KH 2 PO 4 solution was adjusted to pH 7.5-11.2 at 25 °C, using a small volume of KOH. The precise pH value was selected by design so that the reaction solution would have an initial pH ∼7.4 at 37 °C upon mixing all solution components. Aliquots of calcium and pH-adjusted phosphate solution were sequentially added to protein solutions to yield final concentrations of 2.5 mM Ca 2+ , 1.5 mM P i , and 0.2-2.0 mg/mL protein, with a final volume of 100 μL, as previously reported 18 . Control experiements were also done with only protein extract (protein ageing effect) in addition to only calcium phosphate (−ve control) and with BSA. A special control for ruling out collagen biased mineralization of calcium phosphate was done with Cisplatin injured nephrotoxic rat kidney tissue sample (A kind gift from Cardiovascular Pharmacology Laboratory, AIIMS, New Delhi). ELISA was done with rat collagen 1 antibody kit (CUSA BIO) followed by a similar procedure of mineralization and NTA experiments as described above.
Samples were then placed in a thermostatic water bath adjusted to 37 °C. Initial pH values were set to ∼pH 7.4. To minimize evaporation, the reaction tube was tightly sealed with a cap or Parafilm. Each experiment was carried out using two identically prepared samples. One sample was visualized on the light microscope attached with polarization accessory and the other one proceeded for nanoparticle tracking analysis (NTA) at different time points; at 0 hours, 6 hours and 12 hours. Experiments were repeated 8-10 times.

Enrichment of intrinsically disordered proteins in protein teeth extracts.
Total soluble teeth protein extract was subjected to acid treatment by the addition of 60% Perchloric acid (PCA, Sigma) to give the desired final concentration of 3%. After incubation on ice for 15 mins, the reaction was stopped by centrifuging at 4500 g for 20 min at 4 °C 25 . After treatment and clarification, an aliquot of the soluble protein extract was kept for in vitro mineralization and size measurement experiment by NTA. The soluble extract was then concentrated by precipitating with 8 volumes of acetone at −20 °C overnight. Samples were run on SDS-PAGE and subsequently silver stained. The remaining pellet down fraction was also checked through in vitro mineralization followed by NTA size-based measurements.
Nanoparticle tracking analysis (NTA). NTA measurements were performed with a NanoSight LM20 (NanoSight, Amesbury, United Kingdom), equipped with a sample chamber with a 640-nm laser. The samples were injected into the sample chamber with sterile syringes (BD Discardit II, New Jersey, USA) until the liquid reached the tip of the nozzle. All measurements were performed at room temperature. The software used for capturing and analyzing the data was the NTA 2.0 Build 127. The values obtained were concentration (particles/ml) and size and these were plotted in SigmaPlot 12.0 software. Statistical analysis. Statistical analysis was performed with Stata 12.1 (Texas, USA). Data represent mean ± SD of 4 independent experiments. Parametric data were analyzed by two sample Student's unpaired t-test (mean comparison test) between control vs different protein extract at all-time points. P values less than 0.05 were considered statistically significant (*<0.05; **<0.001; ***<0.0001). Graphs were plotted by using GraphPad Prism; Version 5.03; San Diego, CA 92108.

