Skip to Content


The Next Immuno-oncology Frontier?

Here’s some strong evidence for what could be the next wave of immuno-oncology: combining a TLR9 ligand with an OX40 antibody. We’re all going to have to get familiar with that sort of talk, so here’s what’s going on:

As those who have worked in inflammation or infectious disease know well, the TLRs (Toll-like receptors) are key sensors in the innate immune system. (As is often the case, especially for proteins of a certain era, the “Toll” nomenclature comes from the Drosophila fruit-fly model where they were first identified, and the human form was named as the analog of that one. hERG is another famous example). They’re found in large part out on the so-called “sentinel cells” (such as macrophages), and the ten TLR subtypes recognize common pathogen motifs. They’re tuned, in other words, to a list of common default settings. For example, TLR5 recognizes bacterial flagellin, which we don’t have, and other TLRs pick up on bacterial lipopolysaccharides or viral nucleic acid varieties. TLR9 recognizes the “CpG” nucleic acid motif that’s found far more in bacteria than it is in higher organisms.

OX40, for its part, is a sort of second-wind system for activated T cells. It’s not even expressed on the surface of the resting cells, but kicks in after 24 hours or so. Its ligand, OX40L, is similar – the antigen-presenting cells only start expressing after a delay as well. T cells start dying off after a couple of days of activation unless they get juiced back up by OX40 activation (the receptor is also known as CD134). I think of it as a “Keep cranking, this is a real one” signal – it’s an immune checkpoint to keep things from going full blast too often. As important as it is that the immune system activates when it should, it’s just as important that it stand down when it should, too (as witness autoimmune disorders, anaphylactic shock, cytokine storms et very much al.)

What this new paper (from the Levy group at Stanford) is doing is injecting tumors with CpG, for starters. That sets off the local CD4 T cell population, and the authors can then see these cells start to express OX40 on their surfaces. This only happens after the addition of exogenous CpG; it doesn’t take place in the normal tumor environment. If you take tumor-infiltrating T cells out in vitro and expose them to CpG, nothing happens on the OX40 front, and nothing happens even inside the tumor if you deplete macrophages and dendritic cells. So it’s these “sentinel cells” that are picking up the presence of CpG and communicating it to the T cells, apparently through IL-12, interferon-gamma, and TNF-alpha or some combination of those (determined through a whole set of other experiments.

So the next step was to do the CpG injection and follow it up with an OX40 agonist to set off those newly expressed receptors. There’s an antibody that binds to OX40 as an agonist, fortunately, and the effects were dramatic. In implanted tumor models in mice, injecting CpG alone showed strong effects at the site of tumor injection, but only there, not at distal tumors. Injecting the OX40 agonist antibody alone showed some effect at all tumor sites, but not enough to be useful. But the combination of the two showed complete regression of both the injected and noninjected tumors. The local CpG injection site went down first, followed after a few days by the distal tumors, consistent with the time it takes to raise a response in both cases.

Only 3 of the 90 mice treated had a recurrance of the distal tumors, but (importantly) these all responded to another round of CpG and OX40 agonist. Remarkably, this combination was effective across a range of implanted tumor types: lymphoma, breast carcinoma, colon, and melanoma. In case you’re wondering, the team did compare the OX40 antibody to PD-1 and PDL-1 antibodies. The latter two did delay tumor growth in the nontreated tumor sites, but were not curative: OX40 treatment was.

The team then went on to an even tougher test, a rodent line that spontaneously develops metastatic tumors in the mammary fat pads. That’s the MMTV-PyMT model, which has been widely used in the field for over ten years now. Injecting the first tumors with CpG and following with the OX40 antibody not only led to regression at the site of injection, it also protected against the development of further tumors. Mice so treated had 70% survival at 25 weeks, compared with 100% mortality at 15 weeks in the control group, which is how you get a P value of <0.0001. Clearly, the immune response raised was both durable and general, not something that was too specific to the first site of injection.

But the injection site was still crucial – a further experiment showed that when more than one antigenically distinct tumor was implanted, injection at either one cleared all the others of its type, but not the other variety. But simultaneous treatment cleared them both.

