A cryo-CLEM–guided workflow for quantitative assessment and optimization of cryo-ET specimen quality

TL;DR

This paper introduces a workflow using cryo-CLEM to improve the quality and efficiency of cryo-ET specimen preparation.

Contribution

A quantitative workflow is proposed to assess and optimize cryo-ET specimen quality using cryo-CLEM.

Findings

The Leica EM cryo-CLEM system offers higher signal-to-noise ratio and faster tile imaging.

The Thermo iFLM system provides a wider field of view, showing complementary advantages.

Fluorescence intensity correlates with lamella thickness, aiding early specimen quality assessment.

Abstract

Cryo-correlative light and electron microscopy (cryo-CLEM) is an essential technique for precisely targeting regions of interest in vitrified biological specimens for in situ cellular cryo-electron tomography (cryo-ET). Despite its widespread use, the potential of cryo-CLEM remains underexploited because its imaging performance under cryogenic conditions has not been characterized. We evaluated the imaging capabilities of 2 commercially available cryo-CLEM systems—a stand-alone Leica EM cryo-CLEM microscope and the integrated Thermo Fisher Scientific iFLM coupled to a cryo-focused ion beam platform—by quantifying their spatial resolution and signal-to-noise ratio using vitrified cellular specimens expressing green fluorescent protein (GFP)-labeled proteins. The EM cryo-CLEM setup provided a higher signal-to-noise ratio and faster acquisition via tile imaging, whereas the iFLM offered a wider field of view, revealing complementary strengths between the 2 instruments. Furthermore, we established a quantitative correlation between fluorescence intensity and lamella thickness, demonstrating that cryo-CLEM imaging can serve as a practical indicator of lamella quality during early specimen preparation, before cryo-ET data collection. Collectively, our results show that cryo-CLEM not only facilitates precise targeting but also provides quantitative feedback for optimizing lamella quality, thereby streamlining the overall cryo-ET workflow.

ga1

Figures

D](#fig0005), Supplementary Table 2). First, the PSF of each system was measured to characterize the intrinsic optical response. Both microscopes produced well–defined bead profiles across all excitation wavelengths, allowing Gaussian fitting of intensity distributions (Supplementary Fig. 1). The lateral PSF width (FWHM) at 470 nm was 0.52 µm for EM cryo-CLEM and 0.50 µm for iFLM, while the axial PSF extended to 2.32 and 2.11 µm, respectively ([Fig. 1D](#fig0005)). These results confirm the typical anisotropy of widefield optics, with slightly worse resolution than the theoretical calculation

E and F](#fig0010)). In both datasets, GFP-labeled structures appeared noisy and partially merged, characteristic of in situ widefield cryo-fluorescence imaging with limited resolution. To improve image interpretability, we applied postprocessing—EM cryo-CLEM Thunder computational clearing for EM cryo-CLEM datasets and standard deconvolution for iFLM images. After computational clearing, cellular puncta appeared sharper and more distinct, and fine structures within dense clusters became more discernible ([Fig. 2E](#fig0010)). Intensity-profile analysis revealed narrower peak widths and a marke

B](#fig0015)), with the steepest drop observed between 800 and 500 nm. This range coincides with the typical size of mammalian autophagosomes (approximately 500-1,500 nm in diameter), implying that the observed intensity loss originates primarily from the mechanical truncation of individual organelles during milling. Because each milling step effectively sections the same cellular region, the measured decrease in intensity reflects a proportional reduction in the amount of fluorescent material contained within the remaining volume of the lamella. Thus, the fluorescence decay provides a direct

F](#fig0020)), confirming a consistent correlation between fluorescence signal and lamella thickness. Based on this relationship, lamellae were categorized into 2 groups according to a defined normalized fluorescence threshold: low-intensity (thinner) and high-intensity (thicker) subsets. From each group, approximately 700 ribosome particles were randomly extracted and reconstructed independently. Subtomogram averaging of particles from low-intensity lamellae yielded a map at ∼25 Å resolution, whereas those from high-intensity lamellae reached only ∼37 Å, corresponding to a ∼1.5-fold decrease

Full text

INTRODUCTION

Cryo-electron tomography (cryo-ET) has revolutionized structural cell biology by enabling direct visualization of macromolecular assemblies within their native cellular environments at nanometer resolution (Obr et al., 2022, Turk and Baumeister, 2020). This technique bridges the gap between molecular and cellular scales, providing in situ insights into the spatial organization and dynamic architecture of complex biological systems (Pyle and Zanetti, 2021). Advances in direct electron detection, phase plate technology, and subtomogram averaging have further extended cryo-ET toward near-atomic resolution (Danev et al., 2014, Galaz-Montoya and Ludtke, 2017, Li et al., 2013, Wang and Fan, 2019). Despite these achievements, obtaining high–quality cryo-ET data remains technically demanding, as it requires the preparation of lamellae that not only meet the optimal thickness (typically 100-200 nm) for electron transparency (Sartori et al., 2007, Tuijtel et al., 2024), but also contain the precise cellular region of interest (ROI) to be imaged. Identifying these ROIs within intact vitrified cells and accurately targeting them for focused ion beam (FIB) milling demand a delicate balance of spatial precision and mechanical control, and the difficulty of reliably correlating fluorescence signals with subsequent electron imaging coordinates often leads to low yield and variable data quality—making the accurate isolation and imaging of the correct ROI one of the most persistent bottlenecks in cryo-ET workflows (Hylton and Swulius, 2021).

