An Optimized Approach for Protein Residue Extraction and Identification from Ceramics After Cooking

Abstract

Extraction of protein residues from archaeological matrices, such as pottery clay, lithics, and grinding stones, has proven to be methodologically challenging. Protein residue analysis is fraught with technical challenges in analytical chemistry. In cooking pottery, protein residues are thought to bind to clay surfaces in vessel walls through a variety of primarily non-covalent interactions. Removal of protein residues requires the disruption of these interactions, and a diverse set of tools has been proposed and applied. Here, we test extraction procedures through varying combinations of physical parameters and solvents to derive an optimal approach yielding efficiencies of recovery from experimental pottery above 60%. The utility of our extraction approach was further validated through liquid chromatography–mass spectrometry analysis of experimental residues. We have identified several hurdles to developing a successful study of protein residues from pottery, each of which is surmountable with additional method development in the realm of archaeological chemistry.

Appendix

A common challenge faced by archaeological chemists is to balance the reporting requirements necessary for the evaluation of methods and results by analytical chemists with clear communication required by archaeologists. It is imperative that both be accomplished because the veracity of results and claims in archaeological chemistry rely upon data quality, which stems from the analytical procedures used. In this appendix, we report the analytical details of the methods and results published in general terms in the body of the article.

Methods

The phase 1 (protein–ceramic binding evaluation), phase 2 (extraction of isolated/mixture protein residues from pottery), and phase 3 (LC-MS identification of phase 2 extracts) experiments reported herein were conducted at the University of North Texas Lab (UNT). Phase 4 (evaluation of residues from experimental foodstuffs) preparation, extraction, and initial cleanup steps were also done at UNT, with proteomic characterization of residues completed at the University of South Florida Lab (USF). Because there are multiple experiments at two labs, we cover each phase in order; however, we discuss each lab’s liquid chromatography mass–spectrometry equipment and procedures in “High-Performance LC-MS/MS” below. In general terms, there are two types of analysis that are discussed throughout this paper—total organic carbon (TOC) analysis and liquid chromatography–mass spectrometry (LC-MS).

TOC analysis involves the combustion of a sample in an oxygen-rich, carbon dioxide-purged environment, resulting in the production of carbon dioxide from any inorganic and/or organic sources present. Finely ground samples are first pretreated with acid to convert inorganic carbonates to carbon dioxide and then elevated to 800°C to convert the remaining organic carbon to carbon dioxide, which is quantified by infrared spectrometry. We use TOC analysis to estimate the total amount of protein bound to experimental pottery.

LC-MS is a well-established instrumental approach for the identification of whole proteins and protein fragments based on the characterization of predictable products of enzymatic cleavage, which results in small peptides (typically <3,000 Da). The resulting peptides are separated by liquid chromatography and analyzed by tandem mass spectrometry, which consists of determining the mass of the peptide in addition to another stage of mass spectrometric analysis in which the peptide is fragmented into smaller daughter ions. Peptide ion selection is based on data-dependent acquisition that selects dominant precursor ions for fragmentation by collision-induced dissociation, resulting in fragment ion patterns that are a function of the amino acid sequence of the peptide. These “MS/MS patterns”¹ from multiple peptides can be compared collectively with the patterns predicted as protein products from genomic information. Genomic information is available for an increasingly wide range of taxa and assembled in publicly accessible databases that are searchable through a variety of algorithms (such as Matrix Science MASCOT or Sequest, see below). The search result is a probabilistic estimate of the quality of the match between the MS/MS spectra of unknown peptide amino acid sequences and those predicted from known genomic data. LC-MS, as described above, is used to characterize protein residues extracted from experimental and archaeological cooking pottery.

Protein–Ceramic Binding (Phase 1)

The binding procedures used in this study are modeled after those presented by Craig and Collins (2002) and are also reported in Stevens et al. (2010). Briefly, white earthenware tiles (Gare, Inc., Haverhill, MA, USA) composed predominantly of kaolinite clays were pulverized using a combination of mortar and pestle and a commercial coffee grinder. The resulting powder was sieved through a 710-μm screen to remove large particles and then fired at 800°C to remove trace organics. Forty-gram samples of this prepared ceramic were cooked for 5 days at 85°C in 200 mL of Milli-Q (MQ) water in combination with 20, 100, 1,000, or 2,000 mg of bovine serum albumin (BSA; product no. A2153-50G), bovine collagen (COL; product no. C9879-5G), horse myoglobin (MYO; product no. M0630-1G), bovine casein (CAS; product no. C3400-500G), or an equal-parts mixture (MIX) of the above (all were purchased from Sigma Aldrich, St. Louis, MO, USA). A method blank, prepared without the addition of protein, was also made.

