Modular microfluidics enables kinetic insight from time-resolved cryo-EM

The study addresses how to capture transient, functionally relevant structural intermediates of macromolecules with both spatial and temporal resolution. While cryo-EM can resolve multiple conformational states at high resolution, it does not directly reveal the temporal sequence of states or enrich short-lived intermediates under equilibrium conditions. Existing time-resolved approaches are limited to slow processes or temperature-sensitive systems and often rely on labels, mutants, or crosslinking that can introduce artefacts. The authors aim to develop a generally applicable, label-free, time-resolved cryo-EM (trEM) sample preparation workflow that rapidly mixes reactants, synchronizes reactions on millisecond timescales, and vitrifies samples without blotting to preserve native intermediates and improve sample integrity.

Prior work demonstrated that cryo-EM can separate multiple states within a sample but lacks direct temporal ordering, and short-lived intermediates remain underrepresented at equilibrium. Early time-resolved structural studies in electron crystallography diffused ligands into crystals, akin to time-resolved X-ray crystallography, but were limited to crystalline samples and small ligands. Attempts within single-particle cryo-EM followed slow processes (seconds to minutes) compatible with conventional blotting and freezing or relied on temperature shifts to trap states. Microfluidic mixing combined with gas-assisted spraying has been explored and yielded insights, yet challenges persist: complex fabrication and operation, insufficient characterization of gas-assisted aerosol generation, uncertain millisecond mixing efficiency, and unknown errors in incubation times within microfluidic delay lines. These gaps hinder well-controlled trEM comparable to time-resolved X-ray methods.

The authors developed an integrated, modular trEM platform enabling rapid, blot-free vitrification with controlled reaction initiation and incubation: - Microfluidic 3D mixer: An in situ three-dimensional passive mixer with sequential junctions induces secondary flows and chaotic advection, thinning striations to accelerate diffusion and achieve near-complete mixing. Mixing was quantified by confocal microscopy using two fluorophores (FITC, Rhodamine 6G) at various flow rates and numbers of junctions. Optimal conditions selected: 333 µl/min per inlet (total 666 µl/min), seven 3D mixing junctions, yielding a calculated mixing time of ~3.1 ms based on measured channel dimensions and pump performance. - Gas-assisted spraying nozzle: A modular nozzle with a 100-µm liquid orifice and concentric N2 gas stream atomizes the mixed sample into droplets directed at the grid. Gas pressure was tuned; 0.8 bar N2 provided small, sufficiently uniform droplets and high droplet velocities without grid damage. High-speed imaging quantified droplet size and velocity versus gas pressure. - Rapid plunge freezing: Grids held in tweezers mounted on a servo motor were radially plunged into liquid ethane. An optimized plunge speed of ~1.1 m/s and nozzle-to-grid distance of 8 mm consistently produced thin vitrified ice suitable for single-particle analysis. Time metrics from high-speed video: ~2.7–3 ms grid integration time; <2 ms spray flight and vitrification; combined with ~3 ms mixing and transport, total dead time ~30 ms from mixing to vitrification. The control electronics (Arduino-based) and custom software synchronized pump, gas valve, and plunging to minimize sample consumption (~20–30 µl per grid after pre-filling lines). - Delay-line incubation: Residence time—and thus reaction time—was set by the channel length downstream of the mixer. A series of PDMS microfluidic devices with varied incubation path lengths provided nominal times from ~10 to ~1330 ms at fixed flow. Residence time distributions (RTDs) were estimated via COMSOL CFD and particle tracking, considering laminar flow (Re ~100), drag, and Brownian motion. - Sample assessment: The platform was tested on various grids (holey Quantifoil, Quantifoil with ultrathin carbon, lacey with ultrathin carbon) and samples (apoferritin, 20S proteasome, CSN5H138A–SCF–N8Skp2/Cks1 complex). Tomography quantified particle distributions across ice thickness. High-resolution single-particle datasets were collected on a Titan Krios (300 kV, FalconIII) and processed with MotionCor2, CTFFIND4, crYOLO, RELION/Scipion; CSN 2D classification was done on a Talos Arctica with cryoSPARC. Mixing efficiency and adsorption in PDMS were quantified; PDMS adsorption over 1 s was measured for RecA (protein, 19%), ssDNA (29%), and ATP (42%). - Kinetic test system: RecA filament growth on ssDNA with ATPγS was chosen as a model. Manual mixing (minutes) yielded equilibrium length distributions. trEM with delay lines (e.g., nominal ~20 ms, and a time course spanning 20–1000 ms) captured pre-steady state filaments whose lengths reflect incubation time. Filament lengths were measured from cryo-EM micrographs (FIJI) to extract growth kinetics across replicates.

