Facilitation of glutamate, but not GABA, release in Familial Alzheimer's APP mutant Knock‐in rats with increased β‐cleavage of APP

Abstract Amyloid precursor protein (APP) modulates glutamate release via cytoplasmic and intravesicular interactions with the synaptic vesicle release machinery. The intravesicular domain, called ISVAID, contains the BACE1 cleavage site of APP. We have tested the functional significance of BACE1 processing of APP using App‐Swedish (App s) knock‐in rats, which carry an App mutation that causes familial Alzheimer's disease (FAD) in humans. We show that in App s rats, β‐cleavage of APP is favored over α‐cleavage. App s rats show facilitated glutamate, but not GABA, release. Our data support the notion that APP tunes glutamate release, and that BACE1 cleavage of the ISVAID segment of APP facilitates this function. We define this phenomenon as BACE1 on APP‐dependent glutamate release (BAD‐Glu). Unsurprisingly, App s rats show no evidence of AD‐related pathology at 15 days and 3 months of age, indicating that alterations in BAD‐Glu are not caused by pathological lesions. The evidence that a pathogenic APP mutation causes an early enhancement of BAD‐Glu suggests that alterations of BACE1 processing of APP in glutamatergic synaptic vesicles could contribute to dementia.

modulation of SV function. Indeed, interactome and electrophysiology evidence suggests that cleavage of APP in the ISVAID segment by either β-or α-secretase may reduce or abolish the intravesicular interactions of APP with SV proteins, leading to facilitation of glutamatergic SV release (Yao et al., 2019). The acidic pH of SVs favors β-cleavage, and there is an enrichment of β-cleaved APP metabolites in SV subcellular fractionations (Del Prete et al., 2014). Thus, it is reasonable to hypothesize that processing of APP by β-secretase, rather than α-secretase, may cut inside the ISVAID of APP and destabilize these interactions, thereby facilitating excitatory neurotransmission. The evidence that a BACE1 inhibitor causes strong reduction in the frequency of sEPSC/mEPSC (Filser et al., 2014) and that BACE1 KO mice show an increase in PPF ratio, which is indicative of a reduction in presynaptic release (Wang, Song, Laird, Wong & Lee, 2008), are consistent with this hypothesis.
However, β-secretase may process several synaptic substrates that may affect synaptic transmission; thus, whether the effects of pharmacological and genetic ablation of β-secretase activity are due to inefficient β-processing of APP is unclear. To directly test this hypothesis, a system in which App processing by BACE1 is increased without changing BACE1 activity is needed. A naturally occurring APP mutant allele, the pathogenic Swedish APP allele, codes for an APP protein (herein called APPSw) carrying the amino acids substitutions K670N/M671L. These mutations, which are localized at the NH 2 -terminus of the β-cleavage site of APP, cause increased cleavage of APP by β-secretase (Citron et al., 1992(Citron et al., , 1994Johnston et al., 1994). Thus, a model organism in which APPSw replaces wild-type APP should be apt to test whether APP processing by β-secretase facilitates excitatory neurotransmission.
To test this hypothesis, we introduced the Swedish APP mutations into the genomic App rat locus to generate App S Knock-in (KI) rats. Rat and human APP differ in the Aβ region by 3 amino acids.
Given that aggregated or oligomeric forms of Aβ are by and large considered the main pathogenic entity in AD and given that human Aβ may have higher propensity to form toxic Aβ species as compared to rodent Aβ, together with the Swedish mutations we introduced mutations to "humanize" the rat Aβ sequence. As controls, we produced rats carrying only the humanized Aβ sequence (App h rats, APPh protein). We choose a KI approach because (a) KIs mimic the genetics of familial dementia and make no assumption about pathogenic mechanisms (except the unbiased genetic one); (b) expression of mutant genes is controlled by endogenous regulatory elements in physiological quantitative-spatial-temporal manner, thereby allowing us to test the hypothesis in a biologically relevant model organism system.

