In recent years, the need for highly sensitive, fast, label-free, portable, and low-cost biosensing platforms has motivated extensive research and different transduction strategies1,2,3,4,5,6. Among them, solution gated field-effect transistors (SGFETs) have become a current trend7,8 as they allow for sensitive and label-free identification of different analytes, such as proteins (including enzymes, antibodies and structural proteins) or nucleic acids9,10,11. This biosensing system is based on chemical conformation changes of the probe-target pair upon molecular recognition, which are transduced into an electrical signal. Recently, graphene's outstanding electrical properties and chemical inertness made it a valuable transduction platform for these devices12,13,14, resulting in improved limit of detection (LoD) and biocompatibility15. However, the poor stability on the surface of the immobilized probe biomolecule leads to graphene-SGFET (g-SGFET) biosensors lacking of the reproducibility and robustness required for stepping into the market16.
The fabrication of g-SGFET platforms requires graphene functionalization, which is a difficult process due to its inherent chemical inertness. Non-covalent functionalization 11,17 has been presented as a better alternative for the construction of g-FET-based biosensors, arguing that covalent chemical functionalization jeopardizes the aromatic lattice of the graphene network and thus its electrical performance18. Here, we use a highly controlled covalent functionalization strategy based on ion-sputtering in vacuum and p-aminothiophenol (pATP) linker molecules that preserves graphene properties and allows a precise control on the density of immobilized probe molecules19. In parallel, in vitro selected nucleic acid aptamers have shown several advantages over antibodies as molecular probes for developing biosensors with higher sensitivity, reproducibility and robustness20,21. In recent years, high-affinity aptamers that detect relevant molecular targets, including proteins and nucleic acids from pathogenic viruses, have been obtained22,23.
Here, we report a g-SGFET aptasensor with improved reliability, specificity and attomolar sensitivity to detect the hepatitis C virus (HCV) core protein, whose molecular probe is a chemically modified version of an in vitro evolved, high affinity DNA aptamer24. HCV is a very relevant human pathogen, the etiological agent of chronic hepatitis C25,26. It is estimated that around 58 million people have chronic hepatitis C worldwide and up to 1.5 million new HCV infections occur each year27.
The remarkable performance of our device is understood in the light of theoretical calculations showing that the covalently bound pATP linker is susceptible of the local polarization of graphene due to the proximity of the aptamer/protein system. The unprecedented sensitivity of the HCV core protein sensor is due to the linear energy-momentum band dispersion with vanishing density of the states at the Dirac point. This characteristic reduced electron density of states at the Dirac cone results in a high sensitivity of the Dirac point shift to a charge transfer at the graphene / molecule interface providing the base for the sensing mechanism. From this perspective the pATP linker can be considered as a molecular antenna, which is able to capture a subtle charge transfer at the linker/graphene interface driven by local polarization of graphene by proximity of the aptamer/protein system (Fig. 1a).
The global COVID-19 pandemic has made it evident that advanced ultrasensitive biosensors are required for early diagnosis of viral infections and, as a consequence, they constitute an essential tool for containing the spread of many infectious diseases. Our highly-controlled strategy to functionalize graphene platforms and the theoretical description of their complex mechanism pave the way for an extended use of g-SGFET covalent biosensors with a growing number of clinical applications. Additionally, we anticipate that the molecular antenna effect described here can be generalized to other graphene-based devices.
Ultra-sensitive aptasensor
The functionalization and testing process of the g-SGFET aptasensor developed in this work is schematized in Fig. 1b to d. The first step is the fabrication of the device, consisting of a sensor array of 20 graphene channels with a shared gate contact and an individually addressable channel, using clean-room technology28 (see methods). The golden gate electrode is covered by a self-assembled monolayer of dodecanethiol (DDT), following the protocol indicated in ref29 (see supplementary information). Once the chip is fabricated, it is placed in an ultra-high vacuum chamber (UHV) for the physical covalent functionalization of graphene, using a two-step protocol recently developed19,30 (Fig. 1b). This methodology preserves graphene's electronic properties associated with its atomic structure while ensuring atomically precise control and cleanliness. In the first step, single-atom vacancies are formed in the graphene layer by low-energy Ar+ ions. As a result, the inert graphene turns into a highly reactive surface due to the generated dangling bonds at carbon vacancies. At this moment, pATP linker molecules are evaporated in the UHV chamber by physical vapor deposition (PVD). The linker gets covalently coupled to the surface, its N atom anneals the vacancy, and the thiophenol motif adopts an upright configuration onto one of the neighbouring C atoms, exposing thiol-free groups towards the vacuum as explained below. This structural geometry is essential for the performance of the device. First, it will keep the aptamer (and, then, the aptamer/protein complex) at a short distance of the surface, and second, it will play the role of a molecular antenna that will locally induce polarization on the graphene in such a way that the g-SGFET will record a change in channel conductance.
