I’m asked a lot about the onco-vaccine field, and if immune checkpoint inhibitors will be the key to unlocking the potential of this long-suffering therapeutic class. The answer is never simple, since we are often looking at thin patient data that can contain compelling hints of efficacy – those immunized late-stage patients who not only regressed but stay in remission, month after month and year after year. The problem for companies and investors is that such observational data can be very misleading, and the vaccine candidates most often go on to fail in later and larger clinical trials, sometimes spectacularly. These big failures burden the field with a high evidentiary bar.
Data have emerged that suggest several issues with most vaccines, and these issues are both distinct and related.
At the end of November Nature published two interesting papers that asked a very simple question: what immunogenic antigens are present in common mouse tumor models. Yadav et al from Genentech and Immatics Biotechnologies (link 1) used a genome-wide exome and transcriptome sequence analyses, mass spectrometry and structural modeling to identify immunogenic neo-antigens in the widely used MC-38 and TRAMP-C1 mouse syngeneic tumor models. These models are considered poorly immunogenic in wild-type syngeneic (C57Bl6) mice. The sequencing analysis was used to identify mutated proteins that were present at >20% allelic frequency. From the MC-38 model, 1290 expressed mutations were identified of which 170 were considered to be neo-epitopes, that is, modeling suggested they would be expressed by MHCI and sufficient residues would be solvent exposed to allow immunogenicity. Only 67 expressed mutations were found in the TRAMP-C1 model, and of these 6 were considered to be potential neo-epitopes. Of this total of 170 (MC-38) and 6 (TRAMP-C1) only 6 bound MHC1 by Mass Spec, with a predicted IC50 for MHCI < 500nM. Of these, 3 were actually immunogenic in vivo (using C57Bl/6 mice) and could protect wild-type mice from tumor challenge. The neo-epitopes were found in the proteins Dpagt1, Reps1 and Adpgk. Here is a schematic of the filtering scheme:
Working the other way, the authors confirmed the immunogenicity of neo-epitope peptides by analyzing tumor-infiltrating lymphocytes (TIL) and staining with peptide–MHCI dextramers to identify bound T cells. CD8+ T cells specific for Reps1, Adpgk and Dpagt1 were enriched in the tumor. Using the Adpgk neo-epitope, the TIL were further investigated and found to express PD-1 and TIM-3, inhibitory receptors associated with anergic or “exhausted” CD8+ T cells, showing that the murine immune system had indeed recognized and responded to the neo-epitopes, a response that was then actively immunosuppressed in the tumor microenvironment. Importantly, none of the identified neo-epitopes would qualify as tumor-antigens, that is, they are not specifically overexpressed at a sufficient level to qualify. The neo-antigen-specific TIL are also pretty rare, suggesting that use of tumor lysate as an immunogen to elicit an anti-tumor response may “miss” by failing to present enough of the right antigen to the immune system.
We noted at the top that this paper came out of labs at Genentech and Immatics. An agreement between Roche/Genentech and Immatics will focus on the use of this technology. The two companies will develop new tumor-associated peptides (the neo-epitopes as cancer vaccine candidates, initially targeting gastric, prostate and non-small cell lung cancer. The most advanced candidate is IMA942, a peptide vaccine for the treatment of gastric cancer, in late preclinical development. Immatics CEO Paul Higham has publicly stated that a Phase I study of IMA942 with Roche’s PD-L1 inhibitor MPDL3280A in gastric cancer will be initiated soon. Immatics will also conduct research to identify neo-antigens for the additional indications. For those keeping score, Immatics received $17 million upfront, committed research funding, and potential milestones
The second work, led by Schreiber’s group at Wash U, used mouse models to ask a different but related question: what tumor antigens are recognized after immune checkpoint blockade with anti-PD-1 or anti-CTLA4 antibodies (link 2). The sarcoma lines d42m1-T3 and F244 were rejected in wild-type mice treated with either anti-PD-1 or anti-CTLA4 antibodies, in a CD4/CD8/IFNy/DC-dependent manner. As in the first paper, a filtering system built from diverse technologies was used to identify potential neo-epitopes. Mutations were identified by cDNA sequencing, translated to corresponding protein sequences, then tested against MHCI binding algorithms. Neo-epitopes were ranked by predicted median binding affinities and likelihood of productive immunoproteasome processing and antigen display. Using these methods, two MHCI restricted neo-epitopes were identified in Alg8 and Lama4.
