Evaporation-induced stimulation of bacterialosmoregulation for electrical assessment ofcell viabilityAida Ebrahimia,b and Muhammad Ashraful Alama,b,1

aSchool of Electrical and Computer Engineering, Purdue University, West Lafayette, IN 47907; and bBirck Nanotechnology Center, Purdue University, WestLafayette, IN 47907

Edited by David A. Weitz, Harvard University, Cambridge, MA, and approved May 9, 2016 (received for review April 15, 2016)

Bacteria cells use osmoregulatory proteins as emergency valves torespond to changes in the osmotic pressure of their external environ-ment. The existence of these emergency valves has been known sincethe 1960s, but they have never been used as the basis of a viabilityassay to tell dead bacteria cells apart from live ones. In this paper, weshow that osmoregulation provides a much faster, label-free assess-ment of cell viability compared with traditional approaches that relyon cell multiplication (growth) to reach a detectable threshold. Thecells are confined in an evaporating droplet that serves as a dynamicmicroenvironment. Evaporation-induced increase in ionic concentra-tion is reflected in a proportional increase of the droplet’s osmoticpressure, which in turn, stimulates the osmoregulatory response fromthe cells. By monitoring the time-varying electrical conductance ofevaporating droplets, bacterial cells are identified within a fewminutes compared with several hours in growth-based methods.To show the versatility of the proposed method, we show detectionof WT and genetically modified nonhalotolerant cells (Salmonellatyphimurium) and dead vs. live differentiation of nonhalotolerant(such as Escherichia coli DH5α) and halotolerant cells (such asStaphylococcus epidermidis). Unlike the growth-based techniques,the assay time of the proposed method is independent of cell concen-tration or the bacteria type. The proposed label-free approach pavesthe road toward realization of a new class of real time, array-formattedelectrical sensors compatible with droplet microfluidics for laboratoryon a chip applications.

bacteria | label free | osmoregulation | droplet | electrical microdevice

Low-cost, rapid viability analysis of bacterial cells in food, water,and/or clinical samples is critically important in a variety of

fields, such as bioscience research, medical diagnosis, and foodscreening, especially for low- to middle-income populations (1–4).Under a microscope, bacteria cells are amazingly alive and performa whole host of physiological functions, namely, multiplicationthrough cell division, searching for resources by chemotaxis (5–9),controlling water pressure by exchange of ions (through the os-moregulatory system) (10–12), etc. However, since the introductionof the plate counting method almost 130 y ago, traditional viabilityassays, such as impedance microbiology, rely on cell multiplica-tion to differentiate between dead and live cells (13–19). Celldivision time can vary from hours to weeks depending on the bacteriatype (e.g., 10–20 min for Escherichia coli vs. 15–16 h for Mycobacte-rium tuberculosis), which makes fast, real time detection of cellschallenging, especially at low concentration.Bacteria, the very first form of life, have been occupying Earth for

about 4 billion y, and survived through the harshest environments,including floods and droughts. There is a tight bond between lifeand water, with content that defines the osmotic pressure. Whenexposed to osmotic shocks, bacteria survive by regulating the osmoticpressure difference across their cell envelope. The pressure dif-ference (also known as turgor pressure) is defined by ΔΠ=Πcyt −Πout,where Πcyt and Πout are the cytoplasm and external osmotic pressures,respectively. The turgor pressure is regulated through activationof specific “emergency valves,”which rapidly modulate the concentration

of the solutes (including ionic species) in both the external andcytoplasmic solutions (10, 12, 20, 21).When ΔΠ increases above the natural turgor pressure (under

osmotic downshift), mechanosensitive (MS) channels open to releaseintracellular osmolytes to the surrounding within a fraction of a second(11). These proteins, which in the case of E. coli, are majorly MscLand MscS channels, pump out different osmolytes (including ions,ATP, lactose, etc.) into the surrounding medium without any damageto the cell envelope and/or cell lysis (10, 22, 23). In contrast, on os-motic upshift, another group of osmoregulatory transporters is acti-vated to restore the natural turgor pressure (e.g., by uptaking solutesfrom the surrounding) (12, 21, 24). The bacterial response to differentexternal osmotic pressures is summarized in SI Text, 1. Bacterial CellsUnder Osmotic Shock.Although bacterial osmoregulation has been known for over

half a century, it has not been used as the basis of a biosensingassay. Here, we propose a radically different concept: osmoregu-lation, which is equally universal as the cell multiplication but muchfaster, can be used as an effective, real time monitor of bacteriacells. Toward this goal, we confine the cells in a droplet placed onan impedance sensing unit. The droplet forms a tunable, preciselycontrolled microenvironment for bacteria cells. As the dropletevaporates, the reduction in volume forces the analytes toward thesensor surface. Evaporation-induced beating of the diffusion limitresults in much higher sensitivity and shorter response time of thisdroplet-based sensor compared with the classical impedance sensors(25). During evaporation, concentration of the droplet’s solutes in-creases, and so does the osmotic pressure. This dynamic microenvi-ronment stimulates the osmoregulatory system of live cells, resulting

Significance

Conventional bacterial viability assays rely on cell multiplicationuntil they are detectable by optical, electrical, or other sensors.Consequently, the assay time of classical growth-based tech-niques ranges from hours to weeks depending on bacteria type.In contrast, we present a fundamentally different paradigmbased on bacterial osmoregulation to identify viable cells inminutes. Our label-free platform relies on two key advances: (i)the fact that osmoregulation is as universal as cell division and(ii) the ability to create a microliter-sized environment on aspecially designed multifunctional structure. Our contributionadvances the field with a collage of ideas from diverse disciplines(e.g., biology, microfluidics, surface energetics, and impedancespectroscopy) and therefore, will attract a broad audience ofphysicists, material scientists, biologists, and engineers.

