Makhlouf, Shorouk
(2025)
Computation pathology for a better understanding of breast cancer behaviour.
PhD thesis, University of Nottingham.
Abstract
Background
Oestrogen receptor (ER)-positive tumours constitute more than 70% of all breast cancer (BC) and are typically treated with endocrine therapy (ET). While ER positivity is currently defined by ≥1% nuclear staining on immunohistochemistry (IHC), this threshold does not fully capture the biological and clinical heterogeneity of ER-positive tumours. Quantitative ER expression levels and ER-regulated gene profiles are significantly correlated with tumour characteristics, patient outcomes and response to therapy. Furthermore, ER-low positive BC and invasive lobular carcinoma (ILC) represent diagnostic and therapeutic challenges within ER-positive BC subgroup. This study aims, first, to comprehensively assess the clinical, molecular, and morphological characteristics of ER-positive BC, and second, to identify biomarkers predictive of ET response and to investigate the role of tumour microenvironment (TME) and histological subtype in prognostication and treatment stratification. The addition of a computational pathology approach plays a key role in extracting quantitative insights from digital pathology data, enabling a deeper understanding of tumour heterogeneity and aiding in the development of predictive models for personalised treatment in ER-positive BC. The use of artificial intelligence (AI) has revolutionised cancer research, providing significant value in assessing subtle and complex histological and molecular features. This project also utilised AI to decipher specific features associated with ER-positive BC.
Methods
This study utilised multiple large, well-characterised cohorts, including more than 7,000 Nottingham BC patients, as well as transcriptomic data from The Cancer Genome Atlas (TCGA) and the METABRIC datasets. ER expression was evaluated semi-quantitatively using IHC percentage, staining intensity, and the histochemical score (H-score), and further analysed using RT-qPCR and in situ hybridisation (ISH) to quantify ESR1 mRNA. Differential gene expression analysis was carried out to explore key ER-regulated genes. Prognostic and predictive genes were validated at the mRNA and protein levels. Stromal morphology and tumour-infiltrating lymphocytes (TILs) were assessed using whole slide images and AI algorithms. ILC subtypes were histologically evaluated, and their nuclear morphology was further characterised using digital image analysis. The survival outcomes and prognostic biomarkers of ILC were also investigated.
Results
Quantitatively higher ER expression was associated with favourable prognostic features and a better response to ET therapy, with maximal benefit observed at 100% expression. Tumours with ≥10% ER staining behaved more like ER-positive cancers, while those with 1–9% expression demonstrated clinicopathological similarity to ER-negative BC. Repeat IHC and molecular validation confirmed that a substantial proportion of ER-low positive tumours were, in fact, ER-negative. ESR1 mRNA analysis further improved diagnostic accuracy and identified additional ER-positive cases that were diagnosed as ER-negative BC on IHC. The ER H-score proved to be an independent prognosticator, with outcome stratification at thresholds of 30, 100, and 200. Following analysis of a large number of ER-regulated genes, progesterone receptor (PR) and GREB1 emerged as robust and strong predictors of ET response. SUSD3 was also highly expressed in ER-positive BC and was associated with favourable outcomes and ET responsiveness. High stromal TILs and fibroblast content were associated with a more aggressive tumour characteristics and poor prognosis in luminal BC. Conversely, a high stroma-to-tumour ratio was associated with better outcomes. AI-based quantification of TILs offered greater objectivity and prognostic value compared with visual assessment alone. ILC, accounting for 11% of BC, exhibited marked heterogeneity. Classic ILC (cILC) was associated with more favourable survival compared with invasive ductal carcinoma of no special type. Contrasting this, pleomorphic (pILC) and high-grade solid ILC (sILC) variants comprised a distinct, aggressive subgroup characterised by poor chemotherapy response and higher rates of recurrence and BC-specific mortality. The morphology of pILC was heterogeneous and showed apocrine and non-apocrine subtypes with distinct features. Furthermore, high expression of INO80C and RECQL4 was associated with poor outcomes in ILC.
Conclusions
This study demonstrated that ER-positive BC is not a biologically uniform entity. Quantitative ER assessment, molecular profiling, and identification of novel biomarkers can significantly enhance the prediction of ET response. An ER expression threshold of ≥10% more accurately defines true ER positivity. RNA-based methods, RT-qPCR and ISH can complement IHC to enhance diagnostic accuracy and treatment decisions. The prognosis of ILC could be further refined by identifying aggressive subtypes and prognostic biomarkers, thereby guiding personalised treatment strategies. Digital image analysis and AI algorithms have proven significant value in assessing specific features that can be challenging for pathologists and can improve the objectivity of specific prognostic and predictive variables in BC. Including these assessments in routine practice can help improve risk stratification for patients with ER-positive BC.
Actions (Archive Staff Only)
 |
Edit View |
Tools
Tools
![[thumbnail of Thesis-minor changes]](https://eprints.nottingham.ac.uk/style/images/fileicons/application_pdf.png "Thesis-minor changes")