Results and Discussion
Each type of protein extract was successfully purified upon demineralization of respective human teeth powder. These extracts were run on a 12% SDS-PAGE gel and subsequently stained using silver staining (Fig. 1). The gel showed clear bands of all four teeth protein extracts. As per our knowledge, this is the first report showing a differential proteome in four types of human teeth extracts i.e. Molar, Premolar, Incisor, and Canine. In terms of proteins, there is a clear qualitative as well as quantitative difference in all four of these, this may account for the differences present in their teeth shape which in turn, dictates functionality. Proteins such as bone morphogenic protein (BMP) is reported to control the shape of the crystal formed during teeth biomineralization 26 . These protein extracts constitute proteins from enamel, dentin, and cementum of individual type of teeth's and some of these constituent proteins are known to be unstructured/disordered and can be characterized as IDP's 12 . This view is well supported by Fig. 2A, which shows the ultraviolet spectral scanning of the different protein extracts and compares it with bovine serum albumin (BSA). This comparison signifies two things. Firstly, the proteins present in different extracts are devoid in aromatic residues like tyrosine, tryptophan, and phenylalanine, which itself is a typical trait of IDP's and the primary reason for their "open" structures 27 . Secondly, the post-translational modifications like glycosylation of proteins in these extracts (also validated through FT-IR in Fig. 3A). All protein extracts showed higher absorbance above 300 nm as compared to BSA protein. In most cases, a protein remains optically inactive beyond 300 nm unless its interacting partner is a sugar molecule 28 or the protein contains a www.nature.com/scientificreports www.nature.com/scientificreports/ higher number of charged residues 29 . In general, both are typical characteristics of IDP's. Absorption around 220 nm is common in both which is a resultant of n-π* transitions of the peptide bond 30 .
Circular dichroism (CD) of protein extracts from human teeth. CD spectroscopy has been used to infer the secondary structure content of proteins 31 . Far UV CD spectral absorbance of different protein extracts are shown in Fig. 2B. The negative ellipticity near 205 nm clearly suggests that the majority of the protein molecules are random coil /unstructured in nature 32,33 . In addition to this, some residual secondary structure in the form of polyproline II (PPII)-type structure is also seen. It is well known now that globally unfolded proteins may possess some regions where residual secondary structures reside 34 . This phenomenon is seen predominantly in the case of the premolar and canine protein extracts. The slightly positive band near 220 nm confirms the presence of these PPII type structures in the extracts 35 . However, in the case of molar and incisor protein extract, these bands at 220 nm (n-π* transitions) are absent, which shows that in the extracts PPII helix is not the dominant conformation. Although, collagen I itself is known to have a similar PPII helix structure due to the presence of high proline content. Several other tooth proteins, including amelogenin (which constitutes 95% of developing enamel matrix) has shown the presence of residual local secondary structures in the form of PPII type helices 36 .
Moreover, the data obtained from circular dichroism is in full agreement with the secondary structure calculation obtained by curve fitting of the amide I region (Fig. 3B-E).  www.nature.com/scientificreports www.nature.com/scientificreports/ Fourier Transform Infrared Spectroscopy (FTIR) to characterize the human teeth. FTIR is a well-established technique for studying secondary structural characteristics of proteins and peptides 37 . Figure 3A shows the comparative FTIR profile of all four-different tooth protein extracts within the mid-infrared spectral region. The absorption range 1700-1600 cm −1 is attributed to the amide I vibration and arises mainly from the C=O stretching vibration with the minor contributions from protein backbone. The amide II mode range from 1600-1500 cm −1 is due to the out-of-phase combination of the NH in-plane bend and the CN stretching vibration. Amide III vibrations were also observed at ~1320 cm −1 . In general, amide I vibration is most commonly used for calculating the secondary structural details as this region is less affected by the amino acid side chain 38 . Apart from amide group vibrations, these protein extracts also confirmed the presence of post-translational modification like glycosylation. Carbohydrate moiety whether monomeric or polymeric evoke strong band in 1200-900 cm −1 region. The bands observed at 1180 cm 1 and 1044 cm −1 can be ascribed for C-O and C-C stretching vibration of carbohydrates respectively. Asymmetric and symmetric phosphate (P-O) vibrations are present at 1255 cm −1 and 1080 cm −1 .
We also estimated the secondary structural content of these different protein extracts by performing curve fitting in the amide I region using the programs Kinetics 39 and OPUS. The curve fitting is based on the second derivatives of the respective original absorption spectrum of all four extracts. This is a resolution enhancement method to separate the overlap bands. In the curve fitting, we input the wavenumber range for each peak based on the second derivative spectrum and then did the auto run by using the mixture of Lorentzian and Gaussian line shapes. This procedure resulted in the multi-component bands and their resultant was well overlapped (blue) with the original spectrum (black). The red line indicates the quality of the fits which is the subtracted resultant of the original and the overlapped band, we have observed straight lines in all the four extracts which implies a good fit. Figure 3B-E shows the fitted amide I component bands of all four protein extracts. We observed four prominent bands in these four protein extracts and their band assignments are as follows: band near 1678 cm −1 assigned to irregular structure (random coils) 40 , band near 1660 cm −1 assigned to alpha helices and irregular structure 10 , band near 1642 cm −1 assigned to irregular structures and beta sheets 38 , band near 1628 cm −1 assigned to irregular www.nature.com/scientificreports www.nature.com/scientificreports/ structure/PPII helix or extended beta pleated sheets 38 . We also assigned the band near 1690 cm −1 to beta turn and near 1615 cm −1 to irregular or side chains contribution. The secondary structure contents are summarized in Table 1. It can be clearly concluded that more than 50% of proteins constituting these extracts do not have stable secondary structure characteristics and the remaining are either in extended β structure or in turns and loops. A very small portion of the plot is indicating the presence of typically ordered structures like α-helix and β-sheets. The curve fitting results of amide I region of all four protein extracts underline primarily unstructured/ random conformations, but still, some differences at the level of secondary structure contents can be observed. The absorbance around 1690 cm −1 is only present in the molar protein extract.
This probably means that β-turns frequency is higher in proteins present in the molar extracts as compared to others. Similarly, amino acid side chain absorption is visualized in molar and incisor protein extracts. However, the functional significance of these differences at the level of teeth is not possible.

Bioinformatic predictions for structural disorder. Experimental approaches including CD and FTIR
can only provide the gross structural estimates whether the protein is "mostly ordered" or "mostly disordered", without shedding light on the residue-level resolution. The CH-plot ( Supplementary Fig. 2) clearly reveals the residual level details of the disorderedness. Out of twenty investigated proteins, most lie on the left side of the borderline, which corresponds to a region of low mean hydrophobicity in addition to the presence of high values of mean net charge. These are the typical traits of disordered proteins. Moreover, we plotted the disordered pattern of some selected proteins involved in tooth biomineralization ( Supplementary Fig. 3) which again demonstrate the perfect alliance among the two disorder predictors in almost all cases. It is now revealed through structure predictions, almost all proteins in the Swiss Protein database that have a role in mineralization are IDPs 13 .