Author contributions: A.E. and M.A.A. designed research; A.E. performed research; A.E.and M.A.A. analyzed data; and A.E. and M.A.A. wrote the paper.

The authors declare no conflict of interest.

This article is a PNAS Direct Submission.1To whom correspondence should be addressed. Email: alam@purdue.edu.

This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10.1073/pnas.1606097113/-/DCSupplemental.

www.pnas.org/cgi/doi/10.1073/pnas.1606097113 PNAS | June 28, 2016 | vol. 113 | no. 26 | 7059–7064

APP

LIED

PHYS

ICAL

SCIENCE

S

Dow

nloa

ded

by g

uest

on

Sep

tem

ber

27, 2

020

in uptake (“stealing”) of osmolytes from the droplet and “hiding”them inside the cells. Therefore, although the droplet conductanceincreases with evaporation (because of the increased ion concen-tration), the presence of bacteria suppresses the net increase byshielding a fraction of ions away from the electric field. An ele-mentary theoretical model, to be discussed later, explains the resultsconsistently. In addition, as a reference, we analyzed another groupof cells that lost their osmoregulation ability due to heat treatmentand therefore, are “osmoregulatory dead” (defined as “dead”throughout this work). Overall, we find that the droplet-based im-pedance sensors (25–28) can selectively differentiate live vs. deadE. coli cells with 10–15% accuracy (comparable with the colonycounting method) in less than 20 min.Apart from differentiation of live and dead nonhalotolerant

cells, such as E. coli, our experiments confirm that the techniquecan successfully detect halotolerant cells, such as Staphylococcusepidermidis, as well (SI Text, 5. Detection of Halotolerant Cells:Staphylococcus epidermidis). These results confirm that the tech-nique is versatile and can be used for label-free electrical detectionof a broad range of bacteria. Moreover, our results on differentmutants of Salmonella typhimurium confirm that activation of K+

transporters (triggered by the continuous increase of osmoticpressure due to evaporation) is responsible for uptake of ions by thecells. These results demonstrate, for the first time, that activation ofosmoregulatory transporters can be probed electrically through alabel-free technique based on a microdevice (that can potentially beintegrated in a massively parallel array). Finally, apart from the useas a sensor, the droplet-based, label-free platform is well-suited (as acomplementary probe to traditional techniques) for answering abroad set of open biophysical questions, such as the hypothesisrelated to temperature-sensitive integrity of cell membrane (29).

MethodsFig. 1 illustrates the experimental protocol used in this study. Live samplesare prepared from overnight cultures. To eliminate the parasitic effects ofthe growth medium on the conductance signal, these cells are resuspendedin deionized (DI) water (Fig. 1A). As a reference, dead cells are prepared byheating an aliquot of live samples at 80 °C for 20 min. In this case, we in-tentionally did not lyse the cells but just damaged their protein channels byheat to impair cells’ ability to osmoregulate (29). Details of the samplepreparation are given in SI Text, 2. Sample Preparation. It should be notedthat resuspension of cells in DI water or the testing medium (TM) used in thiswork does not cause noticeable damage to cell integrity and viability, asconfirmed by optical imaging and colony counting results, respectivelyshown in Fig. S1 A and B, and also by their motion as observed in Movie S1.After resuspension of cells in the TM, droplets of 3-μL volume are depositedon a pair of specially designed superhydrophobic electrodes (that hold theshape of the droplets), and the droplet conductance is monitored as itevaporates from ti* to tf* (Fig. 1B). Fig. 1 C and D plot conductance signals as afunction of evaporation time, Gðt*Þ, for live and dead samples with con-centrations ranging from 104 to 107 cells per milliliter; t* is the time normal-ized to the total evaporation time. Clearly, live and dead samples have differentconductance values as explained in Discussion, Conductance Increases with CellConcentration. As we will discuss in Discussion, Repeated Sampling DuringEvaporation Significantly Improves the Accuracy, the sampled data will be usedto extract the unknown parameters [i.e., the total cell concentration (ρtot) andthe ratio of live cells (α) in a given sample].

Designof the superhydrophobic electrodes is critical for creatingdropletswith ahighly reproducible geometric shape, which is essential for high-precision mea-surement of impedance of evaporating droplets. As explained in detail elsewhere(25, 28), there is a universal behavior of the sensor signal with respect to the initialdroplet volume and total evaporation time. The method only relies on the nor-malized differential conductance (correlated to the ionic modulation), andtherefore, the experimental results are insensitive to evaporation rates. Moreover,the technique is simple and inexpensive: apart from the initial cost of fabrication,a sensor can be reused hundreds of times with no loss in performance.

Conductance Increases Because of Activation of MS Channels. As illustrated inFig. 2, when cells are initially resuspended in DI water at t0*, Πout =ΠDI ∼ 0. Asa result, they experience a significant turgor shock beyond the natural turgorpressure, ΔΠ0. To restore the natural turgor, the MS channels open up and

pump out the cytoplasmic osmolytes, including ionic entities. The gating process(opening and closing of the MS channels) and restoring of ΔΠ0 happens in lessthan a millisecond (t*MS) (10, 11, 22, 23). Release of ions from cytoplasm to theexternal solution results in a conductance increase with respect to the analyte-free solution, consistent with our results in SI Text, 3. Quantification of theConcentration of the Ions Released from Cells.

