Global Analysis of Protein Folding Thermodynamics for Disease State Characterization

Abstract

Current methods for the large-scale characterization of disease states generally rely on the analysis of gene and/or protein expression levels. These existing methods fail to detect proteins with disease-related functions and unaltered expression levels. Here we describe the large-scale use of thermodynamic measurements of protein folding and stability for the characterization of disease states. Using the Stable Isotope Labeling with Amino Acids in Cell Culture and Stability of Proteins from Rates of Oxidation (SILAC-SPROX) technique, we assayed ~800 proteins for protein folding and stability changes in three different cell culture models of breast cancer including the MCF-10A, MCF-7, and MDA-MB-231 cell lines. The thermodynamic stability profiles generated here created distinct molecular markers to differentiate the three cell lines, and a significant fraction (~45%) of the differentially stabilized proteins did not have altered expression levels. Thus, the differential thermodynamic profiling strategy reported here created novel molecular signatures of breast cancer and provided additional insight into the molecular basis of the disease. Our results establish the utility of protein folding and stability measurements for the study of disease processes, and they suggest that such measurements may be useful for biomarker discovery in disease.

INTRODUCTION

Gene and protein expression level analyses using DNA microarrays and protein mass spectrometry methods have been widely utilized over the past two decades for the characterization of cancer and other disease-related states.^1–15 While differential expression profiling studies can provide insights into the cellular pathways and proteins associated with a disease state, the biological significance of proteins with altered expression levels is sometimes dubious because of the indirect link between a gene or protein’s expression level and its function. Functionally relevant proteins with the same expression levels in different biological states also go undetected using the regnant paradigm of expression level profiling to characterize such states. Recently, activity based protein profiling studies have been used to characterize disease states.¹⁶ Such studies are attractive because they can probe the functional significance of proteins with a range of expression levels. However, activity-based profiling studies require the development of specialized probes that target specific enzyme activities.

Protein folding and stability measurements can report on a number of different biologically significant phenomena associated with proteins in different disease states. Point mutations, post-translational modifications, and new and/or altered binding interactions with cellular ligands (e.g., other proteins) can all produce thermodynamic stability changes. Indeed, such thermodynamic stability changes have been measured for specific disease related proteins^17–23, including several proteins linked to cancer.^24–26 The thermodynamic analyses of disease-related proteins performed, to date, have largely involved purified protein systems that were already found to be associated with disease states. However, such thermodynamic analyses have not been used for the global analysis of disease states. This is in large part because methods for making protein folding and thermodynamic stability measurements on the proteomic scale have only recently been developed.^27–37 Because protein stability changes can arise for such a wide range of functionally relevant reasons, they have the potential to be a general probe of protein function and to be sensitive to a number of different types of functionally relevant and disease-related changes. The characterization of such functionally relevant changes in different disease states has the potential to create new diagnostic signatures of disease states and to produce a better molecular level understanding of the disease, ultimately facilitating the discovery of novel drug therapies.

Recently, we reported on the use of SILAC-SPROX for making differential measurements of protein folding and stability.³⁰ The SILAC-SPROX technique is an extension of the SPROX methodology in which the chemical denaturant dependence of a hydrogen peroxide-mediated oxidation of methionine side chains in proteins is used to measure the thermodynamic properties of proteins and protein-ligand complexes.³⁸ The SPROX technique has been used to measure the folding free energies and dissociation constants of several model proteins and protein-ligand systems.^27,29,38 To date, SILAC-SPROX has been used to analyze the protein folding behaviors of proteins in cell lysates both in the presence and absence of a specific ligand to identify protein target of the ligand.³⁰

Evaluated here is the potential of thermodynamic stability measurements to characterize disease states. The SILAC-SPROX technique is used for the differential thermodynamic analysis of proteins in three well-established cell culture models of breast cancer including the non-tumorigenic MCF-10A breast cell line and the MCF-7 and MDA-MB-231 breast cancer cell lines that have different molecular features and display different degrees of invasiveness. The thermodynamic stability profiles generated here successfully differentiated the three cell lines. The described thermodynamic profiling strategy has the potential to create a new avenue for biomarker discovery.

MATERIALS AND METHODS

Cell Culture and Cell Lysate Preparations

The MCF-10A, MCF-7, and MDA-MB-231 cells were cultured in a humidified 37°C incubator with 5% CO₂. All cell lines were cultured following the ATCC guidelines except for MCF-7 heavy SILAC- labeled cells which were cultured using heavy-labeled lysine and arginine according to established protocols (see Supporting Information for additional details).^39–41 The heavy labeled lysine was enriched with six ¹³C atoms and two ¹⁵N atoms. The heavy labeled arginine was enriched with six ¹³C atoms.

