At A Glance
Name: Naftali Kaminski
Title: Director, Simmons Center for Interstitial Lung Disease
Medical School: Hebrew University-Hadassah Medical School, Jerusalem, Israel
Residency Training: Department of Medicine, Hadassah Mount Scopus University Hospital, Jerusalem, Israel
Fellowship Training: Institute of Respiratory Medicine and Physiology, Sheba Medical Center Tel-Hashomer, Israel, and Cardiovascular Research Institute and Lung Biology Center, University of California, San Francisco
Clinical Interests: Interstitial lung diseases, pulmonary hypertension
It's every new chip vendor's dream. The respected director of an established microarray core lab decides that his old platform just isn't cutting it anymore and it's time to switch.
Yet, while this ideal customer is the target of at least half a dozen chip providers that have sprung up over the past decade, the reality is that when it comes to core labs, switching platforms is a rare occurence. Many labs integrate newer platforms into their experiments, but most seem resigned to carry out their experiments with the platform they initially invest in.
So few have actually done it that when BioArray News went fishing for labs that had completely switched platforms, it came away with only one name.
Naftali Kaminski is the director of the Simmons Center for Interstitial Lung Disease in Pittsburgh, Penn. In late 2002, Kaminski decided to abandon the Affymetrix platform he had been using since 1997 to use CodeLink bioarrays, then owned by Motorola.
BioArray News spoke with Kaminski this week to discuss the pros and cons of switching platforms.
What kind of studies are you involved at the Simmons Center?
My main research is in understanding transcription networks in pulmonary fibrosis. We have sort of basic science studies in which we try to identify key regulators of fibrosis using a variety of tools and techniques including microarrays. The other side of it is sort of the translational research in which we try to identify biomarkers of diseases.
How long have you been there at the Simmons Center and when did you decide to stop using the Affymetrix platform?
I moved to Pittsburgh two years ago and that's when I made the decision to stop using the Affymetrix platform. I basically did Affy almost from the beginning, so from 1997, during my training, and when I moved to Israel [to serve as director of functional genomics the Sheba Medical Center in Tel Hashomer, Israel] had an Affy system in my lab. Basically it was the only system.
What kind of studies were you doing in Israel?
We were more focused on lung cancer and cancer-related research. But, actually, we collaborated on two things that got a lot of attention - using microarrays to identify diagnostic gene expression patterns in peripheral blood in patients with post-traumatic stress disorder and in multiple sclerosis.
So what made you decide to switch platforms when you got to Pittsburgh?
There were several things. One is that I was not completely excited about the methods in which you detect expression in Affy - specifically the combination of multiple probes in multiple oligos per gene - the fact that you were sort of dependent on a variety of probe-level normalization stages. At that time what Affy recommended was basically comparing mismatches and perfect matches. And I wasn't very excited about it, actually. We published a paper in which we showed the different normalization methods affected the results. We weren't very excited about it, and I was looking for a platform in which I could see the signal. And we looked at both Agilent and CodeLink - these were the two options. The other thing I wanted was to use an open system, and to have a scanner that basically could do everything, and not be dependent. So if Agilent came out with something more interesting we would be able to scan slides and not just be able to scan a single manufacturer. So that's been working very well for us.
So you are able to use other platforms in addition to the main platform in your studies?
That's what we are doing routinely. And also we are using it for other applications, such as array CGH, or ChIP-on-chip methods. Basically I was being more efficient with my space.
Were there other incentives for you to switch, such as price?
I don't think so, I think that all of them come to the same price in the end. I think we were able to get the same deal from all companies. I think it was mostly the fact that I thought that Affy was more dependent on normalization, you could never really see the signal, and the fact that I wanted an open platform. There's a third reason that I think many go with, and that's if you have a big core facility or institute [in a region], you sort of say 'Why waste $250,000 in setting up a second one?' So I wanted to have a complementary system.
Is there another core lab in Pittsburgh?
There is a core facility in Pittsburgh [The Genomics and Proteomics Core Lab at the University of Pittsburgh]. So when we came to establish the Lung Translational Genomics Center, I didn't feel the need to get the Affy system, just because if somebody wanted to use it, they could.