Following cooking, each sample was stripped of unbound protein by repeated washing in Milli-Q water. After centrifuging to separate ceramic from the wash solution, a final wash using a small volume (<10 mL) of Milli-Q water was performed and the supernatant evaluated using a conventional Bradford assay (product no. PI-23236, Thermo Fisher Scientific, Waltham, MA, USA). Washing was terminated when Bradford λ ₅₉₅ absorbance of this final wash in comparison to a Bradford blank was observed to be below 0.01. Although less than the lower limit of the linear range of the Bradford assay, this value reflects a negligible protein concentration (<10 μg/mL, equivalent to <100 μg total per wash), indicating the removal of unbound protein.

After a final centrifugation, spiked ceramic was dried at 103°C until no weight change was observed and then allowed to cool in a desiccator. One gram was then thoroughly mixed with 400 μL of 85% phosphoric acid (H₃PO₄; product no. A242-1, Thermo Fisher Scientific) on a large watch glass and dried again at 103°C. This step served to remove any inorganic carbon. From these treated samples, 10- to 50-mg subsamples were assessed for TOC content using a Rosemount-Dohrmann model 183 Total Organic Carbon Boat Sampler set at 800°C coupled to a Tekmar-Dohrmann Phoenix 8000 Analyzer (Mason, OH, USA). Each ceramic sample was evaluated for TOC a total of five times and the replicates averaged. A total of 25 method blank samples ranging from 10 to 50 mg each were also evaluated for TOC in order to establish background organic levels. Because the mass of phosphoric acid remaining in samples resulted in the dilution of the spiked ceramic, we applied a correction factor, resulting in the same rank order but different absolute TOC values than those previously reported in Barker (2010b).

Protein Extraction (Phase 2)

One gram BSA-spiked ceramic and blank ceramic were subjected to a variety of chemical and physical/temporal extraction parameters in order to determine an optimal approach for protein removal. A total of seven different extraction solutions consisting of Milli-Q water, 2% (w/v) sodium dodecyl sulfate (SDS; product no. 436143-25G), 2% (w/v) sodium ethylenediaminetetracetate (EDTA; product no. E5134-50G), 4 M urea (product no. U5128-500G), 2% (v/v) trifluoroacetic acid (TFA; product no. 91707-250ML), 18% (v/v) formic acid (product no. 56302-1L-F), or 2 M hydrofluoric acid (HF; product no. 339261-800ML; all purchased from Sigma Aldrich and then diluted to respective concentrations in Milli-Q water without further pH adjustment) were evaluated against the parameters of time, temperature, microwave energy exposure, ultrasonic energy, and pressure. For each of these, both “HIGH” and “LOW” exposure levels were tested (Table 6 in the Appendix). For the temperature parameter, an additional exposure level (room temperature), “MID,” served as a reference to which other extraction methods could be compared. The ratio of ceramic to solvent used in extractions was kept at 100 mg to 1.2 mL, with the exception of the formic acid treatments. For these, a ratio of 250 mg ceramic to 1 mL of 18% formic acid was determined (via monitoring with a pH probe) to be necessary in order to match the optimal protein cleavage pH reported by Li et al. (2001). Each solution/physical parameter combination was tested twice. Following extraction, the solutions were centrifuged to remove suspended ceramic and then immediately evaluated for protein content.

Extraction Evaluation (Phase 2)

In order to determine the quantitative success of the extraction methods employed, Bradford- (595 nm) and/or UV-based (260, 280, and 320 nm) protein assays were used, as appropriate, based on their compatibility with the extraction solution employed. SDS, for example, is incompatible with Bradford assay, necessitating the use of a UV-based assay. For each solution type, standard curves were generated using serial dilution of BSA-spiked extraction solutions. Despite the potential for interference from extraction solution components, which resulted in non-zero values for some method blanks (Table 3), quantification of extracted protein was not substantially hindered. In all cases, r ² values of the standard curve plots were at least 0.99.

After the protein content of each extract was evaluated, the extraction efficiency was determined by comparing the observed quantities of the extracted protein to the theoretical maximum yield ascertained via TOC analysis. The values are reported here as percent recovery.

