Thank you, it’s a real pleasure to be here. Maybe next year there’s going to be an option to have a musical background to the presentation, I like that. These are my disclosures.

To give you a little bit of background on neratinib as well as HER2 mutations, HER2 mutations are typically mutually exclusive with what we would otherwise consider HER2 positive malignancies, these are cancers that overexpress or have amplified wildtype HER2. These mutations are seen at relatively low frequencies across multiple tumour types and you can see here we believe bladder cancer is the cancer type with the highest rate of HER2 mutations but a number of other common cancer types, including breast cancers, biliary cancers, colon cancers, endometrial and lung cancers all have frequencies of these mutations estimated between 1-3%.

Unlike many oncogenes where mutations occur predominantly at one amino acid position, HER2 mutations tend to be more widely distributed across the gene and localised to one of three domains, mainly the extracellular domain, the transmembrane domain and the kinase domain. In aggregate a subset of all these HER2 mutations we see do result in constitutive kinase signalling, activation of growth promoting and survival pathways and oncogenic transformation in a number of preclinical models. You can see that here: these bar graphs show the mean colony count in an amplified but wildtype model versus models where the mutation has been transfected. The other point here is you can see that a prior generation HER2 kinase inhibitor, lapatinib, which is approved in breast cancer, does not effectively abrogate signalling in a number of these mutations.

This is the design of the SUMMIT study which looked across essentially any HER2 mutant as well as HER3 mutant malignancy. Initially there were five disease specific cohorts that were specified, listed here, but the study also had an other solid tumour cohort for any cancer that was not one of these five. The design allowed for breakout of additional disease specific cohorts if a sufficient number of patients with that tumour type enrolled to this other cohort. You can see that happened four times, so there was a breakout of breast, cervical, biliary and lung cancer cohorts. The cohorts that are shaded in grey are now closed to enrolment and the cohorts that are shaded in light blue have ongoing enrolment. Just one note there: the breast cancer cohort is ongoing only for triple negative; the ER positive breast cancer cohort has now moved into a combination strategy which I can touch on at the very end. Each one of these cohorts operated almost as a freestanding phase II study with a Simon two-stage design based on the number of responses seen in stage I and stage II. Otherwise the typical endpoints were used for this study. This shows the enrolment per cohort with lung cancers, breast cancers, bladder cancers and colorectal cancers being among the most common.

This really illustrates the complexity of HER2 mutations seen in our patients. In total there were 35 unique HER2 mutations observed in this cohort of patients and you can see, as anticipated, these tend to cluster in the extracellular domain, this S310 mutation being far and away the most common in the transmembrane domain and in the kinase domain. What I’ve done here is annotated the mutations based on whether they are hotspots. Hotspots are mutations that we see as statistically significantly recurrent above background noise. You can see not every one of the mutations enrolled in this study were hotspots, there was also one fusion patient with a GRB7/ERBB2 fusion.

I’m now going to walk you through the efficacy in the study and, just to orient you to what you’re looking at, this is a waterfall plot and directly below it is a swimmers lane turned sideways. This is a vertical swimmers plot and each bar here corresponds to the same patient vertically. So this patient had a regression and this is their duration of time on study. What we’ve also done here with the asterisks is show you patients who were not evaluable. These were patients who discontinued therapy before their first CT scan. We’ve included small bars here so that you can see the mutations present in these patients. We’ve encoded in the colour of the bar the individual mutations; there’s multidimensionality to these data and it takes a while to take in.

At a high level you can see that a number of the breast cancer patients responded and some quite durably. The other point about breast cancer is that you really see the full compendium of ERBB2 mutations or HER2 mutations in breast cancer from extracellular to main mutations, kinase to main mutations and even some exon 20 insertions. There is no clear pattern of greater sensitivity with one mutation, one ERBB2 mutation versus another, in breast cancer.

Moving on to lung cancer you can immediately see a difference here. Most of these bars are orange and that’s because most lung cancer patients have exon 20 insertion mutations. Interestingly enough, the only patient with lung cancer to have an objective response was a patient with a kinase domain mutation, suggesting that maybe the ability for neratinib to inhibit exon 20 mutations is there as we do see some in breast cancer respond but maybe could also be lineage context dependent. So in lung cancer there may be relatively a greater resistance. Interestingly though, when you look at the progression free survival data these stable disease patients appear to have quite stable disease, a prolonged stable disease, and the median progression free survival in the lung cancer cohort was 5.5 months. These are in patients that tend to have quite heavily pre-treated disease, having had one, two or even greater than three lines of prior chemotherapy.

Moving on to two additional cohorts, bladder and colorectal cancer, you see a general lack of clinical activity in these two cohorts and, as such, the monotherapy is closed for both of these. There were two bladder cancer patients that had responses ongoing for greater than two or three years but again not a tremendous amount of activity here. In bladder cancer, as anticipated, you see primarily extracellular domain mutations. So there are mutations that are more likely to appear in certain cancer types.
Then with biliary and cervical cancer we’ve completed stage 1 accrual to these. You can see responses in both cohorts, some of which are durable and we view these as potentially promising additional disease indications for this drug. Finally, to round everything else out, these remaining cohorts – endometrial, gastro-oesophageal and ovarian – have yet to reach a formal stage 1 analysis and then you can see all other patients here.

