Tuberous sclerosis complex‐1 (TSC1) contributes to selective neuronal vulnerability in Alzheimer's disease

Selective neuronal vulnerability of hippocampal Cornu Ammonis (CA)‐1 neurons is a pathological hallmark of Alzheimer's disease (AD) with an unknown underlying mechanism. We interrogated the expression of tuberous sclerosis complex‐1 (TSC1; hamartin) and mTOR‐related proteins in hippocampal CA1 and CA3 subfields.


INTRODUCTION
Alzheimer's disease (AD) is a devastating neurodegenerative disease reaching epidemic levels. Pathological hallmarks of AD include cerebral atrophy, extracellular amyloid-beta (Aβ) plaques and intraneuronal neurofibrillary tangles (NFTs) composed of hyper-phosphorylated tau [1]. A peculiar aspect of AD pathology is the striking predisposition of particular brain regions to neuronal and synaptic loss. One such area is the pyramidal cells in the Cornu Ammonis (CA)-1 subfield of the hippocampus [1,2].
Examination of brain areas known to differ in their vulnerability to AD-related pathology may enhance our understanding of pathogenic molecular mechanisms. Within the hippocampus, the CA1 subfield exhibits the greatest extent of neuronal loss compared with the CA3 subfield [2]. Neuronal and volumetric losses in CA1 uniquely predict AD severity and duration [3], and conversion from mild cognitive impairment to AD [4]. Interestingly, enhanced susceptibility of CA1 to neurodegeneration is not unique to AD, with similar patterns observed in cerebral ischaemia [5] and epilepsy [6]. Unravelling the processes underlying this region's specific neuronal death is central to understanding AD, and neurodegenerative disease more broadly.
Recent data suggest that the mammalian target of rapamycin complex 1 (mTORC1) signalling in CA1 may explain selective neuronal vulnerability to cerebral ischaemia [7]. mTORC1 is a pleomorphic complex involved in protein synthesis, cellular growth, proliferation and differentiation and has an inhibitory effect on autophagy [8]. The activity of this complex is negatively regulated by a host of factors, including tuberous sclerosis complex-1 (TSC1, also known as hamartin) physiologically and the drug rapamycin [8]. In an experimental ischaemia model, hippocampal CA3 neurons, which are resistant to cell death, increased expression of functional TSC1 in the setting of ischaemic injury, whereas CA1 hippocampal neurons, which are vulnerable to death, did not. TSC1 knockdown in CA3 neurons led to a vulnerability of these neurons to ischaemia in vivo [7]. These findings suggest that TSC1, through its inhibitory effects on mTORC1, may confer protection against hippocampal neurodegeneration and that a lack of mTOR inhibition within CA1 may explain its relative vulnerability.
In AD, there is growing interest in mTORC1 dysregulation, particularly overactivity [9]. mTORC1 plays a role in memory and cognition through modulating synaptogenesis and synaptic plasticity [10].
Dysregulation of mTOR signalling has even been suggested as an early event in AD pathogenesis [11,12] with implications for tau pathology through decreased autophagy [9,[13][14][15]. Given that TSC1 is a key inhibitor of the mTORC1 pathway, we hypothesised that TSC1 may be a crucial molecular link between selective hippocampal CA1 vulnerability, mTORC1 dysregulation and AD pathogenesis.
Through the combination of a detailed neuropathological study on human post-mortem hippocampal tissue, in vitro experiments on rodent hippocampal neuronal cultures and transcriptomic data, we showcase a novel role for TSC1 in AD pathogenesis. We highlight a mechanism through which TSC1 loss-of-function may mediate selective neuronal vulnerability of CA1 hippocampal neurons in AD.

Human cohort
Human post-mortem brain material derived from the Oxford Project to Investigate Memory and Aging (OPTIMA) cohort of the Oxford Brain Bank was used, according to Human Tissue Authority guidelines and Oxford Brain Bank ethics approval (REC 07/0606/85).
Twenty-six samples were examined: 10 severe AD cases, seven mild AD cases and nine non-neurologic controls. All groups were age-and sex-matched with minimal evidence of small vessel disease. Indicators  Table 1.
Formalin-fixed paraffin-embedded hippocampal tissue of 6-μm thickness containing CA1 and CA3 subfields was used to obtain neuropathological measures. Antibodies used are described in Table 2,   and staining procedures are detailed in the Supplementary Methods.   Fluorescent double labelling using the same primary antibodies, TSC1 with CkIδ, TSC1 with Tau, Tau with LAMP-1 and TSC1 with LC3 was performed on a subset of severe AD (n = 2) and control (n = 2) cases.

