Bone protein “extractomics”: comparing the efficiency of bone protein extractions of Gallus gallus in tandem mass spectrometry, with an eye towards paleoproteomics

Sample preparation

Two samples were evaluated using each extraction method; one containing ground, defatted chicken bone, and one “blank” sample, consisting of extraction buffers (but no bone samples) to serve as a negative control. Chicken tibiotarsi were defleshed, then degreased whole in 10% Shout (Johnson Co.) with stirring for 1–2 days to remove lipids and blood from the medullary cavity and periosteal surfaces. Bones were then sectioned into large (∼1 cm) pieces and degreased for an additional day. Concentrated Shout contains small amounts (0.001% –1%) of the enzyme subtilisin, which has been shown to have collagen-cleaving properties (Ran et al., 2013); however we applied a diluted form the detergent (0.0001–0.1% enzyme), and only to whole bone pieces prior to grinding or homogenization. Thus, we think it is unlikely that the trace amounts of enzyme significantly affected the proteome of the tissue deeper in the cortex. After triple washing in 18.2 MΩ water to remove any remaining detergent from the bone surfaces, pieces were frozen with liquid nitrogen for 20 s and then ground with a mortar and pestle to the consistency of coarse sand. Bone powders from all bones were homogenized before aliquotting to ensure uniform samples of tissue for all extractions.

Protein extractions

The basic steps and parameters of each protein extraction protocol are summarized in Table 1, and details of each method are provided below. For all protocols, “samples were centrifuged” refers to centrifugation at 7,200 rcf for 20 min at 4 °C unless otherwise stated. Volumes of reagents applied are given as “mL/g,” referring to the mL of reagent applied per gram of bone.

Buffer and salt removal and concentration of proteins

After extractions were performed and supernatants collected, different post-extraction desalting and buffer removal methods were employed to test the efficacy of each for removing buffers, bone minerals, residual salts, detergents, and other non-protein extraction products that can interfere with downstream analyses (e.g., pellet weight calculation, protein quantification assays, ionization in MS) (Visser, Lingeman & Irth, 2005) while retaining protein. All extraction supernatants collected were centrifuged to pellet any debris, then divided equally into two fractions, each placed into pre-weighed tubes and stored at 4 °C. For all fractions of all protocols, half of the extraction supernatant was dialyzed and lyophilized to dryness, and half was precipitated, except as noted below. At the end of each method, post-weighing sample tubes allowed for calculation of pellet mass.

Bicinchoninic acid (BCA) protein assay

To determine total protein content in the final, post-lyophilization or precipitation pellet, a Pierce BCA Protein Assay Kit (Thermo Scientific) was used according to the manufacturer’s specifications (Thermo-Scientific, 2002). Aliquots (25 µg) of lyophilized, precipitated, or dried (by speed vacuum) extract from each fraction (see above) were resolubilized in 25 µL phosphate buffered saline solution (PBS) and plated (in duplicate) on a 96 well microtiter plate. 200 µL of working reagent (50 parts BCA kit reagent A to 1 part reagent B) were added to all wells, and plates were incubated at 37 °C for 30 min. Absorbance was measured on a Multiskan Spectrum 1500 microplate reader (Thermo Scientific) at a wavelength of 562 nm using SkanIt 2.2 software. Sample protein content was determined against a standard absorbance curve calculated from a serial dilution of bovine serum albumin (BSA) in PBS to the following concentrations (25 µL plated per well): 2 mg/mL, 1.5 mg/mL, 1 mg/mL, 750 µg/mL, 500 µg/mL, 250 µg/mL, 125 µg/mL, 25 µg/mL, and 0 µg/mL (PBS only). Assays were repeated 4–7 times for each sample and total protein content per 25 µg of pellet was assessed as the average of these multiple trials. Overall protein recovery and pellet purity were calculated based on these results and the total weight of the recovered pellet for each fraction. Extract products from “buffer only” samples were also tested to confirm that interference from left-over reagents did not cause spuriously high results in bone samples.

Protein digestion and mass spectrometry

Duplicate aliquots of 20 µg protein were prepared from the following fractions: all fractions of Methods 1 and 4–6, the solubilization fractions of Methods 2 and 3 (dialysis only of 3), and the demineralization fractions of Method 7. Methods that had fractions identical to some already tested (HCl from 2 and 3), extreme amounts of non-protein precipitation in the blank controls (precipitated “Urea” from 3), or solubilization fractions with extremely low amounts of total protein yield (acetic acids from 7) were excluded. Samples were reduced in 20 µL of 8 M urea (previously deionized with BioRad AG 501-X8 resin according to the manufacturer’s protocol) and 2 µL of 100 mM DTT diluted in 100 mM ABC (final concentration ∼9.1 mM DTT) for 20 min at RT, then alkylated with 3 µL of 300 mM iodoacetamide diluted in 100 mM ABC (final concentration ∼32.14 mM iodoacetamide) for 30 min at RT in the dark. Proteins were digested overnight with 200 ng of Trypsin Gold (mass spectrometry grade; Promega, activated for 15 min at 30 °C prior to application) diluted with 135 µL of 100 mM ABC and 1 µL of 1 M DTT. Digestion was terminated with 2 µL of 100% formic acid (FA) and resulting peptides were desalted and concentrated with C18 ZipTips (Millipore) as follows: Tips were activated with 90% acetonitrile (ACN), 0.2% FA, and equilibrated with 0.2% FA. Peptides were bound to tips, then washed with 0.2% FA and eluted into 30 µL of 70% ACN, 0.2% FA. Samples were dried by speed vacuum to remove acetonitrile, then stored at −80 °C until analysis.

