Eric J. Deeds

Developmental variation has often been appreciated as a major phenotypic substrate of evolutionary innovation in multicellular organisms. Although quantification of developmental variation in isogenic populations has a long history, it has been difficult to accomplish across natural populations until recently. Given its invariant cell lineage, Caenorhabditis elegans represents an ideal model system for studying how natural diversity in development is expressed at the resolution of single cells, cell lineages, and the whole embryo. Using embGAN, a deep-learning pipeline that enables label-free lineage tracing without the bottleneck and potential confounding effects of generating transgenic strains, we reconstructed the early embryonic lineage of 78 embryos spanning eight genetic backgrounds, including six wild strains, and quantified cell-cycle timing with a lineage-aligned graph metric that separates global rescaling of developmental rate from localized changes in lineage structure. We found that cell-cycle timing differs reproducibly among genotypes, with between-strain variation consistently exceeding within-strain variation. The magnitude of this variation was itself genotype-specific and not correlated with levels of genetic divergence. Unlike the global shifts in the tempo of development that dominate divergence between species, these differences are predominantly in lineage patterning, changing the timing of specific divisions, and they align to cell fate. Thus, the invariant lineage of C. elegans is built on a structured, genotype-specific landscape of developmental variation, a potential source of trait variation from which drift in developmental systems can emerge.

Fibrosis severity is routinely inferred from collagen abundance, although whether collagen quantity determines pathological fibrosis remains unclear. In Duchenne muscular dystrophy (DMD), chronic muscle injury and inflammation drive extracellular matrix accumulation, making these processes difficult to disentangle. We exploit sarcospan overexpression in mdx mice, a model of DMD ( mdx ^TG ), which improves membrane integrity and muscle function despite persistent matrix remodeling. mdx ^TG muscle accumulates more collagen than mdx yet lacks its dense macrophage-rich scars. Matrisome proteomics and spatial transcriptomics reveal compositionally and spatially distinct matrix states, while decellularized mdx ^TG matrix protects myotubes from membrane damage relative to mdx matrix. Despite these differences, both dystrophic matrices remain stiff and induce nuclear YAP in fibro-adipogenic progenitors. Verteporfin suppresses collagen production and reduces fibrosis in vivo , while nuclear YAP is increased in FAPs from patients with DMD. Thus, collagen abundance alone does not define pathological fibrosis; matrix organization, biological activity, and mechanosignaling distinguish functionally distinct fibrotic states.

Tubuloglomerular feedback (TGF) is essential for the renal auto-regulation of flow. TGF is known to induce spontaneous oscillations in single-nephron tubular fluid flow in male kidneys. However, male-female differences in this dynamic behavior have not been studied. Leveraging intravital two-photon microscopy, resting-state magnetic resonance imaging, ultrasound-based and transdermal recordings, we found TGF-mediated oscillations across spatial scales in the rodent kidney, from single-nephron to whole-organ levels, and that male kidneys exhibited higher operating frequencies than females. To understand the mechanisms involved, we developed a dynamical systems model of TGF that agrees with physiological observations. Analysis of the mathematical model indicated that higher reabsorption rate and fluid flow efficiency in male proximal tubules not only result in higher frequencies, but also render male nephrons more susceptible to lose TGF-mediated oscillations. Furosemide abolished TGF-mediated oscillations in male kidneys and upregulated tubular injury marker KIM-1, suggesting that the propensity to lose TGF-mediated oscillations underlies the heightened risk for injury in males. Our analysis also indicated that SGLT-2 inhibition confers renoprotection by preserving TGF-mediated oscillations in hyperglycemia. Combining quantitative imaging and mathematical modeling, this study provides mechanistic insights into the transition from normal physiology to pathophysiology in the kidney.

Manifold learning builds on the “manifold hypothesis,” which posits that data in high-dimensional datasets are drawn from lower-dimensional manifolds. Current tools generate global embeddings of data, rather than the local maps used to define manifolds mathematically. These tools also cannot assess whether the manifold hypothesis holds true for a dataset. Here, we describe DeepAtlas, an algorithm that generates lower-dimensional representations of the data’s local neighborhoods, then trains deep neural networks that map between these local embeddings and the original data. Topological distortion is used to determine whether a dataset is drawn from a manifold and, if so, its dimensionality. Application to test datasets indicates that DeepAtlas can successfully learn manifold structures. Interestingly, many real datasets, including single-cell RNA-sequencing, do not conform to the manifold hypothesis. In cases where data is drawn from a manifold, DeepAtlas builds a model that can be used generatively and promises to allow the application of powerful tools from differential geometry to a variety of datasets.

Manifold learning builds on the "manifold hypothesis," which posits that data in high-dimensional datasets are drawn from lower-dimensional manifolds. Current tools generate global embeddings of data, rather than the local maps used to define manifolds mathematically. These tools also cannot assess whether the manifold hypothesis holds true for a dataset. Here, we describe DeepAtlas, an algorithm that generates lower-dimensional representations of the data's local neighborhoods, then trains deep neural networks that map between these local embeddings and the original data. Topological distortion is used to determine whether a dataset is drawn from a manifold and, if so, its dimensionality. Application to test datasets indicates that DeepAtlas can successfully learn manifold structures. Interestingly, many real datasets, including single-cell RNA-sequencing, do not conform to the manifold hypothesis. In cases where data is drawn from a manifold, DeepAtlas builds a model that can be used generatively and promises to allow the application of powerful tools from differential geometry to a variety of datasets.

