Atallah, Nehal Mohammed
(2025)
Computational pathology and deciphering the complexity of HER2 expression in breast cancer.
PhD thesis, University of Nottingham.
Abstract
Background
Breast cancer (BC) is the most common cancer and second leading cancer-related death in women. Despite significant advancement in the detection, prognostic and predictive classification of BC, there remains a proportion of
patients who recur and die of their disease. Further research on BC diagnosis and classification is needed. Digital pathology (DP) and molecular assays, including whole slide imaging (WSI), digital image analysis (DIA) and artificial intelligence (AI), are revolutionising BC diagnosis and treatment, enabling detailed cyto-morphometric assessments that enhance diagnostic accuracy, personalise treatment, and improve outcomes. Although the therapeutic roles of HER2 protein in BC have been defined for more than two decades, the spectrum of HER2 expression in BC is evolving, and therapeutic modalities are expanding. HER2-positive cases are candidates for targeted anti-HER2 therapy, but patient response varies, with recurrence rates ranging from 20-40%.
The introduction of HER2-Low BC and antibody-drug conjugates (ADCs) has
transformed the management of HER2-negative BC but added complexity to
HER2 assessment, particularly regarding the lower limits of ADC therapy eligibility.
This thesis evaluates the integration of computational pathology and advanced
molecular approaches in identifying novel BC prognostic parameters and deciphering the complexity of HER2 expression, including its biological and clinical characterisation. It highlights differential responses to targeted anti-HER2
therapies among HER2-positive classes, the role of hormone receptor (HR)
expression, and refines the definition of HER2-Low BC.
Methods and Results
A large cohort of 40,160 breast WSIs was analysed to assess the accuracy
and utility of WSI compared to traditional microscopy. The assessment focused
on the presence, frequency, location, and tissue type of missing tissue and its
clinical implications. Additionally, scanning time, specimen type, time to implement WSIs and quality control measures were evaluated. The results indicated
that while missing tissue's frequency, area, and location varied, all cases involved fat tissue, and none affected final diagnoses. Quality control measures
significantly improved image quality and reduced WSI failure rates by sevenfold.
We also used AI to identify novel prognostic parameters in breast cancer. All
slide images were analysed using AI-based DP tools to extract stromal and
tumour features, focusing on the tumour-stroma ratio and spatial distribution
of stromal cells. Deep learning algorithms were employed for feature extraction
and spatial analysis, following extensive annotation at the cellular and region
levels. The tumour-stroma ratio was found to be a significant prognostic factor,
with higher ratios associated with worse outcomes. AI-based spatial analysis
revealed patterns of stromal cell distribution that correlated with tumour aggressiveness and patient survival. This demonstrated the utility of AI in enhancing understanding of the tumour microenvironment and its impact on BC
prognosis.
A substantial BC cohort with clinical, transcriptional, and survival data was utilised to develop a HER2-driven morphometric signature reflecting HER2-positive tumours with active oncogenic signalling and better response to targeted
anti-HER2 therapy. WSIs of invasive BC cases with HER2-positive and HER2-
negative controls were analysed. Image acquisition, segmentation, and spatial
distribution analyses were performed using DIA algorithms. Statistical correlations between HER2-driven morphometric features and patient outcomes were
analysed. A total of 57 morphometric features were evaluated, with 22 significantly associated with HER2 positivity. HER2 IHC score 3+/oestrogen receptor
(ER) negative tumours were significantly associated with HER2-related morphometric features and showed the least intra-tumour morphological heterogeneity. The HER2-driven morphometric signature was validated on PAM50
HER2-Enriched molecular subtype (HER2-E) cases, correlating with prolonged distant metastasis-free survival (DMFS) post-adjuvant anti-HER2 therapy.
For deciphering the response of HER2-positive BC to targeted anti-HER2 therapy, we first compared both HER2-positive classes regarding clinicopathological features, pathological complete response (pCR), patient survival, and the
role of ER status on therapy response. Differentially expressed genes and molecular pathways within both classes were identified, with TAF10 and BCAT1
selected for further validation through transcriptomic and proteomic analyses.
Tissue microarray blocks were constructed, and IHC staining, scoring, and
analysis were performed. The value of concomitant use of multigene assays
that assess ERBB2, ESR1, PGR, and MKI67 (MammaTyper® assay) with
HER2 IHC for predictive stratification of HER2-positive BC patients to anti HER2 therapy was investigated. RNA was extracted from formalin-fixed paraffin-embedded (FFPE) tissue, quantified using RT-qPCR, and discordant HER2
status cases were validated through staining on excision specimens and patient record reviews.
Our results revealed that compared to HER2 IHC 3+ tumours, patients with
IHC 2+/Amplified BC had a significantly lower pCR rate (22% versus 57%,
p<0.001) and shorter survival regardless of HER2 gene copy number level. ER
positivity was significantly associated with decreased response to anti-HER2
therapy in IHC 2+/Amplified but not in HER2 IHC 3+ BC patients. Compared
to IHC 3+, HER2 IHC 2+/Amplified tumours were less frequently classified as
HER2-E molecular subtype (16% versus 49%, p<0.001), with downregulation
of HER2 oncogenic pathway genes, upregulation of trastuzumab resistance
genes, and enrichment in ER signalling pathway genes.
