Analysis of copper-induced protein precipitation across the E. coli proteome

Abstract

Metal cations have been exploited for their precipitation properties in a wide variety of studies, ranging from differentiating proteins from serum and blood to identifying the protein targets of drugs. Despite widespread recognition of this phenomenon, the mechanisms of metal-induced protein aggregation have not been fully elucidated. Recent studies have suggested that copper's (Cu) ability to induce protein aggregation may be a main contributor to Cu-induced cell death. Here, we provide the first proteome-wide analysis of the relative sensitivities of proteins across the Escherichia coli proteome to Cu-induced aggregation. We utilize a metal-induced protein precipitation (MiPP) methodology that relies on quantitative bottom–up proteomics to define the metal concentration–dependent precipitation properties of proteins on a proteomic scale. Our results establish that Cu far surpasses other metals in promoting protein aggregation and that the protein aggregation is reversible upon metal chelation. The bulk of the Cu bound in the protein aggregates is Cu¹⁺, regardless of the Cu²⁺ source. Analysis of our MiPP data allows us to investigate underlying biophysical characteristics that determine a protein's sensitivity to Cu-induced aggregation, which is independent of the relative concentration of protein in the lysate. Overall, this analysis provides new insights into the mechanism behind Cu cytotoxicity, as well as metal cation–induced protein aggregation.

Graphical Abstract

Graphical Abstract.

Copper surpasses other metal cations in inducing protein precipitation. A quantitative proteomics strategy was used to define the Cu precipitation midpoint, or C_m value, of individual proteins across the E. coli proteome, providing insight into attributes that influence protein susceptibility, or tolerance, to Cu-induced protein misfolding. (Graphic created with BioRender.com)

Introduction

The susceptibility of proteins to precipitation by metal cations is a well-documented phenomenon, with the effect of copper (Cu) salts, and to a lesser extent other d-block metals, noted as early as 1907 in experiments to differentiate proteins from serum and blood.^1,2 Like the precipitating effects of electrolytes in the Hofmeister series,^3–5 the biophysical basis behind metal-induced protein precipitation (MiPP) remains poorly understood, even though it has been used for practical applications.^6–8 Recently, metal-induced protein precipitation has even been applied as a method to identify protein targets of drugs.⁹

Understanding how metal-induced protein aggregation relates to cellular dysfunction has been investigated intensively, particularly in the context of neurodegenerative disease.^10–12 Several recent studies have identified impaired protein folding and aggregation as inherent features of Cu-induced cytotoxicity in both prokaryotic and eukaryotic cells.^13–16 These recent studies suggest there may be a more general connection between metal impaired protein folding and aggregation with cytotoxicity. These studies, as well as those showing iron–sulfur cluster proteins as targets of excess Cu,^17,18 have expanded the mechanistic understanding of cellular Cu toxicity beyond conventional models of oxidative stress to include aberrant Cu–protein interactions and misfolding as key contributors.^19,20 Here, we refer to Cu without specifying oxidation state when its redox status is unknown, ambiguous, or in an environment where it could redox cycle between Cu¹⁺ and Cu²⁺.

To prevent unfavorable metal interaction and reactivity with biomolecules, cells rely on transcriptional regulation of various metal homeostasis proteins to control the import, export, storage, and utilization of metals by cells and organisms.²¹ The stringency of transcriptional control generally follows the Irving–Williams series, with labile concentrations of metals highest on this list of relative complex stability being held to the lowest levels.²² Cu²⁺ being the highest on the Irving–William series of divalent metal cations is therefore thermodynamically favored to mis-metallate other proteins. The transcriptionally controlled restriction of labile Cu to attomolar levels in cells implies its inherent danger to cellular function if it becomes available for adventitious interactions.

The potential connection between aberrant Cu binding and protein misfolding has inspired numerous studies aimed at understanding the mechanistic and biophysical origins of individual proteins prone to aggregation induced by Cu.^23–26 The relative susceptibility of cellular proteins to this effect, however, has not been elucidated at the proteomic scale. What are the consequences across the proteome when Cu homeostasis is breached? Which proteins preferentially interact with Cu, and to what consequence? Answering these questions is relevant to understanding aspects of how our immune system works, as it is thought to deploy Cu intoxication mechanisms to intensify pathogen killing.^19–23 The cytotoxic effects of Cu have also generated significant interest in developing compounds that leverage its antimicrobial, antiviral, and anticancer activity in targeted ways.^27–29

Here we examine the relative susceptibility of proteins across the Escherichia coli proteome to metal-induced precipitation with Cu. As part of this work, we examine the amino acid composition and secondary structure of proteins that may influence their relative sensitivity to metal-induced precipitation. The susceptibility of proteins to precipitate upon addition of Cu is probed using a MiPP methodology that relies on a quantitative bottom–up proteomics readout using an isobaric mass tagging strategy to define the metal concentration–dependent precipitation of proteins on a proteomic scale. The MiPP methodology allows for the determination of Cu precipitation midpoints, C_m values, for each identified protein. The resulting C_m values provide quantitative means to assess the sensitivity of proteins across the E. coli proteome to Cu precipitation. Subsequent analyses of the proteins that were more and less tolerant to MiPP revealed information about the biophysical properties that correlate with the MiPP tolerant and sensitive proteins.

Materials and methods

Materials

All materials were purchased and used without any further purification from Sigma Aldrich, Thermofisher Scientific, Corning, BioSpec Products, and Merck.

Preparation of cell lysates

Escherichia coli MG1655 expressing β-lactamase CTX-M-1³⁰ was streaked onto Luria Broth (LB) agar containing 100 μg/mL ampicillin and 50 μg/mL kanamycin. A single colony was used to inoculate 7–10 mL LB medium containing 100 μg/mL ampicillin and 50 μg/mL kanamycin, which was then incubated at 37°C, 200 rpm, for 16–18 h. This overnight culture was diluted 1:100 in LB medium and grown to an OD₆₀₀ of 0.8–1.4, pelleted, and washed twice with 5 mL of water. The resulting pellets were frozen at −20°C for later use. Cell pellets were thawed, then lysed in 200 μL of PBS (without Ca or Mg, 600 mM KCl added, pH 7.4) with 1 mM of the protease inhibitor phenylmethylsulfonyl fluoride. Six cycles of cell lysis were accomplished by sonication at 30% amplitude for 10 s followed by a 50 s incubation period on ice. The lysed cells were centrifuged at 14 000 ×g for 10 min at 4°C. The total protein concentration in the supernatant from each cell lysate sample was determined by a Bradford assay and ranged from 25 to 40 mg/mL. Lysate concentrations were adjusted to 10 or 25 mg/mL.

Total protein precipitation curves after exposure to metal salts

Concentration-adjusted cell lysates were divided into 10 μL aliquots, to which 10 μL of a metal salt solution was added to reach final concentrations of 0–50 mM, as indicated in Fig. 1, and allowed to equilibrate while shaking at room temperature (RT) for 1 h. The samples were then diluted by addition of water to a final volume of 250 or 500 μL and centrifuged at 14 000 ×g for 20 min at 4°C. The dilution step was necessary to create a supernatant with sufficient volume for protein quantitation assays. The protein concentration of the supernatant was then determined by a Bradford assay or nanodrop reading.

Fig. 1.

