FOXO3 targets are reprogrammed as Huntington's disease neural cells and striatal neurons face senescence with p16INK4a increase

Abstract Neurodegenerative diseases (ND) have been linked to the critical process in aging—cellular senescence. However, the temporal dynamics of cellular senescence in ND conditions is unresolved. Here, we show senescence features develop in human Huntington's disease (HD) neural stem cells (NSCs) and medium spiny neurons (MSNs), including the increase of p16INK4a, a key inducer of cellular senescence. We found that HD NSCs reprogram the transcriptional targets of FOXO3, a major cell survival factor able to repress cell senescence, antagonizing p16INK4a expression via the FOXO3 repression of the transcriptional modulator ETS2. Additionally, p16INK4a promotes cellular senescence features in human HD NSCs and MSNs. These findings suggest that cellular senescence may develop during neuronal differentiation in HD and that the FOXO3‐ETS2‐p16INK4a axis may be part of molecular responses aimed at mitigating this phenomenon. Our studies identify neuronal differentiation with accelerated aging of neural progenitors and neurons as an alteration that could be linked to NDs.

Interestingly, NDs have been linked to cellular senescence, particularly that of glial cells (Bussian et al., 2018;Chinta et al., 2018;Musi et al., 2018;Zhang et al., 2019). However, the temporal dynamics of cellular senescence in NDs and the role of FOXO gene regulation in this context are unresolved, limiting our capacity to target the detrimental effects of cellular senescence in NDs.
We hypothesized that FOXO gene regulation might be able to oppose cellular senescence in ND conditions. We tested this hypothesis in human cell models of Huntington's disease (HD), a genetic yet a primarily late-onset ND caused by CAG expansion in HTT.
We focused on FOXO3, a FOXO factor that is neuroprotective in HD (Tourette et al., 2014). Although FOXO3 is pivotal to neuronal homeostasis in HD, human FOXO3 targets are unknown, including in ND conditions. Here, we found that human HD induced pluripotent stem cell (iPSC)-derived neural stem cells (NSC) reprogram FOXO3 targets in the context of cellular senescence features that are acquired at the time of neuronal differentiation and that are more pronounced in medium spiny neurons (MSNs). These features include the increase of p16 INK4a , a key inducer of cellular senescence (Baker et al., 2016). Remarkably, FOXO3 target reprogramming represses the transcription modulator and p16 INK4a activator ETS2 (Irelan et al., 2009), which antagonizes p16 INK4a expression and which may represent an adaptive response as p16 INK4a promotes the senescence of human HD NSCs and MSNs. Together, these data reveal that cellular senescence may develop during neuronal differentiation in HD, affecting striatal neurons, and that FOXO gene regulation may tip the balance away from the detrimental consequences of cellular senescence via ETS2-p16 INK4a , providing a rationale and strategy for targeting cellular senescence during the early phases of NDs, before the onset of overt neuronal injuries and cell death, a crucial need in HD and other NDs.

| Ryk-ICD binds to Armadillo repeats 9-10 of ß-catenin
In HD, FOXO3 neuroprotection is altered by increased mRNA and protein expression of Ryk (Tourette et al., 2014), a Wnt receptor important for axon guidance and neurogenesis (Andre et al., 2012).
This effect, a consequence of gene deregulation in HD, is mediated by the Ryk intracellular domain (Ryk-ICD) in the nucleus where Ryk-ICD binds to the FOXO3 partner ß-catenin (Tourette et al., 2014).
We overexpressed a Myc-tagged Ryk-ICD fragment, as these cells normally produce relatively small amounts of this gamma-secretase cleavage product (Tourette et al., 2014). The Myc-tagged Ryk-ICD fragment co-precipitated with ß-catenin when endogenous FOXO3 was targeted by the immunoprecipitating antibody ( Figure 1a) as well as with FOXO3 when endogenous ß-catenin was targeted by the immunoprecipitating antibody ( Figure 1b). Thus, Ryk-ICD may be an integral part of the ß-catenin/FOXO3 complex.

