|Home | About | Journals | Submit | Contact Us | Français|
Aβ (amyloid beta peptide) is an important contributor to Alzheimer’s disease (AD). We modeled Aβ toxicity in yeast by directing the peptide to the secretory pathway. A genome-wide screen for toxicity modifiers identified the yeast homolog of phosphatidylinositol binding clathrin assembly protein (PICALM) and other endocytic factors connected to AD whose relationship to Aβ was previously unknown. The factors identified in yeast modified Aβ toxicity in glutamatergic neurons of Caenorhabditis elegans and in primary rat cortical neurons. In yeast, Aβ impaired the endocytic trafficking of a plasma membrane receptor, which was ameliorated by endocytic pathway factors identified in the yeast screen. These links between Aβ, endocytosis, and human AD risk factors can be ascertained using yeast as a model system.
Yeast cells lack the specialized processes of neuronal cells and the cell-cell communications that modulate neuropathology. However, the most fundamental features of eukaryotic cell biology evolved before the split between yeast and metazoans. Yeast studies of the cell cycle, DNA damage repair and checkpoints produced pivotal advances in cancer biology (1). More recently, the conservation of protein-homeostasis networks, vesicular trafficking, mitochondrial biology, autophagy, and apoptosis facilitated the development of yeast models for protein-misfolding pathologies (1). When human diseases impinge on common features of eukaryotic cell biology, yeast’s unequaled toolkit offers an attractive discovery platform, as established for multiple aspects of α-synuclein toxicity (2–7).
Here, we wanted to create a yeast model of cellular toxicities elicited by the amyloid β (Aβ) peptide. According to the still hotly debated “amyloid cascade” hypothesis, Aβ is causal in both sporadic and familial Alzheimer’s Disease (AD) (8). The oligomeric forms of the peptide appear to be the most toxic (9–12). Similar toxic oligomers, formed by unrelated proteins but all recognized by the same conformation-specific antibody, are associated with other neurodegenerative diseases and with yeast prions (13, 14). Thus, the toxicity of such oligomers is an ancient protein-folding problem.
In addition to Aβ, neurofibrillary tangles (NFTs) of tau, a microtubule-binding protein, are hallmarks of AD pathology (15). Aβ seems to act upstream of tau (16, 17). Genetic AD risk factors are now being identified through genome-wide association studies (GWAS), but their relationship to Aβ is unknown.
The most toxic form of Aβ, Aβ 1-42, is generated by proteolytic cleavage of APP, the transmembrane amyloid precursor protein (18, 19). APP processing occurs in the secretory pathway, which releases Aβ into the trans-Golgi, endosomal compartments, and extracellular space. Aβ then interacts with the plasma membrane and is subject to endocytosis and further vesicular trafficking (18). To recapitulate this multi-compartment trafficking in yeast, we fused an ER targeting signal to the N-terminus of Aβ 1-42 (referred to as ssAβ 1-42, Fig. 1A). Without an ER retention signal, after cleavage of the signal sequence Aβ 1-42 should simply transit through the secretory pathway to the plasma membrane (20). The yeast cell wall will restrain secreted peptides from diffusing into the culture medium, allowing Aβ to interact with the plasma membrane, undergo endocytosis, and thereby transit through endocytic compartments potentially relevant to AD (Fig. S1A).
When expressed from a galactose-inducible (GAL1) promoter and a multi-copy plasmid, ssAβ 1-42 decreased cell growth (Fig. 1A). Using the same plasmid, Aβ 1-40 was less toxic, as were Pdi1 (an ER resident protein), BPTI (a small protein commonly used to study secretion), and even BPTIC51A (a variant that misfolds in the ER (21)) (Fig. 1A).
For genetic screens, strains with uniform stable ssAβ 1-42 expression were constructed by integrating tandem copies in the genome (Fig. S1B). We targeted a locus where insertions have no deleterious consequences and selected strains that grew slowly upon galactose induction, but with no major increase in lethality (Fig. S1C and Table S1). The peptide produced was of the expected size for processed Aβ (Fig. S1D), verified by mass spectrometry (Fig. S1E). Localization to secretory compartments was confirmed by immunofluorescence (Fig. 1B).