Once all the induced graphene surface vacancies are covered with the linker, the g-SGFET is removed from UHV. We checked by a colorimetric method (details in the supplementary, Fig. S3) that the functionalized g-SGFET maintains its free thiol groups in a reduced and reactive state for months. Subsequently, a 10 μL drop of the renatured, thiol-modified aptamer AptD-1312 in a buffer solution (see methods) is placed onto the sensor and left to react in a humidity chamber for 30 min at 25 °C (Fig. 1b). The aptamer binds the linker via covalent disulfide bonds, and thus the functionalized platform becomes an aptasensor with high affinity and specificity for the biorecognition of HCV core protein. Each sequential adsorption step was followed by atomic force microscopy (AFM) images (see supplementary, Fig. S4). Analysis of the surface shows that the graphene surface is not damaged by the chemistry performed in every one of the steps indicated in Fig. 1b.
Attomolar detection of HCV core protein
Fig. 2a shows graphene's charge neutrality point variation detected as the transistor transfer curve minimum (VDirac) in the successive stages before biorecognition. The starting point (bare graphene) is slightly p-doped due to the exposure to atmospheric conditions and hence to the p-doping effect of oxygen31. Subsequent steps consistently shift the charge neutrality point towards zero gate voltage: first, when the DDT treatment is complete, and secondly, after covalent functionalization with the pATP linker molecule. When the g-SGFET is functionalized with the aptamer, VDirac slightly shifts in the opposite direction. Two chips are prepared for subsequent incubation of target molecules in buffer and in human plasma. Trends are similar in both chips (Fig. 2), indicating good reproducibility of graphene surface functionalization procedure. Further reproducibility experiments can be found in the supplementary information (Fig. S5).
For the analytic detection of the HCV core protein, the g-SGFET aptasensor was exposed to either a buffer or human blood plasma supplemented with HCV core protein at different concentrations (see methods). Fig. 2b and 2c show the characteristic transfer curves (TCs) obtained after HCV virus core protein incubation in buffer and plasma at different concentrations (from 10-18 to 10-11 M). In both cases, consecutive VDirac negative shifts result for every 10-fold increase of the core protein concentration, as shown in Fig. 2d, which plots the position of VDirac for the range of HCV core protein concentrations studied. This shift indicates an effective n-doping of the graphene sheet. The calibration curve shows a linear device response for protein concentrations over four decades or orders of magnitude, from 10-14 to 10-18 M, with a 14 mV/decade sensitivity. The saturation point in the buffer occurs at 10 fM. The calculated LoD is 15.6 aM.
We also exposed the biosensor to human blood plasma (which contains a great variety of proteins and other biomolecules) from a HCV-negative donor, supplemented with HCV core protein at concentrations between 10-18 and 10-11 M. Figure 2e shows the linear device response for HCV core protein concentrations spanning five decades, from 10-18 to 10-13 M. A sensitivity of 15mV/decade and a saturation point at 100 fM are found with the corresponding LoD at 90.9 aM. The result of this experiment is similar to that obtained with HCV core protein in a buffer solution, which shows that our aptasensor allows the detection of HCV core protein in human plasma with high sensitivity (attomolar level) and specificity (without side effects potentially due to the proteins or other biomolecules contained in the plasma). Therefore, the g-SGFET-based aptasensor developed here meets the requirements for testing in the clinical setting as a diagnostic tool for HCV in infected patients. The high reproducibility of the results has been demonstrated by measuring in different aptasensors as shown in the supplementary information (Figs. S6 and S7).
The specificity of the aptasensor to HCV core protein has been further studied by its response to another protein which is unspecific for the aptamer, bovine serum albumin (BSA), at a representative concentration of 10-14 M (see methods) (Fig. 3a and 3b). The measured VDirac did not show a significant variation in the presence of BSA. Another control experiment was performed by functionalizing the graphene surface with the DNA molecule D-ACTG (see methods). As in the previous specificity test, the measured VDirac showed no differences before and after exposition against the HCV core protein (Fig. 3c and 3d). Further specificity tests can be found in the supplementary (Figs. S8 and S9).
The molecular antenna amplification effect
To gain more insight into the understanding of the covalent functionalization of graphene and the operating mechanism of the aptasensor, we carried out density functional theory (DFT) and quantum mechanics/molecular mechanics (QM/MM) simulations (see methods) to monitor how different covalently linked functional groups affect the position of the Dirac cone relative to the Fermi level, i.e., the VDirac shift.
According to the total energy DFT calculations, incorporating pATP via its amine group into a single C atom vacancy of the bare graphene is energetically unstable. Namely, we found a configuration consisting of a nitrogen atom filing the vacancy and a covalently linked thiophenol group bound to a carbon atom next to the substitutional nitrogen (see Fig. 4a and Fig. S10), which is more stable (by ~ 1.8 eV) than the direct incorporation of an amine group directly incorporated into the vacancy. Moreover, the free activation energy for the shift of the thiophenol group from the N-defect to an adjacent carbon atom is almost negligible (dE = 0.14 eV). Based on these findings, we propose the following scenario: the direct incorporation of the amine group into the graphene vacancy is accompanied by an amine dehydrogenation process that destabilizes the covalent C-N bond on the phenyl group; this instability causes breakage of the C-N bond accompanied by the shift of the thiophenol onto the adjacent carbon atom, adopting an energetically more stable configuration (Fig. 4a).