Cells were lysed by mechanical disruption using a disruptor genie (Scientific Industries) and 1 mM diameter zirconia/silica beads (Biospec) in 20 mM phosphate buffer (pH 7.4) containing a cocktail of protease inhibitors that included: 1 mM AEBSF, 500 µM Bestatin, 15 µM E-64, 20 µM Leupeptin, and 10 µM Pepstatin A (Thermo Pierce). The cell lysates were centrifuged at 15000 X g for 15 min at 4°C, and the supernatants were used in the SILAC-SPROX analyses. The total protein concentration of the cell lysates used in each cell line comparison was normalized to the same total protein concentration, which was 3–10 mg/ml depending on the experiment.

SILAC-SPROX Analyses

Two different SILAC-SPROX comparisons were performed here including one that involved the MCF-7 and MCF-10A cell lines and one that involved the MCF-7 and MDA-MB-231 cell lines. Three biological replicates of each SILAC-SPROX experiment were performed. The SILAC-SPROX³⁰ in solution protocol was followed. Briefly, in each biological replicate the test cell lysates were diluted into a series of twelve denaturant-containing buffer stock solutions. The denaturant-containing buffer stock solutions contained 20 mM phosphate buffer (pH 7.4) with urea concentrations ranging from 0 to 9 M, and 75 µL volumes were combined with the 20 µL sample aliquots of the lysates from respective cell lines. The final denaturant concentrations in each set of denaturant-containing buffers ranged from 0 to 7 M. The samples were equilibrated 16–18 hours in the denaturant-containing buffers. The methionine oxidation reactions were initiated by addition of hydrogen peroxide (0.54 M (Sigma)) to the samples in each set of denaturant-containing buffers. The oxidation reactions were allowed to proceed for 6 minutes before quenching each reaction with the addition of 760 µl of a 375 mM solution of L-methionine (Sigma). After quenching each reaction, the samples from the two different cell lines in buffers containing the same denaturant concentration in each SILAC-SPROX analysis were combined. The combined samples were precipitated with TCA (Sigma), and the resulting protein pellets in each of the combined samples were subjected to a quantitative, bottom-up proteomics analysis using SILAC quantitation.

Proteomic Sample Preparation

The dried protein pellets from the SILAC-SPROX analyses were dissolved in 60 µl of 0.5 M triethyl ammonium bicarbonate (TEAB) (Sigma) containing 3 µl of a 2% stock solution of sodium dodecyl sulfate (SDS). The disulfide bonds in each protein sample were reduced upon addition of 5 µl of 50 mM tris(2-carboxyethyl)phosphine hydrochloride (TCEP) (ThermoFisher) and treatment for 1 hour at 60°C. The protein samples were each reacted with 2.5 µl of 200 mM methyl methane thiosulfonate (MMTS) (Sigma) for 10 minutes at room temperature to block cysteine side chains. Ultimately, 3 µl of a 1 µg/µl trypsin solution was added to the protein sample in each tube and the samples were incubated overnight (~15 hours) at 37°C before 5 µl of 10% trifluoroacetic acid (TFA) was added to quench the trypsin digestion.

Liquid Chromatography-Tandem Mass Spectrometry Analyses

The tryptic peptides were desalted using C18 resin (The Nest Group, Southborough, MA) according to the manufacturer’s protocol and analyzed by liquid-chromatography-tandem MS (LC-MS/MS) using an EASY-nLC 1000 system coupled to a Q-Exactive (quadrupole-orbitrap) mass spectrometer (Thermo Fisher Scientific). Columns were packed in-house with Jupiter 4µ Proteo 90Å reversed phase resin (Phenomenex). Peptides were concentrated and desalted on a trapping column (100µm ID x 20 mm) and eluted on an analytical column (75µm ID x 150 mm), operating at 300 nl/min and using the following gradient: 5% B for 3 min, 5–35% B in 120 min, 35–80% B in 2 min, and 80% B for 9 min [solvent A: 0.1% formic acid (v/v); solvent B: 0.1% formic acid (v/v), 80% CH₃CN (v/v) (Fisher Scientific)]. The Q-Exactive was operated in a data-dependent MS/MS mode using the top 10 most intense precursors detected in a survey scan from 300 to 1,800 m/z performed at 70K resolution. Tandem MS was performed by HCD fragmentation with stepped normalized collision energy (NCE) of 20%.

