Amyloid-β reduces the expression of neuronal FAIM-L, thereby shifting the inflammatory response mediated by TNFα from neuronal protection to death

The brains of patients with Alzheimer's disease (AD) present elevated levels of tumor necrosis factor-α (TNFα), a cytokine that has a dual function in neuronal cells. On one hand, TNFα can activate neuronal apoptosis, and on the other hand, it can protect these cells against amyloid-β (Aβ) toxicity. Given the dual behavior of this molecule, there is some controversy regarding its contribution to the pathogenesis of AD. Here we examined the relevance of the long form of Fas apoptotic inhibitory molecule (FAIM) protein, FAIM-L, in regulating the dual function of TNFα. We detected that FAIM-L was reduced in the hippocampi of patients with AD. We also observed that the entorhinal and hippocampal cortex of a mouse model of AD (PS1M146LxAPP751sl) showed a reduction in this protein before the onset of neurodegeneration. Notably, cultured neurons treated with the cortical soluble fractions of these animals showed a decrease in endogenous FAIM-L, an effect that is mimicked by the treatment with Aβ-derived diffusible ligands (ADDLs). The reduction in the expression of FAIM-L is associated with the progression of the neurodegeneration by changing the inflammatory response mediated by TNFα in neurons. In this sense, we also demonstrate that the protection afforded by TNFα against Aβ toxicity ceases when endogenous FAIM-L is reduced by short hairpin RNA (shRNA) or by treatment with ADDLs. All together, these results support the notion that levels of FAIM-L contribute to determine the protective or deleterious effect of TNFα in neuronal cells.

One such protein is the long form of Fas apoptotic inhibitory molecule (FAIM) protein (FAIM-L).
The FAIM gene gives rise to two isoforms, the short (S) and the long (L) form. FAIM-S is widely expressed in most cells and tissues. 22 However, in the nervous system FAIM-S does not exert an anti-apoptotic function. 23 FAIM-L is expressed exclusively in neurons, where it serves as an antagonist of death induced by TNFR1 and FAS. 23 In this study, we found that FAIM-L expression is reduced in hippocampal samples from AD patients and also in a transgenic mouse model of the disease, PS1 M146L xAPP 751sl (PS1xAPP). In primary cortical neurons, Aβ reduced the expression of FAIM-L, thus suggesting that the expression of this protein is associated with the progression of the disease. We also show that the TNFα protection against Aβ toxicity is suppressed when FAIM-L expression levels are low (by RNA interference (RNAi) or by treatment with Aβ). Therefore, the response of neurons to the same inflammatory stimulus (TNFα) varies depending on the quantity of FAIM-L in the neuron. Thus, we provide a molecular basis through which to explain the protective or deleterious effect of TNFα in neuronal cells during the progression of AD.

Results
FAIM-L is reduced in hippocampal samples from AD patients and in the entorhinal and hippocampal cortex in the transgenic PS1xAPP mouse. The pathogenesis of AD involves multiple factors. In this regard, there are several lines of evidence indicating that TNFα signaling makes a considerable contribution to this disease. 24 Here we used quantitative PCR (qPCR) to systematically analyze in postmortem hippocampal samples from AD patients the proteins implicated in this signaling pathway, including the antagonists of DRs: CASP8 and FADD-like apoptosis regulator (CFLAR, aliases cFLIP-L), Lifeguard, and FAIM-L (Supplementary Information and Supplementary Figure S1).
