Our mission is to increase endometriosis awareness, fund landmark research, provide advocacy and support for patients, and educate the public and medical community.
Founders: Padma Lakshmi, Tamer Seckin, MD
×
Donate Now

Linda Griffith - Towards mechanism-based molecular classification of women with endometriosis

Linda Griffith - Towards mechanism-based molecular classification of women with endometriosis

EFA’s Sixth Annual Medical Conference: Ending Endometriosis Starts at the Beginning

Towards mechanism-based molecular classification of women with endometriosis

Linda Griffith, PhD

 

Thank you so much. As you can see from this slide I do not treat patients of any kind but I do see them. I teach physical chemistry at MIT and this is what Dr. As-Sanie would call a stressor. It is a mid-term exam, the most difficult one. They have to use the grand canonical partition function to solve multi equilibria binding problems and they are very stressed. And actually a non-trivial number of girls in my class, it is a very large class of sophomores, do come to me with problems that ultimately end up being diagnosed as endometriosis. They are typically around 18, 19 or 20 years old and have suffered in the environment of MIT and not been taken seriously, not been able to cope and some of them drop out of MIT and were even worried about coming in the first place. I am really, really glad to be at this meeting because it was really the Endometriosis Foundation that stimulated me to think in the context of MIT a school that wants to do all these things to get women into the stem fields, a very practical thing could be to actually start a center at MIT that studies endometriosis.

 

Padma Lakshmi’s interview in Newsweek magazine really made me take the leap out of my very comfortable environment. I had a huge lab, very well-funded in other areas and see if there was something that we could do to bring some of the tools from engineering and basic sciences at MIT into the study of endometriosis. This meeting has been the junior high, high school and college for me learning about it and I am very, very grateful to Tamer and Padma and everyone else involved in the meeting for having me here so many years.

 

What I want to talk to you about today is some work that we have done and things that we want to translate out in the community both ideas, methods and even tools. This has been done mainly with Keith Isaacson and Doug Lauffenburger,  who is leader in systems biology and I will mention work with two clinical collaborators in addition, Mauricio Abräo and Hilde Jørgensen, who is here. She actually is a surgeon who went back to school to get her PhD and I am her thesis co-supervisor. We are just starting a study together.

 

As Dr. As-Sanie this morning pointed out so beautifully there is this very mysterious enduring disconnect between lesion presentation and symptomology in many patients. This we find fascinating. She brought up a lot of beautiful, beautiful things about brain pain processing that are part of this story and what has intrigued us is that in many other chronic diseases, for example, breast cancer there has been a real effort to stratify patients according to underlying molecular mechanisms. If you think ten percent of women have endometriosis roughly it is possible that there could be different disease mechanisms in different sub-groups of patients.

 

If you go to the emergency room with a broken leg you could have had a motorcycle accident, you could have fallen down the stairs, somebody could have hit you with a baseball bat but you have the same need to think about how to treat it. In breast cancer we certainly view the size of the tumor and other surgical appearance features of the tumor as important in diagnosing the patient’s prognosis and therapy but there are three canonical molecular markers related to disease mechanism, estrogen receptor, progesterone receptor and HER2 that are also very important in stratifying patients for their prognosis and therapy. If you are negative for all of these you get a supersize chemo and radiation and your prognosis is poor and if you are positive you can have a much better outcome for exactly the same size tumor.

 

This got us thinking. Are there maybe subsets of patients, and it may be as Rob Taylor and Stacey Missmer brought up that even at different times patients may be experiencing different mechanisms of disease. Therefore, if we are going to treat them with drugs or do clinical trials it would be really helpful to know if there are different stages of disease. A central question for us is, are there ways that we can think about molecular characterizations that go alongside our classical staging?

 

Now, one way to approach this in a very powerful way is to start at the DNA level and what genes did you get from your parents and how might they have been altered soon thereafter, epigenetically or so on to change the flow of information from DNA up to the protein network stakes? In the field of systems biology we think about trying to integrate across this whole path of information flow. Where we at MIT have found there to be some very interesting opportunities to complement the genomics type studies is in the protein and protein activity states because this is your broken leg. Lots of different things could have happened here or here epigenetics to give you the same kind of phenotypic outcome. We are interested in saying are there mechanisms where we can intervene even if maybe there are underlying, and we do not know for sure, but maybe there are underlying genetic mechanisms. 

