Personalized Cancer Vaccines: Why has it been so hard to find the clinical signal?

mRNA-based personalized cancer vaccines (PCVs) have been getting a lot of attention recently due to the Merck-Moderna announcement of a Ph3 success.  (I am eager to see the actual data, which have not yet been released.)  I am cautiously optimistic, but is this really as big of a a deal as it sounds?  As others like Mark McCarron have pointed out, the devil is in the details.

As a champion of new therapeutic modalities, I truly hope that PCVs will one day make a meaningful difference for cancer patients.  As a scientist, I am hesitant to get too excited about the news from Merck and Moderna, and here is why:

1. So far, the clinical successes for PCVs have come in the post-surgery setting, where the tumor is first removed and then the treating clinician looks for extension of disease-free survival (DFS).  In the Merck & Moderna Ph3 trial , the patient population was limited to patients with completely resected Stage IIb-IV melanoma, and the key outcome metrics were RFS and DMFS.  Their PCV intismeran autogene was given with Keytruda, which is a highly effective immune checkpoint inhibitor in this disease setting (so arguably a high bar for the combo therapy compared with Keytruda alone).

So far PCV-driven responses in advanced and metastatic cancer have been limited, and yet this is where the biggest unmet need lies.  On the other hand, a truly efficacious PCV that kept surgically resected tumors from returning would be a big deal for many patients. 

2. There is reason to think that the mRNA-LPN (lipid nanoparticle) vehicle by itself creates an adjuvant effect that contributes to the anti-tumor response (for example, see this study).

While an adjuvant effect can be a good thing, in the clinic it is hard to tease this apart from T-cell-driven neoantigen-specific anti-tumor responses, which is the true goal of PCVs. A lot of engineering has gone into making the mRNA and the LPN immunologically silent, and yet one or both of these components are still likely triggering an innate immune response in most PCV studies.  It is therefore possible that the apparent Keytruda + intismeran autogene clinical responses are not due to the neoantigens that are encoded in the PCV.  Ideally, Merck & Moderna would run a control trial (or trial arm) with “control mRNA” + LPN compared with intismeran autogene to test whether the neoantigens are actually driving the clinical tumor responses.  (However I can’t see Merck or others running such a control study unless regulators forced them to do so, as such a clinical trial would likely be big and expensive, with a big risk if it doesn't pan out as hoped.)

3. Several groups and companies (not just Moderna) have gotten quite good at priming and boosting neoantigen-specific T-cell responses in peripheral blood.  Moderna almost certainly does not uniquely possess a magic algorithm for predicting just the right neoantigens for PCVs.  For example, Genentech and BioNTech published on the robust T-cell responses triggered by their PCV autogene cevumeran, and Lelia Delamarre, Suchit Jhunjhunwala, and team have continued to make major progress in this area. (e.g., see this recent paper)

By the way, the BioNTech cancer vaccine has shown efficacy in a small number of advanced or metastatic solid tumors (see above reference).  Just not enough (yet) to show significance in a large clinical study.

4. Despite PCVs triggering strong T-cell responses in the periphery, most of the time these anti-cancer T-cells fail to get to / into tumors, and they fail to shrink tumors.  Barriers exist that prevent this, ranging from loss of MHC on tumor cells, to physical occlusion by stromal cells and their collagen matrix, to the presence and action of suppressive immune cells within the TME, to the production of immune-suppressive molecules by the tumor itself.  So far PCVs seem largely unable to overcome these barriers, which are prevalent in advanced cancer, and it is not clear that even the mighty Keytruda can help.

I believe that PCVs may in fact work in earlier-stage cancers and in the post-surgery setting, and more work remains to be done to demonstrate pan-cancer, long-term efficacy. I hope that PCVs will also work in advanced cancer patients, when administered in combination with the right therapies that can remove one or more of the aforementioned barriers.

Many thanks to my former Genentech and BioNTech colleagues for their massive efforts developing PCVs.  

Founders, if you are getting into this space, I would be happy to be a resource to you.  The challenges here are many and real.  The hype from some who have invested in PCVs, while appreciated, tends to ignore a multitude of failed companies and clinical studies in this space.

Here’s hoping that the Merck & Moderna Ph3 result is in fact a game changer.  Onward!

Note for context: I worked on the Personalized Cancer Vaccine (PCV) program at Genentech alongside Lelia Delamarre, Ira Mellman, Suchit Jhunjhunwala, Mahesh Yadav, Ina Rhee, Lars Mueller, and many others.  My group helped develop algorithms for neoantigen selection, an effort fearlessly led by Suchit Jhunjhunwala at Genentech and Jonathan Blake’s team at BioNTech, as well as collaborators at TRON.  As a computational biologist, I was immediately drawn to the fact that tumor NGS + mutation calling + neoantigen prediction is a key upstream process for therapeutic manufacturing (the mutations are, in effect, the therapy).  One of the goals of my work with Mahesh Yadav was to analyze clinical data from the Genentech/BioNTech Ph1 pan-cancer study for evidence of T-cell responses and tumor responses to the PCV autogene cevumeran.