Name: Heinrich Roder
Position: CTO, Biodesix, 2006 to present
Background: CTO, Efeckta Technologies, 2000 to 2006; founder and CTO, Data Physics Research 2005 to present; postdoc fellow, Institute of Theoretical Physics, University of Hanover, Germany, 1986 to 1989
Heinrich Roderwas part of a team of scientists from Biodesix, Vanderbilt University, the Scientific Institute University, Hospital San Raffaele in Milan, Italy, and other institutions that recently tested the ability of a molecular diagnostic from Biodesix to determine which patients with non-small cell lung cancer would benefit from two tyrosine kinase inhibitors.
Biodesix’s diagnostic uses an algorithm to analyze data from matrix-assisted laser desorption mass spectrometry. For their research, the team collected patients’ sera before they had been treated with gefitinib, also known as Iressa, or erlotinib, marketed as Tarceva.
The researchers collected the sera from from two institutions and analyzed it with MALDI MS. They then developed an algorithm to predict treatment outcomes with the TKIs in a training set of 139 patients from three cohorts, two independent validation cohorts of 67 and 96 patients who were treated with the EGFR TKI, and 158 patients from three control cohorts.
The study, published in the Journal of the National Cancer Institute, can be found here.
Below is an edited version of an interview ProteoMonitor had with Roder about the study.
Tell me about your diagnostic and how it works.
Targeted therapies in lung cancer [have] opened a new door to the treatment of non-small cell lung cancer and [have become] an alternative to chemotherapy, which is a really rough treatment. The hope is that with this targeted therapy you can give alternative therapies that are better tolerated and lead to better quality of life, and hopefully improve chances of survival.
One example of this kind of targeted therapy are the tyrosine kinase inhibitors of the epidermal growth factor receptor. The drugs that are commonly available are Iressa from AstraZeneca and Tarceva [from OSI Pharmaceuticals, Genentech, and Roche Pharmaceuticals]. (In June 2005, AstraZeneca, based on discussions with the US Food and Drug Administration, updated Iressa’s label to reflect that it can only be prescribed to patients who are benefiting or have benefited from the it. Tarceva monotherapy is indicated for patients with locally advanced or metastatic non-small cell lung cancer who have failed at least one prior chemotherapy regimen— Ed.)
Now, the issue is that these drugs are, in an unselected population, only effective in a subset of the patient population. That leads to the effect that if you are trying efficacy trials, they are marginally improving overall survival. There’s also evidence that some of these patients are actually negatively affected by these drugs.
So the drugs are expensive and the whole task is to find out whom to give these drugs to, and to find whom they would benefit and whom they would adversely affect.
There have been problems since these drugs [hit] the market. Typically, initially, genomic tests [looking for] mutation stages were performed and all of them required fairly large tissue samples. Those are not easily obtained from patients with advanced lung cancer. You try to preserve most of the lung as much as you can.
Of course, it would be very nice if you had a less invasive test. Also, those tests typically are not specific for the drugs, but just give you an idea of the overall prognosis of that patient.
We, together with Vanderbilt University Medical Center, the University of Colorado Health Science Center, a group in Italy, and Japanese co-workers undertook the task to find a test that lets us detect responder stages, or [weed] out people who would benefit from these drugs from a simple blood test, serum or plasma.
Fortunately, we succeeded, and so what we have is something that uses mass spectrometry. It’s a simple mass-spec technique [that] requires about 10 microliters of plasma or serum [and] just the simplest of mass spec techniques, MALDI — just drop it on a plate, mix it with matrix, run it in a MALDI machine, out comes the spectrum.
The spectrum gets [put] into our software, it gets compared with a representative set of files that we’ve developed using training data, and out comes the label [that] says these people are likely to benefit from [these targeted treatments] or they are unlikely to benefit.
What we’ve done at Biodesix is develop analysis tools to make these mass spectra comparable. Attempts to do this have been done before. The most prominent or infamous ones were attempts to find early-detection tests for ovarian or prostate cancer. And they could never be confirmed. And there were technical difficulties, and they put some skepticism in the community.
We made a serious effort to validate the system. Everything was run independently in different labs, and we confirmed that we could reproduce it. And indeed, we found patients that potentially can benefit [from targeted treatment and] those who would not benefit.
The median survival [rate] in the people [who would benefit from targeted treatment] was over twice as long as those [who we didn’t think would benefit from such treatment]. And that was from very simple pre-treatment serum samples.
So, it’s a great benefit because, if this gets confirmed in prospective studies, you will get the right drugs to the patient, [and] you won’t treat the patients that won’t benefit. You won’t deprive them of alternative treatment.
Why did you choose the MALDI platform for the basis of your diagnostic?
A very important issue to us was reproducibility, and with LC-MS platforms, we have not been able to manage reproducibility across multiple labs. And so simplicity is important.
How long did it take you to develop this diagnostic?
We started this nearly two years ago. At Biodesix we’ve been working on analysis of mass spectra for a while and we’ve been given these spectra and we were setting out to do this comparison, so this was from scratch.
Did you come across any setbacks in your development?
Not really. Of course it’s always important [that] you get sufficient data sets and validation sets, and always the rate-limiting step using our techniques is finding sufficient data or sample sets of enough quality with sufficient patient data to validate what we are doing.
The limiting step is the availability of sufficiently characterized sample sets or clinical data sets.
What about the high dynamic range of blood? Was that an issue?
In general, one would worry about, in a simple untreated serum, what it is that you see. But if you see a difference, and you can validate it, it does not seem to matter. And of course, I’m sure there are other things where the dynamic range is important, but in this test, it does not seem important.
The next step is continued validation of the results?
We are currently in the planning phases of prospective trials, and there will be a fairly large phase III study in Europe [with the Scientific Institute Hospital San Raffaele] in Milan, and we’re in discussion with [the Eastern Cooperative Oncology Group] to start a prospective trial in the United States.
What’re you going to be focusing on?
All the trials are designed to test the efficacy of patient stratification using this proteomics classifier for patient selection and compare it to standard of care selection.
When you say trials, do you mean US Food and Drug Administration trials?
We’re currently in discussions with the FDA, and we certainly will plan to go forward and we want to work together with the FDA to, in the long run, obtain approval for this test. …
Certainly, we will have enough data even for a [premarket approval]. Of course, it might be possible to do this at [the] 510(k) de novo level.