Proteomic Data Analysis

Peak lists were extracted from the raw LC-MS/MS data files and the data were searched against the 20265 human proteins in the 2014-04 release of the UniProt Knowledgebase (downloaded on 5/16/2014 at ftp://ftp.uniprot.org/pub/databases/uniprot/current_releases/release-2014_04/knowledgebase/) using Maxquant 1.3.0.5.⁴² The following modifications were used: MMTS at cysteine as a fixed modification, SILAC labeling of lysine (¹³C₆ ¹⁵N₂) and arginine (¹³C₆), and variable (0–1) oxidation of methionine and deamidation of Asparagine and Glutamine (N and Q), and acetylation of the protein N-terminus. The enzyme was set as Trypsin, and up to 2 missed cleavages were permitted. The false discovery rate for peptide and protein identifications was set to 1%, and rest of the parameters were set at the default settings. As part of the default settings the mass tolerance for precursor ions was set to 20 ppm for the first search where initial mass recalibration was completed and a 6 ppm precursor mass tolerance was used for the main search. The mass tolerance for fragment ions was 0.5 Da. We also included match between runs and re-quantification of the searched peptides. The search results were exported to Excel for further data analysis as described below. In cases where identified peptides could be matched to multiple protein isoforms or multiple members of a protein family, the peptide was assigned to the leading razor protein listed by MaxQuant algorithm. The mass spectrometry proteomics data have been deposited to the ProteomeXchange Consortium (http://proteomcentral.proteomeexchange.org) via the PRIDE partner repository⁴³ with the dataset identifiers PXD001847 and PXD001848.

Only the protein and peptide identifications in the Maxquant output with H/L ratios >0 were used in subsequent data analysis steps. The methionine-containing peptides were selected, and those methionine-containing peptides consistently identified in the protein samples derived from six or more denaturant-containing buffers were assayed. For the methionine-containing peptides, a single averaged H/L ratio was calculated for each peptide sequence and each charge state at each denaturant concentration. Excel spreadsheets containing these averaged H/L ratios for the methionine-containing peptides and proteins assayed in these analyses are included in the supplementary material (Tables S1–S6 in the Supporting Information). Similarly, for each analysis a median H/L ratio was determined for each protein using the H/L ratios measured for all the non-methionine-containing peptides identified in all the denaturant concentrations for a given protein. These median H/L ratios were used to select hits with H/L>2 fold in the protein expression level analyses.

For hit peptide and protein selection in the thermodynamic analyses, all the H/L ratios generated for the non-methionine containing peptides from a given protein were divided by the median H/L ratio for that protein in order to generate normalized H/L ratios for each non-methionine containing peptide. These normalized H/L ratios were log₂ transformed. The normalized and log₂ transformed H/L ratios generated for the non-methionine-containing peptides in a given analysis were used to determine the 5^th and 95^th percentile values used in subsequent analysis of methionine-containing peptides (Supplementary Figure S1). The averaged H/L ratios calculated for each methionine-containing peptides were also normalized and log₂ transformed. The methionine-containing peptides and proteins with log₂ transformed H/L ratios less than the 5^th percentile or greater than the 95^th percentile values determined above were selected and then visually inspected to determine which peptides had altered H/L ratios at 2 or more consecutive denaturant concentrations to generate an initial list of protein hits.

In order to differentiate the true-positives (i.e., proteins with altered stabilities in the cell line comparisons) from the false-positives, the hit peptides identified in the differential analyses were analyzed for consistency across the three biological replicates. Peptide hits with inconsistencies in the biological replicates (e.g., a peptide was identified as hit in one biological replicate and not in another, different charge states showed different SILAC-SPROX behavior, or the peptide retention times were not consistent in the LC-MS/MS analyses) were deemed false positives and eliminated from the final protein hit lists.

Protease Activity Assays

The calpain and cathepsin D activity assay kits (ab65308 and ab65302, respectively) were both purchased from Abcam®. The activity assays were performed according to the manufacturer’s protocol using either an Envision® or Victor³ V Multilabel Plate Reader (both from Perkin Elmer). The 405 nm excitation and 515 nm emission filters were used in the calpain activity assay, and the 340 nm excitation and 460 nm emission filters were used in the cathepsin D activity assay. In each case the activity assays were performed in duplicate using cell lysates identical to those used in SILAC-SPROX analyses. In the calpain activity assay, fluorescence measurements were recorded on ~30 µg of the total lysate in the absence and presence of 0.5 µl of the calpain inhibitor provided with the kit. In the Cathepsin D activity assay, fluorescence measurements were recorded on ~20 µg of the total lysate in the absence and presence of 10 µl of a 1mM stock solution of pepstatin A. In each case the specific activity of the enzyme was measured by subtracting the fluorescence intensity of the sample with the inhibitor from the one without the inhibitor and normalizing to the total protein concentration used in each reaction.

