Genomic Characterization of Invasive Lobular Breast Carcinoma

Michael Gatza: Thank you. Today, I present the genomic characterization of invasive lobular breast cancer, on behalf of the TCGA Breast Cancer Analysis Working Group. So we know that breast cancer is the most commonly diagnosed malignancy in women, with about 230,000 cases a year. There are two prominent forms of the disease, invasive ductal carcinoma, which is derived from the ducts, and it accounts for between 50 and 80 percent of all cases, and then invasive lobular carcinoma which arises from the lobule, and it accounts for 10 to 15 percent of the cases; there's somewhere between 25 and 30 thousand cases a year. There's also a smaller subset of patients, maybe 4 to 5 percent, that have a mixed tetralogy, having both ductal and lobular characteristics. So for the goals of this project, in particular for the goals of the project as I will talk about today, is to try to identify, or begin to identify genomic differences between invasive ductal and invasive lobular carcinoma. And then we looked within the invasive lobular carcinoma patients to determine whether this is a single disease, or whether this is a group of diseases characterized by different genomic profiles. So we first started by developing our data freeze, we have 817 samples. Importantly, the pathology of each of these samples was essentially re reviewed by a group of pathologists led by Andy Beck here in the TCGA. Of these 817 samples, 490 are ductal, 127 are lobular, 88 have a mixed topology, and there are 112 patients with other histologies. Within these groups, if we look at the molecular subtypes, these are basal-like or triple negative hormone receptor negative, [unintelligible] luminal A or luminal B, you can see clear differences in the profiles between ductal and lobular, and this to some degree reflects the biology of the disease. Lobular tumors tend to be hormone receptor negative, or estrogen receptor progesterone receptor negative, and have a low mitotic index, tend to be slower growing tumors. In contrast, ductal tumors, as was pointed out yesterday, in the pan can 12 analysis presentation by Josh Stuart, tend to be much more diverse with basal like tumors almost exhibiting a complete novel phenotype. So you can imagine if you do a direct comparison between ductal and lobular, that underlying molecular subtype would clearly confound the analysis, and this is true if we do look at identifying genes that are differentially expressed between ductal and lobular, doing a two class SAM analysis. Here can you see the lobular group of genes that are dysregulated very much mimic what's dysregulated in the luminal A ductile samples. And it's only when you do a direct comparison between luminal a ductal and lobular that you're able to identify genes that are differentially expressed in each sub group, but when compared to normal adjacent tissue, we see no up regulation of those genes. These include a number of genes, including E-cadherin, which is lost in lobular tumors, and this is a characteristic of this disease, it's either lost on the RNA level, or mutated in almost all samples. If we look at the, my data analysis, look at the genes that are dysregulated within these groups. We see that the lobular tumors show up regulation of ATM network signaling, a large number of immune signaling pathways, as well as MAP kinase signaling, whereas the ductal tumors see up-regulation of MYC target pathways and E-cadherin stabilization. So we next asked whether or not we can identify genomic alterations, specifically point mutations within ductal and lobular tumors. One of the concerns we had when looking at DNA whole exome sequencing was that ductal tumors tend to have a much lower cellularity, and that led us to be concerned that we may be under-powering our ability to identify variance in these tumors. So in addition to whole exome sequencing we used data from a program called UNCeqR that was developed by Matt Wilkerson at UNC, that develops -- that combines RNA sequencing analysis with DNA sequencing analysis to identify additional variance, specifically in low purity samples. On top of that, we used a program called ABRA, developed by Lisle Mose, who incidentally has a poster this afternoon on this, which allows us to select small and medium sized variance -- oh, I'm sorry, detect variance in small and medium sized indels. So combining these three approaches, we were able to develop an integrated MAF, comparing the integrated MAF, variant calls and a sample-by-sample comparison against the DNA MAF. You can see that there's an increase in variant calls and pretty uniformly in the integrated MAF. Doing this on a gene-by-gene basis we see a similar profile, where genes that are mutated at a higher frequency, genes like PIC3CA and P53, see an increase in variant calls in the integrated MAF, and even genes that are mutated at a low frequency, here in this inset, you can see also show an increase. In particular, we identified CDH1 or E-cadherin which is important in lobular tumors, where it's mutated in about 7 or 8 percent of the time in, or identified as being mutated using the DNA MAF, but close to 15 percent of the time in the integrated MAF. So using this integrated MAF in combination with justice calls, Giovanni was able to begin to identify, in looking at all ductile versus all lobular tumors, we were able to begin to identify the frequencies of mutations in lobular or ductal samples. So here we see in green CDH1 mutations occurred in about 65 percent of patients, lobular patients, whereas MYC amplification occurs in about 25 percent of ductal tumors. And