The results establish a regional genomic baseline for a selected high-risk clinical subset, rather than a population-wide estimate of resistance prevalence. The State Public Health Laboratory of Aguascalientes processed 3,454 E. coli and Klebsiella isolates from regional hospitals during the study period, including 683 E. coli, 679 K. pneumoniae and 92 organisms initially identified by MALDI-TOF mass spectrometry and phenotypic testing as K. oxytoca. Whole-genome sequencing reclassified two of the latter isolates as K. michiganensis.

Of the processed isolates, 246 met the study definitions for multidrug resistance or extensive drug resistance, and 65 were selected for sequencing after confirmation of species identity, availability of pure clinically relevant cultures and satisfaction of DNA-quality requirements. Almost all the sequenced isolates—94%—were recovered from patients in critical-care units, and clinical records classified 97.1% of Klebsiella cases and 93.3% of E. coli cases as severe illness. This referral pattern is central to interpreting the findings: the dataset represents severe, treatment-refractory or otherwise difficult-to-manage infections and is enriched for highly resistant organisms rather than representing community circulation.

p>Phenotypic susceptibility testing showed broad resistance across major antibiotic classes, including several drugs commonly used for urinary, respiratory and inpatient infections. Every sequenced isolate carried extended-spectrum beta-lactamase genes, consistent with cephalosporin resistance reaching 97.1% among the 35 Klebsiella isolates and 100% among the 30 E. coli isolates. Resistance to ciprofloxacin and levofloxacin exceeded 90% in both groups, while trimethoprim-sulfamethoxazole resistance was 82.8% in Klebsiella and 90% in E. coli.

Carbapenem resistance occurred in 31.4% of Klebsiella isolates and 13.3% of E. coli isolates. Amikacin retained activity against more isolates than most tested agents, although resistance was present in 11.4% of the Klebsiella group. Among Klebsiella respiratory isolates, tigecycline resistance increased from 0% in 2022 to 40% in 2023.

The study also found that E. coli XDR phenotypes were confined to the 2023 collection, while later Klebsiella isolates showed more dynamic carbapenem and tigecycline resistance patterns. p>Integrating phenotypic susceptibility results with genomic predictions produced 78.9% overall concordance across 689 antibiotic-class comparisons. Agreement exceeded 90% for fluoroquinolones, sulfonamides and macrolides, whose resistance determinants are comparatively well characterized, and reached 100% for E. coli against cephalosporins and for both macrolide results in Klebsiella. Concordance was substantially lower for aminoglycosides and carbapenems.

Some Klebsiella isolates were phenotypically resistant to carbapenems without carrying canonical blaKPC, blaNDM or blaOXA-48 genes; combinations of OXA-type beta-lactamases, porin disruption and efflux systems were instead associated with the phenotype. These findings identify possible non-classical contributors but do not directly prove their causal effects. Genomic predictions can also be affected by gene expression, regulatory mutations, epistasis, promoter efficiency and laboratory conditions, so the authors emphasize that gene detection alone cannot replace phenotypic susceptibility testing. p>Phylogenomic analysis placed the Aguascalientes genomes within a broader but unevenly sampled Mexican genomic landscape.

The comparative dataset included 268 public genomes: 139 E. coli, 85 K. pneumoniae and 44 K. michiganensis, the last comprising three Mexican isolates and 41 international reference genomes. Mexico City and Puebla were relatively well represented, whereas more than half of Mexico’s states were under-represented or absent. The 30 E. coli isolates from Aguascalientes fell into 15 sequence types and were distributed across the phylogeny rather than forming one local cluster.

They were interspersed with genomes from other states, indicating that the lineages detected locally overlapped with broader national patterns. Because sampling was neither population-based nor geographically uniform, the national comparison cannot establish population prevalence or a continuous national transmission route. The E. coli collection included established high-risk lineages such as ST131 and ST69, which also occurred most frequently among the public Mexican E. coli genomes.

Six isolates could not be assigned to a recognized sequence type; these untyped genomes carried the largest resistance-gene burdens, averaging more than 100 antimicrobial-resistance genes per genome. The blaCTX-M-15 gene was present in 90% of the Aguascalientes isolates, compared with 18% of public Mexican E. coli genomes. Polymyxin-associated regulators, including arnA, eptA, mgrB and members of the pmrB, pmrC and pmrF family, also occurred at higher frequencies in the referral collection.