 

Let us come back to this inflammation invasion cycle and there have been a number of different representations of it. Here is one – again many different processes acting in a network. Inflammatory processes that are there disturbing the peritoneum, cells, invading in often and we would like to understand this. As engineers we like to take a really complicated landscape and try to organize it into categories and put numbers on things the best we can. If we start with inflammation we think about perhaps the immune cell cytokines, things that the immune cells are secreting and their intracellular signaling. Let us start thinking about how these could be linked together in a network in ways we might intervene. Are there patients with maybe a particular type of network that would benefit from an intervention. So it is a question and we have not solved it, it is sort of a work in progress.

 

This has been published. A lot of you have seen this, apologies if you have seen it. We got ongoing studies. This was an initial small study where we thought peritoneal fluid is where a lot of the action is. Certainly it is driving invasion implementation. It is also very easy to collect and not as easy as menstrual blood but it is certainly easy to teach surgeons all over the world how to do this and really, really prepare it well for subsequent studies. You can take this, get the cells and the fluid and do analysis. Here we analyze 50 different cytokines. Not one or two but 50 saying we are agnostic about which immune networks would be important. Let us think about this in a broad sense and try to capture all the different ways cells are talking to each other. It is a modest size study, not small but not huge and we actually even had a treated…again. This was our starter set to learn something. I was new to endometriosis; we wanted to learn something and then go out and apply it to a bigger patient population. Again, it is very, very important to take these methods into larger patient populations.

 

I want to start with an initial finding. We did the kind of analysis that is often done, which is to compare in a forward supervised way patients with different disease stage to controls and look at all these different things that we measured. Forty-seven of the 50 turned out to be things we could compare between patients. Ten of them were different in this univariate so we are saying here IL8, Interleukin-8, is it higher in patients than controls, higher in stage four than stage two. You can see for this cytokine on average it is higher in the late stage compared to controls or stage one. It did not really vary significantly with cycle phase. But I want to call your attention to this there is a lot of variability here in these patients. So these patients have high levels and these patients here in the severe category have low levels. That made us think, maybe instead of looking at an individual cytokine, and maybe instead of doing a forward analysis where we compare disease states, a different way to do it is take all of our subjects, all of the samples and put them in to a kind of mathematics that looks for patterns among the 50 cytokines and does it in a multi-variant way. Meaning, instead of comparing one to one are there groups of cytokines that change together suggesting they are part of a network? And we will do this agnostic to what the disease state we saw surgically was.

 

We did that and here are our 50 cytokines. We used this multi-variant method and found there were a signature set of cytokines that changed. There were 13 of them in about one third of the patients and in none of the controls. Importantly, the group that had the signature had patients from across the spectrum of disease state. There were stage one patients here and likewise there were severe patient who were not in that group suggesting that we are starting to build a way to molecularly stratify these patients. There was a trend toward increased pain in these patients in an unpublished work we have out of the study with Mauricio where he has a very different demographic. This association will strengthen in the study with Hilde Jørgensen in Oslo will be very powerful because she has a very diverse patient population, very interesting and diverse patient population.

 

We have got these cytokines, what do we do with them? You can use some powerful methods from bioinformatics and computational biology to take that pattern of cytokines and reverse engineer what the network will be. We can go in and ask these are changing in a coordinative fashion so there are receptors and things talking to each other and we can back out what those are by using expression profile databases. We did this and perhaps not surprisingly a particular flavor of activated macrophages was in our predicted network. We predicted this simply based on what we knew about cytokines that cells secrete because these have been published by a number of different institutions. We inferred this network and a particular flavor of macrophages appeared to be driving it. So with this hypothesis we then went back to patient samples, took their macrophages out and asked in culture do these macrophages from patients make these cytokines that seem to be associated with this particular signature. And indeed, they did and the controls did not. So you have your different flavors of macrophages and we found that indeed this seemed to be associated with endometriosis. Moreover, we used the same kind of computational biology to back out and infer that a particular intracellular signaling molecule called Jun kinase was particularly important in governing secretion of these. This had not previously been implicated in this kind of macrophage cytokine secretion and interestingly, a company, PregLem, in Europe already had going on when we published this study with the Jun kinase inhibitor and what this suggests is a way that you might think about classifying patients for looking for outcomes in a study like this. It is very early days it is suggestive this might be a way to study.

 

The next kind of question is so we have inflammation there is invasion processes and certainly there is lymphoneuro angiogenesis and all these things going on. How does this inflammatory milieu both reflect and drive inflammation? Again, a patient of Mauricio Abräo with this terrible bowel involvement. My summer vacation one year was to go see this in person.

 

Now we can bring these same kinds of network approaches measuring lots of molecules in a network and lots of kinds of molecules and phenotypes into looking at how the immune cells and endometrial cells that may be in the peritoneal cavity communicate with each other. This is also linked to then invasion processes involving proteases and growth factors and things that the endometrial cells themselves may be making. Similar to what we would find in cancer and actually much of our work here was influenced by analogous studies we were doing in invasive breast cancer.

