David R. Bell
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
David R. Bell has a total of 41 co-authors. Here are the top co-authors by number of shared publications:
- Ruhong Zhou
Zhejiang University
8 shared publications - Serena H. Chen
Oak Ridge National Laboratory
4 shared publications Yi Song
Institute of Quantitative Biology, College of Life Sciences, Zhejiang University, Hangzhou 310058, China
2 shared publications- Abdel Rahiam A. Hamad
Department of Pathology, Johns Hopkins University School of Medicine, Baltimore, Maryland, USA
1 shared publication Baoyu Li
School for Radiological and Interdisciplinary Sciences (RAD-X), Collaborative Innovation Center of Radiation Medicine of Jiangsu Higher Education Institutions, Soochow University, Suzhou, China
1 shared publicationBarbara Dabek
1 shared publication
Latest Publications
- 1 January 2025
- Published by Springer Nature
- p. 1061-1074
Machine Learning for Protein Structure Prediction and Design
- 1 January 2025
- Published by Springer Nature
- p. 1061-1074
Machine Learning for Protein Structure Prediction and Design
- 1 April 2024
- Published by American Chemical Society (ACS) in Chemistry of Materials
- Vol. 36  (8) , 3861-3874
Impact of Glycosylation on Self-Assembled Peptide Hydrogels
- 1 April 2024
- Published by American Chemical Society (ACS) in Chemistry of Materials
- Vol. 36  (8) , 3861-3874
Impact of Glycosylation on Self-Assembled Peptide Hydrogels
- Open Access
Crystal Structure of a Chimeric Antigen Receptor (CAR) scFv Domain Rearrangement Forming a VL-VL Dimer
Chimeric Antigen Receptor T-cell (CAR-T) immunotherapies are dependent upon designed transmembrane proteins to bind target antigens and stimulate an immune response. The success or failure of these CARs is only partially predictable, yet recent work has highlighted the importance of antigen binding scFvs driving distinct oligomerization states with varied CAR-T efficacy. Here, we sought to determine the extracellular structure of the anti-CD19 CAR 47G4-CD828Z. Unexpectedly, the resolved crystal structure revealed an IgVL homodimer bound along an inverted VL|VL interface. We found that the VL-VH linker, designed to be cleavage resistant, was cleaved, and the VH and CAR hinge domains were absent from the crystal structure lattice. Molecular Dynamics simulations revealed that the inverted VL|VL interface was more stable than the canonical VL|VL configuration. Our work substantiates the need to interrogate the scFv structure and CAR oligomerization state for optimal CAR-T design.
Open AccessCrystal Structure of a Chimeric Antigen Receptor (CAR) scFv Domain Rearrangement Forming a VL-VL Dimer
Chimeric Antigen Receptor T-cell (CAR-T) immunotherapies are dependent upon designed transmembrane proteins to bind target antigens and stimulate an immune response. The success or failure of these CARs is only partially predictable, yet recent work has highlighted the importance of antigen binding scFvs driving distinct oligomerization states with varied CAR-T efficacy. Here, we sought to determine the extracellular structure of the anti-CD19 CAR 47G4-CD828Z. Unexpectedly, the resolved crystal structure revealed an IgVL homodimer bound along an inverted VL|VL interface. We found that the VL-VH linker, designed to be cleavage resistant, was cleaved, and the VH and CAR hinge domains were absent from the crystal structure lattice. Molecular Dynamics simulations revealed that the inverted VL|VL interface was more stable than the canonical VL|VL configuration. Our work substantiates the need to interrogate the scFv structure and CAR oligomerization state for optimal CAR-T design.
