Effort to identify tumor-specific antigens: The University Industrial Complex study results

New study in journal Cell is prime example why utilization of sophisticated high-throughput methods and computer technologies does not guarantee generation of clinically useful results. I imagine the only reason this study was even accepted in Cell was the fact that list of authors included many well-known scientists with links to both academia and silicon valley (Stanford University School of Medicine, Chan Zuckerberg Biohub, Parker Institute for Cancer Immunotherapy).   

Idea of this study was to develop techniques to quickly identify tumor-specific antigens (most likely mutated antigens) that could be used in immunotherapy (though there is no evidence that any cancer vaccines based on mutant protein sequences actually work in humans using available practices). 

For this task, the authors took advantage of yeast-display library expressing random peptide covalently linked to the HLA-A*02:01 molecule, an allele which is present in up to 50% of a number of populations. The authors estimated that "approximately 400 million unique peptides ranging from 8 to 11 amino acids are represented in the combined [yeast-display] libraries."

To validate this approach, they used three recombinant 'blinded' positive control TCRs derived from a melanoma patient (their antigen specificity had been identified independently by exome sequencing, tetramer staining and binding prediction algorithms). However, antigen-specificity of only 1 TCR (NKI 2) could be validated using their yeast-display library. As the authors said "targets of NKI 1 and NKI 3 could not be unambiguously identified through this blinded validation."

Of note, in these validation experiments with NKI 2 (specific for ALDPHSGHFV, a peptide neoantigen derived from CDK4 and other DMF5 TCR specific for EAAGIGILTV derived from the MART-1 melanoma antigen, successful validation [specific enrichment + TCR staining] occurred when HA tagged 10-mer epitope library were used. 

The authors anyway went ahead with this "less than perfect" approach to try to identify tumor antigen specificity of T cells derived from 2 patients with colorectal adenocarcinoma and homozygous for the HLA-A*02 allele. The authors focused on 20 TCR most enriched in tumor tissues (based on frequency of occurrence of the same TCR genes). 

Out of these 20, only 4 TCRs could enrich peptide from the library (only with c-Myc tagged 9-mer epitope library) and only 3 TCR could stain yeast samples.  

Next, the authors try to identify epitopes from potential landscape of sequences for each TCR. Several algorithms were deployed (at least 3 or more such as a modified variant of the previous statistical method using a position weight matrix and a method utilizing a two-layer convolutional neural network). They found 1 peptide sequence EYGVSYEW, which closely matches the peptide motif for TCR 1A, however, neither this exome peptide or the anchor-modified exome peptide (EMGVSYEM), nor the human peptide predictions stimulated the cell line modified to express the TCR 1A. TCR 4B was stimulated with several peptides and as the authors write "true in vivo specificity cannot be unambiguously identified without additional tumor information". Regarding TCR 2A and 3B, only 1 peptide stimulated cell line expressing these TCRs. This peptide was MMDFFNAQM, which is derived from U2AF2, a protein involved in an RNA splicing complex. However, in both patients, no mutations were found in U2AF2.

In summary, the authors wrote "although we cannot definitively determine an immune response targeting the peptide derived from U2AF2, the evidence from the yeast-display screen, prediction algorithm, and in vitro stimulation identify this peptide as the likely target". However, when reading this study it is clear that none of the components worked: yeast-display screen performed suboptimally, prediction algorithms provide little clue and in vitro stimulation made it even more confusing. So, what have we learned from all of these? I would say maybe don't do what they did.

posted by David Usharauli    

Melanoma dendritic cell vaccine is safe but is it effective?

Few days ago journal Science has published yet another study describing procedures how to identify tumor (melanoma)-specific immunogenic neo-antigens through exome-sequencing and combining peptide binding in silico and in vitro peptide recognition analysis. 

There is no doubt that cancerous tissue can yield multiple neo-antigens derived from non-synonymous mutations. Hypothetically and practically (as this new study and others as well have shown), immune system can and is able to detect these minute differences in the mutated proteins. It appears that by combining current in silico algorithms such as NETMHC-3.4 with epitope presentation assays (in vitro assays) provide quite accurate list of potential immunogenic tumor peptides.

