Rescue of GM3 synthase deficiency by spatially controlled, rAAV-mediated ST3GAL5 delivery

GM3 synthase deficiency (GM3SD) is an infantile-onset epileptic encephalopathy syndrome caused by biallelic loss-of-function mutations in ST3GAL5. Loss of ST3GAL5 activity in humans results in systemic ganglioside deficiency and severe neurological impairment. No disease-modifying treatment is currently available. Certain recombinant adeno-associated viruses (rAAVs) can cross the blood-brain barrier to induce widespread, long-term gene expression in the CNS and represent a promising therapeutic strategy. Here, we show that a first-generation rAAV-ST3GAL5 replacement vector using a ubiquitous promoter restored tissue ST3GAL5 expression and normalized cerebral gangliosides in patient-derived induced pluripotent stem cell neurons and brain tissue from St3gal5-KO mice but caused fatal hepatotoxicity when administered systemically. In contrast, a second-generation vector optimized for CNS-restricted ST3GAL5 expression, administered by either the intracerebroventricular or i.v. route at P1, allowed for safe and effective rescue of lethality and behavior impairment in symptomatic GM3SD mice up to a year. These results support further clinical development of ST3GAL5 gene therapy.


Introduction
ST3GAL5 encodes GM3 synthase (ST3GAL5; also called GM3S and SIAT9), the rate-limiting enzyme for production of all a-and b-series gangliosides normally enriched in mammalian brain ( Figure 1A) (1)(2)(3)(4)(5)(6). Biallelic ST3GAL5 loss-of-function variants result in systemic ganglioside deficiency, an infantile-onset neurodevelopmental syndrome characterized by intractable epileptic encephalopathy, auditory and visual impairment, global psychomotor delay, extrapyramidal movements, and untimely death. A number of pathogenic variants have been linked to the GM3 synthase deficiency syndrome (GM3SD) in populations worldwide (2,7). Within Old Order Amish communities of North America, the incidence of GM3SD is enriched to approximately 1 per 1200 births due to a severe ST3GAL5 c.862C > T (p.Arg288Ter) founder variant that abrogates ST3GAL5 activity and results in absence of GM3 and its most important downstream products: GM1, GD1a, GD1b, and GT1b (3).
GM3 and derivative a-and b-series gangliosides are expressed in cytosolic membranes of all mammalian cells, where they contribute to microdomain architecture and activity of intramembrane proteins (cis interactions), as well as ligand binding and intercellular contacts (trans interactions) (8). Disrupted ganglioside synthesis results in neurotoxicity from multiple overlapping mechanisms, including altered receptor interactions, abnormal cellular membrane dynamics, and reduced mitochondrial membrane potential and oxygen consumption (4,9). Oral ganglioside replacement therapy via a powdered buttermilk supplement (G500; Auckland, New Zealand) may transiently improve growth and development during infancy, but low enteral absorption of gangliosides and their restricted transit across the blood-brain barrier (BBB) may ultimately limit the utility of dietary therapy, leading to treatment failures and loss of long-term efficacy (10). At present, no other effective treatment is available for GM3SD.
The development of novel and robust therapeutic modalities requires testing in proper animal models that genetically and phenotypically recapitulate human GM3SD. Homozygous St3gal5 -/mice exhibit GM3 synthase deficiency (GM3SD) is an infantile-onset epileptic encephalopathy syndrome caused by biallelic loss-of-function mutations in ST3GAL5. Loss of ST3GAL5 activity in humans results in systemic ganglioside deficiency and severe neurological impairment. No disease-modifying treatment is currently available. Certain recombinant adeno-associated viruses (rAAVs) can cross the blood-brain barrier to induce widespread, long-term gene expression in the CNS and represent a promising therapeutic strategy. Here, we show that a first-generation rAAV-ST3GAL5 replacement vector using a ubiquitous promoter restored tissue ST3GAL5 expression and normalized cerebral gangliosides in patient-derived induced pluripotent stem cell neurons and brain tissue from St3gal5-KO mice but caused fatal hepatotoxicity when administered systemically. In contrast, a second-generation vector optimized for CNS-restricted ST3GAL5 expression, administered by either the intracerebroventricular or i.v. route at P1, allowed for safe and effective rescue of lethality and behavior impairment in symptomatic GM3SD mice up to a year. These results support further clinical development of ST3GAL5 gene therapy.
