Phage screening identifies candidate tuberculosis peptides for HLA-E

An exploratory screen of a Mycobacterium tuberculosis secretome library identified 30 protein fragments associated with the non-classical human immune receptor HLA-E, and computational modeling prioritized 14 of them as fragments capable of fitting within HLA-E’s peptide-binding groove. The study combined phage-display biopanning against immobilized recombinant HLA-E*01:03 with molecular docking and normal mode analysis, producing a structural shortlist that includes fragments from the secreted antigens Mpt64 and Mpt70, the cell-envelope protein RipD, and members of the PE/PPE family. These results identify experimental candidates rather than confirmed T-cell epitopes: the researchers did not show that the fragments bind native HLA-E complexes in cells, are processed from bacterial proteins, or activate CD8+ T cells from people with tuberculosis or HIV.

The work addresses the uneven reach of conventional tuberculosis vaccine strategies. Current BCG vaccination provides variable and limited protection against pulmonary disease in adults, while many immune responses depend on T cells that recognize antigen presented by major histocompatibility complex molecules. Classical HLA class I molecules are highly variable among people, complicating the design of a broadly matched vaccine. HLA-E offers a potentially more conserved route because it has little genetic polymorphism and can present pathogen-derived peptides to conventional or unconventional CD8+ T cells. It is also reported to resist the downregulation of surface HLA class I caused by the HIV-1 Nef protein, an observation that makes HLA-E especially relevant in settings where tuberculosis and HIV coinfection are common.

HLA-E normally has a specialized immunological role, presenting signal peptides from classical HLA class I molecules to receptors on natural killer cells. Earlier studies nevertheless showed that it can display Mycobacterium tuberculosis-derived ligands. The researchers cited computational screening and stabilization assays that identified 69 candidate peptides, including 36 with measurable HLA-E binding and 11 that activated CD8+ T cells. Mass-spectrometry immunopeptidomics had also found 28 HLA-E ligands from 13 M. tuberculosis proteins that were recognized by T cells from people with active tuberculosis, latent infection or no reported infection. However, conventional approaches can disproportionately recover abundant intracellular or structural proteins and miss low-abundance secreted factors that encounter the host early in infection.

To target that gap, the study used a focused library of 8 × 10^3 variants derived from the genome of the extensively drug-resistant M. tuberculosis F15/LAM4/KZN strain. The library displays exported proteins on bacteriophage surfaces and was created through a structural selection process that eliminates inserts lacking a functional bacterial secretion signal. During biopanning, recombinant HLA-E heavy-chain ectodomain was immobilized on microtiter plates at 10 micrograms per milliliter. The phage library was first incubated in uncoated wells to reduce nonspecific binding to the plate and milk proteins, then exposed to HLA-E for 90 minutes. The researchers then applied 10 washes with PBS containing 0.1% Tween 20 and 10 additional PBS washes. Bound phages were eluted with triethylamine, immediately neutralized and propagated in Escherichia coli TG1, producing 36 colonies after a single round of selection.

Colony PCR, bidirectional sequencing and comparison with M. tuberculosis genome databases identified 30 unique sequences representing 30 different proteins. Four candidates and the known HLA-E-restricted Mtb44 control were too short for the modeling protocol, while 26 fragments of 32 to 85 amino acids proceeded to structural analysis. This design deliberately retained full-length secretome fragments rather than forcing them into the eight- to 11-residue peptides commonly presented by HLA-E. Models could therefore test whether a local core sequence might anchor within the groove while flanking residues projected toward solvent, including possible central bulging. The HLA-E*01:03 structure used in docking was evaluated with MolProbity, receiving a score of 1.78 at the 86th percentile, with 97.79% of residues in favored regions and no Ramachandran outliers.

Docking was performed with PatchDock and refined through FireDock, which considers interface flexibility and side-chain optimization. The resulting poses were examined in PyMOL. Twelve candidates positioned outside the canonical peptide-binding groove were classified as potential nonphysiological artifacts and removed. Fourteen candidates were predicted to occupy the groove’s A, B, C, D, E and F pockets, with pockets E and F used most frequently. Pocket F normally accommodates C-terminal anchor residues, whereas pocket E helps position the middle of a peptide. Across several models, pocket F contacts repeatedly involved HLA-E residues Asn77, Ser143, Lys146 and Thr80, while Glu152 and Ser147 contributed to interactions in pocket E. These are predictions from structural models, not measurements of binding affinity or antigen presentation.

