Structural proteomic profiling of cerebrospinal fluids to reveal novel conformational biomarkers for Alzheimer’s disease

Abstract

Alzheimer’s disease (AD) is the most common representation of dementia, with brain pathological hallmarks of protein abnormal aggregation, such as with amyloid beta (Aβ) and tau protein. It is well established that posttranslational modifications (PTMs) on tau protein, particularly phosphorylation, increases the likelihood of its aggregation and subsequent formation of neurofibrillary tangles (NFTs), another hallmark of AD. As additional misfolded proteins presumably exist distinctly in AD disease states, which would serve as potential source of AD biomarkers, we used limited proteolysis-coupled with mass spectrometry (LiP-MS) to probe protein structural changes. After optimizing the LiP-MS conditions, we further applied this method to human cerebrospinal fluid (CSF) specimens collected from healthy control, mild cognitive impairment (MCI), and AD subject groups, to characterize proteome-wide misfolding tendencies as a result of disease progression. The fully tryptic peptides embedding LiP sites were compared with the half-tryptic peptides generated from internal cleavage of the same region to determine any structural unfolding or misfolding. We discovered hundreds of significantly up- and down-regulated peptides associated with MCI and AD indicating their potential structural changes in AD progression. Moreover, we detected 53 structurally changed regions in 12 proteins with high confidence between the healthy control and disease groups, illustrating the functional relevance of these proteins with AD progression. These newly discovered conformational biomarker candidates establish valuable future directions for exploring the molecular mechanism of designing therapeutic targets for AD.

Graphical Abstract

Introduction

As the most common form of dementia, Alzheimer’s disease (AD) affects an estimated 55 million people globally as of 2020, with rates expected to increase in the absence of effective therapies.^1,2 AD is defined as a slow, progressive neurodegenerative disease, often characterized by initial memory impairment followed by cognitive decline, affecting behavior, speech, and the motor function.³ Peptide amyloid-β (Aβ) aggregates, forming amyloid plaques, were initially suggested to play a central role in AD pathogenicity, defining the premise of the amyloid hypothesis.⁴ The amyloid hypothesis gained more traction after several mutations to genes amyloid-β precursor protein, presenilin 1, and presenilin 2, were found directly responsible for aberrant Aβ accumulation in early onset forms of AD.^5,6 Following the dissemination of the Aβ hypothesis, which emphasizes the imbalanced production and clearance of Aβ in AD, extensive studies have focused on understanding the structural changes of Aβ peptides that results in aggregation, as well anti-amyloid strategies to reduce or reverse aggregation.^7,8 However, recent evidence has suggested that Aβ peptides may only participate in a fraction of the complicated and heterogeneous biological pathway of AD.^9,10 Clinical trials of Aβ therapies have repeatedly failed, suggesting that these theories do not fully recapitulate the biochemical dysfunction that occurs in AD progression.^11–13

Tau aggregation and the formation of neurofibrillary tangles (NFTs) is another hallmark of AD, strongly associated with progressive neuronal loss and cognitive decline.^14–17 Patients at the dementia stage of AD have been shown to display protein abundance changes with elevated unmodified tau protein expression (T-tau), as well as the presence of tau phosphorylated at threonine 181 (P-tau).¹⁸ However, the efficacy of tau-targeting strategies and clinical trials of tau inhibitors in AD have yet to be satisfactory.^19–21 Biochemical changes in the brain are reflected in the cerebrospinal fluids (CSF), making it an appealing sample source for the study of AD.^22–24

Current clinical procedures rely on altered levels of Aβ, T-tau, and p-Tau for AD diagnosis.^25–28 While these biomarkers are well-established, therapeutic efforts that target these proteins have been largely unsuccessful. As there is a dire need for earlier diagnosis of AD, in addition to robust therapeutics, there is much interest in identifying alternative biomarkers that are consistently present in AD disease states. Novel biomarkers also provide an increased capability to understand the biochemical processes that underlie AD progression. While seeking novel biomarkers is not a new endeavor, previous reports have shown conflicting results, lacking documentation of the applicability of these markers in CSF samples, as well as the impact of biomarker expression levels with respect to cognitive impairment.^29,30

Though current utility of biomarkers focuses on the level of abundance, these levels can have subtle differences during different disease state. Our alternative strategy is not reliant on abundance of a particular biomarker, but rather on protein conformational changes, which have been shown to be much starker in disease states. Previous reports have illustrated this with amyloid-β and tau, which display significant conformational changes during the formation of amyloid plaques and NFTs, respectively, in AD cases.³¹ This is a well-documented phenomenon, with a number of reports having established AD to behave according to a prion-like mechanism, where the presence of a misfolded protein will induce a cascade of misfolding on surrounding proteins.^32–36 Current evidence also indicates the presence of insoluble aggregates in instances of AD, which are assembled of misfolded proteins.^37,38 With this evidence, it is proposed that the misfolding of currently-unknown proteins occurs in an independent or synergistic way with Aβ or Tau.^38,39 Therefore, discovery of novel biomarkers based on the conformational change of proteins, other than tau and Aβ, is needed to enable early diagnosis of AD.

