NEW YORK – City of Hope and Translational Genomics Research Institute investigators have pioneered a new cell-free DNA sequencing approach for cancer early detection, which they believe can offer a simpler and more efficient tool than existing epigenetic and fragmentomic tools.
The team published their results in Science Translational Medicine last week, describing what they have dubbed Alu Profile Learning Using Sequencing (A-PLUS), which they developed and validated in a series of case-control sample sets that included 11 cancer types. Although its performance will have to be borne out in additional validation studies, A-PLUS's sensitivity and specificity appear equivalent so far to other leading multi-cancer screening assays like Grail's Galleri test.
A-PLUS is designed to detect the differences in repetitive regions of cell-free DNA, which investigators were able to tune, using machine learning, into a discriminator of normal and cancer-associated signals. Kamel Lahouel, assistant professor in TGen's integrated cancer genomics division and the study's co-first author, said that unlike other methods that derive fragmentation signals from shallow whole-genome sequencing (WGS), A-PLUS uses an amplicon-based approach, which allows for deeper sequencing of target regions of the genome.
The core sequencing technology, called RealSeqS, was initially designed by researchers in the lab of Bert Vogelstein at Johns Hopkins to assess copy number changes across what are known as Alu elements — short interspersed nuclear elements of about 300 base pairs, with more than 1 million copies spread throughout the human genome.
According to the study authors, RealSeqS's advantages over WGS include a simpler workflow with no required library construction, reduced requirement for input DNA, faster computational analysis, and higher sequencing coverage at individual Alu loci. Specifically, the workflow uses a single-primer pair to jointly amplify approximately 350,000 Alu elements. This allows vastly greater coverage of these regions than would be possible with WGS at an equivalent sequencing depth, enabling improved predictive modeling.
Although their role in biology and evolution is an ongoing area of research, Alu elements have been shown to be involved in gene regulation and structural changes and are known to reflect the altered fragmentation patterns found in the cell-free DNA of patients with cancer. Investigators had previously used RealSeqS to track aneuploidy patterns, but Lahouel and colleagues thought that looking for other features could help identify cancer-specific signals.
"The hypothesis is that you would observe a strong signal in these repetitive elements, in cancer versus normal in terms of abundance of the sequences when you amplify," said Lahouel. "If DNA is fragmented inside those repetitive elements, then you cannot amplify them, and you would not see them. If they are not fragmented, then you would be able to amplify them and see them."
In the Science Translational Medicine study, a team led by Lahouel and Cristian Tomasetti, director of City of Hope's Center for Cancer Prevention and Early Detection, applied A-PLUS to 7,657 samples from 5,980 people — 2,651 of whom had cancer of the breast, colon/rectum, esophagus, lung, liver, pancreas, ovary, or stomach.
They divided these samples into four cohorts for model training, determination of the best detection threshold, validation, and reproducibility tests. According to the authors, A-PLUS alone provided a sensitivity of about 40 percent across 11 different cancer types in the validation cohort, at a specificity of more than 98 percent. Combining the fragmentomic discriminator with aneuploidy protein biomarkers boosted the sensitivity to 51 percent at 99 percent specificity.
According to Lahouel, even though broadening their approach offered a boost, the meat of the assay's performance is coming from the fragmentomics classifier.
Since completing the work described in the current paper, Lahouel said that the City of Hope team has created a new version of A-PLUS using a different sequencing method optimized for detecting a signal coming from differences in fragment lengths between cancers and controls.
Using this more specific test, a team let by Tomasetti plans to open a prospective clinical trial this summer aimed at measuring its effectiveness in detecting cancers in adults aged 65-75.
Lahouel said that the group hopes certain aspects of their initial study will ensure that investigators see similar detection metrics in the prospective trial. "I cannot predict the future, but obviously the hope is to get very stable performance," he said.
One thing that bodes well for the prospective validation is that the researchers were able to filter out about two-thirds of the 350,000 target Alu elements, which they determined were sensitive to batch effects, which Lahouel said can be a problem when moving from a case-control setting into a broader population.
Another factor, he added, is that the group tried to mimic what it would look like to move into a new cohort by splitting their study group into enough sections to have an independent validation after locking their assay specificity cutoff.
"When you move to a prospective study or an independent cohort, one major problem is usually [maintaining] the false positive rate at that same fixed threshold," he said. "You expect to start seeing much more false positives, but we did not observe a big significant drop when we moved to cohort three."
High specificity is crucial in the context of multi-cancer early detection, especially if used for population screening, where the harm of false positives is magnified.
"Obviously we would like to have a sensitivity that is as good as what we observed here in this study. But if we have to give priorities, the first priority is really to keep the high specificity," Lahouel said.