The core idea behind CCI-seq is deceptively simple. Instead of fully dissociating a tissue into single cells, the researchers partially break it apart into small clumps of fewer than twenty cells, preserving the native contacts between neighbors. Each clump is then tagged with a cholesterol-modified oligonucleotide that anchors to the membranes of all its constituent cells, followed by three rounds of split-pool barcode ligation that assign a unique combinatorial index to each clump. Only afterward are the clumps fully dissociated, and the resulting single cells are run through a standard droplet-based single-cell RNA sequencing workflow. Because every cell that once belonged to the same clump carries the same combinatorial index—its so-called Cell-ID—the researchers can reconstruct exactly which cells were in physical contact before sequencing began.

The team validated the approach using three-dimensional spheroids of human HEK293T and mouse NIH/3T3 cells, a clever benchmark in which the true composition of each clump was known. The labeling process proved efficient and even, reaching cells in the interior of spheroids as well as the exterior. Sequencing quality was essentially indistinguishable from standard single-cell RNA-seq, with a median of more than 11,000 unique molecular identifiers and roughly 3,400 genes detected per cell. When the researchers mixed human and mouse spheroids, 99.7 percent of human cells and 93.8 percent of mouse cells received the correct species-specific Cell-ID, and only about 0.3 percent of clumps contained cells from both species, indicating minimal cross-contamination. Roughly 95 percent of cells were assigned to a clump of two or more cells, meaning the vast majority of the data carried interaction information.

Applied to the mouse kidney, CCI-seq produced profiles for 11,519 high-quality cells across three biological replicates, detecting a median of 2,822 genes per cell. That depth is a substantial advantage over leading spatial transcriptomics platforms: 6.5-fold more genes than Slide-seq-V2 and 24.5-fold more than seqFISH. Unsupervised clustering resolved 17 distinct cell types spanning all major renal lineages, and the resulting interaction map was highly reproducible across replicates. Critically, it recapitulated known anatomy—glomerular associations among podocytes, endothelial cells and mesangial cells, and collecting duct associations among intercalated, principal and transitional cells—and showed strong concordance with an independent high-resolution seqFISH dataset, with a cosine similarity of 0.86.

The kidney map also revealed interactions that had not been previously appreciated, including contacts between monocytes and podocytes and between B cells and proximal straight tubule cells. The team confirmed these pairings directly in tissue using immunofluorescence and amplified sequential fluorescence in situ hybridization, or asmFISH, demonstrating that the computational map reflects genuine physical proximity rather than artifact. Because monocytes and B cells are known players in kidney immunity, the authors suggest these interactions may contribute to local immune surveillance and epithelial regulation, though the functional consequences remain to be fully explored.

Perhaps the most striking feature of CCI-seq is that it goes beyond mapping contacts to reveal what those contacts do molecularly. In the kidney, distal convoluted tubule cells adopted divergent transcriptional programs depending on which collecting duct partner they touched: those interacting with intercalated cells upregulated calcium transport genes such as Trpv5 and Calb1, while those interacting with principal cells upregulated blood pressure regulation genes including Scnn1g and Slc8a1. Ligand–receptor analysis performed within clumps further grounded predicted signaling in verified proximity, uncovering specific axes such as Timp3–Kdr and Vegfa–Kdr between podocytes and endothelial cells, and Thbs1–Sdc4 signaling between monocytes and podocytes. In effect, the method links physical adjacency to functional specialization at the level of individual genes.

In the mouse small intestine, CCI-seq captured 6,412 high-quality cells and faithfully reconstructed the crypt–villus architecture, with enterocyte subtypes preferentially interacting with spatially adjacent neighbors along the villus axis and canonical niche contacts between Lgr5-positive intestinal stem cells and Paneth cells. The data also exposed a previously unresolved question: transit-amplifying cells, the rapidly dividing progenitors that fuel intestinal renewal, turned out to split into two spatial subtypes. Crypt-bottom transit-amplifying cells expressed stem-like markers such as Sorbs2 and Olfm4 and were enriched for Wnt signaling, whereas crypt-top counterparts expressed maturation-associated genes such as Rbp7 and showed metabolic enrichment. A spatial score built from these genes correlated strongly with differentiation pseudotime, with a Pearson correlation of 0.74, confirming that a cell’s interaction-defined location mirrors its developmental state.

