General consideration of biological gFETs

A typical gFET biosensor consists of a graphene channel between the source and drain electrodes on an insulating substrate. The density of charge carriers in this graphene channel, and hence the current, is modulated by a local electrostatic field, which is itself changed by modulations in the environment around the channel. The fundamental elements of a gFET are the source/drain contacts, the gate electrode, and the 2D graphene channel for sensing (Fig. 1a). Depending on the positing of the gate, one can distinguish a variety of gFET configurations including next to others back-gated, top-gated, and liquid-gated concepts. The interest in using graphene over other materials is due to its atomically thin geometry, which makes its electrical conductance highly responsive to bioreceptor-analyte binding events close to the graphene surface. The way the bioreceptor is immobilized onto the graphene channel together with its charge will influence the sensing performance of the gFET in the same manner as will be the size and the charge of the analyte. The application of a gate voltage (VG) creates an electric field on the graphene channel, modulating the conductivity of graphene and consequently the drain-source current (IDS) (Fig. 1b). In the case of gFET, such transfer curves are characterized by a minimum conductivity, the Dirac point (VDirac), which is generally observed at zero gate voltage (VG = 0) (Fig. 1b). When a target molecule binds to the receptor on the graphene surface, the redistribution of electronic charge generates a change in the electric field across the FET channel region, which changes the electronic conductivity in the channel and the overall device response (Fig. 1b). While similar devices have been fabricated with silicon FETs for years, limited sensitivity and poor selectivity was achieved. Indeed, the sensitive detection of bioreceptor-analyte binding events in a gFET is related to either graphene doping effects by direct charge transfer between the formed bioreceptor-analyte duplex and the graphene channel and/or electrostatic gating effects (Fig. 1b). Gating effects are ascribed to the accumulation of charges on the graphene surface arising from bioreceptor-analyte binding, resulting in a local external voltage drop across the channel. If VG < VDirac, with VDirac being the charge neutrality point equivalent to the minimum conductivity, called Dirac point, then the Fermi level is located in the valence band and holes are the majority charge carriers, conversely, if VG > VDirac then the Fermi level is located in the conduction band and electrons are the majority charge carriers (Fig. 1b). Positively charged analytes result generally in a shift of VDirac to more negative gate voltages. In contrast, the negatively charged target molecules will increase the density of holes in graphene and generate a positive shift. The left branch of the transfer curve (Fig. 1b) represents thus an increasing density of positive charge carriers (holes), while the right branch corresponds to increased negative charge carriers (electrons), both branches extending linearly from the Dirac point according to Eq. (1):

$$_}=}_} (_}-_})\mathrm_}=(W/L) \mu _}_}$$

(1)

with the slope gm (transconductance) being depending on the width (W) and the length (L) of the graphene channel as well as on the mobility of charge carriers (µ) and the gate capacitance (Cg).

Fig. 1

Graphene field-effect transistor assemblies: a schematic illustration of back-gated, top-gated, liquid-gated, and co-planar configurations. b Transfer curves (IDSvs. VG) and change in position of Vdirac upon sensing with positively (green) and negatively (brown) charged analytes

Back-gated gFETs for biological sensing

Due to the influence of the gate capacitance on IDS, the choice where the gate is positioned becomes of high importance for biosensing applications. Si|SiO2 remains the preferred substrate, owing to its well-established compatibility with nanoelectronics and the use of back-gating still seizes attention among the sensor community. Other solid substrates such as quartz [7, 8] and glass have lately been more widely used, but in connection with liquid-gated biosensing [7]. In back-gated configuration, Cg is dominated by the insulating layer separating the graphene channel from the gate which is typically 100 nm to few µm and thus requires the application of rather high VG voltages to drive the FET device. A back-gated gFET for opioid sensing down to 10 pg mL−1 based on chemically bonded μ-opioid receptor proteins was recently proposed by the A. T. Charlie Johnsons’s group (Fig. 2a) [9].

