Summary: FASEB Science Research Conference on Protein Aggregation in Health and Disease

The first FASEB Science Research Conference on Amyloidosis in 1995 brought together protein chemists, clinician investigators, and pathologists who were interested in the group of diseases identified by the tissue-compromising deposition of amyloid fibrils. The meeting has grown organically to become a biennial gathering of scientists who study every aspect of protein aggregation, from the analysis of structure and behavior of single-molecule aggregation precursors to the study of the downstream organismal effects that are manifested in human diseases. Meetings have reflected the evolving nature of the field: broadening its scope to include different forms of aggregation and recognizing the existence of amyloids that serve normal functions in both humans and prokaryotes and that the templating properties of the amyloidogenic prion protein, PrP—once thought to be unique as an infectious protein—may be a general property of all amyloids. After 2009, “Amyloid (or Amyloidosis)” was no longer part of the meeting title, replaced by the more generic “Aggregation.” Of most importance, since its inception, the meeting has served as a venue for the cross-fertilization of ideas and the formation of working scientific collaborations among principal investigators, postdoctoral fellows, and graduate students across institutions, countries, and continents.

As in prior meetings, many of the presentations involved the investigation of amyloidogenic structures. A characteristic of amyloidogenesis is that the various soluble precursors do not share amino acid sequences, despite the fact that fibrils all contain the same cross-β 3-dimensional core structure. The aggregation process is clearly homotypic, and a core fragment of almost all precursors can exhibit the full fibrillogenic potential of the intact molecule. This was confirmed at this meeting with solid-state NMR measurements of the low complexity (LC) N-terminal domain of FUS, a protein that has been known to be involved in some forms of amyotrophic lateral sclerosis. The structure revealed that a specific 57-residue segment forms the core of FUS LC fibrils, although the entire N-terminal domain is 214 residues, and the amino acid composition of the core-forming segment is not much different from the overall amino acid composition. This again raises the issue of why a specific segment forms the fibril core, and, mechanistically, what are the thermodynamic and kinetic factors that are responsible for such a monomorphic process in a sequence with a high degree of quasirepetition and self-similarity? (Tycko). Similarly, in a new model that uses a short, amyloidogenic peptide (HYFNIF), the hydrogen-bonded β-strands form the core structure that is fibril formation competent even when excised from the intact protein (Serpell).

Despite these shared structural characteristics, many homotypic amyloid and preamyloid aggregates that are derived from the same precursor display considerable ultrastructural polymorphism. Several groups have utilized NMR chemical shift data to analyze the structural similarity of the Parkinson’s disease–associated protein α-synuclein (AS) fibril polymorphs that are formed in vitro, which exist mostly within two families of folds, one unique and one shared. The analysis of submilligram quantities of fibrils by using customized isotopic labeling protocols allowed for detailed analysis of the conserved structural features of AS fibrils that are determined uniquely by protein sequence and how these features change as a result of interactions with other proteins in the context of disease (Rienstra).

NMR approaches have also been used to examine the atomic-resolution structure of the aggregates of the polyglutamine-expanded huntingtin (HTT) fragments that aggregate in Huntington’s disease. These data suggest that the aggregation process of nonhydrophobic amyloids may be less prone to polymorphism as a result of their relatively shallow energy landscape compared with those on the basis of hydrophobic interactions. Polyglutamine tends to assemble into aggregates with a highly reproducible core architecture. In this system, differences in the sequestration and dynamics of nonamyloid flanking domains underpin HTT exon 1 aggregate polymorphism. Expansion of such studies should allow for the determination of the most critical factors that control polymorphic aggregation. These may influence the soluble, disordered monomer ensemble; the nature of the primary nucleation event; and/or processes of secondary nucleation, fibril elongation, and fragmentation (Van der Well).

