Antisense oligonucleotides to KIF1A polymorphisms expand targets and rescue patient-derived neurons in vitro

Abstract

Dominant negative pathogenic variants in KIF1A result in an allelically heterogeneous neurodegenerative condition that manifests as a variable clinical phenotype including seizures, cognitive deficits, optic nerve atrophy, spasticity, and peripheral neuropathy. One potential therapeutic strategy is allele-specific knockdown of pathogenic transcripts. However, targeting the over 100 known unique pathogenic variants is challenging. Alternatively, different pathogenic KIF1A variants in multiple patients can be knocked down by targeting shared common polymorphisms with antisense oligonucleotides, provided that the pathogenic variants are in cis with the targeted polymorphisms. Here, we use long-read sequencing data from fifty-six individuals to phase for polymorphisms. We identify four common polymorphisms that, if targetable, would make it possible for 54 of these individuals to receive antisense oligonucleotide therapy. Using patient-derived glutamatergic neurons, we characterize and quantify a cell-autonomous phenotype, dendrite neurite outgrowth length. In vitro we further demonstrate that antisense oligonucleotide-mediated knockdown of the pathogenic transcript rescues the dendrite neurite outgrowth phenotype in neurons from a patient with the P305L variant.

Pathogenic variants in KIF1A give rise to KIF1A-associated neurodegenerative disorder (KAND). Here, the authors use antisense oligonucleotides targeting common polymorphisms to knock down pathogenic KIF1A expression in patient-derived stem cells.

Introduction

KIF1A encodes a molecular motor kinesin¹. The protein forms a homodimer that is highly neuronally expressed and is responsible for axonal anterograde transport^2–4. One aberrant KIF1A subunit is sufficient to disrupt the dimeric molecular motor function through a dominant-negative mechanism^5,6. The motor domain is necessary for transport of critical cargoes such as neurotrophic factors that move along axons to synapses and affects numerous neuronal functions including axon growth, dendrite length, pruning, synaptic function, and plasticity^7,8. Pathogenic KIF1A variants are primarily heterozygous de novo missense variants in the motor domain and often result in decreased processivity or velocity along microtubules^6,9.

Pathogenic variants in KIF1A (OMIM #614255, 610357, 620607, 601255) give rise to a group of degenerative disorders collectively referred to as KIF1A-associated neurological disorder (KAND). KAND has a highly variable phenotype that ranges from mild spastic paraplegia with a normal lifespan to a degenerative neurological disorder that is lethal within the first few years of life. The majority of reported individuals exhibit some degree of intellectual disability and neuronal atrophy in multiple brain regions. Individuals with inherited biallelic variants outside of the motor domain of KIF1A are less severely affected, and their heterozygous parents are generally unaffected¹⁰. Homozygous or compound heterozygous nonsense variants in KIF1A have been shown to contribute to hereditary sensory and autonomic neuropathy type II (HSANII)¹¹. Inherited truncating variants in the plekstrin homology domain are predicted to lead to nonsense-mediated decay and haploinsufficiency with reduced amounts of normal KIF1A protein¹². Humans with KIF1A haploinsufficiency and haploinsufficient mouse models exhibit a mild neurological phenotype which occurs later in life¹³. These reports contrast with phenotypic observations of missense variants which are more commonly observed in clinical cases, suggesting that haploinsufficiency for KIF1A results in a less severe phenotype than missense variants. The milder clinical severity associated with heterozygous loss-of-function variants suggests that a decrease in normal KIF1A is, to some degree, tolerated. Thus, allele-specific knockdown of the pathogenic allele could allow for increased dimerization of normal KIF1A protein and improve overall protein function of KIF1A.

One method to achieve knockdown is employing antisense oligonucleotides (ASOs). These molecules can bind to endogenous RNA to cause preferential allele-specific RNA degradation via cleavage by RNase H¹⁴. ASOs can be delivered to the brain through intrathecal administration, do not require a vector for entry into cells, and can maintain RNA knockdown for prolonged periods in non-dividing cells. We have demonstrated successful treatment of one individual with KAND using an allele-specific ASO¹⁵, and neurons derived from induced pluripotential stem cells (iPSCs) from this patient are included in our studies.

Patient-derived induced pluripotent stem cells provide a renewable source of cells capable of neuronal differentiation and enable in vitro disease modeling^16,17. The majority of the cerebral cortex is comprised of excitatory glutamatergic neurons. Patient-derived neurons differentiated into this specific cell type can model KAND’s neurodevelopmental and neurodegenerative conditions commonly observed in global brain dysfunction^18,19. Measurable cellular phenotypes including morphology and neurite outgrowth length are straightforward to obtain and can be reliably quantified to assess the relative efficacy of potential treatments^20,21.

To date, over 100 different KIF1A pathogenic variants have been identified. With increased awareness and improved access to genetic testing, the number of diagnosed KAND patients is increasing. The extensive allelic heterogeneity in KAND, where no single variant accounts for more than 5% of known cases, makes developing individual allele-specific therapies resource-intensive and impractical. Instead, we targeted common variants in cis with pathogenic variants to allow greater flexibility in ASO design and reduce the number of ASOs required to address the needs of the majority of patients.

Results

Targeting a broader patient population through common polymorphisms in cis with pathogenic variants

Given KAND’s allelic heterogeneity and the resources required to develop a treatment for a single KIF1A variant, our strategy was to develop reagents targeting common polymorphisms that might be useful for the majority of patients (Fig. 1A). We used gnomAD 4.1 to identify common polymorphisms with an allele frequency between 40 and 60% in coding, intronic, and untranslated regions of the KIF1A gene and identified 39 potential targets. We genotyped 97 KAND patients for two common SNPs, rs7578279 (intronic) and rs1063353 (coding) (Figure S1); 73% of genotyped individuals were heterozygous for at least one of the two SNPs (46% for rs7578279 and 40% for rs1063353) (Fig. 1B). Of these heterozygous patients, we used long-read sequencing data from 56 individuals (Data S1) to phase common polymorphisms relative to pathogenic KIF1A variants and to determine additional polymorphism targets in our patients. Based upon these phasing data, aside from rs7578279, an additional six sites were heterozygous in at least 70% of these individuals (Fig. 1C). Four of these sites, which would require 8 ASOs (one for each potential allele), would theoretically be sufficient to treat almost 90% of patients. SNPs rs4676368, rs1529665, rs6725635, and rs4998254 would collectively cover 89.3% of our phased individuals (Fig. 1D).

