Poster abstracts

A1 - Simulating actin networks in synaptic spine heads using Dynamical Graph Grammars

Matthew Hurr and Eric Mjolsness

There is a morphodynamic component to synaptic learning by which dendritic (postsynaptic) spine head size changes can strengthen or weaken the synaptic connection between two neurons, in response to the temporal correlation of signals from the axon of the presynaptic neuron and the soma of the postsynaptic neuron. These morphological factors are in turn sculpted by the graph-like dynamics of the actin cytoskeleton. In this project, we seek to use Dynamical Graph Grammars (DGGs) [1,2] implemented within a computer algebra system to model how networks of actin filaments can dynamically grow or shrink and reshape the spine head. We designed and implemented several DGG sub-grammar mathematical models including actin network growth, anisotropic filament forces, filament-membrane mechanical interaction, and Hessian Boltzmann sampling of random molecular displacements, to regulate the generation and deletion of graph objects. From first principles expressed in about a dozen DGG rules we simulate emergent biomechanics of a simplified network of actin polymers and its interaction with membrane, using very rough parameter estimates, all in two dimensions. We refined this model by incorporating rate constants used in previous models (e.g. [3,4,5]) in each sector, and we are currently recapitulating the previously observed sub-grammar behaviors.

A3 - Semi-Parametric Model Learning using the Pulsatile Stimulation of T Cells to inform Optimal Design of Immuno-therapeutic Activation Protocols

Ke Xu and Jun Allard

T cells - both natural and engineered for therapeutics – are activated by receiving antigenic signals. Recently, multiple research groups have found that stimulating T cells with short pulses of antigen yields categorically distinct activation, in some cases stronger than stimulating them continuously. We use these observations to learn and constrain mathematical models of activation. The model uses semi-parametric neural net differential equations, in which individual terms in the dynamical system are universal neural nets, an approach that allows learning on intermediate data size, obviates the bias arising from choosing functional forms as in traditional differential equation modeling, and yet exposes the signaling network connectivity in a way “closed box” neural nets do not. We find that a single model explains how T cells variously demonstrate band-stop, band-pass, high-pass and low-pass frequency response patterns. These nonlinear frequency responses allow us to predict an optimal activation protocol, i.e., to select the optimal duration, number, and spacing of pulses to maximize T cell activation.

A5 – Divergent Wnt Signaling States Model Distinct Colorectal Cancer Subtypes

Linzi Hosohama, Delia Tifrea, John Lowengrub, Rob Edwards, and Marian L. Waterman

The most common genetic mutations in colorectal cancer (CRC) are those that activate the Wnt signal transduction pathway with over 80% of CRCs harboring such mutations. And yet despite this common mutation pathway to cancer, there are multiple CRC subtypes that differ in their Wnt signature as well as their prognosis and response to therapies. How these subtypes arise, whether they co-exist in tumors, and importantly, how their individual signatures influence the tumor microenvironment and patient prognosis is unknown. To model these differences, we isolated two subtypes from a single, well-studied CRC cell line and carried out RNAseq and xenografting analysis. One subtype is characterized by high Wnt reporter activity (“Wnt-hi”). This subtype also expresses a strong Yap and fetal intestine regeneration transcription signature. Single cell RNAseq analyses of xenografts derived from these cells reveal an inflammatory tumor microenvironment with abundant stromal cell types. In contrast, a “Wnt-lo” subtype from the same cell line (from the same patient tumor and with identical oncogenic mutations) is clearly different with a MYC-proliferative and neuronal/neuroendocrine signature. These cells create highly proliferative, aggressive and immune-suppressive tumors. Paradoxically, “Wnt-hi” and “Wnt-lo” signatures do not correlate with the level of beta-catenin protein, which is similar between the two subtypes. Rather, we propose that distinct signaling networks in each subtype provide different contexts for beta-catenin activity and the set of Wnt target genes that it regulates. While these contexts could represent potential sensitivities to metabolic or kinase inhibitors, interference with these networks might alternatively cause conversion to other CRC subtypes – and an outcome of apparent drug resistance (e.g. Yan et al. 2023. “Multiscale Modeling of Cell-Cell Interactions In Colon Cancer Reveals How Patterned Heterogeneity Emerges And Influences Growth And Response To Treatment”).

