Name: Cristobal Belda-Iniesta
Position: Physician and researcher at University Hospital La Paz, Madrid, Spain
Cristobal Belda-Iniesta is an oncologist and researcher at Madrid's University Hospital La Paz, where he heads the institution's Biomarkers and Experimental Therapeutics for Cancer Group.
In a paper published in the December edition of the Journal of Proteome Research, he argued that, despite advances in biomarker research and the emergence of molecularly targeted drugs, physicians are still plagued by a lack of good information as they try to select therapies for their cancer patients.
In the paper, Belda-Iniesta called for an international effort to gather proteomic data on cancer patients worldwide and to compile it in a database that would serve as a source of information oncologists could consult when determining therapy. Three years ago he started such an effort at University Hospital La Paz, collecting proteomic data on all of the institution's cancer patients. Currently, the hospital has roughly 10,000 blood samples from more than 2,000 people.
He spoke recently to ProteoMonitor about this work and about what the broad collection of proteomic data could do to improve physicians' cancer treatment options. The following is an edited version of the interview.
In your opinion, right now the methods oncologists have for identifying cancer types are inadequate?
I think that all medical oncologists understand this problem very well. When we attend to patients we have to decide what therapy, what drug, we will use to treat this patient. This is decided by the pathology, and the pathology [analysis] focuses on the histological diagnosis – [for instance,] 'In my microscope I see that [the tumor] cells are rounded' or 'They look like a square.' So we use a lot of drugs, well-designed drugs, for patients where all we know is the shape of the [tumor] cell, no more – and probably the shape of the cell is not enough to decide what therapy [to use] against the cancer. So there's a big gap between our drugs and the actual data that we use to decide what drug to use for what patient.
What sort of information is needed for doctors to make better decisions?
I think that we have to classify patients based on their possibility to respond to current therapies. Right now we have 10, 20, 30 drugs that our patients will receive this year, next year. All these efforts to classify cancer [by molecular profile] are very exciting; however, the first thing we need is to classify cancer patients based on their ability to respond to current cancer therapies. The question is very easy. What is the drug, what is the patient, and when do I have to give this drug to this patient?
Rather than focus on classifying patients by the molecular signature of their cancer, for instance, you're more interested in classifying them simply on how they respond to existing therapies?
The number one thing is: What is the best way to treat this patient? This is a very practical question, but this is a question that needs to be answered by clinicians and scientists. The relevant thing for a patient is — I have cisplatin, or erlotinib, or gefitinib — what is the genomic, proteomic classification of a cancer that responds to this drug? If you identify the pathway, the critical kinases, for instance, this is very useful, and I agree with all these approaches and probably all clinical oncologists agree with this research. However, the question is the same: What is the effectiveness of [existing] tyrosine kinase inhibitors in our patients?
Why has it been so difficult to translate protein biomarker research into information that's helpful to oncologists in guiding therapy?
Every day, every week, hundreds of groups identify a potential biomarker. However these biomarkers are based on these groups' areas of expertise. They identify the proteins they are looking for. If you perform a statistical analysis and you have a lot of patients, you'll probably be able to find evidence for whatever you want. This is very similar to many epidemiological studies. You identify several traits in a cancer population and you can correlate the cancer with the use of whatever. I think we need to look at all proteins in a cancer patient, and then probably we can identify real biomarkers in subsets of patients. And probably the population of cancer patients will be divided into thousands of subsets.
You'd anticipate that there are many more varieties of cancer than are commonly talked about?
Sure, there are probably hundreds of types of lung cancer, [for instance]. Nowadays we have four types of lung cancer. But this isn't true – there are probably hundreds of types of lung cancer.
How do you compile and share proteomic data specific to potentially thousands of kinds of cancer? It seems like a tremendous undertaking.
It's very difficult to coordinate. For example, if we have 100 different types of lung cancers, you would never have enough patients for a clinical trial with a concrete subset if you don't collaborate internationally. So it is very difficult. However, we have to do this for our patients. We have to work together to provide to our basic scientists all the clinical data that we have, and they have to provide us all the biological, molecular data they have.