Fig. 1. Common polymorphism targets for ASO treatment.

A Schematic representation of phasing and the ability to target polymorphisms with ASOs. B Schematic of ASO target selection process (left) and how many genotyped individuals in our cohort are heterozygous for each of the two common targetable polymorphisms or at least one of the two polymorphisms. C Quantification of the number of polymorphisms for which different percentages of individuals are heterozygous for target polymorphisms in the subset of patients who were phased. D Cumulative fraction of individuals who are heterozygous and could potentially be targeted by the incremental SNPs targeted.

However, these data are from our largely US-based cohort, primarily of a single self-reported ancestry (51/56). The 1000 Genomes Project data represent genetic ancestries around the world. To estimate the minimum number of ASO target sites required for a global population, we utilized a greedy algorithm. Based on 1000 Genomes Project data²², a few common haplotypes in the KIF1A genomic region are present in the majority of individuals. In the dataset, most individuals are heterozygous for at least one among a small number of common polymorphic sites. Therefore, targeting the small number of common polymorphic sites should cover the majority of KAND patients. We calculated the minimum number of polymorphic sites required for various fractions of the 1000 Genomes Project individuals to have at least one heterozygous genotype. We found that three polymorphic sites (rs10196375, rs62187837, and rs4414678) can cover about 80% of individuals globally, and that six polymorphic sites (rs79612768, rs2288746, and rs544175909 in addition to the previous three SNPs) can cover more than 90% (Figure S2).

Patient-derived neurons exhibit decreased neurite outgrowth length

iPSC lines were derived from three individuals with different de novo pathogenic KIF1A variants: c.452 G > A (p.C151Y), c.757 G > A (p.E253K), and c.914 C > T (p.P305L); all three individuals were heterozygous for rs1063353. The cell lines p.E253K and p.P305L, associated with more severe clinical phenotypes (rapid disease progression, loss of mobility, seizures), were edited using CRISPR/Cas9 to correct the pathogenic variant and generate isogenic control lines with the respective nonpathogenic allele (E253E and P305P). Lines were also derived from an individual mosaic for c.609 G > T (p.R203S) as an additional set of KAND and isogenic control cells without the need for gene editing (Figure S3). Cells were induced into neural stem cells (NSCs), confirmed by morphological inspection and immunofluorescent staining for SOX1 and NESTIN (Fig. 2A, B). Established NSCs were then differentiated into glutamatergic cortical neurons without supplementation or co-culturing of additional cell types (Fig. 2C). We focused on glutamatergic cortical neurons as a representative model because the brain is central to the most severe phenotypes and likely a good representative model for other neuron types. Neurons were cultured for up to 35 days in vitro (DIV) and analyzed via immunofluorescent staining for the neuronal markers VGlut1 and MAP2 (Fig. 2D). All patient-derived cell lines were capable of successful neuronal differentiation. We sought to identify a KAND phenotype in these differentiated glutamatergic neurons. Abnormal outgrowth morphology has been observed in neurological disorders, and individuals with KAND have been reported to present with severe neuronal atrophy from MRI data^21,23–25. We quantified differences in patient-derived neurons and isogenic controls for neurite outgrowth length via immunofluorescent staining for MAP2 and found that neurons with pathogenic KIF1A variants exhibited  a statistically significant reduction in neurite outgrowth lengths compared to isogenic controls (Fig. 2E–G, Figure S4).

Fig. 2. Patient-derived neuronal outgrowth phenotype.

A Representative images of immunofluorescent characterization of neural stem cells (NSCs) at 20x magnification. Scale bars: 50 µm. B Quantification of NSC marker staining where the number of biologically independent induction experiments is 6 per genotype, and each experiment includes an n of 120 cells assessed per experiment. Data represented as min to max whiskers, horizontal line indicates center, box represents interquartile range. C Schematic of differentiation protocol for cortical neurons. DIV = Days in vitro. D Representative images of immunofluorescent characterization of iPSC-derived cortical neurons with MAP2 (red), vGlut1 (yellow), KIF1A (green), and DAPI (blue). Scale bars: 50 µm. Characterization analysis performed across 5 biologically independent differentiation experiments per genotype. E Representative images of immunofluorescent staining of MAP2 used for neurite outgrowth tracing and length measurement, a representation shown in (F), and quantification (G). The number of biologically independent differentiation experiments is 5 per genotype, where each point represents data from an individual MAP2-positive outgrowth (outgrowths were averaged per neuron where n = 76, 78, 71, 58, 75, 63, 73 for Control (R203R), C151Y, R203S, E253K, Corrected E253K, P305L, and Corrected P305L, respectively). Two-sided one-way ANOVA was used for (G) with Dunnett’s multiple comparisons test; alpha = 0.05, DF = 486. p-values: ns = >0.05 (0.8404 and >0.9999 for Corrected E253K and Corrected P305L, respectively), **** = p < 0.0001. Source data are provided as a Source Data file.