Quantitation of Protein Stability Changes

The protein folding free energy changes (ΔΔG_f values) determined in this work were calculated using equation 1, which has been previously described.³⁸

$$\mathrm{\Delta }\mathrm{\Delta }{\mathrm{G}}_{\mathrm{f}}=-m\times \mathrm{\Delta }{\mathrm{C}}_{1/2}$$ (Eq. 1)

In equation 1, m is the δΔG_f/δ[Denaturant] where ΔG_f is the folding free energy and ΔC_1/2 is the range of denaturant concentrations at which the H/L ratios of the hit peptide deviate from the protein baseline (i.e., the relative expression level of the protein in the two disease states (see Figure 1)). All the ΔΔG_f values in this work were calculated using an estimated m-value of 1.3 kcal mole⁻¹M⁻¹. This estimated m-value was based on an average protein domain size of 100 amino acids and an average contribution to the m-value of 0.013 kcal mole⁻¹M⁻¹ per residue. This per residue value was empirically derived by Myers and co-workers.⁴⁴

Figure 1.

Schematic representation of the experimental workflow used in this study. Aliquots containing equal amounts of total protein from the cell lysates generated in this study were distributed into a series of denaturant-containing buffers. The protein samples were equilibrated in the denaturant-containing buffers and reacted with hydrogen peroxide under conditions that enable the selective oxidation of solvent accessible methionine residues in proteins. The oxidation reactions are quenched, and the light- and heavy-labeled samples in buffers with matching denaturant concentrations are combined. The combined protein samples are submitted to a conventional bottom-up proteomics analysis in which the H/L ratios of the methionine-containing peptides are evaluated as a function of denaturant concentration. Shown at the bottom are the protein unfolding curves that correspond to the SILAC-SPROX behaviors shown for the wild type methionine-containing peptides. Protein unfolding curves are not directly measured in SILAC-SPROX. However, shifts in the unfolding curves can be determined and are represented by ΔC_1/2 values.

RESULTS AND DISCUSSION

Global analysis of protein folding thermodynamics

The experimental workflow used in this study is outlined in Figure 1. The estrogen positive (ER+) and poorly invasive MCF-7 breast cancer cell line was heavy-labeled and used in two different comparative studies. In one study protein stabilities in the heavy-labeled MCF-7 breast cancer cell line were compared to those in a non-tumorigenic MCF-10A cell line. In the second study protein stabilities in the heavy-labeled MCF-7 breast cancer cell line were compared to those in a highly invasive, triple negative (i.e., estrogen, progesterone, and human epidermal growth factor 2 receptor negative) breast cancer cell line, MDA-MB-231. Each comparative study utilized the SILAC-SPROX method³⁰ to compare the thermodynamic stabilities of the proteins in each cell line. Proteins with different chemical denaturation behavior in the two cell lysate samples were identified based on the H/L ratios determined for the methionine-containing peptides identified in a quantitative bottom-up proteomics analysis of the combined protein samples generated in each comparative analysis (see Figure 1).

Shown in Table 1 is a summary of the proteomic data obtained in the MCF-7 versus MCF-10A and MCF-7 versus MDA-MB-231 cell line comparisons. In each of the comparative analyses approximately 800 protein stabilities were surveyed using ~2500 methionine-containing peptide probes (see Table 1). Approximately 10–12% of the methionine-containing peptides assayed in each of the comparative analyses were identified as hits (i.e., yielded different chemical denaturation curves in the comparative analyses). This hit rate is significantly above the false positive rate of 3.5% that we have previously reported for hit discovery using SILAC-SPROX³⁰, indicating that a number of the detected hits are true positives.

Table 1.

Proteomic Coverage of peptides and proteins assayed for thermodynamic stability changes in the MCF-7 Vs MCF-10A and MCF-7 vs MDA-MB-231 cell line comparisons.

MCF-7 Vs MCF-10A
Biological Replicate	Total Peptides (Proteins) Identified	Total Peptides (Proteins) Assayeda	Total Peptides (Proteins) Hitsb
1	5582 (1394)	987 (432)	120 (80)
2	12578 (1788)	1751 (576)	155 (92)
3	11039 (1863)	1570 (538)	128 (70)
Total	16582 (2431)	2472 (772)	336 (175)
Biological Replicate	MCF-7 Vs MDA-MB-231
1	12504 (2249)	1457 (609)	118 (82)
2	3051 (1210)	548 (359)	35 (34)
3	11299 (1852)	1595 (531)	132 (98)
Total	18147 (2872)	2602 (881)	274 (169)

^aAssayed peptides are the methionine-containing peptides that were consistently identified in SILAC-SPROX samples from at least six denaturant concentrations in the SILAC-SRPOX experiment. The assayed proteins are the proteins to which the assayed peptides map.