Leveraging the high resolution and sensitivity of advanced mass spectrometry (MS), a method that couples limited proteolysis with MS (LiP-MS) was developed by the Picotti Lab in 2014, offering a powerful tool for proteome-wide structural analysis.⁴⁰ This strategy can effectively detect structural changes in aggregation-prone proteins, and can also probe both subtle and pronounced structural changes of proteins on a large scale without compound modification or labeling. The development of the LiP-MS platform is grounded in the understanding that subtle protein structural changes can have pronounced differences in proteolytic accessibility, yielding specific peptides originating from the misfolded region, known to be conformotypic peptides.⁴⁰ This method involves the use of broad-specificity proteases with designed conditions, where the initial cleavage sites are dictated by the structural features of the protein. In the experimental LiP-MS workflow, proteins are extracted from cells or tissues under native conditions. The proteome extract is split into a control sample for complete digestion with LysC/trypsin and another experimental sample to be subjected to limited proteolysis with a broad-specificity protease K to generate structure-specific protein fragments. The samples previously subjected to limited proteolysis will be denatured and fully digested with LysC/trypsin to generate peptides amenable to bottom-up proteomics. The use of a double-digestion step makes this approach compatible with standard bottom-up proteomic analyses, simplifying downstream data analysis.^40,41 LiP-MS was first applied to a yeast system and was used to assess the structural features of more than 1,000 yeast proteins, resulting in the detection of altered conformations for ~300 proteins upon a nutritional change.^40,41 The LiP-MS strategy was then applied to multiple experimental systems, including bacteria undergoing nutrient adaptation as well as yeast response to acute stress. This structural approach is able to capture different molecular events including enzyme activity changes, enzymatic substrate site occupancy, allosteric regulation, phosphorylation, and protein-protein interactions with a resolution that pinpoints single functional sites.⁴² Of late, this method has been successfully applied to in situ analysis of the structural proteome in individuals with Parkinson’s disease, ultimately resulting in the identification of a new class of conformational biomarker,⁴³ supporting the potential value of this method for global analyses of the human structural proteome in different diseases. Unique from conventional concentration-based disease biomarkers, conformational biomarkers are proteins that undergo structural changes associated with disease state, which can provide insight into protein activity and reveal their function during disease progression.^40–43 When it comes to the global analysis of clinical biofluids, which frequently necessitates the assessment of large cohorts and analytes spanning wide dynamic ranges, these features would be extremely helpful in addressing biological issues.^44–47

In this study, the LiP-MS platform was employed to globally probe protein structural changes in human CSF to discover new misfolding- or aggregation-prone proteins, to provide a new understanding of the molecular mechanism of AD. Mild cognitive impairment (MCI) refers to relatively minor impairments in memory, an early stage of memory loss, or other cognitive ability loss in individuals.⁴⁸ The main difference between AD and MCI is the degree of cognitive impairment and decline, with the symptoms of AD typically beginning with MCI.⁴⁹ Therefore, the structural alteration among AD, MCI and healthy individuals may also possibly align with disease progression and severity. To characterize the misfolded peptides, we first determined the optimal ratio of protein concentration to proteinase K concentration and incubation time of proteinase K using pooled human CSF samples. We then implemented the LiP-MS platform with the optimized digestion conditions to explore significant protein structural changes in the CSF of a cohort of 5 healthy patients, 5 patients with mild cognitive impairment (MCI), and 5 AD patients, to understand the relationship between identified structural changes and disease progression. By comparing the intensity change of fully tryptic and half-tryptic peptides, the change in the accessibility or flexibility of the corresponding protein region was elucidated. Significantly up- and down-regulated peptides associated with MCI and AD cohorts were discovered by comparing protein expression in AD compared to control (AD:Ctrl), AD compared to MCI (AD:MCI), and MCI compared to control (MCI:Ctrl) populations, indicating potential structural changes that occur in AD progression. It was determined that 12 proteins exhibited 53 structural changes with high confidence, indicating functional relevance and warranting gene ontology analysis, with several of these proteins implicated in AD, including transthyretin (TTR), complement C3 (C3), and clusterin (CLU). These findings provide a foundation to further utilize conformational biomarker discoveries to complement conventional neuroimaging and pathological studies with respect to AD. Through this system-wide investigation of protein structural changes in human CSF collected at different stages of AD, we identified novel candidate conformational biomarkers, which will improve our understanding of AD pathogenesis and provide new avenues for the design of therapeutic strategies and pharmaceutical targets for AD.

