Mass-Spectrometry based GEE Footprinting Characterizes Kinetic Mechanisms and Sites of Conformational Change in Amyloid Beta 1–42 Aggregation

Abstract

Understanding the dynamics of Aβ aggregation is critical for elucidating Alzheimer’s disease (AD) progression. This study extends our previous work on Aβ42 using fast photochemical oxidation of proteins (FPOP) and pulsed hydrogen/deuterium exchange and introduces mass spectrometry (MS)-based glycine ethyl ester (GEE) footprinting, combined with kinetic modeling, to characterize Aβ42 conformational changes and elucidate polymer populations along its aggregation pathways. We investigated Aβ42 conformational changes by analyzing three distinct peptide regions generated by Lys-N digestion, revealing three different views of aggregation behaviors. The middle and C-terminal regions are identified as primary aggregation sites; in contrast, the N-terminal peptide exhibited only minor changes in GEE modification, supporting its limited involvement in intermolecular interactions during aggregation. Amino-acid-level analysis provided higher spatial resolution: D1 underwent relatively constant footprinting throughout aggregation, whereas E3/D7, E22, and D23 showed more substantial decreases in modification, underscoring their critical roles in aggregation. By integrating these findings with kinetic modeling, we identified four predominant polymeric populations involved in Aβ1–42 aggregation. This study reports, for the first time, a stable, specific, and slow chemical footprinting approach to characterizing Aβ1–42 aggregation, offering new insights into Aβ1–42 polymerization dynamics and enhancing our understanding of its role in AD pathology. The solvent accessibilities features of the six acidic amino acids and the C-terminus found with footprinting agree with those observed in the final, fibril state structure of Aβ42.

Graphical Abstract

INTRODUCTION

Alzheimer’s disease, a prevalent neurodegenerative disorder, is marked by progressive cognitive decline, memory loss, and ultimately independent-function loss. A prevailing theory in Alzheimer’s pathology is the Amyloid Cascade Hypothesis^1–4, positing that the accumulation of amyloid-beta (Aβ) polymers in the brain initiates a cascade of events leading to neurodegeneration and cognitive impairment. Supporting this hypothesis is the approval of recent therapeutic interventions showing modest efficacy of FDA-approved Aβ targeted monoclonal antibodies, Aducanumab, Lecanemab, and Donanemab^{5, 6}. Aducanumab binds to aggregated Aβ, promoting its clearance from the brain, and potentially slowing disease progression^{7, 8}. Similarly, Lecanemab⁹ and Donanemab¹⁰ function through mechanisms aimed at reducing Aβ plaques in the brain.

Despite challenges in clinical trials and ongoing debates regarding efficacy of drugs, research on Aβ and its role in Alzheimer’s pathology is needed. Elucidating Aβ aggregation dynamics is important for understanding Alzheimer’s disease and developing effective therapies. This is particularly important because there is good evidence that soluble oligomers are the toxic populations¹¹, and they may constitute early causes of the disease. If so, understanding the aggregation from monomer to small and large soluble oligomers is of high relevance.

Other biophysical tools (e.g., Circular Dichroism (CD), Nuclear Magnetic Resonance (NMR), X-ray Crystallography, and cryogenic electron microscopy (cryo-EM)) have been used to investigate Aβ’s structure and its polymerization kinetics. For example, CD was applied to characterize Aβ peptide secondary structures^{12, 13}, yet it is constrained by its inability to provide high-resolution structural information (i.e., to identify the regions of Aβ that undergo structural change) and to distinguish subtle conformational changes. Solid-state NMR, X-ray crystallography, and Cryo-EM provide atomic-resolution views of insoluble fibrils^14–19, affording a structural characterization of the Aβ42’s final fibril structures. These latter tools, however, are challenged to afford structural information on the mid-sized MW oligomers that form as heterogeneous intermediates in the aggregation²⁰. Moreover, the methods are unable to characterize Aβ42 in complex milieu (e.g., live cells) and follow protein dynamics along their several possible aggregation pathways. Although each of these biophysical tools exhibits unique strengths, they also have serious limitations in elucidating Aβ42’s structure and aggregation mechanism.

Similarly, widely utilized techniques for investigating Aβ42, such as atomic force microscopy (AFM), transmission electron microscopy (TEM), and electron paramagnetic resonance (EPR), face limitations as discussed above.

Mass spectrometry-based footprinting is now playing a role in protein higher order structure (HOS) characterization²⁰, allowing informative analysis via differential measurements of protein structures and interactions at the molecular level by mapping changes in solvent-accessible surface areas (SASA). This field is often referred to as “structural proteomics”. Our goal is to develop and apply methodologies to gain insights on important biomedical problems such as amyloid formation. Such MS-based approaches can allow real-time monitoring of Aβ aggregation and promise to identify novel binding sites of drugs or inhibitors of aggregation on Aβ polymers, thereby supporting the development of effective therapeutics.

In our first work on Aβ aggregation, we applied MS-based pulsed hydrogen-deuterium exchange (HDX) to test the influence of temperature and presence of Cu²⁺ on the aggregation²¹. We then compared the aggregation rates across Aβ42 mutants²². Our findings show a significant difference between Aβ42 and Aβ40 in aggregation and suggest a self-catalyzed aggregation mechanism and highlight the significant role of Aβ’s central region in controlling the aggregation²¹. Owing to the need to limit the time of proteolysis and thereby minimize HDX back-exchange, the extent of the proteolysis to gain peptide-level information was never 100% and varied during the aggregation time as aggregates became larger and more stable^{20, 21}.

