Monitoring of Lactococcus Lactis Growth Based on Reduced-Graphene Oxide TFT for Dairy Industry Applications

Thin-film transistors (TFTs) are a cutting-edge technology for biosensing applications due to their fast response time, low signal-to-noise ratio, and straightforward miniaturization into electronic circuitry. Liquid-gated reduced graphene oxide TFTs (rGO-TFTs) are particularly promising as biosensors since they allow direct coupling between the active material and any (bio-) chemical species dissolved in aqueous solution. In this work, we explore the use of ambipolar rGO-TFTs as transducers for biosensing, focusing on the detection of bacteria proliferation in laboratory-made samples. We choose Lactococcus lactis (L. lactis) as a reference bacterial species due to the potential final applications for the dairy industry. We performed electrical characterization of two different rGO-TFTs based on Si substrate (Si-TFT) and PET substrate (PET-TFT), assessing the devices’ stability, and extrapolating the parameters through a suitable model. The characterization is carried out using M17 broth as the liquid gate. The L. lactis detection is based on the evident parametric shifts of threshold potential caused by the bacterial presence at the liquid gate. The experimental results are compared with optical absorbance measurements at 600 nm to validate the proposed method. Experimental tests with laboratory-made L. lactis samples confirm that PET-TFTs have a higher detection capability compared to Si-TFTs, being also suitable to monitor the bacteria proliferation during observation of 370 min. rGO-TFTs show evident shifts of threshold potential of more than −100 mV in L. lactis spiked M17 broth compared to the pristine one.


Monitoring of Lactococcus Lactis Growth Based
on Reduced-Graphene Oxide TFT for Dairy Industry Applications

I. INTRODUCTION
I N THE recent years, biosensors have attracted consid- erable research efforts since they allow rapid, accurate, and cheap detection for a wide range of applications [1], allowing the development of easy ready-to-use sensing devices suitable for nonspecialized users, in contrast to the bulky, complex, and costly high-throughput machines of microbiology laboratories [2].Cheap and cost-effective biosensors are of great interest, especially for clinical diagnostics [3], [4], [5], e.g., biomarker detection [6], and for agri-food quality analysis [7], [8], [9], e.g., phage detection in the dairy industry [10].
In recent years, field-effect-transistor (FET) biosensors have become a rising technology due to their outstanding performances, e.g., fast measurements, low signal-to-noise ratio, and the possibility to implement the device into compact and portable instrumentations [11], [12], [13], [14], [15].In [13], a liquid-gated thin-film transistor (TFT) based on carbon nanotubes (CNTs) semiconductor was coupled with oligonucleotide aptamers to efficiently detect the presence of blood biomarkers of Alzheimer's disease.The biosensor reported high selectivity and a large dynamic range, with rapid response time and low variation.In [14], an FET biosensor based on carboxylated multiwalled CNTs (MWCNTs) and reduced graphene oxide (rGO) has been reported for the ultrasensitive detection of the ovarian cancer antigen.
Liquid-gated rGO-TFT is particularly promising as biosensors since they allow to directly couple the active material and the biological element [16], [17] by exploiting the remarkable proprieties of graphene and graphene oxide (GO).The latter bears different oxygen groups (viz.hydroxyl, epoxy, carboxy, etc.) that allow a wide range of chemical functionalization to anchor chemical receptors for a specific analyte of interest.Furthermore, the reduction process to obtain rGO offers the possibility of tuning its electrical conductivity as well as the density of oxygen groups [18], [19], thereby leading to high-capacitance transducing devices optimal for biosensing applications [20], [21].
In this work, we want to explore a possible application of the rGO-TFTs as transducing devices for the detection of Lactococcus lactis (L.lactis), whose biological activity affects milk fermentation in dairy plants.The proper growth of these bacteria in milk lots can aid the production of high-quality products due to their lactic fermentation process.Usually, the bacteria monitoring is performed through microbiological methods, such as real-time quantitative polymerase chain reaction (RT-qPCR).Very few sensing solutions have been proposed to monitor L. lactis cells growth.In [8] and [10], a screenprinted electrochemical biosensor monitoring the growth of L. lactis to detect the presence of its bacteriophage was proposed.The device was based on electrochemical measurements, and the different bacterial growth on the electrodes was determined by the variation of charge transfer resistance.Due to the final application in the industrial field, a more scalable and portable approach is desirable for this method.Hence, we hereby present the development, fabrication, and characterization of a liquid-gated rGO-TFT biosensor for the monitoring over time of the L. lactis for dairy industry application.The sensing performances are reported, and the feasibility of complex biological detection systems through rGO-TFT transduction is discussed.

