{"paper_id":"42d9de7c-23ed-47d7-9315-e9510085d8c9","body_text":"1 \n \nParallelised detection of bacteria viability using an electrode \narray and the Exeter Multiscope \nKa Kiu Lee 3,4, David Horsell1, James Stratford2, Magdalena Karlikowska2, Salman Khattak2, Tailise de-\nSouza-Guerreiro-Rodrigues3,4, Junqing Jiang 5, Mike Shaw 5,6, Stefano Pagliara 3,4, Alexander D.  \nCorbett1,4* \n1. Department of Physics and Astronomy, University of Exeter, UK. \n2. Cytecom Limited, Coventry, UK. \n3. Department of Biosciences, University of Exeter, UK. \n4. Living Systems Institute, University of Exeter, UK. \n5. National Physical Laboratory, Teddington, UK. \n6. UCL Hawkes Institute and Department of Computer Science, University College London, London, UK. \n \nAbstract  \nAntimicrobial resistance remains a global existential threat. Given that antimicrobial therapy \ncommonly starts before pathogen identification, rapid and scalable methods capable of determining \neffective antimicrobial compounds are needed.  In this paper , we demonstrate a 2 × 2 array of \nparallelised microscope s that uses  low numerical aperture (NA=0.25) detection optics and LED \nexcitation to determine bacterial viability based on their fluorescence response to an electrical \nstimulus.  Following a 2-hour incubation, the fluorescent viability readout requires less than one \nminute. We use K-means clustering to classify pixels in a time lapse sequence of widefield fluorescence \nimages and extract changes seen within bacterial clusters. We demonstrate sufficient sensitivity to \nmeasure fluorescence changes after electrical stimulation in a bacterial monolayer. To capture these \nsubtle fluorescence changes at high signal -to-background ratios, we place a limit on the minimum \noptical density of the bacteria l sample . This novel approach is scalable to 96 -well formats using a \nsuitable consumable electrode array.  \n \nIntroduction \nAntimicrobial resistance (AMR) remains a major health threat  to the global population . The annual \ncost of AMR in Europe alone is currently estimated at €1.5 bn with AMR predicted to cause 10 million \ndeaths annually by 2050 1. Developing a rapid and scalable AMR test would reduce this burden by \nensuring patients receive the appropriate antibiotic at the right concentration sooner.  \nThe current clinical gold standard for determining antibiotic resistance in a patient sample is to culture \nthe bacterial isolate with a panel of antibiotics across multiple concentrations for 24 to 48 hours. The \nefficacy of an antibiotic and its minimum inhibitory concentration (MIC) are established by monitoring \nbacterial growth (turbidity) in in each well. The limitation of this approach is that, in acute cases, results \nare required within much shorter time frames (a few hours) for effective treatment. Th e long \nincubation period also limits throughput and contributes to a backlog of testing.  \nIn recent years, there has been a rapid expansion of antimicrobial susceptibility testing (AST) methods \navailable (see Reszetnik, et al. [2] for a review). Recent advances include direct-from-specimen testing, \nsingle-cell imaging, miniaturized growth chambers, microfluidics, and novel detection modalities \nincluding plasmonic and nucleic acid -based methods that shorten turnaround times by accelerating \ngrowth detection or viability assessment under antimicrobial exposure. Several commercial platforms \nnow provide phenotypic AST directly from positive blood cultures or urine specimens significantly \n.CC-BY 4.0 International licenseperpetuity. It is made available under a \npreprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in \nThe copyright holder for thisthis version posted March 11, 2026. ; https://doi.org/10.64898/2026.03.10.710830doi: bioRxiv preprint \n\n2 \n \nfaster than traditional methods  (such as the electric field detection method employed by iFast \nDiagnostics3). Emerging non-commercial approaches show promise for further reductions in time-to-\nresult, including methods that combine rapid imaging or metabolic sensing with deep learning for \nenhanced signal interpretation. Challenges remain in extending these technolog ies to diverse \nspecimen types, standardising inoculum effects, and achieving broad clinical validation.  \nOne recent technique developed by the Munehiro lab4 measures bacterial viability via electrical \nstimulation of the bacteria under investigation, whilst monitoring changes in fluorescence intensity, \nwith positive  and negative changes indicating viable  and non-viable bacteria, respectively . After \nincubation in a medium containing fluorescent dye thioflavin  T (ThT), electrical stimulation \nhyperpolarises the membrane  of healthy cells by opening voltage -sensitive ion channels. Viable \nbacteria rapidly accumulate surrounding (positively charged) ThT through these ion channels, leading \nto increased fluorescence brightness. In contrast, non-viable bacteria with disrupted membranes and \na weak ened potassium ion gradient will undergo depolarisation and expel any intracellular ThT \nthrough electrical stimulation. This translates into either no change or a reduction in fluorescence \nbrightness.  \nThis technique has since been refined and commercialised 5. Whilst significant developments have \nbeen made  to improve sensitivity and repeatability , there is clear potential (as with many of the \ntechniques summarised in [2]) to greatly increase sample throughput . This could be achieved by \ncreating arrays of stimulation electrodes, each with their own bacterial sample. However, in studies \nperforming fluorescence imaging of multiple fields of view, 75% of the total acquisition time was spent \nmoving the sample 6, with similar estimates for fast axial scanning systems 7. Throughput can be \ndramatically increased by eliminating the need to move samples by combining electrical stimulation \nwith parallelised microscopy.  \nParallelised microscopy, in which an array of samples can be imaged individually without the need to \nmechanically translate the sample, has gained popularity in recent years. Unlike macroscopic imaging \nof the entire sample array, where each sample occupies a small fraction of the whole image, in parallel \nmicroscopy, each sample is imaged by the entire detector area, greatly increasing the achievable \nresolution from ~100 µm to less than 10 µm. Versions of parallelised microscopes include schemes \nthat have a separate excitation and detection path for each sample, using arrays of light sources, \nlenses, and cameras to capture image data from each sample simultaneously8,9. This approach has the \nadvantage of true simultaneity for up to 24 samples , but the machine vision cameras used lack the \nsensitivity required to detect the fluorescent emission from small bacterial microcolonies, which may \nbe less than 10 µm across. In addition, the large data rates associated with collecting data from 24 \ncameras at once (45 Gbps compared to 10 Gbps for USB 3.1 Gen 2.0) require custom data transfer \nmethods, limiting scalability and impact . An alternative method is to use a galvanometer mirror to \nsweep a macroscopic image of all samples across a single detector to increase sampling at the desired \nlocations, but this low N.A. (i.e. 0.04) imaging method lacks spatial resolution10.   \n“Random Access Parallel” (RAP) microscopy 11 uses an array of LED light sources to illuminate each \nsample individually, but uses a common detection path and therefore requires only one camera. This \nsequential acquisition approach has recently been applied to the measurement of cardiomyocyte \nmonolayer contraction (implemented as the “Exeter Multiscope”)12. The utility of using one detection \ncamera becomes particularly apparent in fluorescence imaging, as the prospect of having an array of \nlow-noise, high  quantum efficiency fluorescence cameras quickly becomes impractical in terms of \nmechanical integration and cost. We have developed a proof-of-concept parallelised microscope with \nthe ability to distinguish between viable and non-viable bacteria using a custom-built UV-fluorescence \nMultiscope that can read out a 2 × 2 array of bacterial samples, testing up to four different conditions, \n.CC-BY 4.0 International licenseperpetuity. It is made available under a \npreprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in \nThe copyright holder for thisthis version posted March 11, 2026. ; https://doi.org/10.64898/2026.03.10.710830doi: bioRxiv preprint \n\n3 \n \nin rapid succession, without any moving parts  in less than 5 minutes . K-means clustering is used to \nextract fluorescence intensity changes associated with bacterial clusters  from time lapse data . We \ndemonstrate the fluorescence detection sensitivity required to scale the readout to larger numbers of \nbacterial samples, including standard 96-well plate formats.   \n \nResults \nBaseline results from an inverted microscope   \nTo provide a baseline measurement of bacterial fluorescence changes in response to electrical \nstimulation, we first performed the viable/non-viable assay on an inverted microscope using a multi-\nelectrode array composed of gold electrodes on a coverglass-bottom dish (Figure 1A-B). The electrodes \nwere stimulated individually (100 Hz, 4V peak -peak, 2.5 seconds) using an external control box \n(‘CytePulse’, Cytecom Ltd.) connected to the electrode array via an IDE ribbon cable. Two of the six \nelectrodes were used to collect reference data (electrode #5 and #6, Figure 1B).  \nTo measure bacterial viability, Bacillus subtilis was cultured to log-phase and an optical density (OD) of \n1.5. 1 μL of this culture was pipetted onto one side of 5 × 5 mm agarose pad s containing ThT.  