| App mRNA expression is normal in App h and App s KI rats but is greatly reduced in App δ7 rats
Founder (F0) rats carrying the h, s, and δ7 mutations were generated as described in the Experimental Procedures section. F0-App h/δ7 and F0-App s rats were crossed to WT (App w/w ) Long Evans rats to generate F1-App δ7/w , F1-App h/w , and F1-App s/w rats. The δ7 mutant allele was a product of aberrant homology-directed repair but is useful. As shown in Figure 1a, this 7-bp deletion in exon 16 causes a frameshift that would produce a truncated soluble protein (sAPPδ7, missing the transmembrane region of APP) with a novel COOH-terminal sequence. Alternatively, this mutation could also produce a hypomorphic allele. F1-App δ7/w , F1-App h/w , and F1-App s/w rats were crossed to WT Long Evans to generate F2-App δ7/w , F2-App h/w , and F2-App s/w rats. These crossing were repeated three more times to obtain F5-App δ7/w , F5-App h/w , and F5-App s/w rats. The probability that F5 rats carry unidentified off-target mutations (except those, if present, on Chr. 11) is ~1.5625%. Male and female F5-App δ7/w , F5-App h/w , and F5-App s/w rats were crossed to obtain App δ7/δ7 , App h/h , and App s/s rats.
To verify that the humanizing and Swedish mutations were correctly inserted into App exon-16, we amplified by PCR the App gene exon-16 from App w/w , App h/h , and App s/s rats. Sequencing of the PCR products shows that the humanizing mutations and the humanizing plus Swedish mutations were correctly inserted in the App h/h and App s/s genomes, respectively ( Figure 1a).
To determine whether the introduced mutations disrupt App expression, we examined App mRNA levels in 21-day-old App w/w , App δ7/δ7 , App h/h , and App s/s rats (2 females and 3 males for each genotype). In App δ7/δ7 rats, App mRNA was virtually absent, indicating that the App δ7 mutation is a hypomorphic mutation. In contrast, expression of App in App h/h and App s/s brains was comparable to that detected in App w/w rats ( Figure 1b). Following are the statistical data obtained by ordinary one-way: ANOVA summary: F = 9.588, p = .0007 (significant = ***). Post hoc Tukey's multiple comparisons test: App w/w vs App s/s , p = .9800 (not significant); App w/w vs App δ7/δ7 , p = .0022 (significant=**); App w/w vs App h/h , p = .9970 (not significant); App s/s vs App δ7/δ7 , p = .0048 (significant = **); App s/s vs App h/h , p = .9352 (not significant); App h/h vs App δ7/δ7 , p = .0015 (significant = **).

