Analysis of Brain Protein Stability Changes in a Mouse Model of Alzheimer’s Disease

Abstract

The Stability of Proteins from Rates of Oxidation (SPROX), Thermal Proteome Profiling (TPP), and Limited Proteolysis (LiP) techniques were used to profile the stability of ~2500 proteins in hippocampus tissue cell lysates from 2- and 8-month old wild-type (C57BL/6J; n=7) and transgenic (5XFAD; n=7) mice with five Alzheimer’s Disease (AD)-linked mutations. Approximately 200 - 500 proteins hits with AD-related stability changes were detected by each technique at each age-point. The hit overlap from technique to technique was low, and all the techniques generated protein hits that were more numerous and largely different than those identified in protein expression level analyses, which were also performed here. The hit proteins identified by each technique were enriched in a number of the same pathways and biological processes, many with known connections to AD. The protein stability hits included 25 high-value conformation biomarkers with AD-related stability changes detected using at least 2 techniques at both age-points. Also discovered were subunit- and age-specific AD-related stability changes in the proteasome, which had reduced function at both age-points. The different folding stability profiles of the proteasome at the two age-points are consistent with a different mechanism for proteasome dysfunction at the early and late stages of AD.

Graphical Abstract

Introduction

Alzheimer's disease is a progressive and irreversible brain disorder characterized by a decline in cognitive function and memory. It is the most common cause of dementia, affecting millions of people worldwide. The hallmark pathological features of Alzheimer's include the accumulation of beta-amyloid plaques and tau tangles in the brain.¹ Beta-amyloid plaques are extracellular deposits of a sticky protein fragment, while tau tangles form inside neurons, disrupting their structure and function. These abnormal protein aggregates interfere with communication between brain cells, leading to cell damage and death.^2,3

To-date, systems-level analyses of AD have largely relied on protein expression level analyses.^1,4-14 Such expression level studies have identified AD biomarkers in cerebral spinal fluid (CSF) (e.g., amyloid beta peptides,^15,16 tau,^17-19 differentially phosphorylated tau,²⁰ alpha- and beta-synuclein,²¹ SNAP-25,²² SYT-1,²³ and LAMP2,²⁴ neurogranin,²⁵ NPTX2,^26,27 VGF,²⁶ Trem2,^28,29 and apolipoprotein E^30,31). Some of these CSF biomarkers (e.g., amyloid beta,³² tau,⁴ apolipoprotein E and beta synuclein³³) have also been found to be useful at detecting AD in blood-based assays. Unfortunately, the routine use of these differentially expressed AD biomarkers in the clinic has been complicated due to standardization issues associated with lab-to-lab variabilities and inconsistencies across different analytical methods.^34,35 A few of these differentially expressed biomarkers have also been exploited as drug targets in AD.^5,36,37 In particular, drugs targeting two of the most well-established proteins in AD pathogenesis to-date (AB³⁶ and tau³⁷) have been developed. However, clinical trials of such drug therapies have generally failed.³⁸ These clinical trial results, together with a growing number of results from genetic, proteomic cellular, and biochemical studies of AD pathogenesis, suggest that AD pathogenesis is more complex than once thought.³⁹ Therefore, new approaches are needed to study AD pathogenesis at the systems-level.

Recently, a new toolbox of mass spectrometry-based proteomic methods has enabled large-scale measurements of protein folding stabilities in unpurified biological samples such as cell lysates.^40-50 In this toolbox are denaturant-based methods including Stability of Proteins from Rates of Oxidation (SPROX)⁴⁰, Pulse Proteolysis (PP)^41,42, Thermal Proteome Profiling (TPP)⁴³, and chemical denaturation and protein precipitation (CPP)⁴⁵, as well as native state methods including Limited Proteolysis (LiP), Fast Photochemical Oxidation of Proteins (FPOP) and Covalent Protein Painting.^44,51 Although the above methods all rely on quantitative bottom-up proteomics-based approaches to probe a protein’s folding stability, they use different experimental workflows and mass spectrometry-based readouts. SPROX relies on the detection of methionine-containing peptides to probe the chemical denaturant-induced protein folding/unfolding equilibrium at the level of individual protein folding domains. In TPP all peptides from a specific protein are used to probe thermal stability at the protein level (i.e., it is a measurement of global stability). In contrast, LiP, FPOP, and the Covalent Protein Painting approach probe protein conformational changes in the absence of denaturant and report on more local conformation changes in proteins via analysis of peptides that map to the modification site.

In addition to their widespread use in the analysis of protein ligand binding interactions,^41,42,52,53 the above methods for profiling protein folding stability on the proteomic scale have been used in an increasing number of applications focused on understanding disease biology.^47,48,54-56 These applications in disease biology have demonstrated that protein folding stability profiling methods can produce novel protein biomarkers and help elaborate the molecular basis of different disease phenotypes.^48,49,54-57 Here, the SPROX, LiP, and TPP approaches are used to profile the stability of proteins in hippocampus tissue cell lysates derived from a wild-type mouse strain (C57BL/6J) and a transgenic 5XFAD mouse strain (Tg6799) that contains five AD-linked mutations. Protein folding stability profiles were generated at 2 age-points corresponding to early and advanced stages of the disease phenotype in the AD mice. By identifying proteins with changes in their folding stability during AD progression, we aimed to better understand the direct and indirect effects of β-amyloid (Aβ) aggregation across the proteome. The differentially stabilized proteins identified in this work help elucidate the molecular basis of AD progression. They also have the potential to be novel biomarkers and therapeutic targets of AD that are more biologically significant and clinically useful than those currently generated in conventional systems-level analyses of gene and protein expression levels.

