Amyloid β Proteoforms Elucidated by Quantitative LC/MS in the 5xFAD Mouse Model of Alzheimer’s Disease

Abstract

Numerous Aβ proteoforms, identified in the human brain, possess differential neurotoxic and aggregation propensities. These proteoforms contribute in unknown ways to the conformations and resultant pathogenicity of oligomers, protofibrils, and fibrils in Alzheimer’s disease (AD) manifestation owing to the lack of molecular-level specificity to the exact chemical composition of underlying protein products with widespread interrogating techniques, like immunoassays. We evaluated Aβ proteoform flux using quantitative top-down mass spectrometry (TDMS) in a well-studied 5xFAD mouse model of age-dependent Aβ-amyloidosis. Though the brain-derived Aβ proteoform landscape is largely occupied by Aβ1−42, 25 different forms of Aβ with differential solubility were identified. These proteoforms fall into three natural groups defined by hierarchical clustering of expression levels in the context of mouse age and proteoform solubility, with each group sharing physiochemical properties associated with either N/C-terminal truncations or both. Overall, the TDMS workflow outlined may hold tremendous potential for investigating proteoform-level relationships between insoluble fibrils and soluble Aβ, including low-molecular-weight oligomers hypothesized to serve as the key drivers of neurotoxicity. Similarly, the workflow may also help to validate the utility of AD-relevant animal models to recapitulate amyloidosis mechanisms or possibly explain disconnects observed in therapeutic efficacy in animal models vs humans.

Graphical Abstract

INTRODUCTION

Alzheimer’s disease (AD) patients often present a high burden of neuritic plaques and tangles¹ in the brain with the phenotypic expression of neurodegeneration, brain atrophy, and memory loss.^2,3 Neuritic plaques are primarily composed of aggregated amyloid-β (Aβ) peptides,^4–6 which are derived from the cleavage of the amyloid precursor protein (APP). Aβ aggregation is believed to be a complex process with monomeric species capable of aggregating into multiple aggregated species, including soluble Aβ oligomers (AβOs) that are believed to be neurotoxic and larger, insoluble fibrils⁷ that further assemble into the macroscopic plaques that are readily detected by histological stains.⁸ In addition to the complexity in the quaternary structure of its aggregate forms, numerous Aβ proteoforms (i.e., primary amino acid sequence truncations or extensions, single nucleotide polymorphisms (SNPs), post-translational modifications, or combinations thereof)⁹ have been identified in the human brain and cerebrospinal fluid (CSF).^4,10–16

Despite a recognition that the accumulation of AβOs is one of the central toxic events in AD,^17–19 significant controversy remained due to the poor performance of past Aβ-targeting therapeutics in clinical trials.²⁰ However, therapeutic targeting of Aβ recently has been reinvigorated by the clinical successes of donanemab, an immunotherapy targeting a relatively low-abundant N-terminal pyroglutamyl Aβ proteoform (pE3), and lecanemab, an immunotherapy targeting soluble Aβ protofibrils and oligomers.²¹ Donanemab has been shown to reduce amyloid plaque load²² and slow cognitive and functional decline in early AD²² in the Phase 3 TRAILBLAZER-ALZ 4 study. The continued development of other next-generation Aβ targeting therapeutics presents an immense technical challenge because of a lack of knowledge regarding the precise identity of the most toxic AβO structures^15,23–25 whose chemical makeup may be impacted by a heterogeneous Aβ proteoform landscape that can vary spatiotemporally within AD.

Along with serving as therapeutic targets, the Aβ1−42 and Aβ1−40 peptides have been elevated as diagnostic biomarkers indicative of AD pathology (e.g., Aβ42/40 in the cerebrospinal fluid (CSF) and brain).^26,27 Similarly, they often serve as model peptides for aggregation studies that seek to define key structural elements associated with Aβ peptide aggregation by probing the folding free-energy landscape of Aβ through structural biology assays and computer simulations. However, such assays often consider compositionally limited systems (i.e., single Aβ peptide sequence) and do not account for complexities of the brain macrosystem.^7,23,28–30 Proteomics data show that canonical Aβ1−42 and Aβ1−40 represent only two of many Aβ proteoforms (i.e., protein-forms) in the AD brain.^4,10–14,31 For example, LC-MS/MS-based differential expression studies of soluble and insoluble Aβ fractions from post-mortem tissues obtained from patients with severe dementia and pathologically confirmed AD revealed numerous N- and C-terminal truncations and different classes of post-translational modifications (PTMs) (e.g., pyroglutamate and oxidation).¹⁵ As noted for pE3, various other PTMs have been reported to increase Aβ toxicity, for example, C-terminally extended Aβ43,³² N-terminal extensions,³³ aspartic acid isomerization,³⁴ and Ser26 phosphorylation.³⁵ While such investigations suggest that distinct proteoforms preferentially segregated between the soluble and insoluble pools, the limited cohort size combined with high proteoform diversity and biological variability has presented a substantial barrier to achieving statistical relevance in correlative analyses.