Samples were resuspended in 27 µL of “Buffer A” (95% Optima grade water (Fisher Scientific), 5% Optima grade ACN (Fisher Scientific), 0.2% FA), centrifuged at 21,000 × g for 10 min at 4 °C, and transferred to autosampler vials (RSA™ AQ™; MicroSolv) with caps containing PFTE septa (Fisher Scientific). A Dionex Ultimate 3000 UHPLC System (Thermo Scientific) injected 6 µL of each sample onto a self-packed C18 Aqua (3 µm particle size, 125 Å pore size; Phenomenex) trap column (4 cm L, 150 µm ID). Peptides were washed and desalted for 10 min at a rate of 2.5 µL/min, then transferred to and eluted from a self-packed C18 Aqua analytical column (20 cm L, 75 µm ID) and spray emitter (12 cm L, 15 µm ID, New Objective, self-packed with 2 mm C18 Aqua resin) at 300 nL/min with the following gradient: 5% B at 0 min, 5% B at 12 min, 40% B at 45 min, 85% B at 47 min, 85% B at 49 min, 5% B at 51 min., and 5% B at 60 min. (A: 95% Optima grade water (Fisher Scientific), 5% Optima grade ACN (Fisher Scientific), 0.2% FA; B: 95% Optima grade ACN, 5% Optima grade water, 0.2% FA). Eluted peptides were then introduced into a custom 12T LTQ Velos FT-ICR mass spectrometer (Thermo Scientific) for analysis by tandem MS/MS. Full-scan FT MS1 spectra were obtained with a 400–1800 m∕z scan range at a resolving power of 85.7k. The top 8 most abundant peaks per MS1 scan were then selected for fragmentation by collision-induced dissociation (CID). MS2 scans were performed in the ICR cell (FT/FT) in centroid mode with an isolation window of 4 m∕z, a normalized collision energy of 35%, an activation q of 0.25, and a duration of 15.0 ms at a resolving power of 42.9k. Dynamic exclusion was enabled with the following parameters: repeat count of 2, repeat duration of 45 s, exclusion duration of 30 s, and an exclusion list size of 500. Total instrument acquisition time was 50 min for each sample run.

Data analysis

Spectra were searched in PEAKS (version 7.5; Bioinformatics Solutions Inc.) using a precursor mass tolerance of 10 ppm and a fragment ion mass tolerance of 0.2 Da. Two missed cleavages were allowed, as well as nonspecific cleavage at one end of the peptide. The following post translational modifications were allowed: fixed—carbamidomethylation [C]; variable—oxidation [M], oxidation or hydroxylation [RYFPNKD], [G]@C terminal, pyro-glu from Q, and deamidation [NQ]. Spectra were searched against the UniProt_chicken database, and PEAKS PTM and SPIDER were enabled to account for unspecified post-translational modifications (PTMs) or mutations, respectively. Results were filtered using ≤1% FDR for peptide spectral matches (PSMs), and either a protein score of −log₁₀p ≥ 20, or ≤1% FDR (whichever was more stringent) plus at least 1 unique peptide for proteins. Effective FDR values and associated expect score cut-offs for each file searched, as well as the number of MS and MS/MS scans performed, is provided in Table S1. Search results were analyzed by hand: peptides corresponding to common lab contaminants (e.g., keratin) were eliminated from consideration, and duplicate matches of one spectrum to multiple protein accession IDs were reduced. Additionally, proteins identified by only 1 peptide in 1 injection were eliminated (23 proteins), and only those found in multiple injections, or as multiple peptides from a single injection, were used for comparison. Finally, data from duplicate trials of the same fractions were combined to account for variation between trials.

Enzyme-linked immunosorbent assay (ELISA)

After total pellet protein concentration was calculated (see methods), aliquots of protein extract from each pellet were solubilized with PBS into 10 µg/mL solutions. 100 µL aliquots of sample solutions (1 µg of extracted bone protein) or 100 µL of PBS were added to the wells of an immulon U-bottom 96-well microtiter plate (Thermo Scientific) and allowed to incubate for 1–4 h at RT. After removal of antigen, to prevent non-specific binding of antibodies directly to the plate, wells were blocked with 200 µL of 5% BSA diluted in PBS with Tween20 and thimerosal (hereafter “5% BSA”) and the plate was incubated either 1–3 h at RT or overnight at 4 °C. A portion of wells from each sample was then incubated with 100 µL of the following treatments, diluted in 5% BSA: polyclonal chicken specific anti-collagen I (US Biological, C7510-13B) diluted 1:1000, polyclonal anti-alligator hemoglobin (Bio-Synthesis, Inc., host # BYSN 6941, lot # AB1421-3) diluted to 1:700 or 1:850, or 5% BSA only. We applied antibodies raised against alligator hemoglobin (HB), rather than commercially available polyclonal antibodies raised against chicken HB, because: (1) they were readily available in the time allotted for the experiment; (2) previous tests on antibodies against chicken HB were found to demonstrate non-specific binding to multiple proteins (M Schweitzer, 2006, unpublished data); (3) chicken and alligators are both archosaurian taxa (Brochu, 2001), therefore, substantial overlap in their hemoglobin epitopes is predicted.

Plates were incubated 1–3 h at RT or overnight at 4 °C, then washed 20 times in ELISA wash buffer (10% PBS diluted 18.2 MΩ water with Tween20). Wells received 100 µL of secondary antibody (alkaline phosphatase conjugated goat anti-rabbit IgG (H + L)) diluted 1:2000 in PBS and were incubated 1–2 h at RT, then washed an additional 20 times with ELISA wash buffer. 100 µL of substrate (diethanolamine substrate buffer + p-Nitrophenylphosphate, Pierce PNPP Kit) was added to each well, and absorbance at 405 nm was measured with a microplate reader (Multiskan Spectrum 1500; Thermo Scientific). Data were acquired using SkanIt 2.2 software. Measurements were taken at time = 0, 10, 20, 30, 40, 60, 90, 120, 150, 180, 210, 240, and 270 min.

