Pathogenic Variants in HEPACAM Alter Protein Localization and Interactome in Astrocytes of the Developing Mouse Cortex

ABSTRACT

Megalencephalic leukoencephalopathy with subcortical cysts (MLC) is a rare leukodystrophy characterized by early‐onset macrocephaly, white matter edema, seizures, and motor and cognitive decline. Approximately 25% of MLC patients carry HEPACAM pathogenic variants, many of which are dominant missense variants causing remitting MLC Type 2b. HEPACAM encodes hepatic and glial cell adhesion molecule (hepaCAM), also known as GlialCAM, an astrocyte‐enriched transmembrane protein with important roles in astrocyte territory establishment, gap junction coupling, branching organization, synaptic function, and development of the gliovascular unit. The molecular mechanisms through which pathogenic variants in HEPACAM alter hepaCAM protein function in vivo and facilitate MLC pathogenesis during brain development remain largely unknown. Here, we used new viral tools and proximity‐based proteomics to examine how three different dominant pathogenic variants alter hepaCAM subcellular localization and protein interactome in astrocytes of the developing mouse cortex. We found dramatic changes in hepaCAM distribution throughout the astrocyte, which were common to all mutants tested. We also observed significant changes in protein interactome between wild type and mutant hepaCAM, including decreased association with previously described hepaCAM‐interacting proteins Connexin 43 and CLC‐2. Moreover, we identified the epilepsy‐associate potassium channel KCNQ2 as a novel hepaCAM interaction partner and found reduced association between KCNQ2 and pathogenic variants. Collectively, our data provide new insights into hepaCAM protein function in astrocytes during brain development, reveal altered protein dynamics of pathogenic variants, and provide a new resource to explore the molecular underpinnings of MLC pathogenesis.

• Pathogenic variants alter hepaCAM protein distribution.

• Astrocyte‐specific hepaCAM TurboID reveals KCNQ2 as a new interaction partner.

• Pathogenic variants alter hepaCAM association with key transmembrane proteins.

1. Introduction

Megalencephalic leukoencephalopathy with subcortical cysts (MLC) is an early‐onset and slowly progressive leukodystrophy characterized by macrocephaly, white matter edema, seizures, and motor and cognitive decline (Hamilton et al. 2018; van der Knaap et al. 1995). A majority of MLC cases are caused by homozygous or compound heterozygous recessive mutations in MLC1, an astrocyte‐specific transmembrane protein of unknown function (Leegwater et al. 2001). Approximately one quarter of MLC cases are caused by mutations in HEPACAM, which encodes the hepatic and glial cell adhesion molecule (hepaCAM) (Hamilton et al. 2018; Pla‐Casillanis et al. 2022; Lopez‐Hernandez, Ridder, et al. 2011), an astrocyte‐enriched cell adhesion molecule with important roles in astrocyte territory establishment, gap junction coupling, branching organization, synaptic function, neurite outgrowth, and development of the gliovascular unit (Baldwin et al. 2021; Gilbert et al. 2019; Jin et al. 2023). At the cellular level, MLC appears to be a disorder of astrocyte ion and water regulation (van der Knaap et al. 2012). A growing body of evidence supports this notion, including clinical findings (Hamilton et al. 2018; van der Knaap et al. 2012; Ridder et al. 2011), and phenotypes of MLC mouse models (Bugiani et al. 2017; Dubey et al. 2015; Kerst et al. 2025; Gilbert et al. 2021). Understanding the molecular basis of this dysfunction is hindered by an incomplete understanding of how pathogenic variants impact MLC1 and hepaCAM protein function in astrocytes during brain development.

HepaCAM is a single‐pass transmembrane protein and member of the immunoglobulin superfamily. In the mouse cortex, hepaCAM is strongly enriched in astrocytes (Zhang et al. 2014, 2016) and is abundantly expressed throughout the membrane, including major branches, leaflets, endfeet, and cell–cell junctions (Baldwin et al. 2021; Gilbert et al. 2019). HepaCAM is also expressed in oligodendrocytes, though at lower levels than astrocytes (Zhang et al. 2014, 2016). HepaCAM is required for proper expression and localization of MLC1 both in vitro and in vivo (Bugiani et al. 2017; Capdevila‐Nortes et al. 2013). An in vitro study using HeLa cells and U251N, a tumor cell line, found that hepaCAM functions as a chaperone‐like protein for MLC1, stabilizing its expression both in the endoplasmic reticulum (ER) and at the membrane (Xu et al. 2021). In addition to its regulation of MLC1, hepaCAM is required for proper localization of a growing list of transmembrane proteins that play important roles in ion and fluid homeostasis, including the chloride channel CLC‐2 (Jeworutzki et al. 2012; Hoegg‐Beiler et al. 2014), and gap junction protein Connexin 43 (Cx43) (Baldwin et al. 2021; Wu et al. 2016). During normal brain development, hepaCAM is required for astrocyte territory establishment and proper gap junction coupling (Baldwin et al. 2021). HepaCAM also functions at synapses to regulate synaptic strength (Baldwin et al. 2021) and on astrocyte‐secreted exosomes to promote axonal outgrowth (Jin et al. 2023).

Most pathogenic variants of HEPACAM are missense mutations located in the extracellular IgV domain (Elorza‐Vidal et al. 2020; Capdevila‐Nortes et al. 2015; Lopez‐Hernandez, Sirisi, et al. 2011). These mutations may be dominant or recessive depending on the location and specific amino acid substitution. Homozygous recessive variants cause MLC Type 2a, a progressive disease similar to MLC Type 1, which is caused by mutations to MLC1 (Passchier et al. 2024). In contrast, heterozygous dominant variants cause MLC Type 2b, a milder form of the disease with a remitting phenotype and more frequent comorbidity with autism spectrum disorder (Hamilton et al. 2018; Lopez‐Hernandez, Ridder, et al. 2011; Bosch and Estevez 2020). A strong body of in vitro studies using HeLa cells or primary mouse astrocytes has shown that hepaCAM is enriched at cell–cell junctions, where it interacts with itself both in cis and in trans; furthermore, MLC‐causing hepaCAM mutations impair the ability of hepaCAM to localize to these cell–cell junctions (Lopez‐Hernandez, Ridder, et al. 2011; Capdevila‐Nortes et al. 2015; Lopez‐Hernandez, Sirisi, et al. 2011; Arnedo et al. 2014). Recessive HEPACAM variants, and some dominant variants, are thought to impair cis interaction based on their location within the IgV domain, while other dominant variants impair trans interaction (Elorza‐Vidal et al. 2020). In vivo, mice that express the dominant G89S mutation showed altered hepaCAM localization in the molecular layer of the cerebellum (Hoegg‐Beiler et al. 2014). To date, how pathogenic variants in HEPACAM impact protein localization and function in astrocytes in vivo, particularly within the context of brain development, remains unexplored.

Here, we investigated the impact of dominant pathogenic variants on hepaCAM protein dynamics in astrocytes of the developing mouse cortex. Our findings reveal altered subcellular localization of pathogenic variants in astrocytes and significant changes in association with key membrane proteins. Through proximity‐based quantitative proteomics, we describe how the hepaCAM interactome changes across mutations and provide a resource for exploring the molecular basis of MLC pathogenesis.

2. Results

2.1. Dominant Pathogenic Variants Alter hepaCAM Localization in Astrocytes Co‐Cultured With Neurons

We previously investigated the function of hepaCAM in astrocyte development using an astrocyte‐neuron co‐culture system and exogenous expression of hepaCAM (Baldwin et al. 2021). Using this system, we observed substantially altered subcellular localization of both dominant G89S and recessive R29Q variants of human hepaCAM protein. Both variants showed strong, relatively homogenous expression throughout the cells, in stark contrast to the punctate expression of endogenous hepaCAM or exogenously expressed wild type (WT) hepaCAM (Baldwin et al. 2021). Both G89S and R92Q impair hepaCAM cis interaction (Elorza‐Vidal et al. 2020). To test whether mutations that impair hepaCAM trans interaction behave similarly, we performed site‐directed mutagenesis to produce different dominant pathogenic variants in vectors that expressed human hepaCAM‐HA under control of the gfaABC1D promoter (Figure 1A). Both Q56P and D128N impair trans interaction, and D128N introduces a new N‐glycosylation site (Elorza‐Vidal et al. 2020). We expressed these variants in astrocytes co‐cultured with neurons and examined their localization in comparison to WT hepaCAM‐HA and G89S‐HA using HA immunolabeling (Figure 1A,B). We also co‐transfected a plasmid expressing cytosolic green fluorescent protein (GFP) to visualize cellular structure.

FIGURE 1.

MLC‐causing mutations alter hepaCAM localization in astrocytes co‐cultured with neurons. (A) Diagram showing wild‐type (WT) and mutant hepaCAM interactions at astrocyte‐astrocyte contacts, created in https://BioRender.com. WT hepaCAM interacts with itself in cis and trans. G89S mutation impairs cis interaction, which may impair trans interaction. Q56P and D128N mutations impair trans interaction. (B) Co‐culture workflow, created in https://BioRender.com. HA tagged hepaCAM constructs are transfected into astrocytes after 8 days in vitro (DIV 8) prior to co‐culture with neurons (DIV 10). (C) Representative images of astrocytes transfected with GFP (green) and HA‐tagged hepaCAM constructs (magenta) and co‐cultured with neurons. Scale bar 20 μm. The region within the white box for each image is enlarged in panel (D). (E, F) Quantification of (E) fraction of GFP area occupied by HA signal, (F) HA signal outside of GFP cell area, displayed as a percentage of GFP area, and (G) GFP area. N = 4 experiments (triangles), 5–8 cells/condition/experiment (dots). One‐way ANOVA with Tukey's HSD (E, G) or Kruskal‐Wallis test with Dunn's multiple comparisons test (F).

While WT hepaCAM showed a characteristic punctate distribution throughout the branches, G89S, Q56P, and D128N variants were broadly and homogenously distributed throughout the cell (Figure 1C). At the tips of branches, mutant hepaCAM was visible in lamellipodia‐like protrusions that exceeded the territory of the cytosolic GFP signal (Figure 1D). Analysis of HA signal area normalized to cell (GFP) area revealed a significant increase in total HA signal area with all three variants (Figure 1E), as well as a significant increase in HA signal outside of the cytosolic GFP signal (Figure 1F), but no significant differences in cell area (Figure 1G). These results demonstrate that in astrocytes co‐cultured with neurons, dominant pathogenic variants that impair either cis or trans homophilic hepaCAM interaction similarly alter hepaCAM protein localization.

