Evaluating cell-free DNA-based blood tests for early detection of multiple cancers

By Neha MathurReviewed by Aimee MolineuxNov 21 2022

In a recent study published in Cancer Cell, researchers assessed several approaches for a circulating cell-free deoxyribonucleic acid (cfDNA)-based multi-cancer early detection (MCED) test. Defining the clinical limit of detection (LOD) based on circulating tumor allele fraction (cTAF) enables the comparison of different approaches.

An MCED test is a blood test that helps early detection of a shared cancer signal across multiple cancers using blood samples. Currently, available MCED tests have a low false-positive rate of less than 1%.

Background

The discovery that DNA from various tissues in the human body exists in the blood and other bodily fluids outside of cells (cfDNA) has led to several blood-based cfDNA tests. Such tests find clinical applications, such as facilitating interrogating specific genomic abnormalities, for instance, actionable tumor-derived mutations for targeted cancer therapy selection.

Recent modeling work predicted that adding an MCED test to standard care may improve early-stage detection and prevent 39% of all cancer-related deaths within five years of diagnosis.

Complementary MCED testing might also allow population screening across numerous deadly cancer types at once.

An abundant background of non-cancer cfDNA in the blood relative to the genomic material shed from the tumor, and the prevalence of somatic biology (e.g., clonal hematopoiesis (CH)) might confound specific cancer signal detection. Thus, researchers continuously pursue new approaches, such as machine-learning techniques, to overcome the signal-to-background ratio challenges associated with blood-based MCED tests.

About the study

A total of 2,800 participants, 1,628 with cancer and 1,172 without cancer, participated in the first circulating cell-free genome atlas (CCGA) substudy. The researchers randomly assigned 1,414 and 847 participants to independent training or validation sets and obtained samples that met prespecified laboratory quality control standards. Also, they ensured with cancer screening that they enrolled only those participants with cancer who had not begun their cancer therapy.

The team used de-identified blood samples from participants at sites in the United States and Canada and extracted cfDNA from plasma. The median time between blood collection and plasma isolation was less than two days. They processed a cfDNA blood sample using three sequencing methods: whole-genome bisulfite sequencing (WGBS), TS, and WGS. To calculate the LOD for the second CCGA substudy, they extracted genomic DNA (gDNA) from scrapings of formalin-fixed, paraffin-embedded (FFPE) tissue in-house.

Study findings

The current study had several important findings. First, more than cancer type and clinical stage, cTAF was responsible for most of the cfDNA variations from cancer signals. Thus, the researchers observed extensive variation in cTAF between cancer types and within single stages, which indicated that its clinical stage alone might not be the sole predictor of the amount of tumor-specific genomic features.

Cancer types might have dramatically different shedding rates even after controlling for the stage. Based on the distribution of tumor shedding across cancers, tumor biopsy sequencing could estimate cTAF across cancer types and their clinical stages. Likely due to increased tumor shedding in advanced-stage cancers, cTAF typically increases with the clinical stage. As expected, cancer signal detection improved for each cancer with increasing stage and cTAF, although this approach could not reliably detect all stage IV cancer signals. A plausible explanation is that molecular factors in undetected stage IV cancers, such as low mitotic activity, are associated with lower tumor DNA shedding and lower cTAF. Likewise, physical factors, the lesser surface area of the tumor and microscopic tumor extent (i.e., its access to the blood supply) result in lower tumor DNA shedding and lower cTAF.

Second, the researchers observed that clinical staging might not fully capture tumor behavior. On the contrary, even if a stage I–III cancer showed higher cTAF due to active proliferation and high tumor DNA shedding, it could well detect tumor behavior. There might be a correlation between cfDNA cancer signals and cTAF with more hostile tumors; thus, cfDNA assays might better detect clinically significant cancers. Indeed, cancer prognosis is better with lower cTAF. Furthermore, the Surveillance, Epidemiology, and End Results (SEER) program reported that cancers not detected by cfDNA-based tests showed significantly better survival.

Third, based on the observations that there is a strong correlation between classifier signal detection and cTAF, a clinical LOD using cTAF could be developed. It could provide a metric for classifier optimization that would account for the extensive tumor-shedding variations within a clinical stage and between cancer types. This clinical LOD-based metric could directly compare the performance of cancer signal detection across multiple cfDNA-based assays. However, the condition is that cTAF estimates use tumor-biopsy-verified features at equivalent test specificity levels.

Finally, the WG methylation was the most promising option in this study because, with ≈30 million CpGs, it was a pervasive signal across the genome. Of all cfDNA features assessed in this study, it was among the most sensitive methods for the following reasons:

i) did not require WBC sequencing,

ii) exhibited one of the lowest clinical LODs, and

iii) had the highest cancer signal origin (CSO) prediction accuracy.

Also, methylation patterns along each gene segment contain a robust tumor-specific signal, readily identifiable above normal genomic background variation. It, in turn, facilitates the detection of the methylation signal at lower cTAF levels than the other cancer features of the genomes tested in the study. The WGBS assay generated WG methylation features, thus, also showed the most likelihood for improvement among the three assays.

Another study performed WGS and TS with sequencing depth and breadth of 30× and 60,000×, covering 507 genes, respectively. Since they removed technical noise using unique molecular identifiers and CH suppression using WBCs from this data, the results likely represented the upper limit of performance for WGS and TS assays for a practical MCED test. The bisulfite sequencing assay targeted the CpG-containing regions most likely containing cancer- and tissue-specific WG methylation patterns in cfDNA. It allowed increased sequencing depth while controlling complexity facilitating improvements in clinical LOD.

The researchers used a methylation-based approach for further development because of its optimization potential and superior performance. Compared to WG methylation, the clinical LOD for the targeted methylation classifier validated in the second CCGA substudy showed nearly an order-of-magnitude improvement. Subsequent improvements in specificity, sensitivity, and CSO accuracy of the targeted-methylation-based assay and classifier supported the clinical implementation of the Galleri® MCED test. Overall, the study findings add to the evidence on methodologies deploying cfDNA methylation patterns for cancer detection.

Conclusions

According to the authors, none of the prior studies have reported systematic comparisons of various genomic features from cfDNA for MCED testing.

The first CCGA substudy showed that clinical LOD is a valuable benchmark to assess classifier performance. It could enable comparison between studies, provided specificity and detection probabilities are equivalent. The study data also suggested that cTAF may be a more direct and accurate measure of the underlying tumor biology. It is a better driver of cfDNA cancer signal detection than current prognostic indicators, such as stage and cancer type.

Thus, strategies for MCED test optimization should include efforts to improve detection at lower cTAF levels. Furthermore, WG methylation from cfDNA used in a prototype MCED test provided the best performance among the approaches characterized for cancer signal detection and CSO prediction without requiring additional sequencing to correct for WBC background.

Finally, the study results following the evaluation of the top-performing prototype tests informed the design and performance of the recently reported targeted-methylation-based cfDNA MCED test, the Galleri® MCED test. It showed marked improvements relative to all the tests evaluated in the study.