Distinct sub-cellular autophagy impairments occur independently of protein aggregation in induced neurons from patients with Huntington’s disease

Huntington’s disease (HD) is a neurodegenerative disorder caused by CAG expansions in the huntingtin (HTT) gene. Modelling HD has remained challenging, as rodent and cellular models poorly recapitulate the disease. To address this, we generated induced neurons (iNs) through direct reprogramming of human skin fibroblasts, which retain age-dependent epigenetic characteristics. HD-iNs displayed profound deficits in autophagy, characterised by reduced transport of late autophagic structures from the neurites to the soma. The neurite-specific alterations in autophagy resulted in shorter, thinner and fewer neurites presented by HD-iNs. CRISPRi-mediated silencing of HTT did not rescue this phenotype but rather resulted in additional autophagy alterations in ctrl-iNs, highlighting the importance of wild type HTT in neuronal autophagy. In summary, our work identifies a distinct subcellular autophagy impairment in aged patient derived HD-neurons and provides a new rational for future development of autophagy activation therapies.


Introduction
Huntington's disease (HD) is an autosomal dominant neurodegenerative disorder caused by an expanded polyglutamine tract within the first exon of Huntingtin (HTT) (1). Clinically, HD is characterized by involuntary movements together with cognitive impairment, psychiatric disturbances as well as metabolic and sleep problems, a result of extensive cell impairment and death within the central nervous system (CNS). Genetics and age in combination are key components of HD pathology as the length of the CAG repeat expansion in HTT correlates with age of disease onset, and manifest disease is more prevalent with increased age, independent of CAG repeat length (1-3). Most HD patients have CAG repeats in the range of 40 -45 CAGs and are diagnosed around the age of 50 (4). HTT is ubiquitously expressed, yet the presence of a mutated Huntingtin allele (mHTT) results, at least early on, in dysfunction and death of neurons specifically in the striatum and cortex (5). Mutant HTT has a propensity to aggregate and form insoluble protein inclusions, but it is still debated as to how protein aggregation influences, if at all, neuronal dysfunction and ultimately cell death. In general, the molecular and cellular basis for the pathology and the age-related disease process remains poorly understood and treatment of HD remains a major challenge.
Several studies have documented altered autophagy in neurodegenerative disorders including HD, a phenomenon thought to contribute to the failure of clearance of aggregating proteins (6)(7)(8)(9)(10)(11)(12). Autophagy is a lysosomal protein degradation pathway that is present at a basal level in all cells, including neurons and is essential for their survival (13,14). Boosting autophagy through pharmacological or genetic manipulation successfully reverses disease-associated phenotypes in various mouse models of neurodegenerative disorders, including models of HD, and is associated with a reduction of the protein aggregate burden (6,7,11,15,16). These pre-clinical findings have led to the initiation of clinical trials to activate autophagy in HD and other neurodegenerative disorders (17)(18)(19)(20). While these initial studies have shown that this approach is feasible and well tolerated, it is also evident that therapeutic approaches to activate autophagy need to be optimized and tailored for different neurodegenerative disorders. In particular, a clear understanding of exactly how and why alterations in autophagy appear in HD (and other neurodegenerative disorders) and how this contributes to neuronal dysfunction and death is currently lacking.
In this study we have used direct reprogramming of human fibroblasts to generate patientderived induced neurons (iNs) that retain age-associated epigenetic marks (21)(22)(23). When performing a combined transcriptomic, proteomic and automated microscopic analysis on iNs obtained from patients with HD (HD-iNs), we found a distinct impairment of autophagy characterized by a failure to transport late autophagic structures from neurites to the cell body.
This subcellular autophagy impairment was directly linked to a reduction in the neurite complexity of HD-iNs. The autophagy impairment in HD-neurons appeared without the presence of mHTT-aggregates, demonstrating that this phenomenon lies upstream of overt protein aggregation. Finally, inhibition of HTT-expression (both wt and mutant) using CRISPRi rescued some of the autophagy-related impairments but also resulted in additional new autophagy alterations suggesting that the disease phenotype is driven by a combination of both loss-of-function and gain-of-function mechanisms. In summary, our results provide a novel understanding of the HD disease process by demonstrating a specific subcellular autophagy impairment localised to the neurites. Our findings have clear translational implications.