When lysates were not subjected to boiling, which disrupts oligomeric species, we detected Aβ oligomers on Bis-Tris gels (Fig. S1D). These forms reacted more strongly with the antibody NAB61, which preferentially recognizes toxic Aβ oligomers in AD patients (Fig. 1C) (10). These species disappeared upon boiling in lithium dodecyl sulfate (LDS) buffer. Assaying lysates by size-exclusion chromatography with a monoclonal IgM anti-Aβ antibody specific for Aβ oligomers detected a broad range of oligomeric species (Fig. 1D and Fig. S1F) (22). Eliminating preparation artifacts, these were not seen when purified monomeric Aβ peptide was added to control cultures prior to lysis (Fig. 1D). In strains that produced the same levels of Aβ 1-40 and 1-42 monomer after boiling (Fig. S1G), oligomers were much more abundant for Aβ 1-42 prior to boiling (Fig. 1E). Thus, oligomeric Aβ forms contribute to toxicity in yeast, as in neurons.
We transformed a screening strain with an overexpression library of 5532 full length open reading frames (ORFs) (~90% of yeast ORFs) under control of the same promoter used for ssAβ 1-42 (Fig. S2A). Individual transformants were arrayed in media that prevented induction of either ssAβ 1-42 or the library constructs, then plated (four replicates each) onto several types of inducing media, chosen to support different levels of mitochondrial respiration (Fig. S2B; (23), Table S2). Intermediate levels of Aβ toxicity enabled the identification of enhancers and suppressors in the same screen (Fig. S1B; (23), Table S2). Genes that decreased or increased growth (Fig. S2B) were retested in an independently derived screening strain. Secondary screens eliminated genes that simply altered expression of Aβ from the GAL1 promoter (Fig. S3) or growth in the absence of Aβ.
We identified 23 suppressors and 17 enhancers (Table S2). Only a few modifiers were strongly affected by the state of respiration (Table S2). The screen hits comprised a wide range of cellular functions. Numerous hits had sequence similarity to human genes, and twelve had very clear human homologs (determined by HomoloGene or by analogous functionality [SLA1 – SH3KBP1] (24)) (Table 1). We focused further analysis on these.
Three of these twelve genes had functions related to clathrin-mediated endocytosis (YAP1802, INP52 & SLA1; P=3.89e-4) and seven were functionally associated with the cytoskeleton (YAP1802, INP52, SLA1, CRM1, GRR1, KEM1 & RTS1; P=6.06e-8). None were identified in our previous screen for modifiers of α-syn toxicity (5, 7), establishing their specificity for the type of toxicity caused by Aβ 1-42.
Several human homologs of our yeast hits had connections to human AD risk factors, particularly those involved in clathrin-mediated endocytosis (Table 1). The human homolog of yeast YAP1802, PICALM, is one of the most highly confirmed risk factors for sporadic AD (25, 26). Another AD risk factor, BIN1, is involved in synaptic vesicle endocytosis and is believed to interact with synaptojanin, the human homolog of yeast INP52 (27). The functional homolog of yeast SLA1, SH3KBP1 (28), directly interacts with the risk factor CD2AP (29, 30). CD2AP links endocytosis to cytoskeletal dynamics and our other major class of screen hits.
To assess the potential clinical relevance of other screen hits with highly conserved human homologs we examined association with AD susceptibility using data from a published family-based GWAS (31, 32). Using a family-based association test, we discovered a suggestive association of XPO1 (CRM1 homolog, rs6545886, P=0.003) with AD susceptibility ((23), Table S3).
Next, we leveraged genotyping with extensive clinical and pathological data from two large epidemiological studies of aging, cognition, and AD (33, 34) ((23), Table S4–7). Using a quantitative summary measure of global AD pathologic burden available in these cohorts, counting both amyloid plaques and NFTs, we found that two additional loci identified by our yeast screen, ADSSL1 (ADE12 homolog, rs1128880, P=0.001) and RABGEF1 (VPS9 homolog, rs17566701, P=0.002) showed evidence of association with AD neuropathology (Table S7). Both loci also harbored suggestive association signals with episodic memory decline (Table S6). Thus, our yeast screen connects multiple human AD risk factors, and suggested risk factors, to Aβ toxicity.
To directly test our modifiers for effects on Aβ toxicity in neurons, we created a transgenic C. elegans model that expressed Aβ 1-42 in glutamatergic neurons, a neuronal subtype particularly vulnerable in AD. (A previous model expressed Aβ in body-wall muscles (35).) We used the eat-4 promoter, which regulates the BNPI glutamate transporter and, again, targeted Aβ to the secretory pathway (35). The Aβ transgene was integrated into chromosomal DNA to ensure the same Aβ 1-42 expression levels in all animals.
C. elegans has highly stereotyped cell lineages. Wild-type worms invariably have five glutamatergic neurons in their tails, which we visualized with eat-4-regulated GFP. Crossing worms expressing ssAβ to worms carrying this reporter resulted in the loss of GFP-marked glutamatergic neurons in an age-related manner: at day three only 48% of worms had five intact glutamatergic neurons, and at day seven only 25% did (Fig. 2A).
