Circulating tumour DNA and cell-free methylated DNA

ctDNA obtained via liquid biopsy has shown serious potential not only in the early diagnosis of cancer but inter alia as an effective marker for its recurrence, longitudinal monitoring and response to therapy.12 Current methods aimed at detecting ctDNA, however, are mostly based on sequencing somatic mutations from cell-free DNA (cfDNA), a process constrained by relevant limitations in terms of clinical applicability due to (a) expectable low sensitivity in early stage cancers given the limited number of recurrent mutations in ctDNA,8 13 (b) the vast heterogeneity of DNA mutations occurring in a single tumour together with non-specific mutational profiles along different patients and cancer types,12 13 (c) current cost-prohibitive impact of cfDNA next generation sequencing (NGS) for mutation assessment.14 On the other hand, the assessment of cfMeDNA may overcome the limitations outlined above. DNA methylation is commonly involved in cellular development, tissue-specific gene expression, and regulation of imprinted alleles, with widespread effects on cellular growth and genomic stability.15 Changes in methylomes in cancer are usually associated with alterations of the transcriptional outcome and genomic instability, priming or enhancing carcinogenesis. Given the impact of methylation changes on cellular equilibrium, different methylomes may be associated with specific biological features, providing useful information for the early diagnosis of cancer.16 Shen et al recently developed a sensitive, immunoprecipitation-based protocol to analyse the methylomes of small quantities of cfDNA, providing an efficient method to detect large-scale DNA methylation changes that are enriched for tumour-specific patterns.8 cfMeDNA immunoprecipitation and high-throughput sequencing (cfMeDIP-seq) showed high accuracy in the detection of cancer-derived DNA methylation events in cfDNA, managing to distinguish multiple cancer types from healthy controls and discriminate different and specific methylation patterns across diverse cancers.

RNA-based biomarkers

RNA-based biomarkers can be referred either to coding or non-coding RNAs. Non-coding RNAs, including miRNAs and long non-coding RNAs, have been extensively studied over the last years as promising markers for cancer early diagnosis and monitoring, being of particular interest for their stability and quantity in the bloodstream.17 On the other hand, coding-RNAs include the cell-free messenger RNA, whose evaluation may allow the non-invasive assessment of the whole-body transcriptome18 with relevant implications for several clinical purposes.

Circulating proteins

Proteomic analyses in neoplastic patients have been evaluated in regard to diverse clinical purposes with suboptimal results, given the challenging characterisation of circulating proteins. The assessment of alterations in protein-mediated signalling networks is of great interest for its possible implications in the early diagnosis and monitoring of cancer, besides the detection of potentially actionable targets for therapeutic purposes.11

Peripheral blood and urine collection, processing and storage

One investigator, a research nurse, and two biologists are collaborating in the first 18 months of the project to collect at t0 and t1 two PAXgene Blood ccfDNA Tubes CE-IVD (PreAnalityX, GmbH), one BD Vacutainer K2E (EDTA) Plus Blood Collection Tube CE-IVD (BD life Sciences), one Tempus TM Blood RNA tube (Applied Biosystems) and one urine specimen in a sterile container. Blood samples are processed to extract and store cfDNA from plasma, and to collect proteins, exosomes, peripheral blood mononuclear cells (PBMC), and total RNA. PAXgene tubes are first centrifuged for 15 min at 1900 rcf at room temperature (RT), then the collected plasma is further centrifuged for 10 min at 1900 rcf at RT. EDTA tubes are centrifuged for 15 min at 1600 rcf at RT and the collected plasma is further centrifuged at 1900 rcf for 10 min at RT. Plasma is then aliquoted in cryovials. Tempus tubes for total RNA extraction are immediately stored at −80°C. At the moment of collection, urine is mixed with Cell-Free DNA Urine Preserve (Streck) in order to stabilise cfDNA in samples for up to 7 days at a temperature ranging from 6°C to 37°C. Urine is centrifuged at 2680 rcf for 10 min, and the supernatant is aliquoted into 15 mL tubes and stored at −80°C until cfDNA extraction. Samples are stored at −80°C in a dedicated, Eppendorf CryoCube F740hi ULT Freezer, three compartments.