- Open Access
- 11 April 2023
- Published by National Academy of Sciences in Proceedings of the National Academy of Sciences
- Vol. 120  (16)
A mutagenesis study of autoantigen optimization for potential T1D vaccine design
A previously reported autoreactive antigen, termed the X-idiotype, isolated from a unique cell population in Type 1 diabetes (T1D) patients, was found to stimulate their CD4+ T cells. This antigen was previously determined to bind more favorably than insulin and its mimic (insulin superagonist) to HLA-DQ8, supporting its strong role in CD4+ T cell activation. In this work, we probed HLA-X-idiotype-TCR binding and designed enhanced-reactive pHLA-TCR antigens using an in silico mutagenesis approach which we functionally validated by cell proliferation assays and flow cytometry. From a combination of single, double, and swap mutations, we identified antigen-binding sites p4 and p6 as potential mutation sites for HLA binding affinity enhancement. Site p6 is revealed to favor smaller but more hydrophobic residues than the native tyrosine, such as valine (Y6V) and isoleucine (Y6I), indicating a steric mechanism in binding affinity improvement. Meanwhile, site p4 methionine mutation to hydrophobic residues isoleucine (M4I) or leucine (M4L) modestly increases HLA binding affinity. Select p6 mutations to cysteine (Y6C) or isoleucine (Y6I) exhibit favorable TCR binding affinities, while a swap p5-p6 tyrosine–valine double mutant (V5Y_Y6V) and a p6-p7 glutamine–glutamine double mutant (Y6Q_Y7Q) exhibit enhanced HLA binding affinity but weakened TCR affinity. This work holds relevance to potential T1D antigen-based vaccine design and optimization.
- 11 April 2023
- Published by National Academy of Sciences in Proceedings of the National Academy of Sciences
- Vol. 120  (16)
Open AccessA mutagenesis study of autoantigen optimization for potential T1D vaccine design
A previously reported autoreactive antigen, termed the X-idiotype, isolated from a unique cell population in Type 1 diabetes (T1D) patients, was found to stimulate their CD4+ T cells. This antigen was previously determined to bind more favorably than insulin and its mimic (insulin superagonist) to HLA-DQ8, supporting its strong role in CD4+ T cell activation. In this work, we probed HLA-X-idiotype-TCR binding and designed enhanced-reactive pHLA-TCR antigens using an in silico mutagenesis approach which we functionally validated by cell proliferation assays and flow cytometry. From a combination of single, double, and swap mutations, we identified antigen-binding sites p4 and p6 as potential mutation sites for HLA binding affinity enhancement. Site p6 is revealed to favor smaller but more hydrophobic residues than the native tyrosine, such as valine (Y6V) and isoleucine (Y6I), indicating a steric mechanism in binding affinity improvement. Meanwhile, site p4 methionine mutation to hydrophobic residues isoleucine (M4I) or leucine (M4L) modestly increases HLA binding affinity. Select p6 mutations to cysteine (Y6C) or isoleucine (Y6I) exhibit favorable TCR binding affinities, while a swap p5-p6 tyrosine–valine double mutant (V5Y_Y6V) and a p6-p7 glutamine–glutamine double mutant (Y6Q_Y7Q) exhibit enhanced HLA binding affinity but weakened TCR affinity. This work holds relevance to potential T1D antigen-based vaccine design and optimization.
- 7 July 2022
- Published by American Chemical Society (ACS) in The Journal of Physical Chemistry B
- Vol. 126  (28) , 5151-5160
Binding Affinity Calculations of Gluten Peptides to HLA Risk Modifiers: DQ2.5 versus DQ7.5
Free energy perturbation (FEP) calculations can predict relative binding affinities of an antigen and its point mutants to the same human leukocyte antigen (HLA) with high accuracy (e.g., within 1.0 kcal/mol to experiment); however, a more challenging task is to compare binding affinities of wholly different antigens binding to completely different HLAs using FEP. Researchers have used a variety of different FEP schemes to compute and compare absolute binding affinities, with varied success. Here, we propose and assess a unifying scheme to compute the relative binding affinities of different antigens binding to completely different HLAs using absolute binding affinity FEP calculations. We apply our affinity calculation technique to HLA-antigen-T-cell receptor (TCR) systems relevant to celiac disease (CeD) by investigating binding affinity differences between HLA-DQ2.5 (enhanced CeD risk) and HLA-DQ7.5 (CeD protective) in the binary (HLA-gliadin) and ternary (HLA-gliadin-TCR) binding complexes for three gliadin derived epitopes: glia-α1, glia-α2, and glia-ω1. Based on FEP calculations with our carefully designed thermodynamic cycles, we demonstrate that HLA-DQ2.5 has higher binding affinity than HLA-DQ7.5 for gliadin and enhanced binding affinity with a common TCR, agreeing with known results that the HLA-DQ2.5 serotype exhibits increased risk for CeD. Our findings reveal that our proposed absolute binding affinity FEP method is appropriate for predicting HLA binding for disparate antigens with different genotypes. We also discuss atomic-level details of HLA genotypes interacting with gluten peptides and TCRs in regard to the pathogenesis of CeD.
