Advancing disease detection and risk prediction is a key focus of ongoing medical research and has had a prominent presence on front pages during the recent trial of Theranos founder Elizabeth Holmes. While the Theranos product – claiming to detect disease using only few drops of blood – did not deliver on its promise, the sheer amount of attention and investment the company attracted highlights the need for new diagnostic tools that can detect or predict diseases and are minimally invasive.
Genetic risk scores cannot account for dynamic changes nor indicate environmental exposure
Several other research projects were luckily more successful and made tremendous progress in identifying biomarkers of disease (and risk thereof), especially cancers. For risk prediction, a good example is the breast cancer polygenic risk score which looks at 313 genetic variants and can aid clinical risk stratification (area under receiver operating characteristic curve=0.63 1, in combination with other tools up to 0.68 2). However, despite advances in genetics-based risk prediction for several cancers, we know that cancer development is largely driven by non-genetic factors, such as lifestyle, environment, or hormone exposure. Genetics-based risk scores are static in time and space (they will likely not change over time and are also not tissue-dependent). In contrast, individual disease risk will likely change over time and depends on exposure to certain risk factors, for instance obesity, smoking, or viral infections. It seems to be the sum of all parts – genetics and environment – that dictates whether an individual will develop a malignancy. Genetic risk scores cannot account for dynamic changes nor indicate environmental exposure.
However, there are other molecular signals that can integrate and record the effect of non-genetic factors in our cells. Lifestyle, hormonal, or environmental factors leave epigenetic modifications on our DNA in the form of DNA methylation (DNAme). DNAme therefore lies at the interface of genetics and the environment and may provide a piece of the puzzle we were missing so far: integration of genetic and non-genetic risk factors that can dynamically change and potentially also enable us to monitor disease risk over time.
Epigenetics for detection and prediction the risk for women’s cancers
Our group at the European Translational Oncology Prevention and Screening (EUTOPS) Institute (Universität Innsbruck and University College London) is dedicated to reducing the incidence and mortality of women’s cancers. Two of these cancers in particular, breast and ovarian cancer, pose a significant global health and socioeconomic burden. Breast cancer is the most common cancer in women and accounts for the second most cancer-related mortalities. Ovarian cancer is less frequent than breast cancer but is the most common cause of death from gynaecological malignancies. It is often only detected late as there are no screening or risk prediction tools (a recent large-scale trial in the UK recently failed to describe a survival benefit when measuring longitudinal changes of CA-125 3), and has a poor 5-year survival rate. Screening is available for breast cancer, but the benefit of mammography for mortality reduction has recently been questioned 4 and the abovementioned polygenic risk score cannot be used for longitudinal monitoring of changes in risk.
Breast and ovarian cancers are epithelial-derived and largely hormone-sensitive, and we therefore aimed to find a sample that can fulfil both these criteria and is additionally easily accessible
To assess whether DNAme signatures in surrogate samples are associated with either of these cancers, the choice of surrogate is vital. DNAme is highly tissue specific – it is a key regulator of cellular phenotypes after all. Breast and ovarian cancers are epithelial-derived and largely hormone-sensitive, and we therefore aimed to find a sample that can fulfil both these criteria and is additionally easily accessible. A large majority of women aged 25-64 years (exact age ranges are country-specific) already participate in regular cervical cancer screening programmes, where a cervical sample is taken and assessed for the presence of human papillomavirus (HPV) or cellular changes. We were interested to see whether these samples, which include hormone-sensitive epithelium, could be suitable for detection and possibly risk prediction of breast and ovarian cancer. This could have the added benefit that not one, but several cancers could be detected using a single sample.
This premise was at the heart of the FORECEE (4C) project, spanning several European centres including the UK, Italy, Germany, the Czech Republic, and Norway. Our UCL team collected cervical samples from women presenting with poor prognostic breast or ovarian cancers as well as controls and assessed whether we could distinguish between cases and controls. Poor prognostic cases were selected to identify signals associated with more aggressive tumours which would benefit from earlier diagnosis or risk prediction. Applying penalised linear regression on DNA methylation data, we identified specific breast (WID-BC - Women’s cancer risk identification – breast cancer) and ovarian cancer (WID-OC) signatures which could detect these anatomically distant cancers in external validation sets with high accuracy (AUC=0.81 for WID-BC, AUC=0.76 for WID-OC).
Presence of tumour DNA or systemic epigenetic programming defect?