Targeting common polymorphisms enables allele-specific ASO-mediated knockdown of pathogenic variant expression in patient-derived neurons

Because mutations arise de novo, two individuals with the same pathogenic KAND-causing variant may not have the same phasing with respect to the targetable polymorphism (Figure S5). As such, we sought to determine the degree of knockdown for ASOs targeting both alleles at multiple polymorphism sites. Neurons differentiated from iPSC lines were treated with ASOs that targeted different common polymorphic sites. Each cell line was treated separately with one ASO targeting the reference SNP allele (Ref) and one ASO targeting the alternate SNP allele (Alt) (Data S2). qPCR analysis for KIF1A expression in the p.C151Y, p.E253K, and p.P305L lines was performed to determine KIF1A knockdown, combined with next generation sequencing to determine the relative allelic expression between KIF1A pathogenic and reference alleles (Fig. 3A–D, Figure S6). While nonspecific knockdown was observed in every cell line for every ASO, ASO-rs7578279-Ref, ASO-rs7578279-Alt, ASO-rs7598218-Ref, and ASO-rs7598218-Alt exhibited significant allele-specific knockdown in the P305L line which was heterozygous for both polymorphic target sites. Significant specific knockdown was also achieved in lines C151Y, E253K, and P305L (R203S was not heterozygous) for the targeted allele when using ASO-rs1063353-Ref and ASO-rs1063353-Alt compared to untreated and scramble controls. These data demonstrate that allele-specific knockdown can be achieved by targeting a common polymorphism with similar efficiency across different individuals with genetically different pathogenic variants as well as different genomic contexts.

Fig. 3. ASO-mediated allele-specific knockdown.

A Total KIF1A mRNA expression quantified from qPCR of C151Y neurons treated with individual ASOs that target either the reference (Ref) or alternate (Alt) allele for rs1063353 (left) with relative allelic expression (right). B Total KIF1A mRNA expression quantified from qPCR of E253K neurons treated with individual ASOs targeting rs1063353 (left) with relative allelic expression (right). C Total KIF1A mRNA expression quantified from qPCR of P305L neurons treated with individual ASOs targeting rs1063353 (left) with relative allelic expression (right). D Comparison of total mRNA expression of all lines treated with individual ASOs. Data is from three independent differentiations for each genotype, with each point representing the average of three technical replicates per biological replicate. Two-sided one-way ANOVA was used with Dunnett’s multiple comparisons test for left panels A-C (alpha = 0.05, df = 8 for each). Two-sided two-way ANOVA with Šídák’s multiple comparisons test was used for right panels (A–C) (alpha = 0.05, df = 16 for each). Two-sided one-way ANOVA was used with Tukey’s multiple comparisons test for D (alpha = 0.05, df = 24). p-values: ns = >0.05 (0.8387, >0.9999, and >0.9999 for A; 0.7962, >0.9999, and 0.7266 for B; 0.6534, >0.9963, and >0.9992 for C; 0.5439 for D), **** = p < 0.0001. A–D error bars represented as mean with SD.

ASO targeting a common polymorphism in cis with pathogenic variant rescues neurite outgrowth phenotype

We sought to investigate the effects of pathogenic allele-specific reduction on our cell-autonomous phenotype via ASO-mediated allele specific knockdown directed to rs1063353 (Fig. 4A, B); this SNP is in cis with the pathogenic allele in our P305L cell line, which has different phasing from the C151Y and E253K lines (Figure S5, Data S2). ASO design optimization yielded two additional sequences to target rs1063353 in the P305L line. P305L neurons were differentiated and treated with either ASO or a scramble control, after which cells were either harvested for mRNA expression analysis or immunofluorescent staining (Fig. 4C–E). Compared to untreated and scramble control neurons, the ASOs targeting rs1063353 in cis with the mutant allele had 74% and 60% allele-specific knockdown of the pathogenic allele (Fig. 4C). After four weeks of differentiation post-treatment, morphology assessed via immunofluorescent staining of MAP2 showed that neurons treated with either of the two ASOs (both targeting the same SNP) rescued neurite outgrowth (Fig. 4D, E).

Fig. 4. Treatment of patient-derived neurons with ASOs rescues neurite outgrowth.

A Schematic of differentiation protocol for cortical neurons. B Schematic of ASOs used to target in cis polymorphism rs1063353 (indicated by yellow highlight). DIV = Days in vitro. C qRT-PCR quantification of relative allele-specific mRNA expression in p.P305L DIV35 cortical neurons treated with two different ASOs (targeting the same alternative SNP allele) for 4 weeks. Data is from three independent differentiations. Two-sided two-way ANOVA with Šídák’s multiple comparisons test was used for right panels (A–C) (alpha = 0.05, df = 16 for each). p-values: ns = >0.05 (0.9993 and 0.7784 for Untreated Control and Scramble, respectively), **** = <0.0001. Error bars represented as mean with SD. D Quantification of neurite outgrowth length from immunostaining. Data is from five independent differentiations per genotype, where each point represents data from an individual MAP2-positive outgrowth (n = 82, 77, 103, 96, 86 for Untreated P305L, Scramble ASO, ASO 1142, ASO 1144, and Corrected – P305P, respectively). Two-sided one-way ANOVA was used with Dunnett’s multiple comparisons test was used (alpha = 0.05, df = 439). p-values: ns = >0.05 (0.9999 and 0.9999 for C; 0.9979 for D), **** = p < 0.0001. Source data are provided as a Source Data file. Thick dashed line represents the median. Thin dashed lines represent quartiles. E Representative images of immunofluorescent staining of MAP2 used for neurite outgrowth length analysis. Scale bars: 50 µm.

Discussion

We have previously demonstrated the ability to use allele-specific ASOs to treat KAND¹⁵. However, the current number of known KIF1A pathogenic variants is large, and consequently we are attempting to design a strategy to develop a modest number of ASOs to treat the majority of KAND patients. We designed ASOs targeting common polymorphisms in cis with pathogenic KIF1A variants. We demonstrate a reproducible phenotype of decreased neurite outgrowth length in glutaminergic neurons differentiated from iPSCs of individuals with pathogenic KIF1A variants. The pathogenic allele was knocked down via ASO-mediated allele-specific targeting of a common polymorphism. Furthermore, pathogenic allele knockdown rescued the neurite outgrowth length phenotype in patient-derived neurons. The key advantage of ASOs that target common polymorphisms is that a relatively small number of ASOs can target the majority of the patient population, obviating the need for bespoke ASOs for each unique pathogenic KIF1A variant. Additionally, these results may be generalized to other diseases that are caused by heterozygous dominant negative variants.