To overcome this targeting problem, cryo-correlative light and electron microscopy (cryo-CLEM) was developed to integrate fluorescence light microscopy with cryo-electron microscopy (cryo-EM) (Pierson et al., 2024, Tuijtel et al., 2019, van Driel et al., 2009). By combining the molecular specificity of fluorescence labeling with the ultrastructural precision of cryo-EM, cryo-CLEM enables the localization of tagged biomolecules within vitrified cells, guiding subsequent cryo-FIB milling and tomographic data acquisition (Li et al., 2023a, Li et al., 2023b, Li et al., 2023c, Yang et al., 2026). Over the past decade, cryo-CLEM has evolved from a simple navigation tool to a key workflow component that improves targeting efficiency and success rates in in situ cryo-ET. Recent studies have applied cryo-CLEM to locate rare autophagosomal intermediates, viral assembly sites, or cytoskeletal remodeling events, demonstrating its growing importance for investigating specific molecular states within heterogeneous populations (Bieber et al., 2022, Chakraborty et al., 2020, Hou et al., 2025, Sexton et al., 2022). Moreover, emerging correlative approaches, such as correlated voltage and electron tomography, now extend this concept further by integrating functional readouts—such as membrane potential dynamics—with cryo-EM imaging (Jung et al., 2025), highlighting a broader shift toward multimodal, quantitative correlative imaging.

Several commercial cryo-CLEM platforms are currently available, each reflecting a distinct design philosophy. The Zeiss cryo-CLEM combines a cryo-stage with conventional confocal optics (de Beer et al., 2023); the Leica EM cryo-CLEM functions as a stand-alone widefield cryo-fluorescence microscope equipped with a high numerical aperture (NA) objective and independent cooling system (Hampton et al., 2017); and the Thermo Fisher Scientific iFLM integrates a fluorescence module directly into a cryo-FIB–scanning electron microscope (cryo-FIB–SEM) instrument (Yang et al., 2026) (Fig. 1A, Supplementary Table 1). These configurations entail different trade-offs between sensitivity, field of view, and targeting precision (de Beer et al., 2023, Kaufmann et al., 2014, Moser et al., 2019, Yang et al., 2026). Stand-alone microscopes offer higher photon-collection efficiency and faster acquisition across large grid areas, while integrated systems enable in situ imaging during milling and direct spatial correlation between optical and ion beam coordinates (Boltje et al., 2022). Despite the increasing use of these instruments, however, their quantitative imaging performance under cryogenic conditions—such as spatial resolution, signal-to-noise ratio (SNR), and field-of-view characteristics—has not been systematically evaluated, leaving users without clear guidance for instrument selection or workflow integration.Fig. 1. Benchmarking imaging performance of 2 cryo-CLEM systems at the cryogenic condition. (A) Schematic design of EM cryo-CLEM (Leica) and iFLM (Thermo Fisher Scientific). (B) The experimental scheme of the resolution measurement using 200 nm fluorescent beads in vitrified ice. (C) The representative snapshot of 200 nm fluorescent beads on the grid for each microscope at 470 nm wavelength. Vitrified beads are zoomed in (Scale bar: 50 µm) (left) (Scale bar: 100 µm) (right) The XY, YZ central slices of the point spread function of 3 fluorescence channels of each microscope at different wavelength (EM cryo-CLEM: 405, 470, 565 nm, iFLM: 385, 470, 565 nm) are shown. (D) The measured resolution performance of each microscope. The table shows the measured average resolution for each channel. (E) Measured signal-to-noise ratio of EM cryo-CLEM and iFLM.Fig. 1

In this study, we systematically compare 2 widely adopted cryo-CLEM systems—the stand-alone EM cryo-CLEM and the integrated iFLM—to quantitatively assess their fluorescence imaging performance under vitrified conditions. We first determine their intrinsic optical resolution and evaluate postprocessing–based resolution enhancement. We then establish a quantitative relationship between fluorescence intensity and lamella thickness, demonstrating that cryo-CLEM imaging can serve as a nondestructive proxy for lamella quality before transmission electron microscope (TEM) imaging. Finally, by integrating the complementary strengths of these systems, we propose an optimized high-throughput pipeline that combines rapid grid-scale screening (EM cryo-CLEM) with in situ milling-stage feedback (iFLM), thereby streamlining the overall cryo-ET workflow. Collectively, our findings position cryo-CLEM as both a targeting and quantitative analytical tool, bridging the emerging paradigm of multimodal in situ structural biology.

MATERIAL AND METHODS

Resolution Measurement of Cryo-CLEM, and iFLM

The point spread function (PSF) of both EM cryo-CLEM (Leica) equipped with HCX PL APO 50x/0.9 CLEM objective (WD = 0.28 mm) and Leica DFC310 FX digital camera and iFLM (Thermo Fisher Scientific) equipped with 20× Zeiss Epiplan-Apochromat objective (NA = 0.7, WD = 1.3 mm, Piezo-driven) and Basler ace 2 (2A4504-5gmPRO; Sony IMX541 CMOS sensor) camera was performed according to the previous reports (Cole et al., 2011). Briefly, 200 nm TetraSpeck beads (Thermo Fisher Scientific) were diluted 2× in distilled water and loaded on R1.2/1.3 holey carbon grid (Quantifoil) and blotted back side and vitrified using Leica EM GP2 (Leica, SNU CMCI). Then, grids were loaded to EM cryo-CLEM or Aquilios 2 equipped with iFLM (Thermo Fisher Scientific, SNU CMCI), and Z-stack images were acquired using 3 channels in EM cryo-CLEM (405, 470, and 565 nm) and iFLM (385, 470, and 565 nm). Then, 5 individual fluorophores were cropped in ImageJ, and the representative images of fluorophore, PSF, and FWHM were measured using MetroloJ (Faklaris et al., 2022) and 5 FWHM was averaged in each channel, respectively.

