Aβ42 oligomers modulate β-secretase through an XBP-1s-dependent pathway involving HRD1

The aspartyl protease β-site APP cleaving enzyme, BACE1, is the rate-limiting enzyme involved in the production of amyloid-β peptide, which accumulates in both sporadic and familial cases of Alzheimer’s disease and is at the center of gravity of the amyloid cascade hypothesis. In this context, unravelling the molecular mechanisms controlling BACE1 expression and activity in both physiological and pathological conditions remains of major importance. We previously demonstrated that Aβ controlled BACE1 transcription in an NFκB-dependent manner. Here, we delineate an additional cellular pathway by which natural and synthetic Aβ42 oligomers enhance active X-box binding protein XBP-1s. XBP-1s lowers BACE1 expression and activity indirectly, via the up-regulation of the ubiquitin-ligase HRD1 that acts as an endogenous down-regulator of BACE1. Thus, we delineate a novel pathway by which cells could compensate for Aβ42 oligomers production and thus, associated toxicity, by triggering a compensatory mechanism aimed at lowering BACE-1-mediated Aβ production by a molecular cascade involving XBP-1s and HRD1. It thus identifies HRD1 as a potential target for a novel Aβ-centered therapeutic strategy.

established that exogenous application of Aβ 42 monomers or transient transfection of cDNA encoding Aβ 42 enhanced BACE1 mRNA levels via NFκ B 32 . This data was conforted by our demonstration that cells overexpressing Swedish-mutated β APP and producing supraphysiological levels of Aβ displayed enhanced NFκ B-dependent BACE1 transcription and activity 17 . Although these studies consistently indicate an indirect control of BACE1 by its own APP-derived product, it did not definitely demonstrate the capability of Aβ 42 oligomers (Aβ o), currently recognized as the most toxic Aβ species 33,34 to mimic this effect. Furthermore, it did not examine whether alternative NFκ B-independent pathways could govern Aβ -linked control of BACE1.
We examined first the influence of synthetic oligomeric Aβ o (Fig. 1a) and we show that they increase BACE1 mRNA levels in SH-SY5Y cells (Fig. 1b) while a trend of enhanced BACE1 protein expression was observed that did not reach statistical significance (Fig. 1c,d). It was unclear whether this apparent dichotomy between Aβ o-associated influence on BACE1 mRNA and protein expressions could be accounted for by the experimental procedure where acute exposure to oligomers or by kinetic delay between transcription process and traduction. Thus, we envisioned determining whether naturally occurring Aβ o chronically produced by chinese ovary hamster cells (CHO) stably engineered to express either wild type β APP or β APP harboring the London mutation (V717I; APP LDN ) 35 could modulate BACE1 expression and activity. Figure 1 shows that CHO APP LDN cells indeed yield high levels of Aβ o (Fig. 1e) concomitant to increased BACE1 expression (Fig. 1f,g) and enzymatic activity (Fig. 1h,i). Overall, this set of data indicates that both synthetic and naturally occurring Aβ o modulate BACE1 mRNA and protein expression and activity although we cannot totally preclude the possibility of a marginal contribution of Aβ monomers.
Synthetic and naturally occurring Aβ oligomers up-regulate mRNA levels of spliced X-Box binding protein 1. The intimate link between cellular stress, Aβ and BACE1 23,36 led us to investigate whether Aβ o effects on BACE1 expression could be governed by cellular sensors involved during the unfolded protein response in response to Aβ aggregation. Since Aβ o affect BACE1 transcription, we envisioned XBP-1s, a transcription factor activated upon stress-associated Ire1α -mediated maturation, as a putative target. Interestingly, XBP-1s mRNA levels were increased upon synthetic Aβ o treatment of SH-SY5Y cells (Fig. 2a). We ruled out the possibility of a cell-dependent artifact by monitoring Aβ o-induced modulation of XBP-1s in murine fibroblasts. Thus synthetic Aβ o also triggered enhanced XBP-1s mRNA levels in wild-type ( Fig. 2b) but not in XBP-1s null fibroblasts (Fig. 2c). Of importance, we show that naturally occurring Aβ o also increase XBP1s mRNA levels ( Fig. 2d) even if we did not examine the precise nature of oligomers species involved in such effect. Overall, this data demonstrates the ability of both synthetic and naturally occurring monomers/Aβ o mix to up-regulate XBP-1s mRNA levels.