Fig. 2

Back-gated gFET for biosensing: a (left) transfer characteristics before and after exposure to naltrexone of a back-gated gFET array and (right) change of VDirac with increasing naltrexone concentration (reprint with permission of ref. [9]). b Exosome sensing on back-gated gFET modified with anti-CD63 antibody via 1-pyrenebutyric acid N-hydroxysuccinimide ester (PBASE) linkers alongside the change in Vt, being the position of the Dirac point at a given time point, over time upon addition of different concentrations of exosomes (reprint with permission from ref. [10]). c Buried-gate-based gFET for sensing of IL-6 by recording changes in the equilibrium ΔV/ΔVmax as a function of IL-6 concentration (dashed lines are a least-squares fit to the Hill − Langmuir equation, yielding equilibrium dissociation constants (KD) in gargle solution (green) and 1 × PBS (blue) (reprint with permission from ref. [11])

The gFET sensor was functionalized with a computationally designed water-soluble variant of the human μ-opioid receptor (G protein–coupled receptor) using 4-carboxybenzenediazonium tetrafluoroborate, which produced carboxylic acid sites on the graphene, which were further activated and stabilized with 1-ethyl-3-[3-dimethylaminopropyl] carbodiimide hydrochloride/sulfo-N-hydroxysuccinimide (EDC/sNHS). Electronic measurements of the source-drain current as a function of the back-gate voltage following each step of the functionalization procedure showed reproducible shifts in conductance. The proposed mechanism for the concentration-dependent change in VDirac is linked to a conformational change in the binding pocket of the μ-opioid receptor upon naltrexone binding, which alters the electrostatic environment of the gFET and results in a “chemical gating effect. The possibility for the sensing of exosomes (0.1–10 μg mL−1) with a back-gated antibody-modified gFET was recently demonstrated (Fig. 2b) [10]. As the exosomes are negatively charged and interact with the antibody-modified graphene channel acting as a dielectric layer, a positive charge accumulation in the graphene is detected. The maximum thickness of the functionalization layer was estimated at about 12 nm, 2 nm of the surface linkage, and 10 nm of the antibody, and sensing in low ionic strength solution (0.001 × PBS) was performed only to overcome Debye length screening limitations. Interestingly binding between the exosomes and antibodies occurred under these low ionic strength conditions.

Recently, a buried-gate electrode formed via a bilayer lift-off photolithography process was proposed for interleukin-6 (IL-6) sensing [11] and overcame the requirement for the application of high VG voltages to drive the FET device. The planar gate electrode consisting of a Cr/Au structure (2 nm/43 nm in thickness, respectively) was defined on SiO2/Si, coated with a 30 nm HfO2 thick dielectric using atomic layer deposition (ALD) system onto which a gold-based drain/source electrode was formed and chemical vapor deposition (CVD) graphene transferred (Fig. 2c). The electrical signal was wirelessly transmitted to a smart-phone through a Wi-Fi connection for visualizing the trend of the cytokine concentration change reaching a LoD of 12 pM for interleukin-6 (IL-6) in saliva. This is one of the several examples of the potential of a portable gFET for disease diagnostics at an early stage in complex high ionic strength solutions such as saliva paving a new avenue for monitoring conditions of the high-risk population. Still, the construction of the device requires adapted clean room facilities making its wider application out of research laboratories currently of limited use.

Flexible electronics has become a very active research field, driven by a potentially enormous market for smart and wearable devices. Polymeric substrates such as polyimide have become attractive for flexible electronics [12-16] and flexible gFETs are an active research direction [12]. Initial steps toward developing flexible aptamer-based FET biosensors, for example, for neurotransmitter monitoring are that by Zhao et al. [14] based on nanometer-thin-film In2O3 back-gated FETs.

Top-gated and co-planar gated gFET-based biosensors

Top-gated gFETs are widely found in publications related to radio frequency applications [17, 18] and are indeed a suitable approach to applications where a thin oxide layer gate is advantageous to exert more control over the electrostatically doped carriers in the channel with lower gate bias required to modulate the channel. A top-gate-based electrode (10 nm/60 nm Ti/Au) (Fig. 3a) was defined for example on graphene-coated Si/SiO2 post-modified with source and drain electrodes (10 nm Ti/50 nm Au) by a series of steps that include photolithography, e-beam evaporation, and lift-off, resulting in a top-gate channel length of 48 μm [19]. Strain sensing was performed on this device, but there is currently no biological application reported. The difficulty of growing an oxide on top of the graphene, without damaging its lattice and thus degrading the mobility of its free carriers, remains a hurdle and limits this approach strongly.