2.2. Dominant Pathogenic Variants Alter hepaCAM Protein Distribution in Astrocytes of the Developing Mouse Cortex

To understand how pathogenic variants of HEPACAM impact hepaCAM localization and function during brain development, we transitioned our studies to an in vivo model. We chose the developing mouse visual cortex as we have previously characterized the expression and function of hepaCAM in this brain region during postnatal development (Baldwin et al. 2021). To express WT human hepaCAM and pathogenic variants in astrocytes of the developing mouse cortex, we packaged hepaCAM‐expressing plasmids into PHP.eB serotype adeno‐associated virus (AAV) under control of the gfaABC1D promoter. Due to the nature of our viral overexpression approach, we focused our studies only on dominant variants. We also fused TurboID to the C‐terminus to enable proximity labeling in the presence of biotin, followed by an HA tag to detect exogenous protein expression (Figure 2A). To confirm that the addition of TurboID to the hepaCAM C‐terminus does not alter hepaCAM protein localization, we compared expression of hepaCAM‐HA and hepaCAM‐Turbo‐HA in astrocyte‐only cultures and observed characteristic enrichment at cell–cell junctions for both constructs (Figure S1A). In astrocyte‐neuron co‐cultures, we observed characteristic punctate expression throughout the arbor for hepaCAM‐Turbo‐HA, similar to Figure 1C,D and in a previously published study (Baldwin et al. 2021) (Figure S1B).

FIGURE 2.

Dominant pathogenic variants alter hepaCAM localization in vivo. (A) Schematic of hepaCAM‐TurboID‐HA fusion proteins used for in vivo expression. (B) P1 mice were administered intracortical AAV to express hepaCAM‐Turbo‐HA fusion proteins. Brains were collected at P21 and astrocytes in the mouse visual cortex (blue box) analyzed. (C) Representative maximum‐projection images (three‐slices, 0.3 μm step size) of transduced astrocytes in layer 5 of the mouse visual cortex at P21 with HA in green and hepaCAM in magenta. Scale bar 5 μm. (D) Analysis of HA puncta density and (E) average puncta area in contiguous astrocytes transduced with WT hepaCAM or pathogenic variants. (F) Representative maximum‐projection image (three‐slices, 0.3 μm step size) showing endogenous hepaCAM (magenta) in a non‐transduced astrocyte in layer 5 of the mouse visual cortex at P21. Scale bar 5 μm. (G) Analysis of hepaCAM puncta density, (H) hepaCAM puncta area, and (I) hepaCAM puncta intensity in contiguous astrocytes transduced with WT hepaCAM or pathogenic variants and non‐transduced (NT) astrocytes. For (D), (E), (G), (H), and (I) n = 3 animals/condition (triangles), with 5–6 cells/animal (dots). One‐way ANOVA with Tukey's HSD.

We performed bilateral intracortical injection of hepaCAM‐expressing AAVs at postnatal day 1 (P1) and collected brains for immunohistochemistry at P21 (Figure 2B), a juvenile timepoint when synaptogenesis and astrocyte maturation are largely complete and roughly corresponding to 1 year of age in human brain development (Zeiss 2021; Semple et al. 2013). Immunohistochemical analysis confirmed robust expression of hepaCAM‐Turbo‐HA and pathogenic variants in astrocytes of the mouse cortex, particularly in the deeper cortical layers (Figure S2A,B). Western blot analysis of whole cortical lysates showed a trend towards increased expression of the G89S and Q56P mutants, but this did not reach statistical significance (Figure S2C,D). Interestingly, both the Q56P and D128N mutants showed an upward band shift, which could indicate changes to protein glycosylation or other post‐translational modifications.

We next performed a detailed analysis of the cellular distribution of WT human hepaCAM‐Turbo‐HA and pathogenic variants in contiguous transduced astrocytes, along with endogenous mouse hepaCAM in transduced and non‐transduced astrocytes in layer 5 (L5) of the mouse cortex. We focused on L5 because astrocyte development is well‐characterized and our viral transduction is robust in this layer (Figure S2A). In the mouse cortex, hepaCAM is broadly expressed throughout the astrocyte arbor, appearing as discrete puncta via immunolabeling in major branches, branchlets, leaflets, and endfeet (Baldwin et al. 2021). Strong expression is also found at cell–cell junctions, though it does not appear enriched compared to the rest of the non‐junctional arbor (Figure S3A). The cellular distribution of astrocytic hepaCAM in human brain tissue has not been characterized, though the amino acid sequence is highly conserved throughout (94.26% similar) and identical in the IgV domain, the site of cis and trans homophilic interaction (Figure S3B).

We performed immunolabeling for HA to detect WT human hepaCAM‐Turbo‐HA (hereafter referred to as WT) in transduced astrocytes and observed a punctate distribution pattern throughout the astrocyte arbor and at endfeet, similar to endogenous mouse hepaCAM (Figure S4A,B). We next quantified HA puncta density and HA puncta area in contiguous transduced astrocytes for all three pathogenic variants in comparison to WT. All three variants showed significantly increased HA puncta density (Figure 2D) and significantly decreased puncta area (Figure 2E), indicating an altered distribution pattern. This pattern was also observed in isolated transduced astrocytes expressing mCherry‐CAAX to visualize astrocyte membranes (Figure S4B).

To compare the distribution pattern of endogenous and exogenous hepaCAM in contiguous transduced cells with the distribution pattern of endogenous hepaCAM in non‐transduced cells, we used a hepaCAM antibody that recognizes both human and mouse hepaCAM (Figure 2F, Figure S4A). We found no difference in the density of hepaCAM puncta between transduced and non‐transduced cells (Figure 2G). Puncta area, however, was substantially increased in WT transduced cells compared to non‐transduced cells (Figure 2H), and average puncta intensity was significantly increased in transduced cells compared to non‐transduced cells (Figure 2I). While increased puncta size and intensity are likely artifacts of protein overexpression, they could also reflect altered regulatory mechanisms of human hepaCAM that impact hepaCAM protein organization or expression. In either case, we interpret our results in the context of unphysiologically high hepaCAM protein levels for mouse astrocytes.

All three pathogenic variants displayed significantly increased hepaCAM puncta density and significantly decreased hepaCAM puncta area compared to WT (Figure 2G,H). Across all conditions, roughly 60% of hepaCAM puncta were HA positive, with the remaining hepaCAM puncta HA negative, representing endogenous mouse hepaCAM that is not associated with human hepaCAM. Notably, the distribution pattern of endogenous mouse hepaCAM in transduced cells appeared altered in some of the variant conditions, in particular Q56P, where we observed small looping structures that stained positively for hepaCAM but not HA (Figure S4C). Collectively, these results reveal altered protein distribution in all three dominant pathogenic variants in vivo.

2.3. Decreased Co‐Localization of Pathogenic Variants With Connexin 43

HepaCAM expression is required for the proper localization of a growing list of transmembrane proteins in vivo (Alonso‐Gardon et al. 2021; Capdevila‐Nortes et al. 2015), including the astrocyte gap junction protein Connexin 43 (Cx43) (Baldwin et al. 2021). Frequent colocalization of hepaCAM and Cx43 is observed in visual cortex astrocytes via confocal and super resolution microscopy, and deletion of hepaCAM from astrocytes alters subcellular Cx43 localization (Baldwin et al. 2021). In vitro, R92Q and R92W pathogenic variants of hepaCAM show impaired Cx43 co‐localization at cellular junctions (Wu et al. 2016). To determine whether G89S, Q56P, or D128N pathogenic variants impact hepaCAM co‐localization with Cx43 in vivo, we co‐labeled tissue sections with HA, hepaCAM, and Cx43 and assayed the degree of co‐localization in transduced astrocytes in L5 of the mouse visual cortex (Figure 3A).

FIGURE 3.

Dominant pathogenic variants show reduced co‐localization with Connexin 43 accompanied by altered Cx43 distribution in vivo. (A) Representative maximum‐projection images (three‐slices, 0.3 μm step size) of contiguous transduced astrocytes in layer 5 of the mouse cortex at P21, labeled with Cx43 (magenta), HA (yellow), and hepaCAM (Hep, cyan). Scale bar 5 μm. (B) Percentage of HA puncta co‐localized with Cx43 puncta. (C) Density of colocalized Cx43 and HA puncta. (D) Percentage of Cx43 puncta colocalized with HA puncta. (E) Representative maximum‐projection images (3‐slices, 0.3 μm step size) of endogenous hepaCAM (magenta) and Cx43 (green) in non‐transduced astrocytes of the P21 mouse cortex. Scale bar 5 μm. (F) Percentage of hepaCAM puncta colocalized with Cx43 puncta, (G) Percentage of Cx43 puncta colocalized with hepaCAM puncta, and (H) density of colocalized Cx43 and hepaCAM puncta in contiguous astrocytes transduced with WT hepaCAM or pathogenic variants and NT astrocytes. (I) Density of Cx43 puncta, (J) average area of Cx43 puncta, and (K) relative frequency distribution of Cx43 puncta in contiguous astrocytes transduced with WT hepaCAM or pathogenic variants and non‐transduced (NT) astrocytes. n = 3 animals/condition (triangles), with 5–6 cells/animal (dots). (C, D, F, G, H, and J) One‐way ANOVA with Tukey's HSD. (I) Kruskal–Wallis test with Dunn's multiple comparison's test.

For all three pathogenic variants, we found a significant reduction in the percentage of HA puncta that co‐localized with Cx43, compared to WT (Figure 3B). The density of co‐localized HA and Cx43 puncta was also significantly reduced for all three variants (Figure 3C), indicating that the reduced percentage of co‐localized HA puncta was not an artifact of the increased HA puncta density in variant‐expressing cells. Accordingly, the percentage of Cx43 puncta co‐localized with HA was significantly reduced with the Q56P and D128N variants and nearly reached significance with the G89S variant (p = 0.0523) (Figure 3D). Consistent with previous findings (Baldwin et al. 2021), in non‐transduced cells (Figure 3E), we observed that approximately 15% of hepaCAM puncta co‐localized with Cx43 (Figure 3F) while 20% of Cx43 puncta co‐localized with hepaCAM (Figure 3G). WT transduced cells demonstrated a significantly higher percentage and density of co‐localized hepaCAM and Cx43 puncta compared to non‐transduced cells (Figure 3F–H), along with reduced Cx43 puncta density (Figure 3I) and increased Cx43 puncta area (Figure 3J,K). This suggests an enhanced capacity for co‐localization of Cx43 with WT overexpressed human hepaCAM protein and serves as a reference point for determining how pathogenic variants impact co‐localization of hepaCAM and Cx43.

Cx43 puncta density was not significantly different between WT and pathogenic variant‐expressing astrocytes (Figure 3I), though Cx43 puncta area was significantly decreased in all variants compared to WT (Figure 3J). The percentage of hepaCAM puncta colocalized with Cx43 was also reduced in all three variants (Figure 3F), though only G89S showed a significant reduction in the percentage of Cx43 colocalized with hepaCAM (Figure 3G). Both G89S and Q56P, but not D128N, showed significantly reduced density of co‐localized hepaCAM and Cx43 puncta (Figure 3H). Collectively, these results demonstrate that reduced co‐localization with Cx43 is a common feature of all three dominant pathogenic variants examined in vivo.