| The proteins encoded by the App h and App s alleles contain the humanizing and Swedish mutations, while the App δ7 is an hypomorphic App allele
To verify whether the protein products of the App h and App s alleles contain the humanizing mutations, we used the following anti-APP antibodies. Y188, a rabbit polyclonal raised against the COOH-terminal 20 amino acids of APP, an epitope that is unchanged by the humanizing and Swedish mutations. M3.2, a mouse monoclonal raised against the rat APP sequence between the β-and α-secretase cleavage sites (DAEFGHDSGFEVRHQK); this antibody will only recognize APP molecules containing the rat Aβ sequence. 6E10, a mouse monoclonal raised against the corresponding domain of human APP (DAEFRHDSGYEVHHQK, the 3 amino acid differences with the rat sequence are underlined); this antibody will only recognize APP molecules containing the human F I G U R E 1 Characterization of App h , App s , and App δ7 KI rats. (a) Left panel. To verify that the humanizing and Swedish mutations were correctly inserted in the App exon-16, we amplified by PCR the App gene exon-16 from App w/w , App h/h , and App s/s rats. Sequencing of the PCR products shows that the humanizing mutations and the humanizing plus Swedish mutations were correctly inserted into the App h/h and App s/s genomes, respectively. Sequencing analysis of genomic DNA confirms the expected G to C, T to A, and GC to AT substitution in App h/h rats and the GA to TC, G to C, T to A, and GC to AT substitution in App s/s rats. Substituted nucleotides are highlighted in gray. The amino acid sequences are indicated above the DNA sequences and the amino acid substitutions introduced by the mutations are highlighted in gray (GA to TC=KM to NL; G to C=G to R; T to A=Y to F; and GC to AT=R to H). Right panel. Predicted sequences of WT and App δ7 cDNAs and proteins. For space reasons, only Exon 16 (black) and Exon 17 (red) are shown. The 7-bp deletion (in white characters and boxed in black), the transmembrane domain of APP (underlined), and the novel predicted COOH-terminal of sAPP δ7 (bold and italic) are indicated. (b) Levels of App mRNA were measured in 21-day-old App w/w , App δ7/δ7 , App h/h , and App s/s rats (2 females and 3 males for each genotype). App mRNA expression was normalized to Gapdh mRNA expression. The δ7 mutant allele was a product of aberrant homology-directed repair but is useful because this mutation will either produce a truncated soluble APPδ7 protein (sAPPδ7) or a hypomorphic allele. Data were analyzed by ordinary one-way ANOVA followed by post hoc Tukey's multiple comparisons test when ANOVA showed statistically significant differences and presented as average (App/Gapdh) ± SEM. (c) Schematic representation of APPWT rat, APPh, APPSw, and the metabolites derived from α-and β-secretase processing. The amino acid changes that humanize the Aβ region of are in red and underlined (G>R, F>Y, R>H); the amino acids changes that introduce the Swedish mutation are in blue and underlined (K>N, M>L); the epitopes recognized by antibodies are highlighted as follows: Y188 is highlighted in gray, M3.2 is highlighted in green, 6E10 highlighted in yellow, anti-sAPPα is highlighted in cyan, anti-sAPPβWT is highlighted in red, and anti-sAPPβSw is highlighted in magenta. The αCTF and βCTF derived from APPh and APPSw are identical; thus, in the paper, we will refer to both as αCTF or βCTF. The App δ7 allele could produce sAPPδ7: The sequence in white highlighted in black indicates the novel amino acid sequence produced by the deletion of 7 bp, which causes a frameshift. Casp. Indicates the site of cleavage of APP in the cytoplasmic region by caspases, which leads to the generation of JCasp (Gervais et al., 1999;Pellegrini, Passer, Tabaton, Ganjei & D'Adamio, 1999). (d) Summary of the expected immunoreactivities of the anti-APP antibodies used in this study: + = positive reactivity; − = no reactivity. (e) Western blot analysis of postnuclear supernatant isolated from App w/w , App δ7/δ7 , App h/h , and App s/s rats with Y188 (detects mAPP, imAPP, αCTF, and βCTF from all animals except the App δ7/δ7 rats). (f) M3.2 (detects only rat WT mAPP, imAPP, and βCTF). To verify that APPSw contained the Swedish mutations, we used two antibodies raised against the COOH terminus of wild-type human sAPPβ (sAPPβWT), one from IBL the other from Covance, and an antibody raised against the COOH terminus of Swedish sAPPβ (sAPPβSw) (the KM>NL mutations are in the 2 COOH-terminal residues of sAPPβ). The two anti-sAPPβWT should detect sAPPβ produced from APPh (called sAPPβh) but not sAPPβ produced from APPSw (called sAPPβSw); conversely, the anti-sAPPβSw should only detect sAPPβSw and not sAPPβh. The specificities of these antibodies are summarized in Figure 1c,d. As expected, analysis of soluble brain fractions (S70) isolated from one App h/h , one App s/s , and one App δ7/δ7 showed that the two anti-sAPPβWT detected sAP-Pβh only (Figure 1h), while the anti-sAPPβSw only detected sAP-PβSw ( Figure 1i); none of the antibodies gave a positive signal in the App δ7/δ7 sample.
To complete our preliminary biochemical analysis of soluble fractions, we performed Western blots with 22C11, a monoclonal antibody raised against the ectodomain of APP that will recognize all sAPP and sAPPβ species as well as sAPPδ7, and an anti-sAPPα monoclonal antibody, which is raised against the COOH terminus of sAPPα and is specific for sAPPα (see Figure 1c,d). 22C11 (Figure 1j) detected equal amounts of sAPP species in both App h/h and App s/s but gave no specific signal in the App δ7/δ7 sample (no sAPPδ7 detected), confirming that δ7 is a hypomorphic App allele. Anti-sAPPα detected both sAPPαh (sAPPα derived from APPh) and sAPPαSw (sAPPα derived from APPSw) ( Figure 1k). Overall, these data indicate that (a) the pattern of immunoreactivity of APP and its metabolites F I G U R E 2 Increased processing by β-secretase and decreased processing by α-secretase of APPSw. To test whether the App s/s rats present the expected changes in APP metabolism, we analyzed brain samples isolated from 21-day-old App h/h and App s/s rats, 2 females and 3 males for each genotype. (a) Western blot (WB) of postnuclear supernatant from App s/s and App h/h rats with Y188 and 6E10 (bottom). (b) WB analysis of soluble brain fractions with an ant-sAPPα antibody showed that levels of sAPPα are significantly lower in App s/s compared to App h/h brains. (c) Because sAPPβh and sAPPβSw cannot be detected by the same antibody (see Figures 3a,b and 4d,e), to compare sAPPβh and sAPPβSw amounts, we used 6E10 and 22C11. Signal intensity was quantified with Image Lab software (Bio-Rad). Quantification is shown on the right. (d) Aβ is produced by γ-cleavage of βCTF, which is increased in App s/s brains. ELISA measurements of endogenous Aβ40 and Aβ42 in brain homogenates of 21-day-old animals. Levels of steady-state endogenous Aβ40 and Aβ42 are increased in App s/s brains as compared to App h/h brains. The Aβ42/Aβ40 ratio did not change. These ELISA kits are specific for human Aβ40-42, as attested by the fact that they gave no signal on brain homogenates from App w/w (expressing rat Aβ) and App δ7/δ7 (expressing no Aβ) rats (not shown). Data are represented as mean ± SEM. Data were analyzed by Student's t test and shown as average ± SEM. **p < .01; ****p < .0001 confirm that the predicted mutations have been properly integrated into APPh and APPSw proteins; (b) App δ7 is a hypomorphic App allele.