Among the proteins analyzed, FAIM-L was most clearly altered during the progression of BRAAK stages. BRAAK staging describes the amount and distribution of neurofibrillary tangles (NFT). 25 This postmortem analysis is widely used because it has been found to correlate well with the severity of dementia. [26][27][28] As FAIM-L is expressed only in neurons and has been described as an antagonist of TNFα-induced death, 23 we focused on the role of this protein in AD. A notable decrease in FAIM-L mRNA was observed with the progression of the disease, being significant between control individuals (BRAAK 0 and BRAAK II) and demented AD patients (BRAAK V and BRAAK VI) (Figure 1a). Compared with nondemented patients, FAIM-L mRNA at BRAAK V decreased, although not reaching statistical significance. At BRAAK stage VI, FAIM-L mRNA was reduced to 30% of BRAAK 0 (individuals with no histopathological alteration of the brain but younger) or BRAAK II (asymptomatic subjects with similar age and with some histopathological modifications) (Figure 1a). Protein quantification of FAIM-L using the neuronal marker NeuN for the normalization revealed similar trend. Using NeuN allows discarding that the reduction of FAIM-L was due to neuronal death. In BRAAK V, we observed a diminution of FAIM-L levels but we were not able to detect significant difference, as there was one out of range sample. Nevertheless, at BRAAK stage VI, FAIM-L quantification showed a reduction of 50% compared with normal cognition individuals (BRAAK 0 or BRAAK II), confirming the results obtained by mRNA analysis (Figures 1b and c).
The results in human samples prompted us to perform similar analysis in the AD transgenic mouse model PS1xAPP. These animals reproduce the temporal and regional neurodegeneration and neuroinflammation that occur in the brains of AD patients. At 6 months of age, these animals show degeneration in principal neurons and somatostatin/neuropeptide Y (SOM/NPY) interneurons in the entorhinal cortex. 29 At this age, the analysis by qPCR in microdissected entorhinal cortex showed a significant reduction of FAIM-L mRNA in the transgenic animals compared with wild-type (WT) mice ( Figure 2A). In addition, the immunodetection of FAIM-L in this cortical region displayed a marked reduction with age in the transgenic animal ( Figure 2B).
In the hippocampus, PS1xAPP mice develop pyramidal neurodegeneration at ∼ 17-18 months of age. [30][31][32] Western blot quantification of the expression of FAIM-L in the hippocampus of these animals revealed a noteworthy decrease at 18 months ( Figures 2C and D). Thus, in both brain regions the decrease in FAIM-L expression occurs before the increase in neuronal death. 29,32 Aβ reduces the expression of FAIM-L. In order to analyze the factors affecting the FAIM-L expression, we treated primary mice cortical neurons with soluble fractions from the cortex of PS1xAPP animals of different ages. By western blot, we observed a dose-dependent reduction of FAIM-L, but not FAIM-S, in neurons treated with the soluble fractions of the transgenic animals but not those treated with the soluble fractions of WT animals ( Figure 3a). This reduction was significant for the soluble fractions from both 6-and 18-month-old animals at a final protein concentration of 100 μg/ml in the culture media ( Figure 3b). Thus, our data further confirmed in vitro what we have already observed in AD patients and in an AD animal model. We have previously reported the presence of oligomeric Aβ (oAβ) in the soluble fractions of these transgenic animals. 32 According to this, in cultured neurons these soluble fractions also induced death ( Figure 3c). Nonetheless, the death of neurons did not explain the decrease in FAIM-L levels as FAIM-L protein quantification was corrected with the neuron-specific marker NeuN (Figures 3a and b).
The hippocampi of PS1xAPP mice show an increase in soluble oAβ with age, specially the low-n oligomers. 32 Therefore we questioned whether oAβ causes the reduction in FAIM-L expression. To address this point, we treated primary neurons with increasing amounts of Aβ-derived diffusible ligands (ADDLs) and examined FAIM-L expression by western blot. The expression of FAIM-L, but not that of FAIM-S, in cultured neurons decreased in a dose-dependent manner in response to increasing concentrations of ADDLs (Figures 4a  and b). Moreover, the decrease in endogenous FAIM-L in response to ADDL treatment also occurred in the neuron-like cell model PC12 ( Figure 4c). As in human samples and the PS1xAPP mice, the downregulation of FAIM-L was also detected at the mRNA level in the cultured neurons (Figure 4d), thus suggesting that Aβ is modulating the expression of this protein rather than its degradation.