one of the things we identified when we were doing this direct comparison of all ductal against all lobular was genes like P53 and also ERBB2 were identified and we know that P53 is mutated in almost 90 percent or maybe 100 percent of basal-like tumors, whereas ERBB2 is specifically amplified, which is shown here in red, in HER2-amplified or HER2-positive tumors. So we reduced our analysis to look specifically at the luminal A ductile and luminal A lobular tumors, here again are frequency of mutations in luminal A lobular versus luminal A ductile tumors. We see CDH1 mutations occur more than 65 percent of the time. We still see P53 mutations and MYC amplification, albeit at a lower percentage than what we saw when we looked at all samples. One of the interesting things that we learned is that GATA3 is mutated about 20 percent of ductal tumors, whereas its immediate downstream target FOXA1 is mutated in about 10 percent of the lobular tumors, adding a layer of complexity to that when we look at the protein expression analysis, we see that lobular tumors have a loss of GATA3 protein, so what this suggests to us is that dysregulation of this pathway is very important in these tumors, but whether the tumor is derived from the lobular or from the duct, there may be two different mechanisms deactivating the same pathway. Taking a step back and looking at this on a more global scale, we used paradigm analysis that was introduced yesterday by Chris Benz to look at patterns of associated signaling pathways, and my point here is not for you to read all of this, but more to point out that there are nearly 2,000 features that were dysregulated between the two subtypes of cancer. In particular, in ductal tumors, we see a loss of CDH1 signaling, which is consistent with mutations that occur in about 65 percent of patients, as well as loss of mRNA that occurs in these patients and protein of this gene -- or protein of loss. And again here we see loss of MYC signaling, relative to ductile, which is consistent with amplification of MYC in the ductile samples. We see low XBP1 signaling which is consistent with FOXA1 mutations -- estrogen receptor in addition to FOXA1 results in XBP1 activation. We see increased P53 damage response in the lobular tumors, consistent with increased P53 mutations in ductal tumors, and we see increased immune related signaling, which is consistent in not only the RNA expression that I showed a few slides back, also some methylation data that I don't have the time to get into. So as we were going through these different analyses, looking at ductal versus lobular, one of the things that we observed is that lobular tumors seem to have increased variability, it doesn't seem to be a single disease, so we asked whether or not we could begin to identify molecular subtypes initially based on RNA sequencing. So we used consensus cluster plus to identify three initial groups, you can see those here in a sample by sample comparison. To turn this from a qualitative into a quantitative subgroup classification strategy, we took those samples that had a positive silhouette with those samples we had that were best representatives of each group, and used a centroid -- used a [unintelligible] called ClaNC, developed with dino gene centroid-based classifier, so now we can quantitatively assign, not only these group samples, but other samples to these groups. When we began to look at the characteristics of these groups, we did it first at at two class SAM analysis to identify differentially expressed genes in each group. We found that with an FDR of zero, 988 genes were dysregulated. The majority of these, about 700, were dis-upregulated in class one; these include EGFR and a number of other oncogenes, a number of keratins, kallikreins, and claudins, as well as a number of ATM DNA damage response proteins -- or genes, and also some reactive stromal type proteins or genes. In class two we see 268 genes that are upregulated, including a large number of immune related genes, including LCK and interferon gamma which are master regulators of this process. And then interestingly, in group three, we see almost a decrease or a void of upregulated genes, suggesting that these genes -- this group is characterized not by what's being overexpressed but by what's being lost, in that case, perhaps DNA damage response or tumor suppressors. We could then compare these 988 genes that are up dysregulated in the lobular groups to the ductile or normal adjacent tissues, two things became very apparent. First is at class one has a very similar gene expression profile to the normal adjacent tissue, and then secondly, that ductal tumors seem to have a profile very similar to both class two and three. If we quantitatively assess this [unintelligible] 90 genes centroid based predictor, we actually see that's true, all 94 adjacent normal tissue samples are classified as class one, whereas 60 percent of ductile samples are classified as class two and remaining 30 percent are classified as class two. Interestingly, the same profile was examined when we looked at microRNA. Where class one looked very similar to adjacent normal, in class two and three looked more like the ductal tumors. Coincidentally, also we see that reduction in microRNA reduction in microRNA expression in class three samples. So when we looked at -- I'm sorry. So when we looked at the genes that were dysregulated in class one, not only do we see that there are a number of oncogenes, but we also saw that there are a number of genes that are associated with the reactor stromal or reactor stromal subtype that had been previously identified by