In the current study, we utilized the 5xFAD mouse model to investigate age- and solubility-dependent flux of Aβ proteoforms in the brain. Unlike the human AD population, the 5xFAD mouse offers a model to study processes that contribute to severe Aβ-amyloidosis,^36,37 mirroring many aspects of AD in an age-dependent manner^37,38 without confounding factors like varied genotypes, comorbidities, medications, etc. This model overexpresses five familial AD mutations that produce human Aβ1−42, including three APP mutants (Swedish, K670N/M671L; Florida, I716 V; and London, V717I) and two PSEN1 mutants (M146L and L286V), which supposedly impact the function of the catalytic component of γ-secretase.³⁹ Within one year, 5xFAD mice exhibit histological, biomolecular, or clinical features^36,37,40 that parallel the human AD pathology (Figure 1A), where the neurobiological course includes early AβO buildup,³⁸ plaque accumulation,³⁷ neuronal loss,³⁶ and cognitive impairment.³⁸ Several Aβ proteoforms pertinent to those expressed in the human brain and CSF have been observed in this model system.^41–45 However, to date, a systematic evaluation of Aβ proteoforms in separated soluble and insoluble fractions over the course of the disease, in 5xFAD mice or human AD, has not been performed. Understanding factors that drive the formation and distribution of soluble Aβ monomers or AβOs vs insoluble aggregates could have a major impact on the development of therapeutic targets. To address this, we have applied ultracentrifugal fractionation to separate soluble lower-molecular-weight oligomers from insoluble aggregates, followed by advanced top-down mass spectrometry and hierarchical clustering to identify the characteristics of the proteoform landscape of the soluble and insoluble Aβ isolated from 5xFAD mice at ages spanning these parallel AD milestones.

Figure 1.

Amyloid β (Aβ) proteoform discovery and quantitation from 5xFAD mouse brain tissues. (A) Neurocognitive decline and pathological progression in 5xFAD mice marking parallels to Alzheimer’s disease (a: Buskila et al. 2013, Oblak et al. 2021, b: Oakley et al. 2006, c: Viola et al. 2022). Sampled age groups are demarcated with arrows. (B) Study design for MS analysis of Aβ from 5XFAD transgenic mouse brain tissues. The order of preparation and analyses were randomized. (C) Schematic for immunoprecipitation of Aβ from brain tissue, followed by nano-LC/high-resolution mass spectrometry (MS). (D) MS data are first acquired in discovery mode, where an intact precursor ion is quantified, and its corresponding fragmentation spectrum is used for identification. Targeted mode MS (tMS2) acquisition is used for proteoform quantitation with product ions from fragmented proteoform precursor ions.

MATERIALS AND METHODS

5xFAD Transgenic Mouse Model

Transgenic mice were bred from adult 5xFAD heterozygous male mice (RRID: MMRRC_034840-JAX) on the B6SJL background with wild-type females as per the IACUC-approved protocol. The mice were maintained on an unrestricted diet of standard mouse chow and water. The LC/MS study age groups were selected based upon previously characterized AD-relevant critical time points in 5xFAD mice where increases in amyloid burden and/or neurodegeneration have been documented, including ages 2, 4−5, 8−9, and 12−14.5 months (N = 3/sex/age group) (Figure 1B). Euthanasia was performed by CO₂ overdose, followed by decapitation, and the brains were harvested, weighed, and flash-frozen in liquid nitrogen.

Tissue Preparation and Aβ Immunoprecipitation (IP)

Aβ was immunoprecipitated from each ultracentrifugation fraction prior to MS investigations (Figure 1C). Optima LC/MS-grade water or acetonitrile (Fisher Scientific, Hampton, New Hampshire) was used in all aspects of sample preparation and LC/MS analysis. Tissues were Dounce-homogenized (25−30 strokes) in 1:10 homogenization buffer (HB) consisting of 1× cOmplete protease inhibitor (Roche, Basel, Switzerland) and 0.45% CHAPS in Ham’s F12 medium (Caisson Laboratories Smithfield, UT). Homogenates were centrifuged at 17,000g for 30 min at 4 °C, followed by ultracentrifugation at 100,000g for 1 h at 4 °C. Pellets were resuspended in 300 μL of HB. The supernatant and pellet, referred to as soluble and insoluble fractions, respectively, were subsequently aliquoted for single-use and flash-frozen. The Aβ IP protocol used was similar to that previously published.^15,16 Generally, the soluble and insoluble fractions were thawed with the addition of 8 M guanidine to a final concentration of 1.5 M and 4 M, respectively, and denatured overnight at 4 °C. Blocking buffer, consisting of 0.05% α-casein in HB, was used to dilute guanidine concentration to <0.5 M and spiked with an internal standard consisting of 5 ng of N15 stable isotope labeled Aβ1−42 (rPeptide, Watkinsville, GA). The IPs were performed with 10 μg each of antibodies 6E10 (RRID: AB_2564652, BioLegend, San Diego, CA) and 4G8 (RRID: AB_2565324, BioLegend, San Diego, CA) immobilized on protein A/G magnetic beads (Pierce, Thermo Fisher Scientific, Waltham, Massachusetts) with overnight end-to-end incubation at 4 °C. The beads were sequentially washed with 1 mL of HB, followed by 1 mL of 50:50 HB/water, and then 3× 1 mL of water on the Kingfisher Flex (Thermo Scientific, Waltham, MA) using BindIt software (v.3.9). Beads were released in water, transferred to fresh Eppendorf LoBind tubes, the water removed by centrifugation, and then flash-frozen and stored at −80 °C.