ASOs act through the recruitment of RNase H1 in both the nucleus and cytoplasm²⁶, and can target both exonic and intronic sequences²⁷. Targeting intronic sequences allows the ASO to target a wider pool of polymorphisms as introns are more likely to contain common variants than exons²⁸. Targeting nonpathogenic, common polymorphisms in cis with the pathogenic allele should simplify the regulatory approval process of a modest number of ASOs for KAND. In KAND, almost all pathogenic mutations occur de novo in patients, so there is no phasing bias with common variants. Our strategy assumes developing two ASOs for each common variant target to knockdown the allele that is in cis with the pathogenic variant. The SNPs we have identified are largely intronic and not close to the coding region for the motor domain of KIF1A and therefore should not interfere with binding due to the mutation itself. KIF1A is a relatively large gene, spanning just over 106 kb. As some sequences lend themselves more readily to being targeted with ASOs, a large gene sequence can be advantageous for ASO design because of the theoretical larger pool of polymorphisms available to target. Other genes that are smaller, have sparse or small intronic regions, or are less genetically variable may have fewer targeting options. ASO gapmers targeting introns can yield more efficient knockdown compared to exon targeting, possibly due to steric hinderance from splicing machinery²⁹. The intron-exon differences observed have also been proposed as a product of higher densities of RNA binding proteins in the chromatin patterns of exons, as well as inherent RNA structures³⁰. However, as suitable predictive models for ASO specificity do not currently exist, empirical testing of potential ASOs is an acknowledged technological limitation.

We identified common variants across the KIF1A locus. If successfully targeted, just four targets (eight ASOs) could be used to treat 97% of the patients for which we have phasing data. A limitation of our current dataset is that the majority of our study population is of a single ancestry. Since the majority of patients have de novo variants, there is no specific ancestral enrichment for KAND globally. We require greater heterogeneity in our available sequence data to ensure that ASOs are being developed that will address the global treatment needs by targeting polymorphisms common for a global population. In lieu of direct access to the total worldwide KAND population, we took an algorithm-based approach to estimate how many ASO targets would be needed for a global therapeutic application. Based on data from the 1000 Genomes Project, three common variants, could cover about 80% of individuals, and six targets (12 ASOs) could cover over 90%. An important distinction to consider is that phasing occurs randomly, and any given polymorphism target may be in cis or in trans with the patient pathogenic variant. As such, each polymorphic target site will require the development of two ASOs to account for both allelic possibilities.

Accurate and physiologically relevant modeling of human diseases remains a challenge for many neurogenetic disorders. Many conditions rely on intricate network assays or resource-intensive organoid differentiation for an observable phenotype. In human iPSC-derived neurons, morphological changes have been associated with pathogenic function^31,32. KAND has been associated with neurodegeneration and neuronal atrophy, especially in individuals with more severe variants²⁵. Abnormalities in neurite outgrowth morphology, including length, have been reported in multiple conditions including Rett syndrome, and Fragile X syndrome^33–37.

Several stem cell models of neurodevelopmental disorders have been shown to exhibit a neurite outgrowth phenotype including Fragile X Syndrome, tuberous sclerosis, Rhett Syndrome, and Kleefstra syndrome^38–43.

We used MAP2 staining to assess neurite outgrowth length in neurons harboring a pathogenic KIF1A variant. Our patient-derived neurons, compared to nonpathogenic and CRISPR-corrected controls exhibited decreased neurite outgrowth length. Our hypothesis was that outgrowth morphology was a potential cell-autonomous phenotype since KIF1A is the primary transporter of dense core vesicles which are transported to plasma membranes in dendrites via the Golgi aparatus^44,45. Dendrite outgrowth morphology is critical for normal function, and morphological features, including length, are affected by a plethora of cues which including extrinsic signals⁴⁶. It is therefore not surprising that pathogenic variants in the axon motor protein KIF1A can yield a neurite outgrowth phenotype, considering that BDNF, a well-established axonal cargo of KIF1A, directly impacts dendrite morphology^47,48. Additional cargoes such as Rab3 and Synaptotagmin may also facilitate signaling that impacts neurite outgrowth morphology⁴⁹. The exact mechanism behind the observed neurite outgrowth length phenotype requires closer examination of morphological signaling. This quantifiable phenotype was replicated in cell lines derived from multiple individuals with different pathogenic KIF1A variants. A limitation of this study is the indirect connection between an in vitro and in vivo phenotype. To date, there are limited data available regarding the neuropathology of KAND, although loss in brain tissue has been reported²⁵.

We treated neurons with ASOs designed to target heterozygous alleles and observed allele-specific knockdown. ASOs knocked down the targeted allele in neurons from three different individuals harboring genetically distinct pathogenic KIF1A variants with similar degrees of efficacy and specificity, demonstrating that the outcomes are not attributed to cell line variability that can be observed in stem cell models. Several candidates demonstrated substantial knockdown of the targeted allele, with all ASOs tested exhibiting at least some degree of knockdown for the non-targeted allele. An important consideration, especially with regards to potential treatments, is the degree of nonpathogenic allele knockdown tolerated in the genetic condition of interest. We utilized the allele-specific knockdown based upon the existing clinical evidence that KIF1A haploinsufficiency is less severe than de novo dominant negative missense variants that comprise the majority of individuals with KAND. Maintaining sufficient expression of the normal KIF1A gene is critical since absence of KIF1A in mouse models is lethal⁵⁰. The implications of haploinsufficiency need to be considered on a gene-by-gene basis given the potential risks when utilizing ASO-mediated knockdown of alleles. Since there are potentially many common SNPs within a gene and ASO efficiency depends on the target sequence, the number of possible targets should support multiple ASO designs.