On-Grid Culture and Cryo-Specimen Preparation

HeLa cells labeled with LC3-GFP were gifted from Dr Chungho Kim at Korea University. HeLa cells were cultured and maintained in supplemented DMEM complete medium (Gibco). R2/2 200 mesh holey carbon Au grids (Quantifoil) were glow-discharged and irradiated with UV. The grids were coated with 50 μg/ml fibronectin for 1 hour and washed with a PBS buffer. Grids were then placed onto 35 mm imaging dishes (Mattek). HeLa cells were trypsinized and seeded at a density of 0.5 × 10^5^ cells/ml on the grid and incubated for 24 hours at 37°C, 5% CO_2_ to allow adhesion of cells on the grids. Grids were then loaded on Leica EM GP2 (Leica) at 37°C, and 95% humidity and 5 to 10μl of PBS buffer with 10% glycerol were mounted on grids. Grids were blotted from the back for ∼6 seconds and plunged into liquid ethane.

Leica EM Cryo-CLEM and Thermo Fisher Scientific iFLM Imaging

Vitrified HeLa cells expressing GFP-tagged LC3 on grids were imaged using EM cryo-CLEM and iFLM in the GFP channel, respectively. After acquisition of cell images in EM cryo-CLEM, image analysis and postprocessing were carried out using LAS X software with the Thunder imaging system. The iFLM images were deconvoluted using Richardson-Lucy method (Liu et al., 2025), which is instituted in DeconvolutionLab2 in Fiji (Sage et al., 2017). The PSF was generated based on the system’s optical parameters. The number of iterations was experimentally optimized by testing multiple iteration values and selecting the minimum number that provided sufficient resolution improvement without introducing visible noise amplification and deconvolution. Finally, the deconvolution was performed with 10 iterations. The image was also deconvoluted using fast iterative shrinkage-thresholding with iteration 10, step γ 0.5 for parameters (Beck and Teboulle, 2009). The intensity profiles in lateral and axial were plotted in ImageJ using Plot profile.

Measurements of Acquisition Time for the Tile Imaging

All experiments were performed using the 488 nm channel, and both the single-plane tile imaging and z-stack tile imaging were conducted for 3 × 3 tiles originating at the center of the grid mesh. Due to the integration with cryo-FIB-SEM instruments, iFLM can only image specimens at a specific angle. Thus, iFLM requires individual stage positioning sequentially and adjustments to the specific specimen angle for the tile imaging. The total acquisition time required for the stage movement and adjustments of angles was included for iFLM. For z-stack tile imaging, a total of 10 µm z-step with the 0.5 µm z-stack was used for the imaging.

SNR Measurement

For each microscope, 5 ROIs were selected either from fluorescent beads or from intracellular puncta in cell images. SNR was calculated as the ratio of the mean pixel intensity of the signal region to the standard deviation of the background noise. The same calculation procedure was applied to both images, and the averaged SNR values across all ROIs were plotted in the graph.

Lamella Preparation

Lamellae were prepared using an Aquilos2 Cryo-FIB (Thermo Fisher Scientific, SNU CMCI). To target LC3-GFP puncta, iFLM images were aligned with grid atlas maps from the SEM image using Maps 3.20 software (Thermo Fisher Scientific). Then, grids were sputtered with an initial platinum coat (10 seconds) followed by a 7-second gas injection system to add an extra protective layer of organometallic platinum. Samples were tilted to an angle of 10° and ∼10 µm wide lamellae were prepared. The milling process was performed with an ion beam of 30 kV energy in 4 steps: (1) rough milling: 0.5 nA with offset 700 nm, (2) 0.3 nA, with offset 300 nm, (3) 0.03 nA, with offset 150 nm, and (4) 0.03 nA, with offset 20 nm. For the serial imaging during the milling process, the iFLM images were acquired using the same exposure time and LED intensity for each milling step. For fluorescent intensity measurements, lamellae were prepared with the same procedure, and the final thickness was set to 700, 500, 300, and 150 nm without polishing steps.

Cryo-ET Data Collection

A total of 50 tilt series were acquired on Titan Krios (Thermo Fisher Scientific, SNU CMCI) equipped with a Falcon 4i direct electron detector and Selectris X energy filter (Thermo Fisher Scientific). Images were recorded in movies of ∼10 frames at defocus ranging from 4 to 5 µm and a pixel size of 3.06 Å. Tilt series were acquired using Tomo5 (Thermo Fisher Scientific), and a dose–symmetric tilt scheme was used covering a range of ±60° with a pretilt of ±10° and an angular increment of 3°. The cumulative total dose was ∼120 e-/Å2 per tilt series.

Data Processing and Subtomogram Averaging

For tomogram reconstruction, frames were aligned in Relion 5.0 (Burt et al., 2024) using the Relion implementation, and CTF was estimated using CTFFIND4 (Rohou and Grigorieff, 2015). Dark or bad tilt series were discarded by an eye inspection, and the remaining tilt series were aligned using Aretomo (Zheng et al., 2022). Tomograms were reconstructed at 8× bin and used for visualization and segmentation purposes.

