NEW YORK – An international team of researchers and Roche scientists have developed a method that they believe could improve clinical tumor molecular profiling by generating accurate tumor mutation burden scores. Roche anticipates commercializing the assay following further large-scale feasibility and validation studies.
The method, called representative sequencing (Rep-Seq) and described last week in Cell Reports, involves homogenizing chunks of leftover tumor tissue not used for pathology analysis in a blender followed by next-generation sequencing (NGS) to capture an unbiased molecular profile of the tumor.
When pathologists extract tumor tissue to identify potential biomarkers for downstream therapy, they deal with issues related to tumor heterogeneity. By analyzing a single sample from a fixed location, they do not always get a representative sample of the cancer, and a tissue biopsy usually only presents less than 0.0005 percent of the tumor.
"Spatial heterogeneity is an existential threat to cancer research, as well as cancer diagnostics and personalized healthcare," Nelson Alexander, cancer biologist and research lead at Roche Sequencing Solutions, and a corresponding author of the publication, explained. "We therefore started talking about how many samples, using current techniques, we would need to capture a rough estimate of the heterogeneity of solid tumors at the cellular level, and found that would be way too many samples."
"Taking a small sample from a solid cancer, which contains millions or billions of cells, is very problematic, as you deal with the issue of reproducibility," said Samra Turajlic, clinical investigator at the Royal Marsden Hospital in London and another corresponding author. "We needed a way to create a more representative sample, capture intratumor diversity, and create a sample that you could repeatedly go back to and always get the same solution."
Rep-Seq requires at least 1 gram of patient leftover tissue following pathology extraction. The process starts with a pathologist macro-dissecting a sample to remove normal tissue that is at least 5 cm away from the tumor. Tumor and normal tissue are separately incubated and then blended with a specific solution inside a homogenizer. Tumor and immune cells are further separated using fluorescence-activated cell sorting, enriching the representative sample for tumor cells.
"Downstream analysis can involve DNA, RNA, or protein extraction, followed by library prep and NGS," Turajlic said. "Then, with sequencing, you don't need redundancy coverage, as all of the sample originates from the tumor."
For their study, Turajlic, Alexander, and their colleagues first analyzed data for 1,667 samples across six tumor types from the National Cancer Institute's Cancer Genome Atlas (TCGA). For each sample, the researchers assessed how much of each tumor was sampled.
The group found that tumor sequencing protocols had a high under-sampling bias. Sampling bias was likely also affected by the levels of heterogeneity and purity of tumor cells in the sample.
Known as "clonal illusion," Alexander explained that when a small sample is taken and clonal variants are identified, the tendency is to assume the clonal variants are in the entire tumor. However, in the study, the researchers found that the sample of the tumor that is tested often leads to inaccurate generalizations about the tumor's mutational profile.
"We looked at data from tumors, taking multiple samples from different areas, and found that you get different results every time, whether they're used for prognostic or predictive purposes," Turajlic said.
To determine the effect of spatial bias in single-biopsy sampling, Turajlic's team pooled extracted DNA from 1,184 multi-region biopsies, taken from 79 primary renal carcinomas (RCCs), to create "cocktail samples." Subjecting the samples to NGS, the team compared mutation calls with previously generated single-biopsy and multi-region biopsy data.
They found that the cocktail samples discovered all of the true-set mutations, compared with single biopsies, which only discovered 73 percent. The team therefore believes that a more representative sample can lead to improved variant detection.
The researchers also found that using a single, more representative sample offers an improved estimate of the true cellular mutational prevalence across the total tumor mass, which is a crucial consideration for prognostic and predictive factors.
They also examined whether cocktail sequencing could determine clonal versus subclonal somatic copy number alterations (CNAs) using the RCC dataset. They found that known RCC clonal events had better coverage than subclonal alterations, suggesting that they could be distinguished using a pooled sequencing approach.
Using tumor masses from leftover material post-surgery, Turajlic and her colleagues then applied Rep-Seq on 11 tumors from breast, lung, colorectal, and RCC cancers.
Choosing a large clear cell RCC tumor (RS1), the team collect 68 fresh-frozen biopsies from the primary tumor and later homogenized the rest using Rep-Seq. The team performed whole-exome sequencing (WES) on seven biopsies and the Rep-Seq sample, identifying 76 unique mutations.
Turajlic's group then compared the Rep-Seq results to the single-biopsy regions and found that the variant allele frequencies (VAFs) from Rep-Seq closely matched the overall tumor VAFs in the 68 biopsies. The group only failed to detect three of the 76 mutations in the Rep-Seq sample.
Because clonal diversity has been linked to cancer prognosis, Turajlic's team then examined Rep-Seq's ability to determine clonal structure. Calculating cancer cell fractions (CCF) for the mutations in the RS1 primary tumor biopsy set, they identified four distinct tumor clones. Repeating the clustering process for the Rep-Seq sample, the group found that Rep-Seq clustered all 41 clonal mutations together into all truncal and subclones.