Experimental

Materials

S-methylmethanethiosulfonate (MMTS), guanidine hydrochloride (GdmCl), trifluoroacetic acid (TFA), tris(hydroxymethyl)aminomethane hydrochloride (Tris-HCl), triethylammonium bicarbonate buffer (TEAB, 1M, pH 8.5), hydrogen peroxide (H₂O₂) (30% w/w), acetic acid, and 2-mercaptoethanol were from Sigma-Aldrich (St. Louis, MO). Acetonitrile (ACN, LC-MS grade), TMT 16-plex isobaric labeling reagent set, and porcine pancreas trypsin (proteomics grade) were from ThermoScientific (Waltham, MA). Methanol (gradient grade OmniSolv^®) and tris(2-carboxyethyl) phosphine (TCEP) were from Santa Cruz Biotechnology (Dallas, TX). Phosphate buffered saline (PBS, pH 7.4) was from Gibco (Gaithersburg, MD). MacroSpin columns (silica C18) and pi^{3 ™} methionine reagent kits were from Nest Group (Southborough, MA). The 10 kDa Ultra Centrifugal Filter Units were from Millipore Amicon (Cork, Ireland).

The 2-month-old (n=7) and 8-month-old (n=7) female wild type (strain: C57BL/6J) and 5XFAD AD mice (strain: B6.Cg-Tg(APPSwFlLon,PSEN1*M146L*L286V)6799Vas/Mmjax) used in this study were purchased from The Jackson Laboratory and handled in accordance with the handling procedures approved by Duke University Institutional Animal Care and Use Committee approval for Protocol A083-22-04.

Mouse Euthanasia and Tissue Lysis

The mice used in this study were euthanized by cervical dislocation at Duke University. After mouse brain samples were harvested, hippocampus samples were separated from the brains on ice, and placed into centrifuge tubes with 800 μL phosphate buffered saline. The samples were processed by tip-sonication (FB120 Sonic Dismembrator, Fisher Scientific) for three cycles on ice of 15 s each, under 40% amplification. The cell lysates were centrifuged at 14,000 rcf for 20 minutes at 4 °C, and the supernatant recovered. A Bradford assay was used to determine the protein concentration in the supernatant, and the final protein concentration was adjusted to 2 mg/ml for the hippocampus cell lysates.

One-pot SPROX Analysis

The one-pot SPROX experiment was performed as previously described.⁵² Briefly, 10 μg of total protein from each mouse brain sample was aliquoted into a series of 12 PBS buffers (pH 7.4) containing increasing GdmCl concentrations. The final GdmCl concentrations in the buffers were equally spaced at 0.1 M intervals between 0.8 and 1.9 M. The protein sample in each GdmCl-containing buffer was equilibrated at room temperature for 1 h, before the methionine oxidation reaction was initiated by adding 30% (v/v) H₂O₂ such that the final concentration of H₂O₂ was 1 M. The oxidation reaction in each buffer was allowed to proceed for 3 min before quenching with 125 μL of 0.5M TCEP. Equal volumes of each GdmCl-containing sample from the same mouse were combined to generate 14 protein samples (n=7 WT, n=7 5XFAD) at each time point. Two additional samples were also generated at each time point. These additional samples, which ultimately served as protein expression level controls, were generated by combining equal amounts of the 7 WT or 7 5XFAD lysate samples. The resulting 16 samples were subjected to a quantitative bottom-up proteomics analysis using an isobaric TMT 16-plex and the iFASP protocol, which has been previously described.⁵⁸ Finally, the methionine-containing peptides in the sample were enriched using the Pi^{3 ™} methionine reagent kit according to manufacturer’s protocol. The enriched sample was desalted using a Macrospin (C18) column and the solvent evaporated.

One-pot TPP Analysis

The one-pot TPP experiment was performed similarly to that previously described.^52,59,60 Briefly, each mouse brain lysate sample was distributed into a series of 12 tubes. Each tube, which contained 15 μL of lysate, was heated for 3 min at one of 12 temperatures equally spaced at 2 °C intervals between 40 and 62 °C, equilibrated for 3 min at room temperature, and cooled on ice. The lysate samples in each series of 12 tubes were combined. The resulting 14 samples (n=7 WT and n=7 5XFAD at each of the 2 age-points) were centrifuged at 48,000 rcf for 20 min at 4 °C. As described above in the one-pot SPROX analysis, two additional protein expression level control samples were also generated. The resulting 16 samples were subjected to a quantitative bottom-up proteomics analysis using an isobaric TMT 16-plex and the iFASP protocol, which has been described previously.⁵⁸ Lastly, the combined sample was desalted using a Macrospin (C18) column and the solvent evaporated.

STEPP-LiP

The mouse brain cell lysates were subjected to a limited proteolysis analysis using the semi-tryptic peptide enrichment strategy for proteolysis procedures (STEPP)-LiP protocol that has been previously described.⁶¹ Briefly, 80 μg of total protein from each mouse hippocampus tissue lysate (n=7 WT, n=7 5XFAD, 2-month or 8-month) was treated with 1 μg of Proteinase K for 5 min at room temperature before the proteolytic digestion reaction was quenched with 5 mM phenylmethylsulfonyl fluoride (PMSF). The Proteinase K digested protein sample from the 7 WT and 7 5XFAD mice were labelled with 14 isobaric mass tags from a TMT 16-plex before they were combined and digested with trypsin. The resulting semi-tryptic peptides (peptides resulting from proteinase K cleavage at one terminus and trypsin cleavage at the other terminus) from the resulting peptide mixture were enriched using an enrichment strategy we have previously described previously.⁶¹ Ultimately, the solvent was evaporated.

Proteasome Activity Assay

Aliquots of the cell lysates used in the SPROX, TPP, and LiP analyses were also submitted to a proteasome activity assay using the Proteasome Activity Fluorometric Assay Kit (abcam, ab107921). Mouse samples (n=3 to 7) from four different stages (2 or 8 months, AD or WT) were examined. In each case 50 μL of lysate containing 100 μg of total protein was used to evaluate the chymotrypsin-like activity of the proteosome according to the manufacturer’s protocol which involved the use of a fluorogenic peptide substrate (Suc-LLVY-AMC). The fluorescence generated upon substrate cleavage was quantified every 5 minutes for 60 minutes using a PerkinElmer Victor3 V multilabel plate reader with excitation and emission set at 340 and 460 nm, respectively.

