Introduction

Huntington’s disease (HD), with a worldwide prevalence of 5–10 cases per 100,000, is one of nine polyQ related disorders. These include six spinocerebellar ataxias (SCA): types 1, 2, 3 (a.k.a. Machado-Joseph disease/MJD/SCA3), 6, 7 & 17; dentatorubral pallidoluysian atrophy (DRPLA); spinal and bulbar muscular atrophy (SBMA), and HD [1]. Except for the X-linked SBMA, all polyQ diseases are autosomal and dominant. Their common characteristic is the expanded cytosine-adenine-guanine (CAG) repeats inside the coding regions of nine different genes. In HD, the affected gene (and protein) is Huntingtin (HTT), located in chromosome 4p16.3 [2]. The normal function of the HTT protein is not entirely understood, but it was shown to participate in a large variety of cellular processes including endocytosis, trafficking, autophagy, and transcription [3]. In the normal population, the repeats are polymorphic, ranging from 6 to 35 repeats. Between 36 and 38 repeats penetrance is partial, and those with 39 repeats and above have a fully penetrated mutation (100%), with a strong reverse correlation between repeat length and age-of-onset [4]. Symptoms begin with motor impairments, including involuntary movements, abnormal eye movements, impaired gait, speech, and swallowing problems. Eventually, patients deteriorate to cognitive and psychiatric impairments, including learning and memory difficulties, insomnia, depression, and more. Inhibitory (GABAergic) medium spiny neurons (MSNs) are the most susceptible to the mutant HTT, primarily in the striatum and later also in cortical neurons [5]. Expanded CAG repeats are unstable, and once they reach a certain threshold, they are prone to expand further in different tissues [6, 7] and in successive generations. Consequently, an earlier age of onset and/or a more severe disease course may exhibit through intergenerational transmission [8].

The expanded CAG repeats encode an aberrantly long polyQ tract in the corresponding protein products, leading to their misfolding; these proteins are thus prone to aggregation, eventually leading to the degeneration of specific neuronal subpopulations in specific regions of the nervous system [9]. The abnormal stretch of glutamines in the mutated proteins in polyQ diseases causes the mutant proteins to stick and form polyQ Inclusion Bodies (IBs). However, the relationship between these IBs and the disease pathology is still under debate [10]. It was originally suggested and widely accepted that IB formation is detrimental and correlates with pathology [11,12,13]. It was shown, for example, that IBs can function as a sticky trap that sequesters not only the faulty proteins containing stretches of polyQ but also other healthy proteins, especially those containing short polyQ stretches [14]. Several essential proteins in the cells, including some transcription factors (TFs), have short polyQ stretches in the normal protein [15]. Thus, these polyQ IBs potentially affect many proteins and can disrupt the precise mechanisms within the nerve cells [16]. However, polyQ IBs were shown to be protective in primary cultures of rat striatal neurons infected with the first exon of the mutant HTT gene. Through specialized automated microscopy, the authors revealed that the diffuse form of mutant HTT is toxic, and that the formation of IBs serves as a protective response, ultimately reducing neuronal death [17].

To test this in a human neuronal system, we developed human iPSC-derived neuronal cells enabling to compare genetically identical IB+ and IB- cells. We identified several inflammatory-related cytokines expressed exclusively in the IB+ cells by ATF3. Live imaging of Tunicamycin-treated IB+/− cells demonstrate that polyQ IBs confer significant protection against death, and ATF3 knockout cells did not form polyQ IB and displayed increased sensitivity. Taken together, we uncover a role for ATF3 in polyQ IB formation in human cells, show that polyQ IBs are protective against cellular challenges in human neurons, and suggest that the ATF3 pathway may be targeted for potential therapeutic applications in HD and other IB-related diseases.

Materials and methods

Plasmid constructs and lentivirus generation

Constructs containing the first 21 exons of HTT with different lengths of CAG repeats (18 or 105), in-frame with GFP (kindly provided by the Tamar Juven-Gershon lab), were amplified by PCR and ligated into the viral vector N174 using EcoRI and NotI restriction enzymes. For standard virus generation, 50–70% cell confluence of HEK293 cells were transfected with lentiviral packaging vectors; pMD2.G (Addgene #12259), psPAX2 (Addgene #12260), vector of interest in a ratio of 1:3:4, respectively, following the Mirus reagent (transIT-LT1) protocol. 65 h post transfection, the medium containing the viruses was collected and filtered using 0.45 µm filters. Filtered supernatant was centrifuged in a swing bucket rotor (Beckman Coulter) at 24,000 rpm for 2 h at 4 °C. The pellet was quickly dried before being reconstituted in cold hESC growth medium (1 ml) and frozen in aliquots at −80 °C. Virus titres were determined as previously described [18]. The target cells will then be infected with 8 g/ml Polybrene.

Cell lines and genotyping

The 180Q isogenic system: 180CAG and 96-ex (180CAG-CRISPR-corrected) were obtained from the Pouladi lab [19]. The 96-ex (180CAG-corrected) and HEK293 cell lines were infected with GFP-HTT (21exons)-105Q construct for ectopic expression of polyQ proteins. 180CAG and HEK293 cells were infected with GFP-HTT (21exons)-18Q to monitor endogenous IB formation. Knockout of Atf3 was performed using CRISPR/Cas9 using two gRNA as two separate oligos (provided by IDT):

5’-CACCGGGTGTCCATCACAAAAGCCG-3’; 5’-CACCGAAAGTGCCGAAACAAGAAGA-3’. The oligos were annealed in a thermocycler by mixing 100 μM of each oligo with 10X T4 Ligation Buffer (NEB) and T4 PNK (NEB) for a total 10 μl reaction. Thermocycler parameters were 37 °C for 30 min, followed by 95 °C 5 min and then ramp down to 25 °C at 5 °C/min. The annealed gRNAs targeting Atf3 were then diluted 1:200 and cloned into pU6-(BbsI) CBh-Cas9-T2A-mCherry (Addgene #64324), which was digested using BbsI (NEB) in 37 °C for 1 h. T4 Ligase (NEB) was used for ligations. The final products were then transformed into DH5α competent bacterial cells. Transfection was performed using the Mirus reagent (transIT-LT1) protocol with two different pairs of gRNAs, targeting different sites of the genes. NPCs were plated 16–24 h before transfection, and 48 h after transfection mCherry-positive NPCs were sorted and collected as single cells into a 96-wells plate using FACS.