FAIM-L does not modify Aβ toxicity. FAIM-L is a potent antagonist of death induced by the activation of the death receptors (DRs) TNFR1 and FAS. 23 DRs are cell surface receptors that belong to the TNF superfamily and that trigger apoptosis upon ligand binding. There are conflicting reports regarding the mechanism by which Aβ kills neurons, with some describing the activation of the extrinsic DR-mediated apoptotic pathway. 13,[33][34][35] We found that 60% of the neuronal death induced by Aβ is apoptotic as cotreatment with ADDLs and the pancaspase inhibitor (Q-VD) led to a significant reduction in neuronal death compared with that induced by ADDLs alone (Figure 5a). Notice that the vehicle by itself produced some death, and that death was not reverted with Q-VD, indicating that the reduction in neuronal death in the combined treatment (ADDL+Q-VD) is due only to the prevention of apoptosis induced by ADDL. We next addressed whether FAIM-L serves as a functional antagonist of Aβ-induced neuronal death. For this purpose, we modulated the endogenous levels of FAIM-L using recombinant lentiviruses (Lv) for the overexpression (FAIM-L) or downregulation by short hairpin RNAi (shFAIM-L) (Figure 5c). In cultured neurons infected with these lentiviruses or the corresponding scrambled or empty forms, the neuronal death induced by Aβ toxicity did not show any relevant differences when FAIM-L expression was modified (Figure 5b).
FAIM-L is required for the protective effect of TNFα against Aβ-induced neuronal death. AD brains are characterized by the presence of neuroinflammatory mediators, including increased TNFα during the progression of the disease. [6][7][8][9] The contribution of TNFα to AD pathogenesis is highly controversial. It has been described that inflammation (c) FAIM-L protein expression normalized with the neuronal marker NeuN, thereby discarding the possibility that the reduction was caused by neuronal loss. Mann-Whitney nonparametric t-test. *P = 0.0382 BRAAK 0 versus BRAAK VI and & P = 0.0305 BRAAK II versus BRAAK VI. In both cases, data are mean ± S.D. of three independent experiments may be protective during onset of the disease, 32 although the persistence of inflammation may become detrimental. 36 We have previously described that TNFα activates pro-survival intracellular signaling pathways through cFLIP-dependent MAPK/ERK activation. 15 Thus, in the context of Aβ toxicity, we addressed whether TNFα has a protective or deleterious effect and also how this effect is regulated by FAIM-L. A relative high concentration of TNFα was used to reproduce the elevated levels of this cytokine in pathological conditions. [37][38][39] Co-treatment of neurons with TNFα and Aβ resulted in a significant prevention of cell death by 47%, considering the death induced by the vehicle as basal death (Figure 6a). More importantly, the protective effect of TNFα was dependent on the endogenous levels of FAIM-L as FAIM-L short hairpin RNA (shRNA) blocked the protective effects of this cytokine (Figure 6a).
The cleavage of caspase-3 gives a good correlation with neuronal death, as generally it is an indicator that the caspase cascade is triggered. Caspase-3 was activated in the cultures treated with ADDLs and abrogated in neurons co-treated with TNFα ( Figure 6b). When FAIM-L was reduced by shRNA, once again, we observed the appearance of the active caspase-3 fragment indicative of neuronal apoptosis ( Figure 6b). All these results support the notion that FAIM-L is required for TNFα protection against ADDLs.
To ensure the specificity of the FAIM-L reduction for the results obtained regarding TNFα protection, we generated a lentivirus that expresses FAIM-L with a silent mutation (smFAIM-L) that cannot be recognized by shFAIM-L, thus impeding its downregulation when both lentiviruses are co-infected ( Figure 6c). The smFAIM-L greatly increased the expression of FAIM-L (lane 3, Figure 6d), and this level was maintained when shFAIM-L was expressed simultaneously (lane 6, Figure 6d), in spite of shFAIM-L being highly efficient in reducing endogenous FAIM-L (lane 5, Figure 6d). In these conditions, that is, when the levels of FAIM-L were restored by smFAIM-L, TNFα regained its capacity to prevent the effects of Aβ toxicity in neurons ( Figure 6e). On the basis of our findings, we conclude that TNFα protects these cells against Aβ toxicity, but that this protection occurs only when FAIM-L is present in the neuron.