In 2016, we utilized fast hydroxyl-radical footprinting with a pulsed laser, scavenger, and a flow system (termed Fast Photochemical Oxidation of Proteins or FPOP) in combination with analysis by MS proteomics methods to follow Aβ42 aggregation²³. Optimized LysN digestion gave effective digestion of the polymers, and the FPOP results provided a new perspective of the polymerization, especially during the early stages, providing a nearly unbiassed view. FPOP, however, faces challenged as a routine method owing to the requirement for specialized laser equipment. Furthermore, the high reactivity with methionine and aromatic ring-containing residues, diminishes the coverage of less reactive residues.

Therefore, we wish to explore applications of robust but slow chemical footprinting that complements FPOP and builds upon the optimizations and insights gained from previous approaches. We are seeking a fuller perspective that offers different spatial resolution on Aβ conformational changes than provided by •OH and HDX footprinting on the dynamics driving the polymerization. We realized that slow, specific-amino-acid footprinting (sometimes called “covalent labeling”) may offer advantages particularly by not requiring a laser. Previous studies demonstrated the effectiveness of slow footprinting in other amyloid aggregation characterizations; Vachet group^{24, 25} utilized diethylpyrocarbonate (DEPC) footprinting and structurally characterized β−2-microglobulin (β2m) dimer and tetramer interfaces. Valentine²⁶ applied DEPC footprinting coupled with ion mobility spectrometry–mass spectrometry in the investigation of N-terminal 17-residue segment (Nt17) oligomerization.

Because slow footprinting has high specificity and, consequently, limited coverage, we plan to repeat the footprinting with several reagents (e.g., GEE, DEPC, benzoyl fluoride). This study, however, is a necessary precedent for those studies being the first example of slow footprinting applied to the Aβ aggregation field.

Here, we employed mass spectrometry-based glycine ethyl ether (GEE) footprinting to follow Aβ42 aggregation. The GEE footprinting method labels solvent-accessible carboxylic groups (Asp D, Glu E and the C terminus) and was previously utilized to study protein structures and protein binding interactions owing to its robustness and versatility (Figure S1)^27–31. We chose GEE footprinting because Aβ42 contains 7 Asp and Glu residues (including the C terminus) (Figure S2), providing an opportunity to cover nearly 20% of the Aβ42 sequence.

Following introduction of GEE footprinting by Akashi et al.³² in 1993, we implemented it to determine the orientation of the FMO protein in photosynthetic membranes³³. We then employed it to compare ApoE structures across different isoforms³⁴, characterized the binding interactions between calmodulin and zinc²⁷, and integrated it into an automated footprinting platform³⁵, demonstrating some effectiveness in protein structural studies. We expect new perspectives will arise from GEE footprinting because the initial reacting facilitator is larger in size and subject to more steric constraints than •OH and exchange with D₂O.

RESULTS

Aggregation of Aβ1–42 followed by ThT fluorescence: Thioflavin T (ThT) assay is a widely utilized fluorescence-based technique employed to monitor protein aggregation^{36, 37}, particularly for amyloid fibril formation. It exploits thioflavin’s ability to bind selectively to β-sheet-rich structures, thereby facilitating fluorescence and providing real-time detection and quantification of protein aggregation kinetics. As a reference point for our investigation of Aβ42, we used the thioflavin T assay and observed the expected sigmoidal curve (Figure 1) that delineates three distinct phases along the aggregation pathway: (1) initial lag, (2) elongation, and (3) saturation phases. This sigmoidal curve represents the kinetics of Aβ42 aggregation, highlighting nucleation as the rate-limiting step. The ThT assay in this study served as a critical “quality control” experiment post Aβ1–42 monomerization to provide a benchmark for comparing the outcome of Aβ1–42 aggregation reported by slow chemical footprinting.

Figure 1.

Aβ42 Kinetics Characterized by our Thioflavin T assay. Aβ1–42 aggregation occurs in three phases: (i) lag, which is thermodynamically unfavorable and is the rate-limiting step, (ii) elongation, which is more favorable and occurs quickly, and (iii) saturation phase.

GEE analysis of Aβ1–42 Aggregation at the Peptide Levels.

Our footprinting study focused on two key regions near salt bridges in the aggregation: the middle domain region (containing reactive acidic amino acids 22 and 23) and the N-terminus region (containing reactive acidic amino acids 1, 3, 7, 11). These regions are allegedly responsible for vital electrostatic intramolecular and intermolecular interactions^{16, 38}. Mutation studies^39–44 involving substitution of these residues alters the local electrostatic repulsions, modifies the overall changes of the protein, and significantly impacts the aggregation rates, underscoring the importance of these regions in influencing the structural stability and aggregation propensity of Aβ1–42.

As the aggregation of Aβ42 is a time-dependent phenomenon, our study employed pulsed irreversible footprinting during aggregation to capture “snapshots” of the changing solvent accessibility of polymerization intermediates in real time. In our triplicate experiments, we first monomerized the synthetic Aβ1–42 peptide by treating it with trifluoracetic acid (TFA) and hexafluoro-2-propanol (HFIP), a standard treatment used in many previous studies²³. Prior to inducing aggregation, we subjected the monomerized Aβ1–42 to sodium hydroxide (NaOH) treatment. Formation of Aβ1–42 aggregates was initiated by diluting Aβ1–42 monomer preparation in PBS buffer (pH 7.4) and allowing Aβ to aggregate for various times from 0 s up to 30 h. We initiated the footprinting of the protein at these several time points by pulsed introduction of labeling reagents during the incubation. The amounts of footprinting reagent GEE and EDC were optimized prior to the footprinting to achieve nearly complete GEE and EDC modification instead of zero-length crosslinking (Figure S3). Quenching was achieved by adding excess ammonium acetate. Intact protein MS analysis gave representative mass spectra of Aβ1–42 labeled by GEE reagents (Figure S4). The footprinting occurs at solvent-accessible Aβ1–42 acidic side chains and at the C-terminus to give a mixture of labeled proteins.