JCI Insight 2023;8(9):e168688 https://doi.org/10.1172/jci.insight.168688 a-and b-series ganglioside deficiency, insulin hypersensitivity, and hearing loss, but in contrast to human patients, do not suffer from early mortality or clinically relevant neurological disease (11,12). In mice, biochemical defects caused by ST3GAL5 deficiency seem to be compensated by remaining minor gangliosides species, resulting in minimal physiological alternations ( Figure 1A). However, mice with KO of both St3gal5 and B4galnt1 are unable to synthesize any gangliosides and more closely mirror clinical hallmarks such as CNS pathology, developmental delay, motor impairment, and early death (Supplemental Figure 1; supplemental material available online with this article; https://doi.org/10.1172/jci.insight.168688DS1) (13). Thus, St3gal5 single KO and St3gal5/B4galnt1-double KO mice serve as complementary models for evaluating the actions of novel therapies (Table 1).
Because GM3SD is a monogenic loss-of-function disease, gene replacement therapy may be a promising approach. Recombinant adeno-associated viruses (rAAVs) have emerged as powerful gene delivery tools for the treatment of monogenic diseases and, to date, have been tested in 17 clinical trials targeting CNS disorders (14,15). An ideal rAAV vector should deliver its therapeutic cargo into specific target cells to restore an appropriate spatial, quantitative, and temporal pattern of protein expression. However, a limiting factor for successful and safe rAAV-mediated gene therapy is the broad tropism of common AAV serotypes. The naturally isolated AAV serotype 9 (AAV9) can cross the BBB and transduce neural tissues, but it also efficiently transduces multiple peripheral tissues such as liver, skeletal muscle, and heart (16,17). Hepatotoxicity after high-dose systemic AAV9 delivery (18,19), including several patient deaths due to acute liver failure (20,21), has raised legitimate concerns about the overall safety of gene therapy. Therefore, regulating the tissue specificity of transgene delivery and expression may preserve the therapeutic benefits of rAAV while minimizing associated risks. Currently, this can be approached through a combination of variables, including route of administration (e.g., regional tissue injection), use of CNS-favorable viral capsids, and inclusion of cell type-specific promoters and tissue detargeting miRNA binding sites within the therapeutic genome sequence.
In this study, we first examined ST3GAL5 replacement cassettes for their ability to reconstitute gangliosides in cortical neurons produced from GM3SD patient-derived induced pluripotent stem cells (iPSCs). We then administered the AAV9 vectors by intracerebroventricular (ICV) injection to St3gal5 -/and St3gal5 -/-/B4galnt1 -/mice. Treatment with rAAV9-ST3GAL5 extended survival, restored CNS ganglioside production, improved growth, and partially rescued motor function of experimental animals. When delivered systemically, however, this therapy led to hepatic injury and death caused by high off-target ST3GAL5 expression in the liver. We therefore designed a second-generation rAAV9 vector using a CNS-specific promoter (human Synapsin1 [Syn1]) in combination with liver-specific miRNA targeting sequences (miR-122) to optimize both transcriptional and post-transcriptional regulation (22,23). In GM3SD mouse models, this strategy eliminated liver toxicity while preserving neurotherapeutic effects.
Finally, we examined if data from the St3gal5 -/-/B4galnt1 -/--double KO mouse underrepresented the therapeutic potential of GM3SD gene therapy, because it might apply to humans and, therefore, we co-injected St3gal5 -/-/B4galnt1 -/mice with both ST3GAL5 and B4GALNT1 rAAV vectors. Vector co-injection completely eliminated behavior impairments in St3gal5 -/-/B4galnt1 -/mice. Overall, our study supports potentially translating safe and effective ST3GAL5 gene therapy for clinical development.

