AACR 102nd Annual Meeting, 2—6 April 2011, Orlando, Florida

Full genome sequencing of 50 breast cancer patients

Professor Matthew Ellis – Washington University, St Louis, USA

The genetics of breast cancer now, here at the American Association for Cancer Research. Dr Matthew Ellis, you have been looking at this and it’s complex isn’t it?

Very complex. In precise terms what we are looking at is what we call the somatic genomes, so these are the changes that occur in the breast cell cancer cells as they develop in the breast; we are not talking about the inherited genome, we are talking about all the corruptions in the DNA that occur as the tumour forms.

So these are changes that you don’t want?

Well yes. Of course there are changes like this occurring in all our proliferating tissues in the body all the time but what we are tracking here is the ones that accumulate in the final stage in tumorigenesis which are in the invasive cells.

And you and your colleagues have made one of the biggest investigations of the cancer genome yet, and it came out of a clinical study. What was the study and what in fact did you do?

Well there is a lot of cancer genome sequencing going on right now but we really felt we wanted to focus this technology on a precise clinical question, which is why it’s in a trial. The clinical question is why do some patients when treated with oestrogen lowering agents for hormone receptor positive breast cancer have a good prognosis because they respond and then, importantly, what’s the genomics behind the non-responders? Because, remember, the majority of women actually die of hormone receptor positive endocrine therapy resistant disease, so this is a huge problem. 
And so we did a clinical trial where we started an oestrogen lowering agent before surgery to shrink tumours, to promote better surgical outcomes, but also we got immediate read-outs on whether the tumours were responsive or resistant to the oestrogen lower agent, the aromatase inhibitor. So about half the patients of the fifty had resistant disease, as marked by ongoing proliferation of the tumour despite the drug, and half were responsive, that is to say the proliferation had been suppressed by the drug, and therefore that was in a better category of patient. And then what we did was what’s called full genome sequencing, so again for the terminology many people doing partial genome sequencing; that is perhaps just the protein coding genes. In this experiment we looked at the full genome, which is what makes it a very big experiment involving about 10 trillion base pairs of sequence, because we actually sequence the patient’s germ line and their matched tumour about 30-40 fold depth which means you are generating about 100 billion base pairs per experiment.

And as anti-oestrogen therapy has, in fact, been responsible for massive improvements in outcomes and if you can get at those 50% who are not responding to your AIs then that is a big potential gain.

Absolutely, critical. It is not just a question of predicting who will do poorly to give them today’s drugs, it is predicting who will be too poorly and why so you can give them tomorrow’s drugs. And what we found is probably there are as many therapeutic opportunities there we can address with drugs that are already approved.

Now I know the situation is complex, there are a lot of genes that you have found are controlling this, but some big ones have emerged. Could you give me the list please?

Well I like to talk about the top three, because they interact in interesting ways. So the most frequently mutated gene, and so remember this is an unbiased experiment, will gain functions in the catalytic sub-unit of PI3 kinase; about 50%. So, if you like, as B-Raf is to melanoma, PIK3CA is to hormone receptor positive breast cancer; that is fully half the cases. So then the next most frequent hit was p53; no big surprise there because that is currently mutated, although at a lower frequency than in other tumours, about 20%. And then the third most frequently hit gene was a gene that is called MAP3-kinase 1. It’s very interesting because that’s a kinase that actually is subject to loss of function mutations. The tumour actually goes to the trouble of actually knocking this gene out through frameshift mutations mostly, so that’s a little bit of a surprise. The other surprise to me, and at first glance a disappointment, is that those top three genes only really account for half the cases. So half the cases don’t have the most frequently recurrently mutant genes; so what that really tells you is that the rest of breast cancer is comprised of recurrent mutations and tumour unique mutations, and so that’s what makes it complex.

In the future, of course, you may be able to target those smaller mutations present in smaller numbers, but let’s have a look at the three big ones. Could they perhaps represent therapeutic targets?

There are two things about extracting medical value from sequencing; one, what does it inform you as to the likely clinical behaviour of the tumour with today’s drugs? And then the second is do they give you therapeutic clues? So the first piece is, we found something rather interesting. I was intrigued by the fact the MAP3-kinase 1 mutation had not being reported before which made me think that perhaps it had been missed from other experiments because this could have been associated with better prognosis, the kinds of tumours that are not present in cell lines and that kind of thing. And that is what we found, it’s a sort of luminal A or better prognosis type of mutation that when present, particularly with PI3 kinase and in the absence of p53, is marking for a very indolent and endocrine responsive form of cancer.

And 10% of patients have it?

Yes, so you would say that actually the targeted drug for them is the endocrine therapy, it works extremely well. And then in terms of therapeutic opportunity, obviously the PI3 kinase hit is a big therapeutic opportunity, and in the tumours without PI3 kinase catalytic sub-unit mutations we find other rarer mutations in the pathway, either in receptor tyrosine kinases upstream or AKT and even further downstream, which will probably produce a PI3 kinase pathway activation event. So there are common ways to make the pathway active and then less common, but they all become something that you can sort of drug up perhaps as a class.

So there are insights into how to plan your strategy, your therapeutic strategy, for patients from these, also some possible therapeutic targets there. What about all the smaller genes, the less frequent genes; what can you say about those at this stage?

Well one thing we did that I thought was useful is we crossed the mutations with what’s called the druggable genome. So the druggable genomes are essentially genes that there are already drugs available for. And what we found was there are a lot of mutations in genes that are receptors for benzodiazepines, anti-psychotics, opiates. Now it’s possible that those are all carrier mutations and not important, they are not drivers, they just accumulate because, for example, g protein linked receptors are very big and they tend to accumulate somatic mutations that are not important. But we did find mutations in receptor tyrosine kinases for which there are existing drugs for. So, for example, we found mutations in genes that could be targeted with imatinib which is actually a leukaemia drug, but maybe there is sub-section of breast cancer patients who could benefit from these. Now don’t go starting treating your breast cancer patients with imatinib because these mutations in the receptor tyrosine kinases could still be passive gene mutations like the ones I mentioned for the g protein linked receptors.

But more work to be done.

So there’s a lot of functional annotation that has to happen, but I think looking at it you can see that there’s only one way forward, period, on how to do treatment in the future and that’s genome forward. You have to have the sequence of each individual patient’s tumour to design the individual trials; you can’t do it in retrospect.

I am beginning to get the picture. What you are saying is that if you don’t have this it’s like doing it with a blindfold on.

Well that’s what randomised trials developing, sort of, alphabet soup type of chemotherapy regimens are exactly that. So a modern chemotherapy trial to improve, say, survival or relapse free survival by 5%, where the background cure rate is 70-80%, requires thousands and thousands of patients because you are treating the tumours blind to their genomic structure. Whereas if you knew the genomic structure up front, you would do something completely different wouldn’t you? You would tailor a therapy to each particular tumour per the therapy hypotheses that shout at you from the genome.

You certainly would. Could I get you, then, to distil for us briefly what doctors should take home from this; what they should take note of in planning their therapy and also looking to the future?

Well I think obviously we have to extract the medical value from all this sequencing, but I think there’s a seismic shift in our approach to cancer because now that we can do genome sequencing in an unbiased way, perhaps starting with just the coding region genes, but as we extract medical information from the non coding regions of the genome this will be the entry level diagnostic quite quickly I would suspect. And so you are going to see a lot of new trials where the sequencing is done up front and that will probably translate into clinical care quite fast I suspect.  So your pathology report in, you know, five years’ time is going to look very different to what it looks like today.

Well Matthew, thank you very much for joining us, very interesting work, for joining us here on ecancer.tv.

It was great fun, thank you.