Assessing Surface Binding Events on Golgi Microsomes Isolated from Cancer Cells by Capillary Electrophoresis Authors: Clara Davoine, Gaël Cobraiville, Lionel Pochet, Marianne Fillet Published: 2026 ABSTRACT Intracellular membrane proteins represent a vast and pharmacologically essential class of therapeutic targets, yet their characterization within native pathological environments remains experimentally challenging. Here, we report what is, to the best of our knowledge, the first direct affinity measurements performed at the surface of Golgi microsomes isolated from cancer cells. By integrating affinity capillary electrophoresis (ACE) with a refined mathematical framework, we overcome long-standing obstacles associated with membrane protein analysis, including the need for target modification, intrinsic enzymatic activity, labeled probe, or large protein excess. The microfluidic format further minimizes sample consumption while enabling both direct and indirect detection modalities, thereby expanding applicability to impure compounds and ligands lacking charge or UV absorbance. Applying this strategy to UGCG, a Golgi-resident membrane protein of oncological relevance, we demonstrate quantitative ligand-binding measurements on fairly monodisperse vesicular fractions (∼150 nm) that preserve native intracellular architecture. Collectively, this work establishes a robust, broadly applicable platform for target engagement studies in complex biological systems and highlights the potential of ACE to accelerate drug discovery efforts involving intracellular or extracellular membrane-bound targets. Introduction Evaluation of drug-target interactions is a key component in the development of new drugs. The main issue when developing in vitro assay is to ensure the integrity of the target. This is particularly true for membrane proteins which play crucial roles in numerous cellular processes, such as signal detection and transduction, transport, enzymatic reactions, and cell–cell communication. Dysfunctional membrane proteins lead to various pathological conditions and serve as targets for many approved drugs. − Both mechanistic studies and drug discovery campaigns focusing on a membrane protein require reliable structural and functional information. Although many in vitro characterization assays exist, they usually require detergents that remove the native environment. However, membrane proteins rely heavily on the membrane environment. , Their interactions with lipids and neighboring membrane proteins determine both their structure and functional behavior, underscoring the need for assays that preserve the native local environment. Recent studies have highlighted affinity capillary electrophoresis (ACE) in partial-filling mode as a powerful tool for validating target engagement and determining dissociation constants (K D). − ACE detects binding events by monitoring changes in the electrophoretic mobility of an analyte upon interaction with its target. As a microfluidic biophysical method, it requires only minimal amounts of unmodified proteins, allows quantification over a wide range of affinities under near-physiological conditions and on nonenzymatic targets, and can analyze impure samples thanks to its inherent separation capability. , Although ACE has proven effective for measuring affinity with purified soluble proteins, its application to membrane proteins has not yet been explored due to their dependence on a native lipid environment. The Golgi apparatus serves as the central trafficking nexus of the cell, orchestrating the post-translational modification, sorting, and targeted delivery of secretory and membrane proteins as well as lipids. In mammalian cells, this indispensable membrane-bound organelle is organized as a polarized stack of five to seven flattened cisternae. The cis face of the Golgi receives newly synthesized cargo from the endoplasmic reticulum, whereas the trans face and the trans-Golgi network mediate its sorting and export to distinct intracellular compartments or the extracellular space. Post-translational processing of proteins and lipids is mediated by a diverse set of Golgi-resident enzymes, including glycosidases, glycosyltransferases, and luminal proteases. Accumulating evidence indicates that disruptions in Golgi architecture and function are implicated in a broad spectrum of diseases. In cancer, aberrant glycosylation patterns are closely associated with disease progression and poor prognosis. , Accordingly, quantitative determination of the binding affinities of candidate ligands to Golgi-resident proteins is essential for elucidating mechanisms of cellular regulation and for the discovery of therapeutic agents. A range of analytical approaches has been developed to probe such interactions, including time-resolved fluorescence resonance energy transfer (TR-FRET)-based assays, radioligand binding assay, fluorescence polarization assay, surface plasmon resonance (SPR), or affinity selection mass spectrometry (AS-MS). Each of these methods, however, presents limitations. Displacement assays may fail to detect interactions occurring outside the labeled probe. SPR necessitates immobilization of one binding partner, which can alter binding behavior. AS-MS may show reduced sensitivity for ligands that dissociate on rapid time scales. In this context, ACE may reveal itself as a versatile and complementary strategy for Golgi-targeted drug discovery. In this study, UDP-glucose ceramide glucosyltransferase (UGCG, also known as glucosylceramide synthase), the rate-limiting Golgi