Phenotypic Antibiotic Susceptibility Testing at the limit of one bacterial cell

Phenotypic Antibiotic Susceptibility Testing at the limit of one bacterial cell | bioRxiv /* */ /* */ <!-- <!-- /*! * yepnope1.5.4 * (c) WTFPL, GPLv2 */ (function(a,b,c){function d(a){return"[object Function]"==o.call(a)}function e(a){return"string"==typeof a}function f(){}function g(a){return!a||"loaded"==a||"complete"==a||"uninitialized"==a}function h(){var a=p.shift();q=1,a?a.t?m(function(){("c"==a.t?B.injectCss:B.injectJs)(a.s,0,a.a,a.x,a.e,1)},0):(a(),h()):q=0}function i(a,c,d,e,f,i,j){function k(b){if(!o&&g(l.readyState)&&(u.r=o=1,!q&&h(),l.onload=l.onreadystatechange=null,b)){"img"!=a&&m(function(){t.removeChild(l)},50);for(var d in y[c])y[c].hasOwnProperty(d)&&y[c][d].onload()}}var j=j||B.errorTimeout,l=b.createElement(a),o=0,r=0,u={t:d,s:c,e:f,a:i,x:j};1===y[c]&&(r=1,y[c]=[]),"object"==a?l.data=c:(l.src=c,l.type=a),l.width=l.height="0",l.onerror=l.onload=l.onreadystatechange=function(){k.call(this,r)},p.splice(e,0,u),"img"!=a&&(r||2===y[c]?(t.insertBefore(l,s?null:n),m(k,j)):y[c].push(l))}function j(a,b,c,d,f){return q=0,b=b||"j",e(a)?i("c"==b?v:u,a,b,this.i++,c,d,f):(p.splice(this.i++,0,a),1==p.length&&h()),this}function k(){var a=B;return a.loader={load:j,i:0},a}var l=b.documentElement,m=a.setTimeout,n=b.getElementsByTagName("script")[0],o={}.toString,p=[],q=0,r="MozAppearance"in l.style,s=r&&!!b.createRange().compareNode,t=s?l:n.parentNode,l=a.opera&&"[object Opera]"==o.call(a.opera),l=!!b.attachEvent&&!l,u=r?"object":l?"script":"img",v=l?"script":u,w=Array.isArray||function(a){return"[object Array]"==o.call(a)},x=[],y={},z={timeout:function(a,b){return b.length&&(a.timeout=b[0]),a}},A,B;B=function(a){function b(a){var a=a.split("!"),b=x.length,c=a.pop(),d=a.length,c={url:c,origUrl:c,prefixes:a},e,f,g;for(f=0;f<d;f++)g=a[f].split("="),(e=z[g.shift()])&&(c=e(c,g));for(f=0;f<b;f++)c=x[f](c);return c}function g(a,e,f,g,h){var i=b(a),j=i.autoCallback;i.url.split(".").pop().split("?").shift(),i.bypass||(e&&(e=d(e)?e:e[a]||e[g]||e[a.split("/").pop().split("?")[0]]),i.instead?i.instead(a,e,f,g,h):(y[i.url]?i.noexec=!0:y[i.url]=1,f.load(i.url,i.forceCSS||!i.forceJS&&"css"==i.url.split(".").pop().split("?").shift()?"c":c,i.noexec,i.attrs,i.timeout),(d(e)||d(j))&&f.load(function(){k(),e&&e(i.origUrl,h,g),j&&j(i.origUrl,h,g),y[i.url]=2})))}function h(a,b){function c(a,c){if(a){if(e(a))c||(j=function(){var a=[].slice.call(arguments);k.apply(this,a),l()}),g(a,j,b,0,h);else if(Object(a)===a)for(n in m=function(){var b=0,c;for(c in a)a.hasOwnProperty(c)&&b++;return b}(),a)a.hasOwnProperty(n)&&(!c&&!--m&&(d(j)?j=function(){var a=[].slice.call(arguments);k.apply(this,a),l()}:j[n]=function(a){return function(){var b=[].slice.call(arguments);a&&a.apply(this,b),l()}}(k[n])),g(a[n],j,b,n,h))}else!c&&l()}var h=!!a.test,i=a.load||a.both,j=a.callback||f,k=j,l=a.complete||f,m,n;c(h?a.yep:a.nope,!!i),i&&c(i)}var i,j,l=this.yepnope.loader;if(e(a))g(a,0,l,0);else if(w(a))for(i=0;i (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0];var j=d.createElement(s);var dl=l!='dataLayer'?'&l='+l:'';j.src='//www.googletagmanager.com/gtm.js?id='+i+dl;j.type='text/javascript';j.async=true;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-M677548'); Skip to main content Home About Submit ALERTS / RSS Search for this keyword Advanced Search New Results Phenotypic Antibiotic Susceptibility Testing at the limit of one bacterial cell Irfan Ahmad , Lisa Johansson , Jimmy Larsson , Spartak Zikrin , Petter Knagge , David Fange , Johan Elf doi: https://doi.org/10.1101/2025.04.13.648565 Irfan Ahmad 1 Department of Cell and Molecular Biology, SciLifeLab, Uppsala University , Sweden Find this author on Google Scholar Find this author on PubMed Search for this author on this site Lisa Johansson 1 Department of Cell and Molecular Biology, SciLifeLab, Uppsala University , Sweden Find this author on Google Scholar Find this author on PubMed Search for this author on this site Jimmy Larsson 1 Department of Cell and Molecular Biology, SciLifeLab, Uppsala University , Sweden Find this author on Google Scholar Find this author on PubMed Search for this author on this site Spartak Zikrin 1 Department of Cell and Molecular Biology, SciLifeLab, Uppsala University , Sweden Find this author on Google Scholar Find this author on PubMed Search for this author on this site Petter Knagge 1 Department of Cell and Molecular Biology, SciLifeLab, Uppsala University , Sweden Find this author on Google Scholar Find this author on PubMed Search for this author on this site David Fange 1 Department of Cell and Molecular Biology, SciLifeLab, Uppsala University , Sweden Find this author on Google Scholar Find this author on PubMed Search for this author on this site For correspondence: johan.elf{at}icm.uu.se david.fange{at}icm.uu.se Johan Elf 1 Department of Cell and Molecular Biology, SciLifeLab, Uppsala University , Sweden Find this author on Google Scholar Find this author on PubMed Search for this author on this site For correspondence: johan.elf{at}icm.uu.se david.fange{at}icm.uu.se Abstract Full Text Info/History Metrics Supplementary material Preview PDF Abstract The blood of a septic patient contains only a few bacteria per milliliter. Recently, various techniques have been developed for extracting these few bacteria from a blood sample. Independent of how these bacteria are separated from the blood cells, we want to learn from them how to treat the infection. Here, we describe how a phenotypic Antibiotic Susceptibility Test can be executed with a single bacterial cell by making averages over time instead of populations, if we account for the experimental noise and cell-to-cell variability. We use the method to make preliminary estimates for how long it takes to distinguish a single susceptible from a resistant cell with statistical confidence. We also exemplify how it is possible to sequentially test different antibiotics, or different concentrations of the same antibiotic, on one cell lineage until a susceptible phenotype is detected. The assay can be followed by single-cell species identification using FISH probes. Background Every year, 10 million people die from sepsis ( Rudd et al. 2020 ). 