Diamonds in the doublets

Interacting cells are characterized by sequencing FACS clusters.

For decades, advances in immunology and hematopoiesis have relied on technologies for studying single cells, most notably flow cytometry and cell sorting. Fluorescence-activated cell sorting (FACS) uses fluorescence emission and light scatter to distinguish and retain single cells of interest for analysis while routinely discarding clusters of two or more cells. In multicellular life, of course, cells do not exist in isolation. On the contrary, they participate in many dynamic interactions with other cells and with extracellular signals. In this issue, Giladi, Cohen et al.1 explore what can be learned from the cell clusters, known as ‘doublets’, that immunologists discard (Fig. 1). They apply advanced single-cell sequencing methods and deconvolution analysis to investigate the nature of doublets, revealing the identity and molecular programs of immune-cell pairs in a model parasitic infection as well as new immune–epithelial cell interactions in the developing lung.

Fig. 1: Cell clusters in flow cytometry data.

a, The PIC-seq workflow. Giladi, Cohen et al. devised an optimized tissue dissociation protocol to create cell suspensions while preserving PICs. Cell-restricted markers are then used to identify mixed cell clusters by FACS for sequencing analysis. Leveraging data from single-cell (non-PIC) sorted controls, the PIC analysis pipeline provides an identity estimate for each cell in the cluster, a gene expression set enriched with that PIC, and an approximation of each cell type’s contribution to that program. Measurements of forward-scatter height (FSC-H) and area (FSC-A) distinguish single cells (diagonal events) from clusters of two or more cells (off-diagonal events). b, Flow cytometry data distinguish single cells from clusters. c, A ‘doublet-free’ mass cytometry analysis of CD3+ T cells and B220+ B cells in Balb/c mouse tissues4. The box in the top right quadrant shows the frequency of cell clusters expressing both markers. B cell–T cell PICs, as a fraction of all immune cells, are more abundant in the lung and lymph node than in colon or bone marrow.

One of the best-documented intercellular interactions in mammalian cell biology is the immunological synapse that forms between effector lymphocytes and professional antigen-presenting cells2. The sheer number of immune cell types and receptor–ligand combinations can result in a dizzying number of possible interactions. Depending on the specific configuration, immune cell interactions may serve to maintain homeostasis or respond to viral infections, or they may fall out of balance in autoimmune disease. The importance of the immunological synapse has driven the development of new analytic methods. For instance, a recent approach called LIPSTIC (labeling immune partnerships by sortagging intercellular contacts) enables an engineered cell to enzymatically mark its target cells in living organisms3. Such methods have generated enthusiasm for dynamically mapping transient immune interactions, but implementing them requires cell engineering and animal models.

Interactions between cells are not as transient as one might think. Cells can remain attached to one another for minutes to hours, and clusters of two or more cells can readily survive tissue-dissociation procedures. Figure 1b shows flow cytometry analysis of single cells from our laboratory, indicating gating to distinguish single cells from clusters. Figure 1c shows mass cytometry data from a mouse immune atlas4. Although the data were acquired using a ‘doublet-free’ barcoding strategy5, the top right quadrants appear to show rare interactions between CD3+ T cells and B220+ B cells, with more interactions in the lymph node and the lung than in the colon and the bone marrow, as might be expected.

Giladi, Cohen et al. describe an approach to systematically study the FACS doublets normally discarded as ‘junk’ data. In their method, named physically interacting cell sequencing (PIC-seq) (Fig. 1a), they use FACS to prospectively isolate PICs of interest on the basis of coexpression of lineage markers delineating each of the interacting cell types. Single-cell sequencing of PICs and deconvolution analysis, together with single-cell controls, reveal the identity and molecular program of the cells composing the PICs.

The authors first applied PIC-seq to a mouse model of helminth infection. By isolating and sequencing doublets expressing CD11c and T cell receptor (TCR)-β, they could study interactions between antigen-presenting myeloid cells and T cells. In doing so, they were able to identify a discrete number of enriched gene modules associated with these interactions and to deconvolve the identity of each of the interacting cells in the pair. In control cells that were not immune-activated, the authors were surprised to see an abundance of interactions between myeloid cells and T regulatory cells, highlighting the active role of tolerogenic T regulatory cells under homeostatic conditions.