Cell culture

hiPSCs were cultured on mitomycin C-treated mouse embryonic fibroblast (MEF) feeder layer in hESC media - 85% KnockOut Dulbecco’s modified Eagle’s medium (KO-DMEM, Gibco-Invitrogen) supplemented with 15% KnockOut SR (a serum-free formulation) (Gibco-Invitrogen), 1 mM glutamine, 0.1 mM β-mercaptoethanol (Sigma-Aldrich), 1% nonessential amino acids stock (Gibco-Invitrogen), penicillin (50 units/ml), streptomycin (50 µg/ml) and 4 ng/ml basic fibroblast growth factor (bFGF) (Peprotech). Cells were passaged using dispase 1 mg/mL (STEMCELL technologies). NPCs were cultured on Matrigel with neural induction media (DMEM HAM’s F12, with 50 μg/ml penicillin-streptomycin, 2 mM L-glutamine, 1:100 of N2 (Invitrogen: 17502-048), 1:50 of B27 (Invitrogen: 12587-010) supplemented with 40 ng/ml 15 bFGF (Peprotech: (AF)100-18B), 40 ng/ml hEGF (Peprotech: (AF)100-15), and 7500 units hLIF (Sigma-Aldrich: L5283). Cells were passaged using accutase.

hiPSCs differentiation to NPC

hiPSC colonies were separated using accutase and passed through 70 μm cell strainer followed by 2 passages of 25 min to get rid of MEF cells, and then seeded on Matrigel as single cells in hESC conditioned media, which was replaced the next day with human neural induction media (DMEM HAM’s F12, with 50 μg/ml penicillin-streptomycin, 2 mM L-glutamine, 1:100 of N2 (Invitrogen: 17502-048), 1:50 of B27 (Invitrogen: 12587-010)), supplemented with 2 SMADi (20 μM SB, 100 nM LDN) and 1 μM human recombinant DKK1 (or its equivalent XAV-939) for 10 days. From day 10 to 20 of differentiation, cells were incubated with neural induction media supplemented with 50 nM SAG and 1 μM XAV-939, inducing the cells to become NPCs. NPCs were sorted according to a previously published protocol (Mallon et al., 2014). At about 80% cell confluence, NPCs were separated using accutase and passed through 70 μm cell strainer. Tubes were centrifuged (14 RPM for five minutes), and the pellet was re-suspended with 100 µl 10% fetal bovine serum (FBS). 2 µl conjugated anti-NCAM antibody (1:100) was added to 1 million cells and tubes were incubated in 4 °C in the dark for ten minutes. Later, 1 ml 10% FBS was added to each well, tubes were centrifuged (14 RPM for five minutes), and the pellet was re-suspended with 1 ml 2% FBS and kept on ice. Cells were counted with a cell counter prior to being analyzed in FACS (BD FACS Aria-III) in order to avoid a ratio greater than 5*10^6 cells per mL of 2% FBS.

NPC to neurons

NPCs were plated in 12-well plates at a density of 5000–10,000 cells per well and immediately cultured in differentiation media. The differentiation media consisted of NPC basic media supplemented with 0.5 mM dcAMP, 50 ng/mL Brain-Derived Neurotrophic Factor (BDNF), and 0.5 mM valpromide. The media were changed by replacing half of the volume every 2–3 days.

Tunicamycin treatment

After 8–10 days of neuronal differentiation of GFP-105Q cells, the media was completely replaced with fresh differentiation media containing the original supplements (as mentioned above), along with the following additional supplements: 50 µM Tunicamycin (Sigma-Aldrich, SML1287), 1:1000 dilution of a 1 mg/mL Propidium Iodide solution (PI) (Sigma-Aldrich, P4864), and 5 µM DRAQ5 (1:1000) (Abcam, ab108410). These cells were then subjected to time-lapse live imaging. For the conditioned media experiments (Supplementary Fig. S2D), we sorted IB⁺ and IB− cells by FACS, plated at equal densities, and allowed to recover overnight. Conditioned media was collected separately from a population of IB+ NPCs growing for 36 h and applied to the IB⁺ and IB− NPCs 24–48 h following plating. Cells were treated with Tunicamycin and stained with DRAQ5 at a 1:2000 dilution and imaged over a period of 48 h. ATF3+/+ and ATF3−/− NPCs were first exposed to Tunicamycin for 72 h. Since Tunicamycin is dissolved in DMSO, we used the same volume of DMSO as control. We then collected the cells and stained them with 1 μM propidium iodide (PI) to identify the dead cells and depicted the percentage of dead cells using FACS.

Time-lapse imaging

Cells were imaged using the FV-1200 confocal microscope, equipped with an incubator controlling temperature (37 °C) and CO2 (Olympus, Japan). Depending on multi well plate the following objective were used: 20×/0.45 or 20×/0.75. Images were captured every 15 min using multiple dyes sequential scanning mode in order to avoid emission bleed-through. QuPath image analysis software [20] was used to analyze and quantify the cells in each group.

 

Excitation (nm)

Emission (nm)

GFP

488

505–550

IP

561

575–620

DRAQ5

635

655–755

Immunofluorescence

For Immunofluorescence (IF), cells were fixed in 4% paraformaldehyde (PFA) at room temperature for ten minutes, permeabilized with 0.1% Triton X-100 for five minutes and blocked with 5% bovine serum albumin (BSA). ATF3 (ab207434, 1:500) primary antibody was incubated overnight at 4 °C. Alexa, 568, conjugated secondary antibodies was diluted 1:1000. Cells were incubated for thirty minutes at room temperature on shaker. Then, cells were mounted on slides using mounting media containing DAPI. Images were acquired using an Olympus confocal disk microscope, analyzed and processed using ImageJ.

Western blots

NPC dry pellet from one well of six-well plates were lysed using 100 μl RIPA buffer (50 mM Tris HCL pH 8, 150 Mm NaCl, 1% Triton X-100, 5 mM EDTA, 0.5% sodium deoxycholate, 0.1% SDS) and 1:1000 freshly added protease inhibitor for 15 min on ice. Followed by 10 min centrifugation in a pre-cooled centrifuge at 4 °C for 10 min at 15,000 rpm. The supernatant was collected and mixed with the appropriate amount of 4X Bolt™ LDS Sample Buffer and 10X Bolt™ Sample Reducing Agent. Then, the sample was heated to 70 °C for 10 min and loaded to Invitrogen Bolt Bis-Tris Plus 4–12% (10 or 17 wells). The iBlot dry plotting system by Invitrogen was used to transfer the samples to PVDF membrane. Followed by a 1-h incubation of the membrane in 5% BSA then it was exposed to the primary antibody overnight. 16–18 h later and after washing the membrane 3 times 5 min each, secondary antibody was added for 30 at room temperature, dilution 1:10,000. The following primary antibodies were used: Rabbit-anti ATF3 (ab207434, 1:1000); Rabbit-anti GAPDH (ab181602, 1:10,000).

