A variety of gene signatures have been reported that claim to be predictive of metastasis in various tumor types. However, more recent studies illustrate that non-genetic factors, such as epigenetics, the tumor microenvironment, and dysregulated cell signaling may also drive metastasis.
Created by Sara Gosline sgosline @Kaifu @dgilkes We have recently developed some new analysis techniques specifically to determine inter-sample heterogeneity of genomics response in comparison to a covariate (https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4218688/). We have extended these to integrate ChIP-seq marks and RNA-seq expression. It may be that combining your system for determining driver epigenetics, @Kaifu, with our system for inter-sample heterogeneity in RNA that we could be a robust outcome for this system. @dgilkes 1000 cancer cells from the blood or bone marrow are good enough. RNA-Seq and ChIP-Seq both can be conducted using several hundred cells. For RNA-Seq, we can use as low as 100 cells (ultra-low input RNA-Seq). For ChIP-Seq, the smallest cell number has been optimized to 200 (e.g., for H3K4me3), although 500 is a much easier number. @dgilkes we have been developing methods to control the heterogeneity of a given protein's levels from cell to cell without altering the mean of expression. This requires integrating slightly different genetic constructs with an inducible regulator into otherwise identical genomes. We are also trying to tag genes with fluorescent reporters in their native loci. So maybe we can assess heterogeneity of expression for a given gene, and we can also try to control heterogeneity of the same gene's expression. We don't have these yet in MDA-MB-231 cells, but we are working on that. We could then possibly ask if the single-cell MDA-MB-231 clones you studied have different heterogeneity of expression for markers of metastasis, and if increased heterogeneity contributes to metastatic ability.
@Kaifu This sounds very feasible. The only caveat is that we can only recover about 1000 cancer cells from the blood or bone marrow of a mouse. We would probably have to culture these cells ex vivo to increase the number of cells for the RNAseq and Epi profiling. @Kaifu This sounds very feasible. The only caveat is that we can only recover about 1000 cancer cells from the blood or bone marrow of a mouse. We would probably have to culture these cells ex vivo to increase the number of cells for the RNAseq and Epi profiling.
@gabor I like Gabor's ideas above but I am not sure how we could control or modulate the intra-tumor heterogeneity to determine the threshold a priori Hi @dgilkes, before I saw your work, something I was also thinking is to inject the same human tumor cell population to mouse and wait for metastasis to happen. Then we can isolate (1) tumor sample from the injection site, (2) circulating tumor cells in the blood, and (3) tumor samples from several different metastatic sites on the same mouse. By subjecting these samples to RNA-Seq and epigenomic profiling, the data will allow us to define differentially marked (epigenetic) or expressed genes between these groups. If the tumor samples can include some mouse cells or neighboring mouse tissue, the sequencing will generate data for both the human tumor cell and the neighboring mouse cells, and therefore, allow us to analyze epigenetic or expression change of both tumor cell and microenvironmental cells. From the bioinformatics side, data analysis will not be a problem, human and mouse genome are different and the sequencing reads can be mapped back to each genome. From the wetbench side, does this sound feasible? Yes, metastasis is not simply about growth in a secondary site, but a whole process, part of which is getting from the primary to the secondary site. The question "whether cells with different genetic make ups arrive to the same epigenetic state and microenvironmental modulation?" is quite interesting - especially if asked at the single cell level. Fidler did bottlenecking, meaning seeding single cells or a few cells. His results mean that cell populations are heterogeneous. It would be interesting to ask what is the type and level of intra-tumor heterogeneity required for metastasis, what causes the heterogeneity, and can we control the heterogeneity to reduce the chance of metastasis? Hi Kaifu,
Yes we could do the studies you suggested. To @gabor we also have massague's cell lines that have the distinct target organ preference. So it may be interesting to compare cells that underwent "natural selection" in vivo to those that were randomly derived. It kind of argues that either selection or pre-existing intrinsic mechanisms can drive metastasis. Albeit @JAguirre-Ghiso pointed out that the MDA-MB-231 system may not be fully relevant system. I don't think we could develop these types of matched sets from patient biopsies.
We could get guidance from mentors on whether testing cells that have been "selected" for tropism in vivo_ versus those that randomly (random clones from a very highly-passaged cell lines) are more fit to metastasize and whether they have the same epigenetic signature in the absence of genetic drivers?