Cu induces reversible protein precipitation to a greater extent than other biorelevant metals. (A) Addition of Cu²⁺ precipitated significantly more proteins from E. coli than other metals at 10 mM. n = 3. (B) Precipitation curves indicated that Fe²⁺, Co²⁺, Ni²⁺, and Zn²⁺ did not induce complete precipitation even at 50 mM. n = 2 for all metals except Cu²⁺, for which n = 3. (C) Structures of glutathione (GSH), ethylenediaminetetraacetic acid (EDTA), dl-dithiothreitol, 1,4-dithiothreitol (DTT), and bathocuproinedisulfonic acid disodium salt (BCS). (D) EDTA (5 mM), BCS (50 mM), DTT (50 mM), and GSH (20 mM) allowed protein to return to the supernatant after treatment of E. coli lysate with 10 mM CuCl₂. Water did not rescue protein solubility, for which n = 3. After precipitation with heat, neither 5 mM EDTA nor water was able to rescue protein solubility. Lysate protein concentrations for all experiments were 12.5 mg/mL.

Total protein precipitation measurements after exposure to hydrogen peroxide

Concentration-adjusted cell lysates were divided into 10 μL aliquots, to which 10 μL of 200 mM hydrogen peroxide or water were added and allowed to equilibrate while shaking at RT for 1 h. The samples were then diluted by addition of water to a total volume of 500 μL and centrifuged at 14 000 ×g for 20 min at 4°C. The protein concentration of the supernatant was then determined by a Bradford assay.

Protein solubility rescue experiments

Escherichia coli lysate was prepared and normalized to 25 mg/mL and divided into 10 μL aliquots. To one sample was added 10 μL of an 18 mM CuCl₂ solution, as described earlier. For the thermal denaturation experiment, 10 μL of water was added to 10 μL lysate and heated at 65°C while shaking for 30 min. The samples were then diluted by addition of 480 μL of water and centrifuged at 14 000 ×g for 20 min at 4°C prior to measuring the protein concentration of the supernatant by a Bradford assay. The pellets were then resuspended in the supernatant by vortexing and treated with final concentrations of 5 mM ethylenediaminetetraacetic acid (EDTA), 50 mM bathocuproinedisulfonic acid disodium salt (BCS), 50 mM 1,4-dithiothreitol (DTT), 20 mM glutathione (GSH), or water (all pH 7.4–8.5) for 1 h at RT. The samples were centrifuged at 14 000 ×g for 20 min at 4°C to pellet precipitates and the protein concentration of the resulting supernatant was quantified by a Bradford assay and compared to an untreated sample to determine percentage protein rescue. Protein solubility rescue experiments were performed in biological triplicate.

Large-scale MiPP analysis

Escherichia coli lysates were subjected to a large-scale MiPP analysis using a protocol similar to that performed for the total protein precipitation curves. The only variation from the previous protocol is that aliquots of the supernatant (soluble protein) at each concentration of metal underwent bottom–up proteomics sample preparation and subsequent quantitative liquid chromatography mass spectrometry/mass spectrometry (LC–MS/MS) analysis instead of quantitation by a Bradford assay. Such analysis made it possible to obtain precipitation curves for individual proteins identified and quantified in the LC–MS/MS sample.

The MiPP analysis involved combining 10 μL aliquots of a cell lysate (10 or 25 mg/mL) with 10 μL of Cu stock (0–18 mM) to obtain samples with final concentrations of 5 or 12.5 mg/mL lysate and CuCl₂ concentrations ranging from 0 to 9 mM across 10 samples. Lysates were allowed to incubate with metals for 1 h at RT. At this point in the protocol precipitated protein was visible in samples containing the highest concentrations of metals. Samples were diluted with 480 μL of water, vortexed, and centrifuged at 14 000 ×g at 4°C for 20 min.

A fixed volume of the resulting samples (containing a maximum of 100 μg protein for the 0 mM metal treatment condition) was taken from the supernatant of each of the 10 samples for bottom–up proteomics sample preparation using an Isobaric Mass Tagging with Filter-Aided Sample Preparation (iFASP) protocol described by McDowell et al.³¹ Each sample was transferred into a 10 kDa Molecular Weight Cut-Off (MWCO) centrifugal filter unit. Buffer exchange was performed by adding 8 M urea in 0.1 M Tris-HCl pH 8.5 followed by tris(2-carboxyethyl)phosphine (TCEP) reduction, methyl methanethiosulfonate (MMTS) alkylation, digestion with trypsin, and TMT10-Plex labeling according to manufacturer's protocol. Labeled peptides were centrifuged through the filters after addition of 0.5 M NaCl. Equal volumes from each TMT10-Plex labeled sample were combined into one final sample. The final sample was transferred to a C18 Macrospin column for desalting prior to LC–MS/MS.

Inductively coupled plasma–mass spectrometry analysis

Samples were prepared for inductively coupled plasma–mass spectrometry (ICP–MS) by adding 10 μL of CuCl₂ stock solution to 10 μL of lysate (10 mg/mL) to reach final concentrations of 5 mg/mL protein and 0–14 mM CuCl₂. Samples were incubated at RT while shaking for 1 h, then diluted with 230 μL de-ionized (DI) water and centrifuged at 14 000 ×g at 4°C for 20 min. The supernatant was separated and saved for further analysis. The pellet was washed with metal-grade water and then digested in 100 μL trace-metal grade nitric acid at 90°C for 1 h. The pellet samples were diluted with 900 μL of 1% trace-metal grade nitric acid.

Samples were run on an Agilent 7900 ICP–MS in He mode. Both the supernatant and pellet digestion were diluted into a solution of 2% (v/v) trace-metal grade nitric acid with 0.5% (v/v) trace-metal grade hydrochloric acid spiked with 20 ppb ⁷²Ge and ¹¹⁵In as internal standards. Elements were measured under He atmosphere (collision cell) to reduce polyatomic interferences with a He gas flow rate of 4.3 mL/min. Data were quantified using weighed, serial dilutions of a multi-element standard [CLMS-2AN (SPEX CertiPrep, CL51-042CRYA) Cu]. Seven replicates were measured per sample. A pre-mixed drinking water standard (CRMTDWA from High Purity Standards) was used as an initial calibration verification (20 μg/L). The average value was 19.00 μg/L (n = 12), within 10% of the expected value. Internal standards of ⁷²Ge and ¹¹⁵In were kept within ±10% of the expected values.

Electron paramagnetic resonance spectroscopy

Two E. coli samples were prepared by mixing an aliquot of cleared cell lysate (prepared as described in preparation of cell lysates section) with a solution of CuCl₂ to make a total of 250 μL with final concentrations of 6 mM Cu and 5 mg/mL protein. Samples were allowed to equilibrate while shaking at RT for 1 h, then centrifuged at 14 000 ×g and 4°C for 20 min. The supernatant was separated from the pellet. One of the supernatant samples was treated with 100 mM hydroxylamine for 15 min (to inhibit catalase activity) followed by addition of 1 M H₂O₂ for 10 min, while the other sample was treated with DI water prior to transfer into electron paramagnetic resonance (EPR) tubes, which were carefully submerged in liquid nitrogen. The pelleted samples were subjected to a wash in DI water followed by centrifugation at 14 000 ×g and 4°C for 20 min. The final washed pellets were resuspended in DI water (done in nitrogen-filled glove box for sample without H₂O₂). Either hydroxylamine and H₂O₂ or an equivalent volume of water was added to the resuspended pellets, similar to the supernatant samples. Samples were frozen in liquid nitrogen for analysis.

