So, you want to monitor Measurable Residual Disease? What are the challenges?

Part 1 of 2

Background

Measurable Residual Disease (MRD) monitoring – for purposes of this blog – will be the act of looking for somatic variants in a liquid biopsy sample by analyzing circulating cell-free DNA (ccfDNA). This is done to monitor the disappearance of a metastatic solid tumor during treatment and to follow any future reemergence of that cancer. Analyzing ccfDNA assumes that circulating tumor DNA (ctDNA) will be present, and the median ctDNA frequency in patient samples across cancers seems to be around 0.5 to 1 %. Thus, the median variant allele frequencies (VAFs) of the somatic variants will start around this range, and the goal of MRD monitoring is to be able to detect them at much lower VAFs. This can be challenging and if we are going to design an assay for MRD monitoring, then we need to be aware of them and overcome them.

Challenges

The challenging parts of MRD monitoring are:

• Need to follow variants that may start close to the limit of detection (LOD) of a typical ctDNA assay
• Detection of the variants at VAFs that can drop well more than an order of magnitude below the starting VAFs

For example, if VAFs were around 0.5 % at the start, then a 90 % reduction in tumor content could move them to 0.05 %, which could indicate that the chosen treatment is working – but, 10 % of the tumor would still be present. At some point, one may want to detect them at 0.005 % – and, ideally, not at all (but not because assay does not work!). Since the LOD – defined as the ability to detect something 95 % of the time – for a variant in a typical ctDNA assay can be at and above 0.1 % VAF^1,2, attempting to detect variants at VAFs that are well below such an LOD can require a different type of ctDNA assay.

As VAFs drop lower, some somatic variants may not even be present in a given liquid biopsy sample. While some cancer patients have been observed to have higher concentrations of ccfDNA, during MRD monitoring, the assumption is that ccfDNA levels will return to normal levels upon tumor elimination. Thus, let’s assume a yield of about 30 ng from a single patient’s sample.  An MRD assay should work with this amount of cfDNA input. In theory, one could collect more blood, but that has additional challenges and we may not want to analyze more ccfDNA. The 30 ng represents about 8,000 genomic equivalents and a given variant at 0.1 % VAF would be expected to be at about 8 copies. At 0.01 % VAF, it would be expected to be present less than once. Since a variant at 0.01 % VAF would not be expected to be present in a 30 ng sample 95 % of the time, the LOD for a single variant has to be higher than 0.01 %. Thus, an MRD assay needs to monitor many variants in the same sample in order to obtain an LOD for the tumor at levels where most variants would be at around 0.01 % VAF. Due to the need to monitor many variants in the same sample, Next Generation Sequencing (NGS) is a preferred method. When developing a clinical assay,  it is important to keep in mind that the FDA would like MRD assays – at least for hematologic malignancies – to have LODs that are 10-fold below the clinical decision point³. So, if one would change treatment (or start treatment, etc.) at ~0.01 % tumor content, then the LOD of a clinical MRD monitoring assay should be ~0.001 % tumor content. That is very low.

If we are going to monitor variants at very low VAFs where some variants will not be present in the sample, then we may also need to remove some of the safeties that are built into typical ctDNA NGS assays. For example, some ctDNA assays will not report variants below 0.1 % – but, that becomes necessary for MRD monitoring. Additionally, at 0.01 % tumor content, some variants will not be present in a sample and many variants will only be present at 1 or 2 copies. To account for the possibility of errors in the assay related to unexpected observations of variants, ctDNA assays can require a minimum number of observations of a variant, which can be 3 or 4. At this point, we should become familiar with Poisson distributions⁴. Observing a variant at least 4 or 5 times when it is expected to be present less than once is very unlikely. So, we might need to lower the minimum number of observations to 1 or 2. At those levels, it is important to be very sure that any observed variants were actually present in the sample and were not introduced by the assay workflow – especially, given the error rate of NGS sequencing.

Finally, ctDNA assays can be expensive to run – especially, those with large panels. Monitoring implies repeat testing; and, if assays are expensive, then there may be less repeat testing or no testing at all. So, we need to manage the per-assay cost.

Stay tuned next week for part two of this blog where we will discuss how the challenges encountered in MRD screening can be overcome.

References:

1. https://www.accessdata.fda.gov/cdrh_docs/pdf20/P200010C.pdf, last accessed 25th September 2020
2. https://www.accessdata.fda.gov/cdrh_docs/pdf19/P190032C.pdf, last accessed 25th September 2020
3. Hematologic Malignancies: Regulatory Considerations for Use of Minimal Residual Disease in Development of Drug and Biological Products for Treatment Guidance for Industry. https://www.fda.gov/media/134605/download, last accessed 29th June 2020
4. https://en.wikipedia.org/wiki/Poisson_distribution, last accessed 29th June 2020