Fig. 3

Top and liquid-gated gFET for sensing: a top-gated gFET design coupled with biodegradable piezoelectric material–based dynamic pressure sensor (reprint with permission of ref. [19]). b IDSvs. VG of a liquid-gated gFET based on gold interdigitated electrodes (IDE) covered with CVD graphene modified with ethynyl functional groups as well as aptamers [22]. c Influence on IDE architecture on (left) IDSvs. VG curves, (middle) Ratio of ID/IG (dark blue) and I2D/IG (bright blue) peak intensities and (right) shift of the Dirac point upon the addition of MPP-9 in 1 × MOPS to aptamer-modified chips (unpublished results)

A flexible co-planar gated gFET, applied for consistent and time-resolved detection of cytokines in human biofluids and allowed for the sensitive detection of TNF-α and IFN-γ with limits of detection down to 2.75 and 2.89 pM, respectively [20]. However, this area is still in its infancy, in particular, when compared to organic field-effect transistors (OFETs) a focused research hotspot in recent years because of the fast development of flexible electronics [21].

Liquid-gated gFET-based biosensors

As biomolecules such as proteins and nucleic acids are present in biological fluids, co-planar and liquid-immersed-gate configurations are largely favored in biological gFET design as it allows for sensing in the fluid sample directly without intermediate drying steps. It is the electrical double layer (EDL) formed at the graphene/electrolyte interface rather than the position of the gate electrode which determines the capacitance as it acts like a very thin dielectric layer. The resulting capacitance (Ctotal) is much larger than that of back-gated dielectric and is determined by Eq. (2):

$$\begin_}=_}+1/_}]}^& \mathrm\;_}=_}_}A/_}\end$$

(2)

with Cq being the quantum capacitance originating from changes in total charge to chemical potential (Fermi level, EF) dQ/dEF of the 2D material, Cdl is the double layer capacitance of the 2D/electrolyte interface, εr corresponds to the relative permittivity of the electrolyte, εo is the vacuum permittivity, A is the area of the graphene channel, and λD is the Debye length. 2D materials such as graphene exhibit a high Cq and a small change in its density of state results in a significant change in its Fermi level. As Cdl is usually one order of magnitude larger than Cq, it leads to a dominant contribution of Cq to the total capacitance. Any accumulation of the analyte on the bioreceptors will lead to a considerable shift of the Fermi level, resulting in the high sensitivity of these gFET biosensors. This means that gate potentials applied across the EDL are at least two orders of magnitude more efficient than through the back-gate with a much smaller sweeping range of VG required to capture the linear p and n branches being in the order of ± 1 V compared to ± 10 V for thin oxides and ± 100 V for thicker oxide layers in the back-gated configuration. The avoidance of unwanted water hydrolysis as a side reaction during the gate bias sweep in liquid media and other electrochemically driven reactions motivates the choice of the liquid or in-plane gated gFET configuration.