Only the P305L line possessed both the ASO target and pathogenic variant on the same haplotype, allowing us to analyze the ASO treatment and phenotype rescue in only a single line. ASOs targeting rs1063353 knocked down pathogenic KIF1A allele expression sufficiently to rescue the neurite outgrowth length phenotype in P305L neurons. This rescue establishes a relative degree of knockdown that is both tolerated and results in in vitro phenotypic rescue, which is an important finding for future studies, particularly those aimed at optimizing potential therapeutic agents. One limitation is that this targeting strategy could not be implemented for the other existing lines in our possession. Additional patient lines with the same phasing will be required to achieve replication of these findings. Optimizing mutant knockdown both in terms of efficacy⁵¹ and specificity⁵² is critical to ensure the nonpathogenic allele remains sufficiently expressed to retain normal protein function^53,54. Off-target effects are also an important consideration, especially for neuronal cultures which may not immediately reveal morphological indicators. Compensatory changes in RNA expression of other genes and even cell death can occur⁵⁵. While the ASO specificity we achieved was sufficient to change the cell-autonomous phenotype in vitro, this level of specificity may not be sufficient in other disorders. Other strategies for ASO optimization include utilizing a mixmer by incorporating additional RNA bases based on secondary structure predictions for both mutant and wild-type alleles to enhance ASO accessibility to mutant transcripts can achieve a 3-fold increase in specificity⁵⁶. Specificity can be further improved by incorporating a nucleotide mismatch as a structural modification to achieve a 10-fold enhancement in specificity compared with the gapmer design.

Ideally, treatment will be administered to individuals as early as possible given the progressive, degenerative nature of KAND. However, currently, timing is limited by symptom presentation and subsequent genetic testing required to make the diagnosis⁵⁷. Furthermore, it will be of the utmost importance to have access to phased patient-specific variants to enable accurate design and implementation of therapeutic ASOs. An important consideration moving forward, especially for clinicians, is that patients eligible for a given ASO will likely require phasing of rare and common variants, which is information not typically provided by clinical genetic testing as these reports focus on reporting the pathogenic variant and are usually performed on short-read platforms. To perform phasing, high molecular weight DNA will need to be extracted from patient samples to provide sufficient material for the long-read sequencing required to accurately map and determine phase of these variants^58–60.

This approach of targeting dominant negative mutations has long been known⁶¹, and can be significant in its clinical outcome¹⁵. It is potentially generalizable to other disorders caused by dominant negative or gain of function mechanisms, provided haploinsufficiency is tolerated. Conditions similar to KAND, in which dimerization is necessary for proper function, could make for suitable therapeutic targets. Trafficking proteins such as endoglin can have dominant-negative variants, which are observed in hereditary hemorrhagic telangiectasia type 1 (HHT1)⁶². Genes such as SCN2A⁶³, SCN8A⁶⁴, and CDKL5⁶⁵ are known to have gain of function variants that impact the brain and nervous system, but they also have known loss of function variants with haploinsufficiency, and are thus not ideal for allele-specific knockdown It is important to know the consequences of knockdown considering the tight regulation that often accompanies proteins⁶⁶. Much like the specificity and efficacy of each individual ASO, the targeted gene(s) of interest with ASOs should be carefully considered and evaluated. There are several ASO treatments with allele specific targets for alleles with a gain-of-function mechanism including SOD1⁶⁷, C9orf72⁶⁸, and FUS⁶⁹. There are also the well-known Huntington’s studies that propose polymorphism targets^61,70,71. Additionally, ASO targeting of a polymorphism has been successfully carried out in an in vitro model of Sca3 in human-derived neurons⁷².

Given that an ASO targeted to a common polymorphism can be used for patients with many different KIF1A pathogenic variants and at many different disease stages, the clinical trial design will be quite complex. Since the severity of phenotypic presentation is heterogeneous, it is likely that each patient will need to serve as their own control pre and post treatment. Given the progressively complicated developmental and neurodegenerative nature of this condition over time, it is critical to begin collecting detailed phenotypic clinical data as soon as possible for KAND patients that are going to be treated.

We demonstrate here that common polymorphisms can be utilized to design ASOs to target a broader patient population rather than relying on bespoke treatments for each individual pathogenic variant. ASOs can be utilized to rescue an in vitro phenotype associated with pathogenic KIF1A variants in patient-derived neurons. Given the progressive, degenerative nature of KAND, this work provides critical data for a therapeutic ASO approach to treatment. Furthermore, these findings may be applicable to the broader field of neurogenetic diseases due to heterozygous dominant negative mechanism including some forms of autism, epilepsy, ALS, and Parkinson’s^73–78

Methods

Ethics

The study design and conduct complied with all relevant regulations regarding the use of human study participants and received ethics committee approval. Columbia University IRB AAAT8830 and Boston Children’s Hospital IRB P00046386 reviewed the protocol and provided ethical approval for this study. This work was conducted in accordance with the criteria set by the Declaration of Helsinki.

Participant recruitment and characteristics

DNA samples from 97 individuals of our cohort were used in this study. Age range of individuals was 2 years old to 55 years old, with 54 males and 43 females. Consent was obtained from all individuals. In the event an individual was a minor or could not consent, consent was provided by a parent/guardian. Sex or gender was not a consideration of this study because the causative variants are de novo and there is no known difference in occurrence rates between males and females.

Patient DNA extraction and long-read sequencing

High molecular weight (HMW) DNA was extracted from whole blood using New England Biolabs Monarch^® HMW DNA Extraction Kit for Cells & Blood (T3050L). DNA was also extracted using Puregene Blood Kit (Qiagen 158023) according to the manufacturer instructions.

