Johan Elf

Antibiotic resistance is an increasingly serious problem worldwide. Identifying the bacterial species can help guide antibiotic selection and improve patient outcomes. In sepsis, where every minute counts, rapid identification is particularly important, and new approaches are needed. In this study, deep learning models were trained to identify bacterial species from phase-contrast time-lapse images of cells growing in microfluidic traps, without fluorescent labels. We demonstrated this approach in two previous studies using laboratory strains. Here, the previous work is extended to clinical bacterial isolates, evaluating performance on isolates not seen during training. Furthermore, it was tested whether autofluorescence could aid species identification. Two validation sets were acquired in separate experiments from the training data, the first containing clinical-isolate traps with five species ( Escherichia coli , Enterococcus faecalis , Kleb-siella pneumoniae , Staphylococcus aureus and Enterobacter cloacae ), imaged every two minutes, and the second containing clinical isolates of two species ( E. coli and K. pneumoniae ) recovered from spiked donor blood, imaged every minute. Training with the best previous CNN method achieved unsatisfactory results. Therefore, vision transformers were explored, and a self-supervised adaptation procedure was developed for pretrained DINOv3 using a frozen teacher, followed by supervised classification using time-lapse montages. Mean macro-F1 was 93.6 ± 1.3% for isolates precultured in Mueller–Hinton medium using 60 frames (118 minutes) and 87.2 ± 2.7% for isolates precultured in blood using 55 frames (54 minutes) (mean ± sample standard deviation across training seeds). Performance was evaluated over time by increasing the number of frames available to the model during testing, and Fourier-domain filtering approximated reduced numerical aperture. The results are promising and suggest that the aperture can be decreased while retaining good results. However, further evaluation on independent isolates, experiments and patient samples is needed to establish generalization. Together with previous findings, these results bring us closer to rapid image-based diagnostics that could transform the initial treatment of acute bacterial infections by enabling better-informed antibiotic selection.

Transposon mutagenesis is a powerful method to create deep libraries of genetically diverse cells. However, it has previously not been possible to analyze transposon libraries with respect to complex phenotypes as characterized by intracellular spatial dynamics. Here, we use optical pooled screening to characterize a transposon mutagenesis library via live cell single-particle imaging. The library is analyzed in real time, which allows us to use an optical tweezer to isolate cells with interesting phenotypes. We used the method to identify mutants with perturbations in replication initiation control in Escherichia coli , but it can in principle be used to identify genetic elements associated with any type of complex or dynamic single-cell phenotype.

Transcription factors (TFs) efficiently locate their target DNA sequences by combining three-dimensional diffusion and one-dimensional sliding on nonspecific DNA. To balance rapid sliding with strong specific binding, TFs were proposed to switch between search and recognition conformations. For Escherichia coli lac repressor (LacI), the folding of the hinge helices has been implicated in the conformational switch. Here, we tested how mutations in the hinge region impact the search speed and binding stability. Based on molecular dynamics simulations, we selected two LacI mutants favoring either search or recognition conformation. We measured the binding kinetics of the mutants both in vitro on DNA microarrays with 2479 different Lac operators and in vivo via single-molecule experiments. We identified a mutation that enhances the specificity but reduces binding strength globally, and another mutation that makes the operator binding stronger but also reduces the specificity. However, the altered specificity impacts the search time less than expected. Instead, the major effect was impaired dissociation in response to Isopropyl β-D-1-thiogalactopyranoside (IPTG) induction for the strongly binding mutant. Together with earlier reports of affinity–inducibility trade-offs in LacI, our data support the model in which the trade-off is between binding stability and inducibility rather than between speed and binding stability.

Approximately 50 million people suffer from sepsis yearly, and 13 million die from it. For every hour a patient with septic shock is untreated, their survival rate decreases by 8%. Therefore, rapid detection and antibiotic susceptibility profiling of bacterial agents in the blood of sepsis patients are crucial for determining appropriate treatment. Here, we introduce a method to isolate bacteria from whole blood with high separation efficiency through Smart centrifugation, followed by microfluidic trapping and subsequent detection using deep learning applied to microscopy images. We detected, within 2 h, E. coli, K. pneumoniae, or E. faecalis from spiked samples of healthy human donor blood at clinically relevant concentrations as low as 9, 7 and 32 colony-forming units per ml of blood, respectively. However, the detection of S. aureus remains a challenge. This rapid isolation and detection represents a significant advancement towards culture-free detection of bloodstream infections.

