Electrostatic bimodality of human cytosolic DNA sensor cGAS-dsDNA condensates revealed by millisecond flashing electrophoretic separation Authors: Xiaoting Ling, Xingrui Song, Dapeng Zhang, Weiyi Lai, Cong Liu, Guibin Jiang, Qiang Zhao, Hailin Wang Published: 2026 ABSTRACT AbstractElectrostatic landscapes within biomolecular condensates are critical regulators of their structural stability and functional plasticity. To quantitatively resolve these elusive electrical signatures, we develop a millisecond Flashing Electrophoretic Separation (msFES) platform enabling nanoliter-scale profiling of biomolecular condensates at near single-droplet resolution—with a focus on zeta potential characterization. Applying msFES to condensates formed by human cytosolic DNA sensor cGAS (hcGAS) and DNA, we unexpectedly discover an intrinsic electrostatic heterogeneity: droplets formed from identical components under identical conditions exhibit two distinct zeta-potential populations. This heterogeneity dictates differential susceptibility to TREX1, an exonuclease that selectively disassembles negatively charged droplets while sparing or amplifying positively charged assemblies, thereby potentially modulating cGAS activation. Pathogenic hcGAS variants (K432T, G303E) disrupt this electrostatic control, rendering hcGAS-DNA droplets hypersensitive to TREX1-mediated DNA degradation. msFES thus uncovers a hidden layer of charge-based regulation in phase-separated systems, potentially linking condensate biophysics to innate immune control.Subject terms: Biotechnology, Chemical biology Introduction As a sentinel cytosolic DNA sensor, human cyclic GMP-AMP synthase (hcGAS) detects viral DNA or stress-induced self-DNA1–4. Upon engaging double-stranded (ds) DNA (> 45 bp)5,6, hcGAS simultaneously activates two interlocked defense programs: enzymatic production of 2′,3′-cGAMP to initiate STING-dependent interferon responses via endoplasmic reticulum signaling3,7, and liquid-liquid phase separation (LLPS) that organizes hcGAS-DNA condensates into functional hubs. These dynamic compartments selectively exclude inhibitory factors (e.g., TREX1, BANF1/BAF)8–11, while concentrating hcGAS activation platforms to achieve local cGAMP concentrations exceeding STING activation thresholds. By spatially constraining reaction-diffusion dynamics, the condensates enforce digital immune signaling—preventing signal leakage yet enabling rapid, all-or-none responses to viral infections3,12–16 and cellular stress17–21. This spatial organization proves critical for cellular homeostasis, as its collapse triggers pathological cascades: leaked self-DNA sensors drive autoinflammation22–24, dispersed antiviral effectors blunt interferon responses3,25,26, and maladaptive stress signaling promotes oncogenic transformation27–29. Thus, LLPS emerges not merely as a structural scaffold but as a kinetic proofreading system essential for balancing immune vigilance and self-tolerance through precise spatiotemporal control. The characterization of LLPS droplets leverages multimodal methodologies to probe their structural and dynamic properties: optical/fluorescence microscopy5,7,9,30, fluorescence cross-correlation spectroscopy31,32, turbidimetry6, dynamic light scattering33–35, small-angle X-ray scattering36,37, single-molecule detection38–41, microfluidics42–44, capillary flow experiments (Capflex)35, electrophoresis (electrophoretic mobility shift assay (EMSA)6,7, free-flow electrophoresis (μFFE)45, laser Doppler electrophoresis46–48, microelectrophoresis49,50), and tweezer experiments51,52, and field-effect transistor (FET) sensors53. Although these approaches collectively map condensate assembly/disassembly dynamics and interaction networks, critical gaps persist in resolving charge-mediated interactions, a fundamental determinant of condensate stability and biological activities. Intrinsic surface charges and charge blocks54–56 govern droplet stability through inter-droplet repulsion (F ∝ ζ2/d2)45,57 and selective biomolecule partitioning. Despite evidence that imaging and computational simulations estimate pH gradients58,59 or interfacial potential gradients60,61 between condensate and dilute phase, the ζ potential landscape of LLPS droplets remains largely unmapped. Moreover, correlations between droplet electrostatic potentials and biological regulations in physiological environments remain unestablished, with techniques such as μFFE45, microelectrophoresis50 underutilized for this purpose. This knowledge gap underscores the need for next-generation analytical platforms capable of resolving transient charge states in native condensate populations and biological fates. To overcome the intrinsic limitations of conventional ζ potential analysis—which demands a large sample volume (≥ 500 μL) and non-physiological protein concentrations (≥ 1 μM) incompatible with fragile condensates, here we report a millisecond Flashing Electrophoretic Separation coupled Laser-Induced Fluorescence (msFES-LIF) detection platform. This innovation integrates high separation efficiency and the physiological buffer of capillary electrophoresis with ultra-sensitivity of LIF (down to single molecule levels)62–70. This platform enables quantification of ζ potential profile of fragile biomolecular condensates at pseudo-single-droplet resolution using nanoliter volumes (~ 10 nL, 104 times reduction in volume) and nanomolar-concentration molecules. With this msFES platform, we demonstrate the bimodal ζ potential distribution of hcGAS-dsDNA droplets, and that the droplets with positive ζ potential render themselves to be resistant to negative regulatory factor TREX1. Overall, this platform pioneers quantitative condensate electrophysiology, transforming our capacity to dissect charge-encoded biological regulation. Results High-resolution separation of cGAS-DNA droplets by Mg2+-tuned msFES A conventional EMSA verified hcGAS binding to dsDNA (120 bp, named as ds120) (40 nM ds120, Kd = 120 ± 6 nM, Fig. S1) but failed to resolve the condensates themselves, likely because the gel matrix (pore size ≤ ~30 nm71) disrupts droplets whose diameters span ~50 nm to several micrometers31. In contrast, our msFES-LIF platform employs charge-to-mass differential migration and enables matrix-free separation of macromolecular complexes in open tubular capillary (25 μm i.d.) (Fig. 1)63,72,73, thereby preserving native condensate architectures. Schematic diagram illustrating ζ potential measurement for biomolecular droplets revealed by millisecond flashing electrophoretic separation at near single droplet resolution. Implementing our msFES-LIF platform with TMR-labeled ds120 (TMR-ds120, 4.0 nM) and hcGAS (25 nM) in sample buffer (20 mM Tris-HCl, pH 7.4, 20 mM NaCl, 5 mM MgCl2), we detected two distinct migratory populations (annotated with red dots, Fig. 2A): fast-migrating and slow-migrating flashing species. Both types of flashing species were presumably attributed to the hcGAS-DNA droplets (Fig. 2A). In this separation, neutral rhodamine B (RB) (or baseline drop caused by sample buffer, Fig. S2) marked electroosmotic flow (EOF). Since positively charged species migrate faster than EOF and negatively charged ones migrate slower, the fast population is positively