Experimental section

Chemicals and materials.

MS grade trypsin/LysC mixture and dithiothreitol (DTT) were purchased from Promega (Madison, WI). Urea, tris base, ACS grade acetone, ACS grade methanol (MeOH), ACS grade acetonitrile (ACN), Optima UPLC grade ACN, Optima UPLC grade water, and Optima LC/MS grade formic acid (FA) were purchased from Fisher Scientific (Pittsburgh, PA). Iodoacetamide (IAA), trifluoroacetic acid (TFA), and ammonium formate (NH₄HCO₂) were purchased from Sigma-Aldrich (St. Louis, MO). Sep-Pak C18 cartridges and Bridged Ethylene Hybrid C18 particles were purchased from Waters (Milford, MA). Quantitative colorimetric peptide assay was purchased from Thermo Scientific (San Jose, CA).

Participants.

All study procedures involving human subjects have been approved by the University of Wisconsin Institutional Review Board and abide by the Declaration of Helsinki principles. Each enrollee was provided a signed informed consent form before participation. Fifteen enrollees (including five individuals in AD stage and equal number of healthy controls and MCI) from Wisconsin Alzheimer’s Disease Research Center (ADRC) participated in this study. Detailed subjects’ information can be found in Supplemental Table S1. All AD participants were diagnosed via applicable clinical criteria in standardized and multidisciplinary consensus conferences.^50,51 Cognitive normalcy was determined based on intact cognitive performance by a comprehensive battery of neuropsychological tests, lack of functional impairment, and absence of neurological or psychiatric conditions that might impair cognition.^52,53. 1 mL CSF samples were collected through lumbar puncture.

CSF protein tryptic digestion.

Concentrations of the CSF proteins were measured by BCA assay. Human CSF samples (N=5 for either healthy control, MCI or AD dementia group) with equal amounts of protein (100 µg) were reduced by adding DTT to a final concentration of 5 mM and incubated at room temperature (RT) for 30 min. 15 mM of IAA was used for alkylation of cysteines by incubating for 30 min at RT in the dark. 8M urea was added to dissolve the pellets and 50 mM Tris buffer was used to dilute the samples to a urea concentration of <1 M. On-pellet digestion was performed with LysC/trypsin mixture in a 50:1 ratio (protein to enzyme, w/w) at 37 ℃ overnight. The digestion was quenched with 1% TFA and samples were desalted with Sep-Pak C18 cartridges. Concentrations of peptide mixture were measured by peptide assay following the manufacturers’ protocols.

CSF protein limited proteolysis (LiP).

Time course LiP experiments were conducted using of 100 µg pooled CSF sample extracts with incubation times of 1-, 5-, 30-, and 60-min. Proteinase K from Tritirachium album (Sigma) was added to the protein extracts at a fixed enzyme-to-protein ratio of 1:100. Similarly, four different fixed enzyme-to-protein ratios (1:10, 1:50, 1:100 and 1:1,000) were assessed with a fixed reaction time of 5 min. A fixed enzyme-to-protein ratio of 1:1000 and 1 min incubation at RT was selected for the cohort CSF samples. The digestion was stopped by boiling for 5 min. The digestion mixtures were then subjected to complete tryptic digestion as described above.

LC-MS/MS analysis.

Samples were analyzed on an Orbitrap Fusion Lumos Tribrid mass spectrometer (Thermo) coupled to a Dionex UltiMate 3000 UPLC system. Each sample was dissolved in 3% ACN, 0.1% formic acid in water before loaded onto a 75 μm inner diameter homemade microcapillary column packed with 18 cm of Bridged Ethylene Hybrid C18 particles (1.7 μm, 130 Å, Waters) and fabricated with an integrated emitter tip. Mobile phase A was composed of water and 0.1% FA while mobile phase B was composed of ACN and 0.1% FA. LC separation was achieved across a 100 min gradient elution of 3% to 30% mobile phase B at a flow rate of 300 nL/min. Survey scans of peptide precursors from 200 to 1500 m/z were performed at a resolving power of 60k (at m/z 200) with an AGC target of 2 × 10⁵ and maximum injection time of 100 ms. Top 20 abundant precursor ions were selected for fragmentation with a normalized collision energy of 30. MS2 acquisition was performed with an isolation window of 1.6 Da, a resolving power of 30k, an AGC target of 5 × 10⁴, a maximum injection time of 54 ms, and a lower mass limit of 120 m/z. Precursors were subject to dynamic exclusion for 45 s with a 10-ppm tolerance. Each sample was acquired in technical triplicates.

