Rare CASP6N73T variant associated with hippocampal volume exhibits decreased proteolytic activity, synaptic transmission defect, and neurodegeneration

Caspase-6 (Casp6) is implicated in Alzheimer disease (AD) cognitive impairment and pathology. Hippocampal atrophy is associated with cognitive impairment in AD. Here, a rare functional exonic missense CASP6 single nucleotide polymorphism (SNP), causing the substitution of asparagine with threonine at amino acid 73 in Casp6 (Casp6N73T), was associated with hippocampal subfield CA1 volume preservation. Compared to wild type Casp6 (Casp6WT), recombinant Casp6N73T altered Casp6 proteolysis of natural substrates Lamin A/C and α-Tubulin, but did not alter cleavage of the Ac-VEID-AFC Casp6 peptide substrate. Casp6N73T-transfected HEK293T cells showed elevated Casp6 mRNA levels similar to Casp6WT-transfected cells, but, in contrast to Casp6WT, did not accumulate active Casp6 subunits nor show increased Casp6 enzymatic activity. Electrophysiological and morphological assessments showed that Casp6N73T recombinant protein caused less neurofunctional damage and neurodegeneration in hippocampal CA1 pyramidal neurons than Casp6WT. Lastly, CASP6 mRNA levels were increased in several AD brain regions confirming the implication of Casp6 in AD. These studies suggest that the rare Casp6N73T variant may protect against hippocampal atrophy due to its altered catalysis of natural protein substrates and intracellular instability thus leading to less Casp6-mediated damage to neuronal structure and function.

Recombinant Caspase-6 N73T (Casp6N73T) zymogen showed similar self-processing as Caspase-6 wild type (Casp6WT) zymogen. To observe the self-processing of Caspase-6 N73T (Casp6N73T) zymogen, recombinant protein from three independent clones expressing Casp6N73T zymogen were compared with Caspase-6 wild type (Casp6WT) zymogen and catalytically inactive triple mutant Casp6D (23,179,193) A. The Casp6WT zymogen undergoes self-processing at D23, D179, and D193 to remove the pro-domain and the inter-subunit linker, which allows the formation of active Casp6 consisting of two large subunits (LS) and two small subunits (SS) (Fig. 2a). Purified recombinant Casp6N73T and Casp6WT zymogens both generated the expected small subunit (SS) and large subunit with (LSL) or without (LS) inter-subunit linker. Casp6WT Table 1. Demographic and clinical characteristics of cognitively normal (CN) and Alzheimer disease (AD) participants for different brain region analysed. TCX temporal cortex, PHG para-hippocampal gyrus, IFG inferior frontal gyrus, STG superior temporal gyrus, FP frontal pole, DLPFC dorsolateral prefrontal cortex, CER cerebellum, RIN RNA integrity number.  AD  CN  AD  CN  AD  CN  AD  CN  AD  CN  AD  CN  AD   Number  71  80  16  78  18  102  21  98  22  111  86  155  72  79 Sex (F/M) 35 Figure 1. Differential expression of CASP6 and CASP1 in Alzheimer disease (AD) and cognitively normal (CN) brains. The logarithm of read counts per million total reads (logCPM) values for CASP6 mRNA (a-g) and CASP1 mRNA (h-n) generated from RNA-Seq data for AD and CN temporal cortex (TCX; a,h), superior temporal gyrus (STG; b,i), para-hippocampal gyrus (PHG; c,j), dorsolateral prefrontal cortex (DLPFC; d,k), cerebellum (CER; e,l), frontal pole (FP; f,m) and inferior frontal gyrus (IFG; g,n). Each dot represents data from one individual and the horizontal bar denotes the mean. Statistical evaluations were done with the R package limma between AD and CN. *P < 0.05, **p < 0.01, ***P < 0.001, ****p < 0.0001. www.nature.com/scientificreports/ and Casp6N73T zymogens generated more LSL than LS (Fig. 2b) indicating preferred self-cleavage at D193, as previously observed with Casp6WT 14,33,46 . The catalytically inactive Casp6D (23,179,193)A was not processed, as expected, and migrated as full-length (FL) Casp6 at 34 kDa. The Casp6N73T zymogen generated LSL, LS, and SS at levels comparable to those of the Casp6WT zymogen (Fig. 2b,c). FL, LSL, LS, and SS of Casp6 were confirmed with western blotting against anti-FL and anti-LSL Casp6 antiserum (Fig. 2d,e), neoepitope antiserum recognizing Casp6 cleaved at D179 (Fig. 2f,g), and anti-SS antibodies (Fig. 2h,i). No difference in the levels of each subunit between Casp6WT and Casp6N73T was detected (Fig. 2e,g,i). These results indicate that Casp6N73T zymogen self-processing is similar to that of Casp6WT zymogen.   