| Human HD NSCs reprogram FOXO3 targets
Having shown that Ryk signaling may modulate FOXO3 gene regulation through spatial/allosteric modifications of the ß-catenin/ FOXO3 complex, we used massively paralleled RNA sequencing (RNA-seq) and chromatin immunoprecipitation followed by sequencing (ChIP-seq) to identify FOXO3 direct targets (F3Ts) in human HD cells. A human-induced pluripotent stem cell (iPSC) model of HD, in which isogenic cells express mutant (72Q/19Q) HTT (HD) or CAGcorrected (21Q/19Q) HTT (C116) was used Ring et al., 2015). HD and C116 NSCs were treated with Ryk or scrambled sequence siRNAs. As expected, HD NSCs showed increased (1.2 fold) Ryk mRNA levels ( Figure S1A) and a 2 fold in human HD MSNs ( Figure 6b) whereas ß-catenin and FOXO3 mRNA levels are similar in HD and C116 cells (Figure S1A, middle and right panels).
We then induced FOXO3 nuclear translocation (see Section 4) in Ryk siRNA-silenced NSCs (Figure S1B,D) prior to collecting RNA-seq and FOXO3 ChIP-seq data.
Much of FOXO3 transcriptional activity can be due to binding enhancers (Eijkelenboom, Mokry, de Wit, et al., 2013), and there is a significant association between gene regulation and FOXO3 binding up to 20 kb from transcriptional start sites in human cells (Eijkelenboom, Mokry, Smits, Nieuwenhuis, & Burgering, 2013). We thus defined FOXO3 direct targets as genes that (a) show FOXO3 binding at promoter and enhancer regions (±20 kb) as determined by ChIP-seq data and (b) are up-or down-regulated upon FOXO3 induction into the nucleus as determined by RNA-seq data (Table S1/ sheet 1). Additionally, we used RNA-seq data upon FOXO3 knockdown (Table S1/sheet 2). However, ß-catenin transcriptional activity may bypass the absence of TCF/LEF (Doumpas et al., 2019) and, possibly, that of FOXO3 (Essers et al., 2005). Hence, FOXO3 nuclear induction and knockdown could differently alter gene regulation, which calls for caution in using FOXO3 knockdown data to prioritize F3Ts. We thus used these data as a bona fide criterion for defining two classes of F3Ts, that is, those identified by (a) FOXO3 nuclear induction (F3T-IN) and (b) FOXO3 nuclear induction and FOXO3 knockdown (F3T-IN-KD) ( Table S1).  (Table S2/sheet-1). Increased FOXO3 occupancy was also true for F3T-INs that are gained (Table S2/ Sheet-2) or conserved (Table S2/ Together, these results suggest that F3Ts are reprogrammed in response to HD during neurogenesis and this response cannot be fully attributed to higher FOXO3 occupancy. Rather, Ryk signaling may act as a significant modifier of FOXO3 activity.

