High-dimensional (HiDim) cytometry is revolutionizing the way we study single immune cells, immune cell populations, diseases and immune responses. Generating, processing and interpreting this complex data requires considerable research efforts from wet lab scientists, physicians, and computer scientists alike.

Our group utilizes HiDim cytometry derived from mass cytometry and high parameter flow cytometers to characterize the immune composition in several diseases including cancer. Our research interest encompasses three interconnected fields:

  1. Designing, testing and using high dimensional cytometry panels to characterize the immune composition in several diseases such as cancer.
  2. Deciphering the regulatory circuits underlying cancer immune responses, which carries the potential of predicting response to immunotherapy.
  3. Developing computational tools to examine and model complex immune responses in close collaboration with computer scientists.

Using “Immune Instagrams” to predict and monitor response to immunotherapy


Immunotherapy is a clinical reality for oncologists. As immunotherapy with anti-PD-1 is used in a wide array of cancers it becomes clear that not all patients respond and only a small fraction achieves durable responses. Thus biomarkers or a cell type that stratifies patients most likely to respond to therapy is in urgent need. Given the diversity of the immune system, containing several cell types, activation subtypes and/or differentiation subtypes, and considering that most immunotherapies target the immune proteome, we used single cell high dimensional mass cytometry (MC) in order to visualize all cell types and stages. We utilized peripheral blood mononuclear cells (PBMC) from samples of twenty patients with metastatic melanoma before and twelve weeks after anti-PD-1 immunotherapy to compare responders and nonresponders.

Experimental workflow using CyTOF. Experimental setup for the processing of frozen PBMC from matched samples before and after PD-1 immunotherapy using metal-labeled antibodies and acquisition by mass cytometry.

Using three optimized MC immune marker panels and data driven artificial intelligence supported bioinformatics analysis tools our initial analysis revealed that hierarchical Ward clustering by marker expression on all leukocytes in each patient before and after therapy stratified patients into responders and non responders, indicating a possible pre-therapeutic biomarker. Subsequent clustering and algorithm assisted annotation of cellular clusters using FlowSOM revealed that responders had a lower frequency of CD4+ and CD8+ T cells and higher frequencies of CD19-HLA-DRhigh myeloid cells in peripheral blood before therapy.

Immune instagrams were generated using the FlowSOM algorithm over millions of cells from patients from all groups. Cells are colored according to the cluster they were assigned by manual annotation using the exemplified heatmap. The heatmap represents the median arcsinh-transformed marker expression normalized to 0-1 range of respective markers within the 7 cellular clusters. In a top to bottom approach samples are then allocated to the different analysis groups: non-responders (NR), responders (R), and healthy donors (HD).

The lymphocyte landscape during therapy clearly revealed a CD8+ T cell centric response to anti-PD-1 but did not provide a predictive biomarker. Since we observed an increase in of CD19-HLA-DRhigh myeloid cells in peripheral blood before therapy, we now wanted to explore this cell population in depth by using a myeloid-centric MC panel. FlowSOM analysis revealed a higher frequency of CD14+CD16-HLA-DRhi circulating blood monocytes in responders and suggested this signature as a robust predictive biomarker in patients most likely to respond to anti-PD-1 immunotherapy.

Bar graphs showing individual patient sample composition before immunotherapy after extraction from the algorithm generated healthy donors (HD) non-responders (NR) and responders (R) clusters.

In order to translate our finding to the clinic we designed a conventional flow-cytometry based validation of our findings using fewer markers on a second independent cohort of 31 patients before therapy. Results from the validation dataset conformed that patients with lower circulating T cells and and increased frequency of CD14+ blood monocytes are more likely to respond towards therapy and that the frequency of classical monocytes was robustly associated with progression-free survival and overall survival in response to anti-PD-1.

Classical monocyte frequencies above (blue) and below (red) 19.38% over time (months) correlate with cumulative hazard and overall survival in patients with.

We currently are expanding our studies to larger patient cohorts in order to see whether the response-associated immune signature holds true and want to understand whether myeloid cells actually actively contribute to the immune response.

Monitoring the immunome in patients treated with novel combination immunotherapy

Mass cytometry analysis of the immune cell response to a novel superagonistic IL-15 plus anti-PD-1 immunotherapy in patients with refractory non small cell lung cancer (NSCLC).

In the second article, published in April 2018 in Lancet Oncology, together with MUSC faculty members, first author John Wrangle and last author, Mark Rubinstein, we reported the first-in-human experience of an IL-2Rbetagamma agonist with a drug targeting the PD-1/PD-L1 pathway.  This report demonstrates that these two classes of drugs can be administered safely, that there is evidence of clinical efficacy, and a comprehensive single cell high dimensional analysis of the immune profile in these patients.