enzyme for glucosylceramide synthesis, was chosen to benchmark our ACE approach. ,, Large-scale production of this intracellular integral membrane protein remains challenging, , and its structure and mechanistic details are still incompletely characterized. The D1, D2, D3, (Q/R)­XXRW motif defines its active site (Figure ). UGCG function is modulated by its environment, forming heterodimers or higher-order oligomers with proteins such as sphingomyelin synthase 1, c-Fos, and Reticulon-1C. ,− Oligomerization may influence its localization and activity. ,, UGCG is frequently upregulated in multiple cancers, where its inhibition reduces malignancy, mitigates drug resistance, and may enhance immunotherapy by reversing tumor-associated immunosuppression. ,− These characteristics establish UGCG as both a biomarker and a promising therapeutic target. , Traditional evaluation relies on enzymatic assays with fluorescent or deuterated substrates, and AS-MS has been applied for high-affinity binder identification. , However, no method currently enables quantification of fast-equilibrium interactions. UGCG, a therapeutic target in oncology. Predicted structure from TmAlphaFold database (membrane plane determined by TMDET algorithm). In pink, D92, D144, D236, RMIRW272-276 reported active sites. Here, we report the first demonstration of ACE applied to Golgi membrane fractions isolated from cancer cells, preserving endogenous protein and lipid microenvironments. This approach extends the scope of ACE to low-abundance, membrane-embedded targets and simultaneously broadens drug discovery strategies toward physiologically relevant membrane proteins. Experimental Section Materials Cell culture consumables and the Micro BCA Protein Assay Kit were obtained from Thermo Fisher Scientific (Waltham, MA). OptiPrep Density Gradient Medium, bovine serum albumin (heat shock fraction, protease-free, fatty acid–free, essentially globulin–free, pH 7, ≥98%), 3-[(3-Cholamidopropyl)­dimethylammonio]-1-propanesulfonate hydrate (CHAPS), UDP-α-d-glucose disodium salt, 4-methylmorpholine (NMM), and trypsin inhibitor from Glycine max (soybean, 1 mg inhibits 1.8 mg of trypsin with activity of approximatively 10,000 BAEE units per mg of trypsin) (SBTI) were purchased from Merck (Darmstadt, Germany). Eliglustat hemitartrate and miglustat were supplied by BLD Pharmatech GmbH (Reinbek, Germany), and ceramide C8 was obtained from Avanti Polar Lipids (Alabaster, AL). Sequencing grade modified trypsin (10,868 BAEE units per mg) was acquired from Promega Corporation (Madison, WI). Poly­(vinyl alcohol) (PVA) capillaries were provided by Agilent Technologies (Santa Clara, CA). The wash medium (WM) consisted of 10 mM triethanolamine and 250 mM sucrose, adjusted to pH 7.4 with acetic acid. The homogenization medium (HM) was prepared by supplementing WM with Halt Protease Inhibitor Cocktail and 25 U/mL Pierce Universal Nuclease (Thermo Fisher Scientific, Waltham, MA). A 50% iodixanol working solution was generated by mixing five volumes of OptiPrep with one volume of diluent (60 mM triethanolamine, 1.5 M sucrose, pH 7.4) and subsequently diluted with WM to yield a 15% iodixanol solution. The background electrolytes (BGE) used in capillary electrophoresis consisted of 30 mM sodium phosphate and 250 mM sucrose, pH 7.4. To prepare 100 mL of BGE, 321.1 mg of Na2HPO4 (anhydrous), 101.8 mg of NaH2PO4 (monohydrate), and 8558 mg of sucrose were added to approximately 80 mL of Milli-Q water. After dissolution, the pH was adjusted to 7.4 with 1 M NaOH. The solution was then quantitatively transferred into a 100 mL volumetric flask and the volume was adjusted with Milli-Q water. Finally, the BGE was filtered through Chromafil Xtra 0.2 μm PVDF syringe filters (Macherey-Nagel, Duren, Germany). Conductivity was measured using a SevenCompact conductimeter (Mettler-Toledo, Columbus, USA). Physicochemical properties of small molecules and the BGE were calculated using MarvinSketch 19.27 (ChemAxon Ltd., Budapest, Hungary) and PeakMaster version 6.0f8, respectively. Cell Culture Caki-1 cells (ATCC HTB-46) were obtained from the American Type Culture Collection (Manassas, VA) and maintained at 37 °C in a humidified atmosphere containing 5% CO2. Cells were cultured in McCoy’s 5A modified medium (Gibco, Grand Island, NY) supplemented with 10% fetal bovine serum, 100 U/mL penicillin, and 100 μg/mL streptomycin. Prior to seeding, culture flasks were coated with type I collagen (rat tail, Sigma). The stock collagen solution was diluted 30-fold in PBS, applied to the flask surface, and incubated for 20 min at 37 °C. The coating solution was then aspirated and replaced directly with culture medium without washing. Golgi Microsomes Golgi isolation was performed following established protocols (Figure S1). − Nineteen T175 flasks, corresponding to ∼2 × 108 Caki-1 cells, were washed three times with 3 mL PBS and twice with 5 mL wash medium (WM) at room temperature. Cells were scraped in 5 mL ice-cold WM, pooled into two 50 mL Falcon tubes, and pelleted by centrifugation at 4 °C (300 g for 20 min followed by 1000 g for 15 min). The supernatant was removed, and cell pellets were combined in 2 mL homogenization medium (HM). Cells were disrupted by 20 passes through a 27-gauge needle. The homogenate was centrifuged at 2000 g for 10 min at 4 °C; the pellet was resuspended in 500 μL HM, centrifuged again, and the resulting supernatants were pooled. A continuous gradient was prepared in two Ultra-Clear centrifuge tubes (SW40 rotor, Beckman Coulter, Brea, CA) by mixing equal volumes