Fig. S1a-f). Two to four hundred nM active Casp6N73T exhibited comparable VEIDase activity (defined as cleaved pmol AFC per minute) than similar concentrations of active Casp6WT (Fig. 3a). The velocities of product formation by 2, 10, and 50 nM Casp6N73T with 20 µM Ac-VEID-AFC substrate were similar to those of 2, 10, or 50 nM Casp6WT (Fig. 3b). Furthermore, to better understand the kinetics of Casp6N73T, the K m , V max , k cat and k cat /K m were determined based on Michaelis-Menten kinetics measurements ( Supplementary Fig. S2). No differences were found in the K m , V max , k cat or k cat /K m between Casp6N73T and Casp6WT (Fig. 3c).

Recombinant
Casp6N73T showed increased proteolytic processing of natural Casp6 protein substrate, Lamin A/C. To assess if Casp6N73T affects Casp6 activity on natural protein substrates, the cleavage of Lamin A/C by Casp6N73T and Casp6WT was assessed. Lamin A/C was extracted from Casp6 KO mouse tissue, and was incubated with varying concentrations of active site titrated Casp6N73T or Casp6WT. Both Casp6N73T and Casp6WT processed Lamin A/C and generated the expected 28 kDa fragment (Fig. 4a) 47 . The % FL Lamin A/C and cleaved Lamin A/C generated by Casp6N73T at concentrations from 50 to 500 nM were similar to that of Casp6 WT after 1 h incubation (Fig. 4b). In order to calculate the initial reaction velocity, cleaved Lamin A/C was measured every 5 min for the first 15 min of reaction, followed by every 15 min until 60 min. The amount of cleaved Lamin A/C increased with incubation time with both Casp6N73T and Casp6WT (Fig. 4c). Initially, at the 5 and 10 min time points, 200-500 nM Casp6N73T consistently generated slightly higher amounts of cleaved Lamin A/C than Casp6WT (Fig. 4d). The slope of the linear phase of the curve in Fig. 4d determined the initial velocity of the enzymes on Lamin A/C. The initial velocity of Casp6N73T was significantly higher than that of Casp6WT (Fig. 4e). These results suggest that the N73T substitution may modify Casp6N73T interaction with Lamin A/C.

Recombinant
Casp6N73T showed decreased catalytic efficiency on α-Tubulin. To further understand the effect of Casp6N73T on different protein substrates, the catalytic activity of Casp6N73T and Casp6WT on α-Tubulin was compared. Both Casp6N73T and Casp6WT processed α-Tubulin and generated a fragment 2 kDa smaller than the FL (Fig. 5a top panel), consistent with Casp6 cleavage at VGVD438 in α -Tubulin. Cleavage of α-tubulin at D438 was confirmed with the neoepitope antibody GN20622 (Fig. 5a bottom panel). The % Tub∆Casp6 by Casp6N73T at concentration 15.6-250 nM was significantly lower than that of Casp6WT after 4 h incubation (Fig. 5b). Kinetically, 15.6-62.5 nM Casp6N73T consistently generated less Tub∆Casp6 compared to Casp6WT during the first 4 h of incubation (Fig. 5c&d). Analysis of the initial velocity determined by the slope of the linear phase of the cleavage curve in Fig. 5d, showed a significant 50% reduction  www.nature.com/scientificreports/ of Casp6N73T catalytic efficiency (Fig. 5e). These results show that Casp6N73T catalytic efficiency on α-Tubulin is lower than that of Casp6WT. As observed with the Lamin A/C substrate, the results suggest that Casp6N73T may interact differently than Casp6WT with α-Tubulin.