charged, and the slow one is negatively charged. To validate this observation, msFES was applied to condensates with known surface charges50,60. The ζ-potentials of DDX4-IDR droplets with a positive charge, mCherry-tagged FUS-LC with a negative charge, and [(RGRGG)5/polyU75 with a neutral charge, were all measured with msFES and show concordance with previous reports (Fig. S3). High-resolution Mg²⁺-mediated millisecond Flashing Electrophoretic Separation coupled Laser-Induced Fluorescence (msFES-LIF) analysis of hcGAS-DNA droplets. A Mg²⁺-mediated msFES electrophoretogram of hcGAS-DNA droplets. Samples were prepared in the buffer of 20 mM Tris-HCl (pH 7.4) + 20 mM NaCl+ 5 mM MgCl2 and included 4 nM TMR-labeled dsDNA (120 bp, TMR-ds120) and 25 nM hcGAS. msFES buffer (1× TG) contained varying Mg²⁺ concentrations (0–0.5 mM). Red dots denote flashing droplet peaks. A. U. means arbitrary unit for fluorescence intensity. RB rhodamine B, neutral electroosmotic flow marker. B Half-height peak width analysis of flashing droplet peaks. The whiskers of the violin plots represent the mean ± SD, and the distribution is fitted to a Gaussian distribution (droplet numbers: 71, 100, 107, and 113, respectively; pooled from 3 biological replicates). C Half-height peak width analysis of free DNA peaks (mean ± SD of 3 biological replicates). D The number of flashing droplet peaks with a S (signal)/N (noise) > 3. Mean ± SD of three biological replicates. Two-sample t-test assuming equal variances, two-tailed, p-value = 0.015, 0.015, 0.011, respectively. E Fluorescence intensity of flashing droplet peaks with a S/N > 3. Each dot represents fluorescence intensity of a single droplet (droplet numbers: 71, 100, 107, and 113, respectively; pooled from 3 biological replicates). Bars represent the mean, and whiskers represent the SD. The F-test was used to confirm unequal variances, followed by a two-sample Welch’s t-test (unequal variance t-test) with two-tailed analysis, p-value = 0.13, 0.65, 0.29, respectively. No adjustments for multiple comparisons were performed. Not significant (n.s.) p > 0.05, *p ≤ 0.05, **p ≤ 0.01, ***p < 0.001 (D–H). Source data are all provided as a Source Data file. Building on previous work showing that Mg²⁺ supplementation in electrophoresis buffers enhances oligonucleotide separation through charge shielding effects74, we developed an innovative msFES incorporating Mg²⁺ modulation. Systematic evaluation of Mg²⁺ concentrations (0.1–0.5 mM) revealed significant improvements in resolving hcGAS-dsDNA droplets, as evidenced by a few well-resolved and discrete flashing droplet peaks (Fig. 2A). These characteristic flashes exhibited remarkably consistent temporal profiles, with a mean half-height width of 0.16 ± 0.04 s (~160 ms, Fig. 2B), contrasting with the broad peaks of free DNA molecules (half-height widths: 0.6–5.1 s, Fig. 2C). Dose-response analysis identified 0.1 mM Mg²⁺ as the optimal concentration, yielding maximum fluorescence intensity (1.4-fold increase compared to the 0.5 mM conditions) while maintaining comparable peak frequency to 0.2–0.5 mM conditions (~ 34 flashing droplet peaks in total, Fig. 2D, E). This concentration was therefore adopted for subsequent experiments. Whereas conventional electrophoretic peaks typically follow Gaussian distributions consistent with Fickian diffusion75, the observed ultra-narrow flashing droplet peaks (~ 160 ms) demonstrate non-Boltzmann separation behavior, indicating detection at the single-droplet level. These discrete fluorescence transients imply two key properties of the tested LLPS droplets: (1) Unlike the field-induced collapse reported for low-viscosity droplets76, DNA is retained within our condensates without dissociation or broadening, even at high field strengths (714 V/cm; 20 kV/28 cm); and (2) the collective migration patterns are consistent with solid particle–like behavior of intact droplets. This unique separation mechanism allows the gel-free msFES-LIF platform to simultaneously preserve native droplet morphology through minimized shear forces77 (estimated 0.48 nN/μm², see Fig. S4) and achieve high-throughput analysis with separation efficiencies of 0.5–17.5 million theoretical plates, completing separations within 2.5 min. Flashing peaks are associated with the cGAS-DNA droplets To unequivocally establish the association of the transient electrophoretic signatures (flashing peaks in msFES) with cGAS-dsDNA droplets, we conducted systematic mutagenesis targeting critical LLPS regulatory domains (Figs. 3A and S5–S7). Structural perturbation analysis revealed that the R300E/K301E double mutations in DNA-binding Site C6, a key electrostatic interface, caused near-complete LLPS suppression (Fig. 3B), manifested as 56% reduction in the droplet number (34 ± 4 vs. 15 ± 2, p = 0.0044, 3 biological replicates, Fig. 3E) and 5.6-fold increase in dilute-phase DNA concentration (0.05 ± 0.003 vs. 0.28 ± 0.07 nM, Fig. 3H). Fluorescence imaging confirmed the reduction in the droplet number (Fig. 3C). The msFES flashing peaks exclusively originate from hcGAS-DNA droplets. A Schematic representations of wild-type (WT) and mutated hcGAS proteins. B, C The msFES electrophoretogram (B) and corresponding fluorescence imaging(FI) (C) of wild-type and mutated hcGAS-DNA droplets. The samples were prepared with 20 mM Tris-HCl, pH 7.4, 20 mM NaCl, and 5 mM MgCl2, and included 25 nM hcGAS (wild-type or mutated) and 4 nM TMR-ds120. Red dots denote flashing droplet peaks. A. U. means arbitrary unit for fluorescence intensity. The scale bar: 10 μm. Each experiment was repeated three times independently with similar results. D Half-height peak width analysis of wild-type and mutated hcGAS-DNA droplets, free DNA, and rhodamine B (3 biological replicates). E The number of flashing droplet peaks corresponding to (B) (mean ± SD of 3 biological replicates). Two-sample t-test assuming equal variances, two-tailed, p-value = 0.0022, 0.0002, 0.0002, respectively. F Distribution of flashing droplet peaks across defined fluorescence intensity intervals corresponding to (B) (mean ± SD of 3 biological replicates). G Fluorescence intensity of flashing droplet peaks corresponding to (B). Each dot represents fluorescence intensity of a single droplet (total droplets number 102, 46, 309, 7, respectively, from 3 biological replicates). Bars represent the mean, and whiskers represent the SD. The F-test was used to confirm unequal variances, followed by a two-sample Welch’s t-test (unequal variance t-test) with two-tailed analysis, p-value = 0.0046, 0.0013, 0.0001, respectively. No adjustments for multiple comparisons were performed. H Free DNA concentration in the dilute phase corresponding to (B). (mean ± SD of 3 biological replicates). Two-sample t-test assuming equal variances, two-tailed, p-value = 0.0044, 0.0005, 2.0 × 10−7, respectively. Not significant (n.s.) p > 0.05, *p ≤ 0.05, **p ≤ 0.01, ***p < 0.001 (E, G, H). Source data are all provided as a Source Data file. Truncation mutant Δ146 (residues 1–146 removed), which deletes the N-terminal disordered region6,7,9,78, significantly altered LLPS: droplet number increased 3.0-fold (p = 0.0005, 3 biological replicates) while mean fluorescence intensity decreased by 72% (Fig. 3B, E, G), suggesting