1.5 million of these are associated with antibiotic resistance ( Murray et al. 2022 ). The initial choice of antibiotic (AB) treatment is based on the experience and judgment of the treating physician (empirical treatment) because standard of care Antibiotic Susceptibility Tests (ASTs) are too slow; for the patient with septic shock, every hour of delay before an effective antibiotic is administered increases the risk of a deadly outcome 7% ( Kumar et al. 2006 ). Current phenotypic ASTs (pASTs) require a positive blood culture (PBC) as the starting point. Here, a PBC means that patient blood has been mixed with broth and cultured until there are ≈10 8 colony-forming units (CFU)/ml ( Azrad et al. 2019 ). It typically takes 12-24h for the bacteria to reach this density since the bacterial viable count in the bloodstream is very low (<10 CFU/ml) ( Wain et al. 1998 ; Kreger et al. 1980 ). The AST and species identification (ID) possibilities following PBCs have improved radically in the last five years. Phenotypic AST, i.e. AST methods based on how the bacteria respond to the drug, used to depend on plating assays, with a typical two-day turnaround time, including isolation streak from PBC. Now these tests can be completed in 6-8 hours with the rapid EUCAST protocols ( Jonasson, Matuschek, and Kahlmeter 2020 ). Companies such as Q-linea, Gradientech and QuantaMatrix market instruments and cartridges that can perform AST from a positive blood culture in 4-6 hours ( Reszetnik et al. 2024 ). The bottleneck for a truly rapid AST, which is clinically meaningful for a patient with bacterial bloodstream infection, is the time to obtain a PBC. To remove this bottleneck, we need to determine the antibiotic susceptibility from the handful of cells that can be extracted from the patient’s blood without PBC. In ( Baltekin et al. 2017 ), we showed that it’s possible to determine if a bacterial isolate is susceptible to an antibiotic in less than one generation by: (1) Monitoring the growth rate impact by the length extension of cells rather than measuring growth by an increase in the number of cells; (2) Averaging the growth rate over several (≈1000) bacteria to eliminate the measurement noise and biological cell-to-cell variation; (3) Normalizing the antibiotic impact to an untreated reference population of the same sample to eliminate isolate-to-isolate variation. The question was whether we could perform AST in less than 30 minutes, and the number of bacteria was not a critical concern. Now, we ask how fast it is possible to perform a phenotypic AST if we only have one bacterial cell. The problem of extracting a single bacterium directly from a patient’s blood sample has been addressed in other papers, for example ( Kim et al. 2024 ; Miguélez et al. 2024 ). Here, we focus on the subsequent steps of AST and species ID. At first glance, one may think that sampling a single cell gives a biased representation of the bacteria that cause the infection. However, this is also the basis for conventional blood culture, which commonly starts with a single cell that is grown until there are enough cells to pass the limit of detection of the specific assay, such as broth microdilution, agar dilution, MALDI, etc. This limit is so high that cell-to-cell variation does not significantly impact the properties of interest. The question here is how long it takes to accurately estimate the same properties starting from one cell and accounting for the cell-to-cell variation. The approach is to work with single cells and make temporal averages instead of averaging over many cells. We will evaluate a method for single-cell phenotypic AST based on three consecutive steps. (i) monitor the cell before antibiotic treatment and estimate the growth rate or division sizes of the untreated cell, to be able to normalize the response to antibiotic treatment for different isolates and species, (ii) monitor the cell grown with antibiotic to calculate the normalized impact of the antibiotic, (iii) determine the species ID using FISH probes, such that the antibiotic response can be evaluated against the correct CLSI/EUCAST clinical breakpoints for susceptible, intermediate/”susceptible, increased exposure”, or resistant (SIR). (i) and (ii) independently become more accurate the more time they are allocated, thus there will be a trade-off between how much time is used to determine the pre-antibiotic treatment properties and the post-antibiotic treatment response. The experiments are facilitated by growing and imaging the bacteria in a microfluidic chip with narrow channels, called traps, which are constricted such that the bacteria can’t exit the channel at one end ( Baltekin et al. 2017 ). The consequence is that the growing mother cell at the bottom of the channel pushes out the daughters, and after a few generations, all cells in the same channel are of the same lineage ( Wang et al. 2010 ). The bacteria are imaged in time-lapse phase contrast microscopy ( Figure 1a ) and segmented using an AI model (methods). The segmented cells are tracked, and the time evolution of the cell area ( Figure 1b ) is used to calculate growth rates for single cells ( Figure 1c ). Download figure Open in new tab Figure 1. Illustration of single-cell imaging and tracking. (a, top) Example of time-lapse phase contrast images of E. coli cells in a single trap of the microfluidic device, (a, bottom) Output of cell segmentation of the time-lapse images in (a, top). (b) Cell area vs time for three examples (in different colors) of cell lineages. (c) Growth rates vs time for the cell areas shown in (b). Growth rates are estimated using exponential regression of a 10 min sliding window. Methods Media and batch growth Overnight cultures (ONCs) were inoculated from glycerol stocks or bacterial plates in Mueller-Hinton (MH) media (70192; Sigma-Aldrich) and grown at 37°C overnight with shaking (200 rpm). For the microfluidic experiments, ONCs were diluted 1:1000 into fresh MH media supplemented with a surfactant (Pluronic F-108; 542342; Sigma-Aldrich) at