Urologic Oncology

SLIDES & TRANSCRIPTS
Friday, December 5, 2003

Proteomics: Translational Tools For Clinical Applications

Emanuel Petricoin, Ph.D.

Slide 1:

Thank you very much. I am going to turn our attention to the protein side of things. Translational tools for clinical applications, our FDA-NCI clinical proteomics program is focused on developing technologies for bedside applications and trying to understand both the promises and the warts of the technology as they exist and starting off I think Sir William Osler was right a very long time ago and it still is germane today, “If it were not for the great variability among individuals medicine might actually be a science instead of an art.”

Right now it is an art because of the heterogeneity really of the disease and the patients themselves. Trying to understand biomarkers and discovering biomarkers is very complex because specificity is always the killer. We see this over and over again passing Kleenex box after Kleenex box across the table at the FDA to promising biomarkers that have failed and held up in larger validation trials.

Slide 2:

The goals of our program are to develop new tools for early detection of disease. So, we came up with a technology several years ago, proteomic pattern diagnostics which utilizes a drop of blood and looks at the complicated pattern of hundreds of millions of signatures perhaps of proteins and peptides in a drop of blood and using artificial intelligence type computer programs to sift through that information to detect diagnostic signatures, applying proteomic information basically to move from correlation to causality, developing molecular profiling tools, developing circuit profiling tools and pathways analysis tools from clinical samples.

The goal is to develop a circuit diagram, a wiring diagram at the proteomic level since these are the drug targets. GPCRs for example make up about 80 percent of all that we see at the FDA. These are protein-based therapeutics. So, what is happening at the proteome is something we are very interested in looking at within the agency.

Personalization of therapy, everyone is talking about it. How do you actually find and treat patients effectively? Protein based therapeutics may be developed in combination so that you get an increased efficacy and decreased toxicity if you knew beforehand the interconnections that exist within the circuitry of the cell.

Slide 3:

So, very briefly over the next 6 minutes I am going to talk to you about the types of technologies that we have been developing both at the tissue level using new types of novel protein microarrays that can look at hundreds of phosphorylation events specifically from as few as 5000 cells from a fine needle biopsy specimen. We are doing this in the clinic in clinical trials today at the CCR and using proteomic pattern diagnostics to detect disease earlier.

Slide 4:

So, this concept initiated several years ago and developed through a CRADA with Corelogic Systems uses high throughput high reproducible mass spectrometry signatures to generate a proteomic image in high dimensional space and using pattern recognition tools to develop a signature that can distinguish cancer from healthy or cancer from benign.

The initial type of approach utilized a low resolution mass spectrometer

Slide 5:

that had a very high throughput based application using liquid robotic handling systems and protein chip surfaces for on-chip chromatographic separation of the serum proteome

Slide 6:

and then utilizing these types of pattern recognition tools to develop a diagnostic fingerprint by looking and culling through hundreds of billions of combinations of patterns generated from these approaches.

Slide 7:

So, along with David Ornstein who is now out in California we published a paper in JNCI where we looked not for determining case versus control in prostate cancer but actually to determine the sensitivity and specificity of a new type of approach in the diagnostic indeterminate range.

So, we looked to see if we could be sensitive for the cancer which we could, 95 percent sensitivity. We also had a very high specificity in the indeterminate range of PSA. So, PSA has zero percent specificity here. Everyone is getting a biopsy and this is in a fairly large blinded set of samples.

We switched our technology over to more high resolution technology because of the machine-to-machine reproducibility issues that would hamper potential clinical applications of low-resolution instrument. This was more of an approach to determine can we build a plane to fly. Now, we are attempting to see if we can start taking this cross country.

Slide 8:

So, the low-resolution cell that gives you something like this on a high-resolution cell would be looking at the exact same spot on a Ciphergen chip looks something like this

Slide 9:

but importantly if you look over months on the same machine using the same serum sample you see tremendous reproducibility in the ion species at the exact same molecular weight and this is very important as we go to the clinic.

Slide 10:

We do a lot of up front clinical evaluation and quality control of the spectra before we even model it. We do a lot of reference standard control charting. We check the process of about 30 things simultaneously during the runs and during this time. Then we can develop acceptance criteria, failure criteria, basically develop release specifications and process controls before we do our modeling.

Slide 11:

[No text is associated with this slide]

Slide 12:

In work with David Ornstein and Walter Rayford in a high-resolution QSTAR mass spectrometer in work with them we got serum from men with PSA in the 2.5 to 15 nanogram per ml range.

Slide 13:

Men had biopsies of at least 10 cores and then we went to mass spec analysis, developing a training set of 40 men with 26 cancer biopsies, 14 at least 2 negative biopsies in the training set and then in 136 men that were held back in a blinded testing set that either had one negative biopsy or a smaller percentage that had two negative biopsies

Slide 14:

we were able to accurately classify 34 to 35 of the men in the blinded set for 97 percent sensitivity, 55 percent specificity with one negative biopsy and 16 out of 18 men with two or more negative biopsies correctly identified.

