2 The microbiome as an active partner in human health

Microbial communities inhabit many parts of the human body, including the skin, oral cavity, respiratory tract and urogenital tract. However, the gastrointestinal tract contains by far the largest and most diverse microbial population, comprising trillions of microorganisms that play fundamental roles in digestion, vitamin production, immune maturation and host physiology regulation. The molecules that are released by the microorganisms can act as microbe-associated molecular patterns (MAMPs) and be recognized by specific pattern recognition receptors (PRRs) of the host organism. These molecules provide the immune system with continuous information about the presence and physiological state of bacterial communities. Under normal conditions, the organism is trained to their recognition and the signaling contributes to immune tolerance and homeostasis. However, a condition that disturbs this balance, like the presence of a pathogenic microorganism or the abnormal increase of the release of naturally occurring molecules, may promote reactions of the host organism, such as excessive inflammation or immune dysfunction.

3 Peptidoglycan: More than a structural barrier

Peptidoglycan provides bacteria with mechanical strength, allowing them to withstand the high internal osmotic pressure generated by the cytoplasm. This macromolecule consists of glycan strands formed by alternating residues of N-acetylmuramic acid (MurNAc) and N-acetylglucosamine (GlcNAc); the strands are cross-linked to one another by short peptide stems. Gram-positive (monoderm) and Gram-negative (diderm) bacteria differ markedly in the organization and thickness of their PGN layers. Gram-positive bacteria possess a multilayered PGN matrix (20–80 nm) and is often associated with teichoic and lipoteichoic acids. In contrast, Gram-negative bacteria contain a thinner PGN layer (2–7 nm) located within the periplasmic space between the inner cytoplasmic membrane and an outer membrane enriched in lipopolysaccharides (LPS). The composition of the PGN stem peptide also varies among bacterial taxa as certain patterns are more common among Gram-positive or Gram-negative bacteria. The diversity of PGN modifications observed is so extensive that, historically, the chemical composition of PGN was used as a chemotaxonomic marker to distinguish bacterial genera, a method that has now been replaced by molecular phylogenetic methods.

Rather than functioning as a rigid shell, the bacterial cell wall is continuously remodeled throughout growth and cell division. This remodeling depends on numerous enzymes collectively called autolysins. In the model organism Escherichia coli, there are four groups of autolysins, based on their specificity for different bonds in the PGN backbone: lytic transglycosylases, amidases, endopeptidases and carboxypeptidases. Their coordinated activities create space for cell elongation, septum formation during division, and recycling of old cell wall material. In every generation, approximately half of the PGN of E. coli is remodeled by autolysins, a process named “turnover”. The majority of the PGN fragments generated by autolysins (≈90%) reenter the cytoplasm in order to be recycled for PGN synthesis or for metabolic purposes, while a smaller fraction of the fragments (≈10%) is being released into the external medium. These released fragments become important biological signals that influence bacterial physiology and host immune responses.

4 -lactam antibiotics and peptidoglycan fragment release

-lactam antibiotics—including penicillins, cephalosporins, monobactams, and carbapenems—remain among the most widely prescribed antimicrobial agents worldwide. Their primary mechanism of action consists of inhibiting penicillin-binding proteins responsible for PGN cross-linking. Previous studies have demonstrated that -lactam exposure even at sublethal doses of the antibiotics trigger the increased release of soluble PGN fragments in the external medium of the cells, a phenomenon known as “peptidoglycan storm”. Although PGN is sensed by a variety of PRRs, the most well-defined host sensors are the cytosolic NOD-like receptors NOD1 and NOD2 (NOD: nucleotide-binding oligomerization domain). NOD1 is expressed in intestinal epithelial cells, hepatocytes, macrophages, dendritic cells and neutrophils, whereas NOD2 is mainly expressed in monocytes, macrophages, dendritic cells and Paneth cells. As cytosolic sensors, NOD1 and NOD2 can either detect bacteria that enter the cytosol, or PGN fragments released by bacteria and transported into the cell. Bacterial PGN turnover fragments are released into the medium either in the form of soluble molecules or as part of outer membrane vesicles (OMVs) and can subsequently enter the host cell cytoplasm through multiple uptake mechanisms. Activation of NOD1 and NOD2 by soluble PGN fragments triggers signaling cascades that mainly activate the transcription factor NF-κB inducing the expression of pro inflammatory genes for the production of cytokines and antimicrobial peptides. Alternative pathways include MAPK signaling, autophagy responses and inflammasome activation through interactions with other NLRs, resulting in caspase-1 activation and IL-1 production. This enhanced release of immunostimulatory PGN fragments has been associated with excessive inflammatory responses, the development of chronic inflammatory diseases and the emergence of disorders such as rheumatoid arthritis, psoriasis and other autoinflammatory conditions.