As a second potential interest, we have developed a color-marking cell line to fate-map hypoxic cells during metastasis (similar to what @Gaodch has developed for EMT). Our initial data suggests that
cells exposed to hypoxia in the primary tumor are the first to seed metastatic sites but that soon after non-hypoxia exposed cells can arrive. We don't know how these cells can colonize first and we plan to do RNAseq to compare hypoxic cells that metastasized versus non-hypoxic cells that metastasized as well as cells at the primary site. If we find differences, my guess is that they must be occurring in a non-genome altering way. The hypoxia signal would be gone as cells traversed through the oxygen rich blood and arrive in more well-oxygenated organs. It would be very exciting to pair your methods with this project.
best!
@dgilkes
It is exciting to see that your student has done the MDA-MB-231 experiment!
Could you be interested in generating RNA-Seq data and ChIP-Seq data for H3K4me3, H3K4me1, H3K27ac, and H3K27me3 using the different clones? We have developed computational epigenetics model to identify tumor suppressor genes and oncogenes, and can easily enhance the model to pinpoint metastasis-specific genes using these epigenomic data. The hypothesis is that "epigenetic mutation", either together with or without genetic mutation, can trigger transcriptional reprograming for metastasis.
I am proposing this because our recent work indicated that cancer genes bear their common epigenetic signature, which does not appear on most other genes. For example, tumor suppressor genes in normal cell tend to bear a broad H3K4me3 domain of 5 to 100kb that covers both promoter and gene body region, whereas housekeeping genes tend to bear only a sharp H3K4me3 of less than 1kb wide in promoter only (Chen, et al, Nature Genetics, 2015. http://www.nature.com/ng/journal/v47/n10/full/ng.3385.html). We demonstrated that broad H3K4me3 domain is associated with enhanced enhancer activity and stable transcription elongation status. The shortening of broad H3K4me3 is associated with down regulation of tumor suppressor in cancer cells. In broad H3K4me3 domains, most other active epigenetic marks also show broad enrichment, whereas most repressive epigenetic marks show broad depletion.
Using our computational epigenetic model, we successfully recovered known cancer genes, and further identified cancer genes that failed to be identified by expression analysis or genetic mutation analysis. The MDA-MB-231 cell and mouse model that you have developed, if can be used to generate the epigenetic data described above, will be a perfect system for application of the new bioinformatics model that we are developing to pinpoint metastasis-specific genes.
Did we set up a time for a teleconference? @gabor
Julio Aguirre-Ghiso (JAguirre-Ghiso)
your point about the relevance of highly-passaged cell lines and their relevance to clinically occurring metastasis is well taken. However, as you see above, a clear hypothesis can be formulated regardless of the cells being used. A metastasis is still metastasis provided that the cells start growing in a location that is distant from the primary site.
** related to your above comments - if growth at a secondary organ was all that defines a metastasis then the strategy of using anti-proliferatove therapies used on primary tumors should work on mets and it rarely does - metastasis are very different depending on the target organ and even with different genetics across patients they still grow in the same organ. The sequencing studies in pancreatic cancer for example show no pattern in genetics between patients' liver metastasis - still all patients had liver metastasis, suggesting a non-genetic driver. Maybe a question would be whether cells with different genetic make ups arrive to the same epigenetic state and microenvironmental modulation? More sequencing studies are revealing this pattern **
Of course, such experiments would be worth repeating with primary cells from patient biopsies.
**Yes because here the genetics would vary.**
Once we see evidence for different metastatic capacities of genetically identical cells, we can formulate secondary hypotheses on what is causing this.
**This in many ways was done already by fidler and massage using a homogeneous parental line and selecting variants that have distinct target organ preference...these were expression profiled but never sequenced to my knowledge. so it could be asked if the "selection" was genetic or epigenetic in nature** Hi @gabor, @lanilonzo, @jungwoo, @dgilkes, @jjbravo, and @scarc,
This is a great discussion! To move it along further I suggest signing up for a teleconference this week. The signup sheet can be found [online](https://docs.google.com/spreadsheets/d/1hY53jRaqoBMnb9HhuE8dN3k4ejX37gG-4C4pattxkuM/edit#gid=1701101959) and as many of you can sign up for as many topics as you'd like. Just put the 'call topic' in column F. The workshop organizing committee is ready to host and facilitate these calls as need be, all you need to to do is call in the number in the column A. You can use this forum or communicate offline to find a time when all of you can call in to discuss further.