Extraction Confirmation (Phase 3: Isolated Proteins and Mixtures)

In order to confirm the binding and successful extraction of peptides from the ceramic matrix, we used conventional overnight trypsin digestions and LC-MS to evaluate the quantitatively most successful extracts from each category of extraction solvent (urea LOW pressure, SDS LOW pressure, TFA HIGH microwave, HF HIGH microwave, and formic acid HIGH microwave; see “Results”). Because HIGH pressure extracts yielded negligible gains relative to LOW pressure and likely resulted in further undesired denaturation and/or degradation of protein, LOW pressure SDS and urea extracts were used instead. In addition, MQ and EDTA extracts were not further evaluated as a result of poor quantitative performance.

Due to potential interference from solution components, cleanup steps were used to prepare samples for digestion and analysis. For phase 3 SDS-containing samples, an initial cleanup step was employed that is chemically similar to the phase 4 cleanup step discussed below, but more expensive (see next section). Briefly, disposable detergent cleanup spin columns (catalog no. 17100, Norgen Biotek, Thorold, Ontario, Canada) were used to remove SDS from protein extracts. Following this step, samples were treated in the same manner as the non-SDS-containing ones.

For all phase 3 samples, molecular weight cutoff (3 kDa) spin columns (Millipore, catalog no. UFC800324, Billerica, Massachusetts, USA) were used to remove contaminants and isolate peptides via a series of three washes. The purified samples were then lyophilized to remove water. Although there was no quantitative way to assess contaminant removal, the cleanliness of the samples was evaluated by visual inspection after lyophilization; if any large quantities of salt were observed, samples were filtered again using cutoff spin columns until demonstrated to be clean, though this was typically unnecessary. Cleanup was immediately followed by LC-MS analysis at UNT (see “High-Performance LC-MS/MS” below).

Experimental Foodstuffs (Phase 4)

To evaluate the behavior of proteins derived from a more realistic cooking event, white earthenware bowls (Gare, Inc., two in total) were filled with a mixture consisting of 200 g of locally produced sausage containing a 50:50 mix of ground white-tailed deer (O. virginianus) and ground pork (S. domesticus), 40 g of dried beans (P. vulgaris), 50 g of fresh corn (Z. mays), and 250 mL of MQ water. A third bowl (method blank) was filled with MQ water only. Each bowl was covered in aluminum foil to minimize water loss and then placed in an oven set to 90°C for a total time of 5 days. The bowls were removed daily, stirred, and then covered and placed back in the oven. When necessary, additional MQ water was added to ensure that the broth level remained at approximately 1 cm below the rim of the bowl. Following 5 days of cooking, bowls were removed from the oven and gently rinsed with MQ water. Subsamples of both the solid and liquid portions of the “stew” were retained for analysis.

After being allowed to air dry for 24 h, each bowl was cut into roughly 1-cm-wide vertical transects using a Dremel® tool fitted with an abrasive cutoff wheel (product no. 420, Dremel, Racine, WI, USA). The blank control sherds and the sherds from food-exposed bowl were wrapped in foil and stored at −20°C until ready for further processing. TOC analysis was conducted in triplicate at 1-cm intervals on an individual transect so that residue deposition patterns could be determined. This procedure was repeated for an adjacent transect from which the surface (approximately 1–2 mm) had been removed so that the permeation of residues into the ceramic matrix could be evaluated.

For the LC-MS analysis of protein residues, samples were pulverized using a sterile mortar and pestle, extracted using 2% SDS and “low”-pressure treatment at 108°C and then processed as previously described (see “protein extraction” and “extraction confirmation” above). Protein was isolated from SDS, salts, and other buffer components from phase 4 extracts using a chloroform/methanol extraction method adapted from Wessel and Fugge (1984). Briefly, an amount of methanol equal to four volumes of each sample was added. This step was followed by the addition of chloroform equal to the original sample volumes and HPLC-grade water to three times the original volumes. Thorough mixing was done after each addition by at least five inversions and 1 min of vortex mixing. Samples were then centrifuged at 14,000×g for 2 min to separate the organic from the aqueous phase—leaving protein at the interface. The aqueous phase was removed followed by the addition of three times the original sample volumes of methanol with inversion and rigorous vortexing. Protein was pelletized by centrifugation at 14,000×g for 5 min. The supernatant was removed, allowing the pellet to air dry in the fume hood prior to resuspension in reduction/alkylation buffer (6 M urea, 25 mM ammonium bicarbonate).