^bHit peptides are the methionine-containing peptides with altered H/L ratios at 2 or more consecutive denaturant concentrations in the SILAC-SPROX experiment. The hit proteins are the proteins to which the hit peptides map.

In order to differentiate the true-positives from the false-positives, the hit peptides were analyzed for consistency across the three biological replicates. Approximately 50% of the peptide hits in Table 1 had inconsistent data in the biological replicates. The final hit list contained 138 peptide hits from 84 unique proteins and 128 peptide hits from 102 unique proteins in the MCF-7 versus MCF-10A and MCF-7 versus MDA-MB-231 cell line comparisons, respectively (see Table S7A and B in the Supporting Information). In these final hit lists 38 and 23% of the protein hits were identified as hits in two or more replicates and/or with multiple methionine-containing peptide probes in the MCF-7 versus MCF-10A and MCF-7 versus MDA-MB-231 comparisons (respectively). The remaining protein hits included those that were only identified as hits in only one experimental replicate with a single methionine-containing peptide.

The peptide hits displayed one of three different SILAC-SPROX behaviors (see Figure 2). About half of the final hits detected in each of the two comparisons showed the behavior expected for methionine-containing peptide probes from globally protected regions in proteins (or protein folding domains) that were differentially stabilized in the disease states (Figure 2A). The other hits in each comparison had methionine-containing peptide probes with H/L ratios that were significantly altered across all the denaturant concentrations used in each experiment (Figure 2B and C). Some of these also showed a denaturant dependence to their H/L ratios at least in one cell line (Figure 2B), while others did not (Figure 2C). The SILAC-SPROX behaviors observed in Figures 2B and C would be expected of methionine-containing peptide probes derived from regions of protein structure that are either solvent exposed or only locally protected in one cell line and globally or sub-globally protected in the other cell line (Figure 2B) or one exposed and the other locally protected (Figure 2C) in the disease states.

Figure 2.

Representative SILAC-SPROX data obtained in the two cell line comparisons in this work. (A) Data obtained on peptides from calpain small subunit 1 in the MCF-7 versus MCF-10A comparison including the wild-type (filled circles) and doubly oxidized (filled squares) forms of the methionine-containing peptide SMVAVMDSDTTGK (detected in biological replicates 3 and 2, respectively) as well as the median data for all the non-methionine-containing peptides from calpain small subunit 1 that were detected in biological replicate 2 (open squares). (B) Data obtained on peptides from splicing factor 3B subunit 3 in biological replicate 2 of the MCF-7 versus MCF-10A comparison including the wild-type form of the methionine-containing peptide AVMISAIEK (filled circles) and the median data for all the non-methionine-containing peptides (open squares) from the protein. (C) Data obtained on peptides from peptidyl-prolyl cis-trans isomerase proteinin biological replicate 3 of the MCF7 versus MDA-MB-231 comparison including the oxidized form of the methionine-containing peptide (ac)MVNPTVFFDIAVDGEPLGR (closed squares) and the median data for all the non-methionine-containing peptides (open squares) from the same protein. Shown on the bottom panels are schematic representations of the expected unfolding curves that produced the observed SILAC-SPROX behavior in each example.

Quantitation of Protein Stability Changes

The protein stability changes were quantified for hit peptide probes from globally protected regions in proteins (or protein folding domains) with SILAC-SPROX behavior like that shown in Figure 2A. A total of 129 peptide probes from the two cell line comparisons in this work displayed SILAC-SPROX behavior like that shown in Figure 2A. The quantitative assessment of ΔC_1/2 values (and ultimately ΔΔG_f values using equation 1) requires the successful detection and quantitation of peptide probes in SILAC-SPROX samples covering enough denaturant concentrations to evaluate both the departure and return of H/L ratios to the protein baseline. A subset of 52 of the above 129 peptide probes included sufficient data points in the SILAC-SPROX experiment to enable a ΔC_1/2 value determination. Approximately half of the 52 estimated ΔC_1/2 values were the result of destabilizations in the MCF-7 cell lysates, with the other half being stabilized in the MCF-7 cell lysate. The ΔC_1/2 values determined for the stabilization and destabilizations detected for the 52 peptide probes ranged from 1.3 to 4.6 M, which corresponded to ΔΔG_f values of 1.7 to 6 kcal/mol. Only a qualitative assignment of either stabilizations or destabilizations of the remaining 77 peptide probes were possible. These quantitative and qualitative assignments of the destabilizations and stabilizations detected in both cell line comparisons are summarized in Tables SI–7A and B.