Cells “Steal” Ions from the Droplet as It Evaporates. When exposed to anincrease in Πout, during droplet evaporation from ti* to tf* (see Fig. 2), bacteriarespond by uptake of osmolytes either from the environment or by synthesis.The fastest response involves the uptake of K+ ions from the environmentthrough turgor-responsive transport systems, such as TrK (12, 21, 24). The up-take of ions from the droplet is reflected in an effective decrease in ioniccontribution of each cell (σ*; defined as per-cell conductivity) with time asschematically shown in Fig. 2B. Such stealing of ions from droplet reducessolution conductance over time compared with the scenario where cells areirresponsive to the modulation of osmotic pressure (dashed and dotted line inFig. 2B). In Discussion, Electrical Probing of Physiological Processes, we will usethe time-dependent conductance data to calculate the rate of ion stealing bycells. In addition, our preliminary studies on different mutants of S. typhimuriumshow that, depending on the osmoregulatory channels present, cells responddifferently to droplet evaporation.

DiscussionConductance Increases with Cell Concentration. Plotted in Fig. 3A arethe time-averaged signals ΔYl and ΔYd as a function of ρtot down to104 cells per milliliter for live and dead E. coli samples, respectively.We define ΔY ≜

Pkj=1Gðtpj Þ−Grefðtpj Þ=Grefðtpj Þ, where k is the total

sampling number during droplet evaporation (here, k= 9), andGrefis the conductance of cell-free droplets; ΔYl and ΔYd are calculated

Fig. 1. Graphical representation of the experimental steps. (A) At t0*, whichis when the cells are initially resuspended in DI water, the MS channels opento efflux the intracellular content (shown by arrows) and eventually, restoreΔΠ0. This process (opening and closing within t*MS) is very fast and happenswithin a fraction of a second. (B) After mixing the cell-containing solutionswith the TM, a droplet is placed on the sensor surface for conductancemeasurement. As time passes, the droplet size reduces, which causes ρout (thedroplet’s ion concentration) to increase. This evaporation-induced increasecauses ΔΠ to decrease. Decrease of ΔΠ activates the osmoregulatory trans-porters to uptake ions from the solution (indicated by one-sided arrows). Incase of dead cells with disintegrated proteins, the ions can diffuse freely inboth directions (indicated by double-sided arrows). (C and D) Droplet-basedconductance measurements on droplets containing live and dead cells. Thearrows indicate the increase of ρtot: 10

3, 104, 105, 106, and 107 cells per mil-liliter. With increasing the number of cells in a given volume, the number ofreleased ions increases proportionally. Dead cells with permeable membraneshow higher conductance increase than the live cells. These plots are aver-ages of at least three measurements.

7060 | www.pnas.org/cgi/doi/10.1073/pnas.1606097113 Ebrahimi and Alam

Dow

nloa

ded

by g

uest

on

Sep

tem

ber

27, 2

020

from the plots shown in Fig. 1 C and D, respectively. Fig. 3A showsthat (i) samples with dead cells generate larger electrical conductanceand (ii) conductance increases with cell concentration, ρtot. Theresults can be explained as follows.Live samples. It has been reported that ion (more precisely,osmolyte) release from bacterial cells in a hypotonic solution (whenΔΠ>ΔΠ0) explains the change of the solution impedance with cellconcentration. Yang (1) reported that impedance of the suspensionsof Salmonella in DI water decreased with increase of cell concen-tration (consistent with our observation in Fig. 3A). Although it wassuggested that the change of impedance is caused by the chargesassociated with cell wall and release of ions from cells, it was notquantitatively justified by Yang (1). In this regard, we calculated theeffective density of the species released to the solution throughthe correlation between osmotic pressure and concentration of theosmolytes. Our calculations are summarized in SI Text, a. Concen-tration of the released ions from live cells. We estimated that, aftersample preparation and reaching the equilibrium (from tpMS to tpi ), E.coli cells with ∼3 × 108 cells per milliliter produce an average con-ductance increase of Gcalc

l ∼ 3.42  μS. Remarkably, this simpleestimation is in excellent agreement with the measured valueGexp

l = 2.56  μS (indicated in Fig. S2).Dead samples. In heat-treated samples, the cell envelope becomespermeable, and there is no barrier for the solutes to diffuse across(30, 31). Therefore, the intracellular contents of cells, including ions(K+, Na+, and Mg2+), DNA, RNA, amino acids, and enzymes, leakto the surrounding environment. As a result, the solution conduc-tance increases significantly, more so than in the live samples. Thisincrease is proportional to the number of cells in a given volume. Byassuming that nearly all of the cytoplasmic content is released to thesurrounding solution on heat treatment, E. coli cells with ∼3 × 108

cells per milliliter produce a conductance increase of around

Gcalcd ∼ 6.6− 7.92  μS, which is consistent with the measured

value Gexpd = 6.34  μS (indicated in Fig. S2). Details of these

calculations are provided in SI Text, b. Concentration of thereleased species from dead cells.In practice, a sample under study may contain a mixture of live

and dead cells. Therefore, the ability to distinguish between them isof critical importance for practical applications. In the followingsection, we first provide a simple but comprehensive conductancemodel of a droplet containing a mixture of cells. Then, we validatethe model by the experimental data and show that the approach candetermine, with high precision, the fraction of live cells in amixture of dead and live ones.

