Introduction

Huntington’s disease (HD) is a devastating neurodegenerative disorder characterized by irreversible neuronal loss in the striatum, resulting in profound motor dysfunction and cognitive decline(Kernich, 2002; Saudou and Humbert, 2016; Walker, 2007a). The condition is driven by the expansion of a polyglutamine (polyQ) tract within the Huntingtin (HTT) protein, rendering it highly prone to aggregation (Gutekunst et al., 1999). Notably, HTT aggregates exhibit significant instability following chemical fixation (Ferrante et al., 1997), a histological challenge that can lead to the misinterpretation of polyQ biology and pathophysiology if standardized protocols are not strictly applied.

While polyQ aggregate formation and the presence of intranuclear inclusion bodies are traditionally considered pathological hallmarks of HD (DiFiglia et al., 1997; Gutekunst et al., 1999), their precise role remains controversial.

Emerging evidence suggests a paradox: some studies indicate that aggregate formation may actually increase single-neuron survival (Arrasate et al., 2004), exerting a protective rather than toxic effect (Kim et al., 1999; Saudou et al., 1998). Conversely, soluble monomeric HTT and its transition into oligomeric species—which precede fibril and inclusion body formation—are the primary drivers of cellular toxicity (Hatters et al., 2013; Jimenez-Sanchez et al., 2017; Lajoie and Snapp, 2010; Legleiter et al., 2010; Mukai et al., 2005; Poirier et al., 2002; Takahashi et al., 2008). These observations suggest that mature HTT aggregates may possess critical biological functions rather than being purely pathogenic.

HTT is a stiff protein composed of multiple ɑ-helical and PolyQ (Dougan et al., 2009; Guo et al., 2018; Saudou and Humbert, 2016; Sivaramakrishnan et al., 2008; Tabrizi et al., 2020). Expanded polyQ (>35) and short polyQ form a rigid fibril (Bauerlein et al., 2017; Dougan et al., 2009; Isas et al., 2017; Scherzinger et al., 1999); however, expanded polyQ fibrils have been reported to physically disrupt organelle membranes, such as the endoplasmic reticulum (ER), and alter membrane dynamics (Bauerlein et al., 2017). Functionally, HTT acts as a versatile scaffold involved in a wide array of cellular processes, including vesicular transport, autophagy, gene transcription, and embryonic development (Saudou and Humbert, 2016). HTTs are an essential protein in embryonic development (Dragatsis et al., 1998; Duyao et al., 1995; Zeitlin et al., 1995). The presence of mutant HTT (mHTT) precipitates widespread cellular dysfunction, ranging from early transcriptional dysregulation and synaptic impairment to mitochondrial distress and disruption of the nuclear pore complex (Costa et al., 2010; del Toro et al., 2009; DiFiglia et al., 1997; Grima et al., 2017; Kim et al., 2010; Mehta et al., 2018; Milnerwood et al., 2006; Saudou and Humbert, 2016; Shirasaki et al., 2012). mHTT induces extrasynaptic excitotoxicity(Milnerwood et al., 2010), higher levels of brain inflammation(Lee et al., 2020; Silvestroni et al., 2009), abnormal fetal brain development(Barnat et al., 2020), and aging(Machiela and Southwell, 2020).

HAP40 stabilizes HTT(Guo et al., 2018; Harding et al., 2021). HTT interacts with and activates dynamin, a protein that catalyzes membrane fission (Caviston et al., 2007; Caviston et al., 2011; Morlot and Roux, 2013), microtubule motors, and actin-associated adaptors to support cellular vesicle transport, including autophagy, endosomes, and lysosomes (Caviston et al., 2011; Olenick and Holzbaur, 2019).

Using HD family-derived cells and super-resolution scanning, we revealed a knitted-fabric-like network formed by HTT polyQ that encircles flat Golgi stacks/cisternae and dynamically couples with both mitotic and stress-related Golgi fragmentation. We demonstrate that while normal polyQ assemblies support the structural integrity of the Golgi, pathogenic mHTT "crisps" these assemblies, destabilizing the complex and impairing Golgi activities. Our findings reveal that the polyQ assembly-Golgi complex is a highly organized functional unit, the disruption of which represents a central mechanism of mHTT-induced toxicity.

Results

Characterization of endogenous polyQ aggregate structure

We identified an HD family in the Mongolian population of West China. 24 persons have identifiable syndromes, including an adult-onset, abnormal gait, uncontrolled arm, leg, and head movements, and psychiatric abnormalities; some patients with severe syndromes showed mania and pica (Table S1). The self-reported symptoms appeared at ages 24 to 50; patients died from 35 to 56 (Table S1). Genetic testing revealed that the patients in this family harbor 44 to 59 CAG repeats in HTT (Fig. 1A, B; Table S1), compared with 18-19 CAG repeats in healthy individuals. Consistent with HD, the lengths of CAG repeats define the disease onset age, symptom severity, and death age (Fig. 1C; Table S1).

Figure 1.

To assess polyQ status in fibroblasts from this HD family, we stained them with the 3B5H10 antibody, a polyQ antibody that equally recognizes both HTT and mHT polyQ (Owens et al., 2015). We observed that the polyQ of fibroblasts, including patients and healthy siblings, exhibits well-organized assemblies that resemble knitted fabrics formed by two paralleled spindles that interfused at a regular interval, structurally resembling an assembly of multiple rings (Fig. 2A). Contrary to other observation(Tousley et al., 2019), phalloidin counterstaining showed that the polyQ assembly did not regularly parallel actin filaments and randomly interact with action filaments (Fig. 2A). 3D rendered tomography showed that the polyQ assembly in fibroblasts interacts with the nuclear membrane (Fig. 2A). Notably, the size of polyQ assemblies in HD is similar to that in healthy fibroblasts (HD: n=4; healthy: n=4) (Fig. 2B, C). Mitochondria did not preferentially enrich in the polyQ assembly site: some of them were present in the polyQ assembly network’s space, others were not (Fig. 2D). The polyQ assemblies were fragmented in mitotic cells (Fig. 2E). Strikingly, the polyQ assemblies fragmented after 4% PFA fixation: the immediate staining after fixation is the only way to visualize an intact polyQ assembly (Fig. 2F). These findings indicate that polyQ assemblies are a unique, dynamic polyQ assembly with a specific structure in cells.