XBP-1s regulates BACE1 expression and activity at a post-transcriptional level. We reasoned that since Aβ o increase both BACE1 and XBP-1s, the latter could behave as an intermediate cellular effector mediating Aβ o-associated increase in BACE1 transcription. Thus, we investigated the effect of a modulation of XBP-1s on BACE1 expression and protease activity. Surprisingly, we found that overexpression of XBP-1s reduces BACE1 expression (Fig. 3a,b) and enzymatic activity (Fig. 3c,d) in HEK293 cells. Conversely, XBP-1s depletion increases BACE1 expression (Fig. 3e,f) and activity (Fig. 3g,h), a phenotype that can be rescued by XBP-1s overexpression in XBP-1s knockout cells (Fig. 3i,j).
We then examined whether XBP-1s-linked modulation of BACE1 expression and activity could be accounted for BACE1 promoter transactivation by XBP-1s. As expected, XBP-1s overexpression yields quantifiable amounts of XBP-1s mRNA (Fig. 4a) while it neither altered BACE1 mRNA levels ( Fig. 4b) nor BACE1 promoter transactivation (Fig. 4c). To rule out any cell specific artifacts, these results were confirmed in fibroblasts genetically invalidated for XBP-1. As was observed in HEK293 cells, restoration of XBP-1s expression in null cells (Fig. 4d) did not modify BACE1 mRNA levels ( Fig. 4e) or promoter transactivation (Fig. 4f). This post-transcriptional XBP-1s-mediated regulation of BACE1 was also observed in cells of neuronal origin (SH-SY5Y) ( Supplementary Fig. 1a-c). Altogether, this set of data indicates that XBP-1s down-regulates BACE1 at a post-transcriptional level in both human neuroblastoma cells and mouse fibroblasts.   and d), respectively). A Student one-tailed T-test is applied for statistics (*p-value = 0,02 and n.s = non significant, in (b and d), respectively). (e) Western blot analysis of media recovered from CHO cells stably overexpressing either wild type β APP (APP WT ) or β APP bearing the London mutation (APP LDN ) compared to mock-transfected CHO cells. (f,g) Western blot analysis (f) and densitometric analysis of BACE1 (g) protein expression in the indicated cell lines. The graph shows the mean of 5 independent experiments (10 biological replicates per group). For statistics, a one-way ANOVA followed by a post hoc Bonferroni-Holm test is applied (**p-value = 0,0053 alpha 0,01; n.s non significant). (h) Specific BACE1 activity measured in the indicated cell line. BACE1 activity corresponds to the inhibitor-sensitive fluorimetry and is expressed relatively to control and normalized by protein quantification using Bradford method. The graph represents the mean BACE1 activity of 4 independent experiments, with 8 biological replicates plotted per group. (i) Represents the slope value of the curves in (h). Statistic analysis is as in B (***p-value = 0,001 alpha 0,01; n.s non significant).
Scientific RepoRts | 6:37436 | DOI: 10.1038/srep37436 XBP-1s regulates HRD1 expression at a transcriptional level. Our data indicate that Aβ o increases BACE1 and XBP-1s but that the latter lowers BACE1 by post-transcriptional mechanisms. We thus reasoned that such cascade of events could only be explained by an intermediate effector between XBP-1s and BACE1 that should fulfill three criteria: (1) to be up-regulated by XBP-1s; (2) to act as a repressor of BACE1 and; (3) to trigger its effect at a post-transcriptional level.
Yamamoto and Colleagues demonstrated that HRD1, an E3 ubiquitin ligase, was up-regulated under ER stress conditions through an XBP-1 dependent pathway 37 . This led us to envision HRD1 as the cellular effector linking XBP-1s to BACE1. First, we show that XBP-1s mRNA levels are enhanced in the presence of thapsigargin, a well known ER-stress inductor (Fig. 5a), and that this activation, as expected, is indeed fully abolished by XBP-1 genetic invalidation. HRD1 mRNA levels were also enhanced by thapsigargin and this increase was abolished by XBP-1s depletion (Fig. 5b). To strengthen this observation, we examined the influence of XBP-1s expression on HRD1 mRNA levels in human cells. As expected, XBP-1s overexpression enhances both HRD1 mRNA (Fig. 5d) and protein (Fig. 5e,f) expressions. These results were conforted by the ability of XBP-1s to increase HRD1 (Fig. 5h) in MEF cells devoid of XBP-1s (Fig. 5g). The effect of XBP-1s on HRD1 transcription was also confirmed in SH-SY5Y cells ( Supplementary Fig. 1d). Overall, our results show that XBP-1s up-regulates HRD1 transcription in our cellular models.