Our team together with others have extensively investigated interdigitated gold electrodes printed onto glass slides as a base for the construction of liquid-gated gFETs (Fig. 3b) [22-26]. Interdigitated gold electrodes (IDE) consisting of 90 electrode pairs of 10 µm in length with a 10 µm separation and a total surface area of 9.62 mm2 (r = 1.25 mm) were coated with CVD graphene in this case. We could show that electrochemical reduction of 4 [(triisopropylsilyl)ethylenyl]benzenediazonium tetrafluoroborate (TIPS-Eth-ArN2+) followed by the chemical deprotection of the triisopropylsilyl (TIPS) function leads to gFET devices with largely improved drain-source current (IDS) as a function of gate voltage (VG) with hole and electron mobilities reaching 1739 ± 376 cm2 V−1 s−1 and 1698 ± 536 cm2 V−1 s−1, respectively [22] and well adapted for further sensing. To get a better understanding of the influence of the IDE design, the IDS/VG characteristics of gFETs formed on interfaces with a varying number of IDE (Fig. 3c) were compared. Decreasing the number of IDE further resulted in gFET with significantly lower charge mobilities in the electron and hole regions. The intensity ratio for the D, G, and 2D Raman bands is often used as a criterion to assess the graphene quality. The Raman intensity ratios, ID/IG and I2D/IG, are comparable for all devices and indicate little defects and thus high-quality transferred graphene (Fig. 3c). Devices 1, 2, and 6 were in addition modified with an aptamer specific for MMP-9 as reported previously [27]. While devices 2 and 6 showed smaller changes in VDirac, the sensing sensitivity for MMP-9 in 1 × MOPS buffer remained comparable in all cases, suggesting no influence of the width/length ratio of the sensing channel area and charge mobility.

Gao et al. [28] reported lately the development of a flexible liquid-gated biosensor for ultra-sensitive and specific detection of miRNA with LOD as low as 10 fM within 20 min. The device was fabricated on a flexible polyimide substrate and integrated with a microfluidic chip containing an inlet and an outlet for sample loading and gate electrode placement in the liquid-gate solution. The work provides hope for developing flexible and wearable biosensor platforms for future POC diagnostics.

For liquid-gated gFETs, the Debye length λD (Fig. 4a), however, plays a key role in the sensing performance of the sensor. The Debye length is the distance at which the potential of a net charge is screened to 1/e of its maximum value by mobile ions in the medium. According to the Debye-Hückel model, charged molecules in solution are screened by mobile counter-ions such that their electrical potential is exponentially dampened with distance λD being the decay constant called Debye length which is given by Eq. 3:

$$_}\left(\mathrm\right)=\sqrt_}_I^}}, I(\mathrm\;}^)=\frac\sum __\left[\mathrm_}=0.304/ \sqrt\right]$$

(3)

with kB being the Boltzmann constant, T the absolute temperature, NA the Avogadro’s number, e the electron charge, I the ionic strength, and ρi and zi the density and valence of the ion species i, respectively. Charges located outside λD are considered out of range for electrostatic gating-based detection by a FET sensor (Fig. 4a). Under physiological conditions (> 150 mM or 1 × phosphate-buffered saline (PBS)), this accounts for λD = 0.7 nm, and increases to 2.4 nm (0.1 × PBS) and 7.4 nm in lower ionic strength solutions (0.01 × PBS). With the length of antibodies being around 10–15 nm or with a 30-base aptamer of about 10 nm, there is an intrinsic mismatch in dimensions between the bioreceptor and the charge screening. As pointed out lately by Soh and his group [29], the concept of double-layer crossing (orange cross section in Fig. 4a) is often overseen. Despite this challenge, the application of gFET sensors in science has been possible via the implementation of innovative strategies. To generate an electrical signal change, the target molecule’s double layer (pink in Fig. 4a) must interact with the graphene double layer (yellow in Fig. 4a) (Fig. 4a). If the target creates sufficiently high double-layer potentials, signals can be detected even if the target binds many Debye lengths away from the electrode. While this concept has not been further developed until now, it was in 2015, when Gao et al. [30] demonstrated that a poly(ethylene glycol) (PEG) surface (Fig. 4b) coating decreased charge screening on FET-based biosensor and increased considerably the λD. It was shown that the detection of prostate-specific antigen (PSA) in 0.1 M PBS buffers is possible by co-immobilizing a PSA-specific aptamer on the electrode alongside the PEG coating (Fig. 4b) [30]. Indeed, next to the well-organized charged layer on top of graphene and guiding the Debye length, in the presence of a PEG layer on graphene, any charges or dipoles leading to a charge within that immobilized ion-permeable layer requires an extra accumulation of a counter-ion within the layer to maintain charge neutrality. This difference in the concentration of ions between the bulk solution and that in the immobilized protein layer creates a Donnan potential extending beyond the Debye length, in this ion-permeable PEG [31, 32]. The partially hydrated PEG changes the interfacial capacitance of the gFET with higher molecular weight PEG being preferable over lower weight ones. Too long PEG chains will however increase binding kinetics due to diffusion limitations between the bioreceptor and the analyte via the PEG layer [33]. Since then, PEG has become essential in gFET sensing for overcoming fouling issues as well as Debye length screening effects.

