ApoE-associated modulation of neuroprotection from Aβ-mediated neurodegeneration in transgenic Caenorhabditis elegans

ABSTRACT Allele-specific distinctions in the human apolipoprotein E (APOE) locus represent the best-characterized genetic predictor of Alzheimer's disease (AD) risk. Expression of isoform APOEε2 is associated with reduced risk, while APOEε3 is neutral and APOEε4 carriers exhibit increased susceptibility. Using Caenorhabditis elegans, we generated a novel suite of humanized transgenic nematodes to facilitate neuronal modeling of amyloid-beta peptide (Aβ) co-expression in the context of distinct human APOE alleles. We found that co-expression of human APOEε2 with Aβ attenuated Aβ-induced neurodegeneration, whereas expression of the APOEε4 allele had no effect on neurodegeneration, indicating a loss of neuroprotective capacity. Notably, the APOEε3 allele displayed an intermediate phenotype; it was not neuroprotective in young adults but attenuated neurodegeneration in older animals. There was no functional impact from the three APOE isoforms in the absence of Aβ co-expression. Pharmacological treatment that examined neuroprotective effects of APOE alleles on calcium homeostasis showed allele-specific responses to changes in ER-associated calcium dynamics in the Aβ background. Additionally, Aβ suppressed survival, an effect that was rescued by APOEε2 and APOEε3, but not APOEε4. Expression of the APOE alleles in neurons, independent of Aβ, exerted no impact on survival. Taken together, these results illustrate that C. elegans provides a powerful in vivo platform with which to explore how AD-associated neuronal pathways are modulated by distinct APOE gene products in the context of Aβ-associated neurotoxicity. The significance of both ApoE and Aβ to AD highlights the utility of this new pre-clinical model as a means to dissect their functional inter-relationship. This article has an associated First Person interview with the first author of the paper. Summary: We report new humanized nematode models of amyloid-beta and apolipoprotein E gene expression, examining readouts for neurodegeneration, behavior, cellular function and survival, to elucidate the relative contributions of the proteins to the underlying pathology of Alzheimer's disease.


INTRODUCTION
Alzheimer's disease (AD), characterized by the formation of insoluble amyloid-beta peptide (Aβ) plaques in the brain, accounts for nearly 70% of all late-life dementia. Although the causes, whether genetic or environmental, are not clearly defined, it is evident that the most predictive genetic association is variation in the gene encoding apolipoprotein E (ApoE). Although estimates vary based on study and ethnicity, ∼40% of AD cases harbor the ε4 allele of APOE (Spinney, 2014). This allele is a significant risk factor for late-onset AD, where two copies of APOEε4 increases AD risk up to 15-fold relative to APOEε3. The APOEε2 allele appears to provide protection against AD via a mechanism that consists of more than the absence of the APOEε4 allele (Corder et al., 1994;Talbot et al., 1994). Indeed, there may be opposing actions of the APOEε2 and APOEε4 alleles, which would not be unprecedented, as APOEε2 and APOEε4 appear to have opposing activities in lipidation and aggregate stabilization (Hu et al., 2015). Despite this correlation, the mechanisms by which differences in APOE allelic function modify AD risk are not entirely understood.
There are many mechanisms proposed to explain how APOEε4 increases AD risk, including altered glucose and lipid metabolism. Most commonly, however, Aβ-dependent effects are considered within the context of the APOE alleles, where neurotoxicity and aggregation are examined. For example, mammalian models have yielded significant information on how ApoE and Aβ interact to affect cellular function and animal behavior, but the scale and complexity of the mammalian nervous system frustrate examination of quantifiable effects on individual neurons and their functional connectivity. The nematode Caenorhabditis elegans has been employed to generate models of neurodegenerative disorders, including AD (Griffin et al., 2017), Huntington's disease (Muñoz-Lobato et al., 2014) and Parkinson's disease (Martinez et al., 2017a). Because C. elegans is the only animal for which a connectivity map of its entire nervous system exists, it provides an unparalleled platform for the examination and quantitative characterization of neural interactions. Further, the genetic tractability of C. elegans offers a model receptive to genetic manipulation and transgenics. Importantly, specific worm models have proven highly predictive of both genetic and small molecule modifier results obtained in mammalian systems, including genome-wide association studies and induced pluripotent stem cells from patients (Cooper et al., 2006;Matlack et al., 2014;Mazzulli et al., 2011;Su et al., 2010;Tardiff et al., 2013Tardiff et al., , 2017Treusch et al., 2011).
Here, we present new neuronal models to assay ApoE activity in vivo that consist of nematodes expressing human APOEε2, APOEε3 or APOEε4 along with Aβ. Glutamate is a major excitatory neurotransmitter in the brain, and dysregulation of the glutamatergic system can lead to excitotoxicity, which, when chronic, has been hypothesized to play a role in neurodegeneration (Lewerenz and Maher, 2015). Because the glutamatergic circuitry is severely disrupted in the brains of AD patients (Francis et al., 1993;Greenamyre et al., 1988), the eat-4 (glutamate transporter) promoter was chosen for glutamatergic neuron-specific expression of Aβ and the respective APOE alleles. Effects on neuronal integrity were examined through quantitative fluorescent imaging of neurodegeneration and behavioral assays. Additionally, we modulated neurodegenerative effectors via pharmacological treatment and RNA interference (RNAi). By combining neuronal expression of APOE alleles with a transgenic nematode model of human Aβ toxicity, we can further understand the clinically significant relationship between ApoE and Aβ in neurotoxicity. Using these C. elegans models of progressive Aβ-mediated neurodegeneration, a strong attenuation of Aβ-mediated toxicity is revealed by the APOEε2 allele, as well as a modest, yet significant, intermediate protection phenotype by APOEε3 as animals age, in vivo. Strikingly, the neuroprotective activity of ApoE was abolished in animals co-expressing Aβ and APOEε4. Furthermore, this shows that the allelic profile reflects the well-established clinical observation of ApoE-associated susceptibility. Pharmacological and post-transcriptional manipulation further demonstrate differential activities of APOE alleles observable through multiple phenotypic outputs. Though limited as an invertebrate system, C. elegans provides a platform that accelerates attainment of a more mechanistic understanding of how ApoE protein variants function to modulate neuronal degeneration and establishes a new pre-clinical model of AD to accelerate future drug discovery.