LC-MS/MS Analysis

Data-dependent acquisition (DDA) and target MS² (tMS2) acquisition modes on a Fusion Lumos Tribrid mass spectrometer (Thermo Fisher Scientific, Waltham, MA) were used for proteoform discovery and quantitation (Figure 1D). For both LC/MS modes, the beads were treated with 30% formic acid at room temperature for 30 min. The eluent formic acid concentration was diluted to 15%, followed by centrifugation at 20,000g for 5 min prior to transfer to α-casein blocked vials prepared by incubation of 0.05% α-casein at room temperature for 20 min, followed by 1× wash with water, 1× wash with 15% formic acid, and 3× washes with water. For each analysis, 2 μL of IP eluent was loaded on a custom-packed trap column containing PLRP-S reversed-phase stationary phase (Agilent, Santa Clara, CA) (150 μm × 20 mm; 1000 Å, 5 μm particles) at 2.5 μL/min with 5% v/v acetonitrile in water and 0.2% v/v formic acid with trapping for 20 min. Separations were performed on a custom-packed PLRP-S column (75 μm × 200 mm; 1000 Å, 5 μm particle size) at 300 nL/min using a 40 min gradient of 5−35% acetonitrile and 0.2% formic acid (v/v).

The mass spectrometer was operated in a positive-ion polarity; a spray voltage of 1.8 kV was applied on the nanoelectrospray ionization (nESI) source with an ion transfer tube temperature of 320 °C. The data collected by DDA were acquired with a top-2 method. The MS¹ scans were performed in profile mode at 120,000 resolving power (at 200 m/z), AGC target of 2E5, 4 μscans, and in-source activation energy of 15V. The MS² scans were performed by high-energy collisional dissociation (HCD) on precursor ions with charge states 3−24, with a default charge of +5, 3 m/z isolation window, 23% normalized collision energy (NCE), and dynamic exclusion and isotope exclusion on. Dynamic exclusion was applied for precursor ions after 1 instance of fragmentation, with a duration of 60 s. The fragments were detected at 60,000 resolving power, AGC target of 2E5, maximum injection time of 800 ms, and 4 μscans. The proteoform quantitation by tMS2 involves MS² analysis on predefined targets established from pilot DDA studies on soluble and insoluble Aβ extracts. Mass spectra for the tMS2 studies were acquired at 60,000 resolving power (at 200 m/z), AGC target of 1E6, maximum injection time of 256 ms, 1 μscan, and HCD fragmentation with 23% NCE. An isolation width of 3.0 m/z with an isolation offset of 1.0 m/z was used for precursor selection from the isolation list.

Bioinformatic Identification of Aβ Proteoforms

DDA .raw files were processed using a cloud-based search on TDPortal (http://galaxy.kelleher.northwestern.edu, Code Set 3.0.2) against a Uniprot human proteome database^46,47 (Taxon ID, 9606 Proteome ID, UP000005640) with curated histones (June 2020). Two different search types were performed. The “absolute-mass” search assigned terminal fragment ions to the theoretical proteoforms with a precursor mass tolerance of 2.2 Da and a fragment ion tolerance of 10 ppm. The “biomarker” search sought to detect unknown truncations using a precursor and fragment tolerances of 10 ppm. A false-discovery rate (FDR) of 1% was used to filter out lower confidence matches at both the protein and proteoform levels. From the biomarker data sets, mass shifts that were found but not annotated with a modification or localization were manually identified and annotated with TDValidator v1.0, build 21.⁴⁸ Prosight Lite v1.4, build 1.4.8, was used to generate fragment maps and their P-Scores. Murine Aβ was manually identified and annotated but not included in any quantitative analyses.

Proteoform Relative Quantitation

An SAS input sheet for label-free quantification (LFQ) with DDA data was created during the TDPortal search without any normalization. For the tMS2 data set, extracted ion chromatograms (XICs) for the most abundant fragments were generated, and their peak areas were determined using Skyline software v21.1.0.146.⁴⁹ The cyclization of glutamate to form pyroglutamate involves a loss of a water molecules, the quantitation of which at the MS1 level could correspond to water loss at any of the positions due to dehydration or loss of a labile modification like phosphorylation since the loss of phosphoric acid results in the same parent ion as with dehydration or cyclization into pyroglutamate. Therefore, pE3−42 was identified from MS² with the b5 ion being used as a qualifier, such that the quantified proteoform was confirmed to be (Nt−H₂O) rather than other possible isobaric proteoforms. Quantitation was performed using an average of peak areas for the five most intense product ions, with confirmation of identity using all b and y ions.