The calculation of the following stages towards the biorecognition cannot be performed with the whole aptamer/protein system because the protein structure is unknown and the number of atoms involved is enormous. Thus, we used a simple model that mimics the underlying physical process occurring in the proximity of the sensor surface. An aptamer is interacting with its target protein through H-bonds (and other non-complementary interactions) established between the DNA backbone and nucleobases of the first and the amino acids of the later. It is striking that the mere presence of H-bonds induces a large shift of the VDirac, as charge cannot flow to (from) the graphene surface. To simulate this process, we have simplified the system to the minimum configuration that retains most of the physical features. Thus, we model the aptamer by its first three DNA nucleotides bound to the graphene layer through a pATP molecule. The presence of the target protein, to which the aptamer specifically binds through non-covalent bonds, is simulated by single DNA nucleobases that interact with the complementary nucleobases of the aptamer through H-bonds (Fig. 4b). Although it may seems surprising the use of nucleobases instead of amino acids to mimic the aptamer-protein complex, we have preferred to use those as the interaction of the apatamer with the amino acids from the protein is actually established through H-bonds, whose direction are dictated by the structure of the complex. As this structure is unknown in our case, the sole and simple way of simulating a stable H-bond interaction is the use of the complementary nucleobase in the calculation. Thus, we have considered cytosine in the aptamer structure and guanine to mimic at first approximation the interaction with an amino acid of the core protein.
Figure 4e shows the variation of the PDOS in the graphene channel along the different steps of the functionalization and detection process, namely the addition of: i) pATP linker; ii) aptamer; iii) aptamer with one complementary nucleobase bound; iv) aptamer with three complementary nucleobases bound relative to the pristine graphene. The Fermi level is aligned to zero energy to ease the direct comparison of the VDirac shift. According to the DFT simulations, the covalent functionalization of graphene by thiophenol and the incorporation of the substitutional nitrogen atom leads to charge transfer to the graphene, causing a slight shift of the Dirac cone above the Fermi level (Fig. 4e). Interestingly, this shift is opposite to that expected for a substitutional N-doping of graphene. In this case, there is charge transfer from the N-defect to the neighbor carbon atoms32,33, causing a shift of the Dirac cone below the Fermi level. The covalently linked thiophenol to a carbon atom adjacent to the substitutional N-defect overcomes this n-doping effect pushing the Dirac cone slightly above the Fermi level. Thus, the covalent bonding of the linker (thiophenol) and the presence of the substitutional nitrogen atom cause the direct charge transfer from the graphene sheet via the covalent bonding. The effect is also demonstrated by the intense polarization of graphene atoms near the covalent bond (Fig 4c and Figs. S11 and S12 of the supplementary). Interestingly, this local polarization of atomic charges surrounding the covalently bonded linker remains at first glance almost intact after further functionalization of the graphene (Fig. 4d, Fig. S11 and S12). On the other hand, we found that the shift of VDirac correlates nicely with the total accumulated atomic charge in the graphene (Fig. S12). The proximity of charged groups of the aptamer backbone (in particular, phosphate moieties) and nucleobases (amino groups) gives rise to local polarization of the graphene sheet (Fig. 4d), which modulates the position of the Dirac cone with respect to the Fermi level (Fig. 4e). Moreover, the simulations predict that adding more complementary nucleobases (as a proxy to the effect produced by the protein-ligand) leads to the further polarization of the graphene sheet underneath, causing additional downwards shift of VDirac.
To understand the detailed mechanism of this effect, we carried out simulations of a minimal system, approaching a polar guanine nucleobase oriented by its ketone group towards either the pristine graphene or the covalently functionalized graphene with the thiophenol group (Fig. 4f). The accumulated atomic charge remains zero in the pristine graphene and, consequently, the position of the Dirac cone does not change. However, in the case of the pATP-functionalized graphene, the approximation of the polar guanine causes an accumulation of the charge in the graphene (Fig. 4g), which shifts downward the position of the Dirac cone below the Fermi level. Note that in the absence of covalent bonding, no changes are expected, as experimentally confirmed in Fig. S9.
The local polarization of graphene by the proximity of the polar molecule disrupts the detailed charge balance between the graphene sheet and the pATP linker, causing the VDirac shift. Therefore, the linker behaves as a kind of molecular antenna, amplifying the local polarization in graphene's vicinity. Notably, the characteristic vanishing density of states of graphene near the Dirac cone makes its position very sensitive to the local charge transfer. This effect provides an optimal mechanism for biosensing and underlines the advantages of using aptamers instead of antibodies (whose size and molecular weight are much larger) as moleculecular probes in graphene-based biosensors.