this region modulates droplet maturation via charge-hydration balance. Morphometric analysis further confirmed that Δ146 increased droplet density by 60% (40 ± 2 vs. 63 ± 8 particles/6711 μm², p = 0.0013; Fig. S7A) and reduced mean diameter by 50% (1.61 ± 0.97 μm vs. 0.80 ± 0.35 μm, p < 0.001; Figs. 3C and S7B). ThT selectively stains amyloid-like solid aggregates but not liquid droplets61. Consistent with the reported formation of solid aggregates at micromolar level78, high concentrations (2.0 µM Δ146 cGAS) indeed yielded ThT-positive aggregates with dsDNA (Fig. S8B). However, at physiologically relevant low concentrations (100 nM Δ146 cGAS) as we investigated using msFES assay, droplets formed with dsDNA (10 nM) were largely ThT-negative (Fig. S8A), confirming their liquid state. Interestingly, even at high concentrations (2.0 µM), full-length hcGAS formed exclusively ThT-negative condensates. Since ThT staining distinguishes solid-like aggregates from liquid droplets, it provides a complementary method to assist the msFES assay in identifying these distinct physical states. Combining Δ146 with R300E/K301E mutations produced synergistic suppression of LLPS: droplet number decreased by 93% (p = 0.0002, 3 biological replicates; Fig. 3B, E, F), mean fluorescence intensity dropped by 86% (Fig. 3G), and dilute-phase DNA increased 60-fold (Fig. 3H). These systematic perturbations confirm that the electrophoretic transients (flashing peaks) originate from discrete LLPS droplets. Species-specific analysis revealed that mouse cGAS, which has a structurally weaker DNA-binding Site C interface, showed 68% fewer flashing droplet peaks (10 ± 5 vs. 32 ± 2) and 73% lower mean fluorescence intensity than human cGAS (Fig. S5D), consistent with fluorescence morphometry (Figs. S5B and S7A, B). Intriguingly, the E401A mutation (Site A), which enhanced dsDNA binding affinity (Kd = 54 ± 3 nM vs. WT 120 ± 6 nM; EMSA in Fig. S1A–C), unexpectedly impaired LLPS, reducing flashing droplet peaks by 38% (21 ± 2 vs. 34 ± 4) (Fig. S6C) and fluorescence intensity by 57% (Fig. S6D). Mutant droplets were 48% smaller in diameter (0.83 ± 0.43 μm vs. 1.61 ± 0.97 μm; Fig. S7B). This inverse relationship between binding affinity and phase separation suggests that LLPS requires dynamic multivalency rather than excessive stability. Quantitative msFES profiling demonstrated concentration-dependent amplification of transient flashing droplet peaks (as indicated by the number of flashing droplet peaks) across the physiologically relevant hcGAS concentration range (0–25 nM, Fig. S9A). Remarkably, this response exhibited exceptional linearity (R² = 0.99, Fig. S9C). Mean fluorescence intensity of droplets also increased proportionally with hcGAS concentration (Fig. S9D), consistent with morphometric data showing larger droplets at higher protein levels (Fig. S9B, E), confirming that signal amplitude reflects droplet volume. This quantitative correlation establishes msFES as a precision digital analyzer for phase-separated architectures with near single-droplet resolution. All qualified flashing droplet peaks (SNR ≥ 3) showed consistent temporal profiles, with half-height widths averaging 0.15 ± 0.02 s, 5–30 times narrower than those of control analytes (RB: 4.6 s; free TMR-ds120: 0.72 s; Fig. 3D). Peak width remained constant throughout migration (0.6–1.5 min; Fig. 3D). With 60 Hz acquisition (Fig. S10), each transient comprises ~10 digitized points, enabling clear resolution of individual droplets (minimum interval 0.08 s at Rs = 0.5). Bimodal ζ potential distribution of hcGAS–dsDNA droplets revealed via msFES As derived in Eq. (8) in Supporting Information50,79,80, we applied msFES to quantify the ζ potentials of hcGAS–DNA condensates, revealing two electrophysiologically distinct subpopulations: one positively charged and one negatively charged (Fig. 4B–D). Near-neutral droplets were rarely detected. ζ potentials broadly distributed from −40 to +30 mV, with negative droplets (centered near −32.5 ~ −37.5 mV) predominating in both abundance and fluorescence intensity across all hcGAS concentrations tested (Fig. 4C, D). Positive droplets exhibited a narrower distribution around +22.5 mV. Under these low-salt conditions, positively charged cGAS-DNA droplets accounted for approximately 24.9%, while negatively charged droplets represented about 75.1% (Fig. 4C). When the electric field strength in msFES was changed from 714 to 179 V/cm, the peak ζ potentials of both positively and negatively charged droplets, as measured by msFES-LIF, remained unchanged (Fig. S11). Profiling of ζ potentials of hcGAS-DNA droplets by msFES. A Theoretical framework for calculating droplet ζ potentials via msFES. B Fluorescence intensity distribution across ζ potentials of hcGAS-DNA droplets separated by msFES with an electrophoretic buffer of 1 × TG + 0.1 mM MgCl2. The scatter points represent individual droplets. n = the total number of droplets from 3 biological replicates. A. U. means arbitrary unit for fluorescence intensity. C Distribution of hcGAS-DNA droplet counts across ζ potential intervals with bimodal Gaussian fitting. n = the total number of droplets. Data were pooled from 3 biological replicates. D Distribution of summed fluorescence intensities across ζ potential intervals with trimodal Gaussian fitting. n, the total number of droplets, is consistent with that in (C). The source data are provided as a Source Data file. Notably, fluorescence intensity covaried with ζ potential in a charge-specific manner. Highly negative droplets (ζ ≈ −37.5 mV) displayed higher fluorescence intensity, whereas less negative populations (ζ ≈ −22.5 mV) showed reduced signal. The positive cluster peaked at ζ ≈ +22.5 mV with high intensity (Fig. 4D). This multimodal distribution suggests a maturation process wherein small and low-|ζ| assemblies evolve into larger condensates with enhanced charge and fluorescence—supporting a role for electrostatic regulation in condensate stability. Temporal analysis revealed that both negatively and positively charged droplets were present in freshly prepared samples and persisted in aged samples (Fig. S12). In fresh droplets (10-min incubation), both populations exhibited low fluorescence intensity, indicating smaller sizes. Initially, negatively charged droplets comprised two main populations: a larger one with higher fluorescence intensity and a peak ζ potential around −37.5 mV, and a smaller one with a peak between −22.5 and −17.5 mV. Positively charged droplets were smaller in size than their negative counterparts, with a peak ζ potential near +22.5 mV (Fig. S12B). With increased aging time, the intensity of negatively charged droplets with higher absolute ζ potential increased substantially (Fig. S12A, B), indicating droplet growth. Concurrently, the subpopulation with lower absolute ζ potential diminished in intensity and showed broader distribution, suggesting a gradual maturation into droplets with higher charge magnitude and larger size. For positively charged droplets, aging led to increased droplet size and intensity but a decrease in droplet number, consistent with a coarsening process where smaller droplets coalesce into larger, more mature ones. The total droplet number decreased