Figure 2.

PolyQ assemblies of HD patients and healthy individuals differentially respond to stress and drugs

Antisense oligonucleotides (ASOs) reduce the transcription of HTT(Kordasiewicz et al., 2012). To test how the polyQ assemblies in fibroblasts respond to ASO, we treated the HD family fibroblasts with ASO (20 and 40 µM) and found that the ASO equally reduced the polyQ assemblies in healthy and HD fibroblasts without affecting the structure of polyQ assembly (Fig. 3A-C). Autophagy-enhancing agents were reported to reduce mHTT(Martin et al., 2015). We also treated HD fibroblasts with onjisaponin F (5, 10, 20, and 40 µM), an active component of the polygala tenuifolia(Nagai et al., 2001). Onjisaponin F reduced polyQ assemblies in HD fibroblasts, but not those in healthy fibroblasts, in a dose-dependent manner, without fragmenting polyQ assemblies (Fig. 3D, E). To see how the polyQ assemblies respond to energy deprivation, we cultured the fibroblasts from 4 patients with HD (CAG: 48, 49, 55, and 59) and 2 healthy siblings in a low-glucose medium. We fixed cells at 24, 48, and 72 h and evaluated the polyQ assemblies. Under glucose deprivation, the polyQ assemblies in HD fibroblasts are diminished and fragmented over time. In contrast, polyQ assemblies in healthy fibroblasts are stable, and the smaller polyQ fragments in healthy cells are decreased (Fig. 3F-H). After culturing in the low-glucose medium for 72 h, we switched to a high-glucose medium. After culturing for an additional 48 h, we observed that the polyQ assemblies in HD fibroblasts had been restored (Fig. 3F-H).

Figure 3.

Together, the presence of pathogenic polyQ shifts the response of polyQ assemblies to energy deprivation and to an autophagy-enhancing treatment in fibroblasts.

PolyQ assemblies circle the Golgi apparatus

HTTs are present on the surface of the Golgi, and mHTT impairs Golgi trafficking (del Toro et al., 2009; del Toro et al., 2006; DiFiglia et al., 1995). To our knowledge, the appearance of polyQ assemblies resembles Golgi ribbons. GM130, a Golgi marker, and 3B5H10 coimmunostaining revealed that polyQ assemblies circle flat Golgi ribbon/stacks; the flat Golgi stack stands between the internal space of two parallel spindles of polyQ assemblies, and an isthmus-like hollow in Golgi appeared at where polyQ circled Golgi cisternae (Fig. 4A), indicating the physical interactions of the Golgi apparatus with polyQ assemblies. The size of the Golgi apparatus in HD fibroblasts is smaller than that of healthy cells (Fig. 4B). Clathrin-coated vesicles sort on Golgi(Klumperman, 2011). Clathrin+ vesicles preferentially distributed along or conjugated with polyQ assemblies (Fig. 4C), and polyQ assemblies in HD fibroblasts attached fewer clathrin⁺ vesicles than polyQ assemblies in healthy fibroblasts (Fig. 4D). To localize the C-terminal of HTT in the polyQ assemblies, we stained the polyQ with a 3B5H10 antibody and the C-terminal with a C-terminal-specific antibody (LSB11011) and found that discontinuous C-terminal signals are present in the polyQ assemblies, and the C-terminal of HCC reside inside the assembly (Fig. 4E). We also noticed that some smaller polyQ assemblies which conjugate with abundant actin filiaments are present in the cytoplasm or ruffluring membrane of fibroblasts (Fig.S1A-C).

Figure 4.

Huntingtin-associated protein 40 (HAP40), a protein coevolved with HTT, binds to all three domains of HTT and stabilizes HTT(Guo et al., 2018; Harding et al., 2021). HAP40 resided on the corner of the network in polyQ assemblies, polyQ assemblies in HD, and healthy cells attached the same amount of HAP40 (Fig. 4F, G).