Hit Proteins Map to a Wide Range of Protein Expression Levels

The SILAC-SPROX approach is especially attractive for disease-state analyses because the quantitative proteomic data collected on non-methionine containing peptides can be used to evaluate protein expression level changes. This allowed for a concurrent analysis of the expression changes associated with the differentially stabilized protein hits identified in this work. The protein hits with significantly altered thermodynamic stabilities showed a wide range of expression level changes (Figure 3 and Table S7A and B in the Supporting Information). The protein expression levels measured in the MCF-7 and MCF-10A comparison in this work were generally in good agreement with those reported in a previous protein expression level study on the same cell lines (Figure 3 and Table S7A in the Supporting Information).⁴⁰ Approximately, 45% of the protein hits identified with significantly altered thermodynamic stabilities in the two comparisons had expression levels that changed less than two-fold in the comparative analyses (see Table S7A and B in the Supporting Information). These proteins with similar expression levels and different thermodynamic stabilities are likely to have altered functions in the cancer cell lines studied here. These functional differences would go undetected in a typical expression level-based proteomic study.

Figure 3.

Subset of protein hits detected in the comparative analyses performed here. Shown in (A), (B), and (C) are the gene names of proteins that were detected as hits in both cell line comparisons, only the MCF-7 versus MCF-10A comparison, and only the MCF-7 versus MDA-MB-231 comparison, respectively. The numbers in the boxes represent the protein expression level data (i.e., Log₂(H/L) values) either measured in this work or previously reported in the literature for each protein hits. The H/L values were averaged in cases where the hit peptide was identified in replicate experiments. RSD values were typically +/−15%. Errors were not reported for the literature values, which are taken from reference.⁴⁰ The protein hits highlighted here are those that were assayed with the same peptide probe in both cell line comparisons.

Protein Signatures for Disease State Differentiation

The MCF-10A, MCF-7, and MDA-MB-231 cell lines used here are well-studied cell line models of breast cancer, and a number of the protein hits identified in this work have been previously associated with cancer. All the protein hits in our comparative studies that were identified in multiple biological replicates have been previously associated with tumorigenesis and/or the progression of cancer in expression level and/or other studies (see Supplementary Table S7A and B and references therein). All but two (protein disulfide-isomerase A4 and stress-induced-phosphoprotein 1) have been previously associated with breast cancer in expression level and/or other studies (see Supplementary Table S7A and B and references therein).

Several proteins involved in cell building and metabolic processes also displayed differential thermodynamic properties in our study. These included cytoskeletal proteins such as vimentin, keratin, and myosin that are known to have differentially modulated expression level in cancer.^45–47 A significant number of glycolytic enzymes also showed differential thermodynamic stabilities in the normal to cancer transition. These included proteins such as GAPDH, LADHB, TPI1 and ENO1 that are reported to be dysregulated in many cancers.^48–52 These glycolytic enzymes have been implicated in tumorigenesis and proliferation of cancer cells including in breast cancer and have been widely studied for diagnostic and therapeutic purposes.^48,49,52 This helps validate the biological significance of the protein targets discovered here using the described approach.

While many of the protein hits discovered here have been previously associated with cancer, not all associations were discovered using gene and protein expression level changes. For example, the increased catalytic activity of calpain has been associated with ER+ breast cancer even though the protein’s expression level is unchanged.⁵³ Similarly, the post-translational modification of myosin-9 has been shown to regulate the invasiveness of cancer cells in studies on MDA-MB-231 cell lines.⁵⁴

The protein hits with altered thermodynamic stabilities in the two comparative studies described here were classified into three groups based on whether or not they were detected as hits in either one or both of the cell line comparisons described here. A total of 80 of the 162 protein hits with altered thermodynamic stabilities in the two cell line comparisons, were assayed with the same methionine-containing peptide probe in both cell line comparisons. Thus, only these 80 proteins hits could be unambiguously classified into one of the three groups described below. These 80 protein hits included 13 proteins and 16 methionine-containing peptide probes with altered thermodynamic stabilities in both the MCF-7 versus MCF-10A and MCF-7 versus MDA-MB-231 cell line comparisons (Figure 3A and Table S7C in the Supporting Information). A total of 15 of these 16 probes had similar SILAC-SPROX behavior in each comparison (see Figure S2 in Supporting Information), indicating the thermodynamic stability changes in the proteins or protein domains probed by these 15 peptide probes (Figure S2) were specific to the MCF-7 cell line. Thus, these proteins have the potential to serve as biomarkers for MCF-7 specific disease-related changes.