- 7 July 2022
- Published by American Chemical Society (ACS) in The Journal of Physical Chemistry B
- Vol. 126  (28) , 5151-5160
Binding Affinity Calculations of Gluten Peptides to HLA Risk Modifiers: DQ2.5 versus DQ7.5
Free energy perturbation (FEP) calculations can predict relative binding affinities of an antigen and its point mutants to the same human leukocyte antigen (HLA) with high accuracy (e.g., within 1.0 kcal/mol to experiment); however, a more challenging task is to compare binding affinities of wholly different antigens binding to completely different HLAs using FEP. Researchers have used a variety of different FEP schemes to compute and compare absolute binding affinities, with varied success. Here, we propose and assess a unifying scheme to compute the relative binding affinities of different antigens binding to completely different HLAs using absolute binding affinity FEP calculations. We apply our affinity calculation technique to HLA-antigen-T-cell receptor (TCR) systems relevant to celiac disease (CeD) by investigating binding affinity differences between HLA-DQ2.5 (enhanced CeD risk) and HLA-DQ7.5 (CeD protective) in the binary (HLA-gliadin) and ternary (HLA-gliadin-TCR) binding complexes for three gliadin derived epitopes: glia-α1, glia-α2, and glia-ω1. Based on FEP calculations with our carefully designed thermodynamic cycles, we demonstrate that HLA-DQ2.5 has higher binding affinity than HLA-DQ7.5 for gliadin and enhanced binding affinity with a common TCR, agreeing with known results that the HLA-DQ2.5 serotype exhibits increased risk for CeD. Our findings reveal that our proposed absolute binding affinity FEP method is appropriate for predicting HLA binding for disparate antigens with different genotypes. We also discuss atomic-level details of HLA genotypes interacting with gluten peptides and TCRs in regard to the pathogenesis of CeD.
- Open Access
Evolution of Thyroglobulin Loop Kinetics in EpCAM
Epithelial cell-activating molecule (EpCAM) is an important cancer biomarker and therapeutic target given its elevated expression in epithelial cancers. EpCAM is a type I transmembrane protein that forms cis-dimers along the thyroglobulin type-1A-like domain (TYD) in the extracellular region. The thyroglobulin loop (TY loop) within the TYD is structurally dynamic in the monomer state of human EpCAM, binding reversibly to a TYD site. However, it is not known if this flexibility is prevalent across different species. Here, we conduct over 17 μs of all-atom molecular dynamics simulations to study EpCAM TY loop kinetics of five different species, including human, mouse, chicken, frog, and fish. We find that the TY loop remains dynamic across evolution. In addition to the TYD binding site, we discover a second binding site for the TY loop in the C-terminal domain (CTD). Calculations of the dissociation rate constants from the simulation trajectories suggest a differential binding pattern of fish EpCAM and other organisms. Whereas fish TY loop has comparable binding for both TYD and CTD sites, the TY loops of other species preferably bind the TYD site. A hybrid construct of fish EpCAM with human TY loop restores the TYD binding preference, suggesting robust effects of the TY loop sequence on its dynamic behavior. Our findings provide insights into the structural dynamics of EpCAM and its implication in physiological functions.
