Endocytosis in the axon initial segment maintains neuronal polarity

Neurons are highly polarized cells that face the fundamental challenge of compartmentalizing a vast and diverse repertoire of proteins in order to function properly1. The axon initial segment (AIS) is a specialized domain that separates a neuron’s morphologically, biochemically and functionally distinct axon and dendrite compartments2,3. How the AIS maintains polarity between these compartments is not fully understood. Here we find that in Caenorhabditis elegans, mouse, rat and human neurons, dendritically and axonally polarized transmembrane proteins are recognized by endocytic machinery in the AIS, robustly endocytosed and targeted to late endosomes for degradation. Forcing receptor interaction with the AIS master organizer, ankyrinG, antagonizes receptor endocytosis in the AIS, causes receptor accumulation in the AIS, and leads to polarity deficits with subsequent morphological and behavioural defects. Therefore, endocytic removal of polarized receptors that diffuse into the AIS serves as a membrane-clearance mechanism that is likely to work in conjunction with the known AIS diffusion-barrier mechanism to maintain neuronal polarity on the plasma membrane. Our results reveal a conserved endocytic clearance mechanism in the AIS to maintain neuronal polarity by reinforcing axonal and dendritic compartment membrane boundaries.

. Additionally, a microtubule minus-end-binding protein, patronin-1 (an orthologue of CAMSAP), was largely absent from the UNC-44-enriched region of the PVD neuron (Fig. 1d,e), consistent with observations of the AIS in vertebrate neurons 24 . Therefore, the PVD neuron has molecular hallmarks of a vertebrate AIS after the last dendritic branch and before the axon enters the ventral nerve cord (Fig. 1f). We also found evidence of an AIS in other C. elegans neurons, such as the bipolar DA9 motor neuron that has an enrichment of UNC-44 at the proximal end of its axon ( Fig. 1a and Extended Data Fig. 1d,e).
To test the molecular conservation of the C. elegans AIS architecture, we expressed a rat isoform of neurofascin-186 (NF-186), an AIS-resident protein that is enriched at the AIS through a FIGQY ankyrinG-binding motif 9,25,26 . We found that wild-type rat NF-186 is enriched at the AIS of the PVD neuron, whereas a mutant version of NF-186 that does not bind ankyrinG 25 (NF-186(FIGQD)) is not enriched at the AIS (Fig. 1g-i).
We also found functional evidence of the AIS as both an intracellular vesicle filter and a diffusion barrier by examining vesicular trafficking and membrane protein dynamics, respectively, in the dendrite versus the AIS (Extended Data Fig. 1f-i and Supplementary Videos 1 and 2). On the basis of these results, we consider the axonal segment delimited by the last dendritic branch and the point of entry into the ventral nerve cord to be a bona fide AIS with conserved molecular components and similar functions as in vertebrate systems (Fig. 1f).
Having established that C. elegans neurons have an AIS, we next investigated its role in establishing and maintaining neuronal polarity. We initially focused on the compartmentalization of the DMA-1 transmembrane receptor because it has a key role in establishing morphological polarity by mediating the extensive dendritic branching of the PVD neuron 27 (Fig. 1j and Extended Data Fig. 1j). To study DMA-1 compartmentalization, we used single-cell endogenous labelling of The same imaging conditions and formatting were used in both panels. Scale bars, 5 µm. Red arrows indicate aberrant axonal branches. o, Mean normalized GFP fluorescence in the axon of C. elegans animals described in m,n. p, Escape behaviour in C. elegans animals in response to a harsh touch stimulus. C. elegans animals carrying mutations in degt-1 and mec-3, which encode a DEG/ENaC channel and a homeobox transcription factor that controls the differentiation of touch receptor neurons, respectively, were included as control animals with known behavioural defects. Data are shown as mean ± s.e.m. n represents the number of individual animals. e,i One-way ANOVA with Dunnett's test. o, Two-tailed unpaired t-test with Welch's correction. p, Two-tailed unpaired t-test.

Article
DMA-1 in the intact C. elegans animal. Consistent with its branching function, endogenous DMA-1 was highly polarized to the dendrite and excluded from the axon (Fig. 1k-m). However, overexpression of DMA-1 increased its fluorescence in the dendrite as expected, but also caused DMA-1 to enter the axon (Fig. 1n). DMA-1 mislocalization to the axon caused aberrant axonal branching that was dependent on the same ligands that activate DMA-1 in the dendrite, demonstrating that a loss of DMA-1 molecular polarity causes morphological polarity defects in the PVD neuron (Extended Data Fig. 1j,k). Thus, the mechanisms that control DMA-1 compartmentalization are easily overwhelmed by its overexpression. We next investigated the roles of known proteins in AIS function. A loss-of-function mutation in UNC-44, a large AIS scaffold molecule that is required for neuronal polarity 28 , caused a complete loss of DMA-1 dendritic polarity and induced the formation of aberrant axonal branches that were dependent on DMA-1 (Extended Data Fig. 1l-o). Filamentous actin and non-muscle myosin Va at the AIS are important for maintaining neuronal polarity 14,29,30 . Consistently, loss-of-function mutations in gex-3, a WAVE complex component that mediates branched filamentous actin 31 , and hum-2, an orthologue of myosin Va that mediates vesicular trafficking in the AIS 14,29,30 , both caused a partial loss of DMA-1 dendritic polarity (Extended Data Fig. 1p,q). We then examined behavioural consequences of a loss of polarity by using a harsh touch escape behavioural assay that provides a readout of PVD neuronal function 32 . We found that overexpression of DMA-1, which caused mislocalization of DMA-1 to the axon and aberrant axonal branches (Fig. 1n,o), was also associated with a behavioural deficit in response to a harsh touch stimulus, indicative of a loss of cellular function (Fig. 1p). Therefore, the strict dendritic polarization of DMA-1 is critical for proper neuronal function.