We characterized the conformational changes in aggregation for three distinct regions represented by the three peptides resulting from Lys-N digestion (N-terminal region 1–15, middle domain 16–27, C-terminal region 28–42), as was done previously for FPOP footprinting²³. The fraction modified of each digested peptide was calculated from the ratio of the peak area of the modified peptide (P_mod) to that of the total peak area of the modified and unmodified peptides (P) (eq 1), as described in previous studies. For the footprinting fraction calculation, both the GEE modification and EDC modification were taken into consideration.

$$\%\text{modified}=\frac{\Sigma P_{\mathit{\text{mod}}}}{\Sigma P_{\mathit{\text{mod}}}+\Sigma P}\times 100.$$ (eq 1)

Owing to the intrinsically disordered nature of Aβ1–42², the unfolded monomer exhibits larger solvent accessibility and undergoes more modification (Figure 2) at early aggregation times. As time increases and Aβ1–42 gradually aggregates into polymers, the structure becomes more compact and ordered, and some of the side chains become more buried. Aβ1–42 loses solvent accessibility, gains protection, and consequently, undergoes correspondingly decreased GEE modification (Figure 2a–c) with time. Eventually, as the Aβ1–42 aggregates form mature aggregates, the aggregated protein becomes highly resistant to modification, as reflected as the end plateau of the plots. This change in footprinting extents reflects the conformational transformation during the aggregation, consistent with a structural transition from more exposed, more reactive oligomers to compact, protected mature aggregates including large oligomers and fibrils.

Figure 2.

GEE labeling results and kinetic simulations for Lys-N digested Aβ1–42 peptides. Aggregation curves of (a) N-terminal region 1–15, (b) middle region 16–27, and (c) C-terminal region 28–42. Points represent experimental data (10 μM, 25 °C, pH 7.4, no agitation, in triplicate) All solid-line curves are simulations that afford rates constants for each peptide treated independently. Different stages (a-g) for each peptide are labeled.

By analyzing the GEE footprinting data at the peptide level, we observed similar trends but different details in the behaviors of these three peptides comprising Aβ1–42. Significant decreases in modification level occurred for the middle peptide (60% decrease, p = 7.1E-4, two-tail heteroscedastic t-test) and for the C-terminal peptide (78% decrease, p = 1.1E-4, two-tail heteroscedastic t-test) (Figure S5) through the aggregation window, indicating these regions contain major aggregation interfaces and become protected as the soluble oligomers form. On the other hand, the N-terminal peptide displays a smaller extent of decrease in GEE modification (32% decrease, p = 3.9E-3, two-tail heteroscedastic t-test), confirming that the N-terminal peptide is less involved in the intermolecular interactions than are the middle and C-terminal peptide regions, as was also shown in our FPOP study²³.

Looking more closely at the beginning of the curves (Figure 3, points), we see that the N terminal peptide 1–15 showed an overall constant extent of GEE modification during the first 2 h of aggregation (Figure 3a), whereas the peptide 16–27, representing the middle region, and peptide 28–42, representing the C terminal region 28–42, underwent more rapid decreases in modification extents as aggregation began (Figure 3b and 3c). This detail is not revealed by the ThT assay. The GEE results indicate that the N-terminal region, different from middle and C-terminal regions, starts to participate in polymerization from the start of aggregation, maintains more solvent accessibility, and has little involvement in the aggregation for the first 120 min, consistent with solid-state NMR results^{18, 19}. Ultimately, this region showed an overall decrease in modification extent.

Figure 3.

GEE labeling results and kinetic simulations for Lys-N digested Aβ1–42 peptides. Zoomed-in aggregation curves for 0–225 min: (a) N terminal region 1–15 shows little involvement in the first 2 h of aggregation, (b) middle domain 16–27 and (c) C terminal region 28–42 show considerable participation from the beginning of the aggregation.

The results align with those from a previous FPOP study in our laboratory²³. In that study²³ we used hydroxyl radical footprinting on a FPOP platform to characterize Aβ1–42 aggregation. The results showed that the terminal region showed minimal change with aggregation, whereas the middle and C terminal regions underwent 6 and 2.5 times decreases in the modification extent, respectively. In this study, we expect that the reagents (GEE and EDC) footprint Aβ42 differently from hydroxyl radicals because they have larger sizes. Owing to steric size, GEE/EDC is expected to be more discriminative of Aβ1–42’s conformational changes.

GEE analysis of Aβ1–42 at the residue level.

To achieve an increased spatial resolution of footprinting, we determined the extent of aggregation for some amino acid residues and employed a combined proteolysis approach using Lys-N and chymotrypsin digestions. The same footprinted samples were aliquoted into two Eppendorf vials. One aliquot was treated with Lys-N protease, and the other one was treated simultaneously with chymotrypsin protease. The extent of GEE modifications at the residue level was determined by using the equation previously discussed (eq 1). The measurement involved determining the modification extent by analyzing the peak area ratio of peptides containing the footprinted residues to the total peak area of all peptides that included those residues²³.

Notably, chromatographic resolution was sufficient to distinguish modifications at D1, E11, E22, E23, and the C terminus from each other, providing clear and measurable extents of GEE modification at these sites (from extracted ion chromatograms). For the residues E3 and D7, however, there are two isomers, and the chromatogram edges are overlapped (Figure S9). Although deconvolution could result in an approximate quantification of each residue (E3 ~24% at 0 s and D7 ~16% at 0 s), to ensure a more accurate measurement to be used in the kinetic modeling, the composite, somewhat overlapped extracted chromatographic signals for peptides containing modified E3 and D7 were integrated in total.