- Rapid, reliable mixing and delivery: - Near-complete mixing achieved in ~3.1 ms using seven 3D junctions at 333 µl/min per channel (total 666 µl/min), validated by confocal imaging. - Gas-assisted spraying at 0.8 bar N2 produced small, fast droplets; droplet velocity increased with gas pressure. Optimized plunging at ~1.1 m/s and 8 mm nozzle-to-grid distance yielded consistent thin ice. - Total dead time from mixing to vitrification ~30 ms; grid integration time ~2.7–3 ms; spray flight plus vitrification <2 ms. - Sample quality and resolution: - Apoferritin prepared by trEM yielded near-atomic resolution (2.77 Å) comparable to standard Vitrobot prep (2.89 Å), with similar particle counts and processing outcomes. - Tomography revealed markedly more uniform z-distribution of particles in trEM-prepared ice versus biphasic accumulation at air–water interfaces in Vitrobot-prepared samples. - Sensitive CSN5H138A–SCF–N8Skp2/Cks1 complex, which aggregates on standard holey grids, appeared monodisperse over holes with trEM and produced intact 2D classes. - Across >20 grids per type, many grid squares exhibited thin ice, typically 30–75 squares with ~600–800 collectable holes each. - Kinetic trapping and analysis: - trEM captured pre-steady state RecA-ssDNA filaments at nominal ~20 ms with significantly shorter and narrower length distributions than equilibrium (manual) prep. - Time-course (20–1000 ms; multiple replicates per time point) produced a linear growth curve with an average growth rate ~20 nt/s (assuming 3 nt per RecA monomer and 1.53 nm per monomer), in close agreement with prior biophysical measurements over far longer timescales; linear fit R² ≈ 0.973. - RTD analysis via CFD and particle tracking matched observed dispersion in filament lengths; laminar flow-induced residence time scattering explains longer outliers, while slow nucleation explains shorter ones. - Practicality and versatility: - The modular PDMS device and nozzle are simple to fabricate without clean-room facilities; electronics and software coordinate operations and reduce sample consumption to ~20–30 µl per grid after pre-fill. - Compatible with common grid types; humidified chamber and voltage-assisted spraying were not required to achieve high-quality micrographs.

The platform addresses the central challenge of synchronizing and trapping transient biochemical intermediates for cryo-EM without labels or perturbative stabilization. By combining millisecond 3D microfluidic mixing, controlled delay-line incubation, and blot-free gas-assisted spraying with rapid plunging, the method achieves ~30 ms dead time and sub-second time resolution over multiple orders of magnitude (10–1000+ ms). The improved, uniform particle distribution reduces deleterious air–water interface interactions, enhancing structural integrity for fragile complexes and mitigating preferred orientation. The RecA model reaction demonstrates that structural snapshots alone can yield kinetic parameters consistent with established biophysical data, validating temporal fidelity. CFD-informed understanding of residence time distributions clarifies the dominant sources of temporal dispersion, guiding future device optimizations and potential computational deconvolution. Overall, the integrated workflow is reproducible, versatile, and broadly applicable for time-resolved studies of diverse macromolecular processes.

This work introduces a robust, modular trEM sample preparation system that enables label-free kinetic insight from structural snapshots. It delivers near-atomic spatial resolution with millisecond temporal control, improves sample integrity via blot-free rapid vitrification, and quantitatively captures pre-steady state intermediates. The platform’s simplicity and accessibility (PDMS fabrication, standard components, shared designs/code) facilitate adoption. Future directions include reducing sample consumption (e.g., injection valves, gas-dynamic virtual nozzles), refining microfluidic geometries to narrow residence time distributions or employing deconvolution based on measured RTDs, and applying the method to complex reaction pathways (e.g., CSN-mediated CRL deneddylation, homologous recombination specificity) while advancing computational tools to resolve and quantify coexisting transient states.

- Residence time distribution in laminar microfluidic delay lines causes temporal dispersion, yielding a long tail of effective incubation times and overlapping state populations. - Sample adsorption to PDMS can differentially deplete reactants (measured over 1 s: protein ~19%, ssDNA ~29%, ATP ~42%), introducing variability depending on reactant properties. - Continuous-flow operation and equilibration time before stable spraying increase sample consumption (~20–30 µl per grid after pre-fill), higher than some alternative methods. - Slow nucleation kinetics in certain reactions (e.g., RecA) introduce initiation-time variability that broadens early time-point distributions. - Analysis of pre-steady state cryo-EM data remains challenging when multiple similar intermediates coexist; advanced algorithms are needed to deconvolute and quantify populations. - Device timing accuracy is influenced by flow control and channel geometry; RTD corrections may be needed for precise kinetics.