For subtomogram averaging, all datasets were motion corrected and CTF estimation was performed in Warp (Tegunov and Cramer, 2019). The stack tilt series was then aligned using Aretomo (Zheng et al., 2022) and IMOD (Mastronarde and Held, 2017) and full tomograms were reconstructed at 8× bin (24.48 Å) in Warp. For template matching, Pytom template matching (Chaillet et al., 2025) was performed using a mammalian 80S ribosome density map, resulting in 27,344 coordinates. Subtomograms and corresponding CTF volumes were extracted at 4× bin (12.24 Å) in Warp and subjected to a few rounds of 3D classification in Relion 4.0 to discard bad particles. The remaining particles were aligned in Relion 4.0 (Kimanius et al., 2021) and imported to M (Tegunov et al., 2021) to re-extract at 2× bin (6.12 Å). Further refinement was done in Relion, yielding 20.4 Å consensus 3D map using 4,016 particles. Approximately 700 particles were randomly picked from each grouped tomogram according to the normalized fluorescence intensity and reconstructed in a 3D map without alignment, respectively.

Signal-to-Noise Ratio and Fluorescence Quantification

Fluorescence signals in lamellae were quantified using ImageJ (Schneider et al., 2012). For serial FIB-milled lamellae, the mean fluorescence intensity of each lamella was measured and subsequently normalized. To compare fluorescence intensities across 9 lamellae (approximately 50 tilt series), the measured mean intensities were normalized to the experimentally determined minimum and maximum values, resulting in a normalized intensity range of 0.08 to 0.27. Assuming a lamella thickness range of 100 to 300 nm, lamellae were categorized using an intensity threshold of approximately 0.18. This threshold corresponds to the arithmetic mean of the normalized intensity range and roughly to a lamella thickness of ∼200 nm under the assumption of a linear fluorescence-thickness relationship. Normalized fluorescence intensities were subsequently plotted against lamella thickness values measured directly from reconstructed tomograms using IMOD (Mastronarde and Held, 2017).

Segmentation and Visualization

Segmentation of the individual cellular feature, including autophagosomes, was performed in deep learning–enabled program MemBrain v2 (Lamm et al., 2025). Tomograms were rescaled to a pixel size of 10 Å before segmentation and default parameters were used.

To calculate and compare the volume of centrally located autophagosomes in 2 tomograms, we first generated masks corresponding to the z-thickness of each tomogram using Relion 4.0 (Kimanius et al., 2021). Subsequently, we created subtractive volume masks by segmenting the membranes and excluding the internal membrane density of the autophagosomes. The edited volume was then processed in Amira (Thermo Fisher Scientific) using the edit volume tool to generate voxel densities that account for both the z-thickness of each tomogram and the morphology of the autophagosomes. Finally, the voxel densities obtained from the 2 tomograms were calculated and compared using a volume fraction tool in Amira.

RESULTS

Benchmarking Cryo-CLEM Imaging Performance Under Cryogenic Conditions

To assess the imaging performance of cryo-CLEM under vitrified conditions, we selected 2 representative commercial systems that differ in configuration and integration level: the stand-alone EM cryo-CLEM microscope and iFLM, which is optically integrated into a cryo-FIB-SEM platform (Fig. 1A). This pairing allowed a direct comparison between a dedicated widefield cryo-fluorescence instrument and an in situ integrated system, reflecting the 2 major approaches currently adopted for cryo-CLEM workflows.

To quantitatively compare their imaging performance, we used vitrified calibration grids coated with 200 nm multicolor fluorescent microspheres, which emit across the visible spectrum and are widely used for assessing spatial resolution and signal-to-noise characteristics under cryogenic conditions (Fig. 1B). Each grid was imaged under cryogenic imaging conditions optimized according to the manufacturer’s specifications, using the excitation channels available on each system (EM cryo-CLEM: 405, 470, 565 nm; iFLM: 385, 470, 565 nm). Both microscopes clearly resolved individual fluorescent beads across all channels, enabling direct measurement of spatial resolution, SNR, and effective field of view for quantitative comparison.

EM cryo-CLEM captured smaller imaging areas, typically covering 2 to 4 grid squares (each square of a grid corresponds to ∼90 µm in width) (Fig. 1C), whereas the iFLM encompassed 12 to 14 grid squares under its standard imaging configuration (Fig. 1C). Considering their respective camera sensor sizes, pixel dimensions, and objective magnifications, this corresponds to an approximate field of view of ∼270 × 270 µm for EM cryo-CLEM and ∼500 × 500 µm for iFLM, representing a ∼3.5-fold difference in area coverage. The broader view of iFLM is consistent with its lower magnification (20×) and NA (0.7), while EM cryo-CLEM’s 50× objective (NA = 0.9) provides a narrower but more photon–efficient optical path.

We next performed a quantitative evaluation of imaging performance under cryogenic conditions using these calibration grids (Fig. 1D, Supplementary Table 2). First, the PSF of each system was measured to characterize the intrinsic optical response. Both microscopes produced well–defined bead profiles across all excitation wavelengths, allowing Gaussian fitting of intensity distributions (Supplementary Fig. 1). The lateral PSF width (FWHM) at 470 nm was 0.52 µm for EM cryo-CLEM and 0.50 µm for iFLM, while the axial PSF extended to 2.32 and 2.11 µm, respectively (Fig. 1D). These results confirm the typical anisotropy of widefield optics, with slightly worse resolution than the theoretical calculation and Z-resolution roughly 4-fold poorer than XY.

