FDA/NIST Sponsored Workshop
In Vitro Analyses of Cell/Scaffold Products
December 7, 2007
National Transportation Safety Board
Conference Center
429 L'Enfant Plaza SW
Washington, D.C.
These transcripts have not been edited or corrected, but appears as received from the commercial transcribing service. Accordingly, the Food and Drug Administration makes no representation to its accuracy.
Printable Version (PDF - 435 KB)
Note: Documents in PDF format require the Adobe Acrobat Reader®. If you experience problems with PDF documents, please download the latest version of the Reader®.
Welcome
Anne Plant, PhD (Moderator)
Session 2
Fred Heineken, PhDKen Giuliano, PhD
Lani Wu, PhD
Dan Martin, PhD
Andres Garcia, PhD
Michael Sacks, PhD
John Elliott, PhD
Robert Nerem, PhD
P R O C E E D I N G S
DR. PLANT: Good morning. I think it's time for us to get started. I'd like to introduce myself. I'm Anne Plant from NIST, National Institute of Standards and Technology. And, of course, we're co‑sponsoring this workshop with the FDA.
I hope that you‑all are enjoying this. This is I think really productive workshop already. And today will be something a little different, where we're going to focus on some of those challenges that were identified yesterday and how to approach some of the measurement challenges that occur in the issues of tissue engineering, both in R&D and also in what might be useful for regulatory purposes.
I'm leader of a group called Cell and Tissue Measurements at NIST in the Chemical Science and Technology Laboratory. And NIST has a number of measurement science functions that impinge on tissue engineering, both in our laboratory and in the materials science laboratory, particularly in the polymers division. Some of those people from NIST are around today as well.
So hopefully, you'll get to contact and communicate with folks at NIST if you're interested in doing so. And if you'd like more information, please let me know. Give me your card, and I can send you more.
Without any further adieu, we'd like to start our program today. Before we get into some of the more analytical aspects, we're going to warm up this session with a presentation from Dr. Fred Heineken of the NSF on some of the work that has gone on recently among the different agencies within the federal government who have technologies and interests in tissue engineering.
Fred is going to talk about the strategic plan that was put together by the federal agencies for tissue engineering to help advance tissue engineering, particularly through funding and research within the federal agencies.
Dr. Heineken received his BS degree in chemical engineering from Northwestern University and has a PhD in chemical engineering from the University of Minnesota.
He worked for Monsanto for five years where he did enzyme production research, and then joined the University of Colorado, did research on respiratory physiology and taught chemical engineering, and then joined another industrial laboratory, COBE Laboratories, where he worked on human dialysis research and product development.
So he's had a lot of very diverse experience, both in industry and in academic research. And then after nine years at COBE, he joined the National Science Foundation as a program director funding biotechnology and biochemical engineering in the engineering director of NSF.
He's recently received the NSF award for emeritus service. And in addition, I'd like to say that Fred has really been a key player in the development of tissue engineering and funding for tissue engineering, as well as sort of helping the whole field find itself and identify what tissue engineering is, and has been really a facilitator in workshops and in bringing the community together and helping to create the field of tissue engineering.
So it's with great pleasure that we have Fred Heineken, who's the chair of the Multi‑Agency Tissue Engineering Science Inter‑Government Working Group, to give us a summary of the strategic plan for tissue engineering.
DR. HEINEKEN: Thank you, Anne.
Good morning. It's nice to see you‑all here so early in the morning. And thank you to the organizers for inviting me to talk about the Multi‑Agency Tissue Engineering Science.
As you see up here on the slide, we call it MATES, the interagency working group, which put together a strategic plan advancing the tissue science and engineering. There are copies on the table in the back. We had a few copies available yesterday. If you haven't received or gotten a copy yet, there's some more back in the entrance area.
So it's an interagency strategic plan. We have a number of different agencies that have participated in putting this plan together.
So tissue engineering as a term as far as we can determine was first coined in 1985 in a proposal to NSF on an engineering research center proposal. And since then we've had various conferences on tissue engineering, the first being at Granlibakken on Lake Tahoe in 1988, where the first definition that we had for tissue engineering was generated.
So as we've defined it through that workshop at Lake Tahoe, the application of principles and methods of engineering, and the life sciences toward fundamental understanding of structure/function relationships in normal and pathological mammalian tissues, and the development of biological substitutes to restore, maintain and improve tissue functions.
So that came out of the Granlibakken conference in 1988. There's a book published authored by Dick Skalak and Fred Fox that was a summary of the proceedings of that conference.
Since that conference, we've had various calls for proposals and awards in tissue engineering. And interagency contacts have sprung up through the biotechnology research subcommittee activities, the subcommittee activity of the National Science and Technology Council. In the strategic plan, we discussed regenerative medicine. It's an overlapping field with tissue engineering, tissue engineering science as well. And let me just read what we've got in the plan here.
So we look at regenerative medicines as self healing through endogenous recruitment of exogenous delivery of appropriate cells, biomolecules and supporting structures. And it's different from other disciplines by its focus on cures rather than treatments. And there's a HHS publication that's referenced up here, and you can see the URL for the website on that.
In the plan, we expand tissue engineering to tissue of science and engineering to give it a broader field of interest. And here we identify or define the term as the use of physical, chemical, biological and engineering processes to control and direct the aggregate behavior of cells.
So it's much a broader definition than we had for tissue engineering originally. And it includes advances in complex biological applications requiring input from the physical and chemical sciences. We're looking more at systems biology‑type approaches to tissue engineering. That is the computer simulation of cell behavior, and I look forward to advances in the way of looking at complex cell functions.
Here are the agencies that are participating in our working group. I won't read them all. You can see that we have a fairly broad participation, all the way from basic science, to more applied technologies, to regulation and approval processes, to reimbursement for the technologies that are to be put into practice. We feel it's important that all these factors are part of our interagency working group, to give heads‑up and early indications on new technologies that are coming along.
The working group itself was first established with a five‑year plan in the year 2000, although we've had contacts prior to that time. We've had other types of activities with various agencies that had an interest in tissue engineering, but we just ‑‑ we were first formally established with the five‑year plan in the year 2000, which the plan was approved by the subcommittee on biotechnology. It was revised and renewed in 2002.
And now since July of this year, we have what's called terms of reference. And that's also been approved by the subcommittee on biotechnology, one of the committees of the National Science and Technology Council. And our overall goal of the working group, as you can see, is to maximize the benefit of the federal investment in tissue science and engineering.
Some of the accomplishments we've experienced over the last five years, it's been referenced yesterday. We had a panel report on the comparative international assessment of tissue engineering. That's the WTEC study, the World Technology Evaluation Center study, which was published in 2002.
We have a website so you can find out what we're doing. The federal government and the tissue engineering, as indicated up on the slide there ‑‑ NIH has issued a RFA for tissue engineering in 2003 based on some of the information from the assessment that we had. There's a report on the emergence of tissue engineering as a research field, and all this information's on the website, if you want to look further into this.
We have an ongoing, right now funding opportunity announcement on enabling technologies for tissue engineering and regenerative medicine. So there are three submission dates each year for this funding opportunity announcement, FOA.
We had a workshop in February of this year on stem cell research for regenerative medicine and tissue engineering. Here we tried to get the tissue engineering people to establish better contacts with the stem cell folks, and it worked very nicely.
We had presentations on various tissues from a stem cell point of view, from a tissue engineering point of view, and from a implementation point of view, translational point of view. So it worked out very nicely, and the proceedings from that workshop are also on the website. And now we have the strategic plan that was issued in June of this year.
So why have a multi‑agency activity? In tissue engineering and tissue science engineering, the field is not the purview of any particular one agency. Certainly, we've been very active in the area at the National Science Foundation, and all the agencies that were listed on that slide that I showed you earlier have some sort of tissue engineering activity. And the idea here is to try to coordinate those activities in some way or another.
Tissue engineering and science will require a close collaboration with the physical and life sciences. So we need to get the various scientific people and engineering folks talking together more frequently. There's also a need for bioethics, logistics, pre‑market review, standards and patient reimbursement considerations. And the earlier this is all looked at in the research of tissue engineering, the better.
Recently in the last few years, the Office of Science and Technology Policy and the Office of Management and Budget have been issuing a guidance memo for interagency research and development activities. Among the activities that are highlighted in that memo is a deeper understanding of complex biological systems as a priority for interagency activities in the federal government. So that's another item that you may want to look at more closely.
Why a strategic plan now? Well, there's just a lack of tissues and organs to replace those lost to disease, aging and trauma. Heart transplants, kidney transplants, liver transplants, there's a shortage of these organs for people, and tissue science and engineering is seen as a possible way of mitigating that problem.
Another use of tissue engineering is to replace animal testing for various products, cosmetic uses and medical use of products. So use tissues and tissue engineering to replace animal testing.
Tissue engineering could be used to produce vaccines and other complex drugs. There's many scientific and engineering and regulatory disciplines that require some sort of integration. And fully functional tissue products will depend on accurate, reliable and reliable measurements of many scales.
So we've heard some very good discussions about biomarkers yesterday and the need to provide reliable measurements for tissues in determining how effective these tissues are. And that's one of the areas that we see as important to carry out in research.
Some of the overarching goals for the strategic plan, we need to understand better the cellular processes that are involved in tissues. We need to formulate better means of scaffolding and matrix environments. You heard a lot about the scaffolding yesterday for the various tissues that are of interest to tissue engineering.
We need to develop the enabling tools, mathematical modeling, as well as markers and sensing technologies to get a better handle on how to better design tissues for various purposes.
Keep in mind that when we talk about tissue engineering, we're not only talking about therapeutic uses. We're talking about replacing animal testing. We're talking about sensors based on tissues. So tissue science and engineering goes beyond the medical uses that many people associate with them. And we want to help along the process of scale‑up and commercialization of tissue engineering.
In the plan itself, we have eight priority areas, strategic priorities that address the goals that I just presented to you. So there's a desire to understand cell biology much better; to identify the biomarkers and assays to characterize tissues as you heard yesterday; imaging technologies, improving those, advancing those; refining the cell environment interactions and do some more with the computational modeling systems, the system biology aspects of tissues.
Can we get computers to simulate cells, tissue behavior, in a way that helps the advancement of the field and minimizes experimental needs to advance the field? If you can do it in a computer and you have the models validated, you can save yourself a lot of time, as many of you know.
One of the other priority areas is assembling and maintaining complex tissues. So we're talking about mixture of cells and how to design those cells, tissues. Then there's the need for tissue preservation and storage. So if you're going to have products of tissue engineering, you have to have some way of getting it out to the patients in a way that's useful and rapidly available. And we are also interested in application development and commercialization techniques.
Some of the implementation plans that we have in the plan include convening workshops like the one we have right here today and other types of workshops and conferences to issue agency‑specific funding opportunity announcements or interagency announcements of some kind or another, which we have interagency activity right now. That's an NSF/NIST/NIH joint announcement that's currently on the streets right now enabling tissue engineering technologies.
