Ciliary beating patterns map onto a low-dimensional behavioural space

The study asks how the geometry of the behavioural space of ciliary beating maps to underlying molecular and biophysical mechanisms. Although cilia maintain motility across broad genetic and environmental perturbations (ATP, Ca2+, temperature, viscosity, pH, mutations), their spatiotemporal waveforms vary. Prior work across organisms shows that behavioural repertoires often occupy low-dimensional spaces, suggesting a few features define most variability. Leveraging cilia as a tractable model of motor-driven oscillations, the authors investigate whether the diversity of ciliary waveforms also collapses to a low-dimensional space and how those dimensions map onto dynein motor mechanochemistry. They analyze isolated and reactivated Chlamydomonas axonemes across many genetic and environmental conditions to quantify variation in beat frequency, amplitude, and detailed waveform shape, and to link this variation to parameters in a mechanochemical model.

The paper situates its work within a body of research showing low-dimensional behavioural spaces in nematode postures, microbial trajectories, and animal movements, implying few dominant features underlie behavioural variability. Mechanistically, ciliary and flagellar beating arises from dynein-driven microtubule sliding within the axoneme, with constraints (nexin links, radial spokes, basal anchoring) converting sliding to bending. Multiple mechanochemical models have reproduced oscillations in cilia and synthetic filaments, including curvature-controlled dynein regulation and related frameworks. Prior studies have characterized effects of temperature, viscosity, ATP, calcium, and dynein mutations on beat frequency and shape, and documented structural variability in axonemes. These set the stage for mapping waveform variability to specific motor response parameters.

Experimental system: Isolated and reactivated Chlamydomonas reinhardtii axonemes were prepared from cultured cells. Cilia were isolated (dibucaine), purified (sucrose cushion), demembranated (HMDEK buffer with detergent), and stored. Reactivation occurred in flow chambers with HMDEKP buffer containing an ATP-regeneration system; standard conditions were 24 °C and 1 mM ATP unless varied. Environmental and genetic perturbations: Environmental series included ATP (50–1000 µM), temperature (increments from 24 °C), viscosity (Ficoll 400 at 1%, 5%, 10%), Ca2+ (100 µM free), and taxol (10 µM). Genetic mutants included oda1 (outer-arm dynein deficient), ida3 and ida5 (inner dynein arm defects), mbo2 (microtubule inner proteins; symmetric beat), and tpg1 (reduced polyglutamylation). Imaging and tracking: Phase-contrast microscopy with high-NA objectives and a high-speed CMOS camera recorded movies at 1000 fps, up to 3000 frames. Axoneme shapes were tracked with FIESTA software at nanometre precision using 25 equally spaced segments along arc length; tracking included background subtraction and Gaussian fitting of segments. Signal representation: The tangent angle γ(s,t) (relative to a co-swimming frame) parameterized shapes. After removing the static (n=0) mode, the fundamental (n=1) Fourier mode dominated (>90% of power), enabling representation as w(s,t)=a(s)cos(2πft+φ(s)), with arc length normalized by total length L. Dataset: 498 axonemes across conditions, totaling ~40,000 beat cycles. Up to ~200 cycles per axoneme were analyzed. Feature extraction and dimensionality reduction: Amplitude a(s) and phase φ(s) profiles were decomposed into shifted Legendre polynomials. For amplitude, coefficients a0 (mean amplitude), a1 (linear/asymmetry), a2 (quadratic/parabolicity) were computed; for phase, φ0 set to 0, φ1 relates to wavelength (−L/λ), and φ2 quantifies parabolicity. Amplitude coefficients were normalized: ā1=a1/a0, ā2=a2/a0. Principal component analysis assessed variance explained by components. Mechanical modeling and fitting: A linearized mechanochemical model (Machin-type force balance) included a passive sliding stiffness k and complex curvature response β=β′+iβ″, where β′ quantifies instantaneous curvature response and β″ the dynamic (rate) response. The dimensionless Machin number Ma encapsulates the ratio of viscous to elastic forces. For given parameters (k,β′,β″), the model yields discrete waveform solutions; the lowest-wavenumber solution was selected. Model fitting maximized a waveform similarity score R with a principal-axis algorithm from multiple initializations. Fits mapped each empirical waveform (48D: 24 amplitude, 24 phase samples) into a 3-parameter mechanical space (k,β′,β″). Goodness-of-fit R² was reported. Analytical results in the low-friction limit (Ma→0) related k and β″ and predicted λ=−4π/β″ and amplitude asymmetry controlled by β′. Quantitative analyses: Beat frequency f, mean amplitude a0, axoneme length L, and wavelength λ were measured; Q-factors of oscillations were computed; within- vs between-condition variances were compared. Correlations between polynomial coefficients, principal components, and model parameters were assessed.