The authors in this new paper, for example, were able to identify and confirm in a complementary in vitro assays (T2 cell peptide binding assay and tandem minigene constructs expression assay in DM6 cell line) the presence of several neo-antigens in melanoma tissue derived from 3 patients.

Re-injection of CD40L+TLR ligand matured, melanoma-peptide pulsed autologous dendritic cells back into patients yielded antigen-specific CD8 T cells expansion.

This was a small Phase I clinical study to determine the safety of the DCs vaccine. Since science fundamentals are strong behind this trial (especially considering the authors focus on IL-12p70 producing DCs as a source of cellular vaccine), the results were expected. Of note, these 3 patients were treated with ipilimumab (humanized α-CTLA4 antibody) prior to the experiments described in this paper. It is not clear how this could have skewed the [positive] outcome of DC vaccine here. Anyway, successful tumor treatment would require simultaneous approach from several directions (DC vaccine, checkpoint inhibitors, small drug tyrosine inhibitors).

David Usharauli      

Algorithm-based immune-epitope discovery

The goal of a personalized tumor immunotherapy is to identify tumor epitopes that are immunogenic and specific for a given patient. To do it, however, would require the knowledge of exact sequencing of a given "healthy" individual's proteome prior to tumor development. It is safe to assume that generic, public database would not provide the adequate tool.

In addition, the development for robust algorithms able to accurately predict the strength of binding of processed epitopes to MHC class I or II (or HLA) is an absolute must.

It appears that we are getting very close achieving this goal. I have already mentioned in my earlier post that journal Nature has published 5 separate papers about tumor immunology in December issue. However, my review showed that 3 of them were merely based on observation-type clinical data, more suitable for publication in journals like JAMA or Lancet, but not Nature. I am not sure why Nature has decided to include those 3 clinical papers in its publication list. However, 2 other papers that were mouse studies provided sufficiently robust experimental results for my analysis.

This 2nd study, led by Robert Schreiber at Washington University School of Medicine, has analysed mice immune response to aggressive sarcoma tumor cell line following anti-PD1 and/or anti-CTLA4 antibody therapy.  

Initial set of experiments indicated that anti-PD1 and/or anti-CTLA4 antibody therapy "helped" wild-type mice to reject the aggressive sarcoma cell lines in a T cell-dependent manner.

To understand the basis of tumor immunogenicity, the authors has applied 3 different epitope predicting algorithms to the available exome-sequencing data from one of the sarcoma cell line (d42m1-T3). With these in-silico generated epitope prediction methods, the authors were able to identify two dominant mutant epitopes, one from laminin alpha subunit 4 (Lama4) and another from asparagine-linked glycosylation 8 (Alg8).

Parallel experiments with TIL functional assay, tetramer binding and MHC peptide eluates mass spec analyses confirmed that these two mutations from sarcoma cell line (d42m1-T3) were the dominantly recognized by TIL following anti-PD1 therapy.

Additionally, Lama4 and Alg8 mutant peptides, but not wild-type variants, were immunogenic.

Moreover, mutant peptides + adjuvant immunized mice were able to reject sarcoma cell line.

Mechanistically, combination of anti-PD1 and/or anti-CTLA4 antibody therapy (but not anti-PD1 alone) augmented existing tumor-specific T cell numbers and their effect differentiation.

Note that anti-PD1 therapy alone were still able to protect mice from d42m1-T3 (see Fig. 1a), even though its effect on CD8 T cells was minimal, implying that the mode of action of anti-PD1 antibody therapy is less clear.

Here is my opinion: there is no doubt that tumors express mutant proteins (epitopes). If a mutant protein has a critical, obligatory function in fueling tumor growth, then the tumor cannot afford to lose it in an "escape phase" so targeting such epitopes would be useful. If however mutant protein does not have such function tumor will lose it easily once under pressure from T cells thus making immunotherapy less productive. Since we still do not know the function of many proteins it would be very difficult to identify "escape-proof" mutant epitopes for tumor immunotherapy.        

David Usharauli