We want to promote interagency personnel exchanges through participation in other laboratories and postdoc programs and so forth, sabbaticals, to foster technology transfer and translation via SBIR, small business innovative research and other types of joint ventures. Coordinate the policy and development, especially participation in industry‑wide standards, the ASTM standards‑type activities. Exchange knowledge on living databases. So databases are a real issue here in trying to get information transferred among the various people interested in tissue engineering. And then to track R&D activity worldwide.
So some of the expected outcomes that we have in mind, as mentioned already, we look for additional conferences, calls for proposals and grants awarded, looking for publications in the field to further the advancement in the field, patents, entry of new companies in the field, FDA‑approved products, centers for CMS reimbursement decisions and evaluations of the state of the field worldwide and further interagency collaborations.
So those are some of the things that we hope will result from our interagency working group activities.
So thank you for listening to this, and if you have any comments or questions, I'll be glad to try to address those right now. Otherwise, there's further information available at the website that you see on the slide there. So thank you.
DR. PLANT: Thank you, Fred. We have time for one or two quick questions.
MR. DALY: I think this is a very important initiative to get interagency cooperation. It's particularly important ‑‑ tissue engineering ‑‑
DR. PLANT: I'm sorry. Could you please introduce yourself?
MR. DALY: I'm sorry. Mike Daly, Tigenics, Inc.
It's particularly important for tissue engineering, one component that quite often is left out, that we focus on the science, et cetera. But bringing these products eventually to delivery to patients or whatever, it's really important and relative to reimbursement and CMS.
And so is there any ‑‑ I didn't see it on your list of things ‑‑ try to get them involved in terms for reimbursement ‑‑ appropriate cost reimbursement for the technology to get them more onboard of understanding evidence‑based medicine, et cetera, those types of initiatives across these agencies as well as from the tissue engineering basic science perspective?
DR. HEINEKEN: Yeah, that's the objective. That's the goal, is get the CMS people involved right at the beginning and have them involved in the whole research and development process so that they are aware and know what the technologies are. And I think that will ease their reimbursement decisions on these types of products that are coming, yeah.
DR. PARENTEAU: Hi, Nancy Parenteau.
Fred, I looked at the proposal last night, the pamphlet. And what was missing is the history of what products, what has been done in tissue engineering. And I also realize from my talk yesterday some people didn't realize I was talking about an approved product that directly led to a profitable company. And that kind of surprised me, but it was my fault.
However, I think there's a lot to be learned from what we did well and what we didn't do very well. And I would have liked to have seen at least an overview of some of the things that Circe, Advanced Tissue, Advantagenesis (phonetic), Curis (phonetic), all of us that were in the field ‑‑ Genzyme, what have we in your view done properly or what's needed to go to the next step to be more successful.
And I would have liked to have seen some ‑‑ or have maybe a workshop or something in translational issues than can help people in the next generation of these companies do a much better job, say, than we did the first time around.
DR. HEINEKEN: Okay. We'll put that item on the agenda for our next working group meeting.
DR. HOPKINS: Richard Hopkins from cardiac surgery, Children's Mercy Hospital. One of the problems that those of us in the translational research field have faced is that the traditional RO1 NIH study groups have not been a particularly receptive place for this kind of research.
Are you addressing this with NIH? Do you have specific study groups to which tissue‑engineered products should go within NIH, and what are your recommendations on that?
DR. HEINEKEN: Well, there is the RFA. That's the FOA on enabling tissue science and engineering that's currently on the streets. It involves three agencies and six NIH institutes. So that is available. That is currently available. We ‑‑
DR. HOPKINS: But they still go to the same study groups.
DR. HEINEKEN: Well, we have tissue engineering at the National Science Foundation. And we just made five major awards in tissue engineering through what's called Emerging Frontiers for Research and Innovation, the EFRI process, five two‑million dollar awards in tissue engineering. So there are other options.
MS. SEAVER: Sally Seaver. Just a really quick question on your very nice pamphlet. Congratulations.
I think you've got in 2002 a five‑year charter to go ahead, and this is now five years later in 2007. Could you just update us on ‑‑ can you continue on what's happening just really quickly?
DR. HEINEKEN: That five‑year plan has been replaced by what's called terms of reference. That's the official document of the federal government to charter or legitimize what we're doing, and that was signed in July of this year for another two years or so.
MS. SEAVER: Thank you.
MR. RATCLIFF: Tony Ratcliff. Just to clarify the study section issue. Besides the regular study section, there is a study section dedicated to tissue engineering, as well as the regenerative medicine, as well as the study section looking for enabling technologies. So I think the agencies have been addressing that, at least in part, as well as tissue engineering going to the regular study sections.
DR. PLANT: Okay. Let's thank Fred again.
Okay. So now we're going to start the main focus of today's session, which is to discuss some of the trends for tools and strategies for quantifying biological response.
And I think what we saw yesterday ‑‑ and there was some just really excellent talks and some really good discussion. And it really struck me that one of the themes that kept coming up over and over again is how complicated things can be, how difficult it is to know what to measure, how difficult it is to measure anything, and then do we have the correct biomarkers and do we know what biomarkers should be measured.
So what we're going to try to do in this session today is look at what some of those tools might be, either tools that are directly applicable to tissue engineering or tools that have been maybe not yet applied to tissue engineering but could have application, or that are just beginning to be applied to tissue engineering and how they might be useful to tissue engineering.
And to just sort of set ‑‑ I just want to briefly provide a few comments to sort of set the tone of one way of thinking about this very big problem of how do we define biomarkers and how do we discover them. And I'm going to take a page from Anand Asthagiri, who's a ‑‑ he just visited NIST last week. And so I extracted from him the ability to steal some things off of his website for this purpose.
Anand is a systems biologist in the school of engineering at Caltech, and he focuses his system biology problems on developmental biology and tissue engineering. I think we heard a little bit yesterday about how these two fields intersect.
And I thought Anand on his website has a really clear and very simple way of thinking about the nature of this problem, the focus being for tissue engineering to engineer higher order structure and function, to take cells and organize them into structures that are complex, high order, and have some complex function.
And in order to do this, because this is a very difficult endpoint to try to get to, a challenging endpoint ‑‑ in order to do this, one really has to be able to predictably manipulate cell behavior. So you have to be able to figure out how to get cells to do what you want them to do so that you can achieve this complex endpoint. And that in order to do that, it would be really helpful to understand those intracellular mechanisms that drive cell behavior.
And I think I'd like to add for this venue that understanding these molecular mechanisms that drive cell behavior will not only help in the R&D process for developing hypotheses and testing hypotheses and trying things in a systematic way, but that also will help inform the regulatory process by providing some underpinning understanding of how you might have gotten to this complex structure and what its functions might be and what its fate might be.
And, of course, this is maybe the nature of the problem. is that intracellular signaling pathways in the cells are very complex. This obviously is just a small subset of what is known about intracellular signaling. Every day there are new intracellular signaling molecules being identified, and we don't know how many of them we have identified and how many are yet to be discovered.
Now, of course, that's not only part of the challenge but maybe the larger part of the challenge is that all of these pathways share components and there's a great deal of crosstalk between these different pathways. So it becomes a very complex analytical problem not only to identify and measure the specific biochemical molecules within these pathways but to understand how these pathways intersect with one another, such that if you alter one pathway, how that might have an unintended consequence with respect to another pathway. And so this is sort of a huge challenge.
Of course, for tissue engineering, one of the things that really has to be kept in mind is that the extracellular components ‑‑ and particularly with respect to this meeting, the cell scaffolding or the extracellular matrix has a huge role to play in terms of poising these intracellular pathways.
And so how cells respond to their matrix, the scaffolding or extracellular matrix is going to set them up for how they're going to respond to other extracellular materials hormones, growth factors, et cetera, in their environment. They're going to respond differently on some extracellular matrices than on other extracellular matrices. And how they're going to respond to mechanical properties will depend on what integrins they're engaged with their extracellular matrix with.
So these become very complicated issues. And so it's challenging to say, okay, what should we measure, and how should we measure, and how should we understand how to deal with this level of complexity. And so one of the things that we want to try to address today is how do you go about understanding this kind of a complex system and making it work for you.
Now, I'm sure that for small laboratories and independent investigators it's a really formidable task, and I think that that's exactly true. I don't think that every individual is going to be able to understand everything that they need to know ‑‑ to discover everything that they need to know to make their products work and to know what directions they should be moving and then how to ‑‑ what mechanisms are going to be involved when it comes to the evaluation process.
So this is really a team effort. And I think that that was one of the things that Fred was ‑‑ the point that Fred was making with respect to the interagency plan is that this isn't the kind of endeavor that any individual laboratory is going to do. It's going to require a pooled effort among the disciplines and among different laboratories in order to be able to ferret these complex interactions.
And so we do look to the future for increasingly interactions between the physical, chemical and biological communities and computational sciences and engineering in order to ferret out some of these things. And so that sort of is part of what we're going to explore today.
I'd also like to bring up that part of the effort of helping to provide this infrastructure, and to further understanding of these complex events, and what do you measure and how do you measure it is a new subcommittee through ASTM International, which is on cell signaling. And there are a number of people in this room who contribute to that committee, and I would encourage anybody else in this room who might be interested to please find out more about it and be contributors to this.
But part of the genesis of this committee is the realization that directing cells to migrate and differentiate and assemble in some desired fashion and have some desirable function requires some understanding of the intracellular pathways that go into directing these events. And that is going to require quantitative measurements of cell signaling biomarkers such that we can pool these data and really understand the big picture.
Again, this is going to aid both R&D, and it's going to aid quality assurance and quality control. And, of course, as everybody has said already, it's going to require more than one kind of measurement and it's going to require more than one biomarker in order to get an idea of where the cells are in their plan and be able to predict their fate.
So at NIST, some of the ‑‑ one of the things that is being proposed through the ASTM is a guidance document that sort of describes how you might go about making quantitative measurements, for example, at the cellular level. And this comes from work that's been done at NIST. And I think Dr. John Elliott will maybe touch on some of these things.
But one of the things that we have been trying to provide is methodologies for quantitative automated microscopy. One of the issues of automation, being to try to remove bias from the measurements to take data on large number of cells. And we could see that, as if you look at any particular field, you might get one view of things. If you look at lots of cells, you'll see that there are lots of different things going on. And so you can't really draw a conclusion based on one look‑see.
You want to be able to also quantitate things, parameters that have to do with morphology of cells. I mean we've heard a number of people talk about how cell morphology is probably a really important indicator. They're many parameters associated with cell morphology.
We don't necessarily know how that sort of way downstream phenotypic property relates necessarily to all of the details of intracellular signaling or what its impact might be. But certainly just by measuring morphology, we know we can get a very good robust measurement that will provide information from day‑to‑day, from lab‑to‑lab that will allow normalization.
Once you can query cell morphology, then you can also think about applying that same kind of approach to quantifying intracellular markers, like gene reporters, for example, or amino histochemical stains, and then start to then build up a dataset of parameters that tell you something specific about the biochemistry and the signaling pathways going on inside of cells.
One of the things, of course, that is really critical about automated microscopy is that you can measure large numbers of cells, and this is really critical.