Aβ42 oligomers increase HRD1 expression through XBP1-dependent mechanisms. Since
Aβ o increase XBP-1s that up-regulates HRD1, we assumed that a linear cascade should be reflected by an up-regulation of HRD1 by Aβ o. We examined whether naturally secreted and synthetics Aβ o indeed modulate HRD1 expression. Figure 6 shows that HRD1 protein (Fig. 6a,b) and mRNA (Fig. 6c) levels were enhanced in CHO APP LDN cells. This data was corroborated by the fact that synthetic Aβ o also enhanced HRD1 protein (Fig. 6d,e) and mRNA levels (Fig. 6f). Of most interest, Aβ o-associated increased amount of HRD1 mRNA expression in wild-type fibroblasts (Fig. 6g), was fully abolished by XBP1 gene invalidation (Fig. 6h). Thus, this data demonstrates a direct link between Aβ o-mediated increases in XBP-1s and HRD1.  Figure 7 shows that it was indeed the case. Thus, HRD1 overexpression decreased BACE1 protein expression (Fig. 7a,b) and activity (Fig. 7c,d) by about 30%. However, HRD1 expression (see increased HRD1 mRNA expression in Fig. 7e) was unable to modulate BACE1 promoter transactivation (Fig. 7f) Student one-tailed T-test is applied for statistics (*p-value = 0,02 (a) and 0,03 (b,c); n.s non-significant). (d) Quantitative PCR analysis of XBP-1s mRNA in indicated CHO cell lines. XBP-1s mRNAs are normalized as above and graph shows the mean of 3 independent experiments (9 biological replicates per group). Statistical analysis is performed by a one-way ANOVA followed by a post hoc Bonferroni-Holm is applied (*p-value = 0,02 alpha 0,05; n.s, non-significant).
Scientific RepoRts | 6:37436 | DOI: 10.1038/srep37436 and mRNA levels ( Fig. 7g). Thus, HDR1-linked down-regulation of BACE1 likely occurs at a post-transcriptional level. This was supported by the reduced lifetime of BACE1 upon HRD1 expression (Fig. 7h,i) in experimental conditions where neosynthesis of endogenous BACE1 was blocked by cycloheximide (see Methods), in agreement with the HRD1 function as an ubiquitin ligase 38 . Overall, the above-described data demonstrate a linear molecular cascade by which Aβ o decrease BACE1 via an XBP-1s-mediated and HRD1-dependent mechanism.
Additional XBP-1s-independent control of BACE1 by Aβo oligomers. At first sight, the above data linking Aβ o to a decrease of BACE1 appear contradictory to our initial observation that both synthetic and naturally occurring Aβ o increased BACE expression and activity (see Fig. 1). This led us to question whether Aβ o could modulate BACE1 exclusively via XBP-1s or if there exist alternative pathway by which Aβ o could control BACE1. Thus, we examined the influence of synthetic Aβ o on BACE1 in absence of XBP-1s. Figure 8a-e clearly shows that Aβ o could still increase BACE1 expression in XBP-1s depleted cells (Fig. 8a-c). Interestingly, this appeared to occur at a transcriptional level since Aβ o also increased BACE1 mRNA levels in XBP −/− cells (Fig. 8e). This indicates that besides, XBP-1s-mediated HRD1-linked post-translational events aimed at reducing BACE1, Aβ o could also increase BACE1 by alternative XBP-1s-independent transcriptional pathway. Thus, both Aβ o-linked XBP-1s-dependent and -independent pathways could occur and one can envision that the former likely aims at preventing acute Aβ o-induced toxicity and increased BACE-1-associated self-production, at least at early stages of the pathology.

