A tricyclic antidepressant, amoxapine, reduces amyloid-β generation through multiple serotonin receptor 6-mediated targets

Alzheimer’s disease (AD) is a major and devastating neurodegenerative disease, and the amyloid-β (Aβ) hypothesis is still the central theory for AD pathogenesis. Meanwhile, another major mental illness, depression, is one of the risk factors for AD. From a high-throughput screening (HTS), amoxapine, a typical secondary amine tricyclic antidepressant (TCA), was identified to reduce Aβ production. A follow-up investigation on antidepressants showed that most of the TCAs harbour similar activity. Previous studies have indicated that TCAs improve cognitive function in AD mouse models as well as in preliminary clinical data; however, the underlying mechanism is controversial, and the effect on Aβ is elusive. Thus, we developed a secondary screening to determine the molecular target of amoxapine, and serotonin receptor 6 (HTR6) was identified. Knockdown of HTR6 reduced the amoxapine’s effect, while the HTR6 antagonist SB258585 mimicked the activity of amoxapine. Further mechanistic study showed that amoxapine and SB258585 reduced Aβ generation through multiple HTR6-mediated targets, including β-arrestin2 and CDK5. Taken together, our study suggests that amoxapine, though no longer a first-line drug for the treatment of depression, may be beneficial for AD and further structural modification of TCAs may lead to desirable therapeutic agents to treat both AD and depression.

1 (BACE1) activity in an in silico screening 17 . All of these pieces of evidence indicate that TCAs may improve AD symptoms by somehow modulating APP processing. Here, following our HTS data, we have performed cellular experiments to determine the molecular mechanism of TCA's action on Aβ generation.

Results
High-throughput screening identifies amoxapine as an Aβ-reducing agent. A commercially available chemical library composed of 1280 pharmacologically active compounds was assigned to the high-throughput Aβ screening using a sandwich Enzyme-Linked ImmunoSorbent Assay (ELISA). In HEK293 cells stably expressing APP Swedish mutant (referred to as HEK293-APPsw), 69 chemicals at 10 μM showed the ability to reduce the extracellular Aβ amount (≥20%), and among them, amoxapine suppressed the Aβ level by approximately 20% (Fig. 1A) without influencing cell viability (data not shown). We then validated the data in SK-N-SH, a human neuronal cell line. As the direct inhibition of the secretases of the amyloidogenic pathway leads to the decrease in Aβ generation, 10 μM of a BACE1 inhibitor, BACE inhibitor IV (BSI IV) and 10 μM of a γ-secretase inhibitor, L685,458, were used as positive controls. Amoxapine dose-dependently reduced the amount of Aβ secreted into the medium, reaching 37.32 ± 2.75% (mean ± s.e.m.) reduction at 10 μM without affecting cell viability ( Fig. 1B and Supplementary Figure 1A). Amoxapine is a secondary amine tricyclic antidepressant and was approved for treating major depressive disorder in the US in 1992 18 . It was curious to us whether other tricyclic antidepressants harbour similar activity towards Aβ generation. In SK-N-SH cells, amitriptyline, protriptyline and trimipramine also dose-dependently suppressed extracellular Aβ levels with no obvious cytotoxicity ( Fig. 1B and Supplementary Figure 1A). Since TCAs are no longer the first-line drug for depression therapy, we also tested another major type of antidepressants that is currently in use, the SSRIs, in the same system. Compatible with the in vivo data of previous reports [19][20][21] , 10 μM of citalopram, 10 μM of fluoxetine and 3 μM of sertraline slightly reduced cellular Aβ generation with 16.57 ± 2.89%, 25.95 ± 2.92%, and 27.32 ± 2.72%, respectively, without cytotoxicity (Sup. Fig. 1B,C). We further investigated the effect of amoxapine on the generation of two major Aβ species, Aβ40 and Aβ42. The data showed that amoxapine reduced both species of Aβ with similar potency (Sup. Fig. 1D,E), indicating that amoxapine did not work as a gamma secretase modulator (GSM). We then monitored the extracellular sAPPα and sAPPβ levels by using ELISA (Fig. 1C and D). TAPI-1 is an α-secretase inhibitor and, consistent with previous reports, significantly reduced the extracellular sAPPα level at 100 μM, while BSI IV (10 μM) significantly reduced the extracellular sAPPβ level 22,23 . In addition to the alteration in the total extracellular Aβ level, the sAPPα level increased, and the sAPPβ level decreased with amoxapine (10 μM) or amitriptyline (10 μM) treatment (sAPPα-amoxapine: 121.10 ± 3.87%, sAPPα-amitriptyline: 120.00 ± 1.61%; sAPPβ-amoxapine: 80.30 ± 1.76%, sAPPβ-amitriptyline: 76.04 ± 4.87%), indicating that an activity shift occurs between ADAM10 and BACE1. By using a fluorogenic substrate secretase activity assay, we determined the compounds' effect on the activity of α-secretase or BACE1. The enzymatic activity of α-secretase and BACE1 was significantly inhibited by TAPI-1 (100 μM) or BSI IV (10 μM) treatment (Fig. 1E). For amoxapine and amitriptyline, α-secretase activity remained unchanged with cellular (i.e., application of 10 μM of amoxapine or amitriptyline to the cell before membrane extraction) or in vitro (i.e., application of 10 μM of amoxapine or amitriptyline to the extracted membrane fraction) treatment (Fig. 1E). Interestingly, BACE1 activity was significantly reduced with cellular amoxapine (82.128 ± 3.64%) or amitriptyline (75.992 ± 4.68%) treatment, while it remained unchanged with in vitro treatment (Fig. 1F). We then examined whether the protein level of ADAM10, BACE1 or full-length APP was modulated by amoxapine (10 μM) using western blot analysis. As shown in Fig. 1G-K, all expression levels were intact. All of these data suggest that amoxapine reduces Aβ generation possibly through indirect modulation of BACE1 activity.
Amoxapine reduces Aβ generation through HTR6. Amoxapine and other TCAs are antagonists of dozens of G-protein-coupled receptors (GPCRs). To sort out the potential target(s) responsible for the Aβ-reducing activity of amoxapine, we first monitored the expression profile of amoxapine-targeted molecules in SK-N-SH cells. Amoxapine is mainly regarded as an antagonist of several dopamine receptors and serotonin receptors 24 . The most abundant subtypes of dopamine receptors and serotonin receptors expressed in SK-N-SH cells are DRD2, HTR6, HTR2B and HTR4 (Sup. Fig. 2A). Thus, we created a pool of shRNAs to knockdown individual GPCRs. Packaged as lentivirus, equal titers of specific shRNAs were applied to SK-N-SH cells. The knockdown efficiencies of the shRNAs were determined by quantitative RT-PCR. As shown in Fig. 2A and Supplementary Figure 2B, mRNA levels of these GPCRs were reduced by 60.361 ± 3.30% (shDRD2), 64.905 ± 14.56% (shHTR6-1), 57.158 ± 6.28% (shHTR6-2), 56.76 ± 7.85% (shHTR2B-1), 74.66 ± 6.77% (shHTR2B-2), 60.366 ± 6.36% (shHTR4), and 51.267 ± 2.48% (shHTR7), respectively, 72 hours post infection (h.p.i.). Then, we tested which receptor is responsible for the effect of amoxapine. We treated SK-N-SH cells with amoxapine 72 h.p.i. Twenty-four hours later, the supernatant was collected, and the Aβ within was detected by ELISA. The knockdown of HTR6 significantly reduced the activity of amoxapine in its suppression of Aβ generation (from 39.28 ± 2.77% with shNC to 23.441 ± 3.03% with shHTR6-1 and 23.432 ± 4.64% with shHTR6-2) (Fig. 2B), while none of the other three GPCRs showed a similar property ( Fig. 2C and Supplementary Figure 2C-E), suggesting that amoxapine may