| FOXO3 binding sites are enriched for coregulator motifs
FOXO gene regulation involves other transcription factors that synergize with or antagonize FOXO proteins (Webb et al., 2013). The  Figure S3), which could be due F I G U R E 1 FOXO3, ß-catenin, and Ryk-ICD form a protein complex in HEK293T cells. The antibodies used for immunoprecipitation (IP) and for immunoblotting (IB) are indicated across panels. (a) FOXO3, ß-catenin, and Ryk-ICD co-precipitate in pull-down experiments. For negative control, an IgG isotype was used. Representative Western blots for IP of endogenous FOXO3. (b) FOXO3, ß-catenin and Ryk-ICD co-precipitate in pull-down experiments. For negative control, an IgG isotype was used. Representative Western blots for IP of endogenous ß-catenin. (c) Deletion mapping of the Armadillo repeat region in ß-catenin. Ryk-ICD binds to Armadillo repeats 9-10 of ß-catenin. HEK293T cells were transfected with Myc-Ryk-ICD construct and the indicated deletion mutants (∆277-488, ∆489-593). For negative control, an IgG isotype was used. (d) Representative Western blots for IP of wild-type and mutant ß-catenin-FLAG constructs shown in (c) F I G U R E 2 FOXO3 binding and gene regulation in human NSCs expressing normal or mutant HTT with or without Ryk silencing. (a) Enrichment of FOXO3 binding around the transcriptional start sites (TSSs) (±2 kb) in human NSCs expressing normal (C116: 19Q/21Q) or mutant HTT (HD: 72Q/21Q) and treated with Ryk siRNA-1 (siRyk) or scrambled RNA (scramble). The color scale is chip signal intensity with maximum set as 8.0. (b) FOXO3 binding at specific loci in human C116 or HD NSCs. The upper panel is a FOXO3 binding site present in C116 and HD cells at the TMEM132C locus. The lower panel is a FOXO3 binding site gained in HD cells at the STK4 locus. (c) Venn diagrams depicting F3 gene regulation across the 4 conditions tested. The left panel shows FOXO3-dependent genes (RNA-seq data). The middle panel shows FOXO3 binding (ChIP-seq data). The right panel shows the distribution of F3Ts, highlighting an increase in the number of F3Ts upon Ryk silencing in C116 (p < 2.2e-16) and HD (p < 2.2e-16) cells. (d) FOXO3 binding for the FOXO3-dependent genes. The left panel shows the percentages of genes with binding or no binding. Multiple chi-square tests were performed using the R function pairwise.prop. test. The right panel shows the signal (peak score) distributions. Chi-squared test was performed for global and pairwise comparisons of the distributions with the R function chisq.test to better sampling of DNA fragments near FOXO3 binding sites owing to chromatin modifications in HD cells (Achour et al., 2015). Silencing Ryk did not alter co-regulator motif profiles in either HD or C116 cells, except for Tbrain factor motifs as detected in HD cells with Ryk silencing ( Figure S3), supporting a model in which Ryk signaling modulates FOXO3 gene regulation by altering the stoichiometry of the FOXO3/ßcatenin/Ryk-ICD complex (Figures 1 and 2).

| FOXO3 binding sites overlap between human C116 and mouse NSCs
Some FOXO3 target families, for example, those responding to stress and proteotoxicity, may be conserved across species and cell types (Webb, Kundaje, & Brunet, 2016) while others are not conserved (Webb et al., 2016). We compared FOXO3 binding sites in human C116 NSCs with those previously reported in mouse NSCs (Webb et al., 2013) as both studies similarly analyzed F3Ts. To this end, we considered the best human orthologs of mouse genes bearing FOXO3 binding sites. A significant overlap was detected (20.25%, 446 genes, p = 3.15 × 10 −8 ), suggesting that FOXO3 gene regulation and functions have common features in human and mouse NSCs (Table S3, Figure S4).  (Table S5: 111 genes) showed no enrichment for pathways, but displayed low-significance enrichment for the GOBPs Positive regulation of JNK cascade (p = 1.88 10 −04 ) and Regulation of cell cycle (p = .0054).

| F3T reprogramming in human HD NSCs implicates regulators of cell senescence
To enhance the precision of F3T prioritization, we performed network analysis using F3T-INs as seeds for extracting high-confidence networks from the STRING database (Szklarczyk et al., 2015).
This analysis highlighted interconnected F3T-INs that implicate core FOXO3 functions such as for example transcription, translation, and protein quality control (Webb et al., 2016). These F3Ts included those in the conserved ( Figure S5A) or reprogrammed ( Figure S5B,C) group(s) that belong to Wnt, Hippo/TGF-ß (e.g., LATS2), Toll-like receptor and mTOR signaling. This analysis also highlighted Rykindependent and Ryk-dependent F3Ts that in HD NSCs are relevant to neuron differentiation, synaptic function, and cell cycle ( Figure   S5B,C). In the Ryk-dependent group, network analysis (here, F3T-INs showing the strongest regulation by FOXO3 and at least 3 out of 4 classes of F3T-INs connected to the same node) predicted that, in HD NSCs, FOXO3 (a) no longer activates CDKN2AIP (also known as CARF), a co-activator of p14 ARF (b) activates SERTAD1 (also known as p34(SEL1)), an inducer of neuronal apoptosis when in excess that renders CDK4 resistant to inhibition by p16 INK4a (Li et al., 2005) and (c) represses ETS2, a transcription factor that positively regulates p16 INK4a expression (Ohtani et al., 2001) and a F3T-IN-KD gene.
Together, these FT3 changes suggest suppression of the CDKN2A locus, particularly the p16 INK4a segment, in HD NSCs.

