TFG regulates secretory and endosomal sorting pathways in neurons to promote their activity and maintenance

Significance

Pathological mutations in TFG have been implicated in a variety of neurodegenerative diseases, including early-onset forms of hereditary spastic paraplegia (HSP). In this work, we demonstrate that the TFG p.R106C mutation, which has been identified in several children suffering from HSP, is directly responsible for progressive gait abnormalities, central nervous system pathology, altered muscle electrophysiology, and axonal degeneration within one of the major neuronal pathways involved in coordinated movement. At the cellular level, we additionally demonstrate that the TFG p.R106C mutation impairs two distinct membrane trafficking pathways required for cargo movement within neurons, leading to defects in neuronal function. Our studies strongly suggest that TFG regulates both secretory and endosomal cargo sorting to enable neuronal maintenance.

Abstract

Molecular pathways that intrinsically regulate neuronal maintenance are poorly understood, but rare pathogenic mutations that underlie neurodegenerative disease can offer important insights into the mechanisms that facilitate lifelong neuronal function. Here, we leverage a rat model to demonstrate directly that the TFG p.R106C variant implicated previously in complicated forms of hereditary spastic paraplegia (HSP) underlies progressive spastic paraparesis with accompanying ventriculomegaly and thinning of the corpus callosum, consistent with disease phenotypes identified in adolescent patients. Analyses of primary cortical neurons obtained from CRISPR-Cas9–edited animals reveal a kinetic delay in biosynthetic secretory protein transport from the endoplasmic reticulum (ER), in agreement with prior induced pluripotent stem cell–based studies. Moreover, we identify an unexpected role for TFG in the trafficking of Rab4A-positive recycling endosomes specifically within axons and dendrites. Impaired TFG function compromises the transport of at least a subset of endosomal cargoes, which we show results in down-regulated inhibitory receptor signaling that may contribute to excitation-inhibition imbalances. In contrast, the morphology and trafficking of other organelles, including mitochondria and lysosomes, are unaffected by the TFG p.R106C mutation. Our findings demonstrate a multifaceted role for TFG in secretory and endosomal protein sorting that is unique to cells of the central nervous system and highlight the importance of these pathways to maintenance of corticospinal tract motor neurons.

The corticospinal tract (CST; also known as the pyramidal tract) is one of the major pathways involved in relaying information from the cerebral cortex of the brain to the spinal cord to enable coordinated walking as well as other movements (1, 2). Injury to the upper motor neurons that form the CST, which can result from genetically inherited pathogenic mutations or traumatic injury, impairs motor function and may lead to lifelong disability (3, 4). The mechanisms that underlie neurodegeneration following an insult remain largely unknown, at least in part due to an insufficient understanding of the fundamental systems that act normally to maintain neuronal function. However, key insights into these homeostatic pathways have been gleaned from the identification and characterization of rare genetic variants implicated in neurodegenerative disease. In the case of hereditary spastic paraplegias (HSPs), which are characterized by progressive weakness and spasticity of the limbs, mutations in more than 80 distinct genes have been linked to neurodegeneration within the CST, with a majority encoding factors that regulate subcellular organelle dynamics and function (5, 6). In particular, multiple recessive point mutations in the Trk-fused gene (TFG; also called SPG57), which regulates early secretory pathway organization, have been suggested to cause early-onset forms of HSP, although direct impacts on neuronal physiology and behavior have yet to be defined (7–11).

TFG was originally identified as part of an oncogenic fusion with the TrkA receptor tyrosine kinase (12). Subsequently, its native role during the earliest stages of biosynthetic membrane protein transport was revealed. Specifically, TFG functions to cluster and/or uncoat COPII transport carriers following their budding from subdomains on the endoplasmic reticulum (ER) (13–16). Consistent with this activity, inhibition of TFG kinetically delays the movement of cargoes from the ER to the ER–Golgi intermediate compartments (ERGICs), elevating ER stress, reducing cell proliferation, and promoting apoptosis when depletion is highly penetrant (14–19). Germ line deletion in metazoan systems, including Caenorhabditis elegans and rodents, results in early embryonic lethality (13, 15), but numerous point mutations have been identified in patients suffering from various neurological disorders, including HSP, amyotrophic lateral sclerosis (ALS), hereditary motor and sensory neuropathy with proximal dominant involvement, Charcot-Marie-Tooth disease type 2, and Parkinson’s disease (7–11, 20–24). Based on in situ hybridization studies and immunohistochemistry, TFG expression has been observed throughout the brain and spinal cord, but it is also seen in many other tissues (7, 25, 26). Thus, it is unknown why pathological TFG variants primarily impact the nervous system, especially considering the gene is expressed ubiquitously.

The inheritance patterns of TFG mutations that have been identified in the population vary, as do the disease phenotypes associated with each. Mutations found within the carboxyl-terminal disordered region appear to act dominantly and encode proteins that exhibit a propensity to form amyloid fibrils and have been implicated in late-onset neuropathies (20, 22, 24, 27–31). In contrast, those in the structured amino-terminal PB1 and coiled-coil domains are autosomal recessive, impair the ability of TFG to assemble into octameric ring structures, and have been associated with early-onset, complicated forms of HSP (7–11, 14, 32). Likely due to its repetitive identification in several independent families, multiple studies have specifically attempted to define the phenotypic impacts resulting from the TFG p.R106C mutation found within the coiled-coil domain (7, 11, 14, 32–34). Using patient-derived fibroblasts homozygous for the variant, roles for TFG in autophagy and mitochondrial function have been suggested (33, 34). Additionally, ectopic overexpression of TFG p.R106C in primary mouse hippocampal neurons leads to mitochondrial fragmentation (11), while induced pluripotent stem cell (iPSC)–derived cortical neurons natively expressing TFG p.R106C in a homozygous manner exhibit defects in axon fasciculation (32). Other studies have further implicated TFG in innate immunity and regulation of the ubiquitin-proteasome system (29, 35). Despite these many efforts, however, it continues to remain unclear whether a mutation in TFG is sufficient to cause the motor dysfunction and spasticity phenotypes associated with HSP, nor it is evident how neuronal function is negatively affected by pathological variants.