5 Why are autolysins difficult to study?

Despite decades of research, the precise functions of many autolysins remain poorly understood. It is often difficult to assign a distinct function to an autolysin for several reasons. First, the autolysins that bacteria possess appear to have redundant roles and second, an enzyme may have more than one function. E. coli possesses more than twenty enzymes capable of degrading PGN, generating characteristic fragments, with multiple redundant analogous enzymes belonging to the same group. When an enzyme from one group is knocked out through classical deletion approaches, other enzymes of the same group compensate for its loss, masking its biological role. Traditional genetic approaches rely on deleting one gene at a time, however single autolysin gene knockouts rarely produce clear phenotypes due to redundancy. In this context, the rapid generation of mutant strains for multiple autolysin-encoding genes becomes technically demanding, if we want to cover all combinations required to fully elucidate the genetic basis of PGN turnover and the regulation of its release.

6 Developing a multiplex genome editing strategy

To overcome this challenge, our project validated experimentally the genetic constructions of a CRISPR-dCa9 multiplex genome engineering system based on a protocol described by Banno et al. (2018). The idea is that, by applying this system, we would be able to generate a panel of combinatorial mutant strains, since the simultaneous inactivation of all targeted genes across the entire population is unlikely. The principle of the system is based on the expression of a chimeric protein containing a catalytically inactive Cas9 domain (dCas9) fused to a cytidine deaminase base editor toward selected genomic sites, mediated by the expression of a single guide RNA for the specific genes of interest. The target cytidine is originally converted into uracil that will result to the establishment of a C-to-T transition in the genome after DNA replication, introducing a nonsense mutation for the generation of a premature stop codon simultaneously in the targeted genes. Additionally, a uracil-DNA glycosylase inhibitor fused to the chimeric protein suppresses the base excision repair pathway, preventing the removal of uracil intermediates and thereby increasing the efficiency of C-to- T transition fixation. Concurrently, a short degradation tag is appended to decrease the half-life of the fusion protein, reducing the toxicity of the system to tolerable levels for the host bacteria.

The maximum number of genes that can be targeted simultaneously depends on the number of sgRNAs introduced in a single editing round. Given the relatively large number of autolysin-encoding genes in Escherichia coli, together with the structural and functional diversity of the corresponding enzymes, we initially focused on lytic transglycosylase-encoding genes to optimize the system. Since the mutants for lytic transglycosylases do not have a visible phenotype, the validatation of the system’s functionality is done by including an sgRNA targeting a reporter gene, initially lacZ. This strategy enables the visual identification of successfully edited bacteria by blue–white screening on Luria–Bertani (LB) agar supplemented with IPTG and X-gal, where disruption of lacZ results in the formation of white colonies. Sequencing following the experiment confirmed successful introduction of nonsense mutations within the expected editing window. Importantly, each edited colony displayed a different combination of mutations, demonstrating that the system can generate diverse mutant collections in a single experiment. Although our work confirmed existing studies regarding off-target events and decreased efficiency as additional genes are targeted, the study established a practical framework for constructing extensive libraries of autolysin mutants that would be extremely laborious using classical gene deletion techniques.

7 Characterizing autolysin mutants under antibiotic stress

A second objective of the project was to investigate through genetic approaches, the contribution of specific autolysins to PGN fragment release in E. coli by using mutant strains for autolysin-encoding genes. The strains were also exposed to representative antibiotics of the major -lactam classes (penams, cephalosporins, monobactams and carbapenems), grown in the presence of concentrations corresponding to half the minimum inhibitory concentration (MIC) for the antibiotics tested. The aim of these experiments was to evaluate the ability of -lactams to induce the release of PGN fragments in the mutants tested and quantify the latter, so as to compare them to the parental strains. To accomplish this, culture supernatants collected after culture growth were analyzed by LC-MS in the laboratory of Ivo Boneca at the Pasteur Institute, for the identification and quantification of fragments and processed with PGNFinder, a web-based tool for automated PGN fragment detection and annotation.

