Collagenase 1

No statistical methods were used to predetermine sample size. The experiments were not randomized and investigators were not blinded to allocation during experiments and outcome assessment.
Patient samples. All tissue samples used for this study were obtained with written informed consent from all participants in accordance with the guidelines in The Declaration of Helsinki 2000 from multiple centres.
Human embryo, fetal and decidual samples were obtained from the MRC and Wellcome-funded Human Developmental Biology Resource (HDBR^{43 (link)}, http://www.hdbr.org), with appropriate maternal written consent and approval from the Newcastle and North Tyneside NHS Health Authority Joint Ethics Committee (08/H0906/21 5). The HDBR is regulated by the UK Human Tissue Authority (HTA; www.hta.gov.uk) and operates in accordance with the relevant HTA Codes of Practice. Decidual tissue for smFISH (Extended Data Fig. 7c) was also covered by this ethics protocol.
Peripheral blood from women undergoing elective terminations was collected under appropriate maternal written consent and with approvals from the Newcastle Academic Health Partners (reference NAHPB-093) and HRA NHS Research Ethics committee North-East-Newcastle North Tyneside 1 (REC reference 12/NE/0395)
Decidual tissue for immunohistochemistry (Fig. 3b, c, Extended Data Figs. 7a, 9c, d) and flow cytometry staining for granule proteins was obtained from elective terminations of normal pregnancies at Addenbrooke’s Hospital (Cambridge) between 6 and 12 weeks gestation, under ethical approval from the Cambridge Local Research Ethics Committee (04/Q0108/23).
Decidual tissue for smFISH (Fig. 3d, Extended Data Fig. 6b, 7b) was obtained from the Newcastle Uteroplacental Tissue Bank. Ethics numbers are: Newcastle and North Tyneside Research Ethics Committee 1 Ref:10/H0906/71 and 16/NE/0167.
Isolation of decidual, placental and blood cells. Decidual and placental tissue was washed in Ham’s F12 medium, macroscopically separated and then washed for at least 10 min in RPMI or Ham’s F12 medium, respectively, before processing.
Decidual tissues were chopped using scalpels into approximately 0.2-mm³ cubes and enzymatically digested in 15 ml 0.4 mg/ml collagenase V (Sigma, C-9263) solution in RPMI 1640 medium (Thermo Fisher Scientific, 21875-034)/10% FCS (Biosfera, FB-1001) at 37 °C for 45 min. The supernatant was diluted with medium and passed through a 100- m cell sieve (Corning, 431752) and then a 40- m cell sieve (Corning, 431750). The flow-through was centrifuged and resuspended in 5 ml of red blood cell lysis buffer (Invitrogen, 00-4300) for 10 min.
Each first-trimester placenta was placed in a Petri dish and the placental villi were scraped from the chorionic membrane using a scalpel. The stripped membrane was discarded and the resultant villous tissue was enzymatically digested in 70 ml 0.2% trypsin 250 (Pan Biotech P10-025100P)/0.02% EDTA (Sigma E9884) in PBS with stirring at 37 °C for 9 min. The disaggregated cell suspension was passed through sterile muslin gauze (Winware food grade) and washed through with Ham’s F12 medium (Biosera SM-H0096) containing 20% FBS (Biosera FB-1001). Cells were pelleted from the filtrate by centrifugation and resuspended in Ham’s F12. The undigested gelatinous tissue remnant was retrieved from the gauze and further digested with 10–15 ml collagenase V at 1.0 mg/ml (Sigma C9263) in Ham’s F12 medium/10% FBS with gentle shaking at 37 °C for 10 min. The disaggregated cell suspension from collagenase digestion was passed through sterile muslin gauze and the cells pelleted from the filtrate as before. Cells obtained from both enzyme digests were pooled together and passed through a 100- m cell sieve (Corning, 431752) and washed in Ham’s F12. The flow-through was centrifuged and resuspended in 5 ml of red blood cell lysis buffer (Invitrogen, 00-4300) for 10 min.