over incubation time (Fig. S12C), further supporting the coalescence of small droplets into larger ones. Collectively, the temporal data indicate that both positively and negatively charged droplets coexist from the fresh state through aging, and undergo a gradual maturation process over time. Dissection of molecular forces underlying hcGAS–dsDNA droplet formation and ζ potential To elucidate the molecular drivers of hcGAS–dsDNA condensate nucleation and maturation, we systematically probed three major forces governing phase separation81: electrostatic complementarity, hydrophobic interactions, and hydrogen bonding. This framework rationalizes the observed charge bifurcation—cationic, hcGAS-rich droplets form via charge-balanced coacervation, while anionic, DNA-dense droplets emerge through counterion-mediated shielding—and clarifies how these forces collectively influence droplet growth and surface potential. Charge distribution analysis revealed distinct ionic sensitivity between charge polarities (Figs. 5 and S13). Increasing NaCl from 20 to 150 mM (at fixed 5 mM Mg²⁺) reduced droplet number (Figs. 5A and S14) and size (Figs. 5D and S14), while elevating dilute-phase DNA (0.13 ± 0.10 nM to 3.22 ± 0.20 nM; Fig. S13D), indicating suppressed binding and impaired droplets growth (Figs. S13 and S14). Bimodal ζ potentials persisted from 20 to 100 mM NaCl. Notably, the proportion of positively charged droplets increased with NaCl concentration (Fig. 5C). At 150 mM NaCl, droplets became predominantly positive (ζ = +6.3 ± 2.7 mV), with negatives nearly absent (2.4%; Fig. 5A–D). This asymmetry suggests negative potentials rely on long-range counterion correlations, while positive potentials originate from localized charge patches resistant to screening. Separation buffers contained 1.0 mM MgCl₂ for the minimization of EOF and the enhancement of droplet resolution at high salt concentrations, with no observed effect on ζ potential distributions (Figs. 5A, 4C). Synergistic effects of Na⁺ and Mg²⁺ on the generated positive ζ potential of hcGAS-DNA droplets via msFES. A Bimodal Gaussian fitting of droplet count distribution across ζ potential intervals at varying concentrations of Na⁺ as indicated. n = the total number of droplets. Data were pooled from 3 biological replicates. Samples were prepared in the buffer of 20 mM Tris-HCl (pH 7.4) + 5 mM MgCl2 + 0–150 mM NaCl and included 10 nM TMR-ds120 and 100 nM hcGAS. 1 × TG + 1.0 mM MgCl2 was used as electrophoretic buffer for msFES analysis. B The number of hcGAS-DNA droplets at varying concentrations of Na⁺ as indicated (mean ± SD, from 3 biological replicates). p-value = 0.0037, 0.0029, 0.0001, 0.30, 0.0005, 0.054, 0.0009, 0.0013, respectively. C Positively charged droplets fraction at varying concentrations of Na⁺ as indicated (mean ± SD, from 3 biological replicates). p-value = 0.0025, 0.00003, 0.0004, 0.0001, respectively. D Mean fluorescence intensity of positively and negatively charged droplet peaks at varying concentrations of Na⁺ as indicated. Mean ± SD of 3 replicate averages. p-value = 0.0004, 0.023, 0.0004, 0.0001, 0.014, 0.095, 0.0086, 0.0010, respectively. E Bimodal Gaussian fitting of droplet count distribution across ζ potential intervals at varying concentrations of Mg²⁺ as indicated. n = the total number of droplets. Data were pooled from 3 biological replicates. Samples were prepared in the buffer of 20 mM Tris-HCl (pH 7.4) + 20 mM NaCl and 0–5 mM MgCl₂ and included 10 nM TMR-ds120 and 100 nM hcGAS. 1 × TG + 0.1 mM MgCl2 was used as electrophoretic buffer for msFES analysis. F The number of hcGAS-DNA droplets at varying concentrations of Mg²⁺ as indicated (mean ± SD, from 3 biological replicates). p-value = 0.0033, 0.014, 0.0022, 0.0094, 0.00005, 0.0011, respectively. G Positively charged droplets fraction at varying concentrations of Mg²⁺ as indicated (mean ± SD, from 3 biological replicates). p-value = 0.090, 0.065, 0.0030, respectively. H Mean fluorescence intensity of positively and negatively charged droplet peaks at varying concentrations of Mg²⁺ as indicated (mean ± SD, from 3 biological replicates). p-value = 0.0064, 0.0026, 0.0003, 0.020, 0.0003, 0.0044, respectively. I Bimodal Gaussian fitting of hcGAS-DNA droplet count distribution across ζ potential intervals under physiological conditions. n = the total number of droplets. Data were pooled from 3 biological replicates. 1 × TG + 1.0 mM MgCl2 was used as electrophoretic buffer for msFES analysis. Total physiological ion concentrations: 10 mM NaCl, 140 mM KCl, 5 mM MgCl₂, 1 mM CaCl₂; free physiological ion concentrations: 10 mM NaCl, 140 mM KCl, 0.5 mM MgCl₂, 100 nM CaCl₂. J The number of droplets under physiological conditions (mean ± SD, from 3 biological replicates). p-value = 0.035, 0.0088, respectively. K Positively charged droplets fraction under physiological conditions (mean ± SD, from 3 biological replicates). p-value = 0.0038. L Mean fluorescence intensity of positively and negatively charged droplet peaks under two different physiological conditions. Mean ± SD of 3 replicate averages. p-value = 0.90, 0.11, respectively. Two-sample t-test assuming equal variances, two-tailed. Not significant (n.s.) p > 0.05, *p ≤ 0.05, **p ≤ 0.01, ***p < 0.001 (B–D, F, G, H, J–L). Source data are all provided as a Source Data file. We next assessed non-electrostatic contributions using 1,6-hexanediol (1,6-HD; 0–8%), which disrupts weak hydrophobic interfaces82–84. Concentrations ≤5% had no effect (msFES: Fig. S15A; fluorescence imaging: Fig. S15B), but 8% 1,6-HD reduced mean droplet intensity by 66% (Fig. S15C) and diameter by 44% (1.30–0.73 μm; Fig. S15D), indicating that hydrophobic interactions are dispensable for nucleation but essential for maturation and fusion. An emergent microdroplet population at high 1,6-HD suggests fission-dominated remodeling. Urea (1–5 M), a hydrogen-bond disruptor, showed no effect (Fig. S16). These results establish electrostatics as the primary driver for nucleation and growth of biphasic hcGAS–dsDNA droplets, with hydrophobic forces playing a secondary role in maturation and structural integrity. Synergistic effects of Na⁺ and Mg²⁺ on positively charged hcGAS–DNA droplet formation The ionic environment critically regulates charge polarization in hcGAS–DNA condensates. In the absence of Na⁺ or Mg²⁺, positively charged droplets were nearly absent, and negatively charged species (ζ = −45 to 0 mV) predominated (Fig. 5A, E). Under Mg²⁺-deficient conditions, droplets exhibited smaller diameters in msFES analysis (Fig. 5H and S13B). Among the Mg²⁺ concentrations tested (0–5 mM), higher levels (5 mM) increased the number (Fig. 5E–G) and formation (Fig. 5H) of positively charged droplets. These results indicate that Na⁺ and Mg²⁺ act synergistically to promote the formation and support the growth of positively charged droplets. Stoichiometric bifurcation governs electrostatic polarization of hcGAS–DNA droplets Quantitative two-color fluorescence imaging revealed significant NaCl-dependent compositional remodeling within LLPS droplets (Fig. S17A). In the absence of either Na⁺ or Mg²⁺ ions, the dsDNA/hcGAS fluorescence intensity ratio remained high (mean ratio values: 0.95–0.99) and followed an approximately unimodal normal