Optical pooled screening is an important tool to study dynamic phenotypes for libraries of genetically engineered cells. However, the desired engineering often requires that the barcodes used for in situ genotyping are expressed from the chromosome. This has not previously been achieved in bacteria. Here we describe a method for in situ genotyping of libraries with genomic barcodes in Escherichia coli. The method is applied to measure the intracellular maturation time of 84 red fluorescent proteins.

Drug-resistant tuberculosis (DR-TB) kills ~200,000 people every year. A contributing factor is the slow turnaround time (TAT) associated with drug susceptibility diagnostics. The prevailing gold standard for phenotypic drug susceptibility testing (pDST) takes at least two weeks. Here we show that growth-based pDST for slow-growing mycobacteria can be conducted in 12 h. We use Mycobacterium tuberculosis variant bovis Bacillus Calmette-Guérin (BCG) and Mycobacterium smegmatis as the mycobacterial pathogen models and expose them to antibiotics used in (multidrug-resistant) tuberculosis (TB) treatment regimens - i.e., rifampicin (RIF), isoniazid (INH), ethambutol (EMB), linezolid (LZD), streptomycin (STR), bedaquiline (BDQ), and levofloxacin (LFX). The bacterial growth in a microfluidic chip is tracked by time-lapse phase-contrast microscopy. A deep neural network-based segmentation algorithm is used to quantify the growth rate and to determine how the strains responded to drug treatments. Most importantly, a panel of susceptible and resistant M. bovis BCG are tested at critical concentrations for INH, RIF, STR, and LFX. The susceptible strains could be identified in less than 12 h. These findings are comparable to what we expect for pathogenic M. tuberculosis as they share 99.96% genetic identity.

Transcription factors (TFs) efficiently locate their target DNA sequences by combining three-dimensional diffusion and one-dimensional sliding on nonspecific DNA. To balance rapid sliding with strong specific binding, TFs were proposed to switch between search and recognition conformations. For E. coli lac repressor (LacI), the folding of the hinge helices has been implicated in the conformational switch. Here, we tested how mutations in the hinge region impact the search speed and binding stability. Based on molecular dynamics simulations, we selected two LacI mutants favoring either search or recognition conformation. We measured the binding kinetics of the mutants both in vitro on DNA microarrays with 2,479 different Lac operators and in vivo via single-molecule experiments. We conclude that a hinge region mutation causing less helix propensity enhances the specificity but reduces binding strength globally, while a hinge region mutation causing higher helix propensity has opposite effects. However, altered specificity impacts the search time less than expected. Instead, the major effect was impaired dissociation in response to IPTG induction for the strongly binding mutant. Together with earlier reports of affinity-inducibility trade-offs in LacI, our data support the model in which the hinge region governs a trade-off between binding stability and inducibility rather than between speed and binding stability.

The blood of a patient with bacteremia contains only a few bacteria per milliliter. Recently, techniques have been developed for extracting these few bacteria from whole blood. Here, we evaluate if a phenotypic Antibiotic Susceptibility Test (AST) can be performed on a single bacterial cell by monitoring changes in its phenotypic distributions sampled over time rather than across populations, while also accounting for experimental noise and cell-to-cell variability. We use the method to estimate the time required to correctly classify single susceptible or resistant bacteria when treated with antibiotics that affect growth, cell division, or cause lysis. We find that due to heterogeneous single cell responses, it is necessary to consider all bacteria close to the susceptibility breakpoint as resistant. However, it is possible to follow up resistance calls by sequentially exposing the cell or its lineage to new antibiotics until a susceptible phenotype is detected.

The rate at which transcription factors (TFs) bind their cognate sites has long been assumed to be limited by diffusion, and thus independent of binding site sequence. Here, we systematically test this assumption using cell-to-cell variability in gene expression as a window into the in vivo association and dissociation kinetics of the model transcription factor LacI. Using a stochastic model of the relationship between gene expression variability and binding kinetics, we performed single-cell gene expression measurements to infer association and dissociation rates for a set of 35 different LacI binding sites. We found that both association and dissociation rates differed significantly between binding sites, and moreover observed a clear anticorrelation between these rates across varying binding site strengths. These results contradict the long-standing hypothesis that TF binding site strength is primarily dictated by the dissociation rate, but may confer the evolutionary advantage that TFs do not get stuck in near-operator sequences while searching.