Molecular and clinical correlates of quantitative morphological atypia

Primary supervisor: Mariam Jamal-Hanjani, UCL

Secondary supervisors: Anita Grigoriadis, King’s College London, David Moore, UCL

Project

Background:

Nuclear atypia is a near-universal feature of solid malignancies [1]. The degree of atypia observed often correlates with prognosis and is used in grading some cancers. Traditionally, atypia has been assessed subjectively or semi-quantitatively with microscopy. Advances in image analysis now allow automated and objective morphological characterisation of cancer cells at scale [2]. Common genomic alterations, such as whole-genome duplication, can be inferred from whole-slide images (WSIs) and are associated with changes in nuclear morphology [3].

Most studies examining atypia with automated approaches have sought to predict mutation status or prognosis from WSIs or images of nuclei, often using deep learning. The relationship between specific, interpretable aspects of atypia and genomic or transcriptomic alterations is therefore poorly understood. Furthermore, spatial and temporal variation in atypia within and between tumours and metastases has not been systematically described. It is also unclear why the degree of atypia does not always relate to prognosis. High-throughput characterisation of nuclear atypia and its relationship to genomic and transcriptomic features may yield new insights into cancer biology and inform prognostic stratification [1].

TRACERx (TRAcking Cancer Evolution through therapy (Rx)) is a multi-centre study which aims to enhance understanding of tumour evolution in non-small cell lung cancer (NSCLC) through multi-regional histological, genomic and transcriptomic profiling of over 800 primary tumours and recurrence or progression samples [4]. 25 participants have co-participated in PEACE (Posthumous Evaluation of Advanced Cancer Environment), a unique study involving extensive post-mortem sampling of metastatic tumours. Data collected during these studies includes thousands of WSIs with paired whole-exome and RNA-sequencing data that has been processed using established bioinformatic pipelines.

Workstream 1: Develop computational pipeline to characterise and quantify nuclear atypia across primary and metastatic tumours in TRACERx and PEACE
Machine learning classifiers will be used to identify cancer nuclei and characterise and measure features of nuclear atypia such as nuclear membrane irregularity, aberrant chromatin distribution and micronucleation. Different approaches will be optimised on subregions of WSIs and compared with semi-quantitative manual scoring.

Workstream 2: Nuclear atypia and genomic and transcriptomic tumour profiles
Metrics of nuclear atypia derived from TRACERx and PEACE samples will be related to genomic features, such as whole-genome doubling events, to determine whether such factors explain inter- and intratumour variation in atypia. Multi-spatial and multi-temporal sampling in TRACERx and PEACE will be leveraged to identify whether tumours from different patients or tumour subclones evolve certain nuclear features convergently or in parallel. Associations between morphological changes and changes in gene expression will be explored.

Workstream 3: Prognostic impact of nuclear atypia and application to external datasets
Patient outcome prediction, based on specific characteristics of nuclear atypia, will be assessed in the TRACERx cohort. The features of cases with poor outcome but relatively low nuclear atypia will be further investigated, in an attempt to determine biological commonalities between these tumours. The analysis pipeline will be applied to public datasets to determine whether any observations made are applicable in other cancer types.

Hypotheses: Two broad hypotheses will be considered:

• Measurable features of nuclear atypia are associated with chromosomal instability in NSCLC
• Measurable features of nuclear atypia reflect tumour evolution

Candidate background

Candidates should have an interest in cancer genomics and digital image analysis. This project would be suited to trainees in histopathology, oncology or surgery. Basic knowledge of R and/or Python is desirable.

References

1. Singh I, Lele TP. Nuclear Morphological Abnormalities in Cancer: A Search for Unifying Mechanisms. In: Nuclear, Chromosomal, and Genomic Architecture in Biology and Medicine. 2022. p. 443-67.
2. Pocock J, Graham S, Vu QD, Jahanifar M, Deshpande S, Hadjigeorghiou G, et al. TIAToolbox as an end-to-end library for advanced tissue image analytics. Communications Medicine. 2022 Sep 24;2(1):120.
3. Fu Y, Jung AW, Torne RV, Gonzalez S, Vöhringer H, Shmatko A, et al. Pan-cancer computational histopathology reveals mutations, tumor composition and prognosis. Nat Cancer. 2020 Jul 27;1(8):800-10.
4. Frankell AM, Dietzen M, Al Bakir M, Lim EL, Karasaki T, Ward S, et al. The evolution of lung cancer and impact of subclonal selection in TRACERx. Nature. 2023 Apr 20;616(7957):525-33.