Repeated Sampling During Evaporation Significantly Improves theAccuracy. In a given sample, the numbers of live and dead cellsare nl (ρl ≡ nl=V0) and nd (ρd ≡ nd=V0), respectively, with V0 beingthe initial droplet volume. The ratio of live cells to the totalnumber of cells is, hence,

α≡ρl

ρtot  ð=ρl+ρdÞ=

nlntot  ð=nl+ndÞ

,

where ρtot ≡ ntot=V0. By assuming that the total conductance,GmðtpÞ, is a linear combination of conductance of live and deadcells, we write Eq. 1 as

GmðtpÞ= ðσlðtpÞ+ σdðtpÞÞ Hz

gðtpÞ=

V0HzρtotV ðtpÞgðtpÞ

�ασpl ðρtot, tpÞ+ ð1− αÞσpdðρtot, tpÞ

�. [1]

Here, σpl ðtpÞ and σpdðtpÞ are the contributions of individual liveand dead cells, respectively, to the total conductivity, and Hzrepresents the time-invariant length of the deposited droplet;gðtpÞ is the geometry factor defined exclusively by the (time-)evolving shape of the droplet. Additional details on the droplet

Fig. 3. (A) ΔY is defined as the time-averaged relative conductance withrespect to the analyte-free (reference) solution at various total cell concen-trations. ΔYl and ΔYd are obtained for live and dead samples, respectively.These values are calculated from the time-resolved conductance datashown in Fig. 1 C and D, respectively. The error bars are the SDs fromsample-average with k = 9. Averaging is over nine data points obtainedduring evaporation. Because of the time-multiplexing capability of theapproach, the error bars are comparable to the symbol size. Inset shows thetwo measurements of a given sample: conductance measurement onthe as-prepared sample gives Gmðt*Þ and consequently, ΔYm

* . Additionally,a postheating step is carried out to deactivate all cells and pin α to zero. Thepostheating step results in the upshift of the signal (ΔYm

* to ΔYh* ). We es-

timate ρtot by intersecting ΔYh* and the ΔYd curve as schematically shown.

The estimated value is denoted by ρtot* . (B) ρtot* vs. ρtot is shown for five differentsamples, where ρtot* is calculated by using the algorithm described in A; ρtotand α of S#1–S#5 are ∼ 8.65× 104, ∼ 8.65× 104, ∼ 8.65× 104, ∼ 8.65× 105, and∼ 8.65×105 cells per milliliter, respectively, and 70%, 25%, 75%, 50%, and70%, respectively.

Fig. 2. (A) Time variation of the turgor pressure (ΔΠ) as cells go through thesteps shown in Fig. 1. Insets show a droplet as it evaporates from ti* to tf*.(B) The solid curve represents the per-cell conductivity (σ * ) as cells go throughthe steps shown in Fig. 1. Had the cells been like impenetrable dielectric struc-tures, one would have expected the dotted curve (almost constant). This dif-ference is because the cells are constantly trying to adjust their turgor pressure,and therefore, a time-variable per-cell conductivity is observed. Decrease ofconductivity is equivalent to cells reducing their ion release with time. All of thecurves are merely graphical representations of the trend of the parameters.

Ebrahimi and Alam PNAS | June 28, 2016 | vol. 113 | no. 26 | 7061

APP

LIED

PHYS

ICAL

SCIENCE

S

Dow

nloa

ded

by g

uest

on

Sep

tem

ber

27, 2

020

circuit model and conductance formulation can be found else-where (26).To determine the viability ratio (α) of a sample, we conducted a

pair of impedance measurements as follows (Fig. 3A, Inset): (i) onthe sample as prepared,GmðtpÞ, and (ii) after heating the sample at80 °C for 20 min, GhðtpÞ. The heating step kills all of the cells andpins α to zero (independently verified by the colony countingmethod). Therefore, GhðtpÞ is given by setting α= 0 in Eq. 1:

GhðtpÞ= V0Hzρ*totV ðtpÞgðtpÞ

�σpd�ρ*tot, t

p��. [2]

To estimate ρtot in an unknown sample, we need to predeterminetwo calibration curves for ΔY ðρtot, αÞ [e.g., live (α≜ 1) and dead(α≜ 0)]. Fig. 3A shows the results for E. coli samples. After theunknown sample is heated and bacterial cells are heat-killed, thetime-averaged signal shifts up from ΔY p

m to ΔY ph , such as in Fig.

3A. The intersection of ΔY ph [average of GhðtpÞ over tp] with ΔYd

gives the estimated total cell concentration, denoted by ρtot* . Fig.3B shows ρtot* vs. the prepared cell concentration ρtot (indepen-dently determined by standard plate counting) for five differentE. coli samples: S#1–S#5. The correlation between the esti-mated and actual values is strong, and ρtot* differs from ρtot by lessthan a factor of two. This excellent reproducibility is attributed to(i) the time-multiplexing aspect of droplet-based sensing and (ii)sampling a small volume (microliter droplets).By inserting ρtot* into Eq. 1, we obtain

αpðtpÞ=

h1−GmðtpÞ

GhðtpÞi

h1− GlðtpÞ

GdðtpÞi . [3]

