Correlative light-electron microscopy using small gold nanoparticles as single probes

The study addresses a key challenge in correlative light-electron microscopy (CLEM): obtaining robust single probes that are simultaneously visible with high contrast in both light microscopy (LM) and electron microscopy (EM), enabling accurate mapping of molecular locations within cellular ultrastructure. Traditional dual probes (fluorophore plus gold nanoparticle) suffer from fluorescence quenching and potential degradation inside cells, while quantum dots pose toxicity, size, and blinking issues. Fluorophore-only CLEM is hampered by incompatibilities with EM preparation, photobleaching, and technical complexities of cryogenic LM. Moreover, many workflows require additional fiducial markers to register LM and EM images, adding complexity and potential artefacts. The purpose here is to demonstrate that small gold nanoparticles (AuNPs) can serve as single, photostable probes detected background-free by resonant four-wave mixing (FWM) LM and directly correlated with TEM on EM-ready sections, eliminating the need for additional fiducials and avoiding cryo-LM, while achieving nanometric localisation and high correlation accuracy.

Prior work commonly pairs fluorescent dyes with AuNPs for CLEM, but fluorescence is quenched in proximity to AuNPs and probe integrity inside cells has been questioned. Quantum dots offer dual LM/EM visibility but have cytotoxic cores, require bulky shells, and exhibit blinking that complicates tracking. Fluorophore-based single probes face incompatibility with EM fixation/staining and benefit from cryo-LM to preserve fluorescence and enable super-resolution, though cryo-LM demands specialised setups and risks sample devitrification at high illumination levels. High-accuracy correlation often uses spherical bead fiducials to correct for distortions via coordinate transformation, but these add complexity and potential artefacts. Separately, resonant FWM has been developed to detect individual small AuNPs in cells and tissues with background-free contrast and shot-noise-limited sensitivity, suggesting AuNPs as promising single CLEM probes.

• Probe and sample preparation: Human cancer (HeLa) cells were labelled with individual gold nanoparticles (AuNPs) of nominal 5–10 nm radius bound to epidermal growth factor (EGF). To preserve ultrastructure without artefacts from chemical fixation, cells were vitrified by high-pressure freezing (HPF), followed by freeze substitution and Lowicryl HM20 resin embedding without heavy metal stains. Thin (approximately 300 nm) resin sections were prepared for EM.
• FWM light microscopy: A two-beam degenerate resonant four-wave mixing (FWM) setup was implemented on an inverted microscope using ~150 fs pump, probe, and reference pulses from the same laser source, tuned to the AuNP localised surface plasmon resonance. Pump and probe were focused with a high-NA objective; the FWM signal was collected in epi-geometry. Heterodyne detection isolated the FWM at radiofrequency sidebands by amplitude-modulating the pump (vm) and frequency-shifting the probe (v2), detecting interference with a reference at v2 ± vm, providing suppression of linear scattering and incoherent background. The probe arrived ~0.5 ps after the pump to maximise FWM from hot-electron dynamics. Simultaneous confocal reflectance images were acquired for comparison.
• Polarisation-resolved detection: Probe/pump linear polarisations were converted to circular at the sample using waveplates. Dual-polarisation balanced detection measured co- and cross-circular components of reflected probe and FWM fields, yielding amplitudes and phases (A±, A_FWM±; Φ±, Φ_FWM±). Sensitivity of the cross-circular component to small AuNP shape asymmetries was analysed using an ellipsoid model; ellipse fits to high-magnification TEM images provided in-plane aspect ratio and orientation (γ) to correlate with FWM amplitude ratio and phase difference at the NP centre.
• Correlative imaging workflow: FWM was performed directly on EM-ready resin sections prior to TEM. The same sections were then imaged by TEM. Individual AuNP centroid positions in FWM images were obtained by 2D Gaussian fits of the A_FWM profile; corresponding TEM particle centres were assigned by shape geometry. An affine transformation (shear, scale, rotation, translation) was estimated by least squares to map FWM coordinates to EM coordinates. Correlation accuracy S was computed as the RMS of residual distances between transformed FWM and TEM coordinates across N particles. Focus quality and PSF width were used to identify out-of-focus NPs; TEM contrast assessed atypical NP structure.
• Axial localisation: FWM phase in reflection was used to encode axial displacement, enabling z-position estimation without axial scanning; localisation precision was derived from Gaussian fits and shot-noise analysis.