Gordon Mills' [spelled phonetically] group. So we did a direct comparison, and what we see is that in class one, almost 90 percent -- over 90 percent of the samples in class one also can be classified as the RBPA reactive subtype. Significantly, showing significant enrichment of this relationship, whereas classes two and three, there's no obvious relationship. Consistent with this, when we looked at RBP data -- this is goofed up a little bit here, what we saw was that Annexin 1, Caveolin 1, Collagen IV and MYH11 were all over expressed in class one, whereas RBM15 was downregulated, consistent with the profiles of these, expected profiles of these genes and their reactive subclass. Beyond the relationship between class one and the reactive subtype, we also show by paradigm analysis that a number of pathways are altered, particularly we see overexpression of PDGF receptor signaling, and loss of FOXA1 signaling, in particular FOXM1 was the only network that was identified that had more than five downstream nodes being dysregulated. Consistent with that, when we look at RPPA data, and this is messed up, we see increased pSRC and pSTAT3 in group one, consisting with PDGF receptor signaling and down regulation of FOXM1 expression, consistent with downregulation of the FOXM1 sub-network. So we then looked at class two and by two class SAM analysis, we identified a number of genes that were dysregulated, that were related to immune related signaling, so we then ran a number of gene expression analysis signatures that could measure B cell, T cell, macrophage, etcetera signaling, and we found that there's a significant upregulation of these signatures in class two, compared to the other two classes. Beyond that we also see an increase proliferation as measured by PAM50 gene expression signature, measure of proliferation in class two. Importantly, what we find is that both of these features can be reproduced in an independent meta brick data set. So what this data suggest is that whereas our direct comparison between ductal and lobular tumors, show that lobular tumors were enriched for this immune related signaling, we've now been able to restrict this immune related signaling specifically to class two tumors. Consistent with this analysis, a paradigm analysis by Kristina Yow [spelled phonetically] showed upregulation of interferon gamma signaling, consistent with immune related signaling, as well as increased FOXM1 expression, or signaling related to high proliferation. So collectively, what I've shown here is that we've developed a unique integrative MAF that utilizes both DNA exome sequencing and mRNA sequencing to identify unique variance in both ductal and lobular tumors. Comparing ductal versus lobular tumors, we see that FOXA1, CDH1 mutations, as well as GATA3 underexpression is associated with lobular tumors, and GATA3 mutations are expressed specifically in ductile patients. We see altered signaling to a number of pathways, and in data that I didn't have time to discuss, we've identified differentially expressed microRNA and methylation patterns in these two sets of tumors. Likewise, looking specifically at the lobular classes, beyond genomic alterations, we were able to show that there's a strong micro-environment condition that seems to be effecting tumor genesis, where the class one is associated with the reactor stromal subtype and class two has a strong immune component. So this is work that was done by the Breast Cancer Analysis Working Group. I've tried to highlight which each person has done along the way, and I apologize to those people whose work I wasn't able to include today, and I'd be happy to take any questions at this time. [applause] Tom Giordano: Tom Giordano -- just a quick -- you know, so I sign out breast cancers, lobular carcinoma is what I consider sort of a sneaky cancer, it infiltrates and so I would think that there is a lot of opportunity for, you know, having to correct for varying proportions of stroma, normal breast tissue, and so how did you guys deal with that in this project? Michael Gatza: We haven't really addressed that in too much depth yet, but when we look at the lobular classes, we see that there's basically similar amounts of tumor components. Class three is a little bit increased in tumor component Tom Giordano: Okay, so you looked at like tumor purity, and -- Michael Gatza: By both -- using absolute, we looked at tumor purity. Tom Giordano: Yeah. Michael Gatza: If we look at the pathology calls, group one seems to have a slightly increased amount of normal adjacent tissue, or normal tissue, maybe five percent higher. Tom Giordano: Okay, so it's -- Michael Gatza: But in terms of actually dealing with it on a practical matter, we've tried to incorporate micro -- mRNA sequencing to boost our calls in variants. In some of the direct comparisons we've done some normal background, like I said the RNA sequencing analysis, we've done background subtraction to get rid of some of the normal adjacent genes that'd be upregulated. Tom Giordano: Okay, and then the other point is just a question, there's a subtype called pleomorphic lobular carcinoma. Michael Gatza: Right. Tom Giordano: And do you guys have any of those, any hope to get some, or? Michael Gatza: You know, I actually just got an email about that from Chuck this morning. I'm not sure, there may be a few in there, but not very many. Tom Giordano: That'd be interesting to see. Okay, great. Thank you. Michael Gatza: Thank you. Male Speaker: Thanks, Mike. Our next presentation is Brady Bernard on lineup, identifying deleterious mutations using protein domain alignment. Brady? [end of transcript]