Overexpression of Glutaminyl Cyclase, the Enzyme Responsible for Pyroglutamate Aβ Formation, Induces Behavioral Deficits, and Glutaminyl Cyclase Knock-out Rescues the Behavioral Phenotype in 5XFAD Mice*

Pyroglutamate-modified Aβ (AβpE3–42) peptides are gaining considerable attention as potential key players in the pathology of Alzheimer disease (AD) due to their abundance in AD brain, high aggregation propensity, stability, and cellular toxicity. Overexpressing AβpE3–42 induced a severe neuron loss and neurological phenotype in TBA2 mice. In vitro and in vivo experiments have recently proven that the enzyme glutaminyl cyclase (QC) catalyzes the formation of AβpE3–42. The aim of the present work was to analyze the role of QC in an AD mouse model with abundant AβpE3–42 formation. 5XFAD mice were crossed with transgenic mice expressing human QC (hQC) under the control of the Thy1 promoter. 5XFAD/hQC bigenic mice showed significant elevation in TBS, SDS, and formic acid-soluble AβpE3–42 peptides and aggregation in plaques. In 6-month-old 5XFAD/hQC mice, a significant motor and working memory impairment developed compared with 5XFAD. The contribution of endogenous QC was studied by generating 5XFAD/QC-KO mice (mouse QC knock-out). 5XFAD/QC-KO mice showed a significant rescue of the wild-type mice behavioral phenotype, demonstrating the important contribution of endogenous mouse QC and transgenic overexpressed QC. These data clearly demonstrate that QC is crucial for modulating AβpE3–42 levels in vivo and prove on a genetic base the concept that reduction of QC activity is a promising new therapeutic approach for AD.

neurofibrillary tangles. The discovery that certain early onset familial forms of AD may be caused by enhanced levels of A␤ peptides has led to the hypothesis that amyloidogenic A␤ is intimately involved in the pathogenic process (1).
Besides full-length A␤ 40 and 42 isoforms starting with an aspartate at position 1, a variety of different N-truncated A␤ peptides have been identified in AD brains. Ragged peptides including phenylalanine at position 4 of A␤ have been reported as early as 1985 by Masters et al. (2). In contrast, no N-terminal sequence could be obtained from cores purified in a SDS-containing buffer, which led to the assumption that the N terminus could be blocked (3,4).
The presence of A␤pE3 (N-terminally truncated A␤ starting with pyroglutamate) in AD brain was subsequently shown using mass spectrometry of purified A␤ peptides, explaining at least partially initial difficulties in sequencing A␤ peptides purified from human brain tissue (5). The authors reported that only 10 -15% of the total A␤ isolated by this method begins at position 3 with A␤pE3. Saido et al. (6) and others (7) subsequently showed that A␤pE3 represents a dominant fraction of A␤ peptides in AD brain.
Overexpression of A␤pE3-42 in neurons of TBA2 transgenic mice triggers neuron loss and an associated neurological phenotype (8). N-terminal pE formation can be catalyzed by glutaminyl cyclase (QC) and is pharmacologically inhibited by QC inhibitors, both in vitro (9) and in vivo (10). Moreover, QC expression was found up-regulated in the cortex of patients with AD and correlated with the appearance of pEmodified A␤. Oral application of a QC inhibitor resulted in reduced A␤pE3-42 burden in two different transgenic mouse models of AD as well as in a transgenic Drosophila model. Interestingly, treatment of these mice was accompanied by reductions in A␤x-40/42, diminished plaque formation and gliosis, as well as improved performance in context memory and spatial learning tests (10). Thus, A␤pE3-42 reduction is a promising target for therapy of AD. In the current work, the contribution of QC was studied for the first time using genetic means by human QC overexpression and endogenous QC-knock-out in an AD mouse model.

EXPERIMENTAL PROCEDURES
Transgenic and Knock-out Mice-5XFAD (11) mice have been described previously. All mice were backcrossed for more than 10 generations on a C57BL/6J genetic background and housed at a 12-h day/12-h night cycle with free access to food and water. For generation of hQC transgenic mice, an expression vector containing the cDNA of human QC under control of the murine Thy1 promoter sequence was constructed, applying standard molecular biology techniques and verified by sequencing. The transgenic founder was generated on C57BL/6J/CBA background by pronuclear injection (JSW, Graz, Austria). The resulting offspring were further characterized for transgene integration by PCR analysis and after crossing to C57BL/6J wild-type mice for transgene expression by RT-PCR (more than 10 generations). QC knock-out mice (QC-KO) were generated on the basis of a classical homologous recombination approach at Genoway, Lyon. The targeting vector contained the mouse chromosomal QC region ranging from intron 3 to exon 6. This region was modified by insertion of two LoxP sites in intron 3 and 5, respectively. In addition, a neomycin resistance cassette flanked by two flippase recognition targets was inserted immediately upstream of the LoxP in intron 5. After homologous recombination and chimera production, the neomycin selection cassette was removed by breeding with Flp-expressing mice followed by breeding of the pups with Cre-expressing mice for deletion of QC exons 4 and 5. The deletion of exons 4 and 5 causes a frameshift in the QC open reading frame generating a stop codon in exon 6. Successful manipulation was confirmed by PCR and Southern hybridization. Absence of murine QC in 5XFAD/QC-KO comparison with 5XFAD and 5XFAD/hQC was further confirmed by RT-PCR (supplemental Methods and supplemental Fig. 1). Animals were handled according to German guidelines for animal care and studies were approved by the local legal authorities (LAVES). Only female mice were used.