Fig. 4

Debye length screening hurdle and different strategies to overcome this for gFET biosensors: a correlation of bioreceptor size with Debye screening length in solution of different ionic strengths. The electrical double layer formed on the graphene interface is in yellow and that of the analyte is shown in pink. The cross section of these electrical double layers is in orange. b Effect of high molecular mass poly(ethylene glycol) (PEG)-modified gFET on the Debye length and its effect on the concentration-dependent measurements of PSA (10 nM) in 100 mM PB (reprint with permission from ref. [30]) as well as the obtained change of IDSvs. the PBS concentration of the aptamer only, and for the aptamer/PEG gFET [23]. c Scheme of crumpled gFET DNA sensor where the blue dot lines represent the Debye length in the ionic solution. SEM images of crumpled graphene: scale bar is 5 µm (left) and 500 nm (right) (reprint with permission of ref. [37]). d (left) Use of nanobodies instead of antibodies and (right) a combined strategy of nanobodies and PEG chains to overcome Debye limitations

We together with others have followed a similar approach for the sensing of cardio troponin I (cTnI) in serum of patients [23] using a liquid-gated gFET configuration. The device responses were, as excepted, sensitive to the ionic strength of the solution as seen upon the change in IDS of the aptamer:PEG-modified GFETs upon exposure to 240 pg mL−1 of cTnI in PBS at different concentrations (Fig. 4b). The change in IDS was smaller upon increasing ionic strength solution with signals detectable in 0.1 M clearly, while an aptamer-only-modified gFET revealed no current change already in 0.01 M PBS. To push further and realize sensing in serum samples, de-salination of the serum through a Sephadex® (cross-linked dextran gel) G-25 column using gravity for separation allowed to identify correctly cTnI concentration in 15 patient samples grouped according to the magnitude of perioperative myocardial injury risk for a myocardial infraction as mild (cTnI < 15 pg mL−1), moderate (15 pg mL−1 > cTnI < 500 mL−1), and severe (cTnI > 500 pg mL−1) [34]. In addition, the PEG unit worked as an antifouling matrix. As-deposited graphene is commonly hydrophobic and promotes adsorption of species possessing hydrophobic components such as proteins. These hydrophobic interactions are entropically favorable in an aqueous electrolyte, because water molecules are released from the solvation shell around hydrophobic analytes and are typically irreversible in an aqueous electrolyte under mild conditions. Therefore, many antifouling strategies aim at reducing fouling by increasing the hydrophilicity of the electrode surface and thus limiting or even preventing direct contact of antifouling compounds with the electrode-anchored bioreceptors to minimize false positive responses [35, 36].

Another concept was lately put forward by Hwang et al. [37] using a deformed monolayer graphene channel (Fig. 4c). Computational simulations revealed that the nanoscale deformations could form “electrical hot spots” in the sensing channel which reduce the charge screening at the concave regions. The increased Debye length at the convex region of the crumpled graphene brings more DNA strands inside the Debye length, making crumpled graphene electrically more susceptible to the negative charge of DNA and the change upon hybridization. This device achieved an ultra-high sensitivity of detection in buffer and human serum samples down to 600 zM and 20 aM, respectively, which correspond to ∼18 and ∼600 nucleic acid molecules. Moreover, the deformed graphene could exhibit a band gap, allowing an exponential change in the source-drain current from small numbers of charges.

Besides these materials and surface chemistry–based consideration, replacing a typical antibody probe of 10 nm with smaller bioreceptors, such as aptamers and nanobodies, is another way to limit signal screening issues (Fig. 4d).