BCA protein assay

The purity of the final pellet extracted from each method after precipitation, dialysis and lyophilization, or speed vacuuming, and thus the total recovered protein for each sub-extraction, was determined by BCA assay. Graphs of the final pellet weight compared to the calculated total protein recovered are shown in Figs. S1A and S1B. In every instance, dry pellet weight did not reflect actual extracted protein yield; total pellet weights were up to 15 times greater than the weights of their calculated protein content. The maximum purity of any pellet in this study (i.e., percent of total pellet weight that was protein) was 44.8% in 20-H/Urea-D, followed by 43.8% in 20-H/ABC-SV. Significantly, all extractions and post-extraction buffer removal methods conducted here produced a pellet that comprised more than 50% non-protein solids by weight.

The total protein yields per gram of bone analyzed for each demineralization and solubilization fraction were calculated from the BCA results and are shown in Figs. S2A and S2B, respectively. Demineralization fractions (Fig. S2A) recovered 0.51–6.3 mg protein per gram of bone analyzed. The highest recovery was observed from 20-N/EDTA-D (6.30 mg/g), followed by 20-HCl (whether precipitated or dialyzed) (4.59–4.74 mg/g), 20-EDTA-D (4.30 mg/g), 8-EDTA-D (2.77 mg/g), 5-HCl (whether precipitated or dialyzed, 0.74–1.03 mg/g), and finally, a second serial incubation of 20-N/EDTA2-D (0.51 mg/g).

Solubilization fractions (Fig. S2B) recovered between 0–58.39 mg of protein per gram of bone analyzed; the highest recovery values in the solubilization fractions exceeded that of the demineralization fractions by an order of magnitude. 20-ABC-SV recovered the most protein regardless of demineralizing agent (51.59–58.39 mg/g), followed by 20-H/ABC-D and GuHCl regardless of demineralizing agent, buffer removal method, or molar concentration (22.77–28.17 mg/g). Additionally, 20-E/A/GuHCl-D recovered 13.76 mg/g protein after an ABC solubilization step (Method 6). This was similar to 20-H/Urea extractions, which recovered 11.78–13.5 mg/g of protein regardless of buffer removal method, although an anomalously high amount of non-protein precipitant interference in the blank, buffer-only samples was observed (see below). 20-H/SDS-D recovered 8.08 mg/g, and 20-H/SDS-P less than a quarter that amount (1.52 mg/g). Acetic acid yields were below the limits of detection for this assay.

Mass spectrometry

For full consideration of their utility to bone proteomic research, mass spectrometry data obtained from all fractions were analyzed in three different contexts: (1) quantity of peptide-spectral matches (PSMs)/peptides recovered; (2) diversity of identified proteins and the redundancy of protein identifications between fractions within a method; and (3) the depth of sequence coverage (i.e., percent of recovered sequence length vs. total sequence length for select proteins).

PSMs/peptides

The highest number of PSMs (450–550) were recovered primarily from demineralization fractions (Fig. 2A), specifically 20-HCl-D, 20-HCl-P, 5-HCl-P and both 20-N/EDTA-D fractions. The one solubilization fraction that was in this range was 20-H/SDS-P, which recovered 468 PSMs. 20-H/ABC recovered ∼350 PSMs whether dialyzed or dried by speed vacuum. All other fractions recovered less than 250 PSMs; the poorest overall recovery was observed in 20-EDTA-D and 20-H/SDS-D fractions (<50 PSMs).

When only unique peptides are analyzed (i.e., when PSM duplicates and PTM variation are reduced) (Fig. 2B), demineralization fractions recovered the highest numbers of peptides (∼150–250), including 20-HCl fractions, 5-HCl-P, and both 20-N/EDTA-D fractions, with 5-HCl-P recovering the greatest number overall (238 peptides). The only solubilization fraction that recovered a total number of peptides within this range was 20-H/SDS-P, with 178 peptides. 5-HCl-D, 6-EDTA-D, 5-H/GuHCl-D, 20-H/Urea-D and 20-ABC (all variants) recovered 50–150 peptides. All remaining fractions recovered less than 50 unique peptides across two combined proteomic analyses.

Protein diversity and redundancy

A total of 55 unique proteins (identified from at least two unique peptides) were recovered across all fractions (Table 2). Of the 55 proteins identified in this study, 14 had accession numbers that corresponded to “uncharacterized” proteins. However, these proteins all contained substantial overlap with bone proteins from other avians when searched against SwissProt using UniProt’s basic local alignment search tool (BLAST) (Liu et al., 2012). For ease of discussion, these “uncharacterized” proteins will be referred to by their matches in BLAST, marked with brackets to differentiate them from the fully characterized proteins that were also recovered (e.g., [kininogen], [Pigment epithelium derived factor (PEDF)]). A full list of proteins identified with their corresponding accession numbers, protein descriptions, and BLAST matches is provided in Table S2.

Table 2:

Breakdown of proteins identified in each fraction, and the number of peptides recovered for each (including variations in PTMs).

Demineralization and solubilization fractions are divided by the centerline. Proteins for which 5+ peptides were recovered in any fraction are bolded. Of these proteins, fractions that resulted in the most peptide identifications are marked in dark purple, and fractions with peptide identifications within one standard deviation of the highest value are marked in light purple. The highest diversity in identified proteins was observed predominantly in the demineralization fractions (left), which also resulted in the greatest numbers of peptides from most proteins. One notable exception was collagen I, alpha 2; various solubilization fractions (e.g., ABC, GuHCl) resulted in more peptide identifications for collagen I, alpha 2 than found in any demineralization fraction.