Statistical Analysis

For LFQ data, the precursor ion intensities for all Aβ proteoforms, with age groups as the only factor, across both soluble and insoluble fractions were subjected to ANOVA and paired-t tests. Data with FDR < 0.05 was considered statistically significant. Z-scores for each proteoform across age groups were used to generate heat maps. A correlation matrix was generated with hierarchical clustering of signal intensities in R-Studio for each proteoform from soluble and insoluble fractions for all age groups. Correlations with p-value ≥0.05 were marked with a cross (x). To test the statistical significance of differences between proteoform groups, the average Z-scores from proteoform groups for each fraction were then subjected to ANOVA, with post hoc Tukey’s correction for multiple comparisons on GraphPad Prism v9.1.0. From tMS2-based quantitation, proteoform ratios of modified proteoforms to unmodified were also subjected to ANOVA with Bonferroni correction for multiple comparisons. For all statistical tests, adjusted p-values <0.05 were considered statistically significant.

RESULTS

Similarities in age-related Aβ-amyloidosis manifestation in the 5xFAD mouse model to that in human pathology (Figure 1A) provide a compelling rationale for examining molecular heterogeneity and abundance of the Aβ proteoform landscape. At key transition points in modeled disease progression (Figure 1B), we performed LC-MS/MS examinations (Figure 1C) of 5xFAD whole brain extracts and confirmed results reported by Bugrova et al. that both human- and mouse-derived Aβ-proteoforms can be identified in this model.⁴⁵ Due to substantial differences in sample preparation and analytical methodologies, for example, LysC-based digestion of formic acid extracted monomeric, oligomeric, and fibrillar Aβ⁴⁵ in Bugrova et al. publication, only limited comparisons of intact Aβ proteoforms can be made. However, the dual IP protocol in combination with differential ultracentrifugation employed here (Figure 1D) sought to emulate Wildburger et al. for differential mass spectrometry (dMS) of 26 Aβ proteoforms from the analogous soluble and insoluble frontal cortex brain extracts from human patients diagnosed with severe AD.¹⁵ Nt and mid-domain of Aβ were enriched per the original protocol,¹⁵ with minor variations in the use of magnetic beads and sample cleanup. Here, LC-MS/MS analysis of mouse-derived immunoprecipitates from the two ultracentrifugation fractions resulted in 33 highly confident (P-scores ≤1 × 10⁻²⁰, mass accuracy ±5 ppm) Aβ-proteoforms unique to the human sequence (Figure 2A, Supporting Figures 1, 2).

Figure 2.

(A) Mapping of immunoenriched Aβ proteoforms identified from the 5xFAD mouse brain using antibodies 6E10 and 4G8. Site-specific post-translational modifications are color-coded per legend. All proteoforms except murine Aβ1−42 refer to the human Aβ sequence. (B) Frequency of observations for start and end Aβ residue positions with respect to canonical Aβ1−42 are shown with an asterisk (*) identifying those common to the human AD brain (Wildburger et al., 2017). (C) Number of human Aβ proteoforms identified by LC/MS-MS in soluble vs insoluble fractions across ages 2, 5, 8, and >12 months old (m.o.) mice.

Aβ Proteoform Identifications by Mass Spectrometry

Twenty-five of the identified proteoforms had unique primary sequences, while eight others harbored various PTMs, including pyroglutamyl Aβ3−42 (pE3−42), 4-hydroxynonenal (4-HNE), phospho-Aβ1−99 (P-C99), glycated Aβ1−42, oxidized M35 Aβ1−42, and oxidized M35 Aβ1−40 (Figure 2A). Glycated and 4-HNE Aβ1−42, as well as P−C99 have not been identified in any studies pertinent to the 5xFAD model or human AD by mass spectrometry. Glycated Aβ1−42 was identified from robust fragmentation data that included characteristic water losses observed with MS/MS on the advanced glycation end product, i.e., nonenzymatically glycated protein (Supporting Figure 2E). A methionine sulfone at M35 was also identified along with other dioxidized proteoforms believed to derive from noncanonical mono-oxidation localized to different positions to the N-terminal side of OxM35 (Supporting Figure 2F).

Comparison of Aβ Proteoforms Identified in the Transgenic Mouse Model to Human AD

Murine Aβ1−42 was also observed in our mass spectrometric analysis; however, because the immunoprecipitation (IP) protocol included both human-specific N-terminal (Nt)-directed antibody (6E10) along with mid-domain-directed antibody (4G8), our analysis focused on the uniquely human-derived Aβ sequence. Relative to the canonical human Aβ1−42, the other 24 Aβ-proteoforms with unique primary sequences included 6 Nt truncations, 12 Ct truncations, 4 combined Nt and Ct truncations, and one Ct extension (C99). The truncation sites were localized within either the flexible and hydrophilic Nt from Aβ1 to Aβ11 (Figure 2B, top) and/or the highly hydrophobic Ct beyond Aβ21 (Figure 2B, bottom). Of note, the Nt start positions observed were the same as those characterized in AD brain tissues¹⁵ (Figure 2B, top), while a greater number of Ct truncations were observed here, with 12 of the 15 end positions being unique to the 5xFAD tissues (Figure 2B, bottom). A comparison of proteoforms identified in the two ultracentrifugation fractions showed 32 Aβ proteoforms (14 unique) in the insoluble material versus a total of 19 proteoforms (1 unique) in the soluble fractions (Figure 2C). The data also show an increase in proteoform diversification in an age-dependent manner, with 97% of the proteoforms identified in the >12-month-old (m.o.) mice (Figure 2C). The Aβ1−(34, 37, 38, 39, 40, and 42) were the only proteoforms observed at all time points regardless of solubility (Supporting Figure 3), an observation previously reported in AD tissues.^10,39