APOE allele-selective mitigation of Aβ-mediated neurodegeneration
The Aβ peptide is the product of sequential cleavage of the amyloid precursor protein (APP) either at the cell surface or within endosomes. Cleavage of APP is known to produce multiple peptide products, such as Aβ(1-40) and Aβ(1-42); however, the Aβ(1-42) peptide is the most toxic. Extracellular deposition of insoluble Aβ plaques is a pathological hallmark of AD, but intracellular Aβ has been shown to be far more toxic (Burdick et al., 1992;Cha et al., 2012;Esbjörner et al., 2014;Hu et al., 2009;Kounnas et al., 1995;Li et al., 2012;Liu et al., 2013b;Naj et al., 2011;Nakagawa et al., 2000;Okoshi et al., 2015;Reinders et al., 2016;Snyder et al., 2005;Takahashi et al., 2002;Treusch et al., 2011;Ulrich, 2015;Wang et al., 2000;Yang et al., 1998;Zhao et al., 2015). To reproduce the intracellular accumulation of Aβ in C. elegans, Aβ was cloned with promoters for tissue-specific multicopy expression and scored for toxicity. In C. elegans muscle expression models of Aβ toxicity, Aβ was found to form plaques (Link et al., 2001) and intramuscular inclusions (Fay et al., 1998;Link, 1995), and to induce paralysis via cytotoxicity (Dostal and Link, 2010;Fonte et al., 2002). Furthermore, we have shown that expression of Aβ in glutamatergic neurons results in progressive, age-dependent, neurodegeneration modulated by endocytic and endosomal regulators, including the established AD modifier PICALM (Griffin et al., 2018;Treusch et al., 2011), and is amenable to pharmacological treatment (Matlack et al., 2014;Tardiff et al., 2017). To examine the relationship between ApoE and Aβ, we utilized a C. elegans model in which an Aβ(1-42) construct, hereafter referred to as Aβ, was cloned for expression in the glutamatergic neurons and neurodegeneration was quantified with precision in the five glutamatergic neurons in the tail (Matlack et al., 2014;Treusch et al., 2011). Expression in the glutamatergic neurons was achieved using the promoter for the glutamate transporter eat-4, which does not significantly change in expression across larval stages (Lee et al., 1999).
To model ApoE activity in C. elegans, complementary DNAs (cDNAs) encoding the three distinct human APOE alleles (APOEε2, APOEε3 and APOEε4) were recombined with the artificial constitutive her-1 secretion signal, and expression was driven by the glutamatergic neuron-specific eat-4 promoter. These three constructs were microinjected into wild-type (N2) animals, integrated into the genome and crossed with Aβ-expressing animals after outcrossing. Overexpression of Aβ induced neurodegeneration of glutamatergic neurons (Fig. 1A), as has been observed previously (Griffin et al., 2018;Tardiff et al., 2017;Treusch et al., 2011), while expression of APOEε2, APOEε3 or APOEε4 in glutamatergic neurons did not impact neurodegeneration in the absence of Aβ (Fig. 1A).
Because the ε2 allele is associated with protective phenotypes (Bu, 2009;Liu et al., 2013a), we hypothesized that co-expression of APOEε2 with Aβ would attenuate Aβ-induced neurodegeneration. At both days 3 and 7 post-hatching, nearly 100% of all animals expressing GFP alone have all five normal glutamatergic neurons. However, when co-expressed with Aβ, the APOEε2 allele suppressed Aβ-mediated neurodegeneration by ∼30% at days 3 and 7 post-hatching ( Fig. 1B,C). Furthermore, because the APOEε3 allele appears functionally neutral in humans, and ε4 is associated with increased neurotoxicity (Bu, 2009;Corder et al., 1993;Huang and Mucke, 2012;Liu et al., 2013a), we hypothesized that APOEε3 would elicit marginal or no neuroprotective effect, while APOEε4 would increase neurodegeneration. At day 3, there was no statistically significant difference in neurodegeneration between animals expressing Aβ alone or co-expressing APOEε3, but, at day 7, APOEε3 significantly reduced Aβ-mediated neurodegeneration by ∼10%, which was significantly less than the protection afforded by APOEε2. This protection was also significantly greater than that provided by the ApoEε4 strain, in which there was no change in neurodegeneration at either day 3 or day 7 (Fig. 1B,C). Although co-expression with APOEε4 did not enhance neurodegeneration in this model, it was not statistically different from Aβ alone at days 3 or 7 (P=0.9579, P=0.9369, respectively), but was significantly different from the Aβ+ApoEε2 strain at day 7, thereby confirming earlier reports that there may be alternative mechanisms of action between these two alleles (Corder et al., 1994;Talbot et al., 1994).
As a secondary readout for glutamatergic neuronal dysfunction, we turned to a behavioral assay, as altered mechanosensory touch response is indicative of glutamatergic neuron dysfunction. In C. elegans, a pair of glutamatergic tail neurons have processes extending from the tail to the mid-body, to control forward escape in response to posterior gentle touch (Chalfie et al., 1985). In worms expressing Aβ in glutamatergic neurons, this posterior gentle touch response is defective (Fig. 1D). However, in worms expressing APOE alleles without Aβ, gentle touch response is not defective, indicating that the APOE alleles on their own are not pathogenic (Fig. 1D). When worms co-overexpressing Aβ and ApoEε2 were assayed in the touch response assay, there was a significant mitigation of this mechanosensory defect (Fig. 1E). Recovery was not observed by ApoEε3 or ApoEε4 co-expression since they were not significantly reduced compared with the Aβ+ApoEε2 strain (Fig. 1E). These data also suggest that, since there is a significant difference between Aβ+ ApoEε2 and Aβ alone, but not between Aβ alone and either the Aβ+ ApoEε3 or Aβ+ApoEε4 strains, there might be alternative mechanisms of action among these alleles that can be teased out using this assay. For example, although Aβ+ApoEε3 appeared to have a neuroprotective effect at later stages (day 7; Fig. 1B), the seemingly protected neurons in animals co-expressing ApoEε3 demonstrated reduced mechanosensory sensitivity. This suggests that ApoEε3 may confer moderate protection of neuronal structure that does not ameliorate loss of neuronal function by Aβ.
To ensure that the APOE-allele-specific phenotypes we observed are functionally driven and are not simply due to transgenic expression level differences, APOEε2, APOEε3 and APOEε4 mRNA levels were quantified by reverse-transcription quantitative polymerase chain reaction (RT-qPCR: Fig. 1F). There were no statistically significant differences in relative normalized APOE transcripts between ApoEε2 and ApoEε4 samples (P=0.2107) or ApoEε3 and ApoEε4 samples (P=0.1280). However, APOEε3 transcripts were significantly higher than APOEε2 transcripts (P=0.0127). Taken together with the neurodegeneration analyses, these results indicate that ApoEε2 neuroprotection is likely not due to disproportionate overexpression compared with ApoEε3.