Immunohistochemistry-Mouse tissue was processed as described previously (12). In brief, 4-m paraffin sections were pretreated with 0.3% H 2 O 2 in PBS to block endogenous peroxidases, and antigen retrieval was achieved by boiling sections in 0.01 M citrate buffer, pH 6.0, followed by a 3-min incubation in 88% formic acid. Primary antibodies were incubated overnight, followed by incubation with biotinylated secondary antibodies (DAKO) before staining was visualized using the ABC method with Vectastain kit (Vector Laboratories) and diaminobenzidine as chromogen. Alternatively, fluorochromated secondary antibodies (anti-mouse Alexa Fluor 594 and anti-rabbit Alexa Fluor 488; Invitrogen) were used for immunofluorescence detection.
ELISA of A␤ Levels in Brain-Frozen brains (n ϭ 4 -8 per group) were weighed and subsequently subjected to a sequential A␤ extraction. In a first step, brains were homogenized in TBS (120 mM NaCl, 50 mM Tris, pH 8.0, containing complete protease inhibitor (Roche Applied Science)) using a Dounce homogenizer, sonified, and subsequently centrifuged at 27,000 ϫ g for 20 min at 4°C. The supernatant was removed and stored at Ϫ80°C. The pellet was dissolved in 2.5 ml of 2% SDS, sonificated, and subsequently centrifuged at 80,000 ϫ g for 1 h at 4°C. Supernatants were directly frozen at Ϫ80°C. The resulting pellets were again resuspended in 0.5 ml of 70% formic acid, sonified, and neutralized using 1 M Tris. Aliquots of the neutralized formic acid fraction were directly frozen at Ϫ80°C. SDS lysates were diluted at least 10-fold for determination of A␤x-42 and A␤pE3 using ELISA. All dilutions were carried out using EIA buffer (IBL Co.). The neutralized formic acid fraction and the TBS fraction were applied directly or after dilution using EIA buffer. ELISA measurements were performed in triplicate and according to the protocol of the manufacturer (IBL Co.; catalog nos. JP27716 and JP27711). Samples were run in triplicate.
Quantification of Plaque Load-Extracellular A␤ load was evaluated in mouse brain using an Olympus BX-51 microscope equipped with an Olympus DP-50 camera and the ImageJ software (V1.41; National Institutes of Health). Serial images of 40ϫ magnification (hippocampus) and 100ϫ (cortex) were captured on six sections/animal (n ϭ 5/group), which were at least 30 m afar from each other. Using ImageJ the pictures were binarized to 16-bit black and white images, and a fixed intensity threshold was applied defining the DAB staining.
Behavioral Testing-Spontaneous alternation rates were assessed using Y-and cross-maze as described previously (11,14). The alternation percentage was calculated as the percentage of actual alternations to the total number of arm entries. Balance and general motor function were assessed using the balance beam task. A 1-cm dowel beam is attached to two support columns 44 cm above a padded surface. At either end of the 50-cm long beam, a 9 ϫ 15-cm escape platform is attached. The animal is placed on the center of the beam and released. Each animal is given three trials during a single day of testing. The time the animal remained on the beam is recorded and the resulting latencies to fall of all three trials are averaged. If an animal remains on the beam for whole 60-s trial or escapes to one of the platforms, the maximum time of 60 s is recorded (14). For the string suspension test the animals are permitted to grasp the string by their forepaws and are released. A rating system from 0 to 5 is used during the single 60-s trial to assess each animal's performance in this task: 0 ϭ unable to remain on the string; 1 ϭ hangs only by fore-or hindpaws; 2 ϭ as for 1, but attempts to climb onto string; 3 ϭ sits on string and is able to hold balance; 4 ϭ four paws and tail around string with lateral movement; 5 ϭ escape. The following numbers of animals were analyzed in this task (14). The following number of female mice was used at the age of 6 months: 5XFAD, 11; 5XFAD/hQC, 8; hQC, 11; 5XFAD/QC-KO, 4; QC-KO, 6; wild type, 12.