Detection of these genes establishes their genomic presence, not phenotypic colistin resistance or the transfer of a particular plasmid. Across 169 E. coli genomes, the researchers identified 345 mobile genetic elements in the LESPA isolates, including 26 plasmid-associated elements and 319 of uncertain chromosomal origin. MITEEc1 had the largest number of reported associations, including links with ompT and virulence-associated loci.

p>Resistance genes were frequently located in mobile contexts, a pattern consistent with possible horizontal gene-transfer dynamics in clinical environments. The analysis catalogued 211 mobile genetic elements among the sequenced K. pneumoniae isolates and identified associations involving plasmids, transposons and insertion sequences. K. pneumoniae served as a principal sharing hub in the cross-species network, linking genes including the CTX-M-15, sul1, sul2, tetA, mphA and dfrA families with both E. coli and Klebsiella.

The two K. michiganensis isolates carried NDM-7, VIM-1 and MCR-9.1, which were absent from the sequenced E. coli and K. pneumoniae mobilomes. These observations identify mobile contexts and candidate routes of dissemination, but the study did not perform long-read sequencing or reconstruct complete plasmids. Genomic proximity alone therefore does not demonstrate patient-to-patient transmission or prove that a particular element transferred between species. p>The K. michiganensis isolates expanded the limited genomic record for an undercharacterized member of the K. oxytoca complex in Mexico.

One belonged to ST180 and clustered with genomes from China and the United States, while the other had an unknown sequence type and grouped with Mexican and North American isolates. Their phylogenetic positions were consistent with separate introductions into the regional collection, but the sample of two was too small to establish the species’ broader epidemiology. They were also the first multidrug-resistant K. michiganensis genomes reported from Aguascalientes and occupied separate resistance-gene clusters.

Pangenome analysis showed open, accessory-rich genomes for all three species. Cloud genes, defined as genes present in fewer than 15% of genomes, accounted for 81.7% of the E. coli pangenome, 76.1% of K. pneumoniae and 63.8% of the K. oxytoca complex dataset, illustrating substantial genomic diversity without directly demonstrating its origin or rate of evolution. p>Clinical complexity accompanied the molecular findings. Coinfections were recorded in 60% of Klebsiella cases and 36.7% of E. coli cases, with Acinetobacter baumannii, Enterococcus faecalis and Pseudomonas aeruginosa among the most frequent co-occurring organisms.

Among E. coli, virulence analysis identified seven units found only in the LESPA collection, including a shu hemin-uptake operon, afaC-VIII and galU; no K. pneumoniae virulence units exclusive to the new isolates were found. Temporal comparisons showed richer resistomes, more XDR phenotypes and more frequent coinfections among post-2020 isolates than in the older public collection, including a rise in carbapenem resistance among Klebsiella respiratory isolates from 0% in 2022 to 83% in 2023. These observations cannot be attributed to the COVID-19 pandemic or antibiotic use from this referral-heavy dataset, and they do not demonstrate population-wide temporal change.

The study instead supports routine whole-genome sequencing, standardized metadata and continued phenotypic testing as components of more representative antimicrobial-resistance surveillance across Mexico.

Subject of Research: Genomic surveillance of multidrug-resistant E. coli and Klebsiella in Aguascalientes, Mexico

Article Title: Genomic surveillance uncovers high-risk resistance profiles in Escherichia coli and Klebsiella spp. from Central Mexico

Article References: Lozano Gamboa, M. S., Muñoz Gomez, B., Maldonado Gonzaga, A. Y., Alejo Hernández, M. A., Cazares, A., Diaz, D., Marmolejo, A. P., & Ceapă, C. D. (2026). Genomic surveillance uncovers high-risk resistance profiles in Escherichia coli and Klebsiella spp. from Central Mexico. Frontiers in Cellular and Infection Microbiology, 16, 1939440. https://doi.org/10.3389/fcimb.2026.1939440

Image Credits: AI Generated

DOI: 10.3389/fcimb.2026.1939440

Keywords: antimicrobial resistance, genomic surveillance, Escherichia coli, Klebsiella pneumoniae, Klebsiella michiganensis, whole-genome sequencing, resistome, mobile genetic elements, Mexico, clinical microbiology, Genomic, surveillance