There was a little bit of discussion yesterday about the difference between accuracy and precision. Accuracy is getting the right answer, and precision is being able to do the same experiment and get the same result over and over and over again.
This is an important concept, I think, for biological processes because everybody knows you look in a microscope and you see lots of different phenotypes from your cells. And no cell is identical to the cell next to it. And is this experimental error? It's really important to be able to distinguish what's experimental error from what is natural biological variability.
Any group of cells is going to have variable responses. Probably any tissue construct from sample to sample is going to have some variability associated with it. It's really important to know how much of that variability is biological in nature, is real variability, represents the real answer and how much of it is experimental noise.
And, in fact, there's a great deal of variability just in biological response. And you can see that from if you take enough data, you can see that there's always a distribution. Regardless of what phenotype you're measuring, there's always a distribution of responses.
So it's important to have good measurement capabilities so that you know what is the range of responses that you expect to see in your population. It's not all experimental noise, but you have to know what the difference is between experimental noise and biological variability in order to understand what the implications of that is. And, of course, there's information in that biological variability that tells you something about the processes, the mechanisms that are going on in the intracellular signaling events.
So we're going to have two talks today that are going to handle at the cell‑based level how do you take lots of cell data and then what do you do with all those data. How do you develop models based on those data?
And so, again, this is going to be information that is not directly applied in these talks to tissue engineering but is applied to toxicity and to other kinds of models to understand intracellular signaling pathways. And, of course, it's a small step to go from there to particular applications in tissue engineering.
There are, of course, other ‑‑ every scale of biology is important in this game. So I sort of started out by talking about the cell scale. But the tissue and the organism scale is also important, and it's important to have tools at those scales as well.
Can you apply these kinds of things, these kinds of analyses at different scales? And, in fact, this is a very recent example of applying DNA methylation to trying to understand a cartilage product and what is the cell type most predominant in this cartilage product using DNA methylation screens to evaluate this cartilage product, to evaluate what cells are, and whether or not there's been fibroblast overgrowth in this product.
Again, this is another example of collecting lots and lots of data and then using sophisticated analysis to understand the result. In fact, Buddy ‑‑ it was Buddy Ratner yesterday who mentioned using principal component analysis to understand materials properties. Well, materials being complicated, yes, cells being even more complicated, it's really important to have good statistical evaluations and analyses to understand all of these complex data.
Another technique that is going to be very important in tissue engineering and I think it's been alluded to a little bit already in this meeting is proteomics. And we'll hear a talk today about proteomics, again, not necessarily applied to tissue engineering. But there are plenty of examples ‑‑ or examples I shouldn't say plenty ‑‑ but examples showing up now of interest in applying proteomics to tissue engineering.
So basically, proteomics is what are all the proteins that are being secreted that might show up in a tissue or in the surrounding media of the tissue that might tell you what is the signature that is expected for this tissue. If it's operating normally or if it's functioning abnormally, these signatures might be very diagnostic to evaluating that tissue product.
Again, on the organism level, when you implant a tissue, what is the proteome that might be available to you in the blood? These are very complex analytical problems, both from the measurement point of view and from the analysis point of view. But they may provide signatures that give you a really good handle on how is that organism, how is that patient responding to the implanted material and is that normal response, is that abnormal response, and catching that early in the game.
In fact, I found that there's now a ‑‑ I guess that this is a job advertisement that I've taken some of the verbiage out of through the McGowan Institute and the Windber Research Institute that acknowledges that broad clinical implication of cellular therapy will require patient specific understanding ‑‑ we're talking here about personalized medicine ‑‑ and requires exploring the genome and the proteome of engineered tissues, and using bioinformatics in order to really advance the frontiers of regenerative medicine and provide a good assessment.
So it's tools like this that we want to address today. Again, we're sort of on the forefront of what tools are applicable and how to apply them. But at least we can start initiating a discussion here.
And I picked this slide to end on from Tony Ratcliff, where he sort of puts everything into one box. What are all the challenges? And clearly, it's obvious that understanding mechanism is a huge challenge and a very, very important component of being able to be successful at bringing regenerative medicine and tissue‑engineered products to market.
If you don't understand the biological processes, it's very difficult to define biomarkers and to assess patient response. And, of course, that idea of defining biomarkers is probably key to this whole thing. It's a function of what do you measure, how do you measure it, and how do you interpret what you've measured so you know how it relates to an outcome that you're interested in.
That's all I want to say with respect to introducing this session. So if you'll permit me, I'd like to go right on to our first speaker of this session who is Dr. Ken Giuliano.
Dr. Giuliano serves as the principal scientist at Cellumen, Inc., where he's involved in the development and integration of new cell‑based reagents and cellular systems biology profiling assays.
Dr. Giuliano was formerly a principal scientist at Cellomics, also an assistant professor in the department of neurology, neurological surgery and cell biology and physiology at the University of Pittsburgh Cancer Institute.
He's authored many papers, particularly on the use of advanced fluorescence‑based reagents, high content screening, and cellular systems biology. And he's also authored and co‑authored 15 patents. He received his PhD in biochemistry from Colorado State University.
So let's welcome Dr. Giuliano, and I'll try to get his talk up here.
DR. GIULIANO: Thank you, Anne. Thank you for inviting me to describe our technology, and I apologize ‑‑ letting me substitute for Lance Taylor, who's the CEO who really just wanted to be here but just couldn't today.
What I'm going to talk about is what Anne mentioned, is the cellular system biology. We talked some about system biology already. But what we do is use cells as the simplest system, and what I'm just showing you there briefly in the title there is a heat map and also some of the cell dynamics.
These are actually cells where the mitochondria have been labeled and they're treated with a toxin. You can see that the mitochondria potential goes down. So it's really time/space activity in defining cellular systems biology.
So what I want to talk to you about today is just introduce the company and cellular systems biology as we've defined it and take a little piece out of what we call cellular system biology and talk about the cytotoxicity profiling as an example application of CSB or cellular systems biology.
So this is very similar to what Anne just showed, just a little different order, where we start with the continuum of molecules, cells and tissues. And this is mainly where we work here at this interface, cells and tissues. But, again, the systems go all the way up to organs and whole animals. And we know that the single cell is the simplest system that we can define as a system because it has emergent properties.
So the challenge now is that now we know the cell is an integrated, interacting network. And what I'm showing you here is a tumor cell, a brain tumor cell where actin has been labeled, and I'm looking at the dynamics over time of the cell moving. And, again, a fairly simple function such as this, it involves gene expression, the integration with membrane receptors acting all the time, membrane pumps.
How does a cell attach to a surface? It has to be attaching in the front and releasing in the back so it can move forward. You can see some of those attachments being laid down. What signaling pathways are involved? What phosphorylations or other post‑translational modifications are going on so the cell can move?
And, like obviously, the cytoskeleton has to be coming apart and assembling at the right place and the right time for the cell to move; and what molecules are being synthesized and degraded, and how's it using energy?
So it's really ‑‑ what we're trying to attack it as, as an integrated, interacting network and how do we define that network and use it mainly in what we interact with pharma in terms of increasing efficacy and decreasing toxicity of potentially lethal compounds.
So how did we get to where we are today in terms of emergence of cellular systems biology where it started about a little more than ten years ago now, where if you wanted to do cell‑based assays in pharma or in industry, the cell population responses, there was HTS methods and there was whole plate readers.
And we were looking at cells in a whole plate, in a well in a whole plate. And you were looking at, as Anne just mentioned and I'm sure that Lani will talk about ‑‑ you look at the population average.
That came with essentially with the advent of Cellomics, and high content screening was automated cellular imaging. And there was the first generation of HCS readers that one to two cells, two features, were measured in the cells.
Then came multiplexed HCS, which was automated cellular images. Now three to four features were measured. Now we're starting to get a moderate amount of data. So that had to come with some kind of data management.
But now where we are and want to get to is cellular system biology now, where we optimally multiplex the cell and tissue imaging. And we need new reagents for this. And what we're doing now is generally measuring greater than ten features. And now we need much more automated data analysis, and that includes classifier informatics.
So it's really ‑‑ the cell is an integrated, interacting network. And the reagents and tools to look at ‑‑ to measure those processes and the informatics then to really simplify the interpretation of those data because otherwise you just end up with this mountain of data.
So the cellular systems biology approach has as its foundation genomics, proteomics and metabolomics. And what we've done is ‑‑ everybody has their own definition ‑‑ come up with functional biomarkers and use cell arrays and also tissue microarrays to look at the cellular system biology in single cells on really a two‑dimensional substrate, as well as cells in tissue microarrays.
So the idea is this cellular system biology profiling approach where you've seen a perturbant, like, for example, a compound being added. And you can see that as it cycles through the different things that are going on in the cell in terms of organelle changes, translocations within the cell, changes in transcription, for example.
What this does is give us a profile then, which is shown on the left of this, what we're calling proprietary cellular biomarker panels. And we get a readout of multiple parameters, multiplex parameters. But then also we have a proprietary profile database then with classifiers. And what it does is go through and match the profile that we get for a particular compound, for example, a particular perturbation, and then match that with our proprietary database, and then use the classifiers then to classify the response that we get using these multiple parameters. And I'll show you an example of that, how that seems to work a lot better than just doing these simple assays.
So what are the tools then that we're using for the cellular systems biology? So we start out with, again, like I told you, these cell models, which are mostly two‑dimensional but also patient cells and tissues. And what we've been doing is using existing imagers. So we're not developing new high content readers. We use existing ‑‑ the Cellomics and other platforms ‑‑ so we're platform independent ‑‑ to look at time, space activity within single cells.
So these are imaging systems that read single cells, multiple colors within a single cell. And, for example, 384 well plates is our main modality of analysis.
Reagents and profiles. So I won't talk much about reagents. But we have a series of reagents where we can manipulate the cell with, for example, gene switch proteins and siRNAs. For example, we can change the concentration of a protein and then measure the effects of that protein on cells. And that includes ‑‑ which I'll just show one example here of biosensors, of positional biosensors, that report their activity as a position in the cells.
So we can report activity such as kinases, proteases and also protein‑protein interactions using these biosensors and add those into the profiles that we're measuring for cellular systems biology.
So it's imagers, reagents. Also now, the important part then is this informatics and classifiers that we're building now with this huge amount of data that this generates. And you can imagine if you measure four colors in half a million cells at ‑‑ and I'll show you ‑‑ different time points, you very quickly get a very large dataset that you need to manage and be able to interpret the data out of it.
That's why we say that we can then build this systems knowledge, and then since it's mainly pointed to our pharma right now in terms of them being able to make decisions on efficacy and toxicity as early as possible on the discovery pathway ‑‑ I think that's what I showed, start here.
So how do we implement this cellular systems biology? And what we're attacking are really the three different parts of the drug discovery process, the early drug discovery, drug development and clinical trials. And our products then range from ‑‑ and I'll go into a little bit more detail on each one of these ‑‑ cellular models disease for early drug discovery, cytotoxicity profiling, which is an example I'll give you more information about for during drug development. And then for clinical trials, really patient sample profiling ‑‑ so we can take patients' tissue samples and help to initially really stratify them for clinical trials.