Aptamers, artificial single-stranded oligonucleotides, typically 15–70 bases in length, and generally designed by Systemic Evolution of Ligands by Exponential Enrichment (SELEX), are well adapted for gFET biosensing due to their good thermal stability, low-cost (once the sequence has been identified by SELEX), and tuneable affinity to the analyte [38, 39]. With a probe length of < 5 nm for aptamer sequences of less than 30 bases, they are well adapted for gFET sensing even in higher salt solutions [40] (Fig. 4d). One of the first aptamer-modified gFET sensors is that reported by Ohno et al. [41] using an immunoglobulin E (IgE) aptamers with an approximate height of 3 nm and a KD = 47 nM. Kim et al. [42] showed that replacing a typical antibody receptor of 10 nm with a 4 nm aptamer probe, on otherwise similar gFET sensors, improved the sensitivity to the target antigen by 1000 times from 12 fM to 10 aM in 10 µM PBS (λD = 23.6 nm). The signal was, however, completely screened in 1 mM PBS (λD = 2.3 nm) even with the small aptamer probes [42]. The possibility to tune the aptamer density on gFET sensors was assessed by Hao et al. [40] by using a novel immobilization method based on the application of an electrical field (Fig. 5a). Using 1-pyrenebutanoic acid succinimidyl ester (PBASE) as a linker representative, application of an electrical field arranged the electron-rich pyrenyl group toward graphene, resulting in regularly aligned PBASE in the solution due to electrostatic repulsion. The LOD for interleukin-6 (IL-6) could be significantly improved to 618 fM, by applying an electric field at − 0.3 V for 3 h during PBASE immobilization [40].

Fig. 5

Aptamer-based gFETs: a IL-6 detection in 1 × PBS with and without applied electrical field during PBASE immobilization. Without an applied electrical field, PBASE is randomly immobilized on the gFET surface, while a negative potential arranges PBASE regularly with pyrene groups forced toward graphene (reprint with permission of ref. [40]). b Matrix metalloproteinase-9 (MMP-9) sensing in swab samples from patients with diabetic foot ulcers on an aptamer-modified liquid-gated gFET device recording a LOD of 478 pM in MMP-9 spiked wound fluid composed of 5.8 g NaCl, 3.3 g NaHCO3, 0.2 g KCl, 0.2 g CaCl2, 33.0 g BSA in 1L 1 × MOPS buffer/1 mM CaCl2 [27]. c Anchoring of cTnI-specific DNA or PNA aptamers on PEG-modified graphene and electrically determined dissociation constants against cTnI target [26]

We have lately validated that aptamer/PEG-based gFET architectures allow for the detection of metalloproteinases such as MMPs-9, zinc-dependent endoproteinases upregulated in non-healing wounds from 1.5 to 912 pM (0.1–60.8 ng mL−1). A LOD of 478 pM was determined in simulated wound fluids from the change in drain-source current at a VGate = 0 V [27] (Fig. 5b).

While DNA-based aptamers are widely employed as bioreceptors in gFET, the use of peptide nucleic acids (PNAs) has proven challenging, due to the absence of selection methods for the discovery of suitable bioreceptors and the difficult mimicry of established DNA-based ones. Despite the fact that PNAs exceed homologous DNA or RNA in terms of complementary base pairing, they can fail to reproduce alternative modes of binding, because of their different structural features. The remarkable stability and charge distribution of PNAs were lately proven to be beneficial for gFET-based sensing of cTnI (Fig. 5c) [26]. While the affinity of the DNA aptamer toward cTnI depends both on the ionic strength and pH of the medium, with KD increasing upon enhancing ionic strength and pH (Fig. 5c), the PNA aptamer recorded unchanged KD values under all experimental conditions. Though the PNA aptamer has not a dramatically higher affinity for cTnI and no better detection limit, 6.0 ± 1.0 pg mL−1 (PNA aptamer) and 3.3 ± 0.7 pg mL−1 (DNA aptamer), it does help in reducing the influence of pH and ionic strength parameters, showing more stable and consistent KD values, thus allowing for more adaptability to different conditions