Abstracts

The navigational behavior of an animal consists of several steps of information processing such as the detection of the surrounding environment, integration of multi-modal sensation, generation of action, etc. Understanding how each process of navigation is implemented in a neural circuit is a big topic in neuroscience. A combination of theoretical and experimental approaches is necessary to tackle this question. We use Caenorhabditis elegans which provides a compact neural circuit consisting of 302 neurons and simple behavioral experiments for dissecting neural code. Particularly, we are focusing on thermotaxis behavior which involves associative learning and memory. Based on our quantitative measurements of behavior and neural activity, we modeled the sensory encoding of a thermosensory neuron based on a systems identification framework. By considering the stochastic components of both behavior and neural activity, we are trying to extend our model to involve the entire neural circuit. Infortaxis strategy might be the clue for the extension of our approach, and I will discuss how it fits our data and model. The direction of the whole brain neural recoding approach will be mentioned along with this paradigm. 

References:

1. Tsukada Y, Yamao M, Naoki H, Shimowada T, Ohnishi N, Kuhara A, Ishii S, Mori I.J Neurosci. 2016 Mar 2;36(9):2571-81. doi: 10.1523/JNEUROSCI.2837-15.2016.

2. Vergassola M, Villermaux E, Shraiman BI.Nature. 2007 Jan 25;445(7126):406-9. doi: 10.1038/nature05464.PMID: 17251974

Our lab is focused on the algorithms and computation selected by evolution to perform biological decision-making. We address this topic with an interdisciplinary approach mixing statistical physics, Bayesian machine learning, information theory and various experimental biological setups. In this talk I will address two topics associated to 2 different physical scales and 2 different biological systems. 

In the first, we discuss robust statistical testing in single biomolecule dynamic experiments dominated by noise and under-model uncertainty and intractability. We discuss simulation-based inference and associated amortised inferences, which enable variational posterior distribution estimation of physical parameters[1] and physics-informed learned latent space [2].  We address optimality regarding Cramer-Rao [3] bound in a tractable case. We present a two-step statistical testing scheme for comparing biomolecule dynamics observed in different experimental conditions without relying on specific models and in any conditions of experimental noise and trajectory length variability. We provide a platform to run automated statistical tests on any 2D-3D experiments https://tracktor.pasteur.cloud/ and biological applications[4].

In the second, we discuss initiative to explore the constraints driving the design of the Drosophila melanogaster larva. The central nervous system produces diverse behaviours, like muscular responses, observable via video recordings. Current advancements in genetics, large-scale behaviour tracking, and machine learning facilitate the understanding of how behaviour and neural activities correlate. In organisms like the Drosophila larva, it's now feasible to map this at large scales, covering millions of animals and individual neurons [5,6]. Yet extracting nuanced behaviours from these screens and interpreting them at a broader scale remain challenges. We will introduce a mix of physics-informed generative models, unsupervised continuous learned behavioral latent space and statistical testing to link individual neurons to subtle behavioral change. We will discuss the use of these approach to investigate neuromodulation and operant learning [7] in the larva. Finally, we will conclude by recent initiatives in simulating the full body of the larva.

References:

1. Cranmer, K., Brehmer, J. & Louppe, G. The frontier of simulation-based inference. Proc. Natl. Acad. Sci. 117, 30055–30062 (2020).

2. Verdier, H. et al. Learning physical properties of anomalous random walks using graph neural networks. J. Phys. Math. Theor. 54, 234001 (2021).

3. Verdier, H., Laurent, F., Cassé, A., Vestergaard, C. L. & Masson, J.-B. Variational inference of fractional Brownian motion with linear computational complexity. Phys. Rev. E 106, 055311 (2022).

4. Verdier, H. et al. Simulation-based inference for non-parametric statistical comparison of biomolecule dynamics. PLOS Comput. Biol. 19, e1010088 (2023).

5. Vogelstein, J. T. et al. Discovery of Brainwide Neural-Behavioral Maps via Multiscale Unsupervised Structure Learning. Science 344, 386–392 (2014).

6. Masson, J.-B. et al. Identifying neural substrates of competitive interactions and sequence transitions during mechanosensory responses in Drosophila. PLOS Genet. 16, e1008589 (2020).

7. Croteau-Chonka, E. C. et al. High-throughput automated methods for classical and operant conditioning of Drosophila larvae. eLife 11, e70015 (2022).