A. Biological Materials and Chemicals
The L. lactis was purchased from DSMZ (Germany).The selected culture medium for the bacterial growth was M17 broth (pH 7.1), provided by HiMedia Laboratories.The M17 broth is necessary to stimulate the bacteria proliferation and its composition is reported in Table I.The L. lactis bacteria were quantified by measuring the optical density (OD) of solutions in cuvettes through a spectrophotometer (ONDA, UV-30 SCAN) at 600 nm.The starting concentration was 10 7 CFU/mL as reported for dairy production [22].
All the used chemicals were of analytical grade.The water was filtered at 0.22 µm and deionized to ultrapure Milli-Q water (MQ), with a final conductivity < 2000 µS/cm.The calcium chloride (CaCl 2 ) was purchased from Sigma Aldrich (Germany).The CaCl 2 was diluted 100 mM in MQ and used as a phage activity enhancer.

B. Fabrication of the Devices
We fabricated and characterized as solution-gated transistors and two different types of rGO-TFTs based on interdigitated gold electrodes.

1) TFTs fabricated on commercially available n-doped
silicon substrates (Si-TFT), with a resistivity of 0.01-0.03•cm.The gold interdigitated electrodes had a thickness of 100-150 nm (width W -to-length L ratio W/L = 560).These devices were provided by Fondazione Bruno Kessler (Italy), and they are shown in Fig. 1(a).2) TFTs fabricated on commercially available polyethylene terephthalate (PET-TFT) substrate (Metrohm, Spain).The gold interdigitated electrodes had a thickness of 10-20 nm and W /L = 285.The device is shown in Fig. 1(b).
The GO (4 mg•mL −1 , monolayer content > 95%, graphene) deposition was performed on the two devices according to the substrate characteristics.The Si-TFT was first cleaned with subsequent rinsing of acetone, isopropanol, and MQ water to remove the protective layer.The PET-TFT underwent a 15-min cleaning in UV Ozone Cleaner.A 10-µL drop of a poly(diallylammonium chloride) solution (PDDA, Sigma Aldrich), 1% w/w, and 0.5-M NaCl, was deposited on the interdigitated areas for 30 min.The devices were carefully rinsed with MQ water, and then, a GO solution (1 mg/mL) was deposited on the interdigitated areas for 3 h, thereby avoiding any solvent evaporation.The final devices were rinsed with MQ water and dried by a nitrogen stream.The GO layer was then electrochemically reduced using an external gold squared plate (0.5 cm 2 ).The two interdigitated electrodes were short-circuited and a drop of MQ water was placed on the samples.A potential sweep between 0 V and −3 V was applied until reaching the desired level of GO reduction, namely, a resistance check was performed by sweeping the potential from −100 to 100 mV.The ohmic behavior of the dry device shows resistance between 10 and 60 k .
The experimental measurements on these rGO-TFTs were performed in a probe station using an external gold squared plate as a gate electrode and solutions of M17, L. lactis bacteria in M17, and phages and L. lactis in M17 as a gate medium.The devices' general structure is schematized in Fig. 1(c).The gold interdigitated electrodes act as drain and source on Si or PET substrate, and the gate comprises the solution drop and the external gold plate.All the reported measurements were repeated at least three times on different devices.