Table 1. Summary of control and HD patient biopsies.
Overview of the cohort used in the study specifying the age, sex and CAG repeats of 10 healthy control and 10 HD patient fibroblasts lines.
We reprogrammed the 20 fibroblast lines to iNs using our previously described protocol (21,24). In brief, this methodology includes a single lentiviral construct that expresses the transcription factors Achaete-scute homolog 1 (Ascl1) and POU Class 3 Homeobox 2 (Pou3f2v or Brn2) with two short hairpin RNAs (shRNA) targeting RE1-silencing transcription factor (REST1) (21) (Figure 1a). Upon transduction, the fibroblasts rapidly developed a clear neuronal morphology with a reduction in the size of both the nuclei and cell body and the formation of long, elaborate neurites (Figure 1b, SFigure 1a-b). Over a time period of a few weeks, the reprogrammed fibroblasts transformed into mature iNs and started to express the neuronal markers MAP2 (neuron specific cytoskeletal protein enriched in dendrites) and TAU (a highly soluble microtubule-associated protein abundant in neurons) (Figure 1b, SFigure 1a-b). In addition to which they became electrically active as we have previously shown (21).
We analyzed the reprogramming capacity of fibroblasts derived from HD-patients in detail using a high-content automated microscopy analysis of the reprogrammed iNs. By quantifying the number and proportion of MAP2 + and TAU + cells (as defined by DAPI) we found that the HD fibroblasts converted into iNs four weeks post-transduction with similar purity (number of iNs / number of DAPI cells) and conversion efficiency (number of iNs / number of starting fibroblasts) (Figure 1c, SFigure 1c-d) as to that seen with control fibroblasts. Neuronal purity and conversion efficiency were not affected by passage number (SFigure 1e-f), and there was no difference in the rate of cell death between control and HD-iNs at four weeks, as determined through surveying the number of iNs and DAPI + cells at this stage (SFigure 1g). Together, these data demonstrate that fibroblasts obtained from HD patients can be reprogrammed to iNs with the same efficiency as that seen for healthy control individuals.

Transcriptome, proteome and epigenome profiling of iNs
To investigate molecular changes during the reprogramming process as well as molecular alterations in HD-iNs, we performed transcriptome and proteome analysis using RNA sequencing and shotgun proteomics on ctrl and HD-iNs, as well as the unconverted fibroblasts.
To obtain a pure population of iNs for these analyses (in order to reduce background transcriptional noise), we established a procedure to FACS-purify iNs at four weeks post conversion using Neural cell adhesion molecule (NCAM+), a mature neuronal cell surface marker (24) (Figure 1a).
We first analyzed the RNA-seq transcriptome data across seven ctrl and seven HD iNs and fibroblasts and found that fibroblast and iNs samples (both ctrl and HD) were clearly distinguishable ( Figure 1d, SFigure 2a, STable 1, 2). We found high-level RNA expression of numerous genes that are known to be expressed in neurons in the iNs but not in the fibroblasts confirming successful neuronal conversion. Gene ontology analysis confirmed that transcripts enriched in iNs represented cellular processes related to neuronal functions, such as synaptic signaling and regulation of membrane potential (Figure 1e). On the other hand, transcripts enriched in fibroblasts were related to cell proliferation ( Figure 1f). We also investigated the presence of transcripts related to specific neuronal subtypes in the iNs and found genes related to several different neuronal subtypes (Figure 1g), as well as an absence of neural progenitor markers both in the fibroblasts and in the iNs (Figure 1g). This is in line with previous results indicating that these types of iNs represent a mixed population of maturing neurons (21)(22)(23)25).
Next, we analyzed the shotgun proteomics data from the unconverted seven ctrl and seven HD fibroblasts as well as the resulting iNs. The proteome analysis resulted in 7,001 proteins being quantified and identified at high confidence in the majority of samples in at least one group ( Figure 1h, STable 3,4). When we compared the abundance of individual proteins, we found that fibroblast samples and iNs (both ctrl and HD) displayed a high degree of proteomic difference when compared each other (SFigure 2b), similar to what was observed in the transcriptomic analysis. In particular, proteins linked to neuronal function, such as synaptic vesicles proteins, were highly abundant in iNs, while proteins related to proliferation pathways, such as cell cycle and DNA-replication were downregulated compared to fibroblasts ( Figure   1i, SFigure 2c-e). Additionally, the metabolic reprogramming observed in iNs, involved the upregulation of pathways like glycolysis, the lysosome and phagosome, demonstrating that these cells, to a large extent, mimic the metabolic state normally found in neurons (Figure 1i-j, Several previous studies have demonstrated that iNs retain age-dependent molecular features (22,23,(26)(27)(28)(29). To confirm this in our iNs, we investigated if we could detect age-dependent epigenetic signatures in these cells. We used the Illumina Epic Methylation array to profile global DNA methylation patterns in 6 ctrl and 9 HD-iNs. A penalized regression model using a set of 353 CpGs defining the biological age by the Horvath epigenetic clock allows the prediction of the age of the donor (30). We converted ctrl and HD donor cell lines into iNs and estimated the biological age of the resulting iNs. We found that in the ctrl-iNs, the DNAm predicted biological age strongly correlated with the donor's actual real age (Pearson correlation coefficient R 2 = 0.9639, Figure 1k). A previous study performed on postmortem brain tissue indicated an increase in epigenetic aging rates in patients with HD (31). In line with this, we also found significantly increased DNAm predicted biological age in the HD-iNs compared to the ctrl-iNs (p = 0.038, Figure 1k). Taken together, these data confirm that both ctrl and HD-iNs retain epigenetic signatures consistent with aged neuronal cells and that iNs derived from patients with HD have an increased biological age.