| Increased β-processing of APP in Familial Alzheimer disease App-Swedish Knock-in rats
As noted, several reports indicate that APPSw is a better βsecretase substrate as compared to WT APP (Citron et al., 1992(Citron et al., , 1994Johnston et al., 1994). To test whether these APP metabolic changes are reproduced in our KI rats, we analyzed brain samples isolated from 21-day-old App h/h and App s/s rats, two females and three males for each genotype. App s/s rats showed increased βCTF and decreased αCTF/mAPP levels-while levels of imAPP were unchanged ( Figure 2a). α-Secretase cleaves mAPP in the secretory pathway and on the plasma membrane releasing sAPPα into the extracellular fluid. WB analysis of soluble brain fractions showed that levels of sAPPα are significantly lower in App s/s compared to App h/h brains ( Figure 2b). β-Secretase cleaves mAPP in acidic intracellular organelles, such as late endosomes and synaptic vesicles (SV), and a fraction of sAPPβ is released extracellularly during exocytosis.
Thus, significant amounts of sAPPβ are present in both intra-and extracellular compartments. Hence, total brain homogenates were used to quantify sAPPβ levels. Because sAPPβh and sAPPβSw cannot be detected by the same antibody (Figures 1c, d, h and i), to compare sAPPβh and sAPPβSw amounts we used 6E10 and 22C11 antibodies: 6E10 detects imAPP, mAPP, and sAPPα proteins; 22C11 detects imAPP, mAPP, sAPPα, and sAPPβ proteins. These APP metabolites are close in size cannot be separated by SDS-PAGE. As shown in Figure 2c, the imAPP + mAPP + sAPPα signal revealed by 6E10 was significantly lower in App s/s as compared to App h/h brains. This is consistent with the finding that mAPP ( Figure 2a) and sAPPα ( Figure 2b) are significantly lower in App s/s brains. In contrast, the signal revealed by 22C11 (imAPP + mAPP + sAPPα + sAPPβ) was significantly higher in App s/s animal ( Figure 2c). Hence, levels of sAPPβSw must be significantly higher than levels of sAPPβh. Data were analyzed by Student's t test, and significant differences are shown in the Figure (*p < .05; **p < .01, ***p < .001, ****p < .0001). In summary, the evidence that App s/s brains contain reduced amounts of αCTF/sAPPα and increased levels of βCTF/sAPPβ, indicates that APPSw is cleaved more efficiently by β-secretase and less efficiently by α-secretase as compared to APPh. Whether this reduction in sAPPα is due to altered affinity of APPSw for α-secretase and/or reduced availability of substrate (APPSw) for α-processing remains to be determined.
Aβ is produced by γ-cleavage of βCTF, which is increased in App s/s brains, thus explaining why App s/s rats produce significantly F I G U R E 3 The Swedish mutations do not alter the effect of Ex/TM on glutamate release. (a) Sequence of the Ex/TM and Ex/TM-Sw peptides. (b) Average PPF at 50 and 200 ms ISI. Representative traces of EPSCs evoked at 50 ms ISI are shown. Ex/TM and Ex/TM-Sw significantly decrease PPF. (c) Ex/TM and Ex/TM-Sw significantly increase mEPSC frequency. Amplitudes and decay time of mEPSCs were not changed by these peptides. Representative recording traces of mEPSCs are shown. Data were analyzed by ordinary one-way ANOVA followed by post hoc Tukey's multiple comparisons test when ANOVA showed statistically significant differences. Four male and four female rats were used for each group. The number of recordings analyzed for each group is indicated inside the bars. All data represent means ± SEM more human Aβ40 and Aβ42 as compared to control animals ( Figure 2d). The Aβ42/Aβ40 ratio did not change: This is predictable since the βCTFs derived from APPh and APPSw are identical and γ-processing of βCTF should not change. Overall, App s KI recapitulates the biochemical changes of human APPSw metabolism, that is increased processing by β-secretase (Citron et al., 1992(Citron et al., , 1994Johnston et al., 1994). Also, we describe previously unknown decreased processing of APPSw by α-secretase and decrease in mAPPSw levels.

| App-Swedish rats show normal brain development and no observable neurodegeneration and AD-like pathology at 3 months of age
We used histological and immunohistochemical (IHC) analyses to test whether these alterations in glutamate release precede and or co-occur with AD-like pathology. NeuN stained tissue was used to assess neuronal density in the frontal cortex, retrosplenial, piriform, and entorhinal cortex, as well as the anterior and posterior hippocampus.

| D ISCUSS I ON
Human genetic data suggest that APP processing plays a significant role in the pathogenesis of both familial and sporadic dementias.  Tamayev, Giliberto et al., 2010). BACE1 gene polymorphisms as well as increased BACE1 expression/activity are associated with sporadic dementia (Cheng et al., 2014;Hampel & Shen, 2009;Hebert et al., F I G U R E 5 The App s/h and App s/s rats do not show AD-like histopathology at 15 days and 3 months of age. Representative IHC and histology from the anterior hippocampus in a representative 15-day-and 3-month-old subject for each genotype. Three female and three male rats per each genotype and age were tested 2008; Holsinger, Lee, Boyd, Masters & Collins, 2006;Jo et al., 2008;Kan et al., 2005;Long, Ray & Lahiri, 2014). In contrast, humans carrying the Icelandic APP variant, which codes for an APP protein that is inefficiently cleaved by BACE1, are protected from dementia and normal cognitive decline (Jonsson et al., 2012). This genetic link between APP processing and dementia has provided some of the foundation for the "amyloid hypothesis," which indicts Aβ as the main pathogenic factor responsible for neurodegeneration.