So in the drug discovery realm then, the cellular models disease is what I told you ‑‑ disease relevant cellular models, and the biosensors for measuring activities and the profiles, and then what we believe then, this improves the quality and quantity of lead compounds that drug companies want to move forward.
And really, the profiling tools ‑‑ and by using this high content approach with the biosensors, we can now start looking at targets that were really intractable before. And because we do collect information about the system, we can start flagging off‑target effects at the earliest stage during the drug discovery process.
For drug development, we have cytotoxicity profiling. So what we want to do, is the goal here is to really identify potential toxicity before entering expensive preclinical testing. So we can at least prioritize lead compounds to give the drug companies an idea of which drugs are much more likely to show toxicity. And then they're able to rank those before they move into those more expensive trials.
And finally, clinical trials as patient sample profiling and what we want to do really is to improve trial enrollment and really therefore new drug efficacy, so by stratifying patients and then using profiles with their own tissues.
This will also very close couple to what I showed you before, is make these proprietary panels of cellular biomarkers that can also be used as diagnostic tools. Our collaborative partner there is the Mayo Clinic. So let me just talk about this one example then, the CellCiphr cytotoxicity profile.
So as I told you a little bit before, then ‑‑ so the advantages, then, what we do is a systems biology approach now. We monitor multiple functions at multiple time points and multiple doses of compounds. So we can see it's a combination of all of those. And, of course, that leverages the sensitivity and throughput of high content screening.
What we've done is validate this now to high throughput screening standards. And that was a requirement that we had to do ‑‑ to enable pharma to accept this. And what we use is a 3D four wheel capacity plates and extensive quality control, especially on the data analysis side, classifier software now, to simplify the interpretation and predictivity of this cytotoxicity profiling approach, because we are profiling the system itself; not just look at the cell depth but give us some insights on the mechanism of action of the potential toxicants.
So the first panel we developed was in collaboration with Millipore. And what we attacked were several different ‑‑ what we defined as biomarkers then ‑‑ for example, a stress pathway activation, that's a biomarker. And then the feature then, for example, would be ‑‑ I'll show you on the next slide ‑‑ would be, for example, a stress kinase phosphorylation activation, organelle function, oxidated stress, DNA damage to the cell cycle, and cytoskeletal integrity.
What I've show you here are four different colors from the same field. So this is one of the plates as these four colors, which gives us one set of biomarkers. And the second set of plates has another four colors in the same well that gives another set of biomarkers, and there's more detail then.
For example, Panel 1 now was built with HepG2 cells, which are human hepatocellular carcinoma cell line that's widely used for toxicity measurements because it's easy to grow up a lot of these cells. And I'll show you some primary cell results that we also have.
For example, by measuring the ‑‑ we saw some images of Hoechst or DAPI labeling the nucleus. What we can measure are several features: how many cells are lost, a simple assay like that, or is there a cell cycle arrest going on or DNA degradation, what are the changes in the nuclear morphology? And there's another host of reagents now for measuring oxidated stress, which would be measured with the phosphorylation of a particular histome.
DNA damage response, we can look at tumor suppressor activation. We have a couple mitochondrial function assays, a mitosis marker, another measure of where in the cell cycle a compound may act. And a microtubule, we can measure microtubule stability in cells by the destabilization of the microtubules, by looking at the distribution of the component tubule and protein.
So that's Panel 1. Before I get into results, what we found is that there's a better result ‑‑ we get better results if we combine Panel 1 and Panel 2 data. So let me just tell you about Panel 2. Now, Panel 2 is designed with primary rat hepatocytes, which are much more functionally metabolic and, of course, they're primary and they do give you a different answer ‑‑ a complementary answer than the human hepato tumor cells, the HepG2s.
Again, multiple maximum of toxicity again, we measure 11 parameters. And we do this at three time points: acute, early, at 48‑hour exposure. So I did mention that with the HepG2s, those were acute, which is in about one to two hours early, which is 24 hours. And HepG2s, we go out to 72 hours for our chronic exposure. So we found that that's essential because compounds have different effects with these different time points.
So these are all fixed endpoint assays. Now, again, we have the four colors, two‑plate assay, where now we've changed out a couple parameters. And we look at apoptosis, peroxisome proliferation within the hepatocytes, phospholipidosis. And, again, mitochondrial functions is very important for primary hepatocytes as we learned yesterday. Stress pathway activation and cytoskeletal integrity again in this panel.
So how do we look at the data now from these two panels right where we're adding a compound going at three different time points adding ‑‑ again, the compound is added in a 10 concentration dose curve, and we're collecting four colors or at least 11 biomarkers out of each well.
So, for example, this acute profile then this ‑‑ now we can start color coding things to make it a little easier to visualize. This could actually be the response of a single cell. But one of the things we do is look for the AC50 or the EC50 response. And I'll show you some of the details there.
But this would be, for example, the color‑coded response of the early, acute and chronic profiles for a single drug at a single concentration. You can see very quickly then you could start building up a large array of these. And we show it here as a heat map in terms of known and unknown compounds that have been got ‑‑ that have gone through both profiles at different times.
What we'll show you here is that you could start clustering these compounds based on the CellCiphr profiles of knowns and unknowns. And then as shown yesterday in terms of classifying the scaffolds themselves, we can start classifying the compounds or the toxicants now in terms of these different biomarkers. For example, this compound C24 would fall into more of a stress kinase induction classification. So the idea is collect this mountain of data on time/space activity within the cells, cluster the compounds by function using known and unknowns and then classify those.
So we get an automated readout of the experiment. And then this just shows a little more detail where each of the compounds is tested at a ten‑point dose curve. And these are just some images, just a couple example images shown at each of those concentrations. Again, we found this very important because we've also found that drugs have ‑‑ compounds have different mechanisms of action depending on the concentration and the time that they're exposed to the cells.
So we measure 11 features each and measured in each cell, three time points, ten‑point dose curve. And then we curve, fit the data in this case. In this case, we tabulate the AC50 values. That's shown in B here. We could then build heat maps out of this data. And then this is ‑‑ I'll show you a little more detail on this. It's called a mountain map where we can actually start now comparing the response of particular compounds to the total set. So let me just describe then a profiling case study then within the cytotoxicity profiling that we've done.
So what we have, we've taken 137 compounds, and that's 101 unknown compounds and 36 control compounds. But we have drug safety data for 137 compounds. This was done in collaboration with our CHA partner. And what we did was use this human drug safety data and then score each of those compounds on a scale of zero, 1, 2, 3, 4 for in vivo toxicity. So now we have a known set of human toxicities on each of these compounds. And then we tested them in the CellCiphr HepG2 Panel 1 and CellCiphr rat hepatocyte Panel 2.
One of the results here then was that CellCiphr, the Panel 1 then, that showed similarity to known controls, the group toxicity rank order and a safety index ‑‑ are really our goals in terms of what do we want to get out of the profile.
This just shows you just one way of looking at the data. What you're looking across here are each of those across the top are the compounds and down the Y axis here are the parameters, the features that we measure. So this should be about 30 of those, and they're color coded by the AC50s, then, in terms of the cooler color's blue as a millimolar. Yellows and reds then move from micromolar to nanomolars. So the more potent or more toxic the compound, the redder it will show up on this heat map.
The way they're ranked across from left and right are what I told you. We have the in vivo toxicity data from those that are minimally toxic to significant toxicity, and also then there's some that didn't have any of the safety data at all.
This is one way to look at the data in terms of just looking at one compound. There's a couple of compounds that are boxed here, and then you can see the responses of each. But, again, that's just one way to visualize the data. It's not a really easy way to look at it.
Another way here, this is really a mountain plot, for example, that we call it. On the Y axis now, we look at the ‑‑ again, the AC50 goes from millimolar up to nanomolars. So the more potent the toxin, the higher this mountain will be.
What's listed on the bottom then are the different features. And what I'm showing you here is that a known compound, etoposide, is shown in red and the response profile for an unknown compound is shown in the blue. That's been coded to us as H25. We're blinded to it, but we can see that it's ‑‑ in terms of its activities, it's not quite as efficacious. But we can see that it has a very similar profile to etoposide.
So that's one way of visualizing the data to look at comparisons between them. And shown in the background there is the gray which you can't really see on the screen here but which is just the maximal response for the entire set. So that's how it just ‑‑ how these two compounds then compare to the response of the entire set, 137 compounds.
So how do we classify the response from these 137 compounds? So those that I mentioned here ‑‑ so that produced over 4500 dose response curves. So, again, we wrote it here. But it is very difficult to apply manual scoring methodology to handle this analysis.
What we then did was take the assay data from the compounds and we used that to ‑‑ with the in vivo scorings to construct a classifier now, a first‑generation classifier to rank the compound toxicities.
So now we can take all of these data ‑‑ and I showed you the analysis, that of the clustering. And we used the principal component analysis to build this first‑generation classifier. And what I can say right now is that had an improved performance over simple cytotoxicity assays. For example, MTT which just shows you a cell loss or cell death assay.
The idea is that is it really worth getting all these extra data. Let me just show you here then ‑‑ so this is a table now of the results of all those analyses.
So if you look at the in vivo toxicity as we rank them from the compounds, the number of compounds then in each of these ranks overall was 137. If you look at a very simple assay, which is 24‑hour cell loss, how well then did the classifier then pick the significant moderate, minimal toxicities in terms of ‑‑ it did fairly well. It got 84 percent.
But if we start looking now at the CellCiphr classifier, which collects all these extra data on the toxicity, then, we did very well. We caught all 100 percent of the significantly toxic compounds, which is really one of the things that pharma wants to know. How do we rank these?
We would then rank those in terms of lower. It's pretty well guaranteed that it will give you in vivo toxicity. So you want probably to rank those lower as you're deciding about what will go into a clinical trial or preclinical, even into expensive animal treatments.
In each case, you can see here that the classifier did a little bit better than just the very simple assay for cell loss in terms of just what kills cells. And overall then, it had ‑‑ this first‑generation classifier had 82 percent accuracy in matching what we know from the in vivo toxicity of these compounds.
So with that said then, what we are working now is developing ways now to get these data to a customer. We can give summaries of the data. We can give all the data. Some actually want to see all the data, and we can give summaries.
Then what we're really working on is this ranking, being able to rank them. So if all they want to know is how it's ranked and then what we're calling a safety index, it'll give them an idea of just a couple parameters where they have a choice of looking at everything but how to actually rank the data and what kind of probability it has of going forward in the drug development process.
So those are just the first Panel 1 Panel 2 that I've showed you. We just showed you just a ‑‑ I'm almost done. This is Panel 3 now, which is rat hepatobiliary and kind of taking the cue from ‑‑ that we've been hearing about quite a bit lately of going from two dimensions to something that's almost three dimensional now by doing a sandwich system instead of single cell overlays, which our Panel 1/Panel 2 is.
Now if we overlay these with collagen or Matrigel, what we have then is really hepatobiliary model now where you get the highly differentiated rat hepatocytes. You get chronic exposure. Now we can do three days plus. We can do a 384‑well plate. And what we can now measure is cholestasis, steatosis and mitochondrial potential.
