Ovarian Cancer

Genetic variability of cells in tumours is one of the biggest challenges in cancer therapy. As cells in a tumour become more heterogeneous, new mutations can result in resistance to treatments. Whole-genome sequencing is often used to find past mutations present in the majority of tumour cells but recent mutations often remain undetectable. Methods to uncover rarer mutations have been difficult, expensive, and prone to errors that can be mistaken for cancer mutations. In a paper published in eLife, Dr Mohammed KaramiNejadRanjbar and Professor Ahmed Ahmed from the MRC WIMM, along with collaborators from the United Kingdom and Germany, have developed a new sequencing technique to overcome these limitations and detect rare mutations from small numbers of cells.

Finding clone-specific mutations

During its lifetime, a cell may acquire genomic mutations that are specific to it, until they are passed on to any daughter cells.  Mutations specific to a particular cell clone (the group of parent and daughter cells) are of interest to researchers because they indicate mutational processes that are happening now or that have occurred in the very recent past in these cells. Identifying and interfering with these active mutational processes could be useful for therapies for conditions such as cancer.

But uncovering clone-specific mutations has not been straightforward: standard next-generation sequencing, a commonly used technique, is very good at finding past but not active mutational processes. This is because mutations common to many clones of cells are initiated in the distant past in an ancient ’grandparent’ cell. These mutations are then propagated through the progeny cells and end up being present in most cells.

So, applying standard sequencing methods to individual clones of cells results in profound overestimations of the number of mutations present in a clone. False-positive mutations cloud the data because of the propagation of initial DNA damage that occurs as an artefact of the extraction and sequencing process.

The method developed by Dr KaramiNejadRanjbar, Professor Ahmed and their colleagues (called DigiPico/MutLX: Digital whole-genome sequencing of picogram quantities of DNA (DigiPico) instead uses a DNA preparation technique that enables the separation of initial DNA material to near single molecules, by distributing the picogram amount of DNA to 384 individual compartments. The researchers then made many copies of DNA from the original DNA molecules in each compartment. They then bar-coded each compartment, before sequencing the DNA from all compartments together.

The advantage of this method is that a mutation that is present in all daughter molecules from the template original molecule is highly likely to be ‘true’. A mutation that is present in some but not all daughter molecules is highly likely to be false since artefacts due to DNA damage propagate randomly during the amplification process.

Elimination of false mutations

This process helps to eliminate a large proportion of false mutations, but not every single one. So, the researchers then implemented an artificial neural network-based algorithm, which they call Mutation Learning (MutLX). MutLX takes as input parameters all the quality measures of the resulting sequencing information (reads) and their distribution pattern across the original compartments. By learning the patterns of true mutations and those of false mutations, MutLX gives a probability for a mutation being a true one and also estimates how uncertain it is about the probability. The mutations that have high probability and low uncertainty are the true positives.

The researchers found that this reduced the number of false positives from tens of thousands to less than ten mutations while keeping more than 70% of the true ones.

The team then applied the method to an individual group of cancer cells from a recurrent tumour of an ovarian cancer patient and found that this clone had an active mutational process that was not detectable by standard sequencing. 

The research team now hope to apply this method to discover active mutational processes in cancer and pre-cancer conditions. For example, this method could be used to study mutational processes in the disease ‘left over’ after chemotherapy, to understand why these cancer cells are resistant to chemotherapy. The techniques could also be used for genomic characterisation of circulating tumour cells and of a small number of cells from needle biopsies.

Detecting rare mutations in tumour cells.

(A) Cancer usually begins with a mutation (dark blue shape in the top cell) in a single tumour cell, passed on to daughter cells that may also gain new mutations (in pink), which then divide and can acquire still new mutations (various colours). Over time this leads to a population of cells that are genetically distinct from each other: the initial mutation is present in all the cells, whereas mutations that occurred later are present in a smaller number of cells (bottom row).

(B) Researchers extracted genetic material from a single cell, diluted it down to single DNA molecules, and plated these onto individual wells (top panel), then amplified each well is amplified to create individual libraries, which are then combined and sequenced (bottom panel). The MutLX algorithm then determines which of the genetic variants (in dark red) are mutations that appear later during tumour evolution (in dark blue) and which are artefacts generated by the amplification process. 

Source of figure and caption – eLife article by Nadine Bley.

This study was funded by Ovarian Cancer Action and The Oxford Biomedical Research Centre, National Institute of Health Research.