distribution (Fig. S17B). In the presence of Mg2+, as NaCl concentration increased up to 150 mM, the ratio significantly decreased (0.53 at 150 mM NaCl vs. 0.95 at 0 mM NaCl control; p < 0.001), while retaining a unimodal distribution (Fig. S17B). This stoichiometric shift, coinciding with the charge polarity inversion observed via ζ potential (positive ζ dominance at high ionic strength, Fig. 5A, C), suggests that cation-enriched droplets adopt hcGAS-dominant architectures. Thus, elevated ionic strength (≥ 100 mM) triggers a phase separation mode switch: negatively charged droplets destabilize due to electrostatic screening, while positive ζ assemblies maintain structural integrity through non-electrostatic interactions, accounting for the ionic resilience of cationic condensates. Dual-channel fluorescence imaging further uncovered an ion-dependent stoichiometric bifurcation. In systems containing both Na⁺ and Mg²⁺ across the 20–100 mM Na⁺ range, the dsDNA/hcGAS ratio exhibited bimodal distributions (R² > 0.85 for Gaussian fit; Fig. S17B, C). A minor subpopulation mirrored the ratio profile seen under high-salt conditions (150 mM Na⁺ + 5 mM Mg²⁺; ratio 0.20–0.70; Gaussian α = 0.05; Fig. S17B, D), consistent with the charge bifurcation detected by msFES (Fig. 5C). Next, we quantified spatial arrangement via the DNA-to-cGAS intensity ratio—positively charged droplets showed lower ratios ( ~ 0.36–0.65, blue), while negatively charged droplets exhibited higher ratios ( ~ 0.7–1.5, red) (Fig. S18). Imaging analysis indicated that oppositely charged droplets are spatially segregated (average intermixing distance >0.63 μm), whereas droplets of the same charge type tend to cluster. Fusion is consistently rapid, as illustrated in Fig. S19A. Correspondingly, we observed that positively charged (DNA:cGAS = 0.7) and negatively charged (DNA:cGAS = 1.18) droplets underwent complete fusion within approximately 1 min (Fig. S19B). These results suggest that oppositely charged droplets undergo fusion upon close contact and subsequently re-disperse into stable, charge-homogeneous domains. In-cell msFES reveals the predominance of positively charged hcGAS–DNA condensates msFES measurements under cytosolic-like ionic conditions showed that positively charged droplets accounted for 36 ± 2% of the population under low free Mg2+ and Ca²⁺ conditions (10 mM NaCl, 140 mM KCl, 0.5 mM MgCl₂, 100 nM CaCl₂), increasing to 79 ± 9% at physiological ionic strength (10 mM NaCl, 140 mM KCl, 5 mM MgCl₂, 1 mM CaCl₂) (Figs. S13C and 5I–K), whereas both ionic conditions exhibited similar droplet number (Fig. 5J), size (Fig. 5L) and dilute-phase DNA concentration (Fig. S13E) to 100 mM NaCl conditions. This was corroborated by a bimodal DNA/hcGAS fluorescence ratio in imaging data (Fig. S17B), supporting the prevalence of positively charged hcGAS–dsDNA condensates under physiological conditions. To validate that hcGAS–DNA condensates are predominantly positively charged in cells, we performed live-cell imaging in HEK293T cells overexpressing hcGAS-mCherry (isolated by fluorescence-activated cell sorting, Fig. S20) and transfected with ds120 (0.5–2.5 μg/mL; Fig. 6A). Gaussian fitting (R² = 0.74) of the FAM/mCherry fluorescence ratio within intracellular condensates revealed two distinct populations: dsDNA-enriched and hcGAS-enriched subpopulations (Fig. 6B, C). This bimodality mirrors the distributions observed under simulated physiological conditions both in vitro (Fig. S17) and with msFES (Fig. 5). We propose that hcGAS-enriched droplets, with a fluorescence ratio below 0.9 (left Gaussian fit, p < 0.05), represent positively charged condensates and account for ∼68% of the population, while dsDNA-enriched (likely negative) droplets constitute ∼32%. Intracellular hcGAS-DNA droplets tend toward positively charged under sufficient DNA stimulation. A Fluorescence imaging of intracellular hcGAS-dsDNA droplets. Exogenous ds120 DNA solution (total DNA concentration: 0–2.5 μg/mL) were prepared with FAM-labeled ds120 (0–0.5 μg/mL) and unlabeled ds120 (0–2.0 μg/mL) and transfected into HEK293T cells overexpressing hcGAS-mCherry. Yellow markers indicate colocalization of hcGAS and DNA, while green markers indicate that only DNA was observed in cells. Each experiment was repeated 3 times independently with similar results. The scale bar: 10 μm. B Analysis of fluorescence signal ratios of FAM-labeled DNA to hcGAS-mCherry in colocalized foci. The green curve represents the DNA fluorescence intensity, and the red curve represents the hcGAS fluorescence intensity. The scale bar: 1 μm. C Bimodal Gaussian fit of the fluorescence intensity ratio of hcGAS-DNA droplets corresponding to (A) (unlysed cell; 73 droplets from 3 biological replicates). D Bimodal Gaussian fit of the fluorescence intensity ratio of hcGAS-DNA droplets (lysed cell; 125 droplets from 3 biological replicates). Lysates of the cells overexpressing hcGAS-mCherry were prepared from 5 × 10⁵ cells in 20 μL culture medium with 1% Triton X-100. E Bimodal Gaussian fitting of the distribution of number of droplet peaks and the distribution of summed fluorescence intensity across ζ potential intervals detected (total droplets number 4, 5, 22, 23, respectively, pooled from 3 biological replicates). Source data are all provided as a Source Data file. Cell lysis with 1% Triton X-100 (which does not disrupt hcGAS–dsDNA condensates; Figs. S21 and 6D) enabled separate msFES analysis of hcGAS-mCherry signals. msFES confirmed the presence of both positively and negatively charged droplets (Fig. 6E), with a higher frequency of positive droplets after transfection with 2.5 μg/mL ds120. Analysis of droplet counts and peak heights indicated a shift toward positive charge dominance with increasing DNA concentration (Fig. 6E), suggesting that DNA promotes hcGAS binding and subsequent positive condensate formation. Using detergent-free sonication lysis, we confirmed the coexistence of two electrostatically distinct droplet populations with opposite charges (Fig. S22): a positively charged population (centered at ζ ≈ +17.2 mV) and a negatively charged population (centered at ζ ≈ −20.8 mV). The positive population accounted for approximately 59% of the detected droplets (vs ~ 64% Triton X-100 lysed, p > 0.05) (Fig. S22C). Dynamic shift of the hcGAS-dsDNA droplets from negative to positive ζ potential in response to TREX1 cleavage To investigate how enzymatic activity influences condensate electrostatics, we incorporated TREX18,9, a 3′-exonuclease known to degrade cytoplasmic DNA (Fig. 7B) but was selectively excluded from the periphery of hcGAS-dsDNA droplets (Fig. S23), into the msFES platform. This revealed charge-dependent susceptibility as a central regulatory mechanism. Over the first 20 min of TREX1 treatment, positively charged droplets showed an increase in fluorescence intensity (Figs. S24B and 7F), indicating resistance to DNA degradation. In contrast, negatively charged droplets decreased 87% in total fluorescence intensity (Fig. 7F). These results establish surface electrostatics as a tunable parameter that governs condensate stability under enzymatic challenge. DNA in positively charged hcGAS-DNA droplets exhibits resistance to TREX1 digestion. A Schematic representation