Other data from single-molecule FRET studies of synthetically produced, fluorescently labeled HTT exon 1 constructs that contain different length stretches of polyQ designed to examine the length dependence of HTT aggregation have suggested that the increased drive to sequester polyQ away from solvent likely accounts for length-dependent aggregation, rather than a length-dependent structural change per se (Ruff).

In vitro analyses of τ aggregation that used intra τ distance measurements revealed a novel oligomeric form of τ-containing β-strand structure in the paired helical filament 6 hexapeptide motif that forms upon its binding to physiologic lipid membranes. Oligomers also incorporated other regions of repeat 2 and repeat 3 into their immobile core, but these regions did not form regular secondary structure, which suggests that these oligomers are on-pathway, potentially toxic intermediates in τ filament formation. A seed for τ aggregation was not defined in these studies; therefore, the molecular nature of the minimal species required for seeding is still unclear. Consistent with other systems presented at the meeting, the formation of τ-RNA droplets at conditions that are close to physiologic were observed, although the in vivo relevance of the observation is uncertain (Han).

Within cells, membraneless organelles, including P-bodies, nucleoli, and metabolic puncta (stress granules), appear as separate phases, some containing reversible fibrils that are formed by interactions of LC protein domains. Atomic structures of 4 segments of LC domains from inclusion-forming proteins indicated that they stack together into sheets similar to amyloid fibrils, but, unlike pathogenic amyloid, the sheets are kinked and display small surface areas of interaction that predominantly involve aromatic side chains and hydrogen bonds. Computationally, there are hundreds of LC segments that are potentially capable of forming such reversible interactions in proteins as diverse as RNA binders, nuclear pore proteins, keratins, and cornified envelope proteins, which is consistent with the capacity of cells to form a wide variety of dynamic intracellular bodies (Eisenberg).

An additional aspect of structural polymorphism was noted in studies of functional microbial amyloids that identified a structure of a full-length bacterial cross-α amyloid-like fibril (PSMα3 of Staphylococcus aureus), which is unprecedented in >100 structures of eukaryotic cross-β amyloids solved to date. Cross-α fibrils are toxic to human cells, which supports their definition as bacterial functional amyloids that may be involved in pathogenicity. These studies confirm the retention or the re-emergence of the amyloid conformation across kingdoms, achieved by using a variety of structural pathways, including both cross-α and cross-β, to reach a final molecular shape—a tribute to its biologic usefulness (Landau).

The relevance of these observations to human disease was consistent with cryoelectron microscopy studies that showed that the paired helical and straight τ filaments from human Alzheimer’s disease (AD) brains are ultrastructural polymorphs, rather than being uniform (Goedert). How this impacts neurodegenerative disease pathogenesis is unclear (see below).

In vitro, the process of fibrillogenesis is clearly homotypic; however, in vivo and in cells there are many opportunities for potentially amyloidogenic precursors to undergo heterotypic protein–protein interaction. Studies of the sequence similarity required for additional downstream events, such as cross-species seeding by prions, indicated that interspersed asparagine and glutamine residues seemed to anchor interactions at the interface between infectious and host protein molecules, which lowered the barrier for aggregation. Whether this applies to the seeding of other amyloids remains to be seen. These data raise the question of whether the initial intermolecular contact sites between monomeric and aggregated PrP^Sc might also serve as critical templating sites, directing PrP into a particular PrP^Sc fold (Sigurdson).

In the case of α-synuclein, a novel c-Abl phosphorylation site at Tyr39 creates partly helical membrane-bound states of the protein that resemble those that are generated by the G51D Parkinson’s disease mutation, a potent driver of membrane-induced synuclein aggregation, which provides a potential link between the dysregulation of c-Abl activity and AS pathology (Eliezer).