Statistical Analysis-Statistical differences were evaluated using one-way ANOVA followed by Bonferroni post hoc test or unpaired t test as indicated. All data are given as means Ϯ S.E. All statistics were calculated using GraphPad Prism version 5.00 software.

RESULTS
Expression and Distribution of hQC in the Brain of hQC and 5XFAD/hQC Mice-To study human QC overexpression in hQC transgenic mice (hQC and 5XFAD/hQC), the rabbit polyclonal antiserum 1301 recognizing human QC was used to detect the transgene hQC expression in different brain regions (Fig. 1). Because the expression of the hQC trangene is driven by the Thy1 promoter, abundant pyramidal neurons expressing hQC were detected in various brain regions of hQC and 5XFAD/hQC mice including the cortex (Fig. 1, A-D), the hippocampus (Fig. 1, E and F), the midbrain and the cerebellum (data not shown). Notably, a massive staining of hQC was observed in plaque-associated dystrophic neurites in 5XFAD/hQC mice (Fig.  1D). Abundant hQC immunoreactivity was detected in mossy fibers of hQC and 5XFAD/hQC transgenic mice (Fig. 1, E and F).

Co-localization of APP and QC in Neurons and the Neuritic
Component of Plaques-Double immunofluorescence demonstrated co-localization of hQC and APP in the same cellular compartments in the brain of 5XFAD/hQC mice (Fig. 2). APP markedly labels dystrophic neurites around plaques and shows abundant co-localization with hQC suggesting that hQC is axonally transported like APP (Fig. 2).
Effect of hQC Overexpression and QC Knock-out on Behavioral Performance in 5XFAD Mice-Motor coordination was assessed by using balance beam and string suspension tasks (Fig. 5, A and B). In both tasks the 5XFAD/hQC performed significantly worse that 5XFAD (p Ͻ 0.001 and p Ͻ 0.01, respectively). Working memory was assessed using Y-and cross-maze alternation tasks. Analysis in the Y-maze revealed  a significantly reduced alternation frequency in 5XFAD/hQC compared with 5XFAD mice (p Ͻ 0.05). The number of arm entries during the test period was not different between the groups (Fig. 5, C and D). Assessment using the more complex cross-maze task demonstrated again a significantly reduced alternation frequency in 5XFAD/hQC compared with 5XFAD mice (p Ͻ 0.05), and 5XFAD versus wild-type mice (p Ͻ 0.05). The latter finding corroborated previous results (14). Moreover, the working memory deficit of 5XFAD mice was rescued in 5XFAD/QC-KO mice (p Ͻ 0.05) showing alternation frequencies indistinguishable from wild-type mice. The number of arm entries during the test period was not different among all groups (Fig. 5, E and F). Schilling et al. have shown that cyclization of glutamate at position 3 of A␤ can be driven enzymatically by QC in vitro (15). In addition, it has been demonstrated that QC inhibition significantly reduced A␤pE3 formation in vivo, emphasizing the importance of QC activity during cellular maturation of pyroglutamate-containing peptides. The pharmacological inhibition of QC activity by the QC inhibitor PQ150, which significantly reduced the level of A␤pE3 in vitro (16) and in vivo (10), suggests that QC inhibition might serve as a new therapeutic approach. Furthermore, the mean level of A␤pE3-IgM autoantibodies was significantly decreased in AD patients compared with healthy controls. In the group of mildly cognitive-impaired patients there was a significant positive correlation between A␤pE3-IgM and cognitive decline (17).

DISCUSSION
Interestingly, APP/PS1KI mice, a model with severe neuron loss in the hippocampus, accumulate a large heterogeneity of N-truncated A␤x-42 isoforms including A␤pE3 peptides coinciding with the onset of behavioral deficits (18,19). More specifically, transgenic mice expressing only A␤pE3-42 developed a robust and lethal neurological phenotype accompanied by Purkinje cell loss (TBA2 mouse line) (8).
Saido et al. suggested that hypothetically the removal of N-terminal amino acids 1 and 2 of A␤ might be carried out by amino or dipeptidyl peptidase(s) (6). Aminopeptidase A may be responsible in part for the N-terminal truncation of fulllength A␤ peptides (20).
