Investigators from the non-profit Friends of Cancer Research (FOCR)’s tumor mutational burden (TMB) harmonization project shared new data this week from their second phase in which they continue to investigate how available assays differ in their results when applied to the same samples and experiment with methods to harmonize tests to a common reference standard.
Diana Merino Vega, FOCR’s director of research partnerships, described the findings in a virtual session of the American Association for Cancer Research’s annual meeting, reporting that the team was able to quantify variability in the association between different panel-based TMB assays and a whole-exome TMB gold standard.
They also found that the newly tested tumor samples had similar variation to human tumor-derived cell lines and TCGA data derived from fresh frozen tumor tissue. Finally, the group evaluated three different calibration approaches and examined how these helped align panel TMB values in clinical samples, finding that two methods derived from TCGA data performed better than an approach relying on tumor-derived cell lines.
The FOCR effort involves a number of commercial and academic NGS test labs, reagent and reference materials providers, and academic investigators working together to study the variability in NGS panel biomarker calculation and how it might be minimized.
Excitement in the oncology field that TMB might be able to serve as a biomarker to guide cancer immunotherapy has grown rapidly in recent years. And despite the fact that it hasn’t yet been approved alongside any specific therapies, companies are exploring it extensively in clinical trials, with Merck most recently making the first move to seek US Food and Drug Administration approval for pembrolizumab (Keytruda) in advanced cancer patients with high TMB.
But this enthusiasm is tempered by the challenge of a complex diagnostics landscape of multiple tests that query different sections of the genome, use different mutation filtering and bioinformatic pipelines, and were validated against different references.
“The current gold standard for measuring TMB is using whole-exome sequencing, but targeted gene panels, which assess a reduced region of the exome, have been shown to be able to estimate TMB, thus providing a more cost- and time-effective way to quantify TMB in the clinic,” Merino Vega said in her presentation.
However, since panel-derived TMB scores are estimated and reported differently from lab to lab “it is imperative to ensure that … values are reliable and consistent to help predict patients who would benefit from immune checkpoint inhibitors.”
The FOCR group published a paper on the first phase of its effort this march, in which eleven labs compared their tests’ TMB calls across 32 tumor types using whole-exome sequencing as a reference.
This first foray took an in silico approach, using whole-exome data from the Cancer Genome Atlas. The eleven participating laboratories all took the same TCGA data, downsampling to the area of the genome evaluated by their individual panels, and then estimated TMB values using their own internal or proprietary methodologies.
Investigators found an overall strong correlation between TMB as determined by individual test labs and a gold standard gleaned from the initial whole exome TCGA data, although this was more evident for a subset of eight tumor types that exhibit a wide range of TMB, including lung and bladder cancers — a group the project has deemed “stratum 1.”
According to Merino Vega, the goals of Phase II include the creation of a “universal reference standard to help calibrate or align TMB scores obtained from different NGS panels and better understand the sources of variability across panels.”
To do this, the researchers split their effort into two parts, with phase IIA examining the variability of different TMB assays in matched human tumor-derived cell lines, and Phase IIB bringing this comparison into a study of clinical samples and then testing three different potential assay calibration approaches: one developed from pan-cancer TCGA data investigated in the project’s first phase, another using just TCGA data for so-called “stratum 1” tumor types, and a third developed using the Phase IIA tumor-derived cell lines.
In the AACR session, Merino Vega said the team was able to collect 25 matched normal-tumor FFPE clinical tissue samples from lung, bladder, and gastric tumors for the Phase IIB clinical sample portion of the study.
After DNA extraction at a reference lab, tumor and normal DNA samples were distributed to 16 participating laboratories. Whole-exome TMB was calculated at a reference lab using a consortium-agreed TMB algorithm, while the participating labs calculated their own panel-TMB values using their own sequencing and bioinformatics pipelines.
Merino Vega and her team had received panel TMB results for 15 of the 16 participating labs in time for the AACR meeting, and used statistical methods to create visual plots representing the association between labs’ individual panel-TMB calls and the gold-standard whole-exome TMB for each of the clinical samples.
To quantify variability, the researchers calculated a median across-panel TMB measure for each sample and translated individual panel TMB estimates to fold-changes relative to this median.
“Two laboratories appear to consistently underestimate panel TMB while the rest appear to either somewhat overestimate it or closely estimate [whole-exome] TMB,” she said during the session. TMB calls also diverged more between labs, and from the whole-exome standard for samples in which TMB levels, or mutational counts per megabase were higher.
Interestingly when the group compared this to what they had seen in cell lines and using TCGA data, it appeared that despite the sample source the variability in the association between panel and whole-exome TMB remained similar.
Finally, investigators looked at which data source – TCGA data or cell lines – seemed better able to support a calibrator to pull different labs’ panel-based TMB calls into better alignment with exome sequencing TMB.
Merino Vega said the team generated calibration curves for each laboratory using a weighted regression analysis between the panel TMB generated by each lab and exome TMB generated by the reference lab — for the TCGA data and for cell line samples.
“When we apply the three calibration approaches to 10 lung clinical samples in our dataset, we can see how these calibration tools impact the variability across panel TMB values,” Merino Vega said during her presentation.
To do this the group plotted TMB values for samples with no calibration approach applied. They paired this with a second plot representing the same data but with TCGA calibration, a third using just stratum 1 TCGA data, and a fourth using cell lines.
Both TCGA-based calibrators seemed to pull the diverging panel TMB lines into the whole-exome reference, according to Merino Vega. “The spread in calibrated panel TMB values looks very similar for the two [TCGA] box plots and the spread tightens compared to the uncalibrated TMB values … bringing the calibrated panel TMB value closer to the whole-exome sequencing TMB,” she said.
In contrast, for at least two samples, calibration with the cell lines actually yielded a greater spread than the uncalibrated samples, though this should be taken with caution because the cell-line sample size was much smaller than the TCGA datasets.
“It is important to note,” Merino Vega added, “that the TMB scores of these particular samples are very low and remain low after calibration — in a range where the differences might not have much clinical significance.”
That said, the data seem to suggest, at least thus far, that “calibration methods that used either all TCGA samples or stratum 1 [samples] may better align across panel TMB values.”
According to Merino Vega, ongoing work will focus on testing the three calibration methods in additional tumor tissues with higher TMB values.
Authors concluded that the findings, though incremental, represent important steps toward an eventual alignment of TMB tests, which many view as necessary to support clinical development programs and ensure the usefulness of TMB in clinical decision making.