The sooner you circle around a project the more time you will have to prepare for a successful project in June. Please feel free to ask me if you have any questions about the process.
-sara @dgilkes, what you describe sounds very interesting. In the meanwhile I also remembered reading this paper on a related topic: https://www.nature.com/articles/ncomms11246
It sounds like in both cases some bottlenecking has taken place, which allowed the establishment of cell populations that were genetically identical, but epigenetically different. Such experiments test the following Hypothesis: Cell populations of recent common clonal descent (i.e., practically isogenic) can have different metastatic ability. If the hypothesis is true then metastasis can occur without any genetic drivers.
@JAguirre-Ghiso: your point about the relevance of highly-passaged cell lines and their relevance to clinically occurring metastasis is well taken. However, as you see above, a clear hypothesis can be formulated regardless of the cells being used. A metastasis is still metastasis provided that the cells start growing in a location that is distant from the primary site. Of course, such experiments would be worth repeating with primary cells from patient biopsies. Moreover, I am not sure how much RPPR and genomic data is available that is relevant to this question. Basically, we would need RPPA and genomic data from genetically identical parental cells that were split, grafted and then allowed to metastasize in twin humans or genetically identical mice. Comparing this data with the metastasized cell data could provide some answers - but are such data available? I favor collecting data that match a question rather than formulating questions that match existing data.
Once we see evidence for different metastatic capacities of genetically identical cells, we can formulate secondary hypotheses on what is causing this. Some possibilities are long-term epigenetic differences or inherent variability. We are developing some tools to test the second possibility. It will be interesting to form a team and think along these lines.
My thoughts on the recent comments. I think that the flask experiment or the clonal analysis of a highly in vitro passaged cancer cell line is not the way to go to address this question and would be phenomenological rather than hypothesis driven, which is what you will need for a grant. Also - it is not clear from the thread what is the exact hypothesis and/or questions. There is wealth of data from cBio databases to analyze transcriptomic, RPPR and genomic data, some from metastasis that could be used to build a model based on human sample data and then walk back to a mouse model that can address that complexity. I think a question has to be structured first - not do an experiment (that sounds pretty random) and see what happens to then build the hypothesis - it may provide a dead end approach. That is my constructive suggestion. @dgilkes Sounds like could be a potential pilot Hi Elana .. nice idea. We do not have genomics data as of yet .. but we have the cells and we can prep mRNAseq libraries in house so getting the data shouldn't be difficult.
@dgilkes Do you have genomics/epigenomics data in those different mice? It would be really interesting to see if we can mine the data to determine what the differences are between the clones both prior to injection and after implantation into the mice. To Gabor's question. A student in my lab generated single cell clones of the MDA-MB-231 cell line and grew them up substantially. He wanted to inject them into immune deficient mice to assess metastatic ability. I totally disagreed thinking that tumor size and metastatic ability would be largely similar. He proved me wrong. Some of the clones didn't form tumors at all. Some grew faster but didn't metastasize a wide range of in vivo phenotypes. This was really just a largely observational study (fun) study but maybe useful for the topic? Although, we don't know if after many cell culture passages can the "metastasizing" phenotype be maintained. Can we prepare two flasks of cells that are genetically identical or maximally similar, but have different metastatic capacities? Has anyone done an experiment along these lines? I am not talking about bottlenecking - lowering cell concentrations close to single cells such as in the classical studies by Isaiah (Josh) Fidler. Rather, both cell populations would be large and genetically similar. Yet have different metastatic potentials. @JAguirre-Ghiso is their evidence that it's epigenetic and/or epigenetic data that you are aware of in the the public domain? Sounds like they may be fruitful candiates. @jungwoo
@jjbravo
Syngeneic systems may be more useful to dissect these micro-environments in bone where you can reconstitute its not only progenitors for osteogenic lineage but also with HSCs and hematopoietic progenitors that may be crucial for controlling DTCs fate. It may also enable using GEMMs and further dissect ho the genetics are controlled by the microenvironment in these modeling systems.