Fig. 4A plots αpðtpÞ for the same samples discussed in Fig. 3B. Wedefine the estimated α as αestim ≜ hαpitp. Fig. 4B shows αestim vs. α (theratio when live and dead cells were mixed manually) for these sam-ples. By averaging the data points obtained in a single run (ninepoints in this report), the noise in the data is reduced dramatically.These plots confirm that different responses of osmoregulatory liveand dead cells to the dynamic microenvironment enable their iden-tification with a very high precision (10–15%) down to α∼ 20%(comparable with the colony counting method as shown in Fig. S3).In addition to E. coli (nonhalotolerant), we have shown the versatilityof our technique by analyzing halotolerant cells, such as Staphylococ-cus. Our results for S. epidermidis cells (SI Text, 5. Detection ofHalotolerant Cells: Staphylococcus epidermidis) confirm that theproposed osmoregulation-based method successfully differentiates

between live and dead S. epidermidis cells (Fig. S4A), and also,the electrical conductance data confirms that S. epidermidis hashigher turgor pressure compared with E. coli cells (Fig. S4B).Finally, it is important to realize that there are several excellent

techniques for detection of bacterial viability. These techniques canbe broadly divided into growth-based vs. nongrowth-based assays.Growth-based techniques [such as colony counting (14), impedancemicrobiology (16–19), monitoring the resonance frequency (32), etc.]rely on cell division to reach a threshold level so that the signalbecomes detectable. The typical incubation time can vary from hoursto days depending on the cell type and initial cell concentration. Ourapproach provides accuracy and dynamic range comparable with thecolony counting method (Fig. S3) or impedance microbiology, but isseveral orders of magnitude faster. Moreover, unlike growth-basedmethods, detection time of our assay is independent of the initial celldensity and cell type.The second category (i.e., nongrowth-based methods) includes (i)

dye-based staining [dynamic range of ∼104–108 cells per milliliterand detection time of approximately minutes (15, 33)], (ii) molec-ular-based methods [involving antibody-based ELISA or fluorescentreadout of amplified DNA ormRNAwith dynamic range of∼104–108cells per milliliter and detection time of approximately minutes to1 h (34–36)], (iii) dielectrophoresis (DEP)-based differentiation[dynamic range of ∼105–106 cells per milliliter and detection timeof approximately minutes (3, 17, 37)], and (iv) light-addressablepotentiometric sensors [dynamic range of ∼104–105 cells per mil-liliter and detection time of approximately minutes (38)]. Com-pared with these assays, our technique offers similar dynamic range(∼105–108 cells per milliliter) and detection time (approximatelyminutes). However, compared with the classical techniques, thenew approach has two advantages. First, this label-free techniquereduces the time for sample preparation and cost/complexity ofattaching labels to analyte molecules. Second, its small size andcompatibility with standard microfabrication techniques suggestopportunities of massive parallelization (39).

Electrical Probing of Physiological Processes.Cells steal ions in response to their dynamic microenvironment. To examinethe hypothesis explained schematically in Fig. 2B (i.e., decrease ofcells’ conductivity over time as droplets shrink), we extracted per-cell conductivities (σpl with α= 1 and σpd with α= 0) from Eq. 1. Fig.5A plots the initial and final conductivity values for each case (σpl,i,σpl,f , σ

pd,i, and σpd,f). These plots convey three important observations:

(i) at all cell concentrations, fewer ions are released from live cells

Fig. 5. (A) The extracted per-cell conductivities (σ*). Using Eq. 1, σl,i=f* andσ*d,i=f values are extracted from Fig. 1 C and D, respectively. The lines justserve as a guide to the eye. This plot suggests that (i) fewer ions are releasedfrom live cells compared with the dead ones, (ii) the number of ejected ionsfrom each cell reduces with time, and (iii) σ* decreases with ρtot. (B) Per-cellconductivities at ti* and tf* for two samples with ρtot ∼ 107 cells per milliliter intwo electrolytes with different salinities: TM × 10 and TM, which is 10 times morediluted than TM × 10; σ*TM×10

l,i=f < σ*TMl,i=f because of the initially largerΠout in solutionwith TM × 10 than the one in TM (green arrow), and σi* >σf* as explained in thetext and shown by the process in ii in A.

Fig. 4. (A) The estimated α (calculated by Eq. 3) as a function of the nor-malized evaporation time, t*, for the same samples discussed in Fig. 3.(B) Average of the estimated α-values over t*. Similar to Fig. 3A, the error bars arethe SDs from themeanwith n = 9 (evaporation time points); ρtot and α of S#1–S#5are ∼ 8.65×104, ∼ 8.65× 104, ∼ 8.65× 104, ∼ 8.65×105, and ∼ 8.65× 105 cells permilliliter, respectively, and 70%, 25%, 75%, 50%, and 70%, respectively.

7062 | www.pnas.org/cgi/doi/10.1073/pnas.1606097113 Ebrahimi and Alam

Dow

nloa

ded

by g

uest

on

Sep

tem

ber

27, 2

020

compared with dead ones (σpl,i=f < σpd,i=f), (ii) the number of ejectedions per cell, σpl=dðtpÞ, decreases with time (σpl=d,f < σpl=d,i), and (iii) σpdecreases with ρtot. These observations can be explained as follows.(i) Because heat-treated cells have a permeable cell envelope,

the number of ions released from individual heat-treated cells ishigher than that of the live ones at all times; therefore, σpl,i=f < σpd,i=f .(ii) As the droplet evaporates, its ionic concentration (ρout)

increases. To explain the decrease of σp over time in both liveand dead samples, we discuss the two cases separately.In the case of osmoregulatory live cells, increase of ρout causes the