Extracted DNA was quantified on a Qubit Fluorometer (Invitrogen) and quality was assessed using a NanoDrop Spectrophotometer (ThermoFisher) and an Agilent Femto Pulse. The Short Read Eliminator kit (PacBio 102-208-300) was used to remove smaller fragments according to the manufacturer instructions for any sample with a large number of fragments <1 kb on the Femto trace. Libraries were prepared using the Oxford Nanopore Ligation Sequencing Kit (SQK-LSK114), loaded onto a R10.4.1 flow cell, and run using adaptive sampling on a PromethION 24. The KIF1A gene was enriched by targeting the region chr2:239,700,000–241,800,000 (GRCh38), while approximately 200 kb surrounding FMR1 and COL1A1 were used as controls (chrX:147,900,000–148,000,000 and chr17:50,150,000–50,250,000, respectively). Following sequencing, libraries were base called using Dorado version 0.5 (ONT) using the super accurate model with 5mCG and 5hmCG modifications. Run performance was evaluated using cramino (v0.14.1)⁷⁹. Base called reads were aligned to GRCh38 using minimap2 small variant calling⁸⁰ and phasing was performed using Clair3⁸¹. Haplotagging was performed using WhatsHap^82,83.

Phasing algorithm

We obtained 1000 Genome Project phase 3 data which includes 2504 individuals from 26 populations in Africa, Europe, East Asia, South Asia, and the Americas²². We consider common variants (overall frequency > 1%) located in the KIF1A genomic region, including both exons and introns. To estimate the minimum number of variants that cover a certain fraction of individuals, we implemented a greedy algorithm as follows:

1. Include all individuals in 1000 Genomes phase 3 (total number: N = 2504), as individuals to be considered.

2. Sort all SNPs based on number of heterozygotes, in decreasing order, among the individuals to be considered.

3. Take the top SNP, remove the heterozygotes among individuals to be considered,

• Calculate the fraction (f) of remaining individuals: f = the number of remaining individuals divided by N (2504). Record the identity of the SNP in a table.

• If f is below the threshold (0.05), then stop.

• Otherwise, go back to step 1, using the remaining individuals as the individuals to be considered.

After the process is finished, there is a selected minimum set of SNPs sorted by the cumulative coverage of individuals. Each extra SNP corresponds to a cumulative coverage (1 – f), up to 95%.

Polymorphism selection

In the gnomAD 4.1 database there are 38,525 variants within KIF1A. Of these, 55 have an allele frequency between 40 and 60%. After removing variants that cannot be targeted by ASOs, tandem repeats (17 variants) and non-unique sequence (3 more variants), there are 39 potential targets for ASO design.

Cell lines

Induced pluripotent stem cell (iPSC) lines were derived from four patients with de novo pathogenic variants c.452 G > A (p.C151Y), c.609 G > T (p.R203S), c.757 G > A (p.E253K), c.757 G > A, and c.914 C > T (p.P305L) with consent. To derive patient-derived iPSCs, peripheral blood mononuclear cells were collected from blood samples and reprogrammed to iPSCs with a non-integrating Sendai virus kit, following the manufacturer’s exact instructions (CytoTune™-iPS 2.0 Sendai Reprogramming Kit, ThermoFisher Scientific). iPSCs were maintained in mTeSR Plus medium (Stem Cell Technologies 100-0276) on −6 well plates (Crystalgen 191-9381) pre-coated with Geltrex (Gibco A1413201) dissolved in DPBS without calcium or magnesium (Gibco 14190144) at a concentration of approximately 16 µg/mL. iPSCs were passaged as needed using Versene (Gibco 15040066). All cell lines were regularly tested for mycoplasma contamination (Invivogen rep-mys-10). iPSC lines from these patients were selected to ensure disease modeling was representative of the patient population. Additionally, two cell lines (p.E253K and p.P305L) from patients with more severe phenotypes were edited using CRISPR/Cas9 to make isogenic lines with the non-pathogenic allele (Figure S5A). The generation of the E253K and P305L iPSC lines were performed using directed delivery of Cas9, sgRNA, and ssODN using previously described protocols^84,85. Briefly, a single sgRNA and donor DNA template were designed to specifically modify the mutation and electroporated into iPSCs using the Lonza Nucleofector 4D system. Homology-directed repair efficiency was assessed by ICE analysis and colonies of interest were selected and confirmed by Sanger sequencing⁸⁶. To target the variant in the P305L line, the commercially synthetized (Synthego) sgRNA GCCAGGTCAACACGGAATC was used in combination with the single-

stranded oligodeoxynucleotide (ssODN) (IDT): CCACAAGTTCTCACCCAGGTTTTCCCGGAGGAGCCAGGTCAACACGGAATCTC(t)GTAC(G)GAATGAAATCTGTCTTCTTCTTTTTCTTGTTC. The PAM was silenced by altering C > t, with the target editing A > G. Similarly, for E253K, the sgRNA: GGAAAGGCCACAAGATGCGC was used in combination with the ssODN: TGCCCTTGGCTCCCGTGGAGTCAGCCCGCT(C)GCTCCCAGCCAGGTCCACCAGGCTGATTTTGCTCACCTGAAACAGTAGATACATCACATGCAGGAAAGGCCACAAGATGCGCG(at)GGCATCCCAGGACCCCTGGGCAAGGTCTCCACAGCTGTTGAAGGAGCCCT, where PAM was silenced by altering GG>at, and the target editing A > C. The ssODNs were designed as a complementary sequence to the sgRNA target strands. 15 μg gRNAs and 15 μg donor DNA were incubated for 25 min at RT in combination with 10 μg of Alt-R S.p. HiFi Cas9 Nuclease V3 (IDT, Integrated DNA Technologies) to generate the RNP complex that was mixed with 1 × 10⁶ of the target iPSC line. Electroporation was performed using the 4D-Nucleofector (Lonza), program CA137. The Inference of CRISPR Edits (ICE) online tool (Synthego) was used to determine cleavage and knock-in efficiency. 48-72 hrs after electroporation cells were seeded at low-density (single-cell) then colonies were manually picked for genotyping. The primer set for P305P was Forward: TTCCCACACTCGCTTCGTTA and Reverse: AGCCTGAGAATGTGTCCAGC to amplify the target region. Sanger sequencing was performed to confirm editing using the sequencing primer: CAGCCCAGAAAGGGTGAGAG. For E253E, the primer set Forward: CCCGAAAGAGGACACCGTC and Reverse: GTGGAAACTGCGGGGTCAG was used to amplify the target region. Sanger sequencing was performed to confirm the target editing by using the sequencing primer: CTATGCAGGCTCACCCTCAG. SNP chromosome microarray using the Cytoscan array of all lines was performed by the Molecular Cytogenomics Laboratory at Columbia. All lines were reported as normal (Figure S5B).