Quantitative LC-MS/MS Analysis

The LC-MS/MS analyses were performed on a Thermo Orbitrap Exploris480 mass spectrometer equipped with a Thermo Easy nanoLC 1200 system. The trapping column was a Thermo Acclaim PepMap 100 75 um x 2 cm, and the analytical separation was a nanoViper 2Pk C18, 3 μm, 100 Å 75 μm x 25 cm column. The dried peptide samples obtained at the end of SPROX, TPP, and LiP protocols above were reconstituted in H₂O containing 1% TFA and 2.5% acetonitrile to a final concentration of 1 mg/mL. Samples containing 1 μg of total peptide were injected on column and analyzed in triplicate. Peptides were eluted using a 90 min linear gradient of 3-30% ACN and 0.1% formic acid at a flow rate of 400 nL/min. The LC-MS/MS data were collected using a data-dependent acquisition (top 20) with a scan range of 375-1500 m/z. Resolution of the instrument was set to 120,000 and 50,000 for MS1 and MS2, respectively. The collision energy was set to 38%. The isolation window was 0.7 and the dynamic exclusion duration was 60 s.

Proteomic Data Search

The raw data generated in the LC-MS/MS experiments were searched against the proteins in the SwissProt Mus musculus database version 2019-03-20 using Proteome Discoverer 2.3 (Thermo). In the search, MMTS modification on cysteine and TMT 16-plex labeling on lysine side chains and peptide N-termini were set as fixed modifications. Variable modifications include oxidation on methionine, deamidation on asparagine, glutamine, and arginine, and acetylation on peptide N-termini. The number of missed cleavages allowed was set to two. The parameters included a 10 ppm mass tolerance window for precursor masses and 0.6 Da for fragment mass tolerance. Intensity was used for reporter abundance and the normalization as well as the scaling mode were set as none. The rest of the software settings were set to default. Spectra with FDR confidence of “high” and “medium” (i.e., FDR <1%) and without a single quantification channel of 0 were used for subsequent analysis. The raw MS data generated in this work has been uploaded and made publicly available through the ProteomeXchange Consortium via the PRIDE partner repository with the data set identifiers PXD051155.

Data Analysis

Protein Expression Level Analysis.

The protein expression level analysis utilized the non-methionine containing peptides from the unenriched samples in SPROX. In this analysis each intensity was normalized by the sum of all the signal intensities in each mouse sample. Following this normalization, all the intensities of the different non-methionine-containing peptides that came from the same protein were summed, and a series of ratios were generated by dividing the resulting protein intensity from one 5XFAD mouse sample by the intensity of that protein in a corresponding WT mouse sample. This generated seven ratios (referred to hereafter as fold-change values). These fold-change values were log₂-base transformed, averaged, and used in a two-tailed Student’s t test to identify differentially expressed protein hits with mean log₂-fold change values that were significantly different than 0. Protein hits were selected as those with a p-value <0.01 and z-score <−1 or >1.

Identification of Differentially Stabilized Proteins.

Each biological replicate was first normalized by the sum of the signal intensities in each mouse sample. The SPROX and LiP data sets were analyzed at the peptide level using only the wild-type methionine-containing peptides and semi-tryptic peptides, respectively. The TPP data was analyzed at the protein level. In SPROX and TPP, two channels in the TMT16plex were used to directly control for protein expression level differences between the control and 5XFAD mice. These channels contained super samples of the wild-type and AD mice. The signal intensities of different peptides (SPROX) and proteins (TPP) were then divided by the corresponding normalization factors generated from the ratio of the signals in the two channels corresponding to the super-samples. In the LiP analysis, intensities from the semi-tryptic peptides were normalized using the protein expression level data generated using the super samples in the TPP analysis. While this normalization effectively controlled for protein expression level differences in the LiP experiment, it did not take into account differences in LiP peptide levels due to post-translational mutations (PTMs). Thus, the hit peptides in LiP could result from differential PTMs, differential stability, or some combination thereof.

Ultimately, seven ratios per technique per age point were generated using the data from the seven wild-type and seven 5XFAD mice. These ratios were log₂-base transformed, averaged, and used in a two-tailed Student’s t test to identify differentially stabilized protein hits with mean log₂-fold change values that were significantly different than 0. Protein hits were selected as those with a p-value <0.01 and z-score <−1 or >1.

Bioinformatics Analyses.

Gene set enrichment analyses (GSEA) and over representation analyses (ORA) were performed using the gseapy Python package to understand the functional categories significantly associated with protein stability differences between the normal and AD mice at the 2- and 8-month time points. The protein identifiers generated in the proteomics analyses were converted to corresponding gene symbols using the UniProt database. In the ORA, genes from proteins with a p-value less than 0.01, derived from t-tests described above, were selected as the input gene lists for each technique (i.e., the SPROX, TPP, LiP and expression level analyses). Separate ORA analyses were performed for each technique using all the genes assayed in that technique a given as the background dataset. The analysis harnessed the Kyoto Encyclopedia of Genes and Genomes (KEGG)_2021_Human gene set database to ascertain the over-represented pathways in the input gene set. Adjusted p-values were calculated to correct for multiple testing.

GSEA were also performed using the data generated using each technique. Proteins assayed in a given technique were converted into their corresponding gene identifiers and ranked by their -log₁₀(p-values), in descending order (i.e., from the most statistically significant to the least significant p-values). This ranked gene list was then analyzed against the AD gene set within the KEGG disease database. A running-sum statistic increments when encountering a gene within the AD set and decrements otherwise. Separate GSEA were preformed using the data from each technique.

A series of ORA were also performed using i) the protein hits appearing in each technique and at each time point and ii) the 25 protein hits appearing in at least 2 stability techniques at both time points. The background data set used in the ORA of the 25 protein hits appearing in at least 2 stability techniques included the total number of proteins assayed in all 3 stability techniques and at both time points (i.e., proteins assayed in 1, 2, or all 3 stability techniques and at either or both of the age-points).

