Transcript

Welcome to AACC’s Expert Access Live Online program

Our topic this month is HISTORY AND OVERVIEW OF MICROARRAYS

This month's expert is Theodore E. Mifflin, PhD, DABCC, Medical Automation Research Center at the University of Virginia. View the presentation and direct your questions to our online expert. AACC would like to thank Bayer HealthCare Diagnostics for making this program possible.

I note that like many others you attribute the genesis of microarrays to Schena and Pat Brown. Is there some reason that the work of Wang and Grayston is not considered relevant? (Amer J. Ophthal., 70 (1970)). My reading of this and subsequent papers by these authors and others is that multiple assays were carried out using arrays prepared using a fine tipped pen These were not as dense as present day arrays but were quite effective for identifying antibodies to different serovars of infections organisms. Could it be that these are not considered microarrays because the concentrations were determined by serial dilution to a limiting detectable signal rather measuring signal intensity?
Seattle, WA

Theodore E. Mifflin, PhD, DABCC: My personal belief about who created microarrays is very dependent upon the definition of a microarray. Typically, it is considered to be a uniform geometric network of 'features' at a density > 400 features/cm2. This orderly pattern is best generated using a mechanical device, certainly when the densities > 10,000 cm2. Pat Brown and Mark Schena were instrumental in developing such a device along with the concept of simultaneously examining the thousands of cell mRNAs. Since the development of the mechanical spoting devices were so crucial for current day microarray development, it seems logical to assume their development is viewed as a seminal event. Although I am not familiar with the article that you cite, it could be viewed as a forerunner to the creation of the current-day technology.

What is the general acceptance from clinical labs in regards to microarray platforms? What things need to improve in order for the acceptance rate to improve?
Morrisville, NC

Theodore E. Mifflin, PhD, DABCC: From my perspective, most clinical labs have had very little experience with microarrays until recenlty . In particular, the expression arrays are viewed as experimental devices and have not demonstrated robust clinical use. A variant of the expressionmicroarray can be used to detect single nucleotide polymorphisms (SNPs) and that is where the first commercial entry has occurred. The first microarray device to be used in a clinical environment, the Roche "CypChip", is designed to look for associated with selected members within the Cyp450 family. Monitoring these SNPs may shown to be useful in adjusting drug therapies and dosages. Since the CypChip only was given FDA apporval early this year, it may too early to tell about its impact. I would watch for some early abstracts at the AACC annual meeting next year or 2007 that summarize user's experience with this device. To improve the acceptance rate, several criteria have to be met as I see it: (1) there has to be a significant increase in demonstrated clinical utilty of using microarry-based testing, (2) their use has to be shown to be cost effective and that outcome analysis reveals a substantial benefit to their being incorporated into the molecular diagnostics (MDx) menu, and (3) their use and operation have to become much more robust than currently is viewed.

Not really a question, just some observations based on my experiences. Things to avoid in grants containing microarray experiments: 1) The easiest way to criticize a microarray experiment is to describe it as a fishing expedition. Do not propose to characterize genes that you have not yet identified. If you have no preliminary microarray data, you don't know what genes or how many genes you will find. However, they may number in the hundreds. Simply saying that you will pick some interesting genes to study is a quick way to get a bad score on your grant. If possible, your experiment should test a hypothesis. For example, you might make the hypothesis that certain genes (e.g. apoptosis genes) will get induced. Then you can propose to use microarrays to test that (and propose real-time PCR or Northern blots as back-up methods). That way you can test a hypothesis, propose expected outcomes and controls (e.g. genes that should go up and down), which is a much better way of doing a microarray experiment (or any other experiment). Simply going fishing for genes is a bad approach and always draws the ire of the review committee. 2) Simply saying that you will use some software program to analyze the data or group the genes into pathways is also going to get you in trouble. Microarray data can be extremely complex, and will require statistical methods for analysis. The pathway data that is known is woefully incomplete. Most genes are not in the pathways, anyway. You will need a well-planned approach for analyzing the data. You should have a way of telling whether the experiment worked or not (e.g. did the expected apoptosis genes get activated?). 3) The microarray experiment should not be merely a paragraph at the end of one of your aims. Whatever you do, absolutely do not add a microarray experiment to the end of a grant application as something that you "will also do". Microarray experiments are big, expensive and complicated and they can't be done as an afterthought. Many, many grants have a one-paragraph description of the microarray experiment that the researchers will also do. That is a lightning rod for criticism from the reviewers. 4) If you are looking for genes, you should be looking for them for a reason. Don't just propose to look for regulated genes without proposing to do something with them. Finding the genes that go up and down is not a significant enough goal. You need to be looking for genes with some purpose in mind (e.g. some hypothesis to be tested).
Santa Fe, NM