2.4. Unbiased Proximity‐Based Proteomic Investigation of hepaCAM and Pathogenic Variant Interactome

To better understand how dominant pathogenic variants impact hepaCAM protein dynamics in vivo, we transitioned to an unbiased approach using quantitative proximity‐based proteomics with astrocyte‐targeted TurboID. In the presence of ATP and biotin, TurboID enables protein labeling within a small (~10 nm) radius through biotinylation (Kim et al. 2014; Cho et al. 2020). Subsequent precipitation with streptavidin‐coated beads enables quantitative proteomic analysis (Figure S5A,B). To perform proximity labeling with WT and variant hepaCAM probes in astrocytes during brain development, we injected P1 mouse pups intracortically with AAVs to express the previously described WT, G89S, Q56P, or D128N TurboID fusion proteins (Figure 2A), or a previously characterized cytosolic TurboID (Takano et al. 2020) control under the control of the human minimal GFAP promoter (Figure 4A). At P18, mice received daily subcutaneous injection of biotin (24 mg/kg) for 3 days with brains collected for analysis at P21 (Figure 4B). This protocol induced robust biotinylation in astrocytes throughout the cortex, as visualized by fluorophore‐conjugated streptavidin (Figure 4C). At the cellular level, Turbo‐HA was expressed broadly throughout the cytosol, including the cell soma and branches, with similarly localized streptavidin labeling (Figure 4D). Consistent with our observations in Figure 2, WT‐HA was expressed in a punctate pattern throughout the cell arbor, with little expression in the soma (Figure 4D). The streptavidin labeling, which occurred over 3 days, covered a larger domain than the HA signal, which detects only the current hepaCAM location at time of tissue fixation. This suggests that hepaCAM is a dynamic protein with a broad presence throughout the membrane. All three hepaCAM variants demonstrated similarly effective biotin labeling (Figure 4D).

FIGURE 4.

Tool validation for hepaCAM TurboID. (A) Diagram of the five different conditions used in TurboID proteomic experiments created in https://BioRender.com. A cytosolic TurboID‐HA driven by the gfaABC1D promoter is used as a control. WT and pathogenic variant constructs are identical to Figure 2A. (B) Timeline of AAV injection and subcutaneous (s.c.) biotin injection. (C) Tile scan confocal images of the mouse cortex from Turbo‐HA and WT‐Turbo‐HA transduced brains following biotin administration and labelling with fluorophore‐conjugated streptavidin (magenta). Scale bar 200 μm. (D) Representative images of astrocytes in the mouse visual cortex following AAV injection and biotin administration. HA in green and streptavidin in magenta. Scale bar 20 μm.

To identify and quantify biotinylated proteins, we dissected the cortex (n = 4 mice/condition) and performed streptavidin pulldowns to isolate biotinylated proteins for liquid chromatography tandem mass spectrometry (LC–MS/MS) (Figure S5C,D). After removing known contaminants and filtering out proteins with only one unique peptide hit, we identified a total of 3720 unique proteins across conditions (Supporting Information File 1). We used log2‐transformed Label‐free quantity (LFQ) intensities for each individual sample to calculate the log2 fold change (log₂FC) for WT and pathogenic variant conditions compared to cytosolic TurboID (Figure S6A–D). In the WT condition, we identified 321 proteins that were significantly enriched over TurboID (p < 0.05, log₂FC > 2) (Figure S6A). We compared our results to a study that used hepaCAM co‐immunoprecipitation from adult whole brain lysates to characterize the hepaCAM interactome (Alonso‐Gardon et al. 2021). Of the 21 interaction partners described in that study, we identified 19 of these proteins in our dataset. Of these, 14 were significantly enriched over TurboID, including MLC1, CLC‐2 (Clcn2), Cx43 (Gja1), GPRC5B, and GPR37L1 (Table 1). One of the proteins that we did not detect, GPR37, is expressed only by oligodendrocytes (Zhang et al. 2014, 2016), providing confirmation of the specificity of our approach for targeting astrocytes.

TABLE 1.

Comparison of hepaCAM TurboID results with previously published hepaCAM interactome.

Previously identified	BioID detected	Statistically significant	Biologically significant
hepaCAM	Yes	Yes	Yes
Mlc1	Yes	Yes	Yes
Clcn2	Yes	Yes	Yes
Gja1	Yes	Yes	Yes
Slc1a3 (Glast)	Yes	Yes	No
Slc1a2 (Glt‐1)	Yes	Yes	Yes
Atp1a2	Yes	Yes	No
Slc4a4	Yes	Yes	Yes
Ttyh1	Yes	No	—
Atp1b2	Yes	Yes	Yes
Slc2a1	Yes	Yes	No
Slc8a1	Yes	Yes	No
Gpr37	No a	—	—
Gpr37l1	Yes	No	—
Gprc5b	Yes	Yes	Yes
Tspan9	No	—	—
Gpm6a	Yes	No	—
Gpm6b	Yes	Yes	No
Stx1b	Yes	No	—
Stx1a	Yes	No	—
Snap25	Yes	Yes	No

^a Gpr37 is an oligodendrocyte‐specific protein that is not expressed in astrocytes.

We next performed gene ontology (GO) analysis of the 321 significantly enriched proteins to gain further insight into hepaCAM function (Supporting Information File 2). Consistent with the observed cellular distribution of hepaCAM, top Cellular Component GO terms included cell junction (FDR 1.16e‑29), plasma membrane (FDR 1.83e‑20), and cell projection (FDR 3.18e‑19). Biological Process Go terms reflected the role of hepaCAM in cell adhesion (FDR 3.08e‑12) and provided support for its role in protein localization (FDR 1.52 e‑13). Top KEGG Pathway terms included adherens junction (FDR 4.8 e‑6) and SNARE interaction in vesicular transport (FDR 1.25e‑05), while top Reactome Pathway terms included signaling by Rho GTPases (1.21e‑12) and cell–cell communication (3.67e‑10). We also compared our results with an astrocyte‐specific sub‐proteome described in a previous study that used Cx43‐BirA2 as bait (Soto et al. 2023). Out of the 193 proteins in the Cx43 interactome, we found 17 proteins in common with our hepaCAM interactome (Supporting Information File 3). Several methodological differences impact the ability to make a direct comparison between these two data sets, including brain region (cortex vs. striatum), collection age (P21 vs. P63), labeling enzyme (TurboID vs. BirA2), and labeling duration (3 days vs. 7 days). Nevertheless, this relatively modest degree of overlap demonstrates the ability of this approach to identify discrete protein networks for two astrocyte membrane proteins that are known to interact with one another.

2.5. Dominant Pathogenic Variants Alter hepaCAM Protein Interactome

Next, to determine how MLC‐causing mutations impact hepaCAM protein association, we compared target protein abundance between WT and pathogenic variants (Figure 5A–C). For each comparison, we considered proteins with a log₂FC > 0.585 in the mutant condition compared to WT (corresponding to a 50% increase over WT) and p < 0.05 to be significantly increased in abundance. We considered proteins with a log₂FC < −0.585 and p < 0.05 to be significantly decreased in abundance. For all comparisons, we excluded proteins that were not significantly enriched above TurboID in at least one condition.

FIGURE 5.

Dominant pathogenic variants alter hepaCAM protein interactome in vivo. (A–C) Volcano plots comparing differentially enriched proteins between WT and (A) G89S, (B) Q56P, and (C) D128N TurboID probes. Red dots indicate proteins significantly enriched in the mutant conditions (p < 0.05, log₂FC > 0.585). Blue dots indicate proteins significantly reduced in the mutant conditions (p < 0.05, Log₂FC < −0.585). Any proteins that were not significantly enriched in either the WT or mutant conditions compared to the cytosolic TurboID control were excluded (gray dots that appear in red and blue sections of the graph). (D) Venn diagram showing number of overlapping and unique proteins that were increased or decreased for each WT‐variant comparison. (E) Heat map comparing log₂FC of increased and decreased protein targets common to all pathogenic variants in comparison to WT. To the right of each row, a plus sign (+) indicates new mutant interactors that don't interact with WT, while a minus sign (−) indicates WT interactors that do not interact with variants. Proteins without a + or −annotation are found in the WT interactome and are increased or decreased in their association with pathogenic variants by the fold change indicated in the heatmap.

Of the 3720 proteins in our dataset, only a small fraction of these were significantly altered between WT and mutant conditions (Supporting Information File 1). Q56P had the highest number of increased proteins (43, 1.2%), followed by G89S (34, 0.9%), and D128N (25, 0.7%) (Figure 5A–C). There were 14 proteins with significantly increased abundance in all three mutant conditions, including known hepaCAM interacting proteins GPRC5B, TTYH1, and GPR37L1 (Pla‐Casillanis et al. 2022; Alonso‐Gardon et al. 2021) (Figure 5D,E). Slightly more proteins showed decreased abundance in mutant conditions, with D128N having the highest (46, 1.2%), followed by G89S (40, 1.1%), and Q56P (33, 0.9%) (Figure 5A–C). There were 22 proteins with significantly decreased abundance common to all mutant conditions, including known hepaCAM‐interacting proteins CLC‐2 and Cx43 (Figure 5D,E). We did not observe any significant change in MLC1 abundance across WT and mutant protein interactomes (Supporting Information Figure S7).

The voltage‐gated potassium channel KCNQ2 was one of the top decreased proteins in all three pathogenic variant conditions (Figure 5E, Supporting Information File 1). Mutations in KCNQ2 cause developmental epileptic encephalopathy (DEE) (Garcia Castellanos et al. 2021; Goto et al. 2019), which is of interest given the prevalence of seizures in MLC patients. While KCNQ2 function has been primarily studied in neurons, a recent study described a role for glial KCNQ channels in controlling neuronal output (Graziano et al. 2024) and several published databases have detected KCNQ2 mRNA and protein expression in astrocytes (Zhang et al. 2016; Soto et al. 2023; Wei et al. 2025; Boisvert et al. 2018; Endo et al. 2022; Clarke et al. 2018). Thus far, no interaction between hepaCAM and KCNQ2 or any other potassium channel has been described. In addition to KCNQ2, we also identified KCNJ3 and KCNJ16 in the hepaCAM interactome, yet the abundance of these proteins was not altered in pathogenic variant conditions (Table 2). To validate our proteomic findings and determine whether KCNQ2 closely associates with endogenous hepaCAM in vivo, we performed three‐color stimulated emission depletion (STED) microscopy in visual cortex tissue sections from Aldh1L1eGFP mice at P21 and observed co‐localization of hepaCAM and KCNQ2 within GFP+ astrocytes at super resolution (Figure 6A). In some instances, we observed two hepaCAM puncta flanking a larger KCNQ2 puncta, a similar arrangement to the previously characterized organization of hepaCAM and Cx43 in super resolution (Baldwin et al. 2021). To further validate KCNQ2 as a potential hepaCAM interaction partner, we transfected HEK 293T cells with plasmids to express hepaCAM‐HA or HA‐tagged pathogenic variants and KCNQ2 and performed an HA co‐immunoprecipitation (co‐IP). Exogenous KCNQ2 protein was present in the co‐IP fraction from cells expressing both WT hepaCAM‐HA and KCNQ2 but absent in the co‐IP fraction from cells that expressed only WT hepaCAM‐HA (Figure 6B, blue arrow). With increased image brightness (Figure S8A), the exogenous KCNQ2 band was weakly detectable in the G89S co‐IP fraction but remained absent in Q56P and D128N co‐IP fractions (blue arrow). The KCNQ2 antibody also detected several bands in the 293T lysate at the molecular weight of known KCNQ2 isoforms (~95 and ~44 kDa, purple arrows) and cleavage products (~65 kDa, yellow arrow) (Kimura et al. 2026; Pan et al. 2001). We observed that some of these bands were present in the WT hepaCAM‐HA co‐IP fractions and appeared reduced in the pathogenic variant co‐IP fractions. KCNQ2 was recently identified as the potassium channel responsible for background voltage‐gated potassium currents in HEK 293 cells (O'Neill et al. 2026). To provide evidence that this antibody recognizes bona fide KCNQ2 bands in 293T cell lysates, we incubated KCNQ2 antibody with a KCNQ2‐specific blocking peptide prior to membrane incubation, which eliminated all antibody binding to the membrane (Figure S8B). This finding suggests that, in addition to exogenously expressed KCNQ2, WT hepaCAM‐HA co‐immunoprecipitates endogenous KCNQ2 in HEK 293 T cells. In sum, this data provides strong evidence that hepaCAM and KCNQ2 interact with one another and supports our proteomic findings of a substantially weakened association between KCNQ2 and pathogenic variants.