From the PSF profiles, the corresponding spatial resolution—the minimum distinguishable distance between 2 points—was estimated to be ∼0.5 µm laterally and ∼2.0 µm axially for both instruments (Fig. 1D). The close agreement indicates that cryogenic aberrations and scattering within vitrified samples, rather than objective specifications, primarily limit achievable resolution. Resolution also decreased slightly with longer excitation wavelengths, consistent with diffraction-limited behavior. Despite EM cryo-CLEM’s higher NA and magnification, the effective spatial resolution remained comparable to iFLM, suggesting that both systems operate near the practical limit of cryogenic widefield microscopy.

To assess practical image quality, we next compared the SNR derived from the same datasets (Fig. 1E). EM cryo-CLEM images exhibited stronger signal contrast and smoother background, yielding an average SNR ∼4.7-fold higher than iFLM. This improvement likely reflects EM cryo-CLEM’s greater photon-collection efficiency and shorter optical path in its stand-alone configuration, whereas iFLM’s integrated geometry sacrifices some sensitivity in favor of a larger survey area.

Collectively, these analyses demonstrate that both systems achieve comparable intrinsic resolution under cryogenic conditions but exhibit distinctive performance characteristics. EM cryo-CLEM provides higher image contrast and SNR under the tested conditions, making it more effective for rapid screening and precise targeting, whereas iFLM offers broader contextual coverage and direct correlation with FIB milling coordinates, facilitating integrated cryo-ET workflows. Together, these complementary attributes highlight that the 2 systems address different experimental priorities within cryo-CLEM applications rather than differing in absolute performance.

Comparative Imaging Performance of Cryo-CLEM Systems in Vitrified Cells

To extend our quantitative comparison to a biological context, we evaluated both cryo-CLEM systems using vitrified HeLa cells stably expressing GFP–tagged microtubule-associated protein 1A/1B-light chain 3 (LC3) (Fig. 2A), a canonical autophagy marker that localizes to autophagosomal membranes. To induce the accumulation of autophagosomes and generate strong punctate GFP signals, cells were treated with a combination of rapamycin and bafilomycin, which stimulates autophagosome formation while blocking lysosomal degradation. As expected, numerous bright GFP-LC3 puncta were observed in cryogenic fluorescence images, enabling a direct evaluation of imaging performance in a cellular setting (Fig. 2B and C).Fig. 2. Comparison of imaging performance for 2 cryo-CLEM systems in vitrified cells. (A) The experimental scheme of the grid preparation and resolution measurement using a HeLa cell line stably expressing GFP-tagged LC3 in vitrified ice. (B, C) The representative image of HeLa cells stably expressing GFP-tagged LC3 on the grid for EM cryo-CLEM, iFLM, respectively. Scale bar indicates 100 µm. (D) Comparison of imaging throughput and acquisition time measurements between EM cryo-CLEM and iFLM. Bar graphs show elapsed time measurements for 3 × 3 tile imaging (ΔZ = 0 µm) and 3 × 3 Z-stack tile imaging (ΔZ = 10 µm). (E, F) Raw fluorescent images and postprocessed images of vitrified cells acquired at 470 nm using EM cryo-CLEM and iFLM. A single HeLa cell is zoomed in. White dotted line indicates the boundary of the cell. Fluorescent signal intensity profiles of raw and postprocessed images were plotted along the yellow dotted line (bottom). Scale bar indicates 5 µm. (G) Bar graphs showing signal-to-noise ratio of raw images and postprocessed images for EM cryo-CLEM and ifLM.Fig. 2

To assess imaging throughput, we compared how efficiently each system could screen and locate cellular ROIs on the grid. The EM cryo-CLEM system incorporates an automated tile-imaging mode, which facilitates seamless acquisition of large areas, whereas this feature has not yet been adopted in the current iFLM configuration, which acquires individual positions sequentially. As a result, imaging workflows differed modestly in overall efficiency. For 3 × 3 tile imaging, EM cryo-CLEM required approximately 40 seconds, while iFLM took about 4 minutes under comparable conditions. For z-stack tile imaging, EM cryo-CLEM completed a typical acquisition in roughly 1.5 minutes, compared with ∼23 minutes for iFLM (Fig. 2D). Although these time differences reflect distinct design philosophies rather than absolute performance disparities, they suggest that automated tiling and streamlined stage movement in EM cryo-CLEM improve the efficiency of grid–wide cellular screening, whereas iFLM’s integrated architecture prioritizes positional precision within the cryo-FIB–SEM environment.

We next compared the image quality of cellular features captured in situ under comparable cryogenic conditions. Despite differences in objective magnification and working distance, the overall clarity and detectability of GFP puncta within vitrified cells were generally similar between the 2 systems (Fig. 2E and F). In both datasets, GFP-labeled structures appeared noisy and partially merged, characteristic of in situ widefield cryo-fluorescence imaging with limited resolution. To improve image interpretability, we applied postprocessing—EM cryo-CLEM Thunder computational clearing for EM cryo-CLEM datasets and standard deconvolution for iFLM images. After computational clearing, cellular puncta appeared sharper and more distinct, and fine structures within dense clusters became more discernible (Fig. 2E). Intensity-profile analysis revealed narrower peak widths and a marked reduction in background signal. In this dataset, the SNR increased from approximately 21.6 to 38, corresponding to about a 1.8-fold improvement, indicating that computational clearing effectively enhances lateral image contrast in vitrified cellular specimens. For iFLM, deconvolution produced no notable change in SNR and apparent lateral resolution (Fig. 2F). We further examined whether postprocessing improved axial (Z-axis) resolution. Consistent with the limitations of widefield optics, XZ projections displayed blurred features that were largely indistinguishable (Supplementary Fig. 2), and postprocessing provided only marginal enhancement without substantially improving visualization of fine 3-dimensional (3D) structures. Thus, while both instruments exhibit inherently low axial resolution under cryogenic conditions, computational postprocessing primarily benefits lateral image contrast rather than depth discrimination, improving overall interpretability of in situ cryo-fluorescence data.