When we investigated the behavior of N-terminal residues, we observed the modification extent differs at 0 s among D1 (5%), E11 (7%), and E3/D7 (40% integrated together) (Figure S5a), showing different solvent accessibilities and microenvironments for these residues. As Aβ42 aggregates, the N-terminal residue D1 showed an overall constant extent of GEE modification during 0–400 min of aggregation (Figure 4a and S5b), indicating that the region containing D1 undergoes few conformational changes and remains solvent accessible throughout the first 400 min. Interestingly, D1 showed a 11% decrease from 400 min to 628 min, and then remained constant until the end of the aggregation. These results indicate that D1 starts to undergo a small conformational change after 400 min of aggregation to gain protection but overall remains solvent accessible. We conclude D1 is not significantly involved in the aggregation.

Figure 4.

GEE labeling results and kinetic fits for Aβ1–42 N-terminal amino acids. (a) D1 shows higher flexibility during the aggregation, (b) E3 and D7 and (c) E11 show large changes of solvent accessibility accompanying aggregation, (d) and (e) the Cryo-EM structure (PDB: 5OQV), indicating salt-bridge formation between E11, H13 and H6, D7 and R5. (GEE modifications on E3 and D7 were summed to improve accuracy in quantification. Figure S9).

In contrast, E3/D7 (50% decrease) and E11 (43% decrease) (Figure 4b, 4c and S5), participated from the beginning in the aggregation and overall underwent larger decreases in the modification extent than did D1 (29% decrease) along the whole aggregation window. These three residues represent regions that participate more strongly than the N terminus in the aggregation process. This interpretation seems at odds with our previous work²³, who observed nearly constant FPOP modification extent of H6 and H13 and concluded that the residues maintain solvent exposure during aggregation. This “disagreement” can be resolved by examining the Cryo-EM structure (PDB:5OQV) of Gremer et al.¹⁶, who built a high-quality, de novo atomic model of Aβ fibrils (Figure 4d and 4e). This fibril structure shows salt-bridge formation between D7 and R5, E11 and H6 with H13 (Figure 4e). This finding indicates that, at the end of aggregation, the N terminal region undergoes conformation changes, resulting in an “L” shape topology that favors electrostatic interactions between D7 and R5, and H6, E11, and H13. Because GEE/EDC reacts with the carboxylic group of D7 and E13, the reaction was less favorable given the stabilization that results from salt-bridge formation of these residues. For FPOP, the hydroxyl radicals have smaller size and react at the carbon atom on the imidazole of histidine via a footprinting reaction that is likely less influenced by such NH-CO interaction. Although GEE footprinting only probes acidic residues, the footprinting results, when aligned with this Cryo-EM structure, suggest that the soluble oligomers formed during in vitro aggregation are adopting conformations similar to those of the final fibrils. The observed agreement indicates that the in-vitro solution-phase aggregation proceeds through intermediates similar in structure to solid-state fibrils.

Examining the residues within the middle domain and C-terminal region (Figure 5), we find notable decreases in GEE modification levels: E22 (Figure 5a) exhibited a 72% decrease (p = 6.8E-4), D23 (Figure 5b) showed a 67% decrease (p = E-5), and the C-terminal COOH (Figure 5c) displayed a substantial 78% decrease (p = 1E-4) (Figure S5) through the aggregation. (We note that the modification extent for the C-terminus is measured in terms of the signal intensities corresponding to peptide 28–42, modified vs. unmodified, and that the ultimate C-terminus is the sole residue footprinted by GEE in that peptide region.) Interestingly, despite E22 and D23 being in proximity, E22 showed a larger solvent accessibility (24% modification extent at 0 s) compared to D23 (13% modification extent at 0 s) (Figure S5). Unlike D23, which gained protection upon the onset of the aggregation, E22 exhibited a small initial lag phase in GEE modification within the first 60 min before gaining protection and undergoing a rapid decrease in footprinting (Figure 5a and S5c). These findings suggest that E22 serves as part of a critical nucleation interface that becomes more solvent-exposed during the early stages of polymerization but quickly loses solvent accessibility as aggregation progresses. This dynamic behavior implies a potential pivotal role for E22 in initiating and driving the aggregation process. Our results are in accord with other Aβ1–42 molecular dynamic studies^45–47, underscoring the significant role of the middle region in the polymerization.

Figure 5.

GEE labeling results and kinetic fits for E22 (a), D23 (b) and C terminus (c). E22, D23 and C termini show similar overall sigmoidal characteristics. The kinetic fitting curves suggest E22 has a small lag phase in gaining protection at the beginning (a) and then showing decrease; D23 (b) and C terminus (c) show decreases from the beginning of the aggregation.

To support our analysis above, we performed two-tailed heteroscedastic t-tests on our experimental results. We report differences only when the p-value is greater than 0.05; otherwise, we state that the modification extent remains constant or undergoes a lag phase (see examples in Figure S5).

Kinetic Modeling of Footprinting Results.

We recognize that “footprints” only of the aggregating ensemble produce a low-resolution picture. Thus, we augmented our strategy to include kinetic modeling of the results to inform on the number of populations of the aggregates. We chose to model the Aβ1–42 aggregation kinetics based on Cohen’s^{48, 49} nucleation-elongation-fragmentation polymerization equations with an adaptation to replace the fragmentation step with secondary nucleation. This modeling incorporates kinetic equations involving two principal components: Aβ1–42 subunit (mass) concentration (M) and polymer number concentration (P). The aggregation process is driven by a series of microscopic reactions (Figure 6a). The initial step is primary nucleation, which follows second order kinetics. In the primary nucleation, soluble monomers aggregate into polymer nuclei that ultimately have a critical length n_c. This is followed by irreversible conversion reactions from one polymer population to the next. As populations transition from one to another, elongation reactions occur in which the polymers just formed incorporate one additional monomer on their end. The elongation reaction is also accompanied by a dissociation reaction, in which the elongated polymer fragments into a smaller polymer and a monomer. Additionally, there is monomer-dependent secondary nucleation; this step is described as soluble monomer aggregation on a surface nucleus of an already formed polymer.