TABLE 2.

Comparison of potassium channel abundance in hepaCAM interactomes.

Gene names		Hepcam_Turbo	G89S_Turbo	Q56P_Turbo	D128N_Turbo	G89S_Hepcam	Q56P_Hepcam	D128N_Hepcam
Kcnq2	Log2 FC	2.54786301	−0.1063066	−0.0056505	0.75831318	−2.6541696	−2.5535135	−1.7895498
	p value	0.0146392	0.91832188	0.99596588	0.370882	0.01476483	0.02739005	0.01081943
Kcnj3	Log2 FC	2.0417943	2.36481142	2.24662924	2.12946367	0.32301712	0.20483494	0.08766937
	p value	0.02553189	0.00249497	0.00398522	0.00895675	0.67619421	0.79400701	0.91569252
Kcnj16	Log2 FC	2.15604591	1.97424126	0.88323498	0.98325968	−0.1818047	−1.2728109	−1.1727862
	p value	0.03071392	0.00712461	0.10535846	0.1133784	0.8069443	0.11518363	0.16315535

Note: Bolded values indicate statistically and biologically significant changes between WT and pathogenic variant conditions.

FIGURE 6.

KCNQ2 is a novel hepaCAM interaction partner. (A) Representative three‐color STED images of astrocytes in L5 visual cortex of Aldh1L1eGFP mice. GFP‐labeled astrocyte (gray), endogenous hepaCAM (green), KCNQ2 (magenta). (B) Immunoblot of input and HA co‐immunoprecipitation (IP:HA) fractions from HEK 293 T cells transfected with hepaCAM‐HA only (WT‐HA), WT‐HA and KCNQ2, G89S‐HA and KCNQ2, Q56P‐HA and KCNQ2, or D128N and KCNQ2. Upper blot: LI‐COR 700 channel pseudocolored in gray scale showing HA signal. Lower blots: LI‐COR 800 channel pseudocolored in gray scale showing KCNQ2 signal. The blue arrow points to the exogenous KCNQ2 band. The purple arrows points to bands at the molecular weight of known KCNQ2 isoforms. The yellow arrow points to a band at the molecular weight of a KCNQ2 cleavage product. (C) Representative maximum projection images (three slices, 0.3 μm step size) of contiguous transduced astrocytes in layer 5 of the mouse cortex at P21. KNQ2 in magenta, HA in yellow, and hepaCAM (Hep) in cyan. Scale bar 5 μm. (D) Representative maximum‐projection images (three‐slices, 0.34 μm step size) of endogenous KCNQ2 (magenta) hepaCAM (cyan) and GFP (green) in non‐transduced astrocytes of the P21 mouse cortex from Aldh1L1eGFP mice. Scale bar 5 μm. (E) Percentage of HA puncta co‐localized with KCNQ2 puncta. (F) Percentage of hepaCAM puncta co‐localized with KCNQ2 puncta in contiguous astrocytes transduced with WT or pathogenic variants and non‐transduced (NT) astrocytes. (G) Density of co‐localized KCNQ2 and HA puncta. (H) Percentage of KCNQ2 puncta co‐localized with HA puncta. n = 3 animals/condition (triangles), with 5–6 cells/animal (dots). (E, F, and H). One‐way ANOVA with Tukey's HSD. (G) Kruskal–Wallis test with Dunn's multiple comparison's test.

To further explore the nature of reduced KCNQ2 levels in pathogenic variant interactomes, we assessed colocalization of KCNQ2 signal with both HA and hepaCAM puncta in transduced and non‐transduced astrocytes (Figure 6C,D). Consistent with proteomic findings, all pathogenic variants showed significant reductions in the percentage of HA and hepaCAM colocalized with KCNQ2 (Figure 6E,F). However, we did not observe a concomitant decrease in the density of co‐localized KCNQ2 (Figures 6G and S8C) or in the percentage of KCNQ2 colocalized with HA (Figure 6H) or hepaCAM (Figure S8D). Similar to Cx43, the percentage of both HA and hepaCAM puncta co‐localized with KCNQ2 in transduced astrocytes was higher than the percent of endogenous hepaCAM co‐localized with KCNQ2 in non‐transduced astrocytes (Figures 6F and S8D), suggesting that overexpressed human hepaCAM has an increased capacity for co‐localization with KCNQ2. KCNQ2 puncta density was also significantly decreased in WT‐transduced compared to non‐transduced astrocytes (Figure S8E). We observed no change in KCNQ2 density between WT, G89S, and Q56P conditions, while D128N showed a significant increase in KCNQ2 puncta density (Figure S8E). KCNQ2 puncta area was unchanged across all conditions (Figure S8F). Collectively, these results show decreased co‐localization between pathogenic variants and KCNQ2 with minimal changes to the overall cellular distribution of KCNQ2.

2.6. Analysis of Increased and Decreased Protein Networks in hepaCAM Mutants

To gain additional insight into our proteomic findings, we performed network analysis of significantly increased (Figure 7) and significantly decreased (Figure 8) proteins in pathogenic variant protein interactomes. For analysis of increased proteins, we included all proteins that were significantly increased in at least one variant condition and visualized these as a single network in Cytoscape with hepaCAM‐Turbo‐HA as a central node. We performed STRING analysis to identify known and predicted interactions between target nodes (Figure 7A,B). This network contained significantly more interactions than expected by chance (protein–protein interaction [PPI] p = 1e–16), indicating that many of the proteins in the group are biologically related. The top two biological processes identified by GO analysis were cell adhesion and cell migration, (Figure 7C) which we visualized within the network by coloring the nodes associated with each term (Figure 7D). These results suggest that MLC‐causing mutations increase association between hepaCAM and other proteins involved in cell adhesion and migration, though the impact that this has on the function of associated proteins remains to be determined.

FIGURE 7.

Network analysis reveals increased association between pathogenic variants and proteins involved in cell adhesion and cell migration. (A) Perfuse force directed layout of proteins significantly increased in pathogenic variant conditions compared to WT. The network is depicted as an interaction network with hepaCAM‐Turbo‐HA as the central node. Blue lines represent STRING analysis to detect predicted protein–protein interactions. (B) Visualization of significantly increased proteins for each pathogenic variant and common to all three variants. (C) Gene Ontology analysis of Biological Process enrichment based on –log(FDR). (D) Location of proteins associated with top 2 GO terms within the network.

FIGURE 8.

Network analysis reveals decreased association between pathogenic variants and junctional proteins. (A) Perfuse force directed layout of proteins significantly decreased in pathogenic variant conditions compared to WT. The network is depicted as an interaction network with hepaCAM‐Turbo‐HA as the central node. Blue lines represent STRING analysis to detect predicted protein–protein interactions. (B) Visualization of significantly decreased proteins for each pathogenic variant and common to all three variants. (C) Gene Ontology analysis of Cellular Component enrichment based on –log(FDR). (D) Location of proteins within the network associated with the Cell Junction GO term.

For analysis of decreased proteins, we followed the same workflow as above, including all proteins that were significantly decreased in at least one variant condition and performing STRING analysis (Figure 8A,B). This network also contained significantly more interactions than expected by chance (PPI p = 0.00427). The top three cellular component GO terms were anchoring junction, cell junction, and tight junction (Figure 8C,D), which could indicate impaired association between hepaCAM and other cell junction proteins via reduced chaperone function and/or the inability of mutant hepaCAM to localize to cell junctions. Collectively, this dataset reveals significant changes in association of dominant pathogenic variants with numerous transmembrane proteins of physiological relevance.

3. Discussion

Here we used new viral tools, super resolution microscopy, and proximity‐based quantitative proteomics to examine the impact of three different dominant pathogenic variants on astrocytic hepaCAM protein dynamics in the developing mouse cortex. Across all variants, we found altered subcellular hepaCAM distribution and significant changes in protein interactomes. We also identified KCNQ2 as a new interaction partner for hepaCAM and showed reduced association between KCNQ2 and pathogenic variants. These findings advance understanding of hepaCAM protein function during astrocyte development and demonstrate the impact of pathogenic variants on hepaCAM protein function in vivo.

Overexpression studies of hepaCAM in primary rodent astrocytes and other cell lines have investigated the function of hepaCAM at cell–cell junctions and established a role for hepaCAM in localizing key transmembrane proteins to junctions, a function that is disrupted by pathogenic variants (Capdevila‐Nortes et al. 2013, 2015; Elorza‐Vidal et al. 2020; Lopez‐Hernandez, Sirisi, et al. 2011). Astrocytes cultured in serum and in the absence of neurons display a fibroblast‐like morphology, with few branches. Upon co‐culture with cortical neurons, astrocytes undergo a period of morphogenesis to establish a highly branched arbor (Stogsdill et al. 2017). Previously (Baldwin et al. 2021) and in this study, we observed that overexpressed human hepaCAM‐HA in astrocytes co‐cultured with neurons was organized into discrete puncta distributed throughout astrocyte branches, and this distribution was substantially altered in pathogenic variants. In the mouse cortex, astrocytes form elaborately branched, spherical arbors with thousands of fine processes. We previously demonstrated that endogenous mouse hepaCAM is broadly distributed throughout cortical astrocyte arbors, with punctate distribution in major branches, fine processes, endfeet, perisynaptic processes, and at cell–cell junctions (Baldwin et al. 2021). In this study, we examined the distribution of overexpressed human hepaCAM‐Turbo‐HA and dominant pathogenic variants in astrocytes of the mouse cortex at P21. Overexpressed WT human hepaCAM protein organized into larger and brighter puncta than endogenous mouse hepaCAM, though the overall puncta density remained unchanged. While these changes in puncta size and intensity are likely artifacts of protein overexpression, the addition of a C‐terminal TurboID could also impact protein distribution in vivo, though we did not observe any impact to protein distribution in vitro. We also cannot rule out differences between mouse and human hepaCAM as a cause of altered protein distribution, as the in vivo expression pattern of human hepaCAM protein has not yet been described. The extracellular domains of human and mouse hepaCAM are highly conserved, with identical IgV domains. In the intracellular region, there are 11 non‐biologically similar substitutions in human hepaCAM. Whether any of these substitutions impact protein organization, turnover, or expression requires further investigation.