Results

Experimental Design

Shown in Figure 1 are the experimental workflows used in the SPROX, TPP, and LiP experiments performed in this work. The analyses were all conducted on the proteins in tissue cell lysates generated from the hippocampus region of the mouse brain, which is known to be one of the earliest affected regions in neurodegenerative conditions.⁶² The 5XFAD mouse model,⁶³ used in this study is one of the most widely used amyloid-based mouse models of AD. The 5XFAD mouse model is a transgenic mouse that expresses mutant forms of both human amyloid beta precursor protein (APP) and presenilin 1(PS1). 5XFAD mice rapidly produce amyloid pathology faster than any other AD mouse model. For example, 5XFAD mice exhibit normal behavior up to the age of 2 months, but by 4–5 months they exhibit cognitive deficits in spatial working memory as well as develop associative learning impairments in fear conditioning.

Figure 1.

Schematic representation of the one-pot SPROX, STEPP- LiP, and one-pot TPP workflows employed in this work to characterize the mouse hippocampus samples.

Protein samples were generated from the 5XFAD mice at 2 age-points (2- and 8-months), which align with the initiation and full development of AD characteristics, as per previous findings.⁶³ Female mice were chosen for this initial study because female 5XFAD mice display a more pronounced AD phenotype than male 5XFAD mice.⁶⁴ We reasoned that this more pronounced phenotype would make protein folding stability changes between wild-type and AD mice easier to detect in the SPROX, TPP, and LiP experiments. The same brain protein samples from the 7 wild-type control and 7 5XFAD mice were analyzed by each technique. The SPROX and TPP experiments utilized a one-pot strategy that enabled the analyses to be highly multiplexed. One 16-plex of isobaric mass tags per technique per age point was used to readout the proteomic results from the 7 wild-type control and 7 AD mice. Each 16-plex also included 2 protein expression level control samples, one of which was the combined cell lysates from all 7 wild-type controls and the other of which was the combined cell lysates from all 7 5XFAD mice. These protein expression level control samples were not subjected to the protein modification reactions in SPROX and TPP, but they did go through the bottom-up proteomics sample preparation (see Figure 1).

The TMT reporter ion intensities obtained in the LC -MS/MS analyses of the proteomic samples generated for each technique were used to identify proteins with disease-related stability changes between the control and 5XFAD mice at each time point. This was accomplished by generating fold-change values (see Figure 1) associated with the peptides (SPROX and LiP) or proteins (TPP) in the control and 5XFAD mice at each age point. Ultimately, peptides (SPROX and LiP) or proteins (TPP) with fold change values significantly different from 1 (i.e., p-value<0.01 and z score >1 or <−1) were selected as hits. These hit selection criteria have been shown to yield a <1% false positive rate of hit discovery in previously published control experiments using SPROX and TPP to analyze the proteins in an MCF-7 cell lysate.⁵⁷

Proteomic Coverage and Hit Identification

The numbers of proteins and peptides assayed in each stability technique and in the protein expression level analysis are summarized in Table 1. Summarized in Table S1-S4 in the Supplemental Material are the peptides and proteins assayed in each technique. Volcano plots summarizing the results from each technique are shown in Figure S1. The number of proteins assayed by each stability technique was similar at each age point. However, the protein hit rates for the stability-based techniques, which ranged from 16-41%, were generally higher at the 2-month age point. The same was true for the protein expression level results at the two age points.

Table 1:

Summary of peptide and protein coverage in the SPROX, TPP, LiP, and protein expression level experiments.

Age	Techniques	Assayed Protein (Peptide)	Hit Protein (Peptide)1	Hit Rate Protein (Peptide)
2-month	SPROX	2109(5449)	515(731)	24% (13%)
	LiP	571(2063)	233(537)	41% (26%)
	TPP2	1620 (NA)	332 (NA)	20% (NA)
	Expression Level2	971(NA)	162 (NA)	17% (NA)
8-month	SPROX	2025(4427)	262(293)	13% (7%)
	LiP	567(2094)	227(454)	40% (22%)
	TPP2	1367(NA)	224 (NA)	16% (NA)
	Expression Level2	1322 (NA)	150 (NA)	11% (NA)

¹Selection criteria: p value < 0.01, z>1 or z<−1

²Only proteins with at least two assayed peptides were included in the analyses.

A total of ~2500 proteins were assayed across all three stability techniques and both time points with ~450 (or ~20%) of the proteins at each time point being successfully assayed by all three techniques (Figure 2). Approximately 200 - 500 proteins hits with AD-related stability changes were detected by each technique at each age-point. The 11-41% peptide (SPROX and LiP) and protein (TPP) hit rates for the stability-based techniques were all significantly higher than the 1% false discovery rate expected for each technique, based on the results of a control experiment, on which we previously reported.⁵⁷

Figure 2.

Venn diagrams showing the assayed protein and protein hit overlap in the one-pot SPROX, one-pot TPP, and STEPP-LiP experiments. Shown in (A) and (C) are the assayed protein overlap for the 2-month and 8-month timepoint, respectively. Shown in (B) and (D) are the protein hit overlaps for the 2-month and 8-month timepoint, respectively.

SPROX, LiP and TPP can all provide information about protein folding stability changes in proteins induced by disease phenotypes. However, not all types of protein conformational changes are amenable to detection by each technique. The proteins successfully assayed by each technique can also be different. For example, SPROX and LiP, which utilize peptide readouts, can be more sensitive to changes in the sub-global and/or local unfolding properties of proteins than TPP. This is because sub-global and/or local conformational changes in one domain of a protein may not be sufficient to alter the thermal denaturation and precipitation properties of the entire protein. On the other hand, because all the tryptic peptides detected from a specific protein in TPP report on the same global unfolding behavior, changes in that behavior can be easier to detect than in LiP, for example (i.e., detection of a specific peptide located at the specific site of the conformational change is not required). Nonetheless, some conformational changes, such as many ligand binding events, can alter the global, sub-global, and local unfolding properties of proteins and be detected by all three techniques. However, not all proteins are efficiently sampled by all three techniques. This is largely due to the different peptide fractions that are sampled in each technique (i.e., all tryptic peptides in TPP, methionine-containing tryptic peptides in SPROX, and semi-tryptic peptides in LiP).