Open AccessEvolution of Thyroglobulin Loop Kinetics in EpCAM
Epithelial cell-activating molecule (EpCAM) is an important cancer biomarker and therapeutic target given its elevated expression in epithelial cancers. EpCAM is a type I transmembrane protein that forms cis-dimers along the thyroglobulin type-1A-like domain (TYD) in the extracellular region. The thyroglobulin loop (TY loop) within the TYD is structurally dynamic in the monomer state of human EpCAM, binding reversibly to a TYD site. However, it is not known if this flexibility is prevalent across different species. Here, we conduct over 17 μs of all-atom molecular dynamics simulations to study EpCAM TY loop kinetics of five different species, including human, mouse, chicken, frog, and fish. We find that the TY loop remains dynamic across evolution. In addition to the TYD binding site, we discover a second binding site for the TY loop in the C-terminal domain (CTD). Calculations of the dissociation rate constants from the simulation trajectories suggest a differential binding pattern of fish EpCAM and other organisms. Whereas fish TY loop has comparable binding for both TYD and CTD sites, the TY loops of other species preferably bind the TYD site. A hybrid construct of fish EpCAM with human TY loop restores the TYD binding preference, suggesting robust effects of the TY loop sequence on its dynamic behavior. Our findings provide insights into the structural dynamics of EpCAM and its implication in physiological functions.
- 10 August 2021
- Published by American Chemical Society (ACS) in Journal of Chemical Information and Modeling
- Vol. 61  (10) , 4857-4867
Toward Guided Mutagenesis: Gaussian Process Regression Predicts MHC Class II Antigen Mutant Binding
Antigen-specific immunotherapies (ASI) require successful loading and presentation of antigen peptides into the major histocompatibility complex (MHC) binding cleft. One route of ASI design is to mutate native antigens for either stronger or weaker binding interaction to MHC. Exploring all possible mutations is costly both experimentally and computationally. To reduce experimental and computational expense, here we investigate the minimal amount of prior data required to accurately predict the relative binding affinity of point mutations for peptide-MHC class II (pMHCII) binding. Using data from different residue subsets, we interpolate pMHCII mutant binding affinities by Gaussian process (GP) regression of residue volume and hydrophobicity. We apply GP regression to an experimental data set from the Immune Epitope Database, and theoretical data sets from NetMHCIIpan and Free Energy Perturbation calculations. We find that GP regression can predict binding affinities of nine neutral residues from a six-residue subset with an average R2 coefficient of determination value of 0.62 ± 0.04 (±95% CI), average error of 0.09 ± 0.01 kcal/mol (±95% CI), and with an receiver operating characteristic (ROC) AUC value of 0.92 for binary classification of enhanced or diminished binding affinity. Similarly, metrics increase to an R2 value of 0.69 ± 0.04, average error of 0.07 ± 0.01 kcal/mol, and an ROC AUC value of 0.94 for predicting seven neutral residues from an eight-residue subset. Our work finds that prediction is most accurate for neutral residues at anchor residue sites without register shift. This work holds relevance to predicting pMHCII binding and accelerating ASI design.
- 10 August 2021
- Published by American Chemical Society (ACS) in Journal of Chemical Information and Modeling
- Vol. 61  (10) , 4857-4867
Toward Guided Mutagenesis: Gaussian Process Regression Predicts MHC Class II Antigen Mutant Binding
Antigen-specific immunotherapies (ASI) require successful loading and presentation of antigen peptides into the major histocompatibility complex (MHC) binding cleft. One route of ASI design is to mutate native antigens for either stronger or weaker binding interaction to MHC. Exploring all possible mutations is costly both experimentally and computationally. To reduce experimental and computational expense, here we investigate the minimal amount of prior data required to accurately predict the relative binding affinity of point mutations for peptide-MHC class II (pMHCII) binding. Using data from different residue subsets, we interpolate pMHCII mutant binding affinities by Gaussian process (GP) regression of residue volume and hydrophobicity. We apply GP regression to an experimental data set from the Immune Epitope Database, and theoretical data sets from NetMHCIIpan and Free Energy Perturbation calculations. We find that GP regression can predict binding affinities of nine neutral residues from a six-residue subset with an average R2 coefficient of determination value of 0.62 ± 0.04 (±95% CI), average error of 0.09 ± 0.01 kcal/mol (±95% CI), and with an receiver operating characteristic (ROC) AUC value of 0.92 for binary classification of enhanced or diminished binding affinity. Similarly, metrics increase to an R2 value of 0.69 ± 0.04, average error of 0.07 ± 0.01 kcal/mol, and an ROC AUC value of 0.94 for predicting seven neutral residues from an eight-residue subset. Our work finds that prediction is most accurate for neutral residues at anchor residue sites without register shift. This work holds relevance to predicting pMHCII binding and accelerating ASI design.