APOE-allele-specific modulation of calcium homeostasis
To observe whether ApoE confers a physiologically relevant effect in our model, we examined the relationship between calcium homeostasis, Aβ and ApoE. In rat hippocampal neurons and chick sympathetic ganglia, ApoEε2 and ApoEε3 have no effect on N-methyl-D-aspartate (NMDA)-mediated calcium influx, but incubation with ApoEε4 results in massive NMDA-mediated calcium influx (Hartmann et al., 1994;Qiu et al., 2003;Tolar al., 1999). In cultured mouse cortical neurons, the opposite effect is observed, wherein NMDA-mediated calcium influx is inhibited by ApoEε4 but exacerbated by ApoEε2 and ApoEε3 (Chen et al., 2010). Nevertheless, in both mammalian scenarios the functional impact of Aβ neurotoxicity was not assessed. To test the relationship between calcium, ApoE and Aβ in our model, we utilized thapsigargin, which increases cytosolic calcium concentrations by inhibiting the endoplasmic reticulum (ER) Ca 2+ -ATPase sca-1. Indeed, calcium influx induced by APOE has been partially attributed to ER calcium stores (Tolar et al., 1999). Animals expressing GFP alone were not impacted by thapsigargin treatment ( Fig. 2A). Thapsigargin treatment of animals expressing Aβ attenuated neurodegeneration by nearly 20% compared with vehicle control at days 3 and 7 ( Fig. 2A). There was no additive reduction in neurodegeneration by thapsigargin treatment with expression of either APOEε2 or APOEε3 at either day 3 or 7, suggesting that ApoE may potentially allay neurodegeneration in the same pathway as ER-derived calcium. As previously observed, the effect of ApoEε4 was significantly reduced when compared with ApoEε2 at both days 3 (P<0.0001) and 7 (P=0.0177), but together with thapsigargin, ApoEε4 showed protection similar to Aβ+ApoEε2 at both time points. No effect was observed from treatment of GFP animals expressing APOE without Aβ (Fig. 2B). These data suggest that ApoEε2 is neuroprotective through an interaction with ER-derived calcium and that this interaction is lost with the ApoEε4 protein variant.
To confirm that the observed effect by thapsigargin is related to its inhibition of sca-1, a Ca 2+ ATPase and target of thapsigargin, we generated a conditional RNAi-sensitive strain, in which RNAi is restricted to the glutamatergic neurons. This strain was then crossed into the Aβ and Aβ+ApoE backgrounds, so that genetic targets can be depleted with co-expression of Aβ and ApoE (Table 1). As previously observed, thapsigargin treatment reduced neurodegeneration in animals expressing Aβ alone and co-expressing Aβ+ApoEε4, but not in either Aβ+ApoEε2 or Aβ+ApoEε3 animals (Fig. 3). Depletion of sca-1 in Aβ alone was neuroprotective when compared with empty vector (EV) control, but there was no additional protection conferred by a combination of sca-1 RNAi and thapsigargin treatment, suggesting that protection by thapsigargin, redundant with ApoEε2 and ApoEε3, is not independent from its target, sca-1. In contrast, sca-1 RNAi was protective in the backgrounds expressing Aβ alone and Aβ+APOEε4. Taken together, these data suggest that there is a genetic relationship between APOEε2 and sca-1 that is lost in the APOEε4 genetic background.

