Representative Sequencing (RepSeq) for whole tumor samples

An interview with Nelson Alexander and Samra Turajlic

Researchers have long since known that tumors are heterogeneous. While the heterogeneity may not be defined by sharp, distinct boundaries, regions can be determined by cells in the tumor that have distinguishable morphologies and phenotypes. Not only can heterogeneity confound characterizing the tumor for diagnostic assessment, it can present difficulties in prescribing effective treatment strategies. Recently, a consortium of researchers published results on a proposed representative sequencing, RepSeq, protocol to quantify heterogeneity throughout the volume of tumors. The protocol allows for a more thorough assessment of distinguishable, spatially‑distributed regions throughout tumors.

We contacted Nelson Alexander from Roche Molecular Solutions and Samra Turajlic from the Francis Crick Institute and Royal Marsden NHS Foundation Trust, two of the senior authors of the work, to ask about the protocol and how it might apply to precision medicine.

Q1. Let’s start with a brief description of the protocol for sample acquisition and preparation prior to the sampling methodology (see Figure 1 for reference).

A. Representative Sampling is a statistically powered sampling  methodology  used  in the mining industry and political polling. Analogously, Representative Sequencing (Rep‑Seq) starts with a surgical pathologist who identifies macroscopic residual tumor tissue sample and collects a tumor sample through dissection. A separate, macroscopic normal tissue sample (at least five centimeters away from the tumor) is also dissected and collected to create a normal control for downstream analysis.

The preparation for next‑generation sequencing (NGS) starts with homogenizing residual tumor and normal tissue samples (separately); DNA is then purified from an aliquot of the tumor homogenate and the normal homogenate. These paired DNA samples are then entered into standard NGS workflows, followed by standard computational data analysis. It is important to note here that we are not using paraffin embedded blocks for this process; rather we use homogenized tissue, which is not used in standard practice and typically incinerated.

Q2. What differences did you see when you compared the Rep-Seq method with standard sample biopsies – e.g., what genes were detected that  would influence a diagnosis or therapeutic decision but might have otherwise been missed with standard sampling?

A. Together with our collaborators from the Francis Crick Institute, we initially simulated how a broader sampling scheme might impact the genomic correlates of immune therapy response using existing data sets. Pooling sequencing data taken from multiple biopsies (simulated representative sample) significantly decreases the misclassification rate of tumor mutational burden when compared to only taking a single biopsy. In our first test case of true representative sampling, Rep‑Seq eliminated “clonal illusion,” or the assumption that a mutation you see in a single biopsy is present in the entire tumor, even though the biopsy is a miniscule fraction of the tumor Further, Rep‑Seq delivers more accurate mutant allele frequencies than do single biopsies and is more reproducible than sequencing from multiple biopsies. Rep‑Seq also generally discovers more variants by increasing the search space,  and this may include those mutations that are not present in all the cells of the given tumor but may yield data of critical importance, such as resistance‑conferring mutations that could inform combinatorial therapies.

Q. Can you cite examples of how a therapy decision might have been altered by the improved sampling of the tumor – e.g., a combination therapy that might have been overlooked due to a missed allele?

A. Although there are many potential scenarios, thresholds have been proposed to decide whether patients are likely to have benefit from, say, immune checkpoint blockade or not. Using tumor mutational burden in non‑small cell lung cancer (NSCLC) as an example, we find misclassification of this biomarker could have consequences on patients – either that they will miss out on a therapy that they can benefit from, or that they receive a therapy that is not optimal for them.

Q. Have you or anyone in the group assayed sectioned, homogenized tumor samples (individually), then compared with the pooled homogenate assays? If so, do adjacent sections share the same or similar variant alleles?

A. We have very similar experiments embedded into our clinical feasibility trial with the Royal Marsden NHS Foundation Trust, which is ongoing. We expect that the math will show that, similar to multiple biopsies, each large portion of tumor will contain distinct mutations not seen in all of the separate portions. A homogenate containing samples from each individual homogenate would allow one to achieve a statistically significant level of sensitivity for the vast majority of all the variants.

We know from the studies in the TRACERx Renal Consortium* that sequential tumor slices differ in genetic composition, therefore one might infer that the best approach remains homogenization of as much left‑over tumor tissue as possible for maximum yield of reliable information.

Q3. Does Roche foresee a companion diagnostics application of this to identify therapies for patients, say, linking assay results to therapeutic treatments, especially for potential combined therapies to treat the heterogeneous tumor instances?  If so, could you please provide examples or the current status?