FAIM-L expression is not regulated by TNFα. TNFα activates the transcription of protective genes by triggering the transcription factor NFκB. 40,41 Thus, we addressed whether TNFα has the capacity to modulate FAIM-L expression. To this end, we treated primary cortical neurons with increasing amounts of TNFα and analyzed the expression of FAIM-L by western blot. TNFα alone did not modify the endogenous levels of FAIM-L at any of the concentrations tested. Neither in the simultaneous treatment with ADDLs was TNFα able to modify the decrease in endogenous FAIM-L induced by these ligands (Figures 7a and b). Accordingly, we conclude that TNFα does not have the capacity to reverse the downregulation of FAIM-L induced by Aβ treatment.
Reduction of FAIM-L by Aβ suppresses the protective action of TNFα. The results obtained hitherto show that Aβ decreases FAIM-L expression and also that this protein is required to ensure TNFα protection against neuronal Aβinduced death. We further examined the capacity of TNFα to protect neuronal cells against Aβ toxicity induced by several amounts of Aβ. To this end, we treated cortical neurons with increasing concentrations of ADDLs in the presence or absence of TNFα and assessed neuronal death by counting apoptotic nuclei (Figures 8a and b). Neuronal death induced by ADDL followed a dose-dependent trend, with notable death at 5 and 10 μM of Aβ. When neurons were also cotreated with TNFα, this cytokine significantly prevented cell death induced by 5 μM of ADDLs but not by 10 μM (Figures  8a and b). Western blot of caspase-3 cleavage showed a similar result; the treatment with 5 and 10 μM of ADDLs activated caspase-3, but co-treatment with TNFα prevented this activation only in the presence of 5 μM of Aβ (Figure 8c). We then sought to correlate these results with FAIM-L levels. In these conditions, as we have previously shown, there was a dose-dependent downregulation of FAIM-L expression when neurons were treated with increasing concentrations of Aβ. That is, FAIM-L expression decreased at 10 μM of Aβ, but expression at 5 μM was similar to that observed in vehicle-treated controls (Figure 8c). Thus, when neurons were treated with a concentration of ADDLs that reduced FAIM-L expression, TNFα was not able to exert a protective effect. Next, we checked whether restoring FAIM-L levels by exogenous overexpression would re-establish TNFαmediated protection. To carry out this experiment we used a concentration of ADDL of 8 μM because this quantity of ADDL ensured a detectable reduction in the levels of FAIM-L without significantly compromising the neuronal viability. The infection of neurons with a lentivirus that drives the exogenous overexpression of FAIM-L gave rise to a considerable increase in the levels of this protein, even when endogenous FAIM-L was reduced by treatment with 8 μM of  (Figure 8d). In these conditions, TNFα, partially but significantly, prevented neuronal death, as assessed by caspase-3 activation and apoptotic nuclei counts (Figures 8d and e). As expected, TNFα did not protect neurons infected with the empty virus when they were treated with 8 μM of ADDLs because in this condition the amount of FAIM-L was greatly reduced by Aβ (Figure 8e).
In conclusion, our results show that soluble Aβ downregulates the expression of FAIM-L in neurons and that this reduction impedes the TNFα-mediated protection against Aβ-induced neuronal death.

Discussion
Here we show that FAIM-L levels (protein and mRNA) are reduced in patients with AD and also in the regions most affected in this disease, namely the entorhinal cortex and the hippocampus, in the transgenic mouse model of the disease (PS1xAPP). Furthermore, when we added the soluble fractions from these transgenic animals to cultured neurons, FAIM-L expression also decreased. This effect was also observed when we treated neuronal cells with ADDLs. These observations thus reveal that Aβ downregulates FAIM-L expression. We also demonstrate that the reduction in FAIM-L, achieved by shRNA or by treating the cells with Aβ, impedes TNFα protection against Aβ-induced cell death. All together, these findings reveal a vicious circle in which, in addition to toxicity, Aβ accumulation alters FAIM-L expression in neurons, thus switching off the protective inflammatory response induced by TNFα. On the basis of these results, we propose a working model whereby normal levels of FAIM-L expression in neurons allow TNFα-mediated protection against Aβinduced death, thus resulting in a beneficial inflammatory response. However, when FAIM-L expression is reduced as a result of accumulation of Aβ, then TNFα cannot protect from Aβ, thus exposing neurons to death and contributing to neurodegeneration. Hence, the response of neurons to the same inflammatory stimulus (TNFα) differs in function of the levels of FAIM-L present. FAIM-L determines whether the TNFα response is beneficial or detrimental, therefore being critical for neuronal fate.