In vitro mineralization experiments in the presence of whole teeth protein extract. In vitro
mineralization studies were done with the physiological stochiometric Ca/P molar ratio of 1.66 and in the presence of different protein extracts. Nanoparticle tracking analysis showed the size measurements at different time points, i.e. at 0 hours, and then after 6 and 12 hours of incubation of these mixtures at 37 °C. As evident from size measurement analysis, these protein extracts are elevating the crystallization process with increasing time. Figure 4A shows the results with all these four protein extracts from different teeth. We can clearly conclude that each of these protein extracts is involved in the promotion of hydroxyapatite crystallization. The control reaction with only calcium and phosphate is also shown. There is a steady increase in size which is a resultant of the forming calcium phosphate mineral phases, but the rate is much slower compared to the case with protein counterparts. This shows that the protein extracts promote the nucleation process and further increase the mineralization rates. In order to validate the specificity of our protein extracts, we used bovine serum albumin (BSA) protein (alone and in presence of calcium phosphate) as a random non-biomineralizing control protein (See Supplementary Fig. 4A) at different time points. We have also analyzed the protein aging effect (only protein extract and no calcium phosphate, Supplementary Fig. 4C). Finally, to remove collagen biased mineralization effects, a cisplatin induced nephrotoxic rat kidney tissue with elevated collagen content [41][42][43] (confirmed by rat collagen I ELISA kit, see Supplementary Fig. 6) was also analyzed by mineralization experiments followed by NTA size measurements at different time points (Supplementary Fig. 4B).We found that none of these control experiments showed an increase in size with increased incubation time. Thus, proving the significance of IDPs in modulating the calcium phosphate mineralization. IDPs and their role in biomineralization has been recently reviewed by Adele Boskey's group. Figure 4B shows the actual images for all these protein extracts as well as control at the molecular level from Nanosight instrument. Figure 4C shows the statistical significance of the data.
These differences in the shape and sizes of the hydroxyapatite crystals were also validated by polarization microscopy (Data not shown). Several other reports have shown the role of amelogenin in crystal nucleation, growth and regulation of shape and size of enamel mineralization 44,45 . In fact, Janet Oldak's group has also looked into the cooperative effects enamel and amelogenin 46,47 and amelogenin and ameloblastin 48 on enamel mineralization and crystal formation 49 .
In order to emphasize the role of IDPs in teeth biomineralization, we successfully enriched our protein extract by virtue of their failure to be precipitated by 3% Perchloric acid treatment 25 (Supplementary Fig. 5A). Moreover, this treated extract behaved similarly to the original, initial extract did in the in vitro mineralization followed by NTA size measurement experiments (Supplementary Fig. 5B). According to this, we can now conclude that the extracts do contain IDPs in the majority, and that IDPs do have prime roles in facilitating teeth biomineralization.

Conclusion
In this study, we demonstrated the efficient isolation of proteins from different types of human teeth. Protein isolation from hard tissues like teeth, shell, and bone is generally a daunting task because of the need for removal of excessive inorganic ions and due to the low concentration of organic material. These protein extracts were used for structural studies which were carried out using FTIR and CD. Our studies show that most proteins exist in the www.nature.com/scientificreports www.nature.com/scientificreports/  www.nature.com/scientificreports www.nature.com/scientificreports/ unstructured state. The presence of such a flexible structure is inevitable in the case of tooth proteins. These structures provide the necessary flexibility which may be required for stabilization the ions or ion clusters, the critical nucleus, provide epitaxial sites for initial mineral deposition, and stabilize the crystal. As our protein extracts contain a heterogeneous mixture of proteins, our result shows the average structural data. Our spectral data are reproducible, and they overall represent the similar structural homogeneity among the teeth proteins. There is no single or group of proteins present in the mixture so as to indicate a different structural propensity. This is the first time that the structural homogeneity of proteins in different teeth is established.
We have also highlighted the importance of the residual level assessment of disorders through bioinformatics approaches like CH plot and disorder prediction software. Bioinformatics analysis was well correlated with the experimental data obtained from FTIR and CD. All of these marked the presence of unstructured proteins in protein extracts obtained from different types of human teeth. In vitro mineralization assays of hydroxyapatite in the presence of protein extracts showed that these protein extracts elevated the nucleation and foster the overall process of tooth biomineralization (Fig. 5).
Prior studies in this area have only investigated the role of individual proteins from the human tooth, addressing their structural disorder and contribution in biomineralization. Our study has focused on the whole extract from different types of human teeth so as to emulate the in vivo condition of tooth biomineralization.