X-band continuous wave EPR spectroscopy was conducted on a Bruker ESP 300 spectrometer equipped with an Exford Instruments ESR 910 continuous helium flow cryostat. Experiments were conducted at 77 ± 1 K, 20 mW microwave power, and 5 G modulation amplitude.

Ellman's reagent assay for measuring free sulfhydryls

Two E. coli samples were prepared by mixing an aliquot of cleared cell lysate (prepared as described in preparation of cell lysates section) with a solution of CuCl₂ or water to make a total of 250 μL with final concentrations of 0 mM Cu or 6 mM Cu and 5 mg/mL protein. Samples were allowed to equilibrate while shaking at RT for 1 h followed by treatment with 20% trichloroacetic acid for a final concentration of 5% (w/v) and centrifugation at 14 000 ×g and 4°C for 20 min to precipitate all proteins from the samples and leave only small molecule thiol-containing compounds to be assayed. Fully reduced controls were prepared by reacting an aliquot of 100 μL of each test sample with 25 μL of a solution of 3.5 M NaBH₄ with 1.5 M NaOH in a 50/50 (v/v) methanol: water. To measure the free thiol content of each test and control sample, 20 μL portions of each sample were mixed with 220 μL of 10 mM EDTA in 0.5 M TRIS buffer at pH 8.2 followed by addition of 20 μL of 10 mM 5,5′-dithio-bis-(2-nitrobenzoic acid) (Ellman's reagent/DTNB). Absorbance readings of the released TNB⁻ were measured at 412 nm. After accounting for dilution, the ratio of absorbance readings of the test samples to their corresponding fully reduced control samples was used to calculate the percentage free sulfhydryls in each condition. Procedure adopted from Alisik et al.³²

Chemical denaturant and protein precipitation analysis

Escherichia coli lysates were subjected to a chemical denaturation and protein precipitation (CPP) analysis using a protocol reported previously by Meng et al.³³ The CPP analysis involved combining 13 μL aliquots of a cell lysate (35 mg/mL) with 7 μL of a range of GdmCl-containing buffers at varying concentrations. This resulted in final concentrations of 23 mg/mL lysate and GdmCl concentrations ranging from 0 to 3 M across 10 samples. Lysates were allowed to incubate with denaturants for 1 h at RT. Protein precipitation was then induced by fast dilution upon the addition of 480 μL of water to each of the 10 samples. After the addition of water, samples were quickly vortexed, and then centrifuged at 14 000 ×g at 4°C for 20 min. A fixed volume was taken from the supernatant of each of the 10 samples for bottom–up proteomics sample preparation using an iFASP protocol described by McDowell et al.³⁴ Each sample was transferred into a 10 kDa MWCO centrifugal filter unit. Buffer exchange was performed by adding 8 M urea in 0.1 M Tris-HCl pH 8.5 followed by TCEP reduction, MMTS alkylation, digestion with trypsin, and TMT10-Plex labeling according to manufacturer's protocol. Labeled peptides were centrifuged through the filters after addition of 0.5 M NaCl. Equal volumes from each TMT10-Plex labeled sample were combined into one final sample. The final sample was transferred to a C18 Macrospin column for desalting prior to LC–MS/MS.

Stability of proteins from rates of oxidation analysis

Escherichia coli lysates were subjected to a stability of proteins from rates of oxidation (SPROX) analysis using a protocol reported previously Strickland et al.³⁵ The SPROX analysis involved combining 10 μL aliquots of a cell lysate (10 mg/mL) with 26 μL of a range of GdmCl-containing buffers at varying concentrations. This resulted in final concentrations of 2.8 mg/mL lysate and GdmCl concentrations ranging from 0.25 to 2.5 M across 10 samples. Lysates were allowed to incubate with denaturants for 30 min at RT. Oxidation of exposed methionine residues was initiated upon addition of 4 μL of H₂O₂ (0.98 M final concentration) to each of the 10 samples. The oxidation reaction was quenched by adding a 500 μL aliquot of 500 mM TCEP. Each of the quenched samples was subjected to a bottom–up proteomics sample preparation using an iFASP protocol described by McDowell et al. and similar to that for the CPP analysis described earlier.³¹ Each sample was transferred into a 10 kDa MWCO centrifugal filter unit. Buffer exchange was performed by adding 8 M urea in 0.1 M Tris-HCl pH 8.5 followed by TCEP reduction, MMTS alkylation, digestion with trypsin, and TMT10-Plex labeling according to manufacturer's protocol. Labeled peptides were centrifuged through the filters after addition of 0.5 M NaCl. Equal volumes from each TMT10-Plex labeled sample were combined into one final sample. The final sample was transferred to a C18 Macrospin column for desalting before undergoing methionine peptide enrichment using a Pi³ Methionine reagent kit according to manufacturer's protocol. The enriched sample underwent one final C18 desalting step prior to LC–MS/MS analysis.

Quantitative LC–MS/MS analysis

The LC–MS/MS analyses for the E. coli metal-induced precipitation and SPROX samples were performed on a Thermo Easy nanoLC 1200 coupled to a Thermo Orbitrap Exploris 480 mass spectrometer system. The dried peptide material generated from each experiment was reconstituted in 1% trifluoroacetic acid (TFA), 2% acetonitrile in H₂O. Aliquots of 2 μL (1 μg peptide) were injected in triplicate into the ultra-performance liquid chromatography (UPLC) system. The peptides were first trapped on a Thermo Acclaim PepMap 100 75 μm × 2 cm, nanoViper 2Pk C18, 3 μm, 100 Å trapping column. The analytical separation was performed using a PepMap RSLC C18 2 μm, 100 Å, 75 μm × 25 cm column (Thermo); the column temperature was set to 45°C. Peptide elution was performed using a 95 min linear gradient of 4–40% B (80:20 acetonitrile: water, 0.1% formic acid) at a flow rate of 400 nL/min. The MS data were collected using a top 20 data-dependent acquisition (DDA) method, which included MS1 at 120 k and MS2 at 45 k resolution. The MS1 normalized Automatic Gain Control (AGC) target was set to 300%. For MS2, the normalized AGC target was set to 300% with a max injection time of 105 ms. The collision energy was set to 36%, and the scan range was 375–1500 m/z. The isolation window was 1.2 and the dynamic exclusion duration was 45 s. The LC–MS/MS analyses for the E. coli CPP samples were performed on a nanoAcquity UPLC (Waters) coupled to a Thermo Orbitrap Fusion Lumos mass spectrometer system. The dried peptide material generated from each experiment was reconstituted in 1% TFA, 2% acetonitrile in HPLC grade water. Aliquots of 1 μL were injected in triplicate into the UPLC system. The peptides were first trapped on a Symmetry C18 20 mm × 180 μm trapping column with a flow rate of 5 μL/min at 99.9/0.1 water/acetonitrile v/v. The sample was then analytically separated using the Acquity 75 μm × 250 mm high strength silica (HSS) T3 C18 column with a 1.8 μm particle size (Waters) T3 with the column temperature set at 55°C. The samples were eluted with a 90 min linear gradient of 3–30% acetonitrile at a flow rate of 400 nL/min. The MS data were collected using a top 20 DDA method, which included MS1 at 120k and MS2 at 50k resolution. The MS1 normalized AGC target was 4.0 × 10⁵ ions and max injection time of 50 ms. The MS2 AGC target was 1.0 × 10⁵ ions and max injection time of 105 ms. The collision energy was set to 38%, and the scan range was 375–1500 m/z. The isolation window was 0.7 and the dynamic exclusion duration was 60 s.