Slide 15:

Results of this study were preventing potentially in theory 66 percent of unnecessary biopsies, 89 percent in this study set if the two times prior negative biopsies were obtained while missing a small percentage of the cancers.

Slide 16:

The conclusion of the study is that perhaps we can in this diagnostic gray zone range have a specific pattern of low-molecular weight fragments of serum peptides and proteins that through pattern recognition can uncover predicted biomarker patterns.

Slide 17:

Where do these patterns come from? What are these ions? Everyone asks these questions at every meeting I go to.

Slide 18:

We are trying to figure this out, and one of the newest advances we have made is we found that all these diagnostic signatures that we are seeing on SELDI is basically bound to carrier proteins such as albumin and fibronectin. We can now take advantage of this.

This is the enrichment you get when you look at what is bound to albumin versus alone.

Slide 19:

You can take advantage of this now by developing nano harvesting agents. We are doing this with Mauro Ferrari now where these biomarker proteins perhaps coming from the tumor-host interface could be even host proteins themselves. They are specifically clipped, liberating through the endothelial cell wall into the circulation and then being picked up by these nano harvesting agents in the future and right now you have endogenous circulating carrier molecules such as albumin.

Slide 20:

We can now take these nanoparticles themselves, incubate them in serum and through very high resolution Fourier transform type mass spectrometry get an exact mass tag of each one of these peptides.

Perhaps in fact in the very near future we will be able to look at hundreds of millions of low-molecular-weight species and know their exact identity, in fact creating patterns faster than you could perform the trial.

Slide 21:

So, low-molecular-weight fragments and peptides derived from proteins secreted from every tissue and cellular compartment have been found bound to albumin. Through work with Tim Veenstra and Tom Conrads we have identified over 2000 of these to date bound to albumin and fibronectin. Most of these proteins exist bound to carrier proteins and are what people are seeing on the low-molecular-weight SELDI analysis. These constitute a pattern of information perhaps concerning the pathophysiology of tissues from a sample of blood.

Slide 22:

Then lastly we are looking at protein array type of technology to be able to see once you find a cancer how do you effectively treat it and target the therapy at the protein level.

Slide 23:

So, unlike gene array analysis we can use these protein-like arrays which can be using small molecules themselves, antibodies or recombinant proteins or phage or aptimers all of the sort of planar array can then bind proteins and then you detect these either by a tagging system or by direct analysis by mass spec or atomic force microscopy.

Slide 24:

We use laser capture microdissectin that was invented by my co-director and colleague Lance Liotta here at the NCI to procure cells directly from tissue biopsies.

Slide 25:

These tissue biopsy specimens then are arrayed

Slide 26:

on a miniature dilution curve on these protein arrays. You will hear more about this from Robert Grubb who is in Marston Linehan's lab, that collaborated with ours that is looking at prostate cancer signal pathway profiling.

Slide 27:

Because each patient sample is arrayed in a miniature dilution curve we can now probe with different phospho-specific antibodies, correct the slopes and be able to get disparatized analysis of the state of the signaling circuitry from very small numbers of cells by looking at kinase substrates

Slide 28:

and this can also be multiplexed using two-color dye. This is work we are doing with Licor. We can look at say ERK and phosphoERK on the same array

Slide 29:

and this way multiplex the analysis very rapidly.

Slide 30:

[No text is associated with this slide]

Slide 31:

This is work we are doing with Elise Kohn for ovarian cancer looking at a Gleevec trial to microdissect the biopsies before, during and after therapy

Slide 32:

and then array those on an array and look at AKT activation as you go from before herceptin, after herceptin,, after Taxol and look at the changes in the signaling circuitry before, during and after therapy in an isogenic background. We can now find for example phospho-AKT levels in these patients actually are higher after they recur, something that Genentech is very interested in.

Slide 33:

Ultimately one can go from a protein array based profile all the way to the same type of heat maps that you see in a gene array

Slide 34:

except these are for example from a breast cancer study set we submitted to Science several weeks ago.

Now, we are looking at red as an increase in protein phosphorylation, green as a decrease in phosphorylation, looking at true signal pathway circuitry now, not gene expression analysis. We are not inferring what is happening at the protein level. We are actually measuring it

Slide 35:

and since these are the drug products themselves we can now envision a future where from a biopsy specimen, from a phosphoproteomic analysis we can get a diagnostic profile that actually is a drug target itself, tailor the therapy and have a surrogate for efficacy monitoring during that time period.

Slide 36:

Combining this information now knowing the interconnections of the cellular wiring diagram

Slide 37:

perhaps we can develop combinatorial therapeutics that can take us to greater efficacy in the future.

Slide 38:

So, the greatest obstacle to discovery is not ignorance. It is the illusion of knowledge and so we should not kid ourselves about thinking we know so much about what is going on in these disease. Some of these omic-based approaches are interesting because they can go in and drill into a system without apriori knowledge and perhaps teach us something at the same time.

Slide 39:

I certainly would like to acknowledge work with Dr. David Orenstein and Walter Rayford, Marston Linehan and Robert Grubb and his group as well as my collaborator and co-director, Lance Liotta , support from Dr. Carl Barrett and Jesse Goodman at the FDA.

Thank you very much.