Blasting into the Microworld of Tissue Proteomics: A New Window on Cancer : Commentary re S. A. Schwartz et al., Protein Profiling in Brain Tumors Using Mass Spectrometry: Feasibility of a New Technique for the Analysis of Protein Expression. Clin. Cancer Res., 10: 981–987, 2004.

At a functional level, cancer is a proteomic disease (1). Although DNA is the information archive, proteins do all of the work of the cell. A pathogenetic defect is selected out during somatic genetic progression because the defect ultimately alters the protein network to generate a selective advantage for the cancer cell. Deranged, hyperactive, or dominating signal pathways may drive cancer survival, growth, invasion, and metastasis (2, 3, 4). Pathogenic alternations in protein networks extend outside the cancer cell to the tissue microenvironment, in which exchange of cytokines, enzymes, enzyme inhibitors, motility factors, and adhesion molecules act to the advantage of the tumor cell (5). Consequently, critical protein nodes within the key signal pathway circuitry constitute the therapeutic targets of the future (1). Proteins closely reflect the ongoing local pathophysiology of the tissue and thereby become candidates for diagnostic and prognostic biomarkers. On the basis of the promise of new diagnostic and therapeutic strategies and of the proximity to the drug target and/or diagnostic biomarkers, the field of cancer proteomics is currently undergoing a tremendous influx of investigative talent as well as institutional and corporate interest.

The biggest challenge in cancer proteomics is the complexity of the tissue microenvironment (5). Proteins coalesce into networks and circuits on command from external and internal stimuli. As stimuli fluctuate, and as feedback loops return information, newly formed protein networks rapidly break apart or are actively degraded. Thus, the cellular proteome is a dynamic entity, not a fixed list. The population state of protein networks is changing from minute to minute within each living cell (4, 6, 7), and is completely dependent on the context of the local cellular microecology. Cancer is a product of the tissue microenvironment. Proteomic analysis of cultured cell lines does not correlate with microdissected tissue cells from the same patient (8). This underscores the need for new microtechnologies that can analyze the state of the proteome in small cellular samples procured from patient specimens. Previously, because of a lack of sensitivity, proteomic technologies were not applied to human tissue specimens. In the past, protein analytical methods including one-dimensional gel electrophoresis, two-dimensional gel electrophoresis, and mass spectroscopy were limited by the need for large amounts of input material and complex sample preparation (9, 10, 11, 12). To overcome these hurdles, researchers have been attempting to develop new technologies, so that proteomics can be applied to the microscopic world of clinical biopsy tissue samples.

At this early stage of the field, the basic questions are 2-fold: (a) can proteomic alternations in cancer cells be reliably measured in tissue?; and (b) do the derived proteomic signatures correlate with biological state, clinical behavior, or therapeutic response? Investigators are using two types of proteomic technology to mine the telltale proteomic signatures in tissue: (a) mass spectrometry (MS) pattern profiling; and (b) protein microarrays (Fig. 1). The technologies are highly complementary because they use opposite starting assumptions. The MS pattern method assumes no a priori knowledge about the identity of the underlying biomolecules. In the current iteration of the MS method, the pattern of ion peaks itself (without knowledge of the underlying identities of the ionized proteins) can discriminate disease states and appears to contain prognostic information (13). In contrast to MS pattern analysis, protein microarrays start with a limited, but specific, knowledge of the identities of important signaling molecules in the cancer cell. With protein microarrays it is possible to measure the activated (e.g., phosphorylated or cleaved) form of key signal proteins and thereby estimate whether the designated signal node is “in use” at the time the proteins are extracted from the cell (Refs. 4, 14, 15; Fig. 1).

The first attempts to develop a MS ion spectral pattern from microdissected human cancer tissue cells, described by Paweletz et al. (16), involved three preparation steps: (a) laser capture microdissection to procure a pure cell population; (b) solubilization of cellular proteins; and (c) drying the protein lysate onto a prepared surface for surface-enhanced laser desorption ionization time-of-flight (SELDI-TOF) spectrometry. In a vacuum chamber, the dried protein mixture was then literally blasted with a laser beam. This ionizes and desorbs proteins in the low molecular mass range (less than 50,000 Daltons). The charged ions were analyzed in a mass spectrometer to produce a “bar code” spectra of charged species representing the resolved masses of the population of ionized proteins launched by the laser. The laser capture microdissection lysate method generates a MS spectral pattern using less than a thousand microdissected cells, with as few as 25 cells, and the pattern of ion amplitudes in the spectra is different, depending on the tumor histopathology. This technique was used to distinguish early-stage premalignancy from advanced disease (16).