MIC measurements consistently showed that strains lacking multiple autolysins became more susceptible to antibiotic treatment than their parental strains. Standardized protocols for collecting bacterial culture supernatants were followed and the analysis of the samples was done by LC–MS. Our preliminary results for treatment of the strains with amoxicillin showed that the presence of the antibiotic induced the release of specific fragments that could be identified both by NOD1 and NOD2 receptors. Further experiments with different antibiotics and the validation of the results with multiple biological repetitions and statistical analysis will allow researchers to identify which PGN fragments are released under different genetic and antibiotic conditions. In a biomedical context, of particular interest is the possibility to identify which fragments does -lactam treatment promote, capable of activating NOD receptors and amplifying inflammatory responses. Such information could guide the development of narrow-spectrum antibiotics that selectively eliminate pathogens while minimizing inflammatory side effects caused by beneficial members of the microbiome.

8 Conclusion & Perspectives

Our study highlights the importance of peptidoglycan turnover as a dynamic process that extends far beyond its structural role in maintaining bacterial cell integrity. By combining multiplex CRISPR-based genome editing with LC–MS analysis of released PGN fragments, we established an experimental framework for investigating how individual and combined autolysins contribute to cell wall remodeling and to the release of immunologically active PGN fragments during -lactam treatment. The successful generation of combinatorial lytic transglycosylase mutants demonstrates the potential of multiplex base editing to overcome the genetic redundancy that has long limited functional studies of bacterial autolysins. Moreover, our preliminary analyses indicate that -lactam exposure promotes the release of NOD1- and NOD2-stimulatory PGN fragments and that alterations in the autolysin repertoire influence both antibiotic susceptibility and fragment release. Together, these findings provide new insights into the genetic and molecular basis of the peptidoglycan storm phenomenon.

Several important questions remain to be addressed. Expanding the multiplex editing strategy to include enzymes from more than one autolysin groups of E. coli will enable a more comprehensive understanding of the coordinated networks regulating PGN turnover. Increasing the number of biological replicates and extending LC–MS analyses to additional -lactam antibiotics will further clarify how different antibiotic classes shape the composition and abundance of released PGN fragments. Investigating the fragment dynamics is crucial for understanding host-pathogen interactions, particularly within the gut. Future work should also directly assess the immunostimulatory activity of the identified fragments using cellular models expressing NOD1 and NOD2, making a direct link between fragment composition and host inflammatory responses. Finally, deciphering the molecular mechanisms underlying antibiotic-induced PGN release may guide the development of next-generation antimicrobial therapies that preserve therapeutic efficacy while minimizing unintended immune activation. Such strategies could contribute to reducing inflammatory complications associated with antibiotic treatment and support the design of microbiome-conscious antibiotics with improved clinical safety.

2 What can engineers learn from gliding mammals?

The most useful form of biomimicry is not necessarily literal imitation. A UAV does not need to resemble a flying squirrel visually in order to learn from one. What matters is identifying functional principles and determining whether they solve an engineering problem.

One such principle is deployability. The aerodynamic surface of a gliding mammal becomes effective when the limbs extend and tension the patagium. A technological analogue could be a UAV whose lifting area changes between compact and deployed configurations. Such a vehicle might remain compact during storage, launch, or powered manoeuvres and deploy a larger lifting surface when entering an efficient gliding phase.

A second principle is compliance. Biological membranes are flexible structures rather than perfectly rigid wings. Their geometry can respond to aerodynamic loading, while local skeletal or cartilaginous elements provide support and tension. Flying squirrels, for example, possess specialized structures near the wrist that support the outer region of the patagium (Jackson and Schouten 2012). Rather than reproducing this anatomy directly, engineers can ask whether flexible membranes combined with strategically placed stiffening elements could offer useful combinations of low mass, deployability, and aerodynamic adaptability.