Analysis of Aβ Proteoform Expression Changes

We observed low analytical and biological variability, as substantiated with tight %CVs across the age groups for all measured proteoforms, ranging from 3 ± 2% at 2 months to 7 ± 4% at >12 months groups. The observed Aβ proteoforms were subjected to data reduction with hierarchical clustering and generation of a correlation matrix of the spectral intensities for the truncation products observed in the LFQ data sets (Figure 3A). Three groups of proteoforms with sequence variations associated with terminal truncations or extensions were readily observed with a hypothesis-free approach. Sequence attributes that commonly, but not exclusively, differentiated the groups into four subgroups comprised of Nt or Ct truncations of varying lengths relative to the Aβ1−42 sequence (Table 1, Figure 3B).

Figure 3.

(A) Hierarchical clustering of Aβ proteoforms with the correlation matrix based on expression levels across all ages and fractions; grouped proteoforms are demarcated with labels as well as boxes around groups of positively correlated proteoforms. Heatmap of Pearson’s R correlation coefficients depicted for correlations with p-values <0.05; p-values ≥0.05 are marked with a cross (X). (B) Proteoform groupings showing positions of cleavages with start and end depicted with arrow directionality. (C) Total signal from proteoforms in soluble vs insoluble fraction calculated by summating peak areas for all identified Aβ sequence variants from respective extracted ion chromatograms (XIC). (D) Average proteoform group peak areas, normalized to the total average signal from the four Aβ proteoform groups, were plotted against fraction and age.

Table 1.

Amyloid β Proteoform Grouping from Hierarchical Clustering Identifying Common Cleavage Sites

proteoform	cleavage position	residues
Aβ1–42
Aβ1a-42	ΔNt (superior)	a = 2–5, 8, 11
Aβ1–41	ΔCt (inferior)
Aβ1-w	ΔCt (proximal)	w = 22, 23
Aβ1–40
Aβ1-y	ΔCt (inferior)	y = 37–39
Afib-z	ΔNt (superior) + ΔCt (inferior)	b = 3, 4; z = 34, 38, 40
Aβ1-x	ACt (proximal)	x = 25, 28, 29, 32, 33, 34
Aβ1–99 (C99)

Effects of sex and age on Aβ proteoform expression and solubility are discussed. The LFQ data sets also showed that the total Aβ signal increased with age in both ultracentrifuge fractions (Figure 3C). Interestingly, the soluble fraction contributed to ~55% of the combined Aβ signal within the 2 m.o. mice but drastically reduced to 3−4% at 5−12 m.o. mice. A more detailed comparison of the relative contributions of the proteoform subgroups to each fraction (Figure 3D) revealed that in the insoluble fraction, group 1 proteoforms, which included proteoforms with mostly intact Ct such as Aβ1−42, became more prevalent with age, while group 2a proteoforms, which comprised Ct truncations including Aβ1−40, became less prevalent. On the other hand, in the soluble fraction, the relative distribution of the proteoform subgroups remained relatively consistent across the time points, particularly after 5 months of age. Statistical analysis of the LFQ data for all study variables revealed that the data variations observed were best explained by ultracentrifuge fraction, followed by age, while there was little contribution from sex (Supporting Figure 3). To better understand the variations observed across all study variables, the Z-scores of intact ion signal intensities (so-called “MS1”) were compared for the individual proteoforms (Figure 4A) and on a group level (Figure 4B,C). On an individual proteoform level, many of the proteoforms could not be evaluated consistently owing to missing data for lower abundance proteoforms; however, the data showed distinct differences in proteoform abundances between the different groups at different ages between the two fractions. Within the insoluble fraction, while each group showed a marked increase between the 2 m.o. and the 5−8 m.o. mice, only group 1 proteoforms continued to increase through the oldest mice. Groups 2 and 3, which included proteoforms with mostly intact Nt (e.g., Aβ1−40 and Aβ1−34, respectively), plateaued and then reduced in the older mice (Figure 4B,C). In contrast, in the soluble fraction, only those proteoforms in groups 2a and 3 showed significant changes with age. Additionally, the marked change observed between 2 m.o. and the 5−8 m.o. mice in the insoluble fraction was not observed.

Figure 4.

(A) Aβ proteoform expression differences across brain tissue fractions from 5XFAD mice at varying ages. The z-score-based heatmap was generated from label-free quantitation with MS1 signal intensities for Aβ proteoforms. The proteoform order was maintained from proteoform groupings performed in hierarchical clustering (Figure 3A). Average Z-scores are plotted for each proteoform group across each age group in insoluble and soluble fractions in panels (B) and (C), respectively. Data represent mean ± SEM with N = 6 per group (two-way ANOVA, post hoc Tukey’s adjusted p-value ≤0.0332 (*), 0.0021(**), 0.0002(***), 0.0001(****)).

