Expitope 2.0: a tool to assess immunotherapeutic antigens for their potential cross-reactivity against naturally expressed proteins in human tissues

Here we provide an overview of our previous study [8], where we analyzed short (8-15 amino acids) peptide sequences from the Cancer Immunity Peptide Database [17] as well as peptides of viral origin. The CR-index calculation was based only on the PaxDB protein abundance database and without tissue weighting.

The peptide dataset consisted of four groups of currently known human MHC class I epitopes including: mutation antigens displayed by tumor cells (40 peptides, group A), cancer-testis (CT) antigens (67 peptides, group B), differentiation antigens (57 peptides, group C), and overexpressed proteins (94 peptides, group D). In addition, 89 epitopes originating from viral sources (group E) were investigated. When matched exactly, the group of “mutation” antigens produced no hits to the proteins normally expressed in human tissues, since the epitopes of the group have sequences that originated from mutations of normal human protein sequences. The second validation is from the CT antigens, which at small numbers of mismatches (0-1), showed few matches to proteins expressed in the majority of human tissues, with the expected exception of ovary/testis, where multiple hits were found. The hit patterns were very similar for all epitopes of this group. This is exactly as expected, since CT antigens are expressed mostly in these two tissues. In contrast to the results for groups A and B, the antigens of the groups C and D showed more hits, both for exact matches and for high numbers of mismatches. This is also as expected as the proteins containing the epitopes are expressed in a wide variety of normal tissues. Finally, the epitopes originating from the viral sources showed noticeably fewer matches to the human proteins compared to the cancer peptides.

We sought to assess quantitatively the extent of potential “background” CR of the epitopes derived from the host individuals having different disease states - ranging from healthy to cancer. Such background CR is not caused by one single therapy but accumulates due to many factors, including an unknown history of diseases.

The I _CR indices of individual epitopes calculated across the seven databases used in this study are highly correlated, since for each database they are obtained by summation of the abundance (or expression) values for the same proteins. There are high correlations between the I _CR values computed for the peptides using the three abundance databases as well as between the I _CR values derived from the four expression databases (data not shown). Similarly, the correlations between the abundance and expression indices are high (Additional file 5: Figure S1), with the Pearson’s coefficients in the range 0.94-0.96. Averaging of the indices allows one to obtain a more accurate prediction of CR due to increased signal-to-noise ratio, as the databases are derived from different data sources.

Figure 2 shows the I _CR indices for the four epitope groups described in Table 2 (group I _CR indices before averaging by databases can be found in Additional file 5: Tables S5-S7). The indices for the epitopes computed from 10% top-scoring NEs (Q=0.02, Fig. 2 left) are on average 3-times lower, compared to those from 50% top-scoring NEs (Q=1e-4, Fig. 2 right), corresponding to lower numbers of matching NEs. Higher thresholds for Q correspond to a higher probability of the selected NEs to be immunogenic. It has been reported that the top-scoring 7-10% epitopes identified by the immunogenicity prediction methods have 85% probability of being immunogenic [18]. In this work we have chosen two thresholds of 10% and 50% of sequence matches. The rationale for this choice was to ensure a low amount of false positives in the immunogenicity prediction for the 10% I _CR index, and to compare it with the 50% value containing medium to high immunogenic peptides. Two groups - ‘Infectious diseases’ and ‘Healthy’ - have average indices close to zero on both plots, indicating low amounts of cross-reactive epitopes in the critical tissues. The groups ‘Autoimmune diseases’ and ‘Cancer’ exhibit approximately 2- to 5-fold higher average index values compared to the ‘Healthy’ group, in each plot respectively, corresponding to considerably higher presentation level of the cross-reactive peptides in these states.

The interpretation of these results is as follows. The epitopes in the ‘Infectious diseases’ group are derived from non-human organisms rather than from human hosts. Thus, compared to the epitopes from the other three groups, which are of human origin, a lower I _CR index is expected, implying low sequence identity to the host and thus a low probability of CR. The slightly elevated index for the ‘Healthy’ group is most likely due to the presence of common pathogens (such as Herpes simplex virus or Epstein-Barr virus) mimicking human sequences, an immune escape strategy known as immune camouflage [19]. A higher I _CR for the ‘Autoimmune’ group compared to the ‘Healthy’ group is not surprising, as autoimmunity is a response of the human body’s immune system directed against human proteins overexpressed or aberrantly presented in healthy tissues. For example, multiple sclerosis, the most frequently occurring disease in this group, is due to autoimmunity to the myelin basic protein (MBP), expressed in the tissues of the central nervous system [19]. Other epitopes in this group with very high index values are derived, e.g. from the proteins actin, myosin-9, septin-2 and vimentin, which are normally expressed in various tissues. Normally, peripheral T cells are trained to recognize pathogen-derived epitopes and ignore self-antigens, however some T cells escape this selection and are able to recognize self-antigens, thus initiating an autoimmune response and becoming self-reactive. Consequently with respect to autoimmunity, the term CR is defined as the recognition by T cell TCRs of many different peptide antigens, presented by the HLA of an individual [20], which can also be referred to as cross-recognition.

The significantly higher CR index for the cancer group compared to the other three groups indicates a presence of a high background level of CR when targeting cancers. Cancer epitopes originate either from wild-type proteins overexpressed in tumors, or as a result of cancer-specific mutations in the genes, named neoepitopes. On average, neoepitopes have lower similarity to self-antigens compared to the wild-type cancer epitopes, thus potentially are less cross-reactive. Since T cells with TCRs binding to self-antigens are negatively selected in the thymus, there will generally be a lack of the T cells that can fight tumors, producing overexpressed wild-type proteins. In contrast, the cancers producing neoepitopes can be effectively controlled by the immune system provided that suitable T cells are available. Thus, different types of cancer produce epitopes of varying cross-reactivity, which explains the larger variance seen in Fig. 2 for the cancer group compared to the other groups.

High I _CR for the ‘Autoimmune disease’ and ‘Cancer’ groups may also be due to an activated state of the immune system, when immunoproteasomes create larger amounts of immunogenic (in comparison to standard proteasomes) epitopes, including those from the residuals of normal cells killed by the immune system [21]. In addition, disruption of the normal functioning of the ubiquitin proteasome system may result in creation of abnormally presented immunogenic epitopes, leading to many types of disorders, including malignancies, neurodegenerative diseases and systemic autoimmunity [22, 23].

Thus, multiple reasons for a high variability in presented CR epitopes appear to exist depending on the host disease state. This CR, which we tentatively call “background” CR, is independent of any immune therapy. Clearly, a collection of epitopes present in a particular individual is different from our datasets obtained from the IEDB database. Likely, it will include only a subset of the peptides, but a statistical distribution in many patients may exhibit a pattern similar to the one reported in this work. Eventually, it remains to be seen if there can be any interference between the background CR and the CR invoked by a therapy, but both types are important to assess the safety of the therapy.