Figure 6. Scheme of the proposed aggregation mechanism.

(a) The aggregation reactions of M1/P1 population: The initial step is the secondary ordered primary nucleation that is followed by the reversible elongation/dissociation and the irreversible secondary nucleation. In reversible elongation reactions, a monomer associates or dissociates from the end of the oligomer. The secondary nucleation is monomer-dependent when a new nucleus forms at the surface of a preformed oligomer. (b) The reversible elongation reactions and the secondary nucleation of population M2/P2, that converts into population M3/P3 and M4/P4 as aggregations progress.

The rate constants for these reactions are treated as adjustable parameters to fit the model-derived peptide/amino acid signals to the experimental data. The best fit of the experimental results provides a specific set of rate constants that define aggregation and enables the construction of a fractional population plot showing the number of populations of the polymers and their time dependences. The rate constants for each peptide region are fitted to enhance the accuracy of the fitting.

The analysis reveals that the best fit to the aggregation involves four new populations denoted as M1, M2, M3, and M4 along with the free monomer population m. The population trends illustrate the kinetic scheme (Figure 6). The smaller sized polymer populations convert into larger sized polymer populations as aggregation progresses (Figure 6b), accompanied by spontaneous conformational and chemical changes.

The kinetic profile of each peptide shows that at least different stages characterize the aggregation process (Figure 2): (1) Initially for stage a-b, there is a rapid, exponential-like decrease, corresponding to the rapid polymerization of Aβ1–42 monomers into small oligomers in the model population M1. As these monomers assemble, their solvent accessibility decreases, leading to decreased side-chain reactivity in the GEE/EDC footprinting. During this period, the early-formed, small oligomers undergo the first model unimolecular conversion reaction for subsequent growth in population M2. Toward the end of this period, a rapid decrease in the mass concentration of M1 is accompanied by a rapid increase in the mass population of M2, providing additional protection as apparent in the experimental signals of peptides 16–27 and 28–42. Athough the N-terminal peptide 1–15 exhibits constant modification extent during the a-b period, as discussed earlier, the kinetic fit demonstrates the region gains protection slowly (Figure 2a, solid line). (2) The rapid decrease in the experimental signal is followed by a plateau (stage b-c), in which the model suggests that M2 polymers form nuclei that serve as “seeds” for growth in population M3. The GEE modification extent does not significantly change owing to the initial slow loss of mass concentration M2 accompanied by a slow gain in mass concentration gain M3, as the modeling suggests. (3) Subsequently, as the dissociation of M2 and elongation of M3 accelerates, the peptide signals drop dramatically at stage c-d, and the monomers continue to be consumed during this process, as reflected as a rapid decrease (stage c-d) in peptide modification level. (4) An invariant stage (d-e) follows stage (3), for which the model suggests that M3 polymers form nuclei that serve as “seeds” for growth of population M4. This stage is accompanied by the slow decrease in the mass concentration of population M3 and the slow increase in the mass concentration of M4. (5) At stage e-f, this process accelerates with formation of large polymers/protofibrils (as suggested by the modeling) that rapidly assemble into mature fibrils M4. Eventually (6) the modification extents remain relatively constant from stage f to g, indicating establishment of equilibrium between population m and M4 (Figure 7). Notably, the peptide region 28–42 is the first to reach stage e-f (Figure 2c), suggesting this region 28–42 plays the important role in mature-fibril formation. All the peptide and amino acid signals allow us to construct mass concentration profiles of various populations over time during the aggregation²² (Figure 7). The results show that four populations of polymers form during the aggregation, each giving way to another in a consecutive manner.

Figure 7.

Mass concentration plots from kinetic fitting to all peptide and residue populations fraction data vs aggregation time. The mass concentration of each modeled population is color-coded as follows: monomer m*: red, M1: blue, M2: green, M3: orange, M4: cyan. GEE labeling results and kinetic fitting of peptide 16–28 is incorporated for better comparison (black curve). Different stages (a-g) are labeled.

An unusual feature is seen from the fitting: small increases in labeling occur for residue D1 during the initial rapidly decreasing stage (a-b) (Figure 4a) and residue E11 (Figure 4c), specifically around 80 min, during which the mass concentration plots (Figure 7) characterize the rapid transition between M1 and M2, suggesting the decrease in monomer slows down while M2 concentration dramatically increases. The rapid transition between populations with different conformations and chemical properties may account for this observation. This feature, however, was not observed for all three peptides and residues E22, D23 and E3&D7. We note that this observation is a local phenonmenon instead of a global effect and suggest the fitting will be more definite with more data points.

Besides the 4-population model, we also tried a 3-population model (the fitting results are shown in Figure S6). Obviously, the results are less satisfactory than the 4-population model discussed above and show that the modeling in sensitive to the results.

Interestingly, unlike the results of the ThT fluorescence assay (Figure 1), which only captures a single sigmoidal phase in the time span from 0 min to the end of the aggregation, our footprinting outcome, coupled with kinetic modeling, reveals three distinct sigmoidal phases. The first phase occurs a-b, the second c-d, and the third e-f (Figure 2, solid lines). The sigmoids revealed by footprinting enable us to characterize early changes in populations such as unimolecular conversions between monomer m* and polymer populations M1, M2, M3 and M4, which are not detected by ThT fluorescence.