We probably need to observe the history of AIDS [treatment]. [AIDS researchers] designed an international database where the genotype of HIV strains, the protein data of HIV [strains], and the clinical response of [HIV strains] to specific therapies were included. So you have the data for your patient diagnosed with HIV and the data on what are the most efficient drugs [for a given patient]. And as the database added more and more information, the quality of the data available to clinicians became better. Probably we have to follow these steps, the steps that our colleagues treating HIV started more than 20 years ago.
What sort of databases are you building with information from your patients?
We are building a database of proteomic data using SELDI-TOF [mass spec] and MALDI-TOF [mass spec]. We have data from more than 2,000 patients. I don't know the exact number of proteins [we have data on] because we're working with SELDI-TOF, so we're observing the spectrometry profiles and comparing the spectrometry profiles of those who respond to specific therapies.
We take [blood] samples when the patient is diagnosed and after any therapy. And we collect [blood] samples when we perform clinical evaluations, histological evaluations of the tumor in order to coordinate the proteomic data with the histological data and clinical data. So we have the exact spectrometry when we perform the CT scan or the PET scan, and so you can evaluate the clinical responses to specific therapies. Currently we have more than 10,000 patient blood samples.
How long have you been collecting this information?
For three years.
Do you plan to continue collecting it indefinitely?
Sure. This is our standard approach at our hospital, and it will be introduced into our routine clinical practice.
Are you using this data to guide therapy yet?
The main goal of this approach is to develop the proteomic data to predict a response to specific therapies. Probably in one or two years we'll begin using this data for that purpose.
Have you begun distributing this patient data to other researchers and physicians?
Not yet, but we will publish this data in a few months — seven to 12 months, depending on the journal — and when we publish it we will share our database with the [medical and scientific] community.
What does proteomic information offer that genomic data – which has been used by oncologists for some time now [with tests like Genomic Health's OncoType DX] – doesn't provide?
We're not able to identify the genomic profile of a microscopic metastasis [for instance]. First of all because you don't know that a patient has a microscopic metastasis. And if you perform a genomic approach only on the primary tumor, it's very hard to assume that the microscopic metastasis has the same genomic profile as the primary tumor – this is a hypothesis that has to be evaluated. Proteomic [data] from blood samples will provide the possibility to identify proteins that are supplied from one metastasis, two metastases, five metastases, so I think that approach will probably be of interest in clinical practice.
But this is not a competition – 'what is the best technology genomic or proteomic?' A combination [of genomic and proteomic approaches] will be used. However, the protein is the actor, so probably the best point [to focus on] is the protein activity. That's the real situation in the patient.
In addition to helping physicians choose therapies, you also mentioned in the Journal of Proteome Research paper that proteomic data could be useful in improving drug trials for cancer drugs. How do you envision proteomic data being used in that regard?
I think that we need to identify which patients will be the most likely to respond [to a given drug], and we have to then test this drug only in this set of patients — not in all patients. Drug trials are [currently] designed using patients that have no possibility of improving under existing therapies. So if you have 20 to 25 patients in a Phase I clinical trial, they have a wide range of cancers. Three with colon cancer, three with breast cancer, one patient with sarcoma, one patient with glioblastoma, and you're testing this drug that targets a concrete molecule [in all of these patients]. If you have from the pre-clinical data the possibility to identify the correct patients that could respond to this drug, why do you test this drug in a wide range of tumors?
It's very hard to modify the design of Phase I trials, of course, because we've been doing them in this way for more than 40 or 50 years. However, I think that companies must design Phase I trials to test drugs only in patients that the pre-clinical data suggests could respond. I think proteomics will help in this. If we are able to identify the target within the tumor with different proteomic approaches, we could modify our [trial design] to identify patients that [will possibly] respond to therapy. I think that's probably the best approach.
Have topics you'd like to see covered in ProteoMonitor? Contact the editor at abonislawski [at] genomeweb [.] com.