In vitro heterotypic interaction between wild-type amylin [islet amyloid polypeptide (IAPP)] and mutant forms of amylin and amyloid-β was examined by using native mass spectrometry and kinetic analysis of amyloid assembly. Studies indicate that the course of assembly can be altered dramatically by the coassembly of different precursors, which results in the formation of unique hetero-oligomeric species with unique, frequently enhanced aggregation propensities. Such species may interact differently with cellular components, including molecular chaperones and/or glycosaminoglycans; thus, it is possible that the coassembly of different protein sequences could explain differences in disease progression in individuals with the same genotype (Radford).

In vitro experiments that examined the effect of transthyretin (TTR)—the precursor protein in the human systemic amyloidoses familial amyloidotic polyneuropathy and cardiomyopathy and senile systemic amyloidosis—on fibrillogenesis by the microbial amyloid scaffold protein, CsgA, demonstrated that, in contrast to the previously described cross-seeding between IAPP and amyloid-β, TTR inhibited CsgA fibril formation and biofilm formation by intact Escherichia coli. A similar inhibitory effect of TTR on amyloid-β fibrillogenesis had been previously reported. These experiments show that, despite the lack of sequence commonalities, the conformational capacity to form and/or inhibit amyloid exists across biologic kingdoms. What determines the direction of the interaction is still unknown (Buxbaum).

Other instances of functional heterotypy—presumably related to structural polymorphism—have been well documented in yeast. Studies that have used the yeast prion, Sup35/[PSI⁺], showed that prion strains coexist in individual yeast cells and compete with one another to direct the conversion of soluble, nonprion state protein to individual strains. It seems that competition is influenced not only by the inherent rate of conversion of each strain, but also by the number of heritable amyloid aggregates that are composed of these conformers, which suggests that competition is dynamic and fluctuating. The question thus remains: What are the proteostatic niches that allow switches among prion strains, and how do these conditions intersect with the different conformers to alter amyloid dynamics in vivo? (Serio). Mathematically analyzing the nucleated polymerization model to include the action of Hsp104 as an enzyme that carries out the fragmentation process, it seems that in the case of the yeast [PSI⁺] prion, Hsp104 becomes limiting, with variable fragmentation rates experienced by aggregates. These may act as barriers to prion propagation and the ability of prion strains to stably coexist (Sindi).

Quantitative super-resolution techniques—previously used to study transcription—have been repurposed to unveil the early steps of the nucleation process that takes place intracellularly. Analyzing the distribution of subdiffractive cluster sizes revealed that an apparent first-order phase transition underlies the formation and growth of α-synuclein aggregates in mammalian cells and a thus far hidden regulatory pathway exists to clear aggregates that reach some critical size in healthy cells. These data are consistent with those reported from the Finkbeiner and Hatters laboratories (below). The question remains as to whether these techniques can be used to observe individual aggregates as they cross the sol–gel barrier and measure their growth or disassembly kinetics in vivo (Cisse).

Newly developed microfluidic and microdroplet approaches to explore protein aggregation on the micro and nano scale offer the possibilities of probing protein aggregation in vitro on scales that are similar to those present in cells. They also allow for new types of experiments that are designed to probe the microscopic steps of protein self-assembly and aggregation in vitro (Knowles).

As an example, measurements of the effects of DNA-activated protein kinase phosphorylation on hydrogel binding and liquid–liquid phase separated droplet formation by FUS-LC indicate that serine and threonine sites in the FUS-LC fibril core are most sensitive to phosphorylation, which suggests that the fibril core structure has relevance to both hydrogel binding and droplet formation (Tycko).

Other examples include A bodies in which the amyloid-converting motif, a discrete peptide motif, interacts with ribosomal intergenic space long noncoding RNA to initiate amyloidogenesis in the nucleolar area. Aggregation is associated with reversible cellular dormancy. The mechanisms that underlie the conversion remain to be elucidated (Lee).

Similarly, cells that undergo stress, such as increased temperature, accumulate protein aggregates. In the case of the poly-A binding protein 1, its apparent heat-induced aggregation is an ATPase-reversible adaptive phase-separation and gelation process. Heat-induced molecular chaperones disperse poly-A binding protein 1 gels—an endogenous substrate—more readily than model misfolded proteins, such as luciferase. The mechanisms by which stress-induced phase separation, gelation, and other quinary processes function to help cells respond to stress is presently unknown (Drummond).