Attenuation of neurodegeneration by starvation is independent of ApoE function
Starvation and caloric restriction increase health and lifespan through multiple pathways that overlap with significant conservation among yeast, C. elegans, Drosophila, rodents and primates (Fontana et al., 2010). Furthermore, dietary restriction reduces Aβ toxicity (Steinkraus et al., 2008). We therefore hypothesized that starvation would attenuate Aβ-mediated neurodegeneration and tested its effect in the context of the three distinct APOE alleles. To test this, synchronized embryos were hatched onto unseeded plates and incubated for 24 h, after which time they were transferred to normal (nematode growth medium; NGM) nematode plates seeded with Escherichia coli. Although early-L1-stage larval starvation attenuated neurodegeneration as expected in worms expressing Aβ alone, this protective effect was also shared indiscriminately with animals co-expressing any of the APOE alleles (Fig. 4A). These data suggest that, in modulating its effects on neuron survival, ApoE operates outside of this starvation-induced rescue response, thus excluding this mechanism of dietary restriction as an APOE-allelespecific means of modulating neurotoxicity. However, alternative dietary restriction regimens in C. elegans have been found to extend lifespan through parallel or overlapping pathways (Greer and Brunet, 2009). The extension of lifespan by dietary deprivation was dependent on heat shock factor 1 (hsf-1), while AMP-activated protein kinase 2 (aak-2) and FOXO/daf-16 were required for lifespan extension by the absence of peptone. Because the dietary deprivation regimen begins dietary restriction at day 2 of adulthood (day 5 post-hatching), animals were washed off food at day 5 post-hatching and moved to unseeded plates until analysis at day 7. Although dietary deprivation reduced neurodegeneration in the background expressing Aβ alone (Fig. 4B), dietary deprivation provided no statistically significant rescue in the Aβ+ApoEε2, Aβ+ApoEε3 or Aβ+ApoEε4 backgrounds, suggesting that the ApoE protein, irrespective of allelic variation, might generally interfere with hsf-1-associated protective mechanisms. In contrast, there was no statistically significant change in neurodegeneration in animals subjected to the absence of peptone regimen at either days 3 or 7 (Fig. 4C).
Survival shortened by Aβ is rescued by ApoEε2 and ApoEε3, but not ApoEε4 Because AD is an age-related disease and APOEε4 homozygosity is associated with earlier onset of AD (Bu, 2009;Corder et al., 1993;Liu et al., 2013a), we examined how the relationship between Aβ and ApoE in the glutamatergic neurons affected survival with aging. Additionally, the Mantel-Cox/log-rank method was used for survival analyses, as it assigns equal weights in statistical calculations for the entire pattern or path of the curve, not just the median or maximum values displayed. Both wild type (Bristol N2) and animals expressing GFP alone exhibited similar survival curves that were not significantly different from each other (Fig. 5A). In animals expressing Aβ, survival was significantly reduced (Fig. 5A), suggesting a relationship between glutamatergic neurodegeneration and aging in the C. elegans neuronal model. In animals expressing APOE alleles alone (encoding ApoEε2, ApoEε3 or ApoEε4), the survival curves were similar to the N2 control ( Fig. 5B-D). However, co-expression of Aβ+ApoEε2 or Aβ +ApoEε3 increased survival (Fig. 5E,F), compared with Aβ alone (Fig. 5A). In contrast, co-expression of Aβ+ApoEε4 had no significant effect compared with Aβ alone (Fig. 5G). These data suggest that integrity of the glutamatergic neurons through the aging process, as differentially modulated by the APOE alleles in the presence of Aβ, affects whole-animal survival.