Results
ST3GAL5 transgene design and in vitro expression. ST3GAL5-1a-2 (NM_003896) is the most abundant mRNA among 4 ST3GAL5 mRNA variants in the human brain ( Figure 1B) (24)(25)(26); we thus focused on this variant for further vector development. The first AUG start codon in ST3GAL5-1a-2 is in a weak translation initiation context (AUUAGUAUGC). Most ribosomes, therefore, skip the first AUG and recognize either of 2 downstream AUG sequences as the start codon (27,28). As a result, 3 ST3GAL5 protein isoforms differing in their N-termini are produced ( Figure 1B). Because of the lack of knowledge about their physiological roles, we designed and tested different human ST3GAL5 replacement constructs carrying each ORF ( Figure 1B). The codon-optimized transgenes were cloned into a ubiquitous expression cassette driven by chicken β-actin (CB) promoter with an intron ( Figure 1C) and transfected into HeLa cells to confirm protein expression. We found that expression of the shortest construct (ORF3) was weak, and adding Kozak sequence GCCACC (construct KORF3) greatly enhanced expression ( Figure 1D).
We next evaluated whether these ST3GAL5 constructs could function in ganglioside synthesis in cultured neurons. To this end, we differentiated normal (ST3GAL5 +/+ ) and patient (ST3GAL5 E332K/E332K ) iPSCs into cortical neurons ( Figure 2, A and B) and infected them with lentiviral vectors expressing different ST3GAL5 isoforms. Although major brain gangliosides (GD1a, GD1b, and GT1b) were absent in untreated patient neurons, they were restored after transduction of any ST3GAL5 isoform ( Figure 2C) (29). We focused on the KORF3 transgene design in future development, because M3-ST3GAL5 is the most stable isoform (24) and its small gene size (1095 base pairs) is amenable to the self-complementary AAV vector design that can further enhance vector potency when packaging capacity is limited (<2.5 kb). ICV injection of rAAV9-ST3GAL5.v1 improved biochemical and phenotypic abnormalities in GM3SD mouse models. Encouraged by in vitro results, we generated an AAV9 vector carrying a CB promoter-driven ST3GAL5 construct (rAAV9-ST3GAL5.v1) to assess therapeutic efficacy in mice after in vivo delivery. We first treated St3gal5 -/mice by unilateral ICV injection of 3 × 10 10 genome copies (gcs) per pup at P1, and we euthanized animals 4 weeks after injection ( Figure 3A). The ST3GAL5 transgene was detected in brain, liver, and heart, where it induced tissue mRNA expression in excess of endogenous levels ( Figure 3B). GM3 and its derivatives (GM2, GD1a, and GD1b) were undetectable in the St3gal5 -/brain. ICV injection of rAAV9-ST3GAL5.v1 restored these a-and b-series gangliosides to WT levels, concomitant with substantial clearance of lactosylceramide (LacCer), the proximate substrate for GM3 synthase (Figure 3, C and D, and Supplemental Figure 2). However, ganglioside deficiency persisted in serum, likely because vector genomes delivered by ICV injection did not sufficiently penetrate peripheral tissues (Supplemental Figure 3A). Note that total GM1 (consisting of GM1a and GM1b) was not affected by the St3gal5 KO (Supplemental Figure 3B). Because gangliosides are 10-to 30-fold more abundant in human brain than in any other tissue (30), we postulated that ganglioside restoration in CNS was the key to preventing neurological morbidity in GM3SD animals.
St3gal5 -/mice have hearing loss but do not exhibit many of the other neurological deficits characteristic of GM3SD in humans. For testing vector effectiveness, we therefore used the St3gal5 -/-/B4galnt1 -/mouse model, which models key aspects of the human GM3SD phenotype, including reduced survival, growth failure, motor impairments, and neuropathology. These animals were benchmarked to St3gal +/-/B4galnt1 -/mice, which were from the same litter as St3gal5 -/-/B4galnt1 -/and did not exhibit significant neurological morbidity during the period of experimental observation (Supplemental Figure 4A). ICV administration of rAAV9-ST3GAL5.v1 to newborn St3gal5 -/-/B4galnt1 -/pups (3 × 10 10 gcs; P1) extended their survival up to 300 days (median survival: untreated, 18 days; treated, 56 days) (Supplemental Figure 4B), improved growth, and partially restored motor function as assessed by the negative geotaxis test (Supplemental Figure 4, C-F). Taken together, these results demonstrated that neonatal ICV injection of rAAV9-ST3GAL5.v1 was well tolerated and could restore endogenous cerebral ganglioside production but did not fully alleviate GM3SD disease in St3gal5 -/-/B4galnt1 -/mice.
JCI Insight 2023;8(9):e168688 https://doi.org/10.1172/jci.insight.168688 All treated St3gal5 -/and St3gal5 +/+ mice unexpectedly died within 3 days after injection ( Figure 4A). To understand why, we injected WT pups with therapeutic vector, capsids containing cDNA for EGFP, empty capsids (AAV9.empty), or PBS ( Figure 4B). We found that in WT mice systemically treated with rAAV9-ST3GAL5.v1, expression of ST3GAL5 in liver was elevated more than 100-fold relative to mouse endogenous St3gal5 level ( Figure 4C), accompanied by activation of cellular death and defense response ( Figure 4, D and E). We confirmed RNA-Seq results with quantitative PCR (qPCR) and ELISA, which revealed consistent activation of pro-inflammatory cytokines (TNF-α, IL-1α, CCL2, and CCL3) (Supplemental Figure 5, A and B). These molecular changes were accompanied by cellular liver pathology, including hepatocyte swelling ( Figure 4F) and cell death ( Figure 4G). Moreover, the toxicity was also seen in liver of St3gal5 -/mice receiving rAAV9-ST3GAL5.v1 (Supplemental Figure 6, A-D) but not in other peripheral organs (Supplemental Figure 6, E and F). We thus speculated that overexpression of hST3GAL5 transgene in liver and its attendant cytopathic effects played a direct role in lethal toxicity of systemically administered rAAV9-ST3GAL5.v1.