Blood samples were carefully layered onto a Ficoll–Paque gradient (Amersham) and centrifuged at 2,000 r.p.m. for 30 min without breaks. Peripheral blood mononuclear cells from the interface between the plasma and the Ficoll–Paque gradient were collected and washed in ice-cold phosphate-buffered saline (PBS), followed by centrifugation at 2,000 r.p.m. for 5 min. The pellet was resuspended in 5 ml of red blood cell lysis buffer (Invitrogen, 00-4300) for 10 min.
Assignment of fetal developmental stage. Up to eight post-conception weeks, embryos are staged using the Carnegie staging method⁴⁴ . At fetal stages beyond eight post-conception weeks, age was estimated from measurements of foot length and heel-to-knee length. These were compared with a standard growth chart⁴⁵ .
Flow cytometry staining, cell sorting and single-cell RNA-seq. Decidual and blood cells were incubated at 4 °C with 2.5 l of antibodies in 1% FBS in DPBS without calcium and magnesium (Thermo Fisher Scientific, 14190136). DAPI was used for live versus dead discrimination. We used an antibody panel designed to enrich for certain populations for single-cell sorting and scRNA-seq. Cells were sorted using a Becton Dickinson (BD) FACS Aria Fusion with 5 excitation lasers (355 nm, 405 nm, 488 nm, 561 nm and 635 nm red), and 18 fluorescent detectors, plus forward and side scatter. The sorter was controlled using BD FACS DIVA software (version 7). The antibodies used are listed in Supplementary Table 10.
For single-cell RNA-seq using the plate-based Smart-seq2 protocol, we created overlapping gates that comprehensively and evenly sampled all immune-cell populations in the decidua (Extended Data Fig. 1). B cells (CD19 or CD20) were excluded from our analysis, owing to their absence in decidua^{46 (link)}. Single cells were sorted into 96-well full-skirted Eppendorf plates chilled to 4 °C, prepared with lysis buffer consisting of 10 l of TCL buffer (Qiagen) supplemented with 1% -mercaptoethanol. Single-cell lysates were sealed, vortexed, spun down at 300g at 4 °C for 1 min, immediately placed on dry ice and transferred for storage at 80 °C. The Smart-seq2 protocol was performed on single cells as previously described^{11 (link),47 (link)}, with some modifications^{48 (link)}. Libraries were sequenced, aiming at an average depth of 1 million reads per cell, on an Illumina HiSeq 2000 with version 4 chemistry (paired-end, 75-bp reads).
For the droplet scRNA-seq methods, blood and decidual cells were sorted into immune (CD45) and non-immune (CD45) fractions. B cells (CD19 or CD20) were excluded from blood analysis, owing to their absence in decidua^{46 (link)}. Only viable cells were considered. Placental cells were stained for DAPI and only viable cells were sorted. To improve trophoblast trajectories, an additional enrichment of EPCAM and HLA-G was performed for selected samples (Fig. 2 only). Cells were sorted into an Eppendorf tube containing PBS with 0.04% BSA. Cells were immediately counted using a Neubauer haemocytometer and loaded in the 10x-Genomics Chromium. The 10x-Genomics v2 libraries were prepared as per the manufacturer’s instructions. Libraries were sequenced, aiming at a minimum coverage of 50,000 raw reads per cell, on an Illumina HiSeq 4000 (paired-end; read 1: 26 cycles; i7 index: 8 cycles, i5 index: 0 cycles; read 2: 98 cycles).
Flow cytometry staining for granule proteins. For intracellular staining of granule proteins, dNKs were surface-stained for 30 min in FACS buffer with antibodies (listed in Supplementary Table 10). Cells were washed with FACS buffer followed by staining with dead cell marker (DCM Aqua) and streptavidin Qdot605. dNKs were then treated with FIX & PERM (Thermo Fisher Scientific) and stained for granule proteins. Samples were run on an LSRFortessa FACS analyser (BD Biosciences) and data analysed using FlowJo (Tree Star). dNKs were gated as CD3 CD14 CD19 live cells; CD56 NKG2A and then KIR and KIR subsets were generated using Boolean functions with the gates for all the different KIRs stained (KIR), and their inverse gates (KIR). Wilcoxon test was used to compare granule protein staining between paired dNK subsets from the same donor. A P value 0.05 was considered to be statistically significant.