a final concentration of 0.34 mg/ml and cultured in a 37°C shaker for 2-5 h (duration depended on bacterial strain) before being loaded onto a microfluidic chip. The Pluronic-supplemented MH medium is used throughout all microfluidic experiments and, when indicated, supplemented with antibiotics Microfluidics fabrication Microfluidics fabrication was performed as previously described in ( Baltekin et al. 2017 ; Kandavalli et al. 2022 ). Briefly, PDMS [polydimethylsiloxane; Sylgard 184] was cast on a silicon-SU8 mold (ConScience). The silicon wafer has structures of the fluidic channels and modified mother machine traps (trap width of 1000, 1125, 1250). The PDMS chip was punched and bonded to a nr:1.5 glass coverslip (Menzel-Gläser) using plasma treatment (HPT-200, Henniker plasma) followed by incubation at 80°C overnight. Cell loading and imaging Tubings (Tygon) were connected to the punched microfluidic chip using metal tubing connectors. Pressures in the chip were regulated using an OB1-Mk3 (Elveflow). Bacterial cells were loaded onto the chip with growth media flowing over the cells at 200 mbar. The microscope enclosure was kept at 37°C. Cells were imaged every minute in phase contrast using custom scripts in Micro-Manager ( Edelstein et al. 2010 ). Switching growth media in the microfluidic device To switch the media that goes into the microfluidic chip, a 3-way split (IDEX) was inserted on the tubing close to the chip media inlet. The third outlet adds a valve-controlled route to a waste container. Switching of growth media in the microfluidic chip was carried out using the following steps: ( i ) The medium in the growth media container was exchanged; ( ii ) The valve on the waste side of the 3-way split (IDEX) was opened, which allowed the old media to be flushed from the tubing directly to waste; ( iii ) The valve was closed and the new media flowed the last two centimeters from the 3-way split to the cells on the chip, which takes approximately 6 mins. To pinpoint when the new media reaches the cells, a fluorescent probe (Cy3) was added to the media. A subset of cells not used for growth estimates were imaged in epifluorescence to map the time point when the new media reaches the cells. If two media switches are done in the experiment, the second medium does not contain any fluorescent probe. Here, the decrease in fluorescence was used to indicate when the second medium reached the cells. FISH ID Cells were first treated with 70% EtOH for 20 min using the media switching system described above. Then cells were rehydrated with 0.5x PBS followed by introducing the probes for species identification (100nM each dissolved in 30% formamide and 2 x SSC) ( Kandavalli et al. 2022 ) for 30 min. The hybridized probes were imaged in epifluorescence on the Ti2 microscope setup, using the appropriate filter combinations for these probes (300 ms exposure). MIC assay Minimum inhibitory concentration (MIC) assays were performed in flat bottom microtiter plates using broth microdilution ( Kadeřábková, Mahmood, and Mavridou 2024 ). Bacteria were grown on LB agar plates or in Mueller Hinton broth (MHB). Bacterial suspension from colonies or logarithmic phase culture were diluted in MHB to achieve a final inoculum of 1×10 6 CFU/mL. Two-fold serial dilutions of antibiotics were prepared in MHB to obtain a range of concentrations by transferring 50 μl of solution from wells in the left to right direction. 50 μl of diluted bacterial solution was added to each well. Plates were incubated at 37°C for 16–20 hours and OD600, used to quantify bacterial growth, was measured using a microplate reader (Tecan Sunrise). Bacterial strains Information about strains used in the study is found in Table S1 . Microscope setups Phase contrast and epifluorescence images were acquired using either a Ti2 (Nikon) or a Ti-E (Nikon) microscope. The Ti2 microscope was fitted with a CFI Plan Apo lambda 1.45/100x oil DM (Nikon) objective, a Sona 4.2B-11 (Andor) camera, a DMK 38UX304 (The Imaging Source) camera, and a Spectra Gen. 3 (Lumencor) light source for epifluorescence. For epifluorescence, a filter cube with FF562-Di03 (Semrock) dichroic mirror, FF01-543/22 (Semrock) excitation filter and FF01-586/20 (Semrock) emission filter, and a filter cube with FF660-Di02 (Semrock) dichroic mirror, and a Brightline Fluorescence 692/40 (Semrock) emission filter was used. The Ti-E microscope was fitted with a CFI Plan Apo Lambda 1.45/100x oil (Nikon) objective, a Ph3 phase plate (Nikon), a DMK 38UX304 (The Imaging Source) camera, and a Sola FISH (Lumencor) light source for epifluorescence. For epifluorescence, a filter cube with F555-Di03 (Semrock) dichroic mirror, FF01-530/11 (Semrock) excitation filter and FF01-575/19 emission filter was used. Both microscopes were enclosed in a Lexan incubator (OKOlab) where the temperature is kept constant. Image analysis The acquired phase contrast images were segmented using an Omnipose deep neural network ( Cutler et al. 2022 ), which has been trained using outlines generated from fluorescently labelled E. coli cells in 4200 traps and manually labelled outlines of mixed species data from ( Kandavalli et al. 2022 ). Segmented cells were tracked from frame-to-frame using the Baxter algorithm ( Magnusson et al. 2015 ). Post-processing of segmented and tracked cells Growth rate estimates Growth rates were estimated using exponential regression within a sliding window of up to 10 minutes, when 10 minutes of preceding data was available. After cell division, the descendant cell areas were summed. When a cell lineage did grow out of the trap, the lineage was restarted from a descendant of the mother cell closest to the trap constriction. Mean growth rates were measured for each cell during 50 minutes prior to switching to antibiotic-containing medium. To remove non-growing segmented objects, traps in which the mean growth rate was less than 0.002 min −1 were excluded from further analysis. Cell-specific growth rates were normalized to the mean growth rate measured in the corresponding cell lineage. Cell size estimates The relative cell areas presented