To address these issues, we used CRISPR-Cas9–mediated genome editing to introduce the TFG p.R106C mutation into Sprague–Dawley rats. Our use of rats as opposed to mice was intentional, since rats have consistently been found to be more comparable to primates, both anatomically and physiologically (36, 37). We demonstrate that rats homozygous for the TFG p.R106C variant develop progressive motor deficits, thinning of the corpus callosum, ventriculomegaly, and hind limb spasticity, each independent of sex, recapitulating the major phenotypes displayed by HSP patients (38). Additionally, we show that trafficking of an integral membrane protein through the early secretory pathway is slowed in primary dissociated cortical neurons expressing the TFG p.R106C mutation, consistent with our previous studies using human iPSC-derived neurons (32). Moreover, we found that TFG is also targeted to another subcellular compartment in neurons beyond the ER/ERGIC interface. Specifically in axons and dendrites, TFG localizes with Rab4A-positive endosomes and regulates their transport, thereby impacting the trafficking of an additional set of cargoes, some of which function in neuronal excitability. In contrast, we failed to identify lysosomal defects resulting from the TFG p.R106C mutation, distinguishing it from several other variants implicated in HSP (39, 40). Together, our findings highlight multiple roles for TFG function in neurons, uniquely sensitizing them to pathological TFG point mutations that otherwise have more limited impacts on other cell types.

Results

The TFG p.R106C Mutation Causes Progressive Gait Dysfunction.

Exome sequencing studies have linked numerous missense mutations in TFG to rare forms of neurodegenerative disease (7–11, 20–24, 30, 31, 33, 41, 42). However, none have been directly shown to be causative of phenotypes that are associated with neurodegeneration. To address this problem, we used CRISPR-Cas9–mediated genome editing in embryos from Sprague–Dawley rats to introduce the TFG p.R106C mutation, which has been associated with complicated forms of HSP in multiple independent families (7, 9, 11, 33). Two additional silent mutations were incorporated in parallel to enhance editing efficiency (Fig. 1A), which were ultimately found not to affect gene expression (SI Appendix, Fig. S1A). In total, three founders were identified (two males and one female), all of which were heterozygous for the point mutation desired (c.316C > T) based on next generation sequencing. The founders were backcrossed independently to control animals, and their progeny were subjected to sequencing. Three animals bred from distinct founders that expressed the TFG p.R106C variant were selected based on the absence of mutations around the 10 most highly ranked potential off-target sites. Each was independently backcrossed an additional four times to reduce the potential existence of off-target effects that may have occurred during the initial editing experiments. Since no differences in fecundity or viability were observed between the three resulting colonies, a single one was selected for continued maintenance. To reduce the likelihood of suppressor mutations arising in the population, only heterozygous TFG p.R106C animals were used in all subsequent breeding.

Fig. 1.

Homozygous TFG p.R106C mutant animals exhibit progressive motor deficits. (A) Schematic illustrating the editing approach used to incorporate the TFG p.R106C mutation into the genome of Sprague–Dawley rats is shown (Top), as well as a chromatogram obtained following Sanger sequencing of a homozygous mutant animal (Bottom). (B) Representative image of a marked animal traversing the MotoRater platform (Top). Schematics illustrate the manner by which hind body sway, tail tip height, and step cycle duration were quantified (Bottom). (C, E, and F) Measurements of hind body sway (C), step cycle duration (E), and tail tip height (F) of control, heterozygous TFG p.R106C, and homozygous TFG p.R106C mutant animals at different ages indicated (n, number of animals assayed; M, male; F, female). Error bars represent mean ± SEM. *P < 0.05, **P < 0.01, and ***P < 0.001, as calculated using an ANOVA test. (D) Representative traces of the tail base point of 25-wk-old animals (control and homozygous TFG p.R106C) as they traverse the platform. WT, wild-type; HDR, Homology-Directed Repair; gRNA, guide Ribonucleic acid; PAM, protospacer adjacent motif.

Diagnostic criteria for HSPs typically include slowly progressing spastic weakness of the legs that results in ambulatory dysfunction (43). To determine whether animals expressing the TFG p.R106C mutation exhibit progressive difficulty walking, we conducted a series of quantitative kinematic gait analyses by recording animals from three different vantage points as they traversed a clear platform (Fig. 1B). Animals of both sexes were marked with small black spots on multiple body parts and examined at 6, 13, and 25 wk of age. The markings were used to train a neural network model based on feature detectors from previously generated pose estimation networks (44). The positional coordinates of each point over time were then used to calculate various kinematic parameters. Consistent with progressive motor dysfunction that predominantly impacts the lower body of HSP patients, rats homozygous for the TFG p.R106C mutation displayed significant gait deficits compared to control animals, which worsened with age (Fig. 1 C–F and SI Appendix, Fig. S1 B–E). In particular, the stereotypic oscillating pattern of side-to-side hind body movement during walking was significantly disrupted in homozygous mutant animals as they grew older. Although initially normal, by 13 wk of age, both male and female rats homozygous for the TFG p.R106C mutation exhibited exaggerated side-to-side hind body movements, which became even more pronounced in 25-wk-old animals (Fig. 1 C–E). Additionally, homozygous mutant animals exhibited a progressively diminished ability to maintain normal tail height during walking (Fig. 1F), a phenotype observed in rodent models of ALS and indicative of a reduced ability to control muscles important for hind body motion (45). Cross-body movement coordination, as defined by forelimb movement on one side of the body being in concert with the hind limb of the opposite side, was similarly impaired by TFG dysfunction (Movies S1–S3).

We also analyzed the stride of animals, demonstrating that rats homozygous for the TFG p.R106C mutation exhibited a progressively extended period of time when each limb made contact with the platform (stance phase) (SI Appendix, Fig. S1 B–D). The duration of the swing phase, however, when no limb contact is made with the platform, was unaffected by the TFG mutation. Average walking velocity was also normal in homozygous TFG p.R106C animals (SI Appendix, Fig. S1 E), and as we expected due to its recessive nature, heterozygous TFG p.R106C animals did not exhibit any gait defects compared to control animals (Fig. 1 C–F and SI Appendix, Fig. S1 B and E). Taken together, our data indicate that the TFG p.R106C mutation is responsible for gait abnormalities and hind limb motor defects in mammals, which progressively reduce hind body stability, consistent with progressive lower limb dysfunction observed in HSP patients (38).

The TFG p.R106C Mutation Causes Alterations in Hind Limb Electrophysiology and Central Nervous System (CNS) Pathology.