There are differences in the aggregation conditions between the ThT assay and the footprinting. Specifically, stirring was required in the ThT assay, accelerating the aggregation, while aggregation in the footprinting occurred without agitation. Despite these differences, the footprinting provides a significant advantage by characterizing early polymerization populations (sigmoids a-b and c-d) that occur prior to ThT fluorescence and are invisible to it. The MS-based footprinting provides a deeper insight into the early stages of aggregation where the important and disease-relevant soluble oligomers are forming.

DISCUSSION

How do these results compare with those from other approaches? Other investigators employed different biophysical tools and kinetic models aiming to delineate the Aβ42 oligomer intermediates along aggregation pathways. Bernstein et al.⁵⁰ utilized ion-mobility coupled with mass spectrometry (IM-MS) for the characterization of small, early oligomer states. Their findings indicate the involvement of tetramers and dodecamers in the initial stages of Aβ42 oligomerization, which is different than Aβ40 oligomerization, in which tetramer intermediates form and resist further monomer addition. Their arrival time distributions (ATDs) and cross-section calculations are evidence that the tetramer population promotes the formation of a planar hexamer paranucleus. These hexameric paranuclei are proposed to associate subsequently into stacked paranuclear dodecamers ((Aβ42)₆)₂, that serve as nucleation sites for the further oligomerization and elongation. Our results, however, are not informative on the distribution of these small oligomers.

Another prior study is high resolution atomic force microscopy (AFM), as determined by Economou et al.⁵¹ Their high-resolution AFM results recapitulated the oligomer populations occurring at the early aggregation. Specifically, hexamers and dodecamers became the dominant populations as early as 5 min. After 10 min, the dodecamers seeded the assembly of linear preprotofibrils, which associated into larger globular structures over time.

Our lab (Li et al.²³), discussed above, modeled FPOP footprinting results with the Finke-Watzky 2-step nucleation mechanism^{23, 52, 53}. The model considers two separate steps driving the aggregation: a slow continuous nucleation followed by fast autocatalytic elongations. We²³. suggested that a critical dimerization transition stage occurs early in the aggregation process, followed by a series of nucleation and autocatalytic reactions. our aggregation kinetic studies characterized four major populations that were assigned as monomer M, perinuclei D, protofibrils D* and fibrils D**. Lower MW populations deplete as they convert into higher MW populations through nucleation and autocatalytic growth reactions. Eventually an equilibrium is reached between M and D**. Notably, Li et al.²³ also commented that the Finke-Watzky mechanism has limitations because the equilibria between monomer M and other populations (D and D*) were not taken into consideration. These studies highlight diversity in experimental approaches used to characterize Aβ42 aggregation, each offering unique insights into different stages of the pathway. Although IM-MS and AFM have identified early oligomeric intermediates, and FPOP footprinting has provided residue-specific protection data, our use of GEE footprinting adds another dimension to understanding Aβ42 structural transitions.

We believe that employing multiple footprinting techniques not only provides complementary perspectives on aggregation dynamics but also achieve higher coverage of residues. Although both GEE footprinting and FPOP capture multiple sigmoidal phases in their kinetic profiles, GEE labeling offers unique features due to its larger reagent size and selective reactivity toward carboxyl groups (D, E, and the C-terminus) not covered by FPOP.

Due to its larger molecular size, the GEE/EDC reagent exhibits greater sensitivity to steric hindrance compared to hydroxyl radicals (FPOP). This is reflected in our aggregation curves, where GEE modifications reach the plateau phase at approximately 1400 min, beyond which no further protection is seen, likely due to restricted access to buried regions. In contrast, the smaller hydroxyl radicals are able to penetrate these regions, allowing continued characterization.

GEE has a limitation on amino acid coverage, selectively modifying carboxyl-containing residues (D, E, and the C-terminus), whereas hydroxyl radicals (FPOP) react with a broader range of side chains. This reactivity selectivity allows each method to probe different aspects of the Aβ42 microenvironment. For example, GEE labeling is particularly insightful on detecting potential salt bridge interactions involving D7, E11, and adjacent histidines, which may not be as readily captured by hydroxyl radical footprinting. Another practical advantage of GEE is that it does not require a laser-based radical generation system, simplifying experimental setup and reducing potential sample oxidation artifacts. A possible drawback of GEE, however, is its slower footprinting time, requiring minutes compared to microsec-scale footprinting of FPOP.

Illes-Toth et al.^{39, 48, 49} adopted the principle of Cohen’s nucleation-elongation-fragmentation mechanism to fit their pulsed-HDX results and characterized an exponential-like increase of fibril mass along the aggregation pathway. Instead of only considering the equilibrium between monomer and the mature fibrils, the nucleation-elongation-fragmentation mechanism accounts comprehensively for the equilibrium between monomer populations and all other oligomer populations. This and the other above studies provide some understanding of Aβ42 kinetic dynamics and enable proposals of potential prominent intermediates along the aggregation pathway.

Our kinetic modeling results, consistent with those of other studies, support a picture having a few dominant polymer populations driving the Aβ1–42 aggregation. By considering the dynamics between all polymer populations, we contend there are four major populations (Figure 7, M1, M2, M3 and M4). Similar to the findings of Bernstein et al.⁵⁰ and Economou et al.⁵¹, the smaller oligomer population M1 first emerges as dominant population, with depletion of monomer m, and seeds the assembly of paranuclei, and subsequently transforms into a larger oligomer population M2. Over time, M2 evolves into M3. Eventually, the M1, M2 and M3 populations deplete to form presumably mature fibrillar aggregates M4. An equilibrium is achieved between fibrils M4 and leftover monomers m, marked by no further changes in mass concentrations within the timescale of our experiments. We emphasize that our results do not provide, however, the structures of each dominant populations.