The authors boldly start off their Discussion section by saying “We have developed a practical strategy for immunotherapy of cancer.” And they may have just done it. I will defer to those with more immunology experience, but personally, I found this paper extremely damned impressive. This treatment mode has a lot of advantages. You don’t need to understand the tumor antigens themselves; you just need a suitable injection site (which will be a limitation for some tumor types, to be sure). The doses required for the injections seem to be quite small, which is also good. The only thing I worry about (always a worry with immune therapies) is what the long-term consequences might be, but when the short term consequences are that all the metastatic tumors disappear that’s not such a bad thing concern to have.

Most interestingly, the exact CpG formulation used in this study is already in human trials. Agonistic OX40 antibodies are also being studied in human trials. This paper immediately and strongly suggests that these two modes of treatment be combined. Cross the streams!

32 comments on “The Next Immuno-oncology Frontier?”

  1. Ilya Yasny says:

    Where can I bring my money?

  2. Interested Immunologist says:

    Nice summation. Just to note that they showed pretty convincingly, and you refer to this later on, that CpG doesn’t actually “sets off TLR9 signaling in the local CD4 T cell population”.

    These kinds of treatments, just like the therapies targeting CTLA4, are always going to scare immunologists, but like you say, the risk of autoimmunity later in life is worth it.

    1. Derek Lowe says:

      I’ll clear that up a bit – thanks.

  3. SP says:

    Cool stuff, except you committed a serious nerd offense, it’s cross the _streams_.

    1. Derek Lowe says:

      Fixed! Right you are.

  4. Jim Hartley says:

    More on the mechanism, with Levy as co-author, is at

    The first clinical trial was recently posted,

  5. Chrispy says:

    Coley’s toxin has been around for a very long time — there have been a lot of intratumorally injected therapies that have worked in mouse models but not in humans. Maybe OX40 is the trick, but you’ll have to forgive me for remaining skeptical.

    1. Derek Lowe says:

      Yeah, one of the things I like about this is that the barriers towards getting human data appear to be lower than usual – and human data are exactly what are needed.

  6. Emjeff says:

    It’s good to be a skeptic in this field, but I agree that these results are dramtic.

  7. R. T Winters says:

    There has been some good results in Melanoma using something called PV-10 a formulation of Rose Bengal. They inject this in melanoma tumors, and have found a similar effect that adjacent non injected tumors become attacked by the immune system as does the primary tumor injected. Of late they have been adding Opdivo and Keytruda to this in their clinical trials. I have been following this for several years. This is all exciting work, but came to late to help my wife. And yes – I have to agree – as a former anticancer drug researcher – we have managed to cure mice many times over. Never fully translates to humans however.

    1. tangent says:

      And yes – I have to agree – as a former anticancer drug researcher – we have managed to cure mice many times over. Never fully translates to humans however.

      So why is this? Or maybe what is this. Is mouse cancer just easier, and we have genuinely solved it? Or if we were mice and we were oncologists would we find that the treatments somehow don’t apply into practice?

      1. DanielT says:

        As far as I am aware we still don’t know how to cure mice of cancer that resembles human cancer (i.e. spontanous cancers in outbred populations). The mouse cancers that are cured in the lab are all artificial cancers using inbred mice.

        In many ways the whole preclinical cancer model is like the joke about the drunk looking for his keys under the lamp post, not the dark alley where he lost them, because the light is better.

      2. MrRogers says:

        There are major differences between human cancer and experimental mouse tumors, both in size and in growth timelines. This means that human cancers are far more heterogeneous than are mouse tumor models. Most solid tumors in humans have been present and evolving for 3-5 years before diagnosis; mice don’t live that long. The typical human tumor is diagnosed at a size of >1 gram. Most animal facilities wouldn’t allow you to let a tumor get that large. Many experiments begin treatment when tumors are 0.1-0.2 grams in size. What all of this means is that the odds of resistance mutations arising in humans are 10-100x those in mice. The net result is that many therapies developed in mice result in brief response in humans followed immediately by resistance and little/no net benefit. When we cure 100-200 mice in an experiment with substantial genetic heterogeneity among the tumors, then you’ve got something to write home about.