C. Modeling of rGO-TFTs
The rGO-TFTs exhibit an ambipolar behavior, modulating both electrons and holes through the channel, and an always-ON state due to the narrow energy bandgap [23].Hence, it is fundamental to adequately model the characteristics of these devices to estimate transistor parameters [24].
Nomenclature summarizes the adopted conventions and terminology for this work.The two electrodes are labeled as negative (N) and positive (P).Since the device is ambipolar, one of the two electrodes behaves like an electron source and a hole drain (n-type source), while the other acts as a hole source and electron drain (p-type source).The potential applied to the P electrode (V P ) is always positive, and the potential applied to the N electrode is always negative (V N ).Therefore, P is the electrode injecting holes (p-type source), and N is injecting electrons (n-type source).The current from P to N (I PN ) is always positive due to the applied potentials.
The operative regions of ambipolar TFTs can be summarized, as described in Fig. 2, going from one region to another while sweeping V G for fixed values of V N and V P [24], [25].The devices are characterized by two different voltage thresholds for p-and n-type channels (V TH p and V THn ), where the condition V THn > V TH p must be satisfied to maintain the holes accumulation and electrons accumulation, respectively, for negative and positive overdrives.If V PN < V THn − V TH p , the TFT operates as unipolar p-type transistor if or in the OFF-state where the solely field-effect modulation current from P to N ( , or in the bipolar state, where both electrons and holes are injected, respectively, from N and P, behaving as a series of p-and n-type in saturation regime.The equations modeling the current of each operative region as a function of V G are reported in (1), as shown at the bottom of the next page, where C l is the liquid gate capacitance, α p/n is the mobility enhancement factor, and µ 0 p/n is the carrier mobility when the overdrive voltage is V 0 p/n , which is assumed as an empirical parameter.
The always-ON state determines the presence of an OFF-state parasitic current, which is independent of V G .The rGO-TFT can be modeled by introducing a constant resistance R PN in parallel with the transistor, as reported in Fig. 3. Hence, the total current is due to the field-effect modulation current (1) and the current flowing through R PN

A. Parameters Extrapolation, Hysteresis Study, and Modeling of the Devices
We analyzed the characteristics of the rGO-TFT under test using the reported model described in (1) and ( 2).The parameters of interest are k, which represents the physical parameters of the device (C l • W • µ 0 p/n )/L, R PN , the threshold potentials V TH p/n , and enhancement factor α p/n .
We defined a function H from the device I PN and I PN_FET in the saturation regions.H is described as a function of V GP , and it is also dependent on the source applied potential (V S ) and whether the device is working as a p-or n-type Fig. 3 shows the H function of a representative measurement of Si-TFT obtained using M17 broth as the liquid gate.The curve can be fitted as a linear equation y = mx + q, where x is V GP .Hence, the parameters of interest α p/n and V TH p/n can be extrapolated, respectively, from angular coefficient m and intercept q.Meanwhile, R PN and k can be retrieved from (2) using the newly found V TH p/n and α p/n as constants.Hence, the devices under test are fully characterized Authorized licensed use limited to the terms of the applicable license agreement with IEEE.Restrictions apply.The hysteresis study in Fig. 4(a) reports a slight shift between the forward and backward I PN -V GP scans of the Si and PET devices, which is consistent with the behavior of rGO-based transistors [26].The different hysteresis and I PN -V GP characteristics of the tested devices are most likely due to the substantial difference of the substrates' roughness, which can influence the charge trapping in the channel.We modeled the TFTs responses assessing if the ambipolarity of the devices was maintained using a culture broth as the liquid gate.Fig. 4(b) shows an excellent agreement between the I PN -V GP curves of Si-TFT and PET-TFT and the mathematical model, which can fully describe all the devices' operative regions despite the usage of a complex medium as the liquid gate.The slight discrepancy between data and fit, visible from 0.1 to 0.2 V, is most likely due to a small hysteresis during transfer characterizations that can introduce some uncertainty on the extrapolated parameters.

