Single-cell electro-phenotyping for rapid assessment of Clostridium difficileheterogeneity under vancomycin treatment at sub-MIC (minimum inhibitoryconcentration) levels
Ali Rohania, John H. Moorea, Yi-Hsuan Sua, Victoria Stagnaroa, Cirle Warrenb, Nathan S. Swamia,⁎
a Electrical & Computer Engineering, University of Virginia, United Statesb Infectious Diseases, School of Medicine, University of Virginia, United States
Current methods for measurement of antibiotic susceptibility of pathogenic bacteria are highly reliant on mi-crobial culture, which is time consuming (requires> 16 h), especially at near minimum inhibitory concentration(MIC) levels of the antibiotic. We present the use of single-cell electrophysiology-based microbiological analysisfor rapid phenotypic identification of antibiotic susceptibility at near-MIC levels, without the need for microbialculture. Clostridium difficile (C. difficile) is the single most common cause of antibiotic-induced enteric infectionand disease recurrence is common after antibiotic treatments to suppress the pathogen. Herein, we show that de-activation of C. difficile after MIC-level vancomycin treatment, as validated by microbiological growth assays,can be ascertained rapidly by measuring alterations to the microbial cytoplasmic conductivity that is gauged bythe level of positive dielectrophoresis (pDEP) and the frequency spectra for co-field electro-rotation (ROT).Furthermore, this single-cell electrophysiology technique can rapidly identify and quantify the live C. difficilesubpopulation after vancomycin treatment at sub-MIC levels, whereas methods based on measurement of thesecreted metabolite toxin or the microbiological growth rate can identify this persistent C. difficile subpopulationonly after 24 h of microbial culture, without any ability to quantify the subpopulation. The application ofmultiplexed versions of this technique is envisioned for antibiotic susceptibility screening.
1. Introduction
The global rise of antibiotic-resistant bacterial strains and antibioticpersistence has spurred an urgent need for technologies capable of ra-pidly screening susceptibility of particular strains to key antibiotics,preferably without requiring time-consuming microbial culture steps.The unavailability of such technologies, especially within point-of-careor resource-poor settings, has been directly implicated in the wide-spread prescription of broad-spectrum antibiotics [1], since currentmethods based on microbial culture to assess antibiotic susceptibilityrequire 48–72 h to go from sample to answer [2]. While conventionalmicrobiological assays, including broth dilution, e-test strips, and diskdiffusion [3] are highly reliable indicators of antibiotic susceptibility,their use of turbidity to measure microbial growth, limits their sensi-tivity to 107 CFU/mL levels of bacterial sample and growth times of16–20 h are required for detectable signals [4]. Molecular methodsbased on amplification of genomic DNA that are essential for the pur-pose of screening during infectious outbreaks [5] are not practical for
broad-spectrum screening, since the genetic markers for emerging an-tibiotic-resistant strains are not often known, while the high frequencyof genetic mutation under selective environmental pressure causes ge-netic markers to be less reliable [6], with mutations that may not bewell-correlated to antibiotic susceptibility [7], and with combinatorialsmall effects of resistance genes that are difficult to measure [8]. Whileantibiotic susceptibility assessment should be possible based on mon-itoring alterations in microbial phenotypic properties, the sensitivityand specificity of these techniques in identifying and isolating persis-tent microbial subpopulations after antibiotic treatment, at levels closeto their minimum inhibitory concentration (MIC) has not been de-monstrated.
Various microfluidic platforms have been implemented to trackantibiotic susceptibility of microbials by using culture-dependent andculture-independent methods. In this context, electrophysiology-basedmethods are especially powerful since electrical polarization is highlysensitive to size and shape variations of the microbial cell, while dif-ferent subcellular regions, such as the cell wall, membrane and
https://doi.org/10.1016/j.snb.2018.08.137Received 3 April 2018; Received in revised form 23 August 2018; Accepted 26 August 2018
⁎ Corresponding author at: 351 McCormick Road, Charlottesville, VA 22904-1000, United States.E-mail address: [email protected] (N.S. Swami).
Sensors & Actuators: B. Chemical 276 (2018) 472–480
cytoplasm respond within different frequency ranges to enable dis-tinction of the cell region affected by the antibiotic in a culture-in-dependent manner [9]. Additionally, the altered polarization frequencyresponse can be used towards dielectrophoretic (DEP) isolation of livemicrobial subpopulations after critical levels of particular antibiotics,for downstream analysis. In prior work, antibiotic-sensitive versus an-tibiotic-resistant strains of Staphylococcus epidermidis [10], as well asthose of E.coli [11] were distinguished based on conductivity differ-ences of their cell wall region after antibiotic treatment, as deduced bytheir dielectrophoretic collection spectra. In another study [12], thegentamicin resistant Staphylococcus epidermidis strain was reported tohave substantially lower dielectrophoretic mobility under DC fieldsversus the respective susceptible strain. In another study of great mi-crobiological relevance [13], dielectrophoretic focusing enabled highlysensitive optical monitoring of E.coli for polymyxin susceptibility,which enabled multiplexing and automated parallel testing of multipledrugs. In other studies on Mycobacterium smegmatis [14,15], the lowfrequency DEP spectra where cell wall characteristics dominate, wereused to show distinct differences in conductivity for active versusdormant and non-viable strains. Since microbial cells can undergo ei-ther elongation or swelling after treatment with various antibiotics[16,17], the respective shape changes interfere with image-based de-tection of cell division [18]. However, these shape changes can besensitively probed based on electrical polarization [19]. In a recentstudy, the susceptibility of E. coli to the antibiotic Cephalexin [20] andof several gram-negative bacteria to β-lactam antibiotics [21] weretracked by using positive DEP (pDEP) to fix bacterial cells at well-de-fined positions along the focal plane for enabling facile and accurateestimation of cell length and elongation rate. While there has beensome prior work on antibiotic susceptibility analysis by dielectrophor-esis [22–24], this work has not been performed at near-MIC antibioticlevels, for determination of alterations to the electrophysiologicalphenotype within a few hours of treatment, as performed in the currentwork using single-microbial phenotyping.