RNA-Seq

RNA was extracted using RNeasy Mini Kit (Qiagen). Deep sequencing libraries were prepared using two library preparation kits. HEK293 library was prepared using Lexogen QuantSeq 3’ mRNA-Seq kit. NPC libraries were prepared using KAPA Stranded mRNA-Seq kit. In both NPCs with and without IBs following sorting, the initial RNA concentration was 150 ng, and 14 PCR amplification cycles were used to amplify material. For other NPCs (without sorting), the RNA concentration was 1 μg and 11 PCR amplification cycles were used. Libraries were sequenced using an Illumina NextSeq 500.

ChIP-Seq

ChIP was performed as previously described [21]. 3 μg of ATF3 (ab207434) antibody was used. Subsequent library preparation for sequencing was performed as described previously [22, 23].

ELISA

After FACS, WT, IB+, and IB− cells were collected. Cells with IBs were subjected to their seventh PulSA. 50,000 cells from each cell type were plated in 3*12 wells. The medium was collected after 40 h and 50 µl was used to run the Human IL-8 ELISA Kit (Abcam ab214030). The plate reader protocol was set to 600, 20 min, with linear shake for 1 min at 410 cpm and 75 s intervals. 12 readings were completed until saturation was reached. The read was completed at 450 after the stop solution was added.

Quantitative reverse-transcriptase PCR (RT-qPCR)

To validate the differences in expression cDNA was prepared from 100 ng of purified RNA using the High Capacity cDNA kit (Applied Biosystems™:4368814). RT-qPCR was performed using Power SYBR™ Green PCR Master Mix (Applied Biosystems™:4368577), according to the manufacturer’s instructions. Primer sequences are shown below.

Gene

Sequence (5′ → 3′)

HSPA5_F

AGG ACA AGA AGG AGG ACG TG

HSPA5_R

GGA GTG AAG GCG ACA TAG GA

IL11_F

CAC AAC CTG GAT TCC CTG C

IL11_R

CGC AGG TAG GAC AGT AGG T

MAFF_F

CAC CTT CTG CAA ACA TGT CTG

MAFF_R

CAC CGA CAG CCC CAT CAG

STK40_F

TGC AAA GAG AGC TGG ACC AT

STK40_R

GTC ATC CGT GCC ATC TTT CC

XBP1_F

AGC TCA GAC TGC CAG AGA TC

XBP1_R

TGG TTC TCA ACT ACA AGG CCA

CXCL8_F

CAA ACC TTT CCA CCC CAA AT

CXCL8_R

CTC TGC ACC CAG TTT TCC TT

Data and statistical analysis

Results section and/or figure legends contain all statistical information pertaining to sample sizes and statistical tests. Unless otherwise indicated, comparisons between two groups were performed using a Student’s t test. Differences were considered statistically significant at p < 0.05; non-significant results are denoted as ‘ns’.

Preprocessing and alignment

Raw fastq files were passed quality control using fastqc software (and raw reads were trimmed to remove Illumina adapters using the Trimmomatic software version 0.39 [24]. Reads were aligned to the human RefSeq reference genome (GRCh38) using STAR aligner version 2.7.1a [25] with default parameters. Raw counts for each gene were generated using --quantMode GeneCounts.

Gene expression quantification and analysis

Raw counts were normalized based on the median of the geometric mean, similar to DE-Seq [26]. Normalized expression values were then log-transformed using log2(x + 1). Genes were considered differentially expressed if the average of the log transformed values resulted in at least 1.5-fold difference and the sum of averages of the log-transformed values was larger than 3 (which ensures at least ~5 RPM for the condition that had the higher expression level). p-values were calculated based on two-tailed t-test between the log-transformed values in both conditions. FDR-corrected p-values were calculated using the Benjamini–Hochberg (BH) correction for multiple hypothesis. For the HEK data, due to the low number of replicates (n = 2 for each condition) which does not allow for a derivation of meaningful p-values, we used a stringent LFC criterion: Genes were considered as differentially expressed if, in addition to the former sum of average criterion, the average of the log transformed values resulted in at least 1.5 fold difference between the lower expression value of the upregulated condition and the higher expression value of the downregulated condition, rather than using the mean expression values in the two conditions.

ChIP

Raw reads were aligned to the human hg38 reference genome using bowtie2 [27], followed by DNA duplicates removal using rmdup from the samtools [28]. Peak calling was carried out using MACS2 software [29] with parameters nomodel and extsize = 200.

Results

Establishment of polyQ-IBs-containing cellular model Huntington’s disease

To establish a human polyQ-IBs-containing cellular model for HD, we began with human embryonic kidney 293 cells (HEK293) as a conventional cellular model. We first introduced a viral vector containing GFP fused to the first 21 exons of the mutant human HTT gene containing 105 CAG repeats (GFP-105Q) into (HEK293) cells (Supplementary Fig. S1A). In parallel, we infected HEK293 cells with a non-pathogenic version of the same vector containing 18 CAG repeats (GFP-18Q), as control (Supplementary Fig. S1A). While essentially all cells were GFP-positive following infection, some of the HEK293 cells infected with GFP-105Q began to form polyQ IBs, while no IBs were observed in the cells infected with GFP-18Q. To enrich for these polyQ-positive HEK293 cells, we used the PulSA (pulse shape analysis) method [30], which enables to select for cells containing fluorescent IBs based on their physical properties. Following several consecutive rounds of FACS using PulSA, we were able to obtain a pure population of polyQ-GFP-IBs in the 105Q-HEK293 cells (Supplementary Fig. S1B, right). Subjecting the control 18Q-HEK293 cells to the same PulSA regiment produced no IBs whatsoever (Supplementary Fig. S1B, left).

