NextCell is an interactive browser for a single-cell lineage tree of the E13.5 mouse embryo — 1,281,141 profiled cells, reconstructed from DNA Typewriter recordings and dated in embryonic time. Every tip is one cell; every internal node is a dated common ancestor.
The first two are the trees themselves; the other three are analyses built on them. Every view is linkable — the bar at the top of the browser always holds a URL that encodes the current tab and all of its settings.
Every QC-passing cell, placed onto the backbone. Click a clade to see its consensus edit pattern, its cell-type composition, and its cells projected onto the UMAP.
≈43 MB ge7 Backbone → 655,701cells · reference coreThe placement-free constrained backbone that every analysis is anchored to. Splits into the two blastomeres at E1.5, 410,925 cells and 244,776 cells.
≈24 MBWhere every cell in the embryo was, at every moment — at germ-layer, trajectory or cell-type resolution.
≈4 MB Origins → 82cell typesFor one E13.5 cell type, how many ancestral lineages it descends from over time, and what those ancestors were.
≈6 MB Traceback → 275qualifying pathsThe qualifying paths into one cell type, as separated bars, a continuous stream, or one non-crossing row per path.
≈12 MB| Tree | Tips | Internal nodes | Notes |
|---|---|---|---|
Whole embryomerged_full_placed.nwk |
1,281,141 | 1,109,894 | All QC-passing cells, placed onto the backbone. 90.9% bifurcating, 9.1% polytomies. |
ge7 Backbonemerged_minB2h_lineage_constrained.nwk |
655,701 | 648,928 | Placement-free reference core; the constrained backbone all analyses are anchored to. |
Both are dated in embryonic time from the zygote (E0) through two blastomeres (E1.5) to all tips at E13.5. The dating places 656 lineages at E6.5; branch lengths are continuous, and the viewer's time axis is computed from them directly.
Each profiled cell carries a cell-type call, a major-trajectory assignment and a germ-layer assignment transferred from the mouse embryo atlas of Qiu et al. 138 cell types, 25 trajectories and 6 germ layers are represented. Germ-layer assignments are lineage-validated; 118 of the 138 types have a validated assignment; the rest are marked excluded or are new to this callset.
Ancestral cell types were imputed by walking up the tree in half-day steps and matching each node to the atlas in UMAP space with a mutual-nearest-neighbour strategy. 926,852 of 1,109,894 internal nodes (83.5%) carry a call.
These per-node calls are not shown in the two tree browsers. Individually they are too noisy to read one at a time, so the browsers display only what is reconstructed — topology, dating, edit patterns, and the composition of a clade's own profiled cells. The imputed layer is used only in the three analysis views, and only behind an explicit filter.
Two properties of the layer bound what those views can say, and are stated plainly rather than buried:
Reversible transitions. Of cell-type pairs with substantial traffic, ~32% are near-reversible (A→B about as often as B→A), which is not biologically possible at that scale and reflects the inference oscillating between co-resident labels. The Traceback view therefore shows only qualifying paths: heterotypic and non-recursive, carrying at least 1% of that cell type's traceable cells and at least 5 cells, with every step directionally asymmetric across the tree (at least 50 observed ancestor→descendant transitions between the two states, the observed direction favoured at least 60:40). All three criteria use only the phylogeny and the imputed labels. This leaves 275 paths across 82 of 136 cell types, covering 117,704 cells; all of them are browsable here, and a further 23 cohorts retain no qualifying path at all.
Systematic confusions survive filtering. A residual class of transitions is frequent, directional, reproducible across both blastomeres — and still crosses a germ-layer boundary. These are systematic atlas-matching errors rather than sampling noise, and no internal statistic separates them from real biology.
Promiscuous sink types. Telencephalon, Glutamatergic neurons, GABAergic cortical interneurons, GABAergic neurons, Sclerotome and Fibroblasts absorb a disproportionate share of near-reversible traffic — between them just over half of it; treat any edge terminating in them with extra caution.