A third principle is morphological control. Animals can alter aerodynamic behaviour by changing limb configuration, membrane tension, body orientation, or tail position. Existing flying-squirrel-inspired robots already illustrate how such concepts may be translated into technology (Figure 2). Researchers have developed multimodal robots using flexible membranes and articulated structures (Shin et al. 2019), flying-squirrel-inspired UAVs with controllable foldable wings (Lee et al. 2025b), and, more recently, the SquirrelDrone, which uses coordinated forelimb, hindlimb, and tail actuation together with a compliant membrane to achieve whole-body morphing (Zheng et al. 2026). These prototypes demonstrate that biological inspiration can influence not only external shape but also the architecture of actuation and control.

3 From biological specimens to engineering models

Identifying an interesting biological mechanism is only the beginning of biomimetic design. Engineers must transform a complex organism into measurable quantities, geometries, models, and ultimately design rules. This translation is particularly challenging for gliding mammals because their aerodynamic performance depends on membrane geometry, limb positions, and the interactions among these structures during flight (Zhao et al. 2019; Zheng et al. 2026).

As a first step in this research, we sought to reconstruct the three-dimensional geometry of different gliding-mammal species in order to characterize their morphology and study it quantitatively. Natural-history collections can play an important role in this process, as museum specimens preserve biological diversity that may be difficult to observe or measure directly in living animals. However, a preserved specimen is not immediately suitable as an aerodynamic model. Taxidermy preservation frequently resulted in partially retracted membranes and limb orientations that did not correspond to natural gliding postures (Thorington and Heaney 1981).

We therefore explored the use of non-invasive photogrammetry using specimens from the collections of the Muséum national d’Histoire naturelle (MNHN) in Paris, as an intermediate step between biological observation and engineering analysis. Each specimen was photographed from multiple viewpoints and digitally reconstructed as a three-dimensional surface. The resulting geometry was then cleaned and dimensionally calibrated before being further processed using digital rigging to approximate a natural gliding posture and reconstruct retracted membrane regions. The overall workflow is illustrated in Figure 3. At the current stage of the research, these three-dimensional models are being used as geometric inputs for aerodynamic analysis and computational fluid dynamics (CFD).

This approach illustrates a broader opportunity for interaction between natural history and engineering. Biological collections need not be considered only as archives of the past; with appropriate digital methods, they can also become sources of quantitative information for the development of future bioinspired technologies.

4 Conclusion: from copying forms to learning strategies

The promise of passive flight naturally connects biomimicry with discussions of energy-efficient technology. A UAV capable of alternating between powered and unpowered flight could, for suitable missions, reduce the use of active propulsion during portions of the flight. Unpowered gliding phases could also reduce propulsion-related noise, which may be advantageous for environmental observation or operation near noise-sensitive areas. Low-speed manoeuvrability and deployable surfaces could be valuable in forests, rugged terrain, or other cluttered environments where conventional fixed-wing aircraft are less appropriate.

Gliding mammals demonstrate that controlled aerial locomotion can emerge from a strong coupling between morphology and aerodynamics. Their deployable membranes, compliant structures, limb-mediated shape changes, and integrated control strategies offer a design space that remains relatively unexplored in aerial robotics.

The objective of biomimetic engineering should not be to construct mechanical replicas of animals. It should be to understand why a biological structure works, identify the physical principle behind it, and determine whether that principle can be translated responsibly into technology.

For future UAVs, this perspective suggests complementing the traditional question; how much power is required to achieve the desired performance?, with another: how much of that performance can first be achieved through morphology?

The answer may lead not only to new deployable-wing aircraft, but also to a broader way of thinking about engineering systems in which structure, aerodynamics, and control are designed together rather than treated as separate problems.

2 Research Valorisation Note

A first-author manuscript arising from the author’s doctoral research, entitled Human Multi-Donor Proteomics Reveals Biologically Active Age-Associated Extracellular Vesicles Released by Aortic Endothelial Cells, has been submitted to Aging Cell and is currently under peer review. As the manuscript has not yet been accepted for publication, it is not presented here as a published research article. The ORCID identifier and accompanying QR code are provided to enable interested readers to follow future publications and other research outputs as they become available.