Mutations in
Yet, experimental evidence from model organisms-in conjunction with the plethora of failed clinical trials targeting Aβ production/ clearance/deposition-questions the central role of amyloid peptides in dementia. Transgenic mice that express mutant APP using a tet-Off vector systems (APPsi:tTA mice) show high amyloid burden and short/long-term memory deficits. Suppression of APP expression after memory deficits ensue causes rapid decline in the brain levels of soluble full-length APP, sAPPα, sAPPβ, αCTF, and βCTF and significantly improves memory deficits in spite of persisting amyloid deposits, soluble, and oligomeric assemblies of Aβ2 (Melnikova et al., 2013). BRI2-Aβ mice produce high levels of Aβ peptides and BRI2-Aβ1-42 mice develop amyloid pathology that is similar to that observed in mutant human APP transgenic models ). Yet, BRI2-Aβ1-42 mice show intact cognitive performance both pre-and postamyloid plaque formation .

Knock-in models of Familial British and Danish dementia (FBD and
FDD KI mice), two AD-like dementias that are believed to be caused by other amyloidogenic moieties (ABri and ADan+Aβ42, respectively), develop long-term potentiation (LTP) and memory deficits in the absence of amyloidosis (Giliberto et al., 2009;Tamayev, Giliberto et al., 2010;Tamayev, Matsuda et al., 2010). Memory and LTP deficits are mediated by APP and/or its non-Aβ metabolites . Selective reduction of APP processing by BACE1 ameliorates memory and long-term potentiation impairments ; in contrast, inhibiting Aβ production worsens them .
While the β-processing of APP has been studied and targeted for its capacity to produce Aβ, we postulate that the β-processing of APP may also modulate the function of the full-length precursor protein, as well. Specifically, we have found that β-secretase cleaves a functional domain of APP called ISVAID, which interacts with synaptic vesicle proteins. Interactomic and electrophysiology studies suggest that β-cleavage may reduce or abolish the intravesicular interactions of APP with SV proteins, leading to facilitation of glutamate release (Yao et al., 2019). Published evidence using BACE1 inhibitors and BACE1 KO mice is consistent with this hypothesis (Filser et al., 2014;Wang et al., 2008). However, BACE1 has many substrates that function at the synapse, and off-target effects of pharmacological and/or genetic ablation of β-secretase activity would obscure the relative effect of the β-processing of APP on synaptic transmission alterations. To test this hypothesis directly, we generated the App s KI rats, which carry the pathogenic Swedish APP mutations. APPSw, the mutant protein coded for by the Swedish APP allele, carries the amino acids substitutions K670N/M671L, which are localized at the NH 2 -terminus of the β-cleavage site of APP. Consistent with what has been observed in cell lines (Citron et al., 1992(Citron et al., , 1994Johnston et al., 1994), in App s/s KI rats, cleavage of APPSw by BACE1 is increased (Figure 2) without altering BACE1 activity. In parallel, these rats show augmented glutamate release at SC-CA3>CA1 pyramidal cell synapses. It is worth nothing that rats carrying one wild-type and one Swedish allele (App s/h ), which genetically mimic patients since the Swedish mutation is pathogenic in heterozygosity in humans, have an intermediate phenotype. In fact, App s/h rats have only decreased PPF at 50 ms ISI, while App s/s KI rats show also increased mEPSCs frequency. These gene dosage-dependent effects suggest that the Swedish mutation may have an accelerated pathogenic effect in humans that may carry two mutant alleles.
In contrast, the APP Swedish mutation did not significantly alter GABA release, which is consistent with the evidence that the SV-APP-interacting networks may be restricted to glutamatergic SV and may modulate excitatory but not inhibitory synaptic transmission (Yao et al., 2019). Thus, the Swedish APP mutation causes an excitation/inhibition imbalance, favoring excitation. In addition, our data indicate that AD-like pathological lesions are not driving the changes in BAD-Glu caused by the Swedish mutation.
What are the molecular mechanisms underlying deregulation of BAD-Glu by APPSw? The Swedish mutation causes increased β-processing and decreased α-processing of APP ( Figure 2). Therefore, the direct products of β-cleavage (sAPPβ and βCTF) are increased while the metabolites formed by α-cleavage (sAPPα and αCTF) are decreased. Since APP-CTFs are substrates of γ-secretase, Aβ, which is produced by γ-cleavage of βCTF, is also increased, while P3, a metabolite produced by γ-processing of αCTF, should be decreased-albeit we do not experimentally show this. In addition, levels of mAPP are also reduced. Thus, these metabolic alterations may contribute to dysregulation of BAD-Glu in App s/s KI rats (Figure 6a).
Another possibility to bear in mind is that the amino acid substitutions K670N/M671L will alter the primary structure of several APP metabolites including full-length APP, sAPPα, and sAPPβ.
Although the K670N/M671L mutations do not seem to impact the intravesicular interactions of APP (Figure 3), it is possible that they may contribute to deregulation of BAD-Glu (Figure 6b).
Finally, β-cleavage of APP in the ISVAID may directly facilitate glutamate release via a negative modulation of intravesicular interactions, which in turn may facilitate excitatory transmission by functionally enabling the cytosolic interactions. In this model, APP would work as a fine-tuning unit of glutamate, but not GABA, release with β-secretase representing the rheostat (Figure 6c). Conditions that augment the rheostat activity, such as the pathogenic APP Swedish mutation, may favor excitation over inhibition, neuronal hyperexcitability, and a pro-epileptogenic condition. Of note, unprovoked seizures occur in AD patients at rates 8-to 10-fold higher than in the general population (Hauser, Morris, Heston & Anderson, 1986;Scarmeas et al., 2009) and at even higher rates in FAD cases (Cabrejo et al., 2006;Mendez & Lim, 2003;Palop & Mucke, 2016). It has also been reported that anti-epileptic drug levetiracetam rescues cognitive deficits in MCI patients (Bakker et al., 2012). These observations are compatible with the idea that increased glutamatergic tone may have an important pathogenic role, at least in a subset of dementia patients.
All these potential mechanisms do not need to be mutually exclusive and may coincide to result in the dysregulation of BAD-Glu seen in Swedish mutants. Future studies, including longitudinal cognitive assessment and pathology analyses of Swedish rats, are needed to assess whether this early synaptic alteration caused by APPSw underlies pathogenic mechanisms leading to neurodegeneration.