Across different scales of biological organization, biological systems often exhibit hierarchical tree-like organization. For networks with such structure, hyperbolic geometry provides a natural metric because of its exponentially increasing number of states. I will describe how the use of hyperbolic geometry can be helpful for visualizing and analyzing information acquisition and learning process from across biology, from viruses, to plants and animals, including the brain. We find that local noise causes data to exhibit Euclidean geometry on small scales, but that at broader scales hyperbolic geometry becomes visible and pronounced. The hyperbolic maps are typically larger for datasets of more diverse and differentiated cells, e.g. with a range of ages. I will describe data showing that neural responses in the hippocampus have a low-dimensional hyperbolic geometry with curvature optimized for the number of available neurons. With time, hyperbolic geometry of neural representations expanded with radius that increased logarithmically with time. This dependence matches the maximal rate of information acquisition by a maximum entropy discrete Poisson process, further implying that neural representations continue to perform optimally as they change with experience. Finally, I will describe the connection between hyperbolic geometry and Zipf’s law.

References:

1. Zhang, Rich, Lee, and Sharpee Nature Neuroscience 2023

2. Sharpee et al Current Opinion in Neurobiology 2019

Recent studies reveal the vital role of cellular mechanical properties. Live cell imaging helps us explore these properties and their impact on cell function. In this workshop, we present two topics: 

1. Explore the link between cell surface tension and cytoplasmic division. Actomyosin contractility, driven by non-muscle myosin II and actin filaments, influences cell processes like motility and cytokinesis. Our optogenetic method reveals that relaxing cell surface tension during cytokinesis accelerates mitotic furrow entry, highlighting the antagonistic relationship between contractile ring tension and cell surface tension.

2. Investigate cytoplasmic fluidity in cell dormancy and germination. Dormancy is when cells halt proliferation in low-nutrition conditions. We found that fission yeast spores reduce cytoplasmic fluidity during dormancy and restore it upon germination. Trehalose accumulation during sporulation reduces fluidity, which rapidly normalizes upon germination. We'll discuss spore cytoplasm structure and its properties through particle tracking analysis.

References:

1. Yamamoto K, Miura H, Ishida M, Mii Y, Kinoshita N, Takada S, Ueno N, Sawai S, Kondo Y, Aoki K. Optogenetic relaxation of actomyosin contractility uncovers mechanistic roles of cortical tension during cytokinesis. Nat Commun. 2021;12: 1–13. 

2. Sakai K, Kondo Y, Goto Y, Aoki K. Cytoplasmic fluidization triggers breaking spore dormancy in fission yeast. bioRxiv. 2023. p. 2023.09.27.559686. 

Tissue repair, immune defense, and cancer progression rely on a vital cellular decision of whether to proliferate or stay in quiescence. Mammalian cells commit to proliferation by triggering a positive feedback whereby the transcription factor E2F activates cyclin-dependent kinase 2 (CDK2), which phosphorylates the E2F inhibitor retinoblastoma (Rb) leading to a further increase in E2F activity to express the genes needed for proliferation. How cells manage to trigger the positive feedback only when needed is a fundamental question since positive feedbacks can inadvertently amplify small perturbations as illustrated by neuronal death by excitotoxicity or hyperinflammatory immune responses. Here we use single-cell analysis of E2F and CDK2 activity dynamics to determine how cells control the positive feedback to safeguard proliferation commitment. Strikingly, cells spend variable times of a few hours to over 20 hours in a reversible state of intermediate E2F activity. The intermediate E2F activity is proportional to the amount of phosphorylation of an evolutionary conserved Threonine 373 (T373) site in Rb. The Rb T373 site is phosphorylated by CDK4/6 or CDK2 before the other Rb sites due to its much slower dephosphorylation rate compared to the other sites. Cells exit this intermediate E2F activity state by slowly dephosphorylating T373 and returning to quiescence, or by increasingly engaging the positive feedback between E2F and CDK2 to fully phosphorylate Rb and start proliferation. Only full phosphorylation releases Rb from chromatin while T373 phosphorylated Rb remains chromatin bound. Together, our study identifies a dedicated molecular state of intermediate E2F activation in which cells integrate fluctuating signals to reliably decide whether to disengage or fully engage the positive feedback that flips the Rb-E2F switch and initiates cell proliferation.