The persistence of a small fraction of microbials to antibiotics hasbeen attributed to heterogeneity in microbial responses at critical an-tibiotic levels [25]. Hence, subpopulations can emerge even though thestarting sample is an isogenic population [26]. Clostridium difficile (C.difficile) infection (CDI) is a toxin mediated intestinal disease [27],which is the single most common cause of antibiotic-induced entericinfection associated with increased morbidity and in-hospital fatalities[28]. Metronidazole and vancomycin are the most commonly usedantibiotics for CDI treatments, but patients often experience diseaserecurrence after standard antibiotic treatments [29] and a great ma-jority of patients with suppressed C. difficile levels during antibiotictreatments were identified as asymptomatic carriers of C. difficile withinweeks of completion of the treatment [30]. Additionally, some C. dif-ficile strains have demonstrated resistance to metronidazole [31]. Tra-ditional metrics to quantify antibiotic susceptibility based on the halfmaximal inhibitory concentration (IC50) or minimum inhibitory con-centration (MIC) are end-point measurements, which are not suitedtowards detecting heterogeneity in responses within a fluctuating en-vironment. Single-cell techniques, on the other hand, for electro-physiology-based phenotypic analysis within a droplet [32] or by usingimpedance [33–35], or electro-rotational analysis [36,37], are capableof quantifying subpopulations, but they have not been applied towardsstudying microbial heterogeneity after antibiotic treatments. In priorwork, we have demonstrated that C. difficile cell wall capacitancemeasurements can identify subtle alterations in S-layer glycoproteins,which can occur due to strain-based morphological differences [38] ordue to cell wall reconstitution under probiotic interactions [39], andwhich are correlated to the colonization ability of C. difficile [40]. In thecurrent study, we demonstrate that alterations in cytoplasmic con-ductivity gauged based on the pDEP level and quantified using single-cell electro-rotational (ROT) analysis can serve as a rapid, sensitive andspecific means to quantify the persistent live C. difficile subpopulation
after sub-MIC level vancomycin treatment, whereas measurement of themicrobiological growth rate or secreted metabolite toxin requires 24 hof culture to identify this persistent C. difficile subpopulation, with noability for its quantification.
2. Experimental
2.1. Microbial culture and antibiotic treatment
Highly toxigenic C. difficile (HTCD) VPI 10463, low toxigenic(LTCD) ATCC 630, and non-toxigenic (NTCD) VPI 11186 strains (ATCC,Manassas, VA) were cultured in brain heart infusion (BHI) broth (BDBBL Brain Heart Infusion) at 37 °C for 20 h under anaerobic conditions.Antibiotic treatment at 1mg/L and 2mg/L vancomycin levels (note:mg/L equals μg/mL) was performed on the HTCD strain with ∼106
cells/mL in the culture media, for a period of up to 4 h to ensurecomplete disinfection. Strain growth following antibiotic treatment wasmeasured based on absorbance at 600 nm (in triplicate) using a spec-trophotometer (Eppendorf AG, Germany) at 0, 4, 8, and 24 h post-treatment. Toxin production was determined using a C. difficile TOX A/B II ELISA (Techlab, Blacksburg, VA) as per user instructions and ab-sorbance (at 450 nm–620 nm) on the Epoch microplate reader (BioTek,Winooski, VT). Two-way ANOVA with Tukey’s multiple comparison testwere performed with Prism (Graphpad, San Diego, CA).
2.2. Dielectrophoresis and electro-rotation measurements
Following vancomycin treatment for 4 h, the BHI broth was re-placed with 8.8% sucrose water and re-adjusted with BHI broth tooptimize the medium conductivity at 105 ± 5% mS/m for electro-ro-tation (ROT) and dielectrophoresis (DEP) analysis. For ROT experi-ments, a PDMS (Poly-di-methyl-siloxane) microfluidic chamber withinlet and outlet tubing connected to flow control systems was bondedon a glass chip that was pre-patterned with planar hyperbolic-shapedgold electrodes arranged within a quadrapole pattern. Electrical con-nections were enabled using a printed circuit board with pins for in-dependent energization of electrodes (0.1–40MHz). A four-channelarbitrary waveform generator (TGA12104–Thurlby ThandarInstruments) was programmed to apply 90° phase shifted signals toadjacent electrodes. Highly parallelized electrorotational measurements[41] of ∼20 C. difficile cells per chip were observed under Zeiss Z1microscope and recorded by Hamamatsu CMOS camera at 100 fps(frames per second). The average rotation rate at each frequency wasobtained by intensity thresholding [42] to identify cell edges for ana-lyzing 80–100 cells per treatment group or strain. All the rotation rateswere normalized with respect to the maximum rotation rate of un-treated C. difficile, for comparison purposes, while the model fits wereperformed using rpm of the rotation rates. For dielectrophoresis (DEP)measurements, PDMS micromolding was used to fabricate micro-channels with sharp lateral constrictions (1000 μm to 15 μm) andbonded to a glass coverslip for easy microscopic viewing of DEP be-havior. Using Pt electrodes at the inlet and outlet, AC fields were ap-plied over a wide-frequency range (50 kHz-1MHz) by utilizing a poweramplifier [43] for particle trapping towards or away from high fieldpoints at the constriction tips. The trajectory of the unlabeled C. difficileafter each antibiotic treatment was observed under this field, as highframe per second movies to quantify the DEP velocity [44]. The dataacquisition was automated to enable the capture of movies at eachfrequency within about ten seconds, with the entire frequency spectrumcompleted within about 5min. The same cells can be measured multipletimes under the DEP field since we use the electrode-less DEP techniqueunder high frequency AC fields, with applied fields less than 300 Vpp/cm, thereby obviating electro-permeabilization effects, which we con-firmed through viability analysis on microbials under the DEP field. TheDEP and ROT analysis were conducted on the microbials following thelog phase stage of their culture period, to ensure insensitivity of the
A. Rohani et al. Sensors & Actuators: B. Chemical 276 (2018) 472–480
473
analysis to the temperature of the culture media and time for the cul-ture.