Looking beyond the anticipated immunological players, Giladi, Cohen et al. also studied cellular interactions during normal developmental processes. In an effort to capture virtually any immune cell–epithelial cell interaction, they examined CD45+EpCAM+ PICs in the neonatal mouse lung. This identified unexpected interactions between epithelial cells and premature alveolar macrophages, monocytes or dendritic cells, which promoted the glucocorticoid response. This example illustrates how PIC-seq can be used to discover uncharacterized interactions between cells of any type, with minimal assumptions or cell engineering required.

Going forward, PIC-seq also lays the groundwork for studies seeking to understand clinically significant structural microenvironments. For example, it could help dissect the role of innate immune cells in driving the positively prognostic, immune-compartmentalized phenotype in triple-negative breast cancer6,7. As the throughput and economics of integrating antibody staining with single-cell sequencing workflows improves (such as in Ab-seq8 and CITE-seq9), it may even be possible to apply PIC-seq in silico without the need for prospective isolations. Still, given the rarity of PICs observed in the present study, this would require a routine throughput of 105–106 cells or more to be practical without prospective enrichment. In certain cases, it may also be possible to generically apply forward- and side-scatter information in FACS (Fig. 1b) to isolate all PICs in an experiment, with no assumptions about cell identity or lineage.

Lastly, sophisticated tissue imaging approaches, such as multiplexed ion beam imaging6,7, are defining an increasing number of complex and biologically significant microenvironments by their static cell compositions and inferred physical interactions. While these imaging methods can be sensitive and high throughput, they are limited to preselected molecular targets that might not reveal the functional nature of the PICs forming these compositions. Therefore, complementary approaches for profiling PICs provide a universal way to dig deeper into the molecular properties of interacting cells discovered in uncharacterized microenvironments, even within primary human tissue. As methods such as PIC-seq roll out, they continfue to remind us that our concept of ‘junk data’ can be misguided, and that putting aside our notions of data purity can lead to the discovery of biological diamonds.


  1. 1.

    Giladi, A. et al. Nat. Biotechnol. (2020).

  2. 2.

    Huppa, J. B. & Davis, M. M. Nat. Rev. Immunol. 3, 973–983 (2003).

    CAS  Article  Google Scholar 

  3. 3.

    Pasqual, G. et al. Nature 553, 496–500 (2018).

    CAS  Article  Google Scholar 

  4. 4.

    Spitzer, M. H. et al. Science 349, 1259425 (2015).

    Article  Google Scholar 

  5. 5.

    Zunder, E. R. et al. Nat. Protoc. 10, 316–333 (2015).

    CAS  Article  Google Scholar 

  6. 6.

    Keren, L. et al. Cell 174, 1373–1387.e19 (2018).

    CAS  Article  Google Scholar 

  7. 7.

    Keren, L. et al. Sci. Adv. 5, eaax5851 (2019).

    Article  Google Scholar 

  8. 8.

    Shahi, P., Kim, S. C., Haliburton, J. R., Gartner, Z. J. & Abate, A. R. Sci. Rep. 7, 44447 (2017).

    CAS  Article  Google Scholar 

  9. 9.

    Mimitou, E. P. et al. Nat. Methods 16, 409–412 (2019).

    CAS  Article  Google Scholar 

Download references

Author information



Corresponding author

Correspondence to Sean C. Bendall.

Ethics declarations

Competing interests

The author declares no competing interests.

Rights and permissions

Reprints and Permissions

About this article

Verify currency and authenticity via CrossMark

Cite this article

Bendall, S.C. Diamonds in the doublets. Nat Biotechnol 38, 559–561 (2020).

Download citation


Quick links

Nature Briefing

Sign up for the Nature Briefing newsletter — what matters in science, free to your inbox daily.

Get the most important science stories of the day, free in your inbox. Sign up for Nature Briefing