All test tubes, from blood collection to storage, are barcoded to increase traceability and anonymity to external personnel.

Resources

A 150 m² fully equipped laboratory supports the pivotal work on the samples collected for our project. All procedures of platelet separation and plasma processing, cfDNA and RNA extraction, quality control, amplification, library preparation and sequencing happen in separate, clean environments optimised for such tasks. Every operation, from sample processing, to nucleic acid extraction and sequencing, is semi-automated in order to minimise cross-sample contamination and ensure optimal reproducibility and consistency of operations. A dedicated freezer is available in the laboratories for sample storage. The laboratory is equipped with a ThermoFisher Scientific Ion S5 XL sequencer. Data collection and the dry-lab part of the analyses are performed on high performance workstations maintained in high-security, dedicated environments to avoid the risk of data loss and sensible data cyber-theft.

Ultrasensitive NGS on ctDNA

We designed a tagged-amplicon NGS panel covering 101 short regions of frequently mutated genes in BC. For this task, we analysed the GENIE dataset V.3.0,19 which includes 3269 sequenced primary and metastatic BC. This dataset was chosen for its size and homogeneity. Since our purpose was looking at the most frequent, non-overlapping (ie, mutually exclusive) mutations, the sample size of GENIE was deemed as adequate for the panel design. Among the most common cancer genes in BC, very few behave as dominant oncogenes (ie, with frequent mutation in limited regions of the gene, like PIK3CA), whereas several behave as tumour suppressors (ie, with rare mutations occurring along the whole gene, like CDH1). By dividing all the genes sequenced by GENIE into 9603 regions shorter than 50 base pairs, we realised that to detect at least one mutation, we would need to sequence only 101 regions to cover 80% of the whole GENIE dataset, but 420 regions to increase such detection rate to 88%. In particular, amplicons covering TP53 (30 exonic regions), CDH1 (50 exonic regions), GATA3 (13 exonic regions), and PIK3CA (7 exoic regions) would optimise the coverage by minimising the size of our panel. TP53 is planned to be analysed, together with CDH1, for all its exons. On the other hand, PIK3CA presents mostly hotspot mutations in exons 9 and 20. Hence, only those two exons are going to be covered by this design. The 101 selected regions will be assessed in plasma ctDNA using a custom design, which leverages the novel, proprietary tagged-amplicon OncoMine cfDNA methodology by ThermoFisher Scientific. Its use allows for a limit of detection of 0.1% at 20 000× sequencing coverage with 20 ng of circulating DNA, well attainable with two peripheral blood samples collected in PAXgene tubes. ThermoFisher defines ctDNA positivity as the detection of three molecular families. However, this threshold is arbitrary and will be adapted depending on the actual class the samples fall in, in order to optimise the accuracy of the multiomic HDI. cfDNA extraction is performed by using the QIAamp Mini Elute cfDNA Mini or Midi Kit (Qiagen, Hilden, Germany) on the Qiacube system, according to the Manufacturer’s instructions. cfDNA is extracted from at least 4 mL of plasma collected in the EDTA tubes and 2 mL of plasma collected in the PAXgene tubes, managing to obtain the requested amount of cfDNA for cfMeDIP-seq (~5 ng) and the assessment of ctDNA mutations (~15 ng). cfDNA is processed on an Ion Chef fluidic handler, and sequenced using the tagged-amplicon methodology with our custom-design panel on a Ion S5 sequencer. Tertiary analysis will be performed on a dedicated Ion Server System available in our laboratory. The library size of our custom panel is half of the commercial ThermoFisher Scientific Ion Torrent OncoMine Comprehensive cfDNA panel, from which the aforementioned performance considerations have been derived. Hence, the use of our custom-design panel appears feasible and should perform better in light of its smaller size and subsequent deeper attainable sequencing coverage.

Methylome profiling of cfDNA

Recent works pointed out encouraging results in terms of accuracy of cfMeDIP-seq in the detection of informative methylation changes of small quantities (1–10 ng) of plasma cfDNA for the diagnosis of renal cell carcinoma and intracranial tumours,20 21 with the former being effectively detected also by performing cfMeDIP-seq on urinary cfDNA. According to the protocol elaborated by Shen et al, cfMeDIP-seq involves four steps: (a) cfDNA end repair, A-tailing and adapter ligation; (b) cfMeDNA immunoprecipitation and enrichment by using an antibody targeting five methylcytosine; (c) library preparation; (d) high throughput NGS on an Illumina platform for cfMeDNA data.22 This approach based on immunoprecipitation allows to avoid cfDNA bisulfite treatment, typically used to study DNA methylation but associated with a high rate of DNA degradation.