Quantitative and semi-quantitative analyses
To provide a global measure of mTOR activity, the distribution and extent of a subset of core mTOR-related pathway proteins (TSC1, hamartin; TSC2, tuberin; mTOR and phos-mTOR) in CA1 and CA3 subfields were assessed. TSC2 expression in hippocampal neurons from AD and control cases was investigated as it forms heterodimers that act in concert to inhibit the mTORC1 complex. The intricate staining pattern of mTOR and phosphorylated mTOR antibodies prompted a semi-quantitative assessment strategy as outlined in • TSC1 loss-of-function leads to cell death in an in vitro model, independent of amyloid-beta toxicity.
• TSC1 dysregulation can lead to AD-related molecular responses. To estimate the extent of AD pathology, neuronal densities were manually counted in each area of interest, and the area of greatest amyloid and tau burden in each subfield was photographed and assessed using optimised methods [16][17][18], by measuring the proportion of pixels positively labelled with 4G8 staining (Aβ) and AT8 staining (Tau) (Supplementary Methods). For each analysis described above, a minimum of three high magnification (400Â) photographs were taken from CA1 and CA3 hippocampal subfields for each case for downstream analyses. The specificity of reactions was confirmed using both positive and negative control sections, the latter where the primary antibody was omitted.

Statistical analyses
Statistical analyses were conducted using GraphPad Prism 9.
Normality was assessed via the Shapiro-Wilk test and through visual assessment of normality plots. Data were not normally distributed so non-parametric tests were used. There were no significant differences in age, sex or post-mortem interval between subject groups.
Mann-Whitney tests were used for between-case comparisons, and the Wilcoxon matched-pairs signed rank test was used for within-case High-content live cell imaging of rate of cell death and caspase-3 activation The cell death rate was monitored by adding a CellTox™ Green intercalating cytotoxicity dye (Promega) in cell medium (1Â) to rat hippocampal cells simultaneously with Aβ 1-42 and rapamycin or control vehicle. Apoptosis was also monitored by caspase-3 activation using Nucview488 (Biotum, dilution 1:800). Cell death and caspase-3 activity rate were monitored in real-time for 120 h by using the incucyte

ZOOM system. The analysis is outlined in the Supplementary
Methods.

In vitro statistical analyses
Statistics were conducted using GraphPad Prism 9. Statistical tests used are indicated in each data figure. All data are presented as mean ± SD in the text and graphs.

RNA-Seq
In a separate experiment, samples were taken for RNA sequencing using optimised protocols outlined in the Supplementary Methods.
Wild type and TSC1 knockdown rat hippocampal neurons were treated with Aβ or vehicle. The RNA-seq reads were aligned to the Rat Rn6 genome using star aligner (v2.5.3a) [19]. The gene level expression was then quantified based on the Ensembl gene model (Ensembl release 90) by using featureCounts (v1.4.6) of the Subread package [20]. Genes with at least 1 count per million (CPM) reads in at least 1 library were considered expressed and were discarded otherwise. After filtering, the expression levels of the remaining genes were normalised by the trimmed mean of M-values (TMM) [21].
normalisation method to adjust for differences in library sizes. It is noted that count data show non-trivial mean-variance relationships.
Therefore, the voom function in the Bioconductor Limma [22] package was used to infer the mean-variance trend of the normalised logarithmic gene count data to adjust for heteroscedasticity. After TSC1 knockdown signature enrichment in the genes correlated with the mTOR pathway To calculate the genes correlated with the mTOR pathway, we used the hippocampal gene expression data from our recent microarray transcriptomic analysis of AD and control brains in the MSBB-AD cohort [24]. We first computed Pearson's correlation (r) between known mTOR pathway genes and the rest of the transcriptome. Next, we selected those genes with r larger than a given cutoff with any mTOR pathway gene. Then, we applied the FET to assess the enrichment of the TSC1 knockdown signature in the genes correlated with the mTOR pathway at a series of correlation cutoffs ranging from 0.5 to 1. When the r cutoff was 1, only the known mTOR pathway genes were selected for enrichment test.