AIS endocytosis of dendritic receptors
Endogenous DMA-1 was highly polarized to the dendrite and localized to multiple puncta in the dendrite and the AIS (Figs. 1k, l and 2a). This observation motivated us to investigate compartmentalization mechanisms for DMA-1. We first investigated the properties of DMA-1 puncta using fluorescence recovery after photobleaching (FRAP). We reasoned that DMA-1 in intracellular vesicles would have different recovery profiles compared with DMA-1 on the plasma membrane, which can recover through lateral diffusion on the contiguous plasma membrane. Indeed, we found evidence for two pools of DMA-1 with distinct dynamics: a non-punctate pool and a punctate pool with a larger immobile fraction compared to the non-punctate pool (Extended Data Fig. 2a-c). Thus, the punctate pool of DMA-1 is likely to represent intracellular vesicles, which are unable to recover owing to being predominantly in isolated vesicles.
We then sought to determine the origin of dendritic proteincontaining vesicles that were present in the AIS. We considered two possibilities: DMA-1-containing vesicles could represent either biosynthetic or endolysosomal vesicles. To distinguish between these possibilities, we used a DMA-1 cell-surface reporter assay in which DMA-1 is tagged with an extracellular SunTag peptide and an intracellular RFP, and a SunTag nanobody fused to GFP is secreted from the surrounding muscle cells (Fig. 2b). This assay distinguishes DMA-1 puncta in biosynthetic compartments (red only) from DMA-1 that has been exposed to the cell surface (red and green). Red-only puncta were visible in the cell body, whereas the AIS contained a strong GFP signal that represented specific binding of the GFP SunTag nanobody to the GFP SunTag peptide, based on the requirement of the GFP SunTag peptide for a visible GFP signal (Fig. 2c,d and Extended Data Fig. 2d). Thus, the AIS contained a pool of DMA-1 that was either currently or previously on the plasma membrane before incorporation into the endolysosomal system and cannot be accounted for solely by the biosynthetic pathway.
Based on the dominant pool of puncta containing cellsurface-exposed DMA-1 in the AIS and the observation that DMA-1 puncta rarely pass through the AIS (Extended Data Fig. 1f,g), we considered whether dendritic protein-containing vesicles were generated in the AIS itself through endocytosis. We performed a series of experiments to investigate DMA-1 endocytosis within the AIS. We first examined the subcellular localization of endogenously labelled core components of clathrin-mediated endocytosis. We found that clathrin light chain (CLIC-1), a coat protein of clathrin-coated pits and adapter protein-2 (AP-2 (called APA-2 in C. elegans)), a clathrin-associated endocytic adapter protein 33 , both formed diffraction-limited puncta in the AIS (Fig. 2e,f and Extended Data Fig. 2e). CLIC-1 was also present on large structures in the cell body that overlapped with the Golgi network marker AMAN-2, and AP-2 colocalized with the synaptic vesicle marker RAB-3 in the axon, consistent with their known functions at the Golgi and presynaptic terminal, respectively 34,35 (Extended Data Fig. 2f-k). We also found CLIC-1 puncta in the AIS of the C. elegans DA9 motor neuron (Extended Data Fig. 2l). Consistent with our light microscopy results, clathrin-coated pits have been reported in electron microscopy studies of the AIS 36 , and AP-2 has been shown to be enriched in the AIS using proximity-label mass spectrometry 18 .
Several lines of evidence demonstrate that the CLIC-1 puncta are functional endocytic clathrin-coated structures. First, they were co-labelled with dynamin-1, the GTPase that pinches endocytic vesicles from the plasma membrane (Extended Data Fig. 2m). Further, the intensity of CLIC-1 puncta was increased in dynamin-1 temperature-sensitive mutants, indicating that their internalization and disappearance is dependent on dynamin-1 function (Extended Data Fig. 2n,o). More than 80% of CLIC-1 puncta in the AIS were stationary and thus were not mobile transport vesicles transiently passing through the AIS (Extended Data Fig. 2p,q and Supplementary Video 3). We also observed the gradual formation and sudden disappearance of AP-2-labelled puncta, consistent with the dynamics of clathrin-mediated endocytosis 37  We then sought to determine whether DMA-1 engages the clathrin-mediated endocytic machinery in the AIS. First, we found that DMA-1 puncta were generated rapidly in a dynamin-dependent manner in the AIS. Acute dynamin-1 inhibition resulted in a loss of DMA-1 puncta in the AIS (Fig. 2g,h). Then, we found that DMA-1 in the AIS was concentrated into AP-2-labelled structures, which have been used as a marker for clathrin-mediated endocytic events 38 (Fig. 2i). Consistent with this, loss of apa-2 function, but not caveolin-1 (cav-1) and caveolin-2 (cav-2) (major components of caveolar endocytosis), caused a decrease in the number of GFP-positive puncta in the AIS and a decrease in the overall GFP signal of the DMA-1 cell-surface reporter (Fig. 2j,k and Extended Data Fig. 2t). This suggests that the GFP signal from the DMA-1 cell-surface reporter largely reflects internalized receptors. Together, these results support that DMA-1 undergoes clathrin-mediated endocytosis in the AIS.