DOI: 10.7717/peerj.2603/table-2

The greatest number of proteins was identified in 20-HCl-P, which recovered 37 proteins (Fig. 2C). Demineralization fractions accounted for the top five recoveries ranging from 25–37 proteins, including both 20-HCl-P and 20-HCl-D, 5-HCl-P, and both 20-N/EDTA1-D and 20-N/EDTA2-D. 20-H/SDS-P recovered 24 proteins, and was the only solubilization fraction in which more than 15 proteins were identified. All other fractions recovered fewer than 15 proteins.

To investigate whether demineralization fractions were recovering a set of proteins distinct from those obtained in the subsequent solubilization fractions, we constructed Venn diagrams from the lists of proteins identified by each method (Fig. 3). In five of the methods, more than 60.0% of the proteins found in the solubilization fractions were redundant with proteins previously identified in the demineralization fractions (60.0–92.9%, Table S3). Method 6, which included only one demineralization fraction (20-EDTA-D) but two serial solubilization fractions (20-E/ABC-SV and 20-E/A/GuHCl-D), showed 45.5% redundancy when the two solubilization fractions were combined (5/11 proteins). Because of their minute yield, solubilization fractions for Method 7 were not tested by MS, and therefore we cannot assess their redundancy. However, the two serial incubations of demineralizing agent in Method 7 (20-N/EDTA1-D and 20-N/EDTA2-D) were very redundant; of the proteins recovered by each (25 in the first, 26 in the second), 23 were the same (92.0% and 88.5% redundancy, respectively).

Sequence coverage

To compare sequence coverage of the most abundant proteins, we ranked the proteins with the highest coverage in any fraction, then generated a heat map (Table 3) of the top 28 (i.e., top half) that compares their coverage in all fractions. Of these 28 proteins, demineralized fractions possessed the highest (or equal to highest) coverage for 23 proteins. 20-HCl-P gave the best (or equal to best) coverage for 9 proteins (12K serum protein, 45.7%; cystatin, 39.6%; decorin, 31.1%; [dentin matrix protein 1], 18.9%; gremlin-1, 16.8%; histone 2A, 30.2%; lumican, 17.5%; matrix Gla, 54.5%; osteocalcin, 29.9%) and 20-HCl-D for five proteins (12K serum protein, 45.7%; [alpha-2-HS-glycoprotein], 59.6%; cystatin, 39.6%; [dentin matrix protein 1], 18.9%; osteocalcin, 29.9%). Lower volume, 5-HCl-P obtained the best or equal coverage of eight proteins (12K serum protein, 45.6%; [EXFAB], 39.3%; gallinacin-7, 61.2%; histone 2B, 20.6%; lysozyme C, 34.7%; myeloid protein, 37.1%; ribonuclease homolog, 18.0%; tetranectin, 38.8%). High volume 20-N/EDTA1-D gave the best/equal coverage for one protein in the first incubation (osteonectin (F1P291), 69.8%) and five additional proteins in the second, 20-N/EDTA2-D (apolipoprotein, 70.8%; [PEDF], 34.4%; serum albumin (P19121), 51.2%; secreted phosphoprotein 24 (SPP-24), 25.0%; vitronectin, 24.1%). Solubilization fractions obtained the best/equal coverage for only six proteins—20-H/SDS-P for (collagen I alpha 1, 23.8%; hemoglobin beta, 36.1%; histone 4, 17.5%; lumican, 17.5%) and 20-H/ABC-SV for (collagen I alpha 2, 43.7%; transthyretin, 37.3%). The most abundant proteins in bone, collagen I alpha 1 and 2, showed a generally similar range of coverage across the board (9.2–23.8% and 23.0–43.7%, respectively), with the exception of EDTA fractions, which obtained coverage ranges of 4.7%–20.3% for collagen I alpha 1 and 6.8–13.5% for alpha 2.

Table 3:

Heat map of the degree of peptide coverage obtained in all fractions for top 28 (top half) proteins with the highest coverage in any fraction.

Demineralization and solubilization fractions are divided by the centerline. Of the 28 proteins listed, the greatest (or equal to greatest) degree of peptide coverage for 23 of them was detected in demineralization fractions. Notable exceptions to this trend include collagen I alpha 1 and alpha 2; best coverage for these abundant bone proteins was obtained in 20-H/SDS-P and 20-H/ABC-SV, respectively. A color legend for the heat map is provided at the bottom of the table.

DOI: 10.7717/peerj.2603/table-3

ELISA

ELISA testing was performed to independently confirm select patterns observed in mass spectrometry results. ELISA assays plated with extracts from demineralization fractions as the antigen (Method 6 not included) with polyclonal anti-chicken collagen antibodies (Fig. S3A) yielded absorbance values as follows: 5-HCl fractions showed the highest absorbance values (0.78–0.84), which were similar (p = 0.1754) regardless of whether demineralization or precipitation was used as a buffer removal method. 20-HCl-D showed an absorbance value similar to that of 5-HCl fractions (0.74, p = 0.1342, 0.2346), but 20-HCl-P was significantly lower (0.58, p < 0.05). All EDTA fractions (20-N/EDTA1-D, 20-N/EDTA2-D, and 6-EDTA-D) showed the lowest absorbance values (0.46, 0.46, and 0.50 respectively), and were significantly lower than the values obtained for 5-HCl fractions and 20-HCl-D (p < 0.05 in all instances). A table of p-values for t-test results between all tested fractions is available in Table S4A.