But what happened to all the data from before that was developed on the Affy platform?
Well remember, that's actually one of the problems with Affy. Whenever they came out with a new batch of chips it was hardly at all comparable.
So you couldn't compare data from one batch of Affy chips to another?
From one generation to the other. This was highly relevant in the beginning - the human 6K array, then the human 12K - I think now they are more stable. So it wasn't like we were losing a lot. You cannot do direct comparisons, you simply cannot do it. But we've been able to look at fibrosis versus normals, and later some people recently published why and where the platforms don't agree. A group at Harvard did this. We now feel I think a little more confident in regards to cross-platform comparisons than we were previously. But in no case are you going to do controls in Affy and your sick patients in another platform. What you'll have is basically balanced groups. You just compare what has changed between the conditions.
Was it a significant capital investment for you to switch platforms?
Well if you move from an open-ended platform to Affy it costs you a lot because setting up the system is around $250,000 to $300,000. If you go the other way around it's actually not very expensive because a good scanner is something between $50,000 to $100,000. You know when you buy a bunch of hybridization ovens. We are just retooling our system to use both CodeLink and Agilent. We've been doing it previously but we're sort of now going to do it on high-throughput and it will cost us around a few thousand dollars.
You mentioned CodeLink and Agilent a lot. Could you use other arrays like ABI or PamGene to do similar studies and get comparable results?
I don't know about PamGene, but for [every array printed] on slides, you can do it. So for instance, you know all of these small companies that sell the targeted arrays? We can do them. You cannot do ABI. You have to get an ABI scanner. That is again a closed system. And I think that's probably the main limitation in the ability of new products to go into the market. [It's not worth it] if you need to buy a scanner every time you change your microarray platform.
Did CodeLink approach you or did you just read up on a number of platforms and pick one?
I read up and selected CodeLink.
What have been some of the downsides to switching platforms?
Bioinformatics support. I've been there with Affy. I've seen the same things. I saw when there were sequence errors with the mouse chip with Affy.
You expect these things to happen. And there were similar problems with CodeLink - so you have to have your finger on the pulse. So that's one of the downsides. At the beginning it was not that easy. If there's a new product, less of the computational software didn't deal with CodeLink at the beginning. So everytime there is a new product there are some issues.
Has it improved over time?
CodeLink has been very receptive. They are not developed like Affy, and they are not as smart as Agilent yet, but they have done a good job.
There's been a lot of talk about standardizing data, and some people have floated the idea of creating a metric for comparing data. Are these going anywhere or is this just talk at the moment?
Well people are talking and actually thinking about it, but here's my concern: If you take exactly the same Affy arrays, and ran experiments this week and next week, you are going to see differences.
And, we downloaded some of the big concert datasets two years ago and the simple stuff, like clustering, and one of the signals in the data was actually the dates that the experiments were run.
What you have to pay attention to is the design of the experiments and then you can compare them. But anything that doesn't contain internal controls and some calibration methods won't work. So, for instance, one of the things that we do in my lab that is fairly simple, is whenever we run a batch of arrays from the same experimental set, we'll run one RNA as a calibrator. So we repeat the same sample on and on and on. This allows us to sort of get a feeling about the range of differences. So that's one way of doing it in a way that will be clinically relevant for what we are doing.
Do you think that core labs such as yours would be willing to switch platforms for a more affordable, newer chip, or do you think once you make your initial investment you are set with what you have?
Well, in the end, price determines everything. So for instance we are negotiating with Illumina, because their prices seem to be much lower than anybody else, and the other thing is because of their unique designs you can run several samples on the same chip therefore reducing batch effects, increasing high throughput. Instead of doing eight samples a day in the regular lab you can do 72. So I think pricing and the ability to upscale operations will determine a lot. Because when people compare head-to-head CodeLink, Agilent, and Affymetrix, the differences in terms of reproducibility were not that huge. I think in general for clinicians an open platform will be preferable. So slides that work with any scanner in the market are probably better than a unique slide that requires capital investment.