of WM and 15% iodixanol using a Gradient Master (Biocomp Instrument, Tatamagouche, Canada; 135 s, 81.5°, 15 rpm). From the top of each tube, 150 μL of gradient was removed, and 1.25 mL of low-speed supernatant was layered. Gradients were centrifuged at 200,000 g for 2 h at 4 °C. Fractions (28 × 500 μL) were collected by upward displacement, and densities were verified with a refractometer. For ACE experiments, fractions F4, F5, and F6 were pooled and buffer-exchanged into BGE using Amicon Ultra-0.5 centrifugal filter units (10 kDa, regenerated cellulose, Merck KGaA, Darmstadt, Germany). Briefly, 35 μL of the pooled sample (Fpool) was loaded into a filter unit prerinsed three times with 500 μL BGE (14,000 g, 30 min, 4 °C). Subsequently, 500 μL of cold BGE was added, and the sample was centrifuged at 4 °C in two steps: 4,000 g for 30 min, followed by 14,000 g for 60 min. The final ∼30 μL of ultrafiltrated sample (Fpool,UF) was recovered by inverting the filter unit and centrifuging at 2,400 g for 2.5 min. Western Blot Samples were separated on a 4–12% Bis-Tris gel using NuPAGE MOPS SDS running buffer (Thermo Fisher Scientific) and electroblotted onto a 0.2 μm PVDF membrane (Thermo Fisher Scientific) overnight at 40 mA. Membranes were blocked for 60 min in 10% milk in phosphate-buffered saline with 0.1% Tween 20 (PBST) and incubated overnight with anti-UGCG antibody (clone 1E5, 1:500 in 5% milk in PBST; Cat# SAB1401393, RRID: AB_10608571, Merck). Membranes were then incubated with antimouse IgG HRP-linked secondary antibody (1:4000; Cell Signaling Technology, Danvers, MA) for 120 min. Protein detection was performed using Pierce ECL Western Kit (Thermo Fisher Scientific) and imaged on an Amersham Imager 600 (GE Healthcare, Chicago, IL). For total cell lysate and low-speed supernatant, 2 μg of protein were loaded per lane, whereas 20 μL of each ultracentrifuge fraction were applied. Nanoparticle Tracking Analysis Particle size distribution and concentration were determined using a NanoSight NS300 (Malvern Panalytical, Worcestershire, UK) equipped with a 532 nm laser and operated with NTA 3.4 Build 3.4.003 software. Samples were diluted with BGE and infused into the microfluidic cell at a flow rate of 40 arbitrary units via a syringe pump. For each sample, five 60 s acquisitions were recorded using a camera level of 13. Particle sizing calculations accounted for the BGE viscosity at 25 °C, measured as 1.222 cP according to a previously reported procedure. Activity Assay Golgi microsomes (3 μg of total protein) were incubated at 37 °C for 2 h in the BGE supplemented with 200 μM UDP-glucose, 500 μM ceramide C8-BSA complex, 5 mM MgCl2, 500 μM maleic acid (internal standard), and either 0 or 100 μM eliglustat hemitartrate. Reactions were quenched with an equal volume of ethanol, and protein aggregates were removed by centrifugation at 20,000 g for 5 min. Samples were subsequently analyzed by capillary electrophoresis. UGCG-specific UDP-glucose consumption was calculated as the difference between reactions conducted in the absence and presence of 100 μM eliglustat. A schematic of the protocol is provided in the Supporting Information (Figure S2). Capillary Electrophoresis All CE-UV experiments were conducted on an Agilent 7100 capillary electrophoresis system (Agilent OpenLab CDS 2.7) with the cassette thermostatically controlled at 25 °C. Separations were performed in PVA-coated capillaries (75 μm i.d., 31 cm total length, 8.5 cm effective length) at a constant voltage of −7 kV for anions and +7 kV for cations. Each day, capillaries were flushed with BGE for 10 min at 1 bar. Prior to each run, capillaries were conditioned by flushing with BGE for 4 min at 1 bar, except for indirect ACE experiments. CZE of Golgi Microsomes The samples were injected by vacuum at −35 mbar for 4 s, followed by a short plug of BGE at −20 mbar for 2 s. The monitored signals were 205.0 and 280.0 ± 2.0 nm with a reference wavelength at 500.0 ± 50.0 nm. The spectrum between 190 and 400 nm (step, 2 nm) was also stored. The run time was 30 min. Quantitation of UDP-Glucose The following injection sequence was used: −20 mbar for 2 s of water, −20 mbar for 4 s of sample, and −5 mbar for 1 s of BGE. The areas under the curve were determined at 205.0 and 260.0 ± 2.0 nm with a reference wavelength at 500.0 ± 50.0 nm for maleic acid and UDP-glucose, respectively. The run time was 5 min. Direct ACE Fpool,UF was first injected at −50 mbar for 17 s. Eliglustat solutions at varying concentrations (3.125, 3.75, 5, 6.25, 7.5, 10, 12.5, 15, 20, 25, 50, 100 μM), in BGE diluted 2 times with water and containing 200 μg/mL NMM, were subsequently injected at −20 mbar for 4 s, followed by a short BGE plug at −5 mbar for 1 s. UV signals were recorded at 205.0, 210.0, 215.0, 220.0, 225.0, 230.0, 260.0, and 280.0 ± 2.0 nm, with a reference wavelength of 500.0 ± 50.0 nm. Full spectra from 210 to 400 nm (2 nm steps) were also acquired. The run time was 7 min. Indirect ACE Prior to each run, the capillary was conditioned by sequentially flushing with BGE for 2 min at 1 bar, followed by BGE containing varying concentrations of miglustat (6.25, 25, 50, 78.125, 100, 156.25, 200, 312.5, 500, 1000, and 2000 μM) for 2 min at 1 bar. Sample injection consisted of miglustat-spiked Fpool,UF at −50 mbar for 17 s, 5 μM eliglustat in BGE diluted 2 times with water and containing 200 μg/mL NMM at −20 mbar for 4 s, and BGE containing different concentrations of miglustat (6.25, 25, 50, 78.125, 100, 156.25, 200, 312.5, 500, 1000, and 2000 μM) at −5 mbar for 1 s. UV detection parameters were identical to those used for direct ACE experiments. The run time was 7 min. Data Treatment The areas under the curve were calculated using Agilent OpenLab CDS 2.7 software. The migration times of eliglustat and NMM were determined with CEVal version 0.6i327 using signal A (205 nm) and deconvoluted signal, respectively. The deconvolution script processed as follow: 1) signal A (205 nm) and signal B (210 nm) traces were smoothed by moving average using a window size of 1.25 s, 2) the baseline value of each signal was obtained by averaging the absorbance before the plateau jump with an exclusion of the 1000 first points, 3) the FA/B ratio was determined by eqs and ) the deconvoluted trace was constructed using eq . 