Theodore E. Mifflin, PhD, DABCC: Your points are well taken. I certainly hope that viewers in the audience today who are considering use of microarrays in their RO1 NIH grant application (or othere) should carefully read your comments. I would also add two other comments. First, as you point out, mimcroarray experiments can be very expensive and budgeted costs can easily pass $(US)50-100,000. Therefore, early and frequenct consulatation with a bioinformatics specialist is essential to get the most for the $$$ being spent. Second, its becoming clear that the expression data can be analyzed via a number of software approachs (supervised vs unspupervised) and so several avenues of data processing should be considered. Finally, when organizing the microarray experiments, consult and follow (if possible) the MIAME website instructions so that the data can be useful in a broader context.

What is the future of microarrays in the detection, identification and direction of treatment for infectious diseases?
Franklin Lakes, NJ

Theodore E. Mifflin, PhD, DABCC: Use of microarrays for detection, identification, and possible drug resistance experssion seem to be a significant opportunity for development. However, there has been little development outside of two examples I can think of now: 1) HIV genotype assessment using a DNA chip developed as a collaboration several years ago by Roche and Affymetrix and 2) Line probe arrays (LPA) for detecting HIV resistence SNPs and HCV genotyping developed by Innogenetics. I would guess that more has not been done because of the developmental cost vs payback analysis that has been done by companies. I don't see microarrays replacing classical clinical microbiology methods (e.g., streaking and identification followed by Kirby-Bauer disc sensitivity analysis) until the cost benefits are much more attractive. They may play a more specialized roles that are more viral related and for organisms that are difficult / lengthy to grow (Mycobacterium spp). A key opportunity is if they can be coupled to microfluidics that allows BOTH growth and identification to occur in a cassette that doesn't need significant operator intervention.

What are the major technical obstacles facing the use of microarrays as clinical diagnostic tests?
Franklin Lakes, NJ

Theodore E. Mifflin, PhD, DABCC: I think I hav addressed some of these isssues in today's forum, but I will quickly summaraize them here: 1) Diagnostic microarrays must show favorable cost : benefits that make them attractive to replace existing technologies if they are to be considered for that purpose, 2) Diagnostic microrrays have to show robust clinical utlitity for specific disease identification and / or treatment. We're not there yet for expression arrays, but coming along for wide scale SNP screening, 3) Diagnostic microarrays need to be performed within a simplified and sturdy format without errors. A package approach probably will become the standard, expecially for use of microarrays with moderate to high density features, and 4) reliability in the form of improved precision and quality control need to be addressed agressively so that QC is at the fron of the technology as compared to being the "caboose at the end of the train".

Is there any advantage to using an antibody array instead of a gene array, or vice versa? Any particular applications where you would select one over the other?
Sofia, Bulgaria

Theodore E. Mifflin, PhD, DABCC: For the measurement of proteins or other large molecules, protein / antibody arrays might be the preferred or only way to monitor these species. Several companies have developed antibody arrays for both investigational use (e.g., BD) as well as clinical use for monitoring allergies and small therapeutic drup monitoring (e.g., Randox). Clearly, if the target of choice is NOT a nucleic acid, then protein / antibody arrays may work well. Protein / antibodys work well if they amont of material to be detected is considerably greater in concentration (> nM) that when using nucleic acid-based arrays. They same scanning systems can be used by both (fluorescence confocal scanners) and they both can quantify the targets. I see a increasing market for them when they are coupled to microluidics for sample processing to create a 'bundled' package similar to the old DuPont ACA packets used in clinical chemistry labs some years ago.

Can you please comment on the FDA's first attempt to regulate microarray products, vis-a-vis the Roche Amplichip decision? How will this decision affect the movement of these devices from R&D to clinical use?
Burlington, VT

Theodore E. Mifflin, PhD, DABCC: From my perspective, this was a learning experience for all of the parties involved and hopefully will be remembered when future microarray-based devices are submitted for condsideration. Since this was the first of these types of devices, then there is an opportunity for later comparison as to operational mechanics. Since not all microarray devices will be based upon the photolithography approach, spotted microarrays still need to be considered as a viable diagnostic platform and that has not happened at this time.