| Rats and ethics statement
Rats were handled according to the Ethical Guidelines for Treatment of Laboratory Animals of the NIH. The procedures were described and approved by the Institutional Animal Care and Use Committee (IACUC).

| Rat brain preparation
Brains were homogenized using a glass-Teflon homogenizer (w/v = 100 mg tissue/1 ml buffer) in 250 mM sucrose, 20 mM Trisbase pH 7.4, 1 mM EDTA, 1 mM EGTA plus protease, and phosphatase inhibitors (ThermoScientific), with all steps carried out on ice or at 4°C. Homogenates were centrifuged at 800 g for 10 min.
Supernatant was collected and labeled S1 and used for Western analysis. Soluble fractions were generated by ultracentrifugation of S1 at 70 000 g for 1 hr to obtain S70 and P70. S70 was used for Western analysis of soluble APP content. Soluble fractions for ELISA were generated by solubilization of S1 with 0.1% SDS and 1% NP-40 for 30 min rotating. Solubilized S1 was spun at 20 000 g for 10 m and analyzed by ELISA.

F I G U R E 6
Modeling BAD-Glu mechanisms and how the Swedish mutation may alter BAD-Glu. (a) APP undergoes complex proteolysis. In the amyloidogenic proteolytic cascade, APP is cleaved by β-secretase/BACE1 into sAPPβ and the COOH-terminal fragment βCTF. Cleavage of βCTF by γ-secretase produces Aβ peptides and the intracellular domain (AID/AICD). Alternatively, APP is processed by α-secretase into sAPPα and the COOH-terminal fragment αCTF. αCTF can be cleaved by γ-secretase to produce P3 and AID/AICD (Passer et al., 2000;Sisodia & St George-Hyslop, 2002). The Swedish mutations cause more β-cleavage with a corresponding increase in direct and indirect metabolites (filled in black), but less α-cleavage with a corresponding decrease in direct and indirect metabolites (filled in white). In addition, mAPP levels are reduced. The quantitative alterations in one or more of these APP metabolites can participate in BAD-Glu dysregulation caused by the Swedish mutations. (b) The Swedish mutations cause changes in the primary sequence of several APP metabolites, including APP (APPSw), sAPPβ (sAPPβSw), and sAPPα (sAPPαSw). The qualitative alterations in one or more of these APP metabolites can participate in BAD-Glu dysregulation caused by the Swedish mutations. (c) APP present in synaptic vesicles can interact with SV proteins and proteins regulating exocytosis via an intraluminal (ISVAID) and a cytosolic (JCasp) domain. Intraluminal and cytosolic interactions may have an opposite effect: the former tunes down glutamate release while the latter facilitates glutamate release. Cleavage of APP by BACE1 inside the ISVAID can abrogate the intravesicular interaction triggering the facilitator function of the cytosolic interaction. Increased β-cleavage of APPSw can dysregulate BAD-Glu dysregulation and facilitate glutamate release Cas9 mRNA, gRNA generated by in vitro transcription, and oligo donor were co-injected into zygotes for production of rats carrying these knock-in (KI) mutations by homology-directed repair. To verify CRISPR-induced mutation, the pups were genotyped by PCR, followed by sequence analysis. The rat App locus was amplified by PCR with the following specific forward (F) and reverse (R) primers: F-CTTTCTCCAGTCTGTTTGCTTGCG; R-GCCTGCTTCCGTGCTTCCTTT.