Among the 14 prioritized fragments, Rv2972c received the strongest predicted global docking score at −87.49 kilocalories per mole, followed by Mpt64 at −73.05, FadD4 at −72.71 and Rv2578c at −72.27. Mpt70 scored −69.30, while the remaining candidates ranged from −61.38 to −44.55. These scores represent relative computational estimates of shape complementarity and atomic packing, including desolvation, van der Waals and electrostatic components; the paper emphasizes that they are not absolute experimental free energies or dissociation constants. Several fragments nevertheless produced detailed anchoring patterns. The Mpt64 model linked peptide residues Tyr50, Asp47 and Ser46 with HLA-E residues Lys146, Ser147, Ser143 and Asn77. Mpt70 showed a core nine-residue region entering pocket E and bridging pockets A, B and E. Rv2972c was predicted to engage pockets D, E and F simultaneously, while the PE_PGRS14 and PPE47 fragments favored pocket E and contacted His155 in pocket D.

Normal mode analysis added a dynamic measure to the static docking analysis by characterizing large-scale motions in each modeled complex. In this framework, a lower eigenvalue indicates less resistance to deformation and therefore greater predicted flexibility. The crystallographic Mtb44–HLA-E complex served as the reference, with an eigenvalue of 1.37 × 10−4. Each of the 14 candidates had a lower value, ranging from 9.13 × 10−6 for Rv2578c to 3.08 × 10−5 for PPE47. Mpt64 and Mpt70 had values of 1.09 × 10−5 and 1.21 × 10−5, respectively. The authors interpreted this difference as evidence that the longer secretome fragments could form more mechanically plastic complexes than the rigid nine-residue benchmark. B-factor and covariance analyses suggested that stable anchor contacts could remain in place even as the larger complexes underwent coordinated movements between peptide and HLA-E helices.

The findings provide a testable map of possible HLA-E ligands, but several limitations sharply separate the structural predictions from biological efficacy. The commercial recombinant HLA-E used for biopanning was a monomeric partial heavy chain without beta-2-microglobulin and therefore lacked the native trimeric architecture. The study also used only one round of panning, which preserved diversity but did not enrich binders through repeated cycles. A positive phage sequence establishes association with the recombinant protein, not necessarily high-affinity binding to native HLA-E. The docking and normal mode calculations used isolated components and did not model proteasomal processing, TAP-mediated transport, intracellular antigen abundance or the environment of infected macrophages. Only HLA-E*01:03 was studied, and the candidate fragments were much longer than conventional HLA-E ligands. Independent binding assays and functional tests, such as peptide stabilization and interferon-gamma ELISpot assays using donor T cells, are therefore required before the fragments can be described as HLA-E-restricted tuberculosis epitopes or assessed as vaccine candidates.

Even with those boundaries, the study expands the experimental search space for a relatively conserved tuberculosis-antigen pathway. Its clearest result is a workflow that pairs real phage selection with structural filtering, reducing 30 secretome-derived hits to 14 candidates with predicted groove-directed binding. The list spans biologically distinct groups, including the well-known secreted antigens Mpt64 and Mpt70; the mycolate transporter MmpL10; FadD4, which contributes to cell-wall lipid metabolism; Mce2B, associated with virulence and detoxification; the magnesium-acquisition protein MgtC; the peptidoglycan-binding protein RipD; PE/PGRS14 and PPE47; and several conserved hypothetical proteins. Follow-up studies can test these fragments in sequence order of predicted structural compatibility while also determining whether shorter core motifs reproduce the modeled interactions. Until then, the work is best understood as an exploratory screening and prioritization platform for unconventional tuberculosis peptides, not as evidence of a new vaccine or a demonstrated route to CD8+ T-cell activation.

Subject of Research: Exploration of Mycobacterium tuberculosis secretome fragments as candidate HLA-E-binding peptides

Article Title: Identification of candidate HLA-E binding peptides from Mycobacterium tuberculosis using phage secretome library screening and computational modeling

Article References: Masondo, N., Gordon, M. L., & Chiliza, T. E. (2026). Identification of candidate HLA-E binding peptides from Mycobacterium tuberculosis using phage secretome library screening and computational modeling. Frontiers in Cellular and Infection Microbiology, 16, 1724374. https://doi.org/10.3389/fcimb.2026.1724374

Image Credits: AI Generated

DOI: 10.3389/fcimb.2026.1724374

Keywords: Mycobacterium tuberculosis, HLA-E, phage display, peptide screening, molecular docking, normal mode analysis, tuberculosis immunology, HIV-TB coinfection, antigen presentation, secretome, Identification, candidate