function through HTR6. We then asked whether specific HTR6 antagonists mimic amoxapine's efficacy for reducing Aβ production. SB258585, SB271046 and SB742457 are selective antagonists of HTR6 with a Ki of approximately 10 nM, while SB215505 and SB206553 are potent and selective HTR2B/ HTR2C antagonists 25 . We applied those compounds to SK-N-SH cells and detected the Aβ level in the medium 24 hours after treatment. Consistent with the effect of amoxapine, all three HTR6 antagonists reduced Aβ generation dose-dependently without an obvious effect on cell viability ( Fig. 2D and Supplementary Figure 2,F); however, HTR2B/HTR2C antagonists treatment did not change the extracellular Aβ level (Fig. 2E). Collectively, our data indicate that HTR6 mediated amoxapine's effect on Aβ generation. confirm the activity of amoxapine in a more relevant system, we used neuronal differentiated human iPSC-derived NSCs as a model, which were primarily taken as fibroblasts from people. The neuronal differentiated cells were evaluated for their neuronal linage property by immunostaining and quantitative RT-PCR. Most of the differentiated cells were positive for Tuj1 and Map2, markers for neurons, and negative for Sox2, a marker for NSCs (Fig. 3A). Furthermore, the transcription level of Nestin and Sox2 was significantly down-regulated in the induced neuronal cells (Sup. Fig. 3A), and meanwhile, Doublecortin (DCX), β-III tubulin (Tubb3), microtubule-associated protein 2 (MAP2), NeuroD1, synapsin 1 (SYN1), neural cell adhesion molecule (NCAM), and microtubule-associated protein Tau (MAPT) were all up-regulated (Sup. Fig. 3B), indicating the formation of neurons in the cell population. Interestingly, the genes encoding HTR6 and the GABAergic neuronal marker, vGAD67, were also up-regulated after the three-week differentiation ( Fig. 3B and Supplementary Figure 3,C). Using this neuronal cell model, we investigated the effect of amoxapine and SB258585 on Aβ generation. BSI IV and L685,458 again served as positive controls and significantly inhibited Aβ generation. In these cells, amoxapine and SB258585 consistently reduced Aβ generation at 10 μM by 36.97 ± 6.61% and 30.64 ± 6.98% (Fig. 3C). In addition, the knockdown of HTR6 diminished such action of those compounds (for amoxapine: from 29.609 ± 3.74% with shNC to 6.69 ± 8.16% with shHTR6-1 and 13.653 ± 7.73% with shHTR6-2; for SB258585: from 23.357 ± 4.22% with shNC to 0.11 ± 1.44% with shHTR6-1 and 3.043 ± 0.654% with shHTR6-2) (Fig. 3D). Taken together, our results indicate that amoxapine and SB258585 reduced Aβ generation through HTR6 in human neuronal differentiated NSCs.

Discussion
After being first marketed in the 1950s, TCAs have long been in clinical use as major anti-depressants for at least three decades. Nevertheless, the severe side effects have TCAs have cast a shadow ever since their development, and resulting in their gradual replacement by SSRIs or other more specific anti-depressive agents gradually in the late 1990s. As every cloud has a silver lining, there are scattered clinical data and several AD animal model results suggesting that TCAs may potentially improve AD symptoms through multiple mechanisms [11][12][13][14][15] . This study unintentionally identified that a typical secondary amine TCA, amoxapine, clearly reduced cellular Aβ generation in an Aβ-targeted high-throughput screening, which was further verified in several cell models. Though we have suggested that HTR6 is a major target for its Aβ-reducing effect, dirty drugs such as TCAs may exert their effect on the central nervous system (CNS) through not only HTR6 but also other GPCRs, transporters or unknown targets. Apart from the mechanism proposed in this study, we are still working on other possibilities that may be involved.