The steady state level of active Casp6N73T subunits is considerably lower than those of Casp6WT in transfected human embryonic kidney 293 T (HEK293T) cells.
To assess if eukaryotically expressed Casp6N73T behaves like the prokaryotically expressed recombinant Casp6N73T, CASP6WT and CASP6N73T cDNAs lacking their pro-domain (∆Pro) to promote self-processing of Casp6 were cloned in pCep4β and the constructs were transfected in human embryonic kidney 293 T (HEK293T) cells. Catalytically inactive, CASP6C163A with its pro-domain sequence, was transfected as a control for full length Casp6. Casp6 mRNA levels were equivalent in CASP6WT, CASP6N73T, or CASP6C163A cDNA transfected cells (Fig. 6a,b). In transfected cells, the ∆ProCasp6 expressed in CASP6WT-and CASP6N73T-transfected cells migrated approximately 2 kDa below the full length Casp6 in CASP6C163A-transfected cells (Fig. 6c). Unexpectedly, CASP6N73Ttransfected HEK293T cells contained consistently lower levels of ∆Pro-Casp6 than CASP6WT-transfected cells (Fig. 6c,d). Compared to CASP6C163A-transfected cells, ∆ProCasp6 levels were lower in CASP6WT-transfected cells, as expected since Casp6WT can be processed into its subunits (Fig. 6d). The levels of LS in CASP6N73Ttransfected cells were significantly lower than those of CASP6WT-transfected cells (Fig. 6e,f). Consistently, CASP6N73T-transfected HEK293T protein extracts did not show significant Casp6 VEIDase activity compared to non-, mock-, or empty vector-transfected cell extracts, while CASP6WT-transfected cells exhibited high VEI-Dase activity (Fig. 6g). Because transfected cells contained similar Casp6WT and Casp6N73T mRNA levels, decreased Casp6N73T full length protein level and VEIDase activity may be the result by enhanced turnover. Indeed, in the presence of translational inhibitor cycloheximide (CHD), the degradation rate of ∆ProCasp6N73T www.nature.com/scientificreports/ was faster than that of ∆ProCasp6WT levels (Fig. 6h,i), suggesting a higher cellular turnover of Casp6 due to the N73T substitution. Since Casp6 LS has previously been reported to be degraded by the proteasome 13,48 , we verified if proteasomal activity may be responsible for FL or LS Casp6N73T lower levels. The epoxomicin proteasomal inhibitor did not significantly alter levels of either Casp6N73T FL or LS in HEK293T cells, excluding proteasomal degradation as an explanation ( Supplementary Fig. S3). Nevertheless, these results indicate that Casp6N73T is unstable and degraded by an alternate non-proteasomal dependent mechanism, thereby limiting the amount of Casp6N73T LS and Casp6N73T activity produced in mammalian cells.