Proteomics analysis

The proposed task is based on the relative quantification of circulating plasma proteins by a novel, highly multiplexed proteomic assay (SomaScan).23 24 SomaScan (SomaLogic) is considered the most comprehensive protein array available so far for the relative quantification of proteins. Over 7500 proteins can be simultaneously assessed from 55 µL of plasma. This technology involves a new type of aptamers, which are single stranded DNA molecules able to bind proteins, called SOMAmers. In addition to their high affinity for individual proteins, SOMAmers have a unique 40-nucleotide sequence tag and a fluorescent label that allows their identification and quantification in high-density microarrays. SOMAmers have been successfully assembled in a commercial product allowing the comparative evaluation of proteins in a quantity of serum or plasma (or other biological fluids) as low as 55 µL. The SOMAscan assay is a highly multiplexed, sensitive, quantitative and reproducible proteomic tool for biomarker discovery and development. SomaScan can detect proteins within a range of 10 logarithms, allowing for an unmatched sensitivity to detect even femtomolar protein concentrations. The analysis of the SomaScan is performed by using classic DNA array data analysis and is based on bioinformatics tools that have been developed for gene array analysis.

Radiomics analyses

A preliminary radiomics classifier has been developed by the Diagnostics Senology team of Ospedale Policlinico San Martino, based on digital breast tomosynthesis (DBT) images of consecutive participants from the ASTOUND (Adjunct Screening With Tomosynthesis or Ultrasound in Women With Mammography-Negative Dense Breasts, ClinicalTrials.gov Identifier: NCT02066142) trial.2 Radiomics analyses were performed on all DBT images within manually selected regions of interest (ROIs) including all the dense parts of the breast and excluding the fatty parts. ROIs were selected by a single radiologist, with proven expertise in quantitative image analysis. Descriptors of the preliminary classifier were selected after initial screening of 104 radiomics features to reduce the risk of over-fitting and according to features previously used to associate breast parenchymal patterns with cancer risk.25 For the present project, images features will be extracted from the same cases and matched controls for whom NGS and proteomics analyses are performed, using an open source software platform for medical image informatics and advanced deep learning methodologies.

HDI classifier

Sensitivity and specificity of ctDNA analysis, cfMeDNA tests, proteomic analysis, and radiomics for the early detection of BC will be assessed. Data from these classifiers will be subsequently processed and used to generate the HDI model, based on the ensemble learning approach methodology. Ensemble learning combines predictions of multiple individual classifiers obtained by different techniques such as random forest, support vector machine or general linear modelling in order to enhance generalisation power,26 avoid overfitting, and increase the strength and reliability of the final outcome.27 Specifically, the outcomes from ctDNA, cfMeDNA, proteomics and radiomics tests will be combined by using a weighted-majority voting approach implemented in the R environment (caret package). We do not envisage RNA-based classifiers to be initially included in our multianalyte model.

Experimental validation and other analyses

On completion of these experiments, we will seek to validate the possible discovery of novel biomarkers and to facilitate their transfer to clinical applicability. Other promising non-invasive biomarkers will be studied, thanks to the unique sample set at our disposal. We foresee the possibility to perform exosome-enriched miRNA sequencing and PBMC transcriptome sequencing by using the Ampliseq Transcriptome solution on our Ion S5 XL sequencer. Eventually, such analyses may be integrated into the HDI classifier to augment the accuracy of BC early diagnosis.

