Single-cell Map of Diverse Immune Phenotypes Driven by the Tumor Microenvironment

Context:

Immunologists have long struggled against, or defended, their reliance on cell surface markers to study subsets of immune cells. Hiding behind cell surface markers, we fear, is great biological complexity that goes unseen. Similarly, gene expression studies on bulk populations often mask biologically relevant variation among individual cells. Thus a growing camp of immunologists have turned to single cell RNA sequencing (scRNA seq) to refine canonical immune cell subsets and measure gene expression changes particular to them (Tirosh et al). Yet with the ability to query individual cells comes the new challenge of clustering them into biologically meaningful populations, despite the huge variation within or between them. With a new kind of measurement must come new methods to normalize and compare the results they provide. This is especially the case for integrating data obtained from patients into clinically meaningful categories such as molecular subtypes or metastatic site from cancer patients.

Data:

The authors obtained tissue from eight breast cancer patients representing different subtypes. For three of these patients the authors also obtained adjacent normal tissue, and from one patient they obtained a lymph node metastasis. They enriched for immune cells and sequenced via the InDrop platform, sequencing over 47,000 immune cells in aggregate. Following sequencing, their first pass analysis showed that immune cells exhibited significant patient-specific batch effects. A major advance put forth by this preprint addresses this challenge using BISCUIT, described in a prior paper, to remove such effects and normalize scRNA seq data across all eight patients, enabling comparisons among them (Prabhakaran et al).

Their most detailed and provocative observations concern tumor-resident T cells, which were not clustered into pools of distinct phenotypes such as naïve or activated. Individual cells rather were smeared along a continuum in which gene modules for these states were coexpressed and vary widely. This continuum was best captured along axes of activation, terminal differentiation and signatures of hypoxia. To capture the total variation among cells the authors described the “phenotypic volume,” i.e. the summed diversity of cell states, which was vastly expanded within tumors vs normal tissue. This observation of continuous cell states agrees with previous work published by the Pe’er group, showing continuous gene expression changes in response to competing cytokine inputs (Antebi and Reich-Zeliger, et al).

After posting these results, the authors later added data from T cell receptor (TCR) sequencing by using both InDrop and 10X platorms, to query TCR clones in three additional patients. Somewhat unsurprisingly, clonotypes clustered with restricted suites of cell states, rather than exhibiting the full panoply of T cell phenotypes. They argue, convincingly, that T cell clones likely respond uniquely to microenvironments of their antigen, yielding spatially distributed T cell phenotypes according to antigen distribution and microenvironmental cues.

Turning to cells of the myeloid lineage, the authors find a similar phenomenon in macrophages. Contrary to models espousing mutually exclusive M1 and M2 subsets, macrophages express M1 and M2-type gene modules simultaneously. Indeed, M1 and M2 genes were positively correlated in macrophages.

Implications:

In light of their observations, the authors argue for refined models of T cell activation and differentiation in malignant tissues that account for their apparent plasticity as they encounter aberrant environments. The notion that the local environment strongly determines T cell phenotypes predicts trouble for T cell therapies that use genetic modifications to hardwired cell activation, as these may not surmount microenvironmental effects within a target tissue. How accurately the signature for hypoxia actually reports on hypoxic environments is not assessed by this study. Yet the broader points – of continuous variation in T cell phenotype, trending with signatures for hypoxia – already presents a set of provocative and testable predictions.

With the new level of detail offered by scRNA seq, biologists promise (and are promised) to advance our understanding of immunology and, in this case, immunotherapy for cancer. However, these more detailed measurements threaten to outpace our ability to meaningfully analyze cell populations, and to apply these analyses across patients. Thus we might temper our enthusiasm for this vast increase in granularity.

Questions:

Does the correlation between hypoxia and differentiation gene expression signatures reflect a direct causative relationship?

Would normalizing hypoxic microenvironments necessarily change T cell functions?

How can we make comparisons between patients, or conclusions on patient cohorts, when our yardstick is high-dimensional datasets that encompass enormous variation?

References

Prabhakaran, S., Azizi, E., Carr, A., and Pe’er, D. (2016). Dirichlet Process Mixture Model for Correcting Technical Variation in Single-Cell Gene Expression Data. In Proceedings of The 33rd International Conference on Machine Learning, B. Maria Florina, and Q.W. Kilian, eds. (Proceedings of Machine Learning Research: PMLR), pp.1070–1079.

Tirosh, I., Izar, B., Prakadan, S.M., Wadsworth, M.H., 2nd, Treacy, D., Trombetta, J.J., Rotem, A., Rodman, C., Lian, C., Murphy, G., et al. (2016). Dissecting the multicellular ecosystem of metastatic melanoma by single-cell RNA-seq. Science 352, 189-196.

Yaron E. Antebi, Shlomit Reich-Zeliger, Yuval Hart , Avi Mayo , Inbal Eizenberg , Jacob Rimer , Prabhakar Putheti, Dana Pe’er, Nir Friedman. (2013). Mapping Differentiation under Mixed Culture Conditions Reveals a Tunable Continuum of T Cell Fates. PLoS Biol. 11(7): e1001616.

 

Posted on: 17 May 2018 , updated on: 5 July 2018

Read preprint (1 votes)