TSC1 cytoplasmic inclusions and phos-mTOR expression are selectively increased in CA1 hippocampal neurons in human AD
In AD, we found TSC1 aggregates in the CA1 through CA4 hippocampal subfields and in the subiculum with an absence of TSC1 aggregates noted in the dentate gyrus ( Figure S2). TSC1 aggregates of smaller size were rarely found in entorhinal cortical neurons. In AD, the distribution and morphology of TSC1 expression in the CA3 subfield did not differ from controls ( Figure 1A,B). This contrasted dramatically with the CA1 subfield where the majority of AD cases had frequent distinct, coarse granular cytoplasmic inclusions containing TSC1 sparsely seen in controls ( Figure 1C,D). A selective increase in the number of hippocampal neurons with TSC1 inclusions in the CA1 subfield was pronounced in severe AD ( Figure 1E). Conversely, TSC2 expression was decreased in the CA1 hippocampal subfield of AD cases compared with the CA1 and CA3 hippocampal subfields of controls ( Figure S3). Alongside this, the intensity and proportion of neurons expressing phos-mTOR in CA1 hippocampal neurons were increased in severe AD compared with controls ( Figure 1F,G). A concurrent increase in phos-mTOR expression was noted with disease severity, though this did not differ between the CA1 and CA3 subfields ( Figure 1H).
As expected, neuronal density was reduced in the CA1 subfield in severe AD compared with controls, whereas hippocampal Aβ and tau burden increased with disease severity, with a predisposition for CA1 compared with CA3 ( Figure S4).

TSC1 knockdown in rat hippocampal neurons results in increased cell death
We developed an in vitro TSC1 knockdown model by using lentiviral knockdown of TSC1 in cultured rat hippocampal neurons based on previous work [7]. (Table S1 and Figures S10-S14).
Cell death was measured in rat hippocampal primary cultures treated with a combination of validated TSC1 knockdown with and without rapamycin (Figures 3 and S15). TSC1 knockdown led to increased cell death throughout the culture period ( Figures 3A and   S15A). Caspase-3 activation-a marker of apoptosis-showed a similar pattern to the total cell death ( Figures 3B and S15B), suggesting that apoptosis mediates at least some of the TSC1 knockdownrelated cell death. Surprisingly, rapamycin treatment did not reduce the extent of TSC1 knockdown-related cell death or caspase-3 activation at any time point, which suggests a mechanism independent of mTORC1. hTSC1 overexpression reduced total cell death and reversed the effects of TSC1 knockdown on caspase-3 activation ( Figure 3C,D).
Unexpectedly, treatment with Aβ 1-42 in TSC1 knockdown or hTSC1 rescue experiments did not have additional effects on mTOR activation (Figures S11-S14), Tau burden (Figures S16 and S17), total cell death or caspase-3 activation ( Figure S18).  Figure 4C). In particular, the TSC1 DEG signatures were significantly enriched for the genes differentially expressed between AD and control in the CA1 and CA3 subfields from the Miller et al. [29] study (up to 2.4-fold, FET adjusted p-value up to 5.0E-33), suggesting that the molecular response to the TSC1 knockdown mimics the gene expression changes in AD, consistent with the cellular phenotype (i.e., increased cell death) due to the TSC1 knockdown.

Pathway analysis reveals disruption of hippocampal neuronal function
Gene ontology and pathway analysis provided further insights into the molecular mechanisms by which TSC1 loss-of-function in AD may play a role in disease ( Figure 4D). In particular, genes associated with with human AD co-expression network modules [33], DEGs in our TSC1 knockdown model significantly overlapped various modules ( Figure 4E). Specifically, the up-regulated genes were enriched in the modules associated with response to biotic stimulus or defence response, including the yellow, cyan and light cyan modules, which were ranked as 1, 7 and 25, respectively, by association to AD pathology in the original human AD network analysis study [33]. On the other hand, the down-regulated genes were enriched in the nerve ensheathment modules. In addition, we found that the down-regulated genes induced by TSC1 knockdown were significantly enriched for the mTOR pathway and the genes correlated with the mTOR pathway at various correlation cutoffs (Table S2), suggesting TSC1 as an upstream regulator of the mTOR pathway. Overall, our findings suggest that TSC1 loss-of-function activates AD-associated immune responses and suppresses neuronal function, thus supporting a role of TSC1 in selective neuronal vulnerability in AD.