Recombinant Casp6N73T caused less neurofunctional damage and neuronal degeneration than Casp6WT in hippocampal CA1 pyramidal neurons. The effect of Casp6N73T on neuronal function was compared with recombinant Casp6WT and catalytically inactive Casp6C163A proteins by patching CA1 pyramidal neurons in hippocampal organ slices with these proteins (Fig. 7a) and measuring the amplitude of excitatory postsynaptic potential (EPSP) (Fig. 7b). EPSP amplitudes from recorded neurons were analyzed when the membrane potential and input resistance were stable (Fig. 7b bottom panel). Activated Casp6WT and Casp6N73T were stable for 5 h after their preparation in internal solution ( Supplementary Fig. S4), within which the recording data was acquired. Ten pg Casp6WT induced a decrease of EPSP amplitude 30 min after patching, and a continually decreasing EPSP amplitude during the 50 min of recording (Fig. 7c). In contrast 10 pg catalytically inactive Casp6C163A did not alter EPSP amplitude within 50 min. Interestingly, 10 pg Casp6N73T decreased the EPSP amplitude less than Casp6WT in pyramidal neurons for approximately 10 min after patching, and EPSP amplitude was maintained at 65% of the original levels for 50 min (Fig. 7c). Paired pulse ratio (∆PPR) of the first and second EPSP wave increased in Casp6WT patched neurons compared to Casp6C163A, suggesting that the decaying EPSP amplitude was at least in part due to a reduced probability of release presynaptically 49 .
In contrast, Casp6N73T ∆PPR was indistinguishable from that of Casp6C163A (Fig. 7d), possibly hinting at an absence of perceivable effects on presynaptic release. Although the change in PPR due to Casp6WT suggested a presynaptic locus of action, the coefficient of variation (CV) analysis of the first EPSP 50 indicated a predominantly postsynaptic effect, although with variable outcome (Fig. 7e). Taken together, the combined outcome of PPR and CV analyses is consistent with a coordinated downregulation on both pre-and post-synaptic sides due to Casp6WT.
The morphology of Casp6WT, Casp6N73T, or Casp6C163A patched CA1 pyramidal neurons was analyzed by manual 3D reconstruction (Fig. 7f). Dendritic map density and convex hulls indicated the average distribution and the maximum extent of dendrites, respectively (Fig. 7g). Neurons patched with Casp6WT exhibited a lower length of apical dendrites in the SLM region compared to Casp6C163A-patched neurons, but the data did not reach statistical significance (Fig. 7h). The apical dendrites length of Casp6N73T-patched neurons was longer than those of Casp6WT-patched neurons. In addition, Casp6WT-patched neurons displayed beading along basal neurites in the stratum oriens region indicative of neuronal degeneration (Fig. 7i), while beading was rarely seen in neurons patched with Casp6C163A, indicating that this beading is linked to Casp6 activity. The beading dendrites of Casp6N73T-patched neurons was approximately 50% lower than in neurons patched with Casp6WT. These results indicate that Casp6N73T is less detrimental to neuron function and causes less neurodegeneration than Casp6WT.

Discussion
Our study demonstrates that a rare human CASP6 variant encoding Casp6N73T, genetically associated with preserved hippocampal CA1 volume in an AD cohort, exhibits altered catalyses on natural protein substrates, lamin A and α-Tubulin. In addition, our study showed that Casp6N73T has less negative impact on neuronal function and neurodegeneration than the Casp6WT in mice CA1 hippocampal pyramidal neurons. These results provide initial evidence for a neuroprotective effect by Casp6N73T as reflected by hippocampal subfield volume preservation but this needs to be validated in independent data sets.
The significant association between Casp6N73T and hippocampal CA1 volume by SKAT-O provides an advantage in identifying associations between low frequency functional exonic SNPs and pathological features of AD compared to genome-wide association studies (GWAS) 45,51 . Different from the single regression model used in GWASs, SKAT uses a multiple regression model to regress the phenotypes of different SNPs in the same genomic region and allows different orientations of the phenotype, in order to assess the cumulative effects of one gene on disease. Another advantage of our approach is the use of a quantitative measure of hippocampal CA1 volume, instead of the numerous variables of clinically diagnosed AD. Hippocampal CA1 volume was chosen because atrophy in this region occurs early in AD progression 52,53 . Large scale longitudinal measures indicate that hippocampal CA1 atrophy is a robust MRI biomarker to distinguish MCI that further develop to AD from stable MCI 54 . Furthermore, rare SNPs also mean lower number of individuals carrying the SNP, therefore, our finding will require testing additional cases to confirm the role of Casp6N73T in MCI and AD. On the other hand, the ADNI study provides precise cognitive and neuroimaging information on the individual carrying CASP6N73T. The Casp6N73T variant was identified in a heterozygous individual at the early MCI stage with unexpected preserved hippocampal CA1 volume compared to those carrying the wild type CASP6.
