Accelerating diagnosis with digital pathology based biomarker analysis

Modern pathology is being transformed by digital pathology. Pathologists have historically examined tissue samples mounted on glass slides, and while that workflow has long been the gold standard for diagnosing many diseases including cancer, it carries a built-in degree of variability from subjectivity and human error.

What is digital pathology?

Advances in digital pathology have solved many of the difficulties posed by traditional glass slide workflows. The term refers to the collection, analysis and dissemination of digitised pathology specimens. Whole slide imaging (WSI) makes it possible to scan an entire specimen at diagnostic quality, and digitised slides occasionally offer better resolution of the features that aid diagnosis. Combined with AI algorithms built on machine learning, the technology’s full potential is realised.

Biomarker analysis in practice

Biomarkers are increasingly crucial to the development of cancer medications, because many new oncology drugs are created as precision medicines in which a patient’s biomarker profile determines which treatment suits them. To increase the likelihood of finding biomarkers that stratify patients by who is most likely to respond or develop resistance, the pharmaceutical industry is integrating biomarker strategies earlier in R&D. Biomarkers can support efficacy, safety, pharmacodynamics and mechanism of action — all difficult to establish, all highly instructive.

The technologies

Several techniques are used to find tissue biomarkers: fundamental stains that distinguish cellular compartments such as H&E, and molecular assays for protein biomarkers such as immunohistochemistry (IHC) or immunofluorescence (IF). In situ hybridisation with chromogenic or fluorescent options remains common for detecting RNA and DNA biomarkers despite advances in sequencing. Because many tissue biomarkers are protein-based, robust IHC and IF protocols exist for several significant cancer targets:

• Tumour-associated antigens (TAAs) such as growth factor receptors — EGFR, HER2
• Markers of cancer mechanism — angiogenesis, tumour growth and apoptosis (Ki-67, cleaved caspase-3, CD31, VEGF)
• Immune checkpoint markers — PD-1 and PD-L1, which can predict a cancer type’s response to immunotherapy

Biomarkers in decisions

HER2

HER2-positive breast cancer is one subtype of breast cancer, in which the gene producing the HER2 protein is duplicated in the cancer cells — this occurs in roughly one in five breast cancers. The protein encourages the growth of cancer cells, and HER2-positive cancers are typically more aggressive than other subtypes. Trastuzumab, a monoclonal antibody, binds to HER2 to prevent the cancer cells proliferating and dividing.

ER and PR

Examining the tumour for both oestrogen and progesterone receptors is widespread practice in breast cancer diagnosis, and the accuracy of the result is crucial because it directs treatment. IHC is currently the most common technique: from a tissue sample — a biopsy, or tissue from a surgical procedure — it can identify oestrogen and progesterone receptors in cancer cells.

PD-L1

Programmed death-ligand 1 is a protein present on the surface of many cells throughout the body. Large amounts of PD-L1, present in some tumour cells but not all, help those cells evade the immune system. A laboratory test determines the quantity of PD-L1 on tumour cells, which may help determine how certain cancers should be treated. T cells, a category of white blood cell, are one of the immune system’s weapons against cancer; certain T-cell subtypes can identify and attack cancer cells directly.

IHC as a companion diagnostic

In breast pathology, IHC is used as a companion diagnostic for a particular therapy as well as for diagnosis and prognostic stratification of malignant lesions. Antibodies recognise the epitopes of a particular target — usually a protein, but sometimes a lipid or carbohydrate. Used with cytological preparations, fresh frozen samples and formalin-fixed paraffin-embedded tissue, IHC detects the presence and expression pattern of a marker in situ through antigen-antibody reactions.

Where manual workflows fall down

Each of these IHC processes can be altered, and there is no universal protocol applicable to all IHC markers. Antibodies, staining protocol, interpretation and scoring all vary, and the data is inconsistent. Each marker’s specificity, sensitivity, staining pattern, validation procedure, antibody clones, detection kits, cross-reactivity, controls and pitfalls differ and depend on several variables. Tissue type, fixation characteristics, the staining kit and reagent, the staining condition, and differences between observers who may apply different thresholds and subjective interpretation can all cause significant variation in IHC performance.

That last source of variation is the one worth dwelling on. It is not a knowledge gap that more training closes. Two competent pathologists applying different thresholds to the same slide will report different numbers, and neither is wrong. It is a measurement problem, and measurement problems are addressed by instrumentation.

AI-assisted diagnosis

Digital pathology-based biomarker analysis offers considerable potential for selecting the cancer drug candidates most likely to succeed in clinical trials, including immuno-oncology agents. The transition to a digital and automated framework lessens the effects of subjectivity and human error, while encouraging the research collaboration that makes diagnosis faster and more accurate.