Having successfully established HEK293 cells containing polyQ IBs encouraged us to advance to a more disease-relevant system. To this end, we used patient-derived genetically corrected iPSCs [19] and ectopically expressed the pathogenic vector (GFP-105Q) (Fig. 1A). Despite repeated attempts, we could not detect polyQ-IBs-containing cells in human PSCs, suggesting that human pluripotent cells can avoid IB formation [31]. Therefore, we differentiated our 105Q-GFP iPSCs into neuronal progenitor cells (NPCs) [32]. While most cells remained devoid of polyQ IBs, some IB+ cells began to appear. To enrich for these rare IB+ NPCs, we used PulSA to sort the cells (Fig. 1B), and repeatedly collected the IB+ NPCs achieving, once again, a homogeneous culture of IB+ NPCs (Fig. 1C).

Fig. 1: Establishing human NPCs with GFP-polyQ-IBs.
Fig. 1: Establishing human NPCs with GFP-polyQ-IBs.The alternative text for this image may have been generated using AI.
Full size image

A A scheme of GFP-105Q-iPSCs generation. HD-iPSCs with normal number of glutamines (corrected) were infected with a GFP-HTT vector containing 105 CAG repeats using lentivirus. B Flow cytograms of 105Q-NPCs depict the enrichment of IBs between the first PulSA (left) and the fourth PulSA (right). Red rectangles indicate the merge of the new IB subpopulation. C A fluorescent image of 105Q NPCs prior to PulSA enrichment (left) and after the 4th sorting (right) shows polyQ IB following PulSA. Scale bar = 200 µm. D A scheme of GFP-18Q-iPSCs generation. HD-iPSCs with 180Q were infected with a GFP-HTT vector containing 18 CAG repeats using a lentivirus.

So far, our cellular systems contain ectopically-induced IBs. We therefore next wished to develop a similar system where we would be able to analyze naturally occurring IBs. To this end, we turned to the most severe HD patient-derived iPSCs available, containing 180 CAG repeats [19, 33]. To enable visualization of the naturally occurring endogenous IBs, we expressed a GFP-18Q plasmid (Fig. 1D), which does not form IBs by itself (Supplementary Fig. S1B), but can join existing IBs [14], and differentiated the 180Q iPSCs into NPCs. For the GFP-105Q system above, we used the isogenic corrected line of the 180Q cells, thus our two ‘exogenous’ (105Q) and ‘endogenous’ (180Q) systems share the same genetic background.

PolyQ IBs are protective

So far addressing this question directly, especially in human neuronal cells, has been challenging. Since following cell division, our cultures contain both IB+ and IB- cells growing side-by-side, our system provides an excellent platform to resolve this long-standing question. To this end, we differentiated our NPCs, containing 5–20% IB+ cells, into post-mitotic neurons. This not only provided us with a more relevant cell type, but also eliminated proliferating cells, which otherwise hinder analysis. We then exposed these neurons to Tunicamycin, an inhibitor of N-linked glycosylation (N-glycans), which gradually leads to the accumulation of unfolded proteins in the cells, triggering endoplasmic reticulum (ER) stress. Cell death was subsequently assessed using propidium iodide (PI). Using live imaging, we monitored changes in nuclear staining to track cell death. We assessed the startpoint (Fig. 2A, Supplementary Video 1) and the endpoint (Fig. 2B, Supplementary Video 1) using QuPath image analysis software [20]. Remarkably, after ~30 h, we observed a striking difference in the number of dead cells between IB+ and IB- cells (Fig. 2B). To quantify this difference, we trained QuPath to classify cells into categories: ‘live’ (red), ‘dead’ (yellow), ‘IB+’ (green), and cells with both ‘dead’ and ‘IB+’ signals (light blue) (Fig. 2B, C; Supplementary Fig. S2A, B, Supplementary Video 1). To calculate the percentage of dead cells in the IB+ and IB- populations, we subtracted the percentage of dead cells at the starting point from the percentage of dead cells for each population (Supplementary Fig. S2C). This procedure was repeated for >1000 cells across several time-lapse movies. Profoundly, the percentage of dead cells in the IB- group was significantly higher than in the IB+ group (Fig. 2D). To test whether the IB⁺ cells can induce damage in the IB− cells, we performed a similar experiment using conditioned media derived from IB⁺ and IB− cells. Pure populations of IB⁺ and IB− cells, sorted from the same parental population, were treated with conditioned media from either IB⁺ or IB− cells. While, as before, IB+ cells were more protected than IB- cells from Tunicamycin, the IB+ conditioned media did not result in increased cell death (Supplementary Fig. S2D), suggesting that IB⁺ cells do not promote cell death in IB− cells via secreted factors. Taken together, these results suggest that, once a mutant polyQ protein is expressed, the formation of IBs protect the cells rather than inflict harm, at least short-term.

Fig. 2: IBs are protective.
Fig. 2: IBs are protective.The alternative text for this image may have been generated using AI.
Full size image

A A snapshot from the start of a time-lapse imaging experiment following treatment of GFP-105Q NPC IB+ (green) with Tunicamycin and staining with DRAQ5 (DNA, blue) and propidium iodide (PI) (dead cells, red). Left: merged image of all channels from the start point. Right: same image with QuPath annotations and detections; red (cells), green (IB+), yellow (dead cells) and lime-green (IB+ and dead cells). B A snapshot from the end point of the same video showing the same features as in (A). C Zoomed-in view of a small field to detail the QuPath annotations. D Box plot representing the percentage of the change in cell death of IB+ versus IB- between the start point and the end point.

IB + NPCs induce neuroinflammatory and UPR related genes

To identify the pathways potentially responsible for the formation of polyQ IBs, and leveraging our system that allows comparison between IB+ and IB− sister NPCs. We separated the IB+/− 105Q-NPCs, which have all been descended from the IB+ NPCs, and performed RNA-seq (Fig. 3A). As expected from sister cells grown side-by-side, most of the genes had similar expression levels between the two populations (Fig. 3B, black dots). However, 44 and 47 genes were reproducibly up/down-regulated, respectively, in three independent biological experiments (with log2 fold change > 1 or <−1 and P < 0.05) in the IB + 105Q-NPCs (Fig. 3C & Supplementary Fig. S3A). Among the reproducible and significantly induced genes were CREB5, ATF3, FOSL1, IL-11, ILR6 and CXCL8 (encoding for IL-8). The latter was the most highly induced gene in the IB+ NPCs. GO analysis of the 44 up-regulated differentially expressed genes (DEGs) using Enrichr [34] highlighted regulation of cell proliferation, endoplasmic reticulum (ER) stress, inflammatory response, and cytokine-mediated signaling pathway (Fig. 3D). The down-regulated DEGs revealed mainly viral related pathways and different interferon responses (Supplementary Fig. S3B).