B. Stability and Reproducibility Tests
We tested the devices Si-TFT and PET-TFT using M17 broth as the liquid gate to assess their stability over time to the test solution.The measurements were repeated on each device for 60 min.
Fig. 5 reports the I PN -V GP and I G -V G characteristics at V P = 0 V and V N = −0.5 V of Si-TFT with V G ranging from −0.5 to 0.2 V and PET-TFT with V G ranging from −0.3 to 0.2 V. Fig. 5(a) shows a slight current increase of the Si-TFT I PN -V GP characteristics, which is particularly evident around −0.3 V (+9%).Moreover, a slight change in pendency is visible after 60 min of observation.Meanwhile, PET-TFTs I PN -V GP characteristics [Fig.5(b)] show a slightly higher current increase (+5%) around the curves' minimum and a slight shift of potential.Despite these shifts, the variation Authorized licensed use limited to the terms of the applicable license agreement with IEEE.Restrictions apply.The curves are retrieved with V P = 0 V and V N = −0.5 V.
over time of the threshold potentials ( V TH p/n ), the variation of a minimum current ( I OFF ), and the variation of transconductance ( g m ) are almost negligible, as reported in Table II, thus proving the stability of both Si-TFTs and PET-TFTs using M17 broth as a liquid gate.It is also worth mentioning that both the devices reported a negligible hysteresis during the measurements; thus, a low leak current at both P and N terminals can be deduced.Fig. 5(c) and (d) reports the I G -V GP characteristics of Si-TFTs and PET-TFTs, respectively.In both devices, the I G currents are around the order of magnitude of 10 −7 A and almost constant throughout the experiments, demonstrating a low leakage at the gate electrode.Within this range of potentials, no faradaic peaks are recorded; hence, the leakage current can be inferred as purely capacitive.
The reproducibility of the measurement is studied as deviceto-device variability, reported in Fig. 5(e) as the mean and standard deviation (m ± sd) of V THn .The devices show an analogous decreasing trend of V THn , also having similar variability as visible from the sd, which, for example, is ∼0.02V at 60 min for both TFTs.

C. Detection of Bacteria Presence in Liquid Gate Solution
The sensing rationale relies on the phage lytic activity in the presence of L. lactis bacteria cells.If the sample under test presents little to no phages, the bacteria are free to proliferate.Meanwhile, if there is sufficient phage contamination, the bacterial growth is hindered by the phage lytic activity, thereby leading to a final solution almost deprived of bacteria and with an enormous number of phages.Hence, two different solutions should be compared: the sample under test and a solution at a known and low-level concentration of bacteria as negative control.As the negative control allows a standard bacteria growth, the corresponding device provides the electrical readout for such a bacteria growth.Therefore, we first tested the devices' capability to detect the bacteria's presence.The final bacteria solution at O.D. = 1 was prepared by adding 1200 µL of L. lactis at high concentration to 1800 µL of M17.A plain M17 solution without any bacteria was used as a control.A 60-µL drop of the prepared solutions was used as a liquid gate to test the devices.We changed the gate drop after each measurement to avoid side effects due to the bacteria deposition on the rGO layer.The devices' surface was accurately washed with MQ before depositing a new drop.6(d)] are almost constant throughout the experiment, suggesting that I G is independent of the presence of L. lactis at the liquid gate.In fact, a massive deposition of bacteria on the gate surface should cause an evident change in capacitance.However, in this case, the I G curves report no signal variation attributable to such an event.Hence, the main mechanism of bacteria detection and monitoring can be considered independent of bacteria proliferation at the gate and should be further investigated in future work.

TABLE II SI-TFT AND PET-TFT ELECTRICAL PARAMETERS VARIATION OVER TIME
We calculated the variation of the electrical parameters between L. lactis and M17 curves in order to retrieve which parameters are most influenced by the bacteria presence and the devices' detection capability.Table III reports the values of I OFF (I OFF−L.lactis− I OFF−M 17 ), the variation of threshold potentials V TH p/n (V TH p/n−L.lactis− V TH p/n−M17 ), and the variation of transconductance between L.lactis and M17 as percentual increase ( g m ).The bacteria presence can be effectively detected through these electrical parameter shifts, which are particularly evident in PET-TFT.Therefore, it can be concluded that PET-TFT has an optimal performance for bacteria presence detection with respect to Si-TFT.

