Patient Outcomes Linked To Biomarker Levels

JonN writes to tell us Science Daily is reporting that researchers at Yale University have discovered that current pathology methods for biomarker detection can be dramatically altered depending on the concentration of antibodies used. From the article: "Biomarkers may have the power to provide diagnostic, therapeutic, and prognostic information for personalized medicine." said Donald Earl Henson, M.D., of the George Washington University Cancer Institute, in "Back to the Drawing Board on Immunohistochemistry and Predictive Factors," an accompanying editorial. "However, immunohistochemistry, a popular technique for evaluating biomarker expression, may contain procedural flaws that jeopardize its promise."

Comments

Computational Molecular Phenotyping

[BWJones]

I am not surprised as most immunohistochemical approaches to biomarkers are optimized for proteins that have notoriously variable levels depending upon sampling method and analytical method. Most basic scientists have known this for some time, and are very careful about interpretation of immunohistochemical results, but the medical field has been slow to pay attention.

As an outcome of our work in the visual system, we have been developing a new approach to biomarker analysis based upon quantitative small molecular molecular phenotyping called Computational Molecular Phenotyping (CMP) that is a much more sensitive and reliable assay for not just eyes, but just about any biological system. Small molecular signals are much more tightly regulated between subjects and even remarkably between species. CMP relies upon 1) quantitative immunohistochemistry 2) computational tools derived from methods originally developed by the CIA and NASA for remote sensing and 3) new technologies developed in-house to assist in the the data processing and analysis.

Applications are in not just in pathology such as histological analysis of oncological tissues, but also in drug development, pharmacology and basic science. Also, as an interesting aside, I have also looked not only at a variety of vertebrate and invertabrate organ systems, but I am also looking at plant tissues with these technologies and there are some very interesting results that could assist in agronomics and bioencryption.

Re:Computational Molecular Phenotyping

[jameyers14]

There seems to be great confusion over this topic, so I am going to chime in with a my attempt to clarify.

Biomarkers in a nutshell: Let's say we believe that there are 6 proteins of interest to some disease. If we find that you have higher than normal levels of proteins A,B, and C and have lower than normal levels of proteins X, Y, and Z then you have a good PROBABILITY of having the disease so we should follow up with imaging or other diagnostic studies. These 6 proteins would be called biomarkers.

Finding proteins that are suitable for use as biomarkers out of the entire human genome is where the challenge is. One of the most popular approaches these days is to use "Gene Chips" which can measure relative expression of virtually the entire human genome in one shot. One can then use a variety of classic AI algorithms against a training data set (patients who have the disease versus those who don't, or good vs poor outcome, etc) to try to look for combinations of genes that are predictive (biomarkers) of the disease or outcome. It's a very difficult problem to say the least.

Oh nooooes.

[Number44]

Coincidentally, I just accepted a job at a company making immunohistochemical staining equipment (Ventana Medical Systems) and all I can say is, look at the stock over the past 5 years and tell me this article's prediction holds water.

Ventana's been making the stuff that runs the staining process for a long time, and has done VERY well by it. Their results are outstanding and have proven to be good medicine!

Re:Oh nooooes.

[Comatose51]

I would not trust the stock market to be the expert on science. I work at a finance company. The traders and analysts study companies and their products quite a bit but they don't get any special information nor do they have any special insights. Most of those guys are not scientists (but a few are engineers). They're not much better than you and I (unless you ARE a scientist or more precisely medical doctor, etc.) at knowing if something will work nor not. Remember the dot-com bubble and how those companies were bidded up and later died? Traders and analysts are specialists in their field, but not all fields. This is why people like Buffet prefer traditional business that they can look at the bottom-line or people like Peter Lynch advocates trading in what you know and understand. Lynch, especially, points this out in his famous book, "One Up on Wall Street".

Re:hmmmm ....

[Anonymous Coward]

Currently, if someone has a disease, doctors use a variety of pathology techniques to characterize a disease. In the age of molecular medicine, doctors often do biopsies and determine the levels of various important proteins, or "biomarkers," made by the tumor. The level of these different proteins can be used for prognosis or treatment; for example, cancers with high levels of a protein called MMP9 tend to be metastatic and should be treated aggressively.

The problem comes when trying to measure the amount of protein. Most proteins are measured using immunohistochemistry, which just means that you "stain" the sample's proteins with antibodies specific for that protein. You then measure the amount of antibody through various methods; the antibodies are often attached to a fluorescent tag, and you measure the level of fluorescence and extrapolate the true protein concentration from that. You usually assume that the more antibody that binds, the more protein there is, and the two are related in a linear fashion. This is an important assumption.

Different pathologists use different concentrations of antibody. The article states that depending on what concentration you use, you can make completely opposite conclusions about the protein levels, and thus about the disease. Essentially, the flaw is that "there is a non-linear relationship between the antibody concentration and its target." In other words, adding a lot of antibody changes the way the antibody binds to the protein, which makes identifying the true protein amount much more difficult.

I hope that helps.