3. Theory and calculation
3.1. Dielectric model fits to compute electrophysiology
Based on the rod shape of C. difficile, we employed the ellipsoidalsingle shell model [45–47] to extract dielectric properties for the cellenvelope (wall plus membrane) and cytoplasm of C. difficile. A rotatingelectric field induces dipoles on the target bio-particles that are sus-pended in electrolytic medium. If angular velocity of the rotating fieldis substantially greater than the time for dipole formation on the par-ticle, then the resulting lag between the dipole moment and the electricfield causes a torque on the particle, thereby resulting in asynchronousrotation within the field. We employed effective moment method toderive expressions for the net torque on an ellipsoid in AC electric field.In all these calculations, frequency of applied field is high enough toensure that particles respond to the time averages. For planar electrodes(x–y), the time averaged torque in the z-direction (<Tz>) and effec-tive dipole (peff) are given by:
⟨ ⟩ = −T Re p E p E12
[( ) ( ) ]z eff y x eff x y0,*
0,*
(1)
= =p πabcε K E α x y( ) 4 ,eff α m α α0, (2)
In order to generate a torque, there is need for a phase differencebetween the applied electric field (
⎯→⎯E ) and dipole (→p ). For this purpose,
the electric field is represented in its complex phasor form, with realand imaginary parts. Hence,
⎯→⎯E represents electric field, E* represents
complex conjugate of the electric field, and E0 represents intensity ofthe electric field. The term “Kα” is the spheroidal complex ClausiusMossotti factor for a single layer particle (similar to fCM for sphericalparticles) along α direction. a, b, and c are the axis lengths in differentCartesian directions (x, y, z) and εm is the permittivity of medium. For arotating electric field = = ±E E and E jEx y0 0, the torque in z directionwill be:
⟨ ⟩ = − +T πabc ε E Im K K23
[ ]z m x y02
(3)
In the single-shell ellipsoidal model, εs represents permittivity of theshell, and εp represents permittivity of the particle core that is envel-oped by the shell. For an ellipsoidal particle with a single shell, this canbe written as:
= =− + + −
+ − + − −=K K
ε ε γK ε L ε εε L ε ε K γL L K ε ε
α x y( ) 3 ( ( ))
3 ( ( ) 9 (1 ) ( )( , )α α
s m α s α m s
m α s m α α α α s m,1
* *,0
* * *
* * *,0 ,0
* *
(4)
Here, “γ” is the ratio of volume of the core to the total spheroid,“Kα,0” is the spheroidal complex Clausius Mossotti factor and Lα is the“depolarizing factor” along the α axis:
=−
+ −=K
ε εε L ε ε
α x y( )
3[ ( ( ))],α
p s
s α s m,0
* *
* * * (5)
The depolarization factor along major axis (Lx) and minor axis (Ly),for a particle of eccentricity “e” are given by:
= ⎡⎣
⎛⎝
+−
⎞⎠
− ⎤⎦
L bca e
ee
e2
ln 11
2x 2 3 (6)
= −L L12y
x(7)
= −e ba
12
2 (8)
For Gram-positive C. difficile, the dielectric properties of the shell orcell envelope are dominated by its cell wall, due to its single-layermembrane. The core particle for C. difficile is formed by its cytoplasmicregion. Hence, we can use εw
* (to denote cell wall) in place of εs*; εcyto
* inplace of εp
*; and media properties: εmedia* for εm
* . The respective complexpermittivity values (ε*) are related to their dielectric permittivity (ε)and conductivity (σ) based on angular frequency (ω) and imaginaryunit (j): = +ε ε σ
jω* . In this manner, the permittivity and conductivity
of the cell wall and cytoplasmic regions were computed by solving Eq. 3and fitting the parameters in Eq. 4 using the geometrical informationabout cell axis and cell wall thickness, which were measured from TEMimages and using established values for cell envelope (σw and εw) andcytoplasm (σcyto and εcyto). The relative permittivity and conductivity ofmedium were set to 78.5 and 105mS/m, respectively.
4. Results and discussion
4.1. C. difficile functionality after vancomycin treatment at near-MIC levels
To confirm MIC levels, we plot measurements of the C. difficilegrowth rate (Fig. 1a) after vancomycin treatment (1 mg/L to 50mg/Llevels) and quantification of the secreted toxin (Fig. 1b) after vanco-mycin treatment (2 mg/L and 1mg/L levels), with untreated C. difficileserving as the positive control. We choose 2mg/L and 1mg/L vanco-mycin treatment levels, since these are close to the previously reportedMIC level [48]. Based on comparison to the control (untreated C. dif-ficile), it is clear that while the 2mg/L treated sample is completelydeactivated, the 1mg/L treated sample exhibits an active subpopula-tion, which becomes apparent only after 24 h of culture. Significantdifferences in growth between untreated, 1mg/L, and 2mg/L vanco-mycin treated C. difficile growth 24 h post-inoculation were confirmedby two-way ANOVA with Tukey’s multiple comparison post-test(p < 0.0001 in each case). Hence, we infer for the highly toxigenic C.difficile (HTCD) VPI 10463 strain, 2 mg/L (or 2 μg/mL) vancomycin is atthe MIC level and 1mg/L (1 μg/mL) vancomycin is at the sub-MIC level.Antibiotic concentrations in excess of 2mg/L did not demonstrate anyincrease in growth suppression. Subsequently, we seek to identify this
Fig. 1. (a) Growth of vancomycin-treated C. difficile with culture time; (b) toxin production of vancomycin-treated C. difficile measured as function of culture time.
A. Rohani et al. Sensors & Actuators: B. Chemical 276 (2018) 472–480
474
persistent active subpopulation at sub-MIC levels at the earlier timepoints and quantify its relative levels by following alterations in themicrobial cytoplasmic conductivity.