Optimized ST3GAL5 vector construct with spatial regulation eliminated liver toxicity associated with systemic administration. We reasoned that CNS-restricted and liver-detargeted expression of an ST3GAL5 transgene might preserve therapeutic efficacy while eliminating hepatotoxicity. We therefore designed a spatially regulated expression cassette that included human Syn1 promoter (35,36) to drive neuronal expression at the transcriptional level, combined with miR-122 binding sites in the 3′UTR, which silence transgene expression in hepatocytes at the post-transcriptional level ( Figure 4C) (22,23). We named this refined construct ST3GAL5.v2 and packaged it into AAV9. Following the same P1 facial vein injection paradigm in WT pups, transgene expression from rAAV9-ST3GAL5.v2 was greatly reduced in liver tissue, and all animals survived with no evidence of liver inflammation, cytopathology, or transcriptomic derangements ( Figure 4 and Supplemental Figure 5).
Interestingly, we noticed that packaging the first-generation construct (rAAV9-ST3GAL5.v1) consistently resulted in low titers (1 × 10 12 to 4 × 10 12 gcs/mL), likely due to transgene toxicity in HEK293 cells during the manufacturing process. In contrast, rAAV9-ST3GAL5.v2 was routinely produced at higher titers of 0.8 × 10 13 to 1.5 × 10 13 gcs/mL (Supplemental Figure 7). We therefore concluded that by tuning tissue specificity, the optimized second-generation construct design eliminated both hepatotoxicity and the manufacturing bottleneck, serving as a clinically translatable candidate for studies that followed.
ICV injection of rAAV9-ST3GAL5.v2 improved biochemical and phenotypic abnormalities in GM3SD mouse models. We cloned the ST3GAL5.v2 construct in self-complementary (sc) configuration to facilitate faster and stronger expression as compared with the single-stranded (ss) transgene ( Figure 5A and Supplemental Figure 8) (37,38). After P1 ICV injection in St3gal5 -/mice, scAAV9-and ssAAV9-ST3GAL5.v2 led to comparable levels of transgene expression in hippocampus 4 weeks after injection, whereas scAAV9 slightly outperformed ssAAV9 in the cerebral cortex ( Figure 5B). Both vectors normalized the brain ganglioside profile in St3gal5 -/mice up to 12 weeks after injection ( Figure 5, C and D), although neither corrected circulating gangliosides, consistent with the neuron-specific expression design (Supplemental Figure 9).
Nevertheless, St3gal5 -/-/B4galnt1 -/mice that survived after scAAV9-ST3GAL5.v2 treatment continued to exhibit hindlimb clasping, a sign of motor dysfunction not seen in their St3gal5 +/-/B4galnt1 -/littermates ( Figure 6H). This might reveal a fundamental limitation of the sequential double KO animal model; that is, isolated KO of B4galnt1 in mice has independent neuropathological effects (39,40) that cannot be fully rescued by replacing ST3GAL5 alone. To test this hypothesis, we generated ssAAV9-CB-B4GALNT1 and co-delivered it with scAAV9-ST3GAL5.v2 by ICV administration. Indeed, this dual-vector treatment regimen completely rescued lethality, growth retardation, hindlimb clasping, and motor impairment in St3gal5 -/-/B4galnt1 -/mice ( Figure 8). These encouraging results suggest that using the St3gal5 -/-/B4galnt1 -/mouse model to test ST3GAL5 gene replacement may be overly stringent and underrepresent the clinical potential of scAAV9 ST3GAL5 gene replacement vectors. IV injection of rAAV.PHP.eB-ST3GAL5.v2 improved phenotypic abnormalities in GM3SD mouse models. To examine whether systemically delivered scAAV9-ST3GAL5.v2 could achieve broader brain transduction and better therapeutic efficacy without causing liver toxicity, we treated St3gal5 -/-/B4galnt1 -/pups with 3 × 10 11 gcs on P1 by facial vein injection. Although we did not observe the acute lethality associated with systemic administration of first-generation vectors, i.v. scAAV9-ST3GAL5.v2 showed limited efficacy with regard to survival (median survival: 34 days), growth, and motor function (Supplemental Figure 11, B-D), likely due to low ST3GAL5 expression and poor ganglioside reconstitution in the brain as compared with ICV injection (Supplemental Figure 11, E-G). We thus packaged the v2 construct into PHP.eB, an engineered AAV capsid that penetrates the murine BBB more efficiently than AAV9 (41). Under the same systemic administration regimen, ssAAV.PHP.eB-ST3GAL5.v2 led to higher transgene expression in the brain, robust CNS ganglioside restoration, and better phenotypic rescue by all measurements (Supplemental Figure 11, B-G). Taken together, these data underscore the importance of restoring ganglioside synthesis in CNS, particularly in neurons, for ameliorating the GM3SD phenotype.