And what you're looking here in the red is a marker of mitochondrial potential in these cells which aren't really changing that much. And then the green, then what you're seeing is the green is a dye that actually gets pumped out from the intracellular compartment into this hepatobiliary space. And what we can measure then is the effect of compounds on the pump that's actually pumping those compounds out.
So that would be an assay then ‑‑ one of the assays that pharma wants to know about in terms of how quickly are compounds being metabolized and then pumped out so we can actually add compounds, and measure the change in the pumping of the compound, and simulate the pumping of the compound out into the hepatobiliary space, and measure the activity on the pump. So that's just one of the parameters then that we can measure in this new Panel 3, which is a hepatobiliary model.
So just a couple more things that we have in development now is human and rat primary cell lines in terms of developing new panels of cytotoxicity as well as new parameters, then, if we're going to measure new features of biomarkers.
We're working on tissue selectivity panels for neuronal cells and cardiomyocytes. We want to branch out from just the liver toxicity. We focused on that first because that's what pharma said was the most important.
We've done a little bit of work now and we're working on stem cell derived cultures into the particular organ or tissues, culturing tissue‑engineered array models. And as I showed you there, we're moving towards more kinetic live cell panels, like the model I just showed you with hepatobiliary and some of the others that I showed you, the kinetic measurement of mitochondrial potential and also cell motility in general.
So let me summarize then, just very briefly then. So the CSB, cellular systems biology approach is being implemented ‑‑ or we're implementing it, again, to improve the efficacy and decrease the toxicity of leads, clinical candidates and drugs.
And I showed you one example here of the cytotoxicity profiling and just showed how the results of that cytotoxicity profiling demonstrates this idea of cellular systems biology where we can measure and manipulate the cellular system, collect the data, and then really the development of the informatics that allows those data to be interpreted.
And with that, I'll stop. Thank you.
DR. PLANT: Thank you, Ken. That was very nice.
Bob.
DR. NEREM: Ken, over here, your last slide was sort of ‑‑ triggered. Can you say anything more about what your concept is for a tissue‑engineered array model?
DR. GIULIANO: I can't say too much, not that it's a secret, just that we haven't worked on it that much. But I mean that's one of the reasons I think that we are here, to learn more about that.
But right now, what we're working on in terms of tissues are the patient sampling profiling. And that's our initial stage of that. So we haven't really done much on the engineering of those at all.
DR. NEREM: Do you see the possibility in the future of actually being able to do the kind of analysis you're doing on a 3D tissue‑engineered construct?
DR. GIULIANO: No, so that's a good question. And a lot of that then depends on moving these ‑‑ the detection system, which we can already on some of the present detections. For example, the array scan has an apotome, which we can do some sectioning.
So I think that, yes, that would be, definitely, because we already know that we're going to get a different response from going from two dimensions to three dimensions. And, yes, I think that that's definitely in the plans. And we really need to address that sooner than later.
DR. NEREM: Thank you.
DR. PLANT: Keep in mind, we will have a roundtable discussion after this session before the wrap up. And so some of these might be the subjects of further discussion.
MR. DALEY: Mike Daley, Tigenics. It's a very fascinating approach.
And my question is have you ever really validated it? And the validation comes from the pharmaceutical industry is ‑‑ we only hear about the successes. But there are a hundred times more, a thousand times, maybe a million times more failures. And many of those failures fail in the development process of various aspects of cell toxicity or whatever it is.
And so, therefore, what you really should do in a blinded fashion is try to pair up and find out whether or not something that historically we knew failed because of X, Y and Z, were you able to detect it in early high throughput screening processes and predict that it would have failed and saved somebody gazillion of dollars. But the real value to you is, then, now you've validated the process for prospectively predicting an outcome
Have you done that, or are you planning to do that?
DR. GIULIANO: So, yes, so that's ‑‑ there's two parts to that then. What I showed you then, the case study, I guess I needed more detail on that ‑‑ was with CHA Cambridge Health Tech Advisors, we worked with ten pharma companies. And they each sent us 10, 20 compounds that they knew the human tox data on. We were blinded to those, and that was the table that I showed there where we could actually predict those pretty well. So that's one thing there.
So we have done that, and we're actually following up with some of those to do more assays for them because we did get a fairly good result there that showed that our approach was ‑‑ at least in this first‑generation classifier was better than just looking at simple toxicity assays.
But we're doing also what you exactly said, is taking some of these fallen angels or other compounds that we know have failed and starting to run those through as a demonstration of our assays and these profiles and help ‑‑ using those libraries to actually develop the next set of profiles then.
So the answer then is two where, yes, we have begun to validate it, and we do have good results in that. And we're working on further validation, so we can go forward and do prospective predictions using new generations of this classifier with new profiles.
MR. HICKMAN: Hi, Jay Hickman, University of Central Florida. I'm a little confused. So, in fact, if I understand you, you've basically identified a number of markers for cytotoxicity, and then you're basically taking a fingerprint with those markers and comparing it to a database.
Now, at other times you sort of alluded to you're relating that back to pathway information inside the cell. How much of that can you do, or really is it more kind of a cytotoxicity fingerprint based upon what you think are known markers for cytotoxicity?
DR. GIULIANO: So that's a good question. And what I didn't talk much about is our cellular models of disease, where we have specific biosensors and other array of biomarkers and features. And what I'd like to see is us marry the two together.
What I showed you were really what we thought were some cytotoxicity or cell stress features that we could measure biomarkers. And those are what's in the panels right now. And that's Panel 1 we developed with Millipore, and that's a kit that pharma can actually use right now.
But what we want to do is take some of the biomarkers and biosensors and things and manipulations that we use for cellular models of disease and marry those with the cytotoxicity and make an overarching big cellular model of disease and toxicity. So we are expanding the biomarkers that we have for the cytotoxicity outside of what we think are just toxic, but more understanding, what is the compound's effect on the cell as a system?
MR. HICKMAN: So you're trying, in effect, to marry up eventually the Cellomics platform with the cytotoxicity platform?
DR. GIULIANO: Well, the ‑‑ or whatever measurement platform you wanted ‑‑ or whatever ‑‑ for example, it could be the INCELL or it could be (inaudible) ‑‑ but it's measure up our reagents and assays with our informatics and really the marriage of those two to make the toxicity more predictive as well as the cellular models of disease.
MS. DONG: Hi, Jiyoung Dong from the FDA Center for Devices. I was wondering if you could comment on sort of the pros and cons between an assay that could give you a lot of information that's very complicated and maybe very sensitive, but requires a specialized user or a lot of training versus some other system that maybe gives you sort of a medium sensitivity, a little less information, but is a lot more accessible, or can be used by people who aren't really trained or little to no training.
DR. GIULIANO: I guess I say so in front of everything. So that's exactly what we're grappling with in terms of ‑‑ pharma for a long time has been using simple assays and they're getting more and more complicated. And the burden is on us to show ‑‑ to tell them. That's what they ask us. What are these new data going to tell us? We have these fairly simple assays. Can you do it better?
So the idea is we keep it as simple as we can. But there's always going to be a space for these very simple assays to give you a yes‑no answer. Where it shows here on the table, they are fairly predictive. But if you want to go to the next step ‑‑ because the drug development process is so expensive what we've done is gone to that next step. And it's because they are more complicated. We can either transfer it to them or offer it to them as a service, which a lot of them have done that because it is a little too much for them to develop in the lab.
So I would say that there's room for all of these assays. It just depends on the amount of information that you want to get out of it and how you want to use it in the end.
DR. PLANT: If I could interject, too, I think that's a really deep question and something that we ought to bring up again this afternoon and discuss in some detail.
DR. NYBERG: Scott Nyberg from Mayo.
Have you tested a common drug like acetaminophen, which is safe at low doses but toxic at high doses? I'm curious to know what your system would show.
DR. GIULIANO: Yes. So we have, and then we need to go, like you say, up to the millimolar amounts to start showing toxicity in this, as well as some other compounds where it's really hard to show some toxicity. And that's why we need to go actually to that 72‑hour time point.
But using that as a lesson, that's helping us design some of these other cellular models or toxicity assays, where we can start seeing more pathway‑specific or something where we can show those drugs that are safe but do have some effect on the cell that we can actually measure then as a baseline.
DR. NYBERG: So you do dose responses as part of the testing?
DR. GIULIANO: So as part of the testing, yes. So we do a ten‑point dose response curve in duplicate, and those are what give rise to the AC50s. So we do go across three or four orders of magnitude of concentration of each of the compounds.
DR. PLANT: Okay. Let's thank Dr. Giuliano again.
All right. So our next speaker is going to be Dr. Lani Wu from UT Southwest Medical Center. Lani comes to biology by way of pure mathematics, computer science and electrical engineering. So this might be something entirely different.
Dr. Wu graduated from University of California at San Diego with a doctorate in mathematics and left mathematics faculty at Princeton University to join the fledgling research division at Microsoft, where she led a number of research efforts, including projects in video compression, semantic search and speech‑noise separation for multi‑microphone input.
Prior to arriving at UT, Dr. Wu was at Rosetta Informatics and was a fellow at the Bauer Center for genomics research at Harvard University.
Today she has two main research areas in her lab. She's interested in understanding network design principles that lead to cell polarization, and she does this by combining mathematical modeling and microscopy‑based experiments. And then second, she's interested in understanding how single cells respond to perturbations. And, again, this involves combination of populations of cells and high throughput immunofluoresence microscopy and high performance computing.
So let's welcome Dr. Lani Wu.
DR. WU: Thank you to the organizers for inviting me here. Before the invitation, I have never thought I would be talking in the tissue engineering workshop.
Coming from mathematics, engineering, we were completely captivated by the complexity of phenotypes the microscopy images captures. For us, when we look at it, it captures the cell morphology. It captures the intensity and the spatial organization of a readout. It also captures the individual cellular response.
We're interested in understanding where we can identify the cellular states by the observed phenotype. If you look at these images, these images are drug‑treated HeLa cells stained with DAPI, PR, and P30A.
What I can see visually ‑‑ just my own interpretation, but you can tell me what your interpretation is ‑‑ is that drugs are single mechanisms induced to cellular response. And drugs of the similar mechanisms induce distinct cellular phenotypes.
So one of the questions we like to ask is that can I look at the cellular response and try to predict the perturbation of mechanism? In our lab, our research is not really looking at toxicity or looking at large data ‑‑ it's not a pharma type of work. But for our lab, the conceptual model is starting from a biological network and monitoring the network components and look at the translocation and its spatial distribution.
What we would like to do is represent ‑‑ within the individual single cell, represent the phenotypes we see mathematically as a point in a high‑dimensional feature space. And I will talk to you about that a little more later. But conceptually, that's how we're thinking about it.
If you perturb a cell with a perturbation, ideally, what we'd like to think is that it will get you a point on a high dimensional feature space at a different point. If you have similar perturbation, it should get similar points that are close together. And if you will have distinct perturbation, you would get points in the representation that are far away from each other.