DISCUSSION
The APOEε4 allele is the strongest risk factor associated with late-onset AD, yet determining precisely how the APOE alleles differentially modulate Aβ toxicity and neuronal behavior remains unresolved. An expedient examination of the relationship between the APOE alleles and Aβ requires a model system in which neuronal dysfunction and loss are amenable to both genetics and tractable neuronal outputs. Our C. elegans model of Aβ-induced neurodegeneration in glutamatergic neurons recapitulates mammalian and cell culture models for AD-associated gene analyses (Griffin et al., 2017;Matlack et al., 2014;Treusch et al., 2011). Furthermore, genes associated with AD have C. elegans orthologs (Mukherjee et al., 2017;Vahdati Nia et al., 2017). Here, we debut a model of ApoE activity in our established neuronal C. elegans background and suggest that it can be exploited to examine the relationship between ApoE and Aβ for neuronal behavior, integrity and proteotoxicity.
Induction of autophagy by thapsigargin is reported to occur through stimulation of ER stress (Bernales et al., 2006;Ding et al., 2007;Høyer-Hansen et al., 2007;Kouroku et al., 2007). Although our data suggest divergent participation in protection between autophagy and ApoE, they do not preclude the possibility of ER stress. Whether ApoEε2 yields protection by inducing ER stress is unclear. It is, however, unlikely, considering that ApoEε4 has been shown to significantly increase ER stress compared with ApoEε3 in mice (Verghese et al., 2013;Zhong et al., 2009). In such a paradigm, increased ER stress by ApoEε2 would presumably recapitulate ApoEε4-associated phenotypes. Further, the protective effect of ApoEε2 might not be attributed to differences in ER stress induction, as ApoEε2 and ApoEε3 have been reported to have no difference in the expression of ER stress targets IRE1 (also known as ERN1), BiP (also known as HSPA5) and CHOP (also known as DDIT3), which increase, instead, with ApoEε4 expression (Verghese et al., 2013). Rather, stress and injury typically increase the expression of ApoE in brains . The effect of increased ApoE expression during stress might be due to mitochondrial interactions, because RNA sequencing of mouse brains revealed Apoe-allele-specific responses in mitochondrial gene expression (Babenko et al., 2017;Xu et al., 2006). Indeed, ApoEε3 is less likely to be retained at the ER (Brodbeck et al., 2011), and although the retention of ApoEε2 in the ER has not been reported, the effect of ER retention is due to the S61R present in ApoEε3 and ApoEε4, but not present in ApoEε2, thus making ApoEε2 far less likely to be retained at the ER. Perhaps the additional cysteine residues in ApoEε2 compared with ApoEε3 or ApoEε4 make ApoEε2 an agent of redox stabilization at mitochondria during stress (Yamauchi et al., 2017). Additionally, variations in the translocase of outer mitochondrial membrane 40 (TOMM40) and ApoE are associated with differences in longevity (Lin et al., 2016). Notwithstanding, the interaction between ApoE and the ER stress pathway is poorly understood and deserves to be more explicitly delineated.
Loss or depletion of the insulin signaling receptor, daf-2, doubles lifespan in C. elegans (Kenyon et al., 1993) in a manner that is independent of autophagy (Greer and Brunet, 2009). In C. elegans models of proteotoxicity, loss of daf-2 reduces α-synuclein-mediated neurodegeneration Ray et al., 2014), paralysisinduced poly-Q toxicity (Steinkraus et al., 2008) and paralysisinduced Aβ toxicity (Cohen et al., 2006;Florez-McClure et al., 2007;Steinkraus et al., 2008). Furthermore, loss of daf-2 decreases Aβ toxicity (Steinkraus et al., 2008) by increasing the autophagic clearance of Aβ (Florez-McClure et al., 2007). However, different longevity association pathways are activated in response to diverse dietary restriction regimens (Greer and Brunet, 2009). Peptone absence extends lifespan through aak-2 and FOXO/daf-16, but yielded no change in neurodegeneration, suggesting that daf-2mediated protection observed in other Aβ models (Cohen et al., 2006) might be engaged through downstream mechanisms in parallel with AAK-2 activation of FOXO/daf-16. For example, reduced insulinlike signaling decreased Aβ accumulation by elevating autophagy and lysosome populations (Florez-McClure et al., 2007). In the dietary deprivation model, the extension of lifespan requires hsf-1. In such a model, it is feasible that the complete absence of food during the first larval stage could activate responses controlled by HSF-1 activity independently of expression of any APOE allele (Steinkraus et al., 2008). However, this would stand in opposition to the hsf-1dependent dietary deprivation model that begins starvation 2 days into adulthood (Greer and Brunet, 2009;Steinkraus et al., 2008), but was only protective with Aβ alone (Fig. 4B), suggesting an interaction between ApoE in the dietary deprivation model that does not take place in the L1 starvation model. We show a potent neuroprotective effect of two different dietary restriction regimens that interact differently with ApoE in Aβ toxicity in vivo (Fig. 4A,B). Considering the robust understanding and utility of C. elegans in aging research, this model opens avenues for more thorough examination of the relationships between longevity pathways, ApoE and Aβ. Fig. 4. Starvation of animals during the L1 larval stage attenuates neurodegeneration in all transgenic strains. (A) Synchronized animals were deprived of food during the L1 stage and then assayed for neurodegeneration as young adults at day 3 posthatching. Animals expressing Aβ (UA198; P=0.001) exhibited reduced neurodegeneration. Additionally, animals expressing Aβ+APOEε2 (UA351; P=0.0156), Aβ+ APOEε3 (UA353; P=0.0003) or Aβ+APOEε4 (UA355; P=0.0201) also displayed reduced neurodegeneration when deprived of food as L1 larvae. n=90 for each line; two-way ANOVA with Sidak's post hoc test. These data are reported as mean normalized to GFP animals±s.d. All nematodes were grown at 20°C. (B) On the second day of adulthood (day 5 post-hatching), animals were moved to plates absent of bacteria, according to the dietarydeprivation regimen outlined by Greer and Brunet (2009), which reported that dietary-deprivation-dependent lifespan extension was dependent on hsf-1. Neurodegeneration analysis of dietary-deprived animals was performed on day 7 