Conference Agenda

Surpassing the diffraction limit has become daily practice in the field of microscopy. This is made possible by methodological breakthroughs in which specific modifications to the optical hardware and labeling biochemistry are combined with computational advances in an integral approach. In the presentation, I will give an overview of current developments in our lab at Delft University of Technology. Topics that will be presented range within limitations imposed by noise, advances for large field or volume of view, and efforts toward cryogenic microscopy. Specifically I will highlight:

(1) Recent results on the classical Richardson-Lucy deconvolution method, in which we use a Cramer-Rao Lower Bound analysis to show that amplification of noise and ill-convergence are a necessary consequence of the method.

(2) The impact of field dependent aberrations, where we use so-called Nodal Aberration Theory to model and fit field-dependent aberrations from single-molecule data, and show how this improves the accuracy of localization.

(3) Cryogenic 4pi-microscopy, where we use the self-interference of fluorescent light collected trough two opposing objective lenses, in order to overcome the poor axial localization precision in low NA localization inherent to cryogenic setups.

Structured illumination microscopy (SIM) is a powerful super-resolution technique that relies on patterned light to surpass the diffraction limit. However, conventional SIM setups require complex optical configurations that limit their adaptability and usability. In this work, we present an integrated optical chip fabricated using femtosecond laser micromachining, designed to simplify the generation of structured light. This chip, coupled with an optical fiber, enables direct generation of structured illumination on its surface, where the biological samples can be positioned for imaging. Our approach eliminates bulky optics, enhances system stability, and offers a compact and scalable solution for super-resolution microscopy. We present the design, the fabrication process, and the first validations of this system for advanced biological imaging.

Live-cell imaging enables observation of dynamic cellular processes, yet many subcellular structures remain inaccessible due to the diffraction limit. Fluctuation-based super-resolution techniques overcome this barrier by exploiting intensity variations caused by fluorescence blinking. Among these methods, Super-resolution Optical Fluctuation Imaging (SOFI) enhances spatial resolution by leveraging the statistical properties of blinking fluorophores, achieving an n-fold resolution gain through nth-order calculations. However, SOFI requires the acquisition of hundreds of frames and substantial post-processing, making it unsuitable for real-time visualization of rapid cellular dynamics.

To address these limitations, we utilize a recurrent neural network (RNN) model, MISRGRU, inspired by SOFI. MISRGRU processes sequences of low-resolution frames to extract correlated temporal signals, effectively enhancing temporal resolution while achieving a twofold increase in spatial resolution. We also compare this approach with a widely adopted fully convolutional architecture, the U-Net, evaluating both in terms of resolution enhancement and computational latency. While both models achieve comparable resolution improvements using significantly fewer frames than SOFI, the U-Net’s requirement for frame interpolation prior to inference limits its suitability for real-time applications. In contrast, MISRGRU offers a 250-fold reduction in reconstruction latency, demonstrating strong potential for real-time super-resolution imaging in live-cell experiments.

Super-resolution microscopy based on the localization of single molecules facilitates the visualization of cellular structures at a resolution approaching the molecular level. However, their low-throughput nature hampers their applicability in biomolecular research and screening. Here, we propose an efficient workflow, starting with the scanning of large areas using fast fluctuation-based imaging, followed by single-molecule localization microscopy of selected cells. We exploit the versatility of DNA oligo hybridization kinetics with DNA-PAINT probes to tailor the fluorescent blinking toward high-throughput and high-resolution imaging. Additionally, we employ super-resolution optical fluctuation imaging (SOFI) to analyze statistical fluctuations in the DNA-PAINT binding kinetics, thereby tolerating much denser blinking and facilitating accelerated imaging speeds. We demonstrate 30–300-fold faster imaging of different cellular structures compared to conventional DNA-PAINT imaging, albeit at a lower resolution. Notably, by tuning the image medium and data processing, we can flexibly switch between high-throughput SOFI (scanning an FOV of 0.65 mm × 0.52 mm within 4 min of total acquisition time) and super-resolution DNA-PAINT microscopy and thereby demonstrate that combining DNA-PAINT and SOFI enables one to adapt image resolution and acquisition time based on the imaging needs. We envision this approach to be especially powerful when combined with multiplexing and 3D imaging.

We present a method to generate vector beams using a single spatial light modulator (SLM) for depletion-based super-resolution microscopy techniques such as STED, RESOLFT, and subtraction microscopy. We demonstrate the generation of circularly polarized Laguerre-Gaussian LG01 beams and vector beams with spatially varying polarization, such as azimuthal beams. Their intensity profiles, polarization distributions, and robustness to optical aberrations are experimentally analyzed under low- and high-numerical aperture conditions to assess their suitability for super-resolution applications.