To test the effects of our screen hits (Table 1) we established three independent transgenic worm lines for each gene, again expressing the protein from the eat-4 promoter. Unrelated control proteins mCherry and LacZ had no effect on Aβ toxicity. All three C. elegans homologs of the genes involved in clathrin-mediated endocytosis, unc-11, unc-26, and Y44E3A.4, increased the percentage of worms with five intact glutamatergic neurons (Fig. 2).
Finally, we tested four hits, three suppressors and one enhancer, involved in a diverse array of pathways. The yeast RTS1 gene encodes a phosphatase regulatory subunit that controls several stress-response pathways. The ADE12 gene product catalyzes the first step in the synthesis of adenosine monophosphate from inosine 5-monophosphate. The C. elegans homologs of each gene suppressed the Aβ-induced loss of glutamatergic neurons (Fig. 2A). We were unable to clone the worm homologs of GRR1, VPS9, or CRM1. However, CRM1 encodes a highly conserved nuclear export receptor. Expression of CRM1’s human homologue, XPO1, protected nematode glutamatergic neurons from Aβ (Fig. 2A). PBS2/MAP2K4, a MAP kinase involved in stress responses, increased neuronal loss (Fig. 2).
The effect of each gene was statistically significant (p < 0.05) for both the modest neuronal toxicity evinced at three days and the more severe toxicity at seven days. Importantly, the modifiers did not alter the levels of Aβ mRNA, which we tested by semi-quantitative RT-PCR (Fig. S5). Thus, every gene we tested in C. elegans glutamatergic neurons modified Aβ toxicity in the same direction (suppression vs. enhancement) as they did in yeast.
PICALM is one of the most highly validated AD risk factors and its efficacy in our yeast and nematode models strongly suggests it alters Aβ toxicity. Modeling this in cultured mammalian neurons is not trivial, because any Aβ peptide expressed in the secretory pathway would simply diffuse away from the cell. Exogenously applied, preformed oligomeric Aβ species are often employed to model toxicity (13, 36), but their relevance is highly debated. We reasoned that a positive result for the highly validated AD risk factor PICALM might not only validate this assay but confirm the role of PICALM in Aβ detoxification.
We analyzed cortical neurons, a neuronal population particularly relevant to AD. Embryos from female rats with timed pregnancies were harvested at 18 days of gestation. Cortices dissected from these embryos were dissociated, plated and cultured for up to 21 days (23). The production of toxic Aβ oligomers is notoriously variable. We prepared oligomers according to several published methods, characterized them biochemically, and tested them for producing consistent levels of toxicity in our cortical neuronal preparations (Fig. 3, Fig. S6 and Fig. S7). The loss of toxicity when the same samples were allowed to form Aβ fibers (Fig. S6A and S6C) confirmed the importance of the oligomeric species (Fig. S6D).
Next, we infected neurons with lentiviruses engineered to express GFP or PICALM. When Aβ oligomers were added to these neurons, GFP had no effect but PICALM partially rescued the cells from Aβ induced cell death in a dose-dependent manner (Fig. 3). Rescue was significant whether assayed by cellular ATP levels (Fig. 3A) or by counting the number of MAP2 positive neurons (Fig. 3B and Fig. 3C). As previously described for midbrain neurons (7), we found that RAB1 protected cortical neurons from α-syn toxicity when this protein was expressed intracellularly by viral transfection. However, RAB1 was ineffective against Aβ oligomers, confirming the specificity of PICALM for the type of toxicity caused by Aβ oligomers (Fig. S7).
The role of PICALM in AD is unknown, but it has been postulated to affect disease by modifying APP trafficking (25). However, our experiments in yeast, nematode and rat neurons clearly establish PICALM as a modifier of Aβ toxicity itself. To investigate the mechanism for PICALM and the other modifiers that affect clathrin-mediated endocytosis, we returned to yeast.
One possibility is that clathrin-mediated endocytosis modulates Aβ toxicity simply by shunting toxic Aβ species to the lysosomal/vacuolar system for degradation. However, immunoblotting of yeast lysates indicated these modifiers had little effect on Aβ levels (Fig. S4). Alternatively, if Aβ specifically perturbs endocytic homeostasis, up-regulation of this pathway might ameliorate the defect.
To determine if Aβ altered clathrin distributions, we used a strain in which endogenous clathrin light chain (Clc1) was tagged with GFP, a fully functional fusion. A control strain exhibited the expected distribution of Clc1-GFP foci (37). Aβ perturbed clathrin localization, increasing both the number and brightness of foci, but decreasing their average size (Fig. 4A).
