William Check, PhD
A user group coalesced in 2008 around a new technology that was then called chromosomal microarrays. Composed of 25 laboratories that were using those arrays for cancer diagnosis and prognosis, the group adopted current terminology and called itself the Cancer Cytogenomic Microarray Consortium, or CCMC. Jill Hagenkord, MD, one of the founding members, says few clinical laboratories at that time had experience with the arrays in multiple cancer or specimen types. “Only a few labs offered these arrays for cancer diagnostics,” she said in a presentation at last year’s meeting of the Association for Molecular Pathology. That presentation was about the group’s first study, called CCMC-Quality Control, a multicenter, cross-platform clinical validation of cancer cytogenomic arrays.
Dr. Hagenkord, who was at Creighton University when the CCMC was formed and who moved to Complete Genomics during the course of the work, told CAP TODAY in an interview last month at this year’s AMP meeting that cytogenomic arrays have “huge advantages” over using only conventional cytogenetics and fluorescence in situ hybridization, or FISH. In contrast to conventional cytogenetics, many cytogenomic arrays work well on formalin-fixed paraffin-embedded tissue. And “often you want to get a view across the whole genome, which is not possible with FISH,” she says. In addition, data from cytogenomic arrays are digital, which makes them powerful. “You can get more precise breakpoints and can easily interface with genome browsers and online databases,” explains Dr. Hagenkord, who recently became senior VP and medical director for the startup InVitae. The greater precision isn’t surprising; arrays can provide information at the gene or exon level, while conventional cytogenetics has a resolution of 5–10 megabases (MB). Other advantages include the ability to detect acquired uniparental disomy, which is now known to be clinically significant in many cancers (Hagenkord JM, et al. J Mol Diagn. 2010;12: 184–196; Maciejewski JP, et al. Br J Haematol. 2009;146:479–488).
The specific aims of the study were twofold, says Federico Monzon, MD, director of molecular pathology in the Cancer Genetics Laboratory at Baylor College of Medicine. First, to compare the performance of cytogenomic arrays versus gold standard assays in a range of specimen and tumor types. Second, to assess the cross-platform reproducibility of the commercially available arrays in clinical use. Dr. Hagenkord sums up the results: “The cytogenomic array results were highly concordant with standard assays, and the labs using different arrays all gave the same diagnostic line.”
Arrays from Affymetrix, Agilent, and Illumina were used in the study. All arrays incorporate single nucleotide polymorphisms, or SNPs, throughout the entire genome. “Vendors were very generous in sponsoring this work and providing arrays,” Dr. Hagenkord notes. Four types of samples were evaluated: fresh peripheral blood for diagnosis of chronic lymphocytic leukemia (CLL); fresh bone marrow for myelodysplastic syndrome (MDS); fresh solid tumor for renal epithelial tumors; and formalin-fixed paraffin-embedded solid tumor also for renal epithelial tumors. Comparison methods (gold standard) were a FISH panel for CLL, conventional karyotype for MDS, and morphology for renal tumors. Thirty samples of each tissue type were provided for evaluation by The Methodist Hospital (renal cell carcinoma; RCC), University of Pittsburgh Medical Center (CLL), and GenPath (CLL and MDS).
Use of microarrays in CLL and MDS is now becoming more common, and they can be applied also to RCC (Monzon FA, et al. Arch Pathol Lab Med. 2009;133:1917–1922), particularly in morphologically challenging or unclassified renal neoplasms (Hagenkord JM, et al. Cancer Genet. 2011;204:285–297). Cytogenomic arrays are also being used to resolve “double-indeterminate” HER2 status in breast cancer (Aaron M, et al. Arch Pathol Lab Med. 2011; 135:544–557) and in leukemias, lymphomas, brain tumors, sarcomas, neuroblastomas, and Wilms tumors, as well as benign neoplasms (Dougherty MJ, et al. Cancer Genet. 2012 Jan-Feb;205(1-2):42–54).
“One of the driving forces of using cytogenomic arrays is that fresh tissue is needed to run conventional cytogenetics,” Dr. Monzon says. “This is not always available, and the arrays can use FFPE material.” Archived tissue can be used for testing with cytogenomic arrays, he adds.