Discussion
In this proof-of-concept study, we show that rAAV-mediated ST3GAL5 gene replacement restores cerebral ganglioside synthesis, ameliorates neuropathology, and improves motor function in 2 different murine models of human GM3SD (St3gal5 -/and St3gal5 -/-/B4galnt1 -/-). Of note, both ICV and i.v. routes of  administration provided meaningful benefits in animal models, illustrating that CNS-directed ST3GAL5 replacement holds promise for further clinical development (Figure 9).
Developmental and functional differences in ganglioside biology of mice, as compared with humans, present a significant experimental challenge. Humans with severe, biallelic loss-of-function mutations in ST3GAL5 exhibit complete absence of GM3 and its downstream derivatives in plasma and, presumably, brain tissue, and present with epileptic encephalopathy and psychomotor stagnation within a few months of life. A similar enzyme disruption in St3gal5 -/mice leads to tissue deficiency of GM3 but a comparatively mild pathological and behavioral phenotype. Although the major documented phenotype in these mice is hearing loss, we were unable to investigate this further due to limitations in our techniques and expertise. In the absence of GM3 synthase, LacCer is shunted into alternative biosynthetic pathways for the production of O-series gangliosides ( Figure 1) from human patients and St3gal5 -/mice (11,42). It is unclear yet if the complex gangliosides are redundant for maintaining membrane physics and signal transduction in different species. Further thorough analysis of gangliosides and related glycosphingolipids may help understand this discrepancy. On the other hand, a more phenotypically relevant murine model requires simultaneous disruption of 2 serial enzymes in the ganglioside synthetic pathway, St3gal5 and B4galnt1. These double KO mice exhibit severe neuropathology and functional deficits concordant with human GM3SD, but suffer from abiding and functionally relevant B4GALNT1 deficiency after successful ST3GAL5 replacement. Thus, although St3gal5 -/-/B4galnt1 -/mice allow us to test the efficacy of different ST3GAL5 replacement vectors, they may underestimate the therapeutic potential of such vectors for treatment of human GM3SD. This scenario underscores the importance of using relevant animal models in preclinical gene therapy studies. Modeling GM3SD in larger gyrencephalic species, such as pigs or sheep, might prove more informative for future studies (43).
A number of other technical hurdles exist for the treatment of neurological diseases via gene replacement. For example, it is increasingly clear that for many neurogenetic deficiencies, successful treatment will depend on efficient and even delivery of transgene across the neuraxis coupled to a pattern of expression that approximates the natural distribution, abundance, and developmental timing of WT protein. As an example, our first-generation vector induced an active unfolded protein response and severe hepatotoxicity caused by off-target hepatic overexpression of ST3GAL5. Organ toxicity caused by transgene overexpression has been observed in other preclinical disease models. For example, adult mice administered AAV9-MECP2 replacement vectors at 10 12 gcs/mouse develop liver damage with drastically elevated liver transaminases levels and disorganized liver architecture. Similar to our findings reported here, the damage is associated with overabundant expression of MECP2 and apoptosis triggered by UPR (44). In another murine system, neonatal mice receiving ICV delivery of AAV9-SMN at 5 × 10 10 gcs/g develop late-onset, dose-dependent motor dysfunction, impaired proprioception, and neurodegenerative changes due to overexpression of SMN and subsequent RNA dysregulation (45).
Working toward safer and more efficacious therapy, we combined a neuron-specific human Syn1 promoter with a liver-specific miR122 targeting sequence in our second-generation vector. This design prevented ST3GAL5 overexpression in liver and thereby eliminated liver toxicity. The same principle of using facilitative cell-specific promoters coupled to inhibitory miRNA binding sites could be applied more broadly for achieving refined expression specificity (46)(47)(48). In general, we believe that optimizing spatial and temporal regulation of transgene expression will enable safer and more effective systemic gene therapy for a number of neurogenetic disorders in humans, and GM3SD provides an important experimental model to test this idea. Furthermore, scAAV, which uses a mutated inverted terminal repeat (ITRs) to generate an intramolecular, double-stranded genome configuration, allows for faster and higher gene expression but limits the packaging capacity to half that of the ssAAV genome (37,38,49). Indeed, the stronger ST3GAL5 expression driven by scAAV restored ganglioside production and improved phenotypical abnormality more efficiently in the present study.
Last, to identify the most effective and clinically translatable route for administering ST3GAL5 to CNS cells, we tested both ICV and i.v. routes commonly used in current clinical trials (50). ICV injection bypasses the BBB, similar in principle to the more spatially delimited intrathecal injection of nusinersen (51), from WT and St3gal5 -/-/B4galnt1 -/mice with ssAAV9.ST3GAL5.v2 or scAAV9.ST3GAL5.v2 or no treatments. (C) Representative images and quantification of anti-GFAP immunostaining of brain sections (cortex, hippocampus, midbrain) from WT and St3gal5 -/-/B4galnt1 -/mice with ssAAV9.ST3GAL5.v2 or scAAV9.ST3GAL5.v2 or no treatments. Mean intensity was quantified by Fiji. Data are reported as the mean ± SD of 3-5 animals. Statistical analysis was performed by 1-way ANOVA, followed by Sidak's multiple comparisons test. sc, self-complementary; ss, single-stranded. Scale bar: 62.1 µm.
Highly neurotropic AAV capsids, delivered systemically via CNS capillaries, may be key to achieving much broader CNS distribution. As a proof-of-concept, we tested i.v. delivery of ST3GAL5 encapsulated in PHP.eB, an engineered capsid that crosses the murine BBB more efficiently than AAV9. The biochemical and phenotypic results were promising. Systemic delivery of PHP.eB is not clinically translatable because of its species-and strain-specific characteristics, but our results can inform future studies of CNS favorable capsids. In conclusion, AAV-mediated CNS gene transfer with ST3GAL5 at a clinically relevant dose provides significant biochemical and therapeutic benefits with limited off-target toxicity. Notwithstanding limitations of current murine models, our second-generation scAAV9-ST3GAL5.v2 replacement vector is a promising candidate for further clinical development of ST3GAL5 gene therapy.