The method’s power for disease biology became evident when the team applied it to a mouse model of intestinal tumorigenesis driven by conditional knockout of the adenomatous polyposis coli gene, a classic initiator of colorectal cancer. In these precancerous intestines, the orderly crypt–villus organization collapsed: Paneth cells, normally confined to crypt bases, were found scattered along villi in direct contact with villus-tip enterocytes, a finding confirmed by asmFISH. The researchers also identified two novel precancerous cell populations, dubbed TA-like_Ly6a-positive and TA-like_Notum-positive cells, the former of which genotyping confirmed to carry the Apc deletion. These aberrant cells lost their normal interactions with stem cells and instead formed a new ectopic signaling hub in the villus, frequently contacting villus-tip enterocytes and each other. Meanwhile, villus-tip enterocytes engaging T cells displayed enhanced antigen-presentation signatures, consistent with active immune surveillance at the earliest stages of adenoma development.

Finally, in human colorectal cancer samples from five patients, CCI-seq captured 37,897 high-quality cells, with nearly 72 percent retained in clumps, and distinguished the clinically relevant iCMS2 and iCMS3 epithelial tumor subtypes. The contrast between them was stark: iCMS2 tumors behaved as immune deserts with minimal immune contact, whereas iCMS3 tumors showed extensive infiltration and frequent engagement between tumor cells and CD8-positive T cells, B cells and natural killer cells—directly visualizing the cellular basis of the immune-hot signature previously attributed to this subtype. Within iCMS3, the method even detected heterogeneity, with myeloid interactions present in one sub-branch but absent in another. Tumor cells in direct contact with monocytes specifically upregulated NF-κB target genes, including the inflammatory cytokines CXCL8 and IL1B, and showed stronger CD44-related signaling, hinting that monocyte-derived signals can push cancer cells toward a pro-inflammatory, stem-like state that may foster invasion and immune evasion.

Together, these results establish CCI-seq as a scalable, unbiased platform that simultaneously maps who touches whom in a tissue and what those touches mean at the molecular level. The authors acknowledge current limitations, including per-experiment cell throughput that may limit resolution of rare interaction states, and note that future iterations could integrate genomic, epigenomic, protein-level and immune-repertoire measurements, all of which are compatible with the standard single-cell workflow. But even in its present form, the method offers something the field has long lacked: a way to watch cells not just as individuals but as conversational partners, and to read the transcriptional consequences of every handshake. For cancer biology, immunology and developmental science alike, that is a conversation worth eavesdropping on.

Subject of Research: High-throughput mapping of cell–cell interaction networks and their molecular consequences in complex tissues using combinatorial cell clump indexing and single-cell sequencing

Article Title: Resolving cell–cell interaction networks and their molecular logic in complex tissues

Article References: Tang, L., Tian, K., Fu, X., Xu, Y., Wu, J., Zhang, J., Wang, X.-W., Ye, C., Wu, Q., Wu, W., Feng, C., & Zhang, Q. C. (2026). Resolving cell–cell interaction networks and their molecular logic in complex tissues. Nature Methods, 23(10), 2042-2053. https://doi.org/10.1038/s41592-026-03237-0

Image Credits: AI Generated

DOI: 10.1038/s41592-026-03237-0

Keywords: CCI-seq, cell–cell interactions, single-cell RNA sequencing, spatial transcriptomics, tumor microenvironment, colorectal cancer, mouse kidney, intestinal crypt, Apc knockout, NF-κB signaling, combinatorial indexing, ligand–receptor analysis