• 
• 

Archive

April 2007

Gene microarray technology yields interesting results in the laboratory

Dr. Matt van de Rijn is a member of the LRG Research Team working to understand and overcome GIST treatment resistance. This is the fourth article in a series to be written by each of the key research team members.

In this issue of the newsletter I would like to describe the type of experiments we perform in my laboratory. Less than ten years ago, Pat Brown of Stanford University and others developed the “gene microarray technology.” I was fortunate enough to start collaborating with Dr. Brown on a number of projects at about the same time. Gene microarray technology allows researchers to look at the expression levels for essentially all human genes in a single specimen using an overnight test. This approach represents an immense advantage over prior techniques in which each gene had to be investigated one at a time, making a “genome-wide” analysis of gene expression levels practically impossible.

Using gene microarray technology, one can therefore (Figure 1) determine the level of message or messenger RNA for each of the approximately 30,000 human genes in a single experiment. In this way one creates extremely large datasets. For example if one would analyze 30 samples of GISTs, one would generate a dataset of 900,000 datapoints. Clearly analysis of such large datasets requires an entirely new set of tools for the researchers.

At Stanford, we are extremely fortunate that the Stanford Microarray Database (SMD, http://genomewww5. stanford.edu/) has developed software that allows researchers such as myself, who do not have an in-depth knowledge of statistical methods, to nevertheless interrogate these enormous datasets and discover interesting gene expression profiles. An example of this is the DOG1 marker (stands for “discovered on GIST 1”). We found high levels of messenger RNA for this protein in all GIST specimens that we analyzed on gene microarrays but in none of the other samples of other tumors that we were examining, such as synovial sarcoma or leiomyosarcoma.

We therefore realized that this DOG1 marker could be a potential novel diagnostic marker for GIST and could be helpful in the diagnosis of these tumors. The identification of this new marker was then verified by using “tissue microarrays,” using immunohistochemistry or in situ hybridization to detect the DOG1 protein and DOG1 messenger RNA respectively. This has been a repeating theme in our work. We use gene arrays to examine one sample at a time but for many markers (30,000 genes) and then, through biostatistical analysis, pick out one or two interesting genes to further study by immunohistochemistry or in situ hybridization on tissue microarrays.

In the principle of tissue microarray (TMA), as developed by Dr. Olli Kallioniemi and his colleagues, a simple instrument removes a core of a paraffin block containing a tumor sample and positions this core in a pre-drilled hole in an empty paraffin block. By repeating this process one can position up to 500 cores taken from 500 different tumor samples in a single TMA.

Essentially, what you would see are cross-sections of the cylinders of tumor tissue that have been placed in neat rows and columns. We keep track of which tumor is represented by which core. A section of such a tissue microarray brings a collection of tumor specimens together on a single microscopic glass slide. This can then be stained for protein expression by an antibody or can be used for the detection of messenger RNA for a particular gene by in situ hybridization.

Using such a TMA, we were able to show that a conventional antiserum, raised by injecting a peptide derived from the DOG1 DNA sequence into rabbits, was able to recognize many GISTs that failed to stain for the KIT marker. Approximately 10 to 15 percent of GISTs do not react for KIT using immunohistochemistry and many of those GISTs that failed to stain with KIT antibodies reacted for DOG1 antiserum.

Unfortunately, the rabbit only yielded a small amount of antiserum before it died. We therefore decided to make a monoclonal antibody against the DOG1 protein, in collaboration with a laboratory of Mike Cleary of Stanford University. In this technique, mice are immunized with a DOG1 protein fragment that was generated in a test tube. The immunized mice are sacrificed and their spleen cells or Blymphocytes are fused to a myeloma cell line. The myeloma cell line provides the Blymphocytes with the ability to be cultured in vitro. Next, we tested approximately one thousand “supernatants” (the culture medium of a single clone from the myeloma/ lymphocyte fusion experiment) for reactivity with the same DOG1 peptide that was used for immunization. Of the one thousand wells that we tested, by a technique we call ELISA, sixty showed reactivity. In Figure 2, panel A shows the appearance of such an ELISA test. The faint yellow color represents the presence of an antibody that binds to the DOG1 peptide. Sixty of such wells were identified and the supernatants from those wells were then used by immunohistochemistry on a small tissue microarray shown in panel B. The five cores that stained brown were samples of five different GIST specimens, while the two cores at the right top-hand corner were taken from leiomyosarcoma samples. The lower right hand core is a background control core taken from normal human placenta. Of the sixty antibodies thus screened, two showed promising staining patterns and were further subcloned and retested. Ultimately, we ended up with a monoclonal antibody that we called DOG1.1.