To quickly run through this data in the opposite fashion, so now what we’ve done is we’ve grouped the data by mutation type and not by tumour type. What you can see is responses in both extracellular domain mutations, the S310, as well as kinase domain hotspots. Interestingly, again, to call this out we see a breast cancer patient with an S310 mutation having a very robust response whereas in bladder cancer, shown here in this light purple, you don’t see any responses in this S310 mutation, again suggesting both a lineage and mutation context dependence. Showing that nicely again, you see this in the exon 20 insertions where the lung cancer patients with exon 20 insertions had no objective responses but several breast cancer patients with exon 20 insertions did have responses.

Finally, just to draw your attention to this patient which I think illustrates an interesting aspect of this type of research. This is a patient with a non-hotspot mutation, in fact this patient is a breast cancer patient with a complex insertion and substitution event in the kinase domain. This particular genomic mutation has never been observed, to our knowledge, in any other patient looking at datasets as large as 50,000 sequenced patients and yet she responded. This raises the possibility that there are certain patients that will have truly private genetic alterations that may be true one-off events and yet that can be the driver in their cancer and they can go on to have benefit from this treatment. So it shows you the complexity of this type of work.

This is a figure that really integrates the efficacy data, both by tumour lineage and by allele. What you can see here is that the size of the circle shows the number of patients at the intersection of any particular mutation and tumour type. The darkness of colouring corresponds to the median percent change. So this is essentially a median of the waterfall plot bars within that intersection in that circle. This is a way of visualising the data that we’re probably not used to and yet is going to become increasingly common as we develop therapies against a complex set of mutations in a gene and multiple tumour types. Really we have two degrees of freedom if you think about it in that way in terms of how patients may or may not respond. Again, I think this shows the data nicely showing that breast cancer is a disease where we see the most activity with biliary and cervical cancer as being the next most common or the next most efficacious. This is a summary of the efficacy data which you can look at in detail later and shows the data I’ve already called out.

Moving on to the adverse event profile, the diarrhoea was the most common adverse event which is known with neratinib and 22% of patients had grade 3 diarrhoea. There were no grade 4 or grade 5 diarrhoea events; these were 22% all having grade 3. To dive into that in a little bit more detail, and this is really the important take home message, is that only 4 patients on this study of 141 patients discontinued due to diarrhoea. The reason that patients don’t discontinue due to diarrhoea is because the diarrhoea, the median number of episodes of grade 3 diarrhoea per patient that experience it is only one and the median duration of that episode is 2 days. This also tends to be a phenomenon seen within the first two weeks of therapy. So really what you see is a patient comes on and approximately one in five patients will have an episode of grade 3 diarrhoea, typically occurring during cycle 1 and lasting for two days. That patient will subsequently be dose reduced and it’s very unlikely that they will experience additional high grade diarrhoea events. So ultimately although it’s a high grade adverse event appearing in 20% of patients it’s not one that they ultimately discontinue therapy or find intolerable.

So these are our conclusions. The activity of neratinib was influenced both by tumour lineage and mutation type. Breast cancer showed single agent activity and combinations with fulvestrant in ER positive disease are underway. We did not see activity in colorectal and bladder cancers. In biliary and cervical cancer there’s preliminary single agent activity, enrolment is ongoing, and in lung cancer, although the response rate was low, we do see a prolonged stable disease which appears to be inconsistent otherwise with the natural history of this disease untreated with neratinib. We do think we see a greater degree of sensitivity in missense mutations compared to exon 20 insertions which is consistent with some preclinical data.

We will obviously follow up these single agent activity signals. We believe ultimately combinations may be needed to improve activity and durability. Of note, this has really been the story of all HER2 targeted therapy which have always been developed in combination with chemotherapy. As single agents those individual agents which have had dramatic activity have very little activity as single agents, so I think this is the same direction we’re headed here. The safety profile is consistent with previous reports where diarrhoea was not a treatment limiting toxicity. Thank you so much.

Thank you very much, Dr Hyman. Please for questions step to the middle microphone. Let me kick this off. To quote another individual, HER2 has been in the news for years but who know HER2 could be so complicated? I think these issues around what HER2 is as a mutation here is different than HER2 amplification which everybody in the country knows is a big part of breast cancer. These are the mutations, you’ve got 141 people with HER2 or HER3 mutations here, how long did it take to get to and what did you learn about how many people you have to screen to find these rare individuals?
There’s a lot of questions to unpack from there. To answer the screening question it’s very difficult to answer because screening was not actually part of this study. We used local testing to enrol patients into this study so the true denominator is not known but it’s likely to reflect the overall incidence of these mutations. So for every one patient enrolled there were probably fifty patients that were tested that didn’t have the mutation.