of wild-type (WT) hcGAS and tumor-associated mutants (G303E, K432T). B Gel electrophoresis analysis of TREX1-mediated dsDNA cleavage over time (12% polyacrylamide gel). Samples were prepared in a buffer containing 20 mM Tris-HCl (pH 7.4), 20 mM NaCl, and 5 mM MgCl₂, with 10 nM TMR-ds120 and 100 nM hcGAS, and incubated for 10 min prior to the addition of 10 nM TREX1 for 0–60 min. C Quantification of degraded DNA by gel electrophoresis (mean ± SD, from 3 biological replicates). D, E Quantification of degraded DNA by msFES (mean ± SD, from 3 biological replicates). Samples were prepared the same as in (B) and analyzed in an electrophoretic buffer consisting of 1 × TG + 0.1 mM MgCl₂. Product 2 (D) and 3 (E) are two truncated DNA fragments corresponding to Fig. S24A–D. F The sum of fluorescence intensity of WT, G303E, and K432T droplets across ζ potential intervals with Gaussian fitting corresponding to Fig. S24A–D (pooled data, from 3 biological replicates). n = the total number of droplets. Source data are all provided as a Source Data file. Pathogenic hcGAS mutations with compromised generation of hcGAS-dsDNA droplets of positive ζ potential are more sensitive to TREX1 cleavage The hcGAS mutants K432T and G303E, which occur naturally in tumor cells, with K432T identified in endometrial carcinoma (Fig. 7A), localize to DNA-binding site C, a region containing the KRKR loop that facilitates droplet fusion during LLPS. Although both mutants undergo LLPS similarly to wild-type hcGAS and retain comparable DNA-binding affinity (Fig. S25A), fluorescence imaging revealed that they form significantly smaller condensates under identical conditions (Fig. S25B, C). Gel electrophoresis analysis showed that dsDNA within K432T and G303E condensates was more susceptible to TREX1 cleavage than that in wild-type hcGAS droplets, which exhibited slower degradation kinetics (Fig. 7B, C). msFES measurements further indicated an increase in degraded DNA in the dilute phase of both mutant LLPS systems compared to WT, with a twofold increase for DNA fragment 2 (Fig. 7D) and a 1.6-fold increase for fragment 3 (Fig. 7E). As observed with WT hcGAS, TREX1 digestion largely eliminated negatively charged droplets in both K432T and G303E mutant systems (Figs. S24C, D and 7F), with a concomitant shift toward predominantly positive ζ-potential populations. However, the enhancement in mean and total fluorescence intensity of positive droplets following TREX1 treatment was significantly weaker in both mutants compared to WT, especially at 20 min (Figs. 7F and S24E). This difference persisted throughout the 60-min digestion period. Since both mutations localize to the Site-C hcGAS–DNA interface, a region critical for multivalent interactions during condensation, we propose that their impaired ability to support the growth of positive-potential droplets underlies heightened TREX1 sensitivity (Fig. 8). These results underscore the functional importance of positive-electrostatic microenvironments in conferring enzymatic resistance and maintaining condensate integrity in vivo. The response of hcGAS-dsDNA droplets to TREX1 cleavage. Negatively charged droplets are susceptible to TREX1 cleavage, leading to their rapid degradation and partial conversion into a positively charged state. This conversion is pivotal because the resulting positive droplets are resistant to TREX1. The interplay between these susceptible and resistant populations collectively regulates digital hcGAS activation. Discussion Multi-scale characterization of biomolecular condensates in a single msFES run msFES enables multi-parametric analysis of hcGAS–dsDNA condensates from nanoliter-volume samples in a single measurement. The number of flashing droplet peaks detected by msFES corresponds directly to the number of droplets present in the sample (Fig. 2D). Using RB, an approximately neutral marker, to account for EOF, the absolute injection volume (about 10 nL) was precisely determined based on electrophoretic velocity, injection time, and applied voltage. The number of droplets within this volume was quantified by counting sharp electrophoretic peaks. Although Mg²⁺ influences the formation of protein-RNA biocondensates85 and bimodal distribution of hcGAS-DNA droplets, its low concentration in the msFES does not alter their core biophysical properties. When DNA or protein components are fluorescently labeled, msFES further provides fluorescence intensity readings for individual droplets, as well as population-averaged values that correlate with droplet size. Consistent with fluorescence imaging data, the mean fluorescence intensity of droplets increased proportionally with hcGAS concentration, reflecting enhanced droplet growth. The system also captures heterogeneity in droplet size via variations in single-droplet fluorescence intensity, allowing assessment of size distribution across the population. Moreover, msFES simultaneously quantifies the concentration of unbound dsDNA in the dilute phase through the corresponding fluorescence peak. This capability was exemplified in mutants with impaired condensation (e.g., Δ146 hcGAS R300E/K301E), where elevated dilute-phase dsDNA levels were readily detected. Importantly, msFES directly resolves the ζ potential of individual droplets, adding electrostatic characterization to its multimodal profiling capacity. Several techniques are available for characterizing the surface charge of biomolecular condensates, including laser Doppler electrophoresis (e.g., Malvern Ζetasizer)46–48, microelectrophoresis49,50, μFFE45, and graphene-based FET sensors53. Among these, our msFES-LIF method offers distinct advantages in minimal sample consumption (nL to low μL volumes), high resolution and multiplexing (simultaneous resolution of single positively and negatively charged droplets in a single run), high throughput (rapid analysis of large numbers of droplets within minutes), and operational simplicity. In comparison, while laser Doppler electrophoresis requires no fluorescent labeling, it typically demands larger sample volumes and provides only an ensemble-average ζ potential. Microelectrophoresis, which tracks individual droplet motion, is suitable for homogeneous systems but faces challenges with heterogeneous droplet mixtures. μFFE allows continuous high-throughput analysis but also generally requires larger sample volumes. In summary, msFES represents an integrated platform for quantitative, multi-scale analysis of biomolecular condensates at near single-droplet resolution, delivering concurrent measurement of droplet number, size distribution, ζ potential, and dilute-phase solute concentration in a single run. Equation derived for calculation of ζ potential For ζ potential determination of hcGAS-dsDNA droplets in msFES-LIF analysis (Eq. (8) in Supporting Information), the hcGAS-dsDNA droplets have much higher viscosity than the solution7, the effects of droplet viscosity, radius of droplets, and Debye length on measurement of ζ potential can be neglected50. For liquid droplets with lower viscosity (< 1.0 Pa·s), however, the more complete formula recommended in the recent study by Van Haren et al. should be used to calculate the ζ values50. Bimodal ζ potentials and heterogeneity of hcGAS–DNA