Structural progression of amyloid-β Arctic mutant aggregation in cells revealed by multiparametric imaging

The 42-amino-acid β-amyloid (Aβ42) is a critical causative agent in the pathology of Alzheimer's disease. The hereditary Arctic mutation of Aβ42 (E22G) leads to increased intracellular accumulation of β-amyloid in early-onset Alzheimer's disease. However, it remains largely unknown how the Arctic mutant variant leads to aggressive protein aggregation and increased intracellular toxicity. Here, we constructed stable cell lines expressing fluorescent-tagged wildtype (WT) and E22G Aβ42 to study the aggregation kinetics of the Arctic Aβ42 mutant peptide and its heterogeneous structural forms. Arctic-mutant peptides assemble and form fibrils at a much faster rate than WT peptides. We identified five categories of intracellular aggregate—oligomers, single fibrils, fibril bundles, clusters, and aggresomes—that underline the heterogeneity of these Aβ42 aggregates and represent the progression of Aβ42 aggregation within the cell. Fluorescence-lifetime imaging (FLIM) and 3D structural illumination microscopy (SIM) showed that all aggregate species displayed highly compact structures with strong affinity between individual fibrils. We also found that aggregates formed by Arctic mutant Aβ42 were more resistant to intracellular degradation than their WT counterparts. Our findings uncover the structural basis of the progression of Arctic mutant Aβ42 aggregation in the cell.

mCherry and carried out direct, in-cell observation of A␤42 aggregation kinetics using FLIM (16) and super-resolution imaging (28 -30). We identified five categories of intracellular aggregate-oligomers, single fibrils, fibril bundles, clusters, and aggresomes-that underline the heterogeneity of A␤42 aggregates and represent the progression of A␤42 aggregation within the cell. We also found that Arctic mutant aggregates form more rapidly and to a far greater extent in cells than WT aggregates and are inefficiently degraded.

A␤42 Arctic mutant in cell model displays a fast and aggressive aggregation phenotype
We constructed single-copy, stable cell lines in Flp-In T-REx293 cells with a cytomegalovirus promoter driving mCherry, mCherry-A␤42(WT), or mCherry-A␤42(E22G) expression (Fig. S1). Characterization by wide-field fluorescence microscopy showed prominent expression of the reporter proteins in these cell lines (Fig. 1, A-C). However, there was a marked difference among them with respect to the amount of protein aggregation. After a week of induction by tetracycline, mCherry was homogeneously distributed in cells (Fig. 1A), whereas mCherry-A␤42 and its Arctic mutant formed aggregates inside cells (Fig. 1, B and C). Cells expressing the Arctic mutant A␤42 displayed aggressive aggregation of the fusion protein with very few or no coexisting soluble fragments, indicating that most of the expressed protein is recruited into aggregates (Fig. 1C). Such aggressive aggregation was not observed in the mCherry-A␤42(WT) cell line. In a time course experiment, visible mCherry-A␤42(E22G) aggregates began to accumulate 24 h after gene induction and within 3 days were present in most cells. However, in the mCherry-A␤42(WT)-expressing line, cells containing microscopically visible aggregates became apparent more slowly and by the end of the time course were only present in ϳ20% of the cell population (Fig. 1D). Therefore, our stable cell lines seem to mimic, albeit on a different timescale, the different assembly and aggregation properties of WT and Arctic mutant forms of A␤42 in the respective forms of AD.

FLIM reveals different conformation states of in-cell A␤42 species
To characterize the nature and dynamics of Arctic mutant A␤42 assembly in cells, we used FLIM (31) to monitor the conversion of soluble protein fragments into amyloid fibrils. In our previous study, a decrease in fluorescence lifetime of a linked fluorophore was correlated with the development of ␤-sheetrich amyloid structures (3,30). In the current study, we first investigated different states of aggregated species by measuring fluorescence decay times of an associated reporter protein, mCherry. After 48 h of gene expression, the fluorescence lifetime of most of the mCherry-A␤42(E22G) protein within the cell varied little throughout the cytosol, indicating a relatively homogeneous population of A␤42 species within cells ( Fig. 2A). Where aggregates were clearly visible, however, a lower fluorescence lifetime was observed (Fig. 2, A and B), consistent with the conversion of soluble species or loose fibrils to more compacted structures (32). To correlate the fluorescence lifetime with the underlying amyloid structure, we next imaged the same cells by SIM (Fig. 2C). In agreement with the fluorescence lifetime results, we showed that the cytosol contained primarily soluble protein, confirming that A␤42 species were either in the soluble state or associated with a single large perinuclear aggregate, which are the large, and thus visible by confocal microscopy, fibrils that also display the low fluorescence lifetime. The higher resolution of SIM demonstrated this large aggregate to be composed of multiple fibrils.