DISCUSSION
In this study, we investigate a novel role for TSC1 in mediating selective neuronal vulnerability in AD. Using a systematic approach guided Overexpression of hTSC1 combined with rat TSC1shRNA (30.9 ± 5.0%) significantly reduced cell death compared with TSC1shRNA + overexpression at 120 h (p = 0.0002), whereas cell death remained increased compared with the control group (p < 0.0001). (D) hTSC1 overexpression more completely rescued the effect of TSC1 knockdown on caspase-3 activation. Caspase-3 activation was increased starting from 48 h with TSC1 shRNA treatment compared with the control group (58.8 ± 23.3% vs 27.1 ± 13.0% at 120 h, p < 0.0001). In the TSC1shRNA + hTSC1 overexpression group, this effect was reduced to the control level throughout the culture period, such that there was no difference between rescue and control conditions at any point (34.9 ± 5.7% at 120 h, p = 0.216). In C and D, markers above the bars indicate the significance of comparisons between TSC1shRNA + overexpression control and control shRNA + overexpression control, whereas significance bars at the right-hand side of plots indicate comparisons between indicated treatment groups at 120 h. (All statistical tests were made using Tukey's test. ns = non-significant, *p < 0.05, ** p < 0.005; *** p < 0.0005; **** p < 0.0001).
by human neuropathological findings, followed by an in vitro rodent hippocampal neuronal culture model and RNA-seq data, our data implicate TSC1 loss-of-function as a driver of selective CA1 neuronal loss in the AD hippocampus.
In human post-mortem tissue, we found striking cytoplasmic TSC1 inclusions and evidence of mTOR activation mainly restricted to CA1 neurons in AD. Importantly, the extent of these inclusions related to decreased neuronal density in CA1, the severity of cognitive impairment during life and markers of AD pathology after death. Further, TSC1 inclusions were mainly observed in cells with granulovacuolar degeneration. However, the TSC1 inclusions observed in our study co-localised to separate compartments with the granules seen resembling not only those found in granulovacuolar degeneration but also in frustrated autophagy, the latter being most prevalent in the CA1 subfield in severe AD [34].
TSC1 inhibits mTOR, so we expected increased levels of TSC1 to result in reduced mTOR activation. Instead, we found a striking correlation between the number of TSC1 inclusions with activated phos-mTOR, suggesting that TSC1 is inactive in these inclusions. As TSC1 exerts its functional activity when associated with the lysosome, our observation of selectively reduced lysosomal activity in the CA1 subfield in severe AD and the association of TSC1 inclusions with markers of autophagy further support this hypothesis. Thus, the presence of TSC1 inclusions must not be confused with increased TSC1 expression, which has been shown by others to be neuroprotective [7]. Our finding of reduced TSC2 expression in the CA1 hippocampal subfield of AD cases is relevant here given in vitro data that show TSC2 down-regulation leads to the formation of TSC1 aggregates with resultant loss of function leading to mTORC1 hyperactivation [35][36][37]. Data derived from whole brain homogenate preparations support a role of mTOR hyperactivation in the mesial temporal lobe in AD [15], but there has been no research into cellspecific mTOR changes in AD. Our findings fill this critical knowledge gap. The fact that mTOR activation was far more pronounced in vulnerable CA1 neurons compared with resistant CA3 hippocampal neurons, which lie only micrometres away, provides an important biological clue to the specific substrate of selective neuronal vulnerability in AD.
Our in vitro experiments provide evidence that TSC1 loss-offunction promotes mTOR hyperactivation and leads to cell death independent of amyloid exposure. Taken together with our pathology findings that TSC1 dysregulation selectively impacts the CA1 subfield in the AD hippocampus, our data support a mechanism of selective neuronal death in the AD CA1 hippocampal subfield not heretofore described. Previous work has implicated mTOR hyperactivation globally in the AD brain but our findings point to a more region-specific role of mTOR dysregulation in the pathogenesis of neuronal loss in AD. What is more, we highlight an important contribution of TSC1 loss-of-function in driving cell death that is separate from amyloid toxicity, which may cast light onto the debated role of amyloids in cognitive decline in AD.