        1. tangent says:

          These make sense, thanks for explaining.

        2. DanielT says:

          Pity we never do these experiments. I doubt you could even get an experiment of 200 outbred mice past any animal ethics committee.

  8. Kendall Square Postdoc says:

    There’s a difference in expression of TLR9 in mouse DCs (mDCs and pDCs) vs. primate DCs (pDCs only) that might be a potential issue for CpG. But there are vaccine adjuvants (AS15, e.g.) out there that use CpG successfully.

    They’ve done some clinical trials with intratumoral injections of poly IC/LC (synthetic RNA, TLR3 agonist), and I think the results have been mixed. Maybe combining with a checkpoint antibody would help? I’d bet on needing to combine it with some sort of antigen though.

  9. Andy B says:

    Perhaps this is unhelpful, but as a lay reader of rationalist inclination, the question I always want to ask about things like this is, what are the probabilities that this goes anywhere? I know they’re low, but how low? Is the chance that this will ever help an actual human being more like 10% or 1% or…? Is the chance that the help will be more than fraction of a QALY more like 1% or 0.1% or…?

    Can questions like that be usefully answered? Are there better questions I should be asking? Or am I just not part of the target audience for a post like this, because I don’t have enough background to use this information to update any of my priors?

    (Not, of course, that I except to be part of the target audience for every one of your posts. You do you.)

    1. Kyle MacDonald says:

      Guess from a fellow lay reader: Your prior should probably just be the historical average QALY per dollar spent on drug discovery, with the last twenty years or so given more weight in the average (or, going even coarser, the historical probability that a given dollar spent on drug discovery exceeds your favourite cutoff in QALY). If you manage to estimate a prior worth updating that way, then you could try looking at the conditional probability that a given project ends up doing significantly better than average, given that Derek writes an excited post about it. (You might be able to get much lower variance, at the cost of some bias, by looking at the posts where he says other people are too excited about something.)

    2. johnnyboy says:

      I understand your need to know this, but no one has the answer to such a question (and those who’ll give you one are wrong). If we knew the probability of success of any therapeutic approach, we’d know where to focus our efforts, but as you can see from the myriads of different approaches to cancer treatment, we just don’t have a clue what will work and won’t. People will make educated guesses on what has the best chance of success, based on their experience, preconceived ideas, and intuition, but there is no way you can put a probability number on that with any rational basis. In medical research (and particularly in cancer), there are a few things we know, there are tons of things we know we don’t know, but all these are massively dwarfed by the things we don’t know we don’t know. It’s what makes medical research both exciting and infuriating.

  10. Anon says:

    Could someone on the thread comment on how Right to Try would apply here? Ox40 and TLR 7/8/9 have been published in phase 1 trials and continue to be developed. Would this be sufficient for Right to Try? Or would a phase 1 trials with the combination need to have been published? The current Ox40 trials are also using systemic dosing off up to 10 mg/kg which is 100 times the dosing that would be used in intratumoral injections.

  11. qetzal says:

    Would be nice to think this would turn out to be as revolutionary as it’s bring portrayed. But we’ve been down similar roads before. Endostatin, anyone?

    To be clear, I really hope this works a tenth as well in humans as it seems to do in mice. I’m just not going to be too optimistic until there’s some human data.

  12. Experienced Immunotherapist says:

    Rather old news. Intratumoral injection (including by the Levy group with CpG) has been done for many years. I cannot believe the press this has been getting.

    1. Barry says:

      As Derek emphasized, what is NOVEL here is the combination of intratumor CpG with an OX40 agonist; both components have been reported before. That it is UTILE requires proof in man. Whether it is NON-OBVIOUS the USPTO will have to decide (I’m pretty sure they will grant the patent(s))

      1. Experienced Immunotherapist says:

        I have no doubt they can get some sort of protection. Who cares? It’s a very minor advance relative to the large body of literature before it. A yawn and fairly minor news.

  13. ajp says:

    This “evidence” is unnecessary – just do the clinical experiment please. Lots of OX40 agonists in the clinic already.

