Saturday, March 16, 2013

Molecular Prognosis in DLBCL

There is a new test for patients with diffuse large B cell lymphoma (DLBCL) that has the potential to alter the way we manage the disease.  R-CHOP is very good treatment for approximately 2/3 of patients with DLBCL who are cured - but that still leaves quite a bit of room for improvement.  What we really need is a way to identify the patients who are high risk right from the outset.  That will enable us to concentrate our efforts to improve R-CHOP in the high risk group - and perhaps even consider ways to make therapy less intense for the low risk patients (in clinical trials only for now).

As you probably know from my prior posts - there are a bunch of different types of lymphoma.  My post about telling them all apart has been the most frequently visited page on my blog.

What makes lymphoma interesting to manage is that a two cases of a specific disease like DLBCL that may look identical under the microscope can be very different biologically.  Those differences may have consequences for treatment, prognosis, etc.  We are now in an era where we can begin to measure those biologic differences and figure out what to do about it.

In the past DLBCL was DLBCL was DLBCL.  If you looked into the microscope and saw larger cells (centroblasts) and they infiltrated the lymph node in a "diffuse" pattern and applied a few extremely simple tests to prove it was of B cell origin - the diagnosis was DLBCL.  We really didn't understand that cases had fundamental biologic differences and they were pretty much treated all the same.  You gave them all CHOP (rituxan wasn't around yet) and about half of patients did well and half didn't.

Understandably docs and patients want to be able to predict how an individual patient will do right from the outset.  Researchers looked at large data sets and realized several "clinical variables" (i.e. easy lab tests or patient characteristics) could segregate patients into risk categories.  By looking at a patients age, stage, ldh (blood test), patient functional status (ECOG score), and the number of places where the lymphoma affected the body outside of the lymph nodes - you could calculate a score called the international prognostic index (IPI).  After rituxan improved things - the IPI was revised - giving the R-IPI which can be calculated online.

IPI and the R-IPI were good for their day, but in 2000 a new technology called "Gene expression profiling" was applied to a large group of DLBCL samples.  A buddy of mine from Stanford named Ash Alizadeh (the smartest I've ever known) published a paper in the journal Nature that helped identify two subgroups of DLBCL called "germinal center DLBCL" (GCB) or "activated B cell DLBCL" (ABC).

This technology measures a differential quantities of a specific type of RNA called mRNA.  mRNA is called "messenger" RNA because it is synthesized in the nucleus of the cell by being copied off the DNA template.  It then goes to the cytoplasm where it serves as a "messenger"for the protein synthesis machinery.  Different types of lymphoma need to make different proteins.  By measuring the mRNA you can tell these lymphomas apart.

One of the cool aspects of this technology is that you can simultaneously measure THOUSANDS of mRNA's at once.  You can let the data sort itself out into patterns (unsupervised analysis) or you can ask questions of the data (supervised analysis).  One thing to shake out of letting the patterns sort themselves out was that there are two main types of DLBCL - the Activated B Cell (ABC) and the Germinal Center B Cell (GCB) subtypes.  Understanding that there are two main types of DLBCL has been a big advancement in the disease.  We are now learning that underneath this differential expression are unique mutations that probably cause the diseases in the first place.

The main categories of DLBCL to emerge from this was the Germinal Center B Cell  / Activated B cell subtypes.  The GCB subtype do a lot better than the ABC subtype.  Since those patients with ABC generally do not do as well, there have already been a number of clinical trials focusing on that subgroup specifically (such as the addition of velcade to R-CHOP for this group).

The were two big headaches with this technology that prevented it from becoming a standard test.  The first is that it has required a chunk of tissue that was taken directly from the patient and put into the freezer instead of being immersed in paraffin which is the standard in most pathology labs.  The only "frozen" samples were pretty much the domain of major research centers (these specialized freezers are expensive).  That meant that the vast majority of patients who had a biopsy to get the diagnosis might have to have a second biopsy if you wanted to get the expression profile.  The second problem is that the "chips" that enabled you to measure thousands of genes at once were expensive.  Those barriers essentially kept this knowledge out of routine practice.

There have been a number of efforts to substitute a commonly used technology called "immunohistochemistry" which is pretty inexpensive and readily available.  Unfortunately several groups have developed different testing criteria, it can be hard at times to get labs to agree on the results etc.  Even with these limitations though, more and more people are getting this testing done as part of their work-up.

In an effort to make the expression profiling technology more applicable to clinical samples, the Ron Levy laboratory (where I proved I have no business working in the laboratory) went back to the ABC / GCB expression profiling data sets and asked the question - which markers best predicted prognosis - both good and bad (supervised analysis).  They pulled out six genes that helped determine which patients had high or low risk.

Unfortunately these samples came from the pre-rituxan era and it was unclear if the gene would hold their value in the rituxan era.  Researchers then went and rounded up a bunch of clinical samples from patients treated with either CHOP or R-CHOP.  They found that a few of the markers were not as good but two main markers stood out in their ability to predict outcome - those were LMO2 (which helped segregate GCB/ABC pretty well and TNFRSF9 (aka CD137 or 4-1BB) which served as a marker of the surrounding immune system.  By focusing on just two markers - you could run the test without using an entire gene expression chip (making it a lot less expensive).  Furthermore, they were able to use a newer technology to extract the mRNA from samples embedded in paraffin.  This allowed access to the test for routine clinical samples.  The results come from a mathematical formula that converts expression into a score and spits out high / intermediate / low risk.

Finally - a test that can be run on routine clinical samples, that is relatively easy, and reflects tumor biology instead of just clinical variables.  The next question was what happens to IPI.  Fortunately the results enhanced each other instead of just replicating one another.  They found that you could combine the IPI and the biologic score to further refine the prediction model.

So where does this go from here?

Right now, a lot of frontline DLBCL studies require certain IPI in order to enroll (commonly an IPI score of 3 or higher)  I think this test could be integrated into clinical trials to enlarge the population of eligible patients.  It is also useful for the newly diagnosed patients who wants to understand their prognosis better.  Right now, outside of a clinical trial, I am not totally sure what to do for the patient who has a high score.  Do I give them R-CHOP or do I try to intensify their therapy.  At this point, I am unclear what alternative therapy to choose - perhaps R-EPOCH?

Anyhow, I think this is a step forward in DLBCL.  Hopefully we will see more data sets that validate the test.  It can be ordered today.  I am optimistic that this will take a lot of science developed over the last 13 years and adjust our management of DLBCL - particularly for that group that isn't cured with R-CHOP.

Thanks for reading.

(Disclosure- I worked in Ron Levy's lab and I personally know Izidore Lossos - I think they are great researchers so I tend to believe their conclusions - I have no financial relationship pertaining to this test to disclose)