 

A place to start, because with all of these they are not hypothesis free, so an unsupervised analysis does not mean you are just like going in with no idea of what is going on. The ETF receptor family has four members shown here, has previously been implicated by many different laboratories in endometriosis. It is actually very interesting to see Asgi show that decorin goes down in his studies because decorin is something that inhibits the EGF receptor from sending a signal. These have multiple ligands in the extracellular environment. They impinge on the growth factor receptors and transmit various signals that govern these various properties. Very important, there are different steps along this process and we can break that down in an engineering sense and again, build a model of the key things going on. It turns out that the receptor, which I am showing here signaling here for just one of the family member EGF receptor, that all of the ligands that activated are made as trans-membering precursors that get cleaved from the cell surface by proteases. The activity of these proteases, or at least the observed activity toward the ligands, is regulated by intracellular signaling. You can see we can auto-stimulate the cells. It could be another signal coming in over here to activate these that stimulate this. So, very interesting network, lots of players, again, EGF receptor family is canonical but is not the only one. We set up, again in an engineering sense you want to set up a network model, so we set this up in a way conceptually you have a growth factor stimulation or cytokine. This could be IL-1, this could be TNF coming in impinging on the cell. It stimulates intracellular signaling and then there are proteases that regulate it and then the cell does something – it invades, it secretes something that causes pain, etc. This is a paradigm, a framework, for how do we start to populate this with measurements and mathematical analysis. It is just an engineering approach to network analysis. Again, this has been published.

 

Let me take you through what we did. We did an integrative study that was part computation, part experimental. We started in cell culture and we used the same 12Z cells that Asgi mentioned because we wanted to understand how to link these things together before we went back in patients. This is even more complicated than the cytokines. We started in culture and said okay, we are going to take these 12Z cells and we are going to watch them migrate in collagen gels because at the time that was all we had in three dimensions and we are going to see how fast they do it in response to all these different signals that could be in the peritoneal fluid. We are going to measure the intracellular signaling and we are going to measure the protease activity. It was really hard to measure the protease activity we had to invent a new method to do that and that has been published in the work by Miller et al. It is really fun a lot of people are using it in cancer research now.  

 

I will show you the data, do not pass out with this! This is just to say we measured phosphate protease signaling. Again, across a huge data set. Protease activities and this was with a method we developed ourselves. We measured what was shed from the surface of the cell. So you shed not only the ligands like amphiregulin you shed proteases. You shed receptors from the surface of the cell. Then we measured the invasion properties of these cells in a quantitative way.

 

We then again use computational biology to put together a network saying what is governing the invasion properties and importantly what could we shut off to inhibit that invasion? Again, this just shows a network we used a bullion type of inferential method to get the network and put things together. This is not really important to understand. The really important thing to understand is that if we just blocked – Jun kinase actually turned out to be very important here as well – if we block just one arm with a MEK inhibitor there is compensatory signaling here. You really have to block two parts of the network.

 

Then we went back in our limited set of patient samples. We took some of the same samples for the cytokine study. It is actually very hard to do these kinds of measurements on the activity of the proteases and we started to say can we start to classify patients according to these network properties as well? This is very early work. It was only a few patients but there is a high ADAM10 disease and a low ADAM10 disease and ADAM being one of these proteases that sheds. We are following  this up and it requires developing a more robust way to measure these high throughput in samples but we are in the process of doing that.

 

I am going to finish up with just a little bit on tissue engineering. We really want to go back and say how can we take patient samples and do a much better job of handling them in culture and in particular building 3D models of various processes involved in endometriosis, the endometrium itself, the peritoneum for invasion but do it in a way they are reagents that we can share with other people and protocols that would let you construct a 3D culture and dissolve the cells and plant the cells and so on. Working on this actually stimulated a lot by early discussions with Kevin Ostein but now by many others.

 

I do not have time to go through it all and it has not been published yet so I am only going to show a couple of illustrative figures so this is a completely synthetic extracellular matrix that can stabilize epithelial cultures. It essentially replaces Matrigel. You can see the beautiful polarization of endometrial epithelial cells, sort of gland like structures. It stabilizes the cell secreted basement membrane, so this is the apical surface, the basal surface and it is a completely synthetic gel. We are now building 3D models of endometrium epithelial stromal. This is a focal plane on an epithelial layer and inside in a stromal layer and we are putting monocytes, peripheral monocytes on there and asking how they invade in, well, not how they invade in because this is not actually a typical presentation, but really what do they do when they are all cultured together? What do these monocytes do in the presence of cells from normal and diseased patients? This is just proof of principle showing that if we put these monocytes on they do move down into the matrix and if I can get the movie to work I will just show this one. They are incredibly migratory. These are the white cells here they really do not do anything when they are just by themselves.