Projected sample size

The Diagnostics Senology Unit of San Martino Hospital is the highest-level referral centre in Italy for a population basin of more than 2 000 000 people. In a single year, approximately 15 000 mammograms are performed (figure 2). Of these, 1500 yield a radiological suspicion of malignancy with an ensuing biopsy. Assuming a 50% refusal rate by our patients to undergo trial-specific blood collection, we foresee to enrol 750 patients with radiologically suspect small breast lesions. Among these patients, we assume diagnosis of invasive BC will be confirmed in one-third. Of these 250 bioptic diagnoses of BC, which will undergo primary surgery, as per internal historical records, 40% are assumed to be pT1 and 60% higher than pT1. Assuming another 10% failure rate in sample processing and storage, we will have the potential to collect imaging data and samples from approximately 90 patients with pT1 BC and 180 radiological size-matched lesions in a single year (figure 2). Patient enrolment is projected in a time frame of 12 months from the beginning of the study, whereas samples collection at t1 is expected to end approximately 6 months after the enrolment of the last patient.

• Download figure
• Open in new tab
• Download powerpoint

Figure 2

Sample size diagram. Approximately 1500 breast biopsies per year are performed at the Diagnostics Senology Unit of San Martino Hospital. Of a projected number of 750 liquid biopsies, we foresee to collect samples and acquire mammograms from at least 49 patients with stage I–IIA BC, and 98 patients with radiological size-matched lesions, along with those samples and images acquired from 100 healthy women with two consecutive negative mammograms. BC, breast cancer.

Sample size calculation

The sample size required for our analyses is N=147, with 49 biopsy-proven stage I–IIA BC cases and 98 biopsy-proven benign lesions of similar radiological size. In particular, assuming that the best non-HDI classifier is, in the end, the one with the highest number of tested variables (worst-case scenario, for example, transcriptome sequencing on platelets with approximately 20 000 transcripts analysed), we would need28 29:

• N=87 samples for a training set, in a 1:2 ratio between histologically proven BCs and benign lesions matched for radiological size, with a standardised fold change of 1.2, n~20 000 features to be assessed, tolerance=0.05 from the best possible classifier, as defined by Dobbin et al.29

• N=26 samples for a testing set (70/30 split of samples between training and testing set) with similar ratio of cases and controls.

• N=34 samples for a validation set (30% of the sum of training and testing sets).

It may be possible that, adopting HDI classification, the actual needed sample size will be smaller than calculated. At present however, we are not aware of well-established statistical approaches to obtain a more robust estimation of the number needed for our experiments other than the one we adopted.

Amendment #1: collection and analysis of urine samples

The first amendment to this protocol, presented on 18 February 2020, involved the collection and analysis of urine samples in addition to blood samples, which was not included in the original version. An efficient method for the early diagnosis of BC based on the assessment of urinary biomarkers would substantially eliminate the invasiveness of the overall procedure, possibly facilitating the execution of large-scale screening campaigns.

Amendment #2: healthy controls

The second amendment, presented on 4 February 2021, involved the recruitment and analysis of a cohort of 100 healthy women with negative mammography (BIRADS-1 or BIRADS-0 with adjunct negative ultrasound) as a healthy control group, in addition to patients undergoing breast biopsy. As the presence of benign breast lesions may determine possible variations in blood and urinary analytes, especially in case of in situ tumours, the purpose of the present amendment is to possibly categorise patients on the presence or absence of non-malignant breast lesions, enhancing the accuracy of the HDI classifier.

Amendment #3: redefinition of the study cohort

The third and final amendment, presented on 27 May 2021, involved several points. According to the original protocol, concerning the cohort of patients with histological diagnosis of invasive BC, only samples collected at t0 from patients with T1N0 tumour at surgery would have been analysed. Moreover, the collection and analysis of t1 samples would have involved only this group of patients, with the exclusive aim of distinguishing between tumour-specific and host-specific molecular alterations in connection with the presence/absence of BC. With the present amendment, samples collected at t0 will be analysed from patients assessed as with stage I–IIA tumour at surgery (T1N0 or T2N0 or T1N1a), allowing an effective expansion of the sample size, while remaining in the setting of early stage BC. Furthermore, the collection and analysis of t1 samples will include all patients diagnosed as with invasive BC and sampled at t1, without limitations in terms of tumour stage assessed at surgery. The analysis of t1 samples, coupled with patients’ longitudinal monitoring performed as per normal clinical practice, will allow the assessment of the accuracy of a multi-analyte evaluation for the prediction of BC recurrence, added as exploratory aim of the present study.