MiPP proteomic data analysis

Proteome Discoverer 2.3 (Thermo) was used to search the raw LC–MS/MS files against the E. coli MG1655 (Proteome ID: UP000000625) proteins in the 2019-09-24 release of the UniProt Knowledgebase. The raw LC–MS/MS data generated in the protein expression experiments was searched using fixed MMTS modification on cysteine; TMT10-plex labeling of lysine side chains and peptide N-termini; variable oxidation of methionine; variable deamidation of asparagine and glutamine; and variable acetylation of the protein N-terminus. Trypsin (full) was set as the enzyme, and up to two missed cleavages were allowed. For peptide and protein quantification, reporter abundance was set as intensity, and normalization mode and scaling mode were each set as none. All other settings were left as the default values. Data were exported at the protein level for quantitation. Only proteins with False Discovery Rate (FDR) confidence labeled as “high” or “medium” (i.e. FDR < 1% or <5%) were used for subsequent analyses. Proteins with any of the first four TMT-tag signal intensities equaling zero were removed before these analyses.

The data were subject to two normalizations prior to curve fitting. First, the intensities for individual proteins across all metal concentrations were normalized to the intensities of the 0 mM metal point. This enabled scaling of all the data within the same range. Second, the intensities of all the proteins within a single metal concentration (TMT tag) were averaged and plotted as a function of metal concentration. This dataset was fit using Mathematica to a four-point sigmoidal equation to extract a supercurve:

(1)

The differences from the fitted supercurve values to the average value at each metal concentration were determined to look for systematic errors and deviations from supercurve behavior. These differences were corrected for across all proteins by adding the calculated difference in average TMT-tag intensity from the supercurve at each metal concentration to all individual proteins at that metal concentration. The resulting normalized data were imported back into Mathematica to fit individual protein precipitation curves to the same four-point sigmoidal equation (eq. 1). Metal precipitation midpoint values (C_m values) of protein precipitation curves were exported for further statistical analyses as well as fitting P-values for variables “b” and “c.” Proteins with poorly fitted curves (fitting P-values <0.01) were removed before further analysis.

Statistical analysis of protein biophysical characteristics

Python scripts were developed to determine the primary and secondary structure of proteins in the E. coli proteome and aid in the identification of structural features that influence MiPP. First, the unique accession number for each E. coli protein was compiled using the FASTA database on UniProt. Next, the accession numbers were used to extract the primary sequence of each protein from the Uniprot protein database. From the primary sequence, the script calculates a variety of characteristics (i.e. number of cysteine, cationic, or hydrophobic residues) for each protein in the E. coli proteome. The Uniprot database also compiles a list of published structures for each protein, and this list is annotated with the proportion of protein resolved in each structure. For proteins with published structures that contained greater than 90% of the protein, the structure with the highest coverage was used for secondary structure analysis. The structures were analyzed using the DSSP program, which calculates the hydrogen bonding energy between all atoms in the 3D structure of a protein to determine the secondary structure at each residue. Using this output, the script identified the proportion of each structure that consisted of helical, stranded, and unstructured regions. In this work, this script enabled the automated analysis of proteins identified as tolerant or sensitive to MiPP, but the script can also be extended to assist in the primary and secondary structure analysis of any protein found in the Uniprot database.

Escherichia coli proteins present in three of the five biological replicates for 12.5 mg/mL data or three of the four for the 5 mg/mL data were used for the analyses. A Welch's ANOVA was performed to determine significant differences in amino acid and secondary structure composition between sensitive and tolerant protein groups (P-values <0.05).

CPP proteomic data analysis

Proteome Discoverer 2.3 (Thermo) was used to search the raw LC–MS/MS files against the E. coli MG1655 (Proteome ID: UP000000625) proteins in the 2019-09-24 release of the UniProt Knowledgebase. The raw LC–MS/MS data generated in the protein expression experiments was searched using fixed MMTS modification on cysteine; TMT10-plex labeling of lysine side chains and peptide N-termini; variable oxidation of methionine; variable deamidation of asparagine and glutamine; and variable acetylation of the protein N-terminus. Trypsin (full) was set as the enzyme, and up to two missed cleavages were allowed. For peptide and protein quantification, reporter abundance was set as intensity, and normalization mode and scaling mode were each set as none. All other settings were left as the default values. Data were exported at the protein level for quantitation. Only proteins with FDR confidence labeled as “high” or “medium” (i.e. FDR < 1% or <5%) were used for subsequent analyses.

To fit individual protein CPP curves, chemical precipitation datasets were fitted to a four-parameter sigmoidal equation, eq. 1, using a Mathematica-based program (developed in house). CPP midpoint values of the curves were exported for further statistical analyses as well as fitting P-values for variables “b” and “c.” Proteins with poorly fitted curves (fitting P-values <0.01) were removed before further analysis.

SPROX proteomic data analysis

Proteome Discoverer 2.3 (Thermo) was used to search the raw LC–MS/MS files against the E. coli MG1655 (Proteome ID: UP000000625) proteins in the 2019-09-24 release of the UniProt Knowledgebase. The raw LC–MS/MS data generated in the protein expression experiments were searched using fixed MMTS modification on cysteine; TMT10-plex labeling of lysine side chains and peptide N-termini; variable oxidation of methionine; variable deamidation of asparagine and glutamine; and variable acetylation of the protein N-terminus. Trypsin (full) was set as the enzyme, and up to two missed cleavages were allowed. For peptide and protein quantification, reporter abundance was set as intensity, and normalization mode and scaling mode were each set as none. All other settings were left as the default values. Data were exported at the protein level for quantitation. Only proteins with FDR confidence labeled as “high” or “medium” (i.e. FDR < 1% or <5%), peptide spectral match (PSM)s ≥2, and having no TMT-tag abundances of zero were used for subsequent analyses. The data were normalized as previously described in reference Strickland et al.³⁵

To fit individual peptide SPROX curves, chemical denaturation datasets were fitted to a four-parameter sigmoidal equation, eq. 1, using a Mathematica-based program (developed in house). This program fit each set of data nine times, once with all eight points and then eight more times, each time leaving out a different one of the eight data points. The fit with the highest R² was chosen as the final output. SPROX midpoint values of the curves were exported for further statistical analyses as well as fitting P-values for variables “b” and “c.” Proteins with poorly fitted curves (fitting P-values <0.01) were removed before further analysis.

Results

Cu²⁺ and other divalent metals cause proteins to precipitate out of E. coli lysate

Initially examined was the metal ion dependence of the proteins in an E. coli cell lysate to precipitation by a panel of divalent metal chloride salts of Mg²⁺, Ca²⁺, Mn²⁺, Co²⁺, Ni²⁺, Cu²⁺, and Zn²⁺, as well as di- and trivalent Fe in the form of ferrous ammonium sulfate and ferric ammonium citrate. After 1 h, measurement of the total protein concentration remaining in the supernatants revealed that exposure to 10 mM Cu²⁺ precipitated all the protein from solution, while the same concentration of Zn²⁺ and Fe²⁺ caused only partial protein precipitation, and the other metals had no effect (Fig. 1A). Replacing CuCl₂ with CuSO₄ resulted in the same level of observed protein precipitation, confirming that the nature of the counterion does not affect the MiPP reaction (Supplementary Fig. S1).