Manifold learning builds on the “manifold hypothesis,” which posits that data in high-dimensional datasets are drawn from lower-dimensional manifolds. Current tools generate global embeddings of data, rather than the local maps used to define manifolds mathematically. These tools also cannot assess whether the manifold hypothesis holds true for a dataset. Here, we describe DeepAtlas, an algorithm that generates lower-dimensional representations of the data’s local neighborhoods, then trains deep neural networks that map between these local embeddings and the original data. Topological distortion is used to determine whether a dataset is drawn from a manifold and, if so, its dimensionality. Application to test datasets indicates that DeepAtlas can successfully learn manifold structures. Interestingly, many real datasets, including single-cell RNA-sequencing, do not conform to the manifold hypothesis. In cases where data is drawn from a manifold, DeepAtlas builds a model that can be used generatively and promises to allow the application of powerful tools from differential geometry to a variety of datasets.

The dystrophin-glycoprotein complex (DGC) is composed of peripheral and integral membrane proteins at the muscle cell membrane that link the extracellular matrix with the intracellular cytoskeleton. While it is well established that genetic mutations that disrupt the structural integrity of the DGC result in numerous muscular dystrophies, the 3D structure of the complex has remained elusive. Two recent elegant cryoEM structures of the DGC illuminate its molecular architecture and reveal the unique structural placement of sarcospan (SSPN) within the complex. SSPN, a 25 kDa tetraspanin-like protein, anchors β-dystroglycan to the β-, γ- and δ-sarcoglycan trimer, supporting the conclusions of biochemical studies that SSPN is a core element for DGC assembly and stabilization. Here, we advance these studies by revealing that SSPN provides scaffolding in δ-sarcoglycanopathies, enabling substitution of δ-sarcoglycan by its homolog, ζ-sarcoglycan, leading to the structural integrity of the DGC and prevention of limb-girdle muscular dystrophy R5. Three-dimensional modeling reveals that ζ-sarcoglycan preserves protein-protein interactions with the sarcospan, sarcoglycans, dystroglycan, and dystrophin. The structural integrity of the complex maintains myofiber attachment to the extracellular matrix and protects the cell membrane from contraction-induced damage. These findings demonstrate that sarcospan prevents limb-girdle muscular dystrophy R5 by remodeling of the sarcoglycan complex composition.

Effective analysis of single-cell RNA sequencing (scRNA-seq) data requires a rigorous distinction between technical noise and biological variation. In this work, we propose a simple feature selection model, termed “Differentially Distributed Genes” or DDGs, where a binomial sampling process for each mRNA species produces a null model of technical variation. Using scRNA-seq data where cell identities have been established a priori, we find that the DDG model of biological variation outperforms existing methods. We demonstrate that DDGs distinguish a validated set of real biologically varying genes, minimize neighborhood distortion, and enable accurate partitioning of cells into their established cell-type groups.

Cells employ many large macromolecular machines for the execution and regulation of processes that are vital for cell and organismal viability. Interestingly, cells cannot synthesize these machines as functioning units. Instead, cells synthesize the molecular parts that must then assemble into the functional complex. Many important machines, including chaperones like GroEL and proteases like the proteasome, are comprised protein rings that are stacked on top of one another. While there is some experimental data regarding how stacked-ring complexes like the proteasome self-assemble, a comprehensive understanding of the dynamics of stacked ring assembly is currently lacking. Here, we developed a mathematical model of stacked trimer assembly, and performed an analysis of the assembly of the stacked homomeric trimer, which is the simplest stacked ring architecture. We found that stacked rings are particularly susceptible to a form of kinetic trapping that we term “deadlock,” in which the system gets stuck in a state where there are many large intermediates that are not the fully-assembled structure, but that cannot productively react. When interaction affinities are uniformly strong, deadlock severely limits assembly yield. We thus predicted that stacked rings would avoid situations where all interfaces in the structure have high affinity. Analysis of available crystal structures indicated that indeed the majority – if not all – of stacked trimers do not contain uniformly strong interactions. Finally, to better understand the origins of deadlock, we developed a formal pathway analysis and showed that, when all the binding affinities are strong, many of the possible pathways are utilized. In contrast, optimal assembly strategies utilize only a small number of patwhays. Our work suggests that deadlock is a critical factor influencing the evolution of macromolecular machines, and provides general principles for not only understanding existing machines but also for the design of novel structures that can self-assemble efficiently. Statement of Significance Understanding the assembly macromolecular machines is important for understanding a wide range of cellular processes. Here, we use mathematical models to study the assembly of stacked rings, which are a common motif in these machines. Our models revealed that these complexes can readily get “stuck” during assembly when the binding affinity between subunits is too strong. This suggests an evolutionary pressure to favor weaker interactions, and our analysis of solved structures confirmed this prediction. Our findings not only contribute to the fundamental understanding of assembly but also offer insights into the evolutionary pressures shaping the architecture of stacked rings, and have implications for both cell and synthetic biology.