NEW YORK – Based on growing evidence from their ongoing TraceRx study in non-small cell lung cancer, investigators believe that the cancer research field is ready for an expansion of adjuvant therapy trials informed by non-invasive, molecular measures of minimal residual disease (MRD).
In an online presentation for the American Association for Cancer Research annual meeting this week, researchers involved in the effort presented data from an expanded investigation of circulating tumor DNA (ctDNA) patterns in pre- and post-surgery samples from patients in the study. They reported, among other things, that their strategy for creating personalized ctDNA MRD assays can reliably detect low-frequency variants, and that MRD surveillance can lead to detection of relapse, often significantly ahead of standard-of-care imaging.
The use of blood-based genomic tests to establish whether early-stage cancer patients have signs of residual cancer in the post-surgery adjuvant setting has become an area of increasing interest in recent years. And although the TraceRx trial has a much broader scope than just ctDNA MRD detection, the study has previously yielded important early evidence for such approaches.
Commercial firm Natera, for example, established its own Signatera assay approach, in part, within TraceRx, launching its method for early research use after the publication of results from the trial in 2017 showing that using ctDNA to gauge MRD could predict relapse in post-operative lung cancer patients an average of 70 days sooner than imaging.
Since that initial work, study researchers have gone on to expand upon their findings by exploring how altering their MRD detection technique beyond Natera's 16-gene strategy affects detection, confirming earlier data that linked ctDNA patterns with different histological tumor subtypes, and more solidly establishing the evidence that ctDNA MRD can improve the ability to conduct adjuvant therapy trials.
In the AACR presentation this week, Chris Abbosh, a University College London oncologist and researcher, said that the continued focus on MRD in TraceRx reflects how much stands to be gained if care for early-stage lung cancer patients can be better personalized.
"If we take a patient population with high-risk early-stage disease who've undergone potentially curative resection of their cancer and we offer these patients adjuvant chemotherapy or chemoradiation therapy, then we can improve five-year survival outcomes," he said. In contrast, if the same treatments are used in the metastatic testing, there is much less benefit.
This means that "if we want to improve outcomes further in non-small cell lung cancer, we really need to focus on innovation in the early-stage space," Abbosh explained.
Such innovation has proved especially challenging, however, because trials have not been able to contend with heterogeneous patient populations "containing some patients cured by surgery and some patients who have residual metastatic disease," which, until recently, researchers struggled to differentiate.
Without such differentiation, the adjuvant trials necessary to establish new standard-of-care treatment paradigms require "hundreds, if not thousands of patients [who must be followed] for many, many years," Abbosh said.
The promise of MRD "is that you can differentiate between these populations and you can conduct trials in smaller, more relevant populations when you are only escalating treatments in patients destined to relapse from their disease."
In his AACR presentation, Abbosh first described experimental validation data the team collected as it explored the sensitivity and specificity boundaries of the personalized ctDNA assay approach with different numbers of target mutations. He also described an MRD calling algorithm the group created to go alongside this chemistry.
Briefly, the personalized assay creation involves whole-exome sequencing of ideally multiple regions of a patient's excised lung tumor. From this data, a panel of mutations is chosen to track in blood, and investigators use anchored-multiplex PCR, a technology commercialized by ArcherDx, to create a PCR panel against these positions of interest.
According to Abbosh, the MRD caller the team has added to this process calculates error rates, based on background sequencing reads outside of the region of actual interest to the assay, to inform the MRD call.
"If we see a mutation, we categorize that mutation based on its flanking trinucleotide context and we come up with trinucleotide error rates for that library," he said. "We then utilize that error rate to ask a simple question: is the MRD signal at our positions of interest higher than what would be expected based on the background error calculated in the library? If the answer is yes, then we make and MRD call."
To explore variables affecting sensitivity and specificity of the approach, the researchers spiked DNA into normal background samples. Abbosh said that, as expected, detection sensitivity does scale with DNA input. For example, at the highest DNA inputs the assays can detect variant fractions "down to 0.003 percent, which is the lower limit of our serial dilution," he said.