To confirm the comparative efficiency of SDS for hemoglobin extraction, we tested a variety of extracts in ELISA with anti-alligator hemoglobin antibodies. ELISA performed on demineralization (Fig. S3B) and solubilization (Fig. S3C) fractions (Method 6 not included) with polyclonal antibodies raised against affinity purified alligator hemoglobin yielded absorbance values as follows: In the demineralization fractions (tested at an antibody concentration of 1:850), very low (but positive, as defined by absorbance at least twice above background levels (sensu Appiah et al., 2012; Ostlund et al., 2001; Tabatabai & Deyoe, 1984) absorbance values were obtained for HCl fractions 20-HCl-D, 20-HCl-P, and 5-HCl-P (0.02–0.09) and EDTA fractions 6-EDTA-D, 20-N/EDTA1-D, and 20-N/EDTA2-D (0.12–0.09). In the solubilization fractions (tested at an antibody concentration of 1:700), 20-H/Urea-D had very low, but positive absorbance (0.02). SDS fractions obtained the highest absorbance values, which were significantly greater (p < 0.05) than all other extracts tested; additionally, 20-H/SDS-D absorbance values (0.75) were three times that of 20-H/SDS-P (0.26). A table of p-values for t-test results between all tested fractions is available in Table S4B.

Comparison of chicken extracts

The data generated by this study reveal a number of patterns that influence proteomic studies of bone. First, the dry weight of the pellet remaining after extraction and buffer removal did not accurately reflect its calculated protein content, in most cases by a very wide margin, and is therefore a poor indicator of actual protein yield (Fig. S1). In fact, pellet weight did not accurately predict relative protein content when comparing various extractions, in that larger pellets did not necessarily correspond to a larger protein yield. For example, 5-H/GuHCl-P had a pellet weight of 151.1 mg and a protein content of 28.15 mg, whereas 8-E/GuHCl-P produced a far larger pellet (357.3 mg) but less protein (23.48 mg) (Fig. S1B). In all cases, less than half the pellet mass could be ascribed to proteins, indicating that none of the buffer removal methods we employed completely removed salts, detergents, and other non-protein solids from the sample. The inaccuracy and inconsistency of the relationship between pellet weight and protein content suggest that pellet weight should not be used as a proxy for assessing the yield of a bone protein extraction or for preparing subsequent samples for further proteomic analyses.

The most important pattern shown by this data is that, although solubilization fractions (e.g., ABC, GuHCl) overall recovered far more bulk protein than demineralization fractions (e.g., HCl, EDTA) (Fig. S2), demineralization fractions contained a greater diversity of proteins in each µg of recovered protein extract (Table S2). This translates to the recovery of more NCPs, and in many cases better peptide coverage of those NCPs. Indeed, in all but one instance, the greatest proteomic diversity in each method resulted from whichever reagent was applied to the pellet first (the exception is Method 6, in which 20-E/ABC-SV extracted 10 proteins compared to the nine extracted in the preceding 20-EDTA-D).

This is critical knowledge when designing studies to elucidate the bone proteome. Although the large disparity in protein yield between the two fractions would seem to make the loss from discarding the demineralizing reagents negligible, our proteomic data shows that these demineralization fractions contain an extractome with a much greater wealth of proteomic diversity per µg. Additionally, as mass spectrometry does not require large amounts of protein (but rather analyzes protein concentrations on the order of µg to ng), the relatively low total mass of protein extracted by demineralizing agents should not preclude proteomic analysis by tandem MS/MS in many cases. Although a greater mass of protein extracted by solubilization reagents may be necessary for downstream applications (e.g., gel electrophoresis), or for maximum recovery from degraded, low-protein samples, we strongly recommend that, when conducting proteomic analyses of bone (fossil or extant) the demineralization and wash products be retained and investigated, rather than discarded.

Beyond that overarching point, we also observed some general trends in the extractome recovered by each discrete step of the extraction protocols. Among demineralization fractions, larger volumes of reagent (relative to bone mass) generally resulted in a higher bulk protein yield and greater numbers of peptides, protein diversity, and sequence coverage identified by downstream MS analyses, with some exceptions and caveats. Increasing the relative volume of either HCl or EDTA from 5 mL/g to 20 mL/g led to double or triple the bulk protein yield (Fig. S2), and HCl or EDTA showed similar protein recovery at similar volumes. Notably, a brief incubation of the pellet in NaOH prior to application of EDTA led to a slightly higher protein yield for this reagent (e.g., 20-N/EDTA1-D vs. 20-EDTA-D), but the reason for this is unclear. Although increasing reagent volume increased protein yield, additional incubations in fresh reagent did not generate a substantial amount of additional bulk protein during demineralization (Method 7, 20-N/EDTA1-D vs. 20-N/EDTA2-D). For HCl fractions, dialysis versus precipitation made virtually no difference for total protein yield (Methods 1–4).

MS analyses showed that 20-HCl (Method 4) incubations resulted in the highest degree of diversity in recovered proteins, with precipitated fractions resulting in a degree of recovered protein diversity only slightly higher than dialysis. Furthermore, only the two 20-HCl fractions recovered [alpha-2 antiplasmin], coagulation factor IX, osteonectin (P36377), and osteopontin (Table 2). When compared with other protocols, 20 mL/g HCl resulted in the greatest percentage of peptide coverage for 10 proteins (12K serum protein, [alpha-2-HS-glycoprotein] cystatin, decorin, [dentin matrix protein 1], gremlin-1, histone 2A, lumican, matrix Gla, and osteocalcin) and resulted in ∼20% and ∼30% peptide coverage for collagen I alpha 1 and alpha 2, respectively (Table 3). 5-HCl-P incubations (Methods 1–3) performed similarly to larger-volume HCl fractions in protein diversity (p > 0.05) and coverage. 5-HCl-P was the only fraction to recover [heterochromatin associated protein], and resulted in the best peptide coverage for 8 other proteins, including 12K serum protein, [EXFAB], gallinacin-7, histone 2B, lysozyme C, myeloid protein, ribonuclease homolog, and tetranectin. Conversely, 5-HCl-D fractions resulted in similar peptide counts and percent coverage for collagen I alpha 1 and alpha 2 (∼20% and ∼30%, respectively) as other HCl fractions, but a much lower number and percent coverage for identified NCPs (Table 3). Although 20-HCl treatments only performed marginally better than precipitated 5-HCl-P, the larger total protein yield of the 20-HCl fractions (∼4.5 mg/g vs. ∼1 mg/g) lead us to recommend larger volumes for extraction over smaller volumes for modern samples (there are additional concerns for fossil samples; see below). Finally, it should also be noted that osteocalcin, a small, acidic NCP intimately associated with bone mineral (Buckley et al., 2008; Ostrom et al., 2006; Ostrom et al., 2000), was consistently found in HCl fractions, but only in HCl fractions (Table 2). This may be the result of osteocalcin decarboxylation in an acidic reagent.