The exciting alternative to laser capture microdissection lysate MS is “direct-tissue” MS (17, 18). This method bypasses the need for microdissection and solubilization of the cellular proteins. Instead, the frozen tissue section is dried onto the matrix-assisted laser/desorption ionization plate, coated with the energy-absorbing ionization matrix, and then blasted with a laser in a vacuum chamber. The mass spectral pattern is developed directly from the cut face of the tissue section. (Fig. 1). Caprioli’s team has previously used this method to develop MS patterns correlating with lung cancer disease stage (19). The tissue MS spectra were reported to classify lung tumors according to lymph node metastasis with 85% accuracy. In the current issue of Clinical Cancer Research, they have now extended this method to human brain tumor tissue sections (20). Schwartz et al. (20) conducted direct-tissue MS on 20 prospectively collected snap-frozen normal and brain tumor specimens. The MS spectral patterns rendered from the tissue, as reported by the investigators, “can reliably identify glial neoplasms of similar histological grade, and differentiate them from nontumor brain tissues.” Bioinformatic cluster analysis algorithms were applied to the spectra data and were able to group tumor and nontumor tissues into separate clusters.

Whereas the speed and ease of direct-tissue MS for brain cancer is appealing, at this stage of development (20), these tools do not provide new information the pathologist doesn’t already know from looking at the brain histological section directly. The number of proteins currently detected by this method is in the range of 200–400 high-abundance proteins, only a tiny fraction of the tens to potentially hundreds of thousands of distinct cellular protein isoforms capable of being queried. Analysis of the 99.99% of the other proteins within the tissue and the important posttranslational modifications (e.g., phosphorylation or cleavage) that drive signaling is outside the current detection limits afforded by tissue MS. Nevertheless, the tissue MS patterns appear to provide reproducible information correlating with histopathology and disease stage [for lung cancer (19)]. Thus, an important feasibility milestone has been reached in the field of tissue MS proteomics. Potentially, these proteomic changes observed in the tissue could give rise to cascades of biomarkers that percolate into the surrounding interstitium and into the circulation. Of course, for clinical utility, one would want these biomarkers to be measurable in a readily accessible body fluid so as not to require a tissue biopsy for diagnosis. Mass spectral pattern diagnostics for cancer and disease detection using serum, plasma, urine, and other body fluids, first described for early-stage ovarian cancer detection (13), seems to be gaining enormous momentum as evidenced from the success in other disease states (21, 22, 23, 24, 25, 26).

In the future, as MS technology becomes ever more sensitive and accurate, it should be possible to obtain protein identity information directly from the MS ion spectra. Instead of just a pattern, the MS output will directly yield a list of differentially expressed protein identities. When this stage of the technology is reached, investigators in the field of tissue proteomics will reap a rich harvest of diagnostic biomarkers and therapeutic target candidates.

The ultimate payoff of clinical proteomics will be to go beyond molecular discovery to reach individualized therapy. Just as patients visiting an oncologist’s office are distinct individuals, with no two being exactly alike, it is likely that their respective tumors have unique molecular circuitry even though they may be given the same pathological description. Molecular profiling (genetic and proteomic) will replace a histological diagnosis with the equivalent of a circuit diagram describing the individual tumor in the context of the originating tissue. A functional map within that patient’s tumor cells will become the starting point for individualized therapy. Under this scenario, therapy can be tailored to the individual tumor’s molecular defect. The same signal pathway portrait becomes a platform that can serve multiple functions: as a diagnostic tool, as a direct window into multiple potential drug targets, and as a means to assess the efficacy of the drug, or combination of drugs, on the entire pathway. A panel of specific inhibitors acting at multiple points along the length of a signal pathway could effectively shut down the pathway. Theoretically, combination therapy of this type could achieve higher efficacy with lower toxicity.