The project has delivered across all the aims:
1. Characterization of the clonal and cellular heterogeneity of primary tumours from the 10 (11) genomic driver-based breast cancer subtypes (ICs)
We used a large breast cancer cohort (METABRIC) to assess tumour diversity grounded in evolutionary theory, including its application into single cell RNA sequencing (scRNAseq) (Hausser et al, Nature Communications 2019). We used METABRIC and other sample sets to characterize DNA methylation, single cell omics, and single cell proteomics using imaging mass cytometry [IMC]. We reported DNA methylation of METABRIC samples within a rich context of genomic, transcriptional, and clinical data (Batra et al, Nature Communications 2021). We generated single nucleus DNA sequencing (snDNAseq) data for a total of 38,880 nuclei in 154 primary cancers from METABRIC revealing the variety of clonal architectures. We completed scRNAseq using 10X Genomics in 80 breast cancer explants, with ~300,000 cells profiled, revealing the complexity of cell types in both cancer cells and the TME. These explants overlap with the 53 where CyTOF/IMC identified 11 distinct types of cancer cell (Georgopoulou et al, Nature Communications 2021). We published a high impact paper using IMC (Danenberg et al, Nature Genetics 2022) to systematically map malignant and TME structures in 693 METABRIC tumours. In conclusion we have defined in unprecedented detail the characteristic patterns of clonality and TME across ICs.
2. Comparative characterization of the clonal and cellular heterogeneity of matched pairs of primary and metastatic cancers
We performed extensive multi-omic profiling of metastases in 10 autopsies (de Mattos-Arruda et al, Cell Reports 2019) to show that metastases propagate and evolve as communities of clones, reveal their neo-antigen landscapes, and show they accumulate HLA LOH. The data identify variable TMEs and reveal, through analyses of TCR repertoires, that adaptive immunity co-evolves with the metastatic genomes. We have samples set from a further 10 cases where we completed WES, RNAseq and ctDNA, a unique dataset to compare in patient, in vivo and in vitro clonal architecture, and its evolution over time and under the selection of therapy. The DETECT clinical study recruited a cohort of 188 patients with metastatic breast cancer with a total of 1098 blood samples from which cfDNA was extracted. In 100 patients we collected 240 PBMC samples characterized by FACS using two panels: 12-colour immunophenotyping and 15-colour T cell activation/exhaustion. We identified a transition point of 7% ctDNA fraction to stratify patients and developed a Bayesian learning model to predict subsequent treatment response. FACS data was analysed using FlowSOM and Metacell with 13 PBMC cell types identified. Correlations with tumour burden and survival will be conducted. A first manuscript has been submitted to JCO. In conclusion we have either published or completed studies describing in unprecedented detail the landscape of metastatic breast cancer and how ctDNA monitoring can help predict outcome.
3. and 4. Characterization of the clonal and epigenetic evolution, and the immune response, across therapy courses
We collected clinical, digital pathology, genomic and transcriptomic profiles of pre-treatment biopsies from patients in Trans-NEO (Sammut el al, Nature 2022). Pathology end points at surgery were then correlated with multi-omic features in these diagnostic biopsies. We collected similar multi-omic data at midway and at surgery. Results from the midway profiling after 9 weeks of therapy reveal the dynamics of the full tumour ecosystem. Dynamics can be used as additional parameters to improve the machine learning predictive model. This work received one of the extremely prestigious AACR Scholar-in-Training awards. From these patients plasma samples were also collected and cfDNA extracted for quantitation of ctDNA using TARDIS. The data generated will provide ctDNA dynamics and this will be added to the predictive model. In another study we recruited 8 patients from a neo-adjuvant chemotherapy study where we generated xenografts, and completed the multi-omic characterization of the donor tissue and the derived explant. We have used these models in co-clinical trials to show concordant model-patient therapeutic responses and to identify a novel epigenetic mechanism of resistance to olaparib. In conclusion we have characterized full tumour ecosystems using multi-omics and showed machine learning can be used to generate a predictor of response, described how tumour ecosystems evolve under therapy and demonstrated that explant models mimic clinical responses and can be used to identify novel mechanisms of resistance.