Proteins identified within EDTA fractions were, on average, less diverse than those identified within HCl fractions, but were in some ways complementary to the regions of the proteome retrieved by HCl. We tested 4 different fractions of EDTA: two combined serial incubations of EDTA (6-EDTA-D, Method 5), 20-EDTA-D (Method 6), and two separated serial incubations of 20 mL/g EDTA, prior to which the bone powder had been briefly incubated in 0.1 M NaOH (20-N/EDTA1-D and 20-N/EDTA2-D, Method 7). The 20-N/EDTA-D incubations out-performed both non-NaOH prepped EDTA fractions for total PSMs and peptides identified, overall protein diversity, and sequence coverage. Notably, the second serial incubation was slightly richer than the first, resulting in one additional protein (gremlin-1) and slightly greater or equal peptide coverage than the initial incubations for 12 of the 16 proteins they shared for which coverage was tracked, despite having a substantially lower overall protein yield as measured by BCA (∼6.5 mg/g in 20-N/EDTA1-D vs. ∼0.5 mg/g in 20-N/EDTA2-D, Fig. S2A). Taken together, 20-N/EDTA-D incubations had the second highest degree of identified protein diversity of all reagents tested, after 20-HCl and 5-HCL-P fractions. Although EDTA fractions did not identify any proteins that were not also identified in other treatments, 20-N/EDTA-D fractions resulted in the highest percentage of peptide coverage for multiple key NCPs, including apolipoprotein, serum albumin (P19121), osteonectin (F1P291), [PEDF], secreted phosphoprotein 24, and vitronectin. However, although this method recovered abundant NCPs, 20-N/EDTA-D extractions recovered only marginal amounts of collagen I relative to other treatments, with 7.2–16.2% peptide coverage of alpha 1 and only 7.2–8.8% of alpha 2 (HCl fractions recovered ∼20% and ∼30% peptide coverage, respectively). This suggests that EDTA recovers less collagen than HCl overall, but contains a greater proportion of NCPs per µg of extracted protein than HCl extraction products. To test this hypothesis, we performed ELISA on 1 µg aliquots of extracted protein from demineralization fractions using antibodies raised against purified collagen. The results (Fig. S3A) are consistent with this interpretation; absorbance values, which are a proxy for antibody-antigen interactions (Paulie, Perlmann & Perlmann, 2005), are highest in HCl fractions, and lowest in 20-N/EDTA-D fractions, suggesting that less collagen I is present per µg EDTA-extracted bone proteins, leaving the presence of other NCPs that comprise bone tissue to account for the difference.

Conversely, 20-EDTA-D fractions, not pre-treated with NaOH, did comparatively poorly, resulting in some of the lowest numbers of PSMs, peptides, proteins, and percent coverage of all fractions. Unlike HCl, lower volumes of EDTA without NaOH pretreatment did better than their higher-volume counterpart (15 proteins identified from 6-EDTA-D vs. 9 proteins identified from 20-EDTA-D). This discrepancy could be caused by the difficulty of removing EDTA from solution (e.g., Pereira-Mouries et al., 2002); thus, fractions with larger volumes of reagent may retain more EDTA after dialysis and lyophilization, causing greater interference during ionization and poorer MS results. However, both high and low volumes of EDTA were much less efficient than either of the two serial incubations of 20-N/EDTA-D. The reason for increased recovery in EDTA fractions pretreated with NaOH is unclear, and must be evaluated more in depth. The purpose of this step in previous published protocols has been to remove NCPs from the bone (Liu et al., 2012) and it is possible that, as opposed to completely removing the NCPs prior to demineralization, brief incubation in NaOH made them more accessible for removal by EDTA, given that 20-N/EDTA-D fractions resulted in greater sequence coverage and more peptides of numerous NCPs and less coverage/fewer peptides of collagen I than all other fractions (Table 3). Since NaOH is used to buffer EDTA from an acidic to an alkaline pH (e.g., pH 8), it is possible that slightly more basic conditions are optimal for NCP extraction from bone by EDTA. This hypothesis, however, remains to be fully tested. In any case, our data suggest that a brief incubation in NaOH increases the potential recovery of NCPs with EDTA demineralization, resulting in even greater peptide coverage of some proteins than HCl or any subsequent solubilization step, and should be considered as a possible component of methods where EDTA will be used for demineralization or initial extractions.