| Generation of rats carrying the App gene with the humanized Aβ sequence and rats with 7 bp deletion in App exon 16
The rat App gene (GenBank accession number: NM_019288.2; Ensembl: ENSRNOG00000006997) is located on rat chromosome 11. We created Long Evans rats with point mutation GGA>CGA, TTC>TAC, CGC>CAT at rat App locus by CRISPR/Cas-mediated genome editing. These mutations will create a rat that carries an App gene coding for rat APP with the humanized Aβ sequence.
The rat App gene comprises 18 exons, with the ATG start codon in exon 1 and TAA stop codon in exon 18; the GGA, TTC, and CGC codons are located in exon 16. Thus, exon 16 was selected as target site. gRNA targeting vector and oligo donor (with targeting sequence, flanked by 120-bp homologous sequences combined on both sides) were designed as follows.
Cas9 mRNA, sgRNA, and oligo donor are co-injected into zygotes, but homology-directed repair can occur even after few cell cycles. Thus, injected rats can have a mixture of correctly targeted alleles and alleles carrying aberrant mutations or no mutations. To identify rats carrying correctly targeted App alleles, the PCR products were cloned into TA vectors and 10 clones were sequenced using forward primer: 5-GTCAATGGTTTCAATCTAGGATG-3′. This analysis showed that RatID#120 had three types of alleles: If properly spliced, the δ7 App allele is predicted to produce a mRNA (App δ7 mRNA) coding for a truncated, soluble protein called APPδ7 (Figure 1a).

| Off-target analysis for gRNA1 and sRNA2
Homology-directed repair can cause off-target mutations in genetic sites that have high homology with the gRNAs. We identified potential off-target sites for gRNA1 and gRNA2. Based on this analysis, RatID#120 (F0-App h rat) has been analyzed for mutations in these most likely off-target mutation sites. Mismatched bases are in red.

| Off-target analysis of targeting sequence gRNA1: GTGAAGATGGATGCGGAGTTCGG
Three potential off-target sites have been identified (mismatched bases with the targeting sequence are in red). These sites have been amplified by PCR and sequenced.

| Off-target analysis of targeting sequence gRNA2: CGAAGTCCGCCATCAAAAACTGG
Two potential off-target sites have been identified (mismatched bases with the targeting sequence are in red). These sites have been amplified by PCR and sequenced.

| Generation of rats carrying the App gene with the humanized Aβ sequence and the FAD Swedish mutations
We created Long Evans rats with point mutations AAG>AAT, ATG>CTG, GGA>CGA, TTC>TAC, CGC>CAT at the App locus by CRISPR/Cas-mediated genome engineering. These mutations will create a rat that carries a humanized Aβ APP sequence plus the FAD Swedish mutation KM>NL. The AAG, ATG, GGA, TTC, and CGC codons are located in exon 16. gRNA targeting vector and oligo donor (with targeting sequence, flanked by 120-bp homologous sequences combined on both sides) were designed as follows.
Cas9 mRNA, gRNA generated by in vitro transcription and oligo donor were co-injected into zygotes for production of rats carrying these knock-in (KI) mutations by homology-directed repair. To verify CRISPR-induced mutation, the pups were genotyped by PCR, followed by sequence analysis. The rat App locus was amplified by PCR. PCR products were cloned into TA vectors and 10 plasmids containing App inserts were sequenced as described above for the humanizing mutations. This analysis showed that RatID#24 had three types of alleles: Thus, RatID#24 was identified as a positive chimeric founder (F0-App s rat).

| Off-target analysis of targeting sequence gRNA3: CTCAGAAGTGAAGATGGATGCGG
Three potential off-target sites have been identified (mismatched bases with the targeting sequence are in red.) These sites have been amplified by PCR and sequenced.