FAIM was first described as an antagonist of FAS. 42 The FAIM gene comprises six exons, with two putative translation initiation sites within exons II and III that generate two protein isoforms. 22 FAIM-S is widely expressed in many tissues, 22 whereas FAIM-L is expressed exclusively in neurons. 23 The mechanism by which FAIM-S expression is regulated in B cells has recently been described. 43 However, the regulation of FAIM-L expression has not been resolved. We have reported on the roles of the two isoforms in the nervous system. The short form is required for neurite outgrowth through the NFκB and MAPK/ERK pathways; however, FAIM-S is unable to protect neurons from death induced by trophic deprivation 44 or DRs. 23 Nonetheless, the neuronal long form of the protein does exert protection against TNFR1and FAS-induced apoptosis. 23 We recently described the molecular mechanism by which FAIM-L protects against the latter in neuronal cells. 45 However, the mechanism by which FAIM-L mediates TNFα protection against Aβ-induced cell death is, at present, still unclear. In this study we show that Aβ decreases FAIM-L expression without altering the expression of FAIM-S. This observation suggests that the expression of these two forms of the FAIM protein in neurons is subjected to differential regulation. In addition, in neurons the quantity of FAIM-L is higher than the FAIM-S, as evidenced in the qPCR in Figure 4. The reduction in FAIM-L is not attributable to neuronal death, as demonstrated by the neuronal loading marker NeuN and also by the observation that at 100 μg/ml the soluble fraction from 6-month-old transgenic mice caused a significant reduction in FAIM-L expression whereas the death induced was similar to that detected in age-matched WT animals. The finding that FAIM-L mRNA levels are also reduced indicates that Aβ modulates the expression of this protein rather than its degradation.
The downregulation of FAIM-L has important consequences for neuronal viability. This protein is a potent antagonist of TNFα-and FAS-induced death in neuronal cells. The knockdown of FAIM-L sensitizes neurons to death induced by these DRs. 23 In this regard, it has been suggested that the expression of FAIM-L is reduced in dopaminergic neurons in Parkinson's disease, thus making this type of neurons more vulnerable to FAS-induced death. 46 However, in general, neurons are resistant to death induced by DRs, 47,48 although DRs and ligands are expressed physiologically in the brain. [49][50][51] Indeed, TNFα-induced neuronal death requires the blockage of NFκB 17,18 or protein synthesis 19,20 or the depletion of FAIM-L expression. 23 Consistent with this, as shown in Figure 6, the knockdown of FAIM-L sensitized the neurons to TNFα; and this effect was reverted when the cells were co-infected with smFAIM-L to restore FAIM-L expression.
In the context of AD, the function of TNFα is controversial. It protects neurons from Aβ toxicity 10,11 but also contributes to their neurodegeneration. 12,13,33 Moreover its role in the aggregation of NFT has been scarcely explored. 52,53 In normal conditions TNFα acts as a modulator of brain functions, 54 participating in the control of synaptic transmission, 55,56 and   homeostatic plasticity also known as synaptic scaling. [57][58][59] The distribution of the glial cells surrounding neuronal synapses modulate these processes. 60,61 The fine-tuning control of the neurotransmission mediated by TNFα could explain the deleterious effects on synaptic transmission when deregulated. However, the neurodegeneration observed in AD probably requires the downregulation of FAIM-L mediated by Aβ. We assessed the neuronal viability with Hoechst counts and cleavage of caspase-3. Despite that activation of this caspase has been associated with other functions different to apoptosis, in our model its cleavage appeared to be linked to neuronal death as demonstrated with the pyknotic nuclei observed with Hoechst staining, and the apparition of the 89 kDa fragment produced by the cleavage of poly (ADPribose) polymerase (PARP) by caspase-3. The latter is a welldefining and an ending event in the apoptotic process. Our results demonstrate that the dual role on viability of TNFα depends on the degree of FAIM-L expression. Although the neuronal death induced by Aβ involves mainly the activation of caspases, FAIM-L has no role on this effect. However, the expression of this protein is required for the protection conferred by TNFα. The overexpression of FAIM-L did not enhance this protective effect, thereby indicating that FAIM-L by itself was not responsible for the prevention of Aβ-induced  ), and counts of nuclei after Hoechst staining were used to determine cell death. Graph represents mean ± S.E.M. ** , ## Po0.01 (*versus same treatment and # between ADDLs and ADDLs+TNFα). One-way ANOVA followed by multiple comparison Newman-Keuls test. In all cases, the experiments were repeated three times death when neurons were treated with TNFα. Interestingly, FAIM-L does not appear to be a target gene of the TNFinduced pro-survival pathway NFκB. Consequently, the treatment of neurons with TNFα did not reverse the downregulation of FAIM-L caused by Aβ.