- Open Access
Towards guided mutagenesis: Gaussian process regression predicts MHC class II antigen mutant binding
Antigen-specific immunotherapies (ASI) require successful loading and presentation of antigen peptide into the major histocompatibility complex (MHC) binding cleft. One route of ASI design is to mutate native antigens for either stronger or weaker binding interaction to MHC. Exploring all possible mutations is costly both experimentally and computationally. To reduce experimental and computational expense, here we investigate the minimal amount of prior data required to accurately predict the relative binding affinity of point mutations for peptide-MHC class II (pMHCII) binding. Using data from different residue subsets, we interpolate pMHCII mutant binding affinities by Gaussian process (GP) regression of residue volume and hydrophobicity. We apply GP regression to an experimental dataset from the Immune Epitope Database, and theoretical datasets from NetMHCIIpan and Free Energy Perturbation calculations. We find that GP regression can predict binding affinities of 9 neutral residues from a 6-residue subset with an average R 2 coefficient of determination value of 0.62 ± 0.04 (±95% CI), average error of 0.09 ± 0.01 kcal/mol (±95% CI), and with an ROC AUC value of 0.92 for binary classification of enhanced or diminished binding affinity. Similarly, metrics increase to an R 2 value of 0.69 ± 0.04, average error of 0.07 ± 0.01 kcal/mol, and an ROC AUC value of 0.94 for predicting 7 neutral residues from an 8-residue subset. Our work finds that prediction is most accurate for neutral residues at anchor residue sites without register shift. This work holds relevance to predicting pMHCII binding and accelerating ASI design.
Open AccessTowards guided mutagenesis: Gaussian process regression predicts MHC class II antigen mutant binding
Antigen-specific immunotherapies (ASI) require successful loading and presentation of antigen peptide into the major histocompatibility complex (MHC) binding cleft. One route of ASI design is to mutate native antigens for either stronger or weaker binding interaction to MHC. Exploring all possible mutations is costly both experimentally and computationally. To reduce experimental and computational expense, here we investigate the minimal amount of prior data required to accurately predict the relative binding affinity of point mutations for peptide-MHC class II (pMHCII) binding. Using data from different residue subsets, we interpolate pMHCII mutant binding affinities by Gaussian process (GP) regression of residue volume and hydrophobicity. We apply GP regression to an experimental dataset from the Immune Epitope Database, and theoretical datasets from NetMHCIIpan and Free Energy Perturbation calculations. We find that GP regression can predict binding affinities of 9 neutral residues from a 6-residue subset with an average R 2 coefficient of determination value of 0.62 ± 0.04 (±95% CI), average error of 0.09 ± 0.01 kcal/mol (±95% CI), and with an ROC AUC value of 0.92 for binary classification of enhanced or diminished binding affinity. Similarly, metrics increase to an R 2 value of 0.69 ± 0.04, average error of 0.07 ± 0.01 kcal/mol, and an ROC AUC value of 0.94 for predicting 7 neutral residues from an 8-residue subset. Our work finds that prediction is most accurate for neutral residues at anchor residue sites without register shift. This work holds relevance to predicting pMHCII binding and accelerating ASI design.