To account for caveats associated with protein overexpression, we compared the distribution patterns of overexpressed dominant pathogenic variants with overexpressed WT human hepaCAM. We performed immunolabeling with an HA antibody to detect exogenous hepaCAM protein and an antibody that detects both human and mouse hepaCAM to detect both exogenous (co‐localized HA and hepaCAM) and endogenous (hepaCAM without HA) protein distribution in transduced astrocytes. All three dominant pathogenic variants that we investigated in this study showed a substantially altered distribution pattern, with increased puncta density and decreased puncta area and intensity. Intriguingly, the endogenous mouse hepaCAM in variant‐transduced cells was visibly distinct from WT‐transduced cells, appearing more broadly distributed throughout the membrane. This was particularly evident in Q56P‐transduced cells, where we observed small, looped structures that stain positively for hepaCAM but were largely devoid of HA. How pathogenic variants impact endogenous protein distribution is unclear, and additional studies are required to determine whether this phenomenon occurs with physiological levels of hepaCAM protein expression.

To determine whether altered distribution of pathogenic variants impacts hepaCAM association with other proteins, including known and unknown interaction partners, we performed proximity‐based quantitative proteomics using TurboID. We detected well‐characterized hepaCAM interactors in our WT hepaCAM interactome, including MLC1, CLC‐2, GPRC5B, and Cx43. With pathogenic variants, we found decreased association with CLC‐2 and Cx43, supporting findings from previous studies (Jeworutzki et al. 2012; Wu et al. 2016) and demonstrating the robustness of our approach. We also found decreased association with several proteins that are exciting targets for further exploration into hepaCAM function and MLC pathogenesis. KCNQ2 is of particular interest, given the causative role of KCNQ2 mutations in DEE (Garcia Castellanos et al. 2021; Goto et al. 2019), and the prevalence of seizures and ion dysregulation in MLC. We validated KCNQ2 as a new hepaCAM interaction partner using co‐immunoprecipitation and observed close association of endogenous hepaCAM and KCNQ2 puncta at super resolution. KCNQ2 function has been primarily studied in neurons, and its function in astrocytes is unknown. Human transcriptomic studies show higher expression of KCNQ2 in fetal astrocytes than in adult astrocytes (Zhang et al. 2016), while mouse transcriptomic studies show that expression in cortical astrocytes is highest in young adulthood (Wei et al. 2025). Whether hepaCAM regulates KCNQ2 channel function remains to be explored. Chloride channel CLIC‐like protein 1 (CLCC1) was another top decreased protein in pathogenic variant conditions. CLCC1 is localized to the ER membrane and is involved in ER calcium ion homeostasis (Nagasawa et al. 2001; Jia et al. 2015). Given the known role of hepaCAM at the plasma membrane in regulating ion channel localization and ion homeostasis, investigating the function of hepaCAM and the impact of pathogenic variants in the ER membrane is an exciting area for future study. Though we did not find a decreased association with MLC1 in our dataset, this does not preclude altered MLC1 localization or impaired function. The altered distribution of hepaCAM signal in the MLC mutant conditions may substantially impact MLC1 localization and protein dynamics, thereby impacting its function.

The organization of endogenous hepaCAM into discrete puncta throughout the astrocyte membrane and a growing list of membrane‐targeted interaction partners underscores its emerging role as a molecular organizer of transmembrane proteins. The upstream mechanisms that regulate the expression and localization of hepaCAM itself remain unknown, though our proteomic results provide some insight. Because we chose to compare hepaCAM‐TurboID to a cytosolic TurboID control, rather than a membrane‐targeted TurboID, we expected to detect enrichment of proteins and GO terms related to membrane protein sorting and transport. Indeed, KEGG pathways analysis of hepaCAM versus Turbo interactomes found significant enrichment of “SNARE interactions in vesicular transport” and “Endocytosis.” Interestingly, proteins in these pathways are significantly reduced in abundance in the pathogenic variant conditions, suggesting that altered sorting or trafficking of pathogenic variants could underly the altered distribution patterns that we observed in our study. Regarding downstream signaling, the hepaCAM C‐terminus reveals little in terms of signaling domains and motifs. Previous in vitro studies have suggested association between hepaCAM and the actin cytoskeleton (Moh et al. 2009), which is of interest given the role of hepaCAM in astrocyte morphogenesis. Our findings also point to RhoGTPase signaling as a downstream pathway of interest and find decreased association of pathogenic variants with proteins related to both actin and RhoGTPase signaling.

There are limitations to our study, which should be considered when interpreting the data. For one, we used viral tools to exogenously express WT hepaCAM‐Turbo‐HA and pathogenic variants in astrocytes. This approach is advantageous for TurboID and more efficient than generating multiple new mouse lines yet introduces the caveat of overexpression. While exogenous WT hepaCAM expression recapitulated the punctate distribution and density of endogenous hepaCAM, it did result in increased puncta area and increased puncta intensity compared to non‐transduced cells. We compared all mutant conditions to overexpressed WT to control for any caveats associated with this overexpression. Still, there may be unknown consequences of overexpression, and future experiments using physiological levels of protein expression are required to fully mitigate these concerns. We also acknowledge that while TurboID experiments can approximate an “interactome,” the presence of a protein in a TurboID dataset is not proof of interaction. We validated our discovery of KCNQ2 as a hepaCAM interaction partner using co‐immunoprecipitation and used STED microscopy to demonstrate co‐localization of these proteins at 50 nm resolution, comparable to the resolution of the proximity ligation assay (Hegazy et al. 2020; Soderberg et al. 2008). Similar validation experiments are required to solidify other candidate proteins as new interaction partners.

A final limitation to consider is that we examined hepaCAM protein function and pathogenic variants solely in protoplasmic gray matter astrocytes of the mouse cortex. We chose this region due to both the effectiveness of our tools in targeting and isolating information from this region and our experience studying hepaCAM protein function in developing cortical astrocytes. Much less is known about astrocytes and astrocyte development in the white matter. This is due in part to the low volume of white matter in the rodent brain (~10%, compared to 50% in human; Krafft et al. 2012) as well as a lack of tools tailored to targeting and studying white matter astrocytes. Because MLC is a leukodystrophy with significant white matter pathology, investigating astrocyte development, hepaCAM protein function, and MLC pathology in white matter astrocytes is an exciting future avenue of study with significant clinical relevance. Adapting available tools and techniques to target white matter astrocytes is therefore necessary to yield new insights into the cellular and molecular drivers of white matter astrocyte dysfunction in MLC.

4. Materials and Methods

4.1. Animals

All mice were used in accordance with the Institutional Animal Care and Use Committee (IACUC) at UNC Chapel Hill and the UNC Department of Comparative Medicine. Mice were housed in standard conditions with 12‐h day/night cycles. Timed‐pregnant CD1 females were obtained from Charles River (RRFD:IMSR_CRL:022). Aldh1l1‐GFP transgenic mice were obtained from MMRRC (RRID:MMRRC_011015‐UCD).

Mice were used for experiments at P21, or as specified in the text and figure legends. For all experiments, mice of both sexes were included in the analysis. Criteria for inclusion, exclusion, and randomization are listed for each experiment in specific subsections of the Methods Details section.

4.2. Plasmids

The QuikChange Lightning Site‐Directed Mutagenesis Kit (Agilent Technologies 210518) was used to introduce Q56P and D128N mutations into pcDNA3.1‐hepaCAM‐HA (Baldwin et al. 2021) using the following primers:

Q56P Forward: GCTCTGCTTTCTGTGCCGTACAGCAGTACCAGCA

Q56P Reverse: TACTGCTGTACGGCACAGAAAGCAGAGCCGACTT

D128N Forward: ATCTCCATCACCAACGACACCTTCACTGGGGAGAA

D128N Reverse: CAGTGAAGGTGTCGTTGGTGATGGAGATCTCGAC

For mutagenesis, 10 ng of plasmid DNA was amplified using the mutagenesis primers, reactions were digested with DpnI and transformed into XL10‐Gold Ultracompetent Cells on LB‐Amp plates. Colonies were isolated and then grown overnight at 37°C in 14 mL round bottom tubes containing 3 mL of Luria‐Bertani (LB) media and 300 μg of ampicillin. The ZymoPure II Plasmid Maxiprep Kit (Zymo Research D4202) was used to isolate DNA, and mutations were confirmed by whole plasmid sequencing (Plasmidsaurus). pcDNA3.1‐hepacam‐HA and pcDNA3.1‐hepacam‐G89S‐HA were generated previously (Baldwin et al. 2021). WT and mutant hepaCAM‐HA constructs were subcloned into pZac2.1‐gfaABC1D‐TurboID‐HA (Takano et al. 2020) using EcoRI and Not1 restriction enzymes. Final plasmids were sequenced using whole plasmid sequencing (Plasmidsaurus). pZac2.1‐gfaABC1D‐mCherryCAAX and pZac2.1gfaABC1D‐GFPCAAX were generated previously (Stogsdill et al. 2017). pCS2_IRES2_EGFP_KCNQ2 var4 was a gift from Al George (Addgene plasmid # 173154; http://n2t.net/addgene:173154; RRID:Addgene_173,154).

4.3. Primary Cell Culture

4.3.1. Cortical Neuron

Purified rat cortical neurons were prepared as described previously (Baldwin et al. 2021). Briefly, cortices were isolated from P1 rat pups of both sexes, digested in papain (7.5 units/mL), and triturated in low and high ovomucoid solutions. Neurons were resuspended in panning buffer (DPBS (Life Technologies 14287) supplemented with BSA and insulin) and passed through a 20 μm mesh filter (Elko Filtering 03‐20/14). Filtered cells were incubated on negative panning dishes coated with Bandeiraea Simplicifolia Lectin 1, followed by goat anti‐mouse IgG + IgM (H + L) (Jackson ImmunoResearch 115‐005‐044), and goat anti‐rat IgG + IgM (H + L) (Jackson ImmunoResearch 112‐005‐044) antibodies, then incubated on positive panning dishes coated with mouse anti‐L1 (ASCS4, Developmental Studies Hybridoma Bank, Univ. Iowa) to bind cortical neurons. Adherent cells were dislodged using a P1000 pipet, pelleted (11 min at 200g), and resuspended in serum‐free neuron growth media (NGM; Neurobasal, B27 supplement, 2 mML‐Glutamine, 100 U/mL Pen/Strep, 1 mM sodium pyruvate, 4.2 μg/mL Forskolin, 50 ng/mL BDNF, and 10 ng/mL CNTF). A total of 100,000 neurons/well were plated onto 12 mm glass coverslips coated with 10 μg/mL poly‐d‐lysine (PDL, Sigma P6407) and 2 μg/mL laminin and incubated at 37°C in 5% CO₂. On day in vitro (DIV) 2, half of the media was replaced with NGM Plus (Neurobasal Plus, B27 Plus, 100 U/mL Pen/Strep, 1 mM sodium pyruvate, 4.2 μg/mL Forskolin, 50 ng/mL BDNF, and 10 ng/mL CNTF) and AraC (10 μM) was added to stop the growth of proliferating contaminating cells. On DIV 3, the media was replaced with NGM Plus. Neurons were fed on DIV 6 and DIV 9 by replacing half of the media with NGM Plus.