Here, Casp6N73T was shown to be less damaging to synaptic transmission and plasticity than Casp6WT. In contrast to catalytically inactive Casp6C163A, Casp6WT rapidly decreased induced EPSP amplitude in CA1 hippocampal neurons, whereas Casp6N73T showed only a minor decrease. In addition, the paired pulse ratio was increased in Casp6WT, but not in Casp6N73T. Furthermore, Casp6WT, but not Casp6C163A, induced basal dendritic beading indicating Casp6 activity-mediated neurodegeneration CA1 neurons, a feature that was reduced with Casp6N73T. Based on the morphological analysis of reconstructed neurons, Casp6WT, but not Casp6N73T, caused a decrease in the total apical dendritic length. These data are consistent with transgenic www.nature.com/scientificreports/ expression of a self-activated form of Casp6WT in CA1 neurons causing age-dependent cognitive impairment 11 , small amounts of active Casp6 causing a rapid depression of CA1 neuronal transmission 22 and impairing longterm potentiation in hippocampal CA1 circuits in vivo 55 . Given that the presence of active Casp6 detected in the entorhinal cortex of aged individuals correlates with decreased cognitive performance 1,7,12,42 and that early cognitive decline is strongly correlated with synaptic loss 56,57 , these data raise the possibility that the active Casp6 observed in neuropil threads, neuritic plaques and neurofibrillary tangles 4,5 causes synaptic transmission problems. Most importantly, the data suggest that Casp6N73T variant would not be as detrimental as Casp6WT for neuronal structure and function and this may protect aged individuals from cognitive decline. In support of this conclusion, Casp6 deficient mice have been shown to be protected against excitotoxicity, nerve growth factor deprivation and myelin-induced axonal degeneration, and showed an age-dependent increase in cortical and striatal volume 58 . The molecular reason for the protective action of Casp6N73T might stem from its altered proteolysis of natural protein substrates compared to Casp6WT. Casp6 can cleave numerous protein substrates, some of which are specific to neurons and synapses 2,59 and involved in neurodegenerative diseases, such as Tau, Vimentin, Drebrin, Spinophilin, α-Actinin 2,5 , valosin containing protein p97 3 , APP 60 , Presenilin 1 and 2 28 , and Huntingtin 61 , that may be responsible for structural damage, synaptic loss, and protein aggregation. Here, we show that compared to Casp6WT, Casp6N73T had increased proteolytic processing of Lamin A/C, decreased proteolysis of α-tubulin and equivalent processing of the Ac-VEID-AFC peptide substrate. These results indicate that either the N73T alters the structure of the Casp6 enzyme thereby affecting the catalytic site, or the N73T alters the interaction of Casp6 with substrates. The altered catalysis of natural protein substrates α-Tubulin and Lamin A/C, but not of the small peptide substrate, suggests that the active site of the Casp6N73T is relatively normal. Therefore, it is likely that Casp6N73T interacts differently with α-Tubulin and Lamin A/C than Casp6WT, thus suggesting the presence of an exosite either within N73, its surroundings, or in a structure that is altered by N73T. Exosites are defined as sites outside the substrate-binding catalytic site which influence structure or function of an enzyme. The N73T substitution is located in the middle of Casp6's helix B with a shorter side chain pointing outward. The nearby negatively charged D72 is conserved among Casp6 orthologues and unique to Casp6 among the caspase protein family, which may also contribute to this exosite. Casp6 exosite 42RRR44, located at the hinge between the core structure and the N-terminus of the large subunit, has been confirmed 62,63 . Furthermore, two rare human variants, G66R and R65W completely eliminate or significantly reduce Casp6 activity through impaired substrate binding, alter the catalytic site activity, and have dominant negative effects on Casp6 WT 46 . The alternatively spliced Casp6b isoform lacking amino acids 13-104 while retaining the catalytic site also has no activity, and inhibits Casp6WT activation 64 . The N73T data presented here further highlights the importance of the N-terminus of the large subunit of Casp6 in substrate recognition. Casp6N73T may change the proteolysis of different neuronal substrates in addition to α-Tubulin, which together display a reduced damage due to Casp6 and favors a neuroprotective phenotype. Future investigation in characterizing the cleavage efficiencies of Casp6N73T on these substrates could expand our understanding of the new exosite as well as the protective mechanism of Casp6N73T. The identification of an additional exosite for Casp6 could be useful in specific drug design against Casp6 activity, which has been shown to cause age-dependent cognitive impairment in mice 11 .