As blood-based cancer screening methods rely on the presence of cell-free DNA or tumour material, we were curious to see whether we detected tumour material or indeed captured a systemic epigenetic programming defect associated with cancer. Inferring the amount of tumour DNA based on tumour methylation profiles, we found tumour material in very few samples of patients with ovarian cancer, while the vast majority of samples did not contain tumour DNA. A cancer classifier defined using breast cancer tissue methylation data from The Cancer Genome Atlas (TCGA) was not able to distinguish breast cancer cases from controls in cervical samples, which further indicated we weren’t picking up cancer-related material but rather detecting an epigenetic footprint associated with cancer risk.
We weren't picking up cancer-related material but rather detecting an epigenetic footprint associated with cancer risk
To evaluate the systemic epigenetic footprint further, we looked at methylation patterns associated progesterone, a driver of breast cancer development, in breast tissue adjacent to cancer and cervical samples from breast cancer cases. There was a significant overlap of hypermethylated progesterone receptor binding sites compared to normal breast tissue or cervical samples from controls, respectively. Likewise, the WID-OC-index mirrored the epigenetic program of cells at risk of becoming cancerous in BRCA1/2 germline mutation carriers (i.e. mammary epithelium, Fallopian Tube fimbriae, prostate). These findings indicated that patterns of risk exposure or systemic epigenetic programming defects could be recorded in the cervix as a surrogate tissue for the breast or ovary.
Prospective assessment in biobanked samples
To validate risk prediction, we needed to obtain samples predating diagnosis. We had the pleasure of collaborating with researchers from the Swedish Karolinska Institutet that run a large biobank of a Swedish cervical screening cohort. Some women in this cohort developed a breast cancer in years following sample collection. This gave us the unique opportunity to investigate whether the WID-BC index would be elevated in samples predating diagnosis… but of course experiments don’t always work as planned. The samples had been biobanked at -25°C in methanol and hence suffered degradation in Open Sea regions, where the majority of CpGs in the WID-BC index is located. We developed a method to quantify this and determine the signal to noise ratio. This enabled us to show that with increasing sample quality (signal to noise), we were able discriminate between future cases and controls for poor prognostic cases with an AUC of 0.7 in the least-degraded samples, indicating the test had the potential to pick up cancer (risk) in samples predating diagnosis. This was not the case for good prognostic cases and was reassuring as we specifically aimed to identify aggressive cancers.
Outlook
Our two new WID-tests based on cervical samples may offer an innovative way of diagnosing and risk predicting breast and ovarian cancer. Some areas where this could be particularly beneficial in the future are:
- ovarian cancer screening – currently no screening in place
- guiding treatment decision for BRCA1/2 mutation carriers – they have a higher risk for both breast and ovarian cancer but could now make an informed decision about their individual risk and whether risk-reducing surgeries are needed
- individual disease risk monitoring – assessment of changes in risk following risk-reducing measures, both pharmaceutical (progesterone antagonists) or lifestyle-based (exercise, smoking cessation)
Further studies are now in planning to evaluate our WID tests in large-scale or even population-based prospective trials to compare these screening methods to current standard of care and investigate whether they can induce a clinically meaningful stage shift without overdiagnosis.
Read our papers
- The WID-BC-index identifies women with primary poor prognostic breast cancer based on DNA methylation in cervical samples. J E Barrett, C Herzog, et al. Nature Communications, 2022.
- The DNA methylome of cervical cells can predict the presence of ovarian cancer. J E Barrett et al., Nature Communications, 2022.
... or watch the video below for a brief visual introduction to the WID tests.
About the European Translational Oncology Prevention and Screening (EUTOPS) Institute
The EUTOPS Institute focuses on real-world implementation of novel methods for personalised primary and secondary cancer prevention. We’re a multidisciplinary team based at Universität Innsbruck, Austria, and University College London, UK.
The work presented was supported by the European Union (H2020 FORECEE Programme and the European Research Council Advanced Grant BRCA-ERC) and we foster a long-standing relationship with the UK Charity The Eve Appeal who has further enabled this research and provided guidance for a patient-centric research approach.
References
1 Mavaddat, N. et al. Polygenic Risk Scores for Prediction of Breast Cancer and Breast Cancer Subtypes. Am J Hum Genet 104, 21–34 (2019).
2 Choudhury, P. P. et al. Comparative validation of breast cancer risk prediction models and projections for future risk stratification. J National Cancer Inst 112, 278–285 (2019).
3 Menon, U. et al. Ovarian cancer population screening and mortality after long-term follow-up in the UK Collaborative Trial of Ovarian Cancer Screening (UKCTOCS): a randomised controlled trial. Lancet 397, 2182–2193 (2021).
4 Autier, P. & Boniol, M. Mammography screening: A major issue in medicine. Eur J Cancer 90, 34–62 (2018).