Fig. 3: IB + NPCs induce neuroinflammatory and UPR related genes.
Fig. 3: IB + NPCs induce neuroinflammatory and UPR related genes.The alternative text for this image may have been generated using AI.
Full size image

A Left: A typical flow cytogram demonstrating IB- (bottom rectangle) and IB+ (top rectangle) cells. Right: Images representing IB− (left) and IB+ (right) NPCs. Scale bars = 200 µm. B Volcano plot of IB+/− 105Q-NPCs RNA-seq data. Blue and pink represent down- and up-regulated in IB+ NPCs, respectively. C Heatmap presenting the significantly altered DEGs between the three reproducible biological replicates of IB+/− 105Q-NPCs. D Gene Ontology (GO) analysis of pathways enriched in the up-regulated genes between IB+ and IB- 105Q-NPCs. E Volcano plot of IB+/− 180-NPCs RNA-seq data. Blue and pink represent down- and up-regulated genes in the IB+ NPCs, respectively. F Heatmap presenting the significantly altered DEGs between the three reproducible biological replicates of IB+/− 180Q-NPCs. G Gene Ontology (GO) analysis of pathways enriched in the up-regulated genes between IB+ and IB- 180Q-NPCs.

We next repeated the whole process to compare DEGs between the naturally occurring IB+/− in the 180Q-NPCs. RNA-seq revealed 33 and 38 genes that were consistently up/down-regulated, respectively, in the IB+ cells (with log2 fold change > 1 or <−1 and P < 0.05) (Fig. 3E, F; Supplementary Fig. S3C). GO analysis showed regulation of cell proliferation, cytokine-mediated signaling pathway and unfolded protein response (UPR) in the upregulated DEGs, and regulation of heart growth, regulation of cellular process, and extracellular matrix organization in the down-regulated DEGs (Fig. 3G; Supplementary Fig. S3D). Quantitative reverse-transcriptase PCR (RT-qPCR) analysis of several key genes validated the RNA-seq results (Supplementary Fig. S3E). Combined, our IB+/− systems now provide a discovery platform for genes induced by polyQ IBs regardless of the cells of origin and regardless of whether the IBs are ectopically or naturally induced. Taken together, the results from both systems suggest a potential involvement of inflammatory response and UPR in human cells containing polyQ IBs.

ATF3 is a prime suspect in regulating IB+ genes

Comparing our DEGs from both the ‘endogenous’ and the ‘exogenous’ systems, we observed that more than half of the genes that were altered in the 180Q-NPCs were also altered in the 105Q-NPCs (Fig. 4A) (P < 2.2 ×10−16, OR = 396.65, Fisher’s exact test), attesting to the specificity and reproducibility of the two systems, and allowing us to focus on the most relevant, reproducible and prominent genes. In the HEK293 cells, across two independent experiments, we identified 145 and 102 up/down-regulated genes (log2 fold change >0.58/<0.58, P < 0.05), respectively (Supplementary Fig. S3F). Of these, 25 genes shared the same expression pattern with our unique set of 37 genes, but only 7 of those exhibited significant changes (P = 3.026 ×10−7, OR = 19.15, Fisher’s exact test). This left us with 30 genes that were exclusively differentially expressed in the IB+ NPCs (Supplementary Fig. S3G). Among these 30 NPC-specific genes were eight transcription factors (TFs), including ATF3, FOSL1, PTX3, and ZNF165 (Fig. 4B, red arrows), as well as several inflammation-related cytokines, such as IL6R, IL11, and PTX3 (Fig. 4B, turquoise arrows). Interestingly, CXCL8 (encoding IL8), which was the most highly induced gene in the IB+ NPCs, was common to all three systems.

Fig. 4: ATF3 is a prime suspect in regulating IB+ genes.
Fig. 4: ATF3 is a prime suspect in regulating IB+ genes.The alternative text for this image may have been generated using AI.
Full size image

A Venn diagram representing the common genes that are differentially expressed in the 105Q and 180Q NPCs. B Heatmap of the 37 common genes from three independent experiments (shown separately) of IB+/− 105Q and 180Q NPCs samples. The turquoise and red arrows denote inflammation-related genes and TFs, respectively. C GO analysis of the 37 genes from (B). D ChIP-seq enrichment analysis (ChEA) of the 37 common DEGs. E Western blot for ATF3 (top) in WT ATF3+/+ NPCs (left lane) and ATF3−/− NPCs (right lane). GAPDH (bottom) served as a loading control. F Immunofluorescence images of WT (Top, ATF3+/+) and Knockout (Bottom, ATF3−/−) NPCs. Blue, DAPI; Red, ATF3. Scale bars = 50 μm. G GSEA enrichment plot of the unfolded protein response (UPR) genes in ATF3−/− 105Q-NPCs, showing downregulation of UPR genes upon ATF3 depletion. Each black line represents a gene. H Heatmap of expression changes of the 79 UPR-related genes between ATF3+/+ and ATF3−/− 105Q-NPCs, in 3 biological replicates.

Although the GO analysis on the 37 shared genes also revealed inflammation-related pathways including different interleukins, cytokine-mediated signaling and PERK-regulated gene expression pathways (Fig. 4C, red arrows), the ATF2 TF network and AP-1 TF network stood out as new entries (Fig. 4C, turquoise arrows). Identifying these TF-related pathways prompted us to seek the specific TFs which might regulate our 37 induced genes. To this end, we used ChEA (ChIP-seq enrichment analysis) on the 37 common genes. Intriguingly, two out of the top three ChEA hits – ATF3 and FOSL1 – were among our 37 upregulated genes (Fig. 4D). Applying the same analysis on the upregulated genes in both systems also highlighted ATF3 and FOSL1 ChIP-seq maps (Supplementary Fig. S4A, B), suggesting functional relevance. These findings may suggest that the TFs that regulate the 37 differentially regulated genes, are among those genes. As the most prominent among the two, we focused on ATF3.