The precipitation behavior was studied in more detail to examine the concentration dependence of Fe²⁺, Co²⁺, Ni²⁺, Cu²⁺, and Zn²⁺ (Fig. 1B). These results did not strictly follow the Irving–Williams series, as Fe²⁺ precipitated more protein than Co²⁺ and Ni²⁺, raising intriguing questions about what factors drive MiPP. The observation that even at concentrations up to 50 mM, only Cu²⁺ caused complete protein precipitation further raises the question: why is Cu so much more effective at precipitating proteins than other d-block metals? Given the uniquely dramatic effect of Cu²⁺ on protein precipitation, an observation consistent with other reports,¹⁴ we focused on this metal for subsequent MiPP proteomics studies.

Cu-induced precipitation is reversible upon addition of metal chelators

Since many factors can influence protein unfolding and precipitation, we were interested in exploring how the properties of the Cu-induced protein precipitation compared to those induced by other unfolding and precipitation mechanisms. To test the reversibility of Cu-induced protein precipitation, the precipitated protein samples were treated with metal chelators. The proteins in an E. coli lysate were precipitated by reaction with 9 mM CuCl₂ as described earlier. The resulting pelleted proteins were resuspended within the supernatant and then treated with a strong Cu²⁺ chelator (EDTA), a strong Cu¹⁺ chelator (BCS), a good disulfide bond reducing agent and Cu¹⁺ binder (DTT), a good Cu²⁺-to-Cu¹⁺ reducing agent and Cu¹⁺ binder (GSH), or by simple dilution with water and allowed to incubate for 1 h (Fig. 1C).³⁶ The resulting samples were centrifuged to pellet any remaining insoluble protein(s), and the concentration of the resolubilized protein in the supernatant was determined. Treatment with 5 mM EDTA, 50 mM BCS, 50 mM DTT, and 20 mM GSH significantly restored protein solubility of the total protein (Fig. 1D). On the other hand, only a small fraction was resolubilized upon dilution with water. Interestingly, these data clearly show that protein precipitation induced by added Cu²⁺ is reversible upon the removal of Cu²⁺ or Cu¹⁺ by chelation. It is likely that during the resolubilization procedure, the presence of EDTA favors the oxidation of Cu¹⁺ to Cu²⁺ upon chelation under these ambient conditions. For comparison, proteins that were thermally denatured and precipitated were not resolubilized upon addition of EDTA or dilution with water (Fig. 1D), consistent with the irreversible mechanism of thermal denaturation and precipitation.

Pelleted proteins reach Cu saturation and contain a mix of Cu²⁺ and Cu¹⁺

The cell lysate samples in the MiPP experiments in this work were treated with Cu²⁺; however, the lysate is a reducing environment that contains ∼5 mM GSH, which is known to reduce Cu²⁺ and bind to Cu¹⁺.^37–39 Therefore, an investigation was carried out to determine the distribution and oxidation state of Cu across both the soluble and insoluble fractions following reaction with Cu²⁺. Supernatants and pellets from samples prepared as described for Fig. 1A were separated and analyzed individually by ICP–MS to quantify Cu content and EPR to determine oxidation state. As shown in Fig. 2A, the amount of Cu in the pellet reached saturation around 6 mM of added Cu. In contrast, the amount of Cu in the supernatant initially plateaued around 2 mM and remained constant until about 5 mM, when the concentration of Cu in the supernatant was directly proportional to the added Cu. Interestingly, at approximately 6 mM added Cu, there was nearly complete protein precipitation at this lysate protein concentration (Fig. 2A: green triangles). This result suggests that Cu binds directly to proteins with a finite number of binding sites that reach saturation. The protein aggregates themselves do not accumulate additional Cu after this saturation point has been reached.

Fig. 2.

Cu precipitates with proteins during Cu-induced protein precipitation and has a mixture of Cu¹⁺ and Cu²⁺ in the soluble and insoluble fractions. (A) The left y-axis shows a comparison of the amount of Cu measured by inductively coupled plasma–mass spectrometry (ICP–MS) found in the pellet vs. the supernatant at each Cu treatment condition for E. coli lysate with a protein concentration of 5 mg/mL. The right y-axis shows the relative fraction of protein in the supernatant for each Cu treatment condition, using the same biological replicates used for the ICP–MS data. Points represent mean and error bars represent ±1 SEM; n = 3. (B) Representative electron paramagnetic resonance (EPR) spectra of resuspended pellet (left) and supernatant (right) with and without hydrogen peroxide. Lysate treated with 6 mM Cu displayed a small Cu signal (red), which became more prominent after treatment with 100 mM hydroxylamine and 1 M hydrogen peroxide (blue) in both the supernatant and pellet. Replicates are shown in Supplementary Fig. S2. (C) Bar graph showing the percentage of reduced sulfhydryls in lysate treated with and without 6 mM Cu assayed by reaction with 5,5′-dithio-bis-(2-nitrobenzoic acid) (DTNB; Ellman's reagent). The bars represent the mean and error bars represent ±1 SEM; n = 3.

Cu²⁺ is EPR active due to its unpaired electron while Cu¹⁺ is EPR silent and therefore not detectable by this method. In order to assess the relative amount of EPR-silent Cu¹⁺, EPR spectra were collected before and after exposure of the samples to hydrogen peroxide, which oxidizes all the Cu in the sample to Cu²⁺. The change in EPR signal thereby reveals the latent Cu¹⁺. Quantitative analysis of the data revealed that the relative amount of each Cu oxidation state differed between the pellet and the supernatant. After peroxide oxidation the pellet generated at 6 mM added Cu²⁺ actually contained 66 ± 9% Cu¹⁺, while the supernatant remained mostly Cu²⁺ with only 37 ± 8% reduced Cu¹⁺ (Supplementary Fig. S2).

In order to probe how the oxidation status of GSH/GSSG was affected by Cu treatment, the percentage of small molecule free sulfhydryls present in lysate after global protein precipitation was determined through a colorimetric assay using Ellman's reagent (DTNB). The assay compares the ratio of DTNB-reactive thiols of test samples to control samples that are fully reduced by reaction with sodium borohydride. The protocol includes a step to precipitate all proteins so that only small molecule thiols are assayed, of which GSH is expected to be the major component. As shown in Fig. 2C, thiols in lysates not treated with Cu²⁺ are already in their fully reduced form. This result confirms that GSH in the lysate samples does not autoxidize under the ambient conditions and timeframe that the cell lysates are handled for these experiments. On the other hand, samples treated with 6 mM Cu²⁺ only contained 15 ± 7% of the total free sulfhydryls, implying that the bulk of the GSH has been oxidized to GSSG. Combined with the EPR data mentioned earlier, these results suggest that GSH is the likely culprit in the redox reaction that reduces Cu²⁺ to Cu⁺ while being oxidized from GSH to GSSG.