Conservation of AIS endocytosis
To directly visualize endocytosis of dendritic receptors in the AIS and to test whether AIS endocytosis of dendritic receptors is conserved, we examined the transferrin receptor (TfR) in cultured mouse and rat neurons. The TfR is a prototypical dendritically polarized receptor that has been used to uncover many principles of AIS function in neuronal polarity 14,16,29,39 . We found that the dendritically polarized TfR is also present in the AIS of cultured mouse neurons and overlaps with AP-2 and clathrin light chain, components of clathrin-mediated endocytic machinery (Extended Data Fig. 3a-e). We then used a pulsed-pH protocol to resolve TfR-containing endocytic events in real time 40 . In this assay, the TfR is extracellularly tagged with the pH sensitive fluorophore super-ecliptic pHluorin (SEP), and endocytic events render SEP-labelled receptors resistant to the extracellular application of a membrane-impermeant acidic buffer. Therefore, by alternating the pH of the imaging buffer between pH 5.5 and 7.4 at 0.5 Hz during imaging, TfR-SEP-containing endocytic events were resolved as acid-resistant fluorescent puncta that abruptly appeared in the pH 5.5 imaging frame (Fig. 2l). We visualized the AIS in living cultured rat neurons using two labelling methods: (1) a genetically encoded construct expressing a voltage-gated sodium channel intracellular domain fused to mScarlet (mScarlet-Na v II-III) that is known to localize to the AIS and (2) incubation with a neurofascin antibody that detects the extracellular domain of neurofascin 41 (Extended Data Fig. 3f). We observed TfR endocytic events in the somato-dendritic region, as expected 42 ; however, we also observed robust TfR endocytosis in the AIS (Fig. 2l,m and Extended Data Fig. 3g,h). Notably, despite being dendritically polarized, the TfR undergoes endocytosis at the same frequency in the AIS as in the dendrite ( Fig. 2n and Extended Data Fig. 3i), whereas we detected few TfR endocytic events in the axon distal to the AIS (Fig. 2m). Of note, AIS endocytosis is unlikely to be owing to exogenous expression of TfR-SEP because we found similar ratios of surface-bound and internalized transferrin ligand in TfR-SEP-expressing and untransfected cells (Extended Data Fig. 3j-l). Thus, transgenic TfR-SEP behaved similarly to endogenous TfR. Directly visualizing TfR endocytosis in the AIS distinguishes endocytic removal of TfR from the plasma membrane from known TfR intracellular sorting events at the AIS 14,16,29,39 .
We also found evidence of TfR endocytosis in the AIS of induced human neurons. These neuronal cultures were properly polarized (as indicated by the localization of the TfR to the dendrite and L1CAM to  GFP-tagged SunTag nanobody is secreted from adjacent muscle cells. c,d, Confocal images (c) and DMA-1 cell-surface reporter fluorescence (d). Scale bar, 10 µm. e,f, Cell-specific endogenous expression of clathrin light chain (GFP-FLPon-CLIC-1) (e) and AP-2-FLPon-GFP (f). Asterisk indicates unrelated gut autofluorescence. Scale bar, 10 µm (main images), 1 µm (expanded selection). g, DMA-1-FLPon-GFP in the AIS of wild-type or dynamin-1 temperature-sensitive (TS) C. elegans animals. Scale bar, 1 µm. h, Line scan analysis of DMA-1 from images in g. i, Endogenous DMA-1-GFP is concentrated into AP-2-labelled structures in the AIS. Scale bar, 1 µm. j, GFP signal from the DMA-1 cell-surface reporter. Scale bar, 10 µm. k, GFP fluorescence of the DMA-1 cell-surface reporter in C. elegans animals. l, Top, illustration of endocytosis and the SEP signal during the pulsed-pH protocol (grey represents quenching). Endocytic vesicle scission generates an acid-resistant fluorescent punctum in the subsequent pH 5.5 step. Bottom, AIS of a cultured rat neuron (DIV9) imaged during the pulsed-pH protocol. An endocytic event (yellow arrow) is indicated by a pH 5.5-resistant signal that corresponds to a pre-existing cluster at pH 7.4 (green arrow). Contrast is increased for the pH 5.5 frames. m, A transfected rat neuron in culture (DIV9). Crosses represent endocytic events detected during a 10-min pulsed-pH protocol in the AIS (yellow, 85 events) and other neuronal regions (blue, 797 events). Scale bars, 10 µm. n, Frequencies of events in the indicated region. Data are mean ± s.e.m. n represents the number of animals or cells. k, Two-way ANOVA with Šidák's multiple comparison test. d,n, Two-tailed unpaired t-test. Article the axon 43 ), had a single ankyrinG-labelled AIS that is known to exhibit AIS plasticity when co-cultured with glia 44 , and had clathrin-mediated endocytic machinery in the AIS (Extended Data Fig. 4a-e). Further, the TfR was present in the AIS and overlapped with AP-2-labelled endocytic structures (Extended Data Fig. 4f,g). Together, these data suggest that dendritic receptor endocytosis in the AIS is conserved from C. elegans to humans.

Function of endocytosis in the AIS
Because endocytosis inherently removes transmembrane receptors from the plasma membrane, we considered the possibility that endocytosis in the AIS functions to remove polarized receptors from the membrane and thus functions in maintaining their axon-dendrite compartmentalization. Therefore, we first examined the consequences of global endocytic inhibition by perturbing dynamin function. Endocytic inhibition using a dynamin-1 temperature-sensitive mutation caused a loss of DMA-1 dendritic polarity and defects in neuronal morphological polarity in the C. elegans PVD neuron (Fig. 3a,b and Extended Data Fig. 4h). In cultured mouse and human neurons, we found that endocytosis contributed to the dendritic compartmentalization of the TfR. Endocytic inhibition by treatment with Dyngo 4a caused an increase of TfR in the axon, thus weakening its polarized dendritic localization, although TfR was still enriched in the dendrite, indicating that a component of its dendritic targeting is independent of endocytosis ( Fig. 3c-e and Extended Data Fig. 4i-k). These results demonstrate that endocytosis contributes to the polarized distribution of dendritic receptors.
To investigate the contribution of AIS endocytosis in maintaining the compartmentalization of polarized receptors, we took a two-pronged approach to perturb endocytosis at the AIS. Both approaches provided evidence that AIS endocytosis contributes to neuronal polarity. First, by investigating the mechanism of DMA-1 endocytosis, we gained molecular access to disrupt DMA-1's AIS endocytosis through inhibition of an LRP protein-mediated endocytic module that is preferential to the AIS (Supplementary Note 1). Loss-of-function mutations in lrp-2 caused a decrease in DMA-1 dendritic polarity and a loss of neuronal function (Fig. 3f,g,j). Second, we designed a more specific and independent method to preferentially antagonize polarized receptor endocytosis at the AIS that can be applied to other receptors. Interaction with ankyrinG is known to prevent the endocytosis of AIS-resident proteins 9,26 . One such AIS-resident protein, NF-186, is endocytosed elsewhere in the neuron, but its endocytosis is inhibited specifically at the AIS by binding ankyrinG through a known motif 9,26 (FIGQY). We genetically altered the endogenous DMA-1 locus by replacing its cytoplasmic tail with the NF-186 cytoplasmic tail. This chimaeric DMA-1-NF-186 receptor was enriched at the AIS, consistent with inhibited endocytosis (Fig. 4a,b). This change in receptor localization was dependent on the ankyrinG interaction. Introduction of a validated point mutation to disrupt ankyrinG binding (FIGQD) reversed the behaviour of the chimaeric receptor: it became punctate in the AIS and was not enriched there ( Fig. 4a,b). We then used this  chimaeric receptor strategy to examine polarity defects. We found that the DMA-1-NF-186 chimaera entered the axon, caused aberrant axonal branching, and elicited behavioural deficits in a harsh touch escape behavioural assay (Fig. 4c-f). Restoring DMA-1-NF-186 endocytosis at the AIS by introducing a single point mutation in the NF-186 ankyrinG-binding motif 9,25 reversed these localization, morphological and functional defects (Fig. 4c-f). We then performed parallel experiments in induced human neurons by creating chimaeric fusion proteins with the delta/notch-like EGF repeat-containing receptor (DNER), a dendritically polarized single-pass type I transmembrane protein 45 . DNER-NF-186 also exhibited increased fluorescence at the AIS that is dependent on ankyrinG interaction (Extended Data Fig. 6x,y). We then examined polarity defects and found that DNER-NF-186 fluorescence was increased in the axon compared with DNER-NF-186 (FIGQD) (Extended Data Fig. 6x-z), consistent with a role of AIS endocytosis in receptor compartmentalization.