Collectively, these results demonstrate that both cryo-CLEM systems can visualize cellular GFP-labeled structures at cryogenic temperature, but differ in practical performance. The EM cryo-CLEM achieves faster acquisition and greater enhancement after computational clearing, making it more suitable for rapid, high-contrast screening of cellular regions. In contrast, the iFLM provides wider contextual coverage within an integrated FIB-SEM environment, advantageous for precise spatial correlation during lamella preparation. Together, these complementary features delineate distinct but synergistic roles of stand-alone and integrated cryo-CLEM configurations in biological imaging workflows.

Cryo-CLEM Fluorescence as a Quantitative Proxy for Lamella Thickness and Quality

The preceding analyses demonstrated that both cryo-CLEM systems can effectively visualize GFP-labeled structures under cryogenic conditions and offer complementary strengths for biological imaging—the EM cryo-CLEM providing faster, high-contrast acquisition and the iFLM enabling wider contextual coverage within an integrated FIB-SEM platform. Notably, the iFLM is uniquely capable of acquiring fluorescence images intermittently during the milling process, allowing direct observation of changes in fluorescence signal as the specimen is progressively thinned (Fig. 3A). This iterative milling and imaging approach provides an opportunity to monitor lamella quality, linking optical information to the physical state of the sample. While cryo-CLEM has traditionally been employed mainly for coordinate correlation between fluorescence and electron microscopy, our results suggested that its imaging capability is sufficiently robust to yield quantitative indicators of specimen integrity. In particular, variations in fluorescence intensity may reflect parameters such as lamella thickness and structural preservation, both of which critically influence the success of downstream cryo-ET imaging.Fig. 3. Fluorescent intensity monitors the thickness of the frozen lamella sample. (A) The FIB views and fluorescence images of the target lamella during the FIB milling. Each fluorescence image was taken after the first milling (∼1,500 nm), the second milling (∼850 nm), the third milling (∼500 nm), and the final milling (200 nm) in series. The zoom-in view shows the XY slice and YZ slice image of the lamella during the FIB milling. The scale bar for the SEM images is 10 µm, and the scale bar for the iFLM images is 20 µm. (B) A line profile of the fluorescence signal intensity corresponding to each milling stage of the lamella. The white boxes show the areas enlarged in the fluorescence images, and the orange and magenta arrows indicate the corresponding fluorescence peaks in the line profiles below. The scale bar indicates 5 µm. (C) Normalized fluorescence intensity as a function of lamella thickness. Linear curve was fitted to the measured intensity using 3 different lamellae. (D) Representative TEM and Fourier transform images from the different lamellae thickness, showing information up to 8.0, 9.1, and 13.6 Å according to CTFFIND4 (indicated as white arrows). The scale bar indicates 200 nm.Fig. 3

Achieving high-resolution cryo-ET requires lamellae typically 100 to 200 nm thick, as excessive sample thickness reduces electron transparency and limits attainable resolution (Tuijtel et al., 2024). However, lamella thickness is usually assessed only after transfer to the TEM, making preselection inefficient. To streamline this process, we explored whether cryo-CLEM fluorescence intensity could serve as an intuitive and quantitative indicator of lamella thickness during FIB milling.

We used HeLa cells expressing LC3-GFP as a cellular model and conducted serial cryo-FIB milling on a single vitrified cell to produce lamellae of approximately 1,500, 800, 500, and 200 nm in thickness, while capturing corresponding fluorescence images at 470 nm excitation using the iFLM (Fig. 3A). This approach allowed us to monitor fluorescence changes within the same cell as material was progressively removed, thereby minimizing variability among different samples. In the initial thick region (∼1,500 nm), numerous bright GFP puncta corresponding to LC3-positive autophagosomes were clearly visible in both XY and XZ planes, appearing as discrete fluorescent vesicles distributed throughout the cytoplasm. As milling proceeded and the lamella thickness decreased, the overall fluorescence intensity gradually diminished, and by approximately 200 nm, the punctate signals were barely detectable above background noise. This progressive reduction in signal suggested a direct link between the physical thinning of the lamella and the total fluorescent volume retained within the imaging plane.

To quantify this effect, we tracked the fluorescence intensity of several representative GFP puncta across successive milling stages. The measured signal showed a gradual and reproducible decline with decreasing lamella thickness (Fig. 3B), with the steepest drop observed between 800 and 500 nm. This range coincides with the typical size of mammalian autophagosomes (approximately 500-1,500 nm in diameter), implying that the observed intensity loss originates primarily from the mechanical truncation of individual organelles during milling. Because each milling step effectively sections the same cellular region, the measured decrease in intensity reflects a proportional reduction in the amount of fluorescent material contained within the remaining volume of the lamella. Thus, the fluorescence decay provides a direct physical readout of the progressive thinning process.

We next plotted the mean fluorescence intensity of the lamellae as a function of their measured thickness (Fig. 3C). The resulting relationship was approximately linear, showing a consistent decrease in signal with decreasing thickness and approaching background levels at around 100 nm. This trend suggests that, within the typical thickness range used for cryo-ET (100-200 nm), fluorescence intensity can serve as a reliable proxy for lamella thickness. Importantly, this relationship offers a rapid, nondestructive means to estimate specimen quality and identify lamellae that are sufficiently thin for high-resolution data collection before transfer into the TEM.