The Golgi apparatus fragments during cell mitosis and reassembles after mitosis (Wei and Seemann, 2009). PolyQ assemblies and Golgi stack/ribbon fragment during prometaphase (28 cells from 6 samples) and anaphase (23 cells from 6 samples), and reassemble at the late telophase (22 cells from 6 samples), and the fragmented Golgi in the mitotic cells conjugate with polyQ patches (Fig. 4H). Consistent with our previous reports(Liu et al., 2024), the polyQ assemblies in HD fibroblasts attached fewer ARF1, and Brefeldin A (BFA) treatments decouple polyQ assemblies from the Golgi apparatus in all fibroblasts (Fig. 4I-M). ARF1, which conjugates to the Golgi after converting its GDP to GTP-ATP in a GTP-ATP-dependent manner(D’Souza-Schorey and Chavrier, 2006), couples polyQ assembly to the Golgi (Liu et al., 2024). Thus, it is rational to hypothesize that polyQ assembly in HD fibroblasts is sensitive to energy deprivation. To test this hypothesis, we also immunostained HD patient fibroblasts and their healthy siblings that had been starved for 48 h with antibodies against 3B5H10 and GM130. Consistent with the changes of polyQ assemblies of HD and healthy cells under stress, the Golgi ribbons of HD fibroblasts loosened and fragmented, and most polyQ fragments/patches attached by Golgi fragments in glucose deprivation (Fig. S1F), whereas the structure of Golgi ribbons is stable, and the small Golgi fragments are reduced in healthy fibroblasts after culturing in a low glucose medium for 48h (Fig. S1F, G). In addition, we observed that HD patient cells with higher CAG repeat counts (55) had more fragmented Golgi than those with lower CAG repeat counts (49) after 48 h of glucose deprivation (Fig. S1H). The size of the Golgi in HD fibroblasts is reduced more quickly than that of its healthy siblings after 48 h glucose starvation (Fig. S1I). Together, our findings indicate that polyQ assemblies form a polyQ assembly-Golgi apparatus complex impaired by the presence of mHTT. To see in vivo Golgi dynamics in a living fibroblast, we continuously monitored fibroblasts (CAG: 49, 55, 19, and 19) stained by Golgi apparatus-selective dye after glucose starvation or H2O2 stimulation. The Golgi apparatus in HD patient fibroblasts is easier to fragment under glucose starvation than healthy fibroblasts (Fig. 4N, O), whereas the Golgi apparatus in patients and healthy fibroblasts uniformly fragments under H202 stimulation (Fig. S1D, E).

Impaired polyQ assembly in HD neurons is associated with deficient golgi function in cortical and striatal neurons

We generated iPSCs from the family fibroblasts and differentiated them into GABAergic projection neurons in 2D ^{(Ma et al., 2012)}. On day 90, long polyQ assemblies are present in striatal neuronal projections, and fragmented polyQ assemblies were deposited in the neuronal nucleus of an HD patient with CAG 59 (Fig. 5A). PolyQ almost completely occupied the nucleus of some HD striatal neurons around 100 days in all patient neurons (CAG 47-59), but PolyQ did not appear in the nucleus of healthy striatal neurons with over 90 days of culture (Fig. 5A, B). DIC images revealed that an intact, observable, deeply invaginated gorge where polyQ assemblies reside or traverse appeared in neurons, and 3D tomographic rendering showed that polyQ spindles also created a gorge on the nuclear surface (Fig. S2A), indicating that polyQ assemblies are an intact and rigid structure. In addition, the striatal neurons with polyQ nuclear accumulation showed a pale-stained nucleus, nuclear membranes pocked by polyQ, and nuclear fragmentations (Fig. S2B-F). The fragmentation of polyQ also existed in human fetal and child brain samples, but not in fresh human brain organoid samples (Fig. S3A-C). Despite the fragmentation of polyQ assemblies in brain tissues, the structure of polyQ assemblies and their interaction pattern with the nucleus are identical to those of the polyQ assemblies in human brain organoids (Fig. S3B; Fig.5A, B). The necrosis-dependent fragmentation of polyQ assemblies in brain organoids was common (Fig. S3D). The proteomic data also showed that nuclear activities in the HD striatal organoids (hStrO), including chromatin and DNA complex assembly, protein-DNA complex, DNA packaging complex, and nucleosome binding, were impaired in hStrO cultured for more than 60 days (Fig. S4A, B).

Fig. 5.

We also characterized the relationship between polyQ assemblies and the Golgi in neurons of hStrO and cortical organoids (hCOs) (Fig. S4A, C) (Chen et al., 2022). PolyQ assemblies include a flat Golgi stacks in hCOs, and the surface of Golgi in HD neurons were irregular (Fig. 5C). Multiple Golgi also resided in the polyQ assemblies of striatal neurons in hStrO, which traversed through the nuclear surface (Fig. 5D). PolyQ assemblies in the HD striatal neurons conjugate fewer Golgi than polyQ assemblies in healthy striatal neurons of hStrO (Fig. 5E). Clathrin+ vesicles preferentially distributed along or conjugated to polyQ assemblies in the HD and healthy neurons (Fig. 5F). Consistent with the results of fibroblasts, polyQ assemblies in HD neurons of hStrO attached fewer clathrin+ vesicles and ARF1 than that of healthy polyQ assemblies (Fig. 5F, G; Fig. S5A-C), and Golgi apparatus in HD neurons also attached fewer ARF1 puncta than that of healthy Golgi (Fig. S5D-F). In addition, ARF1 transcription levels in HD hStrOs are uniformly lower than those of healthy hStrO (60 and 80 days) Fig.S5G, H). Striatal projection neurons have extremely long dendrites and axon(Lanciego et al., 2012).

Extended polyQ assemblies in HD striatal neuronal projections in hStrO, which resemble human striatal neuronal polyQ assembly, attached fewer fragmented/small Golgi than healthy polyQ assemblies (Fig. 5H-I); small Golgi resided in the hole of rings or attached to the outer spines of long polyQ assemblies in hStrO neurons (Fig. 5N). Clathrin+ vesicles interact with the polyQ assemblies and mainly reside on the two sides of the assemblies in the neurons of hStrO (Fig. 5O).

To test the functional impact of deficient polyQ assemblies-Golgi complexes on HD neuronal activities, we performed HD-MEA recordings on 80-day hCOs. Control organoids exhibited robust, highly synchronized network bursts, resembling the coded activity patterns of developing human brains (Fig. 5K). In contrast, HD hCOs displayed erratic firing patterns (Fig. 5K).

Quantitative analysis revealed a significant reduction in both spikes per burst and the inter-burst interval (IBI) in HD hCOs (Fig. 5L). Notably, the elevated IBI CV in HD hCOs signifies a disruption in the periodicity and rhythmic regularity of spontaneous network-wide bursting. These findings demonstrate that polyQ assemblies-mediated pathologenesis disrupts the emergence of functional network oscillations and collective neuronal firing in cortical circuits.