Lower limb spasticity is a hallmark of HSP that is typically diagnosed on physical examination (46). In patients suffering from complicated forms of the disease, including those harboring the TFG p.R106C mutation, electromyography (EMG) targeting a resting leg muscle has been used to highlight lower motor neuron dysfunction (7, 33). To determine whether this mutation in TFG is specifically responsible for altered skeletal muscle electrical activity, we used EMG to examine the hind limb gastrocnemius muscles of our CRISPR-Cas9–modified animals, anesthetized using ketamine. At 6 wk of age, control and TFG mutant animals showed no significant differences in spontaneous electrical activity in these muscles (Fig. 2 A and B). However, 13-wk-old homozygous TFG p.R106C animals exhibited an increasing frequency of EMG spikes, reaching statistical significance at 25 wk of age (Fig. 2 A and B). The irregular firing pattern identified in the homozygous mutant animals suggested muscle fasciculation consistent with decreased inhibition of motor neurons, as opposed to muscle denervation. These findings contrast the effect of deleting TFG in motor neurons, which results in denervation of neuromuscular junctions (47), strongly suggesting that the p.R106C mutation only partially reduced TFG function. Notably, neither control nor heterozygous TFG p.R106C mutant animals showed similar signs of muscle spasticity, consistent with the point mutation acting recessively (Fig. 2 A and B).

Fig. 2.

Homozygous TFG p.R106C mutant animals exhibit spontaneous electrical activity in hind limb skeletal muscles. (A) Representative electromyograms recorded from 25-wk-old male animals (control and homozygous TFG p.R106C). (B) Quantification of the average number of spikes greater than 20 μV over a 40-s recording period in control, heterozygous TFG p.R106C, and homozygous TFG p.R106C mutant animals (n, number of animals assayed; M, male; F, female). Box and whisker plots show data in the 10th to 90th percentile. *P < 0.05 as calculated by a Dunn’s multiple comparisons test following a Kruskal-Wallis test. WT, wild-type.

MRI is widely used in clinical settings to help differentiate distinct groups of HSP patients (48). Using this approach, specific brain MRI patterns have been identified that appear to be characteristic of several complicated forms of HSP (49, 50). To determine the impact of the TFG p.R106C mutation on brain morphology, we harvested tissue from control and mutant animals at various time points (6, 13, and 25 wk of age, identical to those used in gait analysis studies) during development and conducted H&E (hematoxylin and eosin) and LFB (Luxol fast blue) staining of paraffin-embedded sections. Consistent with neuroimaging of adolescent patients with the TFG p.R106C mutation, rats homozygous for the variant exhibited thinning of the corpus callosum, while no such defect was seen in control or heterozygous animals (Fig. 3 A and B). Additionally, we found that the ventricles of homozygous TFG p.R106C animals became progressively dilated (Fig. 3 A and C), a phenotype associated previously with the TFG p.R106C mutation in patients (33), as well as a complicated, X-linked form of HSP known as MASA syndrome (Mental retardation, Aphasia, Shuffling gait, and Adducted thumbs) that results from mutations in SPG1, which encodes the axonal cell adhesion molecule L1CAM (51, 52).

Fig. 3.

Homozygous TFG p.R106C mutant animals exhibit progressive CNS pathology that is consistent with HSP. (A) Representative images of coronal brain sections from 25-wk-old animals (control and homozygous TFG p.R106C) stained with LFB. Scale bar, 1 mm. (B, C, E, and F) Quantification of the surface area of the corpus callosum (B), the surface area of the ventricles based on H&E staining (C), microglia density based on the presence of Iba1-positive cells in the primary motor cortex (E), and astrocyte density based on the presence of S100beta-positive cells in the primary motor cortex (F) in control, heterozygous TFG p.R106C, and homozygous TFG p.R106C mutant animals (n, number of animals assayed; M, male; F, female). Error bars represent mean ± SEM. **P < 0.01 and ***P < 0.001, as calculated using an ANOVA test. (D) Representative electron micrographs of the CST (lumbar region of the spinal cord) from 25-wk-old animals (Left; control and homozygous TFG p.R106C). Scale bar, 2 μm. Quantification of myelin sheath thickness based on the G-ratio is also shown (Right). Error bars represent mean ± SEM. ***P < 0.001, as calculated using an unpaired t test. WT, wild-type.

In addition to brain morphology, we also examined the axons of upper motor neurons within the CST, which are among the longest in animals and have been shown to degenerate in HSP (5). Specifically, we used electron microscopy to measure the thickness of myelin sheaths in the lumbar segments of the spinal cord. These studies demonstrated reduced CST axon myelination in homozygous TFG p.R106C animals relative to controls, consistent with motor deficits being largely restricted to the hind limbs (Fig. 3D). Finally, based on previous work showing an elevated inflammatory response in many forms of neurodegenerative disease (53, 54), we examined the distribution of Iba1-positive microglia and S100beta-positive astrocytes throughout the motor cortex of control and mutant animals using immunohistochemistry. These studies demonstrated increased density of both cell types in 25-wk-old homozygous TFG p.R106C animals compared to controls, suggesting that CNS inflammation and astrogliosis accompany progressive neuronal dysfunction in HSP (Fig. 3 E and F).

The TFG p.R106C Mutation Impairs Neuronal Protein Trafficking from the ER.

Our previous work using human stem cell–derived cortical neurons indicated that the accumulation of a cell surface axonal membrane protein (L1CAM) was diminished when TFG function was perturbed, potentially due to reduced secretory protein transport (32). To independently confirm and extend this finding, we used primary rat cortical neurons harvested from control and TFG p.R106C mutant embryos and quantitatively imaged surface L1CAM at different time points after plating. Following 10 d in culture, homozygous TFG p.R106C mutant neurons exhibited significantly less L1CAM labeling on the plasma membrane compared to controls, despite similar levels of the protein being expressed irrespective of genotype (Fig. 4 A and B). In contrast, TFG p.R106C mutant neurons cultured for 14 d in vitro exhibited no significant differences in surface L1CAM accumulation compared to controls, suggesting only a kinetic delay in secretory cargo transport when TFG is mutated (SI Appendix, Fig. S2A). To directly test this idea, we leveraged an inducible release system in which L1CAM fused to HaloTag is initially trapped in the ER lumen but remains capable of entering the secretory pathway following its disaggregation, which is mediated by an inert rapamycin analog referred to as DDS (dimer-dimer solubilizer) (55, 56). In control primary rat cortical neurons, released L1CAM concentrated quickly within the peri-nuclear Golgi after DDS addition (Fig. 4C). In contrast, neurons homozygous for the TFG p.R106C mutation exhibited a significant delay in L1CAM transport through the early secretory pathway (Fig. 4C).