In a study published in International Microbiology, Rasha Assad Assiri of Princess Nourah bint Abdulrahman University and colleagues describe a computational systems immunology framework that integrates subtractive proteomics, reverse vaccinology and immunoinformatics to design a multi-epitope vaccine, or MEV, against K. michiganensis strain ATCC 8724. The approach begins with the bacterium’s complete proteome of 5,483 proteins and systematically filters it down to a handful of promising vaccine targets. Essentiality screening with the Geptop 2.0 server identified 398 proteins the bacterium cannot live without, and comparison against the human proteome using BLASTP, with sequences showing at least 30 percent identity excluded, left 199 non-homologous essential candidates. Subcellular localization prediction, antigenicity scoring with VaxiJen, allergenicity screening with AllerTOP and transmembrane-helix analysis with TMHMM narrowed the field further, ultimately yielding four envelope-associated proteins: the outer-membrane beta-barrel proteins BamA and LptD, the periplasmic lipopolysaccharide transport component LptA, and the outer-membrane lipoprotein LolB.

The choice of targets is biologically deliberate. LptA shuttles lipopolysaccharide, the inflammatory outer-membrane molecule of Gram-negative bacteria, across the periplasm, while LptD delivers LPS to the cell surface. BamA anchors the beta-barrel assembly machinery that builds outer-membrane proteins, and LolB sorts lipoproteins into the outer membrane. All four are essential for bacterial survival, predicted to be antigenic and non-allergenic, and lack similarity to human proteins, reducing the risk of autoimmune cross-reaction. Homologs of these proteins have been reported as conserved and immunogenic targets in several other Gram-negative pathogens, lending external support to their selection. By focusing on envelope-associated machinery rather than variable surface structures, the designers aimed for a vaccine less vulnerable to the antigenic drift that undermines single-antigen subunit vaccines.

From these four proteins, the team predicted cytotoxic T-lymphocyte, helper T-lymphocyte and linear B-cell epitopes using the Immune Epitope Database analysis tools. Eight CTL epitopes survived filtering for non-toxicity, antigenicity, MHC-I binding within the top two percent of predicted binders and non-allergenicity, although six of them carried negative immunogenicity scores on the IEDB scale and were retained only because they met the other predefined criteria. Four HTL epitopes were selected, three of which were predicted to induce interferon-gamma, a cytokine central to cellular immunity against intracellular pathogens; the fourth, ARFNIDSTQVSLTPD, was kept despite its non-inducer prediction because it satisfied all other binding and safety thresholds. Three linear B-cell epitopes, predicted with an artificial-neural-network-based server, completed the immunological repertoire. Population coverage analysis across 78 global population groups estimated that the combined MHC class I and II epitope set would cover 73.80 percent of the world’s population, peaking at 83.22 percent in South Africa and 82.90 percent in England, but dipping to 67.86 percent in South Asia, a regional variation the authors suggest could inform future vaccination prioritization.

Assembling fifteen epitopes into a single molecule requires architectural finesse. Isolated epitopes are weak immunogens, so the researchers grafted the 130-amino-acid 50S ribosomal protein L7/L12 from Mycobacterium tuberculosis, a known immunostimulatory adjuvant, onto the N-terminus of the construct via a rigid, alpha-helical EAAAK linker that spatially separates the adjuvant from the epitope payload. Within the epitope region, AAY linkers separate CTL epitopes to support proteolytic processing, flexible GPGPG linkers rich in glycine and proline space out the HTL epitopes, and short lysine-rich KK linkers keep the B-cell epitopes accessible. The result is a 386-amino-acid chimeric protein with a predicted molecular weight of 40.53 kilodaltons, a theoretical pI of 8.97, an instability index of 28.71 and a GRAVY value of minus 0.334 indicating overall hydrophilicity. VaxiJen assigned the full construct an antigenicity score of 0.9564, comfortably above the 0.5 threshold, and the in silico safety screens classified it as non-toxic and non-allergenic.