D. Bacterial Proliferation Monitoring
The bacterial proliferation in the M17 solution was monitored over time in order to assess the devices' capability of detecting increasing concentrations of L. lactis.For this test, we selected the PET-TFT due to the optimal parameters' variation.I PN -V GP measurements were performed in a wide time range using a solution of L. lactis as the liquid gate.The bacteria solution was prepared by diluting L. lactis in M17 reaching an O.D. of 0.15.A plain M17 solution without any bacteria was used as the negative control.Before testing, the solutions were incubated at 37 • C in a hot bath to stimulate the bacteria proliferation.The bacteria growth in the solutions was monitored through optical absorbance measurements at 600 nm throughout the experiment [8].We performed the electrical measurements after 100 min of incubation (t 0 ) when  the bacteria solution reached a final O.D. of 0.2.A 60-µL drop of the solutions under test was used as a liquid gate.After each measurement, the drop was removed to avoid side effects due to the bacteria proliferation and deposition on the rGO layer.The device was carefully rinsed with MQ.After every 45 min, another drop was deposited on the gate electrode and the measurement was repeated.Fig. 7 shows the electrical responses of L. lactis-gated PET-TFT observed in a time range going from t 0 to t 0 + 270 min.The I PN -V GP signal [Fig.7(a)] shifts at each reported time due to an evident curve shift toward negative potentials and increasing I OFF , which is particularly noticeable between time t 0 and t 0 + 270 min (+2.4 µA).This characteristic variation over time is consistent with an increasing bacterial concentration in the liquid gate.Notably, the I GP -V GP curves [Fig.7(b)] are almost constant, confirming the leakage current independence from the bacterial growth.
We calculated the variation of the electrical parameters over time to further assess the device detection capability, comparing L.lactis-gated devices at increasing bacteria concentration and M17-gated responses in a time range from t 0 to t 0 + 270 min.Fig. 8 reports the O.D. of the solutions under test and the variation of characteristics parameters as a function of time, i.e., I OFF = I OFF − I OFF−t 0 , V THn = V THn − V THn−t0 , and g m as percentual increase.We consider only V THn since the bacteria presence in PET-TFT causes a potential shift toward negative potentials, hiding the p-type saturation region of the transistor and thus hindering an accurate extrapolation of V TH p .The bacterial growth in the M17 solution is monitored through optical absorbance in a time range from 0 to 370 min [Fig.8(a)].The L. lactis solution (red curve) shows an evident O.D. increase over time due to the proliferation of the bacteria.The bacteria population is at an initial O.D. of 0.15; then after 100 min, the bacteria starts to proliferate exponentially, approaching a plateau (O.D. = 1.25) around 300 min due to the saturation of the solution.The final detected concentration is ∼10 8 CFU/mL, consistent with the RT-qPCR results obtained from dairy samples [27].The monitored concentration range from 10 7 CFU/mL to 10 8 CFU/mL is also in agreement with the detection range of other biosensors for monitoring bacterial growth, which is from 106 to 107 CFU/mL to 10 8 -10 9 CFU/mL according to the considered bacteria species and the application of interest [28], [29].On the other hand, the negative control curve (black) remains unchanged, with O.D. values almost equal to 0 throughout the experiment, indicating the absence of undesired contaminations in the M17 solution.The extrapolated I OFF of L. lactis-gated and M17-gated TFTs [Fig.8(b)] shows a similar increasing trend over time, reaching comparable final values.Therefore, I OFF cannot be considered a suitable parameter to monitor the bacteria growth, showing inconsistent results to the O.D. curves.The M17-TFT g m [Fig.8(c)] shows a slight increase after t 0 , followed by a rapid decrease at 280 min, while the L. lactis g m reports almost the opposite behavior.Nevertheless, the curves are highly unstable, and they do not allow the definition of a clear trend consistent with O.D. results.Hence, we can conclude that I OFF and the transconductance of PET-TFTs are not strongly influenced by the bacteria's presence on the gate electrode.Meanwhile, V THn L. lactis solution [Fig.8(d)] shows a substantial variation of the parameter over time, which decreases to −0.12 V with a well-defined linear trend with p1 = −3.9× 10 −4 V/min and p2 = 0.04 V (R 2 = 0.97, p1 is the slope and p2 is the intercept of the linear equation).On the other hand, M17 reports an almost constant V THn throughout the experiment with negligible variation.These results are consistent with the optical absorbance curves, showing that bacterial proliferation can be effectively detected Authorized licensed use limited to the terms of the applicable license agreement with IEEE.Restrictions apply.
by monitoring the V THn variation.Moreover, the presence of bacteria in the solution is detectable after 150 min and can be continuously tracked throughout the experiment.For instance, in [10], the presence of L. lactis was only detectable after 240 min and quantifying its proliferation was challenging due to signal saturation.Therefore, PET-TFT can offer significant advantages by enabling early detection and continuous monitoring of bacterial growth over time, thereby allowing for signal quantification throughout the experiment.

