Fig 1.
Decision making in chemotaxis of sea urchin sperm.
(A) Helical swimming path of a sea urchin sperm cell (black) with helix centreline (red), while navigating in a concentration field of the chemoattractant resact [16]. The concentration field is cylindrically symmetric with symmetry axis parallel to the z-axis (indicated in blue). (B) Projection of the same swimming path on the xy-plane. Dots mark the beginning (black) and peak (red) of ‘high-gain’ steering phases (or off-responses [16]). The concentration field is indicated by blue circles. (C) From the swimming path and the local gradient direction, we can determine a time-dependent rate γ(t) of helix bending towards the gradient [16]. The beginning of a ‘high-gain’ steering phase is defined as the level-crossing of γ(t) above its median as is indicated by black dots. Peaks of γ(t) are indicated by red dots. (D) Scatter plot of the orientation angle Ψ and local concentration c at the beginning of ‘high-gain’ steering phases (n = 9 cells). ‘High-gain’ steering is predominantly initiated for Ψ > π/2 (grey shading).
Fig 2.
Helical chemotaxis in the presence of sensing noise.
(A) Chemoattractant molecules bind to receptors on the cell membrane. (B) The sequence of binding events defines a stochastic input signal s(t) with rate b(t), Eq 1. (C) A sperm cell swims along a helical swimming path (black), whose centreline (red) can bend in the direction of a concentration gradient (blue). (D) Helical swimming in a concentration gradient causes a periodic modulation of the rate b(t) of binding events (red). Representative realization of input signal s(t) (black, low-pass filtered for visualization). This signal dynamically regulates the path curvature κ(t), here shown in the absence of sensing noise (red) and for stochastic input signal (black). (E) Example swimming paths with and without sensing noise for two values of the gain factor (‘low-gain’ steering ρlow = 1, ‘high-gain steering’ ρhigh = 10). Egg cell (yellow disk). (F) Signal-to-noise ratio (SNR) as a function of distance R from the egg. The SNR defines a ‘noise zone’ spanning intermediate distances R, bounded by a noise zone boundary , where SNR = 1, and a spatial limit of chemosensation
, where c(R) = (λT)−1. (G) Probability to find the egg as a function of gain factor ρ for initial distance R0 = 3 mm to the egg (and random initial orientation). Without sensing noise, the success probability increases monotonically with ρ, while in the presence of noise, this probability displays a maximum at an optimal ρ. Maximum search time 300s. Error bars smaller than symbols. Parameters chosen to match experiment, see S1 Appendix.
Fig 3.
Sperm navigation mapped on a Markov decision process.
(A,B) Binning of (R, Ψ)-phase space and sketch of trajectories for ‘low-gain’ (white) and ‘high-gain’ (black) steering. (C) Illustration of a single decision: Starting in a state 1, the player first chooses between two actions, i.e. ‘low-gain’ steering or ‘high-gain’ steering. This choice determines the transition probabilities Lij for jumping to a different state, here labelled 2 and 3. (D) Illustration of a memoryless decision strategy, assigning a choice of action to each state. The figure shows coarse bins for sake of illustration.
Fig 4.
Chemotactic success with decision making.
Success probability P(R0) for the optimal decision strategy, resulting from switching between ‘low-gain’ and ‘high-gain’ steering, as function of initial distance R0 to the egg for the case of noise-free concentration measurements (A), and physiological levels of sensing noise (B) (red squares). For comparison, success probabilities for strategies without decision making are shown (circles). (C,D) Optimal decision strategies for the cases shown in panel A and B. Greyscale represents prediction frequency of ‘high-gain’ steering, using a cohort of MDPs obtained by bootstrapping, see S1 Appendix for details. Arrows and dashed lines indicate zone boundaries as introduced in Fig 2. (E,F) Spatial sensitivity analysis of optimal strategies: Shown is the change in chemotactic range as function of cut-off distance Rc for hybrid strategies that employ the optimal strategy for R < Rc, and either ‘low-gain’ steering (white circles) or ‘high-gain’ steering (black circles) else. Positive values indicate a benefit of decision making at the respective distance to the egg. Parameters, see S1 Appendix.
Fig 5.
Simple implementation of optimal decision making.
(A) Signalling variables p and q contain information about the helix orientation angle Ψ and distance R to the target. Contour levels for conditional probability densities (red) and
(black) (corresponding to 1%, 10%, 50%, 90% percentiles; R = 1.5mm). (B) Relative frequency of ‘high-gain’ steering predicted by the optimal decision strategy, for given combination of (p, q). We define a decision boundary Θ(p) (yellow) by a piecewise linear fit to the 50%-contour line (up to p = 5ms, corresponding to a limit of sufficiently reliable state estimation, see S1 Appendix). (C) Simulated swimming path using this decision rule with dynamic switching between ‘high-gain’ steering (red) and ‘low-gain’ steering (black); projected on xy-plane. The chemoattractant concentration in this plane is shown (blue gradient), together with the boundary of the noise zone. (D) Success probability P(R0) for full simulations with simple decision making (red) as a function of initial distance R0 to the egg. For comparison, success probabilities for ‘low-gain’ steering (white) and ‘high-gain’ steering (black) are shown. (E) The effective chemotactic range
with decision making (red) is larger than
for an optimal constant gain factor (black). Parameters, see S1 Appendix.