Discussion
Numerous biochemical, anatomical and genetic data led to the claim that the so-called amyloid cascade hypothesis fulfils the major gap in the understanding of Alzheimer's disease (AD) etiology. For years, it has been proposed that senile plaques that accumulate in AD-affected brains correspond to the pathological lesions and thus, their main components, amyloid β -peptides (Aβ ) thought to be at the center of gravity of the neurodegenerative process, drove much attention. This simplistic view of a complex disease has recently drastically evolved. Besides altered ratios in the canonical forms of Aβ 40 and Aβ 42 that are indeed affected in AD brain 39 , recent advances on the structural nature (N-terminally truncated and/or C-terminally trimmed 40,41 , biophysical state (monomeric, multimeric, aggregated, fibrillar 34,42 , subcellular localization (intracellular, secreted 43,44 ) of Aβ species as well as other β APP catabolites (C99, AICD 5,[11][12][13]45 , have significantly modified our view of the genuine trigger of the pathology. It arose from these studies that soluble Aβ oligomers (Aβ o), appear prior to senile plaques and are now considered to be more toxic than monomeric or fibrillar Aβ . Supporting this view, amongst a series of cellular perturbations, Aβ o contribute to neuronal cell death, LTP inhibition 46 calcium homeostasis perturbation 47 , oxidative or ER stresses 48 .
Whatever the toxic Aβ -related trigger, it remains that understanding the upstream enzymatic steps ultimately yielding Aβ was of major importance. Although the nature of the secretases is consensual, the mechanisms by which their activity is finely tuned or regulated remained a matter of investigation.
The choice of the secretase as a therapeutic target is a hard one. Most attention was originally centered on γ -secretase because this enzyme triggers the ultimate cleavage yielding Aβ . However, γ -secretase-mediated breakdowns of numerous additional substrates involved in vital cellular functions 49 have been documented and it has been hard to design specific inhibitory compounds usable in clinic. This likely explains reiterated failures of γ -secretase-centric clinical trials and has severely tempered the optimism for considering γ -secretase blockade as a mean to interfere with the course of the disease. Although BACE1 was recently shown to contribute to several physiological functions [50][51][52][53][54] , it appears that unlike is the case for γ -secretase 55,56 . BACE1 gene ablation is rather well supported in animals, indicating a more narrow substrates specificity than that of γ -secretase or, alternatively, that functions of its additional substrates appear less vital for cells. These considerations suggest that the targeting of BACE1 appears apparently less challenging. Furthermore, this strategy presents several advantages. First BACE1 is the rate-limiting enzyme of Aβ biosynthetic pathway. Second, the blockade of BACE1 not only impairs Aβ production but also theoretically prevents all β APP-related catabolites, some of which appear very toxic, even in absence of Aβ [11][12][13] .
Relatively few data concern BACE1 regulation and most of them linked Aβ load or β APP mutations to altered BACE1 gene transcription 16,17,32 . More often, BACE1 expression or activity have been linked to cellular conditions mostly related to hypoxia 36,57 , oxidative 58 and ER stress 23,59 .  XBP-1 is a transcription factor known to regulate genes involved in ER homeostasis that has also been involved in multiple signaling pathways and diseases (For review see ref. 28). Here we show that XBP-1s lowers BACE1 expression and activity in various cells of human and murine origins. Interestingly, XBP-1s-mediated repression of BACE1 occurs at a post-transcriptional level. This agrees well with a recent study showing that XBP1s did not modify BACE1 promoter transactivation in human cells 60 . In search for a cellular intermediate, we reasoned that it should be either a BACE1 activator that would be transcriptionally repressed by XBP-1s or, alternatively, a BACE1 repressor, the transcription/activity of which would be enhanced by XBP-1s. The latter case stood. Thus we demonstrated that HRD1 mRNA levels were up-regulated by XBP-1s and that HRD1 down-regulated BACE1 protein but not mRNA expressions. This agreed with the well-documented function of HRD1 that acts as an ubiquitin ligase involved in protein ubiquitination and degradation during Endoplasmic Reticulum associated degradation (ERAD), a process known to be under control of XBP-1 61 . The above-described set of data was comforted by the fact that Aβ o increased both XBP1s and HRD1 expressions. This agreed with several independent studies showing that XBP-1 mRNA splicing, and therefore activation, was potentiated by Aβ o in transgenic flies 30 and primary cultured neurons 62 see Suppl. Fig. 12) and demonstrating an Aβ o-mediated post-transcriptional regulation of neuronal BACE1-like immunoreactivity 63 . In addition, previous works indicated a negative correlation between HRD1 expression and Aβ generation 64 and demonstrated that HRD1 suppression leads to enhanced Aβ production 65,66 .