The similarities and differences between the SPROX, TPP, and LiP techniques mean that both overlapping and unique proteins are assayed and protein hits are expected across the techniques. However, we have shown that requiring protein hits to appear in multiple techniques can significantly reduce the false discovery rate.⁵² Requiring the protein stability hits in this work to appear in at least two techniques yielded 138 and 100 hits at the 2- and 8-month age points, respectively. A subset of 25 protein hits appeared at both age point and in at least two techniques. A majority of the 25 protein hits in this subset have known associations with Alzheimer's disease pathology (Table 2). However, none of these 25 protein stability hits were among the 25 protein expression level hits that appeared at both age points (Table S5).

Table 2.

Summary of the proteins selected as hits in at least two techniques at both time points.

Accession Number	Protein Name	Relation to AD (Reference No.)
P20152	Vimentin	83
Q91XV3	Brain acid soluble protein 1	84
O89053	Coronin-1A
P04370-4	Isoform 4 of Myelin basic protein	85
Q6P8I4	PEST proteolytic signal-containing nuclear protein
P20357	microtubule-associated protein 2
P09411	phosphoglycerate kinase 1
P08249	Malate dehydrogenase, mitochondrial	86
Q02053	Ubiquitin-like modifier-activating enzyme 1
P26443	Glutamate dehydrogenase 1, mitochondrial	87
P19096	Fatty acid synthase	88
Q8C1A5	thimet oligopeptidase	89
Q9R0K7	Plasma membrane calcium-transporting ATPase 2	90
O08553	Dihydropyrimidinase-related protein 2	91
Q04447	Creatine kinase B-type	92
P08551	Neurofilament light polypeptide	93
P06151	L-lactate dehydrogenase A chain	94
P47857	ATP-dependent 6-phosphofructokinase, muscle type
Q62261-1	Spectrin beta chain, non-erythrocytic 1	95
P16546	Spectrin alpha chain, non-erythrocytic 1	95
P70663	SPARC-like protein 1	96
Q9DBG3	AP-2 complex subunit beta	97
Q9ERD7	tubulin beta-3 chain	98
Q01853	Transitional endoplasmic reticulum ATPase	99
P47809	Dual specificity mitogen-activated protein kinase kinase 4	100

In addition to the above 25 protein stability hits that appeared in at least 2 techniques at both age points, another 246 proteins were differentially stabilized at both age points in at least one technique. These included 102, 51, and 108 protein stability hits in the SPROX, TPP, and LiP experiments (Table S5). Only a small fraction (14 out of 246) of these protein hits overlapped with the 25 protein stability hits in Table 2. Thus, the three techniques generated largely unique hits in this study. Also, only a small fraction (7 out of 246) overlapped with the 25 protein expression level hits at both age points (Table S5).

Biological Variability

The standard deviations of the log₂-fold change values generated for the peptides (SPROX and LiP) or proteins (TPP and expression level) analyzed in the WT and 5XFAD mice cohorts were used to assess the biological variability of each technique at each timepoint. The frequency distributions of the standard deviations observed in the different techniques for all the assayed peptides/proteins are shown in Figure 3. The averages of the standard deviations in the SPROX and TPP experiments at the 2 age points ranged from 0.16-0.31. This is very close to the range of average standard deviations observed in our previously reported control SPROX and TPP experiments (i.e., 0.25-0.29).⁵⁷ This suggests that the biological variability of the SPROX and TPP measurements in this work was within the experimental error of these techniques. The same also appears to be the case for the LiP and expression level measurements in this work, which had similarly low average standard deviations (0.12-0.21).

Figure 3.

Frequency distributions of the standard deviations associated with the fold-change values generated in the SPROX, TPP, LiP, and expression level analyses on the 7 pairs of WT and AD mice at each age-point. biological replicates analyzed here. Shown in (A) are the results for the peptides analyzed in both timepoints in SPROX. Shown in (B) are the results for the proteins analyzed in both timepoints in TPP. Shown in (C) are the results for the semi-tryptic peptides analyzed in both timepoints in LiP. Shown in (D) are the results for the proteins analyzed in both timepoints in expression level. The blue and orange lines in each plot represent data from the 2-month and 8-month mice, respectively.

Bioinformatics Analysis

GSEA analyses performed on the SPROX, TPP, and expression level data at the 8-month age-point and revealed a significant (p<0.05) enrichment of AD-associated genes (Figure 4). However, the same was not true for the SPROX, TPP, and expression level data at the 2-month age point as there was an even dispersion of the AD-associated genes in the ranked gene list (Figure 4). The GSEA of the LiP data did not reveal an enrichment of AD-associated genes at either of the age-points in this study.

Figure 4.

Gene Set Enrichment Analyses (GSEA) of the assayed proteins in each technique at two time points. The line plot at the top segment illustrates the running Enrichment Scores (ES) for the AD gene set as it traverses the ranked list. The peak of this line plot denotes the ES for the entire KEGG AD gene set, capturing the maximum deviation from zero during this traversal. In the plot's middle segment, each blue line marks a gene acknowledged in the KEGG AD gene set, plotted in correspondence with its position in the ranked list. The lower segment of the plots delineates the -log₁₀(p-values) as the ranking metric. P-values in the plot illustrate the statistical significance of the enrichment. (A) and (E): results for proteins assayed in SPROX at 2-month and 8-month age, respectively. (B) and (F): results for proteins assayed in TPP at 2-month and 8-month age, respectively. (C) and (G): results for proteins assayed in LiP at 2-month and 8-month age, respectively. (D) and (H): results for proteins used in expression level analysis at 2-month and 8-month age, respectively.