Both trees are written directly from the same artifacts this browser is built from, so their topology, dating and node numbering are identical to what you see on screen. Newick format, gzipped. Branch lengths are in days, so a node's age is the sum of branch lengths from the root: the root sits at E0.0 and every tip at E13.5 (the trees are ultrametric to within 3.2 seconds).
Labels. Tip labels are cell_id values, the join key for the single-cell
metadata. Internal labels are n<id>, where the id is the node's preorder index
and matches the node numbering used by this browser — note that the full tree and the backbone
tree number their nodes independently, so n1234 is not the same node in both
files. These are identifiers only: the imputed ancestral cell-type calls are deliberately not
included in these files, since they are model output rather than reconstruction, and are
better read in context in the browser.
| Item | Location |
|---|---|
| Dated lineage trees (full + backbone) | Above, on this page |
| Analysis and figure code | github.com/seidels/dtt-mouse-analysis |
| Raw sequencing (TAPE amplicon libraries) | GEO GSE341627 Private until release — use reviewer token gbwpuyeqhrypxyt |
| Single-cell expression matrices | GEO GSE341627, and as
gene-count matrices (Matrix Market .mtx plus per-cell
and per-gene .csv, 7 sequencing runs, ~15 GB) |
| Processed cell metadata (cell type, trajectory, UMAP) | cell_metadata.annotation.txt · 150 MB |
| Ancestral cell-type calls (per internal node) | Coming — linked here |
| Qualifying traceback paths | Table S7 of the preprint |
| Reference atlas | Qiu et al., single-cell atlas of mouse embryogenesis (E8–E13.5) |
| License | Data CC BY 4.0 · code MIT |
Yu Q*, Kim H*, Seidel S*, Acosta-Clark JF, Martin BK, O’Connor K, Daza RM, Gasperini M, Nathans JF, Lam M, Gamo E, VijayKumar S, Kuo L, Lalanne J-B, Simeonov K, Pepper M, Trapnell C, Gray JM, Choi J, Qiu C#, Shendure J#. In vivo reconstruction of the cell lineage history of a developing mouse with DNA Typewriter, from zygote to late organogenesis. bioRxiv (2026). https://doi.org/10.64898/2026.07.29.741625
* Qi Yu, Haedong Kim and Sophie Seidel contributed equally. # Correspondence: [email protected], [email protected]
We thank the members of the Shendure lab (particularly the Recording subgroup), as well as the Seattle Hub for Synthetic Biology, Allen Discovery Center for Cell Lineage, and Department of Genome Sciences communities for helpful discussions over the course of many years, particularly A. Schier, M. Elowitz, and B. Waterston. We also thank the Transgenic Colony Management, Neurosurgery & Behavior, Lab Animal Services, and the Veterinary Services teams at the Allen Institute for their contributions to animal work.
The University of Washington has filed a patent application related to DNA Typewriter, on which J.C. and J.S. are listed as inventors. J.S. is on the scientific advisory board, a consultant, and/or a co-founder of 10x Genomics, Cellular Intelligence, Guardant Health, Pacific Biosciences and Phase Genomics. All other authors declare no competing interests.
We disclose that data exploration, data analysis, coding and manuscript writing were supported by AI-based tools. The authors take full responsibility for the data, code, analyses, conclusions and writing. This browser was built with Claude Code.
This browser owes an obvious debt to Nextstrain, and the NextCell mark is an intentional homage to theirs: a disc of flat, spectrally coloured facets. The geometry is our own — where Nextstrain’s mark is a globe, ours is a radial dendrogram, one cell at the centre dividing outward, which is how Fig. 4A of the paper draws this tree. The artwork is original and NextCell is not affiliated with, endorsed by, or connected to the Nextstrain project.
NextCell · next-cell.org · E13.5 mouse embryo TAPE lineage tree · built on the v8 constrained backbone. Version 8 of the browser.