Article
These data suggest that when dendritically polarized receptors are not removed from the AIS plasma membrane through endocytosis, they enter the axon, which diminishes their dendritic polarity. To understand how dendritic receptors enter the axon when endocytosis is inhibited, we used time-lapse imaging of DMA-1 to monitor vesicle entry into the axon of the C. elegans PVD neuron. When endocytosis was inhibited in C. elegans animals through a loss-of-function apa-2 mutation, we did not observe robust DMA-1 vesicular trafficking in the axon (Extended Data Fig. 7a and Supplementary Videos 5 and 6), despite observing vesicle movement of axonally polarized GFP-FLPon-RAB-3 in the axon of wild-type C. elegans animals under the same imaging conditions (Extended Data Fig. 7b,c). Further, apa-2 loss-of-function mutation had no effect on the known intracellular vesicle filter or membrane diffusion-barrier functions of the AIS (Extended Data Fig. 7d-g). Together, these results suggest that DMA-1 mislocalizes to the axon when endocytosis is inhibited through lateral diffusion in the membrane and not via intracellular vesicular trafficking pathways.
We then examined whether DMA-1 endocytic clearance from the AIS is a distinct mechanism that works in conjunction with known AIS polarity mechanisms. In principle, intracellular vesicle sorting and endocytic clearance from the plasma membrane could both contribute to neuronal polarity, as they dispatch distinct populations of proteins. To test this, we focused on myosin Va (also known as HUM-2 in C. elegans), which mediates dendritic protein intracellular vesicular trafficking in the AIS 14,29,30 , and LRP-2, which promotes AIS endocytosis of DMA-1 (Supplementary Note 1). Using double mutational analysis, we found that lrp-2 and hum-2 double-mutant C. elegans animals exhibit a further loss of DMA-1 dendritic polarity compared with single mutants, suggesting that these proteins function in separate pathways (Extended Data Fig. 7h,i). These results suggest that AIS endocytosis works in conjunction with the vesicle-filter function of the AIS for the parallel sorting of plasma membrane and intracellular receptors to maintain neuronal polarity.
For endocytosis in the AIS to prevent dendritic protein entry into the axon, endocytosed proteins must be either degraded or recycled back to the dendrite. Several lines of evidence support degradation. In the AIS, DMA-1 fluorescence overlapped with internal RAB-7-positive late endosomes but not RAB-10-and RAB-11-positive recycling endosomes (Fig. 4g,h and Extended Data Fig. 8a-d). RAB-7-positive late endosomes were a distinct population from AP-2-positive endocytic structures and were preferentially enriched in the AIS compared with RAB-10-and RAB-11-positive recycling endosomes (Extended Data Fig.  8e-g). We did not observe a pool of anterograde RAB-7-positive late endosomes in the AIS (Extended Data Fig. 8h), which suggests that DMA-1 is sorted to RAB-7-positive late endosomes for degradation after its endocytosis at the AIS. Consistent with a role for RAB-7, we found that rab-7 mutation, but not rab-10 rab-11 double mutation, caused aberrant axonal branching that coincides with a loss of DMA-1 dendritic polarity (Extended Data Fig. 8i,j). Conversely, rab-10 rab-11 double mutation, but not rab-7 mutation, resulted in a significant decrease in dendritic branching (Extended Data Fig. 8k,l). These results demonstrate distinct roles of Rab-mediated pathways in PVD neuronal morphology and suggest that post-endocytic trafficking of DMA-1 differs between the AIS and the dendrite. A role for RAB-7 implicates degradation through lysosomal pathways, which often involves ubiquitin-dependent sorting of transmembrane proteins for degradation. We next tested whether DMA-1 ubiquitination is required by endogenously mutating the three lysine residues in the cytoplasmic tail of DMA-1. Indeed, lysine to arginine mutations (DMA-1(KtoR)) caused a significant accumulation of DMA-1 in larger puncta in the AIS that colocalized with RAB-7-positive vesicles (Extended Data Fig. 8m,p). We also found that the dendritically polarized TfR partially colocalized with Rab7-positive late endosomes in the AIS of cultured mouse and human neurons (Extended Data Fig. 8q-t). The requirement for Rab7 and localization to the AIS make this endocytic clearance mechanism molecularly and functionally distinct from known trafficking modalities such as transcytosis, that function in polarity 46 .

Generality of endocytosis in the AIS
Finally, we tested whether endocytic clearance of polarized receptors from the AIS is broadly used. We focused on an axonally polarized receptor and diverse dendritic receptors across multiple neuron types and species. We found that SER-1, a serotonin G protein-coupled receptor, was axonally polarized in the PVD neuron and enriched in AP-2-labelled clathrin-coated structures and RAB-7-positive vesicles in the AIS of C. elegans animals (Fig. 4i-k). These data support the notion that axonally polarized receptors are removed from the AIS membrane through endocytosis after initial intracellular vesicle trafficking to the axon distal to the AIS. Inhibition of SER-1 endocytosis through apa-2 mutation caused a loss of axonal polarity and mislocalization to the dendrite (Fig. 4l,m). We found that the dendritically polarized claudin-like transmembrane protein, HPO-30, localized to clathrin-coated structures and RAB-7-positive vesicles in the AIS, and endocytic inhibition caused a loss of its dendritic polarity in the C. elegans PVD neuron (Extended Data Fig. 9a-f). In the C. elegans DA9 neuron, the dendritically polarized ROR1/2 tyrosine kinase receptor, CAM-1, localized to clathrin-coated structures and RAB-7-positive vesicles in the AIS, and endocytic inhibition decreased its dendritic polarity (Extended Data Fig. 9g-n). In cultured mouse and induced human neurons, we found that the dendritically polarized glutamate receptor 47 , GluA1, partially localized to both AP-2-labelled endocytic structures and Rab7-positive vesicles in the AIS, and endocytic inhibition diminished its dendritic compartmentalization (Extended Data Fig. 10a-n). Therefore, we find that diverse polarized receptors in multiple neuron types and species are endocytosed in the AIS to be removed from the AIS plasma membrane.