The protective action of Casp6N73T might also stem from its instability in mammalian cells. Casp6N73T full length levels were decreased significantly relative to Casp6WT, despite similar mRNA levels for Casp6N73T and Casp6WT. This lower level of full length Casp6N73T could be explained by either a lower mRNA translational rate, increased protein processing, or increased degradation by alternate cellular proteolytic activities. A more rapid turnover for full length Casp6N73T compared to FL Casp6WT was observed indicating either increased protein processing into its active subunits or increased degradation by alternate cellular proteolytic activities. The fact that the level of Casp6N73T processed large subunit (LS) was reduced almost tenfold relative to Casp6WT indicated that Casp6N73T was processed less efficiently than Casp6WT or that the Casp6N73T LS was degraded more rapidly than the Casp6WT LS. Proteasomal activity was excluded since proteasomal inhibition did not significantly alter either FL or LS Casp6N73T levels. The fact that self-processing of prokaryotically expressed recombinant Casp6N73T is equivalent to that of Casp6WT eliminates the possibility that the Casp6N73T mutation alters processing. Therefore, the most likely explanation for these findings is that the FL Casp6N73T is unstable and degraded by a non-proteasomal cellular mechanism, thereby limiting the amount of Casp6N73T LS and Casp6N73T activity produced in mammalian cells.
Lastly, we confirm here that CASP1 and CASP6 mRNA levels are significantly increased in specific regions of the AD brains in the ADNI cohort compared to cognitively normal control brains from large sample sizes in the AMP-AD Consortium. We investigated CASP1 expression because it activates Casp6 in human primary neurons 42 . Increased CASP1 mRNA levels have been reported previously in AD cortex and entorhinal cortex (ERC) 65,66 . Our results additionally show increased CASP1 mRNA levels in AD temporal cortex and superior temporal gyrus. CASP6 expression in AD is more controversial. One early study reports low level of CASP6 mRNAs in AD brains 65 . Others indicate increased CASP6 mRNA levels in cortex and cerebellum of AD brains 43,66 . Here, we show increased CASP6 mRNA levels in AD temporal cortex, superior temporal gyrus, para-hippocampal gyrus, and dorsolateral prefrontal cortex. These results support the implication of CASP6 in AD.
This study highlights the importance of assessing the role of the amazing genetic diversity of humans in disease by combining human genetic information associated with well-ascertained neuroimaging and cognitive measures with biochemical and electrophysiological approaches to investigate the potential influence of rare variants on AD-related pathologies and cognition.

