Early response evaluation by single cell signaling profiling in acute myeloid leukemia

Abstract

Background: A fundamental hallmark of cancer cells is their ability to sustain proliferative signaling and cell survival, reflected in a cellular chemotherapy response that is poorly understood. We questioned whether chemotherapy modulated phospho-signaling at 4 and 24 h in vivo could provide information about long-term survival in acute myeloid leukemia (AML), and if the signaling response to therapy was more informative than analysis at time of diagnosis. Methods: Peripheral blood was collected from 32 younger AML patients (age 16-74 years), before, 4- and 24 hours after start of induction chemotherapy. Samples were analyzed by 36-dimensional mass cytometry for assessment of alterations in immunophenotypes and intracellular signaling using unsupervised and supervised machine learning approaches. Results were validated by RNA sequencing and mass spectrometry proteomics (Super SILAC). Targeted sequencing was used to characterize patient samples for recurrent AML mutations. Drug sensitivity and resistance testing ex vivo was compared to activation of relevant signal transduction pathways and mutational profile. Findings: 5-year patient survival was accurately predicted in the leukemic cell population at 24 hours after therapy onset by phospho-proteins p-ERK1/2 (T202/Y204) and p-p38 (T180/Y182). RNA sequencing showed induction of MAPK target gene expression and the AP-1 transcription complex in patients with high p-ERK1/2. Super-SILAC proteomics confirmed an increase in the abundance of p38 prime target MAPKAPK2(MK2) 24 hours after start of induction therapy. Ex vivo drug sensitivity testing demonstrated high sensitivity to MEK inhibitors in the patient cells with high p-ERK1/2 measured at diagnosis or 24 hours after start of chemotherapy. Interpretation: Early single cell signaling response to chemotherapy provided precise prognostic information independent of stratification by genetics. We propose that early functional measurement of chemotherapy-potentiated MAPK pathway signaling could identify non-responders to intensive chemotherapy allowing precise treatment adjustment.

Automated cell type annotation and exploration of single cell signalling dynamics using mass cytometry

“…Considering the Remission status, we observed significant differences for both HSC, and Monocytic populations (Supplementary Figure S8). We also compared abundance distributions between mutated and wild-type patients for known hotspot mutations such as DNMT3A, IDH1/2, TET2, SRSF2 and NPM1 identified by Illumina Trusight targeted sequencing (22). Differences were observed for B-cells between DNMT3A wild-type/mutated patients, as well as for HSC, MPP, and T-cytotoxic cells between NPM1 wild-type/mutated patients (Supplementary Figure S9).…”

“…This addition would also enable the interpretation of biological mechanisms that control survival of cells, and it might be informative with regard to signalling pathways (i.e., signalling relationship), the effect of genetic aberrations (e.g., CNV or SNVs) and their underlying developmental and proliferation states. In addition, developing predictive models of relapse using data collected after therapy initiation seems a powerful option that we could exploit further using the ML-based tools described here (22).…”

“…In addition, we selected 14 intracellular markers were studied to measure intracellular activation/phosphorylation of signalling networks namely pNFkB, pErk, pSTAT1, pSTAT3, pSTAT5, pP38, pHistone3, pRB, CASP3, pAxl, pAkt, pS6, CyclinB1, pCREB. Data pre-processing and normalisation of all CyTOF samples is described in (22). The Helios mass cytometer (Fluidigm) was used for profiling, and all experiments were conducted at the Flow Cytometry Core Facility of the University of Bergen.…”

“…A novel feature engineering approach based on DREMI was utilised to score all pairwise interactions between signalling proteins in the antibody panel. As a case study we analysed 45 samples from leukemia patients (22). By applying LASSO regression to DREMI .…”

“…A novel feature engineering approach based on DREMI was utilised to score all pairwise interactions between signalling proteins in the antibody panel. As a case study we analysed 45 samples from leukemia patients (22). By applying LASSO regression to DREMI scores computed from single-cell populations, we constructed computational models to predict short term survivors at time of diagnosis.…”