The team also looked at how sensitivity and specificity scaled with increasing number of variants tracked, using 50-variant, 100-variant, and 200-variant versions across 400 TraceRx libraries.
"Sensitivity at lower ctDNA fractions does scale with number of variants tracked," Abbosh said. However, this is offset by specificity, he said, noting that specificity when tracking 200 variants was 99.4 percent versus 99.8 percent when tracking 50 variants.
On the heels of these technical experiments, the team set out to analyze some real-world samples from TraceRx patients who did or did not relapse after tumor removal surgery.
Notably, Abbosh said, the group saw one pattern that they reported in their 2017 Nature paper reiterated in these new analyses, which was that pre-operative ctDNA detection in non-small cell lung cancer varies by histological subtype.
"Pre-operative detection when the tumor was in situ occurred in only 19 percent of lung adenocarcinomas in our original data versus 97 percent of squamous cell carcinomas," he said. "And in our current data set, this finding is validated in that we see ctDNA detection pre-operatively in only 49 percent of adenocarcinomas versus 100 percent of squamous cell carcinomas."
Among 45 patients who did suffer relapse of their primary NSCLC, 37 had detectable ctDNA at or before clinical relapse. In these 37 patients the median ctDNA lead time (the time from ctDNA detection to clinical relapse) was 151 days, ranging from 0 days up to 984 days.
Interestingly, in all 10 patients who developed second primary cancers during follow-up no ctDNA was detected, highlighting the specificity of the MRD assay toward the primary tumor.
In 23 patients who remained relapse-free during the study follow-up, ctDNA was detected in 2 of 271 time-points analyzed leading to a specificity of 99.3 percent, Abbosh said.
Closing the presentation, Abbosh also shared some emerging data the group has collected tracking larger ctDNA panels in longitudinal patient samples — up to 483 variants per patient — to try to follow dynamic changes in clonal composition and copy number status. In one case example, researchers were able to track a macroevolutionary copy number event occurring in the patient's tumor.
"At 767 days post-surgery all of the 132 clonal variants we were tracking were within a single distribution, whereas at day 877 we saw the emergence of an outlier cluster of variants that had not increased … to the same extent as other clonal variants," he said.
Going back to primary tumor tissue sequencing data, the researchers could then determine that this outlier cluster of variants had undergone a copy number deletion event. "What's particularly interesting in this case," Abbosh said, "is that the variants involved in the deletion event are enriched for variants that are predicted to encode for neoantigens. So potentially here we're seeing neoantigen editing due to copy number instability at relapse."
In a second case example, the investigators were able to track ctDNA to see "differential dynamics of subclones in response to different therapies given in the post-relapse setting, including [immunotherapy], which leads to rapid expansion of [a specific] clone."
Taking the progress made in full, Abbosh said that he views the field as "getting to the point at which we can start to leverage these biomarkers in clinical trials to direct or escalate adjuvant treatment."
Several studies across a number of cancer types are now seeking to do this, and commercial players have also begun to try to establish MRD assays for clinical use.
In addition to Natera, which launched a commercial service to create patient-specific assays several years ago, liquid biopsy firm Inivata has also recently taken a step into the space, marketing a new service called "RaDaR" (Residual Disease and Recurrence) — which uses its InVision platform to create personalized ctDNA panels of up to 48 mutations gleaned from tissue sequencing data.
Abbosh argued that although the promise of these approaches is clear, to truly prove clinical utility the research community will need to follow through to test and establish adjuvant treatment paradigms that ctDNA MRD tests could then guide. Only then will it be clear whether the whole sum — MRD detection and personalized treatment based on that detection — actually improve outcomes.
"There's lots of very exciting work going on and many different groups across different cancer types working within that focus, in colorectal cancer, for example, bladder cancer, breast cancer, lung cancer. So I do think we're going to see the field … move forward and start to get more clinical data, which will tell us about the utility of these technologies in helping patients and improving outcomes," he said.