Stem cell engineering

Pluripotent stem cells can be directed to become specific cell types, for example insulin-producing beta cells of the pancreas, dopaminergic neurons of the brain, and cardiomyocytes of the heart. Inventing the protocols, or step-by-step process, to direct stem cells to each these cell types took decades of research spread across dozens of academic labs around the world. Today, all of that work is starting to bear fruit as clinical trials are now underway to test the safety and efficacy of therapies based on pluripotent stem cells for diseases such as Parkinson's disease, systemic sclerosis, Huntington's disease.

However, much research remains to be done. We still do not know how to make hundreds of other cells types in the human body. A major challenge is determining whether the cells produced by a newly invented protocol are equivalent to their natural counterparts. A central goal of our research is to develop computational tools that address this challenge. For example, we invented CellNet, the first platform of its kind to score the fidelity of engineered cells and to prioritize the regulators that would improve them. Our follow-up web app PACNet extended CellNet to bulk RNA-seq data and simultaneously compares a user's protocol to other, published protocols. SingleCellNet extended assessment to single cell RNA-seq data and allowed for both cross-platform and cross species analysis.

Applied to hundreds of protocols, these tools revealed three recurring patterns of failure: residual starting-cell programs (or 'memory'), immaturity, and partial activation of off-target identities. Our studies also quantified the striking gains in target cell identity that follow their transplantation into an in vivo, native niche. Our work in developing genomewide, standardized metrics of cell engineering fidelity is leading to improved fate engineering protocols across many cell types.^{used in 25+ studies}

We are actively extending assessment metrics to data modalities other than transcriptomics, we are developing approaches to assess engineered behavior and function, and we are inventing new approaches to map engineered cells to embryonic progenitor states.

Network biology

A cell's identity is written in its gene regulatory network. We develop methods to reconstruct these networks from genomic data, to measure how they are established and rewired during development, and to simulate what happens when they are perturbed. Epoch infers networks whose topology changes over time. OneSC turns an inferred network into a stochastic dynamical model in which in silico knockdowns and overexpression play out along developmental trajectories. Our benchmarking work has set sober expectations for what expression data alone can and cannot reveal about regulation.

Networks also run between cells. We have reconstructed the signaling that surrounds implanted biomaterials, that organizes vascularized tissues, and that patterns organogenesis in the fly embryo, and we use these models to predict how a tissue responds when one of its cell types changes.

Development and diseases of the musculoskeletal system

The vertebrate limb begins as a bud of mesenchyme covered by a thin ectoderm. Signals from the ectodermal ridge and the posterior mesenchyme pattern the bud along its three axes as it grows, and within it the skeleton appears first as condensations of cartilage that lay out the template of every bone. Cartilage, bone, tendon, the tissues of the joint and much of the limb's connective tissue all descend from a single developmental source, the Prrx1-expressing limb bud mesenchyme.

We want to understand how so many tissues arise from one population. Some mesenchymal cells condense into the cartilage template while their neighbors stay outside it and become perichondrium, tendon or dermis. Within a forming joint, the interzone, cells that look alike go on to make articular cartilage on the bone ends or the ligaments that cross the joint space. We use single-cell genomics, gene regulatory network inference and perturbations in the embryo to find the regulators that make these choices.

Our work has centered on the synovial joint. We built the first single-cell transcriptional atlas of early joint development, which defined the interzone and its descendants and the signals that separate them, and we extended it through the later stages when articular cartilage, ligament and the synovial lining take shape, mapping the signaling pathways and transcription factor networks that drive each one. To test candidate regulators directly we developed an explant method in which gene expression is knocked down locally in the developing limb. Alongside this we study how sensory nerves and immune cells shape the skeleton and how bone marrow stromal cells maintain their multipotency.

Osteoarthritis is a failure of the tissues this program builds, and adult joint cells reactivate parts of the developmental program when they are injured. Our aim is to turn a precise map of how joint tissues are specified into a way to understand why they break down, to recognize the states that precede damage, and to steer cells back toward repair.