The different SILAC-SPROX behavior of the HEAM(ox)ITDLEER peptide probe from myosin-9 (Figure S2C) in the two comparisons, indicates that this protein has different protein folding behavior in all three cell lines. The SILAC-SPROX behavior of the HEAM(ox)ITDLEER peptide probe in the two comparisons suggests that myosin-9 is destabilized in both of the cancer cell lines, but more so in the MBA-MD-231 cell line, suggesting that the thermodynamic stability of myosin-9 can be used to differentiate breast cancer subtypes. Our observation that myosin-9 was highly destabilized in the MDA-MB-231 cell lysate could be because of the altered phosphorylation of this protein in MDA-MB-231 cells, which has been previously reported.⁵⁴ The destabilization could result from altered protein-protein interactions and/or conformational changes induced by the altered phosphorylation. However, more studies are still required to connect the stability difference to the altered phosphorylation of myosin-9 in the cell lines.

A total of 30 proteins were assayed with the same 58 methionine-containing peptide probes in both comparative studies and only identified as hits in the MCF-7 versus MCF-10A comparison (Figure 3B and Table S7D in the Supporting Information). The thermodynamic stability changes probed by these 58 methionine-containing peptides were similar in the two cancer cell lines but different with respect to the non-tumorigenic control cell line. These 30 proteins represent general breast cancer biomarkers for the normal to cancer transition. Similarly, 37 proteins were assayed with the same 39 methionine-containing peptide probes in both comparative studies and only identified as hits in the MCF-7 versus MDA-MB-231 comparison (Figure 3C and Table S7E in the Supporting Information). These 37 proteins have MDA-MB-231 specific protein folding properties and may be useful biomarkers for the diagnosis and further characterization of triple negative (i.e., ER-, PgR- and HER2-) tumors. Together, these results show that thermodynamic stability profiles can be used to differentiate the cell line models of breast cancer studied here, and they suggest that such thermodynamic stability profiles can be useful for disease state characterizations.

Correlating Changes in Thermodynamic Stability with Changes in Function

The thermodynamic stability changes observed for the protein hits identified in this work could arise for a wide variety of functionally relevant changes in proteins (e.g., point mutations, post-translational modifications, and new and/or altered binding interactions with cellular ligands). Significantly, the potential effects (both direct and indirect) of such mutations and PTMs on a protein’s folding stability can be detected in the SILAC-SPROX experiment without actually detecting a methionine-containing peptide probe that includes the mutation or PTM site. This is because the chemical denaturation of a single methionine-containing peptide probe can report on the folding behavior of the protein or protein domain to which it maps. However, the SILAC-SPROX experiment does not identify the biological significance of the specific stability changes observed for each protein hit. Additional studies are required to understand the biological significance of the specific stability changes observed for each protein hit.

Biological activity assays performed with hit proteins that have known catalytic functions provide one way in which to correlate a protein hit’s altered stability with a change in function. As part of this work, the catalytic activities of two hit proteins, the calpain small subunit 1 (CAPNS1) and Cathepsin D, were assayed to determine if the altered thermodynamic stabilities of these proteins could be correlated with changes in their known proteolytic activity. The proteolytic activity of CAPNS1, a protein hit with increased stability in the MCF-7 cell line, was evaluated using a known peptidic substrate of calpain that was spiked into MCF-10A and MCF-7 cell lysates. The catalytic activity of calpain was approximately 2.5-fold greater in the MCF-7, ER+ cell line compared to the activity measured in the normal cell line model (Figure 4A). These catalytic activity measurements are consistent with earlier studies of calpain activity in cell lysates from normal and ER+ breast cancer tissues, in which the proteolytic activity of calpain was also found to be greater in ER+ breast cancer tissues than in normal breast tissues.⁵³

Figure 4.

Catalytic activity data for calpain protease in MCF-7 lysate and MCF-10A lysate (A) and for cathepsin D protease in MCF-7, MCF-10A, and MDA-MB-231 cell lysates (B). Error bars represent +/− one standard deviation from duplicate measurements.

The catalytic activity of Cathepsin D was assayed in the MCF-10A, MCF-7, and MDA-MB-231 cell lines using a known peptidic substrate that was spiked into cell lysates derived from each cell lines. The thermodynamic stability of Cathepsin D was decreased in the MCF-7 and MDA-MB-231 cell lysates compared to that in the MCF-10A cell lysate. The catalytic activity of Cathepsin D was slightly enhanced in the MCF-7 cell line and slightly reduced in the MDA-MB-231 cell line relative to the activity measured in the MCF-10A cell line (Figure 4B). Thus, in the case of Cathepsin D there was not a correlation between the protein’s proteolytic activity and its decreased stability in the MCF-7 and MDA-MB-231 cell lysates. The decreased stability observed for cathepsin D in the cancer cell lysates is likely tied to other cancer related functions, or dysfunctions, of the protein. The results of previous studies have shown that cathepsin D’s proteolytic activity is unrelated to cancer cell proliferation and progression in MCF-7 cell lines.⁵⁵ Thus, the altered thermodynamic stability of cathepsin D in the two cancer cell lines may be a more biologically relevant biomarker than the protein’s catalytic activity.