@lanilonzo: I would say that there is no clear genetic alteration( gain, loss, pin mutation, SNP) that is metastasis specific from the sequencing studies out there. This suggests that other mechanisms are dominant in the fate of DTCs and the expansion of metastasis. From our studies there are mRNA signatures that correlate with late relapse in ER+ breast cancer (PloS one 7, e35569, doi:10.1371/journal.pone.0035569 (2012)) there are also other signatures predictive of late relapse (Breast cancer research : BCR 16, 407, doi:10.1186/s13058-014-0407-9 (2014)). Other signatures are associated with indolent DTCs in prostate cancer patients with no evidence of disease after treatment vs. DTCs from patients with active metastatic disease (Oncotarget 5, 9939-9951, doi:10.18632/oncotarget.2480 (2014)). Maybe the helps. We use human bone marrow-derived stromal cells also known as bone marrow mesenchymal stem cells, which can be differentiated into osteogenic and adipogenic lineages. For tumor cells, we are now using prostate (e.g. PC-3) and breast tumor (e.g. MDA-231) cell lines, but we plan to use tumor cells dissociated from PDX mice. For immune cells, we are using human peripheral blood mononuclear cells, although there is a fundamental gap, mismatched HLA via using three different sources human cells. Our model system can be also considered as a pre-metastatic niche model. @jungwoo your system seems fantastic to recapitulate specific BM microenvironments and answer key questions such as how stiffness regulates DTC behavior and activation of key signaling pathways such as RhoGTPases. My lab is investigating how the interaction with the extracellular matrix modulate DTC motility and we use intravital imaging to acquire dynamic information in terms of motility and activation of signaling pathways. We use GTPase FRET biosensors to look at activation of GTPases pathways in real time at single cell resolution. RhoGTPases are major regulators of actin cytoskeleton and they are regulated also in response to mechanical forces, so the changes of GTPase activities can be readout also of how the tumor cells respond to the different mechanical cues in different tumor microenvironment. We don't understand how GTPases are regulated in vivo in response to different stiffness so your technique will help to adquiere real time activity in different type of microenvironments. This information coupled with mathematical modeling to integrate all these different measurement will be instrumental to understand dynamically how cells respond to different tumor microenvironment and what pathways are activated.
@jungwoo which cellular model are you using in your 3D biomaterials? Our 3D biomaterials are based on a synthetic hydrogel. Its mechanical properties can be tuned in a limited range because it is a hydrogel. Current working model is bone marrow which represents a sharp mechanical contrast. The hard bony matrix is mimicked by using glass beads and/or morselized bone matrix. The critical challenge including hard material is the prevention of deep optical imaging. If your microscope technique can overcome this limitation, it would be fantastic. Our tissue engineering platform can be potentially expanded to other tissues leveraging organoid based stem cell engineering such as liver and lung. In that way, varying mechanical properties of specific tissue microenvironment can be recapitulated. The scaffold based biomaterial platform is implantable, which in turn rapidly form vascularized tissue analogs. In doing so, in vitro and in vivo studies can be integrated into a single platform. Subcutaneously implanted biomaterials are optically accessible via window chamber based intravital multiphoton imaging. One hypothesis I propose is that inflammatory response would alter not just local cytokine profile but also mechanical properties of metastatic tumor microenvironments. These phenomena may synergistically facilitate the revitalization of dormant disseminated tumor cells in major tissue sites. If your microscope technique can capture in situ mechanical properties alteration, this hypothesis can be testable even in vivo mouse models. Interesting topic. @jungwoo what type of tissue engineering platforms? Can you modulate their stiffness for example? We have been developing a confocal microscopy technique to characterize stiffness of cells and gels (http://www.nature.com/nmeth/journal/v12/n12/full/nmeth.3616.html). I wonder if it could be a relevant property to measure in this context. @lmheiser Is this something your group has data available for? This is an important and interesting topic that is supported by compelling evidence including donor-derived metastatic tumor development after organ transplantation. Local stromal cells and systemically migrated immune cells are continuously remodeling tissue microenvironments in the course of aging, tissue damage and opportunistic infection, which may be critical in reactivation of dormant disseminated tumor cells. I propose to apply the tissue engineered platforms to simulate metastatic cascade while retaining high analytical power, which may facilitate to distinguish non-genetic drivers in metastasis. This approach could also identify common mediators that support survival and reactivation of dormant disseminated tumor cells across major organs. Mathematical modeling should be essential to integrate multiple pieces of data generated from different aspects in tumor microenvironments and further distill critical factors, but I do not have such extensive set of data. Does anyone in the group have data that is particularly relevant to this project? Seems like it could be a really exciting avenue for data integration techniques.
Drop files to upload
Topic suggestion: Non-genetic drives of metastasis page is loading…