HD-iNs display alterations in proteins linked to autophagy
To identify molecular mechanisms potentially linked to HD pathogenesis, we analyzed HD-iNs for differences in their transcriptome and proteome when compared to ctrl-iNs ( Figure 2a).
Starting with the transcriptome, we found 549 mRNA transcripts that were upregulated in HD-iNs and 373 downregulated out of 13,879 detected transcripts compared to ctrl-iNs, confirming previous findings that mHTT induces major transcriptional alterations (Figure 2b, STable 5, 6).
However, gene ontology and network analysis failed to identify any molecular or biological processes that were significantly enriched in the differentially expressed genes ( STable 7,8).
This suggests that while HD-iNs display transcriptome alterations, these alterations are not linked to distinct gene programs, making it difficult to link transcriptomic changes in HD-iNs to phenotypical alterations.
We next turned our attention to the the differences between the proteomes of ctrl-iNs and HD-iNs and found that 273 proteins were upregulated in HD-iNs while 137 proteins were Interestingly, when we performed gene ontology and network analysis of significantly dysregulated proteins in HD-iNs, we found that many of these proteins were functionally linked (STable 12, 13). Downregulated proteins were enriched for cellular pathways such as the CAMKK-AMPK-signaling cascade as well as autophagy related processes, while upregulated proteins were connected to ribosomal functions ( Figure 2e, STable 12,13). We also performed the same analysis in ctrl and HD fibroblasts and found that ribosomal proteins were also upregulated in the HD-fibroblasts, suggesting that translational alterations may be a ubiquitous downstream consequence of the presence of mHTT (STable 14) (32). On the other hand, the proteins related to CAMKK-AMPK-signaling and autophagy were only downregulated in HD-iNs and not in the corresponding HD-fibroblasts, indicating that these proteomic alterations are linked to neuron-specific cellular functions (SFigure 3, STable 15). We thus focused our further analyses on these neuron-specific proteome alterations.
In HD-iNs, several kinases in the CAMKK-AMPK pathway were downregulated, including CAMKK2, CAMK2G, AMPK and IRS1 as well as the autophagy regulator BECN1 ( Figure   2f). Moreover, suppressors of the AMPK pathway, PPP2R5E and PPP2R1B phosphatases were significantly upregulated in HD-iNs compared to healthy controls ( Figure 2f). Taken together, this omics-based analysis demonstrate that HD-iNs display an altered proteome with links to alterations in autophagy. Noteworthy, these alterations were cell-type specific, only present in the iNs but not in the fibroblasts and mainly due to post-transcriptional mechanisms that could not be detected by transcriptome analysis.

Subcellular alterations in autophagy in HD-iNs
One of the downregulated proteins detected in the proteomic analysis was BECN1, an autophagic regulator protein that plays a key role in autophagosome formation. Several studies support the importance of BECN1 in HD pathology, as overexpression can slow the progression of HD pathology in both cell and mouse models by inducing autophagy, while the expression of BECN1 in the brains of HD patients declines with age (6,7,33,34). The downregulation of BECN1, as well as the other alterations in the CAMKK-AMPK-signaling pathway, suggests that autophagy activity may be impaired in HD-iNs.
To investigate this in more detail, we first verified that in HD-iNs (but not the parental fibroblasts) there was a significant reduction of BECN1 levels using western blot (WB) analysis ( Figure 3a, SFigure 4a). We then assessed autophagy activity in HD-iNs compared to ctrl-iNs by measuring microtubule-associated protein 1A/1B-light chain 3B (LC3B). LC3B conjugates from LC3B-I to LC3B-II during autophagosome formation. We found a reduction in the LC3B conjugation, as determined by assessing the ratio of LC3B-II over LC3B-I using WB in HD-iNs ( Figure 3b) which was coupled to an increase in total LC3B-II levels, suggesting more autophagosomes in HD-iNs ( Figure 3b). We next measured p62 levels, which inversely reflects autophagolysosome degradation (35) and found a non-significant trend for increased levels (SFigure 4b). We also measured the levels of LAMP1, which is present on endosomes and lysosomes including autophagolysosomes and autolysosomes but did not detected significant difference in HD-iNs (SFigure 4b) (36). In summary, there were alterations in basal autophagy in HD-iNs, primarily reflected by alterations in BECN1 and LC3B, in line with an increase in autophagosomes and most likely a reduction in autophagic flux.
In neurons, autophagosomes are formed in the neurites and then transported to the cell body where the active lysosomes are present (37). Moreover, degradative lysosomes in the soma can also be transported to target autophagosomes in the distal axons anterogradely in mature neurons (38). To characterise autophagy alterations at a subcellular level, we performed immunocytochemistry (ICC) analysis of LC3B, p62 and LAMP1 as well as an unbiased quantification of autophagosomes including the subcellular localization using high-content automated microscopy. This analysis revealed an increased number and size of LC3B puncta in HD-iNs, that was particularly apparent in the neurites of these cells (Figure 3c We used immunohistochemistry (IHC) to verify the neurite-specific impairment of basal autophagy in HD neurons using human post-mortem brain tissue (Table 2). We found no evidence of accumulation of p62 positive dots in the neurites of healthy controls identified by co-labelling with a Neurofilament specific antibody (Figure 3i). In contrast, we found clear p62 accumulation in the neurites in all HD patients analysed regardless of disease stage (Figure 3i).