In the CNS, HTR6 is mainly expressed in neurons of limbic regions including the striatum, cortex and hippocampus and plays a role in regulating cognitive function including memory and mood 30,31 . Revealed by a number of animal behaviour studies, the knockdown or antagonism of HTR6 attenuate the memory deficits in AD animal model and shows anti-depressive activity [32][33][34] . Based on our secondary screening data, HTR6 is responsible for the Aβ-reducing activity of amoxapine, suggesting that compounds targeting HTR6 may modulate both cognition and Aβ generation, which directly fine-tunes AD pathogenesis. Additionally, three HTR6 antagonists tested in our experiment were also effective at reducing Aβ generation, although they showed differences in efficacy (Fig. 2D). Interestingly, among those antagonists, two compounds entering clinical phase III trials reduced Aβ generation to a smaller extent than SB271046, which had failed in a phase I clinical trial due to its poor ability to penetrate into the CNS after systemic dosing 35,36 . Considering the recent failure of idalopirdine in a phase III clinical trial, our data suggests that the Aβ-reducing activity may be worth considering when researchers design and modify the HTR6 antagonists for the development of AD drugs. However, whether HTR6 is a perfect molecular target against AD may need further investigation. In addition, SSRIs were less effective at reducing Aβ generation in the same cellular model compared with TCAs, although a minor but significant in vivo efficacy has been observed and reported by several groups 19,20 . All of these results indicate that compounds interfering with serotonin signalling may have an impact on Aβ generation.
In principle, HTR6 is a Gα s -coupled receptor with constitutive activity, and the decreased level of cAMP upon antagonist treatment was obviously correlated with reduced Aβ production, though it may not be the whole story 37 . In AD, there is mounting evidence implicating the participation of GPCRs in the modulation of APP processing and cognitive function. As reported by Ni. et al., Teng. et al. and Thathiah. et al., β2 adrenergic receptor (β2AR), δ-opioid receptor (DOR) and G-protein coupled receptor 3 (GPR3) regulate γ-secretase activity by direct interaction with secretases or via the assistance of β-arrestin2, which finally leads to a change in Aβ generation [38][39][40] . On the other hand, the angiotensin II receptor type 2 (AT2R) and HTR6 modulate AD pathogenesis by regulating glutamate and/or acetylcholine signalling 35,[41][42][43] . In our study, amoxapine modulated Aβ generation through targeting HTR6 in a β-arrestin2-dependent manner, as the knockdown of β-arrestin2 reduced amoxapine's effect (Fig. 4F). In addition, the downstream molecules of HTR6, including CDK5, may also contribute to the amoxapine's effect ( Fig. 4I and J). Our data indicates that compounds antagonizing HTR6 signalling may harbour disease-modifying activity and be beneficial for AD.
In spite of the numerous hypotheses that researchers have established to explain the aetiology and pathogenesis of psychiatric disorders such as depression and neurodegenerative diseases such as AD, the upstream determinants and downstream effects for those diseases always overlap, which indicates that drugs targeting one disease may bring some other effect to the CNS either directly or indirectly 44 . In our previous studies, anti-Parkinson's disease (PD) drugs, istradefylline, levodopa and piribedil, were all found to promote Aβ generation 45,46 . By either modulating the interaction between the adenosine A 2A receptor (A 2A R) and γ-secretase or regulating the D 2 R-mediated β-arrestin2-dependent signalling pathway, those anti-PD drugs may contribute to AD-like pathology. Conversely, we reported here that a group of TCAs harbour the Aβ-reducing activity. Our data suggest that a multi-target-directed ligand may introduce crosstalk between different neurological diseases and it is necessary to evaluate the potency of therapeutic candidates in the pipeline using multiple neurological disease models.
Cell Culture, Plasmids, and shRNA. SK-N-SH cells and HEK293 cells were purchased from ATCC.
The human HA-tagged HTR6 construct was kindly provided by Dr. Xie Xin (SIMM, China). The viral constructs encoding the HT6R or F20 were generated by subcloning the cDNA into a FUGW vector using the BamHI restriction sites. The constructs were verified by sequencing. The primers used are listed in Supplementary Table 1.