Discussion
We propose a mechanism in which axonally and dendritically polarized transmembrane proteins enter the AIS through lateral diffusion, are captured for endocytosis, internalized from the plasma membrane, and targeted for degradation (Fig. 4n). Numerous polarized receptors partially localize to clathrin-coated structures and Rab7-positive late endosomes in the AIS across multiple neuron types and species. Although we find that endocytosis in the AIS is coupled to degradation, both DMA-1 and TfR require recycling in the dendrite 48,49 . This suggests that the post-endocytic fate of receptors differs between the AIS and dendrite. Endocytosis and subsequent degradation of polarized receptors inherently functions as a clearance mechanism, and thus may be critical for reinforcing compartment boundaries in the AIS and maintaining neuronal polarity. Through molecular manipulations of DMA-1, we find that preferentially perturbing AIS endocytosis results in a loss of its dendritic compartmentalization. Our analysis with the DNER receptor suggests this may represent a general mechanism used to maintain axon-dendrite compartmentalization.
Endocytic clearance of polarized receptors from the AIS membrane is likely to work in concert with known polarity mechanisms. We found that AIS endocytosis functions in parallel with myosin Va mediated intracellular vesicle trafficking. The AIS diffusion barrier has been shown to trap proteins in the AIS between 190-nm-spaced periodic ring-like structures that comprise the actin-spectrin-based submembranous cytoskeleton 10,21 . We anticipate that the membrane region between these structures can accommodate endocytic vesicle formation and that endocytosis removes transmembrane proteins that are trapped by the diffusion barrier. The combination of polarized vesicular trafficking, the diffusion barrier and endocytic clearance in the AIS may enable the neuron to maintain polarity over its long lifespan.
Our results elucidate an endocytic clearance function of the AIS in neuronal polarity. Through the study of one of the most polarized cell types, the present results reveal a mechanism by which neurons can achieve strict compartmentalization even along a contiguous membrane. Whether other polarized cell types or cells with specialized functional membrane domains use such a mechanism to reinforce compartment boundaries on the membrane is subject to future investigation. AIS endocytosis, which is critical for neuronal polarity and neuronal function, is therefore likely to have widespread physiological significance.

Online content
Any methods, additional references, Nature Research reporting summaries, source data, extended data, supplementary information, acknowledgements, peer review information; details of author contributions and competing interests; and statements of data and code availability are available at https://doi.org/10.1038/s41586-022-05074-5.

Constructs and cloning
All constructs are described in Supplementary Table 3. Constructs generated during this study for use in C. elegans experiments were created with an isothermal assembly method using overlapping oligonucleotides 51 . Flippase-mediated recombination was used to achieve selective labelling of endogenous fusion proteins in a limited number of cells 52 . C. elegans expression constructs were made in a pSM delta vector. ser2prom3::FLP was assembled from a 4,141-nucleotide 5′ promoter sequence of ser-2 and a 2× nuclear localization sequence (NLS) FLP recombinase sequence from pMLS262 53 . ser2prom3::NF-186 (rat) and ser2prom3::NF-186(FIGQD) (rat) were assembled from a 4,141-nucleotide 5′ promoter sequence of ser-2 and 3,516-nucleotide rat HA-NF-186 sequence 25 (Addgene plasmid #31061). The FIGQD point mutation was introduced via the reverse cloning primer to amplify the NF-186 sequence: CCTTTACTCATAGAGCCGCTTCCACTACCcccgggACTTC-CGCTGCCggccagggaatagatggcattgactggagatgtggcctctgagctctcattgc-cctcggtctcctccttgtcctttctgacagtgtCctggccaataaaggagccatcttc.

C. elegans genome editing using CRISPR-Cas9
Endogenous fluorophore insertions, basepair changes or deletions and chimeric receptors were created by gonadal microinjection of CRISPR-Cas9 protein complexes. CRISPR-Cas9 genome editing was performed using standard protocols 54 . Cas9 protein and tracRNA (IDT) were both injected at 1.525 µM. crRNAs (IDT) were injected at a concentration of 1.525 µM. DNA repair templates were generated using PCR or ordered as ultramers (IDT) and injected at 0.15-3 µM. dpy-10(cn64) was used as a co-injection marker and outcrossed once the desired genome edit was generated. F 2 C. elegans animals were screened for the desired genome edit using PCR and confirmed by sequencing. Guide RNAs used for genome editing are listed in Supplementary Table 4.

C. elegans confocal microscopy imaging and analysis
Images of fluorescently tagged fusion proteins were captured at room temperature in live C. elegans animals. Mid-L4 stage hermaphrodite C. elegans animals were anaesthetized using 10 mM levamisole (Sigma-Aldrich) in M9 buffer and mounted on 5% agarose pads for imaging. C. elegans animals were imaged on either an inverted Zeiss Axio Observer Z1 microscope equipped with a Hamamatsu EM-CCD digital camera, a Yokogawa CSU-X1 spinning-disk unit, controlled by Metamorph (version 7.8.12.0), and using either a Plan-Apochromat 100× 1.4 NA objective or a 63× 1.4 NA objective or on an inverted Zeiss Axio Observer Z1 microscope equipped with a Yokogawa CSU-W1 spinning-disk unit, a Prime 95B Scientific CMOS camera, controlled by 3i Slidebook (v6), and using either a C-Apochromat 40× 0.9 NA or 63× 1.2 NA objective. FRAP experiments were performed on the above microscope using a Vector diffraction-limited laser scanner (3i). In all cases, image settings (for example, exposure time and laser power) were identical for all genotypes across the experiment.
Quantitative image analysis was performed on unprocessed images using Fiji Software 55,56 . Maximum-intensity projections were rotated, cropped, and straightened to generate display images. Brightness and contrast were adjusted in Fiji to show relevant features, treating images being compared in the same manner. Kymographs were generated using Fiji. Line scan analysis of AIS regions was performed on cropped AIS images using the plot profile function in Fiji. Manders' overlap coefficient was calculated using the Coloc2 plugin in Fiji to quantify endogenous RAB fluorescence overlap with endogenous DMA-1 fluorescence in the AIS. Overlap was calculated by drawing a region of interest around the AIS and using Costes' thresholding. Puncta density was calculated using fluorescence thresholding and automated particle counting to determine the number of puncta per micron using the standard thresholding and particle analysis functions in Fiji.
To calculate dendritic polarity index, the dendrite or axon of PVD was traced based on the general mCherry membrane marker using the freehand selection function in Fiji. The mean pixel intensity of the fluorescently tagged protein in the dendrite (DP) and axon (AP) and the membrane-bound mCherry in the dendrite (DM) and axon (AM) were measured separately. The normalized mean intensity of the fluorescence-tagged protein in the dendrite (D) or the axon (A) is calculated by: D = DP/DM and A = AP/AM. We normalized the GFP signal of the cargo to the membrane-bound mCherry because the left and right PVD axonal regions overlap, but the contributions of each neuron to membrane-bound mCherry and cargo GFP are proportional. The polarity index (PI) for each worm was obtained by: PI = D/(A + D). PI is 0 for a completely axonal localization, 1 for a completely dendritic localization, and 0.5 for a non-polarized localization. All dendritic polarity indices shown were calculated based on the proximal anterior dendrite region though no differences were found compared to the proximal posterior dendrite region.

