Fast and direct identification of SARS-CoV-2 variants via 2D InSe field-effect transistors

2.1 Fast and ultra-sensitive detection of SARS-CoV-2 via 2D InSe FETs

2D InSe is a promising FET channel material with low effective electron mass (m_e* = 0.143 m₀), adjustable band gap (1.26–2.11 eV), and high intrinsic mobility (>10³ cm² V⁻¹ s⁻¹).^24-31 However, the deficient chemical stability of 2D InSe has greatly limited its potential applications in biosensing. In this work, an environmentally stable 2D InSe FET with real-time monitoring, high sensitivity, and sufficient stability was developed for the selective recognition of SARS-CoV-2 nucleic acids (Figure 1A).

The fabrication of 2D InSe FETs was achieved through a controlled two-step ALD of Al₂O₃ on 2D InSe flakes. As shown in Figures S1 and S2, the 2D InSe flakes with a thickness of 6–10 nm were obtained from bulk single crystals by mechanical exfoliation. The final structure of 2D InSe FETs (Figures 1B and S3) revealed a channel region exposed for the probe incubation and detection of the target NAs. The regular performances of FETs depicted in Figure S4B and S4C revealed mobility reaching 15.7 cm² V⁻¹ s⁻¹ at 25°C. The stability of devices greatly improved after Al₂O₃ encapsulation (Figure 1C). The I_d–V_gs curves measured every 12 h in phosphate-buffered saline (0.01× PBS) illustrated preserved on-state current (V_gs = 10 V) as high as 73.5% (N_I = I/I₀) of the original current, without significant decline in mobility after soaking in PBS for 48 h. These features would fully meet the device stability requirements for probe immobilization (36 h) and sensing in PBS (Figure 1D). Meanwhile, the electrical performance and stability of devices are also tested with different Al₂O₃ layers, of which the deposit cycle number should consider the balance of InSe passivation and oxidation (Figure S5). Afterward, high-density Au nanoparticles (Au NPs) were fabricated for efficient probe immobilization (Figure S6). The obtained Al₂O₃ layer prevented the devices from the common p-type doping effect and the positive shift in the threshold voltage (V_th) induced by Au NPs.^{32, 33}

During the detection of NAs by FET, a decrease in I_d was observed due to the phosphate skeleton of NAs generating one negative charge per base in PBS. The induced negative charges may ignore the electrical double layer shield within Debye length (λ_D) of 7.5 nm in PBS (0.01×).³⁴ Meanwhile, the thermal stability of target-probe duplexes influenced the proportion of target NAs hybridized and immobilized on FETs, resulting in variable equivalent negative gate voltages on channels (target-gating effect). After hybridization, Fermi energy (E_F) shifted downward, reducing the mobility and density of electrons, thereby resulting in different current drops (Figure 1 E,F, Table S1).

To obtain an accurate relationship between the concentration and the decrease in current, independent NA concentration measurements were conducted and the results are given in Figures S7 and S8. The bio-FET response was defined as a calibrated response (∆V_cal): $∆V_{\mathrm{cal}}=\left|∆I_{\mathrm{ds}}\right|/\frac{d{I}_{\mathrm{ds}}}{d{V}_{\mathrm{gs}}}$ that can diminish the device-to-device variations.^{35, 36} Here, ∆V_cal reflects the equivalent negative gate voltage induced by DNA or RNA (target-gating effect), with the value tending to stability and consistency after 10–15 min incubation (Figure 1G,H). The bio-FET exhibited a wide linear work range (10⁻¹⁴–10⁻⁸ M), covering the detection of both ultra-low and medium-low concentrations (Figure 1I). From the linear fitting data, the LOD was determined as 0.71 fM for DNA and 0.79 fM for RNA based on signal/noise = 3 (Figure S9). Although the long-term stability and repeatability need to be improved due to the strict requirement of the Al₂O₃ thickness (Figure S10), the bio-FET shows enough stability and robustness in a single test.

To confirm the target-gating effect of NAs and the target-probe hybridization efficiency, the surface potentials (Figure 1J–L, top) with or without DNA and RNA hybridization were directly detected through Kelvin probe force microscopy (KPFM) and the resulting band bendings are depicted in Figure 1J–L (bottom). The balanced state corresponded to the surface potential of Au substrate without duplex formation. During NAs detection, the formed negative potential served as the negative voltage to reduce the gate field effect. Accordingly, Fermi energy moved downward, reducing both the electron mobility and density. The DNA induced a larger potential difference (∆V_DNA) of −154 mV than that of the RNA (∆V_RNA = −111 mV, Figure 1K,L). Thus, the double-stranded DNA (dsDNA) generated a more distinct target-gating effect and significant responses than DNA–RNA duplexes due to the different thermodynamic stability of the duplex.