You mention the MIAME reporting algorithm in both your presentation and in one of your earlier responses. Is MIAME now the reporting method of choice for microarray investigations?
Seattle, Washington

Theodore E. Mifflin, PhD, DABCC: It certainly is strongly recommended. Some journals will not accept manuscripts now that contain microarray expression data that do not adhere to the MIAME standards. However, while these standards apply most to basic and applied research, I believe that there are a number of elements within these standards that can be adapted into the clinical laboratory for microarray standards.

Would you advise using cluster analyses in investigations of potentially co-regulated genes, or should we go with a more powerful, supervised method of analysis?
Gaithersburg, MD

Theodore E. Mifflin, PhD, DABCC: This is a question best answered by a bioinformatics person. It probably depends upon a number of factors such as the number of persons in the study, the number genes / targets being measured, the stastical power being sought, etc.

Thanks for a very nice presentation, Dr. Mifflin. What is the longest oligonucleotide probe that you've seen attached to a solid platform, and is this a consideration in the development of microarray devices for clinical use?
Boston, MA

Theodore E. Mifflin, PhD, DABCC: My understanding is that Agilent uses probes that are in the 60 - 70 nt length range for its spotted microarrays. I don'recall seeing longer probes being mentioned by anyone else. I don't see much added benefit by using longer probes as the costs to make probes longer does go up as the synthesis yield decreases. Certainly, shorter probes in the 25 -30 range are used more frequently (e.g., Affymetrix), especially for spotted probe on planar glass surfaces. This is also about the same length one finds on beads as well

With the power of the microarray it seems we hold the potential to diagnose or determine the probability of contracting life threatening diseases. In the future, could these tests be conducted at infancy and potentially used to provide early treatement options? If so, this information would be valuable to insurance companies. Is there any legislation being prepared to protect this vital personal information?
San Diego, Ca

Theodore E. Mifflin, PhD, DABCC: I think you have recognized a crucial problem that is not limited to microarray results but cuts across a wide spectrum of personal information, be it derived from diagnostic laboratory procedures or other sources. I don' see expression microarray results having the same impact as I do when large scale SNP (single nucleotide polymorphism) results are accumulated from a SNP screening chip. There is a recent review on genome-wide SNP genotyping (A-C Syvanen, Nature Genetics, May, 2005) that indicates several different platforms currently can measure 10,000 snps and Affymetrix has announced it will have a 100k SNP GeneChip ready soon followed by the 500K SNP chip. So it is possible that we could be screened for predispostion to a variety of chronic diseases at birth much like infants are now screened for mostly metabolic diseases by mandated pre-natal screening programs. Given the discovery power of monitoring 500,000 SNPs, its clear that some group of SNPs will probably be determined to be informative for some chronic diseases such as stroke, hypertension, diabetes, etc. A significant question then becomes who can have access to this information as screening for a known genetic disease (e,g, cystic fibrosis) may carry substantially different obligations than have a risk factor (and predisposition) to a chronic disease that may be influenced by lifestyle, diet, environment, etc. such as hypertension. I don't know whether the ADA covers this possibility and it may require that Congress have another look at this legislation.

Is it better to pool my RNA samples and do replicate hybridizations from the pooled sample, or to maintain separate samples and do hybridizations from these independent samples?
State College, PA

Theodore E. Mifflin, PhD, DABCC: My own observations are that its better to have multiple RNA samples and do one hybridization / RNA sample. Listed below is a good reference to read about experimental design for microarrays: • “Data Analysis Tools for DNA Microarrays”. S. DRÂGHICI. Boca Raton, Chapman Hall / CRC. 2003, 477 pg. – Include 2 CD’s with microarray processing software THere are other books now that have been published on design and analysis of microarray data, but this one is a good place to start. Certainly, I think 15 -20 minutes spent with your favorite bioinformatics resource person would also be time well spent as well. Remember, "Biology often lies" and so to get at the truth about the real values of gene expression, you are better served with multiple samples rather than replicate analysis of the sample or pooled samples. This is particularly true of you are interested in gene expression that is just beyond the cutoff of noise (usually between 0.5 X to 2X of baseline).

How many replicates should I run for non-clinical investigations, and how should they be done?
Miami, FL

Theodore E. Mifflin, PhD, DABCC: The number of replicates is one of those parameters that is usually best decided before you begin the study. It depends upon a number of factors such as the numer of samples you intend on analyzing, the number of chips / arrays you want / can afford, the statistical power that you have as a goal (p=0.05, vs 0.01, etc). I usually think that replicates should be run of samples and not microarrays. Say you have 90 arrays to work with in your study. Because of the inherent variability in biological samples, your power will be better if you look at 90 samples once vs 30 samples in triplicate (assuming that all of the samples were being treated the same and were derived from the same environment). It would make sense for you to talk with a biostatistician as soon as possible to determine the optimum number for your study to give the most robust results.