What I would say, this study started in the very end of 2013 so it enrolled these patients in about two years. We’ve seen a tremendous uptick in enrolment to this study as this sequencing really comes into the community setting. So it’s increasingly feasible to do this type of research and our ambitions can be likewise greater in terms of the number of patients we think we need to enrol to answer the clinically important questions.

Thank you David. If you could identify yourself for the people on the phone that would be helpful too.

[Audience Member] Elaine Schattner, contributor to Forbes. Thank you, this is fascinating. Do you have any insights as to why this drug is so powerful in breast cancer and not, for instance, in colon cancer where HER2 is also implicated and why you also saw some results in biliary?

This is a really great question is why we see lineage or tumour type dependence in terms of targets. There are two answers to that question, one is that we know that HER2 is a credential target in breast cancer, for example, and this is simply another mechanism of activating HER2 in breast cancer. So it’s not a surprise to me, if I had to bank on seeing greatest activity in one disease it would be the disease where this target, through another means of activation, is the best credential.

This leads to a second question of does this mean that HER2 is not a driver in these other cancers or are we just not drugging it effectively? It’s probably both but I think it’s more the latter. So the story we’ve seen, for example, with BRAF mutations in colon cancer suggest that colon cancers have preprogrammed signalling feedback pathways that mean that you are actually not effectively inhibiting BRAF with vemurafenib alone or BRAF inhibitors alone. I think a very similar story is playing out here so my guess would be that the HER2 mutations are the driver in these cancers but that there are true differences in the way these cancers are programmed to respond to inhibitors and that’s where we really need the physician scientists to take these models into the lab, study them, identify what these feedback mechanisms are and drug them. We’re seeing that happen and there has been a big effort to generate patient derived models from patients enrolled onto the study to create that feedback loop and bring those observations into the clinic.

Any other questions? Yes, please.

[Audience Member] Ted Bosworth from Clinical Oncology News. Will it be increasingly important to subtype these HER2 mutations in terms of treatment? Is that where it’s going, is that where this field is going?

This gets to how we annotate these mutations and ultimately it’s studies like this that really help us do that. So we can annotate them in the lab, although I’ll say that HER2 is difficult to model in the lab because typically what you do with these mutations is you overexpress them. In cell lines when you overexpress HER2 that’s an oncogene in and of itself. But the short answer to your question is studies like this and other resources are creating knowledge bases where we basically annotate the variant level, which mutations are likely to be actionable and which are not likely to be actionable. It’s really important that we do that because these data show that you can’t treat all HER2 mutations the same, you really need to look at them at the variant level and not at the gene level.

Let me just add one question to that. These were only certain variants that were called mutants, right? You already had screened out for SNPs or was any change acceptable in this trial?

We didn’t pre-specify the variants, however interestingly we did go through an effort to retrospectively centrally confirm all the mutations that were locally reported. They were concordant more than 99% of the time. The assay by which we did that central confirmation did match germline sequencing so we know that they are truly somatic events.

The other take home message from me in this is that actually the community testing for this type of alteration is excellent and we’re generally not getting patients with germline SNPs seeping through and coming onto the study.

Thank you. Next question.

[Audience Member] Lynne Peterson with Trends-in-Medicine. Sort of the bottom line is testing for HER2 or HER3 in other cancers just isn’t… either this isn’t the right drug or that’s just not useful. Isn’t it that simple?

I don’t think it’s quite that simple. I’ll tell you why I disagree with that. I think that this is an alteration which will ultimately be druggable in patients and improve their outcomes. Is neratinib monotherapy going to get us there? Probably not in most diseases; I think neratinib in combination with other drugs may. But really we have a chicken and egg problem here. If we don’t test patients, we don’t identify them, we don’t enrol them into studies, we can’t move forward as a field. So what I would say is that this is still research, it’s not practice changing but if we stick at this we will get there, that’s my view.

I’m still struck, though, by the concept that first of all if we think that it’s HER2 action you pick neratinib as opposed to one of Herceptin or something else. We had this idea that we should stop calling things breast cancer and colon cancer and we should talk about it in terms of the mutations that we’re addressing. This sort of takes us backwards, it seems.

I think that’s a very good point. I don’t think we are ready to re-organise the taxonomy of cancer by mutation alone. I will say that the majority of mutations will be exactly like you said, which they will be influenced by the tumour lineage in a very significant way in terms of their drugability. However, there are certain classes of genetic alterations which we may not see this. So MSI high status seems to predict for response to checkpoint inhibitor regardless of tumour type. Similarly we’re seeing some preliminary data with other fusions like Trk which may not be lineage dependent.

The other thing I’ll just say is that I agree that I wouldn’t suggest that every patient go out there and get HER2 dedicated testing, mutational testing, but recognise that they’re not being tested for HER2 in isolation. The tests now are testing for hundreds of actionable or potentially actionable alterations simultaneously so would I suggest that practice be changed and every patient be tested for HER2 mutations alone? No but I think that you can embed that in a highly multiplex test that looks for other credentialed and investigational biomarkers. I’m not ready to give up on that approach personally but I hear your scepticism and I hope to prove you wrong eventually.