Study design
The primary goal of this study was to develop an rAAV-mediated ST3GAL5 replacement therapy to treat GM3SD. Our experimental approach combined cells derived from patients with GM3SD and mouse models to evaluate safety, efficacy, and duration of effect. Molecular and physiological readouts include delivery of rAAV genome, ST3GAL5 expression, restoration of gangliosides, body and brain weight, motor functions, and survival. For each experiment, sample size reflected the number of independent biological replicates and, here, is provided in the figure legends. Mice were assigned randomly to the experimental or control groups. Data from all animals were included in the analysis with no excluded outlier.
iPSC culture and differentiation iPSCs from patients with GM3SD were shared by Michael Tiemeyer in the Department of Biochemistry and Molecular Biology, University of Georgia, Athens, GA. iPSCs were maintained in mTESR1 (STEM-CELL Technologies, catalog 85850), cultured in plates precoated with Matrigel (Corning, catalog 354277), and passaged with Rho kinase inhibitor (Abcam, catalog Ab120129). The cortical neuron differentiation was described by Shi et al. (52). Briefly, iPSCs were cultured in neural maintenance media (DMEM:F12 + GlutaMax, Thermo Fisher Scientific, catalog 10565018; and Neurobasal, Thermo Fisher Scientific, catalog 21103049) and first induced by neural induction media containing SB431542 (Tocris, catalog 1614) and dorsomorphin (Tocris, catalog 3093) for 12 days to form the neuro-epithelial sheet. Then cells were passaged with dispase (Thermo Fisher Scientific, catalog 17105041) to wells coated with laminin (Sigma-Aldrich, catalog L2020) in neural maintenance medium. Cells were passaged and plated until after differentiation day 35 in the final plates precoated with poly-l-lysine (Sigma-Aldrich, catalog P5899). Neurons were infected with lentiviral vectors in the presence of 8 μg/mL polybrene (Sigma-Aldrich, catalog TR-1003-G).

