25 years ago, our understanding of cancer genomics underwent a paradigm shift when it was discovered that cancer cell karyotypes are in a state of constant flux due to underlying chromosome instability (CIN), ie continuous gain and loss of chromosomes and/or acquisition of structural rearrangements. And indeed, it is now widely accepted that CIN is a major driver of tumour heterogeneity, phenotypic adaptation and drug resistance. In the last two decades, we have learnt a great deal about the molecular mechanisms responsible for chromosome replication and segregation, as well as the associated cell cycle checkpoint controls. However, oncogenic mutations in genes directly involved in these processes are extremely rare, meaning our understanding of the basic principles governing the acquisition of CIN, how it drives tumorigenesis and alters trajectories in the face of selective pressures remains more limited. Thus we aim to understand these principles in order to exploit CIN as a therapeutic target. Despite intense efforts, multiple factors have hindered progress. Mechanistic studies typically focus on a limited number of established cancer cell lines, due to experimental tractability, yet they tend to have limited clinical annotation, do not reflect disease heterogeneity, and lack pre-/post- treatment counterparts. Moreover, outgrowth of fitter subclones best suited for long-term cell culture yields relatively stable karyotypes. Additional confounding factors are genetic drift due to extensive in vitro propagation, and a tumour-site agnostic philosophy that ignores the possibility of disease-specific CIN pathways. Another limitation is the lack of non-transformed, karyotypically-stable model systems that represent the cell-of-origin to recapitulate CIN pathways. While cancer sequencing projects can analyse large cohorts of clinically annotated samples, reliance on archival biopsies results in stromal contamination, single-cell approaches are technically challenging, and testing emerging hypotheses with functional experiments is impossible. While sequencing spatially resolved biopsies allows reconstruction of evolutionary trajectories, longitudinal cohorts of matched chemo-naïve, on-treatment and relapse samples are less common. Moreover, isolating progenitors of drug-resistant clones is impossible. Therefore, to define the basic principles governing how CIN drives tumorigenesis and drug resistance, we now propose a fundamentally fresh approach with several key benefits to address the limitations that have hindered progress to date. Importantly, we will take a disease-specific approach, focusing on high-grade serous ovarian cancer (HGSOC), where CIN is the key driver and acquired drug resistance the key clinical challenge. Firstly, we will exploit our living biobank of patient-derived ovarian cancer models (OCMs), possibly the largest and most diverse collection of primary HGSOC cell cultures. OCMs are early passage, purified tumour fractions that possess the hallmarks and heterogeneity of HGSOC. Coupled with a cell culture system that enables extensive proliferative potential, OCMs are amenable to multi-omics, including single-cell omics, high-resolution cell biology studies and drug-sensitivity profiling. As the biobank matures, we are assembling longitudinal cohorts, ie OCMs derived from biopsies taken before, during and after chemotherapy. Secondly, as HGSOC originates from fallopian tube epithelial cells, we have established the FNE1 model system of these cells. FNE1 cells are non-transformed and karyotypically stable, but we have shown that introducing HGSOC-specific genetic lesions is sufficient to induce CIN, yielding karyotypes similar to those of our patient-derived models. And thirdly, to isolate the progenitors of drug-resistant descendants, we will combine a state-of-the-art barcode-based lineage tracing technology with our longitudinal OCMs to study clonal dynamics in response to chemotherapy.

Outcome of psychosis is heterogeneous, with chronic relapsing-remitting course in the majority. Relapse in the early years is the best predictor of a chronic course. Maintenance treatment with antipsychotics effectively reduces relapse risk but has many side-effects. In our changing society, the attitude towards long-time use of antipsychotics has become negative and for many patients and clinicians, maintenance treatment is no longer acceptable. Without maintenance treatment, relapse can still be prevented, if it is predicted accurately, as timely warnings provide a window of opportunity for patients, informal caregivers, and clinicians to take effective actions to prevent relapse. We hypothesize that computational analysis of speech can provide such accurate, timely warnings. Psychosis affects both form and content of thoughts, which are detectable in speech. Natural language processing (NLP) can accurately differentiate people with psychosis from healthy individuals and predict which high-risk individuals will develop psychosis. In this project we will test if NLP can also predict emerging psychotic relapse, to provide a window of opportunity for relapse prevention. We aim to improve outcome of individuals with psychosis by gaining scientific knowledge on speech as a biomarker of relapse and, in a next step, by creating an e-health tool to predict relapse.