Aβ Proteoform Redistribution in Age Groups

Next, we investigated the relative contributions of individual proteoforms with age in more detail (Figure 5, Supporting Figure 4). In the insoluble fraction, Aβ1−42 and Aβ1−40 constituted 78 and 14% of the total Aβ pool at 2 m.o. mice, respectively, which in 12 m.o. mice changed to 95% and <1%, respectively. In the soluble fraction, the contributions of Aβ1−42 and Aβ1−40 to the total pool remained relatively consistent with age, averaging ~65 and ~18%, respectively (right panels of Figure 5A,5B). In the insoluble Aβ pool, correlations between the individual proteoform intensities showed a substantial increase in intact Ct containing group 1 proteoforms with age. This trend in the insoluble pool of Aβ was largely associated with Aβ1−42 at the expense of Ct truncated group 2 and 3 proteoforms, including Aβ1−40, which demonstrated a negative correlation to age (left panel of Figure 5B). As indicated in the group-level comparisons above, the changes of the individual proteoform landscape within the soluble fraction were more subtle, particularly for the 5 m.o. and older mice. A modest increase in Aβ1−40 (~2.5%) at the expense of Aβ1−42 was observed between the youngest versus 5 m.o. mice. Group 3 proteoforms exhibit inconsistent trends in the soluble fraction (right panel of Figure 5C), with 1−34 and 1−99 levels reducing but other proteoforms increasing across age groups. The observed increases or decreases of individual insoluble proteoforms with age were not apparent for the soluble proteoforms, consistent with little change in the overall ratio of the total soluble versus insoluble Aβ pool (left panels of Figure 5).

Figure 5.

Relative abundance for Aβ proteoforms within insoluble and soluble brain fractions for proteoform groups 1−3 are shown in panels A−C, respectively. Proteoforms with relative levels <0.1% of the total Aβ signal were not plotted. Relative levels expressed as a percentage of the cumulative signal are shown with split y-axes to observe trends for low-level proteoforms. Pearson’s coefficient (R) is calculated and reported for −0.70 ≥ R ≥ 0.70.

Quantification of Oxidized and Nt Pyroglutamyl Aβ

Pyroglutamyl Aβ3−42 (pE3−42) and various truncation products oxidized at Met-35 (OxM35) were the most consistently observed PTMs within our LFQ data set (Supporting Figure 1). Previous reports^{14,15,50–52} indicate that differentially modified proteoforms (e.g., pE11, OxM35) could be reproducibly observed in the AD frontal cortex, albeit at different extents on different truncated products. While oxidation as a chemical modification can be controversial, given the possibility for artifactual oxidation ex-vivo during sample preparation, prior work in humans suggested that differences between the oxidation status of Aβ1−42 and Aβ1−40 may occur.⁵³ Here, our platform detected pE3−42 in mouse tissues but failed to detect pE11. Because of the potential clinical significance of these modifications,^54,55 we performed a targeted quantitative (tMS2) assay to monitor OxM35 in Aβ1−42 and Aβ1−40 as well as Nt pyroglutamate in pE3−42 (Figure 6). Notably, Ox-Aβ1−40/Aβ1−40 was consistently observed to be ~20−30% in both ultracentrifuge fractions across all age groups (Figure 6). On the other hand, at the earlier time points, the initial Ox-Aβ1−42/Aβ1−42 was observed to be ~80% in both fractions at 2 m.o. groups. In the soluble fraction, there was a small decrease in Ox-Aβ1−42/Aβ1−42 with animal age with 80% in 2 m.o. versus 58% in >12 m.o. groups. The ox-Aβ1−40 results are in stark contrast to Ox-Aβ1−42 where the reduction of Ox-Aβ1−42/Aβ1−42 was significantly greater in the insoluble fraction with increased age with 80% in 2 m.o. Vs 22% in >12 m.o. groups. In contrast, the pE3−42/Aβ3−42 increases with age in both soluble and insoluble fractions, with no detection at 2 m.o. and 10% at >12 m.o. groups in the soluble fraction, and 1% at 2 m.o. and 12% at >12 m.o. groups in the insoluble fraction (Figure 6).

Figure 6.

Aβ peak areas for modified proteoforms are expressed as a relative ratio of unmodified for oxidized (Ox) Aβ1−40 and Aβ1−42 and N-terminal pyroglutamate (pE) Aβ3−42. Data represent mean ± SEM with <i>N</i> = 6 per group (two-way ANOVA, post hoc Tukey’s adjusted p-value ≤0.0332 (*), 0.0021(**), 0.0002(***), 0.0001(****)).