Harsh touch escape response assay
Harsh touch escape response assays were performed by touching day 1 adult C. elegans animals with a platinum wire pick in the midsection of the body as described previously 57 . The tip of the wire platinum pick was flattened and cut to be 20 mm thick and 30 mm wide. For each genotype, 11-12 C. elegans animals were tested 10 times with a 10 min interval between trials. All C. elegans animals used for this assay were in the mec-4(e1611) light touch mutant background to isolate the harsh touch response except for mec-3(e1338) animals that were used as a control 58 . C. elegans animals carrying mutations in degt-1 and mec-3, which encode a DEG/ENaC channel and a homeobox transcription factor that controls the differentiation of touch receptor neurons, respectively, were included as control animals with known behavioural defects 58,59 . These experiments were performed and analysed blind to genotypes.

Pulsed-pH protocol
The TfR-SEP plasmid was used 42 and described 60 previously. The mScarlet-Na v II-III plasmid was constructed as follows: the fragment corresponding to loop II-III of the Nav1.2 sodium channel was extracted from the YFP-Na v II-III plasmid (Addgene #26056) by PCR (forward: TAGTCGATCGCCAGTTCTTTCAGCTCAGACAACCTG; reverse CACTC-GAGCTAGCTTATCTTGTAGCACGTT), digested with PvuI and XhoI and ligated into a pmScarlet_ABHD6 vector (gift from B. Fakler), which leads to the expression of mScarlet-Na v II-III under the control of a CMV promoter. The construct was verified by sequencing.
Pregnant Sprague-Dawley rats ( Janvier Labs) were housed at 50-70% humidity and 18-22 °C with ad libitum feeding and a 12 h light-dark cycle. Neurons were cultured following the Banker method 61 as described before 42 . All procedures were in accordance with the European guide for the care and use of laboratory animals and approved by the ethics committee of Bordeaux University (CE50) and the French Ministry of Research.
Neurons were transfected at 6 days in vitro (DIV) with TfR-SEP alone or together with mScarlet-Na v II-III by the calcium phosphate method. Live cell recordings were performed at 9-10 DIV. Neurons were perfused with HEPES buffered saline solution (HBS) at 37 °C. HBS contained (in mM): 120 NaCl, 2 KCl, 2 MgCl 2 , 2 CaCl 2 , 5 d-glucose and 10 HEPES, and was adjusted to pH 7.4 and 260-270 mOsm. For the pulsed-pH assay, MES buffered saline solution (MBS) was prepared similarly by replacing HEPES with MES and adjusting the pH to 5.5. All salts were from Sigma-Aldrich. The field of view was centred on the AIS as visualized either with mScarlet-NavII-III or endogenous neurofascin fluorescence. Endogenous neurofascin was visualized by labelling with anti-Neurofascin antibody diluted in HBS (1:250) (Monoclonal mouse (clone A12/18), Antibodies Incorporated 75-172, RRID AB_2282826) for 10 min and then labelled with anti-mouse Alexa568 (Thermo Fisher Scientific, A10037, RRID AB_2534013) in HBS (1:500) for 3 min. The field of view also contained part of the soma and some dendrites (17 out of 21 recordings for mScarlet-NavII-III expression and 12 out of 20 recording for Neurofascin labelling), which were used for comparison. HBS and MBS were perfused locally around the recorded cell using a 2-way borosilicate glass pipette as described previously 40 . Imaging was performed with an Olympus IX71 inverted microscope equipped for total internal reflection fluorescence microscopy with a 150×, 1.45 NA objective (UAPON150XOTIRF), a laser source (Cobolt Laser 06-DPL 473 nm, 100 mW) and an Ilas2 illuminator (Gataca Systems) with a penetration depth set to 100 nm. Emitted fluorescence was filtered with a dichroic mirror (R405/488/561/635) and an emission filter (ET525/50m, Chroma Technology) and recorded by an EM-CCD camera (QuantEM 512C, Princeton Instruments) controlled by Metamorph (version 7.10.3.279). Videos were acquired for 10 min at 0.5 Hz.
Detection of endocytic events and their analyses was conducted using previously described, custom-made MATLAB scripts 40,42 . In short, a sudden punctate fluorescence increase appearing in pH 5.5 images was detected as being an endocytic event if: (1) it was visible for more than 3 frames (that is, 8 s), and (2) it appeared at the same location as a pre-existing fluorescence cluster detectable in pH 7.4 images. Candidate events from each cell were validated using a trained support vector machine as described previously. Event frequency was expressed per cell surface area measured on the cell masks drawn around the AIS, a selected dendrite, or the entire neuron.