Amphifluoric FRET has enabled a flow-cytometric, high-throughput exploration of nucleation barriers that are associated with liquid–liquid demixing, amyloid formation, and functional protein polymerization. This method allowed for the distinction of functional from dysfunctional prion-like protein self-assembly. Functional prions nucleate in switch-like fashion, which is characterized by an absence of intermediate phases, whereas low sequence complexity and disease-associated aggregation-prone proteins tend to form metastable liquid or colloid phases that precede amyloid nucleation. Should these findings be reproducible and extendible to other systems, they would suggest that the topology of intermolecular energy landscapes governs whether self-assembly will have functional or pathologic consequences. This question has previously been raised in the context of fibril formation of the normal amyloid formed by the p-mel protein in melanosomes, and these findings suggest that this may be a general phenomenon in cell biology—that is, the exploitation of nucleation barriers to exert switch-like changes in protein activity in cells (Halfmann).

Cytotoxicity is a critical consequence of amyloidogenic protein aggregation. The active cytotoxic molecules in most tissue culture systems seem to be relatively small oligomers; hence, cytotoxicity is frequently used to monitor early aggregation. In a new, simple animal model (pond snail) that was developed to explore the mechanism of neurotoxicity of amyloid peptides oligomeric amyloid-β_1–42 alone demonstrated the internalization and disruption of neuronal intracellular ultrastructure and resulted in memory loss. The conformational species—or range of species—that is associated with this neuronal toxicity and its relationship to that observed in human AD, remains to be defined, as do their differences in conformation from functional or designed self-assembling, fibrillogenic peptides. It has been shown that amyloid-β toxicity is associated with oxidative stress, which can generate the formation of covalent dityrosine cross-links within oligomers and fibrils. Of interest, these cross-links are also formed in other peptides that are associated with neurodegeneration, including α-synuclein and τ. Do they have a specific role in amyloidogenesis? (Serpell).

Concurrent time-resolved biophysical measurements and cell viability assays have revealed that the most toxic species of the pancreatic β-cell, IAPP (amylin), are transiently populated intermediates that are formed during the lag phase of amyloid formation. Newly developed spectroscopic methods identified in vitro a transient intermediate with a non-native β-sheet structure that may be the toxic species. Biophysical studies in conjunction with studies of cultured cells, intact islets, and transgenic mouse models have identified a receptor-mediated mechanism of IAPP-induced toxicity. A critical open question is how oligomers that are characterized in vitro relate to those formed in vivo (Raleigh).

In contrast to amyloid-β and IAPP toxicity, there are two models for how mutant HTT exon 1 aggregation into inclusions relates to pathogenesis—that is, toxicity—that involves seemingly contradictory mechanisms: one defines aggregation as pathogenic, and the other where aggregation is beneficial by sequestering soluble states of HTT exon 1 that are proteotoxic. A new model was proposed in which inclusion sequestration of soluble HTT exon 1 removes the trigger for apoptosis; however, the inclusions may coaggregate other proteins, which compromises cell function and leads to quiescence and, ultimately, slow death by necrosis. Questions raised to test the model are: Does HTT aggregation in cells arise from the coalescence of nascent HTT proteins that emerge from ribosome? If so, is a stall on the ribosome the trigger for apoptosis? Does this process also involve an aberrant liquid protein phase separation at the ribosome, or possibly in stress-granule like structures? (Hatters).