Solubilization fractions generally extracted a larger yield of bulk protein (per gram of bone analyzed) than demineralization fractions (Fig. S2), but resulted in a lower degree of diversity in identified proteins from every 20 µg protein sample analyzed by MS (Fig. 2 and Table 2). Furthermore, the proteins that were identified from the solubilization fractions were mostly redundant with those recovered in the previous demineralization step (Fig. 3 and Table S3). There were two solubilization fractions that deviated from this trend: 20-H/SDS-P (Method 2) and 20-H/ABC-SV (Method 4). For SDS, the precipitated fraction displayed the most protein diversity of any solubilization fraction (24 proteins identified), despite the fact that, except for acetic acid, it had the lowest overall protein yield (∼1.5 mg/g). Additionally, 20-H/SDS-P obtained the greatest or equal coverage for four proteins (collagen I alpha 1, hemoglobin beta, histone 4, lumican) and resulted in a greater or approximately equal degree of sequence coverage than HCl fractions for four additional proteins, after HCl had already been applied to the bone powder (Table 3). When compared solely to other solubilization fractions, 20-H/SDS-P obtained the greatest sequence coverage of 10 of the 16 proteins that were identified in these fractions for which coverage was tracked. The most significant unique feature of SDS in this study is that it seemed to preferentially extract hemoglobin compared to other fractions. Although various fractions of HCl, Urea, and ABC contained, at most, one peptide of hemoglobin alpha-A or beta, precipitated SDS contained two peptides of hemoglobin alpha-A and seven peptides of hemoglobin beta (Table 2). To support comparative efficiency of SDS for hemoglobin extraction, we performed ELISA on protein extracts from different reagents using anti-alligator hemoglobin antibodies (see above). The low, but positive signal from HCl and 20-H/Urea-D extracts are consistent with the isolated hemoglobin peptides found in those extractions (Fig. S3B), and the corresponding spike of MS hemoglobin data in SDS is also seen in ELISA (Fig. S3C). Interestingly, the most absorbance observed in ELISA is from dialyzed SDS, which showed three times more absorbance than precipitated SDS when incubated with the same antibodies under the same conditions. This is peculiar, because although 20-H/SDS-D extracted much more bulk protein than 20-H/SDS-P, the dialyzed fractions did very poorly in MS, generating some of the lowest peptide, protein, and coverage results (Fig. 2B–2C and Table 3). We suspect that, rather than a lack of protein diversity in the sample, the poor MS results of 20-H/SDS-D were caused by interference from remaining SDS detergent during ionization. As an anionic detergent, SDS is difficult to remove through dialysis (Naglak, Hettwer & Wang, 1990), and even trace amounts can cause signal suppression in MS by forming adducts with peptides, increasing sample conductivity, and other factors (Fridriksson, Baird & McLafferty, 1999; Shieh, Lee & Shiea, 2005). Therefore, although dialysis of SDS may retain more bulk protein than precipitation, potentially with greater richness of hemoglobin peptides than any other studied sub-extraction, our data show that precipitation is a comparatively better SDS removal method for downstream MS applications.

20-H/ABC-SV (Method 4) resulted in the best peptide coverage for collagen I alpha 2 (43.7%) and transthyretin. Although it did not rank in the top five treatments for resulting peptide counts or identified protein diversity, when compared solely to other solubilization fractions, it resulted in the best coverage of certain key bone and serum proteins including [alpha-2-HS-glycoprotein], and osteonectin (F1P291). Dialyzed and speed vacuumed 20-H/ABC resulted in variable identifications of low-abundance proteins, but generally produced the same degree of protein diversity (15 proteins identified) and similar peptide counts/sequence coverage for the major constituent proteins (e.g., collagen I, apolipoprotein, serum albumin) (Tables 2 and 3). Thus, the greatest advantages conferred by speed vacuuming ABC over other treatments are: (1) more bulk protein available for downstream applications, and (2) time saved, as it can be performed quickly over the course of hours instead of days. It should be noted that although ABC performed fairly well after demineralization of the bone powder with HCl (Method 4), identical incubations of ABC after demineralization with EDTA (20-E/ABC-SV, Method 7), resulted in the identification of comparatively fewer peptides and lower protein diversity (Tables 2 and 3), suggesting that HCl is a more optimal demineralizing treatment for bone prior to ABC than EDTA.

In general, GuHCl fractions obtained fairly consistent results, regardless of reagent used for pre-demineralization (e.g., HCl, Method 1 or EDTA, Method 5). Among solubilization fractions, whether precipitated or dialyzed, demineralized with HCl or EDTA, 4 M or 6 M concentration, GuHCl extracted less than or equal bulk protein as 20-H/ABC-D and more than SDS fractions (Fig. 2B), while in MS analyses, fewer peptides and proteins were identified in GuHCl fractions than in ABC or precipitated SDS fractions. Within GuHCl fractions, 5-H/GuHCl-D produced the richest MS data, identifying 5+ proteins compared to other GuHCl fractions and obtaining the best (or equal coverage) for 9 of the 10 proteins identified in GuHCl for which coverage was tracked (Table S3). 20-E/A/GuHCl-D obtained additional bulk protein from bone after the pellet had previously been incubated in EDTA and ABC (Method 6), but the specific proteins obtained were nearly all found in previous fractions of the method (Fig. 3F). Thus, while serial incubations of GuHCl after ABC may obtain additional extracted protein for downstream analyses that require large volumes of protein (e.g., gel electrophoresis), it does not appear to expand access to other regions of the bone proteome that were not previously obtained in other fractions.

20-H/Urea-D extracted less bulk protein than all other fractions save SDS, but resulted in fewer peptides and a lower degree of identified protein diversity in MS analyses than SDS, and was roughly on par with 5-H/GuHCl-D in terms of identified protein diversity and peptide coverage. This method identified only one protein (Histone-lysine N-methyltransferase) that was unavailable in other extractions (Table 2). The most notable thing about the Urea extractions is that samples which did not contain protein (i.e., blank buffer controls) generated large amounts of non-protein solids during precipitation, heavily skewing control data (Fig. S1B). We tested this anomaly multiple times, and found that high levels of non-protein material only precipitated when the supernatant was void of protein—spiking bovine serum albumin into the buffer supernatant prevented this reaction upon addition of acetone. This phenomenon precludes precipitation of Urea as a viable buffer removal method for any bone protein study that requires the inclusion of negative controls.