| FOXO3 represses ETS2 expression in human HD NSCs
We performed validation studies of FOXO3 regulation of SERTAD1, ETS2, and CDKN2AIP in human NSCs subjected to stress.

| Prepatterned HD NSCs show cellular senescence features in striatal neurons
Given that p16 INK4a is a key effector of cellular senescence (Baker et al., 2016), we tested whether F3T reprogramming in HD NSCs might occur in the context of and respond to cellular senescence acquired in HD during neuronal differentiation. Using Activin A-induced dorsoventral prepatterning, which efficiently directs striatal projection neuron differentiation of human iPSCs (Arber et al., 2015), we observed increase of p16 INK4a mRNA and protein levels in HD compared with C116 prepatterned NSCs (Figure 4a,c). In addition, senescence-associated ß-galactosidase (SA-ß-gal) activity was more abundant in HD compared with C116 NSCs (Figure 4d (Figure 5c-e). We also found that other cellular senescence markers including CDKN2AIP, MMP3, SELL, and IGFBP7 were all upregulated in HD MSNs when compared F I G U R E 3 Gene expression analyses in human NSCs. The mRNA levels are normalized to cells treated with nontargeting control (NTC) siRNAs (siRNA tests) or to C116 cells or cells without growth factor (GF) deprivation (other experiments). ns, not significant. (a) ETS2 mRNA levels are increased by FOXO3 reduction in HD NSCs subjected to GF deprivation with no effect detected in basal conditions nor in normal HTT cells (left panel: *p < .05). ETS2 mRNA levels are decreased in HD NSCs (middle panel: **p < .01). GF deprivation does not change ETS2 mRNA levels in C116 and decreases ETS2 mRNA levels in HD NSCs (right panel: *p < .05). (b) p16 INK4a mRNA levels are decreased by ETS2 reduction in HD NSCs in basal conditions and in cells subjected to stress with no effect detected in normal HTT cells (left panel: *p < .05, **p < .01). p16 INK4a mRNA levels are increased in HD NSCs (middle left panel: ***p < .001). GF deprivation does not change p16 INK4a mRNA levels in C116 NSCs and decrease p16 INK4a mRNA levels in HD NSCs (middle right panel: *p < .05). p16 INK4a mRNA levels tend to be increased by FOXO3 knockdown in HD NSCs subjected to GF deprivation (right panel: not significant with p = .0736). (c) ETS2 and p16 INK4a mRNA levels are decreased by overexpression of FOXO3, but not that of FOXO3-TM, in human HD NSCs subjected to GF deprivation. The mRNA levels are normalized to cells treated with empty vector. *p < .05 and **p < .01.  (Ohtani et al., 2001). Human HD MSNs also showed decreased levels of nuclear HMGB1 ( Figure   S9D), which relocalizes to the extracellular space in senescent cells (Davalos et al., 2013), an effect not observed in HD NSCs. The size of a senescent cell increases when compared to nonsenescent cells. We found that human HD MSNs have an increase nuclear area when compared to C116 MSNs ( Figure S9E). Together, these data suggest the differentiation of NSCs into striatal like neurons is accompanied by increasingly pronounced features of cellular senescence in HD.

| FOXO3 and p16 INK4a oppositely modulate the vulnerability of human HD NSCs
Next, we investigated whether FOXO3 activity in human HD NSCs might oppose the effects of p16 INK4a . In cell growth assays, HD NSCs divided more slowly compared with C116 NSCs (Figure 6a In cell vulnerability assays, reducing FOXO3 expression ( Figure   S5F) strongly potentiates the mortality of HD NSCs with no effect in C116 cells (Figure 6d). In contrast, reducing p16 INK4a expression ( Figure S6G) decreased the mortality of HD NSCs, with no effect de-