In subsequent studies, we have now further characterized this monoclonal antibody on a number of TMAs. One of these TMAs was made by Drs. Chris Corless and Mike Heinrich and contains a large number of GIST cases for which they determined the mutation status for the KIT and PDGFRa genes. We have found that the mouse monoclonal antibody DOG1.1 is superior to the original DOG1 rabbit antiserum in that it shows a very high specific reactivity for GIST. It also fails to react with the vast majority of other sarcomas and also carcinomas, melanomas and seminomas. The latter is an important point because KIT antibodies can react with a small number of carcinomas and at a greater frequency with melanomas and seminomas. Thus, the DOG1 monoclonal antibody is an example of how we can use global or full genome screening to identify a single gene that may be clinically relevant. We hope that this antibody will show its usefulness in the clinic in the future (Iñigo Espinosa et al. – manuscript submitted).

One frustrating aspect of performing gene microarray analysis is that, until recently, we were obligated to use freshfrozen tumor tissue. While in theory it appears simple to save a small fragment of a tumor that is surgically resected from a patient and store it in a freezer, this often does not happen for a variety of reasons. As a result, fresh-frozen tissue always has been and probably will remain difficult to obtain.

In the last year we have become familiar with a new type of gene microarray that is different from the prior type based on cDNA spot fragments spotted on glass. The new type, called “HEEBO” arrays (http:// www.microarray.org/sfgf/ heebo.do) uses small, seventy nucleotide-long fragments of DNA as probes for each gene. In contrast to the older cDNA arrays, this array type lends itself to using messenger RNA that is derived from formalin-fixed paraffinembedded material (FFPE). In Figure 3, an example of such an array is seen and the wellformed red, green and yellow dots each represent a different gene that is sampled in this test. A red signal indicates the presence of a large amount of messenger RNA for a particular gene whereas a green signal represents a very low level of messenger RNA.

To date, we have had good results with this type of analysis although formal proof of the efficacy of this method is still some months away. Figure 4 shows an example of the analysis that we performed. In the columns labeled “Benchmark 18”, we have shown the top genes that distinguish six samples of each of the three tumors tested in this experiment. The samples are located in the columns while the rows represent the genes. A red color indicates that a gene is relatively highly expressed in a sample. The three tumors used for this test were: DTF (desmoid-type fibromatosis), SFT (solitary fibrous tumor), and GIST. For each tumor, we asked the microarray database to yield one hundred genes that were exclusively expressed at high levels in each tumor and an additional one hundred genes that were expressed at very low levels in the tumor. Compare for example the top one hundred positive GISTs versus the top one hundred negative GIST genes in column labeled G under Benchmark 18.

We then performed two separate analyses on six tumors (two each of DTF, SFT and GIST) that were different tumors from the ones that were analyzed in the Benchmark 18 group. For each of these six tumors, analysis was performed on messenger RNA isolated from formalin- fixed, paraffin-embedded tissue (FFPE) and from frozen tissue (FS). The signals for each of the six one hundred positive or one hundred negative genesets are quite similar in the FFPE and FS columns and are quite similar to the genes expression levels seen in the “benchmark” set of tumors. Future experiments that are currently being performed, will show the degree with which this technique can yield new results.

In collaboration with Dr. Chris Corless, we are currently studying a set of “wild type GISTs.” These are GISTs without a known mutation in the KIT or the PDGFRa gene. These wild type GISTs are extremely rare and very few have available fresh-frozen tissue. We therefore hope that these studies using paraffin-embedded tissues will yield data that are otherwise unobtainable.

Add Comment