post-hatching. Dietary deprivation reduced neurodegeneration in animals expressing Aβ alone (UA198; P=0.0107), but there was not a significantly additive effect in either Aβ+ApoEε2 (UA351; P=0.9909), Aβ+ApoEε3 (UA353; P=0.9441) or Aβ+ApoEε4 (UA355; P=0.2421) C. elegans. These data are reported as mean normalized to GFP animals±s.d. n=90 for each line; * indicates statistical significance; ns, not significant; two-way ANOVA with Sidak's post hoc test. All nematodes were grown at 20°C. (C) The absence of peptone has been reported to increase lifespan through the AMP-activated protein kinase, AAK-2, and the insulin-like signaling protein, DAF-16. Synchronized animals were grown at 20°C on either NGM with a standard final peptone concentration of 2.5 g/l or NGM without peptone and analyzed at days 3 and 7 posthatching. Absence of peptone had no effect on Aβ alone (UA198; day 3, P=0.9998; day 7, P=0.2740), Aβ+ApoEε2 (UA351; day 3, P=0.9965; day 7, P=0.9230), Aβ+ApoEε3 (UA353; day 3, P=0.2798; day 7, P=0.0653) or Aβ +ApoEε4 (UA355; day 3, P=0.8268; day 7, P=0.3279). These data are reported as mean normalized to GFP animals±s.d. n=90 for each line; * indicates statistical significance; ns, not significant; two-way ANOVA with Sidak's post hoc test.
Although ApoE is associated with longevity (Fuku et al., 2017;Lin et al., 2016;Schächter et al., 1994;Skillbäck et al., 2018), it is not clear how ApoE interacts with other longevity-associated pathways, especially when challenged by Aβ-induced proteostatic stress. A more thorough understanding of transcriptional changes with ApoE expression would shed light on the neuronal effect of ApoE that drives the differences between L1 starvation and dietary deprivation models. ApoEε4 has been observed to translocate to the nucleus and alter gene expression by altering transcriptional regulation (Lattanzio et al., 2014;Theendakara et al., 2013Theendakara et al., , 2016. Many of these genes appear conserved from C. elegans to humans and might have similar implications for metabolism, stress response and aging (Arey and Murphy, 2017; Vahdati Nia et al., 2017). Thus, future studies combining transcriptional profiling of the ApoE-Aβ transgenics with RNAi or genetic depletion of up-or downregulated target genes would be informative.
Aging remains the most definitive risk factor for AD. Therefore, it is significant to note that, in the absence Aβ, none of the APOE alleles had an effect on survival (Fig. 5B-D). However, when independently co-expressed with Aβ, both APOEε2 and APOEε3 attenuated the shortened survival caused by Aβ (Fig. 5A,E,F). Although the observed differences between the survival curves were modest, they were statistically significant. In contrast, APOEε4 did not confer any significant effect (Fig. 5G). One possibility to explain these results is that the shortened survival induced by Aβ (Fig. 5A) is a consequence of glutamatergic neuron failure to accurately control feeding behaviors and fat storage Hamilton et al., 2005;Lee and Ashrafi, 2008;Greer and Brunet, 2009;Greer et al., 2008;Hills et al., 2004;Kindt et al., 2007;Zheng et al., 1999). Because the glutamatergic neuronal circuitry modulates feeding behaviors, Aβ might possibly depress survival through dysfunctional feeding, which is hitherto repressed by the protection of neuronal structure observed with APOEε2 co-expression. The ability of the APOEε2 allele to reduce survival depression by Aβ would therefore be due to restored glutamatergic connectivity through the associated neuroprotective phenotypes. Uninhibited feeding, in combination with the utilization of fat storage from loss of glutamatergic signaling, potentially incites insulin signaling responses that influence longevity (Greer et al., 2008;Gusarov et al., 2017). The connection between insulin signaling and longevity was first realized in C. elegans (Kenyon et al., 1993). Clinical research shows a complex relationship between diabetes, AD and ApoEε4 (Arnold et al., 2018). Administration of insulin facilitated memory recall in patients carrying APOEε2 or APOEε3, but further impaired recollection in APOEε4 patients (Reger et al., 2006). Considering the history and utility of C. elegans in the study of aging, we propose this model would be an effective tool to study the relationship between aging, insulin signaling and ApoE variants in Aβ-induced neurodegeneration.
It should be noted that known functions of ApoE are not limited to the nervous system (McNeill et al., 2010;Rosenson et al., 2017;Zhang et al., 2010). By restricting the expression of alleles to a single cell type, as in our model, the cellular and subcellular effects can be isolated from the emergent effects of endogenous expression that would otherwise compound the complexity underlying Aβ-mediated neurodegeneration. Furthermore, that C. elegans has no endogenous ApoE ortholog allows use of this model for dissection of the interactions between ApoE and evolutionarily conserved pathways without obfuscation from other perturbations, such as immunological and hepatic responses typically associated with ApoE. Because of the genetic and pharmacological amenability of C. elegans, screening for modifiers of ApoE-Aβ activity is tenable. Additional phenotypic outputs might provide further insight into nuances of ApoE-induced effects. Because the glutamatergic signaling that regulates fat storage in response to food also modulates pharyngeal pumping rate (Greer and Brunet, 2009), both fat storage (Yen et al., 2010) and pharyngeal pumping (Sanders et al., 2017) are potential quantifiable outputs of glutamatergic signaling. Likewise, the olfactory circuit is modulated by glutamatergic signaling (Chalasani et al., 2007), exhibiting quantifiable changes in turning and reversals (Bhattacharya et al., 2014;Xiao et al., 2015) in response to specific odors (Chalasani et al., 2010). Furthermore, the C. elegans olfactory circuitry is a workshop for research in the neurobiological basis of learning (Cho et al., 2016). Consequently, candidate compounds can be tested for their effects on neurodegeneration, and also how they affect neuron function and animal health. Thus, this model provides a new medium through which neuronal mechanisms of ApoE can be distinctly probed to expedite the identification of therapeutic targets and risk factors to better address the urgent and unmet societal burden represented by AD.