Impaired Golgi function in HD striatal neurons

We performed single-cell RNA sequencing (scRNA-seq) analyses of human striatal organoids (hStrOs) derived from hESC-H9, hiPSC-8-12 (CAG 19), and HD hiPSC-7-2 (CAG 59) after 110 days of culture. Transcriptomic data of 38,716 single cells were obtained after quality control and filtering (Fig. 6A). Uniform manifold approximation and projection (UMAP) clustering identified four clusters in hStrOs that expressed cell-type-specific markers, including progenitor (VIM, NES, HES1), neuron (SYT1, DCX, STMN2), astrocytes (S100β, GFAP), and oligodendrocytes (OLIG2, SOX10) (Fig. 6B, C, and S6A). We further plotted the expression of brain region-specific markers in these clusters and found that both control and HD hStrOs are primarily composed of LGE progenitors and GABAergic neurons rather than cortical progenitors and glutaminergic neurons (Fig. S6C, D), consistent with our protocols for generating hStrOs(Chen et al., 2022). Based on the characteristics of Golgi in HD fibroblasts, we compared the transcriptomes of mature GABAergic neurons in the control and HD groups after sorting the GABAergic neuron population using GAD1 and GAD2, specific markers for mature GABAergic neurons (Fig. S6D-F). In the GABAergic neurons, the DEGs of HD hStrO extensively overlapped with those of hiPSC-derived hStrO (CAG 19) and hESC-derived hStrO, both for up-regulated (321 genes) and down-regulated genes (550 genes) (Fig. 6D). Up-regulated genes in GO terms are associated with nuclear activities, including ribonucleoprotein complex assembly, translational initiation, and regulation of translation, which might be related to the nuclear poking of polyQ assemblies or deposition of polyQ into the nucleus. Interestingly, the down-regulated genes in GABAergic neurons of HD hStrOs were notably enriched for GO terms associated with Golgi or Golgi-related functions, including Golgi vesicle transport, endoplasmic reticulum to Golgi vesicle−mediated transport, vesicle organization, endomembrane system organization, intracellular protein transport, protein localization to membrane translation (Fig. 6E). The down-regulated genes in HD hStrOs belong to Golgi matrix proteins, including GOLGA8A, a structurally similar protein of GM130 enriched in neurons, GOLGA4, and Golgi matrix-associated GTPs that include RAB6, ARL1/4C, and ARF1/3/4.

Fig. 6.

Onjisaponin F did not mitigate polyQ nuclear accumulation, nor did ASO treatment normalize HD neuronal firing rates

We selected iPSCs from two HD patients (CAG 59 and 49), one healthy sibling, and the H9 cell line, and differentiated them into striatal neurons in 2D to test the effects of onjisaponin F on striatal neurons. We exposed iPSC-derived GABAergic neurons to a single dose of 20 μM onjisaponin F for 24 h on days 42 and 60. After culturing for another 48 h, we fixed the striatal neurons, assessed polyQ changes, and found that onjisaponin F treatment on days 42 and 60 reduced polyQ assembly volume in HD striatal neurons but not in healthy neurons (Fig. 7A-C). We treated induced HD and healthy striatal neurons with 3 doses of onjisaponin F for 24h on days 50, 60, and 70. The immunostaining at day 90 showed that 3 doses of onjisaponin (10 µM) decreased the cytoplasmic and total volume of polyQ assemblies in HD striatal neurons but did not reduce the nuclear deposition of polyQ; the immunostaining at day 110 revealed that 3 doses of onjisaponin (20 µM) did not change the volume of polyQ assembly in healthy striatal neurons (Fig.7D-E).

Figure 7.

Given that ASOs have demonstrated efficacy in promoting mHTT degradation across experimental models(Kordasiewicz et al., 2012) and have reduced the polyQ assembly size in human fibroblasts. We utilized ASO as a benchmark to evaluate its therapeutic impact on the polyQ assembly-driven phenotypes in hCOs. Following treatment, we observed downward trends in both spikes per burst and the IBI (Fig. 7F, G). The IBI coefficient of variation (CV) also trended lower, suggesting a subtle shift toward improved firing patterns (Fig. 7F, G). However, these alterations failed to reach statistical significance across 10 days (Fig. 7F, G). The analysis of polyQ, Golgi, and ARF1 by immunostaining revealed that ASO treatment reduced the size of polyQ assemblies in HD-hCOs but did not alter the Golgi tomography and inclusion, Golgi/ARF1 conjugation to polyQ assemblies (Fig. 7H-K; Fig. S7A-D). These findings demonstrate that, under the tested conditions, the ASO treatment was insufficient to achieve a significant functional recovery of the HD-hCO network dynamics and rescue the Golgipathy in HD-hCOs.

Discussion

Here, we redefine the biology and structure of HTT polyQ aggregates, previously characterized as pathogenic aggregates(Truant et al., 2008). Using cells from a large, isolated HD family and state-of-the-art scanning, we show the polyQ assemblies of HTTs are large, dynamic, and well-organized structural networks composed of two or more long, parallel, and interfused spindles that encircle the Golgi apparatus, bind clathrin vesicles, and form a polyQ assembly-Golgi complex, which is structurally and dynamically coupled. The presence of mHTT shifts the response of polyQ assembly to energy deprivation and to autophagy-enhancing drugs, and destabilizes the Golgi apparatus by impairing ARF recruitment, inducing a Golgipathy in HD patients.