Data analysis.

Raw files were searched against the UniProt human proteome reviewed database using MaxQuant (version 1.5.2.8). For trypsin digestion, trypsin/P and LysC were selected as the enzymes and two missed cleavages were allowed. For limited proteolysis, semi-specific digestion was selected, and three missed cleavages were allowed. Carboxyamidomethylation of cysteines (+57.0214 Da) was defined as fixed modification and oxidation of methionine residues (+15.99492 Da) was chosen as a variable modification. Search results were filtered to 1% false discovery rate (FDR) at both peptide and protein levels, calculated based on a target-decoy approach. Bioinformatics analyses, including hierarchical clustering, protein intensity profiling, volcano plots and box plots were achieved using R packages and Python scripts. For protein intensity profiling, biological processes and pathways were enriched using DAVID bioinformatics resources with an FDR cutoff of 0.05. For structural analysis, PyMol (version 1.8.2.1) was employed to visualize the conformational changes. Upregulated and downregulated peptides were identified from MaxQuant protein and peptide results. For each of the proteins identified in MaxQuant, the mean intensity of five replicates was calculated for each of three sample groups: AD, MCI, and control. For each protein, fold change calculations for AD:Ctrl, AD:MCI, and MCI:Ctrl were calculated (Equation 1). For each of the five protein replicates, the p-value was calculated. If the p-value was determined to be greater than 0.05, the corresponding fold change was adjusted to equal 1. For each MaxQuant-identified peptide corresponding to each MaxQuant-identified protein, the p-value of each of the five peptide intensities (MCI, AD, Ctrl) were calculated. Peptides with a p-value greater than 0.05 were eliminated. Intensity values for all remaining peptides were normalized by dividing the raw intensity of the replicate (n=3) by the associated protein fold change (Equation 2). Mean normalized intensities for each peptide were calculated for AD:Ctrl, AD:MCI, and MCI:Ctrl. The mean control intensity was also calculated. Fold changes for each peptide were calculated (Equation 3). Peptide fold changes with values less than 0.67 were classified as “downregulated”, and peptide fold changes with values greater than 1.5 were classified as “upregulated”. Then peptides with significantly changed abundances were filtered out and the matched peptides were selected as a hit where LiP cleavage maps to the same protein region. All described calculations were carried out in Python (v.3.10.4) and are publicly available (https://github.com/lingjunli-research/LiP-MS-Early-Filter).

$${\mathrm{AD}:\mathrm{Ctrl}}_{\mathrm{FC}}=\frac{\stackrel{\_}{{\mathrm{AD}}_{\mathrm{intensity}}}}{\stackrel{\_}{{\mathrm{Ctrl}}_{\mathrm{intensity}}}};{\mathrm{AD}:\mathrm{MCI}}_{\mathrm{FC}}=\frac{\stackrel{\_}{{\mathrm{AD}}_{\mathrm{intensity}}}}{\stackrel{\_}{{\mathrm{MCI}}_{\mathrm{intensity}}}};{\mathrm{MCI}:\mathrm{Ctrl}}_{\mathrm{FC}}=\frac{\stackrel{\_}{{\mathrm{MCI}}_{\mathrm{intensity}}}}{\stackrel{\_}{{\mathrm{Ctrl}}_{\mathrm{intensity}}}}$$ Equation 1:

$${{\mathrm{AD}}_{\mathrm{n}}:\mathrm{Ctrl}}_{\mathrm{norm}}=\frac{{\mathrm{AD}}_{\mathrm{n}}}{{\mathrm{AD}:\mathrm{Ctrl}}_{\mathrm{FC}}};{{\mathrm{AD}}_{\mathrm{n}}:\mathrm{MCI}}_{\mathrm{norm}}=\frac{{\mathrm{AD}}_{\mathrm{n}}}{{\mathrm{AD}:\mathrm{MCI}}_{\mathrm{FC}}};{{\mathrm{MCI}}_{\mathrm{n}}:\mathrm{Ctrl}}_{\mathrm{norm}}=\frac{{\mathrm{AD}}_{\mathrm{n}}}{{\mathrm{MCI}:\mathrm{Ctrl}}_{\mathrm{FC}}}$$ Equation 2:

