CTransformer: Deep-transformer-based 3D cell membrane tracking with subcellular-resolved molecular quantification

Deep learning segmentation and fluorescence imaging techniques allow the cellular morphology of living embryos to be constructed spatiotemporally. These development processes involve numerous molecules distributed at the subcellular scale, such as cell adhesion (E-cadherin), which accumulate at cell-cell interfaces to regulate intercellular connection. However, quantifying molecular distributions within specific subcellular regions across the entire embryo, where cell movement and molecular redistribution occur rapidly, is challenging due to the need for simultaneous cell morphology reconstruction and lineage tracing due to photobleaching and phototoxicity. We report a transformer-based pipeline, CTransformer, that establishes a 4D cellular morphology map before the 550-cell (late) stage. CTransformer constructed 4D cellular morphology atlases, reaching 80% accuracy at the 550-cell stage. Through this advanced architecture, we use only one channel to reconstruct cell morphology and achieve cell tracing. With each cell’s morphology as a reference, the distribution of specific molecules throughout the cell body and at cell interfaces can be quantitatively measured in another fluorescence channel. We apply this methodology to track E-cadherin during embryonic development of the worm Caenorhabditis elegans, from fertilization to gastrulation. Our results reveal that E-cadherin is tightly regulated across individual embryos, both within single cells and at cell-cell interfaces, displaying an anterior-posterior gradient and cell- and lineage-specific patterns. Furthermore, its spatiotemporal heterogeneity influences cell mechanics and embryonic morphogenesis, helping explain how C. elegans achieves stereotypical developmental patterns at cellular resolution.

💡 Research Summary

CTransformer is a novel deep‑learning pipeline that simultaneously reconstructs 4D cellular morphology and lineage from a single fluorescence channel (membrane label) while enabling quantitative measurement of a second molecular marker across the entire embryo. The framework consists of three core modules. The first, TUNETr (Topology‑constraint U‑shape Nucleus‑prompting Euclidean distance transform Transformer), is a Swin‑UNETR‑inspired U‑shaped Transformer that incorporates relative positional attention, a topology‑preserving loss, and boundary‑aware semi‑supervised learning. These design choices allow accurate voxel‑wise segmentation of cell membranes even with limited ground‑truth annotations, noisy axial resolution, and densely packed cells typical of late‑stage C. elegans embryos.

The second module, m2nGAN, is a generative adversarial network that translates the membrane‑segmented volumes into synthetic “pseudo‑nuclei” images. These artificial nuclei can be fed directly into conventional nucleus‑based lineage‑tracing algorithms, thereby eliminating the need for a separate nuclear fluorescence channel. The authors demonstrate that lineage trees can be reliably reconstructed up to the 550‑cell stage with >80 % segmentation accuracy.

The third module, MolQuantifier, takes a third fluorescence channel (e.g., HMR‑1/E‑cadherin) and, using the morphology and lineage maps as spatial references, quantifies signal intensity within each cell body and specifically at cell‑cell contact interfaces. This enables subcellular‑resolution spatiotemporal maps of molecular distribution.

To train and evaluate the system, the authors assembled a comprehensive dataset of 34 compressed and 14 uncompressed C. elegans embryos, comprising 16,922 volumetric time‑lapse images (dual‑channel: membrane + nucleus). From these, 6,279 cell objects were used for TUNETr training and 2,339 for m2nGAN training, while an additional set of 30,509 cells served for benchmarking. Compared against nine state‑of‑the‑art segmentation networks, CTransformer consistently achieved higher Dice (≈0.82) and IoU (≈0.74) scores, particularly excelling in high‑density regions where other methods often fail.

Applying the pipeline to track E‑cadherin (HMR‑1) dynamics, the authors uncovered a robust anterior‑posterior gradient of adhesion protein concentration and lineage‑specific enrichment at particular cell‑cell interfaces. These spatial patterns correlated with variations in cell mechanics and morphogenetic movements, providing mechanistic insight into how stereotyped C. elegans development is achieved at single‑cell resolution.

Key technical strengths include: (1) reduction of photobleaching and phototoxicity by requiring only one membrane channel for morphology; (2) Transformer‑based attention that captures long‑range spatial dependencies; (3) topology‑constrained loss and semi‑supervised training that mitigate limited annotation; (4) GAN‑generated pseudo‑nuclei that preserve compatibility with existing lineage tools; and (5) modular design allowing easy extension to additional molecular markers. Limitations are noted for very late stages (>800 cells) where membrane contrast diminishes, potential artifacts in synthetic nuclei, and the current focus on C. elegans, which may require domain adaptation for other organisms.

In summary, CTransformer delivers a high‑throughput, accurate, and biologically gentle solution for whole‑embryo 4D cell segmentation, lineage reconstruction, and subcellular molecular quantification. It opens new avenues for developmental biology, biophysics, and quantitative imaging studies that demand simultaneous morphological and molecular information at cellular resolution.