  14. an immunologist says:

    As an immunologist worked on immune tolerance, I am skeptical as well. The trick of immune response is inevitably that it has two phases. At the first step, the system finds numerous ways to increase the magnitude of the immune responses (including TLRs, DC vaccines, cytokine storms, oncolytic viruses are few examples in clinic). At the second step, the system will find numerous ways to shut down the initial responses (including PD1, CTLA4, LAG3 etc.). It is relatively easy to find agents to enhance the first step ( the magnitude) because it is what bacteria and virus do (almost all of us can mount a strong reaction to foreign antigens) . The difficult to achieve long lasting effect in human is at the second step. Tolerance of immune system to self is a powerful process ( otherwise, we all end up with autoimmunities). The break-through of anti PD1 is that it accomplished the second goal in breaking tolerance in human. Injection of CpG and OX40L might simply increase the initial magnitude of an immune response against tumor. Whether the approach can break immune tolerance to tumor is not known.

  15. Immunobiologist says:

    The combination of OX40 and CpG was actually reported by the Levy group almost 10 years ago (albeit systemic OX40, not intratumoral):

    (In the above early paper, they also show equivalent / superior efficacy of IT CpG + systemic agonist anti-GITR to IT CpG + systemic agonist anti-OX40. They also show synergy of IT CpG with systemic anti-CTLA4, but claim *without evidence* in this new paper that IT CpG + IT anti-CTLA4 is not effective – probably due to lack of CTLA4 expression in this tumor model. The lack of data on the IT anti-CTLA4 is frankly disturbing and makes me question the reviewers / editors, when it is claimed in their discussion, even if the results are plausible and consistent with the expression data.)

    The utility of innate activation combined with T cell stimulation (e.g. TLR agonist + T cell agonist) has been suspected for years. The authors even show in their supplemental data 50% survival with TLR7/8 agonist + low-dose (4 µg instead of 8 µg as used with CpG) OX40, which is comparable to their low-dose CpG + high-dose OX40. Considering the TLR7/8 agonist used (resiquimod) is also investigational in humans, and there is even an approved TLR agonist (imiquimod – TLR7), the most important conclusion of this paper is probably even further validation of the already strongly supported belief that targeting multiple facets of the immune response from innate to adaptive is useful. It would also be interesting to compare other co-stimulatory pathways with investigational agonists (e.g. 4-1BB and GITR have both had agonist antibodies enter the clinic) in combination with CpG / resiquimod / imiquimod / other innate agonists (STING agonists are now in the clinic, as one example). The data from this paper don’t actually support the claim that CpG + OX40 will be a superior combination to any other (and indeed earlier work from the Levy group with CpG+GITR and the supplemental data from this paper with resiquimod seem to contradict that claim!), and we have all the in-human compounds to test the various combinations.

    1. BigSky says:

      Thank you for bringing the group’s earlier work up. That’s what I found when I hit hit PubMed 5 minutes after reading the PR release. Didn’t ‘In the Pipeline’ just run a post about hype?

  16. Barry says:

    Does the Levy mechanism require that a single leukocyte display both a TLR9 (innate) and a TCR (adaptive)? Or are we looking for an intercellular (soluble?) signal between a leukocyte of the innate arm and one of the adaptive arm? Is it expected that some killer-T cells will display both a TLR and a TCR? (in mice? in man?)

  17. Vader says:

    “Mammary fat pads.” I’m going to have to add that to my list of useful euphemisms.

  18. Barry says:

    As with other modes of ImmunoOncology, the consequences of disinhibiting the immune system (both innate and adaptive, T-cell in this case) can be daunting:

  19. loupgarous says:

    entolimod is another Toll-like Receptor agonist which created a stir a while back, because it not only stimulated immune response, but activated genes for inhibition of apoptosis, reactive oxygen scavenging, and a bunch of protective and/or regenerative cytokines, making it something worth looking at either as prophylaxis or prompt after-care for acute radiation syndrome (where it prevents or limits damage to hematopoietic and GI tissues).

Leave a Reply

Your email address will not be published. Required fields are marked *

Time limit is exhausted. Please reload CAPTCHA.