Publication
Article
Oncology Live®
Author(s):
Matthew J. Ellis, MB BChir, PhD, FRCP, provides his expertise on the genomics and molecular profiling of breast cancer.
Matthew J. Ellis, MB BChir,
PhD, FRCP
Matthew J. Ellis, MB BChir, PhD, FRCP, is a renowned clinician scientist with expertise in the genomics and molecular profiling of breast cancer. His accomplishments include helping to develop a Genome Atlas and Therapeutic Road Map for estrogen receptor— positive breast cancer and groundbreaking research into activating HER2 mutations.
OncLive: How has our understanding of breast cancer evolved with the availability of high-throughput genome screening?
Ellis is director of the Lester and Sue Smith Breast Center at Baylor College of Medicine in Houston. He also serves as co-leader for The Cancer Genome Atlas (TCGA) Breast Project and as one of the principal investigators for the Clinical Proteomic Tumor Analysis Consortium, which seeks to translate TCGA discoveries into protein-based biomarkers with clinical utility.Ellis: We have certainly made progress in understanding the genomic structure of breast cancer, with an emphasis on the fact that every breast cancer has a unique genome, that there are somatic changes of every conceivable type, that there’s a huge difference between tumors in the degree of genomic change, and that breast cancers are multiclonal diseases whereby not every tumor cell harbors every mutation detected by sequencing.
What have we learned about the central drivers of breast cancer and how is this informing therapeutic development?
This has generated a lot of analytical complexity and challenges for informaticians. Where we have partially failed is to link these somatic mutations in robust ways to clinical outcomes or to drug efficacy. Of course, this failure is related to the complexity of the breast cancer genomes. So in terms of clinical impact, you’d have to say that sequencing technology has had a very modest effect so far, although there’s one or two developments on the horizon, so that maybe if you were to interview me in a year or two I might be more positive.Obviously, we identified some of the key drivers a long time ago. When I say drivers, generally I mean druggable drivers, because they would be the highest-impact findings. By performing these genomic screens, we were looking for other examples of HER2 amplification—like hits. With the advent of next-generation sequencing, we were looking at point mutations rather than just amplicons, in druggable kinases, for example.
Upon initial examination of the data on recurrently mutated genes, activating point mutations were conspicuously absent at high frequency, except for PIK3CA. But we still don’t have a prudent PIK3CA mutation drug match. So thus far, PIK3CA mutations themselves haven’t generated a successful clinical algorithm. Interestingly, though, there is a successful algorithm on the horizon around HER2 mutations.
These occur in non-HER2 amplified, primarily estrogen receptor—positive breast cancer, where point mutations in HER2 have been shown to be activating and in preclinical data shown to be druggable with neratinib.
At ASCO we reported that if we screened more than 600 patients to find 14 patients with mutations that were known activating in vitro, we could get a reasonable clinical response rate to single-agent neratinib.1 So a new treatment for breast cancer is born.
But, of course, you could say that you screened more than 600 patients and you found only 14 whom you could treat, but the point is that small percentages in breast cancer still involve, relatively speaking, absolute patient numbers that are important. For example, our latest data suggest that about 3% of patients with metastatic breast cancer have detectable HER2 mutations in the circulation. The number of patients living with metastatic breast cancer in the United States is about 200,000. Thus, 3% is about 6000 patients, which is more than enough to conduct randomized trials to establish a standard of care for patients with HER2-mutant disease.
How could we broaden the clinical impact?
But what we don’t have is an effective nationwide screening program to find these patients. We have an execution problem with respect to HER2 mutations, which can also be seen with other low-incidence mutations that could potentially be matched to new targeted therapies. So there are many logistical challenges in this respect.I think part of the problem in breast cancer is that we have been too DNA focused. So while you may see a coding change or structural abnormality at the DNA level, it’s hard to know what that change produces in terms of a signaling aberration that drives a hallmark of cancer. The next step was to do RNA sequencing, so that maybe we understand a little bit more because it’s a little bit closer to the phenotype, but ultimately it’s the proteins and the posttranslational modifications of proteins that produce the signaling.
So this is a good point to talk about our recent Nature paper on proteogenomics.2 What we attempted to do was to take tumors that have been annotated by TCGA with respect to the DNA and RNA state and add mass spectrometry-based proteomics and phosphoproteomics data so we could better connect somatic mutations with signaling and then ultimately to breast cancer phenotypes. Because we only analyzed 77 cases fully, statistical associations with phenotypes wasn’t possible, but the idea is that in the future we will do many more of these types of analyses.