Structural progression of A␤42 Arctic mutant aggregates
The structural heterogeneity of the Arctic mutant aggregates suggested by FLIM and 2D SIM (Fig. 2) was next examined using 3D SIM, which can reveal morphological and structural details of aggregated protein species (4). The projection view of

Structural progression of amyloid-␤ Arctic mutant
3D SIM was reconstructed from a series of 2D SIM sectioning by Fiji. Fig. 3A shows representative fibrillary structures of intracellular Arctic mutant proteins at different stages of maturation. 2D SIM images and the projection view of 3D reconstructions show that, within 24 h of gene induction, Arctic mutant monomers nucleated to form single fibrils of ϳ100 nm in diameter and up to ϳ2 m in length (Fig. 3A, day 1). As more fibrils formed in the cell, they started to assemble into loose clusters in which some fibrils were aligned with each other and apparently cross-linked but also with many gaps in the structure (Fig. 3A, day 2). The continued accumulation of fibrils gradually led to the formation of bundles, ϳ5 m in diameter, consisting of multiple linear fibrils aligned in a similar orientation (Fig. 3A, day 3). Although there were still gaps between fibrils, as shown in the 2D section slice, the projection view of its 3D reconstruction showed a dense structure, suggesting the compaction of multiple layers of fibrils. These compact structures were widely observed throughout cells after 3 days of gene induction. By day 6, the fibril assemblies had matured further and showed a very different morphology: the fibril clusters were no longer largely aligned but displayed multiple, tangled branches oriented in various directions. Next, we quantified the size distribution of Arctic mutant aggregates over the 6 days of the experiment by analyzing the 2D SIM images (Fig. 3B). On day 1, ϳ90% of the amyloid aggregates appeared to have an area of less than 0.02 m 2 , which is of the order expected for oligomers. Fibrils are greater than 0.02 m 2 in size, whereas fibril bundles and clusters exceed 0.1 m 2 . As the experiment continued, there was a progressive drop in the proportion of oligomers in the overall population from ϳ90% (day 1) to less than 40% (day 6), probably reflecting the assembly of oligomeric species into fibrils, fibril bundles, and larger clusters. Thus, fibrils and higher-order assemblies derived from them become the predominant species as the amyloid structures mature.
We further compared the intracellular size distribution of aggregate species of the A␤42 Arctic mutant with two other aggregation-prone proteins, mCherry-A␤42(WT) and polyglutamine-containing huntingtin (HDQ72) tagged with GFP ( Fig.  3C) (33). Six days after induction, the Arctic mutant had the lowest level of oligomeric species (i.e. 40%; see above) and the majority of its aggregates in the form of fibrils and related structures, showing it to be the most aggressively aggregation-prone of the three proteins. In comparison, ϳ60% of mCherry-A␤42(WT) aggregates and ϳ70% of HDQ72-EGFP aggregates had an area of less than 0.02 m 2 , i.e. were in the form of oligomeric species. This demonstrates that the nature of the aggre-

Structural progression of amyloid-␤ Arctic mutant
gation-prone protein, even a difference of one amino acid, can dramatically affect its aggregation properties, specifically in terms of dynamic assembly and structural organization. This highlights the importance of assessing aggregation-prone proteins individually rather than assuming that all such proteins follow the same pattern of aggregation and supramolecular assembly.

Heterogeneous state of Arctic mutant aggregates inside the cell
We next applied 3D SIM to a more detailed investigation of the different types of mCherry-A␤42(E22G) aggregate in cells at day 7 postinduction. Because gene expression was continuous throughout this period, the whole range of aggregate species, including oligomers, single fibrils, fibril bundles, fibril clusters, and aggresomes, was present and widely distributed throughout the cytosol in essentially all cells (Fig. 4A, left and middle panels) in which the aggregate species were much more abundant compared with cells at day 2 postinduction (Fig. 2). Compared with 2D section slices (Fig. 4A, left panel), which show only a single layer of the fibrillar structure, full 3D reconstructions (Fig. 4A, middle panels) revealed the high density of the aggregate species, mostly fibrils, in this Arctic A␤42 mutant-expressing cell. This demonstrates the ability of 3D SIM to provide comprehensive details of volumetric structures in cells. The zoomed-in regions in Fig. 4A show the heterogeneity of aggregate structures (Video S1 for image 1 and Video S2 for image 3), including oligomers (image 3, yellow arrows), single fibrils (image 3, green arrows), fibril bundles (image 3, blue arrows), and a compact fibril cluster (image 4) composed of tangled fibrillary fragments in various orientations.
The mature state of protein aggregation in these cells is characterized by the presence of a perinuclear aggresome, which is formed when the protein degradation system is overwhelmed (34). Aggresomes are highly compact structures comprising branched and tangled fibrils positioned at or near the centrosome (35) (Fig. 4B and Video S3). A typical example of an mCherry-A␤42(E22G) aggresome is shown in the top left panel of Fig. 4B with z-stack section slices at the depths indicated in the lower left panels of Fig. 4B. These sections demonstrate the internal organization of tangled fibrils in the aggresome, which is further reconstructed as a 3D structure in the right panel of Fig. 4B. This amorphous morphology is consistent with the random assembly of fibrillar structures, either by diffusion or active transport, rather than a more ordered assembly of monomeric or oligomeric species at a nucleation site.