ORA analyses were performed on the protein folding stability data generated using each technique at each age point (Figure S2). The enriched GO terms and pathways at the different age-points within each technique revealed a consistent presence of neurodegenerative diseases and associated pathways, including Alzheimer's (AD), Parkinson's (PD), Huntington disease, and the pathway of neurodegeneration, among others (Figure S2). In addition to these commonly identified enrichments, unique cellular processes and pathways can be identified in the data obtained at the different age-points and/or across the different techniques. Notably, these unique enrichments predominantly fall into two categories: metabolism (e.g., TCA cycle, fatty acid degradation) and bacterial infections (e.g., Shigellosis, Pathogenic Escherichia coli infection). These findings align with previous research where glucose metabolism was found to be altered in the AD brain^65,66 and proinflammatory bacteria, such as Escherichia/Shigella, were found to be elevated in AD patients.⁶⁷ The unique enrichments observed in the data from the different techniques also uncovered AD-related processes. For example, proteolytic degradation in TPP, fatty acid degradation in SPROX, and phagocytosis in LiP. These results reinforce the value of utilizing multiple stability-based techniques to characterize disease phenotypes.

Included in Figure S2 is a summary of the ORA results from the expression level analysis at the two different age point in this study (Figure S2). Many (i.e., 23 out of 32) of the pathways and biological processes enriched in the protein expression level hits were also found to be over-represented in the hits from at least one of the stability-based techniques. These over-represented pathways and biological processes include (among others) the TCA cycle, pyruvate metabolism, and alcoholism. A total of 9 of the 32 pathways and biological processes enriched in the protein expression level hits are consistently over-represented in all three stability-based techniques. Among these 9 pathways and biological processes are neurodegeneration, Alzheimer's disease, glycolysis, and propionate metabolism. Beyond the pathways and biological processes that are over-represented in both the protein expression level and protein stability hits, are 20 pathways and biological processes that are uniquely over-represented in the protein stability analyses. Intriguingly, the majority of these 20 pathways and biological processes have reported associations with AD (Table S6). These pathways and biological processes include (among others) bacterial invasion, endocytosis, and spliceosome.

Proteasome Activity

Many subunits of the proteasome were differentially stabilized in the TPP experiments performed on the WT and 5XFAD mice at both time points (Figure 5). Because of the close connection between protein folding stability and function, we hypothesized that the function of the proteasome was compromised in the 5XFAD mice. To test this hypothesis, we evaluated the chymotrypsin-like activity of the proteasome in the brain tissue cell lysates used in the above folding stability analyses. Proteasome function not only decreased with age, but at both age points the chymotrypsin-like activity of the proteasome was lower in the 5XFAD mice (Figure 6). One explanation of the altered activity levels of the proteasome in these samples could be that the proteosome was differentially expressed. However, our expression level data collected on the WT and 5XFAD mice at each age point did not reveal a significant change in the proteasome expression level across the WT and 5XFAD mice at either age-point. Therefore, the AD-related changes in function are not the result of altered proteosome expression levels, but rather appear to be due to proteosome dysfunction. The relative expression levels of the proteosome at the 2- and 8-month age-points was not measured as part of this work. Thus, it is possible that the age-depended change in proteosome activity observed in this work may be due to altered proteosome expression levels at the 2 age points.

Figure 5.

Stability changes associated with the 32 proteasome subunits detected in TPP at both time points. Data from subunits in the 20S core particle, the 19S regulatory particle triphosphatase, and the 19S regulatory particle non-ATPase are indicated in blue, green, and red (respectively). A negative z-score indicates a destabilization in AD and positive z-score indicates a stabilization in AD.

Figure 6.

Analysis of proteasome activity of WT (blue) and 5XFAD (red) mouse hippocampus lysates at 2- and 8-month time points. Plotted are the initial rates determined in a fluorometric assay using the AMC-tagged peptide substrate (Suc-LLVY-AMC) for proteasome. The error bars represent 1 standard deviation. *p<0.05, **p<0.01.

Discussion

Conformational Biomarkers of AD

The 25 protein stability hits summarized in Table 2 are high value protein biomarkers of AD. Nearly all 25 proteins have previous connections to AD (see references in Table 2). However, none of the 25 protein biomarkers would have been discovered using conventional protein expression level profiling techniques. The AD-related folding stability changes detected in these 25 proteins provide a unique means by which to exploit these proteins for therapeutic and diagnostic purposes. However, for such therapies and/or diagnostic tests to be successful, the folding stability changes detected in the mouse brain samples studied here would need to translate human samples. Use of the AD-related folding stability changes detected in this work would also require their detection in more clinically relevant biological specimens such as cerebrospinal fluid (CSF) fluid samples.

Notably, 3 of the 25 protein hits in Table 2 (including L-lactate dehydrogenase A chain, Brain acid soluble protein 1, and SPARC-like protein 1) were identified in a recent proteomic study of human CSF fluid.⁵⁴ Consistent with our mouse brain data, none of the above 3 proteins were differentially expressed the human CSF fluid from AD and control patients. This earlier proteomic study on human CSF fluid also included a LiP analysis to identify for conformational biomarkers of AD. One of the above 3 protein hits in our work, SPARC-like protein 1, was assayed in the earlier LiP analysis of human CSF fluid from AD and normal patients, but it did not display hit behavior. However, another one of the 13 conformational biomarkers detected in the previous LiP analysis of human CSF fluid form AD and normal patients, clusterin, did overlap with one of the protein hits identified in the LiP analyses on our mouse brain samples. Moreover, one of the clusterin hit peptides identified in our mouse brain study included clusterin amino acids 392-400, which map to the exact same region as the clusterin peptide that was assayed and also selected to have the same AD-related exposure in the LiP study using human central spine fluid (CSF) to study AD.⁵⁴ Indeed clusterin has previously been connected to AD. For example, earlier studies have shown that clusterin plays a role in potentiating both Aβ aggregation and clearance.^68-70

Despite the utilization of different samples (CSF vs. hippocampus brain tissue cell lysate) across different species (human vs. murine) the consistent detection of a protein associated with AD in both investigations suggests that at least some of the brain protein with stability changes detected in the mouse brain samples in this work have the potential to be detected in human CSF. Another one of the 25 protein hits in Table 2, the tubulin-beta chain, was also identified in a previous study using a covalent protein painting technique to characterize AD human brain lysates.⁵⁵ Similar to LiP, the covalent protein painting technique can be used to investigate more local protein conformational changes induced by diseases like AD. However, the proteomic readouts in LiP and covalent protein painting are different, therefore it is not surprising that the specific hit peptides were different in the two studies.