Induction and differentiation

Cells were induced toward a neural stem cell (NSC) fate through chemically defined culture according to the manufacturer’s protocol (Gibco A1647801). Briefly, iPSCs were plated on Geltrex-coated plates in Neural Induction Media (Gibco A1647801) and treated with 10 µm Rock Inhibitor (Selleckchem S6390) overnight. After six to eight days, or until cells reached maximum confluency, the NSCs were dissociated into single cells with Accutase (Stem Cell Technologies 07920) and expanded in Neural Expansion Media (Inclusion of Gibco 12634010 according to the manufacturer’s protocol). Upon passaging, NSCs were plated in Neurobasal medium (Gibco, 21103049) to promote differentiation into glutamatergic neurons on poly-D-lysine (Gibco A3890401) and laminin-coated coverslips (Neuvitro Corporation GG-12-PDL) or plates coated with laminin (Gibco A29249), supplemented with B27 (Gibco 17504044), 10 µm ascorbic acid (Gibco A15613.22), 1X GlutaMAX Supplement (Gibco 35050061), and CultureOne Supplement (Gibco A3320201). After 3 to 5 days, or until neural progenitors adopted a neuronal-like morphology with clear projections, B27 supplement was substituted for B27 Plus (Gibco A3582801). Neuronal media was changed three days a week by removing half of the media and adding half fresh media.

Immunostaining

Post NSC induction, immunofluorescent staining was performed for SOX1 (R&D Systems AF3369) at 1:500 and Nestin (R&D Systems MAB2736) at 1:200. Secondaries NorthernLights™ 557-conjugated Anti-Goat IgG Secondary Antibody (R&D Systems NL001) and NorthernLights™ 493-conjugated Anti-Mouse IgG Secondary Antibody (R&D Systems NL009), both at 1:200.

Neurons were cultured on glass-bottom dishes (Mattek P35G-1.5-14-C) coated with poly-D-Lysine (Gibco A3890401) and laminin (Gibco A29249) and immunostained for markers MAP2 (Abcam ab5392, 1:200), VGlut1 (Synaptic Systems 135318, 1:500), KIF1A (Abcam ab180153, 1:50). Immunostaining was performed as previously described^87,88. Briefly, cells were fixed at room temperature for 15 min in 4% PFA (Electron Microscopy Sciences, 15710-S) followed by 3 × 10 min washes in PBS (ThermoFisher 10010023). Cells were incubated at room temperature for 30 minutes in PBS with 0.1% Triton X-100 (Millipore Sigma T9284) and 5% FBS (ThermoFisher A5256701), followed by three washes in PBS. Cells were incubated with primary antibodies at 4 ^oC overnight and washed three times in PBS with 2%. The secondary antibody incubation (ThermoFisher, A32731, A32932, A-21450, all 1:500 for neurons) was applied for 1 h at room temperature, followed by three washes in PBS. DAPI counterstain was applied to immunofluorescent stainings of all cell types at 2 µg/mL (ThermoFisher D1306). Images were taken with a Zeiss LSM 710 confocal microscope. Neurite outgrowth length was determined for all countable outgrowths from a minimum of 50 cells per genotype using the skeletonization and manual tracing functions in Image J as well as the neuroanatomy package in ImageJ (Fiji v2.14.0) software. Neurite lengths were calculated for a minimum of three independent differentiations for each line.

Cell culture DNA extraction

DNA was extracted from iPSCs using Quick Extract (QEF81050 Biosearch Technologies).

RNA extraction

Cells were harvested and RNA extracted in Trizol (15596026 Sigma) and purified using a Zymo RNA clean and concentrator kit (R1013) according to the manufacturer’s instructions.

Reverse transcription

Reverse transcription was done with the Roche Transcriptor First Strand cDNA Synthesis Kit (04379012001) using a mixture of random hexamer primers and an anchored-oligo(dT)18 primer.

Quantitative RT-PCR analysis

RNA was isolated from neurons from independent differentiations as mentioned above for each line in TRIzol Reagent (ThermoFisher) and total mRNA was extracted using standard TRIzol RNA extraction protocol and purified using the Zymo RNA-concentrator kit (Zymo). Up to 1 µg of RNA was reverse transcribed into cDNA using the Transcriptor First Strand cDNA Synthesis Kit (Roche Diagnostics). For qRT-PCR, 10 ng of cDNA was amplified using gene-specific primers and LightCycler 480 SYBR Green I Master (Roche Diagnostics) on a LightCycler 480 real-time PCR machine (Roche Diagnostics). At least two technical replicates and three biological replicates were assayed.