condensates Our results definitively demonstrate that hcGAS–DNA condensates can be categorized into two distinct subpopulations based on ζ potential: positively and negatively charged droplets. The bimodal distribution observed indicates the coexistence of droplets with strongly positive or negative surface charges, reflecting underlying differences in stability and composition. Nearly neutral droplets were negligible, underscoring the functional relevance of electrostatic character in condensate formation and integrity. Notably, these two distinct electrostatic states arise from identical components under identical conditions. This discovery challenges the prevailing view of condensate uniformity and reveals an unanticipated layer of complexity in phase-separated systems. Multiple intermolecular forces, including electrostatic, hydrophobic, and hydrogen-bonding interactions, collectively contribute to droplet nucleation, growth, and the emergence of charge bifurcation. Notably, synergistic effects of Na⁺ and Mg²⁺ were essential for promoting positively charged condensates (Figs. 5 and S13). These cationic droplets exhibited superior stability under high ionic strength, whereas negative droplets dissociated due to charge screening. Intracellularly, increasing DNA concentration promoted a shift toward positively charged droplets (Fig. 6), indicating that physiological conditions favor electropositive condensates with enhanced bioactivity. Modulation of bimodal charge distribution potentially enable tunable hcGAS activity The coexistence of positive and negative droplets may serve as a buffering mechanism for DNA-stimulated hcGAS signaling. Despite both populations possessing the capacity for cGAMP synthesis, we propose that this charge-based regulatory circuit likely fine-tunes immune activation: positively charged droplets (with higher hcGAS ratio) resisted TREX1 cleavage and amplified the hcGAS-mediated immune response, whereas negatively charged droplets (with higher DNA ratio) are rapidly attenuated by TREX1 to suppress excessive activity (Fig. 7). Consistent with this model, positive droplets exhibited higher hcGAS-to-dsDNA ratios. Coincidently, Van Haren et al. demonstrated that the charge and stability of coacervates formed by poly-L-lysine and poly-L-aspartic acid depend on the specific lengths and mixing ratios of these polymers50. Notably, TREX1 can also digest DNA in positively charged droplets, given sufficient time. To the best of our knowledge, these findings establish the first direct link between condensate surface charge and nuclease accessibility in the context of innate immune regulation. This observation raises the possibility that electrostatic heterogeneity may influence the duration or localization of immune signaling by modulating the protection of nucleic acids within condensates. Pathological mutations disrupt charge-based regulation TREX1 digestion markedly increased positive droplet abundance in WT hcGAS condensates, but not in K432T or G303E mutants (Fig. 7). While both the K432T and G303E mutants, as well as wild-type cGAS, can form two populations of condensates (negatively and positively charged droplets), the mutants exhibit a markedly smaller population of positively charged droplets (Fig. 7F). This impairment likely stems from the reported reduction in cGAS–dsDNA interaction for the K432T mutant6. The defective growth of positive droplets suggests a mechanism for the heightened TREX1 sensitivity and dysregulated immune activity in these tumor-associated variants. Acting in concert with chromosomal instability86,87, these mutations may consequently influence tumor metastasis and therapy response. The observation that pathogenic cGAS mutants disrupt this electrostatic control provides a potential disease-relevant context and suggests that charge regulation may have physiological or pathological consequences, even if the precise functional readout remains to be fully defined. Future perspective and applications The msFES platform offers a powerful approach for multiparametric LLPS droplet analysis at single-droplet resolution. Its ability to resolve subpopulations by size, charge, and composition, while simultaneously quantifying dilute-phase biomolecules, enables unprecedented insight into condensate heterogeneity and dynamics. Beyond the present study, msFES is readily adaptable to other phase-separating systems, including those involved in neurodegenerative disease and cancer, offering broad potential for deciphering pathological LLPS mechanisms. Future applications may extend to intracellular separation and molecular profiling of specific droplet subtypes based on their electrokinetic properties. Notably, our observation that electrostatic heterogeneity correlates with differential TREX1 sensitivity, presented here as a hypothesis-generating finding, opens avenues for investigation, particularly regarding whether and how charge heterogeneity modulates cGAS-mediated signaling in physiological and pathological contexts. In summary, we have developed msFES—a high-resolution separation platform that enables multi-parametric characterization of biomolecular condensates at near-single-droplet resolution—and applied it to reveal a previously unrecognized bimodal distribution of ζ potential in hcGAS–dsDNA droplets. We demonstrate that positively charged droplets exhibit enhanced stability and resistance to TREX1-mediated cleavage, whereas negatively charged droplets are preferentially disassembled. Pathogenic cGAS mutations disrupt this electrostatic regulation, rendering condensates hypersensitive to TREX1-mediated DNA degradation. These findings establish surface potential as a key regulatory feature of LLPS systems and challenge the prevailing view of condensate uniformity by revealing an intrinsic electrostatic heterogeneity. The vulnerability of this electrostatic control to disease-associated mutations provides a potential context for understanding charge-related dysfunction in pathology. Beyond these biological insights, msFES emerges not merely as a sensitive detection tool but as a platform capable of uncovering hidden regulatory principles in biomolecular condensate biology, with broad applicability across diverse phase-separating systems. Methods This study did not involve human subjects, animal experiments, or any materials requiring specific ethical approval. All experiments complied with the institutional guidelines of the Research Ethics Committee of the Chinese Academy of Sciences. Chemicals and reagents All oligodeoxynucleotides were synthesized and purified by Shanghai Sangon Biological Engineering Technology & Services Co, Ltd (Shanghai, China). The double-stranded DNA (dsDNA, named as ds120), consisting of a top strand (5′-TTACACGCTTCCTCAACCACTTACCATACTCGAGATTACGCCAGCAAAGGCGTGTTTGAGCGCGTTCACTCTGTTAAAAGGAAATCCGCCCCCGCCCCGTTTCCTTCCTCCGCGATACAA-3′, ss120) and a complementary bottom strand, was obtained through gel purification after annealing. Fluorescently labeled ds120 was prepared by using the ss120 with fluorophore-labeled (Fluorescein amidite (FAM), Tetramethylrhodamine (TMR), or Cyanine 5 (Cy5)) at the 5′-end. cGAS protein and TREX1 protein were expressed and purified in our lab. Ultrapure water of 18.2 M.