Researchers from George Mason University have completed an analysis of samples and data from the National Cancer Institute's Cancer Genome Atlas (TCGA) project that indicates that laser capture microdissection of tumor samples could improve proteomic analyses.
Their findings, described in a paper published last month in Cancer Research, suggest that LCM, which allows for enhanced enrichment of cancer cells from tumor samples, could help researchers better detect cancer-related proteomic and phosphoproteomic signatures that in non-dissected samples might be masked by protein expression in the surrounding non-cancerous tumor microenvironment.
Using reverse phase protein arrays to analyze LCM and non-LCM glioblastoma tumors from a TCGA study set, the GMU team found significant differences in the protein profiles detected in the LCM and non-LCM material. They also examined a proteomic dataset generated by TCGA researchers using non-LCM glioblastoma tumors, finding, they wrote, a lack of expected correlations between these samples' genomic and proteomic profiles.
According to GMU researcher Emanuel Petricoin, author on the study, the results indicate that the proteomic data generated through the TCGA project is likely incomplete and inaccurate in some respects.
"We're not by any means saying that all the protein data in TCGA is wrong or inaccurate, but what we are saying is that some of it is likely inaccurate and that [it] is missing [signatures] because you are not really able to drill down [without using LCM]," he told ProteoMonitor.
Gordon Mills, chair of systems biology at the University of Texas MD Anderson Cancer Center, and leader of the TCGA's proteomics efforts questioned the Cancer Research findings, however, noting that a number of the correlations that the researchers failed to observe in the non-LCM data are, in fact, apparent in the larger TCGA dataset.
LCM uses a laser coupled to a microscope to allow pathologists to isolate particular portions of a sample of interest with high precision. GMU researcher Lance Liotta, also an author on the Cancer Research paper, is one of the inventors of the method and is on a number of patents covering the technology. The authors declared no conflicts of interest; however, Liotta and his co-inventors do receive royalties on a license for the technique that is currently held by Life Technologies.
Petricoin and Liotta, who are co-chairs of GMU's Center for Applied Proteomics and Molecular Medicine, have long promoted the technique for preparation of samples for proteomic analysis, arguing that LCM enables better identification of protein and phosphoprotein signals by eliminating background from non-cancer cells typically present in tumor samples.
Proteomic analysis without LCM "is like mapping the topology of the earth with a low-resolution instrument that is only able to pick up the highest mountain tops," Petricoin said. He noted that in the past, he and his colleagues have published analyses finding differences between proteomic data from LCM and non-LCM tumor samples. However, he noted, while those studies identified differences between the two sample types, they didn't speak to which offered a more accurate picture of the tumor proteome.
The Cancer Research study, on the other hand, used genomically characterized TCGA samples, allowing the researchers to investigate whether proteomic data from analyses of LCM or non-LCM samples best correlated with this genomic data.
The researchers investigated 39 glioblastoma samples taken from tissue previously analyzed by the TCGA project. Using RPPA they measured the levels of 133 proteins and phosphoproteins, comparing LCM and non-LCM samples, finding differences in 44 percent of the analytes between the two types.
They followed this broad analysis by looking in more depth at the genomic and proteomic data for epidermal growth factor receptor (EGFR) and phosphatase and tensin homolog (PTEN), two clinically important proteins in glioblastoma.
Examining these two proteins, the researchers found several instances in which the non-LCM proteomic data failed to exhibit the behavior expected based on the underlying genomic data. For instance, Petricoin noted, while they observed in both sample types increased EGFR protein and phosphoprotein levels in patients with increased EGFR gene copy number, they observed the increase in EGFR phosphorylation expected in carriers of EGFR mutations only in the LCM samples.
In the case of PTEN, the GMU team observed the expected decrease in PTEN levels in tumors with deep loss of PTEN or PTEN mutations only in the LCM samples. Additionally, only in the LCM samples did they find the expected correlation between EGRF phosphorylation, PTEN levels, and phosphorylation of AKT, which is regulated by the former two proteins.
Petricoin and his colleagues also examined proteomic glioblastoma data previously generated by TCGA using non-LCM samples, again failing to observe the expected correlation between PTEN copy number or mutational status and PTEN protein levels.
Mills', whose lab generated this data along with RPPA data on thousands of additional cancer samples from TCGA and elsewhere, called into question the GMU findings, however, noting that in some cases the correlations they failed to observe were, in fact, present in the larger TCGA dataset, and suggesting that the relatively small sample sizes used in the Cancer Research study might contribute to the discrepancies.
In an email to ProteoMonitor, Mills noted, for instance, that, while the GMU researchers did not find the expected correlation between PTEN mutational status and AKT phosphorylation in the 39 non-LCM samples they analyzed, a strong correlation is present between PTEN mutational status and phosphorylated AKT levels in the larger 215 non-LCM sample TCGA glioblastoma dataset. The expected correlation is also present in the case of PTEN copy number, he said, noting that the functional consequences of PTEN loss are therefore readily detectable.
The Cancer Research study did, Mills said, correctly point out that detecting loss of PTEN at the protein level is challenging in part due to heterogeneity, as has been previously reported. Further, he noted, as the study indicated, LCM has the potential to increase the ability to detect loss of proteins in tumor cells, particularly where the protein is exposed at high levels in stroma.
However, Mills added with regard to the PTEN data that the PTEN R130 mutation, which he said is the most common recurrent PTEN mutation in glioma, has been shown to cause not protein loss but protein stabilization. Given that, he said, PTEN mutation might not lead to PTEN protein loss to the extent expected by the GMU team. Indeed, he said, across the complete TCGA sample set of over 2500 patients, PTEN mutation is strongly associated with PTEN tumor loss.
With regard to the EGFR data, Mills again noted that while the GMU researchers failed to detect the expected relationship between EGFR mutation and EGFR phosphorylation in their analysis of non-LCM samples, this relationship is apparent in the larger TCGA dataset. And, indeed, in the GMU team's analysis of the existing non-LCM sample TCGA glioblastoma dataset, the researchers did find the expected correlations between EGFR mutation and copy number and EGFR phosphorylation.
Mills further noted that while the Cancer Research authors said that the TCGA glioblastoma dataset they used consisted of 123 samples with known mutation status, in fact the full set contains 215 samples, all with known mutation status.
Mills agreed that in cases where sample tumor content is low, or where the signal is strong in the stromal cells, molecular signals could be masked, and that in such cases LCM could be "theoretically a positive." Applying the technique to all samples, however, would require significant resources in terms of money and time and also has the potential to introduce artifacts, he said.