APP transgenic mouse models have been reported to show no (23) or low A␤pE3 levels (31). Maeda et al. have demonstrated that the localization and abundance of [ 11 C]Pittsburgh compound B autoradiographic signals were closely associated with those of N-terminally truncated and modified A␤pE3 deposition in AD and different APP transgenic mouse brains, implying that the detectability of amyloid by [ 11 C]Pittsburgh compound B-positron emission tomography is dependent on the accumulation of specific A␤ subtypes (39). APP/PS1KI (12,18) and 5XFAD (14) mice harbor abundant A␤3pE levels. Interestingly, both models develop an age-dependent neuron loss and robust behavioral deficits, like TBA2 mice with only A␤pE3-42 expression (8).
The findings of the present work are in good agreement with the previous observations that A␤pE3-42 levels correlate with behavioral deficits in transgenic mouse models. Here, we demonstrate for the first time genetic evidence for QC as a major target for AD. Overexpression of human QC is co-localized with APP in the neuritic component of plaques, leading to elevated A␤pE3-42 levels as detected by ELISA. This finding is corroborated by an increase in the overall plaque pathology including A␤pE3-42 in 5XFAD/hQC mice. Consistently, 5XFAD/hQC mice developed a neurological phenotype demonstrated by learning and memory impairments compared with the 5XFAD mouse model at 6 months FIGURE 4. Effect of hQC overexpression and QC knock-out on A␤ levels in 5XFAD mice. Quantification of A␤x-42 and A␤pE3-42 using ELISA showed significant changes in TBS, SDS, and formic acid (FA) fractions in 5XFAD, 5XFAD/hQC, and 5XFAD/QC-KO mouse brain. SDS and FA fractions were pooled for quantification. A␤x-42 levels were significantly reduced in the SDSϩFA fraction of 5XFAD/QC-KO mice. A␤pE3-42 levels were significantly elevated in all fractions in 5XFAD/hQC mice. Although in the TBS fraction of 5XFAD/QC-KO mice the levels of A␤pE3-42 were below the limit of quantitation, in the SDSϩFA fractions of 5XFAD/QC-KO mice the levels of A␤pE3-42 were significantly reduced. **, p Ͻ 0.01; ***, p Ͻ 0.001. LOQ, limit of quantitation. Error bars, S.E. of age. In addition, we could also show that knock-out of endogenous QC is sufficient to lower A␤ levels, including A␤pE3-42, leading to a concomitant rescue of behavioral deficits in 5XFAD mice. The apparent discrepancies between A␤x-42 and A␤pE3-42 are likely because the A␤x-42 peptides are ϳ1000 times more abundant than A␤pE3-42 (micrograms versus nanograms/g wet weight). Therefore it is unlikely that an hQC-dependent increase in A␤pE3-42 levels is reflected in a concomitant increase of A␤x-42 levels in a stoichiometric manner. It is, however, surprising that the level of A␤x-42 is significantly reduced in 5XFAD/QC-KO mice. This might be due to a reduced seeding effect of A␤pE3-42 on full-length A␤, which is therapeutically of interest. It is hypothesized that A␤pE3-42 elevation does not necessarily lead to increased aggregation of A␤x-42, which might be due to a saturation effect. In addition, the differences in the plaque load of total A␤ and A␤x-42 levels measured by ELISA might be because plaque load was done in the cortex whereas ELISA was performed in whole brain lysates.
Because A␤pE3-42 levels were not completely reduced, we assume that other QC-related enzymes like isoQC are responsible for the residual formation of pyroglutamate in QC-KO FIGURE 5. Effect of hQC overexpression and QC knock-out on behavioral performance in 5XFAD mice. A and B, 5XFAD/hQC mice showed a significantly reduced motor performance in balance beam (A) and string suspension task (B) compared with 5XFAD mice. C, in addition, working memory deficits were detected in 5XFAD/hQC compared with 5XFAD mice using Y-and cross-maze (E). Interestingly, the 5XFAD/QC-KO mice showed a rescue of working memory deficits with alternation frequencies indistinguishable from wild-type mice. D and F, the number of arm entries in Y-and cross-maze did not differ among the groups. *, p Ͻ 0.05; **, p Ͻ 0.01; ***, p Ͻ 0.001. mice. QC and isoQC represent very similar proteins, which are both present in the secretory pathway of cells. The functions of QCs and isoQC complement each other, suggesting a pivotal role of pyroglutamate modification for protein and peptide maturation (40). To analyze a possible contribution of isoQC to the remaining QC-like activity in 5XFAD/QC-KO mice, we performed a Western blot analysis using an isoQC antibody. The protein levels of isoQC were unchanged in different brain regions between WT and QC-KO mice. This observation demonstrates that the finding of residual pyroglutamate A␤ levels in 5XFAD/QC-KO mice is likely mediated by isoQC (supplemental Fig. 3). In conclusion, reduction of QC was sufficient to rescue the behavioral impairments in the 5XFAD mouse model suggesting a crucial role of QC as a therapeutic target for AD.