At A Glance
Name: Kevan Shokat
Position: Professor of cellular and molecular pharmacology, University of California, San Francisco;
Professor of chemistry, University of California, Berkeley
Background: Assistant/associate professor of chemistry, Princeton University — 1994-1999; Postdoc, Stanford University — 1991-1994; PhD, UC Berkeley — 1991
Kevan Shokat and colleagues at the University of California, San Francisco, are among a growing number of biomedical investigators taking a pathway-driven, chemical-genomics approach to understanding cell biology. In the April online issue of PLoS Biology [PLoS Biol. 2005 Apr 5;3(5):e128], Shokat’s group — along with collaborators from drug-discovery firm Cytokinetics — describe in a paper an automated imaging and bioinformatics technique for discovering small-molecule probes for cell biology studies and anticancer drug candidates. Last week, Shokat discussed this work with Inside Bioassays.
How did you come to work with Cytokinetics and the Cytometrix platform?
I just happened to go down there at the invitation of the CEO to give a talk about my research. Since my lab works primarily on protein kinases, they thought it would be interesting to see if protein kinase inhibitors would affect their morphology-based screening platform in a way that would discriminate between different protein kinase inhibitors. From their end, they wanted to get a whole collection of protein kinase inhibitors and see what they did, and from my perspective … in general, I’m really interested in structure-activity relationships, and we’ve been doing a lot of gene-chip analyses and proteomic-type profiling, so this appeared to me to be very good window on the whole set of processes in the cell. So we decided to initiate that collaboration. I think we envisioned it first as just a single-pass assay of about 100 compounds — that included some know protein kinase inhibitors, as well as compounds that were thought to be inhibitors.
In the PLoS paper, you used the technology to screen small molecule modulators of kinase pathways, or a variety of targets?
The [kinase pathway] work was totally undirected in a pathway sense. There weren’t any transcription factor readouts, for example, to tell us that, for instance, the NFkB pathway was on or off. We had no idea about that. It was just the movement of tubulin within cells; the morphology of the Golgi, read out by a lectin fluorescence molecule; and lastly, the morphology of the DNA, just using a Hoechst DNA dye. So that was very pathway-blind, essentially. I think you sort of can’t have your cake and eat it too. If you want to look at things broadly and find out things in pathways you know nothing about, then you can’t really use known readouts for existing pathways. You sort of have to take that how it comes. You don’t get pathway, but you get, probably, a whole lot more information — maybe even integrated readouts; for instance, maybe multiple pathways control the Golgi, which we know they do.
Tell me about using this technology in your research. It’s described as an ‘automated imaging and analysis system,’ but it doesn’t seem as if it’s cutting-edge microscopy. Is it software-driven?
That’s exactly right. I don’t know how commercially available the actual microscope with the autofocus is — the multi-well plate reader — but it’s just a 5X image, so they’re just trying to capture lots of cells in a given field in the well. And then it’s all about the masking. For one thing, it’s masking of all the cells, and identifying which of the bright spots in the field are cells. The next part is determining all the characteristics of the brightness — whether it’s speckled, bright, flat, has a rough edge — things like that. Then the last key thing is the statistics that allow them to find even subtle effects early. It’s sort of like if you would analyze a FACS plot or something like that.
So this is object-recognition software?
Yes, and the last thing that’s important about theirs — which is different from the one described in the Altschuler paper described in Science a few months ago (see Inside Bioassays, 11/23/2004 and 11/30/2004) — is that this technology from Cytometrix allows you to segregate cells. For example, in a DNA stain, when cell are going through mitosis, you get a lot brighter DNA stain. And there are two nuclei right after mitosis very closely separated. So there are ways to teach the computer to say: ‘OK, tell me all the cells in interphase, and analyze them for their brightness, and then analyze … mitosis differently for brightness.’ Because maybe if there is a duller mitotic staining of DNA, you would miss that if you looked at the whole range of interphase and mitotic cells. And that’s something — that algorithm — that the Altschuler paper didn’t take into account. I don’t know enough about whether that was a very big feature in our analysis, and whether it allowed us to see something we would have missed otherwise. And Altschuler’s group points out in that paper that this is one direction that would really improve the analysis, to bin. So I think the Cytometrix one is very, very good in that respect. It’s got very good statistics and discrimination of subsets and populations. That, to me, is the part that’s so powerful about microscopy-based screening, and FACS-type screening, where you’re watching single-cell events.