To further verify the correlation between fluorescence intensity and structural quality, we examined the same lamellae by cryo-TEM. Lamellae that exhibited weaker fluorescence intensities consistently yielded stronger image contrast and better–resolved ultrastructural features and even macromolecules such as ribosomes in cryo-TEM (Fig. 3D), whereas those with higher fluorescence displayed lower contrast and diminished visibility of fine details. These findings confirm that fluorescence intensity not only reflects physical thickness but also provides an indirect measure of structural preservation, integrating both optical and ultrastructural indicators of lamella quality. Collectively, these results establish that cryo-CLEM imaging can offer quantitative, predictive feedback on lamella suitability for subsequent cryo-ET, enabling more efficient non–destructive specimen screening and data acquisition.

Cryo-CLEM Fluorescence Intensity as a Predictive Parameter for Lamella Quality and Cryo-ET Resolution

To assess whether fluorescence intensity could serve as a practical parameter for lamella screening before tomographic analysis, we first selected 2 lamellae exhibiting distinctly different GFP signal intensities (Fig. 4A). Cryo-ET tilt series were collected from both specimens, and 3D tomograms were reconstructed (Fig. 4B). In both tomograms, GFP-positive autophagosomes were clearly resolved, allowing volumetric segmentation and surface rendering for quantitative comparison (Fig. 4C). The autophagosome within the lamella showing stronger fluorescence intensity displayed a larger tomogram Z-height, confirming that fluorescence signal correlates with physical lamella thickness.Fig. 4. Fluorescent intensity enables the selection of the lamella for cryo-ET imaging. (A) Fluorescence image of FIB-milled lamella in Lamella 1 and Lamella 2. White arrow indicates the puncta, which are imaged later in TEM. The zoom-in view shows the targeted puncta in the lamella. The scale bar indicates 10 µm. (B) Tomographic slice views of lamellae with thicknesses of ∼120 nm (Lamella 1) and ∼230 nm (Lamella 2) are shown in the XY plane (top). The white arrows indicate the puncta corresponding to those marked in (A). The scale bar indicates 400 nm. The lower panels show the corresponding XZ slice views, visualizing the actual lamella thickness for each tomogram. (C) The segmentation of 2 tomograms. (Autophagosome: magenta, Microtubule: light green, Mitochondria: cyan, ER membrane: gray) (D) The volume analysis of autophagosomes compared to fluorescence intensity. (E) 3D reconstruction of in situ 80S ribosome. The atomic model (PDB: 7CPU, 7OSM) is fitted. The tRNA in P-site is colored in orange. The Fourier shell correlation curve of the reconstructed ribosome. (F) Normalized intensity is plotted against the thickness of tomograms. Lamella were categorized using a normalized intensity threshold of approximately 0.18 (red box), and ∼700 random particles from lamella above and below this threshold were pooled, respectively, for subtomogram averaging. The corresponding 3D reconstructions and achieved resolutions are shown below.Fig. 4

To further explore whether the fluorescence intensity correlates with the 3D characteristics of the autophagosome, we analyzed the relationship between fluorescence intensity and organelle volume. Since larger autophagosomes are expected to contain more localized LC3-GFP, we compared the fluorescence signal intensity with the measured volume of each autophagosome. Because autophagosomes within tomograms of different thicknesses exhibited heterogeneous sizes, their 3D volumes measured in voxels were normalized by their corresponding 2-dimensional projected areas for accurate comparison. As a result, the fluorescence signal of autophagosomes in the thicker tomogram was approximately 1.25 times stronger than that in the thinner tomogram, accompanied by an ∼8% increase in normalized volume (Fig. 4D). These findings indicate that the volume-to-area ratio of autophagosomes shows a direct correlation with fluorescence signal intensity, suggesting that fluorescence intensity can serve as a reliable indicator of volumetric features under cryogenic imaging conditions.

We next sought to determine whether cryo-CLEM intensity could predict molecular-resolution outcomes in cryo-ET. A total of 50 tomograms were acquired, and subtomogram averaging of ribosomes was performed as a benchmark for structural resolution. The initial consensus ribosome map reached 20.4 Å resolution, revealing clear molecular features such as the P-site tRNA and intersubunit cleft. We then normalized the fluorescence intensity of each lamella and plotted it as a function of measured thickness (Fig. 4F), confirming a consistent correlation between fluorescence signal and lamella thickness. Based on this relationship, lamellae were categorized into 2 groups according to a defined normalized fluorescence threshold: low-intensity (thinner) and high-intensity (thicker) subsets. From each group, approximately 700 ribosome particles were randomly extracted and reconstructed independently. Subtomogram averaging of particles from low-intensity lamellae yielded a map at ∼25 Å resolution, whereas those from high-intensity lamellae reached only ∼37 Å, corresponding to a ∼1.5-fold decrease in attainable resolution (Fig. 4E and F).

Collectively, these results demonstrate that fluorescence intensity measured by cryo-CLEM serves as a quantitative predictor of both lamella thickness and molecular–level image quality. This approach enables rapid, nondestructive prescreening and selection of optimally thinned lamellae that yield higher-resolution subtomogram averages, thereby improving the overall efficiency and reliability of cryo-ET workflows.

