Nanopore technology in liquid biopsy multi-modal analysis

Filippo remembers that his real passion for science came during his internship for his Biotechnology degree. Being a bioinformatician had seemed like “black magic” when he was younger, but Filippo has now been working at the Institute for the Study and Prevention of Cancer (ISPRO), in Florence, for a couple of years as a bioinformatician, after successfully completing his doctoral degree. He now has several publications to his name, mostly focused on exploring the potential of cell-free DNA (cfDNA) from 'liquid biopsies' as a biomarker for cancer.

Applying nanopore sequencing technology to liquid biopsy research

Filippo became experienced in the cfDNA liquid biopsy field during his research as a molecular biologist at the institute where he had worked during his bachelor’s and master’s degrees (IRST, Meldola, Italy). During his PhD, when his lab at ISPRO wanted to start working with nanopore technology, he therefore suggested combining the two. Liquid biopsy analysis then developed as a key focus of their lab.

Filippo explained his progression from lab scientist to bioinformatician over the last few years, although he sees these roles as being tightly linked. For the first paper that his team published on copy number variation (CNV) analysis in cfDNA (Martignano et al. 2021. Molecular Cancer; and read the blog), Filippo performed most of the lab experiments as a PhD student at the time. Now, as a bioinformatician, even if he is generally always at the computer, he still sees molecular biology as a large part of what he does on a day-to-day basis: his colleague, Uday Munagala, who was also a co-author of that publication, now performs a lot of the lab work, but they always brainstorm together and one of Filippo’s strengths is merging the wet-lab side with the bioinformatics, with the bioinformatic data forming the basis of a lot of the lab work.

Filippo was also attracted to the wider potential of nanopore technology: the portability and the low capital cost investment seemed like factors that would appeal to a clinic or small hospital, rather than buying large sequencing equipment. He thinks that one of the reasons that the analysis of cfDNA and CNVs hasn’t been widely adopted by clinicians is because of the centralised nature of the sequencing and analysis. Nanopore technology would allow these sorts of analyses to be more accessible to more people.

Moving from copy number variation analysis to methylation analysis in cell-free DNA with nanopore technology

To Filippo and his team, the CNV project was an experiment itself — it was their first exploration of nanopore technology. CNV analysis is “I don’t want to say it’s the easiest thing…but it’s the most direct thing that you can do with a bunch of raw data”. Methylation calling and fragmentation analysis, which formed the basis of their follow-up work, are “a little more tricky”. They also had a collaboration with Alberto Magi’s group, based in the same building, who had developed an analysis software for CNV calling in nanopore sequence data (NanoGLADIATOR). So, things just came together.

Filippo stated that that project was successful and surprisingly “easier than we expected”! They only needed to make some tweaks to the protocol (read Filippo’s blog to find out more) to work with the smaller cfDNA fragments found in plasma. They could see the fragmentation pattern clearly, and the results were concordant with other techniques; “so far so good!"

Filippo presented this work at the annual Oxford Nanopore event, London Calling, in 2020, after which Benjamin Berman from the Hebrew University in Israel reached out to him (he is one of the senior authors on Filippo’s latest publication (Katsman et al. 2022. Genome Biol.)). Filippo explained how Benjamin is experienced in methylation deconvolution and machine learning, as well as on working with plasma, and he was interested in looking at methylation in nanopore data. “It’s included for free with what you sequence with nanopore!”, which meant that they had no additional experiments to do, they just had to take the sequence data and extract the methylation calls. So, in the end, they could merge the CNV and methylation data, plus some fragmentomics data, from the same dataset — “which is the interesting part, I believe, of this project; the fact that from a single experiment you can get at least three “omic” kinds of data — methylomic, copy numbers, and fragmentomics”. With other sequencing technologies that’s not feasible, he explained, as bisulphite conversion, which is often performed to study methylation using alternative sequencing technologies, tends to degrade the fragments, so you don’t get a sharp signal for fragmentomic analyses. “The cool part” with nanopore technology is how you can obtain all three features together in one experiment (see Figure).

Figure taken from Katsman et al. Genome Biol. 2022.

The real-time aspect of nanopore sequencing is ideal for experiment optimisation

As Filippo’s team are currently doing a lot of protocol optimisation, the real-time aspect has been incredibly helpful in their work. Filippo described how he often analyses nanopore sequence data, such as performing methylation calling or fragmentomic analysis, as the run continues, so that they can determine whether their protocol adjustments have worked, and so if they should continue with the run, or wash the flow cell etc. Filippo explained that while his colleague is doing the experiments in the lab and running the sequencer, Filippo is at home analysing the data in real time, reporting back to his colleague “no stop the run! Let’s do this different clean up!”. This sort of thing is really quite fun; it’s like you are in the field, doing things quickly, to make real-time decisions. The real-time aspect is “mandatory” if you want to be able to adjust things in this way.