Large-scale analysis of MiPP using proteins in an E. coli cell lysate

Having seen the global scale precipitation effects of Cu on an E. coli lysate, we were intrigued to identify the relative susceptibility of individual proteins across the proteome to Cu-induced precipitation. The proteins in an E. coli lysate were subject to the MiPP proteomic workflow outlined in Fig. 3. Ultimately, the relative amount of each identified protein remaining in solution at each Cu concentration was determined in a quantitative bottom–up proteomics readout using isobaric mass tags. For each identified protein, the concentration of Cu at which half of the protein precipitated from solution was determined as the Cu precipitation midpoint, or C_m value (Fig. 3). The earlier analysis was performed on the proteins in E. coli cell lysates prepared at two different concentrations of total protein, 5 and 12.5 mg/mL, to produce two datasets containing nearly 700 and 1000 proteins, respectively. The proteins in each dataset were ordered from lowest to highest C_m (Supplementary Table S1), with the proteins in the top quartile (C_m values of 0.26–1.05 and 1.37–2.49 for 5 and 12.5 mg/mL datasets, respectively) and bottom quartile (C_m values of 2.05–6.54 and 3.96–5.87 for 5 and 12.5 mg/mL datasets, respectively) categorized as sensitive and resistant, respectively, to Cu-induced precipitation. The 12.5 mg/mL dataset resulted in 261 proteins in each of the sensitive and tolerant categories, while the 5 mg/mL resulted in 167 proteins in each of these categories. A summary of the identified proteins in each dataset and their respective C_m values is included in Supplementary Table S1.

Fig. 3.

Proteomic workflow employed in the large-scale metal-induced protein precipitation (MiPP) analysis across the soluble E. coli proteome. Created with BioRender.com.

Analysis of C_m values reveals important trends within MiPP data

An analysis of the C_m values measured in the MiPP experiments using total protein concentrations of 12.5 and 5 mg/mL revealed a protein concentration dependence. The median C_m of the proteins in the 5 mg/mL experiment (1.80 mM) was lower than that observed in the 12.5 mg/mL experiment (3.61 mM) (Fig. 4A). A comparison of C_m values for each of the 621 overlapping proteins between the two datasets showed a consistent shift to a lower C_m value in the 5 mg/mL dataset than in the 12.5 mg/mL dataset, with exception of just 4 proteins (Fig. 4B). The average C_m shift [i.e. ΔC_m (C_m12.5–C_m5)] was 1.52 ± 0.55 mM. The results of the two MiPP experiments performed at different protein concentrations also revealed that a large majority (65–80%) of the MiPP tolerant or sensitive proteins assayed, in both datasets, were consistently identified as such (Fig. 4C and D).

Fig. 4.

Cu midpoint values across identified E. coli proteins decrease as protein concentrations decrease. (A) Box and whisker plot of the C_m values recorded for E. coli proteins across the proteome in the MiPP experiments performed using E. coli lysate samples at 5 and 12.5 mg/mL concentrations of total protein. (B) Histogram showing the distribution of ΔC_m values (ΔC_m = C_m12.5–C_m5). (C, D) Venn diagrams showing the overlap of sensitive (C) and tolerant (D) proteins that were assayed in both the 5 and 12.5 mg/mL datasets.

In order to test the Gaussian nature of the distribution of C_m values, the data collected on the proteins across the E. coli proteome were also subjected to the Kolmogorov-Smirnov test, which revealed P-values for the 5 and 12.5 mg/mL MiPP datasets (Fig. 4A) of 0.001 and 0.010, respectively. This analysis indicates that the protein distributions are significantly different from that of a Gaussian distribution, suggesting that the distributions do not arise from a random property but rather from an inherent feature of the samples.

We hypothesized that the non-Gaussian behavior of MiPP C_m values arises from unique features of the identified proteins; therefore, we set out to find correlations between C_m values and various biophysical properties. Initially investigated was the correlation between the Cu MiPP midpoint C_m values obtained here with midpoint C_1/2 values obtained from protein folding stability measurements obtained by using chemical and thermal denaturation approaches, specifically CPP, SPROX, and thermal protein profiling (TPP).^40–42 As part of this work, CPP and SPROX data were collected to determine the chemical denaturant-induced unfolding properties of the proteins in an E. coli lysate. The unfolding transition midpoints generated in these CPP and SPROX experiments (Supplementary Tables S2 and S3) were compared to the C_m determined in the earlier MiPP experiments. The C_m values determined here by MiPP were also compared to melting temperature data (T_m values) previously reported for proteins in an E. coli cell lysate.⁴³ As shown in Fig. 5, the MiPP data did not correlate with SPROX data (regression P-value 0.438). While CPP and TPP data both had statistically significant correlations with MiPP (regression P-values <0.001), their R² values suggested that these models only explain a small percentage of the variance, 8.7% and 5.1% respectively, indicating the absence of a strong correlation between these datasets.

Fig. 5.

Copper precipitation midpoints lack strong correlation with other proteomic methods. (A) Linear regression comparing Cu midpoint values (C_m) to stability of proteins from rates of oxidation (SPROX) midpoint values (C_1/2). y = 0.02x + 1.2. R² = 0.002. (B) Linear regression comparing Cu C_m to thermal protein profiling (TPP) midpoint T_m values. y = 1x + 48. R² = 0.051. (C) Linear regression comparing Cu C_m values to chemical denaturation and protein precipitation (CPP) midpoint values (C1/2). y = 0.04x + 0.8. R² = 0.087. The regressions displayed here use the 5 mg/mL dataset. The same analyses were done on the 12.5 mg/mL dataset (Supplementary Fig. S3) and resulted in correlations nearly identical to those shown here.

Gene ontology (GO) enrichment analyses performed using protein analysis through evolutionary relationships⁴⁴ failed to reveal any consistent significant enrichments in biological process or molecular function across both MiPP datasets, suggesting that the determining factor of a protein's sensitivity to Cu-induced precipitation is not influenced by its biological function. When using the GO term for annotated metal ion binding proteins provided by European Bioinformatics Institute, metal ion binding proteins were found across all categories: sensitive, average, and tolerant. The lack of correlation between metal–ion binding annotation and C_m value or category suggests that having a known metal–ion binding site is not a defining feature of a protein's sensitivity to Cu-induced precipitation.

Methionine oxidation was also considered as a potential driving force behind the Cu-induced protein precipitation reaction. Indeed, we did observe higher levels of methionine oxidation in the Cu MiPP samples than in untreated samples. A global analysis of the proteomics data collected on the 5 mg/mL MiPP biological replicates revealed that 58% of the methionine-containing peptides were identified in their oxidized form. This is in contrast to that observed in an untreated sample, where only 28% of the methionine-containing peptides were identified in their oxidized form. However, while the overall number of oxidized methionine-containing peptides increased with Cu treatment, methionine oxidation levels in the MiPP sensitive and tolerant proteins were not a distinguishing feature of the proteins in these groups. Proteins in the sensitive and tolerant groups were all found to be oxidized at a wide range of different levels (Supplementary Tables S4a–d), suggesting that Met oxidation is not a differentiating factor in Cu-induced protein precipitation. Furthermore, when H₂O₂ was used to induce Met oxidation no protein precipitation was observed (Supplementary Fig. S4). Collectively, this analysis suggests that the increased Met oxidation within MiPP samples is a consequence of the precipitation process but does not drive it.

Secondary structure and amino acid composition of sensitive and tolerant proteins

To further understand the biophysical nature of MiPP, we investigated the secondary structure and amino acid compositions of the tolerant and sensitive proteins. A Welch's ANOVA, which does not rely on the compared groups having a normal distribution or equal variance, was used to determine if there were significant differences in the secondary structure compositions of the proteins from each group (Fig. 6). This analysis was done on proteins consistently classified as sensitive or tolerant in both the 5 and 12.5 mg/mL proteomic datasets and with known NMR or x-ray structures mapping to greater than or equal to 90% of a protein's amino acid sequence. These factors resulted in the secondary structure analysis of 82 sensitive proteins and 106 tolerant proteins for the 5 mg/mL dataset as well as 134 sensitive proteins and 160 tolerant proteins for the 12.5 mg/mL dataset. Multiple secondary structure classifications were shown to have higher or lower percentages in the sensitive proteins compared to the tolerant proteins as determined by P-values ≤0.05 (Fig. 6A). Unstructured regions were more prevalent in proteins tolerant to metal precipitation, while helical regions were more prevalent in sensitive proteins (Fig. 6B).