Immunohistochemistry. Four-micrometre tissue sections from formalin-fixed, paraffin-wax-embedded human decidual and placental tissues were dewaxed with Histoclear, cleared in 100% ethanol and rehydrated through gradients of ethanol to PBS. Sections were blocked with 2% serum (of species in which the secondary antibody was made) in PBS, incubated with primary antibody overnight at 4 °C and slides were washed in PBS. Biotinylated horse anti-mouse or goat anti-rabbit secondary antibodies were used, followed by Vectastain ABC–HRP reagent (Vector, PK-6100) and developed with di-aminobenzidine (DAB) substrate (Sigma, D4168). Sections were counterstained with Carazzi’s haematoxylin and mounted in glycerol and gelatin mounting medium (Sigma, GG1-10). Primary antibody was replaced with equivalent concentrations of mouse or rabbit IgG for negative controls. See Supplementary Table 10 for antibody information. Tissue sections were imaged using a Zeiss Axiovert Z1 microscope and Axiovision imaging software SE64 version 4.8.
smFISH. Samples were fixed in 10% NBF, dehydrated through an ethanol series and embedded in paraffin wax. Five-millimetre samples were cut, baked at 60 °C for 1 h and processed using standard pre-treatment conditions, as per the RNAScope multiplex fluorescent reagent kit version 2 assay protocol (manual) or the RNAScope 2.5 LS fluorescent multiplex assay (automated). TSA-plus fluorescein, Cy3 and Cy5 fluorophores were used at 1:1,500 dilution for the manual assay or 1:300 dilution for the automated assay. Slides were imaged on different microscopes: Hamamatsu Nanozoomer S60 (Extended Data Fig. 7c). Zeiss Cell Discoverer 7 (Fig. 4d, Extended Data Figs. 6, 7c). Filter details were as follows. DAPI: excitation 370–400, BS 394, emission 460–500; FITC: excitation 450–488, BS 490, emission 500–55; Cy3: excitation 540–570, BS 573, emission 540–570; Cy5: excitation 615–648, BS 691, emission 662–756. The camera used was a Hamamatsu ORCA-Flash4.0 V3 sCMOS camera.
Whole-genome sequencing. Tissue DNA and RNA were extracted from fresh-frozen samples using the AllPrep DNA/RNA/miRNA kit (Qiagen), following the manufacturer’s instructions. Short insert (500-bp) genomic libraries were constructed, flowcells were prepared and 150-bp paired-end sequencing clusters generated on the Illumina HiSeq X platform, according to Illumina no-PCR library protocols, to an average of 30 coverage. Genotype information is provided in Supplementary Table 1.
Single cell RNA-seq data analysis. Droplet-based sequencing data were aligned and quantified using the Cell Ranger Single-Cell Software Suite (version 2.0, 10x Genomics)^{13 (link)} against the GRCh38 human reference genome provided by Cell Ranger. Cells with fewer than 500 detected genes and for which the total mito-chondrial gene expression exceeded 20% were removed. Mitochondrial genes and genes that were expressed in fewer than three cells were also removed.
SmartSeq2 sequencing data were aligned with HISAT2^{49 (link)}, using the same genome reference and annotation as the 10x Genomics data. Gene-specific read counts were calculated using HTSeq-count^{50 (link)}. Cells with fewer than 1,000 detected genes and more than 20% mitochondrial gene expression content were removed. Furthermore, mitochondrial genes and genes expressed in fewer than three cells were also removed. To remove batch effects due to background contamination of cell free RNA, we also removed a set of genes that had a tendency to be expressed in ambient RNA (PAEP, HBG1, HBA1, HBA2, HBM, AHSP and HBG2).