The Golgi ribbon comprises up to 100 flat stacks, composed of 4-11 stapled cisternae, in which newly synthesized soluble cargo proteins and lipids are processed and sorted, linked by tubular bridges in mammalian cells (Klumperman, 2011; Liu et al., 2021). The Golgi matrix, a ribosome-free protein network surrounding the cisternae, supports the stack structure(Ravichandran et al., 2020). The spatiotemporal relationships of polyQ assemblies with the Golgi apparatus, clathrin-coated vesicles, and ARF1 indicate that rigid polyQ assemblies serve as a structural component of the Golgi matrix, providing a scaffold for the Golgi apparatus and Golgi activities, including shaping, assembly, and sorting. ARFIP2, which interacts or conjugates multiple guanine nucleotide exchange factors (GEFs), including ARF1, ARF3, ARF5, and ARF6 (Wu and Zhou, 2009), directly conjugates with HTTs (Peters et al., 2002). Thus, it is rational to hypothesize that the ARFIP2-ARF complex mediates the relationship between polyQ assembly and the Golgi apparatus. The decoupling of the polyQ assembly-Golgi complex by BFA, but not during mitotic Golgi fragmentation, supports that the ARFIP2-ARF1 complex mediates this assembly.

Golgipathy, including structural and functional abnormality, is common in neurodegenerative diseases (El Ghouzzi and Boncompain, 2022; Klumperman, 2011; Liu et al., 2021; Passemard et al., 2017). The correct Golgi positioning is critical for cellular polarization and differentiation (Yadav and Linstedt, 2011); The precise localization of the Golgi apparatus mediates the simple spindle-like cells to form the dentate granule cells (DGCs) (Rao et al., 2018); GM130 deletion causes neurodegeneration and ataxia (Liu et al., 2017).

Pathogenic GORASP1 and ARF3 variants cause a neurodevelopmental disorder with neurosensory, neuromuscular, and skeletal abnormalities via affecting the cell cycle and glycosylation (Fasano et al., 2022; Lebon et al., 2025; Sakamoto et al., 2021). Previously, we reported that polyQ assemblies in HD cortical neurons altered corticogenesis by impairing ARF1 recruitment to the Golgi apparatus in the neuroepithelium(Liu et al., 2024).

The striatal projection neurons, which contain a prominent long polyQ assembly and polyQ nuclear accumulation, are lost early in HD patients (Walker, 2007b). HTT is a rigid protein (Dougan et al., 2009; Guo et al., 2018; Saudou and Humbert, 2016; Sivaramakrishnan et al., 2008; Tabrizi et al., 2020). The conjugation of ARF1 to the Golgi consumes ATP(D’Souza-Schorey and Chavrier, 2006). The glucose metabolism in HD patients is stronger than that of a healthy individual (Tang et al., 2013). Based on the body’s compensation mechanism, the deficient ARF1 conjugation to the Golgi apparatus in HD might contribute to increased brain metabolism. The polyQ assembly-Golgi apparatus complex in HD striatal neurons needs more ATP to stabilize and function. Onjisaponin F, a strong autophagy enhancer, preferentially reduces the polyQ assembly in HD fibroblasts and striatal projection neurons without changing the structure of polyQ assembly. However, it did not rescue nuclear polyQ deposition, a hallmark of HD pathology(DiFiglia et al., 1997). Based on the spatial relationship of polyQ assembly with the neuronal nucleus and the rigid nature of polyQ, it is rational to consider that the fragility of HTT assembly, rather than the size of polyQ assembly, might be the cause of nuclear accumulation in long-term cultured striatal neurons. A well-test ASO(Kordasiewicz et al., 2012) reduced the size of polyQ assemblies in HD cells but did not correct aberrant neuronal firings in HD cortical neurons, indicating that the structure of polyQ assembly-Golgi complex in HD cells drives aberrant neuronal activity. However, the size of polyQ assemblies does not matter.

Our study reveals the structure of polyQ aggregates and their spatial-temporal relationship with the Golgi apparatus, and demonstrates that mHTT destabilizes the polyQ assembly-Golgi complexes, dismantles the recruitment of Golgi activity-related proteins, leading to Golgipathy in HD. Thus, improving Golgi function and stabilizing polyQ assembly in HD might represent another approach to treating HD.

The limitation of this study

While our study provides a new structural framework for Huntingtin biology, several limitations remain. First, the specific mHTT-to-wtHTT ratio likely influences the degree of functional deficiency in the polyQ-Golgi complex, yet the precise incorporation pattern of mHTT into the polyQ assemblies remains to be determined. Second, while we characterized the fragility of polyQ assemblies under stress in HD cells, it is unclear how this structural instability relates to the subsequent nuclear deposition of polyQ. Future studies utilizing super-resolution and longterm imaging in the living HD cells may be required to visualize the real-time assembly and transition of these complexes into nuclear inclusions.

Materials and methods

Ethics

The ethics of this HD work was approved (No. 28) by the Ethics Committee of the Institutes of Biomedical Sciences at Fudan University. All HD family members who participated in this study signed a written consent form translated into Mongolian and consent to enter this study and authorisation for publication has been obtained. Dr. Li Li collected the human brain samples.

The ethics approval number was obtained from the Huashan Hospital Ethics Committee (KY2018-310.01).

Human fibroblast and blood sample collection

After obtaining consent, a surgeon collected skin samples from the upper arm using a Skin Punch (diameter, 0.3 x 0.3 cm). After collecting samples, a nurse and doctor visited the patients daily for 7 days to prevent inflammation. The tissues were harvested into a tube with culturing medium supplied with antibiotics. After being transported to a lab, tissues were moved to a Biosafety Cabinet, sliced into smaller fragments with surgical knives, and then attached to dishes with forceps and cultured in a small amount of medium. After fully attaching to dishes, the tissues were placed in sufficient medium and cultured continuously for several weeks to obtain sufficient fibroblasts.