in Figures 4 and 5 were calculated as follows: First, traps containing less than three segmented objects at the time of switching to antibiotic-containing media were identified. Since the media switch occurred more than 59 minutes from the start of imaging, these traps did not contain growing cells and were removed from further analysis. After switching to antibiotic-containing medium, the maximum area of all segmented cells in each trap for up to four different time points was found. The maximum area for each trap was then divided by the mean area of all segmented cells in the same trap from 0 to 40 minutes prior to switching to antibiotic-containing media, resulting in the relative cell areas that are shown in the histograms in Figure 4c, f and in Figure 5f-g, j-k . For experiments where two antibiotics were supplied sequentially, the mean of segmented cells during the first antibiotic exposure, 40 minutes prior to the second antibiotic exposure, was used for area normalization. Results Pre-antibiotic characterization of species growth and division for robust normalization The first question is how long a time should be spent on estimating the untreated phenotype to make a robust normalization, such that the temporal variability in the phenotype has been averaged out. We will focus on the most characteristic phenotypes of antibiotics used in sepsis treatment, which are growth rate reduction (induced by, for example, fluoroquinolones and tetracyclines), and cell elongation (caused by, for example, beta-lactams). The critical parameter is the correlation time for these properties, i.e ., how long does the single cell remember its current growth rate or division size? If the property is sampled over several correlation times, the average converges to a value that can be used for robust normalization. If the property is sampled shorter than the correlation time, the normalization adds noise to the evaluation of the impact of the antibiotic treatment. Figure 2a shows growth-rate autocorrelation vs time for a few sepsis-relevant pathogens, growing in Mueller Hinton (MH) broth without any antibiotic treatment. We find that the correlation time is <10 min for all three species. The observed correlation time is on the same time scale as the size of the sliding window used to calculate the growth rate, which implies that the measured autocorrelation time is most likely set by limitations in the precision of our cell size estimates and not by the underlying biology. Since there is no long-lived growth rate memory stretching over multiple generations, it’s possible to get a good estimate of the typical growth rate for the cell lineage within the time frame of a single generation. This is corroborated by the low correlation of growth rates between mother and daughter cells ( Figure 2b ). Figure 2b also shows that the cell-to-cell variation in the growth rate of E. coli is in the order of 10%. Download figure Open in new tab Figure 2. Characterising the time scale of temporal variation in the phenotypes. (a) Solid lines are the mean of the growth rate autocorrelation of the time trace in each lineage for 3 different clinical isolates growing at steady state without any antibiotic treatment. Shaded areas indicate SEM. Growth rates are estimated by exponential fits in a sliding window of 10 minutes. Total number of lineages included are 277, 340 and 54 for E.coli, K. pneumoniae and P. aeruginosa, respectively. (b) Correlation plot of growth rates for parent (x-axis) and daughter cells (y-axis). Correlation coefficient are 0.31, 0.23 and 0.18 for E. coli, K. pneumoniae and P. aeruginosa, respectively. Growth rates are estimated by exponential fits from division to the subsequent division using the same cell lineages as in (a). The number of datapoints are 2180, 7372 and 959, for E. coli, K. pneumoniae and P. aeruginosa, respectively. (c) Correlation plot of size at division for parent (x-axis) and daughter cells (y-axis). Total number data points are 1293 for E. coli, 5847 for K. pneumoniae, and 485 for P. aeruginosa, respectively. (d) Correlation coefficient (y-axis) for division sizes which are an increasing number of generations (x-axis) apart. When lag=1, the y-axis shows the correlation coefficient for the distributions shown in (c). Total number of data points are 1293 (gen 1) and 141 (gen 2) for E. coli, 5847 (gen 1), 3979 (gen 2) and 2092 (gen 3) for K. pneumoniae, and 485 (gen 1) and 86 (gen 2) for P. aeruginosa, respectively. Results from repeat experiments are shown in Figure S1. In Figure 2c , we show correlation in size at divisions from one generation to the next under exponential growth in MH broth. The correlation coefficient is between 0.35 and 0.55 for the three strains, resulting in the decay of autocorrelation presented in Figure 2d . Compared to the growth rate case, the division sizes correlate over longer periods of time. Thus, if we use the division size from a single generation, the time needed after antibiotic treatment will be longer to get the same precision in estimating the antibiotic impact for these cases. Based on these considerations, we estimated the growth rate for untreated growing cells during 50 min before treating the cells with antibiotics and used this to normalize the impact of the growth-rate-reducing antibiotics. Here, a growing cell was defined as residing in a trap in which the pre-treatment mean cell growth rate is larger than 0.002 min −1 . Figure 3 shows growth rate responses to ciprofloxacin (CIP), an antibiotic which inhibits growth, for a resistant (blue) and a susceptible (orange) E. coli strain ( Baltekin et al. 2017 ) without ( Figure 3a-b ) or with ( Figure 3c-d ) normalizing to the pre-treatment growth rate. The results illustrate that if we were to make a call for susceptibility for the absolute growth rate using a threshold 0.02min −1 , which corresponds to ~35% average impact for the susceptible isolate, this would incorrectly classify ~50% of the resistant isolate cells as susceptible. Download figure Open in new tab Figure 3. Normalization to pre-treatment growth rate. Growth rate before and after