Developing Heart Cell Atlas

Summary of the Euploid Atlas

We conducted a multi-omic spatial profiling of the human euploid foetal hearts, aged between 4 and 20 post-conception weeks (PCW). We mapped the transcriptomic and epigenomic landscapes of cells and nuclei and identified 63 detailed cell types with distinct identities and location-specific signatures. To gain regional insights, some hearts were dissected into different anatomical regions — such as the node, great vessels, ductus arteriosus, or pericardium — before sequencing.

To resolve where these cell types sit in the tissue, we layered three complementary spatial technologies over the single-cell reference:

Visium SD 55 µm

Cellular niches mapped by inferring the abundance of each cell type per spot with cell2location.

Visium HD 2 µm

Sequencing-based, high resolution. The 2 µm bins were aggregated into cells with the bin2cell algorithm.

Xenium 5K panel

Imaging-based (5K Human Pan Tissue and Pathways Panel), segmented with the Multimodal morphology-based cell segmentation algorithm (10x Genomics).

For both high-resolution technologies (Visium HD and Xenium), we annotated cell types by combining label transfer from reference single-cell/nucleus data (CellTypist) and marker-driven manual annotation.

Trisomy 21 Data

11–14 PCW age of the Trisomy 21 hearts profiled in this atlas

Congenital heart disease (CHD) reflects a wide variety of cardiac structural defects affecting approximately 1% of live births (Nees and Chung, Am J Med Genet C Semin Med Genet, 2020). Its aetiology is complex, with both genetic and environmental drivers. Trisomy 21 (T21) is a strong risk factor for CHD, with a frequency of between 40% and 50% amongst T21 live births, and a particular risk of septal defects (interatrial, interventricular, or both: AV canal). A recent study showed impaired proliferation, compaction and mitochondrial respiration in cardiomyocytes using data from human and murine models, which were associated with heart septation defects (Lana-Elola E and Aoidi R et al., Sci Transl Med, 2024).

To shed light on this, we generated Multiome and Visium data from T21 hearts (aged 11–14 PCW). The dataset encompasses a comprehensive set of cell types, which were annotated using a CellTypist model trained on the euploid atlas. T21 nuclei were integrated with the age-matched euploid data using scVI.

Data

Single-cell/nuclei RNA and ATAC-seq

Summary	Interactive	H5AD (log-norm)	H5AD (raw)	CellTypist	ATAC peaks	ATAC frags
Global				–		–
Trisomy 21 + Euploid				–	–	–
Trisomy 21 + Euploid, subsampled for Milo analysis				–	–	–
CellTypist model — Coarse	–	–	–		–	–
CellTypist model — Mid	–	–	–		–	–
CellTypist model — Fine	–	–	–		–	–

The three CellTypist models above were constructed on a sample of the dataset. The sample was carefully chosen by random stratified sampling of the data at the fine_grain resolution, balancing the donor and cell_or_nuclei fractions between all fine_grains (n = 7485, minimum number of cells in each fine_grain cluster = 111). These fine_grain groups were then collapsed into their mid_grain and coarse_grain groups for the lower resolution models. Although good, this approach may be improved by a further balancing of time points to reduce temporal heteroscedasticity, or by splitting cell or nuclei models apart for cleaner distributions.

Visium — H5AD by age

Dataset	H5AD (log-normalised)	H5AD (raw)

Visium — interactive sections

Each tissue section opens in an interactive viewer with cell2location predictions. Sections are grouped by developmental stage.

Visium HD

Dataset	H5AD (raw)
Bin2cell cells
Bin2cell + TACCO cells

Xenium

Dataset	H5AD (raw)
Xenium-5K
TACCO cells

Code

View the data processing and analysis code on GitHub: View on GitHub