Data availability

The sequencing datasets generated and/or analyzed during this study are available in the Sequence Read Archive (SRA) and Gene Expression Omnibus (GEO) with the following accession numbers for anyone: Day 60 and Day 80 MSN in hStrO for bulk RNA-seq (PRJNA760260), Day 110 MSN culture in hStrO for scRNA-seq (GSE184255).

Supplementary Materials and Methods

Primer and sequencing

A nurse collected blood from the upper arm vein, and a Blood DNA Isolation Kit extracted blood DNA. After PCR amplification, the mutant band separates from the wild-type band on a 4% Agarose gel, and the band is then cut and extracted using a DNA Gel Extraction Kit (GenElute™, Sigma-Aldrich). The extract was sequenced by the Sanger method.

The sequencing primers:

A1: 5’-CAGAGCCCCATTCATTGCC-3’

B1: 5’-TGAGGAAGCTGAGGAGGC-3’

Glucose starvation assays for polyQ assemblies and Golgi

Cultured the fibroblasts in a 24-well plate with cover glasses (Φ20mm). After two days, the medium was replaced by low-glucose medium without FBS. The cells in low-glucose medium were fixed at 24 h, 48 h, and 72 h, and stained with the 3B5H10 antibody and phalloidin. At 72 h, the medium was replaced with high-glucose medium lacking FBS and cultured for an additional 48 h, then fixed and stained with 3B5H10 antibody and phalloidin. ImageJ software was used to measure the polyQ assemblies. Imaris was used to render the image. A group of fibroblasts was stained by GM130 antibodies at 48 h after supplying low-glucose medium for Golgi changes.

Onjisaponin and ASO treatments

The sequence of HuASO: 5’-CTCAGtaacattgacACCAC-3’. 5mg HuASO were synthesized by WuXi AppTec. ASOs were solubilized in 0.9% sterile saline solution or PBS. Onjisaponin F was purchased from Jindu Biotech of Shanghai and dissolved in DMSO to a concentration of 100 mM.

Brain organoids (day 100) were treated continuously for 10 days with ASOs at final concentrations of 40 µg/mL. And fibroblasts were treated continuously for 5 days with ASOs at final concentrations of 20, 40 µg/mL. Vehicle (PBS) were used as controls. ASOs were solubilized in sterile PBS and supplemented directly into the culture medium, with media and ASO replenishment every 24 hours.

Immunostaining

Cells, tissues, or organoids were fixed in fresh ice-cold 4% PFA for 15-30 minutes and overnight, respectively. After fixation, immunostaining was performed as previously described¹. Primary antibodies were used as follows: anti-Tuj-1 antibodies (mouse, Sigma, T8860, 1:5000; rabbit, Covance Research Products, PRB-435P, 1:5000), 3BH10 antibody (Mouse, Sigma, P1874, 1:10000), anti GABA antibodies (rabbit, Sigma, A2052, 1:5000; mouse; Sigma, A0310, 1:500), Clathrin antibody (Rabbit, Abcam, ab21679, 1:1000), GM130 antibody (Rabbit, Abcam, ab52649, 1:1000), IHC-plus™ Polyclonal Rabbit anti-Human HTT / Huntingtin Antibody (C-Terminus, LS-B11011, 1:200). Secondary antibodies used were as follows: Alexa Fluro 488 Donkey anti-mouse IgG (Invitrogen, Molecular Probe, A21202, 1:1000), Alexa Fluro 594 Donkey anti-mouse IgG (Invitrogen, Molecular Probe, A21203,1:1000), Alexa Fluro 594 Donkey anti-rabbit IgG(Chemicon, A21207,1:1000), Alexa Fluro 488 Donkey anti-rabbit IgG (Invitrogen, Molecular Probe, A21206,1:1000), Alexa Fluro 594 Donkey anti-goat IgG (Invitrogen, Molecular Probe, A11058,1:1000) Cy5 AffiniPure Donkey Anti-Goat IgG (H+L) (Jackson, 705-175-147,1:300), Cy5 AffiniPure Donkey Anti-Rabbit IgG (H+L) (Jackson, 711-175-152,1:300), Alexa Fluo 633 Phalloidin (Invitrogen, A22284,1:200), Mito-Tracker™ Green (Invitrogen, M7514), ER-Tracker™ Green (Invitrogen, E34251), DAPI (Sigma, D9542), Phalloidin Alexa-633 or 488 (ThermoFisher, A12379 or A22284, 1:1000), GOLGI ID® Green assay kit (ENZO, ENZ-51028-K100).

Golgi apparatus-selective staining and imaging in live cells

Replated fibroblasts in glass-bottomed 35mm dishes (biosharp, BS-20-GJM). The Golgi apparatus in live cells was labeled with the GOLGI ID Green assay kit (ENZO, ENZ-51028-K100), a Golgi-selective dye. The Assay Solution and the Detection Reagent according to the manufacturer’s guidelines. After preparation, these agents were fully mixed. After washing the cells with Assay Solution, a sufficient volume of Detection Reagent was dispensed to cover the monolayer cells, and the cells were incubated for 30 minutes at 37°C. Before imaging, the cells were washed 3 times and incubated in medium at 37°C for another 30 minutes. For imaging, sequential images were captured from multiple ROIs (regions of interest) in dishes. The time intervals were 30 min, 60 min, 120 min, and 240 min.

2D Neural differentiation

The cells were differentiated according to the neural differentiation protocol described ¹. Briefly, dissociated WT-iPS and HD-iPS cells were plated on Matrigel (BD) at a density of 3.5×104 cells/cm2 and maintained in E8 medium (Thermo Fisher Scientific) for 4-5 days until they reached 90% confluence.