| Off-target analysis of targeting sequence gRNA4: CGAAGTCCGCCATCAAAAACTGG
Two potential off-target sites have been identified (mismatched bases with the targeting sequence are in red). These sites have been amplified by PCR and sequenced.
The δ7 mutant allele was a product of aberrant homology-directed repair, but is useful because this mutation will either produce a truncated soluble APPδ7 protein (sAPPδ7) or a hypomorphic allele. F1-App δ7/w , F1-App h/w , and F1-App s/w rats were crossed to WT Long Evans to generate F2-App δ7/w , F2-App h/w , and F2-App s/w rats. These crossing were repeated three more times to obtain F5-App δ7/w , F5-App h/w , and F5-App s/w rats. The probability that F5 rats carry unidentified off-target mutations (except those, if present, on Chr. 11) is ~1.5625%. Male and female F5-App δ7/w , F5-App h/w , and F5-App s/w rats were crossed to obtain App δ7/δ7 , App h/h , and App s/s rats.

| Immunohistochemistry (IHC) staining
IHC staining was performed in accordance with Biospective Standard Operating Procedure (SOP) # BSP-L-06. Slides were manually de-paraffinized and rehydrated prior to the automated immunohistochemistry. Slides initially underwent antigen retrieval, either heat-induced epitope-retrieval (HIER), or formic acid treatment. HIER was performed by incubation in citrate buffer (pH 6.0) and heating to 120°C under high pressure for a period of 10 min. Formic acid treatment was 15-min incubation in 80% formic acid, followed by washing in water and TBS-T. All IHC studies were performed at room temperature on a Lab Vision Autostainer using the REVEAL Polyvalent HRP-AEC detection system (Spring Bioscience). Antigen retrieval was performed as outlined in Table 1, followed by immunohistochemical staining. Briefly, slides were incubated sequentially with hydrogen peroxide for 5 minutes, to quench endogenous peroxidase, followed by 5 minutes in Protein Block, and then incubated with primary, antibodies as outlined in Table 1. Antibody binding was amplified using the Complement reagent (20 min), followed by a HRP-conjugate (20 min), and visualized using the AEC chromogen (20 minutes). All IHC sections were counterstained with Acid Blue 129 and mounted with aqueous mounting medium (Zehntner, Chakravarty, Bolovan, Chan & Bedell, 2008

| Modified Bielshowski silver staining
The slides were manually de-paraffinized and rehydrated prior to histological staining. Rehydrated tissue was immersed in preheated silver nitrate solution (40°C) for 15 min, followed by a deionized water rinse and an incubation in ammoniacal sliver solution at 40°C for 10 min (American Master Tech). Silver deposition was performed in the developer solution for a period of 15 min, and once a golden brown tissue stain was achieved, the development was stopped by sequential incubations in ammonium water then 5% sodium thiosulfate (American Master Tech). The stained tissue sections were dehydrated in xylene and mounted in Permount (VWR) and cover-slipped.

| Image analysis of IHC sections
The IHC and histology slides were digitized using an Axio Scan.
Z1 digital whole-slide scanner (Carl Zeiss). The images underwent quality control (QC) review and final images transferred to the Biospective server for qualitative image analysis. All qualitative assessments were performed blinded to the tissue genotype.

| Rt-pcr
Total brain RNA was extracted from P21 rat pups with RNeasy RNA Isolation kit (Qiagen 74104) and used to generate cDNA with a High-Capacity cDNA Reverse Transcription Kit (Thermo 4368814). 50 ng cDNA, TaqMan™ Fast Advanced Master Mix (Thermo 4444556), and the appropriate TaqMan (Thermo) probes were used in the real-time polymerase chain reaction. Samples were analyzed on a Roche LightCycler 2.0 Thermal Cycler, and relative RNA amounts were quantified using LinRegPCR software (hartfaalcentrum.nl). The probe Rn00570673_m1 (exon junctions 11-12, 12-13, and 13-14) was used to detect rat App, and samples were normalized to Gapdh levels, as detected with Rn01775763_g1 (exon junctions 2-3, and 7-8).

| Experimental design and statistical analysis
Levels of App mRNA (Figure 1b) were measured in 21-day-old App w/w , App δ7/δ7 , App h/h , and App s/s rats (two females and three males for each genotype). The Western blots shown in Figure 1e-g were performed using brain lysates obtained from 1 male and 1 female 21-day-old App w/w , App δ7/δ7 , App h/h , and App s/s rats. The Western blots shown in Figure 1h-k were performed using soluble brain fractions from one 21-day-old male App δ7/δ7 , App h/h , and App s/s rat. The Western blots shown in Figure 2 were performed using protein extracts obtained from 21-day-old App h/h and App s/s rats (2 females and 3 males for each genotype). The slices used for electrophysiology studies were obtained from 6-to 8-week-old rats. For the experiments shown in

DATA AVA I L A B I L I T Y S TAT E M E N T
The data that support the findings of this study are available from the corresponding author upon reasonable request.