So, what did we find? Well, in a simple classification, there are 3 types of somatic alterations in cancer: amplifications, deletions, and smaller genomic changes involving either point mutations or small insertions and deletions which may or may not alter protein coding sequence.
Proteogenomics produced informative and interesting data in all those 3 classes of genomic abnormality. For example, in the amplicon space, we looked at kinase phosphorylation status by mass spec as a readout of kinase activity. We looked for outlier kinases that were activated according to one definition of threshold autophosphorylation and we compared that with data from gene amplification and RNA overexpression data.
This emphasized the potential importance of PAK1, which is a gene that can be co-amplified with cyclin D1 on the long arm of chromosome 6 and is a therapeutic target related to the RAC/ RHO pathway.
Another finding was that, within the HER2 amplicon, our favorite therapeutic target amplicon, there’s a second kinase called CDK12, a gene that drives transcription of the BRCA1 gene and other members of the Fanconi family. This makes for interesting mechanistic speculations on links between homologous recombination and HER2-positive breast cancer that researchers have been already thinking about.
The proteogenomic data emphasize that amplicons are complicated and you can end up with compound phenotypes. In other words, amplicons deregulate cassettes of genes, they are not monogenic in their effects.
Another interesting finding was in the deletion space. One wonders why there are recurrent chromosomal deletions in breast cancer. Here, we looked specifically at the fact that basal-like breast cancer often exhibits recurrent loss of chromosome 5q (around 80% of cases).
How can we tackle the challenge of heterogeneity?
We used a technique whereby we examined the up or down effects of losing the long arm of chromosome 5 across the proteome and transcriptome and came up with the hypothesis that there were certain negative regulators that were being lost, of which SKP1, a member of a ubiquitin ligase complex that regulates protein degradation of key signaling molecules, was one. The proteomics is beginning to dissect apart chromosomal gains and losses. Using phosphoproteomics, we can also find novel associations between mutations and downstream signaling effects that can then be validated using cellular models. This technology is in its infancy, but if we can do this on core biopsies in model systems before and after treatment, eventually we can get closer to how these aberrant genomes cause signaling effects that lead to drug sensitivity, resistance, metastasis, and abnormal cell growth in cancer.We know that our adversary has considerable tools at its disposal to evade therapy, and the continuation of clonal evolution and the development of resistance mutations is why single drug—mutation matches will not be the ultimate solution. Understanding kinome reprogramming and using drugs that inhibit this, such as BET domain inhibitors, where they are inhibiting epigenetic reprogramming of the kinome, would enhance our ability to more effectively inhibit mutant proteins in breast cancer. Resistance is a big problem, leading one to speculate on whether immunotherapy might be an orthogonal treatment to mutation-directed treatment. The consequences of a mutation can be several-fold—while it drives biology, it may also produce a peptide that could be potentially recognized by a cytotoxic T cell.
What are the most significant unanswered questions?
It’s possible that we could make progress despite heterogeneity by targeting immunogenic peptides in combination with a targeted therapy, so we are attacking the tumor from completely different directions, but both stemming from the somatic mutation structure of the tumor.I think we need to get out of our world of reductionism and understand how the multiple mutations produce integrated signaling effects that drive a phenotype. Scientists tend to take a single mutation out of the context of the cell in which it evolves, introduce it as a single event into an indicator cell line and say, well I’ve understood the effects of that mutation. But of course they haven’t really, because they’ve taken the mutation out of the context of all the other mutations that it co-evolved with.
One of our biggest problems is to understand how complicated somatic mutational structures produce integrated signaling events. That’s going to take a lot of computational biology, and I also think ultimately it’s going to require extensive study of human tumors and comparing the proteogenomics of those tumors to the clinical phenotypes and drug sensitivities.
That’s a logistical challenge because it involves human research and it’s a technical challenge because we have to be able to do proteogenomics on relatively small amounts of material such as a core biopsy. Mass spectrometry is only just getting to the level of sensitivity where that’s a practical proposition. We are just at the beginning, I think, of a new era where we think of proteins and posttranslational modifications as a key readout when we are interpreting genomic information.