PAbFold: Linear Antibody Epitope Prediction using AlphaFold2

Abstract

Defining the binding epitopes of antibodies is essential for understanding how they bind to their antigens and perform their molecular functions. However, while determining linear epitopes of monoclonal antibodies can be accomplished utilizing well-established empirical procedures, these approaches are generally labor- and time-intensive and costly. To take advantage of the recent advances in protein structure prediction algorithms available to the scientific community, we developed a calculation pipeline based on the localColabFold implementation of AlphaFold2 that can predict linear antibody epitopes by predicting the structure of the complex between antibody heavy and light chains and target peptide sequences derived from antigens. We found that this AlphaFold2 pipeline, which we call PAbFold, was able to accurately flag known epitope sequences for several well-known antibody targets (HA / Myc) when the target sequence was broken into small overlapping linear peptides and antibody complementarity determining regions (CDRs) were grafted onto several different antibody framework regions in the single-chain antibody fragment (scFv) format. To determine if this pipeline was able to identify the epitope of a novel antibody with no structural information publicly available, we determined the epitope of a novel anti-SARS-CoV-2 nucleocapsid targeted antibody using our method and then experimentally validated our computational results using peptide competition ELISA assays. These results indicate that the AlphaFold2-based PAbFold pipeline we developed is capable of accurately identifying linear antibody epitopes in a short time using just antibody and target protein sequences. This emergent capability of the method is sensitive to methodological details such as peptide length, AlphaFold2 neural network versions, and multiple-sequence alignment database. PAbFold is available at https://github.com/jbderoo/PAbFold.

Keywords: AlphaFold2; antibody; competition ELISA; epitope-prediction; linear epitope; scFv.

Figure 1.. PAbFold pipeline for linear epitope prediction.

A) Antibody V_H and V_L protein sequences are used to generate scFv sequences, either based on the native antibody sequences or loop grafting complementarity determining regions (CDRs) onto either the 2E2 or 15F11 antibody framework regions (2E2 shown). B) The target antigen sequence is parsed into a list of small overlapping peptide sequences, with peptide step and window size parameters adjusted as needed. Rank ordered peptides are output, and partial epitope sequences are underlined manually to highlight the identification of the correct sequence. C) The scFv sequences from Panel A are co-folded with each of the peptide sequences derived from the target antigen in parallel batch mode on a GPU server. pLDDT scores from each structure prediction experiment are collected and scores are presented in their sliding window, both as a heat map organized along the length of the target antigen sequence and a bar chart that shows the per-peptide average pLDDT (Consensus Method). Additionally, the Simple Max data is presented in the third and final panel.

Figure 2.. The Alphafold2-based PAbFold method predicted the Myc linear epitope in different scFv backbones.

The anti-Myc V_H and V_L antibody sequences were used to generate either A) wild-type Myc scFv or loop grafted chimeric B) Myc-15F11 or C) Myc-2E2 scFv variants. The Myc proto-oncogene protein sequence (Genbank NP_001341799.1) was used as the target antigen and processed into 10 amino acid overlapping peptides with a 1 amino acid sliding window. The structure for each scFv:peptide pair was predicted with AlphaFold2 in batch mode on two NVIDIA A5000 GPUs. Average consensus pLDDT values for each scFv:peptide window are illustrated, as well as the maximum pLDDT observed for each residue in any window (bottom). D) Top ranking binding peptides based on average consensus pLDDT. E) Top ranked binding peptides based on summing per-residue maximum pLDDT. For D and E, underlining represents overlap with the reported Myc epitope (EQKLISEEDL).

Figure 3:. The AlphaFold2-driven PAbFold epitope scan method can accurately identify a linear epitope for a novel SARS-CoV-2 antibody.

Antibody VH and VL sequences for SARS-CoV-2 nucleocapsid protein targeted antibody were used to generate scFv sequences A) WT, B) 15F11, C) 2E2 or native VH and VL sequences D) 3 body). Variant scFv sequence in complex with peptide windows from the SARS-CoV-2 nucleocapsid protein (Genbank Accession: YP_009724397) were subjected to AlphaFold2 structure prediction. The top 5 peptides ranked by either the E) Consensus method or the F) Simple Max method, with the underlined sequence highlighting the experimentally verified sequences and a cartoon schematic for each system shown. G) Competition ELISA schematic for assessing the ability of synthetic peptides derived from the SARS-CoV-2 nucleocapsid protein. H) Amino acid windows showing binding interference, with mBG17 binding to SARS-CoV-2 nucleocapsid protein (n = 3). Percentage of binding values were calculated from the no-peptide control. Alignment of synthetic peptides corresponding to SARS-CoV-2 nucleocapsid a. a. 381–419. Peptide a. a. 401–410, which demonstrated mBG17 competition.

Figure 4.. The Alphafold2-Driven PAbFold method accurately predicts molecular interactions between a linear epitope and a scFv

A) Competition ELISA assessing the ability of synthetic alanine mutant peptides derived from the SARS-CoV-2 nucleocapsid protein (a. a. 401–410: DDFSKQLQQS) to interfere with mBG17 binding to SARS-CoV-2 nucleocapsid protein (n = 3). Percentage of binding values were calculated from the no-peptide control. B) AlphaFold2 model for mBG17–15F11 scFv bound to a. a. 401–410 peptide (the average peptide pLDDT was 83.5). Residues that display sharply reduced binding to mBG17 upon mutation to alanine in competition ELISAs (D2, F3, S4, L7, Q8) are shown as warm-colored thick sticks. Predicted hydrogen bonds between the peptide and the scFv are depicted by yellow bars. Sites where mutation to alanine was less disruptive to binding (Q6A, K5A, S10A, D1A, and Q9A) are depicted as thin sticks with cool colors. The carbon atoms of residues in panel B are colored according to the corresponding data in panel A. C) The same AlphaFold2 model for the mBG17–15F11 scFv bound to a.a. 401–410 colored with confidence (pLDDT) as predicted by AF2.