To further understand the role of p16 INK4a in differentiated HD
MSNs and cellular senescent-like features, we transduced these cells with lentivirus expressing p16 INK4a . We tested for HMGB1, an early responder to cellular senescence. We quantified cytoplasmic HMGB1 levels as a more sensitive measure of a senescent-like phenotype compared with nuclear HMGB1 (Figure 6e). We found increased cytoplasmic HMGB1 basal levels in HD compared with C116 MSNs (Figure 6f). We also found increased cytoplasmic HMGB1 levels upon p16 INK4a transduction in HD and C116 MSNs (Figure 6f).
These data suggest that p16 INK4a increase in HD MSNs may promote senescent-like features in these cells.  Figure   S10C). Further studies in mice and human tissues will be needed to confirm cellular senescence.

| DISCUSS ION
FOXO factors have widespread anti-aging effects via the transcriptional regulation of stress response in multiple cell contexts (Martins et al., 2016;Salih & Brunet, 2008). Several cell maintenance mechanisms under FOXO control are affected in several NDs (e.g., mitochondrial homeostasis, proteostasis, autophagy, immune system, DNA repair). Understanding how FOXO gene regulation modulates brain cell maintenance in NDs may thus have important therapeutic implications. Although FOXO gene regulation has been studied in several cellular contexts (Webb et al., 2016), including in NSCs (Webb et al., 2013) and neurons (McLaughlin & Broihier, 2017), human FOXO targets in ND conditions are unknown as well as the biology of these targets in patient-derived cells. Our data identify FOXO3 targets in human NSCs, suggesting a model in which human NSCs reprogram F3Ts in response to HD. Remarkably, this response takes place in the context of senescence that develops in these cells, involving the repression of the ETS2-p16 INK4a axis, a mechanism that is part of the Ryk-dependent element of F3T reprogramming.
Our data suggest that Ryk signaling is a primary factor that modifies the FOXO3 target space. However, Ryk may signal through multiple mechanisms, including the canonical Wnt, PCP, and Ryk-ICD pathways (Andre et al., 2012;Lyu, Yamamoto, & Lu, 2008;Tourette et al., 2014), and the effects of silencing Ryk on the F3T repertoire might also result from changes in pathways other than the Ryk-ICD pathway. Nonetheless, in HD cells, our data suggest that pathways that signal onto FOXO3 such as Ryk/Ryk-ICD signaling play a primary role in modifying the F3T repertoire, rendering FOXO3 able to fine-tune the expression of key inducers of cellular senescence such as p16 INK4a , whereas the increase of FOXO3 occupancy may primarily reflect the wide-spread effect of HD on chromatin remodeling (Achour et al., 2015). (about 20%) that is elicited by the EST2-p16 INK4a axis, our data indicate that reinforcing the outcome of FOXO3 activity in response to HD, that is, by further inhibiting p16 INK4a levels, may have therapeutic potential to avoid the harmful effects (maladaptation) of a chronic cellular senescence response in human HD neurons. Such an approach might be of interest for promoting adult neurogenesis in HD as adult-born neurons may be depleted in the striatum of human HD brains (Ernst et al., 2014) and for targeting the detrimental consequences of neuronal senescence in other ND contexts. The ability of FOXO3 to tip the balance away from cellular senescence in response to CAG expansion in HTT could persist in adult neurons as the deregulation of senescence markers may be conserved from developmental to adult stages. Consistent with this, our data show F I G U R E 6 FOXO3 and p16 INK4a oppositely modulate the vulnerability of human HD NSCs. Significance was tested using two-way ANOVA (panels a-c), paired t test (panels d) and Mann-Whitney test (panel g). ns: not significant. (A) Human HD NSCs show reduced rates of cell growth. Data are mean ± SEM. (b) Reducing FOXO3 does not alter the growth of C116 NSCs (left panel) and strongly reduces the growth of HD NSCs (right panel), with no change detected in HTT mRNA levels (see Figure S6F, left panel). Data are mean ± SEM. (c) Reducing p16 INK4a slightly increases the growth of C116 (left panel) and HD (right panel) NSCs. Reducing p16 INK4a does not alter HTT mRNA levels in HD NSCs (see Figure S5F, right panel). Data are mean ± SEM. (d) Reducing FOXO3 increases the mortality of HD NSCs with no effect detected in C116 NSCs (left: *p < .05). Reducing p16 INK4a decreases the mortality of HD NSCs with no effect detected in C116 NSCs (right: *p < .05). (e) Lenti-myc-p16 INK4a transduction promotes nuclear release of HMGB1 in cytoplasm of HD and corrected (C116) MSNs. HD and C116 MSNs transduced for 4 days with lenti-myc-p16 INK4a (red) were immunostained with HMGB1 (green). NT: transduction without myc- Our data suggest that neural and neuronal senescence could be set early in HD, a ND associated with chromatin remodeling (Achour et al., 2015), and has potential to be prosecuted in view of early drug trials (e.g., during prodromal disease). However, additional studies in the brain of HD mice and in human HD postmortem brains are needed to test for the relevance of senescence to HD.

