Silo Busting: Can We Build Better Immunotherapies for the Treatment of Prostate Cancer?

A very interesting paper published last week in Cancer Research. Written by investigators at The Johns Hopkins University Sidney Kimmel Comprehensive Cancer Center, the paper addresses unmet need in prostate cancer (PC), and proposes a multi-modality approach to building an effective therapeutic. The authors are Nathaniel Brennen, Charles Drake and John Isaacs and title is Enhancement of the T-cell Armamentarium as a Cell-based Therapy for Prostate CancerThe abstract is here.

The authors review the state of hormone-treatment and chemotherapy refractory, metastatic, and/or relapsed prostate cancer, collectively mCRPC. Approximately 200,000 patients (male) are diagnosed with PC each year in the US. Deaths from mCRPC number roughly 30,000 men/year in the US, and there are very limited treatment options. Potential therapies include PC vaccines like Provenge and others in clinical development (GVAX, ProstVax).  Notable for anticipating the need for broad immune stimulation ProstVax uses viral vectors which encode prostate-specific antigen (PSA) along with costimulatory molecules B7.1, ICAM-1, and  LFA-3. More conventionally, Provenge and GVAX are composed of prostate cancer cells that are irradiated and then engineered to express the immune molecule GM-CSF. Provenge uses patient-derived tumor cells while GVAX uses PC tumor cell lines. Provenge offers a modest survival advantage versus the standard of care chemotherapeutic, docetaxel. GVAX failed to meet its primary end point of overall survival when compared with docetaxel in a phase III study. In a phase 2 study, ProstVax treatment led to a median overall survival (OS) of 25 months, very similar to Provenge (vs. 17-22 months in the placebo groups, respectively). Many other variations on this methodology are in development. Other immunotherapeutic approaches in the pipeline include antibodies such as Vervoy (ipilimumab) and nivolumumab that have shown efficacy in other solid tumors. The anti-CTLA4 antibody ipilimumab, a potent immune checkpoint modulator, is being evaluated in mCRPC, both alone and in combination with chemotherapy, radiotherapy or tumor vaccines. Current phase 3 trials include ipilimumab vs placebo in chemotherapy-naïve patients and a second trial in patients who have first received chemotherapy and bone irradiation. Early results suggest that previously treated patients obtain minimal benefit from ipilimumab but that newly diagnosed patients perhaps fare better. Surprisingly, early clinical studies with the anti-PD-1 antibody nivolumab failed to show any benefit. Another approach will be to build T cell therapies such as those based on chimeric antigen receptors (CARs) that have been successful in treating leukemia and lymphomas. However, while known mCRPC antigens are certainly overexpressed in tumor cells, they are not uniquely expressed. For example folate hydrolase (PSMA), a targeted mCRPC antigen, is also expressed, albeit at low levels, in the kidney, small intestine and both the central and peripheral nervous systems. The prostate stem cell antigen (PSCA), another mCRPC antigen, is also expressed in the bladder, colon, kidney, and stomach. Given the inherent aggressiveness of CAR T cells, targeting these antigens is dangerous business indeed.

Back to the paper. Instead of just stating that we need to mix novel immunotherapeutics until we find the right combination, the authors spend some time deconstructing this particular tumor type, and then rationally build up a therapeutic hypothesis based on what the tumor has shown them. This is of course the essence of the personalized medicine approach, although perhaps based more on critical thinking about the biology rather than cataloging driver mutations (both being needed in the bigger picture). And while their ultimate proposals may be overly engineered, they are certainly on an interesting track.

So what do PC tumors tell us? First, PC is often an indolent and slow growing tumor that is caught early by routine screening for soluble prostate specific antigen (PSA, we’ll come back to all these antigens later). Second, more than 80% of PC tumor biopsies have extensive inflammatory infiltration, showing that the immune system has recognized that something is wrong, but can’t seem to fix it. Third, the majority of cells within the inflammatory infiltrate are T cells, so we have the right cell population sitting there ready to kick off a defense (but they don’t).

This is a situation that resembles that of melanoma, that is, a solid tumor that is filled with disabled immune cells, many of them T cells. The difference is that melanoma can be effectively treated with ipilimumab, nivolumab and, very strikingly, with the combination of the two. We reviewed the synergy of immunotherapy treatments for melanoma in our prior post.

So why are PC tumors different? The T cells appear either “exhausted” and unresponsive (anergic T cells), or are actively immuno-suppressive T-regulatory T cells (Tregs). They are identified by expression of specific transcriptional factors (FoxP3) and cell surface proteins (CTLA4, PD-1, and others). They respond preferentially to specific factors that keep them quiescent or immuno-suppressive (IDO, TGFbeta, IL-10). Basically they are so loaded down with redundant and compensatory off-signals that they simply will not be roused by any single method – not a vaccine, not an immune-checkpoint antibody, not even, it seems, when given in combination. A bigger stimulus is needed.