Plasmid construction
The cDNAs of the human APOE alleles were a generous gift from Susan Lindquist. The cDNAs were cloned by Gateway Technology (Invitrogen) according to the manufacturer's protocol. Briefly, primers 5′-GGGGACA-AGTTTGTACAAAAAAGCAGGCTCCatgcataaggttttgctggcactgttctttatctt-tctggcaccagcaATGaaggtggagcaagcggtgg-3′ and 5′-ggggaccactttgtacaag-aaagctgggtcCTAcagtgattgtcgctgggcac-3′ were used to amplify the APOE alleles and amplica were recombined with pDONR221 by BP reaction to generate entry clones. Entry clones were confirmed by sequencing and recombined with P eat-4 expression vectors by LR reaction. Expression clones were confirmed by sequencing.

Neurodegeneration analysis
Animals for analysis were synchronized with a 3-h egg lay using gravid hermaphrodites and incubated at 20°C, unless otherwise specified. To examine the neurons, hermaphrodites at indicated post-hatching time points were immobilized using 3 mM levamisole on glass cover slips and inverted onto 2% agarose pads on microscope slides. Each analysis was replicated at least three times with 30 animals per condition (30 animals×3 trials=90). C. elegans glutamatergic neurons were analyzed for neurodegeneration as previously described (Matlack et al., 2014;Tardiff et al., 2012Tardiff et al., , 2017Treusch et al., 2011). Briefly, animals were scored for glutamatergic neurodegeneration at days 3 and 7 post-hatching, as reported in the Results and in figure legends. An animal was scored as normal if all five tail neurons were present and without malformities such as distention, apoptotic swelling, axon breaks, separation of the soma or loss of fluorescence.

Mechanosensation assay
Assays were performed as previously described (Chalfie and Sulston, 1981;Chalfie et al., 1985). Briefly, animal populations were synchronized by a 3-h egg lay and progeny were incubated at 20°C until day 4 post-hatching.
C. elegans sensitivity to soft touch was assayed by gently stroking hermaphrodite animals on the posterior and anterior with an eyelash hair glued to the end of a Pasteur pipette. Backward locomotion was induced by gently stroking the head of the animal with the eyelash followed by stroking the tail just below the anus. A positive result for soft touch sensitivity was recorded if the animal ceased backward locomotion or began moving forward. This process was repeated five times per animal, and the number of positive responses to posterior soft touch out of five was recorded. A total of 30 worms per strain were scored per biological replicate and percentage posterior touch response was calculated as the percentage average response within the population. The experiment was repeated at least three times (n=3×30=90) and data represent the average of all three biological replicates with s.e.m., as previously reported (Zhang et al., 2004).