DISCUSSION

The recent success shown in clinical trials of donanemab, an immunotherapy targeting the pE3 Aβ proteoforms, suggests that at least certain Aβ-proteoforms are valuable targets for curbing aspects of disease progression. However, there has been a significant history of Aβ immunotherapies showing promise in AD mouse models but then failing in human trials.^56,57 One hypothesis for this disconnect is the suitability of transgenic mouse models for the evaluation of AD therapeutics. Recently, detailed transcriptomics, electroencephalogram, in vivo imaging, biochemical characterization, and behavioral assessments on the 5XFAD model by the National Institute on Aging (NIA) Model Organism Development and Evaluation for Late-Onset Alzheimer’s Disease (MODEL-AD) consortium concluded that the model successfully recapitulates many aspects of human AD, including recommending the model for amyloidosis investigations.⁴⁰ However, it remains to be determined if a transgenic mouse model of AD that is designed to overexpress human Aβ42³⁷ can successfully recapitulate the Aβ proteoform landscape complexity that is known to exist in humans. For example, proteoforms such as Aβ4−42, Aβ4−40, and pE3−42 have been shown to exist at levels equivalent or near to canonical Aβ1−42 or Aβ1−40 in brain.^14,52 Despite the prevalence of Aβ1−42 in the 5xFAD model, shown here and previously,⁴⁵ the cross-sectional study did show an increasingly rich diversity in proteoforms, with 82% (28 out of 33) of the human Aβ proteoforms observed in the 5xFAD mouse model have also been identified in the frontal cortex of brain tissues from patients with severe AD,^15,58 suggesting that this model may indeed be appropriate for studies directed at the Aβ proteoform landscape. Other Aβ proteoforms may exist in the 5xFAD model brain that have not been identified due to limitations in the enrichment protocol or inherent differences in guanidine solubilization across fractions prior to IP.

The amyloid cascade hypothesis^59–61 has evolved over decades with the observation that soluble Aβ oligomers (AβOs) and not amyloid plaques better correlate with neurodegeneration.^19,62,63 This reinforces the need to better understand the differences of soluble monomeric, oligomeric, and protofibrillar structures versus insoluble fibrillar structures, including understanding the relationship between structures and Aβ proteoform composition in a manner not limited to Aβ1−40 and Aβ1−42.^64–66 Interestingly, our results showed that the relative ratios of the observed proteoforms in the soluble and insoluble fractions were consistent only at 2 m.o mice. In the soluble space, the relative amounts remained static across the age groups and therefore seemingly unaffected by increased production of Aβ in the transgenic model with age. Contrastingly, the insoluble fraction exhibited an increase in diversity and significant changes in relative amounts of Aβ proteoforms with age (Figure 5). The differences in dynamics of proteoforms between the soluble and insoluble fractions may provide insights into how soluble forms of Aβ (such as AβOs)³⁷ drive synaptic dysfunction starting as early as 2 m.o. (Figure 1A). Despite dynamic changes, the insoluble fraction might be less relevant to toxicity; while the relative distribution in soluble Aβ appears to be constant, their proportionate increase in abundance with age correlates with reported neurotoxicity progression in this model.²⁹ Although not studied herein, if also observed in humans, the relatively static nature of the soluble fraction could provide unique insights into therapeutic targets at early stages of neurodegeneration. By grouping proteoforms with hierarchical clustering, we can overcome the challenges presented with missing values associated with low abundance proteoforms, thereby improving consistency and preserving valuable information. Additionally, by reducing the number of variables being tested, we are able to enhance statistical testing by controlling type I errors, i.e., false positives.⁶⁷

Dynamics of Aβ Truncations in the Soluble and Insoluble Fractions

There is mounting evidence that Aβ is cleared with C-terminal truncations.⁶⁸ Aβ processing has been demonstrated in vitro to occur via sequential C-terminal truncations, where γ-secretase can cleave peptides by tri- and tetra-peptides.^69,70 While our data do not substantiate clearance by enzymatic activity per se, it can therefore be hypothesized that even in the 5xFAD mouse model, Aβ is cleared with sequential Ct truncations such that those proximal to residue 25 result in the release of Aβ peptides into the soluble space. While our data align well with prior hypotheses and experiments, the results are not sufficient to confirm the mechanistic model proposed for Aβ clearance nor do our observations determine the chronology of truncation product formation with respect to aggregate formation and dependence thereof. However, overlay of the observed cleavage sites associated with each group onto a cryo-EM-generated Aβ1−42 fibrillar structure found in sporadic and familial AD human brain samples⁷¹ does provide a reference for interpretation of what sequence level factors contribute the most to the soluble and insoluble pools (Figure 7). Within the insoluble fraction, group 1 proteoforms increase with age and generally share the characteristics of truncation of the more polar Nt residues with an intact hydrophobic Ct, including Aβ1, 2, 3, and 5−42. This contrasts those of truncated Ct containing 2a and 3 proteoform groups, which increase with age in the soluble fraction. The core of Aβ fibrils is believed to be stabilized by extended β-strands consisting of residues 12−15, 18−20, 30−33, and 38−42⁷¹ with several hydrophobic amino acids spanning these regions.^72–74 The structure is likely further stabilized by salt bridges between amino acids K28 and A42,^71,72,74,75 as well as pi-stacking.⁷⁶ The near absence of insoluble group 3 proteoforms (e.g., Aβ1−28), where cleavage occurs within the interior β-strands, may suggest limited access of proteases to these sites within aggregated states or that once proteolytically processed, the more polar product is released into a soluble state. Notable exceptions for proteoform inclusion within defined groups include Aβ1−22 and Aβ1−23, which contain aromatic rings that might stabilize aggregates with pi-stacking, making them more aggregation-prone.⁷⁶ Also, unlike other group 1 proteoforms, relative soluble and insoluble Aβ4−42 levels show a trend of increase between 2 and 5 m.o. mice and then decrease. This difference in behavior sets apart Aβ4−42, which is often observed in humans at much higher abundance in the AD brain.¹⁵

Figure 7.