A7 – Single-cell Analysis Implicates Early Dysregulation of Nanog in the Production of Syndromic Heart Defects

Stephenson Chea, Arianna G. Favela, Arthur D. Lander, and Anne L. Calof

Approximately 30% of individuals with Cornelia de Lange Syndrome (CdLS) exhibit congenital heart defects (CHDs). The most common form of CdLS is caused by haploinsufficiency for NIPBL, which encodes a cohesin-associated protein. In Nipbl+/- mice, NIPBL-haploinsufficiency causes hundreds of small gene expression changes in every cell and CHDs similar to those in CdLS. Using the Nipbl+/- mouse as a model system for discovering new causal factors for CHDs, we documented abnormal heart development by cardiac crescent (CC)stage, suggesting that CHDs originate as early as gastrulation. To investigate this, we performed single-cell RNA sequencing on both CC- and gastrulation-stage wildtype and Nipbl+/- mouse embryos. Strikingly, we found that Nipbl+/-embryos dramatically overexpressed Nanog. Nanog encodes a transcriptional repressor required for pluripotency in pre-implantation embryos and is normally transiently re-expressed during gastrulation. In Nipbl+/- mice, however, Nanog fails to shut off appropriately post-gastrulation, so that by CC-stage its levels are many-fold above normal. Interestingly, numerous gene expression changes detected in Nipbl+/- mice involve post-implantation-stage targets of Nanog. These include genes associated with pluripotency (Pou5f1/Oct4), left-right patterning (Tdgf1, Lefty2, Nodal), anterior-posterior patterning (Hox genes), and primitive erythropoiesis (Tal1, Lmo2, Hbb-bh1). Accompanying these changes are changes in the allocation of cells to clusters representing Mesp1-expressing cardiac progenitors, the first and second heart fields, and neural crest. These results suggest that a failure to downregulate Nanog expression after gastrulation, and the transcriptional dysregulation that ensues, lead to misallocation and/or dysfunction of cardiogenic cell populations. Funded by NIH-NHBLI.

A9 – Improved feature selection and correlation detection for single cell RNA sequencing

Kai Silkwood, Emmanuel Dollinger, Josh Gervin, Scott Atwood, Qing Nie, and Arthur D. Lander

Many approaches have been developed to overcome technical noise—reflecting both sparse sampling and variability in sequencing depth—in single cell RNA-sequencing (scRNAseq). For many applications—e.g., clustering similarly sized groups of markedly different cells—most methods perform reasonably well. Yet as researchers dig deeper into scRNAseq data—looking for rare cell types, distinguishing cell states, and learning gene regulatory networks, there is a need for greater accuracy, and lower dependence on ad hoc parameters and thresholds. Impeding this goal is a lack of consensus about the correct null distribution for scRNAseq data, which cannot be learned directly from data if actual ground truth is unknown (as is usually the case).

Here we approach this problem analytically, based on the assumption that scRNAseq data reflect only: cell heterogeneity (what we seek to characterize), transcriptional noise (temporal fluctuations randomly distributed across cells), and sampling error (i.e., Poisson noise). We then analyze scRNAseq data without normalization—as normalization can seriously skew distributions—and analytically calculate p-values associated with key statistics. We develop an improved method for feature selection (an early step in most scRNAseq pipelines) in which p-values associated with a modified corrected Fano factor provide a rigorous determination of which genes are meaningfully highly variable. Similarly, we use p-values associated with modified corrected Pearson correlation coefficients to identify significantly correlated and anti-correlated gene pairs. Using both simulated and real scRNAseq data sets, we show that this method, which we term BigSur (Basic Informatics and Gene Statistics from Unnormalized Reads) can lead to more reliable cell clustering, better rare cell discovery, and more accurate identification of significant, yet weakly-connected, gene-correlation networks.

A11 – A Multiscale Model of Immune Contributions to Right Ventricular Failure

Jose Bermudez and Naomi Chesler

The overarching goal of this research is to experimentally investigate the contribution of cardiac macrophages to RVF and then to use the experimental measurements to build a cardiac-immunological model module that will be incorporated into an established RVF multiscale model to predict the impact of abnormalities at different scales. In Aim 1, Mr. Bermudez will investigate the impact of increased ROS, due to pressure overload induced cardiomyocyte mitochondrial dysfunction, on macrophage activation, and add an immune cytokine dynamics module into the established RVF multiscale computational model. In Aim 2, Mr. Bermudez will investigate the contribution of macrophages to fibrotic remodeling, due to pressure overload induced cardiomyocyte mitochondrial dysfunction, and Mr. Bermudez will incorporate the dynamics of pro-fibrotic cytokines into the established multiscale computational modeling approach to improve predictions of RVF due to pressure overload.

A13 - Mathematically Modeling Glioblastoma and Radiotherapy: Studying the Cancer Stem Cell Hypothesis

Alice Vo and John Lowengrub

Glioblastoma is the most lethal and prevalent form of cancer to the central nervous system. Average life expectancy is fifteen months, and in that time the tumor evolves rapidly while modifying its microenvironment in the process, e.g. inducing hypoxia. When blasted with radiotherapy, it recurs quickly to almost its original size. One major cause of this recurrence is within the hierarchical structure of the tumor. In particular, it has populations of stem cells and differentiated cells. Prior work with cell lineage models demonstrates the salience of feedback on division in recurrence. To reflect recent biological insights in treatment of glioblastoma, the work we're doing seeks to extend that by adding rates and processes related to de-differentiation of non-stem cells back into stem cells. Exploration of this model may provide insight into the efficacy of treatments targeting such a molecule.