Fig. 4.

The TFG p.R106C mutation reduces the kinetics of secretory protein transport. (A) Representative images of primary rat cortical neurons from control and homozygous TFG p.R106C mutant animals grown in vitro for 10 d and stained using antibodies accessible only to cell surface L1CAM (Top). Scale bar, 10 μm. Quantification of the relative levels of surface L1CAM intensity as determined for neurons grown in vitro for 10 d that are heterozygous or homozygous for the TFG p.R106C mutation compared to control neurons (Bottom; three biological replicates each). Error bars represent mean ± SEM. ***P < 0.001, as calculated using an ANOVA test. (B) Representative immunoblot analysis of control, heterozygous TFG p.R106C, and homozygous TFG p.R106C mutant neurons grown for 10 d in vitro using antibodies directed against L1CAM (Top, Left) and beta-actin (Bottom, Left). Quantification of the relative levels of L1CAM in neurons grown in vitro for 10 d that are heterozygous or homozygous for the TFG p.R106C mutation compared to control neurons (Right; four biological replicates). No statistically significant differences were found. (C) Representative images of control and homozygous TFG p.R106C neurons grown for 14 d in vitro that were transiently transfected with a releasable form of HaloTag-L1CAM following dye-labeling with the JFX650-HaloTag ligand (Left). Neurons were fixed and stained using antibodies directed against GM130 before and after the addition of DDS. Scale bar, 10 μm. Quantification of Golgi-localized HaloTag-L1CAM intensity at various time points after release from the ER in control, heterozygous TFG p.R106C, and homozygous TFG p.R106C mutant neurons (Right; three biological replicates each). Error bars represent mean ± SEM. ***P < 0.001, as calculated using an ANOVA test. (D and E) Representative images of control and homozygous TFG p.R106C mutant neurons grown 14 d in vitro and immunostained using antibodies directed against TFG (Left). Scale bars, 10 μm (D) and 2 μm (E). Quantification shows the relative number of high-intensity TFG structures in the soma of heterozygous and homozygous TFG p.R106C mutant neurons grown in vitro for 7 d compared to control neurons in the soma (D, Right) and in neurites (E, Right; at least 30 μm of neurite length examined per sample), where values are shown as a percentage of the total number of TFG structures analyzed (Right; three biological replicates each). Error bars represent mean ± SEM. ***P < 0.001 and *P < 0.05, as calculated using an ANOVA test. (F) Representative images of control iPSC-derived neurons immunostained using antibodies directed against TFG (red; Left) and TRIM46 (green; Left). Scale bars, 10 μm and 5 μm (zoomed panel). Quantification of the number of TFG structures present in axons and dendrites is also shown (Right; three biological replicates each). Error bars represent mean ± SEM. No statistically significant differences were identified. WT, wild-type; DIV10, 10 days in vitroh; IPSC, human induced pluripotent stem cell; AIS, axon initial segment.

We previously used single-particle electron microscopy to demonstrate that TFG assembles into small octameric ring structures, which can further self-associate to form liquid-like condensates at the interface between ER and ERGIC membranes (14–16, 28). Additionally, we found that the p.R106C mutation within the TFG coiled-coil domain disrupts its ability to form rings in vitro, potentially impacting its capacity to phase separate in cells (32). To examine this possibility, we conducted a series of immunofluorescence studies using primary rat cortical neurons harvested from our animal models. In control neurons, TFG was found in high-, medium-, and low-intensity structures in both cell bodies and neuronal processes (Fig. 4 D and E and SI Appendix, Fig. S2B). However, in homozygous TFG p.R106C neurons, the number of TFG-positive structures, particularly those of high intensity, was significantly decreased compared to controls, suggesting a defect in the ability of TFG to concentrate and form condensates (Fig. 4 D and E), Importantly, the loss of TFG-positive structures was not due to diminished stability of the protein based on immunoblotting studies of control and TFG p.R106C mutant brain extracts (SI Appendix, Fig. S2C). Based on the critical role for TFG in the early secretory pathway, these findings offer a potential explanation for the reduced kinetics of L1CAM trafficking and axon plasma membrane accumulation in neurons expressing the TFG p.R106C mutation.

TFG Regulates the Trafficking of Rab4A-Positive Endosomes in Axons and Dendrites.

Localization studies in multiple cell types across a variety of species indicate that TFG functions in the early secretory pathway (13–17, 34, 57, 58). In neurons, the majority of protein synthesis is believed to occur within cell bodies, which are enriched with rough ER and Golgi membranes (59–61). However, we found that TFG additionally localizes to structures within axons and dendrites, with similar densities in each (Fig. 4F). To determine whether all TFG-positive structures correspond to sites of COPII transport carrier formation, we costained neurons with antibodies directed against TFG and Sec31A, a component of the outer COPII coat complex. For these studies, we used human stem cell–derived cortical neurons, which enabled better resolution of individual neuronal processes compared to primary cortical neuron cultures. Within cell bodies, TFG colocalized tightly with Sec31A (Fig. 5A), similar to our findings in epithelial cells (14, 15). However, within neurites, only a minority of TFG-positive structures localized together with Sec31A (Fig. 5A). These data indicate that specifically in axons and dendrites, TFG is distributed beyond the ER/ERGIC interface.

Fig. 5.