Downstream analyses—such as normalization, shared nearest neighbour graph-based clustering, differential expression analysis and visualization—were performed using the R package Seurat^{51 (link)} (version 2.3.3). Droplet-based and SmartSeq2 data were integrated using canonical correlation analysis, implemented in the Seurat alignment workflow⁵² . Cells, the expression profile of which could not be well-explained by low-dimensional canonical correlation analysis compared to low-dimensional principal component analysis, were discarded, as recommended by the Seurat alignment tutorial. Clusters were identified using the community identification algorithm as implemented in the Seurat ‘FindClusters’ function. The shared nearest neighbour graph was constructed using between 5 and 40 canonical correlation vectors as determined by the dataset variability; the resolution parameter to find the resulting number of clusters was tuned so that it produced a number of clusters large enough to capture most of the biological variability. UMAP analysis was performed using the RunUMAP function with default parameters. Differential expression analysis was performed based on the Wilcoxon rank-sum test. The P values were adjusted for multiple testing using the Bonferroni correction. Clusters were annotated using canonical cell-type markers. Two clusters of peripheral blood monocytes represented the same cell type and were therefore merged.
We further removed contaminating cells: (i) maternal stromal cells that were gathered in the placenta for one of the fetuses; (ii) a shared decidual–placental cluster with fetal cells mainly present in two fetuses (which we think is likely to be contaminating cells from other fetal tissues due to the surgical procedure). This can occur owing to the source of the tissue and the trauma of surgery. We also removed a cluster for which the top markers were genes associated with dissociation-induced effects^{53 (link)}. Each of the remaining clusters contained cells from multiple different fetuses, indicating that the cell types and states we observed are not affected by batch effects.
We found further diversity within the T cell clusters, as well as the clusters of endothelial, epithelial and perivascular cells, which we then reanalysed and partitioned separately, using the same alignment and clustering procedure.
The trophoblast clusters (clusters 1, 9, 20, 13 and 16 from Fig. 1d) were taken from the initial analysis of all cells and merged with the enriched EPCAM and HLA-G cells. The droplet-based and Smart-seq2 datasets were integrated and clustered using the same workflow as described above. Only cells that were identified as trophoblast were considered for trajectory analysis.
Trajectory modelling and pseudotemporal ordering of cells was performed with the monocle 2 R package^{54 (link)} (version 2.8.0). The most highly variable genes were used for ordering the cells. To account for the cell-cycle heterogeneity in the trophoblast subpopulations, we performed hierarchical clustering of the highly variable genes and removed the set of genes that cluster with known cell-cycle genes such as CDK1. Genes which changed along the identified trajectory were identified by performing a likelihood ratio test using the function differentialGeneTest in the monocle 2 package.
Network visualization was done using Cytoscape (version 3.5.1). The decidual network was created considering only edges with more than 30 interactions. The networks layout was set to force-directed layout.
KIR typing. Polymerase chain reaction sequence-specific primer was performed to amplify the genomic DNA for presence or absence of 12 KIR genes (KIR2DL1, KIR2DL2, KIR2DL3, KIR2DL5 (both KIR2DL5A and KIR2DL5B),KIR3DL1, KIR2DS1, KIR2DS2, KIR2DS3, KIR2DS4, KIR2DS5 and KIR3DS1) and the pseudogene KIR2DP1. KIR2DS4 alleles were also typed as being either full-length or having the 22-bp deletion that prevents cell-surface expression. Two pairs of primers were used for each gene, selected to give relatively short amplicons of 100–800 bp, as previously described^{55 (link)}. Extra KIR primers were designed using sequence information from the IPD-KIR database (release 2.4.0) to detect rare alleles of KIR2DS5 and KIR2DL3 (KIR2DS5, 2DS5rev2: TCC AGA GGG TCA CTG GGA and KIR2DL3, 2DL3rev3: AGA CTC TTG GTC CAT TAC CG)^{56 (link)}. KIR haplotypes were defined by matrix subtraction of gene copy numbers using previously characterized common and contracted KIR haplotypes using the KIR Haplotype Identifier software (www.bioinformatics.cimr.cam.ac.uk/haplotypes).