And in terms of perturbation, you can imagine we can use drugs or things that are like hormone, growth factor or anything that you guys have been talking about in this workshop such as substrates, collagen and all this different perturbation you would like to put on a cell.
Today what I would like to take you through is the journey we have gone through in the past few years in trying to understand how we can get any information from large quantity of microscopy data. And since this is the first time we're going through it, what we decided to do is try to do a perturbation that we can easily control and that is using drugs.
I'm going to tell you two different approachs that we have used, and they're slightly different, and they are both published. And for both of the studies, we use a same large dataset of microscopy data. We start from 100 compounds. And these 100 compounds were taken from 15 diverse functional categories, also include about three unknown drugs.
In fact, at the time of building this data, our collaborator on purpose blinded ten drugs. And when we think about it now is that there's really no reason for him to blind the drugs on us because we did not know what the drug's about anyway. And when we treat the drug, we use 16 titrations with 3, 4 dilutions.
Then in terms of markers, we just decided that we're just going to take 11 marker, antibodies that there was easy ‑‑ found in the freezer of our collaborator. And, of course, that is complicated to try to multiplex. So we decided of putting three markers, three antibodies in one set on a single cell and creating five marker sets.
Then we treated ‑‑ what we do is we put HeLa cell on 384‑well plates. And the reason we do that is because if you think about, we're doing 1600 compounds. That means 1600 experiments. And as a small lab like ours, we cannot afford to do anything else. So we treated a cell ‑‑ we cultured a cell on a 384‑well plate, and we treated them with drugs and fixed the cells after 20 hours of drug treatment.
Then we spin them with different antibodies, the five different marker sets I have outlined here. Now, what we have done is that we capture from ‑‑ in each well, we acquired nine images because we just want to get enough cell count, like Anne was talking about. We need to have enough cells to be able to do your statistics.
So from there, we acquire in the order of about 100,000 of images. In each image, we circumvent out (inaudible) individual cell region until we got about tens of millions of cells.
And the question that had been raised in this conference is what are the phenotypes should we extract, should you monitor. In fact, that this is not a easy question. And because we know ‑‑ with our experience in microscopy, we know that if we want to implement a particular feature that biologists see is important, a lot of times these particular feature is very, very hard to implement, requires a lot of time.
And not only that, the problem there is if you only go in with the features you know, you will not find anything surprising. So that's not really where we want to go as the first time through.
So what we decided to do is do an unbiased approach, and we were good at malloc. So we decided to capture anything malloc can give us. And also in the ‑‑ and today we also ‑‑ adapt some other tools that our bio ‑‑ Murphy at Carnegie Mellon have developed.
So in short, we have been able to capture four different types of features. And one type of feature is morphology. So it just tell you how big the cells is and how round the shape is. And the other one is texture. And in a sense, it just tell you the pixel pattern. And then the other one is the statistic of the pattern. And, of course, the very important feature everybody looks at is the intensity. So, for example, we can look at total intensity in the cell or total intensity in the nucleus.
With this we were able to extract features. However, this does not really help us to interpret the data. So the question is now what.
With this in mind, I'm going to take you through our two different approaches. The first approach we call the univariate approach, and the reason we call it univariate approach is because we're going to look at individual features independently.
So, for example, with one marker one, you can extract one feature. And with this feature, we're going to look at this feature across all the population ‑‑ across every cell in the population. And then we can build a population's statistic for a control population.
At the same time, we can also build another population's statistic for the drug‑treated population. And then we gather more features in the same marker. Remember, we're doing it individually. And so we can extract the population's statistics. Then we can do that for the other two markers and extract more features and compute the population statistic again.
With this, what we decided to do is that the ‑‑ if the drug‑treated population is shifted to the right ‑‑ and we're just going to mark it right. And the bigger the shift that we're going to make it, the redder it is. Similarly, we're going ‑‑ if it shifts to the left, then we're going to mark it green. The bigger the shift is, we're going to make it greener. And with this, I'm going to put it together with some real data here.
So you can look at this. This is a camptothecin dosage data. And what I'm pointing out here is the DNA intensity. So on the bottom, you have low dosage. You can see an obvious ‑‑ visually you can tell maybe it is G1 and G2 cells, two populations. And then when it goes to higher, the G1 cells disappear and you only see the G2 cells. And that is consistent with what we know about camptothecin effect on G2 arrest. The population statistic I'm talking about is cumulative density function. What it means is that the function here, it goes from zero to 1. And if you look at any X axis ‑‑ and the Y axis tells you what is the percentage of cells that have this feature with value less than the X axis at the point in the X. And so with this, we see that we have a shift when it goes up to about ‑‑ starting at the third concentration and going up. And that is a feature that you can easily get from FAKs.
Now, I'm going to look at a feature. It's a little bit harder to get from FAKs. That is anillan, a cytoskeleton ‑‑ cytokinesis marker. We're going to look at the average intensity. In order to get an average intensity, you need to know the cell size.
With this, we see a shift in the beginning. Even when it's green ‑‑ I mark it green here. It does not have much shift. But the real effect comes much stronger, starting from the fourth concentration.
Then we look at another feature. For example, the p53 cytonucleus intensity, this ratio here you cannot get from FAKs. So this is a feature that is very specific microscopic images. And the special thing you see from this picture here is that the intensity ‑‑ the effect really comes out much later in the high dosage.
Putting all this together, thinking about the 11 markers I have talked to you about, we're going to put it all together. So in this study here, we did a very easy ‑‑ we did about ten features for each marker because we just wanted to test whether this is going to work. And so we just did whatever malloc gives us. We just do it easy without any effort.
So, again, the first area on the top is the recap ‑‑ what we call profile on the DNA intensity. So it's stuck on pretty much black to red and then dies off a little bit. So it's anillan feature and a p53 feature.
And so what is this telling us? When we see this, we thought we've got something here. The reason we've got something here, well, for us is because we see that the drug effect does not really come out at a very low dosage. And if you look at it carefully, the first effect comes out at the third and fourth dosage right away. And then at a later dosage, you can kind of detect by eye where the second dosage effect comes out.
So we thought this was nice for just camptothecin. So we decided to look at it for all our 100 compounds. And for this presentation, I'm going to put the first 50 compound here.
So for each drug, I put ‑‑ there are three fingerprints here. The first fingerprint is a replicate one. We did two replicates. And second one is replicate two, and third one is the average profile. And you also see some white here. We mark it white if the cell count is very low or the two replicates do not really satisfy our statistic significance similarity metric.
So here, I was very pleased to see this picture because there's no way I could have done so ‑‑ all the hundreds of ‑‑ thousands of images to get this sense of what the drug is telling me, what is the drug response. And not only that, what you see is that the drug response is complex. You see lots of white, red features ‑‑ profiles and you also see some very green profiles and some mixed. So this is a really nice way to give you an intuitive utilization of what the drug's doing. It is for us.
Now, come back to my previous question, does drug or similar mechanism give you similar response? And for us, on this mathematical representation, we wanted to be close together in the representation space. So what it means is we want it to look similar.
So what I have done here is I sorted a camptothecin drug response one more time on the features side. I found it very green and very red. That's the first figure. And I keep the same cell order for the feature, and then go across at each rack on the screen here.
So what I can see right now visually is that drugs are similar mechanism. I get similar profiles. And drugs are distinct mechanisms. I get very distinct profiles. At least I can see it by eye.
And, for example, if you look at protein synthesis, all these drugs are supposed to be the same, in the same functional category. But if you look at it carefully, they're slightly different. And, in fact, when you go in and look at the mechanism, they are really ‑‑ these drugs are hitting the cell at a different spot even though they are all categorized as protein synthesis.
So like most computational persons, once you have something like this, you go and try the most obvious thing in malloc again, hierarchal clustering. So in this image here what I show you is not every ‑‑ from the previous image you can see that not every drug that we have done shows some response. So what we've done here is we take the 60 drugs with the highest response, with some technical response.
So on the left‑hand side, this is our drug list and on the top is the functional categorization. And so when the drug is in each functional categorization, we make it black. And the blue ones are the ones that are cut up with a blinding fungus. And then the big ‑‑ there is we did a ‑‑ by clustering out a hierarchal clustering and trying to put a profile, they are similar together. And you can see that they essentially group almost by the functions. And so this was very encouraging for us, and this was the first approach.
So in the second approach, the problem with the previous approach is that we're looking at individual features, one and one time. There's no way in that previous approach I can see the correlation changes among multiple features.
So instead what I'm going to do right now is I'm going to look at all features in one cell together. So going back to the conceptual model of the high‑dimensional feature space again, I'm going to pull an individual cell. I'm going to extract all the features in one time and map it into one point in this high‑dimensional feature space. As a result, I can have many dots that represent the control population. And, for example, I can have the blue dots represent the control population.
Now, what can you do with this? I know that there are many different multi‑variate approaches out there. But for us, when I have two populations, the only question I'm trying to address here is do I see a drop response. And that is the question we interpret, is how can we separate them out.
So one of the most obvious algorithms out there is the support vector machine. What that means is we're going to a best separate plan to separate these two populations out. And so if you think about if you have a control population and then have a drug‑treated population, if the drug is very low, what do you expect this to do? This two population is going to mingle with each other. And no matter how well you can separate out these two populations across sufficient accuracy, it's going to be really low. That is 50 percent because you cannot separate one or the other.
So what this gives us is the separating plank in a classification accuracy. So this is our representation here, and to put it into real data again, so what we're doing here is looking at drug dosage data. And for every dosage, we're going to look at individual cell and put them into a point on high‑dimensional space. So we will have a control population and drug‑treated population. And we try to find the best plan and the normal vector.
The hyperplane can be described uniquely by no more vector and also the classification accuracy. So when you go up the high concentration, you see that the drug‑treated cells and the control cells can be more easily separated. And then you get another ‑‑ so when you go up, then you get all the profile. There's no more vectors across the dosage.
The no more vector, it just tells you really the direction of thickest phenotype change. That's what it's telling you. And because it's a direction ‑‑ so then we can cluster the no more vectors by how close they are pointing to the same direction.
So with that, we can ‑‑ on the top of this graph here, we can cluster into ‑‑ for camptothecin, we were able to cluster into three different directions of clusters. And we call it a zero effect, one effect and second effect. And the reason why it's zero effect is that we also look at the classification accuracy. Remember I told you that in the very low dosage, the low dosage population should be very similar to the control population. And we want to make sure that the accuracy is high enough before we can see that this is an effective drug effect.
With that, we were able to look at the low dosage and the no dosage effect, and the first dosage and the second dosage effect. And that has been documented in the literature that camptothecin has two different effects.
So when when our postdoc was doing this, then we say okay, go ahead and see what you can do with multiple drugs. So he applied his algorithm and get different profile for different drugs. And he handed us this high (inaudible) clustering.
And what this is saying, there are ‑‑ I don't remember how many drugs he can count. But the functional category has been indicated on the parentheses part. So HH, they're all the same functional category.