1 FA/B=averaget∈[tpeak1−60s;tpeak1−30s](AbsA(t)−baselineAAbsB(t)−baselineB) where F A/B is the ratio between the signals A (205 nm) and B (210 nm) of the matrix component, t peak1 corresponds to the time of the first peak (∼4 min) that is embedded in the plateau, Abs­(t) is the absorbance value of the signal mentioned as a subscript at time t, and baseline is the average of the absorbance value before the plateau of the signal mentioned as a subscript. 2 Absdeconvoluted(t)=AbsA(t)−(AbsB(t)−baselineB)FA/B where Abs deconvoluted (t) is the absorbance value at time t of the deconvoluted signal. Results and Discussion Our previous work established that ACE enables reliable hit validation and quantitative determination of dissociation constants (K D) using purified soluble proteins. − However, extending this approach to membrane proteins presents distinct challenges. These proteins frequently display limited stability in solution and a strong propensity for aggregation due to their hydrophobic domains. Moreover, correct folding and functional integrity generally depend on the presence of a lipid bilayer and a defined membrane microenvironment. ,,, Variations in lipid composition, including cholesterol content and other membrane constituents, can profoundly influence membrane organization and protein function. Collectively, these features complicate membrane protein–targeted drug discovery while simultaneously creating opportunities for selective modulation in specific cellular contexts. In this study, we evaluated ACE for membrane protein–ligand binding assays using UGCG, the rate-limiting Golgi enzyme for glucosylceramide synthesis. UGCG has traditionally been isolated from rat liver tissue, as attempts to obtain the recombinant protein in preparative quantities have met with limited success. Solubilization of UGCG using detergents is limited and affects its activity. ,, Moreover, nine splice variants are reported in the Ensembl database for the UGCG gene, which further increases its complexity. In order to provide a source that better reflects human malignant physiology and preserves UGCG in its native conformation, we instead used Golgi microsomes from the Caki-1 cell line, a well-established human model of metastatic clear cell renal cell carcinoma. The Caki-1 cell line was selected based on three criteria: (1) upregulation of UGCG at the protein level, , (2) suitability for drug screening due to genetic and phenotypic stability, and (3) its widespread use as a metastatic disease model. However, the effective target concentration in Golgi microsomes is substantially lower than what can be achieved with purified proteins, meaning that ACE measurements must be operated with high sensitivity under low-concentration conditions while maintaining physiological compatibility. To maximize sensitivity, detection must be performed at low wavelengths. In consequence, minimally UV-absorbing BGE must be used to reduce baseline noise. Phosphate buffer is useful in this respect. Besides its UV transparency, the BGE must be low-conductive and isotonic to avoid the generation of high current during electrophoresis while maintaining the integrity of Golgi microsomes. To achieve the isotonicity without increasing conductivity, sucrose was used as a nonconductive alternative to sodium chloride. The optimized BGE was composed of 30 mM sodium phosphate with 250 mM sucrose (pH 7.4), which provides a low UV cutoff (204 nm), isotonicity, low conductivity (3274 ± 18 μS/cm, representing a 75% reduction relative to phosphate-buffered saline, 13283 ± 161 μS/cm), and high buffering capacity at physiological pH (13.2 mM). Isolation and Characterization of Golgi Fractions from Cancer Cells for Interaction Studies To isolate Golgi membranes from Caki-1 cells, the low-speed supernatant was fractionated using a preformed continuous iodixanol density gradient. Based on the refractive index of the collected fractions (Figure S3) and the reported buoyant density of Golgi membranes (1.04–1.07), we examined fractions to identify UGCG. Western blot analysis confirmed the presence of UGCG, with fractions 2 and 5 exhibiting the strongest signal (Figure ). In both fractions, a prominent band was observed at approximately 45 kDa, consistent with the predicted molecular mass of UGCG derived from its amino acid sequence. UGCG levels in the total lysate, low-speed supernatant, and selected ultracentrifuge fractions of Caki-1 cells. The most intense ∼45 kDa band were fractions 2 and 5. In fraction 11, the ∼45 kDa band was weak, while a strong band at ∼33 kDa was present, consistent with a splice variant of UGCG lacking amino acids known to be critical for its activity. , The blue bands in the two external columns are the molecular weight markers (49 and 38 kDa). CL: total cell lysate, LSS: low-speed supernatant, F: fraction, UGCG: UDP-glucose ceramide glucosyltransferase. Fractions were then analyzed by capillary zone electrophoresis (CZE) to evaluate the quality of the isolated fractions that contained UGCG. Relative to the matrix control, only negatively charged species were detected (Figure S4, blue lines). Under negative polarity conditions, the electropherograms of the fractions 2, 4, 5, and 6 included a broad peak (from 4 to 30 min, Figure ) that is characteristic of heterogeneous nanoparticles. This is known that the Golgi apparatus of most cultured cells forms vesicles of heterogeneous size after cell disruption (i.e., Golgi microsomes). As expected for a fraction of the sample zone (Figure S3), fraction 2 exhibited a peak of Golgi microsomes distorted by contaminants. Because significant distortion of the Golgi microsomes peak was not observed in fractions 4, 5, and 6, these fractions were pooled to obtain sufficient material for subsequent analyses (Fpool). The particle concentration and size distribution of the Golgi microsomes were further characterized by NTA (Table S2 and Figure S5), validating the presence of nanometric vesicles in Fpool. A span value of 0.66 was measured, indicating a fairly monodispersed size distribution of the Golgi microsomes. Electropherograms under negative polarity conditions of the selected ultracentrifuge fractions of Caki-1 cells. In all fractions, a broad peak (from 4 to 30 min) was observed and corresponds to nanoparticles of heterogeneous size. However, this peak was distorted by contaminants in fraction 2 (sample zone). AU: absorbance unit, F: fraction. The next step in our workflow was to validate that UGCG remained in an active conformation. UGCG catalyzes the formation of glucosylceramide by transferring glucose from UDP-glucose to ceramide. Traditionally, UGCG activity assays use a radioactive, fluorescent, or deuterated substrate. Their product is then quantified by scintillation counter, TLC or HPLC with a fluorescence detector, or HPLC-MS, respectively. All these techniques require a labeled substrate that is highly expensive. To monitor the UGCG activity of the prepared Golgi microsomes in a label-free manner, we developed a CZE method that quantifies the consumption of UDP-glucose. The details of the analytical developments can be found in the Supporting Information (Supplemental Data–1.). Briefly, samples were incubated with C8-ceramide, UDP-glucose, Mg2+, an internal standard, and either 0 or 100 μM eliglustat, a specific UGCG inhibitor. Reactions were quenched with ethanol immediately after mixing or following a 2 h incubation at 37 °C, and residual UDP-glucose was quantified by CZE (Figure S6). UGCG-specific activity was determined from the difference in UDP-glucose consumption measured in the absence and presence of eliglustat. Since Golgi microsomes may contain other enzymes capable of using UDP-glucose as a substrate (e.g., UGGT1, UGGT2, PLOD3, ALG5), ,,, it was necessary to determine the fraction of the total UDP-glucose consumption specifically attributable to UGCG. As shown in Figure , Golgi membrane fractions consumed UDP-glucose relative to the matrix controls, confirming that enzymatic activity is preserved following cellular fractionation. At 100 μM, eliglustat inhibited UDP-glucose consumption in Fpool by 27 ± 3%, whereas F11 (Golgi) and F18 (smooth endoplasmic reticulum) displayed no inhibitor-dependent change. This was expected, as F11 and F18 serve as negative controls due to their low 45 kDa UGCG abundance observed by Western blot analysis (Figure ). These results demonstrate that our label-free activity assay specifically reflects UGCG activity. UDP-glucose consumption after 2 hours by membrane fractions isolated from Caki-1 cells. All tested fractions showed UDP-glucose consumption relative to the matrix controls, indicating that the enzymes remain active following cellular fractionation (Fpool, F11, and F18) and ultrafiltration-based buffer exchange (Fpool,UF). As anticipated from their low 45 kDa UGCG content observed by Western blotting, fraction F11 and F18 displayed no detectable UGCG-related activity. F11: ultracentrifuge fraction 11, F18: ultracentrifuge fraction 18, Fpool: pooled ultracentrifuge fractions (4, 5, and 6), Fpool,UF: pooled ultracentrifuge fractions (4, 5, and 6) after ultrafiltration, Matrix UF: matrix after ultrafiltration, UGCG: UDP-glucose ceramide glucosyltransferase. UGCG topology analysis of Fpool (adapted from Coste et al.) revealed that 93 ± 6% of Golgi microsomes are sealed in the right-side-out orientation, as determined from the differential impact of external proteolysis on enzyme activity. Details can be found in the Supporting Information (Table S1 and Supplemental Data–2.). To obtain reliable results using a partial-filling ACE method, it is essential to maintain comparable conductivity between the target plug and the BGE. Consequently, the buffer composition of Fpool must be close to the target-free BGE. To achieve this, an ultrafiltration-based buffer exchange was performed on Fpool, yielding Fpool,UF. The impact of this step, summarized in Table S2, was assessed through total protein quantification by BCA assay, microsome purity analysis by CZE (Figure S4, orange traces), nanoparticle concentration and size determination by NTA (Figure S5), and UGCG activity measurement (Figure ). The ultrafiltration step efficiently reduced matrix components (iodixanol and triethanolamine) while retaining 78 ± 3% of total protein, similar size distribution, 66 ± 3% of particles, and 69 ± 9% of UGCG activity