CCI-seq indexes cell clumps to map which cells sit together in tissue

Single-cell RNA sequencing tells you which cell types a tissue contains, but it loses where they sat ^[15]. Spatial transcriptomics keeps position. The CCI-seq authors list its limits: a restricted number of detectable genes, trouble delineating cell boundaries, and special devices at high cost ^[7]. Their method tries to keep the depth of single-cell sequencing and still recover which cells were next to which ^{[5] [1]}.

The design starts with incomplete dissociation. Tissue is only partly broken up, so cells that were touching or close stay together in clumps ^[11]. The clumps are indexed combinatorially and then put through single-cell sequencing ^[11]. What gets recorded runs from direct physical contact out to short-range spatial neighbours ^[14], a range the authors call a proximity continuum ^[1].

By the authors' account, each earlier approach has a specific weakness. Proximity labelling tags the neighbours of a pre-selected bait cell, so an experiment covers one bait cell type ^[8]. Clump deconvolution also starts from partial dissociation. It then infers each clump's members computationally and inherits the errors of that inference ^[9]. Spatial nuclei indexing recovers few nuclei and misses most of a tissue's interactions. It also reads only nuclear RNA, although a large proportion of mRNA is in the cytoplasm ^[10]. The authors describe CCI-seq's capture as unbiased ^[1]. As described, indexing the clumps directly is the design's answer to deconvolution error ^{[11] [9]}.

I think the mouse work is the strongest evidence here, because of how it is built. In kidney the method recovered known interactions in the collecting duct and glomerulus ^[12]. In intestine it placed known enterocyte and goblet cell subtypes along the crypt-villus axis ^[13]. Both are positive controls: anatomy that is already known, recovered by a new readout ^[2]. The Apc knockout adds a perturbation where the tissue is expected to change, and the method registered that change ^[3].

The colorectal cancer result is the one with clinical reach ^[4]. The abstract and introduction do not give the number of tumours, the classification scheme, cell or clump counts, effect sizes, or a comparison with subtyping by cell-type composition alone.

The transcriptional programs tied to interacting cells are reported as associations ^[6]. A clump shows that two cells were together. That alone cannot say whether contact switched on NF-kB inflammatory signalling in a tumour, or whether cells already in that state are the ones that stay attached. The kidney finding has the same structure: epithelial cells engaging different partners were associated with different ion transport programs ^[12].