AAV vectors
Human ST3GAL5 cDNA isoforms driven by CMV-enhancer/chicken β-actin promoter and human ST3GAL5 cDNA isoforms plus miR122 binding sites driven by Syn1 promoter were cloned into AAV plasmids. The plasmids were sequenced throughout the expression cassette, and the integrity of ITRs was confirmed by restriction enzyme digestion. AAV vectors were produced by transient triple transfection in HEK293 cells and purified by CsCl gradient sedimentation for AAV9 or by iodixanol gradient sedimentation for PHP.eB vectors. Vector titers were determined by droplet digital PCR (ddPCR), and vector purity was assessed by gel electrophoresis followed by silver staining.

Vector DNA extraction and alkaline gel electrophoresis
Extractions of vector DNA from 8 × 10 11 to 1 × 10 13 gcs were performed by phenol/chloroform and ethanol precipitation, as described previously (55). Vector DNA was subjected to alkaline gel electrophoresis stained with ethidium bromide.

Animal use
St3gal5 -/-/B4galnt1 +/male mice were imported from Regeneron Pharmaceuticals, Inc., and bred with C57BL6NTac female mice (Taconic, B6-F). Newborns were genotyped on the date of birth. Briefly, 1 mm was cut from tail tips. Genomic DNA was extracted by boiling in 25 mM NaOH plus 0.4 mM EDTA (pH 8.0) at 100 o C for 90 minutes, followed by mixing with 40 mM Tris-HCl (pH 8.0). St3gal5 and B4galnt1 genes were determined by qPCR using Taqman reagents targeting St3gal5 (Thermo Fisher Scientific, assay ID: APH6DZ6, 9057mTGU; assay ID: APMFZ6Z, 9057mTGD), B4galnt1 (LGC Biosearch Technologies, catalog DLOM-RFB-5, assay ID: 15582TU; assay ID: LacZ) and Tfrc (Thermo Fisher Scientific, catalog 4458367). Primer and probe sequences can be found in Supplemental Table 1. To harvest tissues, mice were anesthetized with isoflurane and transcardially perfused with ice-cold PBS. Tissues were immediately dissected, snap-frozen in liquid nitrogen, and stored at -80 o C. Facial-vein injections were performed on P1 via the right facial vein at 100 μL of 3 × 10 11 gcs per pup. ICV injections were performed on P1 at 4 μL of 3 × 10 10 GC bilaterally per pup or P4 at 4 μL of 6 × 10 10 GC bilaterally per pup. After the procedure, pups were cleaned with 70% ethanol and rubbed with bedding material.
DNA/RNA extraction, quantitative real-time PCR, and ddPCR Total DNA and RNA were extracted from snap-frozen mouse tissues using the AllPrep DNA/RNA Mini kit (Qiagen, catalog 80204). The viral vector gc number was determined in a multiplexed reaction using ddPCR Supermix for Probes (No dUTP) (Bio-Rad, catalog 1863024) and Taqman reagents targeting ST3GAL5 (Thermo Fisher Scientific, Assay ID: APGZHGD) and Tfrc (Thermo Fisher Scientific, catalog 4458367). Total RNA (1 μg) was reverse transcribed into cDNA using the High-Capacity cDNA Reverse Transcription Kit (Applied Biosystems, catalog 4368813). Exogenous human ST3GAL5 and mouse St3gal5 cDNA were quantified in a multiplexed reaction using Taqman reagents targeting ST3GAL5 (Thermo Fisher Scientific, Assay ID: APG-ZHGD), St3gal5 (Thermo Fisher Scientific, Assay ID: Mm00488232_m1), and Gusb (Thermo Fisher Scientific, Assay ID: Mm01197698_m1). ddPCR was performed with a QX200 ddPCR system (Bio-Rad). Quantitative real-time PCR was performed on a ViiA 7 Real-Time PCR system using Taqman gene expression master mix (Thermo Fisher Scientific, catalog 4369016) and Taqman reagents targeting Chop (Thermo Fisher Scientific; Assay ID: Mm01135937_g1) and Tnfa (Thermo Fisher Scientific; Assay ID: Mm00443260_g1).
The analysis of LacCer, GM1, GM2, and GM3 was performed on a Shimadzu 20AD HPLC system and a SIL-20AC autosampler coupled to a 6500QTRAP+ mass spectrometer (AB Sciex) operated in positive multiple-reaction monitoring mode. Data processing was conducted with Analyst 1.6.3 (Applied Biosystems). The relative quantification data for all analytes are presented as the peak ratios of analytes to their internal standard.

Mouse monitoring and behavioral assays
Mice were blindly weighed every other day until weaning at 21 days old. After weaning, each mouse was weighed and evaluated for adverse events weekly by a trained observer.
Negative geotaxis. A negative geotaxis assay was conducted every other day for P9-P15 pups on a 45 o inclined plane. Prior to the test, animals were placed on the plane to acclimate for 1 minute. The mouse head was facing downward; success was marked when the mouse rotated 180 o to the head-up position, and failure was when the mouse dropped off the plane. Whether the mouse could finish the assay was recorded. Each mouse was tested 3 times and the success rate of completing the assay was plotted.
Accelerated rotarod. Coordinated motor functions were examined in treated mice and littermates using the 4-40 rpm accelerating rotarod test. Mice were tested at 6 and 10 weeks old. Tested mice were trained 2 days before the testing day. Prior to the test, the animals were placed on the rotarod machine to acclimate for at least 1 minute. Each mouse was tested 3 times and the best latency to fall was recorded and plotted.

