Skip to main content

Analytical Chemistry

Not So Fast on the PD-L1 Comparisons

We may have to rethink what we think about the recent trial in which Bristol-Myers Squibb’s Opdivo was ineffective in non-small cell lung cancer. The arguing has been about the status of the patients, specifically how much PD-L1 protein they were expressing. But writing in BioCentury, Stephen Hansen points out that comparing the various trials is difficult, and maybe even impossible:

The reaction to the CheckMate-026 data appeared to suggest some investors believe patients don’t benefit from PD-1inhibitor monotherapy when PD-L1 expression is as low as 5% ; however, Opdivo and Keytruda are being used with companion or complementary diagnostics in a setting that’s proving difficult to parse for companies and physicians alike.

Part of the challenge is that positive PD-L1 status has not consistently predicted patient benefit across cancer types, with PD-L1 negative patients also deriving benefit from anti-PD-1 mAbs, albeit typically in smaller numbers.

Compounding the issue, however, is that each of the five mAbs against PD-1 or PD-L1 in Phase III testing or beyond has its own assay to determine PD-L1 expression, with each assay possessing its own scoring algorithm and method of measuring cells.

Given these differences, according to one assay developer who spoke to BioCentury, it is unlikely that a generic cutoff for PD-L1 expression to predict patient benefit can be determined without a prospective trial involving all the therapies and assays.

Now that’s a mess! The article (for subscribers) has a great deal more detail on this problem, but my first thought is “Welcome to immunoassays”, because that’s where this variation is coming from. Everyone has different antibodies and different protocols, and unless you run the same samples through the different assays, there’s really no way to tell how things correlate. Or if they do at all.

The one bright spot is the Blueprint Project, which is trying to resolve these questions. Most of the companies involved in this area are part of it, along with several diagnostics companies, societies like the AACR, agencies like the FDA, etc. A great big notable absence from the group, though, is the Merck KGaA/Pfizer partnership, who are going their own way. Some results from the effort are already coming in, and they mostly highlight the need for more results.

So it’s early, very early, to try to draw head-to-head comparisons between Opdivo and Keytruda (Merck’s antibody), or the other players in this field, based on their individual clinical trial data. Everyone’s brining their own facts, which are valid only in their own trials. A separate question is if you’re a physician, and you use some particular PD-L1 assay (such as Merck’s): can you then turn around and give either their antibody or Opdivo? You’d sort of think so, but “sort of think so” is not a very good rationale on which to base one’s oncology practice. Not even the Blueprint folks are trying to go all the way to figuring out clinical equivalence, so that may be an open question for some time to come. . .

21 comments on “Not So Fast on the PD-L1 Comparisons”

  1. Diver dude says:

    Surely “you’d sort of think so” is the current basis of all of medicine? We’re trying to change it but that’s pretty much where we stand right now.

  2. Barry says:

    Richard Pazdur (head of FDA Oncology) acknowledged 11yrs ago that the FDA’s definition of cancers by tissue of origin is unhelpful, and a new system defining cancers by the driving biology might be better. But that can’t happen until the biology is clearer. Historically, many diseases that had been lumped together only got recognized as distinct when one responded to a new therapy and one didn’t (e.g. Diabetes type I responded to insulin, Diabetes type II did not). But that’s all backward-looking. We need to be able to run Clinical Phase-II studies in e.g. PD-1-driven cancers regardless of whether they manifest in breast, or lung, or prostate, or brain…and that can only happen when that diagnosis is more robust

    1. Mac says:

      Quite right, Barry. It’s encouraging to see that more basket trials are being conducted to further this pursuit. Hey, maybe one day it will be out with “ovarian cancer” and in with “BRCA-mutated solid tumors locoregionally related to the ovaries” or what have you.

      Some nice graphs on basket trials are linked to this post.

      1. K Foley says:

        Isn’t the FDA pushing back on the over-proliferation of basket trials?

  3. Imaging guy says:

    I haven’t seen anyone checking whether therapeutic antibodies actually bind to all or most the cells expressing PD1 or PDL1 antigens, i.e. biodistribution into solid tumors. Maybe it is the reason some patients respond and some not. Of course it would need labelling therapeutic antibodies with biotin or fluorescent markers as all of these antibodies are of human origin.

  4. SP says:

    Expression assays would be more quantitative (not relying on an Ab) but don’t account for translational and post-translational modification, so which is a better predictor?

  5. steve says:

    The problem is that expression is not a static phenomenon. One can easily see the possibility that a tumor that starts out with low expression will up-regulate PD-L1 in the face of attack by cytotoxic T cells. So some patients might score low on their initial biopsy but their tumors increase expression during the course of treatment. This doesn’t even take into account microheterogeneities, subpopulations of metastases with different expression levels, etc. Biology is dynamic and we often get into trouble by thinking that a single snapshot in time tells us useful information.