A15 – Phasor Unmixing to Reveal Organelle Organization and Cellular Response

Songning Zhu, Lorenzo Scipioni, and Michelle Digman

Organelles are a series of subcellular structures that perform various functions inside eukaryotic cells. The coordination between organelles play an important role in understanding the intricate biochemical processes within eukaryotic cells as organelles must work in concert to maintain cell function. Here, we perform simultaneous excitation of seven fluorophores and spectral phasor unmixing to achieve robust emission separation of all seven stained subcellular compartments and organelles including the Golgi apparatus, tubulin, lipid droplets, lysosomes, mitochondria, nuclear and mitochondrial DNA. Our imaging approach uses a single round of two-photon excitation at 780 nm coupled with multicolor organelle-directed staining to obtain multi-parametric physiological profiling of live MCF10A cells, a non-tumorigenic epithelial breast cancer cell line, under stress related treatments. We created an analysis pipeline to spectrally unmix the contribution of each organelle and obtain phenotypic signatures under each type of treatment. Ultimately, we characterized significant differences in organelle phenotypes upon application of a variety of treatments, including Antimycin A (mitochondrial respiration), hydrogen peroxide (oxidative stress), Nocodazole (microtubule depolarization) and serum starvation. The phasor approach provides a robust tool in separating up to seven simultaneous emission spectra and could be useful in separating even more components by complementing with lifetime imaging.

A17 – Does growth control predispose us to cancer?

Mika Caldwell and Arthur Lander

We often portray cancer as what happens when a renegade cell “escapes” control. To explain such events, we must first understand the strategies and goals of normal tissue growth control. Homeostasis (stability) is just one objective of many: tissues also attain precise sizes, regenerate rapidly and accurately from disturbances, minimize overshoots and oscillations, and compensate for stochasticity in the fates that daughter cells adopt after division.

These challenges are usually (and, we argue, necessarily) met by systems of collective integral negative feedback control—wherein a cell’s probability of self-renewal depends on signals from neighboring cells. Mathematical models explain how these systems achieve superb control, yet, when such models are expanded to account for stochasticity in patterns of cell division, and the fact that tissues occupy space (so that collective signals decay over distance), curious things happen: Rarely, controlled proliferation spontaneously switches to unrestrained growth. We find these events result from implicit positive feedback loops that arise when stochasticity and space interact. Similar “escape” phenomena can also result from explicit positive feedback loops, without a need for spatial effects.

These findings show that collective feedback control is structurally fragile: Superb control most of the time comes with a risk of total failure some of the time. Evolution likely adjusts parameters to reduce risk, but cannot remove it entirely. These results suggest that some cancer-causing agents (oncogenic mutations, environmental influences, genetic predispositions) may work less by directing individual cells to grow faster (or die less readily) than by altering the parameters of a feedback control system, boosting the probability of rare, random escape. I will discuss how this idea reflects on current puzzles in cancer biology, such as oncogene expression in phenotypically normal cells, tumor dormancy, and the kinetics of relapse after cancer therapy.

A19 – Widespread increase in enhancer-promoter interactions during developmental enhancer activation 

Zhuoxin Chen, Valentina Snetkova, Grace Bower, Sandra Jacinto, Benjamin Clock, Iros Barrozi, Brandon Mannion, Diane Dickel, Axel Visol, Len Pennacchio,  and Evgeny Kvon

Remote enhancers are thought to interact with their target promoters via physical proximity, yet the importance of this proximity for enhancer function remains unclear. Although some enhancer–promoter loops are cell-type-specific, others are stable across tissues or even display reduced physical proximity upon activation. A major challenge is that relatively few enhancers are functionally characterized in vivo, making it difficult to draw generalized conclusions about enhancer–promoter looping during developmental gene activation. Here, we investigate the 3D conformation of enhancers during mammalian development by generating high-resolution tissue-resolved contact maps for nearly a thousand mammalian enhancers with known in vivo activities in ten murine embryonic tissues. We performed enhancer knockouts in mice, which validated newly identified enhancer–promoter chromatin interactions. The majority of enhancers display tissue-specific 3D conformations, and both enhancer–promoter and enhancer–enhancer interactions are significantly stronger upon enhancer activation in vivo. Less than 14% of enhancer–promoter interactions form stably across tissues; however, corresponding enhancers still display highly-tissue-specific activities indicating that their activity is uncoupled from a chromatin interaction with promoters. We find that these invariant interactions form in the absence of the enhancer and are mediated by adjacent CTCF binding. We also identified putative in vivo target genes for enhancers linked to congenital malformations, neurodevelopmental disorders, and autism, demonstrating the utility of our dataset for understanding human congenital disorders.