per particle. The particle recovery is consistent with previous reported data, and the active conformation of UGCG was preserved throughout the process. Therefore, affinity measurements can be reliably performed using Fpool,UF. Affinity Assessment of Eliglustat Binding to Cancer-Derived Golgi Microsomes The last stage of our developments was to set up a partial-filling ACE method using cancer-derived Golgi microsomes as target source. In this configuration, a defined segment of the capillary was filled with Golgi microsomes (Figure A), while the remaining length contained the target-free BGE. Upon application of an electric field, the potential ligand, namely analyte A, migrated through the microsome-containing zone and was subsequently detected in the microsome-free electrolyte. Consequently, the observed electrophoretic mobility (μ obs ) reflects the contributions from the analyte A mobility within the microsomal plug (μ A,eff ), its mobility in the BGE (μ A ), and the electro-osmotic flow (μ EOF ). Within the microsomal zone, the effective mobility (μ A,eff ) represents a population-weighted average of the mobilities of the free analyte (μ A ) and the analyte-microsome complex. , Among the well-characterized UGCG inhibitors, we selected eliglustat as known UGCG ligand to evaluate our ACE method due to its high potency, high selectivity, and established use in clinical cancer research. , Partial-filling ACE method in direct mode. (A) Principle: First, the capillary is partially filled with the Golgi microsomes. Then, a short plug containing analyte A and the noninteracting compound is injected. After voltage application, the noninteracting compound and the analyte A cross the Golgi microsomes plug and arrive at the detector. Upon binding of the analyte A, a shift is observed in its migration time. (B) Electropherograms: The addition of a 3.5 cm plug of Golgi microsomes shifted the migration time of eliglustat. The shift was reduced with increasing concentration of eliglustat. (C) Affinity determination of eliglustat: The resulting bound fraction calculated from migration times were plotted against the eliglustat concentration to determine the affinity. More details can be found in Figure S8 of the Supporting Information. BGE: background electrolytes, EOF: electro-osmotic flow. Prior to affinity measurements, key ACE parameters, including the Golgi microsome plug length and the electro-osmotic flow (EOF) marker, were optimized. Although the fraction of eliglustat bound is independent of plug length, the migration time shift upon binding increased linearly with plug length. Longer plugs therefore enhanced the magnitude of the migration shift, improving the precision of bound fraction determination. The plug length was set to ∼3.5 cm, corresponding to the maximum length that allowed complete migration of eliglustat through the target plug before detection. In ACE, preventing interactions between the binding partners (Golgi microsomes and analytes) and the capillary surface is essential. For this reason, we used PVA-coated capillaries, which also nearly suppress EOF and thereby enhance the mobility difference between Golgi microsomes and eliglustat, increasing the magnitude of the migration shift upon binding. However, because PVA coatings are physically adsorbed onto silica, they undergo slight but gradual erosion over repeated injections. Over 91 injections on the PVA-coated capillary, the EOF increased from 0.18 to 0.31 × 10–8 m2·V–1·s–1, yet it remained negligible compared with the EOF typically observed with bare fused-silica capillaries (around 4 × 10–8 m2·V–1·s–1). However, the EOF cannot be assumed constant, and its determination at each injection is therefore critical for calculating the bound fraction of eliglustat. A neutral, noninteracting EOF marker could not be used for two reasons: (1) residual iodixanol in the Golgi microsome plug produces a strong UV background in that region of the electropherogram, and (2) neutral markers substantially extend analysis time. Instead, we employed N-methylmorpholine (NMM), a cationic, noninteracting compound that migrates before eliglustat and can be detected at concentrations that do not alter the pH. NMM has already been validated as a buffer component for ACE–MS binding studies and is structurally related to Good’s buffers, which are known to be stable, nonmetabolized, and not act as substrate analogues or enzyme inhibitor. For each run, the EOF was determined from the observed electrophoretic mobility of NMM. Effective electrophoretic mobility of NMM in the BGE was measured at 1.260 × 10–8 m2·V–1·s–1 (RSD = 1.3%, n = 91). In addition, the effect of the applied electric field on Golgi microsomes was evaluated using fraction F9, which is comparable to Fpool but contains a lower abundance of the 45 kDa UGCG protein (Figure ). The sample preparation is described in the Supporting Information (Supplemental Data-3). The response of flexible vesicles to an electric field is known to depend on several parameters, including the conductivity ratio between the media inside and outside the vesicle, vesicle size and shape, membrane elasticity, and field strength. Under the conditions investigated, the electric field did not induce significant alterations in the Golgi microsome integrity. Comparable particle size distributions and concentrations were observed before and after electrophoresis, confirming that the ACE conditions do not cause long-term deformation of the Golgi microsomes (e.g., rupture) or adsorption on the capillary wall (Figure S7 and