After digestion with EDTA, 9000 cells were plated in each well of U-shaped 96-well plates (SUNILON). On the first day, we maintained these cells in E8 and hNM (1:1 mixture). On the second day, the medium was replaced with hNM. From day 0 to day 6, the medium containing 2 μMol/L SB431542 (Tocris) and 0.3 μMol/L LDN193189 (BMP inhibitor, Tocris) was used for culture. These cells were plated in a well on day 7, and 200 ng/ml SHH or 0.65 μM Purmorphamine was supplied on day 10 for the differentiation of GABAergic neurons. The cells were maintained in the same medium for 2 weeks. The medium without SHH or Purmorphamine is used for differentiating glutamate neurons. At 15 days of differentiation, the rosettes were collected and cultured in a suspension condition for 25 days. On day 25, these cells were plated on a cover glass double-coated by polyonithin and laminin after digesting with accutase and then supplemented by a neural induction medium (composed of Neurobasal, N2 100×, B27 50×) (Invitrogen), supplemented with 10 ng/ml BDNF (Peprotech), 10 ng/ml GDNF (Peprotech) and 10 ng/ml IGF (Peprotech).

Human brain organoid construction

Brain organoids were constructed according to our protocol 2. Briefly, ES/iPS colonies were dissociated into single cells using Accutase (Gibco, A1110501). 9,000 dissociated cells were plated into each well of a V-bottom ultra-low-attachment 96-well plate (Sumitomo Bakelite, MS-9096VZ). Induction media contained Human Neural differentiation Medium and E8 media (1:1), supplemented with the two SMAD inhibitors 0.3μM LDN-193189 (STEMGENT, 040074), 2μM SB431542 (Ametek Scientific, DM-0970), and 10 μm Y-27632 (APE.BIO, A3008). On the sixth day in suspension, hCOs were transferred to neural differentiation media containing a 1:1 mixture of DMEM-F12 media and Neurobasal media supplemented with 1% (v/v) MEM-NEAA (Gibco, 11140-050), 1% (v/v) GlutaMAX(Gibco, 35050-061), 1% (v/v) N2 supplement (Gibco, 17502-048), 2% (v/v) B27 supplement (Gibco), 1% (v/v) Penicillin/Streptomycin (Gibco, 2051357). Growth factors 10ng/ml EGF (R&D Systems, 236-EG) and 20 ng/ml bFGF (R&D Systems, 233FB/CF) were used from day6 to day23. Growth factors, the 20 ng/ml BDNF (Peprotech, AF450002) and 10 ng/ml GDNF (Peprotech, AF450010) were used from day 24 to day 45.

Interfacing the brain organoid with HD-MEA

To facilitate robust hCO-electrode coupling, high-density microelectrode array (HD-MEA) chips (Maxlab Biosystems, Switzerland) were first functionalized. The sensing area was treated with a 0.07% (v/v) poly(ethylenimine) (PEI; Sigma-Aldrich) solution followed by a coating of laminin. For the interfacing of hCOs, we adapted the manufacturer’s protocols for brain slices. Individual hCOs were meticulously transferred to the center of the coated electrode arrays. To ensure mechanical stability and intimate contact between the basal surface of the organoid and the electrodes, a small droplet of Matrigel (Corning) was applied to secure the tissue. The chips were then lidded and incubated at 37 °C with 5% CO2 for 30 minutes to allow for matrix gelation. Subsequently, the culture was replenished with BrainPhys neuronal medium (StemCell Technologies) supplemented with the following reagents: 2% (v/v) NeuroCult SM1 Neuronal Supplement (StemCell Technologies), 1% (v/v) N2 Supplement-A (Gibco), 1x GlutaMAX (Gibco), 20 ng/ml recombinant human BDNF, 20 ng/ml recombinant human GDNF (PeproTech), 20 ng/ml NT-3 (PeproTech), 1 mM dibutyryl-cAMP (Sigma-Aldrich), and 1x Antibiotic-Antimycotic (Gibco). The recording medium was changed every 2–3 days.

Electrophysiological recordings

To probe network activity in hCOs, longitudinal HD-MEA recordings were used. The spontaneous activity was recorded sequentially in blocks across the entire array (30 seconds per recording configuration, each comprising 1020 recording electrodes). Signals were sampled at 20 kHz for all recordings and saved in an HDF5 file format. Spike detection was performed using a threshold set to 5× the root-mean-square (RMS) of the noise in the band-pass-filtered signal. After the activity scan, the readout electrodes were selected depending on the requirements of each analysis (single-cell or network-related features). All HD-MEA recordings of hCOs started at least 7 days after plating. The reported electrophysiological results are based on data obtained from n=3 hCOs (CTR group), n=3 hCOs (HD group), and n=3 hCOs (HD group with ASO treatment).

Preparation and data analysis of a single-cell RNA-sequencing library

110 days differentiated hStrOs from a HD patient (CAG 59), a healthy sibling, and H9 in suspension culture are prepared for analysis. A sterile dish with 10 ml 1x Dulbecco’s Phosphate-Buffered Saline (DPBS; Thermo Fisher, 14190144) on ice is used to transport samples after removing the residual solution. Neurosphere Dissociation Kit (Miltenyi, 130-095-943) used to create single-cell suspensions. Dissociated cells were rinsed 1x DPBS containing 2% FBS. 0.4% Trypan blue staining (Thermo Fisher, 14190144) is used to assess viability on the Countess® II Automated Cell Counter (Thermo Fisher).