Lentiviral Constructs and Infection. Human embryonic kidney 293T (HEK293T) cells were seeded at a density of 7.5 × 10 6 cells in 100-mm dishes. On the following day, cells were transfected with 20 μg of shRNA constructs, 16 μg of pSPAX2, and 6 μg of pMD2G. The transfection was routinely performed by using the calcium phosphate transfection method. Cells were allowed to produce lentivirus for 48 hours. The virus-containing supernatant was collected and then centrifuged at 1000 × g for 5 min, before being passed through 0.45-μm filters. The lentiviruses were further concentrated by ultracentrifugation at 27,000 × g for 2 hours. The pellets were then ELISA for Aβ, sAPPα and sAPPβ. HEK293/APPswe cells, SK-N-SH cells, and induced human neuronal cells were treated with chemicals at the indicated concentrations for 24 h. The conditioned medium was then collected and subjected to a sandwich ELISA for the measurement of total Aβ level. The measurement was done according to the manufacturer's guidelines. ELISA kits for total human Aβ were obtained from ExCell Bio. ELISA kits for human sAPPα or sAPPβ were obtained from IBL.

Differentiation of Neuronal Cells from Human NSCs. The differentiation of NSC into neuronal cells
was performed according to the previous reports 47,48 with minor modification. In detail, NSC 13A cells were detached by accutase and resuspended in neural stem cell culturing basal medium containing 1:50 NSC supplement (IxCell). Then, 3 × 10 4 13A cells per well were seeded in 24-well plates coated with laminin (Sigma). On the second day, the medium was changed to neuron differentiation medium (Neurobasal, 1 × B27, 1 × N2, 100 nM cAMP, 1 μg/ml L-ascorbic acid, 10 ng/ml BDNF, 10 ng/ml GDNF and 10 ng/ml IGF-I). The medium was refreshed every two days.
Cell Viability Measurement. Chemical-treated SK-N-SH cells were subjected to the CellTiter-Glo Luminescent Cell Viability Assay (Promega) following the manufacturer's instructions. cAMP Assay. Intracellular cAMP was measured using a GloSensor TM cAMP assay following the manufacturer's instruction with minor modification (Promega). SK-N-SH cells were infected with HTR6 and F20-packaging lentivirus and seeded in white 96-well plates (Costar). Before the cAMP assay, the medium was removed and replaced with fresh medium containing 2% (v/v) GloSensor TM cAMP reagent. After 90 min incubation at 37 °C, cells were equilibrated at room temperature (RT) for 15 min and treated with the ligands at the indicated concentrations for another 15 min followed by the measurement of luciferase activity. Immunofluorescence Microscopy. The induced human neuronal cells grown on a coverslip were fixed with 4% paraformaldehyde (PFA) in PBS for 20 min. Cells were permeabilized and blocked with PBS/0.2% Triton X-100/1% BSA for 30 min followed by incubation with the indicated primary antibodies for 2 hours at RT. After washing with PBS/1% BSA three times, cells were incubated with Alexa Fluor 647-labelled donkey anti-goat IgG, Alexa Fluor 594-labelled donkey anti-rabbit IgG or Alexa Fluor 488-labelled donkey anti-mouse secondary antibodies in the dark for 1 hour, washed with PBS/1% BSA, stained with DAPI (1 μg/ml, 10 min), and mounted on slides. Images were acquired using CellInsight CX7 (Thermo Fisher Scientific) with a 20×/0.40 NA objective (Olympus).

Reverse Transcription and Quantitative Real-Time PCR.
Total RNA was extracted with TRI Reagent (T9424; Sigma) according to the manufacturer's instructions. Random hexamer primers and MMLV Reverse Transcriptase (M5301; Promega) were used for reverse transcription. All gene transcripts were quantified by quantitative real-time PCR performed with a 2 × HotStart SYBR Green qPCR Master Mix (ExCell Bio, Shanghai, China) on a Stratagene Mx3000P (Agilent Technologies). The primers used for the detection of mRNA levels of human genes are listed in Supplementary Table 3. All the primers were synthesized and purified by Shanghai Sunny Biotechnology Co., Ltd.
Statistical Analysis. All experiments were repeated at least three times. Data are representative or mean ± s.e.m. All data were analyzed by Prism 6.0 (GraphPad Software Inc., San Diego, CA). Concentration-response curves were analysed using a three-parameter non-linear regression analysis. Unpaired Student's t-test was applied for the comparisons of two datasets. One-way or Two-way analysis of variance (ANOVA) with post-hoc test was used where more than two datasets or groups were compared. Statistical significance was accepted at p < 0.05.