Table S3). In ACE, dissociation constants (K D) are typically derived from expressions based on the free protein concentration, which are valid under the assumption that the protein is in large excess relative to the analyte. ,, Under our experimental conditions, this assumption was not satisfied, as evidenced by the electromigration dispersion of the eliglustat peak (consistent with a low target-to-analyte ratio) and by the dependence of the eliglustat bound fraction on the total analyte concentration ([A] 0). We therefore applied the binding model described by Swillens. In this framework, eliglustat binds to a single class of specific targets with 1:1 stoichiometry, as well as to nonspecific sites. Specific binding is defined by K D and the total accessible protein concentration ([P] 0), while nonspecific binding increases linearly with the free eliglustat concentration ([A]) and is characterized by the parameter NS, representing the ratio of nonspecifically bound analyte to [A]. Under rapid-equilibrium conditions, the bound fraction of eliglustat is described by eq (Supplemental Data-4). Bound⁢ fraction=(−b+b2−4c2KD(1+NS)−b+b2−4c+NS).(11+NS) 3 where b=(1+NS)KD+[A]0−[P]0 c=−(1+NS)KD[P]0 In our ACE setup, the bound fraction of eliglustat was calculated using eq (Supplemental Data-5): 4 Boundfraction=μobs−(μA+μEOF)(1−p)μobs−(μA+μEOF) where μ obs is the observed electrophoretic mobility of eliglustat in the presence of the target plug, μ A is the mobility of the free analyte, μ EOF is the electro-osmotic flow, and p is the fraction of the effective capillary length occupied by the target plug. The dissociation constant (K D) of eliglustat was determined by testing different concentrations (Figure B), with bound fractions calculated according to eq . The resulting data were fitted to eq to obtain K D, the nonspecific binding parameter (NS), and the total accessible protein concentration ([P] 0) (Figure C). Measurements were performed independently in duplicate. [P] 0 was obtained on the basis of the binding curve, as direct quantitation of UGCG would be unreliable due to its transcript complexity, poor intrinsic solubility (CamSol intrinsic solubility score at pH 7.4: −3.1), and limited membrane extraction. ,, Although [P] 0 could vary due to the ultrafiltration step, K D and NS were consistent across experiments, as confirmed by an extra sum-of-squares F test (α = 0.05). For Caki-1 Golgi microsomes, eliglustat exhibited a K D of 15.49 [10.08–23.76] μM (Figure S8). Reported inhibition potencies vary with cell type, ranging from 24 nM in K-562 cells to 42 μM in JVM-2 cells, consistent with the lack of a detectable bound fraction in AS-MS experiments. , The binding model used in this study explicitly accounts for nonspecific binding as a function of the free analyte concentration, inherently incorporating the analyte partition into the lipid bilayer as well as all other nonspecific interactions that follow this function. The value of the NS parameter depends on the analyte and the composition of the microsomal membranes. Estimating specific binding by subtracting experimentally determined nonspecific binding from total binding is meaningless when a significant fraction of the analyte is bound. As described by Swillens, nonspecific binding is no longer a linear function of the total analyte concentration. Consequently, conventional approaches, such as the addition of excess competitor or the use of control microsomes lacking UGCG, lead to an overestimation of nonspecific binding, as they increase the free analyte concentration. To further validate the specificity of the observed binding, four concentrations of eliglustat were tested using F18 under the same experimental conditions as for Fpool. F18 is a smooth endoplasmic reticulum fraction that exhibits similar vesicle size distribution and concentration as Fpool (Figure S9 and Table S4) but lacks UGCG (45 and 33 kDa; Figure ). Indeed, at 5 μM eliglustat, a high bound fraction of 0.33 ± 0.02 (superior to Fpool) was recorded with F11, suggesting that the inactive 33 kDa UGCG also binds eliglustat. The analysis of multiple eliglustat concentrations enables discrimination between purely nonspecific binding and binding that includes a specific component. In the case of nonspecific binding alone, the bound fraction remains independent of the total analyte concentration and is equal to the constant NS(1+NS) . As shown in Figure S10, the eliglustat bound fraction measured using F18 remained constant across the tested concentrations, in contrast to the concentration-dependent behavior observed for Fpool, thereby supporting the validity of the model. Affinity Assessment of Miglustat by Competitive Displacement Titration The direct ACE method described above is limited to compounds that are both UV-active and positively charged at physiological pH. To address this limitation, we explored an indirect ACE approach using cancer-derived Golgi microsomes, which also allows detection of weaker interactions. In this mode, a potential competing ligand, analyte B, is added to the BGE, modulating the binding of eliglustat to microsomes, and the resulting shift in the migration time of eliglustat is monitored (Figure A). Miglustat was selected as the competing ligand because of its lower potency and selectivity toward UGCG and its chemical structure, which differs markedly from eliglustat. It is a synthetic N-alkylated iminosugar with a pK a of 7.1. Given that molecules with distinct physicochemical properties generally exhibit different nonspecific binding behaviors, miglustat