Inferring maternal or fetal origin of single cells from droplet-based scRNA-seq using whole-genome sequencing variant calls. To match the processing of the whole-genome sequencing datasets, droplet-based sequencing data from decidua and placenta samples were realigned and quantified against the GRCh37 human reference genome using the Cell Ranger Single-Cell Software Suite (version 2.0)^{13 (link)}. The fetal or maternal origin of each barcoded cell was then determined using the tool demuxlet^{57 (link)}. In brief, demuxlet can be used to deconvolve droplet-based scRNA-seq experiments in which cells are pooled from multiple genetically distinct individuals. Given a set of genotypes corresponding to these individuals, demuxlet infers the most likely genetic identity of each droplet by estimating the likelihood of observing scRNA-seq reads from the droplet overlapping known single nucleotide polymorphisms. Demuxlet inferred the identities of cells in this study by analysing each Cell Ranger-aligned BAM file from decidua and placenta in conjunction with a VCF file, containing the high-quality whole-genome-sequence variant calls from the corresponding mother and fetus. Each droplet was assigned to be maternal, fetal or unknown in origin (ambiguous or a potential doublet), and these identities were then linked with the transcriptome-based cell clustering data to confirm the maternal and fetal identity of each annotated cell type.
T cell receptor analysis by TraCeR. The T cell receptor sequences for each single T cell were assembled using TraCeR^{58 (link)}, which allowed the reconstruction of the T cell receptors from scRNA-seq data and their expression abundance (transcripts per million), as well as identification of the size, diversity and lineage relation of clonal subpopulations. In total, we obtained the T cell receptor sequences for 1,482 T cells with at least one paired productive or chain. Cells for which more than two recombinants were identified for a particular locus were excluded from further analysis.
Whole-genome sequencing alignment and variant calling. Maternal and fetal whole-genome sequencing data were mapped to the GRCh37.p13 reference genome using BWA-MEM version 0.7.15⁵⁹ . The SAMtools^{60 (link)} fixmate utility (version 1.5) was used to update read-pairing information and mate-related flags. Reads near known indels from the Mills^{61 (link)} and 1000G^{62 (link)} gold standard reference set for hg19/GRCh37 were locally realigned using GATK IndelRealigner version 3.7^{61 (link)}. Base-calling assessment and base-quality scores were adjusted with GATK BaseRecalibrator and PrintReads version 3.7^{60 (link),63} . PCR duplicates were identified and removed using Picard MarkDuplicates version 2.14.1^{63 ,64} . Finally, bcftools mpileup and call version 1.6^{65 (link)} were used to produce genotype likelihoods and output called variants at all known biallelic single nucleotide polymorphism sites that overlap protein-coding genes. For each sample, variants called with phred-scale quality score 200, at least 20 supporting reads and mapping quality 60 were retained as high-quality variants.
Quantification of KIR gene expression by KIRid. The KIR locus is highly polymorphic in terms of both numbers of genes and alleles^{11 (link)}. Including a single reference sequence for each gene can lead to reference bias for donors that happen to better match the reference sequence. To address these issues, we used a tailored approach in which we first built a total cDNA reference by concatenating the Ensembl coding and non-coding transcript sequences, excluding transcripts belonging to the KIR genes (GRCh38, version 90), and the full set of known KIR cDNAs sequences from the IPD-KIR database^{66 (link)} (release 2.7.0). For each donor, we removed transcript sequences for KIR genes determined to be absent in that individual, which decreases the extent of multi-mapping and quantification. The single-cell reads of each donor were then mapped to the corresponding donor-specific reference using Kallisto^{67 (link)} (version 0.43.0 with default options). Expression levels were quantified using the multi-mapping deconvolution tool MMSEQ^{68 (link)}, and gene-level estimates were obtained by aggregating over different alleles for each KIR gene.
Cell–cell communication analysis. To enable a systematic analysis of cell–cell communication molecules, we developed CellPhoneDB, a public repository of ligands, receptors and their interactions. Our repository relies on the use of public resources to annotate receptors and ligands. We include subunit architecture for both ligands and receptors, to accurately represent heteromeric complexes.
Ligand–receptor pairs are defined based on physical protein–protein interactions (see sections of ‘CellPhoneDB annotations’). We provide CellPhoneDB with a user-friendly web interface at www.CellPhoneDB.org, where the user can search for ligand–receptor complexes and interrogate their own single-cell transcriptomics data.
To assess cellular crosstalk between different cell types, we used our repository in a statistical framework for inferring cell–cell communication networks from single-cell transcriptome data. We derived enriched receptor–ligand interactions between two cell types based on expression of a receptor by one cell type and a ligand by another cell type, using the droplet-based data. To identify the most relevant interactions between cell types, we looked for the cell-type specific interactions between ligands and receptors. Only receptors and ligands expressed in more than 10% of the cells in the specific cluster were considered.