Fig. 6.

Amino acid analysis and secondary structure. (A) Heat map showing Welch's test P-values, which were found from comparing secondary structure composition of sensitive and tolerant protein groups identified from 5 mg/mL lysate samples. (B) Box and whisker plots showing distributions of secondary structure fraction for individual proteins in both the sensitive and tolerant categories. Each individual dot represents a single protein from the 5 mg/mL dataset. Only secondary structures that were found to have a Welch's test P-value <0.05 in both the 5 and 12.5 mg/mL dataset are displayed. (C) Heat map showing Welch's test P-values, which were found from comparing the amino acid composition of sensitive and tolerant protein groups identified from 5 mg/mL lysate samples. (D) Box and whisker plots showing distributions of amino acid compositions for individual proteins in both the sensitive and tolerant categories. Only amino acids that were found to have a Welch's test P-value <0.001 in both the 5 and 12.5 mg/mL dataset are displayed. For a full list of Welch's test P-values and analysis on every amino acid and secondary structure in both the 5 and 12.5 mg/mL datasets see Supplementary Tables S5 and S6.

A comparison of the total number of amino acids per protein in the sensitive and tolerant groups showed a significant difference (P-value = 0.002 and <0.001 for 5 and 12.5 mg/mL datasets, respectively), with sensitive proteins tending to be larger than tolerant proteins (Supplementary Tables S5 and S6). This finding indicates that protein size may play a role in a protein's precipitation tolerance, which could be linked to a protein's solvent accessibility with the sensitive proteins being bigger and thus having greater solvent accessible surface area. The amino acid compositions of the proteins in the sensitive and tolerant groups were also analyzed. Amino acid counts and percentages for each of the canonical amino acids were extracted from UniProt for 165 sensitive proteins and 169 tolerant proteins in the 5 mg/mL dataset as well as 259 sensitive proteins and 261 tolerant proteins in the 12.5 mg/mL dataset. Multiple amino acids were shown to have higher or lower percentages on average in the sensitive proteins compared to the tolerant proteins (Fig. 6C), with the most significant differences being found for histidine, cysteine, and lysine, as determined by P-values <0.001 (Fig. 6D).

Because the ratio of lysine content to arginine content has been reported to impact the solubilities of proteins, with lower ratios indicating poorer solubility,⁴⁵ this characteristic was also investigated in the sensitive and tolerant groups. These ratios were shown to be significantly lower in sensitive proteins (P-value = 0.009 and <0.001 for 5 and 12.5 mg/mL datasets, respectively), indicating that protein solubility may contribute to a protein's susceptibility to Cu-induced precipitation (Supplementary Tables S5 and S6).

In previous work we used the pulse proteolysis approach to study the effect of pharmacologically delivered Cu on the chemical denaturant–induced unfolding properties of proteins across the E. coli proteome.¹⁶ This earlier work identified 31 proteins with stability changes specifically driven by Cu in the chemical denaturant–induced equilibrium unfolding experiments. A significant fraction of these 31 proteins were also found in the tolerant and sensitive groups identified in MiPP (33% and 61% in the 5 and 12.5 mg/mL data, respectively). However, these previously identified protein targets of Cu were nearly equally distributed across the tolerant and sensitive groups. This observation suggests that the existence of specific, tight binding interactions with Cu, which were probed in the pulse proteolysis approach, is not a discriminating feature of relative MiPP sensitivity. However, it is important to note that there are multiple differences between the two experiments, including the previous work probing in vivo treatment of E. coli cells, compared to in vitro MiPP treatment, as well as being at a significantly lower Cu concentration than MiPP treated samples.

Discussion

How does exposure to Cu lead to deleterious biological outcomes? In this study, we address this fundamental question by identifying how proteins across the proteome respond to direct exposure to Cu. The experimental system here is not physiological in the sense that cleared cell lysates were exposed to unnaturally high metal concentrations. These were not conditions in which cells regulate their response to metal by inducing protective measures, rather these experiments were used to characterize the relative susceptibility of proteins across an E. coli proteome to precipitation induced by metal, and from this categorization, to gain insight into mechanisms by which Cu provokes protein misfolding. Indeed, recent work by Zuily et al. found compelling evidence to support a role for Cu-induced protein aggregation as a major contributor to Cu toxicity in E. coli.¹⁴ While in vivo protein aggregation has molecular chaperones and other cellular processes helping cells deal with aggregated proteins, the in vitro experiments used in this paper provide a reductionist approach to better understand specific susceptibilities of proteins to Cu-induced aggregation and how this biophysical propensity may contribute to cellular malfunction.

Our results establish that Cu far surpasses other metals in its capacity to precipitate proteins from cell lysates, that it is a reversible phenomenon for which a finite number of binding sites can be saturated, and that the bulk of the Cu bound in the aggregated protein mixture is in a reduced Cu¹⁺ form, regardless of the aerobic conditions and the fact that exposure started with a Cu²⁺ source, but consistent with concomitant oxidation of GSH. By using a quantitative bottom–up proteomics strategy, we were able to assign a Cu precipitation midpoint, or C_m value, for every protein identified by mass spectrometry. This process has enabled, for the first time, the bulk of the proteins in the E. coli proteome to be rank ordered based on their relative sensitivity to precipitation by Cu. This sensitivity ranking is independent of the relative concentration of protein in the lysate. Rather, the ranking relates to inherent biophysical attributes of the proteins that make them more or less tolerant to misfolding in the presence of Cu.

A comparison of the proteins identified in our data with those found in the recent work by Zuily et al. revealed an overall 75% overlap of proteins identified by Cu precipitation across both studies.¹⁴ Interestingly, however, the distribution of this overlap is uneven, with a 90% overlap found for proteins categorized as sensitive in our work, compared to only 57% overlap for the tolerant proteins. This overlap analysis suggests that the single condition of protein lysate and metal concentration reported by Zuily et al. was sufficient to precipitate most of the sensitive proteins but was not high enough in metal concentration to saturate the proteome and fully precipitate the most tolerant proteins. By covering the full range of Cu susceptibility and ranking the proteins by Cu MiPP C_m values, the data collected here provide a robust starting point for teasing out biophysical and mechanistic attributes that influence protein susceptibility, or resilience, to misfolding by Cu.