Ω cm was obtained from a Purelab Ultra Bioscience system (ELGA, UK). 1,6-HD (≥ 98%) was purchased by Aladdin Biochemical Technology Co., Ltd. (Shanghai, China). The HEK293T cell line was obtained from the Cell Bank/Stem Cell Bank, Chinese Academy of Sciences (Shanghai, China), and was authenticated by short tandem repeat profiling as well as tested negative for mycoplasma contamination. Sample preparation The mixture of all samples normally included 20 mM Tris-HCl, pH 7.4, 20 mM NaCl, 5 mM MgCl2, 50 μg/mL BSA, and 5% glycerol. The wild-type or mutant hcGAS protein (concentration ranging from 0 to 1000 nM) and TMR-ds120 (2.5 to 10 nM) were added depending on the msFES-LIF experiments. The samples were reacted at 37 °C for 10 min. At the end of the reaction, if the samples were used for msFES-LIF analysis, RB (2 nM) was added as internal reference. When the samples were used for 12% polyacrylamide gel electrophoresis, the mixture included 40 nM TMR-ds120, and 0–1 μM wild-type or mutant hcGAS protein. Methods for cell culture, gene overexpression, and cell sorting are provided in the supporting information. dsDNA (0–2.5 μg/mL, containing 20% FAM-labeled ds120 and 80% unlabeled ds120) was transfected into 5 × 105 cells per 2 mL of medium. Cells were trypsinized, centrifuged (1100 × g, 1 min), and resuspended in residual medium ( ~ 20 μL). Cell lysis was performed by incubating the resuspended cell pellet with 1% Triton ×-100 for 10 min at 25 °C. The lysate was then centrifuged at 1100 × g for 10 s, and the resulting supernatant with 2 nM RB was immediately subjected to msFES analysis. msFES-LIF analysis msFES-LIF apparatus was laboratory-built. The excitation wavelength was 543.5 nm. RB and TMR fluorescence were detected at 575 ± 7.5 nm, while mCherry fluorescence was measured at 624 ± 20 nm. Uncoated fused-silica capillary (25 μm i.d., 365 μm o.d.) with a total capillary length of 28 cm, and an effective length of 21.5 cm, was activated with 0.1 M NaOH for 3 h and treated with 1.0× running buffer (25 mM Tris, 192 mM glycine, pH 8.3) for 2 h. The pretreated samples were injected at 15 kV for 5 s and separated using 1.0× running buffer containing 0.5 mM MgCl2 at 20 kV. After each separation, the capillary was rinsed with 20 mM NaOH for 2 min, H2O for 1 min, and 1.0× running buffer containing 0.1–1.0 mM MgCl2 for 3 min. Before using msFES-LIF daily, the capillary was rinsed with 20 mM NaOH for 10 min, H2O for 5 min, and 1.0× running buffer containing 0.1–1.0 mM MgCl2 for 10 min. All operations were performed at room temperature with humidity no higher than 50%. RB (2 nM) and TMR dyes (1 nM) were used as internal references. Electrophoretic profile data were collected using HW2000 (Qianpu, China). Fluorescence microscope imaging The fluorescence imaging assay in vitro was performed on a self-built objective-based total internal reflection fluorescence (TIRF) microscope (Olympus). The capillary loaded on the TIRF microscope (100×, oil objective). The laser light (Coherent Inc., USA) applied to the TIRF microscope was set at 488 nm (80 mW) with a filter of 525/20 nm, 532 nm (80 mW) with a filter of 585/40 nm, and 640 nm (100 mW) with a filter of 692/40 nm. The emitted fluorescence signal was detected by a sCMOS camera (Prime 95B 22MM) (Teledyne Photometrics, USA), with which the images were obtained and processed using Visview (Germany) splitter software. Image processing was performed using ImageJ (USA). Cell imaging was performed using a high-resolution confocal fluorescence lifetime microscope (Leica Microsystems CMS GmbH, model DMi8 automated) with a 63× oil objective. The live-cell nuclear dye Hoechst was excited at 405 nm with fluorescence detection at 450–500 nm. FAM analysis used 488 nm excitation with detection filters at 500–550 nm, and the detection of mCherry fluorescent protein employed 590 nm excitation with detection filters at 600–650 nm. Cell imaging was performed using 8-well chambered cover glasses (No. 1.5 thickness; 0.13–0.16 mm). The acquisition software is LAS X (Germany). Image processing was performed using ImageJ (USA). Statistics and reproducibility All experiments were performed with 3 biological replicates. For statistical significance of the 3 biological replicates, a two‑sample equal‑variance two‑tailed Student’s t‑test was generally used. For statistics involving information from individual droplets, the total number of droplets (n) obtained from the 3 biological replicates was used as the sample size; a two‑sample F‑test for unequal variance followed by a two‑sample unequal‑variance two‑tailed t‑test was generally applied. Not significant: p‑value > 0.05; *p ≤ 0.05; **p ≤ 0.01; ***p ≤ 0.001. No data were excluded from the analyses. The experiments were not randomized. The investigators were not blinded to allocation during experiments and outcome assessment. Data plotting was performed using Origin 2021 (USA). Reporting summary Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article. Supplementary information Supplementary Information Reporting Summary Transparent Peer Review file Source data Source Data Acknowledgements This work was supported by the grants from the National Key Research and Development Program of China (2024YFA0918803), the National Natural Science Foundation of China (22234008 and 22274166), and the Strategic Priority Research Program of the Chinese Academy of Sciences (XDB0750100). Author contributions W.H. conceived project. W.H., L.X., Z.Q. & L.C. designed the experiments. W.H. & Z.Q. supervised the experiments and data analysis. L.X. performed the experiments and collected data. S.X. provided technical expertise in cell biology. Z.D. and L.W. offered technical support. J.G. supplied critical instrumentation. W.H., L.X., and Z.Q. wrote manuscript. W.H., L.C., and Z.Q. reviewed and edited the paper with the input of all coauthors. Peer review Peer review information Nature Communications thanks Merlijn van Haren and the other anonymous reviewers for their contribution to the peer review of this work. A peer review file is available. Data availability The authors declare that data supporting the findings of this study are available within this article and its supplementary information files. All data underlying this study are available from the corresponding author upon request. Source data are provided with this paper. Competing interests The authors have no competing interests as defined by Nature Portfolio, or other interests that might be perceived to influence the results and/or discussion reported in this paper. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Contributor Information Qiang Zhao, Email: qiangzhao@rcees.ac.cn. Hailin Wang, Email: hlwang@rcees.ac.cn. Supplementary information The online version contains supplementary material available at 10.1038/s41467-026-73740-7. References Associated Data Supplementary Materials Supplementary Information Reporting Summary Transparent Peer Review file Source Data Data Availability Statement The authors declare that data supporting the findings of this study are available within this article and its supplementary information files. All data underlying this study are available from the corresponding author upon request. Source data are provided with this paper.