Ten members of the consortium, selected on the basis of their experience and proficiency with one of the array platforms, formed the study group. (Attempts were made to engage other vendors and commercial laboratories.) The CCMC committee members and principal investigators involved in the project, in addition to Drs. Hagenkord and Monzon, were Jackie Biegel, PhD (CHOP), Daynna Wolff, PhD (MUSC), Brynn Levy, PhD (Columbia University), Gokce Toruner, MD, PhD (UMDNJ), Marilyn Li, MD (Tulane/Baylor), Michael Rossi, PhD (Emory), Jennifer Laffin, PhD (University of Wisconsin), and Patricia Miron, PhD (UMass). Laboratories at Creighton and Columbia were Affymetrix sites; Tulane/Baylor and the University of Medicine and Dentistry of New Jersey were Agilent sites; and Children’s Hospital of Philadelphia and the Medical University of South Carolina tested with Illumina arrays. Each site used the same array for all 120 tissue samples, with one exception: Only Affymetrix and Agilent arrays were used in the FFPE renal cell carcinoma arm. Columbia University and the University of Massachusetts Memorial Hospital reviewed and distributed test specimens. Investigators at Emory and Baylor are doing the post-testing analysis.
For each sample type, each array was assessed by three criteria: comparison of cytogenomic array results to the gold standard; intralaboratory reproducibility on duplicate samples; and interlaboratory reproducibility between the two laboratories using each method. In addition, reproducibility of results across all three platforms was evaluated. Intralaboratory reproducibility was rated excellent for all three arrays across tissue types. Agreement between laboratories using the same array was also good.
For CLL, the loci tested were 11q22.3, Chr 12, 13q14, and 17p, which indicate prognosis ranging from “good” to “poor.” Discrepancies consisted of two cases with 13q14 loss detected only on FISH and one case with 17p loss seen only on array. Deletion of 17p is the strongest adverse prognostic factor in CLL, says Dr. Hagenkord, so this is an example of the advantages that cytogenomic arrays can provide. That FISH caught anomalies not seen on arrays is not surprising, Dr. Monzon says, because FISH can identify copy number changes at the 10 percent level, which is more sensitive than arrays whose analytical sensitivity is about 20 percent. Dr. Hagenkord, who reported these data in the AMP session, noted one sample with low-level trisomy 12 (seven percent by FISH) that was called by only three of six laboratories using arrays—one laboratory using each array platform called it. The discrepancies arose due to different laboratory calling thresholds rather than technical failure of any one commercial array. This case highlights the need for standardization of calling thresholds in laboratories using cancer cytogenomic arrays, says Dr. Hagenkord, but provides reassurance in the technical performance of the arrays.
A final feature of microarrays in CLL to which Dr. Hagenkord called attention was genetic complexity, which she describes as “the second most common cytogenomic marker” in this hematopoietic malignancy. As many as 21 percent of CLL cases have genomic lesions detected on cytogenomic arrays that are not tested for on FISH panels (Gunn SR, et al. J Mol Diagn. 2008;10:442–451; Hagenkord JM, et al. J Mol Diagn. 2010;12:184–196). Since it is not uncommon for CLL cells to fail to grow in culture for conventional cytogenetic analysis, oncologists often have only FISH panel results to represent the genetic prognostic information for patient management. “I have seen cases where the FISH panel suggested the CLL would behave indolently, but because we had done the cytogenomic array and detected genomewide genetic complexity, we were able to provide this information to the oncologist to help determine whether to ‘watch and wait’ for this patient’s CLL or to initiate therapy,” she says.
Based on Vysis’ Food and Drug Administration submission for its FISH assay in CLL, this technique has a kappa for interlaboratory reproducibility of 0.86. “Microarrays match or exceed the interlab concordance of FISH,” Dr. Monzon said.
For MDS, the clinical utility of genetic testing is to: 1) assess clonality and help rule out other non-neoplastic causes of dysplasia, and 2) identify pre-leukemic MDS that may advance to acute myeloid leukemia as part of the International Prognostic Scoring System (IPSS). The CCMC study assessed the five loci included in the IPSS that are typically assessed by conventional cytogenetics: -5q/5q-, -7q/7q-, +8, 11q, and 20q-.