Impaired autophagic flux in HD-iNs
We next focused on analysing autophagic flux in HD-iNs by modulating the autophagy pathway using pharmacological agents ( Figure 4a). Treatment with Bafilomycin A1 (Baf), a late-stage inhibitor of autophagy that blocks autophagosome-lysosome fusion, resulted in the expected increase in size of autophagosomes as visualized by LC3B-puncta, in the cell body of both ctrl and HD-iNs (SFigure 5a-c). However, the number of LC3B dots increased significantly in the ctrl-iNs, but not in the HD-iNs. Furthermore, there was an increase in p62-puncta count in both the cell body and neurites in HD-iNs but not in the ctrl-iNs (SFigure 5d-f). Thus, blocking autophagolysosomal formation in HD-iNs resulted in a further reduction in autophagy activity, suggesting that degradation of these structures occurs at a reduced rate in HD-iNs. In line with this observation, we found that the accumulation of autophagolysosomes in HD-iN neurites, as visualised by LAMP1-puncta, was completely abolished upon Baf-treatment (Figure 4b-d).
Thus, when the formation of new autophagolysosomes is prevented in HD-iNs the cells are capable of dealing with the accumulation of these structures in the neurites. We further corroborated these results by treatment with Wortmannin (W), which blocks the initiation of autophagy by inhibiting phosphatidylinositol 3-kinase (PI3K) (Figure 4a) (36), which showed that there was a robust reduction of LAMP1-puncta in the neurites of HD-iNs with this treatment (Figure 4b-d).
To further understand the autophagy impairment in HD-iNs we next used rapamycin (RAP) that activates autophagy at an early stage by inhibiting mTOR signalling ( Figure 4a). In ctrl-iNs, RAP treatment resulted in a clear reduction in LC3B-puncta in both the cell body and neurites, in line with the increased autophagic flux mediated by the treatment (Figure 4e Taken as a whole, these results demonstrate that HD-iNs show impairment in degrading autophagolysosomes. It appears that the cellular machinery is working at reduced rate and cannot degrade autophagy cargo, resulting in an accumulation in late stage autophagic structures. The reason for this impairment is likely to relate to the late autophagic structures getting stuck in the neurites and failing to be transported to the cell body where they should be degraded. This impairment in the last step of autophagy results in an overall reduction in autophagy activity. These observations are important from a therapeutic point of view as treatment paradigms to restore autophagy alterations in HD should aim to enhance autophagolysosome transfer and degradation rather than activating autophagy at an early stage, which could actually worsen the pathology.