The \ninoculated side of each agarose pad was then placed in direct contact with the individual electrodes in \nthe array, ensuring close contact. The pads were then incubated at room temperature for 2 hours to \nallow bacterial proliferation. \nAfter loading the electrode array onto the microscope stage, a widefield fluorescence image of each \nsample (electrodes #5 and #6) was taken to confirm the presence of bacteria colonies (Figure 1C). \nFluorescence time lapse images were acquired for each pad in sequence using a 10× 0.4 N.A. objective \nlens and a UV LED source (400 nm, p300 ultra, CoolLED Ltd.). Each time lapse acquired 15 seconds of \nimages (1 image every second, 500ms exposure) before stimulation to provide a fluorescence \nreference value and then for a further 30 seconds after stimulation to monitor the change (Figure 1F).  \nThe first bacterial assay used proliferating, healthy B. subtilis cells. As t he process of electrical \nstimulation is detrimental to the health of the cell membrane, the best measure of the non-viable cell \nassay was to repeat the experiment on the same site whilst monitoring changes in cell fluorescence \nafter a second stimulation. This repeatable approach avoids complications associated with antibiotic-\nbased killing, such as incomplete loss of viability, while enabling interrogation of the same cell \npopulation on the same pad.   \nTo map out the temporal variations in fluorescence intensity over the field of view, a K-means \nclustering approach was taken. The temporal variation in fluorescence intensity was extracted for each \nindividual pixel in the 256 × 256 pixel region of interest (i.e. over 65,000 fluorescence traces). The K-\nmeans clustering algorithm then identified five fluorescence traces (or ‘clusters’) that best \ncharacterised the majority of traces present (in a way not dissimilar to principal components analysis). \nThe clusters were ranked according to the magnitude of fluorescence change, with the largest changes \nbeing the ‘top’ cluster (=5) and the smallest changes being the ‘bottom’ cluster (=1). The images in \nFigure 1D and Figure 1E have been colour coded according to the cluster to which each pixel has been \nassigned. A total of fi ve clusters were chosen to ensure that there were enough to describe the key \nfeatures anticipated in each field of view: bacteria in the focal plane, bacteria just beyond the focal \nplane, agarose pad, electrodes, plus an addition al cluster for any unanticipated features. Specifying \nsubstantially more than five clusters would result in redundant classifications with highly similar cluster \ncharacteristics. The clusters were not constrained to be equally\n.CC-BY 4.0 International licenseperpetuity. It is made available under a \npreprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in \nThe copyright holder for thisthis version posted March 11, 2026. ; https://doi.org/10.64898/2026.03.10.710830doi: bioRxiv preprint \n\n4 \n \n \npopulated, as the bacterial clusters occupy a relatively small fraction of the total pixel number.    \nAfter clustering, a heat map was produced describing the cluster assignment for each pixel (Figure 1D-\nE). The cluster that was most strongly associated with the in-focus bacteria (‘top cluster’) was used as \nthe best measure of the change in flu orescence intensity over time  (Figure 1F). The absolute \nfluorescence change was obtained from this cluster by calculating the peak fluorescence difference \nFigure 1: (A) Overview of the method used to electrically stimulate bacterial samples in an inverted microscope. (B) \nPlan view of an array of six electrode positions. Only positions #5 and #6 were used. (C) Widefield fluorescence \nimage of a bacterial colony, imaged through the electrodes (black vertical bars) on an inverted microscope. One \nregion of interest (ROI) was selected from within this larger field of view for processing. (D) Cluster assignment maps \nfor pixels within each ROI. The top cluster (most strongly correlated to bacterial clusters) was used to calculate the \nfluorescence change after stimulation. (E) Detail of cluster assignment map indicates the degree of heterogeneity \nwithin each colony of B. subtilis. (F) Changes in fluorescence observed after first stimulation (viable bacteria average \nresponse, black line) , and second stimulation (non -viable bacteria average response, red dashed line). Error bars \nindicate standard deviation (N=2). Scale bars: (C) 500 µm (D) 100 µm (E) 50 µm. \n.CC-BY 4.0 International licenseperpetuity. It is made available under a \npreprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in \nThe copyright holder for thisthis version posted March 11, 2026. ; https://doi.org/10.64898/2026.03.10.710830doi: bioRxiv preprint \n\n5 \n \nbetween stimulation and end points. The peak change in fluorescence for electrode positions #5 and \n#6 are recorded in Table 1. Table 1 also records the  mean pixel value of the traces for the top and \nbottom cluster from which the signal to background ratio is calculated. \nTable 1: Summary of fluorescence changes detected using the inverted microscope. Bacteria at each electrode position were \nstimulated twice before moving to the next electrode. \nElectrode \nposition  \nStimulation \nnumber  \nPeak \nfluorescence \nchange (Δf/f)  \nTop trace \navg. (a.u.) \nBottom \ntrace avg. \n(a.u.) \nSignal to \nbackground  \n5 1 +4.9% 622.3 4.4 141.4 \n5 2 -12.0% 