This tells me that this is a perfect functional ‑‑ this can classify the drug essentially perfectly. And what it is telling, are you sure this is right? Did you just do this in the PowerPoint? And it turns out that he did not. He did not cheat. And so we were very pleased with the accuracy.
With this high‑dimensional representation, the profiles, one of the good things about it is that the classification accuracy was really high. But one of the bad things about it is it's really hard to look at it, to visualize it in high‑dimensional feature space.
So to accommodate that, we pick the three ‑‑ the most important features space and then put all the drugs, and then visualize them in three dimensions. And what that allow us is go in and look at a drug you're interested in by the dosage effect and then look at what are other drugs that is close to it. So this can also provide ‑‑ even though it's in a high ‑‑ it's really done in a high‑dimensional feature space. It can also be visualized in a lower‑dimensional feature space to provide you some insight into the drug mechanism.
Now, to come back to the original question that had been raised by many people in the workshop is what are the phenotypes to capture. Since we go through the approach, our bias ‑‑ collecting dataset, unbiased features. In fact, in the study I just talked to you about, in each individual cell we captured about 300 features. And there's no way I want to go through these 300 features in every cell. And so what we decided to do is that we're going to drop the features systematically and see what our accuracy would still be the same. So what we've done is ‑‑ this is the graph to show you that when we drop the features to about 20 features that we still get reasonable high accuracy.
In fact, this is a very strong result in a sense that this is telling you for almost every marker that you do, you probably only ‑‑ especially in our dataset. I cannot speak for your own biology data. But for our dataset, we only need about 20 features per marker set. And the features are different, depends on which marker set you pick. They have different ‑‑ and the importance has been summarized in this figure on the left here. And it also depends on what are the drug mechanisms you are trying to uncover.
So to summarize what I just talked to you about here is that we have gone through two different approaches. One is the univariate approach and one is multi‑variate approach.
And the pros on ‑‑ the good things about the univariate approach is that it's very easy to ‑‑ I can ‑‑ I just showed you earlier. It was one strike ‑‑ can give you a whole big overview of the drug response. It's very nice. And also that you can ‑‑ we were able to combine all the different marker sets together. However, it's not as sensitive. But in the multi‑variate approach, it's very sensitive in terms of drug mechanism clustering. However, it's much less intuitive. So you cause much more complication of power.
Since I've told you earlier, our lab is really in a basic research direction. And so what we would like to do ‑‑ and found these different approaches that I have shown you about ‑‑ is we want to look at specific biologic networks. Networks that we are looking at ‑‑ biological systems that we are looking at right now in the lab includes understanding chemotaxis, primary human neutrophils, the insulin response in adipocyte, and drug response in cancer cells.
We are also looking at multiple markers. And we need to pick specific markers for their own pathway and extract the exact phenotype we're interested for our biology questions, and going through the perturbations that we need to do.
And this work that I have done by the lab ‑‑ with the great people in the lab and also my long‑term partner, Steven Altschuler, and the univariate approach was developed by Mike Sacks, and the work in multi‑variate approach was developed by Lit‑Hsin Loo.
Thank you. Also I would like to acknowledge my funding source. They have been really helpful and our collaborators. Thank you.
DR. PLANT: Thank you, Lani. That was really interesting. Anybody have any questions?
DR. TUAN: Rocky Tuan, NIH. That was a very fascinating talk. So it just brought some things to my mind.
To some extent, you took a lot of very cell‑specific data and you homogenized them. I mean ‑‑ and you analyzed them and you homogenized it at the end. And I was very fascinated by you saying that when you start dropping the variables that you end up actually ‑‑ were able to use ‑‑ I think you said 20 or something.
So just brings back some memories of sort of classical biochemistry in the early seventies. Christian de Duve, of course, won a Nobel Prize for subcellular fractionation, which could be extremely technologically very specific in terms of recovery of mitochondria, membranes and so forth. Very clean, I mean, so I'm just wondering ‑‑ and that's also homogenizing. It's literally homogenizing. You take a tissue, and you grind it up.
So I'm just wondering if you were to take just a few of the things that you're looking at, namely, whether something is associated with nucleus or cytoplasm or what have you or cytoskeletal things. If you were to just take some of those and go take a look at response to drugs, whether you were able to, I guess, validate using the classic more biochemical approaches.
DR. WU: Okay. I'm not quite sure I understand your word about homogenize it because we did not. All that's ‑‑
DR. TUAN: I guess that's not the right word. I mean taking all the data and then crunching to get a plot that you show us.
DR. WU: Okay.
DR. TUAN: The same thing with taking a cell part, I mean we have all these fractions and you do all these enzyme assays and whatever. At the end it's just a plot. It's some bar little graph. That's what I meant. I don't ‑‑ whatever the term is.
DR. WU: Right. But I think that one of our things is different than for the microscopy approach is because when you see a plot you can go back to the original data and original features. And it's different than Western blot. Everything was lost.
DR. TUAN: No, no. It's not western blot. I'm not talking about Western blot.
DR. WU: Right. But I'm saying that when ‑‑ the main thing about this plot, the way ‑‑ at least for us, is we look at this plot. Oh, this is some interesting features. And I go back to the features, and I can even go back to where are the population of cells, the images, and go back to the origin to go forward. And so that's one point that you raised.
Then the other point is do we look at actin or ‑‑ for example, we have a marker with actin microtubule. And, in fact, you said markers that we would have very good detection ability, too.
DR. TUAN: But still my point is that if you were to go back ‑‑ no, I agree that the power of the analysis is amazing, of course.
I mean but if you were to go back to a cell, that cell naturally can't tell you anything, right? It's the population that tells you the information. That cell alone is one component inside the dataset, which is valid and quantitative and all that. But that one cell doesn't tell you anything. You need the 10,000 or 10 million or whatever to give you the ‑‑ so that's what I was just thinking.
I mean, again, biochemistry is old‑fashioned and what have you. But I'm just curious whether you have taken any of these spots and just kind of use ‑‑ to some extent almost prove whether the old biochemistry was worth anything.
DR. WU: Oh, okay.
DR. TUAN: Just a comment, not a criticism.
DR. WU: I mean that's a really good comment. That's a new work that I did not talk about today, is talking about analyzing heterogeneity ‑‑ is like what everybody have been talking about, no cell is the same.
But the question there is every cell is different. And we have just come out with some algorithms and some analysis to show that it turns out that if you really look at ‑‑ let me rephrase it. The question is do you ‑‑ if you really want to analyze the heterogeneity, do you really need to think about every cell is in their own state or not?
This question, I think it really depends on what is your question you want to answer, what is your success metric. And in the study we just did recently using the same drug data again, the success metric there is really classified drug mechanism.
What we have found, surprisingly, is that it turns out that we only need about four different populations of distinct phenotype to really quantify the drug mechanism. What that means is that even though the complexity of heterogeneity looks daunting, if you look at it the right way, maybe the complexity is not as high as you would expect. And it will provide you a better insight on how to attack this problem.
DR. PLANT: If I could just interject one thing.
I think, Rocky, one of the questions you're asking ‑‑ and it's a really important question ‑‑ is are the cell‑based assays validated with other biochemical methods to know that you're measuring the right thing. And that's where the accuracy question comes in, right? Absolutely, yeah. And that's maybe a separate question. I'm sure that you guys did some of that. But that ‑‑ and that's one of the things that the ASTM committee is very concerned about as well.
But I think also one of the significant things that might be part of your question is that if you're looking at ‑‑ it might be required that somebody look at a whole bunch of data in order to decide which are the most sensitive markers that are going to be most important that, say, people in tissue engineering labs would then want to focus on.
MS. LUMELSKY: Nadya Lumelsky from NIH. Sort of a flip‑side of the same question, talking about heterogeneity, in reality samples of primary cells or tissue‑engineered samples, they're inherently heterogeneous, functionally heterogeneous.
So would your method allow to identify distinct functionally different populations and study them in this multi‑variate parameters?
DR. WU: Let me quantify this. I think what I have shown you today is what can be done. And when you want to address a different question, does not mean that that's the way you should be doing it.
MS. LUMELSKY: But it would be nice if that could be used.
DR. WU: Right. In fact, that we have ‑‑ maybe I can talk to you later about it. We have a different project that have ‑‑ to kind of address this question, but it's really preliminary. So I will probably prefer to talk to you in private about this.
DR. BERTRAM: Thank you, Dr. Wu. Fantastic presentation. Very exciting.
Two questions, one philosophical and one specific and technical. Do you have ‑‑ yesterday Buddy presented and said a little bit of heresy, which was fantastic. He said the way we get our principal component analysis to work real well is we actually go in and we filter it and then we analyze it. By the way, I don't think that's cheating. He said it was cheating. I happen to disagree with him. But be that as it may, I think he's giving us some tremendous insights.
The tremendous insight is this and the question therefore is this: have you considered integrating human intelligence as part of your automated algorithms in order to give you new insights? And if yes, I'm curious how you're actually doing that pre, post and where the integration occurs.
DR. WU: Okay. So that is the part where we're doing human neutrophils right now. And in that particular project, we have very specific questions we want to ask like what is a front and back coordination. And so we have gone in and are really trying to implement the right features to capture what we want to capture.
DR. BERTRAM: I look forward to that publication.
DR. WU: I look forward to it, too.
DR. BERTRAM: The second thing, which is much more technical, it's a little bit of a spin‑off of Rocky's point and maybe it was made previously. One of the things I noticed, and it was probably just for the purposes of presenting here. But you have a tendency to have people focusing on the responders.
In a tissue engineering situation is those that are not responding ‑‑ as a pathologist, I'll say I use the term a "stroma." A stroma's supposed to stay static so that the cells that are responding ‑‑ now, the stroma's dynamic. It's constantly under remodeling. So even that isn't totally static.
But my point is, from a technical level, can your algorithms distinguish those cells that are also stable, if you will, and not changing?
DR. WU: Not in the current two approaches that I have described, but a new approach that hopefully is coming out soon, that it will be able to address that.
DR. PLANT: Was there another question?
And there will be, of course, another question and answer period later.
Okay. Well, let's thank Lani again.
We're going to take a break now and reconvene at 10:30. So it's about a 15‑minute break.
(Whereupon, a recess was taken.)
DR. PLANT: Okay. Time to get started with the next part of the session. We're just running just a little late, but I think that Dan has a presentation that mostly will work now.
So it's a pleasure to introduce Dan Martin, who's our next speaker. Dan's assistant professor and proteomics facility director at the Institute for Systems Biology in Seattle and also a lecturer in hematology and oncology at the University of Washington.
Dr. Martin received a bachelor's degree in mechanical engineering at Cornell and an MD at Yale University, and completed an internship and residency in internal medicine at the University of Colorado in Denver.
He joined the Clinical/Research Fellowship in Hematology and Oncology at the University of Washington in 1998, and after a year of clinical training began basic science research under Rudy Eversol at the Institute for Systems Biology.
And his research has focused on the use of proteomics methods for analysis of the androgen receptor complex in prostate cancer. In addition, he's developed a program to identify biomarker candidates from cultured prostate culture cells and evaluate the presence of these biomarker candidates in the serum of animals xenograft with the same prostate cancer cells.