4.2. Identifying the heterogeneity by dielectrophoresis
The deactivation of C. difficile after vancomycin treatment canpossibly be identified based on the associated reduction in its cyto-plasmic conductivity, as measured by the substantial drop in polariz-ability of the microbial cell under an electric field. To validate thishypothesis, we carried out DEP analysis after varying levels of vanco-mycin treatment, as shown within Fig. 2. Using the device configurationof Fig. 2a, the microfluidic flow cell is used to place C. difficile cells justoutside the constriction region after which the field is turned ON forrecording the DEP velocities of single microbial particles towards oraway from the constriction tip in the 50 kHz to 1MHz region, as pernegative DEP (nDEP) in Fig. 2b or positive DEP (pDEP) in Fig. 2c. Therespective images of DEP transport after ∼30 s at 400 kHz shows uni-form levels of strong pDEP for untreated C. difficile (Fig. 2e), consistentwith strong field-induced polarization due to high cytoplasmic con-ductivity. On the other hand, the 2mg/L vancomycin treated C. difficilesample exhibits uniform levels of nDEP (Fig. 2f). We attribute the in-substantial field-induced polarization to the significantly reduced cy-toplasmic conductivity of the deactivated microbial. However, at in-termediate levels of 1mg/L vancomycin treated C. difficile (Fig. 2g), themajority population shows nDEP, while a minority subpopulationshows pDEP. Based on this, we infer that the DEP data agrees with thegrowth and toxin production data, to indicate persistence of a live C.difficile subpopulation after treatment at 1mg/L levels. The measuredfrequency spectra of the DEP velocities of Fig. 2d show distinct differ-ences for each treatment group, confirming the substantial drop in field-induced microbial polarizability, which is consistent with reduction incytoplasmic conductivity and reduction of cell aspect ratio (major tominor axis).
4.3. Quantifying the heterogeneity by electro-rotation
Electro-rotational (ROT) spectra enable sensitive determination ofthe DEP crossover frequencies, since shifts in the peak for counter-fieldrotation correspond to inflection from nDEP to pDEP behavior, whileshifts in the peak for co-field rotation correspond to inflection awayfrom strong pDEP behavior. In Fig. 3, we seek to establish the corre-lation of ROT spectral shifts to known phenotypic differences betweenrespective strains: highly toxigenic C. difficile (HTCD) VPI 10463 strainwith higher cell wall roughness and higher shape eccentricity versus thelow toxigenic C. difficile (LTCD) ATCC 630 strain and non-toxigenic C.difficile (NTCD) VPI 11186 strain [38]. The rotational rates at differenttime intervals (Fig. 3a-c) for computing the ROT spectra (Fig. 3d) wasacquired using the device set-up of Fig. 3e for the three C. difficilestrains, using homogenous samples. It is apparent from the ROT spectraof Fig. 3d that the peak for counter-field rotation, which corresponds toinflection from nDEP to pDEP behavior, is shifted towards higher fre-quencies from HTCD to LTCD to NTCD. Consistent with prior DEP work[38], we attribute this observation to the reduction in C. difficile cellwall capacitance due to lower surface roughness for HTCD (Fig. 3d)versus LTCD (Fig. 3e) and even lower roughness for NTCD (Fig. 3f)strains. It is also clear that the HTCD strain shows a higher frequencyvalue for the co-field peak and consistently higher rotational rates at allmeasured frequencies versus other strains, which is directly linked to itshigher polarizability that results from the higher eccentricity of itsprolate ellipsoidal shape [9]. The dielectric parameters determined byfitting the ROT spectra of Fig. 3 to a single-shell model (Table 1a), showa gradual reduction in εw from HTCD to the LTCD and NTCD strains thatis consistent with the increasing counter-field rotation peak and a re-duction in σcyto that scales with rpm level for co-field rotation. Theabove correlations allow us to establish the basis for utilizing ROTspectral shifts and ROT rates towards assessing vancomycin-inducedalterations to the HTCD strain, especially since its ROT spectra in Fig. 3exhibits the highest sensitivity to its phenotype versus other strains.Since ROT spectra measure the rotational rate of individual microbialcells over the entire frequency range, they are capable of discerning
Fig. 2. (a) Device configuration to measure DEP velocity under: (b) nDEP and (c) pDEP. (d) DEP velocity versus frequency (50 kHz – 1MHz) for single C. difficile cellsat indicated vancomycin treatment levels. Example images after 30 s of DEP at 400 kHz for: (e) Untreated; (f) 2mg/L treated and (g) 1mg/L treated C. difficile.
A. Rohani et al. Sensors & Actuators: B. Chemical 276 (2018) 472–480
475
heterogeneity in microbial phenotype and quantifying the proportion ofeach subpopulation exhibiting the heterogeneity. Comparing spectrafor untreated VPI 10463 C. difficile (HTCD) to that obtained aftertreatment with vancomycin at 2mg/L (so-called “viable” or “live”versus “non-viable” or “dead” based on growth curve in Fig. 1a), Fig. 4a
shows that the counter-field ROT peak is shifted to higher frequencies(consistent with upshifting of crossover to pDEP in Fig. 2d), while theco-field peak is shifted to lower frequencies (∼3.5MHz shift). Thelatter observation is consistent with the reduced field-induced polariz-ability of microbial cytoplasm that we inferred in Fig. 2d, based on
Fig. 3. Electro-rotational (ROT) spectra of: highly toxigenic C. difficile (HTCD) VPI 10463 strain, low toxigenic C. difficile (LTCD) ATCC 630 strain and non-toxigenicC. difficile (NTCD) VPI 11186 strain to quantify rotational rates at different time points (Fig. 3a-c) for computing the ROT spectra (Fig. 3d), using the device set-up(Fig. 3e). The TEMmicrographs show decreasing cell wall roughness levels for HTCD (Fig. 3f), LTCD (Fig. 3g) and NTCD (Fig. 3h) strains (Scale bar= 0.2 μm). All therotation rates are normalized with respect to the maximum rotation rate obtained for untreated C. difficile. The curves are based on single-shell model fits (Eqs.(1)–(8)) using parameters from Table 1a.
Table 1Fitted dielectric parameters (goodness of fit indicated by Root Mean Square Error or RMSE) based on the ROT spectra in Fig. 3 (Table 1a) and Fig. 4 (Table 1b) usingEqs. (1)–(8) to determine conductivity and permittivity of C. difficile cell envelope (σw and εw) and cytoplasm (σcyto and εcyto). The geometric parameters used in the fitinclude: Major Axis= 3.6 μm, Minor Axis= 0.57 μm, Cell Envelope Thickness (Wall+Membrane)= 35 nm.