DISCUSSION

Cryo-CLEM has long served as an indispensable bridge between fluorescence and electron microscopy, primarily for localizing ROIs before cryo-FIB milling and cryo-ET. Yet, its imaging performance under vitrified conditions has remained largely unquantified. Here, by systematically evaluating 2 widely used cryo-CLEM systems—the stand-alone EM cryo-CLEM and the integrated iFLM—we established quantitative benchmarks for spatial resolution, SNR, and field of view. Both systems achieved comparable lateral resolution (∼0.5-0.6 µm) and limited axial resolution (∼2 µm), which are sufficient for lamella targeting but inadequate for resolving suborganelle details. EM cryo-CLEM, operating as a dedicated widefield microscope, provides higher photon-collection efficiency, rapid postprocessing through Thunder computational clearing, and a convenient tile-imaging mode that accelerates grid-wide screening. In contrast, the integrated iFLM sacrifices some sensitivity and magnification for mechanical stability, optical alignment precision, and direct correlation with the FIB coordinate system. Together, these results highlight the complementary trade-offs of stand-alone and integrated systems—speed and image contrast vs spatial registration and operational convenience.

Beyond its targeting role, our results extend cryo-CLEM toward quantitative assessment of specimen quality. By measuring fluorescence intensity during and after milling, we found a strong correlation between signal strength and lamella thickness. This enables on-site evaluation of lamella suitability before TEM loading, improving the efficiency of cryo-ET workflows. Meanwhile, autofluorescence also showed a linear correlation with lamella thickness, suggesting a potential usage for the absolute intensity–thickness correlation analysis. However, due to its low signal intensity, its application as a definitive reference would require further examination.

The iFLM’s unique capability to record fluorescence in situ during milling further allows real-time monitoring of thinning progress and optical feedback for adaptive FIB control. Importantly, we showed that fluorescence intensity not only reflects geometric thickness but also predicts molecular–level data quality: subtomogram averaging of ribosomes from thinner, lower-intensity lamellae yielded higher-resolution maps than those from thicker, high-intensity specimens. This establishes fluorescence intensity as a quantitative, predictive parameter linking optical and electron imaging modalities.

Integration of EM cryo-CLEM and iFLM into a unified workflow thus provides both high-throughput screening and high-precision targeting (Supplementary Fig. 3). EM cryo-CLEM’s fast tile-scanning capability enables comprehensive mapping of cellular distributions across grids, while iFLM’s integrated design supports in situ imaging and lamella quality evaluation immediately after FIB milling. The complementary optical configurations—EM cryo-CLEM offering higher SNR and localized zoom, iFLM providing a broader field and spatial continuity—collectively streamline specimen selection, lamella production, and downstream data collection. Such an integrated pipeline minimizes unnecessary TEM screening and maximizes the yield of lamellae suitable for high-resolution cryo-ET.

Despite these advances, several challenges remain. While we observed a clear fluorescence-thickness relationship in LC3-GFP–expressing HeLa cells, this correlation should be interpreted within a well-controlled and relatively homogeneous labeling system. Extension of this approach to other cell types or fluorescent probes may be influenced by biological variability in target distribution (Maday and Holzbaur, 2014), microenvironment–dependent photophysical properties of fluorophores (Giepmans et al., 2006), and specimen-specific variations introduced during lamella preparation. Therefore, the present approach is primarily intended for comparative thickness estimation within a defined experimental context, and broader application of the fluorescence-thickness relationship will require further experimental validation across diverse biological systems.

Due to the distinct NA and magnification specifications of the objectives integrated in EM cryo-CLEM and iFLM, a direct comparison of SNR is limited between 2 microscopes. However, even when accounting for their optical discrepancies through theoretical normalization, EM cryo-CLEM consistently exhibited higher SNR. While this indirect comparison provides a reliable benchmark, a more direct comparison considering camera performance, microscope configurations, and objectives with matched optical properties would be necessary in the future. Additionally, technical artifacts during FIB-milling, such as curtaining effects or uneven milling surface, may induce local fluctuations in fluorescence intensity that do not strictly correspond to the nominal thickness. Consequently, the application of the fluorescence-thickness relationship to other biological systems requires careful considerations of cell types, fluorophores, or even specimen surface quality (Al-Amoudi et al., 2005, Franken et al., 2022, Lam and Villa, 2021, Marko et al., 2006, Schaffer et al., 2017).

The fundamental diffraction-limited resolution (∼0.5 µm in XY, ∼2 µm in Z) of widefield cryo-CLEM restricts its use for single–molecule-level localization and limits precise 3D correlation, particularly for features smaller than 200 nm. Continued development of cryo-compatible super-resolution and single–molecule localization microscopy could help overcome these barriers (Chang et al., 2014, Li et al., 2023a, Li et al., 2023b, Li et al., 2023c, Liu et al., 2015, Moser et al., 2019), though maintaining optical performance under cryogenic and vacuum conditions remains nontrivial. Moreover, future systems may integrate in situ lamella preparation and fluorescence imaging within the same cryogenic environment, enabling continuous observation of sample quality during milling and immediate tomographic acquisition—an ideal convergence of optical and electron modalities. Finally, expanding fluorescence–based lamella evaluation beyond LC3-GFP to other organelles, tagged complexes, or small-molecule probes, coupled with automated image analysis and AI–driven predictive models, will further enhance throughput, reproducibility, and data quality in next-generation cryo-ET workflows.

Author Contributions

Minjung Kim: Writing – original draft, Visualization, Validation, Methodology, Investigation, Formal analysis, Data curation. Junsun Park: Writing – original draft, Visualization, Validation, Project administration, Methodology, Investigation, Funding acquisition, Formal analysis, Data curation, Conceptualization. Soung-Hun Roh: Writing – original draft, Supervision, Resources, Project administration, Funding acquisition, Conceptualization.

Declaration of Competing Interests

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.