And so it's a pleasure to introduce him today and have him give us a talk on mass spectrometry‑based proteomic applications for cell/scaffold products.
DR. MARTIN: Thanks very much. Thanks for having me, and please bear with me because of the sliding graphic. That wasn't one of my PowerPoint 2007 graphics. That's the projector, I guess.
The ISB where I work is in Seattle. And in case you haven't heard recently, this is the beautiful Seattle skyline in the summer. This is really what we've been having lately.
I'm here today from Seattle to talk about mass spectrometry‑based proteomics and potential applications to cell/scaffold products. Keep in mind, I'm an academic. I have no financial disclosures with regard to biotechnology products.
And the approach I'm going to take today is one that's more of metrology. The aspects of proteomics that I'm going to try and explain are how we use this technique to measure. And hopefully, you will use your own creativity to figure how will it work for me.
So the first question is what is proteomics. One definition and I like is proteomics includes the identification, quantification of proteins as well as their localization, modification, interactions, activity and ultimately, their function. That's a very broad definition.
And just to make one point very early on, the proteome versus the genome, it's important that we distinguish the features. And the one that I would really allude to is that for proteomics as I'm going to describe it, we do not have PCR. There's no amplification in the study of proteins.
Obviously, the genome is static and the proteome is quite dynamic from cell to cell as we saw in a lot of the images we've seen. It's quite heterogeneous from a monoculture from even neighboring cell to neighboring cell. And there's a tremendous variability in the amount of protein one might see from sample to sample.
So proteomics is a lot of things to a lot of different people. I'm talking about mass spectrometry‑based proteomics, but proteomics might mean two‑dimensional gel electrophoresis to somebody, mass spectrometry to me, protein chips or yeast 2‑hybrid, phage display, antibody engineering. It means a lot of different things to a lot of different people. So I will not say that proteomics isn't any one of these things. I will just tell you about what it is to me. And I'll tell you what it is and what it isn't.
So it is a highly powerful tool for protein identification and quantification, and it's complementary to other technologies and analysis methods. And what it's not, it's not magic. It's not going to give you all the answers. It's not like someone's going to walk in with a little device and take your scaffold, or whatever it is you're studying, and go do, do, do and it's going to have all your proteins and everything in a dramatic list for you.
It's really not all that simple. But it's not that hard compared to some of the other stuff in multi‑dimensional space ‑‑ I honestly can't say. It's not hard, but it's not that simple.
And it's definitely not that cheap. It really depends on the scale of money you think about, but these instruments run on the average of half a million dollars and up, excluding service contracts and the FTEs to operate them.
So mass spectrometry basics 101, what can we measure? We can measure proteins in mixtures. We can do quantitative analyses of protein expression. We can measure post‑translational modifications, such as phosphorylation which is a challenge, and the field is advancing in that aspect. And glycosylation present or absent, not the nature of the glycan at least in mass spectrometry‑based proteomics.
There's a whole field of glycomics. There's all the omics that you would like. One of them is glycomics, metabolomics, lipidomics. We can also measure protein interactions.
So I'd like to give you a flavor for what mass spectrometry‑based proteomics can do for you. And I've talked to a lot of people at this meeting, and I've gotten the impression that there's a room full of very, very intelligent people, but many of whom really don't know all that much about proteomics. So I'm going to focus my talk on the proteomic neophyte. And so here is basically what proteomics can do for you.
If you have a protein gel in a metaphorical way, you can put a name on every band on the protein gel that you can see. And you can also put names on bands that you can't see. So this would be then a theoretical experiment. So this might be your scaffold.
Basically, let's just ‑‑ in this case, let's just talk about cells in culture. Ultimately, you want to take these cells, and you want to get them into the mass spectrometer to figure out what proteins are there. What you do then is your take your cells and you make proteins. This is proteomics. We make proteins, which I'll say is a string of amino acids with a molecular weight of greater than 10,000.
And then you take the proteins. You digest them into peptides. And we typically use trypsin, and I will elaborate a little bit on this because trypsin is an enzyme that cleaves after arginine and lysine. And that will become important. K is lys, R is arginine. And the reason that's important is that both the ends of these peptides now have amines. And these amines support the ability of the groups on either end to carry a charge.
And so ‑‑ now let's just shift gears to say what mass spectrometry primer ‑‑ what does a mass spectrometer measure? It measures charge to mass ratio. So that's ‑‑ we call that M/Z. So you'll see that in a number of slides here.
So you can't measure something in a mass spectrometer unless it's charged. And the basic mass spectrometer configuration looks something like this: you have to ionize your material to make it charged. You have to somehow do a mass separation define ‑‑ separate your analytes according to mass and then you have to collect the ions or count the ions.
So I'm just going to touch very briefly on ionization. So this is electrospray ionization. And so here we have ‑‑ this is a very, very small emitter here. This is a very tiny fuse to let the capillary ‑‑ the diameter of 300 microns into which you typically pack a reverse phase chromatography resin. You put peptides such as this into this emitter, and you put a large voltage between the emitter itself and the mass spectrometer. And this induces a cone. That's what this picture is of.
So basically, your peptides are stuck in a little droplet of liquid. The solvent is charged because of the potential. And as it moves towards the mass spectrometer, this solvent evaporates. And as it evaporates, you ultimately wind up with a higher charge density and the molecule essentially ‑‑ or the droplet explodes when the charge gets too high. And in the end, you wind up with a peptide that is charged. And the charge lives on either end. That's why I drew it this way.
So this is a type of mass spectrometer called a triple quadrupole, and it's the version I'm going to use just to explain mass spectrometry. So there's three of these triple ‑‑ there's three quadrupoles which are four poles arranged around a center, and they're in tandem. And here's your ionization source. Here's your mass separator. Here's your ion collector. It's effectively ‑‑ this is a mass spectrometer.
This quadrupole looks something like this. This is a theoretical set‑up. This is the actual set‑up. And basically, there's a series of ‑‑ there's voltages placed on opposite rods. I don't really ‑‑ you don't need to know the details of this. But basically, you have electric fields that guide the movement of the ions.
And so what that means is when the charged peptide is in this region, it will behave in a certain way. And so I've described it in such a way that it's stable. It's able to stay in this quadrupole region, and it will make it all the way to the end of the quadrupole as it travels from left to right.
You can also set up the electric fields so that it doesn't produce stability for your charged analyte. And when you try and move this analyte down the length of the quadrupole, it hits one of the rods. And when the charged peptide hits a rod, the charge goes to the rod and you still have the peptide there, but it's not charged and you can't measure it anymore.
So the way to think about this without getting too involved is that this system can act as a filter. It's like the tuner in your radio. You can tune it to a certain charge to mass ratio. And so this is another analogy of what I used to do as a kid with the old dial radios. And what you would do is you'd just spin the dial as fast as you could. And as it'd hit each station, you'd hear moments of noise that represent the signal of that radio station.
Now, you can imagine this is the same exact thing that's going on with the mass spectrometer. And it's not just one peptide that's going in because if you digest a lot of protein, there's a lot of peptides. So there's maybe hundreds of peptides. And for a moment, your mass spectrometer is tuned to each one of these peptides. And you get the signal, and you measure a histogram, effectively, of what's there. And so this is how you would say these M/Zs, i.e. peptides, are there.
And so there's another thing, though, that you can do. And this is why there's three of these quadrupoles together. If you have a particular peptide, you can excite it such that in moving left to right, it goes faster. You put energy in.
And so you create a situation where there's a kinetic collision between the peptide and some nitrogen atoms. And so what happens in that collision is the peptide will break. And if you tune the energies right, what happens is the weakest bond, or the most likely to fragment bond, is the amide backbone, none of the side change fragment.
So what you get is this peptide that is ultimately the population ‑‑ there may be thousands of them ‑‑ can be broken in this fashion that any one of the amide bonds. And just to remind you, there are 20 amino acids with 19 different weights. And so that will become important in just a second.
So this is a schematic of what happens. If you have a particular peptide sequence, this is what you would get. These are all the possible bond breakages, and you get pieces; where if you start from the left ‑‑ and remember, this is also why it's important if there's a charge on one side, the amino terminal and the carboxy terminal both halves of the broken peptide have a charge. So they can be measured because they will have a charge to mass ratio.
Now, I just want to redraw it for you. So we had ions on one side. And I labeled one B, and I labeled one Y. That's the nomenclature of the field. But I'm just going to redraw the Y ion for you. So I'm just going to turn it around a little bit. And it's the same sequence, but since we work ‑‑ we're in a left to right universe. I want to redraw it so that this is all the fragment possibilities redrawn.
So what does that mean? What that means is if I actually looked at all the fragments, you can generate a ladder. And the ladder represents combinations. So you can see the N. You can see the NS. And so here's the N. Here's the NS. Here's the NSG. And these are the B ions moving this way. And so it stops at a peak. So here's NSG. Here's NSD. And then here's RGAISG. And it stops at this big, tall peak.
So effectively, you get a ladder. You can think of it as a ladder. You can think of a fingerprint. But what this means is because there's 20 amino acids, I can tell you if you look over here at the V, there's a hundred units between ‑‑ that defines what a V is. And every amino acid has a different separation.
So ultimately, with a good spectra, an intelligent person can sit with a calculator and say okay, I think there's a V here and maybe a V here. And you can deduce the sequence of the peptide that you broke. And so this ultimately is the fundamentals of mass spectrometry, deducing the sequence of a peptide.
So this is how we do shotgun proteomics. And shotgun proteomics is just a high throughput way of doing what I just showed you. You would have a group of peptides that go into the mass spectrometer. And you just turn off the first two of the three quadrupoles. And you take the third one, and you do that radio scanning trick I showed you, which generates this spectra. And you say okay, this is what's there.
We call this an MS survey scan. These are the peptides that were there. And then you decide well, I'm interested in this one over here. And when you decide you're interested in that one, you turn on the first quadrupole. And what it does is it filters just like the ‑‑ I showed you before. And you turn on the second quadrupole, and it accelerates and breaks that peptide. And then you take the third quadrupole, and you do the radio trick where you scan the whole sequence. And lo and behold, at the end you get the spectra. And, again, you now have the spectra. And you can say okay, this peptide was there.
But this is not something that we can actually do in high throughout fashion if you have to sit there with a calculator because this is an old ‑‑ this is an easy to explain type of instrument. The current instruments will do this roughly three times a second. And you need a lot of people with a lot of calculators to actually handle that kind of data.
So what we do is we use a computer program. And one of the variants we use is ‑‑ the name is called SEQUEST. And it's ‑‑ it functions by searching the peptides ‑‑ it assigns a peptide sequence to a spectra. The computer does it for you. And the way it does it is you start out with a scan of something of a mass. M/Z you know. So if it's 750 and you knew the charge date, there was two charges on it. Its molecular weight was 1500.
So what this program does is it goes to the proteome, and it starts at the first amino acid. And you say give me every single possible peptide that has a mass of 1500. And you might have ‑‑ depending on the size of the proteome, you could have quite a few of them.
Then what it does is it fragments