As in the first paper, the authors then turned the system around, asking what neo-epitopes could be identified through analysis of TIL. Alg8 and Lama4 were found in tumor TIL and their frequency was increased by treatment with anti-PD-1 or anti-CTLA4. The neo-epitopes could successfully be used to induce anti-tumor immune responses. As in the first paper, these are not neo-epitopes that would qualify as tumor-antigens using the traditional criteria of selective and high expression.
So these are our two distinct but related issues with the current tumor vaccine landscape: those that have selected antigens have likely selected the wrong ones, while those that use lysates are likely too dilute.
Importantly, we can now compare these model systems to actual data from human patients treated with anti-CTLA4 antibody, as published recently in NEJM (link 3). The clinical group from Memorial Sloan Kettering Cancer Center obtained tumor tissue from melanoma patients treated with anti-CTLA4 antibodies (ipilimumab or tremelimumab). As in the mouse study, whole-exome sequencing was performed, somatic mutations identified and potential neo-antigens were characterized. Here is their schematic:
Baseline analyses showed that there was a significant difference in mutational load between patients with long-term clinical benefit and those with a minimal or no benefit, with higher mutational load associated with response. However the relationship was correlative, since some tumors with a high mutational burden were not responsive to anti-CLTA4 therapy. Using peptides predicted to bind to MHCI with a binding affinity ≤ 500 nM, the authors focused on mutated peptide sequences shared by multiple tumors and shared by patients having long-term clinical benefit. From this analysis a neo-epitope “signature” was derived, consisting of a distinct pattern of mutated peptide sequences. One of the peptide signatures identified matched an amino acid sequence in MART1, a known melanoma antigen. However the bulk of predicted neoantigens were tetrapeptide sequences shared across antigenic peptides, that is, they were encoded by diverse genes (you have to go into the huge supplemental data file to find this list, suffice to say it is very long). To make sense of this curious result the authors note that the some of the predicted sequences have high homology to viral and bacterial antigens, citing a CMV antigenic sequence as an example. They speculate, and here we quote from the paper: ” These data suggest that the neoepitopes in patients with strong clinical benefit from CTLA-4 blockade may resemble epitopes from pathogens that T cells are likely to recognize.”
The unstated null hypothesis is that there is no relationship between the shared tetrapeptide sequences and clinical response and that the association between the two phenomena is due to a productive immune system response to anti-CTLA4 antibody therapy which has released the anti-tumor response as well as many otherwise quiescent immune responses, such as those to pathogenic viruses and bacteria (and to self antigens, as shown by the autoimmune toxicity associated with anti-CTLA4 antibody treatment). The in vitro response assay data sheds no real light on this, since these assays cannot distinguish anti-tumor responses from other immune responses. So we are left with an intriguing correlation, and a nagging sense that only a very few of the vast number of predicted neo-epitopes will actually trigger bona-fide anti-tumor T cells responses. Indeed the weakness of the paper is the reliance on predictive rather than experimental identification of productive peptide/MHC interaction. As we say in the mouse studies the majority of predicted interactions are not confirmed experimentally.
Regardless, the paper is a remarkable and important step forward, and shows us (as do the mouse studies) the level of investigation required to identify neo-antigens that might expand be used to expand patient TIL populations, as we have discussed in other posts. Returning to onco-vaccines, these three papers together show us that neo-antigen anonymity, rarity and variability from patient to patient are critical issues that will need to be addressed if we are to efficiently develop this therapeutic class.