This approach obviously is complementary to other types of studies — you mentioned microarray analysis — but does it replace any technologies you’ve been using?
Like I said, I used it as a potential way to augment the view we get of the way cells work at the transcriptome level. We sort of wanted as many readouts as we could get of what’s going on in the cell due to the presence of a small molecule. There’s really nothing like it, though, that I can think of. The coolest part of it, I thought, was that the phenotype that the computer picked up, which we put in one of the figures — I can’t imagine looking over thousands of wells and remembering the difference between things. So it’s just a very unbiased approach — the binning of the cell morphology. So I can’t think of another way to do that, unless you have a specific readout, and code it like they did in the Altschuler paper. It would be nice if there were many more phospho-specific readouts, and many more biomarkers, essentially, within the cell.
You’re studying kinase pathways with this technology, but Cytokinetics is using it to develop drugs targeting cytoskeletal …
… mitotic motor proteins. That’s a good point, because in any of these morphology-based screening methods, the readouts you choose to include are in general governed by the process you’re interested in. So their tubulin stain is an absolutely perfect readout of mitotic motor inhibition and activity. If I was designing the system to just look at protein kinase inhibitors, I would take every phosphor-specific antibody from cell signaling, and put it into that. So I would design it differently. I worry when people read the paper that they’ll think: ‘Oh, I’ll just take this off-the-shelf method, and put in my compounds that I’m interested in.’ I think everything that people are looking at, they will really need to adapt the system. So in some ways, it’s a very good thing to have — at least, what I hope they do in their business model — to make some kind of core system that analyzes three parameters, but then have the ability to plug in, let’s say, up to ten other fluorescent markers — antibodies, dyes, whatever — and the system would have some basic way of categorizing those binding events. Because then those ten would probably change for everybody’s projects. So it won’t be like a single gene chip, where you put every ORF down; I think there’s going to need to be subsets. But that’s a lot of work. It’s sort of unfortunate that the current venture capital people don’t see the benefit of supporting technology like that.
What’s next for your lab’s research?
We’re really focusing now on the target that we identified [carbonyl reductase 1], and we’re doing two things. We’re doing basic understanding of how that enzyme works and controls apoptosis and acts in the cell. And we’re also expanding to the other 80 members of that protein family, some of which are just really, really hot drug targets. One of them is 11-beta hydroxysteroid dehydrogenase type I, which produces cortisol, and it’s essentially a pro-diabetic target. So if you can inhibit that, you can increase glucose sensitivity, and also reduce obesity. I think there’s a company in Sweden that’s got a big program in that. So we’re excited by this class of enzymes because they’re imminently druggable, and it’s a new category of targets. These enzymes sit, in a tissue-specific way, to control the metabolism of nuclear hormones. So this target is just sitting there in the body, in the right place, so that you can get selectivity at the nuclear hormone receptor level. So you can actually use endogenous nuclear hormones to regulate processes in the right tissue, without having to develop a tissue-specific glucocorticoid receptor agonist or antagonist. So I think of it, potentially, as the body’s way of regulating these small-molecule nuclear hormones. So if we can regulate them, we’ll have it great. I could have probably learned all that without the Cytokinetics screening project, but it’s one of those things where it’s a great inroad to a problem; it’s a nice way to start looking at a new process.
Will there be an ongoing collaboration with Cytokinetics?
We don’t have anything going on at the moment. We did screen many other collections of compounds that we had, and each time we found an interesting SAR. But pulling down the target is so hard, and then crystallizing the structure — there’s a wealth of things that we could follow up, but every follow-up takes so long. I think it’s a great thing, but each round of that ticks off so much work that it’s difficult to do too much screening.