$${\mathrm{AD}:\mathrm{Ctrl}}_{\mathrm{FC}}=\frac{\stackrel{\_}{{\mathrm{AD}:\mathrm{Ctrl}}_{\mathrm{norm}}}}{\stackrel{\_}{{\mathrm{Ctrl}}_{\mathrm{mean}}}};{\mathrm{AD}:\mathrm{MCI}}_{\mathrm{FC}}=\frac{\stackrel{\_}{{\mathrm{AD}:\mathrm{MCI}}_{\mathrm{norm}}}}{\stackrel{\_}{{\mathrm{Ctrl}}_{\mathrm{mean}}}};{\mathrm{MCI}:\mathrm{Ctrl}}_{\mathrm{FC}}=\frac{\stackrel{\_}{{\mathrm{MCI}:\mathrm{Ctrl}}_{\mathrm{norm}}}}{\stackrel{\_}{{\mathrm{Ctrl}}_{\mathrm{mean}}}}$$ Equation 3:

Results and Discussion

Effect of different enzyme-to-protein ratios and incubation time

Previous studies have suggested that, under LiP conditions, the sites of initial proteolytic cleavage are dictated by the structural features of the substrate, emphasizing the necessity of determining an appropriate enzyme-to-protein ratio and enforcing a limited digestion time are critical, as failure to do so can result in the display of slightly different sequences.^40–43 The LiP-MS method has been applied in various biological systems including for the global analysis of protein structural changes upon specific environmental perturbations, targeted analysis of proteins of interest, and large-scale identification of protein–small molecule interactions.⁴¹ Although this method enabled probing of subtle and pronounced structural changes of proteins in numerous applications, the optimal LiP condition in the human CSF system requires further exploration. As a result, we first performed a pilot study to determine the optimal LiP-MS reaction conditions, informed by the conditions that revealed the most pronounced structural changes, specifically with regards to enzyme-to-protein ratio and limited digestion time in human CSF samples. To optimize these digestion factors, protein extracts from pooled CSF samples were subjected to the LiP-MS workflow (Figure 1a) using proteinase K and trypsin at a variety of enzyme-to-substrate (E/S) ratios and incubation times. Higher numbers of peptide and protein identifications were observed at decreasing E/S ratios (Figure 1b, c), while an increase in incubation time resulted in fewer peptide and protein identifications. Although the number of identified peptides at the determined optimal conditions, E/S ratio 1:1000 and 1 min incubation time, were higher than the tryptic peptides, the protein identification numbers with proteinase K were still lower than trypsin. The amount of half-tryptic (HT) peptides mostly increased with a lower E/S ratio, though the number of HT peptides decreased after prolonged incubation, likely due to the occurrence of secondary cleavages (Figure 1d). Although preferential use of a protease K-to-substrate ratio of 1:100 and an incubation time of 1 min was recommended in previous LiP-MS experiments,⁴¹ these recommendations do not necessarily apply when adapting an experiment for an alternative sample type. The example, sample complexity and occurrence of fully tryptic (FT) peptides pose critical factors for consideration. Both trypsin and proteinase K digestions resulted in a fraction of peptides containing missed cleavages when compared to their respective control samples, and a higher E/S ratio was determined to cause more frequent missed cleavages (Figure 1e). In the subsequent large cohort experiment, an E/S of 1:1000 and an incubation time of 1 min was used to achieve the most pronounced structural change information. These conditions ensured the detection of a large number of LiP sites within a proteome, a reasonable quantity of proteins and peptides identified, and acceptable proteome coverage.

Figure 1.

(a) Experimental workflow of LiP-MS. Protein extracts from pooled CSF samples were subjected to the LiP-MS protocol using Proteinase K. The resulting fragments from LiP-treated and control samples were analyzed by LC–MS/MS. (b-e) Effect of different E/S ratios with a fixed incubation time and different enzyme incubation time with a fixed E/S ratio. (b) The total number of proteins and (c) peptides, (d) the percentage of missed cleavages, and (e) half-tryptic peptides out of the total number of peptides identified are reported for the different treatments. Error bars indicate standard deviation values from three replicated analyses.