adding 0.25 μg/ml CIP to the growth medium for a resistant (blue) and susceptible (orange) isolate of E. coli. (a) Examples of single lineages. (b) Population averages (solid lines) of the growth rate distributions for each time point. The shaded areas show +/- 2 standard deviations of the growth rate distribution. (c and d) Same growth rate estimates as in (a) and (b) but normalized to the time-average of the pre-treatment growth rate of each lineage. Results from a repeat experiment are shown in Figure S2. Differentiating a susceptible bacterium from a resistant one Having established that the pre-treatment growth can be used as normalization, the next question is how much time is needed to get a conclusive response from the single cell following antibiotic treatment. It may seem sufficient to sample for as long as it takes to statistically confirm that the response is significantly different from the pre-antibiotic treatment. However, since also resistant isolates, especially with MICs close to the antibiotic concentrations used in the test, may respond at some level to the antibiotic treatment, the more complete requirement is that the single susceptible isolate needs to be significantly different from any resistant isolate. To get a first-order estimate of how long this may take, we grew a resistant and a susceptible E. coli clinical isolate for 70 minutes without and for 120 minutes with CIP. Figure 4a shows how the width (as defined by +-2STDs) of the distribution of normalized growth rates for single cells responds to antibiotic treatment. To determine how long it takes to distinguish one specific single cell from this particular isolate of resistant cells, we determine when the normalized growth rate distributions of resistant and susceptible populations are significantly separated, as defined by when the distributions are separated by more than 2 STDs. Since the distributions are approximately normally distributed, there is less than 5% risk of misclassifying the resistant cell as susceptible at this time point. Considering that the pre-treatment growth is 60 min, the total time for AST for one cell would be approximately 2 hours, for this specific combination of strains. The classification of resistant or susceptible needs to be calibrated based on many susceptible and resistant isolates to include the full width of possible single-cell responses. We also note that the time point at which the distribution separates is based on information only at this time point. By including the entire curve of decaying growth rates following antibiotic treatment, more information can be extracted, which will allow for an earlier detection time of susceptibility. Download figure Open in new tab Figure 4. Single-cell AST. (a) Relative growth rates of one susceptible (MIC=0.008 μg/ml) and one resistant (MIC=50 μg/ml) isolate of E. coli before and after CIP (0.25 μg/ml) treatment. Relative growth rates calculated as in Fig. 3 . Dashed lines show 10 single lineage examples for each isolate. The total number of lineages are 297 for the susceptible and 273 for the resistant isolate. Colors and lines are as in Fig. 3 . (b) Example of a single trap in the microfluidic device visualized as in Fig. 1a . (c-f) Cell area for one susceptible (MIC=0.5 μg/ml) (blue) and one borderline resistant (MIC=8 μg/ml) (orange) P. aeruginosa before and after treating with MER (1.5 μg/ml). Here, cell area is defined as the region enclosed by the segmentation of the phase contrast images. (c) Example of a single lineage from the resistant (blue) and susceptible (orange) isolates. (d) Example of a single trap in the microfluidic device visualized as in Fig. 1a . (e) Distributions of the maximum cell area attained by any cell in a trap up to the given time point (see legend) after MER treatment, normalized to the mean cell area in the same trap 40 mins before MER treatment of the susceptible strain. The total number of traps in the distributions are 146. (f) As (e), but for the resistant isolate. The total number of traps in the distributions are 108. Results of repeat experiments of (a)-(b) are shown in Figure S3. Results using the same isolate and experiment setup as in (e)-(f), but with 2.0 μg/ml MER are shown in Figure S4. We also performed a corresponding analysis using meropenem (MER), an antibiotic that inhibits cell division, on two isolates of P. aeruginosa , one susceptible and one borderline resistant (MIC 8 μg/ml is susceptible, increased exposure; EUCAST breakpoint table v. 15.0). Figure 4c shows the length extension over multiple division cycles. After the cells have been exposed to MER, the susceptible cells are to a large extent inhibited in cell division, and the cells thus keep elongating past their expected pre-treatment division size, as highlighted by the example in Figure 4c-d . The resistant strain is also affected by MER, but to a substantially lesser extent than the susceptible strain. In Figure 4e-f we show the distributions of cell lengths at different time points after adding the antibiotic. The sizes of individual cells are normalized to the average cell size of growing cells in the same trap of the microfluidic device in a 40-minute window before adding the antibiotic. We have not explored whether spending more time in the pre-antibiotic phase to get a better normalization, or spending more time in the antibiotic treatment phase to get a more pronounced phenotypic response, would improve the susceptibility classification. However, as expected, the time necessary to separate the susceptible from the resistant bacteria is longer when the read-out is cell size; after 110 min of antibiotic treatment, there is still an overlap between the distributions for susceptible and resistant isolates. By, for example, using a relative cell area of 4.7 as a threshold between resistant and susceptible, there is an equal probability of 14% of incorrectly assigning a resistant as susceptible and vice versa. Similar as for the use of antibiotics affecting growth rate, the exact area increase needed to make the call that the cell is susceptible has to be calibrated using