Cellular mechanisms underlying the autophagy impairments found in HD-iNs
We next investigated the molecular mechanisms underlying the autophagy impairment in HD-iNs. It has been suggested that protein aggregates are the key driver of the autophagy phenotype observed in neurodegenerative disorders. For example, long-term exposure to protein aggregates could eventually exhaust the autophagy machinery (39,40). On the contrary, HTT has also been suggested to be directly linked to the cellular signalling pathway that controls autophagic activity (41,42). HTT has been reported to directly bind to BECN1 via its polyQtract and modulation of this binding, either by loss of wtHTT levels or by the presence of an expanded polyQ-tract in mHTT, results in a reduction in BECN1 levels and an overall reduction in autophagic activity (7,34). wtHTT has also been reported to directly interact with p62 to facilitate cargo engulfment in autophagy, indicating that the loss-of-function of one wild-type allele of HTT in HD may impair autophagy (42).
To investigate this further, we first looked for the presence of mHTT-aggregates in HD-iNs.
Through the use of WB analysis with several different lysis conditions, we did not detect the presence of any mHTT-containing aggregates (SFigure 6a, b) even though the expression of HTT-mRNA in fibroblasts and iNs was similar in both groups (SFigure 6c). Together, these experiments demonstrate that the autophagy impairments in HD-iNs are present without evidence for overt HTT aggregation.
We next investigated the direct role of HTT in the regulation of autophagy in iNs. Previous studies showed that silencing HTT blocks retrograde transport of late autophagosomes, while depletion of the mHTT results in accumulation of late autophagic structures with undegraded cargo (9,10). Moreover, HTT is also involved in lysosomal transport (43,44). Since both wildtype and mutant HTT have been implicated in the regulation of autophagy we decided to investigate the consequence of transcriptional silencing of wtHTT/mHTT on the autophagy pathway in both ctrl-iNs and HD-iNs (9). To this end we established a lentiviral based CRISPR inhibition (CRISPRi) approach to silence HTT-expression (Figure 5a). The CRISPRi-vector expressed a dead Cas9-KRAB fusion protein that was linked to a GFP reporter as well as a guide RNA (gRNA) targeted to the area around the HTT transcription start site (Figure 5a). This vector design allows for the binding of dCas9-KRAB to the HTT loci, thereby resulting in the establishment of local heterochromatin and subsequent transcriptional silencing. We optimized the vector construct by testing different gRNAs and MOIs in HEK293T cells and human induced pluripotent stem cells (iPSCs) and ultimately found two different gRNAs, targeted to a region just downstream of the HTT transcription start site (TSS), that very efficiently silenced both alleles of HTT (SFigure 7a).
We transduced ctrl and HD-fibroblasts with the CRISPRi-HTT vector and FACS purified GFP expressing cells (Figure 5a). This resulted in efficient silencing of both alleles of HTT in the patient-derived fibroblasts as quantified with qRT-PCR (SFigure 7b). We then proceeded to generate iNs from the CRISPRi-HTT silenced fibroblasts (Figure 5a). After four weeks of conversion, we confirmed that HTT remained silenced in the iNs after conversion and that CRISPRi-HTT treatment did not impact on reprogramming efficacy (Figure 5b, SFigure 7c-f).
The resulting HTT-silenced HD-iNs and ctrl-iNs were then analyzed using ICC for LC3B, p62 and LAMP1 spots in the cell body and in the neurites.
We focused first on silencing of HTT in the ctrl-iNs. Previous studies have demonstrated that HTT has an essential function in autophagy, as it contains an autophagy-inducing domain and it also facilitates axonal trafficking of autophagosomes (9,10). Moreover, HTT functions as a scaffold in autophagy where it physically interacts with p62 and depletion of HTT reduces the association of p62 with LC3B and other substrates of autophagy (42). When silencing HTT in ctrl-iNs we found that while LC3B dot number count or area were not affected, the number of p62 positive puncta significantly increased in the neurites of ctrl-iNs, confirming its role in regulating autophagy or other mechanisms related to p62 degradation (SFigure 8a, b). Notably, the number and area of LAMP1 puncta significantly decreased but only in the neurites of HTT silenced ctrl-iNs (SFigure 8c). Thus, silencing of HTT in the ctrl-iNs resulted in the alteration of autophagic activity characterised by increased p62 accumulation and reduction in the endolysosomal marker LAMP1. These findings are in line with previous studies demonstrating that HTT facilitates cargo recognition by modulating the assembly of the cargo receptors and autophagy proteins. Moreover, these findings highlight that silencing HTT in the ctrl-iNs results in a different autophagy impairment to that found in HD-iNs.
Next, we focused on the effect of silencing HTT in HD-iNs on autophagy. As described above it is important to highlight that CRISPRi experiments resulted in a highly efficient silencing of both healthy and mHTT alleles in the HD-iNs (Figure 5b). Moreover, as described above, HD-iNs display a neurite specific late-stage autophagy alteration with increased LC3B, p62,

The autophagy impairment in HD-iNs results in reduction in neurite complexity
We finally explored the cellular consequences of the impaired autophagy in HD-iNs. It is well established that HD neurons tend to display alterations in neurite arborization and complexity, and these impairments are thought to contribute to the early disease process (26,(45)(46)(47).
Importantly, autophagy has been directly linked to neurite formation, since inhibition of this degradation pathway reduces neurite growth and branching complexity (37,48). To investigate whether HD-iNs have an altered neurite morphology and if this is linked to the autophagy impairments found in the cells, we performed a detailed analysis of neural morphology of the reprogrammed cells using high-content automated microscopy ( Figure 6a, SFigure 9a, b). After four weeks of conversion, we found a significant decrease in neurite complexity in HD-iNs as measured by total neurite area, the number of neurites per cell, neurite length and neurite width (Figure 6b, c, SFigure 9c-f). This phenotype was not a consequence of a slower maturation of HD-iNs, since we observed a similar reduction in neurite number, length and complexity after seven weeks of conversion (SFigure 9g-m). Also, the cell body size was similar in HD-iNs and ctrl-iNs (Figure 6c, SFigure 9m). At a molecular level we found that many proteins that were downregulated in HD-iNs were connected to the microtubule system, which plays a fundamental role in the maintenance of axonal homeostasis by preserving axonal morphology and providing tracks for protein and organelle transport. A significant reduction was seen in proteins belonging to the tubulin protein superfamily, such as TUBGCP2, TUBA1C, TUBAL3, which are all involved in neuronal microtubule migration, axonal assembly and neurodegeneration (Figure 6d, e). Notably, these alterations in the microtubule system were dysregulated at a posttranscriptional level as the tubulin superfamily protein members were not different at the RNA level (SFigure 9n).
Autophagosomes form at the axon terminal and fuse with lysosomes during a dynein-mediated transport to the soma. Moreover, lysosome transport is also mediated via microtubules in the neurites (49). To investigate a direct link between the reduced neurite morphology in HD-iNs and autophagy, we analyzed neuronal morphology after inhibition or activation of autophagy using Baf or W and starvation, respectively. We found a significant reduction in the ctrl-iNs neurite area and length when inhibiting autophagy using Baf or W (Figure 7f). In contrast, HD-iNs did not exhibit any further reduction in neurite area or length after autophagy suppression ( Figure 6f). These data suggest that while ctrl-iNs neurite morphology is affected by autophagy impairment using different pharmacological agents, HD-iNs do not show any further morphological changes. likely due to an already existing autophagolysosomal transport failure.
Next, we used amino acid free starvation to activate autophagy in the iNs. In response to starvation, cells recover nutrients through autophagy by increased AMPK activation and increased mTOR inhibition. This short-term autophagy activation through starvation did not have any major effect on the neuronal morphology of the ctrl-iNs since neurite area and length were not affected (Figure 6g). On the other hand, the neuronal morphology of HD-iNs was significantly affected, neurite area and length significantly decreased after starvation ( Figure   6g), suggesting that HD-iNs could not cope even with this short-term starvation activation of autophagy. Lastly, we analyzed the effect of CRISPRi editing on the neurite morphology after silencing of HTT expression in ctrl-iNs and HD-iNs. CRISPRi silencing did not rescue the reduced neurite area nor neurite length in the HD-iNs (Figure 6h). HD-iNs were significantly shorter and smaller even after silencing both HTT alleles in the HD-iNs compared to the ctrl-iNs (Figure 6h). Together, these results suggest that the abnormal neuronal morphology present in the HD-iNs is directly linked to impairments in autophagy.