TFG regulates the dynamics of Rab4A-positive endosomes. (A) Representative images of control iPSC-derived neurons immunostained using antibodies directed against Sec31A (green; Left) and TFG (red; Left). Scale bar, 10 μm. Quantification of the percentage of TFG that colocalizes with Sec31A in the soma and in neuronal processes is also shown (Right; three biological replicates with more than 3,500 structures analyzed). Arrows highlight a structure where Sec31A and TFG are both present. Arrowheads highlight a structure which harbors TFG but not Sec31A. Error bars represent mean ± SEM. (B) Representative images of control iPSC-derived neurons expressing a GFP fusion to TFG (red) from its endogenous locus and mScarlet-Rab4A (green) transduced following lentiviral infection in different locations along neurites (soma proximal, within 70 µm of the soma; medial, >100 µm from the soma and >150 µm from the growth cone; distal, within 150 µm of the growth cone). Arrows highlight sites of colocalization. Scale bar, 10 μm. (C) Quantification of the percentage of Rab4A that colocalizes with TFG (Top) and the percentage of TFG that colocalizes with Rab4A (Bottom) in different regions of neuronal processes is shown (three biological replicates). Error bars represent mean ± SEM. ***P < 0.001 and *P < 0.05, as calculated using an ANOVA test. (D) Representative images of control and homozygous TFG p.R106C mutant iPSC-derived neurons transduced with lentivirus to express mScarlet-Rab4A. Arrowheads highlight Rab4A-positive structures. Scale bar, 10 μm. (E and F) Quantification of the density (E) and velocity (F) of Rab4A-positive structures in control and TFG p.R106C mutant neurons (three biological replicates each). Error bars represent mean ± SEM. ***P < 0.001 and *P < 0.05, as calculated using an ANOVA test. WT, wild-type; hiPSC, human induced pluripotent stem cell.

To define the localization of TFG in neurites, we first engineered human iPSCs to express a GFP fusion to the endogenous protein using CRISRP-Cas9–mediated editing. Following neuronal differentiation, we examined the relative distribution of a number of different organelle markers and found that TFG colocalized specifically with Rab4A-positive endosomes in neurites, with increasing frequency farther away from the cell body (Fig. 5 B and C). In contrast, TFG and Rab4A exhibited minimal colocalization in processes that were proximal (less than 70 μm) to cell bodies (Fig. 5 B and C).

To assess a functional role for TFG at Rab4A-positive endosomes, we first examined their distribution in control and homozygous TFG p.R106C mutant neurons derived from human iPSCs. These studies demonstrated that the number of Rab4A-positive endosomes was significantly reduced in neurites when TFG function was impaired compared to controls (Fig. 5 D and E). Additionally, the average velocity of Rab4A-positive endosomes was decreased in homozygous TFG p.R106C primary rat cortical neurons relative to controls (Fig. 5F). Together, these data indicate that TFG localizes with Rab4A-positive endosomes to regulate their trafficking in axons and dendrites. In contrast, we failed to identify impacts to Golgi volume, mitochondrial morphology, or lysosome diameter in neurons homozygous for the TFG p.R106C mutation, arguing against the idea that the variant causes pleiotropic defects on organelle structure/function (SI Appendix, Fig. S2 D–F).

The TFG p.R106C Mutation Perturbs Gephyrin Trafficking to Alter Inhibitory Postsynaptic Currents.

To gain further insights into the role of TFG in neuronal membrane trafficking, we immunoprecipitated it from rat brain extracts and used mass spectrometry to define potential interacting proteins. To our surprise, we identified gephyrin, a scaffolding protein important for clustering inhibitory postsynaptic receptors (62–64), at high sequence coverage in TFG immunoprecipitates (Fig. 6A). To confirm this association, TFG and gephyrin were expressed recombinantly in bacteria and shown to interact in pull-down studies (Fig. 6B).

Fig. 6.

TFG regulates the endosomal trafficking of gephyrin in neurons. (A) Representative Coomassie-stained SDS-PAGE gel shows proteins recovered and identified using mass spectrometry following TFG immunoprecipitation from rat brain extracts (Left; three biological replicates performed). The distribution of peptides identified by mass spectrometry together with the percentage sequence coverage for gephyrin and TFG is also shown (red; Right). (B) Representative images of a Coomassie-stained gel (Top) and immunoblot (Bottom) are shown following a GST pull-down experiment from extracts coexpressing His-SUMO-TFG and either GST or a GST fusion to gephyrin (three biological replicates performed). (C) Representative images of control primary cortical rat neurons grown in culture for 14 d, coexpressing an mScarlet fusion to Rab4A (red; transduced following lentiviral infection) and probe for gephyrin fused to GFP (green; following transfection). Arrowheads highlight sites where Rab4A and gephyrin are transported together within neurites. Scale bar, 10 μm. (D) Quantification of the percentage of Rab4A that colocalizes with a probe for PI3P (2xFYVE) and gephyrin in different regions of neuronal processes is shown (three biological replicates; neurons grown 14 d in vitro). Error bars represent mean ± SEM. (E and F) Quantification of the relative intensities of gephyrin (E) and the GABA_A receptor (GABA_AR; F) that were directly juxtaposed to VGAT staining in control and TFG p.R106C mutant primary rat cortical neurons grown in culture for 21 d (three biological replicates each). Error bars represent mean ± SEM. *P < 0.05, as calculated using an ANOVA test. (G) Schematic showing whole-cell patch-clamp measurement of GABAergic currents in culture cortical neurons. (H) Representative traces of mIPSC recordings obtained from control, heterozygous TFG p.R106C, and homozygous TFG p.R106C mutant neurons. (I) Violin plot of mIPSCs recorded from control, heterozygous TFG p.R106C, and homozygous TFG p.R106C mutant neurons (700 to 1,100 events per genotype). Horizontal black bars demarcate quartiles. The plot was truncated at −120 pA for clarity of display. ***P < 0.001, as calculated using a Kruskal-Wallis test. WT, wild-type; MWM, Molecular Weight Marker; IP, immunoprecipitation.

Previous studies have shown that dendritic trafficking of gephyrin is mediated by endosomes containing phosphatidylinositol 3-phosphate (PI3P), a lipid that can be detected on membranes using a green fluorescent protein (GFP) fusion to the FYVE domain from Hrs (65, 66). Based on our work demonstrating that TFG regulates endosomal trafficking, we aimed to determine whether gephyrin uses a Rab4A-dependent pathway to accumulate at synapses. For these studies, a GFP-tagged probe for gephyrin (67) was transiently transfected into primary rat cortical neurons that had been transduced with an mScarlet fusion to Rab4A. Imaging of their relative movements in neurites using spinning disk confocal microscopy demonstrated that gephyrin moves on Rab4A-positive endosomes prior to accumulating at synapses (Fig. 6 C and D), and consistent with previous work, we found that the majority of Rab4A-positive endosomes also harbor PI3P (SI Appendix, Fig. S3A).