A. Rohani et al. Sensors & Actuators: B. Chemical 276 (2018) 472–480
476
reduced pDEP levels after microbial deactivation. Fitting of the ROTspectra using a single-shell model (continuous red and black curves inFig. 4a), confirms a significant reduction in cytoplasmic conductivity(σcyto), from 0.46 S/m for untreated C. difficile (“live”) to 0.275 S/m for2mg/L treated C. difficile (“dead”), as per the fitted parameters sum-marized in Table 1b. Consequently, the lowered cytoplasmic con-ductivity causes inflection away from pDEP at earlier frequencies anddownshifting of the co-field ROT peak, which can serve as a sensitiveindicator of vancomycin-induced C. difficile disinfection. Examinationof the ROT spectra for the 1mg/L vancomycin-treated C. difficile cells inFig. 4b shows far higher standard deviations in rotation rates (bluediamond and green square data points considered together) versus thatobtained for 2mg/L treated C. difficile cells, especially for the co-fieldROT peak region (marked as region of interest). For instance, at10MHz, the 1mg/L vancomycin treated sample shows a combined rpmrange of 480–680 rpm versus the far lower range of 688–720 rpm forthe untreated sample and relatively lower range of 465–530 rpm after2mg/L vancomycin treatment (all ranges within one standard deviationof the mean). This heterogeneity is consistent with DEP responses forthe 1mg/L treated C. difficile sample, wherein one set of cells showedpDEP and another set showed nDEP at 0.4 MHz (Fig. 2g). Choosing10MHz as the particular frequency of interest for identifying andquantifying the subpopulations after 1mg/L vancomycin treatment,since it is in the 3.5MHz region between the co-field peak of the 2mg/L(“non-viable”) and untreated (“viable”), we plot the raw ROT data asthe number of cells at each significant increment of rpm range (Fig. 5a).Histograms were plotted using this data in Fig. 5b to determine: (i) theproportion of each subpopulation; and (ii) the mean and standard de-viation of rpm values (including at least 85% of measured cells) foreach subpopulation. The mean rpm values for each of these sub-populations from Fig. 5b are compared to mean rpm values obtainedfrom statistically relevant histograms for untreated and 2mg/L van-comycin treated C. difficile in Fig. 5c-5e. Based on 100 analyzed cells inFig. 5b, it is clear that we could fit a lower rpm histogram that is
denoted as subpopulation 1 (Sub 1), which is the majority subpopula-tion (∼65%) that resembles the rpm ranges observed for “non-viable”or “dead” cells (Fig. 5d); as well as a higher rpm histogram that isdenoted as subpopulation 2 (Sub 2), which is the minority subpopula-tion (∼65%) that resembles the rpm ranges observed for “viable” or“live” cells (Fig. 5e). A number of instances in prior work [37,41] haveshown a steady reduction in rotational rate upon cell de-activationunder heat or antibiotic treatment, since lowering of cytoplasmic con-ductivity can cause a gradual drop in polarizability and rotation rate.Histograms similar to Fig. 5b can be plotted at other frequencies in the“region of interest” (7.5–30MHz) to separate the subpopulations basedon similar shifts in their co-field rotation rates, with the blue diamondsymbols of Subpopulation 1 corresponding to the fitted ROT curve for2mg/L treated or “non-viable” C. difficile and the green square symbolsof Subpopulation 2 corresponding to the fitted ROT curve for untreatedor “live” C. difficile (Fig. 4b). Through fitting the ROT spectra of eachsubpopulation in Fig. 4b to the single shell elliopsoidal model (fittedparameters in Table 1b), we confirmed that the cytoplasmic con-ductivity of the majority subpopulation after 1mg/L vancomycintreatment (65% of cells corresponding to Sub 1 after 4 h of 1mg/Lvancomycin treatment) was fit to a value of 0.3 S/m that is close to therespective value of 0.275 S/m for deactivated C. difficile, whereas forthe minority subpopulation (35% of cells corresponding to Sub 2 after4 h of 1mg/L vancomycin treatment) the fit cytoplasmic conductivityof 0.42 S/m is close to the respective value of 0.45 S/m obtained forfunctional C. difficile (untreated sample). The mechanism for vanco-mycin action on microbials is attributed to inhibition of cell wallsynthesis and the associated leakage of cytosol upon cell wall inter-ruption. Our observations indicate that vancomycin treatment at levelsabove MIC (≥2mg/L) clearly upshifts the DEP crossover frequency(Fig. 2d) and the counter-field rotational peak (Fig. 4a), which indicatesreduction of dielectric permittivity of its cell wall (εw), as per Table 1b.However, upon sub-MIC level vancomycin treatment (1 mg/L), theshifts in DEP crossover frequency (Fig. 2g) and the counter-field
Fig. 4. (a) ROT shows up-shifting of counter-field peak (cor-responding to crossover frequency towards pDEP) and down-shifting of co-field peak (corresponding to crossover awayfrom pDEP) for 2mg/L vancomycin treated versus untreatedC. difficile, (b) 1mg/L vancomycin treated C. difficile showstwo distinct populations (see Fig. 5 and associated text foranalysis), with subpopulation 1 (blue diamond symbols) re-sembling the fit observed for 2mg/L vancomycin treated C.difficile (black curve) and subpopulation 2 (green squaresymbol) resembling the fit observed for untreated C. difficile(red curve). The curves are based on single-shell model fits(Eqs. (1)–(8)) using parameters from Table 1b. Rotation ratesin Fig. 4a-c were normalized with respect to the maximumrotation rate obtained for untreated C. difficile (For inter-pretation of the references to colour in this figure legend andtext, the reader is referred to the web version of this article).