Overall, our results explain why neurodegeneration is boosted by neuroinflammation when FAIM-L expression is reduced in neurons. The transgenic mouse PS1xAPP, which reproduces the human pathology and shows neuroinflammation and cortical neurodegeneration, also presents temporal and regional age-dependent vulnerability, the entorhinal cortex being the first area affected, followed by the hippocampus and subsequently an extensive part of the cortex. In the transgenic animal, both regions, the entorhinal cortex and the hippocampus, show a temporal correlation between reduced FAIM-L expression and the consequent increase in neuronal death. 29,30,32 Thus, on the basis of the in vitro results, we propose that the increase in soluble Aβ 62,63 promotes the reduction in neuronal FAIM-L expression and consequently the loss of protection afforded by TNFα against Aβ toxicity, hence increasing neuronal death. The same hypothesis could be formulated for patients with AD. Although the progressive reduction of FAIM-L mRNA with BRAAK stage severity may mask neurodegeneration, the quantification of protein using the neuronal marker NeuN also revealed a significant decrease in FAIM-L expression between AD patients and control nondemented subjects, thus indicating that the reduction in FAIM-L is before neurodegeneration. Our observations lead us to suggest that there is a threshold of FAIM-L below which neurodegeneration is accelerated, as documented in advanced BRAAK stages, 25 going from mild cognitive impairment (MCI) at BRAAK III-IV to AD at BRAAK V-VI. Presumably, FAIM-L is not involved in the etiology of the disease, but it can be determinant for the outcome. In this regard, we propose that the degree of FAIM-L expression or its evolution may be indicative of susceptibility to disease or disease progression.

Materials and Methods
Reagents. To prepare ADDLs, the soluble oligomeric forms of Aβ (oAβ), we followed previous protocols 64,65 with some modifications. Cold lyophilized human sequence Aβ 1-42 (Bachem, Weil am Rhein, Germany) was resuspended in icecold HFIP (1,1,1,3,3,3-Hexafluor-2-propanol) (Sigma-Aldrich, Barcelona, Spain) to a final concentration of 1 mM. The solution was incubated at room temperature (r.t.) until it was transparent. Solvent was then allowed to evaporate overnight until a peptide film formed. The films were stored at − 80ºC. The day before use, peptide films were resuspended in DMSO (Sigma-Aldrich) to a final concentration of 5 mM, and were further diluted to 100 μM with DMEM/F-12 Ham's-phenol free (Life Tech, Barcelona, Spain) and incubated at 4°C for 24 h in order to achieve oligomer formation. Before use, the solution was centrifuged for 10 min at 14 000 × g at 4ºC. Supernatants containing the oligomers were diluted in the cell culture media to the indicated final concentrations.
Lyophilized recombinant human TNFα (Biotrend, Köln, Germany) was diluted to 100 mg/ml in sterile DNase/RNase-free water. A final concentration of 100 ng/ml in culture media was used.
The pan-caspase inhibitor Q-VD-OPh (MP Biomedicals, Santa Ana, CA, USA) was diluted to 20 mM in sterile DNase/RNase-free water. A final concentration of 40 μM in culture media was used.