585.4 4.3 136.1 \n6 1 +3.2% 449.8 3.8 118.4 \n6 2 -6.0% 331.6 4.0 82.9 \n \nDetection of bacterial viability using the Multiscope  \nThe RAP/Multiscope approach to parallelised microscopy is outlined in Figure 2. The samples are \narranged on a 9 mm pitch (as in a 96 well plate) and are indicated by the electrode positions in Figure \n1B. Each sample is illuminated by collimated light from a UV LED. Baffles reduce cross-talk between \neach of the illumination paths, ensuring each LED only illuminates one sample. An objective lens below \neach sample collects emitted light to form an image at infinity. The final element in this system has a \ndiameter large enough to cover all of the samples in the array and acts to both steer and focus the \nlight from each sample to a common exit pupil, which is in the same plane as the camera. T his final \nelement could be a transmissive lens, but for the purposes of maintaining a compact setup, a reflective \nparabolic mirror was used. To maintain the camera at the  focus of the parabolic mirror, whilst not \nobscuring the light from the sample, the ax is of the mirror was offset from the centre of the array. A \nphotograph of the prototype system used in this project is shown in Figure 2(B).  \nA parallelised fluorescence microscope was built following the design of the Exeter Multiscope12, which \nin turn is based on the original RAP prototype11. Previous designs had been bright field only, detecting \ntransmitted light with a machine vision camera. The key difference in this work was that we were \ndetecting UV/VIS fluorescence and required UV excitation, and a scientific CMOS camera to achieve \nthe required fluorescence sensitivity. Due to the in -line nature of the fluorescence detection, there \nwas a need for strong filtering to block directly transmitted light, requiring the use of OD6 excitation \n(390 ± 20 nm) and emission (494 ± 10 nm) filters. The objective lenses had a diameter of 9 mm (limited \nby the sample pitch) and a focal length of 18 mm, providing an NA of 0.25 which was not sufficient to \nresolve individual bacteria, but bacterial microcolonies could be clearly discerned (Figure 2E).  \nThe electrode array was loaded onto the sample tray and time lapse fluorescence images were \nacquired of B. subtilis samples at electrodes #1, #2, #5 and #6 (electrode positions #3 and #4 were not \nused as they were not centred on the 9 mm pitch of the objective lens array, Figure 1B). As previously, \ntwo sets of time lapse data were acquired for each electrode position, corresponding to two separate \nelectrical stimulations to compare viable/non-viable responses. The Multiscope provided all -optical \nswitching between the four fields of view by turning on the LED associated with  each electrode \nposition ( Figure 2A). Only one field of view was imaged at any one time. An example widefield \nfluorescence image of electrode #2 is shown in Figure 2 C. Within each 2048 x 2048 raw image , a \n.CC-BY 4.0 International licenseperpetuity. It is made available under a \npreprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in \nThe copyright holder for thisthis version posted March 11, 2026. ; https://doi.org/10.64898/2026.03.10.710830doi: bioRxiv preprint \n\n6 \n \nsmaller 256 x 256 pixel ROI was selected to reduce processing times for the K-means clustering \nanalysis. ROIs were located where the bacterial colonies appeared in sharpest focus. To reduce the \nhigh level of background fluorescence, these ROI images were first processed using a rolling ball \nFigure 2: (A) Overview of the method used by the Multiscope to parallelise imaging of multiple samples . Each sample is \nilluminated one at a time and the images are relayed to the same sCMOS camera. ExF = excitation filter. EmF = emission filter. \n(B) A photograph of the Multiscope system constructed to image bacterial fluorescence (see Methods for full description  of \nindividual components (i) – (viii)). (C) Widefield fluorescence images of the bacteria sample mounted on agarose.  Regions of \ninterest which appeared in best focus were selected for processing. (D) Processed ROIs after K-means clustering. (E) Magnified \nimage of ( D). (F) Averaged fluorescence profiles for the top cluster in each K-means analysis. Profiles corresponding to first \nstimulation (black line) and second stimulation (red dashed line). Scale bars: (C) 500 µm (D) 100 µm (E) 50 µm.  \n.CC-BY 4.0 International licenseperpetuity. It is made available under a \npreprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in \nThe copyright holder for thisthis version posted March 11, 2026. ; https://doi.org/10.64898/2026.03.10.710830doi: bioRxiv preprint \n\n7 \n \nsubtraction (50 pixel diameter) in FIJI 13. The K-means clustering algorithm was then applied to the \nprocessed ROI data and each pixel assigned to a cluster (Figure 2D-E).  \nThe time sequence of each pixel value in the ROI was then analysed using K-means clustering as for \nthe inverted microscope data . The ‘top’ cluster (i.e. that most closely associated with the bacterial \nmicrocolony pixels) was then extracted. The peak difference in fluorescence intensity after stimulation \nwas determined and the results are summarised in  Table 2. The data set for electrode #6 was \noverwritten by the second stimulation and could not be analysed. To provide an indication of the signal \nto background ratio, Table 2Table 2 also includes the mean value of each of the top cluster traces \n(‘signal’) and the bottom cluster traces (‘background’) where the bottom cluster corresponds to the \npixels obscured by the opaque electrodes. The absolute change in fluorescence brightness appears to \nbe approximately correlated with the signal to background ratio.  \nTable 2: Summary of fluorescence changes detected using the Multiscope  at a bacteria density of \nOD=2.5. Bacteria at each electrode position were stimulated twice before moving to the next \nelectrode.  \nElectrode \nposition  \nStimulation \nnumber  \nPeak \nfluorescence \nchange (Δf/f)  \nTop trace \navg. (a.u.) \nBottom \ntrace av g. \n(a.u.) \nSignal to \nbackground  \n1 1 4.0% 61.1 7.9 7.7 \n1 2 -6.7% 60.3 7.9 7.6 \n2 1 13.9% 74.0 8.7 8.5 \n2 2 -10.7% 74.5 8.7 8.8 \n5 1 9.4% 126.4 9.2 13.7 \n5 2 -10.6% 120.3 8.9 13.5 \n6 2 -4.7% 99.5 9.3 10.6 \n \nMultiscope data at lower cell densities \nTo further explore the sensitivity  limits of the UV fluorescence Multiscope, we repeated  the \nexperiments at  a lower  density preparation of B. subtills  (OD=1.5) for agarose pad inoculation . \nReducing the bacterial density used for agarose pad inoculation offers additional clinical benefit by \nenabling earlier identification of effective antimicrobial compounds. Experiments were repeated as for \nthe OD=2.5 samples. Multiscope widefield fluorescence images from each of the electrode positions \nare shown in Figure 3A, together with the selected ROI (red square). The processed data indicate that \nin some cases, such as electrode #1 (Figure 3B-C), the bacteria were very sparse.  \nThe top cluster profiles are shown in Figure 3D. None of the traces indicate a significant change in \nfluorescence brightness. As shown in Table 3, the changes are all small with slightly negative changes \nin fluorescence. It is possible that, even though the signal to background levels are similar to that at \nhigh cell densities, there are simply not enough cells to register a significant change in fluorescence. \nDoubling the number of clusters from five to ten did not significantly affect the results. The more \ndefocused images of the sample led to poorer rolling ball background subtraction leaving higher \nbaseline levels of fluorescence for some electrode positions.  \n \n.CC-BY 4.0 International licenseperpetuity. It is made available under a \npreprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in \nThe copyright holder for thisthis version posted March 11, 2026. ; https://doi.org/10.64898/2026.03.10.710830doi: bioRxiv preprint \n\n8 \n \nTable 3: Measurements of peak fluorescence changes for each of the Multiscope experiments for a \nlow bacteria density (OD=1.5). \n \nDiscussion  \nCombining electrical stimulation with Multiscope readout enables the clear determination of viable \nfrom non-viable bacteria (Figure 2F). Differences in the amplitude of the fluorescence change (Figure \n3E) are linked to both (i) contrast in the sample image and (ii) the strength of the fluorescence signal. \nImage contrast is maintained by keeping the sample in focus and minimising aberrations. Variations in \nsample focusing is thought to be largely due to the mounting of the electrode array in the Multiscope. \nTension in the IDC cable produced a small uplift of electrode positions #2 and #5 relative to #1 and #6 \n(Figure 2B). Focusing the objective array on electrodes #2 and #5 led to reduced contrast in sample \nimages at positions #1 and #6, as reflected in Table 2.  \nElectrode \nposition  \nStimulation \nnumber  \nPeak \nfluorescence \nchange (Δf/f)  \nTop trace avg. \n(a.u.) \nBottom trace \navg. (a.u.) \nSignal to \nbackground  \n2 1 -9.0% 507.4 42.8 11.8 \n2 2 -3.5% 65.3  6.5 10.0 \n5 1 -2.1% 67.2  6.3 10.7 \n5 2 -4.2% 64.2  6.4 10.0 \n6 1 -2.4% 48.9  5.2 9.4 \n6 2 -3.5% 48.9  5.2 9.4 \nFigure 3: (A) Widefield fluorescence images of the bacteria samples mounted on agarose imaged through the opaque \nCytePulse electrodes. Regions of interest (red squares) were selected for processing. Example shown is a live cell response \nafter the first stimulation. (B) Processed ROIs after K-means clustering.  (C) Magnified view of (B).  (D) Change in the \nfluorescence brightness for the top cluster averaged over three electrode positions after the first ( black line) and second \n(red dashed line) stimulation. (E) Change in fluorescence brightness after first (S1) and second (S2) stimulation for the three \nconditions considered above (Multi=Multiscope, Invert=Inverted microscope).  \n.CC-BY 4.0 International licenseperpetuity. It is made available under a \npreprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in \nThe copyright holder for thisthis version posted March 11, 2026. ; https://doi.org/10.64898/2026.03.10.710830doi: bioRxiv preprint \n\n9 \n \nAberrations from the parabolic reflector will also contribute to the reduction in image contrast. \nComatic aberrations increase as (𝐵 𝑓⁄ )2 where 𝐵 is the lateral distance of the sample from the optic \naxis of the mirror and 𝑓 is the focal length of the mirror. This can be improved greatly by increasing the \nfocal length of the mirror from the current value of 100 mm to around 300 mm and thereby reducing \nthe comatic aberrations by a factor of nine. A longer parabolic mirror focal length would also increase \nthe standoff distance of the sCMOS camera, reducing the number of sample positions that might be \nobscured by the camera itself.   \nThe strength of the fluorescence signal has been shown to be dependent on both how well the sample \nwas focused,  and the abundance of bacterial colonies present in the sample, with ODs of 1.5 \ngenerating insufficient fluorescence signal. This could lead to the top cluster expanding to include more \ndim pixels from defocused bacterial microcolonies , thereby “averaging out” the strength of the \nfluorescence response. This is indicated by the number of top cluster pixels remaining approximately \nthe same for the OD=1.5 samples (Figure 3B), despite the approximately two-fold reduction in bacterial \nabundance.  \nFuture improvements in system throughput could be enabled by reducing the current 2.9 second \nintegration time required to accumulate sufficient fluorescence signal from the bacterial monolayer. \nThese long integration times undermine the benefits of all-optical switching between samples. Using \nUV sources that are brighter than the current 35 mW LEDs is one option. As only 1 mW of excitation \nlight currently reaches the sample , there is significant potential to increase the optical coupling \nbetween source and sample by removing the plastic bulb covering the LED emitter and using higher \nNA collection lenses with higher transmission UV coatings. Transitioning from an in-line to an epi -\nfluorescent imaging architecture would also greatly increase contrast in the acquired images.  The \nincrease in signal to background ratio would also open up the possibility of obtaining reliable measures \nof viability at lower bacterial concentrations, even enabling imaging direct from biofluid, bringing the \ntechnology in line with competing AST approaches2.  \nIf the camera integration time can be reduced from 2.9 seconds to around 500 ms (as used on the \ninverted microscope), then it would easily allow all four samples to be imaged in a single 45 second \nexposure by optically moving from one sample to the next between camera exposures. The raw image \nsequence could then be de-interlaced into four time-lapse sequences. If the integration time could be \nreduced further to a more typical 50 ms exposure , this would allow over 50 wells to be imaged \nsimultaneously, including additional time budget for streaming images to disk.  This is a significant \ntarget as it would allow for bacterial viability measurements in the presence of seven concentrations \n(plus controls) of the six main antibiotics administered clinically in acute infections (i.e. Piperacillin–\ntazobactam, Ceftriaxone, Gentamicin, Metronidazole, Vancomycin, Meropenem).  \nFinally, it can be seen in Figure 1C that the bacterial clusters were rarely distributed evenly across the \nagarose pad but would instead accumulate in a ‘coffee ring’ pattern left after the inoculation droplet \nhad dried on the agarose pad. One potential modification to the protocol would be inoculating the \nagarose pad via spraying, as demonstrated in [14] to achieve a more uniform distribution  across the \npad.  \n \nConclusion  \nWith the rapid emergence of AMR there is a growing and urgent need for rapid diagnostic methods \ncapable of identifying antibiotic efficacy. Established ASTs are culture-based and inherently slow: even \nafter a pure bacterial isolate has been obtained, standard AST methods typically require 24-48 hours \n.CC-BY 4.0 International licenseperpetuity. It is made available under a \npreprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in \nThe copyright holder for thisthis version posted March 11, 2026. ; https://doi.org/10.64898/2026.03.10.710830doi: bioRxiv preprint \n\n10 \n \nof incubation to generate results. In contrast, bacterial optical electrophysiology can provide viability-\ndependent fluorescence readouts in under one minute , offering a fundamentally faster route for \nassessing antimicrobial effectiveness. To date, all previous implementations of the fluorescent \nmeasure of bacterial viability using electrical stimulation have used a standalone research grade \nmicroscope and require a high degree of user intervention to obtain bacterial viability data. In this \nwork, we have demonstrated the viability of using an automated Multiscope approach to parallelise \noptical excitation and readout of fluorescence changes in bacterial monolayers. This provides an \nimportant stepping stone towards high throughput screening of bacterial populations using rapid \noptical electrophysiology methods. Whilst the initial test used a relatively modest number of testing \nsites (a 2 x 2 array) there is clear potential for scaling the number of tests to 50 or more wells using the \nsame architecture, providing measures of antibiotic efficacy in one to two minutes. By matching rapid \ntesting with high throughput, the goal of reducing bottlenecks in clinical AST can be achieved.  \n \nMethods \nMultiscope construction  \nThe prototype system in shown in Figure 2B. The light path starts with a 4 × 4 array of LEDs (RS \nComponents #903-3771) which output up to 35 mW in the range 390-425 nm. As the light was emitted \nover a 30 degree cone angle, baffles were used to avoid light spilling over into adjacent well positions \nand creating cross-talk. The baffles were 3D printed as a 4 × 4 array of opaque tubes. These tubes were \n6 mm in diameter and had a length which corresponded to the 9 mm focal length of the collimation \nlenses. The baffle was mounted onto the underside of the custom LED array and together occupied \nthe first tray position (Figure 2B(i)). The second tray position was occupied by a 3 × 3 array of 9 mm \ndiameter, 9 mm focal length collimation lenses (Edmund Optics, #65-549-INK). The short focal length \nwas chosen to maximise the light transmitted onto the bacterial sample. By moving the lens array \nfractionally further away from the LED array, it was possible to focus the excitation light onto only the \nregion of the bacterial sample that lies within the field of view of the objective lenses, further \nincreasing the light intensity at the sample. A 50 mm diameter, OD6 excitation filter with a 390  ± 20 \nnm pass band (Edmund Optics #86-359) was placed after the collimation lenses. The filter was aligned \nto the well positions of the 4 × 4 LED array within a 3D printed mount that occupied the third well plate \nholder position (Figure 2B(iii)).   \nThe CytePulse electrode array occupied the fourth tray position (Figure 2B(iv)). Directly underneath \nthis layer was a 2 × 2 array of imaging objective lenses, centred on the samples (Figure 2B(v)). The four \npositions corresponded to electrode positions 1, 2, 5 and 6 (Figure 1(B)). These lenses (Edmund Optics \n#47-653, 9 mm diameter, 18 mm focal length, NA=0.25) collected the fluorescence emission from the \nbacteria and relayed this to the sCMOS camera (Hamamatsu ORCA Flash4.0 V2) via a parabolic mirror \n(100 mm focal length, 220 mm diameter, Edmund Optics #68-793). A 1” OD6 emission filter with pass \nband 494±10nm (Edmund Optics #84-096) was fixed immediately in front of the fluorescence camera \nto further suppress excitation light.  \nThe 0.25 NA objective lenses provided both better light collection, higher spatial resolution and smaller \ndepth of field to improve the sensitivity to fluorescence changes within the thin bacterial monolayer. \nThe theoretical resolution of 1.2 µm combined with residual aberrations from the parabolic mirror \nmeant it was not possible to resolve individual bacteria. The shallow depth of field required a finer \ndegree of control over the objective lens position. This was provided by a single axis piezo ( Figure \n2B(vi), Mad City Labs MMP1) which achieved a minimum step size of 95 nm over a 25 mm range.   \n.CC-BY 4.0 International licenseperpetuity. It is made available under a \npreprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in \nThe copyright holder for thisthis version posted March 11, 2026. ; https://doi.org/10.64898/2026.03.10.710830doi: bioRxiv preprint \n\n11 \n \nFluorescence then reflected off the parabolic mirror (2B(vii)) and was collected by the sCMOS camera \n(Figure 2B(viii)). The magnification of the system is 100 mm / 18 mm = 5.5 ×. The maximum field of \nview of the system is therefore given by the size of the imaging chip (2048 × 6.5 um = 13.3 mm) divided \nby the magnification (13.3 mm / 5.5 = 2.4 mm square). To minimise stray light incident upon the \ncamera, emanating from outside of the sample field of view, a 3D printed plate was used. This 2 mm \nthick plate contained an array of 3 mm diameter holes that acted as a field stop when centred on the \nCytePulse electrode locations and was aligned with the camera to avoid any clipping of the field of \nview.  \n \nData acquisition and hardware control  \nThe Multi-dimensional acquisition toolbox in MicroManager15 was used to trigger the sCMOS camera \nand define the integration time. For the inverted microscope (IX73, Olympus), images were acquired \non an sCMOS camera (Zyla 4.2, Andor) every second (500 ms exposure), on a 10× 0.4 N.A. dry objective. \nFor the acquisition of Multiscope image data, a 2.9 second camera integration time on a 3 second \ninterval was used for a total of 20 exposures. Electrical stimulation was manually triggered vi a the \nCytePulse graphical user interface 15 seconds after initiating image acquisition (i.e. after 15 frames on \nthe inverted system and 5 frames on the Multiscope). Electrical stimulation of 4V amplitude and 100 \nHz frequency for 2.5 seconds was used in all cases. All images in the sequence were saved directly to \ndisk.  \n \nSample preparation \nB. Subtilis 168 streak plates were prepared from cyrostock by streaking onto lysogeny broth (LB) agar \nplates and incubating overnight at 37 °C. Stationary-phase cultures were prepared by inoculating fresh \nLB with a single colony and incubating for 17 h at 37 °C with shaking. Log-phase cultures were prepared \nby diluting overnight cultures into fresh LB and incubating until mid-log phase (OD600 ≈ 1.5 or 2.5). Log-\nphase cells were used immediately for preparation of imaging samples. \nA defined CytePulse medium (see Supplementary information) was prepared to support bacterial \nviability during imaging. MOPS and potassium phosphate buffers were adjusted to pH 7 and \nautoclaved, whereas trace-ion and carbon/nitrogen supplements (including amino acids, glucose, \nmetal salts, and thiamine), and ThT were filter sterilised to avoid thermal degradation. ThT was \nincluded to enable fluorescence -based monitoring. Agarose pads (2% w/v) were prepared by \ncombining molten agarose with the CyteCom medium. \nAgarose pads were cast under aseptic conditions by  pipetting molten agarose onto 22 × 22 mm \ncoverslips and gently placing a second coverslip on top to form a uniform layer. After solidification, the \nupper coverslip was carefully removed, and each pad was cut into 16 equal sections with a sterile \nscalpel. Log-phase bacteria were spotted onto individual sections with a pipette and allowed to adsorb. \nEach agarose section was then inverted onto a CytePulse electrode such that the bacterial layer \ncontacted the electrode surface while avoiding the central ground electrode. The pads were then \nincubated at room temperature for 2 hours. Loaded electrodes were placed on a custom 3D-printed \nholders and transferred to either the inverted microscope or Multiscope for electrical stimulation and \ntime-lapse imaging. This preparation yielded consistent single -layer bacterial samples suitable for \ncombined CytePulse stimulation and fluorescence imaging. \n.CC-BY 4.0 International licenseperpetuity. It is made available under a \npreprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in \nThe copyright holder for thisthis version posted March 11, 2026. ; https://doi.org/10.64898/2026.03.10.710830doi: bioRxiv preprint \n\n12 \n \nReferences \n1. Interagency Coordination Group on Antimicrobial Resistance. No Time to Wait: Securing the \nFuture from Drug-Resistant Infections. https://www.who.int/docs/default-source/documents/no-\ntime-to-wait-securing-the-future-from-drug-resistant-infections-en.pdf (2019). \n2. Reszetnik, G. et al. Next-generation rapid phenotypic antimicrobial susceptibility testing. Nat. \nCommun. 15, 9719 (2024). \n3. Spencer, D. C. et al. A fast impedance-based antimicrobial susceptibility test. Nat. Commun. 11, \n5328 (2020). \n4. Stratford, J. P . et al. Electrically induced bacterial membrane-potential dynamics correspond to \ncellular proliferation capacity. Proc. Natl. Acad. Sci. 116, 9552–9557 (2019). \n5. Cytecom Limited. https://www.cytecom.co.uk/. \n6. Matsumoto, K. et al. Advanced CUBIC tissue clearing for whole-organ cell profiling. Nat. Protoc. \n14, 3506–3537 (2019). \n7. Gintoli, M. et al. Spinning disk-remote focusing microscopy. Biomed. Opt. Express 11, 2874 (2020). \n8. Harfouche, M. et al. Imaging across multiple spatial scales with the multi-camera array \nmicroscope. Optica 10, 471 (2023). \n9. Thomson, E. E. et al. Gigapixel imaging with a novel multi-camera array microscope. eLife 11, \ne74988 (2022). \n10. Symvoulidis, P . et al. NeuBtracker—imaging neurobehavioral dynamics in freely behaving \nfish. Nat. Methods 14, 1079–1082 (2017). \n11. Ashraf, M. et al. Random access parallel microscopy. eLife 10, e56426 (2021). \n12. Mohanan, S. et al. Automated measurement of cardiomyocyte monolayer contraction using \nthe Exeter Multiscope. Biomed. Opt. Express 16, 4716–4729 (2025). \n13. Schneider, C. A., Rasband, W. S. & Eliceiri, K. W. NIH Image to ImageJ: 25 years of image \nanalysis. Nat. Methods 9, 671–675 (2012). \n.CC-BY 4.0 International licenseperpetuity. It is made available under a \npreprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in \nThe copyright holder for thisthis version posted March 11, 2026. ; https://doi.org/10.64898/2026.03.10.710830doi: bioRxiv preprint \n\n13 \n \n14. Cano, Á. et al. Multiparametric quantification of bacterial cells using digital holographic \nmicroscopy. Sci. Rep. 15, 41051 (2025). \n15. D. Edelstein, A. et al. Advanced methods of microscope control using μManager software. J. \nBiol. Methods 1, 1 (2014). \n \n.CC-BY 4.0 International licenseperpetuity. It is made available under a \npreprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in \nThe copyright holder for thisthis version posted March 11, 2026. ; https://doi.org/10.64898/2026.03.10.710830doi: bioRxiv preprint","source_license":"CC-BY-4.0","license_restricted":false}