4.3.2. Cortical Astrocyte

Rat cortical astrocytes were prepared as described previously (Baldwin et al. 2021). P1 rat cortices from both sexes were micro‐dissected, digested in papain, triturated in low and high ovomucoid solutions, filtered, and resuspended in astrocyte growth media (AGM; DMEM [Life Technologies 11960], 10% FBS, 10 μM hydrocortisone, 100 U/mL Pen/Strep, 2 mM L‐Glutamine, 5 μg/mL Insulin, 1 mM Na Pyruvate, 5 μg/mL N‐Acetyl‐l‐cysteine). Between 15 and 20 million cells were plated on 75 mm² flasks (non‐ventilated cap) coated with poly‐d‐lysine and incubated at 37°C in 5% CO₂. On DIV 3, non‐astrocyte cells were removed by forceful shaking of closed flasks. Fibroblast elimination was performed by adding AraC to the media on DIV 5. On DIV 7, astrocytes were trypsinized (0.05% Trypsin–EDTA) and plated into six‐well (400,000 cells/well) plates. On DIV 8, cultured rat astrocytes were transfected with shRNA plasmids using Lipofectamine LTX with Plus Reagent (Thermo Scientific) per the manufacturer's protocol. Briefly, 2 μg total DNA was diluted in Opti‐MEM containing Plus Reagent, mixed with Opti‐MEM containing LTX (1:2 DNA to LTX) and incubated for 30 min at room temperature. The transfection solution was added to astrocyte cultures and incubated at 37°C for 3 h, then replaced with AGM. On DIV 10, astrocytes were trypsinized, resuspended in NGM Plus, plated (20,000 cells per well) onto DIV 10 neurons or directly onto PDL and laminin‐coated coverslips, and co‐cultured for 72 h.

4.4. HEK 293T Methods

4.4.1. Culture

HEK 293T cells used for co‐immunoprecipitation assays were obtained from the UNC Tissue Culture Facility. Cells were cultured in DMEM supplemented with 10% fetal bovine serum, 100 U/mL Pen/Strep, 2 mM L‐Glutamine, and 1 mM sodium pyruvate (HEK Growth Media). Cells were incubated at 37°C in 5% CO₂ and passaged every 2–3 days.

4.4.2. Transfection

Prior to transfection, HEK 293T cells were passaged into 10 cm plates coated with 10 μg/mL poly‐d‐lysine (PDL, Sigma P6407). The following day, or when cells reach approximately 70% confluency, cells were transfected with expression plasmids using Lipofectamine LTX with Plus Reagent per the manufacturer's protocol. Briefly, 12 μg total DNA was diluted in Opti‐MEM containing Plus Reagent then mixed with Opti‐MEM containing 15 μL LTX and incubated for 30 min. The transfection solution was added to cells and incubated at 37°C overnight. The following morning, cells were washed once with sterile PBS and incubated in HEK Growth Media.

4.4.3. Lysis and Co‐Immunoprecipitation

Forty‐eight hours following transfection, protein was extracted with a co‐IP lysis buffer (10% glycerol, 1% NP‐40, 50 mM Tris/HCl pH 7.5, 200 mM NaCl, 2 mM MgCl2, and protease inhibitors). Cultured cells were washed twice with ice cold PBS and incubated on ice with co‐IP lysis buffer for 10 min with occasional agitation. Cellular lysates were collected, vortexed briefly, and centrifuged at 4°C at high speed to pellet any non‐solubilized material. Protein concentration of lysates was measured using Pierce BCA Protein Assay (Thermo), following the manufacturers protocol. Lysates were all diluted to the same concentrations with co‐IP lysis buffer. Co‐immunoprecipitation was performed using Chromotek HA‐Trap Magnetic Particles (Proteintech) per manufacturer's protocol with minor modifications. Briefly, 20 μL of HA‐Trap Magnetic beads were equilibrated by washing three times with ice‐cold co‐IP buffer. An aliquot of pure lysate was kept for the input condition. Cellular lysates were incubated with equilibrated beads for 1 h at 4°C with over‐end rotation. Following incubation, an aliquot of supernatant was collected for the non‐bound fraction and beads were subsequently washed three times with ice‐cold co‐IP buffer for 10 min at 4°C with over‐end rotation. During the last wash, beads were transferred to a new tube, resuspended in 25 μL 2× Sample Buffer, and heated at 45°C for 45 min in preparation for Western blot analysis.

4.5. Immunocytochemistry

Astrocyte‐neuron co‐cultures and astrocyte‐only cultures were fixed and stained as described previously (Baldwin et al. 2021). DIV 13 co‐cultures were incubated with warm 4% PFA for 7 min, washed three times with 1× Phosphate Buffer Saline (PBS), and blocked in PBS containing 50% normal goat serum (NGS) and 0.4% Triton X‐100 for 30 min at room temperature. Samples were washed once more in PBS and incubated overnight at 4°C in primary antibody. Astrocyte‐neuron co‐cultures were incubated with chicken anti‐GFP (Aves GFP1020, 1:1000) and rat anti‐HA (Sigma 11867423001, 1:500) diluted in antibody blocking buffer (ABB: pH 7.4, 150 mM NaCl, 50 mM Tris, 1% BSA, 100 mM l‐lysine, 0.04% sodium azide) containing 10% NGS. Astrocyte‐only cultures were incubated with chicken anti‐GFP, rat anti‐HA, and rabbit anti‐GFAP (Agilent Z033429‐2, 1:4000). The following day, samples were washed 3× with PBS, incubated with goat anti‐Chicken 488, goat anti‐rat 594, and goat anti‐rabbit 674 (Life Technologies, 1:500) diluted in ABB with 10% NGS for 2 h at room temperature, followed by 10‐min incubation in DAPI (1:50,000) in PBS. Coverslips were washed 3× with PBS and mounted onto glass slides using a homemade glycerol mounting media (20 mM Tris pH 8.0, 90% Glycerol, 0.5% N‐propyl gallate) and sealed with nail polish. Healthy astrocytes with strong expression of GFP and HA, a single nucleus, and minimal overlap with other GFP+ astrocytes were imaged at ×40 magnification in green, red, and DAPI channels using a Nikon Ti2 widefield fluorescent microscope (astrocyte‐neuron co‐cultures) or a Leica Stellaris 8 FALCON microscope (astrocyte‐only cultures). Astrocytes containing multiple nuclei, weak GFP expression, or in areas of dense GFP labeled cells were not imaged. The individual acquiring the images was always blinded to the experimental condition. The area of GFP and HA signal was calculated using ImageJ. The threshold for the GFP channel was adjusted to capture the entire territory of the GFP+ cell. GFP signal from neighboring cells or background was excluded to produce an isolated GFP cell. The area of the GFP+ cell was measured and a mask of the GFP+ cell was created, saved as an ROI, and applied to the HA channel. To calculate HA signal within the GFP ROI, HA outside of the GFP ROI was removed, and signal inside of the GFP ROI was thresholded to capture HA signal. The total area of HA signal within the GFP mask was calculated using “Analyze Particles” and normalized to the GFP area for each cell. To calculate HA signal associated with the cell, but located outside of the GFP ROI, the GFP ROI was applied to the HA channel containing the isolated cell and the HA channel was thresholded to capture all of the HA signal associated with the cell. The HA signal within the ROI was then manually deleted and the remaining HA signal quantified using “Analyze Particles” and normalized to the GFP area for each cell. Data were analyzed in Graphpad Prism 10. Normality tests were run on all data sets (D'Agostino and Pearson and Shapiro–Wilk). Data that passed normality were analyzed using a one‐way ANOVA with Tukey's HSD. Data that failed normality were analyzed with a Kruskal–Wallis test and Dunn's multiple comparisons test.

4.6. AAV Production and Administration

The pZac2.1‐gfaACB1D plasmids containing TurboID‐HA, hepaCAM‐TurboID‐HA, G89S‐TurboID‐HA, Q56P‐TurboID‐HA, D128N‐TurboID‐HA, mCherryCAAX, or GFPCAAX were packaged into AAV2/5/PHP.eB capsids by the UNC BRAIN Initiative Viral Vector Core. Purified AAVs were exchanged into storage buffer containing 1× phosphate‐buffered saline (PBS), 5% D‐Sorbitol, and 350 mM NaCl. Virus titers (GC/mL) were determined by qPCR targeting the AAV inverted terminal repeats. All viruses were adjusted to the same titer (3.4 × 10³ GC/mL) and 1 μL of virus was injected bilaterally into the cortex of postnatal Day 1 (P1) CD1 mice using a Hamilton syringe with a custom removable needle.

4.7. Immunohistochemistry

4.7.1. Sample Preparation

For immunohistochemistry, mice were anesthetized with 0.8 mg/kg tribromoethanol (avertin) and perfused with 1× Tris Buffered Saline (TBS)/Heparin followed by ice cold 4% PFA in TBS. Brains were collected and post‐fixed overnight in 4% PFA. The following day, brains were rinsed twice with TBS, cryoprotected in 30% sucrose in TBS for 2–3 days, frozen in a medium containing 2:1 30% sucrose to O.C.T. (VWR), and stored at −80°C. Coronal sections (40 μm thick) were collected using a CryoStar NX50 Cryostat (Thermo Fisher Scientific) and stored at −25°C in 50% glycerol in TBS. Immunolabeling was performed as described previously (Baldwin et al. 2021). Briefly, slices were washed 3 × 10 min with TBST (1× TBS containing 0.2% Triton), blocked in blocking solution (TBST containing 10% goat serum), and incubated for two nights in primary antibodies diluted in blocking solution at 4°C while shaking at 100 rpm. The following primary antibodies were used: chicken anti‐GFP (Aves GFP1020, 1:1000), rat anti‐HA (Sigma 11867423001, 1:500), chicken anti‐HA (Aves ET‐HA100, 1:500), mouse IgG1 anti‐hepaCAM (R&D Systems MAB4108, Confocal 1:500, STED 1:250), rabbit anti‐Connexin 43 (Cell Signaling 3512, 1:500), guinea pig RFP (Synaptic Systems 390‐004, 1:1000), rabbit anti‐Sox9 (Millipore AB5535, 1:1000), mouse IgG2a anti‐Olig2 (Millipore MABN50, 1:400), and rabbit KCNQ2 (Life Technologies PA1929, 1:500). Following primary antibody incubation, sections were washed in 3 × 10 min in TBST, incubated in secondary antibody solution (see next paragraph for details) for 2 h at room temperature, washed 3 × 10 min in TBST, and mounted onto glass slides with homemade mounting medium (20 mM Tris pH 8.0, 90% Glycerol, 0.5% N‐propyl gallate) and no. 1.5 coverslips. Coverslips were sealed with nail polish and dried at room temperature before storage at 4°C.