CONCLUSIONS

We did not apply GEE footprinting to Aβ1–40 because our previous studies using HDX²¹ and FPOP²³ showed no footprinting changes during the early stages of Aβ1–40 polymerization. Given these findings, we reasonably expected GEE to yield similar results, making it unlikely to provide additional insights into Aβ1–40 aggregation within the experimental timeframe of this study. The significant differences between Aβ40 and Aβ42 observed in footprinting^{21, 23} may reflect their large difference in their toxicity in Alzheimer’s disease.

Our study reports for the first time the implementation of stable, specific, and slow irreversible chemical footprinting to characterize Aβ1–42 aggregation. We integrated kinetic modeling with the results to elucidate the dynamics driving the aggregation. Our GEE footprinting results provide a more comprehensive view of Aβ1–42 polymerization, showing various degrees of involvement across different regions. Specifically, our results verify the important roles played by the middle region and C-terminal region in the aggregation, offering a high-resolution perspective down to the aspartic and glutamic residue level.

The solvent accessibilities features found from the footprinting are in accord with those observed in the fibril solid state structure determined by CryoEM¹⁶, suggesting that the soluble oligomer have similar structures as the that of the final fibril. It is of high interest to probe this question for experiment mimicking in vivo conditions.

For modeling the kinetics, we chose Cohen’s nucleation-elongation-fragmentation mechanism, considering the model’s balance of simplicity and comprehensiveness. The outcome, a four-population model, provides a successful fit to the experimental results, explaining the three-sigmoidal dynamics profile, identifying a key factor for the mechanism driving the polymerization, and delineating the potential reactions/steps at different aggregation times. The ability to model the footprinting results should provide insights into the effects of media on aggregation.

The GEE footprinting approach overcomes the limitations of other biophysical methods and enables a real-time and precise monitoring of aggregation at the “regional” or peptide and residue level in a simple and straightforward experimental setup (no laser or pulsed light sources needed). Nevertheless, this approach is not without limitations; notably, the reagents’ relatively low reactivity compromises reaction yields and necessitates extended reaction times and increased reagent dosage to achieve a reasonable modification extent⁵⁴. Additionally, sequence coverage remains a limitation, particularly in the C-terminal region of Aβ42, where residues such as Gly, Ala, Val, and Leu exhibit low intrinsic reactivity to many common footprinting reagents. A third limitation is that the slow footprinting may perturb the polymer structure although this is less likely in the pulsed mode we used here compared to the long footprinting times in other experiments. In the future, we will use other stable chemical footprinting methods to follow Aβ1–42’s polymerization to obtain a more comprehensive picture of the aggregation mechanism. In the future, we intend to extend this methodology to investigate Aβ-aggregation inhibitor interactions and epitope mapping with potential antibodies both in vitro and ultimately in vivo.

METHODS

Materials.

Synthetic wild type human amyloid beta 1–42 (Aβ1–42) was purchased from AnaSpec (San Jose, CA). Lys-N protease was from Seikagaku Corporation (Tokyo, Japan). hexafluoro-2-propanol (HFIP), trifluoracetic acid (TFA), sodium hydroxide (NaOH), Thioflavin T, glycine ethyl ester hydrochloride (GEE), phosphate buffer saline (PBS, 10 mM phosphate, 138 mM NaCl, 2.7 mM KCl), ammonium acetate, urea, and formic acid (FA) were purchased from Sigma-Aldrich (St. Louis, MO). 1-Ethyl-3-(3-dimethylaminopropyl) carbodiimide hydrochloride (EDC) was from ThermoFisher Scientific (Waltham, MA). Chymotrypsin was purchased from Promega (Madison, WI)

Formation of Aβ1–42 Aggregates.

The synthetic Aβ1–42 as purchased was dissolved in 1 mL of TFA, sonicated for 15 min, and incubated at room temperature for 30 min to ensure complete dissolution of any preformed aggregates. The TFA was then evaporated in a fume hood. Subsequently, the sample was dissolved in HFIP at a concentration of 0.1 mM and incubated at room temperature for 1 h to disrupt any remaining aggregates. This step was followed by evaporation of HFIP in a fume hood. After HFIP evaporation, a clear film of Aβ1–42 formed at the bottom of the tube. This HFIP treatment process was repeated two times, and in the final step, the resulting solution was aliquoted into tubes (32 μL per Eppendorf) and frozen at −80 °C for future experiments. Prior to incubation for aggregate formation, the HFIP-pretreated Aβ1–42 was dissolved in 3 mM NaOH (pH 11.5) to achieve a final concentration of 200 μM in NaOH solution. The solution was incubated without agitation at room temperature for 5 min, followed by sonication for 2 min. To initiate aggregation, the Aβ1–42 NaOH solution was diluted 20-fold (v/v) with 1X PBS buffer (pH 7.4). The final concentration of Aβ1–42 after dilution was 10 μM (pH 7.4). Aggregation was allowed to proceed for varying times (0 to 30 h) at 25 °C in PBS buffer. The extent of aggregation at each time point was studied in triplicate to provide suitable statistics.

Thioflavin T assay.

Thioflavin T powder was dissolved in 1XPBS buffer to yield a 10 μM, pH 7.4 ThT buffer. The previously described 200 μM Aβ1–42 NaOH solution was diluted 20-fold (v/v) with the above ThT buffer. Immediately after the dilution, the sample was submitted to Infinite 200 PRO M PLEX (Tecan, Männedorf, canton of Zürich, Switzerland) for fluorescence intensity measurements. The measurements were carried out at 25 °C with E_x/E_m = 450 nm/484 nm. The read was conducted every 20 min with 15 s shaking between reads. Shaking (orbital) amplitude and frequency were 2.5 mm and 245 rpm, respectively. The Z-position mode was manually 20,000 mm. The integration lag time is 40 ms.