various isolates with different MICs. Sequential antibiotic treatments of a single bacterium instead of parallel treatment on split populations Using 1000s of bacteria, the population can be split into multiple pools, and different antibiotics can be applied and evaluated in parallel. When working with only one bacterium, this is not an alternative. Here, we instead suggest using a sequence of antibiotics on the same cell, or lineage of daughter cells, until an effective antibiotic is found. A caveat of this approach is that the response of a tested cell to an antibiotic may depend on the previous treatments of its lineage, even if it was not classified as susceptible. This is exemplified in Figure 4c-d where the borderline resistant strain shows a slight cell size increase in response to antibiotic treatment. To test if the effectiveness of the antibiotic treatment, or the effectiveness in calling the susceptible strain, is affected by previous treatments, we used an E. coli isolate that is resistant to CIP and susceptible to tigecycline (TIG). In Figure 5a we show that the response to TIG for a cell is similar, independently of whether or not it has been treated with CIP prior to the TIG treatment. Figure 5b-c shows the result of the corresponding experiments using cefotaxime (CEF). Here, the impact of CEF was the same whether or not the bacteria were first treated with CIP, to which both the tested isolates are resistant. The pre-treatment with CIP has no apparent effect in this case, but this may depend on species, resistance mechanism, antibiotics, and concentrations. Download figure Open in new tab Figure 5. The effect of sequential antibiotic treatment. (a and b) Relative growth rates of E. coli resistant to CIP and susceptible to TIG, before and after treatment with either TIG, or CIP and TIG in sequence. In (a) cells are only treated with TIG (0.2 μg/ml) while in (b) the cells are first treated with CIP (1 μg/ml) for 60 min followed by treatment with TIG (0.2 μg/ml). Colors and line specifications as in Fig. 4a . (c) Data from (a) and (b) plotted together, where the time of TIG treatment is set to zero. (d-g) Relative maximum cell area of traps as in Fig. 4c-f , for E. coli isolates, either resistant or susceptible to CEF, and treated with CEF (1 μg/ml). Colors and line specifications and panel content as in Fig. 4c-f . The total number of data points are 301 in (f) and 309 in (g). (f-g) At a relative cell area of 2.08 at 90 min there is an equal probability of 13% of incorrectly assigning a resistant as susceptible and vice versa. (i-l) As (d-g), but for E. coli isolates resistant to CIP and additionally either resistant or susceptible to CEF, first treated with CIP (0.25 μg/ml) followed by treatment with CEF (1 μg/ml). The total number of data points are 315 in (j) and 315 in (k). (j-k) At a relative cell area of 2.39 at 90 min there is an equal probability of 16% of incorrectly assigning a resistant as susceptible and vice versa. Results from the repeat experiment of (a-c) are shown in Figure S5. Results from two repeat experiments of (d-g) are shown in Figures S6 and S7. Results from the repeat experiment of (h-k) are shown in Figure S8. A special case of sequential antibiotic treatment is to use the same antibiotic at different concentrations. This may be needed when different species have different breakpoints for the SIR call. These breakpoints are defined by EUCAST and CLSI, based partly on what has been observed clinically for different species ( Mouton et al. 2012 ). If susceptibility is tested with only one antibiotic concentration, the different breakpoints may be addressed by requiring different levels of impact for different species when deciding if it is susceptible. However, the range of breakpoints this approach can address is limited, and it may be required to perform the AST at different concentrations. In the case of the single-cell assay, where the species ID is unknown at the time of antibiotic treatment, one would start with the lower concentration and then sequentially increase the concentration. At the end of the assay, the species ID can be determined at the single-cell level with FISH probes ( Kempf, Trebesius, and Autenrieth 2000 ) using the protocol we described in ( Kandavalli et al. 2022 ). To exemplify this process, we grew one strain each of E. coli and K. pneumoniae with different CIP MICs. In the experiment, the CIP concentration was increased stepwise, and the two species stopped growing at different concentrations ( Figure 6a-b ). After the AST phase, the cells were permeabilized and hybridized with fluorescent oligos, which species-specifically bind ribosomal RNA ( Figure 6c ). Download figure Open in new tab Figure 6. Single-cell species identification after AST. (a) Relative growth rates of E. coli (blue) and K. pneumoniae (orange) isolates, which are sequentially treated with 0.25 μg/ml CIP for 70 min followed by treatment with 4 μg/ml CIP. Lines and shading as in Fig. 3d . (b) Single trap examples, visualized as in Fig. 1a , for the two isolates that are shown in (a). (c) Fluorescence signal after FISH probing. Fluorescence due to the E. coli- specific probe (labelled with Cy3) is shown with a yellow intensity map (leftmost images) and fluorescence due to the K. pneumoniae -specific probe (labelled with Cy5) is shown with a red intensity map (rightmost images). Discussion Here we have shown that it is possible to perform pAST at the single-cell level within a few hours using the antibiotics ciprofloxacin, cefotaxime, meropenem and tigecycline on rapidly growing E. coli and P. aeruginosa . For antibiotics relevant for treating E. coli , including CIP, we have previously shown that it’s possible to perform AST in ≈15 min ( Baltekin et al. 2017 ). An AST in 15 min was also shown to be the fastest possible, given the underlying biology ( Baltekin et al. 2017 ). However, it requires an analysis based on a few hundred bacteria to be attainable. The additional time for single-cell AST is needed to reach significance in calling susceptibility of a population of cells based on information from a single