• Despite wide variations in beat frequency (~15–160 Hz across conditions) and amplitude, waveform shape space is low-dimensional: two features (amplitude asymmetry ā1 and parabolicity ā2, with corresponding phase features φ1 and φ2) account for ~80% of variance.
• Environmental perturbations (ATP, temperature, viscosity) strongly modulate frequency (about one order of magnitude range) but have relatively little effect on mean amplitude between conditions; within-condition amplitude variability exceeds between-condition variability by ~5-fold.
• Genetic and environmental perturbations yield characteristic shape changes: taxol, ida5, and tpg1 show negative amplitude asymmetry (ā1<0; amplitude decreases distally); Ca2+ shows positive asymmetry (ā1>0); oda1 exhibits near-linear amplitude profiles with small parabolicity (ā2≈0).
• Wavelength scales with axoneme length across all conditions: λ/L=0.97±0.08 (mean±s.d.; N=498), with lengths ranging 7.2–15.4 µm.
• Individual axonemes maintain highly stable frequency over time (Q-factor up to 150), indicating that variability arises mainly between axonemes/conditions rather than within a trace.
• Mechanical model fits are excellent: 92% of axonemes have R²>0.9, reducing empirical waveform dimensionality from 48 to 3 parameters (k,β′,β″).
• Strong correlation between sliding stiffness k and dynamic motor response β″ matches the analytically derived low-viscosity (Ma→0) relation, indicating operation in a low-friction regime. In this limit, λ≈−4π/β″ explains λ≈L requiring β″≈−4π on average.
• Mapping between spaces: β′ controls amplitude asymmetry (ā1), β″ controls parabolicity (ā2) and wavelength; motor response coefficients also correlate with phase features (φ1,φ2). Taxol and Ca2+ occupy distinct regions in (β′,β″) consistent with their distinct waveform features.
• Energetic constraints likely bound the joint amplitude–frequency space: absence of high-amplitude, high-frequency beats is better explained by limits from elastic dissipation than viscous dissipation, consistent with Ma≪1.

The work demonstrates that ciliary waveform variability induced by diverse genetic and environmental perturbations collapses onto a low-dimensional behavioural space defined primarily by amplitude asymmetry and parabolicity (and corresponding phase features). This addresses the central question by linking these geometric features to a simple mechanochemical model in which dynein motors exhibit instantaneous (β′) and dynamic (β″) curvature-sensitive responses and the axoneme has a passive sliding stiffness k. The tight correlation between k and β″ and the consistency with the low-Machin-number analytical limit indicate that Chlamydomonas axonemes operate in a regime where elastic forces dominate viscous forces (Ma≪1). Consequently, a small set of motor response parameters suffices to explain observed waveform diversity across perturbations, including reciprocal changes in amplitude and frequency under taxol vs Ca2+ and the near-equality of wavelength and length. The mapping provides mechanistic insight: changes in dynein curvature sensitivity (phase of the curvature response) underlie observed shape variations. The findings align with broader observations of low-dimensional behavioural phenotypes and suggest that canalized, robust oscillatory mechanisms persist despite molecular diversity and environmental fluctuations.

This study constructs and maps the behavioural space of ciliary beating across extensive genetic and environmental perturbations, revealing that two shape features (asymmetry and parabolicity) explain ~80% of waveform variability. By fitting a minimal mechanochemical model, the authors show that this low-dimensional shape space corresponds to variations in only two dynein curvature response coefficients (β′ and β″), with a correlated passive sliding stiffness k, and that axonemes operate in a low-friction (low Ma) regime. Key contributions include: comprehensive quantification of waveform variability in isolated axonemes (N=498), demonstration of a robust wavelength–length relationship (λ≈L), and a high-fidelity mapping from high-dimensional shapes to a three-parameter mechanical space. Future research directions include establishing causal links between specific perturbations and dynein regulatory changes, extending analyses to intact cells and three-dimensional beating (including twist), exploring alternative motor control mechanisms, and elucidating how beat frequency is selected and regulated across viscosity changes.

• Causality not proven: while waveform changes map to β′ and β″, direct causal modification of dynein curvature sensitivity by each perturbation was not demonstrated.
• Model simplifications: linearized, planar model neglects three-dimensional components (e.g., twist) and uses low-Machin-number approximations; boundary conditions and parameter estimates carry uncertainty.
• Parameter uncertainty: values for friction and bending stiffness vary in the literature; although the low-Ma regime reduces sensitivity, exact parameter magnitudes remain uncertain.
• System scope: measurements on isolated, reactivated axonemes may not fully reflect intact-cell regulation and coupling to the basal body or cytoplasm.
• Frequency selection remains unexplained: Ma was relatively insensitive to viscosity due to compensatory changes in frequency, highlighting incomplete understanding of beat frequency control.