A. Rohani et al. Sensors & Actuators: B. Chemical 276 (2018) 472–480
477
rotational peak (Fig. 4b) are no longer a reliable indicator of a reduc-tion in wall permittivity (εw). As per the counter-field region of Fig. 4b,while the red curve (untreated C. difficile) is significantly different fromblack curve (2mg/L treated C. difficile), there is no significant differ-ence in the measured rotation rates between subpopulation 1 (bluediamond symbols) and subpopulation 2 (green square symbols), whichhighlights that reduction in cell wall permittivity (εw) is not reliable. Onthe other hand, the shift in the co-field peaks are clearly apparent inFig. 4b, with the subpopulation 1 (blue diamond symbols) resemblingthe trend for 2mg/L treated C. difficile (black curve) and the sub-population 2 (green square symbols) resembling the trend for untreatedC. difficile (red curve). Since the electrorotation response in the peak co-field region is highly sensitive to cytoplasmic conductivity (σcyto), weinfer that these alterations are a more reliable indicator of modifica-tions to C. difficile versus the wall permittivity (εw). This set of DEPobservations identifying heterogeneity (Fig. 2g) and ROT resultsquantifying the proportion of each subpopulation (Fig. 5a & b), alongwith the growth curve (Fig. 1a) and toxin production (Fig. 1b) for 1mg/L treated C. difficile after 24 h, allows us to infer that ROT spectra canquantify the minority “live” subpopulation at sub-MIC levels after just afew hours of treatment, whereas the growth and toxin-based methodsrequire 24 h. It is noteworthy that since the antibiotic treatment of C.difficile was conducted in the culture media for 4 h to ensure completedeactivation, the measured proportion of ∼35% live cells after 1mg/Lvancomycin treatment includes the persistent live C. difficile sub-population, as well as additional microbial levels that may occur due toC. difficile growth in the microbial culture media for four hours.
5. Conclusions
For the purpose of enabling rapid assessment of antibiotic suscept-ibility on pathogenic microbials, we present an electrophysiology-basedsingle-cell technique for gauging phenotypic alterations of toxigenic C.difficile (VPI) after treatment with near MIC (minimum inhibitoryconcentration)-levels of vancomycin. Specifically, we demonstrate thatdisinfection of C. difficile after vancomycin treatment at MIC levels(2 mg/L), as validated by microbiological growth assays, can be as-certained rapidly based on reduction in microbial cytoplasmic con-ductivity, which reduces the level of positive dielectrophoresis (pDEP)and down-shifts the co-field electro-rotation (ROT) peak to lower fre-quencies. Furthermore, this single-cell electrophysiology technique canidentify and quantify the persistent live C. difficile subpopulation after4 h of vancomycin treatment at sub-MIC levels (1 mg/L), whereasmethods based on measurement of the secreted metabolite toxin or themicrobiological growth rate can identify this persistent C. difficile sub-population only after 24 h of microbial culture and are unable toquantify the subpopulation. The electro-phenotyping technique de-scribed here differs significantly from prior work by its use of single-microbial analysis to rapidly quantify the emerging subpopulationsafter MIC-level antibiotic treatments, with no need for microbial cultureto amplify their numbers. Microbiological assays or electro-pheno-typing methods that do not utilize single microbial analysis are notcapable of: (i) quantifying subpopulations as we observed at near-MIClevels, which likely get averaged out in methods that measure wholecell populations; and (ii) rapidly ascertaining antibiotic susceptibility atnear-MIC levels due to their need for ∼24 h of microbial culture that is
Fig. 5. (a) Rotation rates at 10MHz for C. dif-ficile (total of 100 cells) after 1mg/L vanco-mycin treatment suggest two subpopulationsbased on the histograms plotted in (b) for theirquantification: subpopulation 1 (Sub 1) of 65cells with lower rpm range (480–580 rpm for85% cells with estimated mean of 523 rpm)and subpopulation 2 (Sub 2) of 35 cells with ahigher rpm range (610–680 rpm for 85% cellswith estimated mean of 645 rpm). (c)Comparison of the respective rpm ranges for2 mg/L vancomycin treated C. difficile whereincomplete loss of viability was assessed (Fig. 1)vs. untreated C. difficile wherein no loss ofviability was assessed (Fig. 1), indicates sig-nificantly lower rpm for non-viable(465–530 rpm for 85% cells estimated mean of498 rpm) versus viable C. difficile(688–720 rpm for 85% cells with estimatedmean of 704 rpm), suggesting the resemblanceof Sub 1 after 1mg/L vancomycin treatment tonon-viable or dead C. difficile and Sub 2 after1 mg/L vancomycin treatment to viable or liveC. difficile. This is further illustrated in (d) thatshows similarity of rpm of the majority Sub 1to the 2mg/L vancomycin treated C. difficileand of the minority Sub 2 to the untreated C.difficile (e). The double sided arrows in (b) &(c) represent variation in the rotation rateswithin each sample (mean +/- 1 standard de-viation).
A. Rohani et al. Sensors & Actuators: B. Chemical 276 (2018) 472–480
478
necessary to amplify the microbial numbers to determine the growthcurve. Hence, this electro-phenotyping technique is best suited for si-tuations wherein there is a need at near-MIC levels, to rapidly ascertainantibiotic susceptibility and quantify any emerging subpopulations.Based on the reported sensitivity within short analysis times, we envi-sion that multiplexed versions of this technique can be applied for an-tibiotic susceptibility screening.
Funding
This work was supported by funds from the NIH Grant1R21AI130902-01 and from University of Virginia’s Global InfectiousDiseases Institute.
References
[1] J. Campbell, C. McBeth, M. Kalashnikov, A.K. Boardman, A. Sharon, A.F. Sauer-Budge, Microfluidic advances in phenotypic antibiotic susceptibility testing,Biomed. Microdevices 18 (2016) 103.
[2] Y. Lu, J. Gao, D.D. Zhang, V. Gau, J.C. Liao, P.K. Wong, Single cell antimicrobialsusceptibility testing by confined microchannels and electrokinetic loading, Anal.Chem. 85 (2013) 3971–3976.
[3] L.B. Reller, M. Weinstein, J.H. Jorgensen, M.J. Ferraro, Antimicrobial susceptibilitytesting: a review of general principles and contemporary practices, Clin. Infect. Dis.49 (2009) 1749–1755.
[4] P. Dalgaard, T. Ross, L. Kamperman, K. Neumeyer, T.A. McMeekin, Estimation ofbacterial growth rates from turbidimetric and viable count data, Int. J. FoodMicrobiol. 23 (1994) 391–404.