 

I will finish up by just saying we are limited in our ability to match the molecular data with the clinical data by the problem that Stacey so eloquently described this morning. We have had tremendous help from her to think about taking the EPHect project and thins she has learned from it along with some things that the surgeons feel are really important for their clinical practice and building an app that will let – and this was driven in large part by Keith Isaacson and now by his fantastic surgical fellow Nyia Noel – but it turns out that there are a lot of really great computer scientists at MIT who know someone with endometriosis, in their family often. They have been helping us, including MIT students. Actually this student has endometriosis and she has been fantastic to help us do this. We are developing this and it brings in a lot of super encryption security things that Frans Kaashock has developed and I can talk to people more about that later if you want.

 

And finally, we are also trying to put an app onto the ENPOWR questionnaire and this happened because Padma, who again was seminal in helping us think about the center and helped up launch it, she visited last summer and the head of the computer science department heard about her and said, “I really want to meet her”. He was really nervous about meeting her because she is so beautiful and so on and it turns out they grew up a few blocks from each other and went to the same school. She was really excited to have her picture taken with him because she said in India everybody in her family wants to go to graduate school in his department. So now in India they will really know she is famous that she got to meet him.

 

I will end there and thank you again all the people who helped. And the last slide which I am putting up for Dr. Gregersen, this is how we should think about together looking at menstrual effluent. It is a way to isolate individual cells. A project that kind of stalled when I got breast cancer and I did not continue the funding for it but we can think together about how to get funding and continue that.

 

Moderator:  Thank you very much Dr. Griffith. Okay, we have some time for some questions, two questions?

 

Linda Griffiths, PhD:  And I will around so we can also…

 

Audience Member:  With regard to the Jun kinase I understand that you guys are using things to try to inhibit it or block it. Have you guys tried to go a step back to find out exactly what is actually causing the growth of the Jun kinase and to try to study it from that aspect as opposed to working – once it is there?

 

Linda Griffiths, PhD:  It is an integrated study and there is certainly evidence in part of our study that we did, you cannot in 15 minutes say everything. There are known activators of Jun kinase in our network. A question we have is what are druggable targets? There are other studies and Asgi has participated in some of them on Jun kinase inhibitors in animal models and there was not a mechanism that was really understood for how it influenced endometriosis. Our contribution is part of an overall bigger picture that the whole community contributes to. There were animal studies done, actually some of them were not published because it was unclear about mechanism. We provided a mechanism for how this is working. It takes a whole community of people looking at facets of it and trying to put a piece of the puzzle out there and then someone else picks it up and puts the puzzle together. We all are very, very curious about what causes this. At the same time in the here and now people need treatment and people need additional treatments or targeted treatments because they are not getting relief. The question is, and we are really just posing a question, are these approaches perhaps translatable to help groups of patients find, because Jun kinase inhibitors are being developed for other chronic inflammatory diseases, is there a way, there is a continued barrier to bringing drugs that are successfully used in other diseases into endometriosis. Is it because it is hard to get patient populations that you think might be able to respond to them? It is a response effect diluted out by having too many patients of too many different kinds in one study. We do not know, we are simply raising the question for consideration.

 

Robert Taylor, MD:  Just a quick question. Beautiful pictures of the 3D culture and the polarization of the epithelial cells. Are those 12Z cells?

 

Linda Griffiths, PhD:  No, no. The ones I showed there were actually Ishikawa cells but we can get the – because when you get the primary cells they are in glandular form but we can keep – I just wanted to – that is like a classic MCF 10-A type assay. But the primary cells, and I could show you more data, they behaved beautifully on top of this matrix with the primary stroma. We have been using less primary cells in the lab just because of the number of people in the lab and trying to really do a lot of…things. At the workshop at SRI we suggested that we try to get these reagents out into the community so people who are growing primary cells can use them and look at how they influence function because we can release the cells using a bacterial enzyme that is not a protease so it does not affect mammalian proteins. It is sort of like having, being able to dissolve Matrigel without a protease.

 

I am happy to talk to anybody who is interested in that. I just showed it today and anyone who is interested Asgi kind of brought this up at his endometrial workshop. We are trying to put together a plan so that next year at SRI we would get them out in the next three or four months after we get papers out and make sure all the kinks worked out. Then let people use them and then we would report out at SRI endometrium workshop next year.

 

Moderator:  We really want to thank our three presenters, three speakers for their very brilliant presentations