    1. Abde says:

      Agree … I wanted to add that given the relatively fast epitope degradation in the case of PD1/PDL1 targeting, patients who provided older biopsies will have less PDL1 expression detected… They’ll be excluded from the treatment, while their real expression of PDL1 might be much higher at the time of screening. Many patients are not fit for a fresh biopsy and thus for a real time assessment of their PDL1 expression!

    2. Barry says:

      a patient whose tumor is PD-L1 negative (if we ignore the parallel CTLA-4 pathway!) is a patient who doesn’t have cytotoxic T-cells targeting his/her tumor. PD-L1 blocking antibodies don’t induce the cytotoxic T-cell response; they merely disinhibit what has been frustrated.

  6. MikeB. says:

    When we will finally learn that epitopes aren’t *just* proteins and that life is vastly more complex than just DNA and protein? Virtually 100% of all cell surface proteins are glycosylated, and lo and behold, tons and tons and tons of studies have shown how changing the sugar coating on cell surface proteins can highly regulate their immunogenecity, half-life, or even expression. One of the first hallmarks ever discovered in cancer was abnormal glycosylation on the cell surface. It’s really amazing how little people have tried developing an antibody that recognizes both the protein portion AND the sugar portion of a cancer associated epitope.

    1. Me says:

      Yup! Add to that protein-protein interactions, membrane lipidomics, various epigenetic states, ubiquitin processing and any other manner of stuff we only vaguely have a clue about (and I have none) and you have the clusterf*ck that says *we don’t know anything at all!! HEEELLLLPPPPP!!!* (An image of the scream painting flashes before my eyes)…

  7. Ran Friedman says:

    “PD-L1 negative patients also deriving benefit from anti-PD-1 mAbs”. I can’t see how. Is this just the sensitivity of the essay?

    1. steve says:

      See my post above. The definition of PDL-1 negative is based on a biopsy of a single tumor at a single point in time. It doesn’t tell you anything about what’s going on at multiple metastatic sites throughout the body over the time period during which the patient is being treated.

      1. Ran Friedman says:

        Thanks for the explanation Steve. In that case, it’s possible in theory to develop better experiments, but this may be difficult in practice with patients.

  8. Chrispy says:

    I guess it’d be cost prohibitive, but if you really wanted to shut down the pathway couldn’t you use BOTH an anti-PD-1 and an anti-PD-L1 simultaneously? Unfortunately, these trials are set up not so much to test a scientific hypothesis as to generate a data package to push a product through the FDA…

    1. Barry says:

      what’s going forward (to my knowledge) is dual blockade of CTLA-4 and PD-1 rather than of PD-1L and PD-1

      1. Maia says:

        durvalumab (MEDI4736) + MEDI0680 = anti-PD-L1 + anti-PD-1

        A Phase 1/2, Open-label Study to Evaluate the Safety and Antitumor Activity of MEDI0680 (AMP-514) in Combination With MEDI4736 and MEDI0680 Monotherapy in Subjects With Select Advanced Malignancies

  9. PD-1 affectionado says:

    I can state with confidence born of first-hand experience that the Roche PD-L1 diagnostic antibody is garbage. And now this:

    How could it be that efficacy of a anti-PD-L1 mAb is independent of presence or absence of its target? There’s the dynamic aspect of target expression previously mentioned in the comments, but I’d bet the many false negatives surely being generated by Roche’s crappy diagnostic antibody to also play a major role.

    Better to be lucky than good!

  10. Dennis Biroscak says:

    The efficacy of the anti-PD-1 MAb could depend on other factors – individual differences in antigen processing and presentation, interferon pathways activated in the immune response, glycosylation, etc. As stated above, it’s a complex process with many actors.

    Another way of looking at ‘predicting’ overall response is to look at the initial (6 week)response to PD-1/PD-L1 blockade. In the Roche-Genentec study cited by PD-1 affectionado, (, overall response was 73% if there was a measureable metabolic response at the 6 week interval. This correlates with another study that found similar overall response based on measuring response shortly after initial therapy:

    A different approach might be to simply measure response at the 6 week interval (perhaps using FDG-PET scan) and concentrate on non-responders with alternate therapy such as VEGF inhibitor combined with immunotherapy.

  11. Anon says:

    The article cited here by Derek is throwing scientific FUD at a very well-established and accepted truth: with a good diagnostic and good biopsy, you can measure PDL1 just fine. BMY should definitely have included a pre-defined sub-study with patients with PDL1 levels identical to the keytruda study. That study would have been positive, then they’d be on equal footing with Merck. The larger study would have been negative, and we’d still know what we know. BMY dropped the ball in this regard, and wall street is punishing them appropriately. The drop is stock price here is justified not by the lost sales in first line lung cancer, but by the fact that management screwed up bad — and they’re all still in charge. So, BMY is likely to continue making bad calls like this until there is a change at the top.

Comments are closed.