Most antibodies have been shown to recognize the three-dimensional structure of epitopes rather than linear AA-sequences of epitopes. Here, when the full-length AA-sequences of a microbial epitopes are found to be identical to those of host protein epitopes, merely having molecular homology at the AA-sequence level could often be insufficient for anti-microbial antibodies to cross-react with host proteins, due to the spatial relationship of the epitope within the entire host protein molecule. In addition, for anti-microbial cross-reactive antibodies to cause cell/tissue damage, the target host epitopes must be on the cell surface for antibody binding. Thus, although Kanduc “discovered” intracellular host protein epitopes possessing molecular mimicries with microbial epitopes, these epitopes cannot be recognized by cross-reactive antibodies, leading to host cell damage.

Theoretically, when an in silico computational analysis shows that the full-length AA-sequences of microbial epitopes are identical to those of human protein epitopes (i.e., molecular mimicry), the microbial proteins have the potential to induce cross-reactive autoantibody production following microbial infections or vaccinations. However, this computational approach has a limitation: Kanduc’s in silico analysis was based on linear AA-sequence databases of protein epitopes, without information on conformational epitopes (such as crystal structures or 3D structure prediction). Thus, antibody binding to the three-dimensional structure cannot be assessed by Kanduc’s methodology. To test clinically relevant molecular mimicry between microbial epitopes and host epitopes, the following immunological assays should be conducted to examine whether anti-microbial antibodies really bind to the host epitopes: an enzyme-linked immunosorbent assay (ELISA) and/or Western blotting with or without adsorption of microbial and host epitopes; cell-based antibody binding assay using cells expressing the target host epitopes; and immunofluorescent antibody binding assay on tissue sections expressing the target host epitopes. These binding assays should be further required to include functional assays using cell lines and animals to determine whether the cross-reacting antibodies can cause antibody-mediated cell and organ damage through molecular mimicry. In our study, using the same approach as Kanduc’s group [16], we examined the antibody epitopes of HPV16L1 and found that none of the entire AA-sequences of the 22 linear epitopes (Table 1) were identical to any AA-sequences of human protein epitopes. Our result was consistent with findings by Kanduc’s group [11, 16], who showed that there were no human protein epitopes identical to the HPV16L1 epitopes, although Kanduc’s group has never emphasized these crucial findings.

Kanduc reported that approximately 2,000 pentapeptides and hexapeptides derived from HPV proteins were screened against all 36,103 human proteins to determine whether each human protein contained at least one identical sequence, and that the corresponding matching rates were 42.5% and 4.7%, respectively. Since peptides are composed of 20 types of amino acids, Kanduc presented the theoretical probability of a given pentapeptide or hexapeptide matching another pentapeptide or hexapeptide as 1/205 and 1/20⁶, respectively. Kanduc then concluded that the observed matching rates were higher than expected based on these probabilities [15]. However, Kanduc’s approach is scientifically flawed because it neither takes into account the length of human proteins nor compares the results with the matching rates obtained using random sequences. In fact, in our analysis, 248 pentapeptides extracted from random amino acid sequences matched 578–972 different human proteins, and 237 hexapeptides matched 28–60 different human proteins (Table 2).

We found that HPV16L1 had fewer partial AA-sequence similarities than HBV and RSV epitopes. Although such “partial” molecular mimicry is irrelevant to the induction of cross-reactive autoantibodies, our results demonstrated that HPV16L1 epitopes were not unique in having a certain number of partial AA-sequence similarities with human protein epitopes, compared with other viral protein epitopes.

We also demonstrated that viral protein epitopes (HPV16L1, HBV, and RSV) had more partial AA-sequence similarities with human protein epitopes than randomly generated AA sequences (i.e., Random array sets 1–4) (Fig. 1). Despite the evolutionary distance, humans and viruses may share specific sequence features that are biologically relevant for protein formation. On the other hand, for HPV16L1 epitope sequences, concordance with human proteins was equivalent to random sequences for AA-sequences of eight or more. Thus, human proteins did not have a highly specific match with the HPV16 L1 epitope sequences, although identical AA-sequences were commonly found at shorter AA-length levels (Fig. 1).

Epidemiologically, HPV vaccinations have been demonstrated not to increase the incidence of immune-mediated diseases or any diseases [21, 22]. Experimentally, no researcher, including Kanduc's group, has ever demonstrated cross-reactivity between the HPV L1 protein and human tissues/proteins by ELISA or other methods using anti-HPV sera, anti-HPV L1 antibodies, or autoantibodies [10, 23]. Clinically, no studies have shown antibody deposition in the organs, increased autoantibody production, or efficacy of immunotherapies in alleged patients with HANS. It is now widely accepted in the scientific community that there is no point in suggesting the risk of autoimmune diseases by demonstrating the sharing of short peptide sequences by in silico analysis without supporting epidemiological data or experimental evidence [24,25,26,27].

Kanduc et al. performed computational analyses on severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) vaccines, using the same computational approach in their HPV vaccine studies, and reported that the SARS-CoV-2 vaccines could be dangerous due to the presence of 5–7 AA-sequences shared between SARS-CoV-2 and human proteins [28,29,30,31]. Kanduc’s group also reported that vaccines against meningococcal B [32], diphtheria toxin [33], RSV [34], and HBV [11, 35] could pose risks due to the presence of 5–7 shared AA-sequences with human proteins; none of these studies, however, experimentally tested whether vaccine-induced antibodies could cross-react with the plausible microbial epitopes or damage/kill the potential target human cell types/organs. As described above, the mere observation that certain AA-sequences of microorganisms have similarities to human proteins should not be used as a scientific basis for asserting vaccine risks.

This study has a limitation. This is an in silico analysis restricted to available databases and does not account for conformational epitopes. In fact, the affinity in the three-dimensional structure cannot be assessed in this current model. However, this limitation also applies to Kanduc's research, and by performing the same analysis as her, we highlighted this limitation in her work.

In conclusion, we demonstrated that the identical AA-sequences to full-length HPV16L1 epitopes were not present among the human protein epitopes. We also demonstrated that HPV16L1 epitopes had fewer partial AA-sequence similarities than HBV and RSV epitopes. Therefore, it is incorrect to assert the risk of HPV vaccinations based on common AA-sequences in the HPV vaccines and human protein epitopes.