Structural credibility was assessed with AlphaFold2 through the ColabFold implementation, followed by refinement in GalaxyRefine. The best model placed 95.6 percent of residues in favored regions of the Ramachandran plot, with 3.8 percent in allowed regions and only 0.3 percent in disallowed regions. ProSA-web returned a Z-score of minus 4.62, consistent with experimentally determined protein structures, and ERRAT reported a quality factor of 99.219. Secondary-structure prediction estimated 34.46 percent alpha helix, 47.93 percent random coil and 17.62 percent beta strand. Notably, the linker regions showed lower pLDDT confidence scores than the structured domains, a caveat the authors carried forward when interpreting downstream docking and dynamics, since flexible segments are inherently harder to model.

To probe whether the vaccine could engage innate immunity, the team docked the construct against chain B of the TLR1-TLR2 heterodimer, a pattern-recognition receptor complex, using ClusPro 2.0. The model drawn from the largest cluster, containing 72 members, was selected for analysis. PDBsum identified 24 hydrogen bonds at the interface, and the PRODIGY server estimated a binding free energy of minus 14.8 kilocalories per mole with a dissociation constant of 1.5 times ten to the minus eleventh molar, values consistent with a favorable predicted interaction. Normal mode analysis with iMODS described a comparatively rigid core with mobile peripheral regions, and a 100-nanosecond molecular dynamics simulation in GROMACS with the CHARMM36m force field showed the complex settling after roughly the first 10 nanoseconds, with the radius of gyration contracting from about 4.2 to 3.7-3.8 nanometers and the solvent-accessible surface area declining from about 580 to 500 square nanometers before stabilizing. The authors are careful to note that these metrics describe structural compatibility under simulated conditions and do not demonstrate receptor activation or downstream immune signaling.

Immune simulation with the C-ImmSim agent-based platform, run with HLA-DRB101:01, HLA-B07:02 and HLA-A*01:01, predicted a coordinated response to a single simulated vaccine dose: rapid antigen clearance accompanied by rising IgM, IgG1 and IgG2 levels, early increases in IL-2 and interferon-gamma, expansion of active and duplicating CD8-positive and CD4-positive T-cell populations, and the persistence of memory-associated helper T cells. The modeled Th1 response predominated over Th2, Th17 and regulatory populations. Codon optimization with JCat produced a Codon Adaptation Index of 0.94 and a GC content of 50 percent, both favorable for expression in Escherichia coli, and in silico cloning into the pET-30a(+) vector between NcoI and XhoI sites yielded a recombinant plasmid of 4,483 base pairs, supporting the theoretical feasibility of recombinant production.

The authors are explicit about the limits of the work. Multi-strain epitope conservation, MM-PBSA binding-energy decomposition, epitope-level cross-reactivity screening against human and microbiome proteins, replicate immune simulations and multi-dose modeling were not performed, and no recombinant expression, in vitro assay or animal study was carried out. Whether responses to the mycobacterial adjuvant might compete with pathogen-directed immunity, a phenomenon known as adjuvant immunodominance, remains an open experimental question. Still, the study represents an early application of an integrated subtractive-proteomics and immunoinformatics pipeline specifically aimed at K. michiganensis, and it arrives amid growing evidence that reverse-vaccinology-designed MEVs can be validated successfully in vitro and in vivo against related pathogens. As carbapenem resistance spreads through hospital and environmental reservoirs alike, computational designs like this one may increasingly serve as the blueprint from which the next generation of antibacterial vaccines is built, provided the laboratory work now required can keep pace with the algorithms.

Subject of Research: Computational immunoinformatics design of a multi-epitope vaccine against multidrug-resistant Klebsiella michiganensis

Article Title: Computational systems immunology and multi-scale modeling for the design of a Multi-Epitope Vaccine (MEV) against emerging multidrug-resistant Klebsiella michiganensis

Article References: Assiri, R. A., Saleh, F. M., Muhammad, S., Fenibo, E. O., & Matambo, T. (2026). Computational systems immunology and multi-scale modeling for the design of a Multi-Epitope Vaccine (MEV) against emerging multidrug-resistant Klebsiella michiganensis. International Microbiology. https://doi.org/10.1007/s10123-026-00903-3

Image Credits: AI Generated

DOI: 10.1007/s10123-026-00903-3

Keywords: Klebsiella michiganensis, multi-epitope vaccine, reverse vaccinology, immunoinformatics, subtractive proteomics, antimicrobial resistance, TLR1 docking, molecular dynamics simulation, epitope prediction, BamA, LptD, codon optimization