Stress response involves p16
We attempted to test for p16 INK4a  Given the tight links between chromatin remodeling, NDs and cellular senescence (Achour et al., 2015;Criscione, Teo, & Neretti, 2016;Jakovcevski & Akbarian, 2012), our data raise the possibility that neural/neuronal senescence could be set early in NDs such as Alzheimer's and Parkinson's. Although senolytics may positively impact on brain activity in mouse models of NDs via removing senescent glial cells (Bussian et al., 2018;Zhang et al., 2019), they might have negative effects by removing neurons and neuronal connections that bear senescence features but retain a proper activity.
Based on our findings, we hypothesize cell-type-specific strategies that can oppose specific detrimental effects of cellular senescence while preserving cellular homeostasis may be safer, particularly in early drug trials.
In conclusion, our data show that cellular senescence features, including increase of p16 INK4a , develop during differentiation of human HD iPSC-derived cells to persist in human HD MSNs. Our data suggest that FOXO3 may antagonize the progression of cellular senescence in ND conditions, repressing ETS2 in human HD NSCs, which reduces the expression of p16 INK4a , in turn fine-tuning stress response. These findings provide a rationale and target, early senescence-like responses, to develop pro-resilience approaches that may be useful for early intervention in HD and other NDs.

| Analysis of the FOXO3/ß-catenin/Ryk-ICD complex
The methods used for protein co-immunoprecipitation and deletion mapping assays are described in the Appendix S1.

| Analysis of FOXO3 gene regulation
The methods used for analyzing FOXO3 targets in human HD NSCs are described in the Appendix S1.

| Analysis of cellular senescence
The methods used for testing cellular senescence are described in the Appendix S1.
When confluent, NSCs were treated with Synaptojuice A medium for 1 week followed by Synaptojuice B medium for 10 d at 37°C (Kemp et al., 2016). 25 ng/ml Activin A was added to both Synaptojuice A and Synaptojuice B media. Half media change was performed every 2 days. The resulting MSNs were characterized

| Cell transfection and cellular assays
The methods used for transfection and for testing cellular proliferation and vulnerability are described in the Appendix S1.

| Statistics
Statistics were performed using Student's t tests or two-way ANOVA. All experiments were repeated at least three times. p < .05 was considered significant. Statistics used for genomic data analysis, overlap analysis, and biological content analysis are described in the Appendix S1.

ACK N OWLED G EM ENTS
We thank Anne Brunet (Stanford University) for providing the ChIP- MV analyzed the data and helped write the manuscript. JC provided essential reagents and advice for senescence tests and edited the manuscript. LME contributed human cell differentiation protocols, designed the research, analyzed the data, and wrote the manuscript.
CN conceived and designed the research, analyzed the data, and wrote the manuscript.

DATA AVA I L A B I L I T Y S TAT E M E N T
RNA-seq and ChIP-seq data are available at GSE109873, subseries GSE109871, GSE109872, and GSE109869.

S U PP O RTI N G I N FO R M ATI O N
Additional supporting information may be found online in the Supporting Information section.