10x library preparation and sequencing

Beads with unique molecular identifier (UMI) and cell barcodes were loaded close to saturation, so that each cell was paired with a bead in a Gel Beads-in-emulsion (GEM). After exposure to cell lysis buffer, polyadenylated RNA molecules hybridized to the beads. Beads were retrieved into a single tube for reverse transcription. For cDNA synthesis, each cDNA molecule was tagged at the 5’ end (i.e., the 3’ end of a messenger RNA transcript) with a UMI and a cell label indicating its cell of origin. Briefly, 10× beads that were then subject to second-strand cDNA synthesis, adaptor ligation, and universal amplification. Sequencing libraries were prepared from randomly interrupted whole-transcriptome amplification products to enrich for the 3’ ends of transcripts linked to the cell barcode and UMI. All remaining procedures, including library construction, were performed according to the manufacturer’s standard protocol (Chromium Single Cell 3ʹ v3). A High Sensitivity DNA Chip (Agilent) on a Bioanalyzer 2100 and the Qubit High Sensitivity DNA Assay (Thermo Fisher Scientific) quantified NovaSeq6000 (Illumina) sequencing libraries (2x150 chemistry).

Single cell RNA-seq data processing and analysis

Cell Ranger 4.0 pipeline with default and recommended parameters processed the reads. Gene-Barcode matrices were generated for each sample by counting UMIs and filtering out non-cell-associated barcodes, and a gene-barcode matrix containing the barcoded cells and gene expression counts was generated. After obtaining the single-cell expression matrix, we used the Seurat (v4.0.5)² R toolkit to perform normalization, clustering, dimensionality reduction, and visualization. Cells that had the number of detected genes less than 200 and the percentage of mitochondrial genes lower than 10% were removed. Gene counts were then log-transformed and scaled, and 3000 shared highly variable genes were identified using the FindVariableFeature function. We applied the Runharmony function with the Seurat-based principal component analysis (PCA) reduction method in the Harmony (v0.1.0)³ package to correct batch effects and generate low-dimensional matrices. A nearest-neighbor graph was calculated using FindNeighbors and clustered using FindClusters, followed by visualization of the cell clusters using uniform manifold approximation and projection (UMAP). We used the well-known marker genes to assign these clusters to corresponding cell types. Differentially expressed genes(|log2FC|>0.25 and adjusted p value<0.01) among GABAergic neurons (CAG 59 vs CAG 19, CAG 59 vs H9) were identified by the Seurat function FindMarkers.

Functional enrichment

Enrichment GO categories for significantly differentially expressed genes were identified by Metascape (https://metascape.org/gp/)⁴, which utilizes hypergeometric test and Benjamini-Hochberg p-value correction algorithm to identify acceptable ontology terms. Adjusted p value of GO biological processes <0.01 was considered to be significant enrichment terms. The gene set enrichment score of Golgi vesicle transport (GO:0048193) was calculated by AUCell(v.1.14.0)⁵. The AUCell_buildRankings and AUCell_calcAUC algorithms were used for ranking model building and calculating "Area Under the Curve" (AUC) scores.

LC-MS analysis

The patients’ iPSCs were differentiated for 50 or 100 days, and the samples were then sent to the core facility of the State Key Laboratory of Genetic Engineering for analysis. The FASP digestion was adapted for the following procedures in Microcon PL-10 filters. After a three-time buffer displacement with 8 M Urea and 50 mM ABC, pH 8.5, proteins were reduced by 10 mM DTT at 37 °C for 30 min, followed by alkylation with 30 mM iodoacetamide at 25 °C for 45 min in a light-proof condition. Digestion was performed with trypsin (enzyme/protein ratio 1:50) at 37 °C for 12 h, after which the sample was rinsed with 20% ACN and buffer displaced three times with digestion buffer (10 mM ABC). After digestion, the solution was filtered, the filter rinsed twice with 15% ACN, and the filtrates were pooled and vacuum-dried. LC-MS analysis was performed using a nanoflow EASYnLC 1200 system (Thermo Fisher Scientific, Odense, Denmark) coupled to an Orbitrap Exploris 480 mass spectrometer (Thermo Fisher Scientific, Bremen, Germany). A one-column system was adopted for all analyses. A home-made C18 analytical column (75 µm i.d. × 25 cm, ReproSil-Pur 120 C18-AQ, 1.9 µm (Dr. Maisch GmbH, Germany) was used to analyze samples. The mobile phases are composed of Solution A (0.1% formic acid) and Solution B (0.1% formic acid in 80%ACN).

The derivatized peptides were eluted using the following gradients: 5–8% B in 2 min, 8–44% B in 98 min, 44–60% B in 3 min, 60–100% B in 2 min, 100% B for 10 min, at a flow rate of 200 nL min. High-field asymmetric-waveform ion mobility spectrometry (FAIMS) was enabled during data acquisition, with compensation voltages set to −45 and −65 V. MS1 data were collected in the Orbitrap (60,000 resolution). Charge states 2-7 were required for MS2 analysis, and a 45 s dynamic exclusion window was used. Cycle time was set at 1.5 s. MS2 scans were performed in the Orbitrap with HCD fragmentation (isolation window 1.6; 15,000 resolution; NCE 30%).

Extended Data

Figure S1.

Figure S2.

Figure S3.

Figure S4.

Figure S5.

Figure S6.

Figure S7.

Table S1.

Additional information

Author contributions

HS and LM set up this study. HS and LM performed most of the experiments; HS analyzed the data and prepared the manuscripts; XC, WY, YL, BH, and JX performed image analysis and immunostaining. XC, LY, and LM constructed organoids.

Financial supports

This work was supported by the National Natural Science Foundation of China (grant no. 32370852 and U24A2014 to LM) and the National Key Research and Development Program of China (grant no. 2021YFA1101302 to LM)

Funding

MOST | National Natural Science Foundation of China (NSFC) (2370852)

• Lixiang Ma

MOST | National Natural Science Foundation of China (NSFC) (U24A2014)

• Lixiang Ma

MOST | National Key Research and Development Program of China Stem Cell and Translational Research (Stem Cell and Translational Research) (2021YFA1101302)

• Lixiang Ma