GEE labeling of Aβ1–42.

Because GEE labeling is a two-step reaction, the tuning of reagent conditions required more effort than for usual specific amino acid footprinting. Previous work indicated the EDC reagent is the limiting reagent in GEE labeling³⁵, whereas different GEE concentrations resulted in similar modification extents. Therefore, for the tuning, a constant concentration of GEE was taken, and various concentrations of EDC were used to define the optimum concentration of EDC. Following that, the GEE concentrations were adjusted at the optimum EDC concentration (Figure S3). At several incubation times, the Aβ1–42 solution was immediately submitted to GEE labeling without a freeze-thaw cycle. GEE (150 mM), and 30 mM EDC were introduced into 10 μM Aβ1–42 to initiate the reaction at room temperature (25 °C) for 10 min. The reaction was quenched by adding 1 M ammonium acetate and incubated at room temperature for 15 min. After the footprinting, all samples were submitted to enzymatic digestion directly. For each aggregation time point, Aβ1–42 was incubated independently in triplicate and subjected to GEE labeling independently.

Proteolysis.

The GEE-footprinted samples were aliquoted into two parts: one part was submitted to LysN digestion, and the other half was for chymotrypsin digestion. For the LysN digestion, 4 M urea was introduced to the footprinted sample solution, followed by incubation at 37 °C for 30 min to allow full denaturation. Then the Lys-N was added to give an enzyme-to-protein ratio of 1:10 (w/w). Samples were incubated at 45 °C for 4 h. The digestion was then quenched by adding formic acid (FA) to give a final concentration of 1% (by volume). For the chymotrypsin digestion, 3 M urea was introduced into the footprinted samples and incubated at 37 °C for 30 min and diluted so that the final urea concentration was 0.9 M. Chymotrypsin was added with an enzyme-to-protein ratio of 1:50. The digestion was conducted at 25 °C for 20 h and then quenched by adding FA to a final concentration of 1% (by volume).

LC/MS/MS.

For peptide and residue-level analysis, the digested samples were analyzed by LC–MS/MS on a nanoElute system (Bruker Daltonics, Billerica, MA) that was coupled to a timsTOF pro 2 mass spectrometer equipped with Captive Spray (Bruker Daltonics, Billerica, MA). Peptides were trapped on a C18 column (PepMap TM C18, 5 mm × 300 μm, 5 μm, 100 Å) (Thermo Fisher Scientific, Santa Clara, CA), desalted for 5 min with 300 μL/min water/0.1% formic acid. Separation was performed on a C18 column (Pepsep C18 15 cm × 150 μm, 1.5 μm, 100 Å) (Bruker Daltonics, Billerica, MA). After desalting, peptides were eluted by using a 45 min gradient of 300 nL/min 2%−90% acetonitrile (ACN)/0.1% FA gradient: 2% ACN (0.1%FA) to 20% in 15 min, increased to 60% ACN (0.1% FA) in 15 min, ramped to 90% in 5 min, held at 90% for 10 min.

For the TimsToF Pro settings, the PASEF method was utilized with mobility-dependent collision energy ramping set to 70 eV at an inversed reduced mobility (1/k₀) of 1.6 V·s/cm² and 20 eV at 0.6 V s/cm². Precursor ions with a target intensity of 20000 counts in an m/z range between 100 and 1700 with charge states 1–5+ were selected for fragmentation. Active exclusion was enabled for 0.4 min.

Data Analysis.

For the peptide identification, the product-ion spectra were submitted to Byonic and Byologic (Protein Metrics by Dotmatics., Cupertino, CA) software. Validation of modification-site was achieved by using MS/MS spectra (Figure S7–S11). Signals for isomeric modified peptides that gave overlapping extracted ion chromatograms were summed for the modification calculation because the MS/MS identification was ambiguous, and the constituent chromatograms could not be integrated with high confidence. All the raw data were submitted to Skyline⁵⁵ for peak-area calculation. For the peptide peaks, the retention time windows for peak area calculation in Skyline were adjusted identically for all the samples. The modification extents for the peptides and residues were calculated based on eq 1, as discussed earlier.

$$\%\text{modified}=\frac{\Sigma P_{\mathit{\text{mod}}}}{\Sigma P_{\mathit{\text{mod}}}+\Sigma P}\times 100.$$ (eq.1)

Kinetic Modeling.

A four-population (including monomer) model (m, M1(P1) to M4 (P4) was used for the kinetic modeling; the model was built as described by Iljina et al. and Illes-Toth et al^{22, 56} (more details are in Supporting Information (SI)). The key components of modeling can be found in the Supporting Information (Table S1 and Excel S2).

Supplementary Material

Supporting Information:

Additional experimental details. Kinetic Modeling section (I); GEE-EDC reaction mechanism (Figure S1); Aβ42 sequence (Figure S2);reagents titration curves (Figure S3); deconvoluted mass spectra for Aβ42 monomer and aggregates (Figure S4); Aβ42 peptides and specific residues GEE modification extent comparison between different aggregation times (Figure S5); 3-population fitting results (Figure S6); LC-MS/MS of LysN digested peptide 1–15 (Figure S7); LC-MS/MS of chymotryptic peptide 1–4 (Figure S8); LC-MS/MS of chymotryptic peptide 1–10 (Figure S9); LC-MS/MS of LysN digested peptide 16–27 (Figure S10); LC-MS/MS of LysN digested peptide 28–42 (Figure S11); Kinetic modeling reaction table (Table S1); References

Kinetic modeling parameters (Excel S2)