CSF proteome reveals candidate biomarkers that are significantly altered in AD

We first explore the CSF protein-level biomarkers reflective of the diverse brain-based pathophysiologies that contribute to AD. Proteins were extracted, denatured, desalted, and then digested with trypsin and were compared according to their relative abundance by label free quantification (LFQ). The LFQ intensity was used for subsequent LiP-MS normalization analyses. In total, 856 and 501 proteins were respectively identified and quantified among all CSF samples. To determine whether the proteome fingerprints in CSF differed among control, MCI, and AD subjects, we first evaluated discrimination/separation between experimental groups using principal component analysis (PCA). Strong group separation was achieved in CSF among all three groups (Figure S1). We then performed hierarchical clustering of all quantified proteins to explore their profiles at different AD stages (Figure 2a and Table S2), which illustrated column-wise clustering of biological replicates in either control or disease groups, suggesting larger intergroup differences than intragroup variations were present. Samples from MCI and AD cohorts were found grouped together, suggesting pathological relevance. One-way analysis of variance (ANOVA) was conducted to compare protein abundance across three cohorts, identifying 200 significantly changed proteins (FDR = 0.05) (Table S3). The profiles of these proteins were assessed through hierarchical clustering (Figure 2b), and differential expression was assessed through a statistical t-test analysis (p < 0.05) (Figure 2c). Compared to normal healthy controls, AD and MCI patients showed significantly increased expression of 11 and 39 proteins, respectively, in CSF. Amongst these, 5 proteins both showed significant increase in both AD:Ctrl and MCI:Ctrl groups, including the complement protein (C1q) and adhesion G protein-coupled receptor b1. Previous reports have shown that C1q binds to amyloid-β fibrils, resulting in the activation of the classical complement pathway which is a key contributor to the defense of infections, clearance of pathogens, removal of apoptotic cells, and maintenance of homeostasis.^54,55 C1q has also been found associated with the bulk of amyloid deposits in the brain.^56,57 Adhesion G protein-coupled receptors have been previously proposed as drug targets for neurological diseases.^58,59 In contrast, the expression of 9 and 23 CSF proteins were found to be significantly decreased in AD patients and MCI patients, respectively, when compared to control samples. 5 down-regulated proteins overlap in the AD:Ctrl and MCI:Ctrl groups, all of which are found to contribute to the mediation of the effector phase of humoral immunity. Altogether, there are fewer proteins with significant abundance changes in AD:Ctrl than other two groups, more unique proteins were found in the AD:Ctrl analysis (p< 0.05).

Figure 2.

(a) Hierarchical clustering of 501 quantified proteins in CSF samples. (b) Hierarchical clustering of ion intensities of 200 significantly changed proteins (one-way ANOVA, FDR 0.05). (c) Volcano plots illustrating pairwise comparisons of protein expression levels. Points above horizontal dashed lines represent significantly altered proteins (two-sided t test, p value < 0.05, p values were adjusted by Benjamini–Hochberg correction for multiple comparisons. Significantly downregulated proteins are shown in green (protein fold change< 0.67) and upregulated ones are shown in red (protein fold change > 1.5). (d) GO functional classification of transcripts according to their assigned molecular function, cellular components, and biological process.

To better discern the cellular localization and biological functions of both the up and down regulated proteins, we performed gene ontology (GO) enrichment analysis (Figure 2d). The identified proteins were primarily localized in the extracellular region, an expected finding, as CSF serves as a biofluid outside of the cells. Proteins were determined to exhibit significant up or down regulation in CSF, which were mainly involved in biological processes such as regulation of antigen binding, activation of complement system, and other immune responses. As expected, studies have revealed that immune system state is strongly associated with the pathogenesis of AD.^60,61 In addition, complement protein expression and complement activation are strongly implicated in the neuroinflammation, neuronal and synapse loss, and subsequent neurodegeneration of observed AD impacts.^62,63