- 8 August 2019
- Published by Royal Society of Chemistry (RSC) in Physical Chemistry Chemical Physics
- Vol. 21  (34) , 18569-18576
Exploration of HIV-1 fusion peptide–antibody VRC34.01 binding reveals fundamental neutralization sites
VRC34.01 antibody binding to a vulnerable site of HIV envelope glycoprotein (Env), the gp41 fusion peptide, renders robust HIV neutralization, but several critical mutations decrease binding affinity and result in unbinding.
- 8 August 2019
- Published by Royal Society of Chemistry (RSC) in Physical Chemistry Chemical Physics
- Vol. 21  (34) , 18569-18576
Exploration of HIV-1 fusion peptide–antibody VRC34.01 binding reveals fundamental neutralization sites
VRC34.01 antibody binding to a vulnerable site of HIV envelope glycoprotein (Env), the gp41 fusion peptide, renders robust HIV neutralization, but several critical mutations decrease binding affinity and result in unbinding.
- 4 January 2019
- Published by AIP Publishing in The Journal of Chemical Physics
- Vol. 150  (1) , 015102
Exploring the binding mechanism between human profilin (PFN1) and polyproline-10 through binding mode screening
The large magnitude of protein-protein interaction (PPI) pairs within the human interactome necessitates the development of predictive models and screening tools to better understand this fundamental molecular communication. However, despite enormous efforts from various groups to develop predictive techniques in the last decade, PPI complex structures are in general still very challenging to predict due to the large number of degrees of freedom. In this study, we use the binding complex of human profilin (PFN1) and polyproline-10 (P10) as a model system to examine various approaches, with the aim of going beyond normal protein docking for PPI prediction and evaluation. The potential of mean force (PMF) was first obtained from the time-consuming umbrella sampling, which confirmed that the most stable binding structure identified by the maximal PMF difference is indeed the crystallographic binding structure. Moreover, crucial residues previously identified in experimental studies, W3, H133, and S137 of PFN1, were found to form favorable hydrogen bonds with P10, suggesting a zipping process during the binding between PFN1 and P10. We then explored both regular molecular dynamics (MD) and steered molecular dynamics (SMD) simulations, seeking for better criteria of ranking the PPI prediction. Despite valuable information obtained from conventional MD simulations, neither the commonly used interaction energy between the two binding parties nor the long-term root mean square displacement correlates well with the PMF results. On the other hand, with a sizable collection of trajectories, we demonstrated that the average and minimal rupture works calculated from SMD simulations correlate fairly well with the PMFs (R2 = 0.67), making this a promising PPI screening method.
- 4 January 2019
- Published by AIP Publishing in The Journal of Chemical Physics
- Vol. 150  (1) , 015102
Exploring the binding mechanism between human profilin (PFN1) and polyproline-10 through binding mode screening
The large magnitude of protein-protein interaction (PPI) pairs within the human interactome necessitates the development of predictive models and screening tools to better understand this fundamental molecular communication. However, despite enormous efforts from various groups to develop predictive techniques in the last decade, PPI complex structures are in general still very challenging to predict due to the large number of degrees of freedom. In this study, we use the binding complex of human profilin (PFN1) and polyproline-10 (P10) as a model system to examine various approaches, with the aim of going beyond normal protein docking for PPI prediction and evaluation. The potential of mean force (PMF) was first obtained from the time-consuming umbrella sampling, which confirmed that the most stable binding structure identified by the maximal PMF difference is indeed the crystallographic binding structure. Moreover, crucial residues previously identified in experimental studies, W3, H133, and S137 of PFN1, were found to form favorable hydrogen bonds with P10, suggesting a zipping process during the binding between PFN1 and P10. We then explored both regular molecular dynamics (MD) and steered molecular dynamics (SMD) simulations, seeking for better criteria of ranking the PPI prediction. Despite valuable information obtained from conventional MD simulations, neither the commonly used interaction energy between the two binding parties nor the long-term root mean square displacement correlates well with the PMF results. On the other hand, with a sizable collection of trajectories, we demonstrated that the average and minimal rupture works calculated from SMD simulations correlate fairly well with the PMFs (R2 = 0.67), making this a promising PPI screening method.