turgor pressure across the cell envelope (ΔΠ) to decrease below thenatural pressure (ΔΠ0) according to Eqs. S1 and S2. As a result,the solution becomes “hypertonic” to cells, causing the osmoregula-tory transporters to activate and uptake ions from the environment.The ion uptake is effectively equivalent to cells decreasing theirion release to the surrounding, and therefore, σpl,f < σpl,i.To confirm this observation, we have performed an experi-

ment with α= 1 and ρtot = 107 cells per milliliter resuspended in adifferent reference solution (TM × 10) with 10 times higher ionicconcentration than the one that we used so far (TM). Fig. 5B plotsthe extracted σpl,i and σpl,f . This plot shows that σpTM×10

l,i=f < σpTMl,i=f ,suggesting that ΔΠTM×10 <ΔΠTMð<ΔΠ0Þ. This observation con-firms our previous statement that, when cells are suspended in asolution with higher ρout (larger Πout), they experience larger de-creases of the turgor pressure and need to steal more solutes fromthe environment to restore ΔΠ0.In the case of osmoregulatory dead cells, the decrease of σpd

with time can be justified by the dielectric behavior of cells at lowfrequencies (37). On increase of ρout during evaporation, ions aresqueezed into the cells, so that they become invisible to the electricfield; therefore, the overall effect is as if the number of the existingions for conduction has decreased.(iii) With the increase of ρtot, the external ionic concentration

seen by each individual cell increases. Parallel to the discussionin ii, cells reduce their ion release because of effective increase ofthe osmotic pressure of their environment, Πout.Droplet-based impedance sensing reveals activation of K+ transporters.To validate that activation of K+ osmoregulatory transporters is themain reason for uptake of ions as droplets evaporate, we studied fourdifferent strains of S. typhimurium, WT, TrkA−, Kdp−, and thedouble mutant, TrkA−/Kdp−. Cells with ρtot ∼ 107 cells per milliliterwere resuspended in 1 μM KCl. The time-varying conductance re-sults of cells are plotted in Fig. 6A. As shown, different strains showdifferent responses to the continuous increase of K+ concentrationbecause of droplet evaporation; more specifically, the TrkA− strainsshow the smallest rate of conductance increase. From these data, we

extracted the ratio of the final per-cell conductivity to the initialvalue, σpf =σ

pi , as plotted in Fig. 6B. The double-mutant samples

lack both the primary K+ responders and hence, show the highestconductance (less K+ stealing from the external droplet). TrkA−mutant, which only has the Kdp transporter (the most selective K+

channel), steals the largest amount of K+ from the solution (i.e.,has the smallest conductance). WT and Kdp− cells show almostsimilar responses, which is because both strains have the TrkAtransporter, which is the first and main responder to the changesof K+ concentration (20). These results further confirm that evap-oration-induced modulation of osmotic pressure because of in-crease of K+ concentration is the main mechanism underlying thetime-dependent uptake of ions by cells.

ConclusionWe have proposed a fundamentally different route for label-freedetermination of bacterial viability based on stimulating their os-moregulatory response in an artificially created microenvironment.The approach is based on the idea that confining cells in a dynamicenvironment with varying osmotic pressure would trigger their os-moregulatory response. Activation of the osmoregulatory emergencyvalves results in exchange of solutes with the external environment,which in turn, is reflected in modulation of the solution conductance.Contrary to the conventional growth-based assays, the proposedmethod has a much shorter assay time. The techniques are notmutually exclusive: the sensitivity/selectivity of the current approachcan be further improved with a pregrowth step (as an initial selectivepreamplification) that precedes the droplet-based assay. Finally, al-though a systematic statistical analysis is required for its adoption as aclinical technique, the proposed label-free miniaturized assay couldfind broader application as a fast, nondestructive, electrical charac-terization technique of various biophysiological processes in micro-organisms, such as in studying different classes of proteins responsiblefor ionic exchange of cells with their extracellular environment.

Materials and Sample PreparationPreparation of Bacterial Samples. E. coli DH5α, S. epidermidis cells, andS. typhimuriummutants (provided as a gift from Laszlo Csonka at Department ofBiological Sciences, Purdue University, West Lafayette, IN) were collected from afrozen stockmaintained at−80 °C and cultured in Lysogeny broth nutrient broth.Bacterial suspensions were prepared by following the procedure described in SIText, 2. Sample Preparation. Briefly, after overnight incubation in Lysogeny brothmedium prepared in house, density of the grown cells was determined throughOD reading. The cells were then washed and resuspended in DI water to removethe residual components from the growth medium. Heat-treated cells were thenobtained by heating an aliquot of live cells for 20 min at 80 °C. By serial dilution,different samples with varying cell concentrations were prepared for the im-pedance sensing. For all of the tests presented here, the bacteria were incubatedoff chip in a macroscale volume before deposition on the sensing electrode arrayfor impedance measurement. The reference medium in our experiments (TM),the medium mixed with cells before the impedance measurements, is composedof 1 μg/mL tryptone, 1 μg/mL yeast extract, and 0.1 μg/mL NaCl.

Device Structure and Testing Apparatus. Fabrication procedure of super-hydrophobic nickelmicroelectrodes is explained in detail in our previous papers(25, 26, 28). The biosensor has a two-electrode configuration, with eachelectrode being an array of electroplated nickel fins (microscale structures)with width of a= 10  μm, height of H= 10  μm, spacing of b= 20  μm, and in-plane length of Hz = 4 mm patterned and deposited on a glass substrate. Foreach impedance measurement, we took out a 3-μL droplet of the sample usinga micropipette and deposited it on the electrode array. The terminal pads areelectrically connected to an HP-A4280 LCR Meter, where the impedance ismeasured by application of a 50-mV ac signal at frequencies ranging from 120 Hzto 1 MHz. The results shown in this work are measured at f = 600 Hz.