cell while accounting for phenotypic cell-to-cell variation. Still, a few hours for a single-cell phenotypic AST is a clear improvement compared to waiting 12-24h to get a positive blood culture and then performing a “rapid” AST method. The method for stepwise increase of antibiotic concentrations can most likely be performed in several steps to determine accurate MIC values, for instance, by calibrating the response observed at different concentrations using isolates with known MIC values, as determined by gold-standard methods. Particular care has to be taken with respect to for how long the impact is evaluated at each concentration for the single cell, since correlation times may increase as the concentration approaches the cell’s MIC. The current study only illustrates principles, and several caveats are not addressed. Heterogeneous populations with a low frequency of antibiotic-tolerant cells, such as in the case of heteroresistance ( Nicoloff et al. 2024 ) and persistence ( Balaban et al. 2004 ) may cause prohibitively long time scales for a single-cell assay. However, these phenomena are also problematic for conventional AST methods ( Heyman et al. 2025 ). More interesting is sub-MIC heterogeneity induced by the antibiotics, i.e ., the broadening of growth phenotypes during sub-lethal antibiotic concentrations ( Brandis, Larsson, and Elf 2023 ). It should be explored carefully if resistant isolates with MIC close to the breakpoint are so heterogeneous that a single cell can appear susceptible if tested after long periods of antibiotic treatment. In this study, we have explored a tiny part of the vast space spanned by combinations of species, MIC, antibiotic sequence, and antibiotic concentration. Therefore, the take-home message should not be that single-cell phenotypic AST is possible for all combinations of drugs and bugs, but rather that it will be possible for some combinations and that it’s important to find which these combinations are. Author contributions IA, LJ & JL made the experiments; SZ & LJ analysed data; JL adapted the fluidic assay; JE, DF & PK wrote the paper with input from the other authors; DF and JE developed statistical reasoning; PK contributed clinical relevance and coordination; JE conceived the idea. Everyone contributed to the evaluation of results, experiments and project planning. Acknowledgements We thank Buu Minh Tran for contributing to initial discussions about the project, Praneeth Karempudi for support with deep learning and Vinodh Kandavalli for support on FISH ID. This research was funded by the SSF (grant no. ARC19-0016), the Swedish Research Council (grant no. 2024-06127), the Novo Nordisk Foundation (grant no. NNF23OC0083419), the Knut and Alice Wallenberg Foundation (grant no. 2023.0531), and the eSSENCE e-science initiative. The computations and data management were facilitated by resources from the Swedish National Infrastructure for Computing at UPPMAX, with partial funding from the Swedish Research Council (grant agreement no. 2022-06725). A patent has been filed for the method with a priority date of April 2024. Funding Swedish Foundation for Strategic Research https://ror.org/044wr7g58 ARC19-0016 Swedish Research Council https://ror.org/03zttf063 2024-06127 Novo Nordisk Foundation https://ror.org/04txyc737 NNF23OC0083419 Knut and Alice Wallenberg Foundation https://ror.org/004hzzk67 2023.0531 References ↵ Azrad , Maya , Yoram Keness , Orna Nitzan , Nina Pastukh , Linda Tkhawkho , Victoria Freidus , and Avi Peretz . 2019 . “ Cheap and Rapid in-House Method for Direct Identification of Positive Blood Cultures by MALDI-TOF MS Technology .” BMC Infectious Diseases 19 ( 1 ): 72 . OpenUrl CrossRef PubMed ↵ Balaban , Nathalie Q. , Jack Merrin , Remy Chait , Lukasz Kowalik , and Stanislas Leibler . 2004 . “ Bacterial Persistence as a Phenotypic Switch .” Science (New York, N.Y.) 305 ( 5690 ): 1622 – 25 . OpenUrl CrossRef ↵ Baltekin , Özden , Alexis Boucharin , Eva Tano , Dan I. Andersson , and Johan Elf . 2017 . “ Antibiotic Susceptibility Testing in Less than 30 Min Using Direct Single-Cell Imaging .” Proceedings of the National Academy of Sciences , 201708558 . ↵ Brandis , Gerrit , Jimmy Larsson , and Johan Elf . 2023 . “ Antibiotic Perseverance Increases the Risk of Resistance Development .” Proceedings of the National Academy of Sciences of the United States of America 120 ( 2 ): e2216216120 . OpenUrl CrossRef PubMed ↵ Cutler , Kevin J. , Carsen Stringer , Teresa W. Lo , Luca Rappez , Nicholas Stroustrup , S. Brook Peterson , Paul A. Wiggins , and Joseph D. Mougous . 2022 . “ Omnipose: A High-Precision Morphology-Independent Solution for Bacterial Cell Segmentation .” Nature Methods 19 ( 11 ): 1438 – 48 . OpenUrl CrossRef PubMed ↵ Edelstein , Arthur , Nenad Amodaj , Karl Hoover , Ron Vale , and Nico Stuurman . 2010 . “ Computer Control of Microscopes Using µManager .” Et Al [Current Protocols in Molecular Biology] Chapter 14 ( 1 ): Unit14.20. ↵ Heyman , Gabriel , Sofia Jonsson , Nikos Fatsis-Kavalopolous , Karin Hjort , Hervé Nicoloff , Mia Furebring , and Dan I. Andersson . 2025 . “ Prevalence, Misclassification, and Clinical Consequences of the Heteroresistant Phenotype in Escherichia Coli Bloodstream Infections in Patients in Uppsala, Sweden: A Retrospective Cohort Study .” The Lancet. Microbe , no. 101010 ( January ), 101010 . OpenUrl ↵ Jonasson , Emma , Erika Matuschek , and Gunnar Kahlmeter . 2020 . “ The EUCAST Rapid Disc Diffusion Method for Antimicrobial Susceptibility Testing Directly from Positive Blood Culture Bottles .” The Journal of Antimicrobial Chemotherapy 75 ( 4 ): 968 – 78 . OpenUrl CrossRef PubMed ↵ Kadeřábková , Nikol , Ayesha J. S. Mahmood , and Despoina A. I. Mavridou . 2024 . “ Antibiotic Susceptibility Testing Using Minimum Inhibitory Concentration (MIC) Assays .” Npj Antimicrobials and Resistance 2 ( 1 ): 37 . OpenUrl CrossRef ↵ Kandavalli , Vinodh , Praneeth Karempudi , Jimmy Larsson , and Johan Elf . 2022 . “ Rapid Antibiotic Susceptibility Testing and Species Identification for Mixed Samples .” Nature Communications 13 ( 1 ): 6215 . OpenUrl CrossRef PubMed ↵ Kempf , V. A. , K. Trebesius , and I. B. Autenrieth . 