ATF3 activates genes involved in UPR

To test the direct involvement of ATF3 in our system, we next used CRISPR/Cas9 to knock-out (KO) ATF3 (ATF3−/−) in the parental 105Q-NPCs (NPCs expressing GFP-105Q before the PulSA sorting, which we will refer to from hereon as ATF3+/+). We confirmed the KO of ATF3 by Western blots (Fig. 4E; Supplementary Fig. 7B, G-I) and immunofluorescence (Fig. 4F). RNA-seq comparing ATF3−/− and ATF3+/+ NPCs identified 262 and 388 up/down-regulated genes, respectively, in three biological replicates (with log2 fold change > 1 or <−1 and P < 0.05) (Supplementary Fig. S4C, D). GO analysis of the upregulated genes revealed RNA binding and ribosome related pathways as the main affected pathways in ATF3−/− cells (Supplementary Fig. S4E). There are no significantly downregulated pathways. We again validated several key DEGs using RT-qPCR (Supplementary Fig. S4F).

ATF3 was previously shown to be induced by ER stress [35], but its direct role in ER stress is not fully understood. Based on our gene expression, GO and ChEA analyses, we focused here on the potential role that ATF3 might play in the UPR. To this end, we tested the effect of ATF3 KO on the expression level of UPR related genes. The Reactome Pathways dataset from Harmonizome [36] lists 79 genes as UPR related. Gene-set enrichment analysis (GSEA) revealed that 48 of them tend to be down-regulated in ATF3−/− cells (P < 0.05; normalized enrichment score [NES] = −1.5) (Fig. 4G, H). These results suggest that ATF3 acts as an upstream activator of UPR related genes in our system.

These results prompted us to analyze the genome-wide ATF3 targets in our NPCs by chromatin immunoprecipitation followed by high throughput sequencing (ChIP-seq). We performed ATF3 ChIP-seq in both ATF3+/+ and ATF3−/− NPCs using ATF3-specific antibodies. Metagene profiles revealed a specific ATF3 peak immediately upstream to the transcription start site (TSS) in the ATF3+/+ NPCs, but not in the ATF3−/− NPCs (Fig. 5A), attesting to the specificity of the ChIP-seq reaction. Zooming in on several ATF3 targets, e.g. MAFF and STK40, we observed specific enrichment of ATF3 in ATF3+/+ (Fig. 5B, red tracks) but not in ATF3−/− NPCs (Fig. 5B, green tracks). Consistent with previously published results, which showed that ATF3 regulates its own expression [37, 38], our data also showed strong enrichment of ATF3 on its own promoter (Fig. 5C). Reassuringly, although this is the first ATF3 ChIP-seq data in human NPCs to date, ChEA analysis of our ATF3 ChIP-seq profiles highlighted the two previously published ATF3 ChIP-seq maps as the most similar profiles to our data (Supplementary Fig. S5A). Moreover, the ChEA analysis further revealed, from third place onwards, ChIP-seq maps of JUND, JUN, SMAD2-4, CREB1, and CEBPB, all of which are members of the direct protein-protein interaction network of ATF3, as evident by STRING (Supplementary Fig. S5B), further attesting to the specificity and biological relevance of our data. Finally, using motif enrichment analysis of our ATF3 ChIP-seq data, we found that our data significantly represented the canonical motif of AP-1 (5’-TGA(C/G)TCA) (Supplementary Fig. S5C). Further inspecting ATF6, XPB1, DDIT3, and HSPA5, all of which are central UPR genes, we observed a selective and specific enrichment of ATF3 in their promoters (Supplementary Fig. S5D).

Fig. 5: ATF3 constitutively activates genes involved in UPR.
Fig. 5: ATF3 constitutively activates genes involved in UPR.The alternative text for this image may have been generated using AI.
Full size image

A ChIP-seq metagene profile and heatmap of ATF3 ± 3 kb around the TSS in ATF3+/+ and ATF3−/− NPCs. B Representative genome browser views of ATF3 ChIP-seq data around MAFF and STK40 TSS for both ATF3+/+ and ATF3−/− NPCs. C Genome Browser tracks of ATF3 binding profile to itself. In panels (B) and (C), green and red tracks show ATF3 ChIP-seq in ATF3−/− and ATF3+/+ NPCs, respectively. Gene annotations are shown at the bottom (blue). Representative genome browser views of ATF3 ChIP-seq data in IB+ (blue), IB- (magenta), ATF3+/+ (red), and ATF3−/− (green) 105Q-NPCs around the genomic regions of UPR-related genes XBP1 (D), HSPA5 (E), HERPUD1 (F) and PPP2R5B (G). Gene annotation (blue) is at the bottom.

So far, our data strongly suggest that ATF3 activates the main UPR genes by directly binding to their promoters, as evidenced by comparing ATF3−/− and ATF3+/+ human NPCs. We then sought to identify the ATF3 binding profile across UPR related genes in IB+ and IB- NPCs. Therefore, we set out to sort our ATF3+/+ NPCs into IB+ and IB- cells and performed ChIP-seq for ATF3 on each of the sorted populations. Interestingly, ATF3 is selectively and specifically enriched in the promoters of the key UPR-related genes, not only in IB+ NPCs but also in the parental ATF3+/+, as well as in the IB- NPCs. This pattern can be clearly observed, for example, in the pathway’s central genes, HERPUD1, PPP2R5B, HSPA5 and XPB1 (P < 0.05, OR = 2.84, Fisher’s exact test) (Fig. 5D–G). This suggests that ATF3 regulates UPR-related genes constitutively, both in cells with IBs and those without.

ATF3 exclusively binds Inflammatory response genes in IB+ NPCs

While the activation of the ER stress response was expected to be one of the primary affected pathways in IB+ NPCs, the involvement of inflammation-related genes in the absence of immune cell systems was intriguing. To validate the observed difference in CXCL8 expression, the most highly expressed inflammation-related gene in our data (Fig. 3B, E), we performed RT-qPCR. As anticipated, CXCL8 expression was significantly elevated in IB+ NPCs compared to both IB- NPCs and ATF3+/+ NPCs (Supplementary Fig. S6A). The product of CXCL8, IL-8, is an interleukin secreted by cells to mediate immune responses. Although immune cells were absent from our tissue cultures, we performed an ELISA assay to assess IL-8 secretion. Remarkably, IL-8 secretion was significantly increased (>8-fold) in IB+ NPCs relative to both IB- NPCs and the parental ATF3+/+ NPCs, in both 105Q-NPCs and 180Q-NPCs systems (Fig. 6A).