Discussion
The pathogenic processes underlying HD have been difficult to elucidate, in part due to the fact that age-dependent human neurodegenerative disorders are challenging to study. Postmortem material is limited, both in availability and experimental possibilities and provides only a static snapshot of the consequence of disease. Animal models poorly replicate the disease process, as their lifespan is shorter in comparison to humans. This has led to the use of transgenic mHTTalleles with very long CAG repeats (sometimes >100 CAGs), where the pathology is accelerated and thus possible to study in mice. However, this many repeats are rarely, if ever, seen in clinical practice. In case they are seen, they are associated with the rare juvenile form of the disease, in which the disease process may be drastically different from HD associated with more typical CAG repeat lengths (50). These challenges to model HD have contributed in part to the lack of effective treatments and it is therefore critical to establish model systems that recapitulate the human disease progression, including age-dependent processes.
Recent advances in cellular reprogramming have allowed for the establishment of induced pluripotent stem cells (iPSCs) that can be efficiently differentiated into neurons, making it possible to obtain human patient-derived HD-neurons (51)(52)(53)(54) with the potential to generate isogenic control lines. However, a drawback with iPSCs is that during the reprogramming process epigenetic marks associated with ageing are erased, thereby transforming them to a juvenile state (55). Thus, the study of iPSC-derived neurons is limited to young cells, which is suboptimal since age is a key determinant of HD pathology (2, 3). As a consequence, most HD-iPSCs studies with well documented phenotypes are of limited utility (51)(52)(53)(54). As an alternative to iPSCs, we and others have recently developed direct lineage reprogramming (25,56). By overexpressing and knocking-down key transcription factors it is possible to reprogram human fibroblasts directly into neurons, without going through a juvenile state. This approach allows for the generation of patient-derived neurons that retain age-associated epigenetic marks (21)(22)(23).
In this study we have used direct reprogramming of patient-derived fibroblasts into iNs to study disease mechanisms in HD. As mentioned above, this approach has many advantages for the study of neurodegenerative disorders as it allows for easy access of patient-derived neurons that retains ageing-related epigenetic marks of the donor and which can be studied in detail using a wide panel of molecular and biochemical technologies. Using this approach, we were able to detect clear disease-related phenotypes when studying iNs from individuals with CAG repeats in the range normally seen in clinic in patients (4).
In HD-iNs we found that there is a post transcriptionally altered cell-type specific autophagy impairment and sought to understand the mechanism by which this comes about. Several studies have demonstrated that the presence of mHTT interrupts autophagy, contributing to the impaired clearance of aggregated proteins (6-10, 57). In various models of HD, different kinds of impairments in autophagy have been described including an increased number of autophagosomes (which sometimes appear empty), disrupted vesicle trafficking and impaired autophagosome-lysosome fusion and dynamics (7,8,10,58). It is also not clear if impaired autophagy directly contributes to the buildup of protein aggregates or if the aggregates themselves influence the activity of autophagy (7-10, 42, 59-63). It has been speculated that defects in the autophagic machinery can lead to a negative feedback loop, whereby mHTT aggregation leads to a further dysregulation of autophagy causing increased mHTT accumulation and neurotoxicity (7,(63)(64)(65). Thus, while there are numerous experimental reports on autophagy impairments in HD, it remains unclear which of these are specific to the model system and which are relevant to the actual disease (7-10, 42, 59-63). This is important given its therapeutic implications and the fact that trials are now starting to appear in the clinic looking at autophagy enhancing agents. In HD-iNs, we found a subcellular, neurite specific autophagy impairment, with an accumulation of LAMP1-positive late autophagic structures.
We also show that this is a consequence of an impaired transport of these structures to the cell soma where they should be degraded. This finding provides an answer as to why neurons are particularly vulnerable in HD and represent a novel therapeutic target -restoration of autophagolysosome transfer to the cell soma.
The underlying molecular mechanism for the autophagy impairment in HD-iNs appear to be linked to the AMPK pathway, since several factors in this pathway were dysregulated. AMPK is a key energy sensor that promotes catabolic pathways while shutting down ATP consuming processes required for cell growth (66)(67)(68). AMPK inhibits cell growth by inhibiting mTORC signaling and protein synthesis downstream of mTORC1. Energy impairments such as decreased mitochondrial biogenesis and trafficking, oxidative stress, increased apoptosis, and ATP deficit all have been implicated in HD pathogenesis (49). Neurons are energetically demanding cells and thus highly vulnerable to abnormalities in cellular respiration. Our findings point towards boosting autophagy by specifically targeting the AMPK pathway. In line with this, we and others have also shown that BECN1 overexpression can rescue some aspects of HD pathology in various models (6,7,33,34). Moreover, genetic and pharmacological activation of AMPK has been shown to protect dysfunctional and vulnerable neurons in HD in nematode, cellular and mouse models (69,70). An impairment of autophagy in neurons will have multiple pathological consequences (13,14). Autophagy is implicated in neurogenesis, synaptogenesis, the control of post-transcriptional networks and protein aggregation (6,(71)(72)(73).
Thus, impairment of autophagy could underlie many of the early cellular disease phenotypes observed in HD (74,75). As such, the development of specific autophagy-boosting therapies is promising as they have the potential to directly restore other dysfunctional intracellular processes.
Since HD-iNs retain ageing epigenetics characteristics, our results indicate that autophagy impairments in HD may be a combination of age-related epigenetic alterations and mHTT-mediated post transcriptional processes. Exactly how the presence of a mHTT-allele results in a reduction in the transport of autophagolysosomes from neurites remains unknown, but a combination of an age-related alteration in autophagy-control together with a direct mHTTmediated protein-protein interaction appears the most likely scenario. For example, mHTT has previously been found to directly interact and destabilize BECN1, which is in line with the reduction of BECN1 protein that we found in HD-iNs (6,7,33,34,57,76). How ageing and the epigenetic alterations influences the disease pathology and autophagy impairments is currently unknown but will be interesting to investigate in order to find mechanistic links between these phenomena.
Our study also has direct implications for the development of therapies working on mHTTsilencing. Such therapies are considered a very promising possibility to successfully treat HDpatients and clinical trials are already underway (17,77,78). Our results suggest that the development of allele-specific silencing of mHTT may be key to the success of such therapies given that wtHTT is directly involved in the control of cellular pathways controlling protein degradation. Thus, while the silencing of mHTT will certainly have beneficial consequences, as demonstrated in our study by efficiently lowering LAMP1 in the neurites, the silencing of wtHTT will also come with loss-of-function consequences on similar cellular pathways.
In summary, we have developed a novel cell-based model of HD that allows for the study of aged patient-derived neurons. We found that HD-iNs display distinct autophagy alterations, characterized by a blockage in autophagolysosome transfer and degradation. Our results thus identify a novel therapeutic target through autophagy while also helps to advocate for the development of allele specific silencing-based HD therapies.