A. Conceptually, Rep‑Seq is a pan‑solid tumor technique rather than fit to a specific tumor type and we look forward to matching this technology to potential companion diagnostic needs in the future. So rather than a specific sequencing panel, Rep‑Seq could serve as a technology that can be implemented for any solid tumor sequencing assay to create panels or signatures for companion drugs.

Q. Have results from the representative sequencing methodology been used to generate specific  molecular  diagnostics   for use in the clinic – that is, a gene panel or a PCR assay?

A. No, this program is in early research and is not being used for any clinical applications at this time.

Q4. Given that the methodology is predicated on identifying heterogeneous variant alleles from a homogenized sample, can you:

Q. Sketch out the informatics and analytics strategy to reconstruct and identify sequences in the homogenized mixture of multiple instances of cancer genomes?

A. The pipeline is bespoke and comprised of freely available software packages and has been previously published (TracerX Renal Consortium*).

Q. Explain the difference with metagenomic methodologies for identifying sequences?

A. There are definitely similarities between Rep‑Seq and metagenomics in that both involve analysis of diversity at the level of the genome and heterogeneity within individual samples. Homogenization of soil in metagenomics is quite similar to our process with solid tumors with the major difference being we are looking for specific mutations within a single patient genome versus detecting sequences from various microbial genomes.

“A homogenate containing samples from each individual homogenate would allow one to achieve a statistically signißcant level of sensitivity for the vast majority of all the variants.”

Q5. Has the RepSeq method been applied to gene expression studies? If so, was the expression pattern likewise more comprehensive than standard sampling techniques?

A. We are currently evaluating Representative Sampling followed by gene expression analysis. Similar to the tumor enrichment prior to sequencing presented in the Rep‑Seq paper, the foundation of our gene expression work is that one should ask expression questions from specific cells rather than from a mixed population of cells.

Q6. What other therapeutic areas could this technology have value? For example, has this approach been used for atherosclerotic samples from different arteries in a patient’s cardiovascular system?

A. For many diseases, heterogeneity isn’t as problematic as it is in solid tumors, so the need for homogenization just isn’t there. Moreover, many diseases do not involve surgical removal of tissue so there’s nothing to homogenize. Even in cancer, many times a patient’s tumor can’t be surgically removed, and the sampling is limited to a single biopsy. While we are interested in expanding Representative Sampling to other surgically removed tissue types, in many instances our process cannot be implemented.

Q7. Final question, what lessons learned (in terms of working through the methodology) and critical points in terms of data and outcomes can you cite?

A. Through our work in partnership with the Francis Crick Institute, especially the implementation of our clinical feasibility study at the Royal Marsden NHS Foundation Trust, I’ve been most surprised and inspired by how straightforward it is to implement Representative Sampling in the clinical space. That’s not to say that it isn’t a significant amount of work, but personally, I’m really excited about applying these learnings to additional labs. In addition, I think this work continues to demonstrate the criticality of academic/ private partnerships when working to address long‑standing challenges in cancer diagnostics.

Thank you for this great discussion, Nelson  61 and Samra

Nelson Alexander received his Ph.D. in Cancer Biology from the University of Arizona, followed by a post-doctoral fellowship at Vanderbilt University. Looking to have a more direct impact on patient care, he joined Roche Diagnostics in 2007 where he has held multiple roles in R&D, as well as Research & Early Development. Nelson became very interested in the interface between tumor heterogeneity and sampling bias. He currently leads a research team within Roche Diagnostics that is pioneering a new sampling process that enables the evaluation and quantification of tumor diversity at the genomic, transcriptomic, and cellular levels from entire solid tumors.

Samra Turajlic completed her undergraduate studies at Oxford University and her clinical training at UCL medical school. She gained a PhD in 2013 from Institute of Cancer Research in the field of melanoma genetics and targeted therapy resistance. In 2014, she was awarded a Cancer Research UK Clinician Scientist Fellowship to study cancer evolution at the Francis Crick Institute. She completed her training in medical oncology in 2015 and was appointed a Consultant Medical Oncologist on the Skin and Urology Units at the Royal Marsden. She became an independent Group Leader at the Francis Crick Institute in 2019, and divides her time between the clinic and her lab. Dr Turajlic is the Chief Investigator of translational studies into melanoma and kidney cancer, and her research goal is to develop an evolutionary understanding of cancer for patient benefit.


* Deterministic Evolutionary Trajectories Influence Primary Tumor Growth: TRACERx Renal, Samra Turajlic and the TRACERx Renal Consortium, Cell, VOLUME 173, ISSUE 3, P595-610.E11, APRIL 19, 2018,