Protein Hit Validation and Analysis

The GSEA and ORA analyses help validate the use of protein folding stability measurements to characterize AD. For example, the ORA analyses performed using the protein stability data collected in this work show that the protein hits detected using each of the three stability-based techniques in this work were enriched in proteins known to be involved with pathways of neurodegeneration and AD. Moreover, these enrichments were generally more pronounced at the 8-month age-point where the disease phenotype was most pronounced. The GSEA analyses also revealed that the protein hits identified in the TPP and SPROX analyses at the 8-month age-point were also enriched in AD-related genes. The absence of such an AD-related gene enrichment in the GSEA of the LiP data is likely due to the relatively small number of assayed proteins and relatively large number of hits. The limited proteomic coverage in LiP may be related to the semi-tryptic peptide readout employed in this work. The relatively large number of hits is potentially related to the fact that our protein expression level normalization step in LiP does not account for differential PTMs; whereas the expression level normalization step using the super-samples in the SPROX and TPP analyses did.

The p-values in the GSEA for the SPROX, TPP, and Expression level analyses at 2-month were also larger than 0.05. This lack of enrichment could be attributed to the lack of true differences within the sample rather than sample size constraints or high biological variability, as the number of assayed proteins and hit proteins at 2-month is comparable to those at 8-month, which did show enrichment, and the biological variability is low and within experimental error, as suggested by the biological variability analysis (Figure 3).

The ORA analysis of the 25 high value protein hits identified in this work (see Table 2) also serves to substantiate the value of these protein hits that were detected using at least 2 techniques across both age-points. These hits were enriched in 2 pathways, 3 tissue expression, 1 subcellular localization, and 1 phenotype ontology (Figure 7 and Table S7). Interestingly, the enriched phenotype ontology term is abnormal brain function, with 14 out of 25 proteins belong to this category. Among these 14 protein hits is mdh2, which is also involved in glycolysis, one of the 2 enriched pathways. Mdh2 was only selected as a stability hit in the AD phenotype at the 8-month age point but under expressed and destabilized in the AD phenotype at the 2-month age point. In previous work, AD patients that died due to AD were found to exhibit an increased activity and protein expression of mdh2.^71,72 While Mdh2 did not pass the hit selection criteria in our expression level study at the 8 month age-point it was indeed overexpressed in AD, albeit with a z-score of 0.7.

Figure 7.

Results of the over-representation analysis of the GO terms enriched in 25 proteins in Table 2.

Of the 14 proteins related to abnormal brain function, 7 proteins are localized to axons, the 1 subcellular localization enriched in the 25 high value protein hits. Interestingly, all 7 of the axon protein hits were destabilized in the AD phenotype at the later age-point. These results are consistent the results of other studies that found axonal degeneration is a common event in AD progression.⁷³ Among these seven identified hits is microtubule-associated protein 2 (map2). MAP2 plays a crucial role in stabilizing microtubules and protecting from axon degeneration.^74,75 Our results indicate that map2 is both less stable and under-expressed in Alzheimer's disease (AD) at 8 months. These two characteristics of instability and reduced abundance may impede the synthesis of microtubules and hinder their cross-linking with other components of the cytoskeleton.

Proteosome Stability Changes in AD

The proteasome was found to be enriched in the ORA of the TPP hits at both age-points. Assayed in the TPP experiments at both age-points were 32 protein subunits from the proteosome (Figure 5). Remarkably, these 32 protein subunits created unique stability profiles at each age-point. The 14 subunits of the 20S core particle of the proteosome were all destabilized (i.e., z<0) in the AD mice at the 2-month age-point and all but one of these 14 subunits were stabilized (z>0) in the AD mice at the 8-month age-point. Moreover, 6 and 11 of these 14 subunits from the 20S core particle of the proteasome were selected as hits (i.e., p<0.01 and z<−1 or z>1) in our TPP analyses at both of the 2- and 8-month age-points, respectively. Only several of the 18 regulatory subunits detected from proteasome appeared as TPP hits in this work (Figure 5). However, it is noteworthy that nearly all of these regulatory subunits (i.e,. 16 out of 18) were destabilized (i.e., z<0) in the AD mice at the 8-month age-point; whereas these same regulatory subunits displayed a nearly equal mix of stabilizations/destabilizations at the 2-month age-point.

Methionine-containing peptide from 11 subunits in the 20S core particle were included in the SPROX analyses at both the 2- and 8-month time point. Remarkably, the z-scores recorded for these peptides were consistent with the TPP results above (i.e., the 20S core subunits were all destabilized at the 2-month time point and, all but two, of the subunits were stabilized at the 8-month time point) (Figure S3). However, the “effect size” (i.e., z-scores) were generally smaller than those observed in TPP, and only 2 subunits at each time point had a z-score large enough to pass the hit selection criteria. The LiP data set only included semi-tryptic peptides from 6 of the 32 proteasome subunits, too few to observe any trends (Figure S3).

The proteasome degrades proteins and prevents the accumulation of damaged and misfolded proteins. Dysfunction of the proteasome is known to correlate with AD.^76,77 The proteasome 20S structure consists of four stacked rings, forming a barrel-like structure.⁷⁸ This cylindrical core particle harbors the proteolytic activity responsible for degrading proteins marked for degradation. This degradation process involves opening of a gate to facilitate the entry of proteins into the channel for degradation. Both in vivo⁷⁹ and in vitro⁸⁰ studies have documented a decline in proteasome activity in the presence of Aβ aggregation, a characteristic feature of Alzheimer's disease (AD). The in vitro study utilized synthetic oligomeric Aβ species to investigate their impact on proteasome activity, and found that direct binding of Aβ oligomers to the 20S core particle decreased proteasome function through an allosteric mechanism that stabilized the closed gate (inactive) conformation of the 20S core particle.⁸⁰ The results of our proteosome activity assays on the mouse brain tissue lysates in this work also showed a significant (p<0.01) decrease in proteasome activity in the 5XFAD mice at the 8-month age point relative to WT mice at the same age point (Figure 6). Notably, our TPP and SPROX data indicate a stabilizing effects on nearly all the subunits in the 20S proteasome from the 5XFAD mice at 8 months. This observation aligns with the hypothesis that oligomers from Aβ overexpression in the 5XFAD mice bind to the proteasome and cause it (i.e., its constituent subunits) to precipitate at a higher temperature (i.e., stabilize all the subunits of the 20S proteosome). The TPP destabilizations we measured for the 18 regulatory subunits of the proteasome at 8 months suggest that these subunits, which together comprise the 19S caps of the proteasome, dissociate from the 20S core particles upon Aβ binding to the 20S core. These destabilizations, which were more subtle than the stability differences observed in the 20S proteasome, were not observed in the SPROX data. However, the similar TPP behavior (i.e., z<0) of nearly all the subunits in the 19S caps at the 8 month suggests that the protein-protein interactions among the subunits in the 19S caps are still intact in the 19S caps.