RNAi
RNAi feeding clones were cultivated initially on LB solid medium containing tetracycline (5 μg/ml) and ampicillin (100 μg/ml) and then individual colonies were grown overnight in liquid LB medium containing 50 μg/ml carbenicillin. Isopropyl β-D-1-thiogalactopyranoside (IPTG) was spread on plates to a final concentration of 100 μM, seeded with RNAi feeding clones and allowed to dry. Induction of dsRNA occurred during a 14-to 18-h incubation at 20°C. Adult hermaphrodites were allowed to lay eggs for 3 h on RNAi feeding clones to produce a synchronized population. Glutamatergic neurons of synchronized progeny were analyzed at least 2 days after hatching, as described above.

L1 starvation
Age-synchronized populations of each C. elegans strain were generated by bleaching. Briefly, 10 ml bleaching solution (1 ml 5N KOH, 2 ml bleach, 7 ml ddH 2 O) was used to isolate embryos. Embryos were then washed three times in 10 ml 1× M9 buffer to remove bleaching solution. Embryos were then transferred to standard NGM plates seeded with 200 μl OP50 bacteria or NGM plates containing no bacteria. After 24 h, animals were moved to NGM plates seeded with 200 μl OP50. Animals were incubated at 20°C for a total of 96 h and then 90 worms per strain were assayed for neurodegeneration.

Peptone absence
Animals and media were prepared as previously described (Greer and Brunet, 2009). Briefly, age-synchronized populations of animals were obtained by allowing gravid adults to lay eggs for 3 h at 20°C on NGM plates containing either the standard quantity of peptone (2.5 g/l), as our control, or no peptone, and seeded with OP50 at a concentration of 5×10 12 CFU/ml. Animals were maintained at 20°C and transferred as necessary until scoring for neurodegeneration.

Dietary deprivation
Animals and media were prepared as previously described (Lee et al., 2006). Briefly, age-synchronized populations of animals were obtained by allowing gravid adults to lay eggs for 3 h at 20°C on seeded NGM plates. Animals were transferred to fresh seeded plates as necessary until day 2 of adulthood (day 5 post-hatching), at which point they were either transferred to seeded plates as they had been previously (ad libitum condition) or transferred to unseeded plates (dietary deprivation). Animals were maintained for neurodegeneration analysis at day 7 post-hatching.

Survival assays
Survival assays were performed as previously described (Hsin and Kenyon, 1999). Briefly, strains were allowed to grow at 20°C in optimal growth conditions for at least two generations before the experiment began. Synchronized animal populations for survival analysis were generated by a 1-h egg lay using gravid hermaphrodites and incubated at 20°C. The L4 molt was defined as t=0, at which time animals were transferred to experimental plates. A total of 200 animals for each of nine strains were examined and all strains were assigned five initial plates with 40 worms each. Animals were then transferred to new plates every day, until the worms' reproductive stage had passed, after which point animals were then transferred every other day to ensure that appropriate amounts of food remained on the plate. Survival was assessed immediately after each transfer, as previously described (Hsin and Kenyon, 1999). To score for death, animals were examined for locomotive response to prodding with a platinum wire. Briefly, animals were touched five times on the head and the tail and assessed for reverse or forward locomotion in response. Animals were classified as dead if they ceased moving and failed to respond to this stimulation. A third category, censored, was utilized for animals that did not die of the natural aging process. Worms were classified as censored if they crawled off the plate, burrowed, or displayed vulval rupture or internal hatching, as previously described (Hsin and Kenyon, 1999). Seeded plates were stored at 20°C until completion. In GraphPad Prism software, the log-rank (Mantel-Cox) method was used to account for differences between survival curves. Specifically, all time points are assigned equal weights in statistical calculations whereby the entire pattern or path of the curve is being analyzed in testing for significance, not just the maximum value displayed (Hansen et al., 2008).

Pharmacological treatments
Thapsigargin (Acros Organics) was dissolved in dimethyl sulfoxide (DMSO) and added to NGM plates to a final concentration of 3 μg/ml, as reported previously (Zwaal et al., 2001), with the modification that thapsigargin was added directly to the medium rather than supplemented on the surface.

Experimental design and statistical analysis
Hermaphrodites were analyzed, which is standard in the C. elegans field, and all animals were incubated at 20°C, unless otherwise specified. In all cases, sample sizes (typically 30 animals per condition; for a total of 90 animals) were standardized within each experiment and examined in a uniform fashion. All experiments used at least three independent replicates per experiment per variable to generate a mean and s.d. In experiments using one independent variable across multiple tested effects (e.g. neuron cell death as a function of construct type), a one-way ANOVA series was used with a multiple-comparisons post hoc test (Tukey's). For grouped analyses, a two-way ANOVA series was used with Sidak's post hoc test. Survival was analyzed by the log-rank (Mantel-Cox) method, as previously described in the survival assay section. P<0.05 was the absolute minimum threshold for statistical significance. Statistics were performed using GraphPad Prism software.