[5] J.A. DuVall, S.T. Cabaniss, M.L. Angotti, J.H. Moore, M. Abhyankar, N. Shukla,et al., Rapid detection of Clostridium difficile via magnetic bead aggregation in cost-effective polyester microdevices with cell phone image analysis, Analyst 141 (2016)5637–5645.
[6] P.D. Stamper, L. Louie, H. Wong, A.E. Simor, J.E. Farley, K.C. Carroll, Genotypicand phenotypic characterization of methicillin-susceptible Staphylococcus aureusisolates misidentified as methicillin-resistant Staphylococcus aureus by the BDGeneOhm MRSA assay, J. Clin. Microbiol. 49 (2011) 1240–1244.
[7] M. Tato, M. Morosini, L. García, S. Albertí, M. Coque, R. Cantón, Carbapenemheteroresistance in VIM-1-producing Klebsiella pneumoniae isolates belonging tothe same clone: consequences for routine susceptibility testing, J. Clin. Microbiol.48 (2010) 4089–4093.
[8] M.R. Pulido, M. García-Quintanilla, R. Martín-Peña, J.M. Cisneros, M.J. McConnell,Progress on the development of rapid methods for antimicrobial susceptibilitytesting, J. Antimicrob. Chemother. 68 (2013) 2710–2717.
[9] R.E. Fernandez, A. Rohani, V. Farmehini, N.S. Swami, Microbial analysis in die-lectrophoretic microfluidic systems, Anal. Chim. Acta (2017).
[10] J. Johari, Y. Hubner, J.C. Hull, J.W. Dale, M.P. Hughes, Dielectrophoretic assay ofbacterial resistance to antibiotics, Phys. Med. Biol. 48 (2003) N193–8.
[11] K.F. Hoettges, J.W. Dale, M.P. Hughes, Rapid determination of antibiotic resistancein E. coli using dielectrophoresis, Phys. Med. Biol. 52 (2007) 6001–6009.
[12] P.V. Jones, S. Huey, P. Davis, R. Yanashima, R. McLemore, A. McLaren, et al.,Biophysical separation of Staphylococcus epidermidis strains based on antibioticresistance, Analyst 140 (2015) 5152–5161.
[13] I. Peitz, R. van Leeuwen, Single-cell bacteria growth monitoring by automated DEP-facilitated image analysis, Lab Chip 10 (2010) 2944–2951.
[14] K. Zhu, A.S. Kaprelyants, E.G. Salina, G.H. Markx, Separation by dielectrophoresisof dormant and nondormant bacterial cells of Mycobacterium smegmatis,Biomicrofluidics 4 (2010).
[15] B.G. Hawkins, C. Huang, S. Arasanipalai, B.J. Kirby, Automated dielectrophoreticcharacterization of Mycobacterium smegmatis, Anal. Chem. 83 (2011) 3507–3515.
[16] D.L. Popham, K.D. Young, Role of penicillin-binding proteins in bacterial cellmorphogenesis, Curr. Opin. Microbiol. 6 (2003) 594–599.
[17] W. Margolin, Sculpting the bacterial cell, Curr. Biol. 19 (2009) R812–R822.[18] J. Choi, J. Yoo, M. Lee, E.-G. Kim, J.S. Lee, S. Lee, et al., A rapid antimicrobial
susceptibility test based on single-cell morphological analysis, Sci. Transl. Med. 6(2014) 267ra174-267ra174.
[19] K. Oh, J.H. Chung, J.J. Riley, Y. Liu, W.K. Liu, Fluid flow-assisted dielectrophoreticassembly of nanowires, Langmuir 23 (2007) 11932–11940.
[20] C.C. Chung, I.F. Cheng, W.H. Yang, H.C. Chang, Antibiotic susceptibility test basedon the dielectrophoretic behavior of elongated Escherichia coli with cephalexintreatment, Biomicrofluidics 5 (2011) 21102.
[21] C.C. Chung, I.F. Cheng, H.M. Chen, H.C. Kan, W.H. Yang, H.C. Chang, Screening ofantibiotic susceptibility to beta-lactam-induced elongation of gram-negative bac-teria based on dielectrophoresis, Anal. Chem. 84 (2012) 3347–3354.
[22] M. Elitas, R. Martinez-Duarte, N. Dhar, J.D. McKinney, P. Renaud,Dielectrophoresis-based purification of antibiotic-treated bacterial subpopulations,Lab Chip 14 (2014) 1850–1857.
[23] W.A. Braff, D. Willner, P. Hugenholtz, K. Rabaey, C.R. Buie, Dielectrophoresis-baseddiscrimination of bacteria at the strain level based on their surface properties, PLoSOne 8 (2013) e76751.
[24] C.-C. Chung, I.F. Cheng, W.-H. Yang, H.-C. Chang, Antibiotic susceptibility testbased on the dielectrophoretic behavior of elongated Escherichia coli with
cephalexin treatment, Biomicrofluidics 5 (2011) 021102.[25] O. Gefen, N.Q. Balaban, The importance of being persistent: heterogeneity of bac-
engineering approaches to eliminate them, Curr. Opin. Microbiol. 14 (2011)593–598.
[27] Y.-S. Fang, S.-Y. Chen, X.-J. Huang, L.-S. Wang, H.-Y. Wang, J.-F. Wang, Simpleapproach for ultrasensitive electrochemical immunoassay of Clostridium difficiletoxin B detection, Biosens. Bioelectron. 53 (2014) 238–244.
[28] K. Carroll, J. Bartlett, Biology of Clostridium difficile: implications for epidemiologyand diagnosis, Annu. Rev. Microbiol. 65 (2011) 501–521.
[29] G.C. Parkes, J.D. Sanderson, K. Whelan, The mechanisms and efficacy of probioticsin the prevention of Clostridium difficile-associated diarrhoea, Lancet Infect. Dis. 9(2009) 237–244.
[30] A.K. Sethi, W.N. Al-Nassir, M.M. Nerandzic, G.S. Bobulsky, C.J. Donskey,Persistence of skin contamination and environmental shedding of Clostridium dif-ficile during and after treatment of C. difficile infection, Infect. Control Hosp.Epidemiol. 31 (2010) 21–27.