How do gene sequencing chips work?
Cambridge, UK

Theodore E. Mifflin, PhD, DABCC: At this time, I don't know of a chip that will give you a primary DNA sequence from an unknown sample. That is, I'm not aware of a chip that can be used to analyze a DNA segment by hybridization with a multiple fragments derived unknown sample, then recreate the original sample's DNA sequence by electronic analysis of the hybridization pattern. I would suggest you can look in a recent book by Mark Schena on microarrays to see how serial sequences are used for specific analysis of samples (e.g., detection of SNPs for example). • “Microarray Analysis”. M. Schena. Hoboken, WILEY-LISS. 2003, 627 pg.

Can you amplify on the sources of imprecision in the use of fluorescence as noted in slide 36?
Potomac, MD

Theodore E. Mifflin, PhD, DABCC: This is one of outstanding issues where the gene expression analysis could use some substantial and significant improvements. In my estimation, it is one of the primary reasons why differences in gene expression cannot be more prescisely measured, especially down near the threshold cutoffs (e.g., 0.5 < ratio of gene expression<2). There are a number of technical and chemical reasons for this imprecision: (1) optical covariance error -- the array scanner cannot measure the fluorescent signal at the precise same spot for both fluors, but instead 'slips' slightly and truncates part of one signal, (2) biochemical error -- the labeling reaction is slightly unequal in its efficiency at labeling the same gene product in both the reference and the test sample, (3) the admixture of labeled gene products is created slightly baised with more of one gene product than another. THere are others, but these highlight the problems, and (4) unequal normalization of fluorescence of fluorescent signals in both scans.

I just received a report from a microarray CRO that listed a gene as absent but noted significant levels of expression. What's going on? I haven't yet approached the company for an explanation, but do you have any ideas about what might have happened?
San Jose, CA

Theodore E. Mifflin, PhD, DABCC: Without knowing more about the microarray itself, I would think one possibility is that there is a transscription error in the listing of the genes spotted vs what was physically spotted on the microarray. I would suggest you ask your CRO to verify the spotting log with the list to ensure that what is on the list of gene sequences spotted is really what was spotted. It could be as simple as just a transposition error (two gene oligos/cDNAs in oppposite locations in the source microplate). Alternatively, there could be a spotting error, but the spotters in general are very reproducible. Another possibility is that a completely different oligo was placed into the source microplate before the microarray was spotted and so gave rise to hybridization signal from a completely different sequence. I would suggest that you ask your CRO to reexamine the spotting logs first to establish what was spotted at the position of the gene in questions, then verify if possible the identity of the oligos / cDNA products in the corresponding source microplates.

How important is data format when setting up a project with a contract group? Are the various software products generally compatible or should I be looking to use the same propritary product for reporting, analysis, etc.?
Sacramento, CA

Theodore E. Mifflin, PhD, DABCC: You have identified a significant issue in the field of microarray analysis: How do I set up my data so that I (and potentially colleagues and others) can realize the maximum benefit from it? It depends a lot on what your end goal is for data processing. Are you going to be doing just routine data analysis (e.g., background correction folowed by normalization such as LOWESS and then unsupervised analysis?) or something a lot more sophisticated? Are you interested in merging the expression data with other data sets to facilitate pathway analysis? Next, you should determine what platform are you going to be using (e.g., Affymetrix or spotted array or home-spotted array or other)? If it is Affymetrix, many of the commercial software packages have been developed to accept Affymetrix scanning files. If you are going to be using a spotted array instead such as Agilent's, then you should check with Agilent to see what software package their microarrray files can be read. If you are using self-spotted arrays, you need to find out which software packages can support unique file formatting. Willyou be creating files that others can use and so need forward compatibility (e.g., MIAME compliant?). What about Gene Onology formatting? Is that important? You may not get all of these features in less sophisticated data processing packages. But the scanner should give you files that can be used and transported without the need for significant manipulation. To answer your question in general, there is significant recongnition by the software vendors that platform specifications (e.g., Affymetrix or non-Affymetrix) are important, but whether you can mix and match the files afterward remains to be seen. I would spend some time on the phone to be sure about some of these issues before I purchased software. Some of these vendors might actually be able to send you a sample access to their program so you can 'try before you buy'.