The results of our proteosome activity assays on the mouse brain tissue lysates in this work also showed a significant decrease in proteasome activity in the AD mice at the 2-month age-point relative the WT. Similar to the results at the 8-month age-point the observed activity difference is likely attributable to proteasome dysfunction rather than differential expression. However, based on the stability profiles we obtained for the detected subunits of the proteasome, the mechanism behind the dysfunction at the 2-month age point appears to be different from the mechanism hypothesized above at the 8-month age point. In contrast to results obtained at the 8-month age point, the 20S proteasome subunits were nearly all destabilized (rather than stabilized) in the AD mice at the 2-month age point. Remarkably, this was true for 13 of the 14 proteasome subunits detected from the 20S catalytic core particle, including 6 subunits that were selected as stability hits in TPP, 3 subunits with z-scores close to (i.e., Z>0.7), but not passing, the hit selection criteria, and another 4 with negative z-scores.

One explanation for the observed destabilization of the 20S catalytic core particle at the early age point may be related to the oxidative stress that AD mouse models (similar to that used here) are known to experience at young ages.⁸¹ During periods of oxidative stress it has been shown that the 26S proteosome is disassembled into the 19S caps and 20S core particle.⁸² The measured destabilizations of all the detected subunits of the 20S core particle in the TPP experiment are consistent with the 20S core particle precipitating a lower temperature when it is not bound to the 19S caps. However, the stability profile of the regulatory subunits in the 19S caps at the 2-month age point suggest the fate of the subunits in the dissociated 19s caps at the 2-month age point is different than that at the 8-month age point. Interesting, it has been noted that upon dissociation of the 26S proteasome after periods of oxidative stress, the 19S caps are sequestered by Hsp70.⁸² Such protein-protein interactions with Hsp70 could explain unique precipitation behavior of the subunits in the 19S caps at the early age point.

The protein folding stability and biological activity results obtained here on the proteasome not only point to involvement of the proteasome in AD at both early and late stages of the disease, but they also help substantiate the close connection between protein folding stability and function. At both age points a number of proteasome subunits were differentially stabilized in the AD mice and the proteasome activity was decreased. However, the different relative stabilities of specific proteasome subunits at the two age-points suggests a different molecular basis for proteasome dysfunction at the early and late stages of the AD.

Stability and Expression Level Profiling are Complimentary

Both protein folding stability and expression level analyses were used to characterize the mouse model of AD used in this work. There was very limited overlap between the specific proteins found to have AD-related expression level changes and the proteins found to have AD-related stability changes. Bioinformatics analyses indicated that some of the same pathways and biological processes were enriched in the hits generated in each approach. However, the enriched proteins did not show consistent stabilizations/destabilizations across techniques, not even when the same protein was assayed by all three techniques, which occurred 16 and 13 times at the 2- and 8-month time points, respectively. The observation of such differences for hit proteins across different techniques is not unusual, at it has been noted in both protein target discovery experiments^53,101 and other disease phenotype analyses employing multiple stability techniques.^56,57

Both the expression level and stability hits were also found to be enriched in unique pathways and biological process (see e.g., Table S6). Thus, the results of this work indicate expression level and stability profiling methods can provide complimentary information about AD. This is similar to that which has previously been noted about the application of protein expression level and stability profiling methods to that analysis of other disease phenotypes.^46,49,56 The close connection between protein folding stability and function makes protein stability profiling techniques especially well-suited to the identification of proteins with disease-related dysfunctions. This was demonstrated here with the proteasome, which displayed both AD-related stability and functional changes despite have no significant AD-related change in protein expression. Moreover, the different protein folding stability profiles of the proteosome subunits at the early and late stages of AD were consistent with different mechanisms of dysfunction.

Supplementary Material

Figures S1-S3 & Tables S6 and S7

Figure S1 includes volcano plots showing all assayed peptides/proteins and the identification of hits in the SPROX, TPP, LiP, and expression level analyses.

Figure S2 summarizes ORA results for hit proteins identified in each technique at the two age-points.

Figure S3 summarizes the stability changes associated with the proteasome subunits assayed in SPROX and LiP at both time points.

Table S6 summarizes the pathways that were enriched in the ORA of the stability hits and not in the expression level hits.

Table S7 summarizes the enrichment terms identified in the ORA of the 25 protein hits identified in at least two stability techniques at both age-points.

Table S1

Table S1 summarizes the methionine-containing peptides and proteins assayed in the SPROX analyses performed on the mouse brain samples in this study.

Table S2

Table S2 summarizes the proteins assayed in the TPP analyses performed on the mouse brain samples in this study.

Table S3

Table S3 summarizes the semi-tryptic peptides and proteins assayed in the LiP analyses performed on the mouse brain samples in this study.

Table S4

Table S4 summarizes the proteins assayed in the protein expression level analyses on the mouse brain samples in this work.

Table S5

Table S5 summarizes the hit proteins identified at both age-points in the SPROX, TPP, LiP, or expression level analyses.

Data Availability

The raw MS data generated in this work have been uploaded to the ProteomeXchange Consortium via the PRIDE partner repository and will be made publicly available with the data set identifier PXD051155 upon publication.