ASO design & treatment

ASOs used in this study were produced by Integrated DNA Technologies (Data S2). ASOs were diluted in DPBS and used at a final concentration of 10 µM to be incorporated via gymnosis. Cells were treated with 10 µm of ASOs on day 7 of differentiation with half-media changes roughly every 3 days. For Fig. 3, cells were harvested on DIV35 for expression analysis to determine ASO efficacy and specificity. For Fig. 4, on DIV35 of differentiation, a subset of cells was harvested for RNA extraction for each ASO treatment and allele-specific knockdown was quantified and cells were fixed and stained for neurite outgrowth analysis.

Transcript quantification

cDNA was amplified using gene-specific primers and LightCycler 480 SYBR Green I Master (Roche) on a LightCycler 480 real-time PCR machine (Roche). Samples were amplified in triplicate (IDT Hs.PT.56a.2993151- KIF1A Fam) and in-well normalization was done using GAPDH (IDT Hs.PT.39a.22214836 Hex). This quantifies the overall transcript level but does not distinguish between the two alleles. To determine the percentage of each allele, we amplified the cDNA across the rs1063353 and subsequently sequenced the product on a NovaSeq using primers. The amplified product was sequenced with approximately 50,000 reads per sample across the region to quantify each allele. To quantify the allelic specificity of the knockdown, we combined the qPCR determined overall expression and the DNA sequencing to quantify allelic distribution using the following formulas:

$$AKD={\left(\frac{A}{A+B}\right)2}^{\left(Cpkc-Cpk1\right)-\left(Cpgc-Cpg1\right)}$$ 1

$$BKD={\left(\frac{B}{A+B}\right)2}^{\left(Cpkc-Cpk1\right)-\left(Cpgc-Cpg1\right)}$$ 2

Where

Cpk1= crossing point KIF1A in sample

Cpkc= crossing point KIF1A in control

Cpg1= crossing point GAPDH in sample

Cpgc= crossing point GAPDH in control

A = NGS reads from allele one

B = NGS reads from allele 2

P305L Primers:

Forward

GGGCCAACATCAACAAGTCG

Reverse

GCGGATCAGCTTGTTGTTGG

E253K Primers:

Forward

GACTCAGGGAACAAGGCCAG

Reverse

GGTCAACACGGAATCTCGGT

C151Y Primers:

Forward

ATCAACTACGCGTCGCAGAA

Reverse

CAGGTCCACCAGGCTGATTT

Statistics

Data analyses were performed using Prism 10 software (GraphPad Software Inc., USA). Multiple unpaired t tests were used with the Holm-Sidak method (alpha = 0.05). Multiple comparisons of ordinary one-way ANOVA performed with Dunnett’s multiple comparisons test (alpha = 0.05). Statistical data was obtained from a minimum of three independent neuronal differentiations for each line and statistical significance was defined by P < 0.05 in all analyses.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Author contributions

All authors discussed the results and contributed to the final manuscript. M.V.Z., N.G., P.S., and W.K.C. conceived and designed the analysis. M.V.Z. and J.H. collected data and performed analyses. E.S. assisted with confocal microscopy and image analysis. A.S. assisted with genotyping and expression experiments and data analysis. D.E.M. performed long-read sequencing and analysis. X.L. and Y.S. assisted with sequencing data analysis. P.L., R.E.Y., and M.V.Z. performed cell culture and genotyping necessary for stem cell derivation and differentiation. T.K. assisted with cell culture experiments and ASO treatments. M.V.Z., C.A.L., and W.K.C. wrote the paper.

Peer review

Peer review information

Nature Communications thanks Shinsuke Niwa, and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. A peer review file is available.

Data availability

The sequencing data generated in this study is made available through the NIH BioProject database under accession code PRJNA1244412, ID: 1244412. Existing data from gnomAD can be accessed and browsed interactively in the browser (https://gnomad.broadinstitute.org/). Existing data from the 1000 Genomes Project can be accessed and browsed interactively in the browser (https://www.internationalgenome.org/). For 1000 Genomes Project data, we download the genotype data file from 20100804 release from NIH FTP site: https://urldefense.com/v3/__https://ftp-trace.ncbi.nih.gov/1000genomes/ftp/release/20100804/__;!!NZvER7FxgEiBAiR_!rw5tB3Oau9SjDPXvP_Rora55IA3XU85XxOQyfYqzeEzon-5vZj-jMC-a0JfPOjsYLrBOc_Tev-pIchqW7UbujulPywgdwEUEbXVLEHk$ The genotype file is: ALL.2of4intersection.20100804.genotypes.vcf.gz Index: ALL.2of4intersection.20100804.genotypes.vcf.gz.tbi We then extract the KIF1A region by tabix: tabix -h ALL.2of4intersection.20100804.genotypes.vcf.gz 2:240737000-240821036 > KIF1A_region.1 kg.vcf For gnomAD data, we obtained allele frequency information of chromosome 2 (VCF file) downloaded from gnomAD website: https://urldefense.com/v3/__https://gnomad.broadinstitute.org/data__;!!NZvER7FxgEiBAiR_!rw5tB3Oau9SjDPXvP_Rora55IA3XU85XxOQyfYqzeEzon-5vZj-jMC-a0JfPOjsYLrBOc_Tev-pIchqW7UbujulPywgdwEUEx5GzeDU$. Source data are provided with this paper.

Code availability

Code for the greedy algorithm utilized is available through Zenodo: Shen, Y. Antisense oligonucleotides to KIF1A polymorphisms expand targets and rescue patient-derived neurons in vitro. KIF1A. 17253681. (2025). 10.5281/zenodo.17253681.

Competing interests

Wendy K. Chung has received research funding from Ovid Therapeutics. Noelle Germaine, Tae Hyun, and Patrick Sarmiere are former employees and consultants of Ovid Therapeutics. Danny E. Miller is on scientific advisory boards at Oxford Nanopore Technologies (ONT) and Basis Genetics, is engaged in a research agreement with ONT, has received research and travel support from ONT, and holds stock options in MyOme and Basis Genetics. The remaining authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s41467-025-67752-y.