Inefficient degradation of A␤42 Arctic mutant aggregates
The aggressive accumulation of Arctic mutant aggregates in cells could result from either its rapid assembly into fibrils, its resistance to degradation, or a combination of both. The A␤42 Arctic mutant has been reported to greatly promote the formation of protofibrils/oligomers (25) and can generate fibrils at much lower concentrations and higher rates than WT A␤42 in vitro (36). In the WT peptide, Glu-22 destabilizes the oligomer structure by electrostatic repulsion between adjacent Glu-22 side chains. In silico modeling revealed that substituting glycine at this position replaces glutamic acid with an uncharged resi-due, resulting in higher oligomer stability (37). The lack of a side chain in glycine may also provide greater conformational flexibility and avoid steric interference among A␤42 peptides (38). Furthermore, glycine is more hydrophobic and ␣-helixdestabilizing than glutamic acid (39) and thus virtually eliminates the ␣-helix propensity in the region adjacent to Gly-22, as shown in molecular dynamics simulations (40). Our results are consistent with the above analyses in the literature and a much faster rate of aggregation of the Arctic mutant than WT A␤42 in living cells.
To test the second premise, that Arctic mutant aggregates are more resistant to degradation than WT, we investigated whether pre-existing mCherry-A␤42(E22G) aggregates (present after 1 week of induction) can be efficiently degraded when no new Arctic mutant protein is produced. When mCherry-A␤42(WT) or mCherry-A␤42(E22G) gene expression was switched off in the respective inducible cell lines, we observed a progressive decrease in the proportion of cells containing aggregates of both types (Fig. 5A). After 4 days, the mCherry-A␤42(WT) cells had degraded almost all of their aggregates; however, ϳ50% of the population of Arctic mutant-expressing cells still contained visible aggregates, mainly accumulated in aggresomes, showing that these aggregates are highly resistant to intracellular degradation. To further address this issue, we used 3D SIM to analyze aggregate structures remaining in cells 4 days after mCherry-A␤42(E22G) gene expression was switched off. As shown in Fig. 5B, a representative cell still contained a compact aggresome, whereas separate fibrillar structures were no longer observed, indicating that these smaller species had been cleared from the cytosol. This large aggregate consists of a highly dense core and a less compact peripheral region containing discernible individual species, as shown in the z-section slice (Fig. 5B, right panel). A large, dense structure of this kind is likely to be more resistant to processing by the ubiquitin-proteasome system and autophagy than looser, smaller assemblies of aggregated protein (41,42).
Because the Arctic mutant cells generate more aggregates than WT A␤42 cells, the apparent resistance to degradation of the former aggregates could simply be due to the degradation machinery becoming overwhelmed. To assess this, we incubated mCherry-A␤42(E22G) cells with tetracycline for 24 h, which resulted in the formation of a limited number of small aggregates and single fibrils (Fig. 5C, day 1). Then we removed the inducer and analyzed the aggregates in these samples over the following 4 days (Fig. 5C, days 2-5). We found that the fibrils and small aggregates formed within 24 h persist throughout this period. This is consistent with Arctic mutant aggregates being intrinsically highly resistant to degradation. Thus, the results of Fig. 5A are unlikely to be due to the degradation machinery being overwhelmed or to a special property of aggregates within aggresomes.

