Deep learning-enhanced Lagrangian 3D Tracking of motile microorganisms

Abstract

How microorganisms respond to and interact with their environment can vary significantly from individual to individual, which can have important microbiological and ecological implications. However, most microscopy techniques can only observe motile microorganisms for short times because of their limited fields of view. Using Lagrangian tracking, a single microorganism can be followed in 3D, potentially indefinitely, allowing to decipher individual phenotypical traits. Current Lagrangian tracking methods use the fluorescence signal emitted by the microorganism as feedback to keep it in focus. However, over long times, epifluorescent imaging can induce photobleaching and photodamage, and importantly, not all microorganisms can easily be made fluorescent. Additionally, traditional algorithms used in feedback loops to determine microorganism position are prone to errors, especially in optically complex media. Here, we present a faster, more reliable, and versatile Lagrangian tracking method that uses deep learning to determine the 3D position of the microorganism. This new method demonstrates enhanced accuracy and speed in tracking fluorescent bacteria with fluorescence microscopy also in optically complex media. Furthermore, we track bacteria with other microscopy modalities, such as brightfield microscopy -- for example, this enables us to track magnetotactic bacteria, which cannot be made fluorescent without degrading their magnetotactic properties. These novel capabilities allow to extract previously inaccessible quantitative information, significantly advancing the study of microorganism behavior -- and thus opening new avenues for research in complex biological and ecological systems.

Figures (6)

• Figure 1: AI-enhanced Lagrangian tracking. (a) Minimalist sketch of a standard Lagrangian tracking setup featuring the feedback loop between the image of a particle (see Darnige et al. darnige2017lagrangian for details on the algorithm) and the command on a set of 3 stages moving in $X,Y,Z$ suited to keep the particle in the central part of the camera image and in the focal plane position $z_F$ (More details in supplementary document Fig S1). (b) Brownian traces $X(t), Y(t), Z(t)$ of a fluorescent latex bead of diameter $d = 1.1 \pm 0.1 \mu$m obtained with the standard tracking algorithm. Inset images are zooms on the particle that highlights a "jump" in $Z(t)$ corresponding to a sudden loss of focus (see Video1b in supplementary documents) . (c) Statistics of the short-time displacements (time lag $\Delta t = 0.0375 s$) showing the statistical importance of spurious jumps due to tracking errors for the standard set-up. (d) Sketch of the neural network (NN) used in the new 3D-tracking version to compute the expected shifts from the central positions $\Delta X$, $\Delta Y$ and of the focal plane $\Delta Z$. This new algorithm is used to feedback on the stages positions. (e) 3D traces of the Brownian motion for a $d = 1.1 \pm 0.1 \mu$m latex bead displaying no spurious "jump". Inset image of the bead, essentially kept in focus. (see Video1e in supplementary documents) (f) Statistics of the short-time displacements (time lag $\Delta t = 0.0375 s$) showing the absence of spurious "jumps".
• Figure 2: Axial position determination. (a) A ZScan : Stack of object images taken every 0,1 µm at 80Fps (b) Vertical offset from the focal plane $\Delta zpred$ estimated by the AI algorithm, versus $\Delta zreal = Z_{real}-z_F$, the real position of the bead with respect to the focal plane for different vertical positions (color code). See corresponding Z-scan in Supplementary Video2 and the other $X$,$Y$ predictions in Fig. S4 in supplementary document. The line is the expected curve for a perfect prediction. (c) At the bottom are displayed the corresponding errors with respect to a perfect prediction.
• Figure 3: Long-term tracking of individual bacteria using fluorescence and brightfield microscopy. Tracking of a E.coli bacterium with the deep learning-enhanced method (a) 3D trajectory of a fluorescent E. coli obtained using fluorescence microscopy, with an inset showing the bacterium in focus (see Video3a in suppl.videos). (b) In red, corresponding fluorescence intensity decreasing over time due to photo-bleaching (right vertical axis); in blue, decline in swimming velocity over time, indicating photo-damage stemming from a high level extended illumination (left vertical axis). (c) 3D trajectory of an E. coli bacterium tracked with brightfield microscopy; inset shows the bacterium in focus see Video3c in suppl.videos). (d) Stable swimming velocity over time, demonstrating the lack of photo-damage and the ability to track individual bacteria in bright field for long times.
• Figure 4: Long-term 3D tracking reveals magnetotactic swimming phenotypes. (a) Trajectory of a magnetotactic bacterium (MRS-1) tracked in 3D under a $B=1.4$T magnetic field (See Video4a in suppl. information). Above, image of the bacterium in focus (63X objective) under bright-field illumination. On the right, sketch of the MRS-1 displaying the two amphitricious flagella and the magnetosome (linear assembly of nanoscale $Fe_20_3$ magnetite crystals). (b) Trace $X(t)$ in the direction of the magnetic field. Inset zoom of the trace showing sequences of run reversals which compound defines a net motion in the direction of the field (this bacterium is called a "north-seeker"). The inset illustrates a dominant persistent time (along the field) $\tau_p^d$ and a sub-dominant persistent time (opposite to the field) $\tau_p^s$. (c) Corresponding distributions of the swimming velocities showing a manifest symmetry-breaking characterizing the north-seeker phenotype. (d) Statistics of run-times in the direction of the $\vec{B}$ field (dominant run times in red ) and opposite to it (subdominant run times in blue). The solid line is the exponent $\alpha = -2$ as a guide to the eyes illustrating the existence of a fat-tail for these distributions.
• Figure 5: Transport properties of a passive tracer in a dense bacterial bath. (a) Image of a suspension of motile bacteria in the "active turbulent" regime , observed at the mid confinement plane $z = H/2$ with $H = 200 \mu m$ (see Video5a in suppl. information). Flow lines in red are reconstructed by PIV measurement on fluorescent tracers (white tracks represent particle streakline). The white horizontal bar is 100 $\mu m$. (b) 3D trajectory of a $3\mu m$ radius red-fluorescent spherical colloid immersed in the turbulent bath and tracked during $450$ s. Inset, image of the red fluorescent tracer surrounded by fluorescent bacteria in green (see video 5b in suppl. information). Fluorescent E.coli are about $10 \%$ of the total number of bacteria. Tracking of the colloid is done with a 2-color technique described in Junot et al.Junot2022 (c) Long-time transverse MSD of the bead showing a ballistic/diffusive crossover at a time $\tau_c = 2.2 s$. Fit of the MSD (red line) yields a mean ballistic velocity of $45 \mu /s$ and an effective transverse diffusivity $D=475 \mu m^2/s$ . (d) Orientation distribution of relative bacteria velocities around the tracer showing the emergence of a local polar order parameter $P(0) = \langle \cos \theta \rangle = 0.2$.