In the case of protein hits for which a catalytic activity is not differentially modulated (such as was observed for Cathepsin D) or for which cell-based activity assays do not exist, other studies are required to connect a protein hit’s altered stability with a change in function. For example, immunoprecipitation strategies can be used to purify the “hit” proteins from the different cell lysates so that additional SPROX experiments can be performed to determine if the originally detected stability differences in the unpurified samples are also observed in the purified protein samples. Detection of the same stability difference would suggest that a differential post-translational modification or other modifications to the “hit” protein (e.g., a point mutation) might be responsible for the detected stability change. In this case, the isolated protein samples could be subjected to bottom-up and/or top-down MS-based strategies to more completely characterize the covalent structure of the protein. On the other hand, if the isolated protein samples do not show similar stability differences in the purified and unpurified samples, it may be that the originally observed stability difference in the unpurified sample was the result of an altered protein interaction network. In such cases, a differential proteomic analysis of the protein complexes pulled down with the “hit” protein in the immunoprecipitation experiments could be used to help dissect the protein interaction networks that may be altered in the biological states being compared.

Classification of Proteins with Altered Thermodynamic Stability

A bioinformatics analysis was performed using PANTHER^56,57 to characterize the distribution of the 84 and 102 protein hits identified in the two cell line comparisons. In general, the biological processes and molecular functions represented in the protein hits identified using the thermodynamic data in this work were similar to the biological processes and molecular functions represented in the protein hits identified using the protein expression level data generated in this work (Supplementary Tables S8–S11). However, one exception was the higher percentage of hits with structural functions observed in the thermodynamic analyses (Supplementary Table S8). Analysis of the protein classes represented in the hits also revealed that cytoskeletal proteins and chaperone proteins were enriched in the thermodynamic analyses (Figure 5).

Figure 5.

Heat map showing the distribution of protein classes observed in the protein hits identified from the thermodynamic and protein expression level analyses performed in this work. The numbers in the boxes represent the percentage of protein hits from each class in each experiment.

The additional molecular space probed using the thermodynamic stability measurements in this study enabled distinctions between the breast-cancer-related disease states in this study that were not possible using more conventional measurements of protein expression levels. The bioinformatics analysis of the protein hits identified in the two cell line comparisons revealed that the fraction of hydrolases hits in the MCF-7 versus MDA-MB-231 cell line comparison was more than double that found in the MCF-7 versus MCF-10A comparison (Figure 5). This is consistent with activity-based profiling studies, in which serine hydrolase activity profiles were found to differentiate MCF-7 and MDA-MB-231 cell lines.¹⁶ Our results add to the growing evidence that hydrolases are important for cancer cell growth and invasiveness. Significantly, the expression level data does not reveal such a signature, as the percentage of hydrolase hits in the expression level analysis was similar in the two subtypes of the breast cancer cell line models (Figure 5). This is especially noteworthy as the fraction of hydrolases in the total assayed proteins in the thermodynamic and expression level analyses was similar (Supplementary Table 12).

CONCLUSIONS

This study represents the first use of global measurements of thermodynamic stability to characterize disease states. This proof-of-principle work involved the analysis of cell-culture models of breast cancer. The use of well-established cell lines was especially convenient in this initial work as the cell lines provide a convenient source of highly homogeneous cells representing initial tumors. While it has been shown that some cell lines, including the breast cancer cell line in this study, can faithfully represent many features of cancer cells in vivo⁵⁸, it is certainly more ideal to directly study clinical samples. In theory the described approach can also be applied to the analysis of tumor tissues cell lysates derived from clinical samples provided sufficient amounts of protein (~2–3 mg of total protein) can be obtained from such samples. It is also possible that protein hits identified with stability changes in cell line analyses could be validated in clinical samples using the SPROX methodology to directly measure the stability of potential hits in tumor tissue cell lysates. Such validation studies could utilize targeted proteomics methods, which require significantly less sample than the shot-gun proteomics methods described here. Alternatively, the physical and/or functional basis associated with changes resulting from the measured protein stability differences in hit proteins could be exploited to validate protein hits in clinical samples.

The differential stability measurements described here create a novel platform for the characterization of disease states. The results of this proof-of-principle work demonstrate that global measurement of thermodynamic stability can be used to detect disease-related changes in mammalian cell lines. The discovery of hit proteins in this comparative study with a wide range of altered expression levels suggests that the described methodology can be used to complement protein expression level studies. The thermodynamic stability measurements strategy presented here shows promise for both validating protein targets from other strategies utilizing proteomic platforms such as protein expression levels and also for uncovering novel molecular targets in future disease-related characterization studies.