Genetically encoded biosensors that report on signaling dynamics in single living cells have revolutionized our understanding of signaling networks. Here, I present novel technological implementations that showcase the power and promise of biosensor imaging. Previously, we showed that dynamically perturbing signaling networks at the exact timescale at which they fluctuate provides unique information about their circuitry. To unlock the full potential of this approach, I describe a genetic circuit comprising a receptor tyrosine kinase (RTK) optogenetic actuator and a spectrally orthogonal biosensor that reports on the activity of the MAPK ERK. This enables us to challenge cells with highly standardized temporal RTK inputs and measure the resulting ERK signaling outputs. Coupled with microscope automation and automated image analysis, this allows for high experimental throughput in studying single-cell signaling dynamics. This leads to large amounts of biosensor time-series biosdatasets that escape visual inspection by humans. To enable scalable data analysis, we present a machine learning approach that enables data-driven analysis of single-cell biosensor timeseries. We showcase how this approach allowed us to analyze large datasets of single-cell ERK dynamics in response to several perturbations, and provide new insight in how the MAPK network is wired to control fate decisions. Additionally, I report on a real-time feedback microscopy platform dedicated to optogenetics that provides on-the-fly image analysis and subsequent automated optogenetic illumination with any desired spatio-temporal light input. I showcase how this platform can be used to study spatio-temporal Rho GTPase and MAPK/ERK network circuitry.

Almost all biological systems possess the ability to gather environmental information and modulate their behaviors to respond to changing environments adaptively. While animals excel at sensing odors, even simple bacteria can detect faint chemicals using stochastic receptors. They then navigate towards or away from the chemical source by processing this sensed information through intracellular reaction systems. 

In the first half of our talk, we demonstrate that the E. coli chemotactic system is optimally structured for sensing noisy signals and controlling taxis. We utilize filtering theory and optimal control theory to theoretically derive this optimal structure and compare it to the quantitatively verified biochemical model of chemotaxis [1,2].

In the latter half, we discuss the limitations of traditional information theory, filtering theory, and optimal control theory in analyzing biological systems. Notably, all biological systems, especially simpler ones, have constrained computational resources like memory size and energy, which influence optimal behaviors. Conventional theories don't directly address these resource constraints, likely because they emerged during a period when computational resources were continually expanding. To address this gap, we introduce the "memory-limited partially observable optimal control," a new theoretical framework developed by our group, and explore its relevance to biological problems [3,4].

References:

1. K. Nakamura, T. J. Kobayashi, "Connection between bacterial chemotactic network and optimal filtering", Phys. Rev. Lett., vol.126 (12):128102 (2021)

2. K. Nakamura, T.J. Kobayashi, "Optimal sensing and control of run-and-tumble chemotaxis", Phys Rev Res. , vol.2 (4): 013120 (2022)

3. T. Tottori, T. J. Kobayashi, "Memory-Limited Partially Observable Stochastic Control and its Mean-Field Control Approach",  Entropy, vol.24 (11): 1599 (2022)

4. T. Tottori, T. J. Kobayashi, "Forward-Backward Sweep Method for the System of HJB-FP Equations in Memory-Limited Partially Observable Stochastic Control", Entropy, vol. 25 (2) :208 (2023)

In this work, we study the intricate network of signaling pathways, demonstrating their active role in resolving conflicts. Through a combination of quantitative experiments and mathematical modeling, we investigate how cells integrate contradictory signals. Our findings reveal novel molecular mechanisms within the signaling networks, shedding light on how cells mitigate ambiguities in response to simultaneous TGFβ and BMP pathway signaling. We identified new, unexpected asymmetric crosstalk biased towards the TGFβ pathway. The resulting integration pattern reduced the ambiguity in pathway activity compared to the extracellular environment. We find this effect to be general, holding across genes, ligand variants, and cell types. Utilizing mathematical models, we predict and experimentally validate a combinatorial interaction between the pathways. Specifically, through combinatorial dimerization of pathway mediators, BMP pathway proteins are rerouted to activate TGFβ targets. This underscores the computational capabilities of signaling pathways in interpreting the cellular environment. Our study extends beyond elucidating cellular responses to conflicting signals, offering a broader perspective on signaling pathways as computational entities capable of shaping cellular decisions during the determination of cell fate. This work opens new avenues for further exploration and control of cellular behaviors in diverse contexts.

The ability of organisms to perform behavioral tasks is limited by how much relevant information they extract from signals, and how efficiently they use that information to perform the task. I will report on our recent theoretical and experimental efforts to address these two questions by calculating theoretical bounds on signal detection and information usage efficiency and by measuring how close to these bounds E. coli cells operate when they are climbing shallow chemical gradients. 

Acknowledgments: 

Theory and measurements were carried out by Henry Mattingly, Keita Kamino, Rafaela Kottou and Jude Ong under the supervision of Ben Machta and Thierry Emonet. This work was supported by the Alfred P. Sloan Foundation under grant G-2023-19668 (HM, TE, BB); by NIH awards R01GM106189 and R01GM138533 (TE, HM, RK), and R35GM138341 (BM); by Simons Investigator Award 624156 (BM); and by PRESTO award JPMJPR21E4 Award (KK and JO).