Transferrin ligand uptake assay
For uptake of transferrin ligand, cultured rat neurons were starved for 10 min in HBS at 37 °C, and then incubated with transferrin-Alexa647 (Thermo Fischer Scientific, T23366,) at 50 µg ml −1 in HBS for 10 min at 37 °C. After a quick wash with cold HBS, cells were stripped of surface-bound transferrin-Alexa647 with glycine buffer (100 mM NaCl, 50 mM glycine, pH 3, 300 mOsm) twice, rinsed with cold HBS and fixed with 4% paraformaldehyde/4% sucrose in PBS at room temperature for 15 min. For surface labelling, neurons were incubated with transferrin-Alexa647 for 10 min at 4 °C, followed by a quick wash with cold HBS and fixed. The cells were then permeabilized with PGT and labelled with anti-TfR antibody (1:1,000) (clone H68.4, Thermo Fisher Scientific, 13-6800, RRID AB_2533029) followed by anti-mouse Alexa568 conjugate (1:500) (Thermo Fisher Scientific, A11004, RRID AB_2534072). The samples were mounted in Fluoromount-G with DAPI and were imaged on a spinning-disk confocal microscope with a 63× objective and 488, 561 and 634 nm illumination. A stack of focal planes, 0.5 µm apart, was acquired for the GFP, Alexa568 and Alexa647 channels. A maximum-intensity projection of all channels was used for quantification of fluorescence measurements. We defined a mask of the cell in the Alexa568 channel and used it for quantification of transferrin-Alexa647 labelling.

Embryonic stem cell culture
Male human embryonic stem cell line H1 was cultured under feeder-free conditions in mTeSR1 medium (Stem Cell Technologies). In brief, medium was replaced every day. When 80-90% confluent, cells were treated with Accutase (Innovative Cell Technologies), collected and centrifuged at 1,000 rpm for 3 min, re-suspended in mTeSR1 with Thiazovivin (2 µM, Bio Vision), and plated onto Matrigel (BD Biosciences)-coated plates. Mouse glial cells were isolated from the forebrain of newborn wild-type CD1 (Charles River) mice. Mice were housed at room temperature (20-22 °C) with 30-70% humidity, and a 12 h light-dark cycle. In brief, newborn mouse forebrain was digested with papain for 30 min and plated onto 10 cm dishes in DMEM (Thermo Fisher) supplemented with 10% calf serum (GE healthcare life sciences), sodium pyruvate (Thermo Fisher), MEM Non-Essential Amino Acids (Thermo Fisher), penicillin/streptomycin (Thermo Fisher), and 2-mercaptoethanol (Sigma). All mouse procedures were approved by the administrative panel on laboratory animal care (APLAC), Stanford University.
Primary mouse hippocampal cultures C57BL/6J mice ( Jackson 000664) were housed at room temperature and 40-60% humidity with a 12 h light-dark cycle with free access to food and water ad libitum. Hippocampi were dissected from P0 neonatal male and female wild-type mice, dissociated by papain digestion, filtered through a 70-µm cell strainer and plated on poly-d-lysine coated coverslips in 24-well plates. Cultures were maintained in Neurobasal A medium supplemented with l-glutamine, B27 and AraC in humidified incubator with 5% CO 2 at 37 °C. The neurons were cultured for 14 days before experiments. All mouse procedures were approved by the administrative panel on laboratory animal care (APLAC), Stanford University.

Immunostaining of cultured mouse and induced human neurons
DIV14 mouse neuron cultures were washed with PBS 3 times for 1 min. Cells were then fixed with 4% paraformaldehyde in PBS at room temperature for 15 min. Cells were washed 3 times for 5 min and then incubated in blocking and permeabilization buffer for 1 h at room temperature (5% normal donkey serum, 0.3% Triton-100, 0.05% sodium azide in PBS). Primary antibodies diluted in TBST (0.1% tween in TBS) were added and incubated overnight at 4 °C. Cells were then washed 3 times with TBST for 5 min before incubation with secondary antibodies diluted in TBST for 1 h at room temperature. Cells were washed 3 times for 10 min with TBST and then mounted with Prolong Diamond Anti-fade Mountant (Thermo) on glass slides for imaging. In experiments with Dyngo 4a treatment, 30 µM Dyngo 4a was used in serum-starved conditions with DMSO used as a vehicle control.
DIV14 or DIV26 induced human neuron cultures were washed with PBS+++ buffer (0.5 mM CaCl 2 , 1 mM MgCl 2 , 4% sucrose in PBS). Cells were then fixed with PFA+ buffer (4% paraformaldehyde and 4% sucrose in PBS) for 10 min at room temperature. Cells were then washed 3 times with PBS. Cells were permeabilized with 0.2% Triton X in PBS for 10 min and washed 3 times with PBS. Cells were then incubated in blocking buffer (4% BSA, 1% cosmic calf serum in PBS) for one hour at room temperature. Primary antibodies diluted in blocking buffer were added and incubated overnight at 4 °C. Cells were washed 4 times with PBS before adding secondary antibodies diluted 1:1,000 in blocking buffer for 1 h at room temperature. Cells were washed 3 times in PBS and then mounted with DAPI Fluoromount-G (Southern Biotech) or Prolong Diamond Anti-fade Mountant (Thermo) on glass slides for imaging. In experiments with Dyngo 4a treatment, 30 µM Dyngo 4a was used in serum-starved conditions with DMSO used as a vehicle control.
The Images were rotated, cropped, and straightened to generate display images. Images were analysed by measuring the mean fluorescence intensity in the indicated region and background subtracted using a nearby region of the coverslip without cells. Only cell regions without overlapping neurites were analysed. Neurons were selected for analysis using the following criteria: (1) having a clearly identifiable cell body with an AIS that projects away from the cell body (based on strong ankyrinG labelling) and (2) having an AIS that was easily followed to the axonal region. Axonal regions were selected as being distal to the strong ankyrinG labelling of the AIS. These axonal regions also had markedly lower MAP2 staining compared to the dendritic domains of the neuron.

