🧬 biology

Genomic signatures and evolutionary dynamics of antimicrobial resistance revealed from 3,163 Klebsiella pneumoniae genomes

By analyzing 3,163 Klebsiella pneumoniae genomes, this study reveals that antimicrobial resistance is driven by a complex interplay of mobile genetic elements and core-genome coding variations, establishing a scalable pan-genome framework that significantly outperforms traditional gene-based models in predicting resistance phenotypes and tracking evolutionary dynamics.

Original authors: Hongru Li, Sifei Wang, Jiabin Li, Han Xia, Kai Ye, Yanyan Liu, Xiaofei Yang

Published 2026-09-22

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Hongru Li, Sifei Wang, Jiabin Li, Han Xia, Kai Ye, Yanyan Liu, Xiaofei Yang

Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). ⚕️ This is an AI-generated explanation of a preprint that has not been peer-reviewed. It is not medical advice. Do not make health decisions based on this content. Read full disclaimer

Bacteria are ancient, adaptable survivors, but when they learn to ignore the medicines designed to kill them, the consequences can be severe. This phenomenon, known as antimicrobial resistance, turns common infections into life-threatening crises. For decades, scientists have fought this battle by looking for specific "bad genes" that bacteria steal or develop to defeat drugs. The prevailing idea was that if you could find these genes, you could predict which drugs would fail. However, this view often misses the bigger picture. Bacteria are not just bags of genes; they are complex organisms where the entire genetic makeup, including the tiny changes in their core machinery and the way their genes are arranged, works together to determine survival. Understanding this full picture is essential for tracking how resistance spreads and for finding new ways to stop it.

A team of researchers set out to map this complex landscape using a massive collection of data. They gathered 1,665 samples of Klebsiella pneumoniae, a common and dangerous bacterium, collected over fifteen years from hospitals in Anhui Province, China. To get a complete view, they combined these new samples with nearly 1,500 existing samples from public databases, creating a dataset of over 3,000 genomes. Instead of just hunting for known resistance genes, they examined the entire genetic code of every bacterium. They looked at the "core" genes that every bacterium in the group shares, as well as the "dispensable" genes that some have and others do not. They also scrutinized tiny spelling mistakes in the DNA and the specific order in which genes were arranged on the bacterial chromosomes. By feeding this vast amount of information into a computer system designed to find patterns, they aimed to see what truly drives resistance to seventeen different antibiotics.

The results showed that looking for single genes was not enough. The researchers found that the most accurate way to predict whether a bacterium would resist a drug was to consider its entire genetic profile. Their computer models, which used both the core genetic changes and the presence or absence of extra genes, correctly identified resistance patterns in more than ninety percent of cases. This was a significant improvement over models that only looked at the known resistance genes. The study revealed that resistance is often a team effort, involving a combination of standard genetic mutations and the presence of specific mobile genetic elements that act like delivery trucks for resistance genes.

One of the most striking discoveries was a long-term shift in how these bacteria organized their defenses. The researchers tracked the arrangement of genes responsible for resisting cephalosporins, a major class of antibiotics, over the fifteen-year period. They observed that a specific genetic configuration, involving a gene called blaCTX-M-15 and a mobile element called ISEcp1, gradually replaced an older configuration centered on a different gene, blaCTX-M-14. This was not just a random change; as the new configuration became more common, the bacteria's resistance to cephalosporins grew stronger. This trend was so consistent that it appeared in a separate group of bacteria from a different part of the world, suggesting it is a widespread evolutionary shift. The study suggests that the bacteria are not just acquiring new genes, but are actively reorganizing their genetic neighborhoods to make those genes work better.

The team also uncovered specific structural changes in the bacteria's proteins that influenced how they reacted to drugs. In some cases, a single letter change in the DNA altered the shape of a protein, making the bacterium more or less susceptible to treatment. In another instance, they found that a section of DNA that acts like a repeating pattern had expanded in length, changing the structure of a protein on the bacterial surface. These findings highlight that resistance is driven by a variety of mechanisms, from the acquisition of new tools to the subtle tuning of existing ones. The researchers noted that while their computer models could predict resistance with high accuracy, the specific new patterns they found are clues for further investigation rather than confirmed causes. They represent a new set of targets for scientists to study, offering a more complete understanding of how bacteria evolve to survive.

Ultimately, this work demonstrates that the story of antibiotic resistance is written in the entire genome, not just in a few famous genes. By looking at the full picture, including how genes are arranged and how the core machinery of the bacterium changes over time, scientists can see the evolutionary path of resistance more clearly. The study provides a scalable method for linking genetic data to real-world drug resistance, offering a powerful tool for surveillance. It suggests that to stay ahead of these evolving threats, monitoring programs must look beyond the presence of resistance genes to the mobile elements and genetic contexts that help them spread and function.