Acetic acid, with and without pepsin, extracted such a low amount of total protein from bone powder that these fractions were not tested in MS analyses (Fig. S2B). Acetic acid has been used as a solubilization agent in protocols designed to extract purified collagen I from skin and bones, to the exclusion of other NCPs, for inoculation to produce collagen I specific antibodies (Liu et al., 2012; Singh et al., 2011). The small amount of overall bone protein recovered from these fractions may be the result of hydrolyzed collagen being lost through the 20K MWCO dialysis membrane along with the smaller NCPs. These results suggest that significant amounts of bone must be processed for a proteomic study that requires multiple assays, and that this method would be particularly ill-suited for studies in which available bone material is limited.

When considered as a whole, these extractome data allow us to make recommendations for how discrete steps in methods may be combined to maximize the breadth and coverage of the bone proteome identified by MS analyses. By combining the two demineralization fractions that recovered the most diverse set of peptides (20-HCl-P and 20-N/EDTA-D) with the solubilization fraction producing a similar diversity (20-H/SDS-P)—only three of the 19 fractions that were analyzed by tandem MS/MS—we can account for 51 of the 55 proteins identified across all fractions, or 92.7% of the total protein diversity identified in the entire study (Fig. 3H). Furthermore, these three fractions include those with the highest (or nearly highest) coverage of all proteins for which coverage data was tracked, thus optimizing this model for degree of identified protein coverage as well as diversity. To combine these sub-extractions into one protocol, Method 2 (Table 1) need only be slightly modified by increasing the initial demineralization volume to bone weight ratio to 20 mL/g, precipitating both fractions as described, and performing a tandem extraction of additional bone material in which the sample is treated with 0.1 M NaOH prior to application of 20 mL/g EDTA.

Implications for paleoproteomics and future directions

Because the extant chicken bones used for these analyses were not subject to the diagenetic factors that can cause chemical alterations in fossil proteins (e.g., deamidation, carboxymethylation of lysine, loss of hydroxylations to proline (Cleland et al., 2016; Cleland, Schroeter & Schweitzer, 2015; Hill et al., 2015)), it is unclear whether the disparity of efficiencies between protocols observed here would be similar if conducted on a fossil sample. Indeed, given the chemical differences in depositional environments experienced by fossils from different localities, it is possible that similar comparison analyses of 10 different fossil specimens would yield 10 different results. Regardless, the current data conducted on chicken bone may have implications for paleoproteomic studies that can be explored in future fossil analyses.

First, the demineralization fractions (e.g., HCL or EDTA), shown here to have higher proteomic diversity per µg in MS, are commonly discarded in paleoproteomics experiments, presumably because their low-yields are considered negligible (e.g., Buckley & Wadsworth, 2014; Cappellini et al., 2012; Wadsworth & Buckley, 2014). This may be counterproductive in studies targeting NCPs of extinct taxa. Collagen I is the most abundant protein in bone tissue, but it contains minimal informative sequence variation between species; therefore, improving access to NCPs in the bone proteomes of extinct taxa may significantly increase our access to phylogenetically relevant data (Cleland, Voegele & Schweitzer, 2012; Schweitzer, Schroeter & Goshe, 2014; Wadsworth & Buckley, 2014). However, studies targeting NCPs in ancient bone that discard, or simply do not analyze, the demineralization fraction and subsequent water “washes” may be discarding the NCPs they seek. For example, a recent study by Wadsworth & Buckley (2014) investigated proteome degradation in archaeological bone up to one million years old by performing tandem MS/MS on protein extracts solubilized by GuHCl after bone mineral was demineralized in HCl. Our data suggest that additional analyses of the HCl products in similar studies may provide an even wider window into the proteome of these specimens than could be accessed by GuHCl alone, potentially allowing access to more NCPs in older fossils.

Additionally, the observed preferential extraction of hemoglobin from bone by SDS warrants testing this reagent for its possible utility in targeting hemoglobin extraction from fossils. Hemoglobin is a serum protein that can potentially shed light on the evolution of endothermy (Schweitzer & Marshall, 2001) and organismal response to climate change (Campbell et al., 2010), in addition to carrying greater phylogenetic and paleobiological signal than other, more conserved proteins (e.g., Weber, 1995). Although the breakdown products of hemoglobin have been identified in dinosaurs (Schweitzer et al., 1997), peptide sequences from dinosaur fossils have remained elusive. When combined with more advanced techniques for anionic detergent removal from the supernatant (e.g., precipitation kits specifically designed for SDS), SDS may be superior to previous methods for hemoglobin extraction.

However, there are some caveats in applying the results of these experiments on extant bone to paleoproteomic studies. For example, while large-volume fractions of reagents (e.g., 20 mL/g) did better in MS analyses than their lower volume counterparts, the comparative dilution of peptides in these sample volumes may cause greater peptide loss from adsorbance on the sides of pipette tips, sample tubes, dialysis tubing, etc. (Chambers, 2013) which may be a serious problem when studying ancient specimens in which preserved protein is already scarce. The tandem extractions using both HCl and EDTA we recommend for modern bone require processing twice as much bone powder as any single protocol, which may be a concern in studies where sample material is extremely limited. In addition, the total weight of the extraction products of these three reagents (HCl, EDTA, SDS) combined was five times less than bulk protein extracted per gram of bone from only one extraction of speed-vacuumed ABC. Thus, although MS only requires micrograms worth of protein, paleoproteomic studies should consider pairing a diversity-rich, low-yield demineralization fraction (e.g., 20-HCl-P) with a higher-yield solubilization fraction (e.g., 20-H/ABC-SV) and analyzing both fractions, or processing large amounts of sample material as feasible. In either case, for studies specifically seeking NCPs in low protein samples (e.g., osteonectin in fossils) it is imperative that any demineralization fractions be analyzed rather than discarded, regardless of comparatively minute amounts of extract. Conversely, extraction protocols that do not include a separate demineralization stage (e.g., Cleland & Vashishth, 2015; Hill et al., 2015) may be particularly well suited for the recovery of NCPs from fossils, and should be further explored.