ability to displace eliglustat from Golgi microsomes would further confirm the specificity of eliglustat binding. Partial-filling ACE method in indirect mode. (A) Principle: The background electrolytes (BGE) were spiked with analyte B. First, the capillary is partially filled with the Golgi microsomes in the presence of analyte B. Then, a short plug of eliglustat (analyte A) and the noninteracting compound is injected. After voltage application, the noninteracting compound and eliglustat cross the Golgi microsomes plug and arrive at the detector in the spiked BGE. The presence of analyte B reduces the eliglustat binding to the Golgi microsomes. Therefore, the eliglustat migration time shifts back. (B) Electropherograms: The addition of a 3.5 cm plug of Golgi microsomes shifted the migration time of eliglustat. This shift was reduced with the increasing concentration of miglustat in the BGE. (C) Affinity determination of miglustat: The best-fit values of the nonlinear regression were 153.1 [119.1–199.8] μM (K B) and 2.641 [2.465–2.824] μM ([P] 0). For the fitting, the [A] 0, K D, and NS values were fixed at 5 μM, 15.49 μM, and 0.03519, respectively. BGE: background electrolytes, EOF: electro-osmotic flow, Without Golgi: in the BGE with no plug of Golgi microsomes, M0: in the BGE with a 3.5 cm plug of Golgi microsomes, M500: in the BGE containing 500 μM of miglustat with a 3.5 cm plug of Golgi microsomes, M1000: in the BGE containing 1000 μM of miglustat with a 3.5 cm plug of Golgi microsomes M2000: in the BGE containing 2000 μM of miglustat with a 3.5 cm plug of Golgi microsomes. To account for competition, the binding model was extended to include a second ligand, analyte B, that binds the target with a 1:1 stoichiometry and competes with eliglustat. A ternary complex (target·eliglustat·B) was assumed not to form. Given the expected weak affinity of analyte B, the free concentration [B] was approximated by the total concentration [B] 0, as the bound fraction of miglustat is negligible. This simplification avoided the implication of a cubic equation during mathematical developments. Under rapid-equilibrium conditions, the bound fraction of eliglustat in the presence of the competing ligand B is described by eq (Supplemental Data-6). 5 Boundfraction=(−b+b2−4ac2KD(1+NS)([B]+KB)−b+b2−4ac+NS)×(11+NS) where a=([B]+KB) b=(1+NS)KD([B]+KB)+KB([A]0−[P]0) c=−(1+NS)KD[P]0KB The dissociation constant of miglustat (K B) was determined by monitoring the effect of varying miglustat concentrations in the BGE on the bound fraction of eliglustat (Figure C). As shown in Figure B, increasing miglustat concentrations led to a reduction in the migration time shift of eliglustat. Eliglustat was injected at 5 μM ([A] 0), and the resulting bound fractions were plotted against the concentration of miglustat. Fitting these data to eq yielded a K B of 153.1 [119.1–199.8] μM, consistent with the reported potency range of miglustat (IC50= 22.9–100 μM). Collectively, the use of a UGCG-depleted fraction (F18) and a competitive binding assay with a structurally distinct UGCG inhibitor provides converging evidence that the observed mobility shifts of eliglustat arise from specific interactions with UGCG rather than nonspecific effects. Conclusions Membrane proteins constitute 20–30% of the proteome and serve as targets for over half of all approved drugs, yet their study remains hampered by difficulties in production and in preserving therapeutically relevant conformations during drug discovery. There is an urgent need for methods that can interrogate interactions in complex, physiologically relevant environments. Here, we demonstrate that cancer-derived Golgi microsomes provide such a system and show that the ACE methodology overcomes key barriers in membrane protein characterization. ACE requires neither target modification nor intrinsic activity and, in its microfluidic format, dramatically reduces sample consumption. In direct mode, the separative power of capillary electrophoresis is preserved, enabling analysis of impure compounds in a single run, while the indirect mode extends applicability to ligands lacking charge or UV absorbance. Importantly, integration with mathematical modeling eliminates the need for large protein excess and corrects for nonspecific binding, addressing limitations that have long constrained traditional ACE approaches. By enabling the assessment of target engagement in complex, physiologically relevant environments, this approach may support more informed compound selection in the early development process. Applying this approach to UGCG, an intracellular Golgi-resident membrane protein, we demonstrated that ACE can quantify ligand binding directly at the surface of cancer-derived Golgi fractions. These fractions, composed of fairly monodisperse vesicles (∼150 nm), underscore the method ability to interrogate complex, membrane-bound systems. By preserving the native protein composition, including components essential to Golgi architecture, this approach prioritizes physiological relevance. Collectively, this work extends the scope of ACE in drug discovery. This strategy is readily transferable to a broad range of intracellular and extracellular membrane-bound particles of therapeutic relevance, such as liposomes and extracellular vesicles, and thus can probe how vesicle curvature and composition influence binding equilibrium. Together, these advances establish ACE as a robust, versatile platform to study biomolecular interactions in pathologically relevant contexts. Supplementary Material