In studies that are related to the pathogenesis of Parkinson’s disease, by using a combination of optical and NMR spectroscopy techniques it was possible to demonstrate that α-synuclein acts as a calcium sensor at the presynapse. In contrast to vesicle associated membrane protein (VAMP)-2⁺ vesicles, which do not form localized clusters at the presynapse upon calcium exposure, a subgroup of presynaptic AS⁺ vesicles are specifically responsive to changes in calcium concentrations via the negatively charged C terminus of AS. It remains to be determined whether a subgroup of AS⁺ synaptic vesicles are responsive to calcium and if they are characterized by a specific lipid composition (Kaminski).

On an organismal level, the question of what instigates the process of aggregation/fibrillogenesis in the tissues of a human or animal destined to develop amyloidosis remains unanswered. Furthermore, the basis of the apparent predilection of neurons and the nervous system to aggregate induced degeneration is also unknown. It is possible that the neuronal proteome is more highly populated by molecules that contain sequences that are prone to aggregation. Alternatively, neurons may be intrinsically less capable of detoxifying amyloidogenic precursors. Studies that have employed automated single-cell methodology revealed that the machinery to mediate a heat-shock response and the ability of neurons to recognize and respond appropriately to aggregation-prone proteins and proteotoxic stress with a canonical heat-shock response differs from the responses to the same stimuli by closely related non-neuronal cells, such as astrocytes, in vitro and in vivo (Finkbeiner).

Studies on the ATF6 regulated arm of unfolded protein response, which has previously been shown to be involved in neuronal responses to several forms of neurodegenerative disease, have indicated that enhancing that response can selectively reduce the secretion of aggregation-prone proteins from the cell without impacting the secretion of normal proteins. This suggests that ATF6-activating small molecules could decrease the secretion of amyloidogenic proteins, providing another mode of neuroprotection that would not be unique to neurons (Wiseman).

The role of a third element of the proteostatic system, the proteasome, and its potential for therapeutic manipulation in tauopathy was explored by using a mouse model in which τ aggregates block the opening gate of the proteasome, thus impairing the degradation of other proteins. Enhanced phosphorylation of the proteasome via the cAMP/PKA pathway by clinically relevant phosphodiesterase inhibitors led to the clearance of τ aggregates, attenuation of early-stage tauopathy, and improved cognitive performance in mice, which suggests that there may be a therapeutic opportunity here. Additional studies have indicated that the stimulation of a GPCR present only on the membrane of dendrites can propagate a downstream signaling pathway, leading to the clearance of τ in the postsynaptic compartments and attenuation of the trans-synaptic spread of τ (Mienstra).

It has been previously demonstrated that the yeast disaggregase, Hsp104, could be engineered to effectively reverse aberrant cytoplasmic TDP-43 and FUS aggregation and restore these RNA-binding proteins to the nucleus. More recent studies have indicated that human nuclear import receptors can also prevent and reverse the aberrant phase transitions of RNA-binding proteins with prion-like domains, including TDP-43 and FUS, by engaging their nuclear localization sequences. Elevated expression of nuclear import receptors mitigates neurodegeneration caused by amyotrophic lateral sclerosis–linked FUS in Drosophila models of the disease (Shorter).

The CsgA system in E. coli is the best characterized of the amyloid networks that function to form the scaffolding for in vivo biofilm formation. Interacting cytoplasmic molecules have been previously identified that keep CsgA from forming curli fibrils intracellularly. Screening of periplasmic extracts has identified CsgC, a 12-kDa, β-rich protein that can inhibit CsgA amyloid formation at substoichiometric molar ratios. In biofilm-forming bacteria that do not produce CsgC, another protein, CsgH, which is 20% similar to CsgC, adopts a structure that is similar to CsgC and is able to efficiently prevent amyloid formation by CsgA. Hence, coli have evolved an efficient system for maintaining functional amyloid precursors in a soluble state until needed (Chapman).