Aiming to understand Rep-Seq's ability to profile lymph node (LN) residual material, Turajlic's team homogenized two LN samples from a patient with metastatic melanoma. In parallel, the team validated the platform by performing single-biopsy sequencing from a single sample and multi-region biopsies from a further 15 samples from eight distinct metastatic sites.
Comparing the results from LN Rep-Seq to single-sample LN sequencing, the researchers found that the Rep-Seq LN sample calculated a polyclonal tumor structure, with only 63 percent of mutations being polyclonal. While not all tumor subclones were detected using the Rep-Seq LN method, the team noted that the tool was accurate enough to distinguish the sample as a polyclonal tumor.
They also applied the Rep-Seq method in 10 additional tissue samples from lung, breast, CRC, and RCC cancer.
Alexander acknowledged that the small cohort size was a major limitation of the study and may have skewed the results. He emphasized that collecting samples can be quite difficult, as "all of that material is considered surgical waste and discarded or bio-banked."
While Rep-Seq may produce accurate and unbiased estimates of clonal and subclonal mutations, the researchers believe that the tool struggles to detect lower-frequency variants.
"[However], this trade-off may be acceptable in a clinical context, in which lower frequency mutations may be less directly actionable than widely expanded clonal or major subclonal driver events," the authors wrote.
Neil Lindeman, an associate professor of pathology at Harvard Medical School who was not affiliated with the study, argued that the study's conclusion conflicts with the fact that a subclone presented in the paper later went on to be problematic.
"When you mix [these tissues] together, you lose the ability to see the small subpopulations," Lindeman explained. "Therefore, it would be interesting to look at samples from five years ago and see if they would have picked up a small subclone that eventually grew and became problematic for a patient later on."
Another question that Lindeman raised about the study is to what extent the admixture of normal cells potentially interfered with the researchers' ability to find a clear signal from the tumor.
"We can do this with [pathology] slides because we can dissect them under the microscope and see the interface between the cancer and normal cells," Lindeman said. "But what happens when you have tumors, when you're looking grossly at it, and can't make the distinction?"
Also, because certain cancers have different growth patterns and mixed patterns of histology, the researchers need to find out whether it is better to homogenize the molecular changes among the different cancer types or to retain them as separate elements, which would allow the team to compare the different patterns, he said.
Meanwhile, Alexander and colleagues have launched a larger feasibility study to validate the Rep-Seq tool in 12 different cancers that they believe reflect the clinical cancer landscape, including breast, lung, melanoma, colon, kidney gastrointestinal, and gynecological cancers. Launched in 2019, the trial has so far collected and homogenized about 200 of 500 tumor samples, Turajlic noted.
"Beyond seeing if there's enough leftover tumor to create a homogenized sample, we want to see if the sample will allow you to predict both prognostic and therapeutic markers robustly and accurately," Turajlic added. "If the readout from the trial, which we plan to complete in the next [two years], is positive, that will be a strong endorsement for this being a new standard of molecular profiling of solid tumors."
Commercial potential
Highlighting that the Rep-Seq method has been in development since 2015, Alexander noted that his team is interested in applying the technology commercially. The group aims to collaborate with researchers and clinicians to better define how Rep-Seq can be applied in a clinical workflow.
"Our focus is to demonstrate the value and which applications can improve data we can extract from solid tumors, relative to providing clinicians the ability to extract diversity from the solid tumors," Alexander said. "The most important aspect of this work is that Rep-Seq, followed by NGS, eliminates clonal illusion, which allows us to detect the variants at the accurate allele frequency. "
Acknowledging that Rep-Seq is a relatively new process, Alexander said his team will need to perform additional validation and feasibility work before further development of a commercial assay for the clinical space.
According to the paper, the authors currently have a patent pending for using Rep-Seq profiling to support detection of minimal residual disease and other patents on representative sampling of solid tumors. However, Alexander declined to provide any information on the pending patents.
He believes the assay will be crucial for targeted therapy, identifying potential resistance mechanisms, as well as to help clinicians and researchers understand genetic correlations with therapy response for a wide range of solid tumors.
Lindeman, who serves as the director of molecular diagnostics at Brigham and Women's Hospital, believes the major obstacle the researchers may encounter while developing a commercialized assay will revolve around what they define as a sample that is "not necessary for patient care," and what needs to be retained for archived analysis.
He said that following pathology analysis and ancillary studies, unprocessed cancer tissue is stored for a period of time — per state and professional regulations — before being discarded. However, he pointed out that fixed tissue that has been embedded in paraffin blocks is stored for years and potentially used for future clinical care.
In addition, Lindeman noted that the approach may only work for tumors that are surgically removed.
"The general trend is to collect smaller and smaller samples these days, so the method they're describing would really only work with surgical resection, not an initial biopsy sample," he said, since a small needle biopsy is not going to be suitable.