Fig. 6: Inflammatory response genes are exclusively bound by ATF3 in IB + NPCs.
Fig. 6: Inflammatory response genes are exclusively bound by ATF3 in IB + NPCs.The alternative text for this image may have been generated using AI.
Full size image

A IL-8 ELISA assay. IB+/− NPCs were sorted 40 h prior to media collection. The parental ATF3+/+ NPCs were used as control. IL-8 secretion is significantly increased only in IB+ NPCs. The image on the right shows the experimental plate, with darker medium indicating higher levels of secreted IL-8. Representative genome browser views of ATF3 ChIP-seq data in IB+ (blue), IB− (magenta), ATF3+/+ (red), and ATF3−/− (green) 105Q-NPCs around the inflammatory response related genes CXCL8 (B), PTX3 (C) and IL11 (D). Gene annotation (blue) is at the bottom. The genome browser views of the RNA-seq results represent the difference in expression between IB+/− NPCs. The red rectangles indicate ATF3 binding sites.

Focusing on the genes which were differentially expressed in IB+ NPCs and those exclusively bound by ATF3 in IB+ NPCs from our ChIP-seq data, we found that five genes out of our 37 DEGs appeared in both lists, suggesting direct induction by ATF3 itself. Notably, three of these genes, IL11, PTX3 and CXCL8, are inflammation related. Zooming in on the promoter region of these genes (±3 kb from the TSS) revealed selective enrichment of ATF3 on the promoters of CXCL8, IL11, and PTX3 in IB + NPC, but not in the parental ATF3+/+, nor the IB− NPCs (Fig. 6B–D). This selective binding pattern was strikingly distinct from that observed in UPR-related genes (Fig. 5D–G). Conspicuously absent from our ATF3 binding list in IB+ NPCs was IL6R, which was one of the top inflammation-induced genes in our system (Fig. 3B, E). However, a closer examination revealed a new clear ATF3 binding site inside ILR6’s second intron. Interestingly, this new ATF3 binding site overlaps with an enhancer region (Supplementary Fig. S6B, blue box, 7_Enh), demonstrating that ATF3 regulates IL6R only in IB+ NPCs by binding to an enhancer and not the promoter (Supplementary Fig. S6B). Altogether, our results suggest that ATF3 binds to, and activates, the inflammation-related DEGs specifically in IB+ NPCs. Although ATF3 was considered as a repressor of proinflammatory cytokines and interleukins in earlier studies [39, 40]. our data strongly suggest that it acts as a prominent activator of neuroinflammatory genes in IB+ human NPCs containing polyQ IBs.

ATF3 acts through direct DNA interaction

ATF3 has a DNA Binding Domain (DBD) through which it binds DNA [41]. However, as mentioned above, ATF3 interacts with several other prominent TFs including ATF4, JUN, JUND, SMAD3, SMAD4, CEBPB as well as P53 (Supplementary Fig. S5B). To test whether, in our system, ATF3 acts by direct binding to DNA or through another interacting TF (Supplementary Fig. S5B), we excised the DBD from ATF3 in our parental ATF3+/+ 105Q-NPCs (Supplementary Fig. S7A). We used two gRNAs flanking both the 5’ and 3’ sites of the DBD and searched for a clone lacking this domain. Sanger sequencing identified a clone with a missing DBD domain, in addition to several knockout clones. Western blots confirmed the expected smaller size protein (Supplementary Fig. S7B, right lane), and immunofluorescence for ATF3 showed that the ΔDBD-ATF3 remains nuclear (Supplementary Fig. S7C). To verify that ΔDBD-ATF3 does not bind DNA, we next performed ChIP-seq for ATF3 in the ΔDBD-ATF3 NPCs. As expected, the removal of the DBD completely abolished ATF3’s ability to bind DNA, and the ChIP-seq metagene profile of ATF3 in the ΔDBD-ATF3 NPCs was similar to that of ATF3−/− NPCs (Supplementary Fig. S7D). Zooming in on the genomic regions of the STK40 and ATF3 genes, which are bound by ATF3, demonstrates the disappearance of the ATF3 peaks in both cases (Supplementary Fig. S7E, F). This confirms that the excision of the DBD abolishes the capacity of ATF3 to bind DNA.

ATF3 depletion impairs IB-mediated protection

To test the functional involvement of ATF3 and ΔDBD-ATF3 in the formation of the protective polyQ IB. We used our PulSA enrichment method, where we follow the kinetics of IB formation over time. Using consecutive PulSA sortings, we first compared IB formation kinetics in ATF3+/+ NPCs versus ATF3−/− NPCs. We subjected our pre-sorted NPCs and ATF3−/− NPCs (coming from the same background, thus both harboring the GFP-105Q cassette) to PulSA FACS sorting and compared the two populations. In the ATF3+/+ NPC, IBs began to appear after the second sorting, and were highly enriched after the third sorting (Fig. 7A, blue). Remarkably, no IBs were formed in the ATF3−/− NPCs after three or more consecutive sortings (Fig. 7A, red). In fact, despite repeated attempts, we were so unable to obtain IB+ populations in ATF3−/− NPCs. The ΔDBD-ATF3 NPCs (Fig. 7A, orange) behaved similarly to ATF3−/− NPCs, indicating that direct binding of ATF3 is required for its action. Taken together, these data establish ATF3’s involvement in the formation of polyQ IBs in human NPCs.

Fig. 7: ATF3 depletion impairs protection.
Fig. 7: ATF3 depletion impairs protection.The alternative text for this image may have been generated using AI.
Full size image

A Three consecutive PulSA sortings in ATF3+/+ (blue), ATF3−/− (red) and ΔDBD-ATF3 (orange) NPCs. B Immunofluorescence images of IB + 105Q NPCs showing GFP (green, IB+) and ATF3 (yellow) staining, with DAPI (blue) for nuclei (left). Quantification of the percentage of cells expressing ATF3 above the mean, with or without IBs, across multiple different images is presented in the boxplot (right), P < 0.00112. C Spearman correlation between the mean intensity of ATF3 and IBs in the images shown in (B). D Percentage of dead cells (box plot) following treatment with Tunicamycin or DMSO for 72 h. NPCs were stained with propidium iodide (PI) and counted using FACS. The experiment was repeated 5 times. Supplementary Video 1: Time-lapse imaging tracking IB + GFP-105Q neurons (green) treated with Tunicamycin and stained with DRAQ5 (DNA, blue) and propidium iodide (PI) (dead cells, red). Images were captured every 15 min over a period of 30 h, displayed at a frame rate of 10 frames per second (fps).