Technical Summary: Genomic Signatures and Evolutionary Dynamics of Antimicrobial Resistance in Klebsiella pneumoniae

Problem Statement
Antimicrobial resistance (AMR) in Klebsiella pneumoniae poses a critical global health threat. While acquired resistance genes (e.g., $\beta$ -lactamases) are well-characterized, they fail to fully explain phenotypic variation; isolates with identical resistance genes often exhibit divergent resistance levels. Current surveillance approaches frequently rely on predefined resistance gene panels or focus on high-risk clones, potentially underrepresenting the broader genomic variation—including core-genome variants and accessory elements in non-dominant lineages—that collectively shapes resistance. Furthermore, existing studies often treat genes, variants, and mobile-element contexts as disconnected signals rather than an integrated AMR architecture, leaving a gap between statistical prediction and mechanistic understanding of resistance evolution.

Methodology
The study established a comprehensive pan-genome framework by integrating 1,665 clinical K. pneumoniae isolates collected over 15 years (2006–2020) via the HuiNet surveillance network in Anhui, China, with 1,498 publicly available genomes from PATRIC. This resulted in a dataset of 3,163 genomes.

Data Generation: Clinical isolates underwent Illumina short-read sequencing (1,665 isolates) and, for a subset (491 isolates), Oxford Nanopore long-read sequencing for hybrid assembly. Phenotypic data included susceptibility profiles for 17 antibiotics across eight drug classes.
Pan-genome Construction: Using Panaroo, the authors defined 3,712 core genes (>90% presence) and 18,866 dispensable genes.
Feature Extraction: The study systematically captured two types of genomic features:
- g-features: Presence/absence of dispensable genes.
- v-features: Non-synonymous core-genome variants (SNPs, insertions, deletions).
Machine Learning Framework: A two-stage selection pipeline was employed:
1. Statistical Filtering: Features were retained if they showed significant association ( $p < 0.05$ ) with resistance phenotypes in both Chi-square and Kruskal-Wallis tests.
2. Ensemble Ranking: Candidates were ranked by four machine learning models (Logistic Regression, Random Forest, SVM, XGBoost). A feature was selected as significant if it ranked in the top tier of at least two models.
Model Validation: Soft-voting ensemble models were trained on the combined dataset and validated on an independent PATRIC cohort. Performance was measured by Area Under the Curve (AUC).
Mechanistic and Evolutionary Analysis: Prioritized variants were analyzed using AlphaFold3 and Missense3D for structural consequences. Temporal dynamics were assessed using Cochran-Armitage trend tests and log-odds ratio (LOR) correlations over the 15-year period.

Key Results

Resistance Landscape: The HuiNet cohort exhibited high diversity, with 74.9% of isolates classified as multidrug-resistant (MDR). While carbapenems retained high activity, resistance was prevalent for $\beta$ -lactam/ $\beta$ -lactamase inhibitor combinations and cephalosporins.
Feature Discovery: The analysis identified 369 AMR-associated features (242 g-features and 127 v-features) across all 17 antibiotics. Only 46 of these matched previously known resistance determinants in public databases (CARD, ResFinder, NCBI).
Predictive Performance: Models utilizing the full set of 369 features achieved a mean AUC of 0.92 on the test set and 0.93 on the external validation set. This significantly outperformed models using only known determinants (mean AUC 0.83) or models restricted to only v-features or g-features, demonstrating that core-genome variants and dispensable genes provide complementary predictive information.
Structural Insights: Structural analysis of prioritized variants revealed specific protein alterations. Notable examples include:
- A missense mutation (S178R) in group_258 (a lysoplasmalogenase-like protein) negatively associated with resistance (associated with susceptibility).
- A hexanucleotide repeat expansion in ompA (encoding the major outer-membrane protein OmpA) negatively associated with resistance (associated with susceptibility), suggesting coding repeat variation as a modulator of susceptibility.
Evolutionary Dynamics: A significant temporal shift was observed in cephalosporin resistance. Over the 15-year period, a specific transposition unit, *ISEcp1-blaCTX-M-15*, progressively replaced blaCTX-M-14-associated genomic configurations. This replacement (Pattern P3) was strongly associated with a sustained increase in cephalosporin resistance (CAZ, CXM, CZO), a trend independently validated in a geographically distinct Houston cohort.

Significance and Claims
The paper claims to establish a scalable, pan-genome-based framework that links bacterial genotype to AMR phenotype more comprehensively than current methods. The study demonstrates that resistance evolution is driven not merely by the acquisition of individual resistance determinants but by:

Mobile-element contexts: The genomic organization surrounding resistance genes (e.g., the ISEcp1-blaCTX-M-15 unit) acts as a unit of selection, potentially influencing expression and dissemination.
Core-genome variation: Coding variations in conserved genes, including missense mutations and repeat expansions, contribute significantly to resistance phenotypes.

The authors emphasize that while the identified features are statistically associated and structurally interpretable, they represent correlative signals requiring functional validation to establish definitive causality. The findings suggest that genomic surveillance should expand beyond tracking resistance genes to include the mobile-element contexts and core-genome variants that shape the clinical resistance landscape.

More like this