• Demonstration of single-probe CLEM: Small AuNPs (radii 5–10 nm) bound to EGF were detected background-free by resonant FWM on EM-ready, resin-embedded sections and correlated to TEM of the same sections without additional fiducial markers.
• Background suppression: Despite strong linear background in confocal reflectance from resin sections, FWM images were free from linear scattering and autofluorescence, clearly revealing individual AuNPs.
• Localisation precision: From Gaussian fits on single 10 nm-radius AuNPs, centroid localisation precision reached ~1 nm in-plane and ~4 nm axially under the demonstrated SNR. General localisation precision was below 10 nm.
• Correlation accuracy: Over areas >10 µm, FWM–TEM correlation accuracy was below 60 nm without fiducials; by reducing systematic errors, it improved to below 40 nm. In a representative dataset, RMS correlation error S was 94 nm when including all particles, improving to 54 nm after excluding outliers (particles with atypically low FWM amplitude due to out-of-focus imaging or atypical structure as indicated by weak TEM contrast).
• Polarisation-resolved shape sensitivity: The cross-to-co circular FWM amplitude ratio at the NP centre correlated with NP ellipticity, and the FWM phase difference tracked in-plane orientation, consistent with an ellipsoid model. TEM ellipse fits of individual NPs showed good agreement with model predictions, supporting the potential for shape-based multiplexing.
• Axial encoding: The linear dependence of FWM phase on axial position enables z localisation without scanning, facilitating 3D correlation.

The results establish small AuNPs as robust single probes for CLEM, directly bridging LM and EM without fluorescent labels or additional fiducials. Background-free FWM detection yields nanometric localisation precision in complex, scattering cellular environments and on EM-ready sections, addressing the long-standing challenge of imaging AuNPs within autofluorescent, resin-embedded samples. The achieved correlation accuracy (≤60 nm, improved to <40 nm by mitigating systematic errors) closes much of the resolution gap between LM and EM for spatial registration, enabling precise mapping of molecular targets (EGF) onto ultrastructural contexts. Polarisation-resolved FWM’s sensitivity to NP shape and orientation, validated against TEM via an ellipsoid model, provides an orthogonal contrast mechanism that can be exploited for multiplexing by NP shape recognition. The photostability of AuNPs and the compatibility of FWM with live-cell conditions obviate the need for cryo-LM, simplifying workflows and reducing risks of devitrification or photodamage that can compromise subsequent EM. Together, these advances suggest FWM-CLEM as a powerful alternative or complement to fluorescence-based CLEM, particularly where probe photostability, background suppression, and precise correlative accuracy are critical.

This work introduces a fiducial-free CLEM workflow leveraging small AuNPs as single probes detected by resonant FWM and correlated with TEM on EM-ready sections. It achieves background-free AuNP detection, nanometric localisation precision (~1–10 nm), and high FWM–TEM registration accuracy (≤60 nm over >10 µm fields, improved to <40 nm with reduced systematics). Polarisation-resolved FWM correlates with NP shape and orientation observed by TEM, opening avenues for shape-based multiplexing. The approach avoids cryo-LM by exploiting AuNP photostability and is compatible with live-cell imaging. Future work may focus on further reducing systematic registration errors across larger fields of view, extending to 3D correlative workflows using FWM phase-based axial localisation, developing libraries of engineered NP shapes for multiplexing, and applying the method to diverse biological targets and tissues.

• Registration outliers arose from particles significantly out of focus or with atypical structure (weak TEM contrast), indicating that focus quality and NP integrity affect correlation accuracy.
• The TEM provides an in-plane projection only, limiting unambiguous discrimination between oblate and prolate shapes and full 3D orientation; shape inferences rely on modelling and are partly degenerate.
• Correlation accuracy depends on mitigating systematic distortions; while improved to <40 nm, further reduction may require additional calibration or correction strategies.
• The method requires specialised FWM instrumentation and polarisation-resolved heterodyne detection, which may limit immediate accessibility.
• Although heavy metal stains were not used to retain compatibility with FWM, this may affect general EM contrast for other cellular features, potentially necessitating protocol optimisation for different specimens.