NEW YORK (GenomeWeb) — Using mouse models, researchers from Genentech have developed a preliminary algorithm for predicting tumor-specific antigens that can elicit an attack from the body's immune system.
If validated in the future, the method may be useful in the development of personalized vaccines.
In a letter published this week in Nature, researchers led by Genentech's Lélia Delamarre described the method, which combines whole-exome sequencing and mass spectrometry to identify tumor antigens in mice and predict which will evoke an immune response.
Researchers today generally don't combine these methods, but each has its strengths and weaknesses. For example, investigators often use whole-genome, -exome, or RNA sequencing to first elucidate tumor mutations. They can then employ an algorithm to predict the mutated peptides that MHC class I molecules bring to the cell surface so that cancer-killing T cells can recognize the antigen and destroy the cell.
Although this approach is "pretty much what we use to predict what peptides are eventually immunogenic," it is limited by the fact that only about 1 percent of the peptides predicted to bind to class I molecules are actually made by the cell, said Delamarre, who works within Genentech's Cancer Immunotherapy and Hematology Department. "So, that is a big hurdle," she told GenomeWeb.
Another way researchers currently predict tumor antigens is by first sequencing peptides using mass spec and then comparing the sequences against public proteomic databases. "The big advantage of mass spectrometry is that it does allow us to identify peptides that are presented on MHC class I [molecules]," Delamarre said. "We know for sure that these peptides are presented [and] it is very accurate."
However, the drawback is that the public databases that the peptide sequences would have to be referenced against usually don't contain such information, but instead only contain wild-type proteins, Delamarre said. Also, mass spec "requires a lot of material" and it may be difficult to implement, she added.
In the Nature letter, Genentech's researchers attempted to draw on the advantages of both approaches. They first employed whole-exome sequencing to identify thousands of variants in two mouse tumor cell lines. They then narrowed their findings to those most likely to be expressed in the majority of tumor cells by cross-referencing RNA-sequencing-based variants that were present at least at 20 percent allelic frequency. This revealed some 1,290 expressed mutations in the cell line dubbed MC-38 and 67 mutations in the TRAMP-CI cell line, from which Delamarre and colleagues identified 170 and 6 neo-epitopes, respectively.
Delamarre and colleagues next performed mass spec analysis to identify MHC class I-presented peptides and then ran those findings against a repository of transcriptomic data specific to the mouse tumor lines. Use of such sample specific databases is a growing trend within proteomics and proteogenomics research, as the approach allows for more complete investigations of sample-specific peptide variants and modifications.
The mass spec analysis revealed a few thousand epitopes in the cell lines. Comparison and validation of the mutational data and the mass spec analysis revealed that out of the more than 1,300 mutations identified in the two mouse cell lines, only seven peptides were presented on MHC class I molecules by mass spec analysis.
That so few mutated peptides were identified, Delamarre noted, may be due to the limited sensitivity of mass spectrometry. "About 30 percent of the peptides that are presented on a MHC class I can be identified by mass spectrometry," she said. And again, the fact that the cell actually makes fewer peptides than are actually predicted by current approaches is another reason for the limited number of mutated peptides researchers identified. "We predicted around 170 peptides were good binders, and we found only seven," Delamarre said.
The researchers went on to gauge if they could predict the immunogenicity of these seven peptides by analyzing their binding affinity with the MHC class I molecules and their ability to interact with the T-cell antigen receptor. "One of the questions we wanted to ask is which peptide that contained a mutation could be immunogenic," Delamarre said. "In other words, is the immune system going to be able to distinguish the mutant peptide from the wild-type peptide?"
By analyzing binding affinity and T-cell antigen receptor interaction, and by modeling the structures of the mutated peptides, researchers predicted that two out of the seven – Reps1 and Adpgk – would evoke an immune response. Sure enough, when investigators immunized mice with these peptides encoding mutant epitopes, Reps1 and Adpgk elicited strong cancer-killing T cell responses, while "the exact counterpart wild-type peptides of Reps1 and Adpgk were not immunogenic," the researchers wrote in the Nature letter.
Additionally, out of the four remaining peptides that the researchers didn't predict to be immunogenic, they found that Dpagt1 induced a weak CD8 T cell response. And when they further investigated the immunogenicity of these peptides by analyzing tumor-infiltrating lymphocytes, they found that "CD8 T cells specific for Reps1, Adpgk, and Dpagt1 were enriched in the tumor bed but not T cells specific for the other mutated peptides."
Finally, Delamarre's team injected healthy mice with these three mutated peptides and tumor cells to assess the immune response. "Tumor growth was completely inhibited in most of the animals in the vaccine group compared to the [control group]," the researchers wrote. "Two of the animals that had substantial tumor growth in the vaccine group also had the lowest frequency of Adpgk-reactive CD8 T cells in blood before tumor inoculation, strongly supporting that CD8 T cell responses specific to mutated peptides conferred protection."
Although Delamarre's team tested this prediction method with only seven peptides, they concluded that their efforts move the field toward developing an algorithm for monitoring T-cell responses in cancer patients and developing personalized vaccines. "It seems to me that our prediction algorithm is pretty accurate," Delamarre said. "Of course the sample is small and we'll have to confirm [this approach] with other tumors."
Cancer immunotherapies are a big focus at Genentech. Last year, its parent company Roche signed an R&D deal with Immatics Biotechnologies to develop cancer vaccines and immunotherapies, which could result in $1 billion in milestone payments for Immatics.
Immatics has a proprietary antigen discovery engine, called XPresident, which enables direct quantification of tumor-associated peptides. Using the technology, which combines mass spec, genomics, biochemistry, and immunology, the company has generated a range of cancer vaccine products in clinical and pre-clinical development.
Within its immunotherapy work, Genentech is particularly interested in identifying biomarkers and using diagnostics to identify best responders and guide treatment strategies. "One theory about immunotherapies is that they could work for everyone, with any type of cancer," Daniel Chen, Genentech global development leader, recently wrote in an editorial describing the company's cancer immunotherapy research strategies. "The data to date show that this isn’t the case. Even though everyone has an immune system, not all patients will respond to the same medicine in the same way."
In order to establish the algorithm for clinical application, Genentech will need to test out its ability to identify immunogenic mutant peptides in human tumors. "Ideally we would like to simplify the whole process," Delamarre said. "As you know, mass spectrometry is a very powerful technology … for identifying peptides that are presented in MHC class I in the tumor cell." However, the "cumbersome" nature of mass spec makes it hard to implement it in the clinical setting, she noted. "We would like to rely entirely on a computer-based algorithm of prediction."