IV. CONCLUSION
In this work, we developed a FET biosensor for detecting L. lactis bacterial cells using an ambipolar rGO-TFT as a transducer.Two different TFTs based on Si substrate and PET substrate were characterized through M17 culture broth as the liquid gate.Both devices were stable throughout the experiments.The parameters of the devices were extrapolated through a mathematical model.The detection method is based on the evident parametrical shifts of V THn caused by the bacterial presence at the liquid gate.Experimental tests showed that PET-rGO-TFTs have the best electrical response for L. lactis detection, showing V THn of more than −100 mV with respect to Si-rGO-TFT.The bacteria proliferation was quantified through the variation of threshold potential in a time range of 270 min.The bacteria in the solution can be detected after 45 min of tracking test.These results demonstrated the feasibility of monitoring L. lactis growth by using the rGO-TFT, which exploits its field-effect modulation.Future work will aim to further test the proposed device in order to detect L. lactis bacteriophages for dairy industry applications, also studying the main mechanism allowing such a detection.Moreover, other rGO-TFT configurations will be fabricated and characterized to design novel FET-based biosensors.

Fig. 1 .
Fig. 1.(a) Image of Si-TFT.(b) Image of PET-TFT.(c) Schematic general structure of the devices under evaluation.

Fig. 3 .
Fig. 3. H(V GP ) function.The red area highlights the linear regions of the function where the parameters of interest are calculated.

Fig. 4 .
Fig. 4. (a) Backward and forward scan of the I PN -V GP characteristics for Si-TFT (left) and PET-TFT (right).(b) I PN -V GP curve fitting with (1).The curves are retrieved with V P = 0 V and V N = −0.5 V.

Fig. 5 .
Fig. 5. Stability and reproducibility evaluation during 60 min of observation of the devices under test.(a) I PN -V GP curves of Si-TFT.(b) I PN -V GP curves of PET-TFT.(c) I G -V GP curves of Si-TFT.(d) I G -V GP curves of PET-TFT.All the curves are retrieved with V P = 0 V and V N = −0.5 V. (e) Device-to-device variability reported as m ± sd of ∆V THn .

Fig. 6
shows the electrical responses of Si-TFT and PET-TFT in the presence of a high concentration of bacteria compared with the devices' signals with the control solution as a liquid gate.The I PN -V GP characteristics evidence a clear difference between the L. lactis-gated TFTs and the M17-gated TFTs.Si-TFT [Fig.6(a)] shows immediately an I OFF increase and a slight shift of −0.02 V toward negative potentials in the presence of L. lactis.PET-TFT [Fig.6(b)] shows a clear difference between L. lactis and M17 characteristics.The L. lactis curve reports an evident shift of −0.145 V toward negative potentials, almost hiding the p-type region of the transistor, while I OFF increases by 1.9 µA.Hence, M17 and Authorized licensed use limited to the terms of the applicable license agreement with IEEE.Restrictions apply.

Fig. 6 .
Fig. 6.Comparison of Si-TFT and PET-TFT responses with L. lactis spiked liquid gate.(a) I PN -V GP curves of Si-TFT.(b) I PN -V GP curves of PET-TFT.(c) I G -V GP curves of Si-TFT.(d) I G -V GP curves of PET-TFT.All the curves are retrieved with V P = 0 V and V N = −0.5 V.

Fig. 7 .
Fig. 7. PET-TFT response during L. lactis growth monitored during 370 min of observation.(a) I PN -V GP curves.(b) I G -V GP curves.The curves are retrieved with V P = 0 V and V N = −0.5 V.

Fig. 8 .
Fig. 8. Parameters analysis of PET-TFT characteristics as a function of time of M17-gated device and L. lactis-spiked M17-gated device.(a) Optical absorbance.The time of observation t 0 is equal to 100 (b) Variation of I OF F .(c) Percentual variation of transconductance g m .(d) Variation of threshold potential V THn with linear fitting of L. lactis response.

TABLE III SI
-TFT AND PET-TFT ELECTRICAL PARAMETERS VARIATION BETWEEN M17 AND L. Lactis