It was reported, somewhat surprisingly, that green fluorescent protein (GFP) binds a variety of amyloid fibrils, including amyloid-β₄₂, IAPP, SEVI, SEM1, insulin, and lysozyme with low micromolar-to-submicromolar affinity and does not bind to the soluble precursors, nor nonamyloidogenic aggregates, such as microtubules or F-actin. It also seemed to be capable of inhibiting fibril formation in vitro at submicromolar concentrations, which suggests that GFP could be a universal amyloid-binding structural motif. Should these data be confirmed, it will be necessary to define the responsible structural determinants and whether they can be redesigned to engineer fibril-specific GFP proteins. It also indicates that studies that use GFP as a biomarker may have to be reinterpreted (Makhatadze).

In the aggregate, these studies suggest that enhancing any or all of these proteostatic processes in neurons may be useful in lowering the susceptibility of the CNS to those neurodegenerative disorders that are related to protein misfolding.

The process of spreading aggregate pathology from cell to cell in neurodegenerative diseases has become a subject of intense investigation, particularly in the context of human tauopathies and their murine models. New data on in vivo seeding using both wild-type and transgenic animals with seeds produced from α-synuclein or τ as well as those generated in tissue culture suggest that there are seeding-competent and -incompetent forms of τ (Lee, Trojanowski, Diamond). It remains to be shown how electron microscopy polymorphisms relate to differences in functional seeding capacities.

Additional tissue culture studies and in vivo analyses of murine tauopathy models have revealed a hitherto unexpected role for apolipoprotein E (ApoE) in pathogenesis, with suppression of pathology in mice that lack ApoE and enhanced disease in mice that also carry human ApoE4 (Holzman).

The necessity of developing methods for detecting protein-only, disease-causing agents became clear with the cross-species transmission of the Bovine Spongiform Encephalopathy agent into humans as variant Creutzfeldt-Jakob disease. It is now possible to diagnose human sporadic Creutzfeldt-Jakob disease by using patients’ cerebrospinal fluid and/or nasal swabs with nearly 100% accuracy using multiwell plate–based seeded polymerization fluorescence assays, called real-time quaking-induced conversion. The assay has now been used for the ultrasensitive detection of disease-associated τ aggregates in brain and cerebrospinal fluid samples from some tauopathy patients, most notably those with Pick disease. Additional adaptations to other types of tauopathies are ongoing to aid in the diagnosis of these disorders and determine the extent of their infectivity (Caughey).

The critical question of prion infectivity was addressed by using a novel method to bioassay prions that are bound to electron microscope grids and 3-dimensional analysis of multiple prion strains showing that authentically infectious ex vivo prions have a more complex assembly state that consists of rods formed from short, paired double-helical fibrils that have so far not been produced synthetically.

Recent findings that amyloid-β pathology may have been transmitted in humans by therapeutic injection with human cadaveric pituitary growth hormone many years earlier suggest that cerebral amyloid angiopathy, and possibly AD, may have iatrogenic and sporadic and inherited forms. Additional biochemical analysis of archived vials of implicated batches of cadaveric human growth hormone for amyloid-β and τ, together with the inoculation of suitable mouse models—in which seeding of amyloid-β pathology and cerebral amyloid angiopathy can be induced by injection with extracts of AD brain tissue—may allow for completion of the causal loop and the definition of the conditions under which amyloid-β pathology is transmissible in humans. It remains to be definitively determined what structural features differentiate distinct prion strains and how their propagation leads to neurodegeneration. Furthermore, can an array of real-time quaking-induced conversion assays be developed for various tauopathies, synucleinopathies, and other proteinopathies to facilitate the early and accurate diagnosis of neurodegenerative diseases as well as therapeutic trials on the basis of the measurement of etiological biomarkers? (Collinge, Caughey).

Glossary

AD

Alzheimer’s disease

ApoE

apolipoprotein E

AS

α-synuclein

GFP

green fluorescent protein

HTT

huntingtin

IAPP

islet amyloid polypeptide

LC

low complexity

TTR

transthyretin

AUTHOR CONTRIBUTIONS

G. J. Morgan and J. N. Buxbaum attended the conference and contributed equally to the writing and editing of the summary.