Discussion
The emergence and proliferation of A␤42 plaques in AD brains could be due to several factors: 1) an increase in production of A␤42 peptides, 2) compromised cellular degradation systems, and/or 3) the rigid structure and stability of A␤42 amyloid. In this study, we investigated the structural progres-

Structural progression of amyloid-␤ Arctic mutant
sion of aggregates of the A␤42 Arctic-mutant peptide in living cells and obtained insights into its fast aggregation and resistance to protein clearance. To do this, we introduced a cell model that expresses the A␤42 Arctic-mutant peptide, tagged by the fluorescent protein mCherry, which mimics intracellular amyloid formation. Our model demonstrates that a single point mutation in A␤42, E22G, has a dramatic effect on A␤42 protein homeostasis, leading to rapid and aggressive amyloid fibril for-

Structural progression of amyloid-␤ Arctic mutant
mation. The reporter protein mCherry is shown here to be an excellent marker for fluorescence imaging, including FLIM, which allows quantitative study of the kinetics of amyloidogenesis, and 3D SIM, which enables us to dissect the morphological details of amyloid species at different stages of development at a resolution of 100 nm. The mCherry tagged to A␤42 could potentially affect the structural progression of aggregates, but other studies (16,32) have shown that similar fluorescent protein tags take up a peripheral location in developing amyloid and are thus thought to exert a minimal influence on the aggregation process. Because oligomers are normally smaller than 100 nm, SIM would not allow us to examine their morphology in detail. However, they are clearly much shorter than single fibrils, as shown in Fig. 4A. Therefore, in this study, we characterized five phases of amyloid development ranging from oligomers to single fibrils, fibril bundles, clusters, and aggresomes (Fig. 6).
We found that, 24 h after mCherry-A␤42(E22G) gene expression, most of the cytoplasm is filled with species displaying low fluorescence intensity and high lifetime, suggesting a soluble pool of Arctic mutant throughout the cytoplasm, whereas the intracellular regions with higher intensity and lower lifetime may reflect the rapid accumulation of oligomers or single fibrils (43). The formation of aggresomes with prominent fluorescence intensity led to a dramatic decrease in fluorescence lifetime, which is an indication of highly compacted structure via the mechanism of mCherry selfquenching (44). This demonstrates the highly compact structures of aggresomes, as further revealed by 3D SIM (Fig.  4). We also characterized the structural progression of amyloid, primarily for the size and morphology of different aggregation states. The 3D structures reconstructed from SIM provide us with an unprecedented volumetric view of aggregates proliferating in the whole cell (Figs. 3 and 4), which provides a more comprehensive model of amyloid fibril organization, distribution, and structures.
In the development of aggregates, we found that fibrils first aligned in the same orientation to form fibril bundles and then assembled in a random way to form clusters and aggresomes as the intracellular fibril concentration increases. We demonstrated that the single residue change in the mutant A␤42 peptide leads to significant differences in amyloid progression and structures, which are resistant to intracellular degradation. How these different amyloid states lead to intracellular toxicity needs to be fully addressed in the future; this model provides a platform for studying the effect of intracellular amyloid on the function of various cellular compartments. In addition, because this model demonstrates various stages of protein aggregation, it can also be used for the screening of antiaggregation compounds.

Cells
Sequence of plasmids used for stable cell line construction are given in the supporting information. Mammalian Flp-In T-REx293 cells were grown in T75 or T25 flasks or 6-well plates by incubation at 37°C in a 5% CO 2 atmosphere (35). Complete medium consists of 90% Dulbecco's modified Eagle's medium, 10% fetal bovine serum, and 2 mM L-glutamine; antibiotics were used as appropriate (35). Cells were passaged on reaching 80 -90% confluence (approximately every 3-4 days) and kept in logarithmic phase growth. Routine cell counting and viability assays were carried out using a hemocytometer and trypan blue. Transfections were performed on cells at 80% confluence. Stable cell line construction has been described previously (32).

Microscopy
The protocol of cell fixation for imaging was as described previously (32). After induction for various times, cells were fixed, and images were recorded with an OMX V3 super-resolution microscope (35).  The aggregation of A␤42 Arctic mutant initiates from oligomerization of monomers, which then rapidly assemble to form fibrils. Multiple fibrils connect with high affinity to form bundles, clusters, and eventually aggresomes. The five phases in the structural progression of Arctic mutant aggregates are shown from left to right. Phase 1, soluble oligomers in cytosol. Phase 2, linear fibrils ranging from 500 nm to 2 m in length, representing amyloid building blocks and the most universal form of intracellular aggregates. Phase 3, fibril bundles consisting of multiple fibrils aligned in a similar orientation. Phase 4, multiple fibrils assembled as a tightly bound cluster. Phase 5, a large number of fibrillary fragments that form highly compacted aggresomes, which are resistant to degradation.

Structural progression of amyloid-␤ Arctic mutant
Confocal microscopy was performed as described previously (35). FLIM experiments were carried out using custom-built time-correlated single photon counting as described before (16). FLIM images were analyzed by FLIMfit (31). Imaging of section slices was acquired from the OMX, and 3D reconstruction from multiple section slices was performed by Fiji Volume Viewer, which produced 3D projection views of the reconstruction.