Such a pattern might indicate a defect in clathrin-mediated secretion as well as in endocytosis. To test effects of Aβ on secretion, we used a halo assay for secretion of the α-factor mating pheromone. As a control, we also tested the effects of α-syn expression, which produces a strong defect in secretion (7) (Fig. 4B). Unlike α-syn, Aβ did not inhibit secretion (Fig. 4B).
To assess the impact of Aβ on clathrin-mediated endocytosis specifically, we examined the well-characterized substrate Ste3, a member of the highly conserved G-protein coupled receptor family. Ste3 is targeted to the plasma membrane. In the absence of its ligand, the yeast mating-factor, it is constitutively endocytosed and degraded in the vacuole (38). As expected, a Ste3-YFP fusion was primarily localized to the vacuole in a control strain (Fig. 4C). In Aβ-expressing strains, endocytic trafficking of Ste3-YFP was profoundly perturbed and the protein was localized to numerous foci ((23), Fig. 4C). Aβ expression resulted in a reduction of vacuolar organelle size without a disruption in morphology, consistent with reduced delivery of cargo to this organelle. Finally, we tested the effects of the three Aβ toxicity suppressors that function in endocytic trafficking, YAP1802, INP52, and SLA1. Each partially reversed the defect in Ste3-YFP trafficking (Fig. 4C).
Our yeast model allowed us to conduct an unbiased screen of an entire genome for modifiers of Aβ toxicity. The emergence of three different genes involved in the process of clathrin-mediated endocytosis from nearly 6,000 tested ORFs confirms that the Aβ peptide in our model is trafficking through the secretory compartments as expected. More importantly, the ability of endocytic genes to rescue Aβ toxicity, together with the effects of Aβ on clathrin localization and the trafficking of a G protein-coupled receptor, establish that within these highly diverse organisms clathrin-mediated endocytosis is a critical point of vulnerability to Aβ.
Aβ oligomers have been reported to increase endocytosis in cultured cells (39), and human-induced neuronal cells derived from the fibroblasts of AD patients exhibit defects in endocytosis (40). Mechanistically, in our Aβ-expressing cells, the increased number of clathrin foci, the internalized foci of Ste3, and the effects of genetic modifiers on vacuolar localization all suggest that Aβ affects this pathway by interfering with the ability of endocytosed transmembrane receptors to reach their proper destinations.
PICALM, as well as two genes whose protein products (BIN1 and CD2AP) interact with hits from our screen are AD risk factors. Given the diversity of pathologies however, their connection to Aβ toxicity was unknown. Our work in yeast, nematodes, and rat cortical neurons clearly places these factors within the Aβ cascade, linking Aβ to the genetics of sporadic AD.
Neurons are particularly vulnerable to perturbations in the homeostasis of endocytosis, because they must constantly recycle both neurotransmitters and their receptors (41). Aβ interacts with, and alters signaling by, a variety of neuronal receptors (42). We propose that the conformational flexibility of these oligomers allows them to interact rather promiscuously with conformationally flexible unliganded receptors, which, in turn, disrupts endocytic homeostasis.
Our yeast screen also identified seven conserved genes functionally associated with the cytoskeleton. Because yeasts do not express tau, our findings may indicate that the connection between Aβ toxicity and the cytoskeleton is more deeply rooted than tau alone, probably involving clathrin-mediated endocytosis. In analyzing human GWAS data we also uncovered suggestive associations between AD and three other genes, XPO1, ADSSL1, and RABGEF1, and confirmed their Aβ relationships in yeast and nematode.
The treatments available for AD are few and their efficacy limited. Determining how best to rescue neuronal function in the context of the whole brain is a problem of staggering proportions. Our yeast model provides a tool for identifying genetic leads, investigating their mechanisms of action, and screening for genetic and small molecule modifiers of this devastating and etiologically complex disease.
We thank L. Chibnik, B. Keenan and D. Ng for helpful discussions. D. Wittrup, V. Lee, M.Vidal and C. Link for materials, J. Corneveaux, M. Huentelman, and other Translational Genomics investigators for assistance with the human study cohorts and participants in the Religious Orders Study and the Rush Memory and Aging Project. This work was supported by an HHMI Collaborative Innovation Award, NRSA fellowship F32 NS067782-02, the Cure Alzheimer’s Fund, NIH grants K08AG034290, P30AG10161, R01AG15819, and R01AG17917, the Kempe foundation and Alzheimerfonden.
Supplemental data are available in the Supplemental Online Material.
Supplemental Online Material
Materials and Methods
Figs. S1 to S7
Tables S1 to S7