Methods
Alzheimer's Disease Neuroimaging Initiative (ADNI). Individuals used in the analysis were ADNI participants o 67,68 . Inclusion and exclusion criteria, clinical and neuroimaging protocols, and other information about ADNI can be found at www. adni-info. org and http:// www. loni. usc. edu/ ADNI/. Written informed consent was obtained at the time of enrollment for imaging and genetic sample collection and protocols of consent forms were approved by each participating sites' Institutional Review Board. Human subject ethical approval was obtained by ADNI and can be found at http:// www. loni. usc. edu/ ADNI/. All methods and experiments were performed in accordance with relevant guidelines.
Whole genome sequencing (WGS) analysis. WGS on the Illumina HiSeq2000 platform with pairedend reads was performed on blood-derived genomic DNA samples obtained from 817 ADNI participants 69 . Briefly, short-read sequences were mapped to the human genome assembly (GRCh build 37.72) using BWA 70 .
During the alignment, we use only bases with Phred Quality > 15 in each read to include soft clipping of lowquality bases, retain only uniquely mapped pair-end reads, and remove potential PCR duplicates. After completing initial alignment, the alignment was further refined by locally realigning any suspicious reads. The reported base calling quality scores obtained from the sequencer were re-calibrated to account for covariates of base errors. All variants with statistical evidence for an alternate allele present among individuals were identified using GATK HaplotypeCaller for multi-sample variant callings.
Neuroimaging analysis. Baseline T1-weighted brain MRI scans were downloaded from the ADNI database. FreeSurfer software was used to process T1-weighted brain MRI scans 71 and extract region of interest (ROI)-based imaging phenotypes 72,73 .
Gene-based association analysis. Since population stratification is known to cause spurious association in disease studies, we restricted our analyses to only subjects that clustered with CEU (Utah residents with Northern and Western European ancestry from the CEPH collection) + TSI (Tuscany in Italy) populations using HapMap 3 genotype data and the multidimensional scaling analysis (www. hapmap. org) [74][75][76] . A total of 757 ADNI participants (259 CN, 219 early MCI, 232 late MCI, and 47 AD) were used for analysis, where late and early MCI were defined as the cognitive performance below 1.5 standard deviations of the normative mean on a standard test and at the range of 1 to 1.5 standard deviations, respectively 77 . After extracting WGS-identified functional exonic SNPs within CASP6, we performed a gene-based association analysis of rare variants (minor allele frequency < 0.05) using SKAT-O software 78 . For hippocampal CA1 volumes, age, sex, year of education, MRI field strength, and total intracranial volume were used as covariates.
RNA-Seq analysis. RNA-Seq data (n = 1,966 individuals for the seven brain regions) reprocessed and realigned in the AMP-AD Consortium using a RNA-Seq pipeline 79  Caspase activity assays on Ac-VEID-AFC. The release of AFC from 20 µM Ac-VEID-AFC by active site titrated Casp6 (2, 10, 20, 50, 100, or 400 nM) was measured as described 46  Casp6 activity assay on protein substrates. To assess Lamin A/C cleavage, nuclear proteins were extracted from Casp6 knockout mice colon tissue 81 as described 46 . Casp6 (50-500 nM) were incubated with 3 µg lysate in Stennicke buffer at 37 °C for 5-60 min.
After the incubation period, samples were prepared in Laemmli buffer and analyzed by Western blot. The initial velocity of Casp6 on Lamin A/C or α-Tubulin was calculated based on the linear portion of the cleavage curve. The number of data points in the linear portion of each cleavage curve used for linear regression depended on the goodness-of-fit when R 2 was closest to 1.00.
Transfection, treatments, and protein extraction of HEK293T cells. The transfections were carried out with Lipofectamine 2000 (Invitrogen, Burlington, ON, Canada) and proteins extracted after 24 h in cell lysis buffer for Casp6 activity or in RIPA for western blots as described 14 . Transfection efficiency of Lipofectamine 2000 was optimized to obtain more than 90% transfected cells by calculating the number of fluorescent cells transfected with pBud-EGFP plasmid (Addgene #23027) over the number of total cells stained with Hoechst (Thermo Fisher Scientific). For protein degradation analysis, cells were treated 24 h after transfection with 75 µg/ml cycloheximide (CHD, Sigma-Aldrich) for the indicated times.