To determine whether TFG dysfunction impairs gephyrin transport, we measured the concentration of gephyrin at inhibitory synapses marked by the γ-aminobutyric acid (GABA) transporter VGAT (68) in control and mutant neurons using quantitative fluorescence microscopy. These studies revealed that gephyrin accumulation at these sites was significantly reduced in neurons homozygous for the TFG p.R106C variant compared to controls (Fig. 6E). Similarly, we found significantly reduced levels of the GABA_A receptor at synapses in mutant neurons compared to control neurons (Fig. 6F).

A predictor of decreased inhibitory receptor concentration at synapses is altered neuronal electrophysiology. To explore this possibility, primary rat cortical neurons from control animals and animals homozygous for the TFG p.R106C mutation were grown in vitro and subjected to patch clamp studies (Fig. 6G). Although we failed to show detectable changes in the frequency or kinetics of spontaneous neurotransmitter release (SI Appendix, Fig. S3 B and C), we identified a significant decrease in the amplitude of spontaneous inhibitory postsynaptic currents in neurons homozygous for the TFG p.R106C mutation compared to controls (Fig. 6 H and I). These data are consistent with TFG dysfunction resulting in diminished inhibitory receptor signaling in neurons.

Discussion

To define the impacts of HSP-associated variants, previous studies have typically relied upon ectopic overexpression of mutant transgenes or approaches that aim to eliminate gene products (69). While these efforts have provided glimpses into the effects of various missense mutations implicated in disease, there are several concerns regarding their use. In particular, overexpression typically perturbs the dynamics and/or activity of proteins, leading to results that lack physiological relevance. Additionally, most mutations that cause HSP alter protein function, as opposed to abolishing it (5, 69). Over the past two decades, more than 20 different mouse models harboring mutations related to those found in HSP patients have been generated, but few have exhibited motor deficits analogous to phenotypes exhibited by humans (6, 70). On the occasions when ambulatory dysfunction was observed, onset typically occurred late during development, often more than a year after birth, creating challenges in defining the etiology of disease (71–73). In other cases, immortalized cell lines or patient-derived fibroblasts have been employed, but these are limited as accurate models given their derivation from diseased tissue or a lineage that has no relevance to CNS function (74). The most frequently used cells are also aneuploid, leading to gene misexpression and genomic instability, which is a source of variation between different isolates of the same cell line and even between cells within a clonal population. Overall, these issues have impeded progress in the field to delineate the molecular basis of phenotypes associated with HSPs.

As an alternative approach, we leveraged the precision of CRISPR-Cas9–mediated gene editing technology to study the specific impacts of a recessive missense mutation associated with a complicated form of HSP using Sprague–Dawley rats as a model system. At a fundamental level, the rat motor cortex exhibits a similar architecture to that observed in humans, subdivided into distinct regions that perform unique functions while still exhibiting a high degree of connectivity (36). Based on our findings, the rat TFG p.R106C model recapitulates many of the pathological phenotypes associated with complicated HSPs, including progressive motor dysfunction with early onset, hind limb spasticity, thinning of the corpus callosum, and signs of axon degeneration within the CST. This contrasts numerous mouse models, including deletion of SPG15, the second-most common cause of autosomal recessive HSP with thin corpus callosum (73, 75). Specifically, SPG15 knockout mice do not exhibit significant motor abnormalities until a year of age, nearly half the normal life span of C57BL/6 animals, and thinning of the corpus callosum has not been observed even after 16 mo (73). Similarly, mouse models harboring pathological mutations in SPG4, the most common cause of HSP, exhibit only mild gait phenotypes with late onset (76–81). Taken together, our data indicate that the TFG p.R106C rat model represents an ideal setting to explore the molecular basis of neuronal dysfunction that underlies HSP.

Previous studies in our laboratory have leveraged human iPSCs that were edited to express the TFG p.R106C mutation and subsequently differentiated into glutamatergic cortical neurons (32). Analysis of these neurons revealed a defect in axon fasciculation that was caused at least in part by impaired accumulation of the cell adhesion molecule L1CAM at the axonal plasma membrane. Similar to TFG, mutations in L1CAM also underlie complicated forms of HSP with early onset, and patients often present with a thin or absent corpus callosum (51, 52, 82). Using primary cortical neurons obtained from our rat TFG p.R106C model, we have now independently confirmed a defect in the normal targeting of L1CAM to the surface of axons. Furthermore, our data suggest that this likely results from a kinetic delay in L1CAM trafficking through the early secretory pathway, consistent with a role for TFG at the ER/ERGIC interface (15, 16). Based on recent work, such a defect would impair the innervation of CST pyramidal neurons by inhibitory chandelier cells, resulting in an imbalance between excitatory and inhibitory synaptic inputs and potentially contributing to motor dysfunction (83).

In addition, our studies have separately uncovered a role for TFG within neuronal processes that further regulates the balance between neuronal excitation and inhibition. Specifically, the TFG p.R106C mutation disrupts normal endosomal trafficking of gephyrin to synapses, thereby impairing inhibitory GABAergic signaling (64). However, gephyrin and L1CAM are unlikely to be the sole cargoes impacted by mutations in TFG. Instead, our data suggest that TFG regulates at least two unique and essential cargo trafficking pathways in neurons, both COPII-mediated secretory protein transport and Rab4A-dependent endosomal sorting, which likely elevates the susceptibility of neuronal cells to mutations in TFG (Fig. 7). Consistent with this idea, patients expressing TFG variants invariably present with nervous system dysfunction, while nonneuronal phenotypes are rare, despite data demonstrating that TFG regulates secretory protein trafficking in other tissues (18, 26).

Fig. 7.

Model for TFG function in neurons. TFG distributed throughout neurons functions distinctly in the soma (regulation of cargo transport from the ER to the ERGIC via COPII-coated transport carriers) compared to axons and dendrites (regulates transport of Rab4A-positive recycling endosomes).