We performed pairwise comparisons between all cell types. First, we randomly permuted the cluster labels of all cells 1,000 times and determined the mean of the average receptor expression level of a cluster and the average ligand expression level of the interacting cluster. For each receptor-ligand pair in each pairwise comparison between two cell types, this generated a null distribution. By calculating the proportion of the means which are ‘as or more extreme’ than the actual mean, we obtained a P value for the likelihood of cell-type specificity of a given receptor–ligand complex. We then prioritized interactions that are highly enriched between cell types based on the number of significant pairs, and manually selected biologically relevant ones. For the multi-subunit heteromeric complexes, we required that all subunits of the complex are expressed (using a threshold of 10%), and therefore we used the member of the complex with the minimum average expression to perform the random shuffling.
CellPhoneDB annotations of membrane, secreted and peripheral proteins. Secreted proteins were downloaded from Uniprot using KW-0964 (secreted). Secreted proteins were annotated as cytokines (KW-0202), hormones (KW-0372), growth factors (KW-0339) and immune-related using Uniprot keywords and manual annotation. Cytokines, hormones, growth factors and other immune-related proteins were annotated as ‘secreted highlight’ proteins in our lists.
Plasma membrane proteins were downloaded from Uniprot using KW-1003 (cell membrane). Peripheral proteins from the plasma membrane were annotated using the Uniprot Keyword SL-9903, and the remaining proteins were annotated as transmembrane proteins. We completed our lists of plasma transmembrane proteins by doing an extensive manual curation using literature mining and Uniprot description of proteins with transmembrane and immunoglobulin-like domains.
Plasma membrane proteins were annotated as receptors and transporters. Transporters were defined by the Uniprot keyword KW-0813. Receptors were defined by the Uniprot keyword KW-0675. The list of receptors was extensively reviewed and new receptors were added based on Uniprot description and bibliography revision. Receptors involved in immune-cell communication were carefully annotated.
Protein lists are available at https://www.cellphonedb.org/downloads. Three columns indicate whether the protein has been manually curated: ‘tags’, ‘tags_ description’, ‘tags_reason’.
The tags column is related to the manual curation of a protein, and contains three options: (i) ‘N/A’, which indicates that the protein has not been manually curated; (ii) ‘To_add’, which indicates that secreted and/or plasma membrane protein annotation has been added; and (iii) ‘To_comment’, which indicates that the protein is either secreted (KW-0964) or membrane-associated (KW-1003) but that we manually added a specific property of the protein (that is, the protein is annotated as a receptor).
tags_reason is related to the protein properties, and contains five options: (i) ‘extracellular_add’, which indicates that the protein is manually annotated as plasma membrane; (ii) ‘peripheral_add’, which indicates that the protein is manually annotated as a peripheral protein instead of plasma membrane; (iii) ‘secreted_add’, which indicates that the protein is manually annotated as secreted; (iv) ’secreted_high’, which indicates that the protein is manually annotated as secreted highlight. For cytokines, hormones, growth factors and other immune-related proteins; option (v) ‘receptor_add’ indicates that the protein is manually annotated as a receptor.
tags_description is a brief description of the protein, function or property related to the manually curated protein.
CellPhoneDB annotations of heteromeric receptors and ligands. Heteromeric receptors and ligands (that is, proteins that are complexes of multiple gene products) were annotated by reviewing the literature and Uniprot descriptions. Cytokine complexes, TGF family complexes and integrin complexes were carefully annotated.
If heteromers are defined in the RCSB Protein Data Bank (http://www.rcsb.org/), structural information is included in our CellPhoneDB annotation. Heteromeric complex lists are available at www.CellPhoneDB.org.
CellPhoneDB annotations of interactions. The majority of ligand–receptor interactions were manually curated by reviewing Uniprot descriptions and PubMed information on membrane receptors. Cytokine and chemokine interactions are annotated following the International Union of Pharmacology annotation⁶⁹ . Other groups of cell-surface proteins the interactions of which were manually reviewed include the TGF family, integrins, lymphocyte receptors, semaphorins, ephrins, Notch and TNF receptors.