2000 . “ Fluorescent In Situ Hybridization Allows Rapid Identification of Microorganisms in Blood Cultures .” Journal of Clinical Microbiology 38 ( 2 ): 830 – 38 . OpenUrl Abstract / FREE Full Text ↵ Kim , Tae Hyun , Junwon Kang , Haewook Jang , Hyelyn Joo , Gi Yoon Lee , Hamin Kim , Untack Cho , et al. 2024 . “ Blood Culture-Free Ultra-Rapid Antimicrobial Susceptibility Testing .” Nature 632 ( 8026 ): 893 – 902 . OpenUrl CrossRef PubMed ↵ Kreger , B. E. , D. E. Craven , P. C. Carling , and W. R. McCabe . 1980 . “ Gram-Negative Bacteremia. III. Reassessment of Etiology, Epidemiology and Ecology in 612 Patients .” The American Journal of Medicine 68 ( 3 ): 332 – 43 . OpenUrl CrossRef PubMed Web of Science ↵ Kumar , Anand , Daniel Roberts , Kenneth E. Wood , Bruce Light , Joseph E. Parrillo , Satendra Sharma , Robert Suppes , et al. 2006 . “ Duration of Hypotension before Initiation of Effective Antimicrobial Therapy Is the Critical Determinant of Survival in Human Septic Shock .” Critical Care Medicine 34 ( 6 ): 1589 – 96 . OpenUrl CrossRef PubMed Web of Science ↵ Magnusson , Klas E. G. , Joakim Jalden , Penney M. Gilbert , and Helen M. Blau . 2015 . “ Global Linking of Cell Tracks Using the Viterbi Algorithm .” IEEE Transactions on Medical Imaging 34 ( 4 ): 911 – 29 . OpenUrl CrossRef PubMed ↵ Miguélez , M. Henar Marino , Mohammad Osaid , Jimmy Larsson , Vinodh Kandavalli , Johan Elf , and Wouter van der Wijngaart . 2024 . “ Culture-Free Rapid Isolation and Detection of Bacteria from Whole Blood at Clinically Relevant Concentrations .” bioRxiv . doi: 10.1101/2024.05.23.595289 . OpenUrl Abstract / FREE Full Text ↵ Mouton , J. W. , D. F. J. Brown , P. Apfalter , R. Cantón , C. G. Giske , M. Ivanova , A. P. MacGowan , et al. 2012 . “ The Role of Pharmacokinetics/pharmacodynamics in Setting Clinical MIC Breakpoints: The EUCAST Approach .” Clinical Microbiology and Infection: The Official Publication of the European Society of Clinical Microbiology and Infectious Diseases 18 ( 3 ): E37 – 45 . OpenUrl CrossRef ↵ Murray , Christopher J. L. , Kevin Shunji Ikuta , Fablina Sharara , Lucien Swetschinski , Gisela Robles Aguilar , Authia Gray , Chieh Han , et al. 2022 . “ Global Burden of Bacterial Antimicrobial Resistance in 2019: A Systematic Analysis .” The Lancet 399 ( 10325 ): 629 – 55 . OpenUrl CrossRef ↵ Nicoloff , Hervé , Karin Hjort , Dan I. Andersson , and Helen Wang . 2024 . “ Three Concurrent Mechanisms Generate Gene Copy Number Variation and Transient Antibiotic Heteroresistance .” Nature Communications 15 ( 1 ): 3981 . OpenUrl CrossRef PubMed ↵ Reszetnik , Grace , Keely Hammond , Sara Mahshid , Tamer AbdElFatah , Dao Nguyen , Rachel Corsini , Chelsea Caya , Jesse Papenburg , Matthew P. Cheng , and Cedric P. Yansouni . 2024 . “ Next-Generation Rapid Phenotypic Antimicrobial Susceptibility Testing .” Nature Communications 15 ( 1 ): 9719 . OpenUrl CrossRef PubMed ↵ Rudd , Kristina E. , Sarah Charlotte Johnson , Kareha M. Agesa , Katya Anne Shackelford , Derrick Tsoi , Daniel Rhodes Kievlan , Danny V. Colombara , et al. 2020 . “ Global, Regional, and National Sepsis Incidence and Mortality, 1990-2017: Analysis for the Global Burden of Disease Study .” Lancet 395 ( 10219 ): 200 – 211 . OpenUrl CrossRef PubMed ↵ Wain , John , To Song Diep , Vo Anh Ho , Amanda M. Walsh , Nguyen Thi Tuyet Hoa , Christopher M. Parry , and Nicholas J. White . 1998 . “ Quantitation of Bacteria in Blood of Typhoid Fever Patients and Relationship between Counts and Clinical Features, Transmissibility, and Antibiotic Resistance .” Journal of Clinical Microbiology 36 ( 6 ): 1683 – 87 . OpenUrl Abstract / FREE Full Text ↵ Wang , Ping , Lydia Robert , James Pelletier , Wei Lien Dang , Francois Taddei , Andrew Wright , and Suckjoon Jun . 2010 . “ Robust Growth of Escherichia Coli .” Current Biology: CB 20 ( 12 ): 1099 – 1103 . OpenUrl CrossRef PubMed View the discussion thread. Back to top Previous Next Posted April 13, 2025. Download PDF Supplementary Material Email Thank you for your interest in spreading the word about bioRxiv. NOTE: Your email address is requested solely to identify you as the sender of this article. Your Email * Your Name * Send To * Enter multiple addresses on separate lines or separate them with commas. You are going to email the following Phenotypic Antibiotic Susceptibility Testing at the limit of one bacterial cell Message Subject (Your Name) has forwarded a page to you from bioRxiv Message Body (Your Name) thought you would like to see this page from the bioRxiv website. Your Personal Message CAPTCHA This question is for testing whether or not you are a human visitor and to prevent automated spam submissions. Share Phenotypic Antibiotic Susceptibility Testing at the limit of one bacterial cell Irfan Ahmad , Lisa Johansson , Jimmy Larsson , Spartak Zikrin , Petter Knagge , David Fange , Johan Elf bioRxiv 2025.04.13.648565; doi: https://doi.org/10.1101/2025.04.13.648565 Share This Article: Copy Citation Tools Phenotypic Antibiotic Susceptibility Testing at the limit of one bacterial cell Irfan Ahmad , Lisa Johansson , Jimmy Larsson , Spartak Zikrin , Petter Knagge , David Fange , Johan Elf bioRxiv 2025.04.13.648565; doi: https://doi.org/10.1101/2025.04.13.648565 Citation Manager Formats BibTeX Bookends EasyBib EndNote (tagged) EndNote 8 (xml) Medlars Mendeley Papers RefWorks Tagged Ref Manager RIS Zotero Tweet Widget Facebook Like Google Plus One Subject Area Microbiology Subject Areas All Articles Animal Behavior and Cognition (7618) Biochemistry (17637) Bioengineering (13864) Bioinformatics (41853) Biophysics (21403) Cancer Biology (18540) Cell Biology (25429) Clinical Trials (138) Developmental Biology (13356) Ecology (19862) Epidemiology (2067) Evolutionary Biology (24287) Genetics (15585) Genomics (22464) Immunology (17701) Microbiology (40300) Molecular Biology (17142) Neuroscience (88440) Paleontology (666) Pathology (2825) Pharmacology and Toxicology (4814) Physiology (7633) Plant Biology (15107) Scientific Communication and Education (2042) Synthetic Biology (4285) Systems Biology (9809) Zoology (2268)