Although previous studies have raised the possibility that lysosomal dysfunction, defined by increased lysosome size and abnormal morphology, represents a unifying pathway for the pathogenesis of genetically diverse HSPs (39, 40), we failed to demonstrate consequences of the TFG p.R106C mutation on lysosome volume or distribution in primary cortical neurons obtained from homozygous mutant animals. However, further studies will be necessary to determine whether mutations in TFG affect lysosomal activity in other ways. Given the intimate relationship between endosomes and lysosomes (84), it would not be surprising if disruptions to Rab4A-mediated endosomal trafficking caused by the TFG p.R106C mutation ultimately culminates in diminished lysosome function in long-lived neurons. Alternatively, TFG may possess additional roles in endolysosomal protein trafficking that contribute to lysosome activity. Indeed, recent studies using immortalized cell lines suggest that TFG accumulates transiently with a subset of early endosomes that harbor the ESCRT-0 subunit Hrs following exposure to Wnt3a or epidermal growth factor (EGF) to regulate downstream signal transduction pathways (85, 86). The mechanisms by which TFG contributes to Wnt and EGF signaling remain unclear, and further studies will be necessary to determine whether TFG functions analogously in neuronal cells. However, based on its proposed role in the early secretory pathway, we speculate that TFG facilitates the clustering of endosomes, similar to its role in COPII carrier clustering at the ER/ERGIC interface (16). In the case of COPII-mediated transport, TFG binds directly to the inner coat component Sec23, thereby enabling carriers to be tethered to one another (15). Future work aimed at defining TFG binding partners on Hrs-positive early endosomes and Rab4A-positive recycling endosomes will be necessary to validate this idea, which will lead to a better understanding of how TFG contributes to neuronal maintenance.

Materials and Methods

Generation and Analysis of CRISPR-Cas9–Edited Sprague–Dawley Rats.

Studies using Sprague–Dawley rats were conducted in compliance with all relevant ethical regulations for animal testing and research and were approved by the Institutional Animal Care and Use Committee of the University of Wisconsin-Madison. For CRISPR-Cas9–mediated genome editing, Sprague–Dawley embryos were microinjected with a mixture consisting of a guide RNA directed at exon 4 of TFG (5′-GTCGAGAACTGATAGAACTT-3′), a single-stranded DNA repair template encoding the p.R106C mutation, and purified Cas9. Appropriately edited animals were outbred with wild-type Sprague–Dawley rats (Envigo) as heterozygotes. For all studies, three males and three females of each genotype were minimally analyzed unless otherwise noted. Littermate controls were used, and data analysis was performed in a blinded manner.

For quantitative gait measurements, animals were evaluated using a MotoRater apparatus (TSE Systems) equipped with a flat platform (87). Multiple anatomical points were marked prior to recording animals as they traversed the walkway (at least three times per imaging session; at least two imaging sessions were conducted for each rat). DeepLabCut was used to extract representative frames from each video, and at least 350 frames were manually annotated and used to train a convolution neural network that was capable of pose estimation tracking (44). Separate neural networks were trained for each age group analyzed. After the initial training, each network was evaluated and validated based on visual assessment, and DeepLabCut-assigned pixel error. Each model was further refined until sufficient accuracy was achieved by using algorithmic identification of frames with outlier points that were corrected manually and used to retrain the model. Calculation of kinematic parameters was performed using custom MATLAB scripts, and assessment of step cycle duration and stance/swing phases was performed using SimiMotion (Simi Reality Motion Systems GmbH).

For EMG studies, animals were anesthetized using intraperitoneal injection of ketamine (100 mg/kg) and xylazine (10 mg/kg). A surface ground electrode was placed on the tail, a reference pin electrode was placed in the thigh, and the active recording electrode was inserted into the gastrocnemius muscle. Electrophysiological recordings (five per limb) were collected using a VikingQuest system (Natus), and spikes (following cessation of insertional activity with a peak-to-peak amplitude greater than 20 μV) were counted.

For immunofluorescence and histological staining of tissues, animals were anesthetized using isoflurane and transcardially perfused with phosphate-buffered saline, followed by 4% paraformaldehyde prior to paraffin embedding. Five-micrometer coronal sections were prepared using a microtome, captured onto glass slides, de-paraffinized with xylene, and rehydrated ahead of staining. For electron microscopy studies, animals were perfused with 5% glutaraldehyde, 4% paraformaldehyde, 9 mM CaCl₂, and 0.08 M sodium cacodylate (pH 7.4), followed by postfixation, embedding, and sectioning. Ultrathin sections were poststained to enhance contrast and imaged using a Tecnai T12 transmission electron microscope (Thermo Fisher Scientific).

For gene expression studies, cortical brain tissue from animals was snap frozen in TRIzol (Invitrogen) and lysed by sonication prior to RNA extraction. Complementary DNA (cDNA) synthesis was performed using the High-Capacity cDNA Reverse Transcription Kit (Applied Biosystems). Real-time PCR amplification was performed on the CFX Opus Real-Time PCR system (Bio-Rad). Primers were validated by assessing their efficiency using serial dilutions of cDNA and confirming that melt curves of amplified products yielded single peaks. Specificity of allele-specific primer pairs was determined using templates that contained a single TFG allele. Differences in amplification between samples that did or did not contain the target allele were used as correction factors when comparing transcript levels in samples from heterozygous TFG p.R106C animals.

Isolation and Growth of Primary Rat Cortical Neurons.

Embryos (embryonic day 18) were decapitated, and cortices were dissected into ice-cold Hibernate A that was supplemented with B27. Cortices were digested using 0.25% trypsin for 25 min and triturated using a pipette following inactivation by Dulbecco’s modified Eagle’s medium (DMEM) growth media containing 10% fetal bovine serum. Cells were plated onto poly-D-lysine–coated coverslips or chamber slides and grown in Neurobasal-A media supplemented with B27, penicillin, streptomycin, and GlutaMax at 37 °C and 5% CO₂. A fraction of the media (one-third) was changed every 7 d. To introduce transgenes, primary neurons were grown in culture for 2 to 9 d and either transduced with purified lentiviral particles or transfected using Lipofectamine LTX with Plus reagent.

iPSC Cultures and Differentiation.