For confocal microscopy, species‐specific Alexa‐fluor conjugated secondary antibodies produced in goat (Life Technologies) were diluted 1:200 in blocking solution. Isotype subgroup‐specific secondary antibodies were used for mouse monoclonal primary antibodies (e.g., goat anti‐mouse IgG1) to prevent excessive background staining. To detect biotin labeling, fluorophore‐conjugated streptavidin (Life Technologies, 1:200) was added to the secondary antibody solution. For three color‐STED, the following secondary antibodies were diluted 1:100 in blocking solution: Goat anti‐Chicken Alexa‐fluor 594 (Life Technologies), Goat anti‐Mouse IgG1 ATTO 647N (Rockland), and Goat anti‐Rabbit CF680R (Biotium).

4.7.2. Confocal Image Acquisition and Analysis

For analysis of HA, hepaCAM, and Cx43 puncta, z‐stack confocal images were acquired on a Leica Stellaris 8 FALCON using a ×100 oil‐immersion objective (1024× 1024; 9 stacks; 0.30 μm step size). For the AAV validation experiments in Figure S2, tiled z‐stack confocal images were acquired on a Leica Stellaris 8 FALCON using a ×20 air objective (512 × 512, tiled; 10 stacks; 1.0 μm step size). For HA and streptavidin co‐labeling images in Figure 4, z‐stack confocal images were acquired on a Leica SPX8 with a ×63 objective (1024 × 1024; 9 stacks; 0.34 μm step size). For co‐localization of HA and KCNQ2 puncta, z‐stack confocal images were acquired on an Olympus Fluoview 3000 with a ×60 objective (1024 × 1024; 9 stacks; 0.34 μm step size). Co‐localized HA, hepaCAM, and Cx43 puncta and HA, hepaCAM, and KCNQ2 puncta in contiguous, transduced astrocytes and in non‐transduced astrocytes were quantified using Synbot (Savage et al. 2024) with the following parameters: 2‐channel, noise reduction, custom ROI, manual thresholding, minimum pixel = 3, pixel overlap. Prior to running Synbot, the ImageJ tracing tool was used to draw an ROI containing HA‐expressing cells in each image, the area of the ROI was measured, and the ROI was saved for use as a custom ROI in Synbot. For each animal, three z‐stack images (nine slices) were acquired and each image converted into three separate maximum projection images (MPI) of three slices each for a total of nine MPIs per animal. Data were analyzed in Graphpad Prism 10. Normality tests were run on all data sets (D'Agostino and Pearson and Shapiro–Wilk). Data that passed normality were analyzed using a one‐way ANOVA with Tukey's HSD. Data that failed normality were analyzed with a Kruskal–Wallis test and Dunn's multiple comparisons test.

4.7.3. STED Image Acquisition

Three‐channel STED images were acquired on the Leica Stellaris 8 FALCON STED using a ×100 objective with ×2 zoom. Channels were acquired in frame sequential mode to minimize crosstalk. The white light laser allowed for precise selection of excitation wavelength. Software‐recommended excitation wavelengths were used for Alexa Fluor 594 and ATTO 647N. The excitation wavelength for CF680R was adjusted to 685 to minimize crosstalk with 647N (Gonzalez Pisfil et al. 2022). A 2D‐STED donut was applied, and the 775 nm STED depletion laser was set at 90%. Z‐stack images were acquired using system‐optimized settings for resolution (4104 × 4104; image dimension of 58.14 × 58.14 μm) and z‐stack (seven steps, 0.18 μm). Raw images were exported directly to Huygens Essential for deconvolution.

4.8. Brain Lysis and Streptavidin Pulldown

AAV‐injected mice were administered subcutaneous biotin (24 mg/kg in PBS) for 3 consecutive days beginning at P18. At P21, mice were anesthetized with avertin, brains were rapidly dissected, and cortices were isolated and flash frozen for storage at −80°C. Brains were collected in batches based on litter, with each litter containing at least two brains per condition. Mice were assigned randomly to each condition. For the proteomics experiment, all 20 streptavidin pulldowns (five conditions, four mice per condition) were performed on the same day in two waves of 10 brains, with each wave containing two brains per sample (one male and one female). Brain samples were homogenized in a lysis buffer without detergent (50 mM Tris/HCl pH 7.5, 150 mM NaCl, 1 mM EDTA, and 50× protease inhibitors) using a dounce homogenizer and ceramic plunger. An equal volume of 2× RIPA buffer was added (50 mM Tris/HCl pH 7.5, 150 mM NaCl, 1 mM EDTA, 0.4% SDS, 2% TritonX100, 2% deoxycholate) to lyse the homogenate and samples were sonicated (3 × 10 s pulses) to further breakdown the tissue. Lysates were centrifuged at 15,000 rpm for 30 min at 4°C, and supernatant collected. Sodium dodecyl sulfate (SDS) was added to a final concentration of 1% SDS and samples were heated at 45°C for 45 min, then centrifuged at 15,000 rpm for 30 min at 4°C. Dynabeads MyOne Streptavidin T1 (Thermo Fisher 65602, 250 μL slurry per sample) were washed three times in RIPA‐IP buffer (50 mM Tris/HCl pH 7.5, 150 mM NaCl, 1 mM EDTA, 2M Urea, 1% NP‐40, 0.25% deoxycholate) and added to the samples to incubate overnight at 4°C with rotation (15 rpm). The following day, beads were isolated using a magnetic rack and washed three times (10 min each, room temperature, rotating at 15 rpm) with RIPA‐IP buffer, followed by three washes with 50 mM ammonium bicarbonate. 10% of each sample was transferred to a new tube and eluted in 70 μL of 2× Sample Buffer (Uezu et al. 2016) supplemented with 5 mM biotin at 60°C for 15 min. The remaining 90% of the sample was resuspended in 50 mM ammonium bicarbonate and stored at −80°C for submission to the UNC Metabolomics and Proteomics Core facility.

4.9. Proteomic Analysis

4.9.1. Sample Preparation for Affinity Purification Mass Spectrometry Analysis (AP‐MS)

Immunoprecipitated samples were subjected to on‐bead trypsin digestion, as previously described (Rank et al. 2021). After the last wash buffer step, 50 μL of 50 mM ammonium bicarbonate (pH 8) containing 1 μg trypsin (Promega) was added to beads overnight at 37°C with shaking. The next day, 500 ng of trypsin was added, then incubated for an additional 3 h at 37°C with shaking. Supernatants from pelleted beads were transferred, then beads were washed twice with 50 μL LC/MS grade water. These rinses were combined with original supernatant, then acidified to 2% formic acid. Peptides were desalted with peptide desalting spin columns (Thermo) and dried via vacuum centrifugation. Peptide samples were stored at −80°C until further analysis.

4.9.2. LC/MS/MS Analysis

Each sample was analyzed by LC–MS/MS using an Easy nLC 1200 coupled to a QExactive HF (Thermo Scientific). Samples were injected onto an IonOpticks Aurora Elite TS C18 column (75 μm id × 15 cm, 1.7 μm particle size) and separated over a 120 min method. The gradient for separation consisted of a step gradient from 5 to 36 to 48% mobile phase B at a 250 nL/min flow rate, where mobile phase A was 0.1% formic acid in water and mobile phase B consisted of 0.1% formic acid in 80% ACN. The QExactive HF was operated in data‐dependent mode where the 15 most intense precursors were selected for subsequent HCD fragmentation. Resolution for the precursor scan (m/z 350–1700) was set to 60,000 with a target value of 3 × 10⁶ ions, 100 ms inject time. MS/MS scans resolution was set to 15,000 with a target value of 1 × 10 (Lopez‐Hernandez, Ridder, et al. 2011) ions, 75 ms inject time. The normalized collision energy was set to 27% for HCD, with an isolation window of 1.6 m/z. Peptide match was set to preferred, and precursors with unknown charge or a charge state of 1 and ≥ 8 were excluded.

4.9.3. Data Analysis

Raw data were processed using the MaxQuant software suite (version 1.6.15.0) for peptide/protein identification and label‐free quantitation (Tyanova et al. 2016). Data were searched against a Uniprot Reviewed Mouse database (downloaded 01/2023, containing 17,137 sequences), appended with the hepaCAM sequences, using the integrated Andromeda search engine. A maximum of two missed tryptic cleavages were allowed. The variable modifications specified were: N‐terminal acetylation and oxidation of Met. Label‐free quantitation (LFQ) was enabled. Results were filtered to 1% FDR at the unique peptide level and grouped into proteins within MaxQuant. Match between runs was enabled. Data filtering and statistical analysis were performed in Perseus software (version 1.6.14.0) (Tyanova and Cox 2018). Detection in at least two of four replicates was required. Proteins with log2 fold change (log₂FC) ≥ 2 in WT or mutant hepaCAM conditions compared to cytosolic TurboID and an uncorrected p < 0.05 were considered statistically and biologically significant. In comparisons of protein abundance between WT and pathogenic variant conditions, proteins with a log₂FC > 0.585 in the pathogenic variant condition compared to WT (corresponding to a 50% increase over WT) and p < 0.05 were considered significantly increased in abundance. Proteins with a log₂FC < −0.585 and p < 0.05 were considered significantly decreased in abundance. For all comparisons, proteins that were not significantly enriched above TurboID in at least one condition were excluded.

4.9.4. Network Analysis

Protein networks of significantly increased and decreased proteins were visualized using Cytoscape (v3.10.3). For each network, hepaCAM‐Turbo‐HA was set as the source node, and significantly increased or decreased proteins were set as the target nodes. The pathogenic variant conditions in which each target node was significantly changed was input as a target node attribute. STRING analysis was performed on networks using the STRING app within Cytoscape to STRINGify network (confidence score = 0.4), followed by Markov clustering (MCL) using the clusterMaker app (inflation = 3). Following clustering, a perfuse forced directed layout was applied to visually organize the network. Gene Ontology (GO) analysis was performed using the STRING Enrichment App in Cytoscape and repeated on the STRING website to generate graphs of top GO terms, prioritized by FDR.