The reversibility of the Cu-induced precipitation process is mechanistically revealing. Results from the chelation-mediated protein solubility rescue experiments suggest that protein precipitation caused by Cu involves a mechanism more similar to that of unfolding reactions induced by chemical denaturants, which are generally reversible, than those induced by heat, which are not generally reversible. Urea, a commonly used denaturant, is proposed to denature proteins either by indirect alterations of the solvent environment around a protein, or through direct interactions that stabilize the unfolded state of the protein, thus shifting the conformational equilibrium to favor the unfolded state.⁴⁶ Cu could be acting by a similar mechanism. By binding to adventitious sites on proteins, Cu could stabilize an alternate conformation or unfolded region of the protein. Further support of a local unfolding mechanism being involved in Cu MiPP comes from the lack of correlation of the MiPP C_m values with midpoint values that report on global unfolding protein stability (SPROX) and protein precipitation (CPP and TPP). The absence of a correlation suggests that MiPP is not occurring through interactions with globally or even sub globally unfolded states. Rather, these results are consistent with MiPP occurring by a mechanism that specifically targets more local unfolding events that may ultimately trigger further unfolding and eventually protein aggregation and precipitation. It is also possible that Cu-induced protein precipitation could be occurring through a mechanism independent of protein unfolding mechanisms, global or partial. One possible mechanism independent of protein unfolding is through Cu-bridged self-assembly or polymerization of proteins. By bridging proteins Cu could be causing protein polymerization and aggregation that leads to eventual precipitation.

The comparison of our sensitive and tolerant categories revealed that unstructured regions were more highly represented in the tolerant proteins, whereas α-helical structures were more highly represented in sensitive proteins. Given that α-helical structures reduce the solvent accessibility of both hydrophilic and hydrophobic residues,⁴⁷ our analysis suggests that adventitious binding of Cu to amino acid residues in α-helical regions could stabilize non-native, non-helical conformations that expose previously buried hydrophobic residues, thereby initiating aggregation. Adventitious binding of Cu to unstructured regions, on the other hand, may cause less disturbance to solvent-inaccessible areas. Therefore, the ability of a protein to retain secondary structure when interacting with adventitious Cu and thereby prevent exposing solvent-inaccessible residues is likely a key diagnostic of how resilient the protein is to Cu-induced precipitation.

The earlier mechanistic hypothesis about MiPP is also supported by the results of our previous biophysical studies on two proteins identified in our earlier semi-tryptic peptide enrichment strategy for proteolysis procedures with pulse proteolysis experiments in which E. coli cells were exposed to Cu pharmacologically delivered by the ionophore pyrithione.¹⁶ In that work, the increased thermodynamic stability of IDH and GAPDH was found to be driven by cellular Cu exposure, which correlated with Cu-dependent inhibition of their enzymatic function in cellulo and in vitro. Interestingly, their secondary structures were affected differently upon direct exposure to Cu, with IDH adopting a non-native structure in the presence of Cu, whereas GAPDH remained well-folded. The Cu MiPP C_m values determined in the current work reveal that these two proteins have very different sensitivities to Cu precipitation, with a C_m of 2.77 mM for IDH indicating it is much more sensitive to Cu-induced misfolding compared to the very tolerant GAPDH with C_m of 5.40 mM (C_m values taken from 12.5 mg/mL data).

The results obtained in our amino acid analyses revealed that His and Cys residues were more abundant in sensitive proteins (Fig. 6D). This finding is consistent with that reported in the Zuily et al. study.¹⁴ As described earlier, the experimental condition used in the Zuily et al. work appears to be one in which MiPP sensitive proteins are preferentially precipitated. Therefore, the similar conclusions regarding His and Cys in the Zuily et al. work and ours is not surprising. These amino acid side chains are known to be good ligands for both Cu¹⁺ and Cu²⁺. While the ligand-binding affinity of Cu²⁺ is high on the Irving–Williams series, that comparison is only relevant across metals of the same oxidation state. Given the high percentage of Cu¹⁺ found in our aggregated samples, combined with findings from Zuily et al. that also found Cu¹⁺ to be the key driver for intracellular protein aggregation and cytotoxicity, we surmise that unique aspects of Cu¹⁺ coordination chemistry are likely factors in the unrivaled sensitivity of proteins to aggregation by Cu. In addition to its preference for binding “soft” ligands like Cys and His, Cu¹⁺ also accommodates low coordination numbers. This unique feature of Cu¹⁺ compared to Cu²⁺ and other d-block biometals implies that only two or three protein ligands may be sufficient to satisfy its coordination sphere for effective binding. If that binding event shifts the conformation of local secondary structure to expose hydrophobic residues, that could initiate aggregation and precipitation.

Conclusion

Whatever the mechanism by which Cu induces protein precipitation, this work supports a growing recognition that protein misfolding may be a primary driver leading to Cu-induced cytotoxicity. Protein aggregation regularly occurs even within healthy cells, which have many mechanisms for dealing with aggregated proteins, including chaperones, proteolytic degradation, heat shock proteins and aggresomes.⁴⁸ So why are cells so sensitive to the threat of Cu-induced protein aggregation that they limit labile Cu to such low levels? As suggested in a perspective piece by O'Hern and Djoko, it could be due to loss of function of Cu-aggregated proteins.⁴⁹ Alternatively, it could be that the amount of aggregated proteins overwhelm the cells’ mechanisms for handling them.⁴⁹

Supplementary material

The following files are available free of charge.

The following supplementary figures, tables, and text are included as a single PDF file: Supplementary Fig. S1, The impact of counterions on protein precipitation; Supplementary Fig. S2, EPR spectra of supernatant and resuspended pellet with and without hydrogen peroxide for all three biological replicates; Supplementary Fig. S3, Linear regressions for comparing SPROX, TPP, and CPP with MiPP proteins at 12.5 mg/mL; Supplementary Fig. S4, Bar graph showing the effect of hydrogen peroxide on protein precipitation; Supplementary Fig. S5, Plasmid map of E. coli used in this study; Supplementary Text, Python Script for extracting amino acid and secondary structure percentages from UniProt.

The following Supplementary tables are included as individual excel files (XLXS): Supplementary Table S1, Complete list of assayed MiPP proteins at the 5 and 12.5 mg/mL protein concentration, including their C_m values, and categorizations; Supplementary Table S2, Complete list of assayed CPP proteins, including their CPP midpoints; Supplementary Table S3, Complete list of assayed SPROX proteins, including their C_1/2 values; Supplementary Table S4, Complete list of peptides used for methionine oxidation analysis and the resulting analysis; Supplementary Tables S5 and S6, Complete list of all analyzed amino acids and secondary structures with Welch's ANOVA P-values for the MiPP proteins at 5 mg/mL (Supplementary Table S5) and at 12.5 mg/mL (Supplementary Table S6).

Funding

This project was supported with funds from the National Institutes of Health (1R01-GM145035 to K.J.F. and M.C.F.).

Disclosures

None declared.

Author contributions

A.T.R.R. wrote the manuscript, performed data analysis, collected ICP–MS and EPR data, performed resolubilization experiments, and helped with the Ellman's reagent assay for measuring reduced sulfhydryls. G.R.S. wrote the manuscript, performed data analysis, and collected proteomics data. J.M.Z-B. collected the large-scale protein precipitation data and helped conceptualize the project. N.W. collected the proteomics data and helped conceptualize the project. A.D. wrote the scripts to collect the amino acid percentage and secondary structure information. I.G.W. helped with and trouble shot the Ellman's reagent assay for measuring reduced sulfhydryls. K.J.F. and M.C.F. conceptualized and designed the research, analyzed the data, and reviewed and edited the manuscript with input from all authors. All authors edited the manuscript.

Data availability

The raw MS data and search outputs generated in this work can be found at the ProteomeXchange Consortium via the PRIDE partner repository⁵⁰ with the dataset identifiers: PXD035711 for MiPP, PXD035834 for SPROX, and PXD036219 for CPP.

Other data underlying this article are available in the article and in its online supplementary material.