To confirm the direct relationship between polyQ IBs and ATF3 expression, we conducted immunofluorescence and quantified cells exhibiting both characteristics. IB+ NPCs displayed significantly higher levels of ATF3 (Fig. 7B), validating the induction of ATF3 in IB+ at the protein level. Furthermore, our data revealed a significant Spearman correlation between the mean intensity of IBs and ATF3 (P < 3 ×10−10, Rho = 0.395, Fig. 7C).

ATF3’s critical involvement in the formation of IBs, together with the realization that IBs are protective, prompted us to investigate ATF3-KO NPCs’ response to Tunicamycin. To this end, we exposed our NPCs to Tunicamycin and compared with DMSO-treated controls. Propidium iodide (PI) staining was used to assess cell death, and the percentage of PI-positive cells was quantified by FACS (Fig. 7D). Remarkably, ATF3−/− NPCs, which are incapable of forming IBs, exhibited a significant increase in cell death compared to ATF3+/+ NPCs. This result integrates key findings from our study, demonstrating that the loss of ATF3 impairs the cellular ability to cope with ER stress induced by Tunicamycin. Furthermore, these findings underscore that the inability to form IBs, due to ATF3 deficiency, increases cellular susceptibility to death under stress conditions.

Discussion

In our study, we have established a unique human-PolyQ-IB-NPC system to study polyQ IB biology in Huntington’s and other polyQ-related disorders. While most studies employed more general cell lines, here we used patient-derived iPSCs and their isogenic corrected controls to generate NPCs with exogenous (GFP-105Q) or endogenous (GFP-18Q expressed in 180Q-iPSCs) polyQ IBs. So far, neither ours nor any other PSC study demonstrated polyQ IB formation in human PSCs [31], likely reflecting the unique proteolytic capacity of PSCs [31]. To obtain IB+ cells, we first had to differentiate the pluripotent cells, followed by sequential enrichment for IB+ cells. Due to asymmetric distribution of IBs to one of the daughter cells during mitosis [42], we were able to obtain genetically identical IB+ and IB− NPCs growing side by side in the same plates in similar conditions. This constituted a great advantage for identifying the genes which are exclusively differentially expressed as an outcome of the presence of the polyQ IB, ruling out all other optional factors.

As expected from sister cells growing together, we found very few DEGs, but the DEGs we identified were highly reproducible in different experiments and in different cellular systems. Among them, ATF3 was one of the highly expressed TFs in IB+ NPCs and was predicted to operate as a potential regulator of other DEGs in our list. Moreover, and crucially, we uncovered its role in promoting IB formation in human cells. This finding aligns with a prior study demonstrating that the knockdown of ATF3 in mouse Neuro2a, using shRNA, suppresses IB formation [43]. While ATF3 was not previously implicated in polyQ IB formation in HD, mutant HTT was formerly shown to increase ATF3 expression and enhance its activity [44]. In the same study, and consistent with our Tunicamycin experiments – where ATF3 KO resulted in increased cell death – ATF3 overexpression protected against cell death, whereas ATF3 knockdown increased mutant HTT toxicity in rat PC12 cells [44]. In another study, ATF3 was implicated in regeneration following injury [45]. ATF3 depletion decreased axon regeneration following injury, whereas increased ATF3 expression promoted axon sprouting, regrowth, and regeneration [45]. In line with these data, overexpressing ATF3 in motor neurons of transgenic ALS mice promoted motor neuron survival, preserved axonal integrity, and slightly increased longevity [46]. Taken together, these findings collectively suggest that ATF3 act as a protective factor in neurodegenerative diseases and/or following injury or stress.

The most upregulated inflammation associated gene in the IB + NPC DEGs was CXCL8, encoding IL8. CXCL8 expression was shown to increase in post-mortem HD patient’s brains [47] and particularly in the striatum [48, 49]. Interestingly, IL8 levels in plasma and CSF were shown to gradually increase with illness severity, and correlated with disease progression [49]. Other cytokines, on the other hand, such as IL6, already increased in the plasma of pre-onset HD patients [49]. Since IBs are formed only at an advanced stage of the disease, these findings support our observation that IL8 is selectively induced in the IB+ NPCs. In addition, we identified ATF3 as the responsible TF for this increase.

IL8 is expressed in humans but not in mice [50]. While this may be considered a nice proof of the uniqueness and advantage of our human system, it is a drawback in terms of research continuation. An additional obstacle is that the current IL8 inhibitors on the market mainly target the IL8 receptors CCXR1 and CXCR2, which are expressed in microglia—the brain’s resident immune cells—yet are absent in our NPC system. This limitation led us to consider co-culturing for follow-up experiments.

Consistent with the gene expression changes observed in our IB+ NPCs, similar alterations were detected in the post-mortem striatum of HD patients. We reanalyzed previously published RNA-seq data from the striatum of three HD patients and three matched controls [48]. Remarkably, 21 genes out of our 37 DEGs were also differentially expressed in the patients’ striatum (P = 6.57 × 10−5). To concentrate on human-specific genes, we analyzed data of DEGs in the striatum of R6/1 mice from the same study [48]. Eight genes out of our 37 genes were common to both patients and mice striatum, leaving us with 13 human-specific prominent genes, including, notably, IL6R, CXCL8, ATF3 and FOSL1. These human post-mortem brain data attested to the specificity and relevance of our findings and reconfirmed our notion that the genes we identified play a potential role in HD and polyQ IB formation.

Finally, we confirm S. Finkbeiner’s observations that polyQ inclusion bodies (IBs) are protective, also in human neurons [17]. Their study suggested that the formation of IBs results in a reduction of soluble mutant HTT, hypothesizing that IBs sequester the toxic, diffuse HTT [17]. We find similar evidence in our system, showing a clear reduction in soluble mutant HTT upon IB formation (Supplementary Video 1). But while our system indeed shares similarities with their approach, our study differs in several key aspects, most notably the use human neurons and the longer HTT construct, which leads to polyQ IB formation mainly in the cytoplasm rather than diffusing into the nucleus. Supporting a protective role for polyQ IBs in vivo, a recent study using the FDNQ175 mice model compared the motor function of males and females separately at 8 months [51]. The study revealed that males were more susceptible to deterioration than females while females had higher number of insoluble IBs in their brain.