Compared to other diagnostic methods, such as RT-qPCR, genome sequencing, and electrochemical biosensors,³⁷ the proposed 2D InSe bio-FETs showed comprehensive advantages in terms of fast detection (10–15 min) and wide detection range (10⁶). These features may potentially facilitate the application of 2D bio-FETs in SARS-CoV-2 nucleic acids testing to distinguish asymptomatic cases and even subclinical infections in early-stage diagnosis.^{38, 39}

2.2 Direct identification of single-nucleotide variations

SNVs with multiple types of mutation are common in SARS-CoV-2 variants. For example, the R203M and R203K in Figure 2A represent in-situ mutations of Delta and Gamma variants, respectively. These variants have different nucleotide mismatched types but are in the same position. In this regard, the proposed 2D InSe FETs can be used to identify SNVs. As shown in Figure S11, FETs were employed to distinguish DNA sequences with different mismatch types in the same 6th position, and the device responses were compared with (non-)complementary sequences. As can be seen in Figure 2B, the ∆V_cal of complementary DNA far exceeded that of noncomplementary and mismatched sequences. All the responses became stable after 15 min, with the signal of C–C mismatch lower than those of C–T and C–A mismatch.

The reliability of the sensing results was verified by analyzing the (thermal) stability of mismatched DNA using high-resolution melting (HRM) analysis. The HRM technology can be employed to monitor the process of dsDNA separation (fluorescence signal attenuation) as a function of the temperature rise. In Figure 2C, the normalized reporter (R_n) of fluorescence intensity represents the dsDNA proportion in the solution, and T_m is the temperature corresponding to the melting curve for R_n = 0.5, considered a crucial indicator for evaluating sequences' thermal stability. The T_m of the complementary sequence was found 11°C higher than that of the C–C mismatch type. By comparison, the T_m values of C–A (55.7°C) and C–T (55.2°C) mismatch sequences were very close, which were highly consistent with the ∆V_cal of 2D bio-FETs (Figure 2D).

The mismatched nucleotide position is also crucial mutation information requiring determination. For example, the core mutation of D614G (Figure 2A) has been detected in Delta, Beta, and Gamma variants, meaning that the location of D614G can be used as a universal signature for identifying VOCs. Similarly, the states of duplexes formed by sequences with different mismatched positions were analyzed through HRM (Table 1). In Figure 2E, the HRM data illustrated the proportion of position-mismatched dsDNA remaining high at 40°C. However, the melting curves showed noticeable differences at 60°C. Note that T_d is the temperature at which the derivative value of −R_n reached a maximum, revealing the most intense separation under the random thermal motion. Hence, T_d was more suitable for revealing the difference in the slighter thermodynamics of mutations with the same mismatched type but various positions. As shown in Figure 2F, the T_d values of the 21st, 13th, and 6th mismatched sequences were determined as 65.3°C, 60.9°C, and 59.3°C, respectively.

Accordingly, the normalized responses of bio-FETs to mismatched DNA sequences were measured in different positions with the same C–T mismatch (Figure S12). At both incubation temperatures of 40°C and 60°C, the ∆V_cal to mismatched sequences in different positions were significantly lower than those of complementary sequences (Figure 2G). However, the average responses of sequences with different SNVs positions at 40°C overlapped when compared to the ∆V_cal measured at 60°C and cannot be identified in the range of 25%–30%. In line with the HRM curves, the exposure of FETs at 60°C resulted in a much more obvious difference in ∆V_cal. Especially, the ∆V_cal of 21st and 6th mismatched sequences presented significant differences (**p < 0.01, †p < 0.05), suggesting the hybridization difference of targets harboring different mismatch positions. Herein, the ∆V_cal of 2D InSe FET was quite consistent with the T_d obtained from HRM analysis.

The as-prepared bio-FETs presented temperature-resolved detection properties, and can directly distinguish SNVs with different types and positions from their (non-)complementary sequences. Since the stability of various target-probe duplexes can be affected by many factors, such as helix conformation and C–G base-pair proportion, the HRM for different target sequences was employed to guide the temperature setting modulation.

Based on the excellent performance of the proposed sensor for identifying SNVs of DNA sequences, it can be concluded that the 2D InSe FETs may be useful for detecting core mutations of the Delta variant. The first mutation L452R (S) accounts for virus escape from antibody LY-CoV555, making the Delta variant more transmissible.¹⁰ The R203M (N) is a signature mutation of the Delta variant, with in-situ mutations with different single-nucleotide types (R203K) found in the Gamma variant.