LiP-MS data processing and structural biomarkers

Although there have been numerous reports investigating the CSF proteome and presenting protein biomarkers associated with pathophysiological pathways, the structural alteration of those biomarker proteins and associated proteins has been less studied.^64–67 Herein, the LiP–MS platform was used to screen protein structural changes and discover novel conformational biomarkers in the CSF proteome. An automated data processing workflow was established to profile the level of peptide abundance change, normalized by corresponding protein-level abundance change (Figure 3 and Table S4). The detailed description of LiP-MS data processing could be found in the data analysis section. The peptides with significantly changed abundance potentially revealed structural alteration. Among the significantly changed peptides, we further match the sequence corresponding to the specific region that undergoes the structural change in one population to a subsequence that exists in another population, representing a high confident hit. In total, 56 and 105 tryptic peptides were significantly upregulated in MCI and AD groups compared to the control group, respectively. Among them, 29 and 82 half-tryptic peptides embedding LiP sites were more abundant in MCI and AD groups compared to the control group. These results complement the theory that a misfolded region will be protected from proteolysis under LiP-MS conditions, particularly informative for disease studies. Interestingly, all the proteins that were identified as protected from proteolysis in the MCI group were also found in the AD group, suggesting the structural alteration is irreversible during AD progression, particularly with respect to apolipoprotein A-II, apolipoprotein D, amyloid-like protein 1, and complement C3 (Figure 4a and Table S5). Amyloid-like aggregates have been reported to contribute to the aging process and accumulate in neuritic plaques in AD.^68,69 The identification of structural changes of amyloid-like protein 1 is consistent with this amyloid aggregation observation. Apolipoproteins have a demonstrated importance in brain cholesterol and amyloid-ß metabolism, and thus have been reported as novel risk markers for Alzheimer’s pathology.⁷⁰ Apolipoprotein A-II also induces acute-phase response associated amyloid A amyloidosis in mice through conformational changes of plasma lipoprotein structure, which is consistent with our conclusion that apolipoprotein A-II could be conformational biomarker.⁷¹ In contrast, 735 and 456 tryptic peptides exhibited significant decrease in relative abundances in MCI and AD groups respectively compared to those of the control group. Among them, 455 and 254 half-tryptic peptides embedding LiP sites were significantly downregulated in MCI and AD groups respectively than that of the control group, as the susceptible regions were exposed to the enzyme-accessible surface, producing tryptic peptides with significantly decreased expression levels in AD and MCI groups in comparison to the control group, including apolipoprotein E, cystatin-C, amyloid-β A4, and fibrinogen. It is noted that ApoE is commonly found in amyloid deposits and its interaction with Aβ peptide has been reported.⁷² The structural alteration of ApoE may have an amyloid-interfering property and association with amyloid-induced cytotoxicity.⁷³ Similarly, most of the proteins displaying this trend in MCI also existed in the AD group, suggesting continuation and permanence of structure alteration during AD progression. To better discern the pathway in which these proteins participate, we performed GO enrichment analysis for Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway⁷⁴ (Figure 4b) for all the proteins with significantly changed peptides. The consensus of the analysis showed these proteins to be over-represented in complement and coagulation cascades pathways, cholesterol metabolism pathways, ECM-receptor interaction pathways, cell adhesion molecule pathways, and many other neurodegenerative disease-potentiating pathways.

Figure 3.

An automated data processing workflow for LiP-MS. Spectral data was pre-processed in MaxQuant and an automated python script was used to identify up- and down-regulated peptides.

Figure 4.

(a) Volcano plots showing pairwise comparisons of peptide expression levels. Points above horizontal dashed lines represent significantly altered proteins. Significantly downregulated peptides are shown in green (protein fold change < 0.67, p value < 0.05) and upregulated ones are shown in red (protein fold change > 1.5, p value < 0.05). (b) KEGG pathway analyses of the significantly structural changed proteins in AD progression. (c) High-confidence candidates that underwent structural changes in AD progression. (d) STRING interaction diagram of high-confidence candidates. (e) Metascape visualization of the interactome network formed by high-confidence candidates. Each term is represented by a circle node, where its size is proportional to the number of input genes falling into that term, and its color represents its cluster identity.