ACKNOWLEDGMENTS. The authors thank Prof. Laszlo Csonka’s generosityto provide the Salmonella mutants, and also the helpful discussions. Inaddition, the authors acknowledge Dr. Lisa Reece for the microbiologytraining and the helpful discussions. This work was supported by the Na-tional Science Foundation Project (high density array biosensors for spatialmapping of cellular gradients and flux; Award 1403582) and Bilsland Dis-sertation Fellowship Award (Purdue University).

Fig. 6. Droplet-based impedance sensing reveals that various mutants ofS. typhimurium show different electrical responses. (A) Different cells show dif-ferent responses to the changes of themicroenvironment (evaporating droplets).(B) The ratio of the final to the initial per-cell conductivities (σf* =σi* ) for the cellsshown in A. The double mutants (Kdp−/TrkA−) show the highest conductancebecause of lack of the necessary gates for K+ transport. Compared with all cases,TrkA−mutant steals themost amount of K+ from the droplet.WT and Kdp− cellsshow similar responses, because both have the TrkA transporters. Cells areresuspended in 1 μM KCl solution, with ρtot ∼ 8× 106 cells per milliliter.

Ebrahimi and Alam PNAS | June 28, 2016 | vol. 113 | no. 26 | 7063

APP

LIED

PHYS

ICAL

SCIENCE

S

Dow

nloa

ded

by g

uest

on

Sep

tem

ber

27, 2

020

1. Yang L (2008) Electrical impedance spectroscopy for detection of bacterial cells insuspensions using interdigitated microelectrodes. Talanta 74(5):1621–1629.

2. Sharma H, Mutharasan R (2013) Review of biosensors for foodborne pathogens andtoxins. Sens Actuators B Chem 183:535–549.

3. Suehiro J, Hamada R, Noutomi D, Shutou M, Hara M (2003) Selective detection of viablebacteria using dielectrophoretic impedance measurement method. J Electrostat 57(2):157–168.

4. Zetsche E-M, Meysman FJR (2012) Dead or alive? Viability assessment of micro- andmesoplankton. J Plankton Res 34(6):493–509.

5. Sourjik V, Berg HC (2004) Functional interactions between receptors in bacterialchemotaxis. Nature 428(6981):437–441.

6. Wadhams GH, Armitage JP (2004) Making sense of it all: Bacterial chemotaxis. Nat RevMol Cell Biol 5(12):1024–1037.

7. Zhang P, Khursigara CM, Hartnell LM, Subramaniam S (2007) Direct visualization ofEscherichia coli chemotaxis receptor arrays using cryo-electron microscopy. Proc NatlAcad Sci USA 104(10):3777–3781.

8. Tu Y, Shimizu TS, Berg HC (2008) Modeling the chemotactic response of Escherichiacoli to time-varying stimuli. Proc Natl Acad Sci USA 105(39):14855–14860.

9. Briegel A, et al. (2009) Universal architecture of bacterial chemoreceptor arrays. ProcNatl Acad Sci USA 106(40):17181–17186.

10. Kung C (2005) A possible unifying principle for mechanosensation. Nature 436(7051):647–654.

11. Louhivuori M, Risselada HJ, van der Giessen E, Marrink SJ (2010) Release of contentthrough mechano-sensitive gates in pressurized liposomes. Proc Natl Acad Sci USA107(46):19856–19860.

12. Balaji B, O’Connor K, Lucas JR, Anderson JM, Csonka LN (2005) Timing of induction of os-motically controlled genes in Salmonella enterica Serovar Typhimurium, determined withquantitative real-time reverse transcription-PCR. Appl Environ Microbiol 71(12):8273–8283.

13. World Health Organization (2003) Heterotrophic Plate Counts and Drinking-water Safety,eds Bartram J, Cotruvo J, Exner M, Fricker C, Glasmacher A (IWA, London), 1st Ed.

14. Kogure K, Simidu U, Taga N (1979) A tentative direct microscopic method forcounting living marine bacteria. Can J Microbiol 25(3):415–420.

15. Bernard L, et al. (2001) A new approach to determine the genetic diversity of viableand active bacteria in aquatic ecosystems. Cytometry 43(4):314–321.

16. Gómez R, Bashir R, Bhunia AK (2002) Microscale electronic detection of bacterialmetabolism. Sens Actuators B Chem 86(2-3):198–208.

17. Gómez-sjöberg R, Morisette DT, Bashir R, Member S (2005) Impedance microbiology-on-a-chip: Microfluidic bioprocessor for rapid detection of bacterial metabolism.J Microelectromech Syst 14(4):829–838.

18. Varshney M, Li Y (2008) Double interdigitated array microelectrode-based impedancebiosensor for detection of viable Escherichia coli O157:H7 in growth medium. Talanta74(4):518–525.

19. Grossi M, et al. (2010) An embedded portable biosensor system for bacterial con-centration detection. Biosens Bioelectron 26(3):983–990.

20. Sleator RD, Hill C (2002) Bacterial osmoadaptation: The role of osmolytes in bacterialstress and virulence. FEMS Microbiol Rev 26(1):49–71.

