Perspective: An outlook on fluorescence tracking

Abstract

Tracking single fluorescent molecules has offered resolution into dynamic molecular processes at the single-molecule level. This perspective traces the evolution of single-molecule tracking, highlighting key developments across various methodological branches within fluorescence microscopy. We compare the strengths and limitations of each approach, ranging from conventional widefield offline tracking to real-time confocal tracking. In the final section, we explore emerging efforts to advance physics-inspired tracking techniques, a possibility for parallelization and artificial intelligence, and discuss challenges and opportunities they present toward achieving higher spatiotemporal resolution and greater computational and data efficiency in next-generation single-molecule studies.

Figure 1:

Widefield offline SPT methods process image stacks by first performing a localization step to identify particle positions within each frame. The number of localized particles retained is typically controlled by user-defined thresholds, specific to each method, such as spot quality, radius, or total intensity. These localized positions are subsequently passed to a linking step, which connects them across frames to reconstruct complete particle trajectories^,,,

Figure 2:

Demonstration of ambiguity between background signal and out-of-focus particle using synthetic data. a, Simulated frames showing a single particle, marked by the red arrow, located at varying axial distances from the focal plane, ranging from 0μm to 1μm. Background noise and camera characteristics are based on real experimental conditions, assuming an EMCCD camera with gain and offset of 100, a numerical aperture of 1.45, refractive index of 1.515, and emission wavelength of 665nm. b, Comparison of log posterior probabilities under two competing hypotheses: the presence of one emitting particle versus no emitting particles. c, Log posterior difference plotted as a function of the particle’s axial distance from the focal plane, illustrating the gradual loss of statistical evidence as the particle becomes increasingly out of focus.

Figure 3:

Blurring artefacts in a simulated single molecule labeled with a Cy3B fluorophore (assuming an emission wavelength of 571nm) become increasingly apparent with higher diffusion coefficients under typical camera exposures of 10 ms. The pixel size is 100 nm, and the objective’s numerical aperture is 1.49. Background noise is excluded to emphasize the effect of blurring for illustration. a, Frames show increasing diffusion coefficients from left to right. Contrast is enhanced at higher diffusion coefficients for better visualization. The total number of detected photons per frame remains constant. A single molecule can appear as multiple at higher diffusion coefficients due to blurring artefacts. b, Representative diffusion coefficient ranges for various biological systems.

Figure 4:

Most real-time SPT methods operate by continuously localizing a single particle in real-time and dynamically adjusting the observation volume to keep the particle centered within the detection volume.

Figure 5:

A sampling step of the physics-inspired SPT framework. Particle track proposals closer to the ground truth (red) than those further away (green) are assigned higher posterior probability densities. And these densities ultimately dictate whether the track proposals will be retained. The relative density is defined as the probability density of a given track divided by the probability density of the ground truth track.

Figure 6:

Demonstration of a physics-inspired SPT framework, BNP-Track, using synthetic datasets modeled with EMCCD camera characteristics. Three consecutive frames illustrate significant PSF overlap and high background noise. The ground truth tracks (orange) indicate the presence of three particles. Conventional offline SPT methods, such as TrackMate, produce tracks (blue) with an incorrect number of particles and mislinked trajectories. In contrast, BNP-Track’s MAP estimate (green) closely matches the ground truth, accurately resolving particle identities and paths despite the challenging imaging conditions.