Representative conformational biomarkers extracted from the LiP-MS analyses

We next explored whether structural peptides could be used to localize the conformational protein regions as belonging to the healthy or AD and MCI groups, with the goal of gaining an understanding as to how the protein underwent exposure or aggregation. We detected 53 structural changes in 12 proteins with high confidence in total between healthy control and disease group (Figure 4c and Table S6). Their possible interactions were further analyzed through the STRING and Metascape protein-protein interaction network analysis (Figure 4d). Some of the hits were associated with cognitive function and biomarkers of neurodegeneration including clusterin (CLU) and complement C3. CLU is a multifunctional glycoprotein that has been implicated in AD.⁷⁵ CLU’s relationship with Aβ has been of great interest to the AD field, including its apparent role in potentiating Aβ aggregation and clearance.⁷⁶ As shown in Figure 5a, the longer half-tryptic peptide (HTSDSDVPSGVTEVVVK) mapped to the β-strand of CLU was exclusively observed in Control compared to AD or MCI. The shorter half-tryptic peptide (SDSDVPSGVTEVVVK) derived from internal cleavage of the same unfolded peptide was detected with elevated level in MCI and AD compared to the healthy control. These data present a specific region of CLU prone to conformational changes. CLU is colocalized with amyloid plaques in AD to regulate Aβ deposition,⁷⁷ suggesting the possibility of CLU as a conformational biomarker accessible through CSF. Complement C3 is activated in the human AD brain and is implicated in neurodegeneration in mouse models of amyloidosis and tauopathy.⁷⁸ Five matched peptides from complement C3 were found through the LiP-MS data (Figure 5b–f) indicating its structural alterations associated with AD progression. For example, as shown in Figure 5c, the fully tryptic peptide (FVTVQATFGTQVVEK) mapped to the β-strand of complement C3 was significantly decreasing in AD compared to MCI. The shorter half-tryptic peptide (ATFGTQVVEK) derived from internal cleavage of the same unfolded peptide was detected increasingly in AD compared to MCI indicating the associated region was protected from proteolysis possibly due to its misfolded structure from MCI to AD progression. Similarly, the fully tryptic peptide (VYAYYNLEESCTR) of complement C3 was found to be significantly downregulated in AD compared to the control, and the shorter half-tryptic peptide (YNLEESCTR), derived from internal cleavage, was detected as significantly upregulated in AD compared to the control (Figure 5f). These observations indicate that this region is more likely to be exposed to proteolysis, possibly due to its misfolded structure associated with control to AD progression. This structural change region is adjacent to the NTR/C345C domain which was found in complement C3. The structure of C3 shows that the NTR domain is located in an exposed position relative to the rest of the molecule, and possibly is the region that covalently binds to cell surface carbohydrates, including components of bacterial cell walls and immune aggregate.^79–81 The altered surface area adjacent to the NTR domain may change its covalent binding affinity and further affect its interaction with other immune-related proteins. Although the function of this domain in complement C3 is poorly understood, this structural alteration may provide a clue to how this domain functions in the activation of the complement systems in the AD patients.^82–84 Although there is no fully characterized structure of CLU available in the literature,⁸⁵ AlphaFold has enabled the prediction of its 3D model and marked structural changes.^86,87 We then further showed all the specific positions that underwent structural changes of CLU (Figure 5g) and complement C3 (Figure 5h) within their protein structures, respectively. Most of the specific regions that underwent structural changes were located on the protein surface which was consistent with the principle of limited proteolysis in the native condition to prioritize the digestion of surface residues. Together, our findings show that LiP–MS can uncover protein structural changes and elucidate potential protein alteration regions in human CSF samples during AD progression.

Figure 5.

Structural changes in selected CSF proteins, clusterin and complement C3. (a) The relative abundance obtained for conformotypic peptides of clusterin in CSF. (b-f), The relative abundance obtained for conformotypic peptides of complement C3 in CSF. “ns” indicates not statistical significance (p>0.05), whereas asterisk (*) and (**) indicate statistical significance at P﹤0.05 and P﹤0.01, respectively. (g-h) Structural changes of clusterin mapped to AlphaFold structural prediction (AF-P10909-F1-model_v2) and complement C3 mapped to crystal structure (PDB: 2a73). Colored regions indicate the protein regions underwent structural changes corresponding to the significantly changed peptides.

Conclusions

Here, we establish a workflow and provide an automated data processing pipeline to explore conformational biomarkers at different AD stages, providing a list of conformational biomarker candidates that may inspire future investigations for diagnosis and treatment of AD. With knowledge of the specific region undergoing structural changes, correlations can be established between the misfolding region and its functional domains. Future studies will need to be conducted to determine the significance of the protein conformational changes with AD pathophysiology. It will be informative to use these functional conformational biomarkers to aid in monitoring pathological progression at the stages of preclinical, MCI, and AD dementia. Moreover, future work should establish a correlation between conformational biomarkers and conventional neuroimaging and pathological features in AD. Through this system-wide investigation of protein structural transitions in human AD CSF, the newly characterized candidate conformational biomarkers will improve our understanding of AD pathogenesis and provide new avenues to design therapeutic targets for AD.

Supplementary Material

Supporting Information

Figure S1: Graphic scattered plot of PCA of protein intensity profiles from CSF samples;

Table S1: CSF sample information;

Supplementary Table 2

Table S2: Raw peptide intensities of healthy, MCI and AD CSF samples measured for trypsin digestion only;

Supplementary Table 3

Table S3: One-way analysis of variance (ANOVA) to compare protein abundance across three stages;

Supplementary Table 6

Table S6: Overview of all matched hits with structurally changed candidate peptides in human CSF among three groups.

Supplementary Table 4

Table S4: Raw peptide intensities of healthy, MCI and AD CSF samples measured for limited proteolysis;

Supplementary Table 5

Table S5: Automated processed peptide intensities with limited proteolysis among three groups;