TAF10 mRNA expression was significantly associated with HR-positive and
HER2-positive tumours, particularly the HER2-E molecular subtype (p<0.001).
High TAF10 mRNA expression correlated with prolonged breast cancer-specific survival (BCSS) in both the TCGA and METABRIC cohorts (p<0.008 and
p<0.001, respectively). Multivariate Cox regression confirmed TAF10 mRNA
as an independent predictor of better survival, which was also validated at the
protein level. High TAF10 protein expression levels predicted a better response
to trastuzumab-based chemotherapy, whereas in chemotherapy-only treated
HER2-positive patients, high TAF10 expression had no predictive role. TAF10
mRNA significantly correlated with low PIK3CA, mTOR, RICTOR and MIB1
mRNA expression, known resistance genes to trastuzumab therapy. At the
same time, BCAT1 showed a poor prognostic and predictive role in HER2-
positive BC cases with high expression associated with short BCSS and DMFS
in trastuzumab-treated patients.
Regarding the role of multigene assays in predicting response to anti-HER2
therapy, ERBB2 mRNA identified 251/287 (87.5%) cases as HER2-positive,
10.8% (31/287) as HER2-Low, and 1.7% (5/287) as HER2-negative. According
to the MammaTyper® assay, ERBB2-positive patients treated with anti-HER2
therapy had significantly prolonged 5-year disease free survival (DFS) and
DMFS (HR=0.56, p=0.003 and HR=0.62, p=0.023, respectively). MammaTyper®-defined HER2-E subtype showed a better response to anti-HER2
therapy compared to IHC-defined subtypes, with significant prolongation in
both 5-year DFS and BCSS. For ERBB2-negative patients, there were no significant differences in survival rates between those treated with trastuzumab
and those who received only chemotherapy (p > 0.05). Validation analysis revealed that 11/36 ERBB2-negative cases were IHC 2+/ISH positive with very
low gene amplification levels.
For HER2-Low tumours, we refined this category by integrating IHC scoring
with mRNA levels and artificial neural network (ANN) models to achieve a refined scoring algorithm with high concordance. HER2 staining intensity, pattern, and subcellular localisation were assessed in detail, and inter-observer
concordance in scoring HER2-Low class according to current guidelines was
evaluated. ANN analysis successfully distinguished HER2 score 0 from 1+ with
high sensitivity and specificity, improving the consistency of HER2 scoring and
pathologist concordance. The study provided a clearer definition of HER2-Low
BC, facilitating better patient stratification for ADC therapy. Most HER2-Low
tumours were HR-positive, enriched with luminal intrinsic molecular subtype,
lacking significant HER2 oncogenic signalling gene expression, and exhibiting
favourable clinical behaviour compared to HER2-negative BC. In HR-positive
BC, no significant prognostic differences were detected between HER2-Low
and HER2-negative tumours. However, in HR-negative BC, HER2-Low tumours were less aggressive, with longer patient survival. Transcriptomic data
showed that most HR-negative/HER2-Low tumours were luminal androgen receptor (LAR) intrinsic subtypes, enriched with T-helper lymphocytes, activated
dendritic cells, and tumour-associated neutrophils, while most HR-negative/HER2-negative tumours were basal-like, enriched with tumour-associated
macrophages.
Conclusion
The research demonstrated the efficacy of integrating DP, including WSI and
AI, in enhancing diagnostic accuracy and identifying novel BC prognostic parameters. HER2 protein overexpression is the primary oncogenic driver and
the main predictor of response to anti-HER2 therapy in HER2-positive BC. ER
signalling pathways are more dominant in BC with equivocal HER2 expression
compared to HER2 3+ tumours. The differential response to anti-HER2 therapy based on IHC classes should inform treatment decisions for HER2-positive BC patients. The concomitant use of multigene assays with IHC could improve the prediction of therapeutic response to anti-HER2 therapies. A more
precise definition of HER2-Low BC was established, integrating protein expression levels and mRNA data. This refined definition aids in identifying patients who may benefit from novel therapies, such as ADCs. Integrating computational pathology and molecular approaches offers a pathway to more precise and effective clinical management of HER2-positive and HER2-Low BC
patients.
| Item Type: |
Thesis (University of Nottingham only)
(PhD)
|
| Supervisors: |
Rakha, Emad
Mongan, Nigel
Allegrucci, Cinzia |
| Keywords: |
breast cancer, computational pathology, prognostic parameters, HER2 expression, hormone receptor expression, HER2 protein overexpression, molecular pathology |
| Subjects: |
W Medicine and related subjects (NLM Classification) > WP Gynecology |
| Faculties/Schools: |
UK Campuses > Faculty of Medicine and Health Sciences > School of Medicine |
| Item ID: |
80899 |
| Depositing User: |
Atallah, Nehal
|
| Date Deposited: |
23 Jul 2025 04:40 |
| Last Modified: |
23 Jul 2025 04:40 |
| URI: |
https://eprints.nottingham.ac.uk/id/eprint/80899 |
Actions (Archive Staff Only)
 |
Edit View |
Tools
Tools