In three cases, lesions were seen on microarrays that were not seen on conventional cytogenetics: trisomy 8, 11q duplication, and a chromosome 7 with uniparental disomy (UPD). Dr. Hagenkord says this is another example of the added information that cytogenomic arrays can provide. The standard techniques used to assess the genetics of MDS cannot detect the UPD7, which is biologically equivalent to a deletion of chromosome 7. The presence of this lesion not only confirms the clonality of this sample, but classifies the patient as poor prognosis in IPSS. Several publications show the clinical utility of detecting UPD in MDS and the diagnostic yield of adding SNP-based cytogenomic arrays. For example, Tiu, et al., showed that combined metaphase cytogenomics/SNP-karyotyping lead to higher diagnostic yield of chromosomal defects (74 percent versus 44 percent, P < .0001) in MDS, compared with metaphase cytogenetics alone (Tiu RV, et al. Blood. 2011; 117:4552–4560; Gondek LP, et al. Blood. 2008; 111:1534–1542). “The take-away from these papers is that each technique has its strengths and limitations, and it’s our responsibility as pathologists to understand them and triage diagnostic testing accordingly,” Dr. Hagenkord says.
Dr. Monzon noted that one platform was a bit more discrepant at the level of data interpretation. On the other hand, he said, raw microarray data agreed between the two laboratories using that array, independent of whether they called specimens abnormal. Dr. Monzon sees this as an artifact of the two laboratories having different thresholds for reporting copy number alterations. In this sample, the genetic abnormality occurred in a low percentage of tumor cells or a low percentage of cells in the sample, so the signal was dampened. “Whether the laboratory called the abnormality depended on how aggressive they were in calling a genetic change,” he told CAP TODAY.
For CLL and MDS, overall concordance with the gold standard (FISH and conventional cytogenetics, respectively) was very high—greater than 90 percent for both tumor types. Interpretive agreement across platforms also exceeded 90 percent for these two types of specimens.
In renal epithelial tumors, the goal was to classify the tumor by tissue type. For this form of cancer, microarrays can provide a “virtual karyotype,” identifying typical genetic -lesions for each histological type: deletion of 3p in conventional clear cell carcinoma, trisomy 7 and 17 in papillary renal cell carcinoma, multiple monosomies in chromophobe renal cell carcinoma, and a nearly diploid genome in benign oncocytomas.
Microarrays had a somewhat lower concordance with the gold standard, histology, in RCC than in CLL or MDS. Correlation was in the 90 percent range on frozen tissue and lower for microarrays on FFPE samples. However, this result does not necessarily reflect decreased microarray performance because some cases might be classified one way morphologically but have a genetic profile that suggests a different RCC subtype. For instance, one specimen was called clear cell RCC on histology and chromophobe RCC on array. However, Dr. Monzon showed that reconsideration of morphology supported the call of chromophobe RCC. “So cytogenomic arrays can be helpful in cases when morphology is not that clear,” he says.
There was also an increased number of “unknown” calls on formalin-fixed paraffin-embedded material with microarrays, largely because histology was not available to the laboratories testing the arrays. This situation was “a bit contrived,” Dr. Monzon points out. “In the clinic the histologic examination and cytogenomic arrays work together.”
Correlation among the platforms was very high for detecting genetic changes in fresh frozen RCC samples. Concordance was also good for distinguishing RCC tissue type when arrays’ performance was compared between fresh frozen and FFPE specimens. (The same specimens, prepared by different methods, were used in these two arms of the study.)
Summarizing these results, Dr. Monzon said the arrays showed excellent performance and reproducibility in all six laboratories for identifying altered chromosome copy numbers. However, different laboratories have different interpretive thresholds. Some standardization of interpretation would be desirable. Dr. Monzon noted that this validation/QC study was the first effort of the CCMC and that a peer-reviewed manuscript and followup studies are forthcoming.
Dr. Hagenkord told the AMP audience that the results of this study have already had an effect on vendors, prompting Affymetrix to incorporate the CCMC gene list into its new cytogenomic array design and stimulating Agilent to market a “CCMC custom array.” Illumina’s SNP microarray already had 99+ percent coverage.
Dr. Hagenkord, whose new employer, InVitae, will offer whole genome sequencing for preventive health purposes, predicts that sequencing will eventually replace cytogenomic microarrays in the cancer diagnostics field. “There are different views of how fast this will happen,” she says.
William Check is a writer in Ft. Lauderdale, Fla.