Histology and immunohistochemistry
Mouse brain and liver were fixed in 10% formalin (Thermo Fisher Scientific, catalog SF100-20). Paraffin embedding, sectioning, H&E staining, TUNEL staining (Roche, catalog 11684817910), and immunohistochemistry (IHC) were performed by the Morphology Core at University of Massachusetts Chan Medical School under standard conditions. Mouse anti-GFAP Ab (EMD Millipore, catalog MAB360; 1:500 dilution) was used in IHC. Images were taken on a Leica DM5500 B microscope. The quantification of GFAP IHC was performed using the Image FIJI software as previously described (57).

ProcartaPlex multiplex immunoassays
Total protein was extracted in ice-cold RIPA Lysis and Extraction Buffer (Thermo Fisher Scientific, catalog 89901) with complete, EDTA-free protease inhibitor cocktail (Roche, catalog 4693159001) from snap-frozen tissues. Protein concentration was determined using a bicinchoninic acid protein assay kit (Thermo Fisher Scientific, catalog 23252). Normalized protein extracts were loaded on a ProcartaPlex Mix & Match panel (Thermo Fisher Scientific). Values were acquired by Bio-Plex MAGPIX (Bio-Rad).

mRNA-Seq
RNA-Seq was carried out by Novogene under standard conditions. RNA was extracted using the Trizol phase separation method from cell debris. Isolated RNA sample integrity and concentration were assessed by the Agilent Bioanalyzer 2100. RNA (1 μg/sample) was used as input material for RNA sample preparations. Sequencing libraries were generated using NEBNext Ultra RNA Library Prep Kit for Illumina (New England BioLabs, catalog E7770L) following manufacturer's recommendations. Briefly, mRNA was purified from total RNA using poly-T oligo-attached magnetic beads. Fragmentation was carried out using divalent cations under elevated temperature in NEBNext First Strand Synthesis Reaction Buffer (5×) (New England BioLabs). Firststrand cDNA was synthesized using a random hexamer primer and M-MuL V Reverse Transcriptase (RNase H-). Second-strand cDNA synthesis was subsequently performed using DNA Polymerase I and RNase H. Final library quantities were assessed by the Agilent Bioanalyzer 2100 system. The clustering of the index-coded samples was performed on a cBot Cluster Generation System using PEE Cluster Kit cBot-HS (Illumina) according to the manufacturer's instructions. After cluster generation, the library preparations were sequenced on an Illumina NovaSeq 6000 platform and paired-end reads were generated.
For data analysis, a 3′ adapter sequence was removed using Trimmomatic with the following ILLUMINACLIP parameters: min_length, 10; seed mismatches, 2; palindrome clip threshold, 30; simple clip threshold, 5. Then, reads were mapped to mouse_mm10_gencode_ using STAR. To estimate expression levels, RSEM55 was used to align reads to a predefined set of transcripts from GEN-CODE. Finally, the RSEM quantification matrix (i.e., estimated counts for each gene and/or for each annotated isoform) was used for differential gene expression analysis. The count matrix was loaded into DEBrowser software for interactive analysis. Data analysis was performed on the RNA-Seq pipeline of the DolphinNext (58).

Statistics
All data are presented as mean (SD) and were analyzed using GraphPad Prism software (version 9). Twotailed Student's t test was used to compare 2 groups, and 1-way ANOVA was used to compare data among multiple groups. Animal weight was analyzed by 2-way ANOVA, and survival was analyzed by log-rank (Mantel-Cox) test. In the figures, P values are indicated by asterisks, as follows: *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001.

Study approval
All animal procedures were reviewed and approved by IACUC at University of Massachusetts Chan Medical School and performed in compliance with all relevant ethical regulations.
Data and materials availability mRNA-Seq data can be found in the National Center for Biotechnology Information's Gene Expression Omnibus (GEO) using GEO series accession number GSE201587. Other data supporting the findings of this study are available within the article or from the corresponding authors upon reasonable request.

Author contributions
GG and KAS conceived the project; HY, DW, KAS, and GG designed the experimental plan; HY performed cell, animal, and mouse tissue experiments; HY analyzed data with critical input from RHB, DW, and GG; HY, DW, and GG wrote manuscript; KAS and GG supervised project.