Human iPSCs (IMR90-4 obtained from WiCell) were cultured on Matrigel-coated substrates in E8 media, as described previously (88). For CRISPR-Cas9–mediated editing to generate HaloTag fusion proteins expressed from their endogenous locus, iPSCs were electroporated using a Gene Pulser Xcell System (Bio-Rad) in Opti-MEM with purified Cas9, a targeted guide RNA, and a plasmid-based repair template. Single cells were isolated using fluorescence-activated cell sorting, and clonal populations were subsequently screened by immunoblotting, Sanger sequencing, and fluorescence microscopy. Edited iPSCs were differentiated into cortical forebrain neurons as described previously. In brief, iPSCs were first treated with Y-27632 (10 nM) and grown in neuronal differentiation media [NDM; composed of DMEM/F12 and Neurobasal-A media (1:1), supplemented with N2, B27, and Glutamax], followed by dual SMAD inhibition. Neurospheres generated were plated onto Matrigel-coated plates to form rosettes, which were grown in NDM in the presence of SMAD inhibitors, and after 14 d, rosettes were transferred into NDM supplemented with fibroblast growth factor 2 (FGF-2) (10 ng/mL) and heparin (2 μg/mL) and grown in suspension for an additional 7 d. Neuronal rosettes were then dissociated and seeded onto plates or coverslips coated with poly-L-ornithine and Matrigel in neuronal maturation media [Neurobasal-A media containing B27, Glutamax, Brain-Derived Neurotrophic Factor (BDNF) (10 ng/mL), Glial cell-Derived Neurotrophic Factor (GDNF) (10 ng/mL), 1 μM Cyclic Adenosine Monophosphate (cAMP), and 0.1 μM compound E] to generate cortical neurons.

Fluorescence Microscopy Studies.

For H&E- and LFB-stained tissue sections, images were acquired using a uScope HXII slide scanner. Fluorescence imaging was conducted on either a Nikon Eclipse Ti2-E spinning disk confocal microscope equipped with a Yokogawa CSU-W1 scanhead, a 60x oil immersion objection, and an ORCA-Fusion BT sCMOS camera or an ImageXpress Micro 4 High-Content Imaging System (Molecular Devices). To measure the size of anatomic brain features, structures were traced in ImageJ for analysis. For cell counting, a region of interest was drawn, and cell number was quantified using Imaris software (Oxford Instruments).

Immunofluorescence studies of dissociated primary rat cortical neurons and iPSC-derived neurons were conducted similarly. Neuronal cells were fixed using paraformaldehyde (4%) and sucrose (4%) and blocked using 10% bovine serum albumin. Permeabilization was conducted using 0.2% TritonX-100 when necessary, followed by incubation with primary antibodies overnight at 4 °C. Following extensive washing, neurons were incubated with dye-labeled secondary antibodies for 1 h at room temperature, followed by additional washing and mounting onto glass sides using VectaShield Antifade media prior to confocal imaging.

Live-cell imaging was conducted within a Tokai Hit Stage Top Incubation System set to 37 °C and 5% CO₂. All image analysis was conducted using Imaris or ImageJ software. For HaloTag-L1CAM trafficking assays, primary cortical neurons were transfected after being grown in culture for 5 to 7 d (DIV5–DIV7). Following 14 d in culture, cells were dye-labeled with JFX650-HaloTag ligand and treated with D/D Solubilizer (DDS) prior to fixation at various time points after cargo release.

Whole-Cell Voltage-Clamp Recordings.

After dissection, cortices were treated with trypsin for 30 min at 37 °C, triturated in DMEM containing 10% fetal bovine serum, and plated onto 18-mm poly-D-lysine–coated coverslips (1/3 cortex per 12-well plate). The medium was changed to Neurobasal-A supplemented with B27 and Glutamax, and no further media changes were performed after the day of plating. Recordings were performed at room temperature on neurons grown for 14 to 19 d in a bath containing 128 mM NaCl, 2.5 mM KCl, 2 mM CaCl₂, 1 mM MgCl₂, 25 mM hydroxyethyl piperazineethanesulfonic acid (Hepes), and 30 mM glucose (∼310 mOsm/L, pH 7.4). For all recordings, the bath contained 50 µM (±)-2-amino-5-phosphonopentanoic acid (APV) and 20 µM 6-cyano-7-nitroquinoxaline-2,3-dione (CNQX); for miniature inhibitory postsynaptic current (mIPSC) recordings, 1 µM Tetrodotoxin (TTX) was also added. Pipettes were pulled from borosilicate glass and filled with a solution containing 130 mM KCl, 10 mM Hepes-NaOH, 1 mM ethylene glycol tetraacetic acid (EGTA), 2 mM Adenosine 5′-triphosphate magnesium salt (MgATP), 0.3 mM Guanosine 5′-triphosphate sodium salt hydrate (NaGTP), and 6 mM Na-phosphocreatine (∼290 mOsm/L, pH 7.3). Recordings were conducted on an inverted microscope using a HEKA EPC 10 combined amplifier and DAQ with digitization at 10 KHz after filtering at 5 kHz. The holding potential was −70 mV, and no liquid junction potential correction was applied. For mIPSC analysis, 60 s of data were analyzed per cell, and a sliding template was applied in Axograph with the following parameters: t(rise) = 1 ms, t(decay) = 25 ms, detection threshold set at 3.5-fold the SD of the baseline signal. Plotting and statistical analysis were performed using Prism (GraphPad). All data collection and analyses of raw data were performed by an experimenter blinded to the genotype of the samples.

Biochemical Interaction Studies.

Either glutathione S-transferase (GST) or GST-gephyrin was coexpressed with His-SUMO-TFG in BL21(DE3) cells and purified onto glutathione agarose resin as described previously (15). The resin was washed extensively and resuspended in an elution buffer containing sodium dodecyl sulfate (SDS). Samples were resolved via SDS–polyacrylamide gel electrophoresis (PAGE), followed by Coomassie staining or immunoblotting.

For immunoprecipitation from rodent brain extracts, antibodies directed against immunoglobulin G (IgG) or TFG were covalently linked to Protein A agarose and incubated with lysates generated by homogenizing brain tissue harvested from adult Sprague–Dawley rats. Following extensive washing, bound proteins were eluted under low pH conditions and separated by SDS-PAGE. Coomassie-stained proteins identified in TFG immunoprecipitates were excised from gels, digested, and subjected to mass spectrometry analysis. Equivalent gel regions were taken from IgG immunoprecipitates and similarly subjected to mass spectrometry analysis as controls.

Statistical Methods.

All P values were determined by paired t test or ANOVA (followed by a Tukey post hoc test) and calculated using Microsoft Excel or GraphPad Prism, and all data are shown as mean ± SEM, unless otherwise indicated. Significant differences were indicated by a P value less than 0.05.

Data, Materials, and Software Availability

All study data are included in the article and/or supporting information.