Human Tissue
Post-mortem human brain tissue was obtained from the Cambridge Brain Bank (Cambridge, UK) and used under local ethics approval (REC 01/177). Severity of HD was graded by a certified pathologist according to the Vonsattel grading system (79) ( Table 2).

Figure 3i
Brainbank ID Age of death

Cell Culture
Adult dermal fibroblasts were obtained from the Huntington's disease clinic at the John van Geest Centre for Brain Repair (Cambridge, UK) and used under local ethical approval (REC 09/H0311/88). The cells were obtained from 10 HD and 10 healthy individuals (Table 1), for more information on the biopsy sampling see (21). CAG repeat length was defined for both alleles using Sanger Sequencing (Laragen Sanger Sequencing Services). The fibroblasts were kept in DMEM Glutamax medium (Gibco) supplemented with 10% FBS (Gibco) and 1% penicillin/streptomycin (Gibco) and passaged when they reached 80-90% confluency using a previously described procedure (21).

Lentiviral production
Third-generation lentiviral vectors were produced as previously described (6). Virus titration was performed, and the titer was determined with qRT-PCR as previously described (6). The virus titers ranged between 2.33E+08 and 9.3E+09. A MOI of 1-20 was used from different lentiviral vectors as specified for each case.

Neural conversion
Prior to the start of conversion, Nunc Delta surface treated plates (Thermo Scientific) were coated as previously described (24). Fibroblasts were plated at a density of 50,000 cells per The guides were inserted into the backbone using annealed oligos and the BsmBI cloning site.
Three guides were designed and tested in HEK293T and iPS cells. HEK293T cells were cultured similarly to the fibroblasts cells as described above. iPS cells (RBRC-HPS0328, 606A1 from RIKEN) were cultured as previously described (82). HEK293T and iPS cells were Millipore; Rapamycin, 20 nM, Sigma-Aldrich; Wortmannin, 100 nM, Sigma-Aldrich) was added to the well followed by fixation for ICC after four hours. Non-treated wells received fresh media with DMSO in equivalent amount to that used in treated cells.
Cells were starved by replacing the media with Hank's Balanced Salt Solution (Thermo Fisher, 14025092) for two hours before fixation.