The XBP-1S-mediated compensatory mechanism would agree with the protective phenotype generally ascribed to XBP-1s. Thus, previous studies showed that XBP-1s protected against Aβ -associated toxicity 30 and could interfere with autophagy process 67 . More importantly and related to Alzheimer's disease, XBP1 depletion revealed memory defects in wild-type mice 29 and XBP-1s expression restored synaptic plasticity and memory control in several AD mice models 62 . Finally, XBP-1S transcriptionally up-regulates ADAM10, the α -secretase constitutive activity involved in the protective non amyloidogenic β APP processing pathway 60 .
According to the time frame of AD development, we anticipate that compensatory mechanisms should occur early and transiently. This notion of temporal window is supported by studies on the expression of XBP1 in various AD mice models and AD-affected human brains. Thus, proteomic analysis revealed enhanced XBP1 protein expression at an early stage of disease progression in 5xFAD mice 68 . Further, Reinhard and colleagues showed that XBP-1s mRNA levels peaked at 8 months of age in APP/PS1 60 . Finally, we observed a transient increase of XBP-1s mRNA in the hippocampus of 3 month-old CRND8 and 3xTg-AD mice 62 . Interestingly, XBP-1s expression was lowered in the frontal and temporal cortices in autopsied AD affected brains, i.e at late stage of the disease, when BACE1 expression has been clearly shown to be enhanced 14 . This is consistent with the claim of a protective increase in XBP1 aimed at reducing BACE1 expression at early stage of the pathology, a transient compensatory process ineffective at later stage of the pathology.
Gene therapy has been proposed as a track to target brain pathologies linked to ER stress 69 . In this context, it is noteworthy that modulating XBP1 has proved efficient to protect against Huntington disease in mice model of this pathology. Whether XBP-1s targeting would be efficient in the case of Alzheimer's disease remains a matter of speculation and still awaits faithful biomarkers that could tag the very early stages of the disease. However, it remains that our study shows that compensatory mechanisms should not be underscored and underlines the need for further understanding of secretases regulation.

Methods
Plasmids. The spliced form of mouse XBP-1 cDNA (XBP-1s) was cloned in pcDNA3 plasmid. The human HRD1-myc cDNA (transcript variant 2) was cloned in pCMV6 and obtained from OriGene Technologies. Luciferase construct of human BACE1 promoter cloned in pGL3-basic vector was kindly provided by Dr. Lahiri D.K.
Protein extraction and western blotting. Cultured cells were collected in PBS-EDTA (5 mM) before centrifugation at 3,000 rpm for 5 min. Pellets were lysed in RIPA buffer with a complete proteases inhibitor, then centrifuged at 13,000 rpm for 10 min. Total protein concentration was assessed in the supernatant by Bradford method. Each sample (50 μ g) was resuspended in Laemli buffer (1x final) and boiled at 96 °C for 5 min before loading and resolving on 10% SDS-PAGE gel. Wet transfer was done using Hybond-C membrane (Amersham Bioscience) at 100 V for 1 h30, then membrane were blocked with 5% PBS-milk solution for 1 h15 and blotted overnight with following antibodies: Mouse monoclonal 3D5 anti-Bace1 (kindly given by Dr. Vassar R.), rabbit polyclonal anti-XBP-1 (M-186 Santa-Cruz Biotechnology), mouse-monoclonal anti-actin (Sigma Aldrich), mouse-monoclonal anti-Myc (9E10; Santa-Cruz biotechnology). Protein immunoreactivity was assayed using peroxidase-coupled antibodies (Jackson immunoresearch), resulting electrochimio-luminescence was detected with luminescence analyzer LAS-3000 (Raytech). Multi-Gauge software (FUJI film) was used for image protein quantification. All densitometric quantifications were normalized using actin as loading control.