Treatment of transfected HEK293T cells with epoxomicin.
Cells were treated with 50 nM of epoxomicin (Enzo) 24 h after transfection for the indicated times and proteins extracted in cell lysis buffer. Proteasomal inhibition was confirmed in a proteasome activity assay as reported previously 48 using 50 µM Succinyl-Leu-Leu-Val-Tyr-7-amino-4-methylcoumarin (Suc-LLVY-AMC; Enzo).

RT-PCR.
Total RNA was extracted using TRIzol (Invitrogen) and converted to cDNA with avian myeloblastosis reverse transcriptase (Roche Diagnostics, Laval, QC, Canada). Casp6 cDNA was amplified with Taq DNA polymerase (New England Biolabs, Ipswich, MA, USA) as described 46 . Electrophysiology. All animal experimentation was approved by the McGill University Animal Care Committee and performed under guidelines and regulations in accordance with the ARRIVE guidelines. Hippocampal slices (300 µm) from 11-16 days old C57BL/6 mice were prepared in ice-cold ACSF 22 .
Multiple whole-cell patches were performed as described 82 . The stimulating electrode was placed in the CA1 stratum orients and delivered five 0.1-µs-long biphasic pulses (10-50 V) at 30 Hz every 30 s to elicit EPSPs in the range 1-3 mV. The recording was done for 1 h unless the data failed to meet the stability criteria defined as potential (± 4 mV), input resistance (± 15%) and temperature (32-34 °C). Offline data were analysed with Igor Pro (WaveMetrics Inc.). EPSP peak amplitudes were measured and averaged every 2.5 min (5 traces). Ensemble time courses were normalized to the first 2.5 min EPSP. Two-photon imaging and neuron reconstruction. Two-photon excitation was achieved using a Chameleon Ultra II femtosecond laser (Coherent, Santa Clara, CA, USA) tuned to 780 nm for both Alexa 594 and 488. Two-photon microscopes were custom-built 83 . Imaging data were acquired using customized versions of ScanImage 2018 (Vidrio Technologies, Leesburg VA USA) 84 running in Matlab (The MathWorks, Natick, MA, USA) via a PCI-6110 or a PCIe-6374 data acquisition board (National Instruments, Austin, TX, USA). When neurons had been well loaded with dye (> 1 h after break-in), the neuronal morphology was acquired as stacks of 512-by-512-pixels slices at 2 ms/line spaced by 2 µm. Maximum-intensity two-photon-imaging stacks compiled www.nature.com/scientificreports/ with ImageJ (NIH) were used for morphological identification (Fig. 7a), and for quantification of basal dendrite beading. Imaging montage of entire neurons was performed by Affinity Designer 1.7 (Serif Ltd, West Bridgford, Notts, UK). Image stacks were used for manual reconstruction of 3D morphologies (Fig. 7f) using the Neuromantic freeware(http:// www. readi ng. ac. uk/ neuromantic/body_index.php). Morphometry of 3D reconstructions (e.g. density maps, hulls, etc. in Fig. 7g,h) was subsequently performed using custom software 83 running in Igor Pro 8 (WaveMetrics Inc.). CA1 layer boundaries were identified using laser-scanning Dodt contrast images acquired simultaneously with 2-photon fluorescence.
Statistic study. Statistical analyses of data were performed using Igor Pro 8 or GraphPad Prism 7 with Student's t-test or one/two-way ANOVA, as indicated in figure legends.

Data availability
Materials generated for this study are available upon request. The datasets supporting the conclusions of this article are available in ADNI database (http:// www. loni. usc. edu/ ADNI/) for whole genome sequencing and MRI scan data the Sage Bionetworks (www. synap se. org) for human postmortem brain RNA-Seq data from seven brain regions.