21. Kempf B, Bremer E (1998) Uptake and synthesis of compatible solutes as microbialstress responses to high-osmolality environments. Arch Microbiol 170(5):319–330.

22. Reuter M, et al. (2013) Mechanosensitive channels and bacterial cell wall integrity:Does life end with a bang or a whimper? J R Soc Interface 11(91):20130850.

23. Levina N, et al. (1999) Protection of Escherichia coli cells against extreme turgor byactivation of MscS and MscL mechanosensitive channels: Identification of genes re-quired for MscS activity. EMBO J 18(7):1730–1737.

24. Dinnbier U, Limpinsel E, Schmid R, Bakker EP (1988) Transient accumulation of po-tassium glutamate and its replacement by trehalose during adaptation of growingcells of Escherichia coli K-12 to elevated sodium chloride concentrations. ArchMicrobiol 150(4):348–357.

25. Ebrahimi A, et al. (2013) Nanotextured superhydrophobic electrodes enable detectionof attomolar-scale DNA concentration within a droplet by non-faradaic impedancespectroscopy. Lab Chip 13(21):4248–4256.

26. Dak P, Ebrahimi A, Alam MA (2014) Non-faradaic impedance characterization of anevaporating droplet for microfluidic and biosensing applications. Lab Chip 14(14):2469–2479.

27. Ebrahimi A, Alam MA (2013) Evaporation-enhanced impedance sensing for highly-sensitive differentiation of dsDNA from ssDNA. Proceedings of the 71st Device Re-search Conference (IEEE, New York), pp 159–160.

28. Ebrahimi A, Alam MA (2015) Time-resolved PCA of “droplet impedance” identifiesDNA hybridization at nM concentration. Sens Actuators B Chem 215:215–224.

29. O’Connor K, Fletcher SA, Csonka LN (2009) Increased expression of Mg(2+) transportproteins enhances the survival of Salmonella enterica at high temperature. Proc NatlAcad Sci USA 106(41):17522–17527.

30. Lapizco-Encinas BH, Simmons BA, Cummings EB, Fintschenko Y (2004) Dielectropho-retic concentration and separation of live and dead bacteria in an array of insulators.Anal Chem 76(6):1571–1579.

31. Cheng X, et al. (2007) Cell detection and counting through cell lysate impedancespectroscopy in microfluidic devices. Lab Chip 7(6):746–755.

32. Huang S, et al. (2008) A wireless, remote-query sensor for real-time detection of Es-cherichia coli O157:H7 concentrations. Sens Actuators B Chem 131(2):489–495.

33. Berney M, Hammes F, Bosshard F, Weilenmann H-U, Egli T (2007) Assessment andinterpretation of bacterial viability by using the LIVE/DEAD BacLight Kit in combi-nation with flow cytometry. Appl Environ Microbiol 73(10):3283–3290.

34. Keer JT, Birch L (2003) Molecular methods for the assessment of bacterial viability.J Microbiol Methods 53(2):175–183.

35. Mannoor MS, Zhang S, Link AJ, McAlpine MC (2010) Electrical detection of patho-genic bacteria via immobilized antimicrobial peptides. Proc Natl Acad Sci USA 107(45):19207–19212.

36. Perelle S, et al. (2004) Comparison of PCR-ELISA and LightCycler real-time PCR assays fordetecting Salmonella spp. in milk and meat samples. Mol Cell Probes 18(6):409–420.

37. Li H, Bashir R (2002) Dielectrophoretic separation and manipulation of live and heat-treated cells of Listeria on microfabricated devices with interdigitated electrodes.Sens Actuators B Chem 86(2-3):215–221.

38. Gehring AG, Patterson DL, Tu SI (1998) Use of a light-addressable potentiometricsensor for the detection of Escherichia coli O157:H7. Anal Biochem 258(2):293–298.

39. Salm E, et al. (2013) Ultralocalized thermal reactions in subnanoliter droplets-in-air.Proc Natl Acad Sci USA 110(9):3310–3315.

40. Raicu V, Popescu A (2008) Integrated Molecular and Cellular Biophysics (Springer,Berlin), 1st Ed.

41. Boulbitch A, Quinn B, Pink D (2000) Elasticity of the rod-shaped gram-negative eu-bacteria. Phys Rev Lett 85(24):5246–5249.

42. Berrier C, Coulombe A, Szabo I, Zoratti M, Ghazi A (1992) Gadolinium ion inhibits lossof metabolites induced by osmotic shock and large stretch-activated channels inbacteria. Eur J Biochem 206(2):559–565.

43. Ausubel FM, et al. (2002) Short Protocols in Molecular Biology (Wiley, New York), 5th Ed.44. Suehiro J, Hatano T, Shutou M, Hara M (2005) Improvement of electric pulse shape for

electropermeabilization-assisted dielectrophoretic impedance measurement for highsensitive bacteria detection. Sens Actuators B Chem 109(2):209–215.

45. Banada PP, Liu Y-S, Yang L, Bashir R, Bhunia AK (2006) Performance evaluation of a lowconductive growth medium (LCGM) for growth of healthy and stressed Listeria mono-cytogenes and other common bacterial species. Int J Food Microbiol 111(1):12–20.

46. Ohno H, ed (2005) Electrochemical Aspects of Ionic Liquids (Wiley-Interscience,Hoboken, NJ), 2nd Ed.

7064 | www.pnas.org/cgi/doi/10.1073/pnas.1606097113 Ebrahimi and Alam

Dow

nloa

ded

by g

uest

on

Sep

tem

ber

27, 2

020