The good consistency between HRM analysis and device response indicated that HRM can be used to optimize the sensor temperature for the detection of the thermodynamic mismatch of specific sequences. For mutant RNA (R203M, N), the melting curves of complementary and mismatched sequences may be distinguished at 60°C, while RNA (L452R, S) curves looked different at 40°C (Figure 2H). Therefore, the bio-FETs were exposed to PBS (0.01×) containing two target RNA sequences at corresponding temperatures (Figure S13). The mean response of the RNA (R203M, N) sequence (30.2 mV) can clearly be distinguished from the mutant sequence (16.3 mV) at 60°C (Figure 2I). For the duplex of mutant RNA (L452R, S), the stability looked significantly different from that of a complementary duplex at 40°C. In Figure 2J, the average ∆V_cal of L452R (S) was estimated to be 43.2 mV, revealing a significant difference (†††p < 0.001) between complementary (68.0 mV) and noncomplementary (7.0 mV) sequences.

The temperature (T)-resolved identification ability could improve the viability of 2D InSe bio-FETs, thereby preventing complex probe design and interference of targets' spatial structures. The trinity of T-resolved, time (t)-resolved, and electrical (∆V_cal) response in our bio-FETs can also realize the variations typing without strictly limiting the mutation positions in the target sequence region. Hence, 2D InSe FET with good feasibility can be utilized to acquire multi-dimensional SARS-CoV-2 variants information for a viable prediction of COVID-19 and prevention of the hidden transmission of VOCs (Table S2).

2.3 Mechanism of realizing variants identification

After a step-wise ALD process (Figure 3A), an Al₂O₃ thin film was constructed as the protective layer on FET. The Raman spectroscopy analysis revealed the InSe channel exhibiting A_g mode at 309 cm⁻¹ corresponding to the In-O vibration (Figure 3B), indicating the formation of InSe_1−xO_x after ALD Al₂O₃.⁴⁰ The x-ray photoelectron spectroscopy (XPS) results further confirmed the presence of InSe_1−xO_x (Figure 3C). Evident broadening and downshift were observed in Se 3d and In 3d binding energy profiles, originating from the electron transfer from the unoxidized (InSe) layers to the oxygen-substituted (InSe_1−xO_x) layers.

To verify the influence of channel oxidation and subsequent improvement in sensing ability, the band structure and density of states (DOS) of pristine 2D InSe (black) and 2D InSe_1−xO_x (red) were calculated within the framework of DFT. In Figure 3D, the O atoms were used to replace Se atoms since Se vacancies (V_Se) may result in excess electrons from exposed In atoms to reduce the dissociation barrier of O (~0.2 eV) and enable the oxidation process.⁴¹ The corresponding oxidations were also directly reflected in the band structure (Figure 3 E,F). The replaced O atom effected the p-orbital of In atom and revealed significant changes in the InSe conduction band, specifically manifested as the band gap narrowing (1.52–0.36 eV) and transition from indirect band gap to direct band gap. After the replacement of the O atom, the DOS curves (Figure 3G) illustrated localized states of InSe_1−xO_x between E_c and E_v of InSe, where the corresponding energy levels acted as trap states to capture carriers.⁴²

Since the DNA and RNA sequences are negatively charged, the hybridization would create stable surface potential on the channel. For 2D InSe, the negative surface potential induced a slight downward shift in Fermi energy. The perturbation energy relationship was determined as ∆E_F ≪ E_c − E_F, so that n₀ can be considered constant. Thus, the dynamic changes of NAs with SNVs were too slight to influence the channels under the band transport mechanism.

However, the value of ∆E_F for 2D InSe_1−xO_x was not negligible when compared to E_t − E_F in Equation (2) since the location of E_t was much closer to E_F than E_c. Consequently, the negative potential could limit electrons from filling the shallow traps and suppress electron hopping transport. The effect of trap state filling restraint was further amplified through Equation (1) to yield a more detectable current drop. The results suggested significantly enhanced sensitivity of FETs with trap state-dominated channels, enabling the bio-FET to identify signals of SARS-CoV-2 mutations at a single -nucleotide level.

In short, the existence of trap states in InSe_1−xO_x during the ALD process was demonstrated, and the mechanism was explained.⁴⁵ These traps were generally known as a factor impairing the electronic conductivity, thereby should be avoided when fabricating devices. However, the negative factor possessed a unique positive side to biosensing since the carrier transport via trap filling and detrapping behavior exhibited robust and energetic sensitivity to ∆E_F caused by NA-induced equivalent gate bias.^{46, 47} Compared to other strategy for the protection and improvement of the sensor performance like 2D metal–organic framework (MOF),⁴⁸ the Al₂O₃ layer did not actively have the functions to enhance selectivity and sensitivity but effectively modulate the sensing property of 2D InSe channels. And the proposed strategy may free 2D materials from the strict fabrication requirement of perfect surface-states, and provide a scalable research strategy for expanding the functional diversity of 2D bio-FETs.