In addition, we considered interacting partners as: (i) binary interactions annotated by IUPHAR (http://www.guidetopharmacology.org/) and (ii) cytokines, hormones and growth factors interacting with receptors annotated by the iMEX consortium (https://www.imexconsortium.org/)^{70 (link)}.
We excluded from our analysis transporters and a curated list of proteins including: (i) co-receptors; (ii) nerve-specific receptors such as those related to ear-binding, olfactory receptors, taste receptors and salivary receptors, (iii) small molecule receptors, (iv) immunoglobulin chains, (v) pseudogenes and (vi) viral and retro-viral proteins, pseudogenes, cancer antigens and photoreceptors. These proteins are annotated as ‘others’ in the protein list. We also excluded from our analysis a list of interacting partners not directly involved in cell–cell communication. The ‘remove_interactions’ list is available in https://www.cellphonedb.org/downloads.
Lists of interacting protein chains are available from https://www.cellphonedb.org/downloads. The column labelled ‘source’ indicates the curation source. Manually curated interactions are annotated as ‘curated’, and the bibliography used to annotate the interaction is stored in ‘comments_interaction’. ‘Uniprot’ indicates that the interaction has been annotated using UniProt descriptions.
Linking Ensembl and Uniprot identification. We assigned to the custom-curated interaction list all the Ensembl gene identifications by matching information from Uniprot and Ensembl by the gene name.
Database structure. Information is stored in a PostgreSQL relational database (www.postgresql.org). SQLAlchemy (www.sqlalchemy.org) and Python 3 were used to build the database structure and the query logic. All the code is open source and uploaded to the webserver.

A schematic overview of the myocyte isolation procedure is shown in Figure 2. An expanded description of the procedure, accompanied with images and videos, and complete materials list is available in the Online Data Supplement, alongside full details of additional methods applied in this study (Appendix A-ix). All animal work was undertaken in accordance with Singapore National Advisory Committee for Laboratory Animal Research guidelines. Relevant national and institutional guidelines and regulations must be consulted before commencement of all animal work.
Buffers and media were prepared as detailed in Appendix D. EDTA, perfusion, and collagenase buffers were apportioned into sterile 10 mL syringes, and sterile 27 G hypodermic needles were attached (Online Figure IA).
C57/BL6J mice aged 8 to 12 weeks were anesthetized, and the chest was opened to expose the heart. Descending aorta was cut, and the heart was immediately flushed by injection of 7 mL EDTA buffer into the right ventricle. Ascending aorta was clamped using Reynolds forceps, and the heart was transferred to a 60-mm dish containing fresh EDTA buffer. Digestion was achieved by sequential injection of 10 mL EDTA buffer, 3 mL perfusion buffer, and 30 to 50 mL collagenase buffer into the left ventricle (LV). Constituent chambers (atria, LV, and right ventricle) were then separated and gently pulled into 1-mm pieces using forceps. Cellular dissociation was completed by gentle trituration, and enzyme activity was inhibited by addition of 5 mL stop buffer.
Cell suspension was passed through a 100-μm filter, and cells underwent 4 sequential rounds of gravity settling, using 3 intermediate calcium reintroduction buffers to gradually restore calcium concentration to physiological levels. The cell pellet in each round was enriched with myocytes and ultimately formed a highly pure myocyte fraction, whereas the supernatant from each round was combined to produce a fraction containing nonmyocyte cardiac populations.
CM yields and percentage of viable rod-shaped cells were quantified using a hemocytometer. Where required, the CMs were resuspended in prewarmed plating media and plated at an applicationdependent density, onto laminin (5 μg/mL) precoated tissue culture plastic or glass coverslips, in a humidified tissue culture incubator (37°C, 5% CO₂). After 1 hour, and every 48 hours thereafter, media was changed to fresh, prewarmed culture media.
The cardiac nonmyocyte fraction was collected by centrifugation (300g, 5 minutes), resuspended in fibroblast growth media, and plated on tissue-culture treated plastic, area ≈ 23 cm² (0.5× 12-well plate) per LV, in a humidified tissue culture incubator. Media was changed after 24 hours and every 48 hours thereafter.