RNA extraction and Quantitative Real-time PCR. Total RNA extractions were performed with RNA easy extraction kit (Qiagen) on Qiacube device following manufacturer's material and methods. RNA quantities were normalized at 2000 ng and reverse-transcripted with GoScript reverse-transcriptase (Promega) on a Biometra thermocycler. Real time quantitative PCR were performed on Rotor-gene6000 (Qiagen) using SYBR Green protocol according to manufacturer's recommendation. Gene expression of BACE1, XBP-1s and HRD1 were normalized using human or mouse ribosomal protein 69 (Rpl 69) gene expressions depending on RNA species. RNA relative expression was calculated following Livak K.J. delta-delta CT method 70  β-secretase enzymatic activity assay. Cells were harvested in PBS-EDTA 5 mM and centrifuged at 3,000 rpm for 5 min. After centrifugation, pellets were lysed on ice with an homogenization buffer (5 mM EDTA, 1 mM Hepes, 0,25 M sucrose) using a 26G syringe. The samples were then centrifuged at 850 g for 10 minutes and the supernatants were centrifuged once more at 20,000 g for 1 h. All preparations were normalized to a protein concentration of 3 μ g/μ l after Bradford quantification using Tris buffer (Tris 10 mM pH 7.5). 10 μl of each sample were incubated in a 96 wells plate with 90 μ l acetate buffer (25 mM, pH 4.5) in presence or absence of BACE1 specific inhibitor (PromoKin). After 5 minutes of incubation, BACE1 fluorimetric substrate [(7-methoxycoumarin-4-yl)-acetyl-SEVNLDAEFRK(2,4-dinitrophenyl)-RRNH 2 ; 10 μ M R&D Systems] was added and the BACE1-like activity was monitored as described 71 every 15 minutes for 2 h30 (320 nm excitation wavelength and 420 nm emission wavelength). Specific BACE1 activity was considered as the inhibitor-sensitive fluorimetric activity.
Luciferase and β-galactosidase activities. Luciferase activities (reporting BACE1 promoter activities) were measured after cells co-transfection with both a luciferase reporter construct and a β -galactosidase encoding construct. Cells were lysed, on ice, using a reporter lysis buffer (Promega) before centrifugation at 2,500 rpm for 10 minutes. Supernatant fraction (20 μ l) was mixed with luciferase reagent (50 μ l) and luminescence was then measured on Varioskan flash reader (Thermo Scientific). Luciferase activity was normalized with either beta-galactosidase activity and protein concentration determined by Bradford method.
Cycloheximide pulse-chase experiment. Cells were transfected with construct encoding HRD1-myc or a control empty vector for 12 hours. Cycloheximide (Sigma-Aldrich) was added to the cell media at a concentration of 1 μ M up to 48 hours of time. Cells were harvested every 12 hours in PBS-EDTA buffer and centrifuged at 3,000 rpm for 5 minutes. Protein extraction and western blot were then carried out as indicated in the corresponding section.
Synthetics Aβ42 oligomers preparation. Synthetic Aβ 42 (Bachem Distribution Services GmbH, Weil am Rhein, Germany) was solubilized in HFIP (hexafluoro-2-propanol) according to previously described (Clarke et al., 2015). Briefly, after solubilization and solvent evaporation, dried films were dissolved and stired in 10 μ l dimethylsulfoxide. After dilution into 500 μ l of cold PBS during 24 hours, insoluble aggregates were removed by Scientific RepoRts | 6:37436 | DOI: 10.1038/srep37436 centrifugation at 14,000 g for 10 min at 4 °C as described (De Felice et al. 48 ). Supernatant was then stored at 4 °C. For each cell treatement, 300 μ l of the preparation were used. Statistical analysis. Statistical analyses were performed with the Excel software. For statistical comparison between two groups, the parametric Student T-test (two tailed) was used. For statistical comparison between three groups, a one-way ANOVA test was performed, followed by the Bonferroni-Holme post hoc correction for multiple comparisons. Graphs represents the mean of the values obtained for each group of samples (biological replicates consisting of independent cells cultures, technical replicates were used in quantitative PCR as internal control for accuracy but are excluded from statistical analysis), error bars represent the standard error of the calculated mean (s.e.m). Criteria for data exclusion were: high background noise and low or heterogeneous specific signal, important loading control variation (Western blot), important modulation of housekeeping gene expression or low accuracy between technical replicates (quantitative PCR), unequal β -galactosidase activity between samples (Luciferase dosage), flat specific enzymatic curve with no exponential phase (β -secretase enzymatic activity assay).