Human samples. The study was performed using samples from 22 cases stored in the BTIN-Tissue Bank for Neurological Research, Madrid, and in the Neurological Tissue Bank of IDIBELL-Hospital of Bellvitge, Barcelona. The cases were scored for BRAAK stage for neurofibrillary tangles (0-VI). BRAAK 0 and II correspond to individuals with normal cognition and BRAAK V and VI to AD patients. There are scarcity of subjects of a similar age to AD patients without any neurofibrillary tangles (BRAAK 0), and thus we also use BRAAK II (individuals with normal cognition who present some histopathological alterations of the brain) for the statistical comparisons. Table 1 provides details of the age, BRAAK stage, gender, and delay postmortem before sample extraction.
Samples of hippocampi (60-80 μg) were homogenized in 1 ml of Tripure Isolation Reagent (Roche, Sant Cugat del Vallès, Spain) as described previously for mouse samples. 29,31,32 After isolation of RNA samples, RNA integrity was assessed by agarose gel electrophoresis. The yield of total RNA was determined by measuring the absorbance (260/280), and the recovery of RNA was comparable in all the groups studied (0.3-0.4 μg/mg of tissue).
Retrotranscription (RT) was performed using random hexamers, 4 μg of total RNA as template, and the High-Capacity cDNA Archive Kit (Applied Biosystems, Barcelona, Spain), following the manufacturer's recommendations.
Laser capture microdissection of the entorhinal cortex was done as described previously. 29 Anesthetized animals (WT and PS1xAPP at 4 and 6 months of age; n = 4/type/age) were killed by decapitation, and brains were quickly removed and frozen. Brains were cut at 15 μm on a cryostat, and sections were collected on Superfrost Plus slides (Fisher Scientific, Waltham, MA, USA) and stored at − 80ºC. The entorhinal cortex from brain sections lightly stained with hematoxylin was identified in stereotaxic coordinates following the mouse brain atlas. 67 Laser capture microdissection was performed using a Pix-Cell II instrument from Arcturus Engineering (Mountain View, CA, USA). Cells were captured with a 30-μm laser setting and CapSure LCM Caps (Arcturus). The laser was set to a pulse of 60 mV for 50 ms. After microdissection, the capture disk was cleared of nonspecific adhering tissue using a CapSure Cleanup Pad (Arcturus). Entorhinal cortex RNA was extracted using the PicoPureTM RNA isolation kit. Capture disks with isolated material were microcentrifuged and incubated in the extraction buffer for 30 min at 42ºC. Cell extracts were then centrifuged and frozen at − 80ºC. For RNA isolation, cell extracts were treated as indicated by the manufacturer (Arcturus). The integrity and amount of purified RNA were analyzed using the RNA 6000 Pico Assay kit (Agilent Technologies, Barcelona, Spain). Retrotranscription (RT) was done using the High-Capacity cDNA Archive Kit (Applied Biosystems), following the manufacturer's recommendations. The RT-PCR was performed basically as previously described. 29 Soluble fractions from mouse brains at the corresponding ages were obtained as previously described. 32 Briefly, brains were homogenated with a Teflon-glass homogenizer in cold phosphate-buffered saline (PBS) with protease inhibitors (Sigma-Aldrich) and ultracentrifuged at 120 000 × g at 4°C for 1 h. Immediately, samples were aliquoted and store at − 80°C until use. Protein content was determined by means of the Lowry assay. Quantitative PCR. For human samples, specific gene products were amplified using the commercial hydrolysis TaqMan probes, following the instructions provided by the manufacturer (Applied Biosystems) using either an ABI Prism 7000 or 7900HT sequence detector (Applied Biosystems). A first run of amplification with three reference genes (18S, GAPDH, and actin) was done in order to determine equal amounts of the three genes used as standard. To perform the relative quantification using the comparative Ct delta method (Relative Quantification of Gene Expression, bulletin 2, Applied Biosystems), three independent reactions for each gene were done in parallel with the reference gene 