Interaction screening and protein biochemistry
The interaction screening was based on the modified extracellular interactome assay strategy as reported previously 64 and adapted to 384-well plates to accommodate the larger number of proteins being screened. 380 proteins chosen from common neuronal cell-surface receptor families were cloned into Drosophila culture expression plasmids in bait format with an Fc tag and in prey format with a pentameric coiled coil domain followed by the human placental alkaline phosphatase. All proteins were expressed and secreted using Drosophila S2 cells following transient transfection with the TransIT-Insect transfection reagent (Mirus, MIR 6104) according to the manufacturer's directions without modification. Bait proteins were captured directly from conditioned medium onto Protein A-coated 384-well plates overnight at 4 °C, washed with 1× PBS including 0.1% BSA, 1 mM MgCl 2 and 1 mM CaCl 2 , followed by a 3-h incubation at room temperature by the conditioned medium for prey proteins, and a second wash. Binding was detected using the absorbance from the chromogenic substrate KPL BluePhos (Seracare, 5120-006) at 650 nm. For analysing results, we followed the trimmed Z-score strategy as previously implemented 65 , where the lowest and highest values (10% each) of measured absorbances were removed to calculate trimmed mean and standard deviations for each collection of measurements, which were then used to calculate Z-scores (that is, s.d. over mean) for all measurements.
For the surface plasmon resonance experiment, DMA-1 ectodomain (amino acids Leu20 to Leu507) was cloned in a modified baculoviral transfer plasmid based on pAcGP67A, designed to secrete proteins with C-terminal Avi tag for biotinylation and hexahistidine tag for protein purification. Similarly, LRPL-1 cDNA lacking the signal peptide was cloned into pAcGP67A with a hexahistidine tag. Both proteins were successfully secreted into medium from High Five cells upon infection with baculoviruses, demonstrating that LRPL-1 is a secreted protein.
Proteins were purified with immobilized metal affinity chromatography with Ni-NTA Agarose resin (Qiagen), followed by size-exclusion chromatography in HBS (10 mM HEPES pH 7.2, 150 mM NaCl). LRPL-1 was incubated with the E. coli biotin ligase BirA for biotinylation. SPR was performed using a Streptavidin chip on a Biacore X100 (GE Healthcare) for capturing LRPL-1 in HBS with 0.05% Tween-20 and 0.1% Bovine serum albumin to decrease non-specific binding of DMA-1 to the chip surface. The sensorgrams were analysed with the manufacturer's evaluation software (Biacore X100 Evaluation software, version 2.0.2) using kinetic and equilibrium models of binding. The kinetic fits yielded on and off rates (k on and k off values) of 4.6 × 10 3 M −1 s −1 and 0.018 s −1 , respectively, resulting in a dissociation constant (K D ) of 3.9 µM. The steady-state analysis gave a similar K D of 6.9 µM. Both analyses suffered from limited non-specific binding, which likely explains the two-fold discrepancy between the two K D values.

Statistics and reproducibility
All data are expressed as the mean ± s.e.m. Sample size refers to the number of C. elegans animals or neurons. Scatter plots show an overlay of mean and s.e.m., and each dot represents an individual C. elegans animal, neurons or neuronal region. Therefore, n is the number of dots in each scatter plot. Statistical comparisons were performed in Prism 8.0 software (GraphPad Software). Statistical significance tests are specified in the figure legends. Each C. elegans in vivo measurement (FRAP or fluorescence intensity) is from a distinct animal. All C. elegans in vivo imaging data were replicated in at least two independent imaging sessions. Each vertebrate neuron measurement is from a distinct neuron. All vertebrate neuron imaging data were from at least three independent neuronal cultures and at least three independent imaging experiment sessions. All experiments showing representative data were repeated with similar results. Independent experiments represent independent biological replicates.

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

Data availability
All reagents and raw data are available from the corresponding author upon reasonable request. Source data are provided with this paper.

Code availability
The MATLAB programs that were used to analyse the pulsed-pH data have been previously described 40,42 and were written for MATLAB2018a with the following MATLAB toolboxes: Image Processing, Wavelet, Statistics and Machine Learning; the programs are formatted as a toolbox, scission_analysis, available at MATLAB Central File Exchange as 72744-scission_analysis (https://fr.mathworks.com/matlabcentral/ fileexchange/72744-scission_analysis). Fig. 3 | Endocytosis of dendritic receptors in the AIS of cultured mouse and rat neurons. (a, b) Confocal images of mouse neurons fixed and stained at DIV14 for the indicated endogenous proteins, including MAP2 (dendrites) and Tuj1 (all neurites). AIS zoom is shown on the bottom, and brightness is increased for better visibility. (c) AP-2 labelled structure containing TfR from the AIS of the image shown in b (white arrow). Brightness is increased for better visibility. (d) Confocal images of mouse neurons fixed and stained at DIV14 for the indicated endogenous proteins. AIS zoom is shown on the bottom, and brightness is increased for better visibility. (e) Clathrin-coated structure containing TfR in the AIS of the image shown in d (white arrow). Brightness is increased for better visibility. (f) Neurons transfected with TfR-SEP and mScarlet-Na v II-III, fixed at DIV10, and stained with anti-neurofascin (NF) antibody (clone A12/18) and anti-mScarlet antibody. Intense staining of both mScarlet-Na v II-III and NF defines the AIS. 18/20 transfected neurons (from 3 cultures) showed colocalization with endogenous NF. (g) Scission/endocytic example events from Fig. 2m, as defined as the appearance of a TfR-SEP cluster (i.e. an endocytic vesicle) at pH 5.5 (time 0) and corresponding to a preexisting cluster at pH 7.4 (time −2). Note that these vesicles remain visible at pH 5.5 for 3 frames or more. Contrast is increased for pH 5.5 frames for better visibility.

Extended Data
(h) Same analysis as Fig. 2m for neurons transfected with TfR-SEP and incubated with anti-NF antibody (clone A12/18). (i) Frequencies of endocytic events in the AIS (20 recordings from 4 different cultures) and selected dendrites (12 recordings). Event frequency is similar between both AIS visualization methods: for neurons expressing mScarlet-Na v II-III, the frequency is 0.080 ± 0.015 events/µm 2 /min (n = 21) and for neurons stained with anti-NF antibody, the frequency is 0.073 ± 0.012 events/µm 2 /min (n = 20); P = 0.72 using an unpaired t test.  Fig. 5i. Western blot was to confirm protein expression and thus not repeated. Gel source data in Supplementary Fig. 1. (k) Surface plasmon resonance experiment for an LRPL-1-immobilized Biacore Streptavidin chip against DMA-1 ectodomain as analyte. Left: Blank-subtracted SPR sensorgrams for DMA-1 at 78 nM to 10 µM with kinetic model fits shown in black. Right: Binding isotherm for the SPR data fit to the Langmuir 1:1 binding model. The estimated K D , 6.9 ± 0.7 µM (±: standard error of the fit), is depicted with a red dashed line. All data are shown as mean ± s.e.m. N-values are indicated on bar graphs and represent the number of animals. P values in b were calculated using a two-way ANOVA with Šidák multiple comparison test. P values in e were calculated using a two-tailed unpaired t test. Scale bars, 2 µm (a, d).