Cross section of a type II Aβ1−42 fibril, observed in familial and sporadic AD brain tissues (PBD identifier 7Q4M (Yang, Arseni et al. 2022)), is depicted with two juxtaposed monomers. Amino acid residues 1−11 (not included in PDB file) contain group 1 cleavage positions Aβ cleavage sites are mapped from proteoforms identified in a mass spectrometric analysis of 5XFAD mouse brain tissues and are color-coded based on proteoform groupings from Figure 3A. Proteoforms 1−99 and those with both N- and C-terminal truncations have not been shown.

Modified Aβ and Biological Relevance

Modifications like cyclization of Nt glutamate to form pyroglutamate (pyro-Glu-3 and pyro-Glu-11), oxidation (OxM35), and racemization (isoAsp-1 and isoAsp-7), among many others, have been shown to modulate Aβ fibrilization and toxicity^77,78 and been observed previously with mass spectrometry in the AD brain.^14,15,51,52 Some of these PTMs were also detected in the 5xFAD model, often exhibiting differential partitioning between the soluble and insoluble fractions. For example, glycated Aβ1−42 and 4-HNE-adducted Aβ were most commonly observed in the insoluble fraction of the 12 m.o. mice (Supporting Figure 1). The observation of glycated Aβ1−42 could be of key importance, given that low circulating insulin levels in this animal model⁷⁹ might indicate an imbalance in glucose metabolism. If translatable to humans, glycated Aβ could hold diagnostic potential due to the established correlation between diabetes and AD.^80,81 Interestingly, while glycated Aβ1−40 is believed to modulate aggregation propensity and seeding activity in the AD brain,⁵⁰ it was not detected here. The observation of 4-hydroxynonenal (4-HNE), an oxidative stress-derived lipid peroxidation product, in the insoluble fraction is consistent with prior evidence that the modification helps accelerate protofibril formation in Aβ plaques in AD brain.⁸² The pE3−42, but not pE11−42, was observed in both the soluble and insoluble fractions, with levels increasing with age. This finding is also supported by quantitative data from LysC-digested Aβ from this animal model.⁴⁵ The formation of pyroglutamate has particular therapeutic interest, given the recent success of donanemab⁵⁵ and could be of biological significance considering its correlation with spatial memory deficits and neuronal degeneration that are observed at older ages.^37,83 Both β-secretase (BACE-1)-dependent C99 and its phosphorylated form (pC99) were observed in the soluble fraction in all age groups, which may reflect the proposed APP processing mechanism, where C99 is expected to be embedded in the cell membrane until cleavage by γ secretase,⁶⁹ indicating that they are less likely to be part of extensive aggregates.

Finally, the role of oxidative stress in AD is of high significance as it can modulate fibril formation,^84–86 as well as neurotoxicity.⁸⁷ Oxidation at methionine 35^88,89 may be induced in response to the inflammatory milieu in AD and has been described to prevent the formation of protofibrils.⁸⁴ Our data showed striking differences in Ox-Met-35 between Aβ1−40 and Aβ1−42 relative to the solubility and age. This may result from differences in the native conformations adopted by the two molecules, which may in part be affected by the presence of different metals^71,84 that may promote or inhibit the rapid sequestration of freshly synthesized Aβ1−42 into aggregates.⁸⁴ However, we are currently unable to conclusively preclude artifactual origins of oxidation, the discernment of which might require use of antioxidants or oxygen-deficient preparation area, for example, a glovebox.

CONCLUSIONS

We employed quantitative top-down proteomics to characterize the Aβ proteoform landscape in the 5xFAD mouse model, correlating proteoform abundances to key transition points across disease progression and their solubility. Changes in Aβ proteoform levels and their pathogenic contribution over the course of AD remain a knowledge gap and highlight the potential of using the 5xFAD model to study age-related Aβ-amyloidosis manifestation in modeled disease progression (Figure 1B). Solubility of Aβ proteoforms in the brain seems to be driven by not only the inherent physicochemical properties of the peptides themselves but also an interplay of biochemical processes involved. While our data do not establish cause and effect, it can be readily deployed on AD patient cohorts to provide a more holistic picture of how Aβ proteoforms change in the AD brain to improve diagnostics and serve as a powerful tool for comparing and validating AD models. By grouping proteoforms with unsupervised hierarchical clustering into proteoform groups, we have established a systematic approach to enhance the statistical significance of observations as well as potentially identifying new mechanisms underlying amyloidosis progression on a molecular level.

ABBREVIATIONS

AD

Alzheimer’s disease

APP

amyloid precursor protein

BACE1

β-site amyloid precursor protein cleaving enzyme

PSEN1

presenilin 1

Aβ

amyloid β

CSF

cerebrospinal fluid

IP

immunoprecipitation

MS

mass spectrometry

LC

liquid chromatography

TDMS

top-down mass spectrometry

DDA

data-dependent acquisition

tMS2

targeted MS2

nLC

nano-liquid chromatography

Data Availability Statement

The mass spectrometry proteomics data have been deposited to the PRIDE Archive (http://www.ebi.ac.uk/pride/archive/) via the PRIDE partner repository with the data set identifier PXD042921.