[31] T. Peláez, E. Cercenado, L. Alcalá, M. Marin, A. Martín-López, J. Martínez-Alarcón,et al., Metronidazole resistance in Clostridium difficile is heterogeneous, J. Clin.Microbiol. 46 (2008) 3028–3032.
[32] J.-C. Baret, O.J. Miller, V. Taly, M. Ryckelynck, A. El-Harrak, L. Frenz, et al.,Fluorescence-activated droplet sorting (FADS): efficient microfluidic cell sortingbased on enzymatic activity, Lab Chip 9 (2009) 1850–1858.
[33] E. Rollo, E. Tenaglia, R. Genolet, E. Bianchi, A. Harari, G. Coukos, et al., Label-freeidentification of activated T lymphocytes through tridimensional microsensors onchip, Biosens. Bioelectron. 94 (2017) 193–199.
[34] N. Pal, S. Sharma, S. Gupta, Sensitive and rapid detection of pathogenic bacteria insmall volumes using impedance spectroscopy technique, Biosens. Bioelectron. 77(2016) 270–276.
[35] N. Mavrogiannis, F. Crivellari, Z.R. Gagnon, Label-free biomolecular detection atelectrically displaced liquid interfaces using interfacial electrokinetic transduction(IET), Biosens. Bioelectron. 77 (2016) 790–798.
[36] R. Hölzel, Non-invasive determination of bacterial single cell properties by elec-trorotation, Biochim. Biophys. Acta (BBA)-Mol. Cell Res. 1450 (1999) 53–60.
[37] C. Dalton, A. Goater, J. Burt, H. Smith, Analysis of parasites by electrorotation, J.Appl. Microbiol. 96 (2004) 24–32.
[38] Y.-H. Su, C.A. Warren, R.L. Guerrant, N.S. Swami, Dielectrophoretic monitoring andinterstrain separation of intact Clostridium difficile based on their S(Surface)-layers,Anal. Chem. 86 (2014) 10855–10863.
[39] Y.-H. Su, A. Rohani, C.A. Warren, N.S. Swami, Tracking Inhibitory alterationsduring interstrain clostridium difficile interactions by monitoring cell envelopecapacitance, ACS Infect. Dis. 2 (2016) 544–551.
[40] M.M. Merrigan, A. Venugopal, J.L. Roxas, F. Anwar, M.J. Mallozzi, B.A. Roxas,et al., Surface-layer protein A (SlpA) is a major contributor to host-cell adherence ofClostridium difficile, PLoS One 8 (2013) e78404.
[41] A. Rohani, W. Varhue, Y.H. Su, N.S. Swami, Electrical tweezer for highly paralle-lized electrorotation measurements over a wide frequency bandwidth,Electrophoresis 35 (2014) 1795–1802.
[42] A. Rohani, W. Varhue, Y.-H. Su, N.S. Swami, Quantifying spatio-temporal dynamicsof biomarker pre-concentration and depletion in microfluidic systems by intensitythreshold analysis, Biomicrofluidics 8 (2014) 052009.
[43] V. Farmehini, A. Rohani, Y.-H. Su, N.S. Swami, A wide-bandwidth power amplifierfor frequency-selective insulator-based dielectrophoresis, Lab Chip 14 (2014)4183–4187.
[44] Y.-H. Su, M. Tsegaye, W. Varhue, K.-T. Liao, L.S. Abebe, J.A. Smith, et al.,Quantitative dielectrophoretic tracking for characterization and separation of per-sistent subpopulations of Cryptosporidium parvum, Analyst 139 (2014) 66–73.
[45] T.B. Jones, T.B. Jones, Electromechanics of Particles, Cambridge University Press,2005.
[46] H. Morgan, N.G. Green, AC Electrokinetics, Research Studies Press, 2003.[47] R. Pethig, Dielectrophoresis: Status of the theory, technology, and applications,
Biomicrofluidics 4 (2010) 022811.[48] J.R. Garneau, L. Valiquette, L.C. Fortier, Prevention of Clostridium difficile spore
formation by sub-inhibitory concentrations of tigecycline and piperacillin/tazo-bactam, BMC Infect. Dis. 14 (2014) 29.
Ali Rohani recently completed his Ph.D. dissertation research in Electrical Engineering atthe University of Virginia focusing on signal analysis aspects of electrokinetic manip-ulation within microfluidic systems. His contributions include the development of anelectrical tweezer for enabling highly-parallelized electrorotational measurements andthe design of nanoslit devices for creating ion conductivity gradients to enhance thespatial extent of negative dielectrophoresis. He is currently a Research Associate in theBiomedical Data Sciences area.
John Moore is a final year Ph.D. student in Electrical Engineering at the University ofVirginia, specializing on AC electrokinetics for microbial analysis within microfluidicsystems. His contributions include development of co-culture and analysis platforms forstudying microbiota interactions on Clostridium difficile.
Yi-Hsuan Su completed her Ph.D. dissertation research in Electrical Engineering at theUniversity of Virginia, with a focus on microfluidics for microbiological analysis. Hercontributions include microfluidic platforms for measurement of dielectrophoretic forceson single cells and microbials. She is currently employed as a scientist with ZeptoLife, Inc.
A. Rohani et al. Sensors & Actuators: B. Chemical 276 (2018) 472–480
Victoria Stagnaro conducted the reported research as part of her undergraduate thesisand recently completed her undergraduate degree in Biomedical Engineering atUniversity of Virginia. She is currently employed as a scientist with Luna Innovations, Inc.
Cirle Warren is an Associate Professor of Infectious Diseases within the School ofMedicine at University of Virginia. Her lab specializes in developing clinical diagnosticsand therapeutics for antibiotic resistant infections.
Nathan Swami Nathan Swami is a Professor of Electrical & Computer Engineering at theUniversity of Virginia. His group specializes in label-free microfluidics for sorting, bioa-nalysis and cytometry of single cells and biological nanostructures, with a focus on im-pedance and deformability-based methods. He seeks to impact detection systems withinpoint-of-care and resource-poor settings for enabling precision medicine, personalizedhealthcare and environmental monitoring systems.
A. Rohani et al. Sensors & Actuators: B. Chemical 276 (2018) 472–480