Immunostaining
Immunocytochemistry to stain iNs was performed as previously described (21). Briefly, the cells were fixed with 4% paraformaldehyde for 10-15 minutes. Following fixation, the paraformaldehyde was aspirated, and the cells were washed carefully twice with DPBS.  Table 3). The cells were washed twice with DPBS and the secondary antibody conjugated to a fluorophore ( Immunohistochemistry staining was performed as described before (6). 10 μm thick paraffinembedded striatal sections were taken from 3 differently graded HD patients and healthy age-matched control brains. Sections were stained using the antibodies listed in Table 3

High-content automated microscopy
The Cellomics Array Scan (VT1 HCS Reader, Thermo Fischer) was used for high-content automated microscopy.
To quantify the number of DAPI + , MAP2 + , and TAU + cells and define neuronal purity and conversion efficiency "Target Activation" (TA) was used. Using this method, we obtained objective, unbiased measurements of the iN cultures. The TA program was used to acquire images of 100-289 fields using a 10x objective of each well to define cell number, neural purity and conversion efficiency. Wells with <50 valid fields were excluded from further analysis. Neuronal Profiling with spot detection (using a 20x objective) was used to determine average LC3B, LAMP1 and p62 dot number and size per cell within MAP2 + or TAU + cell bodies and neurites. LAMP1 and LC3B colocalization was also analyzed by defining the overlapping area as a percentage between the two markers. First valid nuclei were defined by DAPI staining based on intensity and area. Border objects were excluded from further analysis. Next, cell bodies and neurites were defined based on intensity and area of MAP2 + or TAU + as the region of interest. Border objects were excluded from further analysis here also. Autophagy markers were analyzed and defined by intensity and area within the valid neuronal cells. In every case, 150-250 fields were analyzed and wells <50 valid fields were excluded from further analysis.

FACS
To increase the purity of converted control and HD-iNs for RNA-sequencing, the cells were harvested and sorted by FACS as previously described (24). To this end, the cells were The lysed cells were collected in a microcentrifuge tube and incubated on ice for a minimum of 30 minutes, followed by centrifugation at 10,000 g for 10 minutes in 4 °C to pellet cellular debris. Following centrifugation, the supernatant was transferred to new vials. The protein content was quantified with Bradford DC TM Protein Assay (Biorad) and 10-15 µg protein of each sample was used for loading the gel. Gel electrophoresis and blotting was performed as previously described (7). Both the primary and secondary antibodies (   14). The GO analysis between iNs and fibroblasts was performed using the up and downregulated genes found to be significantly different using DESeq2. Genes with basemean more than 10 were used as the background set for the overrepresentation test, and only significant terms are shown (padj <0.05, log2FC >1).
Using the normalized reads (mean of ratios calculated with DESeq2), we tested for difference in expression between HD and ctrl-iNs using unpaired t-test. We defined significantly different genes those with p-value<0.05. Code for tests and visualization is available at Github (https://github.com/ra7555ga-s/iN_HD) (86). Gene ontology overrepresentation tests comparing HD and ctrl-iNs using the RNA sequencing data were performed using PANTHER (version 14) using Fisher's exact test and Benjamini-Hochberg correction to calculate false discovery rates (87). All genes with some expression in any of the conditions were used as background sets for these tests.

Shotgun proteomic analysis
Samples C1-C7 and HD1-2, HD5-7, and HD9-10 were used for mass spectrometry (MS) analysis. Fibroblasts (500,000 cells) and iNs converted in T75 flasks (600,000 fibroblasts plated for conversion per sample) were dissociated as previously described and prepared for quantitative proteomic analysis as follows. The cells were carefully washed off and collected in a tube with either trypsin or accutase and spun at 400 x g for 5 minutes. The supernatant was discarded, and the pellets were washed three times with DPBS. After the final wash, the supernatant was aspirated, and the pellets were frozen on dry ice and stored at -80 °C until use.  (Figure 2e), which was performed with STRING version 11 (89).

Statistical analysis
In iN conversion efficiency and purity analysis each dot represents either one ctrl or a HD adult Correlation analysis were tested with Pearson's correlation coefficient. Correlation between the predicted age (based on Horvath clock) and real age of the donors was tested using Pearson's correlation coefficient. Two-tailed unpaired T-tests or Paired t-tests were used to test differences between two groups. One-way ANOVA or nonparametric Kruskal-Wallis test was used depending on whether the data obeyed a normal distribution as defined by the D'Agostino-Pearson omnibus normality test to test differences between more than two groups. Two-way ANOVA corrected for multiple comparisons using Sidak statistical hypothesis testing was used (*p<0.05; two-tailed unpaired T-tests were used) All data are shown as min/max box plots.   (*p<0.05; Two-tailed paired t-tests were used.) All data are shown as mean ± SEM. Fold changes are presented in all graphs. Scale bar is 25 μm.