The incomplete rescue of the TSC1 knockdown phenotype in our overexpression and rapamycin experiments raises the possibility of off-target effects that contributes to cell death. We provide some evidence that rapamycin acts on mTORC2, another mTOR complex that-when activated-leads to PKC phosphorylation and cytoskeletal changes [38]. It is also plausible that shRNA acts on other genes, and irreversible cell death cascades may have been triggered on which rapamycin and hTSC1 would have no effect. Other downstream targets of TSC1 outside of mTOR may be considered, including Rheb, which has been shown to mediate aggresome formation and cell death in response to misfolded proteins, downstream of TSC1 but independent of mTOR [39]. Our analyses of the transcriptional programs caused by TSC1 knockdown in vitro provides relevance of TSC1 loss-of-function in hippocampal selective vulnerability in AD. In our TSC1 knockdown experiments, we identified AD-related pathway changes, including down-regulation of axon guidance and synaptic transmission and up-regulation of interferon-gamma signalling and lysosome biology. Additionally, we found genes up-regulated by TSC1 inhibition that were enriched for ECM and ECM-related pathway matrisome [40]. Significant changes in ECM have been reported in the early stages of AD [41] and inhibiting hippocampal ECM changes restored both long-term potentiation and contextual memory performance in APP/PS1 transgenic mice [42]. Moreover, the present transcriptomic signatures were significantly enriched in the immune and nerve ensheathment-related human AD gene network modules that F I G U R E 4 Analysis of the transcriptomic response to TSC1 knockdown. (A) Overlap among the TSC1 knockdown gene signatures identified at different time points. (B) AD GWAS risk genes regulated by TSC1 knockdown. *p < 0.05; **p < 0.01; ***p < 0.001. (C) TSC1 knockdown signatures significantly overlapped with known human AD gene signatures including (1) Avramopoulos, signature detected in the temporal lobe from Avramopoulos et al 2011 [25]; (2) Blalock, signature detected in the hippocampus from Blalock et al 2004 [26]; (3) Colangelo, signature detected in the hippocampus CA1 region from Colangelo et al. [27]; (4) Liang, signature detected in multiple cortex areas from Liang et al. [28]; (5) Miller_CA1, signature detected in the hippocampus CA1 region from Miller et al. [29]; (6) Miller_CA3, signature detected in the hippocampus CA3 region from Miller et al. [29]; (7) Myers, signature computed by comparing gene expression between AD and control from the cerebellum region from Webster et al. [32]; (8) Myers_TC, signature computed by comparing gene expression between AD and control from the temporal cortex region from Webster et al. [32]; (9) Satoh, signature detected in the frontal cortex region from Satoh et al. [30]; (10) Wang_PHG, signature detected between AD and control brains as defined by CERAD in the parahippocampal gyrus region from Wang et al. [31]; (11) Zhang_Atrophy_CB, signature correlated with atrophy in the cerebellum region from Zhang et al. [33]; (12) Zhang_Atrophy_PFC, signature correlated with atrophy in the PFC from Zhang et al. [33]; (13) Zhang_Braak_CB, signature correlated with Braak staging in the cerebellum region from Zhang et al. [33]; (14) Zhang_Braak_PFC, signature correlated with Braak staging in the PFC from Zhang et al. [33] Up, up-regulation; Dn, down-regulation. (D) Top GO/pathways enriched in TSC1 knockdown signatures. (E) Human AD co-expression network modules enriched in TSC1 knockdown signatures. X-axis lists the pathway names of the human network modules, with the module name and the ranking in relevance to AD pathology shown in parentheses.
What causes TSC1 to form inclusions in CA1? Amyloid and tau are obvious culprits but, in our study, seem to be less related than expected, and our findings support the idea that reduced expression of TSC2 could be to blame [47]. A gene associated with insulin signalling defects called FTO has been identified as upstream of TSC1/TSC2 and the mTOR pathway in AD [48], which is interesting because type 2 diabetes is an increasingly recognised risk factor for AD. Although Academy of Neurology and MS Academy. MME is a co-founder of Cytox Group, a company that has developed genoscores to assist in AD diagnosis. All other authors declare no conflict of interest.

DATA AVAILABILITY STATEMENT
The MSBB-AD microarray dataset used in the mTOR pathway