Curiosity is my main driver

Regarding his source of motivation, “I think that’s curiosity”, especially in this field of research “where you really don’t know what you can get from an experiment”. When you don’t know if something is even feasible. Filippo explained that that drive is also why they really like to customise things.

Filippo remembers when they got the results of the first nanopore sequencing run on cfDNA samples, when they weren’t really sure if it was even going to work. The fact “that it worked so well, with such an amount of reads, it was like… god, this works actually!” The second highlight was when they performed the methylation analysis in their subsequent work; even though they had always wanted to investigate it (because “it’s free data!”), Filippo was sceptical about what they might actually see, partly because they didn’t have much expertise in methylation analysis at that time. “The cool part” was when they saw the results were concordant with literature — that the markers made sense. They had been concerned that the accuracy wasn’t high enough, or that the coverage wasn’t enough, but when they realised that it was fine, it was like “this really adds a lot to the result that we can get from one single experiment”.

Obtaining sufficient cfDNA and optimising experiments can be challenging

Filippo explained how there are challenges in every part – in the lab, it’s the optimisation of the protocols to get the output that you want from cfDNA samples. From the analysis point of view, it’s trying to understand why things are discordant between different approaches, especially when there is no ground truth to know what is right or wrong. Filippo said that that has been the trickiest aspect for him, but also the most fun.

Filippo provided more detail about the obstacles faced when working with cfDNA samples. For the first publication on CNV analysis the team used research samples obtained from individuals with metastatic lung cancer, and since then they have also been using metastatic colorectal cancer samples. Filippo explained that obtaining cfDNA is easier to do in metastatic cases, as tumour cfDNA is present at higher proportions; when tumour cfDNA is present at less than 10% of the total cfDNA, this can be incredibly difficult to analyse. The most variation in these experiments comes from the prevalence of tumour cfDNA versus healthy cell-derived cfDNA; they are particularly focused on maximising the signal they can obtain from tumour cfDNA – and this is how combining multiple analyses (i.e. CNV, fragmentomics, and methylation analysis) from one sample can really help.

Nanopore is going to be a "big thing" in future

Nanopore technology is going to be a “big thing” in Filippo’s future work as it is central to his research. In future the group plans to take advantage of the ability to obtain so much information from one nanopore sequencing experiment; their recent publication was very much a proof-of-principle project, demonstrating that they could call methylation in cfDNA, distinguish cancer from healthy features, and quantify the cancer-specific cfDNA fraction using methylation information. “One thing that would be very interesting is looking at longer reads because that’s one thing that you cannot do with other techniques at all”; for example, with mitochondrial DNA you cannot get the full molecule in single reads with other technologies, so this could be something to explore.

Calling other modifications, like 5hmC, combined with the more “classical” 5mC cytosine methylation, is also of interest to Filippo. He described how 5hmC is really interesting because it is barely studied. It is present at much lower frequency in the genome compared to 5mC, so, with enough sequencing coverage to study it in cfDNA, it would be interesting to assess its potential as a biomarker, for example.

The potential impact of nanopore technology on the cfDNA field

For the cfDNA field, which Filippo sees as being very important and of great interest to clinicians, nanopore technology can “assess things that cannot be assessed with other techniques”, and therefore he is excited to see what can be discovered with nanopore and where that will lead. In particular, if you can combine everything together from one shallow genome sequencing run — CNVs, methylation, fragmentomics, point mutation signatures, structural variants… — to get the most out of a cfDNA sample, hopefully we can determine just how useful plasma potentially could be in the clinic.

Filippo suggests that such a shallow genome sequencing approach may be less useful in certain contexts, such as in minimal residual disease, where the tumour cfDNA fraction is just too low for this kind of whole-genome profiling, for example. But for tumour characterisation, to understand what the best treatment may be, he believes this approach has great potential, in addition to targeted point mutation analyses, as it “adds a whole new layer” of information that could be used to stratify patients. Filippo pointed out that even though some point mutations, such as those in the EGFR gene, are known to be cancer drivers, with targeted sequencing we can’t determine how the whole milieu of genetic and epigenetic alterations are interacting with such mutations, and how these might impact treatment.

Filippo with the Conticello group (left) and Magi group (right).