4.10. Western Blot

Lysates were mixed with 4× Laemmli Sample Buffer (Bio‐Rad) containing 5% β‐ME and incubated for 45 min at 45°C for denaturation. Precipitated proteins bound to beads were eluted and denatured as described above in the specific method sub sections. For brain lysates, 20 μg of protein or 10 μL of eluate was loaded into 4%–15% gradient pre‐cast gels (Bio‐Rad). For HEK 293Ts, 10 μg protein or all 25 μL of eluted sample was loaded into a 4%–15% gradient pre‐cast gel. Gels were run at 50 V for 5 min followed by 150 V for 1 h. Proteins were transferred to PVDF membrane (Millipore) at 100 V for 1 h and blocked in Intercept Blocking Buffer (LI‐COR). Membranes were incubated overnight in primary antibodies diluted in 3% BSA in TBS‐Tween (1× TBS containing 0.1% Tween‐20). The following primary antibodies were used: rat anti‐HA (Sigma 11,867,423,001, 1:1000), rabbit anti‐tubulin (LI‐COR 92642211, 1:1000), guinea pig anti‐KCNQ2 (Alomone Labs APC‐050‐GP, 1:500). To evaluate antibody specificity for guinea pig anti‐KCNQ2, equal concentrations of primary antibody and KCNQ2 Blocking Peptide (Alomone Labs BLP‐PC050) were incubated at room temperature for 1 h with rotation before the overnight primary antibody incubation. The next day, membranes were washed 3× with TBS‐Tween, incubated in LI‐COR IRDye 680RD or CW800 conjugated secondary antibodies (1:5000 in Intercept Blocking Buffer) for 2 h at room temperature, and washed two times with TBS‐Tween and once with TBS. For streptavidin blotting of eluted TurboID samples, membranes were incubated in LI‐COR IRDye 800CW Streptavidin (1:5000 in Intercept Blocking Buffer) overnight at 4°C. The next day, the membrane was washed twice with TBS‐Tween and once with TBS. Following the TBS wash, membranes were dried and imaged on a LI‐COR Odyssey imaging system.

4.11. Statistical Analysis

All statistical analyses were performed in GraphPad Prism 10, with the exception of proteomics analysis which was performed as described in the proteomic methods subsection. For each experiment, the number of subjects and specific statistical tests are included in the figure legend and data are represented as mean ± standard error of the mean with exact p values shown. Sample sizes were determined based on previous experience for each experiment. No statistical methods were used to predetermine sample size. Specific details for inclusion, exclusion, and randomization are included in specific method subsections.

Author Contributions

R.W.L.: conceptualization, methodology, investigation, formal analysis, visualization, writing – review and editing. B.C.D.: investigation, formal analysis, visualization, writing – review and editing. A.L.S.: investigation, formal analysis, visualization. E.B.E.: investigation, visualization, writing – review and editing. M.G.C.: investigation, visualization, writing – review and editing. H.E.S.‐O.: investigation, visualization, writing – review and editing. K.L.G.F.: methodology, investigation, visualization. A.L.M.: formal analysis, writing (methods). C.A.M.: formal analysis. L.E.H.: conceptualization. K.T.B.: conceptualization, methodology, investigation, formal analysis, supervision, funding acquisition, visualization, writing – original draft.

Funding

The Baldwin Lab is supported by the NIH, DP2 NS136873 to K.T.B. and T32NS007431 to H.E.S and B.C.D. The UNC Neuroscience Microscopy Core is supported in part by funding from the NIH‐NICHD Intellectual and Developmental Disabilities Research Center Support Grant P50 HD103573. The UNC Hooker Imaging Core Facility is supported in part by P30 CA016086 Cancer Center Core Support Grant to the UNC Lineberger Comprehensive Cancer Center. The Leica Stellaris 8 Falcon STED is supported by the NIH Shared Instrumentation Grant 1S10OD030300 to S. Gupton. This research is based in part upon work conducted using the UNC Metabolomics and Proteomics Core Facility, which is supported in part by NCI Center Core Support Grant (2P30CA016086‐45) to the UNC Lineberger Comprehensive Cancer Center and the Nutrition and Obesity Research Center (P30DK056350). The BRAIN Initiative Viral Vector Core is supported in part by the NIH U24NS124025 to K. Ritola.

Ethics Statement

All experimental protocols were performed in accordance with NIH guidelines and received approval from the Animal Care and Use Committee of UNC Chapel Hill.

Conflicts of Interest

The authors declare no conflicts of interest.

Supporting information

Figure S1: hepaCAM‐HA and hepaCAM‐Turbo‐HA display similar localization patterns in cultured astrocytes. (A) Representative images of cultured rat astrocytes transfected with GFP (green) and either hepaCAM‐HA (magenta, left two columns) or hepaCAM‐Turbo‐HA (magenta, right two columns). In both conditions, strong HA signal is observed at sites of astrocyte‐astrocyte contact (yellow arrows), on thin protrusions that extend onto of neighboring astrocytes cell bodies (orange arrows), and at the tips of protrusions that do not contact other astrocytes (blue arrows). Scale bar 20 μm. (B) Representative images of transfected rat astrocytes co‐cultured with rat cortical neurons. In both conditions, punctate HA expression is visible within the astrocyte cell body and throughout the branches. Scale bar 20 μm.

Figure S2: Intracortical administration of AAVs successfully expresses hepaCAM and pathogenic variants in astrocytes of the mouse cortex. (A) Representative multi tile maximum project images (10 stacks, 1.0 μm step size) of mouse cortex at P21 following P1 administration of AAVs to express hepaCAM‐Turbo‐HA or pathogenic variants. HA (magenta) identifies transduced cells. Sox9 expression (blue) in the absence of Olig2 expression (yellow) distinguishes astrocytes. White arrows denote examples of Sox9+ Olig2‐ astrocytes expressing HA. Top row scale bar = 200 μm. All other scale bars = 20 μm. (B) Percentage of Sox9+ Olig2‐ astrocytes expressing HA for each viral condition, n = 3 animals per condition, with three images per animal. One‐way ANOVA. (C) Western blot of P21 cortical lysates, showing expression of HA and beta‐Tubulin. (D) Quantification of HA intensity, normalized to beta‐tubulin and presented as signal abundance relative to WT. n = 4 animals per condition. One‐way ANOVA with Tukey's HSD.

Figure S3: hepaCAM is expressed throughout the astrocyte arbor and at astrocyte‐astrocyte junctions. (A) Representative confocal maximum‐projection images (3‐slices, 0.3 μm step size) of hepaCAM expression (magenta) in P21 mouse cortex with mosaic expression of mCherry‐CAAX (yellow) and GFP‐CAAX (cyan). Punctate expression of hepaCAM is visible throughout the astrocyte arbor and at sites of astrocyte‐astrocyte contact. Top scale bar, 20 μm. Bottom scale bar, 5 μm. (B) Alignment of mouse (top) and human (bottom) hepaCAM amino acid sequence. A plus symbol (+) denotes biologically similar amino acid substitutions. Non‐positive substitutions are highlighted in red and domains as shown in the hepaCAM cartoon on the left.

Figure S4: Exogenous hepaCAM‐Turbo‐HA displays similar subcellular localization to endogenous hepaCAM. (A) Representative confocal maximum‐projection images (3‐slices, 0.3 μm step size) of WT transduced astrocyte in layer 5 of the mouse visual cortex at P21 with HA in green and endogenous hepaCAM in magenta. (i) Inset from A depicting localization of hepaCAM and HA within the astrocyte arbor. Arrow denotes an example of magenta signal that does not co‐localize with green HA signal. (ii) Inset from A depicting localization of hepaCAM and HA at the astrocyte endfoot. Arrow denotes an example of magenta signal that does not co‐localize with green HA signal. (iii) Inset from A showing hepaCAM expression in a neighboring non‐transduced astrocyte. (B) Representative confocal maximum‐projection images (3‐slices, 0.34 μm step size) of astrocytes in layer 5 of the mouse visual cortex at P21 transduced with mCherry‐CAAX (magenta) and WT or pathogenic variant hepaCAM (HA, green). (C) Example of “looping” pattern observed with endogenous hepaCAM labeling in Q56P‐transduced cells. Scale bar, 10 μm.

Figure S5: Quality control for proteomic experiments. (A) Schematic of TurboID workflow. (B) Western blot of precipitated biotinylated proteins, eluted from streptavidin beads and probed with streptavidin 680. (C) PCA plot of all 20 samples. (D) Distribution plots showing number of peptides detected per binned label free quantification (LFQ) intensity values for all 20 samples (5 conditions × 4 replicates) and three pools. Red bars indicate missing values.

Figure S6: WT and mutant proteomes compared to TurboID control. (A–D) X–Y plots depicting significantly enriched proteins over TurboID control for (A) WT, (B) G89S, (C) Q56P, and (D) D128N. Red dots indicate proteins significantly enriched in the hepaCAM‐TurboID condition (p < 0.05, log₂FC > 2).

Figure S7: Heat map of all differentially enriched proteins. (A) Heat map of all proteins with biologically and statistically significant change in at least one pathogenic variant condition compared to WT. (B) Heat map showing change in abundance across pathogenic variant conditions for all 321 proteins significantly enriched over Turbo.

Figure S8: KCNQ2 and hepaCAM puncta metrics, related to Figure 6. (A) Immunoblot of input and HA co‐immunoprecipitation (IP:HA) fractions from HEK 293T cells transfected with hepaCAM‐HA only (WT‐HA), WT‐HA and KCNQ2, G89S‐HA and KCNQ2, Q56P‐HA and KCNQ2, or D128N and KCNQ2. KCNQ2 signal was detected with the LI‐COR 800 channel and pseudocolored in gray scale. The blue arrow points to the exogenous KCNQ2 band. The purple arrows points to bands at the molecular weight of known KCNQ2 isoforms. The yellow arrow points to a band at the molecular weight of a KCNQ2 cleavage product. The area in within the dotted blue rectangle is shown below with increased brightness. (B) Immunoblotting of 293T lysate with KCNQ2 antibody following incubation with a KCNQ2 blocking peptide (+) or no blocking peptide (−). The membrane was cut through the middle of the ladder for separate incubation of the two conditions. The membranes were reunited for imaging. (C) Density of co‐localized KCNQ2 and hepaCAM puncta, (D) Percentage of KCNQ2 puncta colocalized with hepaCAM, (E) Density of KCNQ2 puncta, and (F) Average area of KCNQ2 puncta in contiguous astrocytes transduced with WT hepaCAM or pathogenic variants and non‐transduced (NT) astrocytes. n = 3 animals/condition (triangles), with 5–6 cells/animal (dots). (C) Kruskal‐Wallis test with Dunn's multiple comparison's test. (D–F) One‐way ANOVA with Tukey's HSD.

Figure S9: Uncropped western blots. (A) Uncropped blots of co‐immunoprecipitation experiment from Figure 6B. (B) Uncropped blots of cortical brain lysates from Figure S2C. (C) Uncropped blot of biotinylated proteins eluted from streptavidin beads from Figure S5B.

Data S1: Supporting Information.

Data S2: Supporting Information.

Data S3: Supporting Information.

Data Availability Statement

The data that support the findings of this study are available from the corresponding author upon reasonable request.