18S. The qPCR in primary neuronal cultures was done using cells that were scraped out and washed once with ice-cold PBS. RNA was extracted following the protocol of the RNeasy Mini Kit (Qiagen, Barcelona, Spain). Equal amounts of RNA were converted to single-stranded cDNA using the High Capacity RNA-to-cDNA Kit (Applied Biosystems). The qPCR was performed with Sybr Green PCR Master Mix (Applied Biosystems) in an ABI Prism 7900 sequence detector (Applied Biosystems) with the conditions indicated by the manufacturer. The sequences of the primers, designed with span exon-exon junction, were as follows: faim-l (Mus musculus Fas apoptotic inhibitory molecule, transcript variant 1) forward 5′-GCGCGGAGCAGACTGT-3′, reverse 5′-TCCAGGCTATAGAGAGGGCTT-3′; faim-s (Mus musculus Fas apoptotic inhibitory molecule, transcript variant 2) forward 5′-GCGGAGCAGCCCTCTCTA-3′, reverse 5′-CCTGATGTGGTCCCATGTTCA-3′; l27 forward 5′-AGCTGTCATCGTGA AGAA-3′, reverse 5′-CTTGGCGATCTTCTTCTTGCC-3′; and GAPDH forward 5′-TGTGAACGGATTTGGCCGTA-3′, reverse 5′-ACTGTGCCGTTGAATTTGCC-3′. The results of the qPCR were normalized by l27. Similar results were obtained with GAPDH as the reference gene. Lentiviral particles. The lentiviral particles used to modify FAIM-L quantity in neurons were: short-hairpin RNA (shFAIM-L) to knock down, FAIM-L to overexpress, and the silent mutated (smFAIM-L) to overexpress FAIM-L when the endogenous FAIM-L was downregulated by the shFAIM-L; with their corresponding scrambled (sc) form or empty vector controls. 23 Lentiviral particles were produced as previously described. 23 Cell viability assays. Cell death was analyzed by nuclear Hoechst staining. At 4 h before the end of treatments, cells were stained by adding Hoechst 33258 to a final concentration of 0.3 μg/ml in the media. After this time, pictures were taken in a NIKON Eclipse TE2000-s (Nikon, Tokyo, Japan). Condensed or fragmented nuclei were counted as dead cells. 68 For each condition, 800 to 1000 cells were counted in fields of 100-200 neurons/field. All the counts were repeated at least three times in independent cultures.
Western blot. Neurons were harvested and washed once with ice-cold PBS, then lysed in triton lysis buffer containing 50 mM Tris-HCl, pH 7.4, 150 mM NaCl, 10 mM EDTA, and 1% Triton with a cocktail of protease (Roche) and phosphatase (Sigma-Aldrich) inhibitors for 30 min at 4ºC. Supernatants were obtained after centrifugation at 14 000 × g for 10 min at 4ºC. Protein concentration was quantified by a modified Lowry assay (DC protein assay; Bio-Rad, Hercules, CA, USA). Next, 7-20 μg of protein was resolved by SDS-polyacrylamide gels and transferred onto PVDF (Millipore Ibérica, Madrid, Spain). Membranes were incubated with blocking solution (5% non-fat dry milk in TBS with 0.1% tween-20 (TBS-T)) for 1 h at r.t. Primary antibodies prepared in blocking solution were incubated overnight at 4ºC. After washes with TBS-T, membranes were probed with the appropriate specific peroxidase-conjugated secondary antibody and developed using the EZ-ECL chemiluminescence detection kit (Biological Industries, Kibbutz Beit Haemek, Israel). The primary antibodies used were anti-FAIM-L (antibody generated with the specific sequence presented in FAIM-L) and anti-FAIM-S (both in-house, tested and validated for both WB and immunohisto/cytochemistry 23 ); anti-caspase-3, anti-PARP (Cell Signalling Technologies, Beverly, MA, USA); anti-NeuN, anti-panERK (Millipore); and anti-tubulin (Sigma-Aldrich). Statistical analysis. All the experiments were performed at least three times. The statistics used are indicated in the figure legends with the significance obtained. For the analysis of human samples, we used nonparametric tests: the equivalent to ANOVA test Kruskal-Wallis followed by Dunn's multiple comparison test; and the nonparametric t-test Mann-Whitney. For the statistical analysis of the in vitro experiments, we used parametric tests: one-way ANOVA with multiple Newman-Keuls test to analyze the differences between groups and t-test to compare two sets of data.

Conflict of Interest
The authors declare no conflict of interest.