And this is where Brennan et al. lead us in their recent paper. They begin by introducing a CAR built by the team at Memorial Sloan Kettering Cancer Center in New York, led by Michel Sadelain. As a reminder, most CAR constructs contain a single domain for antigen recognition, usually derived from an antibody. We discussed the technology earlier (see this post, and related posts from March). So, for example, the CAR constructs used to transduce T cells that successfully treat acute and chronic leukemias (ALL and CLL) all recognize the antigen CD19. The CAR proposed by the MSKCC group has two antigen recognition sites, one for PSMA and one for PSCA. As noted above, neither of these antigens alone is specific for mCRPC, nor would I imagine the combination to be 100% specific. As someone tweeted during AACR: “A PSMA CAR? Their gonna kill someone with that drug”. Suffice to say, that’s one problem with the dual-targeted CAR. Another, more theoretical, is presented by Brennen et al. as a problem of effectiveness. This is a curious argument. The author’s state that all CAR therapies “share a dependence on endogenous T-cell effector functions”. This is maybe an unfortunate choice of “endogenous” since the activity of CAR-T cells is certainly not endogenous but rather artificially driven by the construct with which it is transduced. The available data suggest that such activity is considerable and dependent only on the availability of antigen. Whether CARs will shut down once they encounter the immunosuppressive environment of the mCRPC tumor is a different question that has nothing really to do with endogenous effector function. But lets move on and take their point for the sake of discussion as it drives the next interaction of their proposed therapeutic, which is to introduce a non-natural cytotoxic element to the CAR.

I suspect the authors are being provocative to make a point, but this seems a bit overdone. Still, lets soldier on. The proposal is now to create a Trojan Horse out of a CAR T cell, by arming it with a proform of a toxin. The example offered in proaerolysin, a vicious bacterial protein capable of blowing holes in every cell it encounters. The toxin is known from studies in which it was injected directly into tumors, in which setting it destroys every cell in a very small area, thus ineffectively. A second iteration of this strategy was to mutagenize the toxin by introducing a PSA-dependent cleavage site, thereby only releasing the active toxin in the presence of PSA, the PC selective biomarker used in routine testing. This Trojan Horse or “molecular grenade” creates a “kill zone” when cleaved by PSA.

To recap. We now have a dual CAR targeting PSMA and PSCA, carrying proaerolysin, that is activated by PSA. It actually get more complex then that, but lets just stop there, take a step back from this Rube Goldberg approach, and ask if we can find merit in the exercise.

There are two interesting concepts here, useful in their own right. One is to “mask” the CAR, irrespective of Molecular Trojan Kill Zones. One genuinely scary and unpredictable aspect of CAR technology is off-target toxicity. This has been seen repeatedly, and casts a long shodow over every new CAR antigen in development. Why? Because there is the “what if” aspect. What if the antigen is expressed somewhere else, unexpectedly, or what if the domain used for antigen recognition also sees a different, related, antigen on normal tissue? Both of these things have happened, and when these things happen, patients die, very quickly.

So, masking a CAR, even a single domain CAR is a very good idea. Happily this is a well worked field, with companies like CytomX busy working out the utility of using tumor-specific proteases to cleave specific peptide sequences that otherwise effectively mask antibody-drug complexes (ADCs) in an effort to improve their safety. That technology may be transferable in some form to CAR technology, providing an added layer of specificity without having to add up all these different tumor-selective antigens (PSMA, PSCA, PSA).

The second is to “arm” the CAR, masked or otherwise, with a toxin. This takes some thought but is possibly an attractive idea, if one could work out the geography (where to place the thing so it works where one wants it to work). I’m a little puzzled about the idea that you could add a toxin, in proform or otherwise, and avoid killing the transduced T cell. For example, PSA, used in the thought experiment above, exists in systemic circulation, a bad place to blow up your Trojan Horse.

Regardless the take-home message is that we might productively combine technologies across the immunotherapeutic space, to build better individual therapeutics.

Returning to mCRPC, where does immunotherapy actually stand, today. It’s fair although sad to say that progress here is similar to most other solid tumors, that is, there has been limited success. There is a vast amount of work to be done but happily there is an army of researchers and physicians willing to do it. With any luck at all papers like the one discussed in this post will help point the way, as loopy as a Molecular Trojan Kill Zone might sound today.

We’ll talk about iCARs next perhaps. or maybe BiKES. We’ll see.

stay tuned.