High-Performance LC-MS/MS

The chemical reduction and alkylation of disulfide bonds in preparation for trypsin digestion was performed in the 6-M urea buffer solution. Samples were incubated for 1 h at room temperature in 10 mM DTT to reduce disulfide bonds. Alkylation was performed by incubation for 1 h in the dark in 40 mM iodoacetamide. Reactions were quenched by the addition of DTT to 40 mM. Samples were prepared for trypsin by dilution of the urea to <1 M by the addition of 25 mM ammonium bicarbonate. One microgram of research grade TPCK-treated trypsin (USF Lab, catalog no. 4352157, Applied Biosystems, Foster City, CA, USA) or sequencing grade modified trypsin (UNT Lab, catalog no. V5111, Promega, Madison, WI, USA) was added to each sample and incubated overnight at 37°C. These reactions were quenched by adding formic acid to 5%. Samples were desalted in 1 mL C₁₈ solid phase extraction columns (catalog no. 52601-U, Supelco, Bellefonte, PA, USA) and concentrated by centrifugation under vacuum. Digests were resuspended in 0.1% formic acid in HPLC-grade water.

Phase 3 samples (isolated proteins and mixture experiment) were evaluated immediately at the UNT Lab using an Agilent 1100 series LC-MSD trap (Agilent Technologies, Santa Clara, CA, USA) equipped with an Agilent brand Zorbax 300 SB C18 capillary column (part no. 5064-8267) using a gradient elution with solvents A (0.1% formate in MQ water) and B (0.1% formate in acetonitrile). The injection volume was 8 μL and the flow rate was held at 4 μL min⁻¹. Electrospray ionization and positive ion detection were used. A blank (50:50 MQ/acetonitrile) and a BSA tryptic digest standard (8 pmol, catalog no. P8108S, Promega) sample were run alongside processed samples in order to verify system performance. The resulting mass spectra were processed using Agilent ChemStation software and then evaluated using a MASCOT (www.matrixscience.com) search of the SwissProt database. Peptide tolerance was set as ±1.2 Da and up to 2 missed trypsin cleavages were allowed. Error-tolerant searching was not used, meaning that any matches obtained likely reflect proteins that are not substantially modified.

Phase 4 (foodstuff experiment) tryptic digests were analyzed by in-line reversed-phase HPLC-MS/MS in duplicate (one analytical replicate per instrument) on a hybrid linear ion trap Orbitrap instrument (LTQ Orbitrap XL, Thermo Fisher) equipped with XCalibur (version 2.0.7) data acquisition software. Liquid chromatography was done on an Eksigent NanoLC Ultra 2D system (Dublin, CA, USA). Five-microliter injections of each sample were desalted on a ProteoPep II™ 75-μm i.d × 2-cm capillary C₁₈ solid phase trap column (New Objective, Woburn, MA, USA) with 2% acetonitrile and 0.1% formic acid in water prior to a linear gradient to 40% acetonitrile in 90 min at 250 nL/min. Mass spectral acquisition included a full survey scan as well as data-dependent MS/MS scanning of the top five most intense precursor ions with dynamic exclusion for 180 s.

The acquired spectra from phase 4 experiments (foodstuffs) were extracted via the extractMSN.exe program (BioWorks v. 3.3, Thermo Fisher) and analyzed using Mascot (version 2.2.06, Matrix Science, London, UK) program search of the UniProt database (UniProtKB, 518415 entries) with an X!Tandem (version 2007.01.01.1, The GPM, http:thegpm.org) search of a UniProt database subset for validation—both assuming trypsin as the digesting enzyme—allowing for one missed cleavage. Mascot and X!Tandem search parameters included a parent ion mass tolerance of 7 ppm (LTQ XL Orbitrap) and fragment ion mass tolerance of 0.6 Da. Fixed modification of cysteine by iodoacetamide and variable modification of methionine oxidation were also specified.

Phase 4 tandem MS peptide and protein identifications were validated with Scaffold (version Scaffold_3_00_06, Proteome Software Inc., Portland, OR, USA). Peptides identified by the Peptide Prophet algorithm (Keller et al. 2003) at >95% probability were accepted. Protein identifications were generally accepted at >95% probability if they contained at least one identified peptide as specified by the Protein Prophet algorithm (Nesvizhskii et al. 2003). This identification criteria typically established a peptide false discovery rate of <1%, as estimated by Protein and Peptide Prophet statistics in Scaffold.

Table 6 Physical extraction parameters used to evaluate protein extraction success

Supplementary Results

HPLC-MS/MS Confirmation (Phase 4)

A list of proteins that share identical tryptic peptide sequences for different organisms are shown for a few representative proteins identified in this study (Electronic Supplementary Material Tables 7–10). Peptides detected from LC-MS/MS analysis at identification probabilities lower than that employed for identification (<1% FDR) are provided in this table to show the sequence comparison for the entire unfiltered peptide lists of these proteins.